1

Mathematical

Methods

in

Chemical
Engineering

Digitized by tlie Internet Archive

in

2011

littp://www.arcliive.org/details/matliematicalmetliOOjens

Mathematica

Methods

in

Chemical
Engineering
V. G.

JENSON

Lecturer

Department of Chemical Engineering

Tfie University, Birmingliam, England

and
G. V.

JEFFREYS

Senior Lecturer

Department of Chemical Engineering

Faculty of Technology

The University, Manchester, England

1963

ACADEMIC PRESS
LONDON

and

NEW YORK

ACADEMIC PRESS

INC.

(LONDON) LTD

BERKELEY SQUARE HOUSE

BERKELEY SQUARE

LONDON, W.l

U.S. Edition published by

ACADEMIC PRESS
1 1 1

INC.

FIFTH AVENUE

NEW YORK, NEW YORK

Copyright

10003

1963 by Academic Press Inc. (London) Ltd

Third printing 1966

All Rights Reserved

No part of this book may be reproduced in any form by photostat, microfilm,

or any other means, without written permission from the pubhshers

Library of Congress Catalog Number: 63-22092

Printed in U.S.A.

PREFACE
The development of new
more complex and

is becoming
and development

processes in the chemical industry

increasingly expensive.

If the research

of the process can be carried out with confidence, the ultimate design will be
exact, and therefore the plant will operate more economically. In all
facets of such a project, mathematics, which is the language of the quantitative, plays a vital role. Therefore training in mathematical methods is of the
utmost importance to chemical engineers. The present text has been written
with these ideas in mind, and we would emphasize that our aim is to present
mathematics in a form suitable for the engineer rather than to teach engineering to mathematicians. To the pure mathematician an elegant proof is an
end in itself, but to the chemical engineer it is merely a means to an end.
Consequently this book only sketches demonstrations of the validity of
theorems, to encourage the reader to have more confidence in the use of the
technique for the solution of engineering problems. In addition, an attempt
has been made to sort out the useful from the trivial and flamboyant of the
wide variety of mathematical techniques available.
The material presented in this book is based on various undergraduate
and post graduate courses given in the Chemical Engineering Department of
the University of Birmingham. Many of the worked examples have been
selected from research work carried out in the department, and these are
supplemented with problems taken from the chemical engineering literature.
Some chapters of the book (notably Chapters 4 and 5) are almost entirely
mathematical, but wherever possible the text has been illustrated with chemical engineering applications.

more

The book was written when both authors were lecturers in the Chemical
Engineering Department of the University of Birmingham. It is hoped that
the text will encourage chemical engineers to make greater use of mathematics in the solution of their problems.
We wish to express our thanks to Professor J. T. Davies for initiating
the venture, and to Professor F. H. Garner, and Professor S. R. M. Ellis for
their interest and advice in the preparation of the text.
Birmingham
July,

1963

V.G.J, and G.V.J.

CONTENTS
PREFACE
Chapter

THE MATHEMATICAL STATEMENT OF THE PROBLEM

Introduction
Representation OF THE Problem
Solvent Extraction in Two Stages
Solvent Extraction IN A^ Stages
Simple Water Still WITH Preheated Feed
Unsteady State Operation
Salt Accumulation in a Stirred Tank
Radial Heat Transfer through a Cylindrical

1.1

1.2
1.3

1.4

1.5

1.6
1.7

....
....
.

1.8

Conductor

1
1

4
6
8
11

14
16

1.10

Heating A Closed Kettle

Dependent and Independent Variables, Parameters

1.11

Boundary Conditions

18

1.12

Sign Conventions
Summary OF the Method OF Formulation

21

1.9

1.13

17

19

Chapter 2

ORDINARY DIFFERENTIAL EQUATIONS

Introduction
Order and Degree

2.1

2.2
2.3

2.4
2.5

2.6
2.7

Order Differential Equations

Second Order Differential Equations
Linear Differential Equations
Simultaneous Differential Equations
Conclusions
First

....

23
23
24
33
41

66
72

Chapter 3

SOLUTION BY

SERIES

3.1

Introduction

3.2

Infinite Series

3.3

Power

3.4
3.5

Simple Series Solutions

Method of Frobenius

3.6

Bessel's

3.7

Properties of Bessel Functions

Series

Equation

74
74
79
86
90
106
113

CONTENTS

Vlll

Chapter 4

COMPLEX ALGEBRA
4.1

Introduction

4.2

The Complex Number

The Arg AND Diagram

4.3

4.6

Principal Values
Algebraic Operations on the Argand Diagram
Conjugate Numbers

4.7

De Moivre's Theorem

4.4
4.5

4.8

4.9

117
117
118
119
.

123

124
125
126

The mu Roots of Unity

Complex Number Series

Trigonometrical Exponential Identities

The Complex Variable
4.12 Derivatives OF A Complex Variable
4.13 Analytic Functions
4.10

.128
128

4.11

.130
131

4.14 Singularities

132

Integration of Functions of Complex Variables, and

Cauchy's Theorem
4.16 Laurent's Expansion and the Theory of Residues
4.

.120

137

.142

Chapter 5

FUNCTIONS AND DEFINITE INTEGRALS

5.1

5.2
5.3

5.4
5.5

5.6

Introduction
The Error Function
The Gamma Function
The Beta Function
Other Tabulated Functions which are Defined by
Integrals
Evaluation of Definite Integrals
.

149
149
151

154
157

.159

Chapter 6

THE LAPLACE TRANSFORMATION

6.1

Introduction

163

6.2

The Laplace Transform

The Inverse Transformation

163

6.3

6.4
6.5

167

Properties of the Laplace Transformation

The Step Functions

174
179

6.8

Convolution
Further Elementary Methods of Inversion
Inversion of the Laplace Transform by Contour

6.9

Integration
Application of the Laplace Transform to Automatic

6.6

6.7

Control Theory

.170

.180
182
188

contents

IX

Chapter 7

VECTOR ANALYSIS

7.7

Introduction
Tensors
Addition and Subtraction of Vectors
Multiplication of Vectors
Differentiation of Vectors
Hamilton's Operator,
Integration of Vectors and Scalars

7.8

Standard

7.1

7.2
7.3

7.4
7.5

7.6

Identities

Curvilinear Coordinate Systems

7.10 The Equations of Fluid Flow
7.11 Transport of Heat, Mass, and Momentum
7.9

199

200
203
210
216
218
222
227
228

231

236

Chapter 8

PARTIAL DIFFERENTIATION

AND

PARTIAL DIFFERENTIAL EQUATIONS

....

8.1

Introduction

8.2

Interpretation of Partial Derivatives

Formulating Partial Differential Equations
Boundary Conditions
Particular Solutions of Partial Differential
Equations
Orthogonal Functions
Method of Separation of Variables

8.3

8.4
8.5

8.6

8.7

8.9

The Laplace Transform Method

Other Transforms

8.10

Conclusions

8.8

....

238
239
245
252
259
269
272
290
302
306

Chapter 9
FINITE DIFFERENCES
9.1

Introduction

9.2

The Difference Operator,

Other Difference Operators.

9.3

9.4
9.5

9.6
9.7
9.8

Interpolation
Finite Difference Equations
Linear Finite Difference Equations
Non-Linear Finite Difference Equations
Differential-Difference Equations
.

307
307
311

315
321
322
331

338

CONTENTS

Chapter 10

TREATMENT OF EXPERIMENTAL RESULTS

Introduction
Graph Paper
Theoretical Properties
Contour Plots
Propagation of Errors

10.1

10.2
10.3

10.4

10.5

Curve Fitting
Numerical Integration

10.6
10.7

Chapter

349
349
354
355
356
360
369

NUMERICAL METHODS
11.1

Introduction

11.2

First

11.3

Order Ordinary Differential Equations

Higher Order Differential Equations (Initial Value
.

Type)
1 1

.4

11.5

11.6
11.7

380
380
385

Higher Order Differential Equations (Boundary

Value Type)
Algebraic Equations
Difference-Differential Equations
Partial Differential Equations

....

388
397
406
409

Chapter 12

MATRICES
12.1

Introduction

12.2

The Matrix
Matrix Algebra

12.3

12.4
12.5

12.6
12.7
12.8

12.9

Determinants of Square Matrices and Matrix

Products
The Transpose of a Matrix
Adjoint Matrices
Reciprocal of a Square Matrix

....
.

The Rank and Degeneracy of a Matrix

The Sub-matrix

12.10 Solution OF Linear

12.11

437
438
439

Matrix

Algebraic Equations

Series

12.12 Differentiation

and Integration of Matrices

Lambda-Matrices
12.14 The Characteristic Equation
12.15 Sylvester's Theorem
12.16 Transformation OF Matrices
12.17 Quadratic Form
12.13

....

443
443
444
444
446
448
448
449
451
452
454
457
459
461

contents

xi

Application to the Solution of Differential

Equations
12.19 Solutions of Systems of Linear Differential
Equations
12.20 Conclusions
12.18

463
.

465
472

Chapter 13

OPTIMIZATION
13.1

13.2
13.3

13.4
13.5

13.6

Introduction
Types of Optimization
Analytical Procedures

The Method of Steepest Ascent

The Sequential Simplex Method
Dynamic Programming

473
474
475
483
485
486

Chapter 14

COMPUTERS
14.1

Introduction

14.2

Analogue Computers
Active Analogue Computers

14.3

14.4
14.5

Passive

Digital Computers
Comparison of the Uses of Analogue and Digital

Computers

492
493
496
505
509

PROBLEMS

511

APPENDIX

532

SUBJECT INDEX

.543

Chapter

THE MATHEMATICAL STATEMENT OF THE PROBLEM

1.1.

Introduction

consists of performing experiments and

This may be done quantitatively by taking accurate
measurements of the system variables which are subsequently analysed and
correlated, or qualitatively by investigating the general behaviour of the
system in terms of one variable influencing another. The first method is
always desirable, and if a quantitative investigation is to be attempted it is

Nearly

all

applied

science

interpreting the results.

better to introduce the mathematical principles at the earliest possible stage,

may influence the course of the investigation. This is done by

looking for an idealized mathematical model of the system.
The second step is the collection of all relevant physical information in
the form of conservation laws and rate equations. The conservation laws of
chemical engineering are material balances, heat balances, and other energy
balances; whilst the rate equations express the relationship between flow
rate and driving force in the fields of fluid flow, heat transfer, and diffusion
of matter. These are then appUed to the model, and the result should be a
mathematical equation which describes the system.
The type of equation (algebraic, diff*erential, finite difference, etc.) will
depend upon both the system under investigation, and the detail of its model.
For a particular system, if the model is simple the equation may be elementary, whereas if the model is more refined the equation will be a more difficult
type.
The appropriate mathematical techniques are then appHed to this
equation and a result is obtained. This mathematical result must then be
interpreted through the original model in order to give it a physical signisince they

ficance.

In this chapter, only the simplest problems will be considered and the

The more complicated models

be introduced throughout the book in the chapters dealing with the
particular mathematical techniques which are needed for the completion of

ideas of simple models will be introduced.

will

the solutions.
1.2.

Representation of the Problem

simple example of the application of these ideas will be given first.

The apparatus shown diagrammatically in Fig. 1.1 is to be used for the continuous extraction of benzoic acid from toluene, using water as the extracting
solvent.

The two streams

and the mixture

vigorously,

are fed into a tank

is

then

pumped

where they are stirred

B where it is allowed

into tank

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

to settle into

are

removed

two

layers.

separately,

The upper toluene layer and the lower water layer

and the problem is to find what proportion of the

benzoic acid has passed into the solvent phase.

Benzene
+
benzoic ac

Water

q
OO

Benzene
+
benzoic acid

Water
+
benzoic acid

Single-stage mixer settler

Fig. 1.1.

The problem is idealized in Fig. 1.2 where the two tanks have been combined into a single stage. The various streams have been labelled and two
material balances have already been used, (a) conservation of toluene, and
(b) conservation of water. These flow rates have been expressed on a solute
R ft/min toluene

R ft^nnin toluene

>

X Ib/ft^benzoic acid

c Ib/ft^ benzoic acid

S ft/min

water
S ft/min

water

y Ib/ft-^ benzoic acid

Fig. 1.2.

Idealized single-stage solvent extraction

free basis to simplify the analysis.

The concentration of benzoic acid in
each stream has also been stated and this completes the mathematical model.
So far, it has been assumed that all flow rates are steady, and that toluene
and water are immiscible. Further assumptions are now made that the feed
concentration c remains constant, and that the mixer is so efficient that the
two streams leaving the stage are always in equilibrium with one another.
This last fact can be expressed mathematically by

y
where

is

= mx

(1.1)

the distribution coefficient.

The equation

is

now

derived from the model by writing

down

mass

balance for benzoic acid.

Input of benzoic acid

Output of benzoic acid

= Re Ib/min
= i?x -f 5y Ib/min

Since benzoic acid must leave at the same rate as

Rc =
The

pair of equations (1.1)

it

enters,

Kx^Sy

and

(1.2)

contain four

(1.2)

known

quantities

1.

THE MATHEMATICAL STATEMENT OF THE PROBLEM

c, m) and two unknown quantities (x, y), and they can be solved
unknowns as follows.
Re = Rx-\-mSx

(R, S,

the

Re

R + mS'

= mRc
R + mS

Therefore, the proportion of benzoic acid extracted

'-l

(1.3)
^

is

-^^

(1.4)

R + mS
= 12R, m =

Re

for

if 5"
1/8, and c = 0-1; then
and the proportion of acid extracted is 60%.
At this stage, it can be seen that even in this simple problem, two dimensionless groups which are characteristic of the problem, have arisen quite

As

a numerical example,

0-04,

0-005,

naturally as a result of the investigation.

(x

and

E=

Putting

= RlmS

(1.5)

Sy/Rc, equation (1.4) becomes

=
That

is, the proportion extracted

dimensionless group a.

1.3.

l/(a

+ l)

(1.6)

governed solely by the value of the

is

Solvent Extraction in

The above example

Two

Stages

now

be reconsidered, but two stages will be used

for the extraction of benzoic acid instead of one stage. Each stage still consists of two tanks, a mixer and a settler, with counter-current flow through
the stages. The idealized flow system is shown in Fig. 1.3 where the symbols
will

R
c

Stage
S

^1

Stage
2

*o

y-,

Fig.

^2

Idealized two-stage extraction

1.3.

have the same meaning as in the previous example, and the diff'erent concentrations in a particular phase are distinguished by suffixes. In accordance
with chemical engineering practice, the suffix denotes the number of the stage
from which the stream is leaving. The assumptions which were made above
are made again, and equation (1.1) is still vahd for each stage separately,
giving

yi

= mxi

benzoic acid mass balance

is

now

y2

Output of acid (Ib/min)

(1.7)

taken for each stage.

Stage
Input of acid (Ib/min)

= mx2
1

Rc + Sy2
Rx^

+ Sy^

Stage 2

Rx^
Rx2 + Sy2

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

The

fact that benzoic acid

gives the

must enter and leave a stage

at the

same

rate,

two equations

Rc + Sy2 = Rx^-hSy^
Rx^
Using equations

(1.7) to

= Rx2 + Sy2

ehminate x^ and y2, the inter-stage concentrations,

Rc + mSx2 = (Ryilm) + Syi

Ryjm = Rx2 + mSx2

and
Eliminating

>^i

between equations

R{Rc + mSx2)
R^c

.-.

(1.8)

and

(1.8)

(1.9)

(1.9),

= (R + mS) (Rx2 + mSx2)

= X2(R^+mRS + m^S^)

""^^

>^l=

R^ + mlLm^S^

mRc(R + mS)
n2^^pe,
^2e2
R^ + mRS + m^S

Again, the proportion of benzoic acid extracted

^'-''^

(1-11)

is

_ mS(R + mS)
Re ~ R^ + mRS + m^S^

Sy^

Introducing the dimensionless groups

equation (1.12) becomes
a+
a^

+a+1

(1.12)

and a again from equation

(1.5),

a^-l
a^-1

Using the same numerical example as before, i.e. S = 12R, m = 1/8, and
c = 0-1 then X2 = 0-021, ^^^ = 0-0066, and the proportion extracted is 79%.
A greater degree of extraction has thus been obtained with two stages
than with one stage, everything else being the same.
;

1.4.

Solvent Extraction in

Stages

This improved extraction leads to the consideration of more than two

stages in the extraction system. The algebraic treatment was quite simple for
one stage, only requiring the solution of two equations in two unknowns.
The application to two stages involved the solution of four equations in four
unknowns, and following the same procedure for A^ stages, it would be
necessary to solve 2N equations in 27V unknowns. This is too laborious and
would require an individual solution for every integer value of A^, and more
advanced mathematical techniques are obviously needed to reduce the work.
One method using matrix algebra will be illustrated in Chapter 12, but a
second method, using finite differences and anticipating the contents of

Chapter 9

will

be used here.

THE MATHEMATICAL STATEMENT OF THE PROBLEM

1.

The general arrangement is as shown in Fig. 1.4 where the flow rates of
two streams are still denoted by R and S, and the benzoic acid concentrations by X and y. The suffix notation is again used to distinguish between
the

yj

^2

\y

X,

^n-1

Yn-I

Ym-I 1

^N-2

N-1
^N

^N-1

n
S

Fig.

1.4.

^N

Idealized A^-stage extraction

the diff'erent states of each stream, the suffix denoting the stage

which the

This time, a benzoic acid material balance

the general stage n of the system.

applied to

stream has just

left.

= Rx _ + Sy +
= Rx + Sy

Input of acid (lb/mm)

Output of acid (Ib/min)

is

Since entry rate and exit rate must be equal,

Rx.,

The

+ Sy^,=Rx + Sy,

distribution coefficient equation (1.1)

.'.

and equation

(1.14)

yn

is

still

(1.14)

valid for

any value of n

'^^^n

becomes

Rx.i

+ mSx+i =

Introducing a again from equation

(R + mS)x

(1.5),

x_i+x+i =(oc+l)x
or in standard form,
x +

This

is

i-(a +

l)x,

+ x_i=0

(1.15)

a second order linear finite diff*erence equation and the

of solution will be discussed in Chapter

9.

The solution

is

method

quoted here for

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

completeness,

viz.

y^

and

this

may

extracted, E,

= mx, =

mc(a^^^-a")/(a^-'^-l)

be verified by substitution into equation

is

(1.15).

(1.16)

The proportion

given by

which gives equation

(1.13) for the special case

Table

1.1.

Proportion Extracted

(S =

60

in

2.

N Stages

1/8)

10

78-9

87-7

95-2

99-4

EiVo)

12/?,

m=

N=

Table 1 1 gives a few values of E for different values of A^ for the parsystem considered. This shows how the proportion extracted increases
with the number of stages, and indicates that more than ten stages are likely
to be wasteful whereas one stage gives a poor degree of extraction. This type
of problem will be continued in Chapter 13, where the most economical
number of stages will be determined by financial considerations.
.

ticular

1.5.

Simple

Water

Still

with Preheated Feed

Figure 1.5 illustrates a distillation apparatus consisting of a boiler B with

a constant level device C, fed with the condenser cooHng water. The steam is
condensed in
and collected in the receiver D. Some of the latent heat of

B
Fig.

1.5.

Water

still

with heat exchanger

evaporation is returned to the boiler by preheating the feed. Denoting the

condenser feed rate by F Ib/h and the temperature by Tq F, the exit water
temperature by rF, the excess water over-flow rate by Wlb/h and the

1.

distillation

THE MATHEMATICAL STATEMENT OF THE PROBLEM

by

rate

G Ib/h,

performance of the apparatus can be

the

investigated.

Input of water to the still (Ib/h)

Output of water from the still (Ib/h)
Output of steam from the still (Ib/h)

F=W + G

:.

= F
= W
= G
(1.18)

Btu/h, the latent heat of

If heat is supplied to the boiler at a rate
evaporation of water is L Btu/lb, and 0F is taken as the datum temperature,
a heat balance can be taken over the boiler.

= H+ {F- W)T
= 212G+LG
=
H + Gr (212 + L)G

Heat input (Btu/h)

Heat output (Btu/h)
.-.

(1.19)

where equation (1.18) has been used to eliminate W. Equation (1.19) contains two unknown quantities, G and J, and another equation is needed to
complete the description. This is obtained from a heat balance over the
condenser.

Heat gained by cooHng water (Btu/h)

Heat lost by condensing steam (Btu/h)

= F{T Tq)
= G{L + 212 T)

F(T-To) = G{L + 212-T)

In deriving equation (1.20) it has been assumed that the distillate
the exit water temperature.

Eliminating

between

and

(1.19)

(1.20)
is

cooled to

(1.20),

H
T=To + HIF

F(T-To) =

From

(1.19)

and

(1.21)

(1.21),

H=

G(2n + L-To-HIF)

FH
"

^^ F(212 + L-To)-H

^^'^^^

This equation gives the rate of distillation in terms of the heat input, and the
temperature and flow rate of the coohng water.
If an attempt is made to interpret equation (1.22) for a constant heat
supply rate and constant feed temperature, it is seen that as F is decreased,
G increases. In fact when F = i//(212+L Tq) it appears that G is infinite,
which is a physical impossibility. Reference to equation (1.21) resolves the
difficulty as follows. As F decreases, T increases; but T cannot exceed 212 F
as

it

leaves the condenser,

and
To
.-.

does not satisfy

equation (1.19) becomes
If

this

this gives the restriction

+ HIF=T ^212
F^H/(212-To)

inequahty,

will

H = LG
:.

G = HjL

(1.23)

remain constant at 212 and

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

The temperature restriction has a further influence in that the amount of

steam produced by the boiler exceeds the capacity of the condenser.
Denoting the rate of collection of distillate by Z), equation (1.20) becomes

F(212-To)

D = (212-ro)F/L

.*.

The complete

solution

is

for

F^

F(2n + L-To)-H

and

this solution is illustrated

when

i.e.

H 1(212- Tq)

FH

that

F=

(1.24)

thus given by equations (1.22), (1.23), and (1.24);

D = (212- To)F/L
and

= DL

for

F^HI{2n- To)

by the continuous Hne

This shows

in Fig. 1.6.

///(212 Fq), the rate of collection of distillate

is

maximum

H/L.
The above analysis has been made without reference to heat losses, and
an attempt to allow for these would lead to a much more compHcated model.

at a value

would suggest that heat losses in the feed line to

would be detrimental, and more serious at low values of F; but
heat losses in the vapour line can have two effects. If F is greater than the
optimum value, losses from the vapour line will be detrimental, but if F is
smaller than the optimum value, heat losses from the vapour line will actually
qualitative investigation

the boiler

increase the yield.

line

On

the basis of these considerations, the second, dotted

has been drawn in Fig.

Fig. 1.6.

1.6.

1.6.

Variation of distillation rate with feed rate

Unsteady State Operation

In the examples considered so far, the system has been in a steady state,
allowing the material entering the system to be equated to the material leaving
the system and this has always given algebraic equations. In unsteady state
problems, however, time enters as a variable and some properties of the
system become functions of time. In the application of conservation laws it
is no longer true that the rate of entry of material will equal the rate of exit,
since an allowance must be made for material accumulating within the

1.

system.

THE MATHEMATICAL STATEMENT OF THE PROBLEM

The general conservation law now becomes

INPUT -OUTPUT = ACCUMULATION

The example

of Section 1.2 will

now

be reconsidered as

it

(1.25)
starts

from the

following situation. Assume that the single stage contains V^ ft^ of toluene,
V2 ft^ of water, and no benzoic acid. Assume, as before, that the mixer is so
efficient that the compositions of the two liquid streams are in equilibrium at
all times, and, in addition, that a stream leaving the stage is of the same

composition as that phase in the stage

This

ibya^

v,,x

xlb/a^

Sftymin

^2'y

aymin
Ib/ft^

Time dependent

state of the

1.7,

aymin

aymin

Fig. 1.7.

which shows the

all illustrated in Fig.

y Ib/ft^

now

is

single-stage extraction

system at a general time

6,

where x and y are

functions of time.

Since the flow rates of water and toluene are constant, V^ and V2 will
remain constant. The conditions are always changing, and so the material
balance must be applied during a small time increment 36. Any function of
6 can be expanded by Taylor's theorem,t and this allows the state of the
system at a time 6 + 36 to be expressed in terms of its state at time 6. In this
case,

it is

helpful to

draw up

Table

1.2.

the following table.

Condition of the System Before and

After a Time Increment

77?^

Property of the System

Flow rate of toluene phase

Flow rate of water phase
Volume of toluene phase in stage
Volume of water phase in stage
Cone, of acid

in entrance toluene

Cone, of acid in

exit toluene

Cone, of acid in

exit

6+39

Vi

V2
c

'*>
dv

water

Amount

of acid in toluene layer

Vix

V.xArV.pe

Amount

of acid in water layer

Viy

V2y+V2pe

Mathematical functions

exist which cannot be expanded by Taylor's theorem as

but these do not normally occur in chemical engineering applications,
and the statement in the text should not cause any difficulty.

shown

in Section 3.3.7,

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

10

During a time

interval SO,

Input of acid

= Re 39

Output of acid

= r(x +
\

--^^S9)s0 + s( y
Y
J

2de

--^^do]d9

2d9

= Vi-77,S9+V2^S9

Accumulation

d9
Since input output

d9

accumulation

Rcd9-R(x + \%^^y^-^(y +

Taking the

^^^^V =

dx ^^
^1 ~r^^^

d9++
d9

dy

v.
Vi'^.SO
^d9

limit as 59 -^ 0,

Rc-R.-Sy==v/^^+V,f^

(1.26)

In stating the output, the arithmetic mean of the concentrations was taken,
but the later calculations show that this was not necessary; the same equation
(1.26) would have been obtained by using the concentrations at time 9 instead.
Only the first two terms were taken in the Taylor series for x and y,
and these are all that are necessary since the further terms which contain a
factor (S9y will disappear when the hmit 39-^0 is taken.
Equation (1.26) must now be solved in conjunction with the equihbrium
relationship (1.1). Eliminating ;j;,

Rc-Rx-mSx =

dx
+ mFi

dx
Fi

do

do

Rearranging,

dx

dO

Rc-{R + mS)x~
and

V\

+ mV2

integrating,

-\n{Rc-(R + mS)x}
'

R^^

"F. + mF,"^^

^^'^^^

integration A can be evaluated by using the given initial

system that the stage contains no benzoic acid at zero time; i.e.

The constant of
state of the

when

= 0,
x =
-\nRc_
" R + mS~
\nRc-]n{Rc-(R + mS)x} = (R + mS)9l(V,-^mV2)
9

:.

(1.28)

or in exponential form:

Re

R + mSW

,,

^^,

THE MATHEMATICAL STATEMENT OF THE PROBLEM

I.

11

= 0, satisfies the differenThis equation satisfies the condition x = 0, at

equation (1.26) for any positive value of 9, and when 0->oo,
x-i'RcKR + mS). This is the steady state solution which has already been
obtained in equation (1.3).
tial

1.7.

Two

Salt Accumulation in a Stirred Tank

tank
by the following example.
of water. A stream of brine containing 2 Ib/ft^ of salt is fed
into the tank at a rate of 3 ft^/min. Liquid flows from the tank at a rate of
2 ft^/min. If the tank is well agitated, what is the salt concentration in the
tank when the tank contains 30 ft^ of brine ?
further points are illustrated

contains 20

1.7.1.

ft^

Simple Treatment

Firstly, the following

simple model of this system can be set up. Liquid

it leaves, so that Hquid accumulates in the tank

enters at a faster rate than

of 1 ft^/min. The increase in tank contents is 30 20 = 10 ft^, and

hence the operation will last for 10 min.
During this 10 min, 30 ft^ of brine enter carrying 60 lb of salt, and 20 ft^
of brine leave. Assuming that the final concentration in the tank is X, and
that the concentration of the outlet stream increases linearly with time, the
following material balance can be estabhshed.
at a rate

Input of salt during 10 min (lb)

Output of salt during 10 min (lb)
Accumulation of salt in the tank

Using equation

(lb)

=
=
=

60
20 (X/l)
30Jif

(1.25),

60-10X = 30Z
X = l-501b/ft^

(1.30)

.-.

This is a very simple model of the system which yields an answer

by elementary algebra. It does, however, contain a further assumption
beyond those given in the question; the assumption concerns the time
variation of the outlet concentration.
This additional assumption need
not be made if a more detailed model is taken in a similar manner to that
in the previous section.
1.7.2.

More

The

Detailed Treatment

of the system at a general time 6 is shown in Fig. 1.8, where

are functions of 6. Again, a table can be constructed showing
the state of the system at time 6 and at time 6 + 39.

both

state

V and x

^'mm
3fty,
2lb/ft^

Ib/ft^

_^2ftymin
X Ib/ft^

Fig. 1.8.

Accumulation of

salt in

surge tank


MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

12

Table

1.3.

The Condition of the System Before and After a Time Increment

Property of the System

Brine input rate (ft^/min)

Input salt concentration (Ib/ft^)
Brine output rate (ft^/min)

2
2

2
2

Output

salt

Volume of

concentration

(lb/ft 3)

dV

(- de

Vx

Salt content of tank (lb)

"->

liquid in tank (ft^)

3d

\ /

dx

)\

de

During a time interval dO,

Input of brine
Output of brine

Accumulation of brine

= iv+-SO] V

2>de

259

de

dV
339-239 = -39

.-.

d9

material balance must also be satisfied by the

Input of salt
Output of salt

Accumulation of
Since input output

salt

=
=

3.239

2x39

= (^ +

-;i^

^^j (x

+ -^39) - Vx

accumulation,

dV

656-2x50

salt, viz.

dV dx

dx

=Vx + x^5e+ V--5e + -7^:77;^-^'^

do dU
du
d9

dV

6-2x =

d9

dVdx
dx
39
+F
+
d9
d9 d9

^^

(1.32)
^

term in equation (1 .32) vanishes. In deriving

concentration in the outlet stream at time 9 has been
used in the output term since, as in Section 1 .6, any correction to this contains
a factor 39 which will lead to its elimination when the limit 39-^0 is taken.
The first important point is now evident; the accumulation consists of two
terms. The one term is due to a volume change, the other is due to a con-

Taking the

limit 39 -^ 0, the last

this equation, the salt

centration change, and both are essential in the solution.

1.

of

THE MATHEMATICAL STATEMENT OF THE PROBLEM

Equations (1.31) and (1.32) must now be solved for x and

Equation (1.31) can be solved immediately;

Fas

13

functions

6.

V = A+e
But when 6

0,

V=

20

V=

:,

Substituting equations (1.31)

and

20 + 9

(1.33)

(1.33) into (1.32),

6~2x =

x + {20 + e) ^
dO

Rearranging,

and

dx

de

6-3x

20 + 6

integrating,

}\n{6-3x) = \n(20 + e) + B
where

is

the constant of integration.

9

But when
x

0,

(1.34)

|ln6 = ln20 + B
Eliminating B, and combining the logarithms,

= In (1 + 0-05^)
=
l-0-5x
(1 + 0-050)-^
x = 2-2(l + 0-05^)-2

-iln(l-0-5x)
/.
.-.

(1.35)

concentration in the tank at any time, and

equation (1.33) gives the volume of brine in the tank. From (1.33) it is seen
that F = 30 when 6 = 10, as in Section 1.7.1, and therefore from (1.35)

Equation (1.35) gives the

salt

.'.

Z=

2-2(2/3)'

X=

1-407 lb/ft'

(1.36)

Comparison of this result with the previous result (1.30), shows that the
first simple model has a significant error of 6-6% due to the additional
assumption of a hnear variation of x with 9; an assumption which is shown
to be incorrect by equation (1.35).
The two treatments of this example illustrate the important point that
the solution of a problem depends upon the choice of model. A simple
model needs many assumptions and yields an approximate answer quickly,
whereas a more complicated model needs fewer assumptions and yields a
more accurate answer by more advanced mathematical techniques. It is
evident that the most accurate result will be obtained by making the minimum
number of assumptions consistent with obtaining a tractable mathematical
problem. In this way the mathematical techniques which are available to an
investigator control the applicability of any theoretical prediction made by
him.
It is therefore desirable that the widest selection of mathematical
techniques should be at his disposal.

mathematical methods in chemical engineering

14
1.8.

Radial Heat Transfer through a Cylindrical Conductor

two problems the properties of the system were functions of

There are many problems of a similar nature, where the properties of
the system are functions of position instead of time, and the following
example illustrates the application of the method to this type of problem.
Two concentric cyhndrical metallic shells are separated by a solid
material.
If the two metal surfaces are maintained at different constant
temperatures, what is the steady state temperature distribution within the
In the last

time.

separating material ?

This

a steady state problem, but some of the properties of the system

is

depend upon the position at which they are measured. In this case, the
temperature (T) and the heat flow rate per unit area (Q) are both functions
of the radius (r), and the heat balance must be related to a space interval
between r and r + dr as shown in Fig. 1.9. Following the same procedure as
before, but considering variations of r instead of 6, Table 1.4 can be compiled

Fig.

Table

1 .9.

1 .4.

Radial heat flow through cylindrical conductor

of the System on Either Side of a

Space Increment

77?^ Condition

Property of the System

Temperature

Heat

transfer area/unit length

Radial heat flux density

Total radial heat flow

dr

T + fsr
dr

2n(r

Inr

Sr)

+f"

Q
InrQ

2n(r

Sr)(Q

^r)

1.

THE MATHEMATICAL STATEMENT OF THE PROBLEM

Considering the element of thickness

dr,

Heat input to inner surface

= InrQ

Heat output from outer surface

Accumulation of heat

Since input output

2n{r-\-dr)

I
\Q +

dQ

~T~ ^fj

accumulation,

2nrQ-2n(r + Sr)(Q +

rQ-rQ- r^ 3r -

:.

15

^sA=0

QSr

^-^{drY

Cancelling out rQ, dividing by Sr, and taking the limit Sr-*0,

r^ + Q =
But

is

related to

(1.37)

T by

2=-^^
where k

is

Equation

the thermal conductivity.

d^T

dT

d^T
,

becomes,

dT

=
+__
"^

(1.39)

dr

dr^

(1.37)

dr

dr^
or

(1.38)

Referring to Table 1.4, the change in heat flow rate

two terms in this example, one due to a
change in temperature gradient, and the other due to a change in the area
of conduction.
This problem can be solved by treating (1.39) as a homogeneous second
order differential equation (see Section 2.4.3), or by solving equations (1.37)
since

is

constant.

across the element gives rise to

and

(1.38) in succession.

Equation (1.37)

dQ_

gives,

_dlr

"""^

e"
.*.

or

where In

is

\nQ= -\nr + \nA

Q = A/r

chosen for the constant of integration. Putting

this value

of

into equation (1.38),

--

-k

dr

dr

kdT

\nr= -{kTIA) + B

(1.40)

MATHEMATICAL METHODS

16

CHEMICAL ENGINEERING

IN

Equation (1.40) involves two constants of integration,

can be evaluated from the fixed boundary temperatures.

At

at

and

T=
T=

a,
b,

B, and these

and

\na-\nb = k{T^-To)IA
A = k(T, - To)/(ln a-\nb)

.-.

B = (T,\na-To\nb)l{T,-To)

.-.

Putting these values of

- To) In r

^
=

B into equation (1.40),

(In 6 - In a)T+ T^ In a-To\nb
and

= {T-To)(\nb-\na) + (T,-To)\na
TTn In r Ina
In 6 - In a
Ti - To
1.9.

^^'^^^

T,}

lna= -{kTolA)-^B
\nb= -(kTJA) + B

.-.

(Ti

=
=

A and

(1.42)

Heating a Closed Kettle

The problems considered

so far, have only used material or heat balances

of the equations. In the following example, the equation is
derived from a rate equation.
closed kettle of total surface area A ft^ is heated through this surface
lb
by condensing steam at temperature T^ F. The kettle is charged with
of liquid of specific heat C Btu/lb, at a temperature of To^F. If the process
is controlled by a heat transfer coefficient h Btu/h ft^ F, how does the temperature of the liquid vary with time ?
in the derivation

Consider a time interval 39.

Heat output (Btu)

= hA(T,-T)Se
=

Accumulation (Btu)

Heat input (Btu)

Since input output

MC -77.39

accumulation,

hA(T,- T)39 =

MC--39

(1.43)

du
Rearranging,

dT
T,-T
and

hAd9

MC

integrating,
In

(T,-T) = (hAIMC)9 + B

(1.44)

1.

THE MATHEMATICAL STATEMENT OF THE PROBLEM

known

can be evaluated by using the

fact that

= 0,T =

when

17

Tq

-\n{T,-To) = B

.-.

MC

T.-To

In this problem, equation (1.43) could have been written down immediately from the definition of the heat transfer coefficient, thus avoiding
the necessity of taking a heat balance.
1.10.

Any

Dependent and Independent Variables, Parameters

is an algebraic equation involving

These symbols fall into three classes, (a) independent variables,
dependent variables, and (c) parameters.

solution of a differential equation

symbols.
(b)

1.10.1.

Independent Variables

These are quantities describing the system which can be varied by choice
during a particular experiment independently of one another. Examples
are time and coordinate variables.
1.10.2.

Dependent Variables

These are properties of the system which change when the independent
variables are altered in value. There is no direct control over a dependent
variable during an experiment. The relationship between independent and
dependent variables is one of cause and effect; the independent variable
measures the cause and the dependent variable measures the effect of a
particular action. Examples of dependent variables are temperature, concentration,
1.10.3.

and

efficiency.

Parameters

This

is

by

far the largest group, consisting

mainly of the characteristic

and the physical properties of the materials.

The group contains all properties which remain constant during an individual
experiment, but since different constant values can be taken by a property
during different experiments, the correct term for them is " parameters ".
Examples are overall dimensions of the apparatus, flow rates, heat transfer
coefficients, thermal conductivity, specific heat, density, and initial or
boundary values of the dependent variables.
Referring to Section 1.6, the solution to the problem of starting a single
stage hquid-liquid extraction system is given by equations (1.1) and (1.29).
properties of the apparatus

^c

and by ehminating

= mx

exp

I-

(1.1)

R + mS

\1

e]

(1.29)

x,

mRc

R + mS

X]

MATHEMATICAL METHODS

18

The symbols can be

IN

CHEMICAL ENGINEERING

classified as follows

Independent variable

Dependent variables

x,

Parameters

m, R,

y
S,

c,

V^, V2

The parameters

are all fixed in value during an experiment, but can be varied

between experiments for comparison purposes.
is the only variable whose
value can be altered during an experiment, x and y vary during an experiment and between experiments; their values depend upon the choice of
values for both the parameters and the independent variable.

The main use of

tiation process.

We

this classification is in the interpretation

of the differen-

usually differentiate a dependent variable with respect

an independent variable, and occasionally with respect to a parameter.

differentiating either (1.29) or (1.47) with respect to 0, an expression giving
the rate of change of concentration with time can be obtained, but differentiating (1.1) with respect to the dependent variable x, gives dyjdx = w, an
almost useless piece of information which does not throw any hght on the
behaviour of the apparatus.
So far, the problems considered in Sections 1.2, 1.3, 1.4, and 1.5 have
been equations between parameters since no independent variables were
involved, and hence no dependent variables could be present. Each experiment found a set of related constant values, the only variation could be
between experiments when some of the parameters may be altered in value.
The problems of Sections 1.6, 1.7, and 1.9 involved time as an independent
variable, and in Section 1.8, a radial coordinate was the single independent
variable.
These problems all gave rise to ordinary differential equations.
When more than one independent variable is needed to describe a system,
the usual result is a partial differential equation, and this type of problem
will be dealt with in Chapter 8.
to

By

1.11.

Boundary Conditions

An

ordinary differential equation usually arises in any problem which

involves a single independent variable. The general solution of this differential equation will contain arbitrary constants of integration, the number of
constants being equal to the order of the differential equation (see Chapter 2).
To complete the solution of a particular problem, these arbitrary constants
have to be evaluated.
In formulating the equation, the conservation law (1.25) is applied to an
infinitesimal increment of the independent variable, and this yields a
differential equation. There is usually a restriction on the range of values
which the independent variable can take and this range describes the scope

of the problem. Special conditions are placed on the dependent variable at

these end points of the range of the independent variable. These are naturally
called " boundary conditions ", and are used to evaluate the arbitrary constants in the solution of the differential equation.

1.

THE MATHEMATICAL STATEMENT OF THE PROBLEM

19

These conditions have already been used in the examples considered

In Section 1.8, a second order differential equation

eadier in this chapter.

was derived, and its solution (1.40) contains two arbitrary constants
B. The independent variable in this problem is r, and the boundaries
The boundary conditions are given by
are given by r = a, and r = b.
equations (1.41) in the form of restrictions on the dependent variable (T) at
the boundaries defined by values of the independent variable (r).
In the other examples in Sections 1.6, 1.7, and 1.9, 6 was the independent
variable, all of the equations were of first order, and each solution only
contained one arbitrary constant. When time is the independent variable,
the boundary condition is frequently called the " initial condition ". This
(1.39)

A and

is

because

it

of the apparatus.

specifies the starting state

most frequently used boundary conditions in heat transfer are

(1) Boundary at a fixed temperature, T = Tq.
(2) Constant heat flow rate through the boundary, dT/dx = A,
(2a) Boundary thermally insulated, dT/dx = 0.
(3) Boundary cools to the surroundings through a film resistance
described by a heat transfer coefficient, kdTjdx = h{T Tq). k is the thermal
conductivity, h is the heat transfer coefficient, and Tq is the temperature of

The

three

the surroundings.

Boundary conditions
be described

will

in

when they

problems involving partial

arise in Chapter 8.

1.12.

Most

differential equations

Sign Conventions

sign difficulties are a result of trying to think too deeply about them,

The temptation to anticipate the

very strong but it must be resisted. The formulation
of equations can be reduced to a set of rules which can be systematically
apphed and although it is difficult to frame such a set of rules to cover the
wide variety of problems which can arise, an attempt has been made in the
next section. The slavish observance of rules is not to be generally recomthus causing a state of complete confusion.

solution of the problem

is

mended, but it is best in the first instance to do just this because it completely
eliminates any possibiHty of anticipating properties of the solution, and
ensures that the signs are correct. The resulting equation can frequently be
subjected to a physical sign check after it has been estabHshed.
All terms must be treated as positive during the formulation, and negative
signs will only occur due to two causes.
(a) The first cause is the negative sign in the general conservation law
(1.25), viz.

Input Output

Accumulation

The second cause

is by the definition of the rate equations governing

heat transfer, mass transfer, and fluid flow. This will be illustrated below for
the case of heat transfer.

(b)

When an

independent variable

is

defined for a system, an origin and

positive direction have to be included in the definition.

This has not been a

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

20

problem so

serious

direction

is

far,

since

when time

is

the independent variable,

the natural one of successive events

and

its

origin

is

its

always

taken as the instant when the initial conditions are estabhshed. The radial
coordinate (r) the other independent variable which has occurred is measured
outwards from the axis of the body of revolution, the axis being the origin.
In other cases it is always necessary to specify the origin of the independent

and also its positive direction.

Turning to heat flow, the quantity of heat (Q) conducted past a point
through unit cross-sectional area in unit time is counted as positive if it
flows in the same direction as the positive direction of the independent
variable (coordinate), and negative if it flows in the negative direction of
the coordinate. According to Fourier's Law, the rate of conduction of heat
is proportional to the temperature gradient, and the direction of the flow is
down the temperature gradient as shown in Fig. 1.10 where the two possible
variable

f + ve/

-^-ve

\^^

1
1

O + ve

O-ve

^x
Sign convention in heat transfer

Fig. 1.10.

cases are shown.

On

the

left

and as a

hand

side of the diagram, the temperature

heat flows from right to

left and is
diagram where the
temperature gradient is negative and the heat flow positive. In both cases,
the equation can be written in the consistent form

gradient

is

negative.

positive

The

reverse

is

result, the

true in the right

hand

side of the

(1.48)

dx
where k

This equation has already been used (1.38), but

was of no importance.
For mass transfer, material diff^uses down a concentration gradient, and
for fluids, the flow is down a pressure gradient. By analogy with the Fourier
Law, these rate equations also contain a negative sign in the definition.
In conclusion, if the rules which have been illustrated in this chapter are
followed, and the solution is not anticipated, the signs will take care of themselves. This can be illustrated by repeating the example of Section 1.7 with a
diff'erent boundary condition. Instead of the tank containing 20 ft^ of water
The only
initially, let it contain 20 ft^ of brine of concentration 5 Ib/ft^.
is

the conductivity.

in that application the negative sign

physical difference that this makes,

is

that the brine concentration will

now

1.

THE MATHEMATICAL STATEMENT OF THE PROBLEM

21

decrease with time instead of increase. Mathematically, only equation (1.34)

need be altered. The list of properties in Table 1.3 will be unaltered, and the
change in salt concentration with time is still adequately described by

dxjdO without

artificially

changing

its

sign.

In this

new problem, dxjdO

is

a negative quantity since x decreases with time and it would be a mistake

further
to say that the output salt concentration dXO + dO is x-(dxld9)S6.
which
is the
theorem
Taylor's
from
seen
mistake
can
be
why
this
is
a
reason

now

governed by the sign of 39 which is positive.

modified problem, and the solution is
equation (1.34) is reached, and this now becomes:

origin of this term; the sign

Table

is

1.3 is therefore still true for this

identical until

=5
-|ln(-9) = ln20 + 5

when
.-.

0,

(1.49

Here, the logarithm of a negative number is involved due to integrating

3a')~^ in an unconventional manner. This need not give rise to any
difficulty if In (-1) is treated as an imaginary quantity by analogy with
(6

J( 1), and B

is

ehminated as before by subtraction.

:.

\n(^^) = -3 In (1 + 0-05^)
.-.

This

how

is

= 2 + 3/(1 + 0-050)^

and it illustrates
and how the diff'erent
appropriate particular solutions from the general

the correct solution to this modified problem,

signs can be allowed to take care of themselves,

boundary conditions

select

solution of the diff'erential equation.

1.13.
(1)

about
(2)

Summary of the Method of Formulation

Pick a mathematical model of the process by making assumptions

ideal behaviour.
Define dependent variables to measure the properties being

its

investigated.
(3) Define one or more independent variables in terms of which the
dependent variables can be expressed.
(4) Define the parameters of the system which particularize the general
relationship between the dependent and independent variables.
(5) Establish the state of the system at a typical point defined in terms
of the independent variables.
(6) Take an infinitesimal increment in each of the independent variables,
and establish the state of the system at neighbouring points in terms of the
state at the typical point by using Taylor's theorem.
(7) Apply the relevant conservation laws and
ment of the system defined by (6).

rate equations to the ele-

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

22

Cancel out terms where possible, including the incremental intervals.

the Hmit as the size of the increments tends to zero.
These rules include the possibility that there may be more than one
independent variable, resulting in the formulation of a partial differential
equation. The treatment of such problems will be given in Chapter 8.
Although only the first two terms of Taylor's theorem are usually needed
in practice, the infinite series of terms is effectively included, since the later
terms involve incremental quantities raised to high powers and these disappear
when the incremental quantities tend to zero.
The solution given by the analysis is related to the model of the system
being investigated, and not to the system itself. The assumptions made in
order to derive an equation for the model may or may not be vahd, and if the
result does not agree with the experiment, then the model is probably wrong
and not the mathematics. As shown in Section 1.7, a more complicated
model gives a more accurate result, but frequently, if the model is too detailed,
the equation derived defies rigorous mathematical analysis. This difficulty
can be overcome by using a numerical method to find an approximate
solution to the difficult equation as shown in Chapter 11, or by simplifying
the model with a further assumption. It must be realized, however, that as
the number of assumptions increases, the likeUhood of the result agreeing
with practice decreases.
Tables 1.2, 1.3, and 1.4 have been compiled in this chapter to facilitate
the description of the method of formulating equations. These tables should
be compiled mentally and throughout the rest of the book they will be
omitted. If any reader finds difficulty in formulating the equations in the
rest of the book, it is suggested that similar tables should be constructed.
(8)

(9)

Take

Chapter 2

ORDINARY DIFFERENTIAL EQUATIONS

Introduction

2.1.

An

equation relating a dependent variable to one or more independent

variables by means of its differential coefficients with respect to the independent variables is called a differential equation. If there is only one independent variable the equation is said to be an "ordinary differential equation".
If there are two or more independent variables and the equation contains
differential coefficients with respect to each of these, the equation is said to
be a "partial differential equation". Thus

is

an ordinary

differential

equation whereas
;

"^

"*"

<'''^Te='
is

dx^

dy"-

(2.2)

5z^

a partial differential equation. Equation (2.2) contains the partial differenof T with respect to each of the independent variables 0, x,

tial coefficients

y,

and

z.

Partial differential equations will be dealt with in

this

Chapter

8,

whilst in

chapter attention will be confined to ordinary differential equations.

2.2.

The order of a

differential

differential coefficient that

it

Order and Degree

equation

contains.

highest differential coefficient

is

three

is

equal to the order of the highest

Thus in equation (2.1) the order of the

and therefore equation (2.1) is a third

order differential equation.

The degree of a

power of the highest order

differential coefficient that the

equation is the highest

equation contains
after it has been rationalized.
Equation (2.1) is of the first degree, even
though it contains the first order differential coefficient raised to the second
power and the cube of the independent variable. However, as the first order
differential coeflicient is squared, equation (2.1) is referred to as a third order
first degree non-linear ordinary differential equation.
It is non-linear because the equation contains the term {dyldxY and generally differential
equations are said to be non-linear if any products exist between the dependent variable and its derivatives, or between the derivatives themselves. Thus
the following equations are non-linear.
23
22
differential

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

24

^y

+
.

dx

,2

>'^

2,.n

sinA:

(2.3)

dy

-+y=

(2.4)

dx'

Equation (2.3) is non-linear because the dependent variable y appears as y^,

whereas equation (2.4) is non-linear because it is of the second degree.
The standard methods of solving the simplest types of ordinary differential equations that occur very frequently in chemical engineering will be
considered in the present chapter, whereas more difficult types of ordinary
differential equations will be considered later.

2.3.

First

Order Differential Equations

First order equations contain

no

derivatives higher than the

but
Unfortunately there is no general method of solution of these equations because of
their different degrees of complexity, but it is possible to classify the different
types of first order equations so that each can be solved by a particular
procedure. These are
relate the first order differential coef!icient to the

(a)

(b)

Of

two

first,

variables.

Exact equations.
Equations in which the variables can be separated.

(c)

Homogeneous

(d)

Equations solvable by an integrating factor.

these (d)

is

the

equations.

most important and consequently

(a), (b)

and

(c) will

be

treated briefly.
2.3.1.

Exact Equations

Occasionally, a differential equation can be expressed in the form

df(x,y)

where
is

df(x,y)

is

an exact

(2.5)

differential coefficient of

some

function.

When

this

so, the solution is obviously,

f(x,y)

where

C is

(2.6)

an arbitrary constant which must be evaluated from the boundary

conditions of the problem.

an exact

=C

Frequently

differential coefficient

it is

possible to

by inspection. Thus

x^
+y=
dx

show

that df{x,y)

is

in the equation,

(2.7)

is immediately seen to be the exact differential coefficient

of xy with respect to x, and therefore the solution is

the left-hand side

xy

=C

2.

ORDINARY DIFFERENTIAL EQUATIONS

25

However, consider the equation,

+ 2v)-/ =

A''-3sinx + (cosx

(2.8)

cix

Only an experienced mathematician would recognize equation (2.8) to be an

exact equation, but the reader familiar with the elementary concepts of
partial differentiation could easily verify that equation (2.8) is an exact
equation as follows.
(p = (t>{.\',y), equation (8.16) shows that

d^Jldx + ^-^dy
Comparing

this

with equation

(2.9)

cy

dx
(2.8),

C(j)

- = x 3 jsmx
ex
C(f)

z^ = cosx + 2y

and

cy

The most general function which can be

to

A'

to give the

first

of these expressions

cl>

differentiated partially with respect

is

ix^ + ycosx+f(y)

(2.10)

Differentiating equation (2.10) partially with respect to

= cos x

cd)

4- /'(>')

cos X

y gives

2y

cy
.*.

The solution of equation

(2.8)
x'^

2.3.2.

Equations

in

/(>')
is

y^

constant

therefore

+ 4ycosx +

4y^

=C

(2.11)

which the Variables can be Separated

In the most simple

order differential equations, the independent

can be separated from the dependent variable
and its differential by the equahty sign, using nothing more than the normal
processes of elementary algebra.
When this is possible, solution of the
equation follows by straightforward integration of the rearranged equation.
The example in Section 1.6 was of this type and as the procedure of solving
is elementary, no further discussion of these equations will be made.
It will
suflice to state that they frequently appear in engineering problems.

variable

2.3.3.

and

its

first

differential

Homogeneous Equations
differential

equation of the tvpe,

is termed a homogeneous differential equation of the first order.

Such an
equation can be solved by making the substitution y = vx and thereafter

MATHEMATICAL METHODS

26

IN

integrating the transformed equation.

CHEMICAL ENGINEERING

Thus

let

vx

(2.13)

and therefore

dy

dv

Substitution in equation (2.12), rearranging and integrating, gives

lnx=

\-^ + C

(2.15)

Equation (2.15) is the solution of the general homogeneous equation (2.12).

should be pointed out, however, that the integral in equation (2.15) is not
always expressible in terms of elementary functions. Thus the final solution
after substituting for v may not be very useful in practice even though it is
It

correct.

Homogeneous
reactor analyses

differential equations of first order arise in

when

batch chemical

the materials undergo chemical reaction by two

simultaneous second order reaction paths. A typical example is the halogenation of a hydrocarbon and the method of treating such a problem is
illustrated in the example below.

Example

1.

Liquid benzene

is

to be chlorinated batchwise

chlorine gas into a reaction kettle containing the benzene.

contains such an efficient agitator that

all

by sparging

If the reactor

the chlorine which enters the

reactor undergoes chemical reaction, and only the hydrogen chloride gas

how much chlorine must be added

monochlorbenzene. The reaction can be
assumed to take place isothermally at 55C when the ratios of the specific

liberated escapes
to give the

from the

maximum

vessel, estimate

yield of

reaction rate constants are

^ = 8-0
as published

reaction

I,

and

^ = 30-0

by Macmullen (Chem. Eng. Prog. 44, 183, 1948). k^

II, and k^, to reaction III below.

refers to

k2 to reaction

C6H6 + Cl2->C6H5Cl + HCl

CgHsCl + Cl2->C6H4Cl2 + HCl
C6H4CI2 + Cl2-^ C6H3CI3 + HCl

II

III

Solution

Take a basis of 1 -0 mole of benzene fed to the reactor and introduce the
following variables to represent the state of the system at time 6,

p mole
q
r

s
t

=
=
=
=

mole
mole
mole
mole

fraction of chlorine present.

fraction of benzene present.

fraction of

monochlorbenzene present.

fraction of dichlorbenzene present.

fraction of trichlorbenzene present.

2.

Then,

q-\-r

and the

total

27

ORDINARY DIFFERENTIAL EQUATIONS

amount of

+ s + t=l

consumed

chlorine

IV
given by

is

= r + 2s + 3r
volume V is constant.

>'

Now

assume that the reaction

is given by

Therefore, the rate of

reaction of benzene

and the

rate of formation of each

product

is

vfg = k,pq-k,pr

VII

VYQ = k,pr-k,ps

VIII

V%^k,ps

IX

can be eliminated as a variable by dividing equations VII to IX

by equation VI. The first ratio gives

The time

k^q

dq

Sq

This is now a first order homogeneous

solved by the substitution technique.

Let

dr

=
dq

and therefore
Substitution of

XI and XII

K is

Removing

XI
dv

+ Q-r

XII

dq

X and integration of the

= [nK- 8/7 In (Iv + 8)

result gives

XIII

the constant of integration.

the logarithms,

But when ^

0,

and

fir
-+8
K[-

=
=

.-.

and

equation which can be

= vq

into

In q

where

differential

8/7

XIV

0;

K=

8/^

after rearrangement,
r

Similarly,

S{q'^^-q)ll

from equations VI and VIII

it

XV

can be shown that

ds

dq

240q

Sq

XVI

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

28

When

has been eliminated by using equation XV, equation XVI beorder Hnear differential equation which can be solved by the
integrating factor method given in Section 2.3.4. The result is

comes a

first

240

(29g-239g^/

l/240>
+ 210^^/^^")

XVII

7 X 29 X 239

For any value of q it is possible to determine the corresponding values

and s from equations XV and XVII, and then the value of / from equation IV. Similarly by means of equation V the total quantity of chlorine gas
consumed is determined. These last stages are arithmetical and are consequently not reported in detail; but the results of the calculations are summarized in graphical form in Fig. 2.1. in which the mole fraction of each
of

0-6

0-8

1-0

1-2

1-4

Moles chlorine consumed

Fig.

2.1,

Rate of reaction of benzene

halogenated hydrocarbon is plotted against the total amount of chlorine

consumed. The amount of chlorine required for the maximum production of
monochlorbenzene is from Fig. 2.1. equal to 1-0 mole of chlorine per mole
of benzene charged.
2.3.4.

Equations Solved by Integrating Factors

There are many equations which can be solved by a variety of integrating

most useful integrating factor is the one used to integrate the

factors, but the

first

order linear differential equation,

dy
dx
where

P and Q

are functions of

+ Py = Q
only.

(2.16)

29

ORDINARY DIFFERENTIAL EQUATIONS

2.

The basis of the solution of equations of this type is that there exists a factor
by which equation (2.16) can be multiplied so that the left-hand side becomes
This factor
a complete differential coefficient.
Assume that the integrating factor R is

factor".

is
a.

called "the integrating

function, of

only,

and

multiply equation (2.16) by R, thus

R^-^

ax

when

+ RPy = RQ

the left-hand side of equation (2.17)

efficient

(2.17)

the complete differential co-

is

Then

of some product, say Ry.

^iRy) =
dx

R'f^yf
ax
ax

(2.18)

Equation (2.18) represents the left-hand side of equation (2.17)

if

dR
- = PR

(2.19)

dx

The

variables

and x can be separated

in this equation,

and the solution

R = QxpnPdx)
That

is,

(2.20)

the integrating factor for the solution of the

equation (2.16)

tial

first

order hnear differenare given to

The following examples

exp (jPdx).

is

is

illustrate the technique.

Example

dy

2.

Solve

(1

+ x^)- + xy =

dx

X
dy

+ 7r2y-,
dx
+ x^^ ~ +

oi'

The

integrating factor

.2

x'

is

xdx

exp

= exp[iln(l+x-)]

(l

+ x')"

II

Hence, multiplying throughout by (1+x^)^,

(^+-^^>*i +

"'

(T^^ = (T;W

Integrating,

('-^^^^^ =

1(1^

This integral can be evaluated by using the substitution x

obtained from tables. The final solution is

ln[x

+ V(^' + l)] + C
V(i+^')

^^

tan

0,

or

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

30

Occasionally, a

solved in which

order,

first

and

first

degree differential equation has to be

are functions of both the dependent

and the independent variable. When this arises it is often possible to linearize the equation by changing the dependent variable.
Thus, consider the following
example.

Example

3.

xy

Solve

y'^

exp

ax
X

l/y^ thus

/-3x^'

dy

exp

y^dx

y"^

Let z

--4: =

-3

or

changing the dependent variable. Then

dz

dz
dy
=--/

y dx
3

-7-=--i and dx
dy
y
Substitution of II into

gives

3.z
Equation

III

is

II

a Hnear

first

dz
/-3x^
= 3exp(^
^

III

order differential equation which has the

solution,

3\

pexp^j

3x +

:.

dx

3x +

/(3x + C) = exp(^

First order linear differential equations occasionally arise in chemical

engineering problems in the fields of heat transfer, momentum transfer and
mass transfer. Whichever aspect of chemical engineering is of interest the

method of solution is the same. This

taken from the field of heat transfer.
Example

4.

An

is

illustrated

by the following example

elevated horizontal cylindrical tank 5

ft

diameter and

= 2", and is
employed as a maturing vessel for a batch chemical process. Liquid at
200F is charged into the tank and allowed to mature over 5 days. If the
data below applies, calculate the final temperature of the liquid and give a
12

ft

long

is

insulated with asbestos lagging of thickness

plot of the liquid temperature as a function of time.

Liquid film coefficient of heat transfer (h^)

Thermal conductivity of asbestos

(k)

=
=

25-0 Btu/h
0-1 Btu/h

ft^ F.
ft^

F ft" ^

Surface coefficient of heat transfer by convection

and radiation

=1-8

(/^i)

Density of Hquid (/?)

Specific heat of liquid

(s)

=
=

Btu/h

ft^ F.

62-4 lb/ft^
0-6.

ORDINARY DIFFERENTIAL EQUATIONS

2.

31

Atmospheric temperature (t) which can be assumed to

vary according to the relation t = 55+15 cos {nO/ll),
where 9 is the time in hours.
Atmospheric temperature at time of charging = 70F.
Heat loss through supports is negligible.
Solution

Rate of heat

loss

(tt x 5 x 12) + 2 (J-tt x 5^)

by liquid = Ii^A (TT^^).

Rate of heat

loss

through lagging

Area of tank (^)

where

72-57r ft^

-j- (7^,-7^).

represents the bulk liquid temperature,

T^^,

represents the inside wall temperature of the tank,

T^ represents the outside surface temperature of the lagging.

Rate of heat

from the exposed surface of the lagging

loss

h,A{T,-t)

Now,
Rate of heat

Rate of heat
loss

from

Rate of heat

transfer through

loss

from

'

liquid

:.

Rearranging the

first

lagging

h,A{T-

surface

kA

= --(T- r,) = h,A{T-t)

r.)

II

part of this equation to give

-(.,

and substituting into the

last part

or,

+7

+
T,

/j,

r+

of equation

r.

III

II gives

kh^

(T-l)

h^h2l + hjc + h2k_

IV

0-2467 + 0-754r

using the numerical values given earlier.

Considering the thermal equilibrium of the liquid,

=
=

Input rate

Output

rate

Accumulation

rate

h2A{T^-i)

= Vps
dO

h2A{Z-t)=Vps

dT

VI

'dO

Using the numerical values given above, and substituting for T^ from
equation V,
J'TI

- = - 0-0465 (0-246r + 0-754r-0

du

VII

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

32

or in standard form,

dT
- +

0-01147

0-0114/

0-628

ad

+ 0-171

cos(7i0/12)

VIII

Equation VIII is a first order linear differential equation which can be solved
by means of the integrating factor e^'^^^"^^. The solution is

Tei^^^

The second term on

0-628 Je-^^^^J0 + O-171 \e^-''''^'QOs{Tieill)de

the right-hand side can be integrated by parts to give

^^|l^|^^[0.0114cos0.262e +
Hence, the complete solution of equation VIII

r=

55

IX

0-262 sin 0-262^]

is

+ O-O28cosO-2620 + O-653sinO-2620 + Xe"^^'*^

= 0, 7 = 200.
200 = 55 + 0-028 + X
K = 144-97

with the boundary condition that at

.-.

.*.

Therefore, the final solution

T=

55

+ O-O28cosO-2620 + O-653sinO-262^ + 144-97e-''^^

Inspection of equation

by

0-028F

is

at

most.

XI shows
Similarly,

that the second term can only affect

the third

term

will

200

20

40

60

80

100

Time (hr)
Fig. 2.2.

XI

Variation of tank temperature with time

only contribute

2.

ORDINARY DIFFERENTIAL EQUATIONS

33

0-653F at most. Both of these terms are neghgible from a practical point
of view, consequently it is justifiable to express equation XI as

r=
Figure 2.2. gives a plot
120 h, T = 92F.

2.4.

As already

55

ofTwsd

+ 145e--^^^'

in

XII

accordance with equation XII and after

Second Order Differential Equations

stated, a

second order differential equation

the highest order differential coefficient appearing

is

is

one in which
However,

the second.

may contain other terms involving first order differential coaddition to functions of the independent and dependent variables.

the equation
efficients in

Furthermore the equation may be linear or non-linear. Thus,

5^=/W

(2.21)

a simple second order differential equation easily solved by two successive

On the other hand the equation,

is

integrations.

appears to be quite formidable on first inspection. These are both second

ethods for solving them will be described
order differential equations and the
in the following paragraphs. Essentially this will consist of proposals by which
the above type of equation can be reduced to a first order equation which
can then be integrated by one of the procedures already presented. Such a
method is possible if the equation can be placed in one of the following

classes.

Non-linear

(b)

Equations where the dependent variable does not occur explicitly.

Equations where the independent variable does not occur explicitly.

(c)

Homogeneous

(a)

equations.

B. Linear
(a)

(b)

The
The

coefficients in the

equation are constant.

independent variable.

coefficients are functions of the

The non-Hnear equations

be considered in this section, and the

be discussed in Section 2.5,
but the last type of equation is treated separately in Chapter 3.
In class A the method of solution depends upon the existence of a substituent which will reduce the second order differential equation to one of
first order. In classes A(a) and A(b) it is customary to let/? represent the first
order differential coefficient. Substitutions of this kind were suggested by
Clairaut (1713-65) and equations of the type,
will

linear equation with constant coefficients will

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

34

They

are called Clairaut forms.

are solved by differentiation followed by the

substitution.

Returning to second order differential equations, the detailed substituting

procedure depends on whether the equation falls into class A(a) or class A(b),
and each of these will now be considered separately.
2.4.1.

Equations where the Dependent Variable does not occur Explicitly

When
is

the

substitution

replaced by the

ducing a

first

first

is

made

second derivative of >'

in this case, the

y completely and proThe following example

derivative of/? thus eliminating

order differential equation in

p and

x.

will clarify the technique.

dy

d'-y

Example

1.

Solve

dx^^'^dx^'

Let

dx

dp
and therefore
dx
Substitution into equation

d^y

Jx^

gives

dp

--{-

xp

ax

II

dx
II is a first order linear differential equation which can be solved by
using the integrating factor exp
x^).

Equation

pQxp(^x^)

.'.

aexp(ix^) +

.*.

= ax + C

.-.

.'.

xQxp(ix^)dx

+ Cexp(-ix^)
i

C
dy

dx

Qxp( ^x^)dx

III

ax + AQd(-^j+B

IV

where A and B are the two constants of integration. The integration of the
last term in equation III involving the "error function" will be treated in
Chapter 5. Here, it is sufficient to quote the result.
2.4.2.

Equations where the Independent Variable does not occur Explicitly

In this case, the

same substitution of p

for the

the substitution for the second derivative of >'

is

first

derivative

is

made, but

obtained as follows.

2.

35

ORDINARY DIFFERENTIAL EQUATIONS

Let

J
dx

(2.24)

JA^JL = /-I
J^
dy
dx
dx

il

(2.25)

dy

dx^

Hence, when the independent variable does not appear expHcitly, make the
by equations (2.24) and (2.25). The following example
will make the procedure quite clear.
substitutions as given

Example

2.

d^y
y
+

SoUc

fdyV

^-j

Substitution of equations (2.24) and (2.25) gives

dp
dy
Separating the variables,

dy

pdp

r-i

Integrating,

lny

This

is

""

= iln(p'-l)

ay

or

where a and

c are the

=
=

II

^y

Jv(^V+i)

a standard integral, and therefore the solution of equation III

2.4.3.

lna

is

{\la)sm\i~^ {ay) + b

sm\i{ax + c)

two constants of integration.

Homogeneous Equations.

In Section 2.3.3, the

first

order homogeneous differential equation was

expressed in the form.

Regarding x and y as of the same dimensions, equation (2.12) is an equation

between dimensionless groups.
The corresponding dimensionless group
containing the second differential coefficient is {x d^yjdx^). In general, the
dimensionless group containing the th differential coefficient is

dx"

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

36

The second order homogeneous differential equation can be expressed in

a form analogous to equation (2.12), viz.
f2

=/.(;

ff
The

substitution for a

order homogeneous equation

first

y
If this substitution

is

("

s)

made

vx

(2.13)

again,

dy

dv
^2.14)

Tr'^'-'T.
as before,

and

is

- + X-5
^2-22

dx

dx^

(2.27)

dx

Substituting these expressions into equation (2.26) gives

^

dv

d^v

dv\

In equation (2.28), a factor x^ is associated with the second differential coof V, a factor x with the first differential coefficient of v on both sides
of the equation, and v occurs alone. Equation (2.28) can therefore be written

efficient

in the alternative equivalent form,

r2.
,d^v

It is to

be noted that

equation (2.26)

if

same form

left-hand side assumes the

dv\

is

multiplied throughout by x, the

as the left-hand side of equation (2.29).

However, the right-hand sides only take the same form if the right-hand side
of equation (2.26) is a Hnear homogeneous function of its two arguments. A
short cut from equation (2.26) to (2.29) is therefore possible if the given
differential equation is both homogeneous and Hnear.
Equation (2.29) is solved by using the second substitution
X = e

or

dv

\nx

(2.30)

dv dt

dx"'"dtlx
X

(2.31)

-T,
X dt

dv
dv

=
dx
dt

Differentiating equation (2.31) again with respect to

12

dx^~

x^ dt

x"^

dt

xdx\dt)
X dx

dt \dt^

2.

ORDINARY DIFFERENTIAL EQUATIONS

dh

^d^v

dv

-n=-7^--r
dx^
dt^
dt

I.e.

37

(2.33)

Putting equations (2.32) and (2.33) into equation (2.29) gives

dh

dv

dv\

^^^^^

^'-''^

-dr^-jt=f\'^dt)
equation (2.26) has now

been reduced to a form

where the independent variable (t) does not occur explicitly, and it can
be solved by the method given in Section 2.4.2. The actual procedure is

The

original differential

(2.34)

illustrated in the following example.

Example

3.

Solve

-''j*''"i:)
homogeneous since it can be expressed in the form (2.26)
after division by 2xy. Use the first substitution y = vx and its differentiated
forms (2.14) and (2.27).

This equation

is

2vx^

x-4) +

(2-^ +
dx

Simphfying into the form of equation

.d'-v

2vx^

making the second

x^ (v + x-^)

(2.29),

,(dvV

-.+2vx =

x^l-]

dx

substitution

II

dxj

dv

dx

Now

v^x^

dx^l

III

\dx/

e^

and simphfying

.^=(|V
\dtj
dt""

Equation IV does not contain the independent variable explicitly, and

be solved as suggested in Section 2.4.2. Put p dv/dt, and obtain

it

can

dv
This has two solutions,

viz.

. dp
2v-

p,

or

dv

The second

solution, called a "singular solution", yields the result,

y
which undoubtedly

satisfies

= Ax

the original equation

VI
I.

The

first

solution can

be integrated twice to give

= x(B\nx + Cy

VII

MATHEMATICAL METHODS

38

IN

CHEMICAL ENGINEERING

By choosing the arbitrary constants in equation VII to be 5 = 0, C = ^A,

equation VI is seen to be included in equation VII as a special case. Equation VII is therefore the complete solution of equation I.
The value of the substitutions given in Sections 2.4. are rather hmited,
because in many engineering applications the transformed equation cannot
be integrated. Such a difficulty was encountered by Marshall and Pigfordf
in their problem on the dynamics of a laminar jet. The authors of this text
also met the same difficulty in attempting to predict the temperature profile
along a graphite electrode by using the above substitution. The particular
problem had to be solved by the method to be given in Chapter 11. However,
the problem is presented here to illustrate the type of difficulty that can arise
and to show that an approximate yet valuable solution can still be obtained
by elementary procedures.

Example 4.
graphite electrode 6-0 in in diameter passes through a
furnace wall into a water cooler which takes the form of a water sleeve.
The length of the electrode between the outside of the furnace wall and its
entry into the cooling jacket is 1-0 ft; and as a safety precaution the electrode
is insulated thermally and electrically in this section, so that the outside
surface temperature of the insulation does not exceed 120F. If the lagging
is of uniform thickness and the mean overall coefficient of heat transfer
from the electrode to the surrounding atmosphere is taken to be 0-3
Btu/hFft^ of surface of electrode; and the temperature of the electrode
just outside the furnace is 2700F, estimate the duty of the water cooler if
the temperature of the electrode at the entrance to the cooler is to be 300F.
The following additional information is given.

Surrounding temperature

70F.

Thermal conductivity of graphite = Kj

where k^ = 88-9, and a = O-OISO.J

kQ cuT

(These numerical values are calculated as an example in Chapter 10.) The

temperature of the electrode may be assumed uniform at any cross section.
Solution
sectional area of the electrode {A) = 0-196 ft^.
Referring to Fig. 2.3 and using the symbols stated there; a heat balance

The

HotX

1
i

Fig. 2.3.

'

T*g<^xf Cold

(5i

Heat transfer through electrode

t Marshall and Pigford, *'Differential Equations in Chemical Engineering," p. 76.

University of Delaware.
X I.C.T. 5, 86, 1937.

McGraw-Hill Book Co. Ltd.

New

York.

ORDINARY DIFFERENTIAL EQUATIONS

2.

over the length of electrode dx at a distance x from the furnace

39
is

T'T-t

Input

= kjA-j-

(3X

Output

dT
.dT
d
= -^r^- + - [ -kjAdx
dx
dx\
^

d_

= !^(T-To)5x
^l5.
A
dxj

dx

U=

II

Accumulation

where

dx + nDU(T-To)Sx

overall heat transfer coefficient

from the electrode

III

to the sur-

roundings

and

D=

electrode diameter.

Substituting for kj
letting

nDUIA =

p,

and simplifying by dividing throughout by Adx and

III becomes

equation

d^T

fdTV

dx

\dx

(/co-aT)--^-a(

-^{T-T,) =

IV

Equation IV is a second order non-Hnear differential equation which does

not contain the independent variable explicitly. Thus let

dT
^

dx

d^T

dp

dx^

dT

Equation IV then becomes

dp

..7
=
(ko-o^T)p^ - ap'-l^(T-To)

V shows that the variables p and T cannot be separaand one might be tempted to introduce simplifying assumptions at this
stage; or attempt to solve equation IV by some other method. However,
the way in which p occurs in equation V suggests the substitution p^ z.
Putting y = {T Tq) also, gives

Inspection of equation
ted,

dz

[(ko-oiTo)-ay-]- laz-l^y =

VI

dy

which

is

first

order differential equation that can be solved by the inte-

grating factor method.

r

The

integrating factor

2(f.dy

is

,,

ocyy
I

ko-ccTo-ccyj

Substituting the integrating factor into equation

VI and converting back

to

the original variables gives

(ko-oiT)dT

=/ V[C +

/?(fco

- aT) (T- To)^ -|a^(r- To)']

VII

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

40

Equation VII cannot be integrated analytically because of the cubic polynomial under the square root sign, or graphically because the arbitrary
constant C is completely unknown.
Furthermore there is no way of
evaluating C other than perhaps by trial and error which would not be very
accurate.
Hence in order to obtain an estimate of the heat load on the
cooler it will be assumed that the heat lost to the atmosphere per ft^ of
electrode surface (0 is constant, and equal to 400 Btu/h ft^. Equation III
then simplifies to
d_

dx

dx

K=

-dx
TiDQ

VIII

fdTV

d^T
or

where

dx

nDQ/A,

The solution of equation IX

P
^

is

obtainable by the substitution

dT
= -r
dx

dp
p -

and

dT

d^T

= -r^
dx^

which gives

a{ko-(xTf

,2_

where B is the constant of integration.

Reducing the power of;? to unity, and choosing the negative root because
the temperature falls as distance x increases.

= -

(x(kn-(xTy

XI

Equation XI can be integrated with the aid of the substitution z

The

result

a^'

where

is

at the

2700
at

XII

ends of the electrode give

i:^ - = ^0-1

- - KaC^

more convenient

XIV

a\ K

to determine

how B and C

solution, rather than solve these equations directly.

is

XIII

/:
/

^/co_l
Ib
= ^ - -^/ - - Ka(i + Cy
a

cooler

oLTf

1,

300
It is

(Jcq

the second arbitrary constant.

The boundary conditions

at X = 0,

and

is

the rate at which heat arrives at

occur in the required

the water

The duty of

which

is

given by,

2.

ORDINARY DIFFERENTIAL EQUATIONS

H=

-(ko-300a)A-dx

^=

-(ko-300(x)AK(l +

by

differentiating equation

Substituting from equation

41

C)

XII and putting x = I.

XIV for the denominator of

this expression,

H=-AK(1 + C)

XV

XIV for C gives

C = i - 2400 (fco - 1 500a)/K

Solving equations XIII and

Substituting into

XV gives
H = ^X[2400 (fco - 1500a)/X - i]
=

and

this is the

XVI

28800 Btu/h

duty of the water cooler.

2.5.

LINEAR Differential Equations

This type of equation is frequently encountered in most chemical engineering fields of study, ranging from heat, mass and momentum transfer to
apphed chemical reaction kinetics. It is one of the most important types of

equation to be found in applied science, and therefore considerable study

has been devoted to finding solutions to the many different forms it may
take.

The general
coefficients

may

linear differential equation of the nth order

d"v

where

having constant

be written

dv

d"~^v

any function of x.
to be described below for the second order equation
applies equally well to linear equations of any other order. However, in the
case of a higher order equation, the mathematical manipulations may be
cumbersome as the reader will soon appreciate, but since no new principles
are involved, further discussion will be restricted to the second order case.
(f)(x) is

The treatment

2.5.1.

Second Order Linear Differential Equations

The general equation can be expressed

in the form,

P^2 + Q^_ + Ry =
dx^
^dx

(t^(x)

(2.36)

where P, g, and R are the constant coefficients.

Let the dependent variable y be replaced by the sum of two new variables
u and V, i.e.
y

+v

(2.37)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

42

Substitution in (2.36) gives

dhi
P-j:^
If V

du

'l^^-^T.^^"'

^W

(2.38)

a particular solution of the original differential equation, then

is

d^v

P -7-2 +
dx

^dv
Q
+ Rv =
dx

and by comparison with equation

Ppi +

(2.38)

Q'JL

Jx^

That

dv

d'v

+ Q-\-Ru +

is,

dx

it

4>{x)

(2.39)

can be seen that

+ Ru =

u satisfies equation (2.40) which

is

the

(2.40)

same

as equation (2.36) with

replaced by zero. Therefore, since y is the general solution of equation

(2.36) and is the sum of u and v, where y is a particular solution of the original differential equation; u must be the complement of the particular solution
(/)(x)

which must be added

in order to get the general solution.

solution of the linear differential equation will be the

That

sum

is, the general

of a "comple-

mentary function" and a "particular integral". The problem resolves itself

into one of finding some specific solution that will satisfy the original equation if necessary by trial and error, and finding the complete solution of the
The general
same equation assuming that the right-hand side is zero.
solution will be the sum of these two solutions and will contain two arbitrary constants which can be evaluated from the boundary conditions of the
specific problem being considered.
The above remarks on the general approach to the solution of the second
order linear differential equation apply to any linear differential equation
irrespective of its order. However, the number of arbitrary constants that
will have to be evaluated will equal the order of the equation. That is, the
general solution of an nth order differential equation will be the sum of a
complementary function and a particular integral and will contain n arbitrary constants.
The arbitrary constants will arise in the complementary
function which will therefore have n terms. These points should be borne

in

mind when

2.5.2.

solving a linear differential equation.

The Complementary Function

Referring back to the second order differential equation (2.36), the

That is a
is to find a solution of the equation when (/)(x) is zero.
solution of the equation,

problem

^dy

d^y

+ Ry = ^
P^2
^ + Q-r.
dx^
^dx
Initially let

it

be assumed that the solution

^e'"^

(2.41)

is

(2.42)

ORDINARY DIFFERENTIAL EQUATIONS

2.

43

^^yl^me""

Then

dx
--^

and

A^m^e'"''

Substitution of the above expressions for the

coefficients into equation (2.41)

and simpHfying

first

and second

differential

gives

A^ e'"* {Pm^ + Qm + R) =
Equation (2.43)
zero. That is

will

be

satisfied if either

(2.43)

or the term within the brackets

if

is

Pm^ + Qm + R =

(2.44)

A cannot be zero by definition in equation

(2.42) and therefore the values of

determined by equation (2.44) will satisfy the "reduced equation" (2.41).
Equation (2.44) is termed the "auxiliary equation". It will have two roots
which could be unequal or equal, real or complex (see Chapter 4), and these
will be considered individually.
(a) Unequal Roots to Auxiliary Equation. Let the roots of equation (2.44)
be distinct and of values nti and Wj. Therefore, the solution of equation
(2.41) will be
y^A^e'"''' or y = A2e'"^''

and the most general solution of the reduced equation

y

Equation (2.45)

is

A^e'"'''

+ A2e'"'''

be
(2.45)

two terms and two

would be expected for a second order linear differen-

the "complementary function". It has

arbitrary constants as
tial

(2.41) will

equation.

Equal Roots to Auxiliary Equation. If the roots of the auxihary

equation are equal, that is if mj = W2 = m, the auxihary equation can be
(b)

written

Pm^ + Qm + R =
and the solution of the reduced equation
y

(2.46)

is

Ae""""

(2.47)

But as this only contains one term and one arbitrary constant it is not the
complete complementary function of a second order equation. Therefore
let

y
be a second

and

trial

solution in which

dy

dV

dx

dx

^ =
dx^

e"**

Fe'"^

F is

(2.48)

a function of x. Then

- + 2me'"*- +
r-

dx^

dx

m^Ve'"''

MATHEMATICAL METHODS

44

CHEMICAL ENGINEERING

IN

first and second differential coequation (2.41), using (2.46) and simphfying gives

Substitution of these expressions for the

efficients into

-^ = '
Double integration of

this gives

V = Cx + D
and the complementary function becomes
y

The above

= {Cx + D)e'"''

(2.49)

solution contains two terms and two arbitrary constants.

D can be made equal to A in equation (2.47),

can be seen that the previous solution (2.47) is included in the later solution (2.49). Equation (2.49) is therefore acceptable as the complete complementary function when the auxiliary equation has equal roots.
Since the arbitrary constant

it

Example

1.

Solve

-j:^

- 5 3^ +

The

auxiliary equation

6>'

dx

dx^
is

m 2_5m + 6 =
and the roots of

this are

mi =
The

solution to equation I

is

A and B

Example

2.

are the

-^ +

Solve

auxiliary equation

m2 =

Ae^''

+ Be^''

6-^ + 9y =
dx

is

m^^-6m + 9 =
and the roots of

III

two arbitrary constants.

dx^

The

2,

therefore

where

II

II

this are

mj = m2 = 3
The

solution of equation

I is

therefore

where

A and B

are the

(A + Bx)e-^'

III

two arbitrary constants.

Any reader who is unfamiliar

with the elementary concepts of complex numbers is advised to read
Sections 4.1 to 4.10 inclusive of Chapter 4 before proceeding with the
following section.
For those who have become familiar with the elements of complex
numbers it will be immediately appreciated that the roots of the auxiliary
equation could be complex.
This will become clear by considering the
following example.
(c)

Complex Roots

to Auxiliary Equation.

Example

Solve

3.

tt^

- 4^^ +

auxiliary equation

5>'

dx

dx""

The

45

ORDINARY DIFFERENTIAL EQUATIONS

2.

is

m^-4m + 5 =

II

and the roots are

m = 2i
Therefore the complementary function

y
or the solution

may

is

= A exp [(2 + 0^] -f B exp [(2 - /)x]

III

be written as

e^""

(E cos x

+ F sin x)

IV

using the relationships between exponential and trigonometric functions.

In conclusion it will be appreciated that irrespective of the nature of the
roots, the complementary function takes the form

Either

Ae'"'''

+ Be'"'''

(2.50)

for distinct roots whether real or complex.

Or

(A + Bx)e'"''

(2.51)

which must always be real. In all forms the expression will

have arbitrary constants equal in number to the order of the differential
for equal roots,

equation.

2.5.3.

Particular Integrals

To

obtain the complete solution of the linear differential equation,

some particular solution of the equation,

it is

necessary to find

d^'y

^dy
^dx

P-^, + Q-^ + Ry =
dx""

where

4>(x)

(2.36)

may

be a constant, or a function of the independent variable x.

is the complete solution and
therefore nothing further is required. However, in most problems encountered, (f){x) is not zero and a particular integral will have to be found in order
to complete the solution.
There are many general methods of determining the particular integral.
Of these only three will be discussed here, as the three to be presented are
capable of solving most problems arising in applied science.
The three
methods may be termed,
If

(/)(x)

(/)(x)

(a)

(b)
(c)

is

zero the complementary function

The method of undetermined coefficients.

The method of inverse operators.
The method of variation of parameters.

Method

confined to hnear equations with constant coefficients, and

</)(x).
Methods (b) and (c) are of more
general applicabihty. Each of the methods will be discussed below.
(a) is

particular forms of the function

MATHEMATICAL METHODS IN CHEMICAL

46
2.5.4.

Particular Integrals by the

ENGINfEERI>iG

Method of Undetermined

Coefficients

and error procedure, and whilst it is easy to

apply, some experience is required before the form of the particular integral
can be recognized. It is most unlikely that the chemical engineer will have
had this experience and consequently the following rules which choose the
form of the particular integral are submitted as a guide.
This

(i)

essentially a trial

is

Then obviously a

constant, say C.

is

</)(x)

equation (2.36)

= CIR

y
because

(2.52)

the differential coefficients will be zero.

all

particular integral of

is

The complete solution

be

will therefore

Ae"''''

+ Be"'''' + CIR

(2.53)

This rule apphes to any order hnear differential equation with constant
coefficients.
(ii)

all

(i){x) is

a polynomial of the form

the coefficients Oq, a^,

The form of

{ ....

^x";

where

the particular integral will also be a rational integral func-

tion of the independent variable

tial coefficients

a iX-\-a2X^

aQ-\-

a are constants.

x because only functions of x whose

differen-

are positive integral powers of the independent variable are

Consequently the form of the particular

themselves integral powers of x.

integral will be,

<XQ

+ oiiX + (X2X^+

'

+ccx"

(2.54)

Furthermore the degree of the particular integral must equal the degree of
in (pix), otherwise the degree of x on each side of the differential equation
will be unequal and this is impossible in a linear differential equation with

constant coefficients.
The solution procedure is therefore to assume a polynomial of the type
shown in equation (2.54) and obtain the first, second, etc., differential coefficients of this expression.
These differentiated expressions must then be
substituted into the original differential equation and the coefficients ao to
a evaluated by equating coefficients. The following example will make the

procedure

clear.

Example

4.

Solve

4 - 4 -^ + 4;; = 4x + 8x^
dx
d

ax
dx"Because the degree of

(^(x) is 3, let

p-\-qx-\-rx^

+ sx^

II

in order to evaluate the particular integral.

Then

and

=
^
dx
d^y
-^
dx^

+ 2rx-\-2>sx^

2r + 6sx

III

IV

2.

47

ORDINARY DIFFERENTIAL EQUATIONS

Substitution of these expressions for the differential coefficients into / gives

(2r + 6sx)

- 4{q + 2rx + 3sx^) + 4(p + qx + rx^ + sx^) =

Equating coefficients of equal powers of

4x + 8x^

x,

2r-4q + 4p =
6s-8r + 4^ = 4

4r 12s =

4s

.'.

yp

.-.

The complementary function

6,

y
(iii)

(/)(x) /5

y^ + y^

or

and

10,

integral

is

of

e'"''

{A +

The

first

where

VI
is

VII

Bx)e''''

J and

be multiples of

will

VIII

The form of

are constants.

form because

e''"'.

Therefore,

all

if

the

the differen-

the particular

to be

Hence

ar^/"".

(2.55)

oce^^

be

differential coefficient will

be

(A + Bx)e^'' + 7 + lOx + 6x^ + 2x^

y,

coefficient will

is

the form Te^^\

assumed

3s

for this case of equal roots

particular integral will also be of exponential

tial coefficients

=
=

= l + 10x + 6x^ + 2x^

y,

and the complete solution

or

are''^,

and the second

differential

substitution into the original differential

equation (2.36) gives

(Pr^

or

Evaluation of a from the

be determined.
(iv)

(f)(x) is

of the form

+ Qr + Ryue''' = Te'''
T
a = WZTTT^:^^
Pr'^ + Qr + R

known

sin

(2.56)

(2-57)

constants enables the particular integral to

nx +

H cos nx; where G and H are constants,

The form of the
same type
be multiples of sin nx or cos nx.

either of which could be zero, whilst nisa. non-zero constant.

particular integral will also be a trigonometrical function of the

because all the differential coefficients will

Therefore assume the particular integral to be
j^p

The

= L sin nx +

first differential coefficient

-
dx
and the second

M cos nx

of equation (2.58) will be

= nL cos nx nM sm nx

differential coefficient

d^
d^y
dx

= n(L sin nx +
,

M cos nx)

(2.58)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

48

Substitution in the original differential equation gives

L=

{R-n^P)G + nQH
(2.59)

M = {R-n^P)H-nQG
{R-n'^Pf + n^Q''

(2.60)

Equations (2.59) and (2.60) enable the constants L and

to be evaluated
and thus the particular integral can be determined.
The above four forms of the particular integral that would be obtained
from each type of expression of 4){x) are summarized in Table 2.1.

Table

2.1.

Normal Forms of Particular

Right-Hand Side of
Equation ^{x)

Particular

Coefficient

Integral

Values

Differential

constant

Integrals

constant

To

polynomial
a^-^a^x^- ... +G,j.T"

polynomial

be

in

An

exponential

coefficients.

exponential

J^rx

the

equation

differential

and equating

An

by

determined

substituting

"^

Pr'-^Qr-rR

(R-n'P)G+nQH
Trigonometrical

Trigonometrical

sin

nx+H cos nx

P, Q, and

sin

nx^-M cos nx

(R-n^Py+n'Q^

(R-n'P)H-nQG
(R-n'Py+n'Q'

refer to the coefficients in the differential equation (2.36).

Modified procedure when a term in the particular integral duplicates a

complementary function. In each of the above forms of the particular integral, it is possible for one or more terms to be identical with one
or more terms in the complementary function. When this occurs the general
procedure is to multiply the assumed form of the particular integral by the
independent variable, which will destroy the similarity of the terms. However, it is possible that this multiphcation by the independent variable will
make other terms in the particular integral and complementary function
identical. If this should happen the assumed form of the particular integral
should be multiplied by the square of the independent variable, and if
similarities still exist the assumed form should be multiplied by the third
(v)

term

in the

2.

ORDINARY DIFFERENTIAL EQUATIONS

49

power of independent variable and so on until no identical terms remain

between the complementary function and the particular integral.
Identical terms in both parts of the solution of the differential equation

can

arise in

trial

form

many

For instance, when

ways.

(pix) is

a constant, the

first

for the particular integral will be a constant according to Table 2. 1

If in addition,

one of the roots of the auxiliary equation

the complementary function will also be a constant.

is

zero one term in

Hence

the modified

form of the particular integral should be a constant multiplied by the

independent variable, in which case the similarity will have been removed.
The following example will make the procedure clear.

trial

Example
The

5.

+ 6^ = 18
3^
dx^
dx

Solve

auxiliary equation

is

3m^-6m =

II

and the roots are

m=
The complementary function

y,

m=

and

is

Ae"''''

+ Be"'''' = A + Be^""

Since one term in the complementary function

integral

is

III

a constant, the particular

cannot be a constant even though the right-hand side of equation

contains a constant.

Therefore

let

the

form of the particular

y

integral be

= Cx

IV

=-C
dx

Then

^=

and

dx"^

Substitution in equation

gives

3x + A + Be^''

VI

C=

or

and the particular Integral

is

18

(3xO) + 6(C)
3

3x.

Therefore the complete solution

is

The above complication can arise also when a root of the auxihary
equation is equal to the coefficient of the independent variable in the exponential term on the right-hand side of the equation. In this case too, an example will make the procedure clear. Therefore consider

Example

6.

Solve

3^ +
dx"^

10-^ - 8v
dx

7e"^*

MATHEMATICAL METHODS IN CHEMICAL ENGINEERINQ

50

The

auxiliary equation

is

= (3m-2)(m + 4) =
m = 2/3 or 4

3m^ + 10m-

.*.

II

and the complementary function is

^,^_^g2x/3_^^g-4x

The second term

jjj

of the same form as that given in Table 2.1

for the particular integral of equation I. Therefore, take the modified form
in equation III

is

Cxe"^^

IV

for the particular integral.

ax

-^ = (16x-8)Ce-^"

and

ax

Substitution in equation
solution of equation

Another way

and solving for

C gives C = - J,

and the complete

I is

y^ + y^

Ae^""^

+ Be-'^'' -^xe-"^""

which the above comphcation can appear, is when the

roots of the auxiliary equation are both equal to the value m, the coefficient
in the exponent in a term of the form Ce'"'^. Thus for an auxiliary equation
with equal roots the complementary function will be {A-\-Bx)e'^^ where the
first term is similar to the expression Ce""^.
Hence the proposed form of the
particular integral would be Ex^e"^^, since multiplying by x alone would
give a form similar to the second term of the complementary function. E
would be evaluated in a manner similar to that illustrated in the above
in

examples.
Finally, when the roots of the auxiliary equation are complex and the
right-hand side of the diiferential equation contains the terms

{A
the modified trial

sin

nx

form of the particular

x(a sin nx

and the evaluation of a and

2.5.5.

j5 is

Particular Integrals by the

The symbol

+ B cos nx)e^^

(2.61)

integral will be

+ P cos nx)e^''

(2.62)

as above.

Method of Inverse Operators

(dy/dx) representing the differential coefficient signifies that

;^ has been carried out with respect to the

the operation of diflferentiation of

independent variable x. In fact if there is no doubt whatever of the independent variable, it is acceptable to write simply "/)>'" for the differential
coefficient implying that the differentiation process has been carried out.
Hence it is convenient to refer to the symbol *'i)" as the "differential operator", and prefixing it to a variable means that a differentiation has been

2.

ORDINARY DIFFERENTIAL EQUATIONS

51

carried out with respect to an obvious independent variable. The letter

alone has no significance and must be placed in front of the dependent
variable to signify the differential coefficient.

Dy = dyjdx

I.e.

D{Dy)

Similarly

D{D^y)

and

D"y

so that
It

= D^y = d^yjdx^
= D^y = d^yjdx^
= d'^yjdx''

should be noted that

{Dyf = {dyldxf
without ambiguity.

The above symbohc representation in terms of the differential operator

can be extended to expressions involving different order differential coefficients.

may be

Thus the expression,

written

D^y + Wy + 2y = (D^ + 3D + l)y

and the

latter

may

(2.64)

also be factorized,

{p + l){D + l)y

(2.65)

can be treated as an
it appears that the differential operator
ordinary algebraic quantity with certain Umitations.
The three basic laws of algebra are the following.
(a) The Distributive Law. This states that,

Hence

A(B + C) = AB + AC
which also appHes to the

differential operator

D(w +

t;

+ w)

D,

(2.66)
viz.

= Du + Dv + Dw

(2.67)

In fact the distributive law was appHed to the operator in equation (2.64).
(b) The Commutative Law. This states that,

AB = BA
which does not in general apply to the

(2.68)

D. It is conventional to consider that the differential operator differentiates every term

which follows it, and not the terms preceding it. Transferring a term past the
operator thus alters the value of the expression. For example

Dxy
but

it is

xDy

true that the expression (2.65) can be given the equivalent forms,

{D + \){D-\-2)y

The operator
(c)

7^

differential operator

will thus

commute with

The Associative Law. This

{D-\-2){D + \)y

itself

but not with variables.

states that,

{AB)C = A{BC)

(2.69)

MATHEMATICAL METHODS

52

which does not

CHEMICAL ENGINEERING

IN

in general apply to the differential operator

how

ing two equations illustrate

far the

is

D. The follow-

obeyed.

= (DD)y
= {Dx)y + xDy

D{Dy)
D{xy)

but

law

(2.70)

In the first term on the right-hand side of equation (2.70) the closing of the
bracket cuts y off from the operator thus destroying the simple equality.
The basic laws of algebra thus apply to the pure operators, but the
relative order of operators and variables must be maintained. Also, the
influence of the operator must not be hindered by brackets as in equation
(2.70) unless this is specifically desired. Consequently these properties can
be utilized particularly in the solution of linear differential equations thus
simplifying the solving procedure. Some of these properties will be illustrated in order to familiarize the reader with the techniques.
Application of the Differential Operator to Exponentials. Let the differential coefficient of e^"" be required. This expression could be written

De^^

Also for the second

(2.71)

differential coefficient,

D^eP""

and

pe'""

p^e^^

(2.72)

for the nth differential coefficient,

J^n^px

In general,

if

/(i))

pn^px

some polynomial of

is

(2.73)

the differential operator,

it

is

possible to write

/(D)6^-=/(p)e^I.e.

the coefficient of

iff(D)

is

replaces

each of these new terms

and the

D in the function.

expanded and then each term

differentiated

equation (2.64) y

will

This

is

4)

quite obvious, for

in the expansion multiplied

by

e^"",

be similar to the left-hand side of equation (2.73)

form similar

e^"",

(D^

to the right-hand side. Specifically,

then the expression can be written

+ 3D + 2)e^^ =

The above concept can be extended

exponentials.

(2.

For example,

if

(p'
to

+ 3p + 2)e^^

if in

(2.75)

more complex functions involving

the differential coefficient of ye^""

is

required,

then
D(ye''')

e'^'Dy

+ yDe^"" =

Similarly the second differential coefficient

D^yeP"")

Similarly

it

^^^(D

+ p)y

(2.76)

would be

= DleP\D + p)y'] = eP\D{D + p)y + {D + p)y .Die^"")

= e'^iD + p)(D + p)y = e''(D + p^y

(2.77)

follows by comparison of equations (2.76) and (2.77) that

DXye'"")

e'^'iD

+ pfy

(2.78)

2.

or in even

ORDINARY DIFFERENTIAL EQUATIONS

more general terms, iff(D)

is

53

a polynomial in D,

f{D)(yen==e'J(D + p)y
The above property of
most important as will be seen

the differential

(2.79)

operator involving exponentials

is

later.

Finally, the operation performed on the exponential function in equations (2.71) to (2.79) can be extended to trigonometrical functions by using

and trigonometrical functions

Let the nth differential coefficient of sin px be

the complex relationships between exponential

to be given in Chapter 4.

required.

Now

where "

D" (sin px)

Im

= d" Im e'^^
= ImZ)V^^
= Im(ip)"e'^"

(2.80)

(2.81)

" represents the imaginary part of the function

The evaluation of equation

Put n = 2m to make Jt even.

(2.81)

and because

which follows it.

depends upon whether n is odd or even.

e'^""

:.

D^'"(smpx)

= cos x + sin x
= (-p^)"'smpx
i

(2.82)

Similarly, for n odd,

Z)^'" +

= (-p^Tpcospx

^(sinpx)

The corresponding formulae

way are

for differentiating cos px,

(2.83)

which can be

derived in the same

D^"" (cos px)

D^'^^'icospx)

= (- p^y" cos px
= -{-p^Tpsinpx

(2.84)
(2.85)

The principles involved in equations (2.80) to (2.85) can be extended to

operations on more comphcated trigonometrical functions by considering
real or imaginary parts as shown. This is left to the initiative of the reader in
attempting the examples at the end of the book.
The Inverse Operator. The operator

signifies differentiation,

i.e.

Z)[J/(x)Jx] =/(x)

(2.86)

so that

jf(x)dx

(2.87)

D-^f(x)

which suggests that the reciprocal of the operator

D placed before a function

implies integration of that function with respect to an obvious independent

Thus i)~Ms the "inverse operator" and is an integrating operator.

Because of its relationship to the differential operator D it would be expected
that it can be treated as an algebraic quantity in exactly the same manner as
D. This will now be considered.
Equations (2.74) and (2.79) can be extended to an infinite series of positive
powers of D, so that if the function of D can be expanded in ascending
powers of D, the equations still apply.
variable.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

54

Example

7.

Solve
II

Transferring the operator to the other side of the equation,

-'

-m c-

4(1
:.

III

D-4

y^-illHiD)HiDr + iiD)' + -']e'^

IV

Performing the operations on e^^ gives

using the binomial expansion.

3^=-ie'^[i+K(i)'+(i)' + ..-]
The

series in the brackets is

a geometrical progression whose

sum

to infinity

is 2.

is

VI

3^=-ie'"

.-.

the particular integral of equation

I.

Because the inverse operator in equation III can be expanded in a series

of positive powers of Z), it satisfies the condition stated above. Thus, the
general property given in equation (2.74) could have been appHed much
earlier to equation III. Viz.

'

2-4

D-4

as before.

In the above example, equation (2.74) has been used in the form,
e^^

KD)

e^^

(2.88)
^

f(p)

However, if/(/?) = 0, e^^/f(p) is infinite and reference has

more general equation (2.79) in order to resolve this
(p) = 0, (Dp) must be a factor of f(D). Hence
f(D)

where n

is

to be

made

makes

(/>(/?)

to the

When

difficulty.

= (D-pn(D)

the lowest integer which

(2.89)
finite

and not

zero.

Then

^px

^px

(D-py^(D)

f(D)

eP""

(D-pyct^ip)

by applying equation

(2.74).

AppHcation of equation

(2.79) with

gives
1

e^"
1

f(D)

(t>{p)

D"

(2.90)

2.

is

ORDINARY DIFFERENTIAL EQUATIONS

power

raised to a negative

in equation (2.90),

and

55

this implies integra-

tion.

JeP^ =

It is

--

(2.91)

unnecessary to include the arbitrary constants arising in the integra-

tions since these are included in the

Example

8.

complementary function.

Solve

^- 8^
+
dx

16>'

6xe

dx^

In terms of the differential operator, equation

{D^-W + i6)y
and the particular

integral

The complementary function

6xe^^

II

4.

x^g^*

(yH-Bx)e^^

III

is

j,

becomes

is

and the complete solution

= (Z)-4)V =

IV

is

= yc+yp = (^+bx+x>'^^

Application of the Inverse Operator to Trigonometrical Functions. Trigonometrical functions can be written as the real or imaginary parts of e'^"" in
which case they can be treated in that form by the inverse operator. The
analytical procedure to follow would be very similar to that given above for
real exponential functions combined with the principles presented in equations (2.80) to (2.85). Therefore no further discussion need be given to this
particular topic.

Application of the Inverse Operator to Polynomial Functions. The inverse

operator can be made to operate on the polynomial function (/)(x) by expanding the function of the inverse operator in terms of
by the usual binomial
series for negative exponents.
The method is illustrated by the following
example.

Example

9.

Solve

d^y
.2

dx^

"

dy
J-

dx

^y

4x''

+ 3x^

Introducing the differential operator, equation

becomes

{D''-D-6)y = (D-3)(D + 2)>; =

4x^

+ 3x2

II

MATHEMATICAL METHODS

56

and the

particular integral

is

(4x-*

(D-3)(D + 2)

+ 3x^)

Expanding each term of the

D^
"^27
9

yp=

-7

3"^

+ 243

35 ^2

36

216

4x^ + 3x2

D^

12x^
"^

4+T

(4x^

65

(4x^

+ 3x2)

+ 3x2)

IV

7776

1296
7(24x + 6)

13x24

216

1296

36

III

+
"^32
16

+ 6x

(4x^+3x2)

by the binomial theorem gives

D^

D +

(2 + Z)).

(3-D)

partial fractions

144x^

CHEMICAL ENGINEERING

1
>';

IN

-0

VI

+ 36x2 + 132x-10
216

The complementary function

is

Ae^^' +

y,

and the complete solution

y
2.5.6.

yc

+ yp =

Be

VII

is

Ae^'' + Be~^''

Particular Integrals by the

(72x3

+ 18x2 + 66x-5)/108

Method of

VIII

Variation of Parameters

This method of determining the particular integral of a differential

equation is a very elegant procedure, but it depends on the complete complementary function being known and therefore its appHcabihty is somewhat
limited. Furthermore when apphed to differential equations of order higher
than three the solution of the simultaneous equations of the parameters is
time-consuming, laborious and above all hazardous to the unfamihar and
inexperienced.

Consider once again the general differential equation of second order

Let P be unity and Q and R be functions of the independent variable.
The equation can therefore be written
(2.36).

dy

d'y

-^^-\-Q{x)~+R{x)y =
and

let

(j>{x)

(2.92)

the complementary function of this equation be

y,

= Au + Bv

(2.93)

where A and B are the arbitrary constants and u and v are functions of the
independent variable x.
Now assume that the particular integral can be expressed,

yp=f,{x)u+f,{x)v

(2.94)

2.

ORDINARY DIFFERENTIAL EQUATIONS

57

where the functions of the independent variable /^(x) and/2(x) are called the
parameters. There will be two parameters for a second order, three for a
third order and so on. Then

-p=nix)u+Mx)u'+n(x)v+f2(x)v'

(2.95)

dx

where/',

and

u'

v'

signify the first differential coefficient of/, u

and

v with

Since in equation (2.94), two variables have been introduced

to represent one variable, it is legitimate to assume another arbitrary relationship between them. It is convenient at this point to assume that,
respect to x.

f[{x)u+n(x)v
in order to reduce the complexity of the

equation (2.94).

(2.96)

second

differential coefficient

of

Hence,

^V._

u'Ux) + u /;(x) + v"Mx) + v'n(x)

(2.97)

dx-

Substitution of equations (2.94), (2.95)

and

(2.97) into (2.92)

and rearranging

gives

A(x)lu" + Q{x)u' + R{x)uli +f^(x)[v'' + Q{x)v' + R(x)v]

+
+ Q(x)u' + R(x)u] and

But

[u"

and

V are solutions of the

[v"

u'fi(x)

+
+ v'n(x) =

+ Q(x)v' + R(x)v]

reduced equation when

cP(x)

(2.98)

are both zero because u

(/)(x) is

taken to be zero.

Therefore,
u'fi{x)

and solving equations

(2.96)

and

+ vyi(x) =

(2.99)

(l)(x)

(2.99) simultaneously gives

dfiix)

dx

v(t)(x)

(2.100)

uv'

vu'

djp^-u^

and

dx

vu

uv

The parameters fi{x) and fjix) can be obtained from

integration. The final complete solution will be
y

ufi(x)

+ vf2(x) + Aii + Bv

which includes the complementary function, equation

Example

10.

these equations by

(2.102)
(2.93).

Solve

2x^

d^_^

dx^

dy
X -^
dx

x^e

""

Equation I is Hnear, and the solution therefore consists of a particular

added to the complementary function. The left-hand side is homogeneous, and the complementary function can be found using the substitu-

integral

tion

e^

given in Section 2.4.3.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

58

The complementary function

the solution of

is

fl_.d_y + y =
2-^-3-^
i.e.

= Ae^ + Be^^
= Ax + Bx^

y^
y^

:.

II

dt

dt^

III

Applying the method of variation of parameters,

yp

where u and

let

ux + vx^

IV

v are functions of x.

dy

_,

dx

du

,dv

dx

dx

du
,dv
X
+ x^ =
dx

Put

VI

dx

d^y

du

dx^

dx

_^,,

_,dv

VII
Substituting V,

VI and VII

dx

into equation

xu

\vx^-\-x^'^--

Ix'^

dx

gives

hvx^

-\-

ux -\- vx^

x^e~*

dx

which can be simpHfied

to,

2;C^ +
dx

x4" = Xe-'

VIII

dx

Subtracting equation VI from equation VIII gives

du

x- = xe
dx
M

:.

= -e-*

IX

Substituting this expression into equation VI,

xdv

xe

''

+ X'-- =
dx

v=

.-.

-jx^e-^'dx

integral in equation X can be expressed in terms of the "error function"

by the methods to be given in Chapter 5. The solution is

The

Putting equations

IX and XI

= x*e~* ivTT ed\/x

into equation

IV

gives the particular integral,

yp= xe~''-\-xe~^ \\lnxQd\lx

= TT^ erfV X

The complete

solution

XI

XII

is

= Ax + Bx''- i Vttx

erfVx

XIII

2.

ORDINARY DIFFERENTIAL EQUATIONS

59

In the above paragraphs, methods of finding the particular integral and

complementary function have been presented for a second order linear
These methods are quite general and are equally
differential equation.
the

The second order

applicable to any order linear differential equation.
equation was chosen simply for convenience in order to bring out the salient
The same treatment would
principles in the solution of these equations.
have to be given to a third, fourth or higher order equation.
2.5.7.

Illustrative

Problems

Linear differential equations arise very frequently in the solution of

chemical engineering problems. Their appearance is not surprising when
one considers that chemical engineers are chiefly concerned with the prediction of rates of transfer of heat, momentum and material. Consequently the
method of solving such problems will now be illustrated by three examples
taken from the various important fields of the subject.

Problem

1.

Simultaneous Diffusion and Chemical Reaction

in

a Tubular

Reactor

tubular chemical reactor of length

is

L and

-0 ft^ in

cross section

is

order chemical reaction in which a material

converted to a product B. The chemical reaction can be represented,

employed to carry out a

first

A-^B
and the

specific reaction rate constant is

k h"^.

is u ft^/h, the
of A is assumed to be constant at D ft^/h, determine the concentration of ^ as a function of length
along the reactor. It may be assumed that there is no volume change during
the reaction, and that steady state conditions are established.

feed concentration of

is Cq,

and the

If the feed rate

diffusivity

Solution

Take a coordinate x to specify the distance of any point from the inlet
of the reactor section, let c denote the variable concentration of A in the
entry section (x<0), and y denote the concentration of A in the reactor
section (x>0), as

shown

in Fig. 2.4.

The concentration

Reaction section

vary in the

Entry section

will

|_

6x
Fig. 2.4.

Tubular reactor analysis

entry section due to diffusion, but will not vary in the section following the
reactor, t
t Wehner,

J.

F.

and Wilhelm, R. H. Chem. Eng.

Set.

6 ,89 (1956).

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

60

material balance can be taken over the element of length dx at a

x from the inlet.

distance

X
Bulk flow of

x-\-5x

uy+u

dy
-D^
dx

Diff'usionof^

The accumulation

in this case

is

dy
-y dx

dx

-D'^+^[
-D'^
dx dx\

Sx

zero, but the input

must exceed the output

to supply the reaction taking place within the element.

Rate of removal of

A by

= ky dx

reaction

of unit cross-sectional area. Alternatively, this term may

be considered as either an output or an accumulation and the general conservation equation (1.25) applied directly.
Thus,
since the reactor

uy-D-ldx

is

[uy
\

-D-^ + '-(-D^dx\ =kySx

+ u-^dx] dx

dx

dx\

dxj

SimpHfying, dividing by dx, and rearranging,

d'^y

dy

dx^

dx

D-j^^-u-{:-ky =

II

Similarly, the material balance in the entry section gives

dc
_ d^c
=
D
^-M
dx
dx

III

which can also be obtained from II by removing the reaction term. Equations II and III are both second order Hnear differential equations with zero
right-hand side, thus the complementary function is the complete solution
in both cases.

The

auxiliary equation of equation II

.'.

where

is

Dm^-um-k =
m = u{la)l2D
a

= Vl + 4/cD/w^

IV

V
VI

Therefore, the solutions of equations II and III are

and

= ^expg(l + a)]+Bexpg(l-a)]

c^a + pexpiuxjD)

VII
VIII

2.

ORDINARY DIFFERENTIAL EQUATIONS

which contain four arbitrary constants, A, B, a and

p.

61

The four boundary

conditions are
at

at

= 00,
= 0,

at

c
c

IX

Co

dc

y
dy

dx

dx

X
XI

0,

dy
at

=
=

XII

dx

The

first

condition specifies the state of the feed stream, and the second
The third condition, taken with equation

ensures continuity of composition.

X,

necessary to conserve material at the boundary assuming that the

equal in both sections. The final condition forbids diffusion

is

diffusivities are

out of the reactor and is necessary as a conservation law for the section
following the reactor. A full argument supporting the last condition is given
in the reference quoted at the start of the problem. Equations IX to XII
give respectively,

= Co
+p = A+B
2p = A(i + a) + B(l-a)
a

oi

+ a) exp

[uL

XIII

XIV

XV

(1

+ fl) +B(l-fl)exp

[i"-">]

XIV and XV gives

=
A{l-a) + B{l + a)
2co
XVI and XVII for A and B gives

XVI

Eliminating a and ^ from XIII,

Solving

2co(a-l)

XVII

uLa\
XVIII

K
B
where

+ l)

2co(a

/uLa\

XIX

K = (a + 1)^ exp (uLallD) -(a- 1)^ exp - uLajlD)

XX

Putting these values of

r'^p

(a

A and B

into equation VII gives the final result,

+ l)exp<|

ua

(L-x)^ + (a-l)exp

-i<^-"}]
XXI

From equation XXI, if diffusion

equation results thus,
cp-y
after

is

neglected, the

first

exp ( - kLjii)

order piston type flow

XXII

a rather involved application of E'Ropital's rule as D-*0 (see Section

3.3.8).

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

62

Problem

2.

The Continuous Hydrolysis of Tallow

in

a Spray Column

mixed with 2270 Ib/h of high pressure hot water

is fed into the base of a spray column operated at a temperature of 450F
and a pressure of 600 Ib/in^. 4120 Ib/h of water at the same temperature and
pressure is sprayed into the top of the column and descends in the form of
droplets through the rising fat phase. Glycerine is generated in the fat phase
by the hydrolysis reaction and is extracted by the descending water so that
5560 Ib/h of final extract containing 12-16% glycerine is withdrawn continuously from the column base. Simultaneously 8900 Ib/h of fatty acid
raffinate containing 0-24% glycerine leaves the top of the column.
If the
effective height of the column is 72 ft and the diameter 2 ft 2 in, the glycerine
equivalent in the entering tallow 8-53 % and the distribution ratio of glycerine
between the water and the fat phase at the column temperature and pressure
8070 Ib/h of a tallow

fat

10-32, estimate the concentration of glycerine in each phase as a function

of column height. Also find out what fraction of the tower height is required
principally for the chemical reaction. The hydrolysis reaction is pseudo first
order and the specific reaction rate constant is 0-17 sec" ^

is

Solution

Consider Fig.

2.5.

which

the hydrolyser column.

illustrates the flows

of extract and raffinate in

and G Ib/h of extract

represents Ib/h of raffinate

L Ib/h

G Ib/h
Vh

x+(5x

y+<5y

'z + dz

6h

H
t:

'

',

U
i

,^0

yo

Fig. 2,5.

Continuously operating fat-hydrolysing column under steady state conditions

suflfix signifies the position in the hydrolyser column.

In addition to the symbols given in Fig. 2.5, let

whilst the appropriate

2.

=
=
y* =
z =
S =
a =
K=
X
y

m =
=
=
h =
w =
H=
k
p

Weight
Weight
Weight
Weight

ORDINARY DIFFERENTIAL EQUATIONS

63

fraction of glycerine in raffinate.

fraction of glycerine in extract.
fraction of glycerine in extract in equilibrium with x.
fraction of hydrolysable fat in raffinate.

Sectional area of tower.

volume of tower.
Overall mass transfer coefficient expressed in terms of extract
Interfacial area per unit

compositions.
Distribution ratio.
Specific reaction rate constant.
Mass of fat per unit volume of column.
Distance coordinate from base of column,
lb fat per lb glycerine.
Effective height of column.

Consider the changes occurring

These are
(a) glycerine transferred

from

in the

fat to

element of column of height Sh.

water phase,

KaS(y*-y)Sh
(b) rate

of destruction of fat by hydrolysis,

kpSzdh
and therefore the

rate of production of glycerine

is

kpSzShi'w

Then a

II

glycerine balance over the element dh

is

by

Fig. 2.5,

Sh

an

an

J
III

glycerine balance between the element

Lzq

and the base of the tower

Lz

Finally the glycerine equiHbrium between the phases

)^*

From

is

= mx

the last two parts of equation III

and equation

V it can be

KaSmx = KaSy - G -^

kpS

shown

that

VI

ah
and substitution of IV

is

in the element glycerine balance III,

[^ + j{y-yo)-x] Sh-L^3h =
L
Iw
dh
J

-G^Sh
dh

VII

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

64

Multiply equation VII by (KaSm/LG) and substitute for x from VI,

kpS^Ka [mzQ
fmzo

mG

kpS\KaS

dy~\

LG
KaS

dy

_d^
~

dh'i

KaSm

dy

'

Equation VIII contains the following groups of constants which can be

denoted by the following parameters,

mG

KaS

kpS

Substitution of the above groups into VIII gives

d^
dh^

Equation

is

(P

+ ^)jT + M>^ =

on rearranging

7 \ryo

a second order linear differential equation with constant

coefficients.

The

auxiliary equation

The

roots of

is

fnl

XI

+ {p + q)mi-{-pq =

XI

are

mi =

or

and therefore the complementary function

is

y.^Ae-^'^ + Be-"''

XII

The right-hand

side of equation
is a constant and therefore the particular
a constant and equal to the right-hand side of equation X divided
by the coefficient of >^, as given in Section 2.5.4.

integral

is

and the complete solution

y

yp=

ryo-mzo/w

^_i

=^

^^^^
^AAi

is

= Ae~ .^ + Be-^ +

''^^-'"'^/"

r-1

XIV

where A and B are arbitrary constants to be evaluated from the following

boundary conditions.
x =
h = 0,
At

and

at

From

H,

equation VI,

r-ldy
q

dh

and from equation XIV,

dy
dh'

= -pAe-^'-qBe-''

XV

2.

ORDINARY DIFFERENTIAL EQUATIONS

(\pA + qB) + C =

A+B +
and

Therefore at

65

=H

at

XVI

Ae-P^ + Be-^^ + C =

XVII

two simultaneous equations in A and B. The

arbitrary constants are best evaluated from these two equations by defining
a further constant. Thus let
Equations

XVI and XVII

are

= iZZ:i =
q

Inserting this constant into equations

A{ve-^"-re-''')

A and B

Substituting these values of

y(ve-''"-re-'")

XVIII

XVI and XVII and

Bive'^^'-re-'")

and

i+^
KaL

solving gives

= C(e-^^-i;)
= qr-e"^'^)

XIX

XX

into equation XII,

Cl(r-e-^")e-'"' + {e-P" -v)e-'' + ve-^^-re-'"']

XXI

C in

and expressing yQ from

that y = yQ a.t h = 0,
mz,

>'-w(r-e-)

terms of the other variables and using the fact

,-,-V-...(^)

and then

Substituting the above value of >'o into equation XIII

XXI

XXII

into equation

gives

mz
XXIII
Equation XXIII gives the weight fraction of glycerine

in the extract

a function of column height h.

Allowing for the solubility of water in tallow, taking

and using the data given

L=
.-.

4-544,

flow rates,

in the problem, gives

G=

8540,

mean

phase as

Solving equation XXIII for

0-198,

Ka

with

3760,
q

H=

>;o

0-188

0-00348 Ka,
72-0

ft,

?^
Ka

gives the value of the

mass

transfer coefficient as

Ka = 14-2 lb glycerine per hour per ft^

of Ka = 14-2, values of y, y'^ and z can

XXIV

With the value

be determined as
functions of column height with the aid of equations XXIII, VI, V, IV. The
results of the arithmetical calculations are shown in graphical form in Fig.
2.6. Figure 2.6. shows that the chemical reaction is virtually complete in the
bottom 30 ft of the column, or 40% of the column.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

66

73-5c

67-5

cerine at interfoce

O'l

in

extract

0-2 0-3 0-4 0-5 0-6 0-7 0-8 0-9

10

Weight fraction
Fig. 2.6.

Concentration distributions in hydrolysing column

The above analysis of a continuous hydrolysing column was abstracted

from "The Analysis of a Continuous Fat Hydrolysing Plant", by Jeffreys,
Jenson and Miles.

2.6.

Simultaneous Differential Equations

These are groups of differential equations containing more than one

dependent variable but only one independent variable, and in these equations, all the derivatives of the different dependent variables are with respect
to the one independent variable. This distinguishes an equation which is a
member of a group of simultaneous differential equations, from a partial
differential equation which will contain more than one independent variable

and consequently

will also contain partial derivatives.

All the derivatives in

the simultaneous differential equations will be total derivatives.

The

basis of the solution of simultaneous differential equations

is

alge-

braic ehmination of the variables until only one differential equation relating

two of the variables remains. The final differential equation is solved in the
ordinary way by one of the conventional methods. If the equation is Hnear
with constant coefficients, one of the methods already presented in this
chapter will be suitable, otherwise a method to be given later must be
employed.
t Jeffreys, Jenson

and Miles, Trans.

Inst.

Chem. Eng. 39, 389

(1961).

ORDINARY DIFFERENTIAL EQUATIONS

2.

67

The variables to be eliminated depend on the set of equations, and the

ultimate information required from the mathematical analysis. The variable
may be the independent variable or one of the dependent variables. These
possibihties will
2.6.1.

now

be considered.

Elimination of the Independent Variable

Consider the pair of simultaneous differential equations,

and

dx
^=/i(^,>')

(2.103)

^^^^(^'^'^

^^-^^^

is the independent
which x and y are the dependent variables and
two functions oi x and y. Equations of this
kind can be solved by ehminating the independent variable by dividing one
equation by the other, say equation (2.104) by equation (2.103) to give

in

variable. fi{x,y) and/2(x,>^) are

(2.105)

T=^tF\
which

is

an ordinary

differential

equation in x and y, and would be solved

as such.

Example 1 solved in paragraph 2.3.3 was really one of the above type.
EUmination of the time gave a homogeneous equation which was solved by
standard solution methods for homogeneous equations.
The above method of ehminating the independent variable is restricted
to first order first degree equations in which the independent variable appears
only in the derivative. Consequently its scope is very Hmited.
2.6.2.

Elimination of

One

or

More Dependent

Variables

This approach has considerably wider application, however it becomes

very involved with equations of high order and therefore it would be better
to make use of matrices if there are many equations. The matrix methods of
solving large sets of simultaneous equations are presented in Chapter 12,
whereas the following paragraphs will be restricted to the simpler situations.
Thus the independent variable can be eHminated by,
(a) Taking advantage of the algebraic properties of the differential
operator.
(b) Systematic ehmination.

By Use of the

It has been pointed out that the

of algebra and therefore can be used
in the ehmination process. The method is best illustrated by an example.

(a)

operator

Differential Operator.

D obeys the fundamental laws

Example

1.

Solve

+ D-6)y + {D^ + 6D + 9)z =

(D2 + 3D-10)>' + (D2-3D + 2)z =
{D^

and

II

MATHEMATICAL METHODS

68

Equations

and

II

IN

CHEMICAL ENGINEERING

can be written

=
(D-2)(D-h5)y + iD-2){D-l)z=0
by (D + 5) and IV by iD + 3) to give
(D + 3)(D + 5)(D-2)>' + (D + 5)(D + 3)^z =
(D + 3)(D + 5)(D-2)>' + (D + 3)(D-2)(D-l)2 =
(D-b3)(D-2)y + {D + 3yz

and
Multiply

III

and

III

IV

V
VI

Elimination of y and simplifying the remainder gives

(llD + 13)(D + 3)z

whose solution
where

A and B

VII

is

are arbitrary constants.

Substitution of z from equation VIII into equation

(D2 + D-6)y

The terms involving Be

add up
The particular integral of y is

to zero.

D' + D-6

IX

e-^^^/'^

(n) -(-n-) + ^J^

^"^

yp

gives

r/13\2

E=

where

'

13X/11

D-2

D + 3j

XI

Ge-'''^''

G= -^E^^A

where

The complementary function

is

y,

and the complete solution

y

He^''+Je-^''

XII

is

He^''+Je-^''+^Ae-^^''^^^

(b) Systematic Elimination.

This term

is

proposed for the eUmination

amount of algebra involved

to a minimum.
method consists of setting up a table indicating the number
of times and in what form each variable appears in the simultaneous equations. Then the variable that appears in the simplest manner is first eliminated by making the appropriate substitution. Following this the second least
frequent variable is removed from the remaining equations by the appropriate substitution, and the substitution process is continued until only one

process which reduces the

Essentially the

equation remains. Generally terms which appear as derivatives are left to

the final steps of the substitution process.
The following example will
demonstrate the technique.

Example 2.t 10000 Ib/h of sulphuric acid (specific heat 0-36) is to be

cooled in a two-stage countercurrent cooler of the following type. Hot acid
t Suggested

Birmingham.

by W.

M. Crooks, Dept.

of Chemical

Engineering,

University

of

2.

ORDINARY DIFFERENTIAL EQUATIONS

69

at 174C is fed to a tank where it is well stirred in contact with cooUng coils.
The continuous discharge from this tank at 88C flows to a second stirred
tank and leaves at 45C. CooHng water at 20C flows into the coil of the
second tank and thence to the coil of the first tank. The water is at 80C as
To what temperatures would the
it leaves the coil of the hot acid tank.
if due to trouble in the supply, the cooling water
suddenly stopped for one hour?
On restoration of the water supply, water is put on the system at the rate
Calculate the acid discharge temperature after one hour.
of 10000 Ib/h.
The capacity of each tank is 10000 lb of acid and the overall coefficient of
heat transfer in the hot tank is 200 C.H.U./h ft^ C and in the colder tank
130 C.H.U./h ft^ C. These coefficients may be assumed constant.

contents of each tank rise

Solution

The steady
and the steady

coohng system are shown in Fig. 2.7,

data calculated from the illustrated conditions are

state conditions of the

state

Water rate before failure of supply

Intermediate water temperature between the two tanks
Heat transfer area of coil in hot tank
Heat transfer area of coil in cold tank

80

40^*

20
^

7740

Ib/h.

40C.
69-3 ft^.f
95-8

ft^.

_
/

88

i
Fig. 2.7.

=
=
=
=

45

Sulphuric acid cooling system

Water Fails

Water flow

is

zero.

For the purpose of the analysis

acid flow rate be

M Ib/h

acid specific heat be

feed acid temperature be Tq C

acid temperature ex tank 1 be T^

=
=
=

let.

10000
0-36

174

C
acid discharge temperature be T2 C
capacity of each tank be
time be 6 hours.

V lb

10000

t In practice, these figures would be rounded off to 70 and 96 but to reduce errors of
calculation the extra figure has been retained (see Chapter 10).

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

70

Then

the heat balances are as follows.

Output Accumulation

Input

Tank

MCTq - MCT^ =

FC

dO

AT'

MCT^ - MCT2 = VC^

Tank 2

II

dO
These are simultaneous first order
=
solved in succession. Because
becomes

differential equations,

V in

but they can be

the numerical example, equation

which has the solution,

To-Ti =
and when 6

T^

0,

Ke-'=n4-Ti

88C the steady

Therefore when 9

state temperature.

K=

.-.

IV

S6

1*0,

Ti= 174-86e-^'= 142-4C

Similarly for tank 2,

from equation

II,

174-86^-^-72 =

d_T2

-^

VI

dO
This

is

factor

first

order linear equation which can be solved using the integrating

e^.

and solution of equation VI

Substitution of the integrating factor

T2

which gives for

gives

= 174-(860 + 129>-^

VII

1-0,

T2

VIII

94-9C

Water Supply Restored

Water flow
derivation

be

^3

rate

let

is

now

10000 Ib/h but for the purpose of the following

Ib/h and let water supply temperature

the water rate be

C.

water temperature ex tank 2 be t2C.

water temperature ex tank 1 be /iC.

Then

the heat balances are as follows.

Accumulation

Input

Output

Wt2 + MCTo

- (Wt^+MCT^) =

Tank 2

Wt^ + MCT^

rl'T

Tank

(Wt2 + MCT2)

KC
du
dT.

= VC--^
du

IX

ORDINARY DIFFERENTIAL EQUATIONS

2.

The heat

transfer rate equations for the

71

two tanks are

{T,-t,)-{J,-h) 1
A r
^('-'^)^ = n
^'^'
[ln(r.-r,)-ln(r,-J
vv(,

and

^'

<

^^(^,-.3)

These equations simpUfy

C;,^,

''"

[i(r,-r,)-ln(r,-d

to,

exp

-^-^ ==a(say)
(-^)

XIII

and
Equations XIII and

XIV

can be written more simply,

XV

= ti-at2
U^-P) = t2-Ph

T,{l-a)

XVI

Equations IX, X,

XV and XVI are the four equations which have to be solved

simultaneously.

The

table,

Table

Table

of elimination. Hence the following

frequency of the appearance of the four variables.

difficulty is that

2.2, gives the

2.2.

Frequency of Appearance of Variables

Ti

Ti

^1

3 plus

2 plus

In Table 2.2, "1" indicates the appearance of the variable and

From Table

'*

+"

its

should be ehminated first, then t2, because

both Ti and T2 also appear in a differentiated form. Substitution of equa=
tion XV into equation IX, and using the numerical property that
=
derivative.

2.2, t^

V, gives
t2

+ CTo-CT,-(l-(x)T,-oit2 = C--^

XVII

au

Now

eliminate

equations

^2

from the system by substituting from equation XVI

XVII and

C^
do
+ CT,-CT^-{\-P)T,-pt, = C^

(l-a)(l-p)T2 + {l-(x)pt, + CTo-CT,-(l-oi)T,

and

into

X thus,

t,

XVIII

XIX

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

72

Rearranging equation XIX,

^'^'^^Id^
and

XX

(i-^+c)r2-(i-^)r3

differentiating,

^-37r

= ^-j;^ + (i-^+0-^

Using equations XX and XXI to eliminate T^ and

from equation XVIII gives

XXI
differential coefficient

its

C^^ + (2C + 2-a-^)C^^^ + [C^ + C(l-aP) + (l-a)(l-/?)]r2

= [C(l-a^) + (l-a)(l-^)]^3 + C^ro
Equation XXII
coefficients.

XXII

a second order linear differential equation with constant

is

The operator does not have simple

factors

and

it is

advantageous

to introduce the numerical values at this stage.

.-.

U^A^

200x69-3

10000

U2A2

130x95-8

10000

(l-a)(l-i5)

1-386,

/.

0-250

1-245,

/.

0-288

and

0-534

(1-a/?)

0-928

Equation XXII thus becomes

-# +

6-06

dO

ad

7-7072

XXIII

308

which has the solution,

T2

The

XXIV

Ae-^-^^^ + Be'^-^^^-^AQ-O

conditions for this solution are the final temperatures from the
previous operation given by equations V and VIII. Before these can be
initial

used, Ti has to be determined

conditions give

from equations

+B =
-1-27/1 + 1-175 =
yl

and
Solving for

A and

B,

and
T2

XXIV and XX. The boundary

2.7.

XXIV

gives

XXVII

+ 54-3e-^-^^^ + 40-0

so that the acid discharge temperature after

Methods have been presented

XXVI

62-8

substituting in equation
0-6e-''-2^^

XXV

54-9

is

48-9C.

Conclusions
in this chapter for solving the simpler types

of differential equations which the chemical engineer

is

likely to encounter.

2.

There
solve

ORDINARY DIFFERENTIAL EQUATIONS

73

is no general method of solution, but the methods included should

most first order equations.

Non-linear second order equations are frequently insoluble analytically,

but the procedures of Section 2.4 may be useful in providing at least an
approximate yet valuable solution.
The most important type of equation dealt with is the linear differential
equation with constant coefficients. The complementary function can always
be found easily and the methods of undetermined coefficients and inverse
operators are recommended for the determination of the particular integral.
If both of these methods prove to be inadequate, the method of variation of
parameters is always worth trying.

Chapter 3

SOLUTION BY

SERIES

3.1.

Methods of

Introduction

number of
order equations were

solution were presented in Chapter 2 for a limited

types of ordinary differential equation.

Most

first

solved in principle, and higher order equations were classified into linear and
non-Hnear differential equations. It was shown that non-linear equations
can be solved in the special cases where either jc or >^ is absent, or the equation

homogeneous; and that the solution of any linear differential equation

can be resolved into the sum of a complementary function and a particular
integral.
The complementary function was found in all cases where the
coefficients of the derivatives are constant, or the reduced equation is homogeneous, but the general linear equation with variable coefficients was not
solved. Particular integrals were found by a variety of methods, some of
these involving knowledge of the corresponding complementary function.
The purpose of the present chapter is to solve the general linear ordinary
differential equation with variable coefficients. The conditions which these
coefficients must satisfy are given in the chapter, but they are invariably
satisfied in problems of chemical engineering importance.
As is usual in
treatments of this subject, only the complementary function will be found in
most cases, the particular integral can then be found by using the methods
of Chapter 2.
When the differential equation has constant coefficients, the solution
involves the exponential function. This function, in common with many
others, can be expanded into a series of ascending powers of x. The series
itself can be considered to be a solution of the differential equation, a property which may be demonstrated by differentiation and substitution into
is

the equation.

form of

Methods given

infinite series

in this chapter will

can be obtained for

many

show

that solutions in the

linear differential equations

with variable coefficients. Of the series that arise, only a few can be summed
allowing the solution to be expressed in closed form, but the other series can
be accepted as solutions provided that they are convergent. For practical
purposes, the series will be useful only if they are rapidly convergent. It is
therefore necessary to consider the convergence of series, before differential
equations can be solved by their use.
3.2.

A series

Infinite Series

of numbers

"l+W2 + "3 + .- + "n =

is classified

'^n

(3.1)

by determining the behaviour of S as n increases to infinity.

74

SOLUTION BY SERIES

3.

S^S some

(a) If

finite

number

75

as ->oo, the series

is

termed "conver-

gent".

S-*co as ->oo, the series is termed "divergent".

In other cases, the series is termed "oscillatory".
When considering oscillatory series, the value of S riiay oscillate between
either finite or infinite limits and a further classification can be based on this
(b) If
(c)

distinction.

In the application of series to differential equations, u will be a function

of X, consequently S will also be a function of x. Because x may take positive
or negative values, or even complex values it is desirable to allow u to be
complex. The properties of complex numbers will be investigated in Chapter
4, but for those readers who are not familiar with the elementary properties
of complex numbers, the term "absolute value" refers to the numerical
magnitude of the number and is signified by vertical fines.
|i/|

The

series (3.1)

is

u.

said to be "absolutely convergent" if the series

.|"l|

is

absolute value of

|"2|

|"3|

-..

(3.2)

lWn|.

convergent.

Example

1.

w;,

.*.

This series

is

z"

= z + z^ + z^ +

S-

ratio z.

II

1-z

|z|<l, z''-'^-^Oasn-^oo,

(a) If

.-.

Hence the

S-^z/l-z

series is convergent.

(b) If |z|>l, |z'''^^|->oo as

^oo,

therefore the series diverges if z

positive, or oscillates for all other values

(c) If

^_^
1-z

and

+ z"

common

a geometrical progression with

..

,.,

Hence the

1,

of

is

real

z.

n which tends to infinity as n increases indefinitely.

series is divergent.

(d) If |z| = 1 and z 7^ 1, the series oscillates within finite limits.

This series thus illustrates all four types of behaviour.
3.2.1. Properties

of Infinite Series

The following

are three useful properties of infinite series,

a series contains only positive real numbers or zero, then it must be
either convergent or divergent; it cannot oscillate.
(ii) If a series is convergent, then w-^0 as - 00. This is a property of a
(i)

If

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

76

convergent series it is not a proof of convergence. A divergent series for

v^hich u-^0 is illustrated in example 2, part (b) below.
(iii) If a series is absolutely convergent, then it is also convergent.
Statement (iii) can be proved as follows by referring to (i). If u is real,
put u = a-b where a = i(\u\ + u) and b = i (\u\-u). Thus a, the
series of positive terms is separated from b, the set of negative terms so
that neither a nor b is negative.
;

/.

fli

+ fl2 + 3 +

...

+ fl ^

|wi|

+ |w2| +

..-

+ |Wn| ^ S

Since the series a only contains positive or zero terms, and the sum cannot
exceed S, then statement (i) above proves that ^a is convergent.
Similarly,

^b

is

convergent.

between +S and S. If u is complex, a

by separating real and imaginary parts.
Thus, absolute convergence implies more than mere convergence.

must tend to some

finite limit

similar proof can be constructed

3.2.2.

Comparison Test

Only convergent series will be of importance in the solution of chemical

engineering problems, but the following tests will be stated in full as they
determine convergence or divergence. It is important to appreciate in these
tests that if

shown that a particular series is not convergent,

mean that it is divergent; it may oscillate.

it is

not necessarily

The comparison
(i)

test is the simplest

If |w|<i^ for all

convergent, then ^u
(ii)

is

lf\u\'^v for all

n>N

where

of

this

does

and consists of two parts

some finite integer, and 2^n

all
is

is

absolutely convergent.

n>N, and

^v

is

divergent to +00, then J^u

is

also

divergent.

This test involves comparing the series under investigation with known
convergent or divergent series, on a term by term basis, and it becomes more
useful as the behaviour of more series becomes known. Comparison with the
simple geometrical progression with a suitable common ratio as in example 1
is very useful for part (i) of the test, and the divergent series given in example
2 (b) below is useful in part (ii) of the test.

Example

The above
(a)

series

= n~^ (p real)
= 1 + 2-^ + 3-^ + + n"^ =
u

2,

..

X ''

has different properties for different values of p.

p>\.

=1

(ir+HY
{iy+iiyHiYH^y
etc.

< 2(1)" = (!)"-'

< 4(1)" = ay-'

SOLUTION BY SERIES

3.

rth line contains 2''"^ terms

The

and begins with

77
(1)'^

side is a geometrical progression with a finite sum to infinity.

Therefore the series is convergent by Section 3.2.1 (i).
(b) p = I. In order to show that the series is divergent in this case, the
following algebraic relationship is required.

The right-hand

Grouping the terms of the

2n

nk

+k

series as

-k'

>

shown below, and applying

the above

algebraic inequaUty, gives

Xj_J.4.i4.i4.i4._l_4.J^

L._I_4.JL-

= (i+iV)+(i+iV)+(i+TV)+(i+TV)+*
>i + i4--2-4-^4-J- = 1
(l)

.-.

The
the

+ (i + i + i) + (i+...+lV) + (TV+-. + 4V) + .- > 1 + 1 + 1 + 1 +

rth bracket contains 3'"^ terms centred

first

terms is greater than

to exceed any chosen

i{3'' 1)

can be made

on

3^

"'".

By taking

r.

finite value,

Therefore, the

...

sum of
sum

sufficient terms, the

and hence the

series is

divergent.
(c)

p<L
IV

The

series is thus divergent

Section 3.2.2
3.2.3.

>

for

n>

by comparison with part

(b) using the test

of

(ii).

Ratio Test

This also consists of two parts

(i)If
for all
(ii)

n>N,

then the series

is

k>l

absolutely convergent.

<1

If
'n+l

n>N,

then the series is not convergent.

first part of the test it is necessary to find a fixed finite
number k to satisfy the inequality for all n>N. In the second part, if the
series only contains positive terms, the test proves that the series is divergent,
but in general the series may oscillate.
When the second part is not satisfied and k = I in the first part, thus
violating the inequality, the test is not conclusive. That is, the series may be
convergent, divergent or oscillatory, the test will not determine which.

for

all

Regarding the

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

78

Example

S=l+i + i + i +
n+1
^n+l
.*.

this

3.2.4.

+-1

1 for all n>N, but it is impossible to find a value of k such

of the ratio test is satisfied. The test is therefore inconclusive,
series has already been proved divergent in Example 2 (b).

|w/w +

i|>

that part

but

(i)

Raabe's Test

The

ratio test is inconclusive

to resolve this difficulty.

The

when k =

test is

I, but Raabe's test

again in two parts

is

designed

^1 +

(i)if
^n+l

where a >
(ii)

for all

> A^,

then the series

is

convergent.

If

then the series

is

divergent.

The second part of Raabe's Test proves

that the series of

Example

is

divergent.

There is still a case unresolved when a = 1 in part

covered by a further extension of the same technique.
3.2.5.

a case which

(i),

is

Gauss' Test

There

only one part to this

is

u
If

test.

^^i^

'+i

< C for n > N, then the series is divergent,

means that when I + l/n has been subtracted from the ratio of
terms, and the result has been multipHed by n^ if the function

where \f(n)
This

test

successive

f(n) remains finite as n-^oo, then the series is divergent.

These last three tests fall into a natural order of increasing resolving

power, each one testing the behaviour of the ratio of successive terms in
greater detail. As mentioned in Section 3.1, a series is of little practical value
unless it is rapidly convergent; consequently, the more powerful tests are
rarely necessary.
3.2.6.

The Integral Test

If /(x)

x>N,
if

is

where

and only

if

a positive integrable function of x which decreases for all

A^ is some positive integer, then the series ^/() converges
f(x) dx exists.

SOLUTION BY SERIES

3.

79

Each of the above tests is sufficient to prove convergence or divergence,

but the large number of series for which they are inconclusive indicates that
all of the tests need not necessarily be satisfied by any particular series. The
following example illustrates an inconclusive case which is nevertheless
convergent.

Example 4

The

successive values of w/w+i are

4

3
2'

3'

where

16

9'

4>

2 7'

27
8

>

after the first term, alternate values are greater

and

less

than unity.

Therefore, the ratio test "proves" the series to be alternately convergent and
divergent.

The
the

value
3.2.7.

is

series

sum of

of odd terms

the even terms

is

a geometric progression with sum to infinity 2

Therefore, the series I is convergent to the

is 3.

5.

Alternating Series

an alternating

series if

Wl~"2 + W3~"4+--w>0 for all n. If, in

successive terms decreases for

all

(3-3)

addition, the magnitude of

values of n and w->0 as ->oo, then the

series is convergent.

Proof
S'2

(Wl-W2) + (W3-W4)+- + (W2m-l-2m)

is positive and the next term W2m +

added, 52^ and S2ni+i must both be greater than zero.

Since each term in the brackets

^2m+l

J^iust

be

= Ui-{U2-U^)-(u^-U5)-...-(U2m-U2,n+l)

Again, each term in the brackets is positive, and removing the last term to
obtain S2m reduces the value further. Therefore, 5*2^ and Sj^ + i are both
less than Wj.

.*.

Both S2ni and S2m+i increase as

they are separately convergent.

for all n.

Wi

m increases

but never exceed u^, and hence

Because t/^0 as -oo, S2m and S2m+i
must both converge to the same limit S. Therefore the alternating series

(3.3) is convergent.
3.3.

Consider the power

Power

Series

series
00

ao
If

\aJa+i\^R

+ aiZ + a2Z^+

...

(3-4)

Y.^n^"

as -^oo, then the ratio test can be applied as follows.

|w+i

a + iz"-'^

a+i

\z\

\z\

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

80

as -> 00. Therefore, if

|z|

< R,

\uju +

unity and will be convergent by part

(i)

will tend to some limit greater than

of the ratio test. In detail, if |z| =

R-e
where

some small

e is

positive

number

less

than R,

R=

>

R-
(s/R)

= l+(e/i?) in Section 3.2.3, and substituting into equation (3.5)

proves that the series (3.4) is convergent.
If |z|>i?, a similar proof shows that the series (3.4) is divergent. Therefore, ^^^z" is convergent for |z|<i^ and divergent for \z\>R. R is termed
the "radius of convergence" of the power series and as shown above, R is
the hmit of the ratio of successive coefficients. If \aJa+i\-^co as ->oo,
then the power series will always converge. If \aJa + j^\-^0 as ->oo, then the
power series will only converge if z = 0; i.e. if all terms are zero.

Taking k

3.3.1.

Binomial Series

The

series

obtained by expanding

(1

+z)^

is

H.,, + (zi),.^Kzll(zlV^
1.2.3

1.2

where
real.

signifies the

binomial

coefficient, z

be complex, but p

p(p-l)(p-2)(...)(p-n +
1.2.3.

n+l
Therefore, the binomial series

+1
1

p-n
is

l)

...n

as M -> 00

convergent for

|z|

<

1.

Exponential Series
z'

^z"

z'

^=^^^^2-^3-!^- = ?T!
Here,^=

(3.7)

\/(n\)

.'.\aja +

i\

Therefore, the exponential series

3.3.3.

is

In this case
a

3.3.2.

may

=
is

+ 1-^

as -> oo.

oo

always convergent.

Logarithmic Series
ln(l

+ z) = z-^z^ + iz'-JiZ*+

...

(-1)"

(3.8)

3.

SOLUTION BY SERIES

+l

81

>

as n -> 00.

'n+l

Therefore, the series for ln(l+z)

is

There are two

convergent for |z|<l.

further interesting cases of this series.

(a) z

1.

Equation

(3.8) gives

ln2=l-i + i-i +

(3.9)

...

This is an alternating series with decreasing terms, and t/->0 as ->co.

Therefore the series is convergent by Section 3.2.7.
(b) z

1. Equation

(3.8) gives

lnO=-(l+i + i + i +

...)

which was shown to be divergent in Section 3.2.2 (b). The series

"conditionally convergent" and not absolutely convergent.
3.3.4.

Trigonometric Series

sinz

= i:^-f-
- + --...
5!
(2n + l)!

= z-

cos.
These two

series are

z^
r-iyz^"
l-_.--...=i:L_l_
z"^

convergent for

values of

all

(3.11)

z.

Hyperbolic Series
3

These two

If z

^2+l

00

- + - + ...=i;^^^^

sinhz

coshz

z^
z^"
l++ -+...=!

series are

convergent for

all

(3.12)

z"^

Inverse Trigonometric

3.3.6.

values of

(3.13)

z.

and Hyperbolic Functions

tan y, the inverse tangent function

That

(3.10)

3!

3.3.5.

(3.9) is thus

is

defined by

tan"^z.

the angle measured in radians, the tangent function of which

y
has the value z.
The other inverse trigonometric functions and inverse
is,

is

hyperbohc functions are defined

similarly.

The

series expressions for

two

of these inverse functions are

3

tan-'z

oo/_-iy

z-^3 + \-...=I^
5
Q

^3

tah-z =
These two

series are

convergent for

|z|

<

2n +

(3.14)
l

oo^2n+l

^5

-+- +

2n +

;^'

...

1.

12-^

(3.15)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

82

Series expansions can be developed for the functions tan z, tanh z,

z, sinh"^ z, cosh"^ z but their form is a httle obscure and no
purpose would be served by presenting them here. The above series and
many others can be obtained by applying Taylor's theorem, which develops
a power series in ascending powers of the variable.

sin"^z, cos~^

Taylor's Theorem

3.3.7.

If

:3C

is

a real variable, and f{x) can be differentiated n times at x

a,

then

Aa + h)

=f(a) + hfXa) + ihT(a)

+ -+

+ ^^f"\^) + ^l

^^Z^y,^'""'^''^

(3.16)

where dashes denote differentiation with respect to

x,

^^0

and

as h-^0.

Most functions of a continuous independent variable which arise in

chemical engineering can be differentiated indefinitely at all points of the
problem, and the conditions for the application of Taylor's theorem are
always satisfied.
UHopitaVsRule

3.3.8.

It is

form

0/0.

sometimes necessary to resolve an indeterminate fraction of the

This difficulty usually arises for some special value of a variable

in a case such as

.=^

(3.17)

where f{x) and g{x) are differentiable n times, and f{a) =g(a) = 0. A
value of y can be calculated for all values of x except x = a. Expanding
f(x) and g(x) about the point x = ahy Taylor's theorem,

f(x)=f(a)H^-a)r(aHi(x-arr(a) +
g(x) = g(a) + (x-a)gXa) + Kx-a)'g'Xa) +

...

..,

_ f(a) + (x-a)fXa) + i(x-ayf"(a) +

^
g(a) + (x-a)gXa) + i{x-ayg"(a) +
and gXa) # 0, then a factor (x a) can

>^-

If /(a)

g{a)

both numerator and denominator of equation

.3 ^g.
,..

be taken out of

(3.18), giving

naHi(x-a)f"(aH...
^

g'ia) + i(x-a)g"(a)+...

In the limit as x-^a,

V ->

f'(a)
(3.20)

Example 1 Evaluate y = (sin x)/x when x = 0.

The value of >^ can be determined for any value of x no matter how small,
but not for zero. The series expansion of sin x, equation (3.10) can be used
.

3.

SOLUTION BY SERIES

83

in this example, giving

-{'44->
x^

X"

In the limit as x-^0, it can be

is valid for all x no matter how small.
seen that y-^ 1 and this is the obvious value to take for y when x = 0.
This process of resolving indeterminate fractions can be repeated k times

which

provided that k<n.

and

The

g(a)
first

k terms

in

g'(a)

...

g^'-'Ka)

0.

both numerator and denominator of equation (3.18) are

can be taken out of the remaining terms and the limit

zero, a factor (x a)''

as

x^a

will give

y=P'\aWKa)

(3.21)

As soon as a non-zero derivative occurs in either series, both series must be

terminated and the ratio evaluated as zero, infinity, or any finite value.
Example

2.

Find the

limit as

x^O

of

x^ sin X
1

With the above notation, f(x)

.-.

.'

and

and

cosx

x^ sin x, g(x)

= ^(0) =
I'M = 2x sin X + x^ cos x
g'{x) = sinx
fXx) = 2 sin X + 4x cos x
g'X^) cosx

COSX'.

/(0)

as X

When

- 0,

/'(O)
g'(0)

x^ sin x

/"(O)
ff"(0)

fM

/"(O)

=
=
=
=

non-vanishing derivative occurs in the denominator, as

and when the first non-vanishing derivative occurs in
the numerator, the hmit is infinite. If, as in the first example, both derivatives
of the same order become non-zero simultaneously, then the Hmit is finite.
the

here, the limit

3.3.9.

first
is

zero,

Application of Series in Chemical Engineering

Some

of the problems encountered in chemical engineering prove very

by standard analytical methods, and some even defy analytical
solution.
Consequently, methods of successive approximation, and the
numerical methods of Chapter 1 1 are used extensively. All of these methods
can only be used when they are convergent, since in practice, a fundamentally

difficult

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

84

being terminated at some finite point. Later in this chapter,

be developed for some problems, and although a series
solution may be convergent and therefore satisfactory from a mathematical
point of view, it may converge too slowly to be of any practical value.
Unless a series is rapidly convergent requiring only about five terms to
express the solution adequately, the numerical work becomes prohibitive
and the solution is of little practical value. No harm can come from this
situation, since the eventual solution will be valid mathematically, and in
Section 3.5.6, Example 6, illustrates this situation. The danger hes in the
opposite situation, when the first few terms of a series decrease rapidly
giving an apparently useful solution, while the series as a whole is divergent
and therefore not vahd mathematically.
divergent series is of no practical
value, and therefore any series obtained must always be checked for convergence. In this respect the advanced tests given in Sections 3.2.4, 3.2.5,
and 3.2.6 may be needed, but for most practical requirements the compariinfinite process is

series solutions will

son and ratio

tests will suffice.

The next example

illustrates the danger of using a divergent series by

Because the example shows that the method of solution does not
work, there is Httle point in explaining the details, consequently only a
rough outhne of the method is presented.
Example 3. In the mathematical description of a stagewise process involving simultaneous gas absorption and second order chemical reaction in
the Hquid phase, a finite difference equation of the following type was

mistake.

developed.

A = a + ^ + 2-
B = a + (l-)(l + i5)
C = 4l + 2(l-)/a]
D = /l(l-)[l + (l-^)/a]
E = X/a
kGH
GmE''
kKH^

a= ;
^ =
P = -7Y7r

where

UV

L'V

and gas flow rates, m is the distribution ratio, E is

k is the reaction rate constant, K is the nett
flow of the transferable component, V is the volumetric hold up per plate,
and
Any consistent system of units
is the corresponding molal hold up.
which makes a, P, and / dimensionless may be used.
The method of deriving equation I is given in Chapter 9. One method
of solving finite difference equations is to expand y+i as a power series in
y, and hence y^ as the same power series in j_i, viz.

L and G
the

are the liquid

Murphree

plate efficiency,

>'n

aiyn + a2yl + a^yl +

Substitution of the series II into equation

and

...

II

equating the coefficients of

3.

like

SOLUTION BY SERIES

85

powers of j_i, gives

Ca^-Eal-D
A a^ ai
a->,

A-a,-al
= Ca2 + (a2-E)(2a^a2 + ^al) + 3ala2a^
A-a^-aj

Ca + (a2-E)Y,aiaj-\-a2Y.aiaja,, + a^Y^i^j^kai +

-"

+ ^nnai"~^a2
III

where J^a^ay signifies the coefficient of y"^^ in the expression of y + ^^

using equation II;^^^^^^^;^ signifies the coefficient of >^""^^ in j^ + i^; etc.
In a numerical example which was attempted by the authors, the first
few terms decreased quite rapidly and the method appeared to be successful.
It was necessary to prove that the series was convergent by using the result
of Section 3.3 that if \a/a+i\-^R as n-^co, then the series would be convergent for \y.i\<R.
Due to the extreme complexity of the equation for the general term III
it

was

difficult to subject it to

the

test,

therefore

two particular

simplifica-

tions were investigated as below.

(a)

Fast Reaction, k large,

.*.

A^-AB =

^-^oo, yl-^oo.
{a

+ p-Ey + 4pE^

- l- + ^(4a-3^)/4j5

.-.

fli

.'.

a2->^''(4a-3)/4^2

.-.

a3->l2(4a-3)/4j5'

appears that aja + i may approximate to a constant (P/?.) in the

The doubt concerns the rapid increase in the number of "correction" terms on the right hand side of equation III. An expression can be
developed for these terms by using the approximate relationship that
It

early terms.

aJa,^,:^PIX<l

IV

and each of the terms in^flj^y will equal a^a. Similarly, all terms in^flja^flfc
will equal firi^a_i, and so on.
With this approximation, equation III

MATHEMATICAL METHODS

86

IN

CHEMICAL ENGINEERING

becomes

Ca + (a2-E)na^a-hia2n{n-l)ala_i

+ 4

n(n-l)(n-2)

aia_2

j-y-^

+
+ - + anna"i

^a^

V
and the number of terms on the
right-hand side of equation V also increases as n increases. Each term tends
to infinity, and the central term approaches infinity as (w)^", which is a very
rapid rate of growth. In view of these considerations, it is most unhkely
that the series will converge inside any circle of finite radius. It must be
emphasized here that no proof has been provided, but the evidence indicates
Equation IV shows that

a^ increases with n,

probable divergence.
(b)

= 1, P^O, X^O.
A = a+i-p
B=

Slow Reaction, k-^0, take ^^

.-.

(x

A-^A^-4B = 2+

:.

y p
a~l

+...

OL\P
But since

jS

is

small, if

a>

then ^i

>

This result

1.

is

physically

wrong

as

may be

seen by inspecting equation II. For small values of ;^_i, >^ will be
greater than y_^ and the column must therefore be generating the transferable gas rather than absorbing

it.

is useless for one particular case a>l,

and part (a) throws serious doubt upon the convergence of the series, even though the first few terms decrease quite rapidly.
The method should not therefore be used on an equation of the type I.

Part (b) shows that the method

'

1,

and k

small,

3.4.

Simple Series Solutions

In Section 2.5.1, the Hnear differential equation with constant coefficients

was solved and shown to have solutions for the complementary function of
the types
>;

and

By

=
=
=

referring to Sections 3.3.2

in ascending

power

A{\

series

.lie'"^^

+ yl2e'"^"

(2.45)

{A + Bx)e'"''
{A cos a.x

and

of x.

-\-

B sin ax)e^''

3.3.4, all

Thus

(2.49)

of these solutions can be expanded

for the

first

case,

+ m^x-^{m\x^ + ,..) + B{\ + m2X + \mlx^ + ...)

(3.22)

3.

SOLUTION BY SERIES

87

This series form of y can be accepted as a solution of the equation provided that the differential equation is satisfied by it and the series is convergent. It can be shown that the first condition is satisfied by differentiating
the series term by term and substituting the result into the differential equation when all terms will cancel out. The series can be checked for convergence by any of the tests given in Section 3.3.
Because the linear differential equation with constant coefficients always

vaHd series solution, it is natural to expect at least some differenequations with variable coefficients to possess series solutions. Since the
majority of series cannot be summed, it is to be expected that some solutions
must be left in series form.
The general second order linear differential equation can be expressed in
possesses a

tial

the

form

dv

d^y

-^ + PW^_ +

QW>; =

(3.23)

lfP(x) and Q{x) can be expanded in a convergent ascending power series in

then the method of solution is as follows.

X,

P(x)

eW
Put

y
.*.

ax

= Po + PiX + P2^2 +
= Qo + QiX-\-Q2X^ +
= ao + aiX + a2X^ +

(3.25)

...

(3.26)

...

ai

+ 2a2X + 3a2X^ +

d^y
.*.

(3.24)

..-

"2 =

2(22

ux

...

+ 6(33 X+ 12^4 X+...

and

substitute these series into equation (3.23).

(2fl2

+ 63X+12fl4X^ + ...) + (Po+PiX + P2x2 + ...)X

x(i+2a2^ + 3a3X^ + ...) + (2o + 2i^ + 22^^ + ...)x
x{ao + aiX + a2X^ + ...) =

(3.27)

equation (3.26) is to be the solution of equation (3.23), then all terms in

equation (3.27) must cancel out. Thus, coefficients of Hke powers of x can
be equated.
If

+ Poa, + Qoao =
+ 2Pofl2 + i'ifli + Qoi + Qio =
12a4 + 3Pofl3 + 2Pia2 + P2ai + eo2 + 6ii + 62o =
.-.

and
and

2a2

6fl3

(3.28)
(3.29)
(3.30)

etc.

Equation (3.28) can be rearranged to determine a2 in terms of a^ and Oq

thus,

ci2=-iPoa,-iQoao
Substituting equation (3.31) into equation (3.29)
of <3i and Qq gives

(3.31)

and solving

a3=i{Pl-Pi-Qo)ai+i{PoQo-Qi)cio

for a^ in terms

(3.32)

42

MATHEMATICAL METHODS

88

CHEMICAL ENGINEERING

IN

By substituting equations (3.31) and (3.32) into equation (3.30), a^. can be
determined in terms of ^^ and ag. This process can be continued indefinitely,
and thus each a can be determined in terms of a^ and Oq only. Treating a^
and Gq as the two arbitrary constants, equation (3.26) together with equations
(3.31), (3.32), etc., can be accepted as the complete solution of equation (3.23).
This method is only valid when P(x) and Q{x) can be expanded in series
of ascending powers of x which have a radius of convergence R. If P{x)
and Q{x) can be so expanded then the simple series method will give a series
solution which is convergent within a radius R. The next example illustrates
what happens when these convergence conditions are not satisfied.

Example

^x-j^ +

1.

6-/-

P{x) and Q{x) in equation I cannot be expanded in a

powers of x without negative powers of x occurring.

series

of ascending

Nevertheless, put

00

Y,^n^" from equation (3.26) into equation

4x(2a2

+ 6^3 X + 12a4X^ +

Equating coefficients of

..

.)

like

I.

+ 6(ai+ 2(32.\- + 3^3 x^ + ...) +

+ (<7o + fliX + a2X^ + ...) =
powers of

x,

+ aQ==0
Sa2 + l2a2 + a^=0
24a2 + lSa^ + a2=0
6a^

= aol6

.'.

a^

:.

a2=-aJ20

.'.

^3

II

= -^2/42

etc.

In this example, the general coefficient of x" gives

4nin

+ l)a^i + 6(n + l)a + + a =

(2n + 2)(2n + 3)a^, + a, =

III

.-.

'n+l

{2n

IV

+ 3)(2n + 2)

Equation IV must be true for all n^O. The result of equating coefficients of
x""^ can be obtained by replacing n in equation IV by (n\). Thus,

{2n

a_i

+ l)2n

This process can be repeated n times until Oq

Go

is

3.2

which checks with the previous result, equation

of (a7 + 1) equations together, and simplifying,

a^i_ (-l)"^'
oo

reached in the equation

(2

+ 3)!

II.

Multiplying the

last set

SOLUTION BY SERIES

3.

alternatively,

-
(2n

The

89

+ l)!

solution therefore takes the form,

^ (-1)X

V(2 + l)!

The

alternating signs

this series

and the

and odd

o(sin

between
Equation VI can be written

factorials indicate a strong similarity

series (3.10) for sin z.

VII

v^)/V^

But the method has given a solution containing only one arbitrary
Comparing equation II with the general equation (3.28), it can be
seen that the term containing ^2 is absent due to the second derivative in
equation I being multiplied by x. Thus equation II determines fl^ in terms
of Uq leaving Gq as the only arbitrary constant, whereas equation (3.28)
leaves both Qq and aj as arbitrary constants.
The second solution may be found by using the method given in
Section 2.5.2 because one part of the complementary function is now known;
constant.

equation VII.

Put

VIII

v(s'my/x)lyjx

Differentiating twice,

V x dv
Vx dx

dy

sin

dx

dx""

v sin yjx

2x

2xVx

yjxcos^/x

s'm\/xd^v

d^y

v cos v:^

sm\/x

dv

fsinV^.^

"^

^x

dx^

Substituting equations VIII, IX,

/-d^v
2x sm Jx -T
Jx^
^

Jx"^4xVx

xy/x

and

"^ ~

3i;cos\/^

4x2

X
X into equation I and simpHfying gives

/-

/-

r^dv

,^
+ {2Jx cos Jx + smJx)- =
.

dx

an equation where the dependent variable does not occur expHcitly.

Solving by the method given for this type of equation in Section 2.4.1,
gives the complete solution.

which

is

(A

sin \/x

+ B cos \/x)l\/x

XI

The second part of the solution can also be expressed in series form by
using equation (3.11), thus
y

= B(x-^-ix^ + ^x^'^-...)

XII

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

90

now

why

method used so far does not give the comeven though a series solution does exist. The
fundamental series substitution (3.26) only allows positive integer powers of
.Y, and if this restriction could be relaxed, a more general type of equation
could be solved. In the above example, equation XII shows that the successive powers of x still increase by integers even though they are all fractional, and that this series can be represented by the previous series if the latter
is multiplied by x~^. This means that instead of the first term being a constant Oq, it is GqX'^. The series substitution is therefore modified to include
this type of series.
It is

clear

plete solution of

the series

example

1,

Method

3.5.

If the general

second order

of Frobenius

differential

equation can be expressed in the

form
G(x)y

(3.33)

Fo + FiX + F2X^ +

...

(3.34)

x^--i

xF{x) -f

where

F(x)

and

G(x)

= Go + GiX + G2X^ +

...

(3.35)

with equations (3.34) and (3.35) having a radius of convergence R\ then the
equation can be solved completely by the method of Frobenius in the form
of a power series which is also convergent for |x| <i?.
By comparing equations (3.33) and (3.23) it can be shown that equation
(3.33) includes equation (3.23) as the special case i^o = ^o = ^i = ^- The
method of Frobenius will therefore always apply when the simple series
method appHes, but it will also solve equations which the simple method
leaves unsolved. Thus, the simple series method has been superseded by
the

more powerful method of Frobenius.

In order to solve equation (3.33), put
y

.-.

Y.^^x^^'

(o#0)

= ^a> + c)x"^-^
^
ax

(3.36)

(3.37)

d'y

dx

= Xn( + c)( + c-l)jc"-'^-2

(3 38)

Substituting these equations into equation (3.33) gives

Xa( + c)(n + c-l)x"-'^ + (Fo + FiX + F2x' + ...)X( + cK''' +

+ (Go + GiX + G2x2 + ...)En^"''' =

where the extra powers of x in equation
y and its derivatives.

series for

(3.33)

have been absorbed

(3.39)
in the

SOLUTION BY SERIES

3.

91

The lowest power of x which appears

efficients

of

x'^

in equation (3.39)
can therefore be equated, giving

=
c2 + (Fo-l)c + Go =

is x*^,

and co-

aQc{c-i) + FQaQC + GQaQ

.-.

(3.40)

This quadratic equation for the index of the first term

"indicial equation".
Equating the coefficients of the next power x'''^^

(c) is

+ c)c + Fofli(l + c) + FiaoC + Goi + Giao =

ai[(c + l)^ + (Fo-l)(c + l) + Go] + ao(FiC + Gi) =

called the

ai(l

.'.

which determines a^ in terms of Gq and

(3.41)

c.

The coefficient of a^ in equation (3.41) can be obtained from the lefthand side of equation (3.40) by replacing c by (c+1). In general, the coefficient of Gj in the equation obtained by equating coefficients of x'^^^ can
be obtained from equation (3.40) by replacing c by {c-\-j). If the two roots
of equation (3.40) are denoted by c^ and C2 and they differ by a positive
integer 7 so that C2 = Cy +j; when the root c = c^ is used, the coefficient of
Gj on the first occasion that it appears will be zero by the above explanation
since (cj +j) satisfies equation (3.40). But this equation should determine
aj, which is impossible when aj is not present in the equation. The method
thus encounters difficulties if the roots of equation (3.40) diff*er by an integer.
The general solution cannot be considered further, since the methods
diverge into four channels depending upon the properties of the roots of the
indicial equation (3.40). The treatment of each case will now be pursued

separately.
3.5.1.

Case

I.

Roots of Indicial Equation Different, but not by an Integer

In this case, equation (3.41) determines a^ in terms of Cq ^^^ ^- ^Y

equating coefficients of increasing powers of x as they appear in equation
(3.39), all coefficients a can be determined in term.s of aQ and c. The two
different values of c which are solutions of equation (3.40) each give rise to
an independent series having its first coefficient as an arbitrary constant.
The sum of these two series is the complete solution of equation (3.33) for
this case.

the

Example 1. The example from the

method of Frobenius.

last section will

be solved again, using

'^^^2 + ^^. + y = ^
dx

dx^

The

series substitution (3.36)

ax

aox' + a^x'-^^

...

and

its

differentiated

X^n^""^^

forms are

MATHEMATICAL METHODS IN CHEMICAL ENGINfEERlWC

92

ax

and these may be substituted

4laoc(c-l)x'-^

+ ai(c +

into equation

l)cx'

...']

to give

+ 6[aoCx'-'+a^(c +

l)x'

+ laox' + aix'''^ +
Equating coefficients of

like

powers of

x, starting

4aoc(c-l) + 6aoC
+ l)c + 6ai(c + l) + ao
4fl2(c + 2)(c+l) + 6a2(c + 2) + ai =0

.-.

=0

=
=

:.

II

x'^~^,
III

IV

for the coefficient of x''^^

4fl,+ i(c

The

...']

...']

with the smallest

4a^(c

.-.

and

+ r+l)(c + r) + 6a,+ i(c + r+l) + a, =

c

VI

two roots

indicial equation III has the

or

0,

c= -i

VII

When c = 0, the series reduces to the previous simple series solution

with constant first term and the subsequent algebra is identical. The first
solution corresponding to c =
is thus
VIII
yi = flo(sin\/^)/\/^
= J, equations IV and VI become ^i + 3ai +^0 =0
=
a,+ i(2r + l)(2r-l) + 3fl,+ i(2r+l) +

However, when

and

fl,

Rearranging,

(2r+l)(2r + 2)

a,

-1
similarly,

(2r-l)2r

a,_i
etc.

a,

-1

ao

1-2

until

Multiplying these equations together gives

n
'

IX

(2r + 2)!

Substituting into the general series gives the solution

yi

The general

= o^

J(-l)"ax"-*

solution of equation

I is

(A sin V^

(2n + 2)!

thus

+ B cos \/x)l\/x

XI

3.

SOLUTION BY SERIES

93

should
The coefficient Gq which appears in both equations VIII and
be different in the two solutions which are independent. It is good practice
to change both of these coefficients to new arbitrary constants as in equation

XL
Equation VI, which was obtained by equating coefficients of a general
X, is called the "recurrence relation" because it enables any coefficient a^ to be determined in terms of the earher coefficients in the series.
When only two coefficients are involved in the recurrence relation it is a
simple matter to find the general term such as equation IX, whereas when
more than two are involved it is virtually impossible to develop an analytical

power of

expression for the general coefficient.

3.5.2.

Case

II.

Roots of Indicia I Equation Equal

The first part of the complementary function can be found by using the
method described above, but only one series containing one arbitrary constant can be determined because the indicial equation has equal roots. The
second solution can be found by assuming that the two roots are different
and taking the limit as the second root approaches the first through variable
values of c. The complete solution for different roots can be written
y

Au(x,

Ci)

+ Bu(x, C2)

= (A- B)u(x, Ci) + B[w(x, C2) - w(x, Ci)]

(A-B)u(x,Ci) + B(c2-Ci)
Redefining the arbitrary constants

A and

+P

Ci)"

B, in terms of a and p,

W(X, C2)

au(x,Ci)

w(x, C2) w(x,

W(X, Cj)
(3.42)

C9-C1

taken of equation (3.42), the term in square brackets

becomes the partial derivative of u with respect to c and evaluated at c = c^.
The complete solution for the case of equal roots of the indicial equation is
If the limit as C2^Cj^ is

thus

au(x,c,) +

p{^)

An

alternative derivation of equation (3.43) can be seen

considering the following example.

Example

(3.43)

more

clearly

by

2.

d'-y

r.

Jy

^7r^ + (1-^)3^-^ =
Put

Ycix"^'

(3.36)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERIN(j

94

and

and

its

/.

fa(n + c)(n + c-l)x"^^-'+fa(n + c)x"^^-'-

differentiated forms (3.37)

Equate

(3.38) into equation I

coefficients of x''"^ to obtain the indicial equation.

aoC{c-l) + aoC

II

This has a double root, c

Equate coefficients of ;v^,
a i(l

+ c)c + a^ii c) GqC - ao =

a,(l + cy = ao(i + c)
a, = aol(i + c)
-{-

.-.

III

.-.

Equate

coefficients of x'^"^^
fl,+ i(r

Since

r is

value of

+ c + l)(r + c) + a,+ i(r + c+l)-a,(r + c)-fl, =

a positive integer, and c

zero, (r +

is

c+

1)

cannot be zero for any

r,

a, +

.-.

a,/(r

+ c+l)

IV

is the recurrence relation, and it is useful

on the algebraic derivation of equation IV.

Equation IV
as a check

By using the recurrence relation

obtained for the general term.
'''^'

r times,

+ c+l)(r + c)(...)(c+l)

(n

+ c)(n + c-l)(...)(c + l)

u{x,c)

is

now

''''

The function

the following expression can be

(r

^^

to derive equation III

estabhshed by putting equation

into the general

series (3.36).
"^'''

'^

When

0, w(x,c) satisfies

y,

To determine

+1

(c

equation

w(x,0)

o [1

aoe'

and

+ l)(c + 2)
is

the

x^

+^ +

first

x^

I.

solution,
1

^ + ^ + -J

the second solution, substitute equation

side of equation

^^

? ( + c)(n + c-l)(...)(c+l)

VII

VI into the left-hand

x^ +

:.

{l-x)-

95

3.

SOLUTION BY SERIES

aoc(c-l)x^ '+^0^'^

+^0^

dx

dx'^

+ aoCx'-' + aox' + -^x'-'' +

..,

(c+l)(c + 2)

c+1

= aQCx'~^{c-\ + \) + aQx'{c + \-c-\) +

+ -^x^-'^(c+l + l-c-l-l) +
c+ 1

...

because all terms after the first are identically zero. This must necessarily
be so because the terms in brackets are the recurrence relationships, and the
The function
first term is the left-hand side of the indicial equation II.
w(x,c) defined by equation VI thus satisfies the modified differential equation

x^
+ (1-x)^ -u = a^c^x'-^
dx^
dx

VIII

Differentiating equation VIII partially with respect to c gives

d^ /du\

,^

d (du\

du

and taking the Hmit of the right-hand

f^ (du\
/du\
d'

^^

_,,

side of the equation as c-^0,

d (du\
/du\

du

y2 = dujdc also satisfies equation I when c = 0.

has thus been shown in two alternative ways that the second solution
of equation I is

I.e.

It

The following two points of technique are useful in order to perform the
required differentiation efficiently.
^n + c^g( + c)lnx

(i)

which may be
(ii)

then

If

verified

(3 44)

by taking the logarithm of both

y =f,(c) f,(c) /3(c)

Inj;

ln/i(c)

sides.

...fXc)

+ ln/2(c) +

...

+ lnX(c)

(3.45)

MATHEMATICAL METHODS

96

IN

CHEMICAL ENGINEERING

and multiplying throughout by

Differentiating,

dy

fi(c)^n(c)_^

dc

flic)

y,

(3.46)

hie)

/,(c).

where (') denotes differentiation with respect to c.

Continue the example by applying these principles to equation IX,
taking

/,(c)

l/(c

+ l),.../(c)=

c+V

/i(c)

+ n),/+,(c) = e<"^^>'"'
-1
fnic)
/nVl(c)
= Inx
c
n
+
Uc)
fn+lic)
l/(c

"oZ
^(c + l)(c + 2)(...)(c + n)\c + l

yi

o"-\

o!

+2

+n

Inx
c

^oe^lnx
n

71

can be seen that the second solution consists of

the first solution multiplied by In x, added to a new series. This relationship
always exists between the two solutions when the second solution is found
by using the above method.
The complete solution of equation I is thus
Referring to equation VII,

Au(x,0) + B(duldc), = o
Ae^'

3.5.3.

Case

Ilia.

This case

is

it

+B

XI

e^lnx

on\\

nj.

Roots of Indicial Equation Differing by an Integer

very similar to Case

II in that the

partial differentiation with respect to

c,

second solution involves a

thus generating a term Inx.

The

function u{x,c) has to be defined as before such that the recurrence relation
is satisfied but not the indicial equation. The solution can be shown to take
the form

Au(x,C2) + B--l(c-c,)u{x,
cc

The reason

for the

)], = ,

above form of the second solution

(3.47)

best visualized

by

and following the same method

as

is

reference to an example.

Example

dy
d^v
x(l-x)--^ + (2-5x)/-4>' =
dx

With the usual

series substitution (3.36)

before, the indicial equation

dx

is

aoc(c-\) + 2aoC

II

the

first

recurrence relation

is

a,(c+l)

and the general recurrence

97

SOLUTION BY SERIES

3.

relation

ao{c + 2)

III

is

+c+2
'^^^~r + c + l
r

The

roots of the indicial equation are

Ci

As mentioned
tion differ

which

= 1

When

from

relation IV, the general

is

still

first

1,

recurrence relation in

and equation

III is the first

equa-

III gives

CO

is thus of no value, but the series based on

vahd.
By successive use of the recurrence

term becomes

solution with c
fl,+

y^

.-.

by the binomial

the roots of the indicial equa-

and for the

when

0.

equation

1,

a^

.'.

series arising

difficulties arise in the first

In this example, 7

which a^ occurs.

the larger root c

or C2

earlier in Section 3.5,

by an integer 7,

aj appears.

tion in

The

IV

=
=

+c+2
c+ l

0, this simplifies to

+ 2)ao
ao{l + 2x + 3x^ + ...)
^ao(l~xr'
(r

VI

series (3.6).

Define the function u{x,c) again, which satisfies the recurrence relation
IV and its solution V, but not necessarily the indicial equation II.
u(x,c)

= f:^^^^^aox"^^

VII

c-\- 1
is substituted into the left-hand side of the differential equaterms cancel out as in Section 3.5.2 with the exception of the term
identical with the indicial equation. Thus,

If this function

tion

I, all

x(l-x)^ + (2-5x)^-4u =
ax

ax

aoc{c + l)x'-^

VIII

Whereas in example 2 the corresponding equation 2.VIII contained a squared

term on the right-hand side, so that after differentiation a factor still remained
in

common

similar behaviour can be obtained

with the indicial equation; here, this

is

not true of equation VIII.

by artificially introducing a squared

term into equation VIII by multiplying throughout by (c+l). An alternative argument which determines the multiplying factor is available by inspection of equation VII.
The factor (c+l) makes u(x,c) infinite, therefore

98

MATHEMATICAL METHODS

transfer the

awkward

throughout by

By a

CHEMICAL ENGINEERING

factor to the other side of the equation by multiplying

it.

argument

similar

to that used in Section 3.5.2,

the second solution takes the

>'2

But,

IN

it

can be shown that

form

= [(c+l)"(x,c)L=_i

IX

from equation VII

(c+l)u(x,c)

|-(c+l)u(x,c)

.-.

CC

y2

.-.

f^(n

Qq

x{\-x)

{\-xY
I is

l]

+ 3x^ + ...) + ao(x~'^ + l + x + ...)

again, the term multiplying In

solution of equation

+ c+l)aox"^'^

= faoX"+^[(n + c+l)ln.x +

ao(\nx)(l-\-2x

^ aplnx
Once

is

the

first

solution.

The complete

thus

B\nx

^^^

y-{i-xf^{i-xf^^^(x^)
3.5.4.

Case

There
in the

is

Illb.

just

Roots of Indicial Equation Differing by an Integer

one further variation in the type of solution which can

method of Frobenius. In the

and when

= 1,

became

last

infinite.

example, equation

Had

III

arise

gave

the right-hand side also con-

would not have arisen,

would have been automatically satisfied, and both parts of the
solution would have been normal power series as in Case I. An example of

tained a factor (c+1) this particular difficulty

equation

III

this situation follows.

Example

4.

^^>'

With the usual

+ (,_l)^_,=0

series substitution (3.36)

before, the indicial equation

and following the same method as

is
flo

and the general recurrence

c(c

1) flo c =

relation

a,+ i(r + c + l)(r +

11

is

c-l) + aXr + c-l) =

Since the roots of equation II differ by 2, consider the

in

first

III

of equations III

which ^2 occurs. Viz.

a2(c + 2)c + a^c

IV

SOLUTION BY SERIES

3.

99

c = 0, the coefficient of ^2 in equation IV is zero as in Section 3.5.3,

but both terms in equation IV are zero when c = 0, and ^2 remains indeterminate instead of infinite.
Therefore, if c = 2, equation III can be simphfied by cancellation but
if c = 0, the term (r + c-l) can only be cancelled if r> 1.

When

a,+ i{r + c +

.*.

The

first

solution

is

given by c

l)= -a,

2,

fl,+ i(r

and by successive

(unless c

0, r

1)

thus

+ 3)= -a,

substitution,

_(-ir^

"'*'-

(r+3)!

/x^

.3

x^

x'

The second

solution, v^hen

yi

= 2ao(e"*-l + x)
c = 0, becomes

= ao-aoX-ha2X^2

^41--)

'^ (21 -3"!

^2^^

- +

ao(i-x) + 2a2(e-''-l

VI

^2^"^

3.4

+ x)

VII

All coefficients after the first two can be found in terms of ^2 and there are
thus two arbitrary constants in the second solution making three in all.
Inspection shows that the series associated with ^2 in equation VII is
identical with the first solution as given

One

by equation VI.

thus been obtained twice, and in Case Illb

when a

coefficient

determinate, the series obtained by using the smaller root

complete solution.

The complete

solution of equation

is

solution has

becomes

in-

in fact the

can therefore be

written

Iqi, and B =

= A{.-x) + Be-''

where

A =

3.5.5.

Summary of the Method of Frobenius

Qq

VIII

2a2 in equation VII.

The examples given above have been specially selected so that the series
solutions could all be summed. The solutions can be checked quite easily
by differentiation and substitution, thus demonstrating that the series
method gives correct results. It must be emphasized that the methods of
Chapter 2 should always be tried before resorting to the method of
Frobenius.

Too much time should not be devoted

to trying to

sum

the

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

100

series obtained because on many occasions the solution of problems of

engineering importance will not be expressible in closed analytical form.
The method has only been described for expressing the solution as a
series of powers of x. Solutions in powers of {x Xq) can also be obtained
by moving the origin along the x axis to Xq and then proceeding as before.
The method will be appHcable if the convergence conditions are satisfied at

the

new origin.
The differential equation

^ + xt{x)+
^^W:r:
dx

z^
dx^

^My = ^

(3.33)

can be solved by putting

(ao^O)

I.ax"^^

(3.36)

F(x) and G(x) can be expanded in a convergent series of non-negative

powers of x for all \x\<R. The solution obtained will also be convergent
for all \x\<R.
The indicial equation is found by equating the coefficients of the lowest

if

power of X occurring in equation (3.33) after substituting equation (3.36).

The recurrence relation is found by equating coefficients of x'"'^'' in the same
equation.

two roots of the

If the

indicial equation are c

Ci

+j andy>0, then

Case I

(7

not an integer or zero).

with C2

c^

and

C2

the following four cases arise.

Two

separate series of the form (3.36)

c = C2. These two series

added together with arbitrary constants give the complete solution.
Case II (j is zero). Define the function u(x,c) which satisfies the recurrence

are obtained, one for c

Cj

and the other for

relation but not necessarily the indicial equation.

The

solution then takes

the form

Case Ilia
appears.

(j is

When

the equation

is

an

Au(x,c,) + B(duldcX=,^

integer).
c^,

Find the

first

(3.43)

recurrence relation in which Gj

and if the rest of

the coefficient of aj will be zero,

not zero the solution takes the form

where

Au(x,C2)

5^[(c-ci)(x,c)],=,,

(3.47)

u(x,c) satisfies the recurrence relation but not necessarily the indicial

equation.

Case Illb (7 is an integer). When c = c^, the first recurrence relation in

which aj appears is identically zero leaving aj indeterminate. The complete
solution can be found in the form of equation (3.36) by putting c = Cj when
o and aj will be the two arbitrary constants.
3.5.6.

Illustrative

Example

5.

Cross-section.

Examples

Temperature Distribution in a Transverse Fin of Triangular

The shape of the cooling fin is illustrated in Fig. 3.1 where

SOLUTION BY SERIES

3.

the radius of the pipe (a)

is

in,

101

the radius of the rim of the fin (b)

is

6 in,

and the coordinate x (measured in feet) is measured inwards from the rim
of the fin. There are two natural origins for the coordinate, but since the
temperature distribution in the vicinity of the pipe axis is of no interest, the
origin is taken on the rim instead. Assuming that the fin is thin, temperature

Transverse cooling

Fig. 3.1.

fin

normal to the central plane of the fin will be neglected. The thermal conductivity of the fin (k) is 220 Btu/h ft^ F ft~\ and the surface heat
transfer coefficient {h) is 2 Btu/h ft^ F. Denoting temperature by rF with
T^ F representing the air temperature, the heat balance can be taken as follows.
The area available for heat conduction is
variations

2n{h

Anx{h x) tan a

x) {2x tan a) =

= H7ix(Z? x)tana-

Input

ax

riT'

/7

ax

-
ax

= - k4nx{b - x) tan a +

Output

k4nx{b x) tan a

dT
Sx

dx

+ 2h27i(b - x) (3x sec a) (T - TJ

Accumulation

Since Input - Output

d_

k4nx{b x)i2Lncc

dx

Accumulation,

dT

Sx-2h2n(b-x)(dxsQcoi){T-T^) =

dx

x(b-x)

ksinoc

dx
Putting

K=

h/k sin

a,

dx

-Kb-x){T-T^) =

II

= T-T^,

and y

dy

d'y

dT

x(b-x)j^^ + (b-2x)-^-K{b-x)y

The appropriate boundary conditions

at

at

III

are,

= 0, y
= b-a,

remains

T=

finite

IV

Tb

V
where T^

is

the temperature of the pipe contents.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

102

Using the Frobenius substitutions

(3.36), (3.37),

and

(3.38) in equation

III gives

bY^a{n + c)(n + c-l)x"^''-' -Y,ci(n

bY,a{n + c)x"^'-'

+ c)(n + c-l)x"^' +
+ c)x"'-' - KbY,a,x"^' +

2Y,a{n

-{-KY.a^x"'-'^^

The

VI

indicial equation obtained

by equating

coefficients of ;c^~^

baoc(c-l) + baoc
c'

.-.

Equation

III

thus gives a Case

II

boundary condition IV,

violating

=
=

VII

With the knowx which becomes infinite when

solution with equal roots.

ledge that the second solution will contain In

is

may

it

be concluded that the second

Therefore it is only
and the remaining work can

solution has no physical significance in this example.

necessary to find the first solution with c

be simplified by omitting c.

Equating

coefficients of the next

ba^

power of x which

is x*^,

KbaQ =
fli = Kao
.*.

The recurrence

relation

bar+i(r + l)r-a,r(r-l)-\-ba,+i(r +

ba,+ ,(r+iy

.-.

VIII

is

l)-2a,r-Kba, + Ka,_i^ =

a,lr(r

+ l) + Kb']-Ka,.,

IX

This recurrence relation is different from all previous ones since it relates
three of the original coefficients a^+i, a^, and a^-^.
It is thus difficult to
determine a general formula for a^ in terms of ^Q) although a solution may be
possible by the methods of Chapter 9. Unless such a solution is possible,
the second part of the complementary function in this example of equal
roots cannot be found by the method of Section 3.5.3 even if it is required.
The appropriate first solution can be found however on a term by term basis.
Introducing the numerical values given in the problem, together with T^ =
60, Tg = 212, and a = 5 gives

A:= 2/220x0-0872 =

0-1043.

Equations VIII and IX become

= 0-1043ao
a,+ i(r + l)2 = fl,[2r(r + l) + 0-1043]-0-2086^,_i
^2(2)^ = 0-1043ao(4- 1043) -0-2086^0
= 0-2195^0
/.
^2 = 0-0549^0
fli

.'.

Using equation

again,

^3(3)2

.-.

etc.

^3

= 0-0549flo(12-1043)-0-2086x 0-1043^0
= 0-6424^0
= 0-0714flo

3.

SOLUTION BY SERIES

103

It can be seen that the most important terms in equation

two, and the later behaviour of the series will approximate to

a.+

.'.

are the

first

2r

XI

fl.

By Section 3.3, this approximate series has a radius of convergence of J in

agreement with the general statement of the convergence conditions related
to equation (3.33). Because the fin is defined to occupy the range 0^:x:^^,
the series solution is convergent throughout the fin.
Evaluating further coefficients from equation X gives

+ 0-1043x + 00549x^ + 0-0714x^ + 0-1068x^ +

+ 0-171x^ + 0-285x^ + 0-488x^ + ...) XII
becomes;
V
Boundary condition
x = i,
at
y = 112-60 = 152
152 = ao(l + 0-0348 + 0-0061 + 0-0026 + 0-0013 + 0-0007 + 0-0004)
:.
XIII
flo = 152/1-0459 = 145-3
y

flo(l

.-.

Equation XII gives the temperature distribution throughout the fin, and
equation XIII shows that the rim temperature is 205 3F.
The rate of removal of heat by the fin can be calculated from the temperature gradient at the base.

Thus

Q = k4n(b - a)a ian a (dT I dx)^= _ a

= 1344(jr/Jx),=,/3
= 1344(0-1043 + 0-0366 + 0-0238 + 0-0158 + 0-0106 + 0-0070 + 0-0047 + ...)
= 285 Btu/h
I,

by differentiating equation XII, and using equation XI to estimate the continuation of the series to infinity.

Example 6. Tubular Gas Preheater. A supply of hot air is to be obtained

by drawing cool air through a heated cyhndrical pipe. The pipe is 4 in
diameter and 4 ft long, and is maintained at a temperature T^ = 600F
throughout its length. The properties of the air are
Specific heat

(Q

Thermal conductivity

{k)

Density (p)

Flow
(the

rate (w)

above are

all

mean

Inlet temperature

Overall heat transfer coefficient

distance measured in feet

Assuming

=
=
=

0-24 Btu/lb F
0-020 Btu/h ft^ F ft~^
0-050 Ib/ft^

1000 ft^h

values)

=
(/?)

from the pipe

70F.

= 5x~*

Btu/h

ft^

F where x

is

the

inlet.

that heat transfer takes place

by conduction within the gas

in

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

104

an axial direction, mass flow of the gas in an axial direction, and by the
above variable heat transfer coefficient from the walls of the tube, find the
temperature of the exit gas.
Solution

With the symbols given above, and

illustrated in Fig. 3.2, the following

heat balance can be estabUshed.

*

70

F^

>

4 ft

U
tn

T-^
'

t.;t

^"

Jl

Gas preheater

Fig. 3.2.

Output

Input

-kA

By conduction

is

dT

dx

dx

d /

is

dT\
dx

upC^[

T+

Sx

the cross-sectional area of the pipe, and

steady state the accumulation

dx

ndh{T^ -T)Sx

Wall heat transfer

where

dT

upC^T

By mass flow

4in dia

its

diameter.

Since in

zero,

T'T-i

-kA- + upCT
+ 7idh(T^-T)3x
^
dx

kA
dx
dx

dT\

dT

d^T
dT
ndh(T^-T)= -kA-^ + upC -
^

dx

dx"-

Rearranging,

upC.dT

d^T

ndh^^

^^

II

dx'

Putting

= T^-T,
d^t

dx^

upC^

dt

kA dx

4/2

kd

III

Inserting the numerical values,

4 -688o4^-3000x-^r =
dx"^

dx

IV

3.

Put X

SOLUTION BY SERIES

z^ to rationalize the coefficient of

an attempt

is

made
first

dz

dz"^

the results for the

t,

z,-il + 13160z^)^-12000zh =

:.

If

105

to solve this equation

few terms

by the method of Frobenius,

are,

or 2 from the indicial equation

Taking

a^=0
a^

2
^3

3440^0

=
=

800^0
7-9

x lO^ao

coefficients are increasing at an alarming rate, yet a test of equation V

shows that the series must be convergent. This is a case mentioned earher
where a convergent series is of no practical value since more than 100 terms
would have to be calculated to determine even the first solution. The trouble
arises because two of the coefficients in equation IV are much larger than the
coefficient of d^t/dx^. The second derivative arises from the gas conduction
term and if this is neglected, equation IV becomes

The

+ 0'436x-^t =
^
dx

VI

which can be solved by separating the variables. Viz.

= aexp(-0-872x^)
= 600 70 = 530.
at x = 0, is
= 530exp(-0-872x^)
T = 600-530exp(-0-872x")

VII

The boundary condition

.-.

VIII

.-.

The exit gas temperature is therefore 507F.

The approximation of neglecting the second

differential term in equation IV

can be checked by comparing the neglected term with one of the other terms.
Differentiating VIII twice gives

-^= -23Mx-^exp(-0-872x^)
dx

-4 = 23M(ix-3/2 + 0-436x-^)exp(-0-872x^)
dx

d^t Lc.r.r.dt

'6880

dx^l

which

ix-3/2-0-436x-'
^

dx

6880x""
1

13760X

15780;c"

small except when x is small.

10"^, the error ratio = 7^%.

is

If X =
The error has thus been made over a
justifies the

above

result.

negligible part of the pipe

and

this

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING,

106

3.6.

The

differential

equation

x^4^ +
known

Equation

Bessel's

^+

(x^

- k^)y =

(3.48)

where A: is a positive or zero

This equation arises so frequently in practical problems that it
has been studied in great detail and this section will be devoted to giving an
outUne of the derivation of the various solutions by using the method of
Frobenius. Because the equation occurs so frequently, the series solutions
have been standardized and tabulatedf thus eliminating the necessity of
working through the series solution for each individual problem. Many
other second order differential equations can be reduced to Bessel's equation (3.48) by a suitable substitution, and they too can then be solved by
using the standard tables.
Applying the method of Frobenius by putting
is

as Bessel's equation of order k,

constant.

and

its

The

indicial equation

differential

= ^ax"^^

(3.36)

forms (3.37) and (3.38) into equation

(3.48).

is

aQc{c

l)

+ aQC k^aQ =

which has the two roots

The

difference

between the roots

/c

is

or

-k

(3.49)

thus 2k, and the type of solution will

depend upon the nature of k.

Equating coefficients of x^'^^,
fli(c
.'.

Because c
/:

!,

+ l)cH-ai(c + l)-k^fli =0
a^(c + l + k)(c+l-k)=0

= k, and k^O,

c= ^, c+1 =

then c+l+/:>0, but for the special case

Therefore,

relation

is

obtained by equating coefficients of x''^^

unless

The recurrence

0.

/:

andc=-i,

ai=0

aXr + c)(r + c-l) + aXr + c)-k^a, + a,_2 =0

aXr + cy-a,k'=-a,_2
.-.

(3.50)

^r 2

^'~
(r

(3.51)

(3-52)
/"3

c-jA

+ c + k){r + c-k)

t For example, Jahnke and Emde, 'Tables of Functions with Formulae and Curves,"
Dover Publications, New York. Or Watson, "Theory of Bessel Functions," Cambridge

University Press (1922).

SOLUTION BY SERIES

3.

The

first

relation

solution for c

k, will

be normal in

Combining equations

(3.51)

and

ai
r

all

and the recurrence

cases

becomes
r(r

and

107

can be replaced by 2

(3.53) gives

^3

by

(w- 1),

^5

...

=0

(3.55)

in the recurrence relation (3.54)

^""
Replacing

+ 2/c)

(w 2),

2m(2m + 2/c)
etc.,

and making successive substitutions as

before,

^2m

(-ir^o

22'"m!(m + fc)(m + /c-l)(...)(/c+l)

m!r(m + + l)
/c

where r(A:+l) is the Gamma function or generalized factorial whose properties will be discussed in Chapter 5.

The

first

solution of Bessel's equation

is

thus

rom!r(m + /c+l)
The

Bessel function of the

Defining a

new

first

kind of order k

arbitrary constant, the

first

is

defined as

solution of equation (3.48)

is

thus

y,=AUx)
The form of

the second solution, for c

(3.57)
A:,

depends on whether or not

2k is an integer, and the method of solution diverges into four channels

corresponding with the cases of the method of Frobenius.
3.6.1.

Case I (2k

is

not an integer or zero)

In this case, the second solution is obtained from the

in equation (3.56) by k, giving the complete solution

y
3.6.2.

Case II (k

first

by replacing k

AJ,(x) + BJ.,(x)

(3.58)

0)

Returning to the recurrence relation (3.53) and putting

r

= -r-2
(r + c)'j

A:

0,

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

108

= from equation (3.51), r can be replaced by

expression for the coefficient of the general term is
Since a^

^""

2m

again and the

(-ir^o

(2m + c)2(2m + c-2)2(...)(c + 2)'

Defining u{x,c) as given in Section 3.5.2,

"^""'"^

^^"^^^

Jo (2m + c)^(2m + c-2)^(...)(c + 2)^

and putting

Differentiating equation (3.59) with respect to c

gives the

second solution,
l^\2m

"-'^

(-irao(ix)

,n-,-l-l-...-i

(m !)

=-H'"'-i/"sr(-^-^)]
The expression in brackets is the Neumann form of the second solution, but
an alternative form named after Weber is more frequently used. The Weber
form is obtained by adding (y In 2) Jo{x) to the Neumann form and multiplying the result by (2/n). Thus
y2

BYo(x)

where

l^oW

2ri...i..x
.
[ln(ix) + 7] Jo(x)
.

....

..-,

7r,=

71

is

the

and

Weber form of
is

which

+- +

lim (1

m-coV

is,

Case Ilia {k

is

is

-+...

= k,

+:;+

+.2.

Inm

=0-5772...

is

thus
(3.60)

an integer)
first

recurrence relation in which ^2* appears

-2fc-2

ci2]ik-^c){2>k-^c)=

-ajk-i

the left-hand side

is

zero but the right-hand side

is

not zero

an even coefficient. The method for Case Ilia given

must therefore be followed.

because a2k-2
Section 3.5.4

AJo(x) + BYo(x)

a2ji2k-^cf-a2kk^

When

( 1

defined by

Ik is an even integer and the

from equation (3.52),

or

/
(mi;

solution of equation (3.48) with k

3.6.3.

(-l)'"(ix)''"/,

the Bessel function of the second kind of order zero,

Euler's constant

The complete

2
-- ^
E

is

in


SOLUTION BY SERIES

3.

The eventual

109

solution using

IS

y,

= -Bl\nix + y]Ux)--

BY,{x)

^TT^^tt^)"""'

y [2
1+am + k

^^

nj^o

m\(m + k)\

...

ml

+ +-+
1

...

m + kj

where y is again Euler's constant, and the Weber form has been found by
adding agiy In 2) J^ix) to the Neumann form and multiplying by (2/7r).
Ykix) is the Weber form of the Bessel function of the second kind of order k.
The complete solution of equation (3.48) is thus

= AUx) + BY,{x)

when k
3.6.4.

is

an

Case Illb {2k

The

first

(3.61)

integer.
is

an odd integer)

recurrence relation in which a 2k appears

M/c + c)(3/c + c)=-a2/c-2

When

is,

from equation

(3.52),

(3.62)

= k,

the left-hand side is zero, but the right-hand side is an odd

and in all cases considered so far, the odd coefficients have been
zero.
This must now be investigated further. If A: = |, equation (3.50)
shows that a^, the critical coefficient, is indeterminate and the solution
follows Case Illb giving two normal power series. If A: = 3/2, the critical
recurrence relation (3.62) becomes
c

coefficient

3(T + c)(f+c)=

and

-a,

Continuing this argument, ajk will

0, aQ is indeterminate.
always be indeterminate if 2k is odd, and the solution again consists of two
since a^

normal

series.

The general

is

(3.63)

AJk(x)

+ BJ.k(x)

(3.58,63)

not an integer or zero, or

AJk(x) + BJ_ki^)

solution of Bessel's equation (3.48) can therefore be written

y
if

AJk(x) + BYk(x)

(3.60,61)

an integer or zero.
Examples of the use of the above Bessel functions occur quite frequently in chemical engineering when the problem involves partial differen-

if

is

tial equations, but their use in ordinary differential equations is rare in

chemical engineering but more frequent in other branches of engineering.
Rather than present an irrelevant example at this point, the reader is referred

to

Example

3 in Section 8.7.5

form one stage

in the solution

and Section

8.7.3,

where the above solutions

of certain partial differential equations.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

110

Modified BesseVs Equation

3.6.5.

now been solved for all real values of x, and

sum of two linearly independent functions of x

BesseFs equation (3.48) has

the solution given as the

each multiplied by an arbitrary constant. Two kinds of Bessel function

have been defined both of which are needed to express the solution if k is an
integer or zero, but two versions of only the first kind if k is not an integer
or zero.

further type of differential equation can be solved with the aid of

Bessel functions

it is

+ ^^.-i^''^ ^^)y =
^^-J^
dx'^
dx

(3.64)

Equation (3.64) can be obtained from Bessel's equation (3.48) by replacing

/ = ^(1).
The solution of equation (3.64) can thus be

X with ix where
written

y
if

is

AJk{ix) + BJ_,,{ix)

(3.65)

not an integer or zero, or

AJ(ix) + BY^(ix)

(3.66)

is an integer or zero.
In Section 2.5.2 it was found that the solution of the second order linear
differential equation with constant coefficients could be expressed more
conveniently in terms of real trigonometric functions than complex exponential functions when the auxiliary equation had complex roots. Here, instead
of using the Bessel functions of a complex variable, it is similarly convenient
to define modified Bessel functions of a real variable so that the above
solutions can be expressed in real form. The necessary definitions are

if

/,(x)

e-*""A(ix)

(3.67)

\2-^J
_ y
~ %m\r(m + k+l)

K,{x)

e*'*^ ""'[A(ix) + iT,(.x)]

(3.68)
*-,^

(-l)'^i[ln(ix) + v]/,(x) + i

2=

(-l)'"(/c-m-l)!
^

'

m
,

^J^o

The

(-ir(ixy'"^^[^

y
if

is

is

is

'-(ixy

1,1

ml

m-\-k\

thus

= Ah(x) + BI.,(x)

(3.69)

not an integer or zero, or

y
if

[2

m\im-\-k)\

solution of equation (3.64)

\,
fn\

AI,(x) + BK,(x)

(3.70)

an integer or zero.

Ik(x) is called the

and KfXx)

is

modified Bessel function of the

first

kind of order

the modified Bessel function of the second kind of order k.

k,

SOLUTION BY SERIES

3.

111

Example, Heat Loss Through Pipe Flanges. Two thin wall metal pipes of 1 in
and joined by flanges | in thick and 4 in diameter, are carrying steam at 250F. If the conductivity of the flange metal k = 11^ Btu/h ft^
F ft~^ and the exposed surfaces of the flanges lose heat to the surroundings
at Ti = 60F according to a heat transfer coefficient h = 1 Btu/h ft^ F,
find the rate of heat loss from the pipe, and the proportion which leaves the
rim of the flange.
external diameter

Solution.
circular face
is

It is

only necessary to consider one flange with one exposed

and an exposed rim.

more convenient

Because

all

dimensions are in inches,

it

to use inches instead of feet for the length dimension,

therefore take a radial co-ordinate r measured in inches

from the

axis of the

pipe and consider the heat balance over an element of width dr, as

shown

in Fig. 3.3.

Pipe flange

Fig. 3.3.

^
= -2n\r
^
^

Input
^

The
of

dT

nkr

dT

dr

12

dr

^ , ,
Output

nkrdT
dTl
=-__
+
_^___J,, + __(T-r,)

Accumulation

factors 12,

nkr

Inrhdr ^

and 144 are necessary because k and h are defined

in

terms

feet.

Simplification of the heat balance gives

dT

d'-T

Putting y
equation.

= T-T^, and x

-=

dx^

Comparison with equation

r^(h/6k) gives the standard form of the

(3.64)

X -^

dx

V=

shows that equation

II

II is

a modified Bessel

MATHEMATICAL METHODS

112

According to equation

equation of zero order.

The boundary conditions

at

(3.70) the solution

is

thus

AIo(x) + BKo{x)

III

i,T = 250

IV

are,

at

and

CHEMICAL ENGINEERING

IN

"

2,

k dT
h
=
J^"^ J^^CT-TJ
^

where equation V states that the heat conducted through the metal to the rim
must equal the heat lost from the rim surface to the surrounding air.
Changing the variables, and introducing the numerical values, equations
IV and V become
X = 00195,
at
VI
y = 190

and

at

dy

0-078,

-f-

-0-0195y

VII

ax

and using the properties

end of this chapter,

Differentiating equation III,

(3.99),

and

(3.100), Hsted at the

-^
ax
Substituting equations VIII

= AI,{x)-BK,(x)

and

A/i(0-078)-BXi(0-078)

and substituting equation

(3.90),

III into

III into

(3.91),

VIII

equation VII gives

-0-0195[/4/o(0-078) + BKo(0-078)]

IX

equation VI gives

yl/o(0-0195) + BKo(0-0195) =

190

Finding the values of the various Bessel functions from tables! and solving
gives
equations IX and

A=

B=

186-5,

0-8636

Putting these values into equation III and reverting to the original variables
gives the temperature distribution

T=

60+186-5/o(0-039r) + 0-8636Ko(0-039r)

The heat conducted from

the pipe by the flange

is

XI

given by

nk (dT\

=
=
and the heat

lost

M27[BXi(0-0195)-^/i(00195)_,
47-81 Btu/h

through the rim

:2

6
t "British Association
(1937).

is

given by

\drj, = 2

Mathematical Tables," Vol. VI.

Cambridge University Press

3.

=
=

4-507

SOLUTION BY SERIES

113

[5Ki (0-078) -^/i(0-078)]

16-62 Btu/h

35% of this
through the rim.
The above problem is similar to the problem of a flat circular cooling
fin, which has been solved by Gardner! who presents solutions in the form
of Bessel functions for the efficiencies of a variety of shapes of extended
Therefore, the pipe loses 47-8 Btu/h through each flange, and
is

lost

cooHng

surfaces.

Properties of Bessel Functions

3.7.

known properties and relationdo Bessel functions, although these are not so widely known. Only
a few of the more useful properties will be listed here, and the reader is
referred to any of the specialized books on Bessel functions given at the end
of the chapter, for a more comprehensive Hst. The formulae will be given
in sections on limiting behaviour, differential properties, and integral proJust as trigonometric functions have well

ships, so

perties.
3.7.1.

Behaviour Near the Origin

As x-^O,
Jk{x)

and

same Hmit as its leading term.

and are therefore zero when x is zero for k>0.

the Bessel function tends to the

IJ^x) start at x^

Therefore,

when

/c

Jo(0)

0,

^.(0)

when

/c>

The ambiguous

0,

=
=

/o(0)
/,(0)

=
=

(3.71)

(3.72)

J-,(0)=/_,(0)=oo

(3.73)

tions (3.56) or (3.67)

resolved in any particular case by reference to equawhere the sign is determined by the sign of r(A:+l)

given in Section

In

sign

5.3.

is

all

physical cases,

if

x =

is

a point in the system,

no physical property can be infinite and the part of the solution containing
J-ilx) or I-u.{x) must be eHminated anyway. The above physical argument
also appHes to Y^{x) and K^{x) which are both infinite at the origin for all
values of k. If k is not zero, the infinity is due to a negative power of x; if
k is zero, the infinity is due to the term In x. Therefore, for all values of k,
-y,(0)

X,(0)

If the origin is a point in the calculation

only physically permissible solutions.
3.7.2.

(X)

(3.74)

field,

then /^(x) and 4(jc) are the

Asymptotic Behaviour for Large X

If x>/:, then the term k'^ in Bessel's equation (3.48) can be neglected
comparison with x^ in the last term. Equation (3.48) becomes
,

d^d

Gardner, K. A.

dy

Trans. A.S.M.E. 67, 621 (1945).

in

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

114

substitution suggested by the

first

x\v

ax
d'z
*

^
Substituting (3.76)

dx^

'

and

-r2

two terms

,d'y

= x'-^ +
dx^

'

_,dy

ijZ.
X
"
dx

1V--3/2

-A + ^-r-ly

(3.77)

(3.77) into equation (3.75) gives

d^z
dx^

is

(3.76)

ax

x--^ +
Again, IjAx

is

(ix

^+x)z =

small compared with x'lf

d^
+
^

x^\.
z

(3.78)

dx'

This

is

the equation of simple harmonic

motion which has the solution

= ^cosx + 5sinx

Substituting back into equation (3.76),

>^

= ^x~^cosx + 5x~^sinx

(3.79)

With the above simple treatment, it has been shown that for x>l and
x'^k, the Bessel functions behave as damped oscillations of period 271. The
evaluation of A and B to simulate the behaviour of the defined Bessel
functions is an involved matching procedure, the result of which is

A similar procedure may be adopted with equation (3.64) for the modified
Bessel functions.

Thus, 4(x)
as

is

The

result

is

I,{x)--^e^

(3.82)

e-^
k.m-Jy,
2x

(3.83)

the only Bessel function of the four which diverges to infinity

increases; the other three decay to zero.

3.

SOLUTION BY SERIES

115

Differential Properties

3.7.3.

By

differentiating the series (3.56)

Jk{(xx)

The corresponding formulae

X

term by term

it

can be shown that

kJk{ccx)-(xxJk + i(ccxy

(XX Jk -

(ax) kJ,,(ccx)

Yk(ax)

for the other Bessel functions are

kYk(ccx)

-ccxYk+ i(ax)

axy,_i(ax)-/c7^(ax)

(3 84)

Ikiocx)

kI^{ccx)-^ccxIk+i(ocx)
kl^iccx)
+ ccxlk+ i(ocx)
ax/fc_

X - K^(ax)
dx

(3.85)

i(ax)- /c/fc(ax)

(3.86)

OLxKk
axKj^ + i(ax)

kKjJioix)
kKk((xx)

-axKk.i{oix)-kK,,(ax)

(3.87)

Integral Properties

3.7.4.

axV^_ i(ax) dx =

x^J^iax)

(3.88)

oix%_ i(ax) dx =

x'Y,(ax)

(3.89)

ax'^/fc.

i(ax)

dx

xH^iax)

Jax%_i(ax) Jx = -x%(ax)

(3.90)
(3.91)

The following integrals are needed for demonstrating the orthogonahty

of the Bessel functions (see Section 8.6). Since only J^{x) and 4(x) are
finite at the origin and therefore the only functions of practical importance
at the origin, the orthogonality integrals need only be quoted for these two
functions.
A,

Uctx)J,{px)xdx

-^^ [<xJ,(Jix)J,^,(ax)-pJ,{<xx)J,^,{M']

(3-92)

lJ,iccx)yx dx

ix' [Ji(ax)-J,_,iax)J,^ i(ax)]

(3.93)

J
X

jl,iax)I,{px)xdx

= -^^^ [ccI,(Px)I,^,(ax)-l}I^{ax)h^,(fixy]

(3.94)

= ix^ iliiax) - 1,. ,(ocx)I,^ .(ax)]

j lI,(ccx)Yx dx

(3.95)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

116
3.7.5.

Bessel Functions of Negative Order

Tables of Bessel functions are only given for zero and positive order, the
negative order Bessel functions have to be evaluated using recursion formulae
as follows. By rearranging equation (3.84).
axJj^_i(ax)

2kJi^(ccx)-(xxJf^+i{x)

(3.96)

Similar formulae can be obtained for the other Bessel functions by rearranging
equations (3.85), (3.86), and (3.87). If the Bessel functions of order k and

known, the Bessel function of order k 1 can be calculated.

also an integer, substitution into equation (3.56) and using the
properties of the gamma function. Section 5.3, it can be shown that

k+l

If

are

is

^'"^^^'^^^

J_,(ax)

(-l)V,(ax)

(3.97)

Y.,(ax)

= (-l)%(ax)
= h((xx)
= Kkioix)

(3.99)

I., (ax)
X_fc(ax)

which are only true for integer values of

3.7.6.

(3.98)

(3.100)

k,

Bessel Functions of Half Integer Order

Bessel's equations (3.48)

can be solved analytically for the case k

by substituting
z
as in Section 3.7.2.

x^y

(3.76)

This gives

which has the solution

z
.*.

= Asinx + Bcosx
= (Asinx + B cos x)/x^

Writing out the first few terms of the series for J^(x) from equation (3.56)
it can be verified that
J^(x)

and
Similarly,

J-^{^)=kI
it

(3.101)

cosx

(3.102)

nx

can be shown that

I,{x)=

^/ sinhx
'V

and

sinx

I.Jx)
^

(3.103)

nx

/
coshx
nx

(3.104)

y/

Equation (3.96) and the similar equation estabhshed from equation (3.86)
allow the J ov I functions of half integer order to be expressed in terms of
trigonometric or hyperbohc functions.

Chapter 4

COMPLEX ALGEBRA
4.1.

Introduction

In the following paragraphs, the algebra of complex numbers will be

described and the principles and properties of a complex variable introduced.
The concepts of a complex number and a complex variable are utilized

many branches of appHed mathematics and the simplest of these, the

complex roots of a quadratic equation, have already been introduced in
Chapter 2. It will be seen later that the basis of operational calculus is the
Consequently an underintegration of functions of complex variables.
standing of the fundamental properties of complex quantities is considered
essential for the appreciation of many mathematical methods employed in
the solution of chemical engineering problems. However, complex numbers
have no direct application in chemical engineering and the illustrative
in

examples are necessarily of a pure mathematical nature.

4.2.

The Complex Number

A complex number is
y are

real

an expression of the type x + iy, where both x and

numbers and the symbol
/

/ is

therefore not a real

number and

V^
is

(4.1)

defined as the fundamental "imagin-

The product iy will also be imaginary and every complex

have a real part and an imaginary part. In the number x + iy,
the real part is x and the imaginary part is y.
The algebraic operations of addition, subtraction, multiplication, and
division can be performed on complex numbers by treating / as an unknown
constant, and replacing (/^) by (- 1) on every occasion that it appears. Thus,
if two complex numbers are represented by z^ and Z2, where
ary" number.

number

will

Zj

= Xi + i>i

and

Z2

X2 + iyz

then the addition of these two complex numbers

Z1+Z2 =

Xi

+ I>i+X2 +

(xi+X2) + 1(^1

is

(4.2)

as follows.

l>2

+ ^2)

(4.3)

Subtraction gives

Z1-Z2 =

+ i>i-X2-iy2
= (xi-X2) + /(yi->^2)

M.M.C.E.

Xi

117

(4.4)
5

MATHEMATICAL METHODS

18

IN

CHEMICAL ENGINEERING

Multiplication gives

(4.5)

Division gives
Zi

Z2

+ t>i ^ xi + iyi X2-iy2

X2 + I>2
^2 + ^>2'^2-i>2
XiX2 + yiy2
yxX2-y2^
+
')
xl + yl
^2 + >^2
xx

(4.6)

The first three operations are quite straightforward but in the division
operation the complex number was eHminated from the denominator by
multiplying both numerator and denominator by a specially chosen complex
number. This number which is obtained by reversing the sign of the imaginary part of the denominator, is called the "complex conjugate" of the denominator, and Section 4.6 is devoted to the further properties of conjugate
numbers.

The two complex numbers

z^ and Z2 will be equal if their difference is

and by inspection of the subtraction operation (4.4) it can be seen that
only if Xj ^2 = and yiy2 = 0. That is, z^ and Z2 are only
Zi Z2 =
equal if their real and imaginary parts are separately equal.

zero,

In each of the algebraic operations presented in equations (4.3) to (4.6)

was identical to that which would be followed
the same operation was being performed on a real number. The resulting

the mathematical procedure

if

number was

in general

complex containing both a

real

and an imaginary

part; although either part could be zero.

4.3

The Argand Diagram

Early in the nineteenth century Argand suggested that a complex number

could be represented by a line in a plane in much the same way as a vector
The value of the complex number could be
is represented (c.f. Chapter 7).
expressed in terms of two axes of reference, and he suggested that one axis
be called the real axis and the other axis arranged perpendicular to the first

T
1

,1

-5 -4 -3

-2

-1

Fig. 4.1.

N 5/1

q1

<

111.
4

>

The Argand diagram

COMPLEX ALGEBRA

4.

119

be called the imaginary axis. Then a complex number z = x + iy could be

represented by a line in the plane having projections x on the real axis and
y on the imaginary axis. Such a line is illustrated in Fig. 4.1. Figure 4.1 is
called the Argand diagram and the plane in which the line is drawn is called
the Argand plane of z. The Hne OP represents the complex number 4 + 3/
whose real part has a value of 4 and imaginary part 3. On the other hand,
the Hne OQ represents the complex number 4/ 3 with real part 3 and
imaginary part 4. The lengths of OP and OQ are equal but the complex
numbers they represent are unequal because the real parts and the imaginary
parts of each are different.
This emphasizes the fact that any equation
involving complex numbers is equivalent to two real equations obtained by
separately equating the real and imaginary parts.
4.3.1

Modulus and Argument

For each of the complex numbers represented by the

5. That is,

lines

OP

and OQ,

the lengths of the Hnes are

x/42

+ 3^ = V(-3)2 + 42 =

(4.7)

The value of the length of the line representing the complex number is
called the "modulus" or "absolute value" of the number and is usually
written in the alternative forms,

The inchnation of the

positive real axis

and

is

|z|

= mod z = V^^ + y^

line representing the

(4.8)

complex number to the

"phase" of z,

called the "amplitude", "argument", or

is

usually written,

= amp z =

arg z

Solving equations (4.8) and (4.9) for

From

the above

it

rcos6

tan" ^ (y/x)

(4.9)

x and y gives
2indy = rsin6

(4.10)

can be seen that a complex number can be expressed

x+iy, or with the aid of equations

in terms of cartesian co-ordinates as

(4.10) in polar co-ordinates as

where

cis

is

rcosO

ir sin

r(cos

sin 9)

r cis 9

(4.11)

a convenient notation which should be read "cos 9 + i sin 0".

4.4.

Principal Values

When

a complex number is expressed in polar co-ordinates, the principal

value of 9 is always implied unless otherwise stated. That is in Fig. 4.1,
9p

9q
If the angle is

equal to (0-645

=
=

tan- \3I4)
tan"

H- 4/3)

=
=

37

143

=
=

0-645 radians
2-495 radians

its principal value amp (4 + 3/) would be

and amp (4/- 3) would be equal to (2-495 +

not restricted to

+ 27c)

radians,

52

MATHEMATICAL METHODS

120

CHEMICAL ENGINEERING

IN

Inn) radians, where n could be zero or an integer.

When the complex
number is represented on the Argand diagram the principal value is the
smaller of the two angles between the positive real axis and the hne in
question. The sign of the angle depends upon the sense of rotation from the

and this implies that the principal value lies in

shown in Fig. 4.2. It is sometimes more convenient

positive real axis,

the range

to choose

to

71

71

as

Fig. 4.2.

Principal value of the argument

a different range of values for 9 but the range is chosen here for consistency
with the definitions to be used in the theory of complex variables at the end
of this chapter.

4.5.

Algebraic Operations on the Argand Diagram

In Section 4.2 the addition, subtraction, multiplication, and division of

two complex numbers were carried out. If the Argand diagram describes
all the properties of complex numbers it should be possible to carry out the
above algebraic operations on the diagram. Thus consider Fig. 4.3a in which
the complex numbers represented by the lines OP and OQ in Fig. 4.1 are
redrawn. If z^ is the sum of Zj represented by OP and Z2 represented by

OQ

then,

or

Z3

Zi

+ 22 =

(xi

X3

xi

+ X2

and

+ x2) + i(yi + y2)

>'3

(4-12)

j;i+>^2

Equations (4.12) give the co-ordinates of Z3 on the Argand diagram as

shown in Fig. 4.3a by the line OR. Using the same numerical values as
before,

X3

= 4-3 =
.*.

It is easily

shown

Z3

that the point

the parallelogram

OPQR

and y^
X3

= 3+4 =

+ i>3 = l +

R can be located

as in Fig. 4.3a.

'7i'

7
(4.13)

geometrically by completing

COMPLEX ALGEBRA

4.

121

In the same way, the subtraction of two complex numbers can be expressed in the form of the addition of Zj to minus Z2 where

_Z2 = -(4i-3) = 3-4i

(4.14)

By completing the parallelogram OPQT in Fig. 4.3b to locate the diagonal

OT, the difference between the complex numbers z^ and Z2 (i.e. Z4 = Zj Z2)

--6
Addition and subtraction on the Argand diagram

Fig. 4.3.

is

obtained.

The co-ordinates of

Z4 can be verified algebraically as before,

thus,

Z4

From

Zi4-(-Z2)

(4

+ 3) + i(3-4) =

7-i.

the elementary geometrical property of triangles that the

the absolute lengths of any

side, there follows the

two

sides of a triangle

important
|^i

sum of

must exceed the third

result,

+ ^2h|^i| + |^2|

(4.15)

Argand diagram,
and division of two
terms of polar coordinates. Thus

In order to illustrate multipHcation and division on the

it

is

first

necessary to

show how the

complex numbers are expressed

Z5

or

Z5

where

ZiZ2

Tj r2 [(cos ^1

=
=

ri r2

rir2(cos0i

[cos (01

r5(cos05

cos 62

in

multiplication

+ sin 6 1) (cos 6 2 + sin 6 2)

sin 01 sin ^2) + ^(sin Oi cos 62 +

+ 62) +

sin (O^

(4.16)

cos 0^ sin ^2)]

+ ^2)]

isin^s)

r.5-/1,2
r, r-

and

65

9^

+ 62

(4.17)

Therefore, to multiply two complex numbers, it is necessary to multiply the

moduli and add the arguments. Hence the multiplication of two complex

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

122

numbers can be expressed on an Argand diagram by multiplication of the

two real numbers r^ and r2 to give the absolute length r^, and drawing the
line OU of length r^ at an angle (^1 + ^2) with the positive real axis as
illustrated in Fig. 4.4.

f\ a\
^

^1

u"

Fig. 4.4.

The

^2

"^

Multiplication and division

division of z^

by

Zj

on the Argand diagram

can be expressed in terms of polar coordinates

as follows.
Zi

ri(cos0i

4-

isin^i)

Z2

r2(cos02

isin02)

ri(cosOi

sin ^1) (cos 02

r2(cos^02

[cos (01 - 62) +

^'sin02)

sin^02)

sin {6^

- ^2)]

(4.18)

where the denominator was again made real by multiplying numerator and
denominator by the complex conjugate of the denominator.
Therefore, to divide two complex numbers it is necessary to divide the
moduli and subtract the arguments. Hence the division of two complex
numbers can be expressed on the Argand diagram by dividing the real
number r^ by the real number ^2, to give an absolute length of r^, then
drawing the line OV at an angle {0^ 9 2) with the positive real axis. This
is

illustrated in Fig. 4.4.

In the multipHcation and division operations 0^ and Q2 were the principal

However {O1+O2) ^^^ (^1 ^2)

need not be the principal values of the arguments of z^ and Zg. Thus consider the complex numbers Zj = (3/ 4) and Z2 = (/I) with arguments
01 = 143, and 02 = 135. By equation (4.5) Z1Z2 = 1-7/, and 05 = 0i +
02 = 278. The principal value of the argument lying between 180 and
+ 180 is 82 and this value should be used in subsequent calculations
values of the arguments of the numbers.

unless there

is

some

over-riding consideration (e.g. see Section 4.14.3).

COMPLEX ALGEBRA

4.

It

has been shown that

123

possible to represent simple addition, sub-

it is

and division of two complex numbers, on the

Argand diagram. Therefore it is obvious that by a series of geometric
operations it is possible to describe any process involving complex numbers
on an Argand diagram provided that they consist of a combination of the
above four operations.
traction,

multiplication

Conjugate Numbers

4.6.

Two complex numbers

such as x+iy and xiy of which the real and

imaginary parts are of equal magnitude, but in which the imaginary parts
are of opposite sign are said to be conjugate numbers.
On the Argand
diagram they can be considered to be mirror images of each other in the real
axis.
Conjugate numbers have already been used in the definition of the
division process. Usually, the conjugate of a complex number z is written
as z, and throughout this text conjugates of complex numbers will be written
accordingly.
If a

complex number possesses a conjugate, the imaginary part cannot be

the sum and the product of a complex number with its con-

However

zero.

jugate are always real.

Thus,

+ iy) + {x- iy) = 2x

(4.19)
and
(x + iy) (x - iy) = x^- (iy)^ = x^ + y^
(4.20)
/^

= 1. The division of a complex

i.e. the final result is a real number since
(x

number by

its conjugate will not produce a real number.

These relations
between sums and products can be extended to more than one complex
number and its conjugate. Thus if w and z be two complex numbers whose
conjugates are w and z respectively, let

w=
then

vv

also let

then

z
.*.

.'.

But

=
=
=

ri(cos e
ri(cos 6

r2(cos

(f)

r2(cos

(j)

= ri(cos 9 +
= (r^ cos 9 +

+
+

sin 9)

(4.21)

sin 9)

(4.22)

sin

0)

(4.23)

sin

(j>)

(4.24)

+ r2(cos + sin 0)
r2 cos 0) + i{r^ sin 9 + r2 sin
w + z = (r cos 9 r2 cos i{r^ sin 9 + r2 sin
sin
w-{-z = r i(cos 9 sin 9) + r2(cos
= (ri cos + r2 cos - i(ri sin 9 + r2 sin
w-\-z

-{

sin 9)

(j)

(/>)

(j)

(/>)

(4.26)

(j))

(4.27)

equations (4.26) and (4.27)

w+z = w+z
That

(4.25)

(j))

(/>)

From

(j))

the conjugate of the

sum of two complex numbers

(4.28)

is equal to the
conjugates of the complex numbers. In fact this principle can be
extended to any number of complex numbers.
is,

sum of the

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

124

In a similar manner
product of wz. That is

it

can be shown that

jP

p
Also

if

number q

the complex

is

q
q

number p

= wz
= wz

=
=

is

the

(4.29)
(4.30)

related to

then

the complex

if

>v

and

by

v^lz

(4.31)

vv/z

(4.32)

to the reader to verify equations (4.29) to (4.32).

It is left

De Moivre's Theorem

4.7.

This important theorem can be enunciated as follows: For all rational

is one of the values of (cos 0-\-i sin 0)".
The
proof of De Moivre's theorem depends on the value of n. That is, on
whether is a positive or negative integer, or a real fraction. Irrational
numbers such as n are not included, but this is an academic restriction.
The product of two complex numbers z^Z2 was shown to be
values of n, cos nd-\-i sin nQ

ri

r2[cos(0i

and there

in Section 4.5,

+ ^2) +

i'

sin (^i

+ ^2)]

was stated that the algebraic process could be

it

extended to evaluating multiple products. Hence,

Z1Z2Z3

=
=

+ 02) + ^sin(0i + 02)](cos03 + isin^3)

(4.33)

sin {0^ + O^) sin ^3 +

Ti r2 r3[cos (01 + O2) cos ^3
+ {cos 03 sin (9^ + ^2) + sin ^3 cos {6^ + ^2)}]
= ri r^ r3[cos (0^ + 02 + ^3) + sin {0^ + 02 + ^3)]
(4.34)
rir2r3[cos(0i

The above algebraic operations can be extended

when it can be shown that
ZiZ2...z
If zi

rir2...r[cos(0i

Z2-...

+ 02+

...

+0)

to n

ism(0i

complex numbers

+ 02+

...0)]

(4.35)

z,

then equation (4.35) becomes

z"

or

(cos

where
If

is

is

r"(cos

sin 0)"

a positive integer.
a negative integer, say
(cos

"
But using

De

m,

sin 0)"

sin 0)"

r"(cos

n6

cos nO

+
sin

sin 0)

nO

(4.36)

(4.37

then
sin 0)"'"

(cos

(cos0

isin0)"'

^"^'^^^

Moivre's theorem for a positive integer as given by equation

(4.37),

(cos

'

1
/

sin 0)"*

cos

m0 +

sin

m0

(4.39

COMPLEX ALGEBRA

4.

By a

125

by equation

direct application of the division rule as given

cos

mO +

cos

mO

sin

(-me) +
and

Collecting together equations (4.38), (4.39),

sin

(4.18),

- mO)

(4.40)

(4.40), gives the required

result,

(cos 9

sin 6)"

cos nO

n9

sin

(4.41)

De

Moivre's theorem for a negative integer.

a fraction, say p/q, where p and q are both integers with
q positive but p either positive or negative, then by De Moivre's theorem

which proves

Finally, if

for

an

is

integer,
'

Pn

COS -9
\
q

+
.

P ^'
isin-9

cos p9

(cos9

Taking the ^th root of each side of equation

cos- 9

p9

sin

sin- 9

9y

sin

(4.42)

(4.42) gives

(cos 9

sin

9)^^^^

(4.43)

Equation (4.43) proves De Moivre's theorem for fractional values of n;

hence, the theorem is proved for all rational values of n.

The

4.8.

mn

Roots of Unity

The th root of a complex number z is z^^" which can be expressed in

terms of polar coordinates, and it follows from De Moivre's theorem that
r^^"(cos9

9 + 2s7i

sin 9y^"

cos

sm

9 + 2snl

^^ ^^^

(4.44)

where the real number 5

(- 1).
0, 1, 2, ....
Equation (4.44) expresses the nth root of any number z real or complex
in its most general form, where the argument is not restricted to its principal
value. Therefore, if z = 1 -0, r = 1 -0, and 9 = 0, the nth. root of unity is
,

1^/"

cos

sm

n
or

more

specifically,

(4.45)

by substituting values of

s into equation (4.45) the nth

roots of unity are

.

cos

+ sm
z

271

1 -0,

cos

sm

2{n-\)n

cos -^

2n

n
.

cos

An

+
n

.An

2{n-\)n

+ sm -^^

sm

...

^^ ^^^
(4.46)

Inspection of equation (4.46) shows that all the n roots are different, and are
situated at the n corners of a regular -sided polygon centred at the origin.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

126

De

Moivre's theorem as expressed by equation (4.44) can be used to find

number and this is illustrated in the following
example.
the nth root of any complex

Example. Find the square root of 4 + Sv 3.

Using the equations (4.8) and 4,9) to express z in terms of polar coordinates
gives

r^

+ (3^/5y = 16 + 9x5 = 61
r = 7-81
and r^ = 2-195

4^

.-.

=
.*.

Using equation

tan-^(3V5/4)
cos iO

(4.44), the

tan"

1-6771

and

0-8695,

59 12'

III

0-4939

IV

two square roots of 4 + 3/\/5

are

2-795(0-8695 + 0-49390

or

+(2-43

+ 1-380

Complex Number

4.9.

If the

II

iO

sin

Series

complex number p^ can be expressed

as

Pm = ^m + iv

(4.45)

where both u^ and v^ are real, and if the complex number

sum of n complex numbers p^, then

Zn=

t,

m=l

The summations

Pm=

t,

u^

m=l

z is

t, v

Mm

and

^n>

(4.46)

m=l

^ u^ and ^ v are both sums of real numbers

m= 1

equal to the

^m

m=

yn

and

if

(4-47)

and as n tends to infinity both x and y tend to finite limits denoted by x

and y respectively, then the two series in equation (4.47) are convergent.
Their sum must therefore be convergent, or
00

00

z=\imz= Y
n-oo
m=

^m
1

Y. v^

m=

= x + iy

(4.48)

Therefore, for a complex series to be convergent, both the real part and the
imaginary part of the series must each be convergent.
The convergence of complex series is usually discussed by referring to the
associated series consisting of the absolute values of the individual terms.

That

is,

the convergence of the series given by equation (4.46)

^

is

= Pi+P2+... +Pn

(4.46)

discussed by referring to the series

S

|Pi|

+ |p2|+...+|p|

(4.49)

COMPLEX ALGEBRA

4.

If S tends to a finite limit

as -^oo, then the series (4.46)

But equation

absolutely convergent.

(4.

IPi-rPi]
.*.

is

said to be

5) gives

\Pi\-^\P2\

+ P2 + P3I ^

IP1

127

+ |p2l + |P3|

|Pl|

etc.

N^|Pl| + b2|+...|Pn|

.-.

and

it

follows that

if

the series (4.46)

is

(4.50)

absolutely convergent, then

it

must

also be convergent.

The convergence tests discussed in Chapter 3 are applicable to series of

complex numbers and therefore need not be repeated here, but an invaluThe
able result can be obtained from the exponential series as follows.
series for a complex exponent z was given in Chapter 3 as
e^

= l+z +

+
^
n

^'+...
2

which is convergent for

(3.7) becomes

,...,

values of

all

(3.7)

...

In particular,

z.

.L.i + L +

if

z>,

equation

(4.51)

...

Each of the series in equation (4.52) can be recognized by comparing with

equations (3.10) and (3.11) in Chapter 3. Thus
or

more

e'^

e^

+ >

cos y

sin

(4.53)

generally
e^

e^(cos

sin y)

(4.54)

Equation (4.53) can also be proved in the following neat fashion. Consider
f(y)

and

e" '^(cos y

sin y)

(4.55)

differentiate with respect to y.

.*.

-
dy

e~'y( siny

icosy)

=
=

e~'^( sin>'

icosy

ie~'y(cosy
icos>'

= ^ (constant)
(4.55) gives /(>^) = 1.
e~'^(cosy + isiny) = A =

but when

^^

0,

equation
/.

Rearranging gives equation

f(y)

(4.53).

cos

>'

sin

>'

e'^

isinj;)

siny)

Integrating,
*.


mathematical methods in chemical engineering

128

4.10.
It

Trigonometrical

Identities

has just been shown that

e'^

and

Exponential

similarly

it

= cosy +

isiny

(4.53)

isiny

(4.56)

can be shown that

e~'y

= cosy

Addition of equations (4.53) and (4.56) gives

+ e-'y = 2cosy
g-'y
e'^ +
cos>^ =
e^y

or

(4.57)

Subtraction of equation (4.56) from equation (4.53) gives

giy^-iy =
2isin>'

or

sin>'

(4.58)

The identities developed above between real trigonometrical functions

and imaginary exponential functions can be extended to include hyperbolic
functions. Thus, using equation (4.57) to find the cosine of a pure imaginary
quantity,

cos IX

which defines the hyperbohc cosine function,

cos ix

+ =

e^^

cosh X

(4.59)

Similarly,
-;2,

sin ix

sinh x

(4.60)

2i

which defines the hyperbolic sine function.

The above relationships can be extended to define a hyperbolic tangent
function as,
tan ix

sin ix

cos ix
.*.

tan ix

sinh x

tanh x

cosh X

tanh X

(4.61)

from the correspondby using equations (4.59), (4.60), and (4.61).

All properties of hyperbohc functions can be obtained

ing trigonometrical identities

The Complex Variable

In the paragraphs above, the complex number z has been expressed in
terms of the fixed real quantities x and y by the relation z = x + iy. If however the quantities x and y are variables, then so is z. In fact the quantity z
4.11.

4.

COMPLEX ALGEBRA

129

since z is related to the variables x and

x and y can be expressed by a curve in a
The plane is called the "z plane".

becomes a "complex variable", and

y\ the way in which z varies with
plane having coordinates x and y.
If the

variable

complex variable z

related functionally to another

is

complex

so that

\^=f{z)

(4.62)

and the complex variable w can be expressed by the

r^al quantities u and v

u + iv, then when z varies in the z plane, w will follow a curve
corresponding to equation (4.62) in the "w plane" whose coordinates are
u and V. Since z = a: + z> and w = u + iVyU and i; must be real functions of
both X and y. This will become clear in the following example.

so that

vv

Example. If z

= x + iy,

and w

w + /i;, where

w=
express u and y in terms of
Substitute x, y, u,
.*.

and
u

x and

v for

+ iv

z2

y.

w and
(x

z in equation

+ iyy =

I.

x^ + 2ixy y^

II

Equating real and imaginary parts gives

u

x^ y^

and

t;

III

2x>'

which determine u and v as functions of x and y.

Because the complex variable w has a functional relationship to z expressed by equation (4.62), derivatives of w with respect to z will exist
provided that /(z) is continuous in the region where the derivatives are
required. Thus a complex variable vv = /(z) is said to be continuous at a
point p if /(/?) has a finite value at z = p, and if the limit of /(z) as z
approaches /? is /(/?). That is,
lim/(z)=/(p)

(4.63)

If/(z) = z"^ and the point/? is the origin, the value of/(z) at the origin
depends upon the way in which the origin is approached. For instance, if
z approaches zero along the positive real axis,

lim z"^

whereas

if

00

(4.64)

z approaches zero along the negative real axis,

lim

z-^= -00

(4.65)

The Hmit of z~^

as z approaches zero does not therefore exist, and the

function z~^ has a "singularity" at z = 0.
On the other hand, if /(z) = z'^, the function is continuous everywhere
in the z plane but becomes infinite as z becomes infinite. Furthermore, z"^
has only one value at any given value of z and therefore it is said to be a
"single valued" function.

more than one value

valued" function.

However,

if

w is

for each value of

z,

related to z

and w

is

by w

z^, vv

can have

then said to be a

"many

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

130

can arise therefore in a number of ways when treating complex

many of these can be resolved by confining the function to
regions in the complex plane in which the functional relationships apply.
This fact will be better understood later in the chapter.
Difficulties

variables, but

4.12.

Derivatives of a Complex Variable

Consider the complex number

w=f(z)
and

to be a continuous function,

Then

the partial derivative of

(4.62)

w =

let

u+

iv

with respect to

and z = x + iy sls
x can be obtained

before.
in

two

ways, thus

dw
rdx

du

dx

(4.66)^

dx

dw^dldz^df

and

^^^^^

dx
since dz/dx

,dv

+ i^

-z-

dz dx

dz

Equating these two expressions,

Similarly, differentiating

df

du

dz

dx

cv

dx

with respect to y gives

dw

du

dv

dy

dy

dy

,, ,^,

dw
df dz
df

= _
= j
dz
dy
dz dy
.

and

(4.70)

since dz/dy

,df

_du

,dv

dz

dy

dy

"

(Ai\\

Eliminating dfjdz from equations (4.68) and (4.71) gives

Since w, v, x, and y are

must be equated.

,du

dv

,df

du

,dv

dx

dx

dz

dy

dy

all real,

the real

=
^
dx
dy
??!

'
'

If equations (4.73) are satisfied,

single valued function

engineering problems.
certain restrictions

and imaginary parts of equation

and

=-
dy

(4.72)

(4 73)

dx

the derivative dw/dz becomes a unique

in the mathematical solution of

which can be used

Hence the

derivative of a

which derivatives of

real variables

complex variable has

do not have and these

COMPLEX ALGEBRA

4.

131

They are known

are described by equations (4.73).

Riemann" conditions and they must be

plex number to have any meaning.

"Cauchy-

as the

satisfied for the derivative

of a com-

Analytic Functions

4.13.

function w = f(z) of the complex variable z = x + iy is called an

analytic or regular function within a region R, if at all points Zq in the region

it

satisfies
(i)

the following conditions.

valued in the region R.

has a unique finite value.
has a unique finite derivative at Zq which

It is single

(ii)

It

(iii)

It

Riemann

satisfies

the Cauchy-

conditions.

Only functions which satisfy these conditions can be utilized in pure and
applied mathematics, and therefore the whole of the following treatment
with analytic functions. However, when a mathematical analysis
of an engineering problem involving complex functions is being undertaken,
care should be exercised to confirm that the function satisfies the criteria
stated above.
The following examples illustrate the application of these
will deal

rules.

Example 1. If w =
Riemann conditions and

But

w =

z^

w =

=
+

,.

J^
ox

show

z^,

that the function satisfies the

state the region

(x
/v,

+ iyY =

wherein the function

3xy^

real

and

= 3x^-3/

du

dy

By inspection of equations III and IV

Riemann conditions (4.73) are satisfied.
of

z,

is finite;

hence the function

w =

parts,
II

= 3x^-3/
?
dy

III

^
ox

-^

and imaginary
3x^y y^

dv

= 6xy

and

Cauchy-

analytic.

+ 3ix^y 3xy^ iy^

x^

and equating

x^

is

IV

6xy
^

can be seen that the CauchyFurthermore, for all finite values

z^ is analytic in any region of finite
it

size.

Example

2.

If

iv

= z"^ show

Riemann conditions and

that the function satisfies the Cauchy-

state the region

w=

wherein the function

i = -^^^^
x-\-iy

x^

X
x^

+ y^

and

V
^
x^ + y^
^

analytic.

+ y^

is

'

II

MATHEMATICAL METHODS

132

From

equations

CHEMICAL ENGINEERING

IN

II,

du

x^
v^
y^-x"-

dv

III

dx~ dy~{x' + yy

Equations

III

du

dv

dy

dx

Ixy
{x^

+ y^f

and IV show that the Cauchy-Riemann conditions are

satis-

but consider the behaviour at the origin.

dujdx is obtained by keeping ;; constant, and to reach the origin, y must
be constant at zero; hence equation II becomes

fied at a general point

=-

-=-

:.

dx

which tends to negative

infinity as

X^

x tends

to zero through either positive or

negative values.
II

Similarly, dvldy at the origin must be evaluated

simphfies to

when x =

0.

Equation

_ -1
y

which tends to positive

negative values.

'-^

dy

y tends

infinity as

1
/

VI

to zero through either positive or

Therefore, from equations

and VI, dujdx and dvjdy

diverge to opposite extremes at the origin, and one half of the Cauchy-

Riemann conditions
the condition

is

not

satisfied.

du
Xdy

= -

The Cauchy-Riemann conditions

becomes

It

can be shown that the other half of

is satisfied, i.e.

infinite at the origin,

dv^
=

VII

dx
are not fully satisfied at the origin,

and the function w

= z"Ms

everywhere in the z plane with the exception of the one point z

4.14.

therefore analytic

0.

Singularities

has been shown in the examples above that the function w = z^ is

= oo whilst the function w = z~^ is analytic
everywhere except at z = 0. In fact no function except a constant is analytic
throughout the complex plane, and every function of a complex variable
has one or more points in the z plane where it ceases to be analytic. These
points are called "singularities", and they may be classified as follows.
It

analytic everywhere except at z

COMPLEX ALGEBRA

4.

(a) Poles

133

or unessential singularities.

(b) Essential singularities.

Branch points.

(c)

The

4.14.1.

(c) arises

Poles or Unessential Singularities

A pole
becomes
tion

and (b) are characteristic of single valued functions

with multivalued functions. Each will be discussed below.

singularities (a)

whereas

a point in the complex plane at which the value of a function

Thus w = z"^ is infinite at z = 0, and therefore the funcz"^ is said to have a pole at the origin. However, in the neighis

infinite.

w =

bourhood of the

origin, this function

is finite

and consequently analytic but

w = z"-^ is said to be "first

order" whilst the function w = z~^ possesses a second order pole. The order
of a pole is determined in the following manner.
If w = /(z) becomes infinite at the point z = a, then define
at the pole

it

ceases to be analytic.

^(z)

in

(z-a)"/(z)

(4.74)

it is possible to find a finite value of n which makes

then the pole of /(z) has been "removed" in forming
The order of the pole is defined as the minimum integer value of n for

where n

is

an

integer.

g{z) analytic at z
g{z).

The pole

which

g(z)

is

If

a,

analytic at z

The function

w=

-^

(4.75)

where p and q are both positive and finite, is a function containing multiple
That is, it contains a pole of order/? at the origin, and a pole of order
^ at z = ^. If either p ox q'ls not an integer, the order of the pole is the next
higher integer than p ox q respectively.
poles.

4. 1 4.2.

Essential Singularities

Certain functions of complex variables have an infinite number of terms

which all approach infinity as the complex variable approaches a specific

These could be thought of as poles of infinite order, but as the

removed by multiplying the function by a finite factor,
they cannot be poles. This type of singularity is called an essential singularity and is portrayed by functions which can be expanded in a descending
power series of the variable. The classical example is e^'^ which has an
value.

singularity cannot be

essential singularity at z

0.

/(^)

Similarly the series

= Tw^-)"''
=

(4.76)

has an essential singularity at z = a.

Essential singularities can be distinguished from poles by the fact that
they cannot be removed by multiplying by a factor of finite value. Thus
consider the function

w=

e^l'

(AJl)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

134

Using equation

expand the exponential function,

(3.7) to

w=

+-+
z

Since

is

is

some
,\

...

(4.78)

...

n\z"

2!z^

remove the

infinite at the origin, try to

by z^ where p

The

-^ +

singularity

by multiplying

finite integer.

z^w

= zP+z^-^ +

J- +

...

j-

(4.79)

...

equation (4.79) continues indefinitely and eventually n will be

Equation (4.79) thus consists of a finite number of positive
powers of z, followed by an infinite number of negative powers of z. Since
all terms of equation (4.79) are positive, then z^w->oo as z-^0. It is therefore
impossible to find a finite value of p which will remove the singularity in
e^^' at the origin. The singularity is thus "essential".
Finally, it can be shown that the behaviour of a function in the neighbourhood of a pole and in the neighbourhood of an essential singularity is
quite different. At a pole, the modulus of the function increases symmetrically about the pole to give an infinite pinnacle at the pole whereas at an
essential singularity, the function behaves very peculiarly.
For example,
Mclachlanf has shown that the function e^^' does not have the characteristics of a pole as z goes to zero. Thus for a fixed value of the argument of
z in the range
< ^ < ^tt the modulus traces out a spiral of ever increasing
= the function oscillates. This behaviour
radius in the w plane; whilst at
is most unlike that of a function near a pole.
series in

greater than p.

Branchpoints

4.14.3.

The

singularities described above arise from the non-analytic behaviour

of single valued functions. However multivalued functions such as w = z^^"
(where az is a positive integer) frequently arise in the solution of engineering
problems. Therefore this type of function must be accommodated in the
theory of complex variables. Thus consider the function

w=
and

let

z vary in such a

way

that

it

z^

(4.80)

follows a circular path whose centre

the origin. Converting to polar coordinates using equations (4.1

z

where

r is

1)

and

(4.81)

re''

a constant and 9 a variable.

is

(4.53),

Substituting equation (4.81) into

equation (4.80) gives

w=

y/re^''

(4.82)

Hence, when the variable z makes one traverse of the circumference of the
circle in the z plane, 6 passes through 2n radians, but w with a constant

modulus of V^

will

only pass through n radians by equation (4.82).

"Complex Variable Theory and Transform Calculus,"

t Mclachlan, N. W.
Cambridge University Press (1955).

If z
p.

16.

4.

of the circle in the z plane with 6 changing from 2k

complete the circle as shown dotted in the w plane
Consequently for any value of z represented by a point on the

makes a second
to

47r,

circuit

the function

of Fig.

4.5.

135

COMPLEX ALGEBRA

will

plane

Vhn

Tri
\

Fig. 4.5.

Branch point of

vv

r*

on the 2 and w complex planes

circumference of the circle in the z plane, there will be two corresponding

values of w represented by points in the w plane. One value will be on the
solid line semicircle where O<J0<7c, and the other will be on the dotted
Une semicircle where n<\d< In. Hence, by the simple expedient of restrictto the range 0^0 <2n, the function w becomes single
ing the values of
valued, and the Cauchy-Riemann conditions can be expressed as follows.

w
u
..

u-\-iv

\lrQos\Q

\lr e^'^

and

\lrs\n\9

^cosi^
T-=
dr

dv

du

dv

/-

/-sm

2Vr

/-

and
which gives
du

du

dv

dv

dr

cO

dr^ ~rd9

(4.83)

Equations (4.83) are the Cauchy-Riemann conditions in polar coordinates.

final step to equations (4.83) can only be made in this example if 6 is
restricted to the range 0^9<27i.
The restriction which makes a multivalued function single valued within
a given region enables these functions to be employed mathematically. The
region where the function is single valued is called the "branch", and the
particular value of z at which the function becomes infinite or zero is called
the "branch point". In the example above, the origin is the branch point.
The function w = z^ was multivalued because the contour followed by
z enclosed the branch point at the origin. However this is a particular case,
since the complex variable z may travel along any open curve in the z plane
or any closed curve which may or may not enclose the branch point. Thus

The

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

136

shown in Fig. 4.6 which does not

occasion the angle 6, which is the argument
of z, changes from 9^ to 62 and back again to 6^, and does not change by 2n
radians in traversing the contour.
Hence the function is single valued
around the contour in Fig. 4.6 without any action being taken to restrict the

let

the function

vv

z^ follow the contour

enclose the branch point.

Fig. 4.6.

On

this

Contour of w

z* not enclosing branch point

This shows that unless the contour of z encloses the branch

on the contour and
this value cannot be altered by 2n by traversing the contour.
It has now been demonstrated that a function is only multivalued around
closed contours which enclose the branch point, so it is only necessary to
ehminate such contours and the function will become single valued. The
simplest way of doing this is to erect a barrier from the branch point to
infinity and not allow any curve to cross the barrier. With this convention,
no curve can ever enclose the branch point and the function becomes single
valued and therefore analytic for all permitted curves. The barrier must
values of

9.

point, the angle 9 can only have one value at each point

+ 77
-rr

Fig. 4.7.

Barrier erected to isolate branch point

4.

COMPLEX ALGEBRA

137

any direction in the

In most normal applications
the barrier is drawn along the negative real axis as shown in Fig. 4.7 so that
it confines the function to the region in which the argument of z is within
the range n<9<n. This branch is then termed the "principal branch",
and the correct name for the barrier is the "branch cut".
The above discussion refers to a single branch point at the origin. If the
z plane contains only one branch point which is not at the origin, it is a simple
matter to transfer the origin to the branch point. When the z plane contains
more than one branch point, it is necessary to draw a branch cut from each
branch point to infinity. There is much scope for choosing a suitable set of
branch cuts in such cases but no new ideas are involved and this complication does not occur very often in practical applications.
Start

from the branch point but

z plane, and

4.15

may be

it

can go to

infinity in

either curved or straight.

Integration of Functions of Complex Variables, and

Cauchy's Theorem

The successive values of a complex variable z can be represented by a

curve in the complex plane, and the function w = /(z) will have a particular
value at each point on this curve. Therefore the integral of /(z) with respect
to z is the sum of the products /^(z) Sz along the curve in the complex plane,
where /m(^) is the mean value of /(z) in the length Sz of the curve. I.e.
lim

Y,Mz)Sz=

\f(z)dz

(4.83)

where the suffix C under the integral sign specifies the curve in the z plane
along which the integration is performed. However, if w = u + iv = f(z),
and z = x + iy are substituted into equation (4.83),

= j(u + iv) (dx +

jf(z) dz

dy)

= j(udx-

vdy) + ij(vdx

+ udy)

(4.84)

Equation (4.84) expresses the integration of the complex function w = /(z)

with respect to z. If both w and z were real, the integral would become

J
since the values of v

of real functions
4.15.1

is

and y would be

u dx

zero,

which shows that the integration

a special case of a contour integration along the real axis.

Cauchy's Theorem

any function is analytic within and upon a closed contour, the integral
taken around the contour is zero.
The above theorem can be proved using a very concise procedure by
referring to Stokes' theorem, which is
If

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

138

J(P.x + e^.) =

/J(g-^)ixi,
A

where

P and Q

hand

(4.85)

are functions of

x and

y,

and the surface

integral

on the

taken over the area

enclosed by the closed contour C.
Stokes' theorem will be proved in Section 7.7.4 to which the reader is referred.
If
represents a closed curve in Fig. 4.8 and there are no singularities of /(z) within or upon the contour, the value of the integral of /(z)
right

side

is

KLMN

Positive
direction

Contour integration of /(z) around closed curve

Fig. 4.8.

around the contour

is

closed, each integral

on the

expressed by equation (4.84); and since the curve is

right hand side of equation (4.84) can be restated
as a surface integral using Stokes' theorem (4.85), thus

jiud.-vdy)=-jj[^y^),.,y
C

and

J(...

(4.86)

+ ..) = J|(^^-|).x..

(4.87)

But for an analytic function, by the Cauchy-Riemann conditions (4.73)

each integral on the right hand side is zero.
jf(z)dz

(4.88)

which proves Cauchy's theorem.

In Fig. 4.8, let the region of the complex plane within and upon the
Therefore a function of z, /(z)
contour
contain no singularities.
will be analytic within and upon the contour and Cauchy's theorem (4.88)
on the contour
is valid. If now the integration of/(z) is started at the point
and terminated at the point M, then the value of the integral along the curve
will be

KLMN

KNM

J
KNM

mdz

(4.89)

COMPLEX ALGEBRA

4.

139

This will be positive by the conventional order of the suffices of the integral
corresponding with the description of the path of integration. If the integrafollowing the positive direction of the arrow in
tion is continued along
Fig. 4.8, the value of the second integral is

MLK

f{z)dz

(4.90)

MLK
However, the value of the integral around the closed contour
zero by Cauchy's theorem and therefore

J
KNM

f(z)dz+

J
MLK

f(z)dz

KNMLK

is

(4.91)

or reversing the order of the suffices of the second integral and therefore

its

sign,

J
KNM
That

is,

f(z)dz==

J
KLM

f(z)dz

the value of an integral of /(z) between

(4.92)

two points

in the

complex

independent of the parth of integation, provided that the function

is analytic everywhere within the region containing all of the paths.
Cauchy's theorem and its corollary that the value of the contour integral
is independent of the path are of great value for the integration of complex
functions of complex variables. The following examples indicate extensions
of these ideas.
plane

is

Example
6/ is the

1.

Show

that the value of

same whether the intregation

around the path

ACDB

AB

is

is

z^ dz between z

and z

carried out along the path

AB

+
or

in Fig. 4.9.

Fig. 4.9.

The path

Two

equivalent contours

given by the equation

y
.-.

z^

ix

{x + iixy

7 + 24i

16

II

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

140

and

dz
8

= dx + ^idx

+ 6i

III

x^ dx

IV

-44+117/ fx^r

[ii:

64

-352 + 936/

ACDB

Consider the integration along the curve

(a)

Along AC, x

and thus z

in parts,

iy.

1000/

~3~
L

(b)

The

CDB

arc

is

of constant radius
8

+ 6i

-lOi

lOOe^'HOie'^dO

VII

-in

= tan'^ =

a
8

lOe'^.

j z^dz= j
where

10, :.z

VI

JO

-^

36 52'

+ 6i
r

z2jz

= lOOO/TigsJ'
L3/

ioi
lOt

1000
(cos 3a

-in

sin

3a

/)

1000

(-0-352 + 0-936/-/)

-352-64/
VIII

The complete

integral

round

ACDB

is

sum of equations VI and VIII

the

which gives

-352 + 936/

as before.

The

integral of z^

from z

to z

+ 6/ is

thus independent of

the path taken.

Example

2.

Evaluate

z~^ dz around a

circle

with

Let z

re'^,

then dz

ire'^dO since r

is

constant.

its

centre at the origin.

4.

COMPLEX ALGEBRA

141

dO

'mi
=
That

is,

z~^dz =

although the function

is

not analytic at the origin,

z"" Jz

Similarly,

is

always zero for any integer value of other than unity.

Cdz
Example

As

Evaluate

3.

around a

c
in the previous example, put z

''

J
c

circle

r e'^

"J

with

with

its

centre at the origin.

r constant.

re''

= ml''
This result
4.15.2.

is

one of the fundamentals of contour integration.

Cauchy's Integral Formula

Consider Fig. 4.10 and let a complex function /(z) be analytic upon and
within the solid hne contour C. Let be a point within the closed contour

Fig. 4.10.

such

that/(fif) is

Integration contour for

not zero and define a

new

f(z)

-^^ dz

function g{z) by

fiz)

9(z)

za

(4.93)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

142

This function is analytic within the contour C except at the point a where
a simple pole will exist. If this pole is isolated by drawing a small circle y
around a and joining y to C as shown dotted in Fig. 4.10, the value of the
integral around the complete modified contour will be zero by Cauchy's
theorem. The straight dotted Hnes joining the outside contour C and the
inner circle y are drawn very close together and their paths will be synonymous. Since integration along them will be in opposite directions and g{z)
is analytic in the region containing them, the net value of the integral along
the straight dotted lines will be zero.

K^-^^ =

{-^-^

.*.

J
C

z-a

(4.94)

za

J
y

Let the value of /(z) on y be

/W =/() +
where

e is

.-.

Since f(d)
z

(4.95)

a small quantity.

is

a-\-r e^^,

rM^_
za

r/i^^_

z-a

ri^=o
z-a

(4.96)

constant, the second integral can be evaluated by putting

and

as 6 varies

by 2n radians the curve

2k

m-*'A

y is described.

re'Ue
re''

=
As

2nif{a)

(4.97)

the radius of the circle y tends to zero, also tends to zero and the

third integral in equation (4.96)

.-.

becomes

/(.)

insignificant.

^r^
2ni J

za

(4.98)

which

It permits the evaluation of a

"Cauchy's Integral Formula".
function at any point within a closed contour when the value of the function
on the contour is known.
is

4.16.

Laurent's Expansion and the Theory of Residues

The following section of complex variable theory is paramount to the

understanding of the Laplace transformation and its effective application to
chemical engineering problems.
Laurent's expansion is an extension of
Cauchy's integral formula for the case when f(z) has a singularity at some
point within the contour C.
For simplicity, take a coordinate system with its origin at the singularity
of /(z) and draw two concentric circles Ci and C2 centred at the origin so
that/(z) is analytic in the annulus between C^ and C2. This is illustrated in

COMPLEX ALGEBRA

4.

143

Fig. 4.11 where the two circles are also joined by two straight dotted lines
which connect the circular contours into one simple continuous closed
curve, /(z) is analytic within and upon the complete contour and thus

Fig. 4.11.

Cauchy's integral formula

Verification of the Laurent theorem

valid.

is

Therefore

if

is

located between C^

and C2, and the two dotted Hnes are superposed, then

M ^V^-IV^
=

2ni J

2ni J

where the negative sign indicates that C2

is

(4.99)

za

described in a negative or clock-

wise direction.
In the

first

integral of equation (4.99),

expanded by the binomial theorem

1/,

az\
L,

a\-^

zj

and {z a)~^ can be

powers of z, thus

|z|>|a|

in descending

a"
(4.100)

n+l

Similarly, on curve C2, |^|<|| and (zd) ^ can be expanded by the

binomial theorem in ascending powers of z, thus

-1

z\-'
z
z'"-^
1/^
-1-=-+
+... + ^ +
1

aa\

...

(4.101)

m=l
Substituting the expansions (4.100)
()

and

(4.101) into equation (4.99) gives

^.J|/^^" + ^J|^z-V(.)c^.a-"

(4.102)

MATHEMATICAL METHODS

144

CHEMICAL ENGINEERING

IN

any simple continuous closed curve containing the origin with

and upon C with the exception of the singularity at the
origin, then all integrals in equation (4.102) can be taken round the curve C
instead of Cj and C2.
If

is

/(z) analytic within

and putting

m =

-,the two

series

/(a)=

can be combined to give

ba"

f;

n=
.

(4.104)

CO

(4.105)

1-Jf#
c

which

is

the Laurent expansion.

The group of terms for negative values of n is known as the "principal

part" of the Laurent expansion near the singularity. If the singularity of
/(z) at the origin is a pole of order A'', then the second series in equation
(4.103) consists of a finite number of terms; because if m>N, the integrand
consists of positive powers of z and the integrals are zero by Cauchy's
theorem. Equation (4.104) can then be written
/(z)

Cz"

:?!

n=0

^+
Z

...

(4.106)

where B = Z?_ is given by equation (4.105), and a has been replaced in the
expansion by z.
Equation (4.106) can be used to obtain the integral of /(z) round any
contour C enclosing the singularity as follows. The infinite series of positive
powers of z is analytic within and upon C and the integral of these terms will
be zero by Cauchy's theorem. Referring to Example 2 in Section 4.15.1,
the integral of each of the other terms on the right hand side of equation
(4.106) will also be zero except for the term {BJz). Example 3 of Section
4.15.1 shows that this integral will be IniB^, or
jf(z)dz

2niBi

(4.107)

Hence the value of the contour

plied

integral of a complex function is 2ni multiz"^ in the Laurent expansion of the complex
This coefficient is known as the "residue" of the function at the

by the

function.

coefficient of

pole.

The above

when

the singularity is, not at the origin,

Laurent expansion is obtained as a series
of powers of (z Zq) and the residue of the function at Zq is the coefficient of
(z Zq)"^ in the Laurent expansion in the neighbourhood of Zq. If the
Laurent expansion is valid near an essential singularity, the coefficient B^
is still termed the residue at the singularity.

thus

if

analysis

the pole

is

at z

is

valid

Zq, the

4.

COMPLEX ALGEBRA

Ce'dz
Example

145

3- around a contour enclosing

Evaluate

the origin.

Expanding the integrand gives

V.3;i + ^ + ^+-

=M

111

+ -^ + +
^
2z
z^
z^

=
Equation

II is the

cient of z"^

is J.

II

...

Laurent expansion of e'/z^ near the origin and the

Therefore the residue at the pole at the origin is J.

2ni{i)

coeffi-

ni

III

Example

2.

Evaluate

-^

around a

circle

centred at the origin.

0-

c
If

|z|

<

the function

|<3|,

is

analytic within the contour

e'dz

J iz-ay

and

by Cauchy's theorem.
If

|z|

>

||,

there

is

a pole of order 3 at z = a within the contour. There= a by putting A = z a.

fore transfer the origin to z

Then
r

e'dz

r
c

All integrals

on the

right

hand

side are zero except the third

term which

gives the residue \ e

re^^^dX
c
4.16.1.

e'dz

,,

,,

Evaluation of Residues without the Laurent Expansion

With a

large

number of expressions encountered

in

apphed

science, the

not immediately apparent to engineers. When this is

the case, the following procedure is helpful. The complex function /(z) can
be expressed in terms of a numerator and a denominator if it has any

Laurent expansion

is

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

146

singularities.

That

is

f(^)

^^

(4-108)

then (z a)

a factor of g(z) so that

= a. Furthermore,
the Laurent expansion of /(z) will exist even though it is not apparent, and
therefore it can be written in the general form of equation (4.106),

and

if

a simple pole exists at z

^,

is

g(z) can be written (z a) G(z) with G(z) not zero at z

f(z)

-^
+ bo + b,(z-a) +

...

+ bXz-a)"+...

(4.109)

at the first inverse term because there

Multiplying both sides by (za) and then

Equation (4.109) has been terminated

only a simple pole at z
putting z = a gives
is

B,

a.

(z-a)/(z)|,=,

F(a)!G(a)

(4.110)

which evaluates the residue at a simple pole. This is a very convenient

method which will be illustrated by examples below, but certain points
must be emphasized first. Thus in equation (4.108), g(z) must contain one
more factor (z - a) than F(z) for /(z) to have a simple pole. Finally, if g(z)
has a number of factors so that

^<"-

<""'

(z-.,(.-''l..),.-..)

f(z) will have a residue at each pole, z = a^, a2, .... a.

evaluated independently by the method given above.

Each residue can be

Example
^

3.

Evaluate the residues of -^

z^

f(z)

=
+ z-12
and z = 4.
z^

which has poles at z = 3

Using equation (4.110), the residue
z

= -4

is

-4/(-7) =

The sum of the

-r-

+ z-12

+ 4)(z-3)

(z

at z

is

3/7;

and the residue

at

4/7.

residues

is

^4-^=1
Example

4.

Evaluate the residues of

z^
f(z)

=
(z

which has poles

The

at z

= +

residue at the pole z

+ w^

+ iw)(z iw)

iw.

iw

is

z + iw

II

2iw

COMPLEX ALGEBRA

4.

The

residue at the pole z

= w

is

iw
of e'l(z^ + w^)

III

2iw

The sum of the

residues

147

is

IV

2iw

The above method of determining the residue of a function appears to

depend on the factorization of the denominator. However, the method is
generally applicable to the evaluation of the residue at a simple pole even

when

the denominator cannot be factorized.

If /(z)

is

expressed as in equa-

and the denominator ^(z) does not factorize, the same procedure
be followed. Thus the residue of /(z) at z = a is

tion (4.108)

may

still

(z-a)F(2)
^'-'''^

g(z)

which

indeterminate. Therefore, by L'Hopital's rule, Section 3.3.8,

is

_ F(z) + iz-a)F'{z)
~

d{z-a)F(z)ldz

^W

dg{z)ldz

where g\a)
at z

is

^-^

''''''

the differential coefficient of g(z) with respect to

z,

evaluated

a.

C e'dz

Example

5.

Evaluate
J
c

around a

sm nz

and radius \z\ <n/n.

The residue at the pole where z

B.=^^''

is

'

dsmnzldz\^ = Q

4.

6.2.

sm nz

with centre at the origin

circle

given by equation (4.113) as

n cos nz

\n/

Evaluation of Residues at Multiple Poles

If /(z) has a pole of order n dX z

a and no other singularity, /(z) can

be written in the form

/W =
where n is a finite integer, and F(z)
by the Taylor series to give
F(z)

F(a) + {z-a)FXa)

is

(^

analytic dXz

+ ^-^^F'Xa) +
z

(4.114)

...

a,

F(z) can be expanded

+ ^I^lF\a) +
n

...

(4.115)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

148

Dividing throughout by

^(^>

which
is

is

(z

a)"

= (I^" +

gives

+ - + (-l)!(z-a) + -

(7^^-

the Laurent expansion of /(z)

and therefore the residue

the coefficient of (z a)"^ in equation (4.116).

pole of order n situated at z

is

Evaluate

c:

given by

This reduces to equation (4.110) for a simple pole

6.

at z

Therefore, the residue at a

^'=(;r^^^K-a)"/(z)]

Example

^'-'"'^

f cos 2z dz
rj-

around a

when
circle

(4.117)

1.

of radius

|z|>|a|.

cos 2z
,

TT

(z-aY

has a pole of order 3 at z

a,

and the residue must be evaluated

using equation (4.117).

- i_ii \(z-afcos2z\
^'-2\dz^[ {z-af J_
= K-4cos2z)|,.,
= 2 cos 2a

''

cos 2z dz

/
C

{2

-ay

27c/(

2cos2fl) =

4711 COS 2a

Chapter 5

FUNCTIONS AND DEFINITE INTEGRALS

5.1.

Introduction

The study of differential equations is one of the many branches of mathematics in which the object is to determine a functional relationship between
twq or more variables. The symbohsm which is used to describe these
functional relationships is largely a matter of convention when a particular
symbol is found convenient, it enters general usage and its properties are
Elementary functions such as polynomials, powers, logainvestigated.
rithms, exponentials, trigonometric and hyperbolic functions should be
famihar to the reader already, and in Chapter 3 it was shown that the four
kinds of Bessel function were useful for expressing the solutions of a particular class of differential equations. Other functions have been defined as
the solutions of differential equations, and the Legendre polynomials which
will be introduced in Chapter 8 is one such group.
Functions are also defined by integrals which cannot be evaluated in
terms of the above elementary functions, and the purpose of this present
chapter is to introduce some of these functions, enunciate their properties,
and draw attention to their tabulated values as an aid to calculation. It will
be shown that a variety of integrals can be expressed in terms of these new
functions and thus be evaluated from tables.
Methods for evaluating some definite integrals by contour integration
were presented in Chapter 4, and a further method by differentiating with
respect to a parameter will be included in the present chapter
;

5.2

The Error Function

This function which occurs in the theory of probability, distribution of

and diffusion of matter, is defined by

residence times, conduction of heat,

the integral,

erfx

The

error function

is

= ^=

e~'^dz

illustrated in Fig. 5.

(5.1)

which shows erf x as the area

beneath the curve and between the ordinates z =

and z = x when 2/ \/n
exp ( z^) is plotted against z. In equation (5.1), z is a "dummy variable"
because it only enables the curve of Fig. 5.1 to be described and any variable
would do this. The variable z is ehminated by the hmits of integration thus
M.M.C.E.

149

MATHEMATICAL METHODS

150

CHEMICAL ENGINEERING

IN

The

Fig. 5.1.

leaving

The

as the only variable.

error function

factor l/yjn

is

introduced for convenience

so that
erf 00

The

(5.2)

truth of equation (5.2) can be proved by the following argument.

simplify the algebra,

To

remove the normalizing factor 2/v tt and consider two

versions of the integral, thus

/=

{e-'^'dx^ je-y"dy

(5.3)

Considering that x and y are two independent cartesian coordinates, equation (5.3) can be written
R R
/2

|Je-('^-^^^>rfxJy

which can be interpreted as a volume as shown in Fig.

Fig. 5.2.

5.2.

If the variables

Evaluation of erf oo

X and y are changed to polar coordinates

x

(5.4)

rcos6,

and 9 using the equations

rsin6

(5.5)

then the element of area (dx dy) has to be replaced by (r dr dO). Also, the
range of integration has to be altered from the square OABC, to the quadrant
which results in an error denoted by e. Thus, equation (5.4) can

OAC

5.

FUNCTIONS AND DEFINITE INTEGRALS

151

be written
Rnll

P = \]
The volume represented by

maximum

height of exp

and as R-*co,

e-'-'rdrdO

has a base area which

(5.6)

is less

R^). Thus
e<iR^exp(-R^)

than J R^ and a

(5.7)

s-^0.

Evaluating the double integral in equation (5.6) gives

P = in- in exp (-R^) + s

Therefore, as

R^oo, P-^in. Comparing

required result which

5.2.1.

is

equation

(5.8)

equations (5.1) and (5.3) gives the

(5.2).

Properties of the Error Function

Differentiating equation (5.1) gives directly

-^erfx
dx
Using

this result to integrate

erfx^x

= ^e-^'

equation (5.1) by parts gives

xerfx

= xQdx +
where

(5.9)

yjn

x-j:ze~^^ dx + C

-j^Qxp(-x^) + C

(5.10)

Equation (5.10) is sometimes tabu- l/yn so that ierf = 0.

lated under the symbol "ierf x" with
Another related function which is sometimes tabulated is the complementary error function "erfc x'\ This is defined by the equation
is

the constant of integration.

C=

erfcx

erf X
00

tJ'5.3.

This

is

dz

(5.11)

The Gamma Function

defined by the integral

00

T(n)

jt"-^e-'dt

(5.12)

f is again a dummy variable since the value of the

independent of t. The variable t is only used to describe
the function to be integrated. As an illustration of this fact, take the special

for positive values of .

definite integral

is

62

MATHEMATICAL METHODS

152
case n

],

CHEMICAL ENGINEERING

IN

thus

r(i)=
r(i)

:.

which

independent of

is

e-'dt
\

(5.13)

/.

The most important property of

the

gamma

function can be derived by

integrating equation (5.12) by parts.

CO

r()=

t"-'e-'dt
OC

= [-t"-'e-q'^ + (n-\) \f-^e-'dt

*'

'
.-.

since the

first

term

T{u)
is

= {n-\)Y{n-\)

(5.14)

zero at both Hmits of integration provided that

n>\.

Repeated applications of equation (5.14) for integer values of n gives

r(n)

(n-l)(^i-2)(...)(2)(l)r(l)

= {n-l)\

(5.15)

by using equation (5.13). The gamma function is thus a generalized factorial,

and for positive integer values of n, the gamma function can be replaced by a
factorial as given by equation (5.15). For non-integer values of , the defining integral (5.12) has been evaluated numerically for l<n<2 and this
table can be extended to all positive values of n by repeated use of equation
(5.14).

The

gamma

defining integral (5.12) is not valid for negative values of /?, but the
function of a negative quantity can be found by using equation

(5.14) to extend the definition.

Thus, rearranging equation

(5.14),

r(u-\) = r(n)i{n-i)

r(

+ l)/n(n-l)

(5.16)

etc.

This process can be repeated until the gamma function on the right hand
side has an argument lying in the range 1 to 2. However, if n is zero or a
negative integer, the gamma function becomes infinite. An example of this
can be seen by putting n = 1 into equation (5.14), thus

= Or(0)
r(l) = 1. Therefore,

r(l)

But equation (5.13) shows that

to satisfy equation (5.17).

It is

(5.16) will eventually involve r(0)

the

gamma

function

Figure 5.3 shows

is

is

r(0) must be infinite

obvious that the sequence of operations

whenever

ri

is

a negative integer, and thus

infinite for all negative integers.

how

alternatively positive

(5.17)

r{n) varies with n. For negative values of n, r()

and negative between successive integer values, and

5.

FUNCTIONS AND DEFINITE INTEGRALS

r(n

Asymptotes

153

l)

The gamma function

Fig. 5.3.

for positive values of n the curve has a

minimum

value of 0-8856

Sit

1-4616.

r(J) can also be evaluated analytically as follows.

tion (5.12),

By

definition, equa-

r(i)= \t-ie-^dt
The negative square root can be removed by making the
'

dt

r(i)

.-.

(5.19)

and

dx

=V

7t

(5.19)

erf 00

(5.2) gives finally,

= V^

Other half integer gamma functions can be found using equations

and (5.16), as illustrated by the following examples.
1.

x^,

(5.1) gives

r(i)

Example

= Jx-^e-^'2xJx

T{\)

and using equation

substitution

2x dx.

-^
Comparing equations

(5.18)

Evaluate r(3i).

Successive application of equation (5.14) gives

r(3i)

|r(2i)

(5.20)
(5.14)

MATHEMATICAL METHODS

154

=
by equation

CHEMICAL ENGINEERING

IN

|.|.ir(i)

(5.20).

r(3i)

.-.

J/v/7r

Example 2. Evaluate T{ 2\).

Application of equation (5.16) gives

r(-2i) = ^--r(-u)

-2 -2 -2
-8

,-

Example 3. Evaluate F (3-7).

Application of equation (5.14) gives
r(3-7)

From

tables,

it

=
=
=

2-7r(2-7)
2-7(l-7)r(l-7)

4-59r(l-7)

can be found that

F(l-7)
F(3-7)

.-.

5.4.

= 0-90864
= 4-1707

The Beta Function

This is yet another function which is defined by an integral, but the beta
function contains two parameters instead of one, and the integral is taken

between fixed

finite limits.

Viz.
1

B(p,g)

jr''-'(l-0'-'rfr

(5.21)

Both p and q must be positive to ensure convergence of the integral

at its

limits.

An

form can be obtained for the defining

alternative

the substitution

sin^ 0,

\-t =

B(p,g)

cos^

0,

.'.

dt

2 sin

integral

by making

cos OdO.

Thus
71/2

By

substituting

cos^

in

sm^p-^ecos^^'-'ede

(5.22)

equation (5.21), an equation can be obtained

similar to equation (5.22) but with/7 and q interchanged, thus showing that the beta function is symmetrical in its two arguments p and q.

which

is

FUNCTIONS AND DEFINITE INTEGRALS

5.

A further

useful substitution into equation (5.21)

t

1
1

+ x'

dt

+ x'

155

is

dx

=
(1

+ x) 2

which gives

x^-^dx
B(p

5.4.1.

(5.23)

Evaluation of B{\ q, q)

For the

special

case/7+^

1,

equation (5.23) becomes

dx^

/^

B{\-q,q)

(5.24)

This integral can be evaluated by contour integration using the methods

developed in Chapter 4 as follows.
Consider integrating z"V(l-^) round the contour shown in Fig. 5.4
where the complex z plane has a branch cut along the negative real axis
y

R/'^
y\
K/
Barrier

is

Integration contour of beta function

Fig. 5.4.

because q

not an integer.

The contour

consists of a circle

C^ of

large

described anti-clockwise, and a circle

of small radius r described
clockwise. The two circles which cannot cross the branch cut, are joined
on either side of the branch cut by straight Hues. The only pole enclosed by
radius

this

pole

is at z = 1 (provided r<\<R) and the residue at this simple

unity by equation (4.113). Consider the four sections of the integral

contour
is

separately.

OnC^,

<

K-\

z-'^dz

nR l-q

\-z

R-1

and

if

as

>

0,

i? ->

00

(5.25)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

156

On C

<:;

1-r

l-z|

andif^<l,

z-Uz

/ 1-z

as r ->

1-r

(5.26)

Cr

On
/',

the upper side of the branch cut where z

the integral takes the

= x

e'"

and x

varies

from R to

form
r

+
1

dx

{521)

and on the lower

/-

side of the

branch cut where z

= x

e''""

and x

varies

from

to R, the integral takes the form,

--dx
1+x

(5.28)

Putting these four parts of the contour integral together and equating

them

to 2ni (residue at z

gives

1)

00

(^e'^--e-'n
J

where the

limits r->0

and

R^oo

x~^ dx
-= Ini
1

(5.29)

+^

have been taken.

Using equation

(4.58)

to introduce the trigonometrical function gives the result,

rx-^^_^
1

(5.30)

+x

sin ^71

Putting this value for the integral into equation (5.24) gives

B{\-q,q)

^^^

(5.31)

sm^Ti

<^<

for
5.4.2.

Relationship between Beta

Make

the substitution

and

x^

Gamma

Functions

the definition of the

gamma

function

(5.12)
00

T{p)

.-.

Define a second

gamma

sider the product of the

2^x^'-^e-''"dx

function using a similar

two

gamma
no

T{p)Y{q)

dummy

(5.32)

variable

y and con-

functions.

00

= A^ ^ x^'-'

y^^-' e-^'^'^'^Uxdy

(5.33)

5.

FUNCTIONS AND DEFINITE INTEGRALS

157

This double integral can be evaluated as in Section 5.2 by changing to polar

coordinates using equations (5.5). The transformation can be justified as
before and equation (5.33) becomes
nil

T{p)T{q)

= 2^

CO

cos^P-' Osin''^-' ed9.2 j e-'" r^P"-^^-^ dr

(5.34)

first integral with equation (5.22), and the second integral

with equation (5.32) shows that

Comparing the

r(p)nq) = B(p,q)T(p + q)
and

for all positive values of p

In the special case

p+q =

(5.35)

q.
1,

using equations (5.13) and (5.31) proves

the useful formula

r(^)r(l -^)

which

is

valid for

<^ <

7r/sin

qn

(5.36)

Examination of Examples 1 and 2 in Section 5.3

is vaUd for all values of ^, although it is of no

1.

indicates that equation (5.36)

use for integer

q.

Other Tabulated Functions which are Defined by

5.5.

Integrals
There are many other functions arising in engineering which cannot be
integrated analytically in terms of elementary functions.

Since the values of

have been tabulated, much numerical work can be avoided
if the integral to be evaluated can be altered to a form that is tabulated.
This section contains primarily the definitions and terminology of some of
these functions, and a description of their general behaviour. Their properties are neither so well known nor so important as those of the earlier
functions in this chapter, nevertheless, a knowledge of the definitions is
valuable particularly when using the Laplace Transform.

many

5.5.1.

integrals

The

This

Elliptic Integral

of the First Kind

usually defined by

is

d^
V(l-P~sin2a>)
where k

is termed the "modulus" and (j) the "amplitude".

The integral
occurs in the study of periodic motions when the amplitude is not small
and the motion is not therefore simple harmonic. Hence the terminology

k and
k only and
for

(j).

is

The complete

elliptic integral

of the

first

kind

is

a function of

defined by

F(k,in)

K(k)

(5.38)

MATHEMATICAL METHODS

158
5.5.2.

The

This

is

Elliptic Integral

CHEMICAL ENGINEERING

IN

of the Second Kind

defined in a similar

manner by
<t>

E{k,(t))

= ^^\-k^s\n^^d(b

(5.39)

the amphtude. This integral measures

where again k is the modulus and
is the complelength round the arc of an ellipse from the minor axis, where
ment of the normal parametric variable of an ellipse, and k is its eccentricity.
The above connection with the geometry of the ellipse gives rise to the name
"elliptic integral". The complete elliptic integral of the second kind is again

defined as

E(k,^n)
5.5.3.

E(k)

(5.40

The Sine, Cosine, and Exponential Integrals

X

Si(x)

= j^-^dt

(5.41)

Ci(x)^

C cos t
-- -dt

(5.42)

00

(-dt

Ei(x)=

(5.43)

00

00

-Ei(-x)=
Of this group of functions, only

/t'

dt

(5.44)

the sine integral remains finite at the origin.

All of the other functions are infinite at the origin; also, Ei(x)

x->oo, and Ei( x)

is

infinite as

x-> oo.
Although the sine integral cannot be evaluated in terms of elementary
functions, it is possible to evaluate 5/(oo). The integrand of equation (5.41)
is an even function, so that
is

infinite as

S/(oo)

and integrating a

1
_
=
~2^

suitable function

r
I

sm t
'-^^dt

(5.45)

~r

round a closed contour

in the

complex

plane, the integral can be evaluated.

An alternative and simpler method is to introduce a suitable parameter

a and evaluate the integral as follows. Define
00

I((x)
[cc)=

dt

(5.46)

5.

and

FUNCTIONS AND DEFINITE INTEGRALS

159

differentiate equation (5.46) with respect to a thus,

^=

da

e-" sin tdt

(5.47)

Equation (5.47) can be integrated by parts twice, or a complex method can

be adopted.

-]mle-''e''dt

dec

lm[
e-^-H"
icc-i
Jo

^-^"[^l^^'c^^l]
dl

On

da

a'

(5.48)

+l

integration, this yields

da

r
/

.-.

where

result at

""J 0^1
= A-tan-^a

is an arbitrary constant which can be evaluated from a known

any value of a. When a = oo, equation (5.46) shows I (a) = 0.

= ^-^71

,-.

Substituting this value of

into equation (5.49) gives

/(a)

The

= i7r-tan"^a

infinite sine integral (5.45) is

therefore

(5.49)

from equation

(5.50)

given by equation (5.46) with a

0,

(5.50),

m=

in
oo

^.^

and

Si(oo)

5.6.

fsinf

dt

= in

(5.51

Evaluation of Definite Integrals

The method used above to evaluate 5/(00) is of wider apphcation, parsome of the transform methods given in Chapter 8. It is often

ticularly in

unnecessary to introduce a parameter such as a above, because a suitable

parameter is already present within the integral. Consider the following
examples of integrals of this type.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

160

Example

1.

Evaluate
00
^

/(a)

As

^ dx

C sin^

ax

standard functions earlier in this chapter, x is a dummy variable

is a function of a only. Differentiate equation I with respect to a.

in the

so that /(a)

dl

C 2 sin

d(x

ax cos ax

J
uu

sm2ax

dx

II

X
But with a simple change of variable
given by equation (5.51) as

(/

2ax), the integral in equation II

is

5/(oo)

^n

=*'
Integration of equation III gives

where

is

known

the

gives, /(a)

a constant of integration.
result for

0.

Hence the

2.

and

is

a.

can be evaluated by using

if a = 0, equation I

Thus

j!!^^, = ^,

IV

Evaluate
\

/(a)

This

before,

final result,

;()

Example

As

any particular value of

much more

difficult

r e -x2sin2ax
""
dx
\

example, however, following the same method

differentiating equation I with respect to a.

.'.

dl

=
dec

The primary

"r
\

2e

_ ^^2

cos 2ax dx

II

effect of this differentiation has been to remove the term x

from the denominator of the integrand. A further differentiation will introduce a factor x into the numerator and facilitate an integration by parts,

thus.

5.

-^

FUNCTIONS AND DEFINITE INTEGRALS

dec'

161

4xe~^^ sin2(xxdx

III

2 e~*^ sin lax

4a

e"""^

IV

cos lax dx\

On the right hand side of equation IV, the first term vanishes at both Hmits,
and the second term is a constant multiple of the right hand side of equation
II.

Therefore,

d'l

dl

The dependent variable does not occur explicitly

method of Section 2.4. 1 can be used. Thus, put

in equation

and the

^^

VT
VI

da
separate the variables, and obtain the solution for

p
where
case a

=V

-da

= ^ne-''
^
da

71

a.

Viz.

can be evaluated for the special

erf 00

yjn

VIII

and using equation

Integrating again with respect to a,

terms of

VII

in

Ae-''"

A is the constant of integration which

= 0, when equation II gives

.-.

i7cerfa + B

(5.1),

IX

where B, the constant of integration, can be evaluated by using equation

when a =

0.

Thus,
a

if

The evaluation of

0, I

0;

.-.

^=

the integral in equation

0.

I is

thus

00

r,

/(a)

_^2sin2ax

r
e

dx

inQTta

The above method consists essentially of finding a differential equation V

which the given integral satisfies, solving the differential equation and hence
obtaining an alternative form for the solution. This process will inevitably
introduce arbitrary constants of integration and these must be evaluated by
referring to the original integral which is the desired particular solution of
the differential equation.
In the first example, the integral

was shown

to be a solution of the simple


MATHEMATICAL METHODS

162

IN

CHEMICAL ENGINEERING

first order differential equation III; and in the previous section, Si(oo) was
shown to be a solution of equation (5.48). In all three cases therefore, the
method is the same but the details are different. It is not always possible to

example the gamma function defined

any differential equation with rational
Nevertheless, the above method and the method of contour
coefficients.
integration are two advanced techniques which will successfully evaluate
find a suitable differential equation; for

by equation

many

(5.12) does not satisfy

definite integrals.

Tables of the functions discussed in

this

chapter are readily available

in:

Dwight, H. B. ^'Mathematical Tables" (2nd Ed.), Dover (1958).

Jahnke, E. and Emde, F. "Tables of Functions" (4th Ed.), Dover (1945).
"Standard Mathematical Tables" (11th Ed.), Chemical Rubber Publishing

Company

An

(1957).

index of published tables has been compiled by Fletcher, Miller and

Rosenhead.t
t Fletcher, A.,

Tables," Scientific

Miller,

J.

Computing

C. P. and Rosenhead,
Services,

London

(1946).

L.

"An

Index of Mathematical

Chapter 6

THE LAPLACE TRANSFORMATION

6.1.

The Laplace transformation

is

Introduction

the

name

given to a particular "operational

Essentially, an operational
method" of solving differential equations.
method is a technique whereby an ordinary differential equation is converted into an equivalent algebraic form w^hich can be solved by the laws
of elementary algebra. The method will also convert a partial differential

equation into an equivalent easily solvable ordinary differential equation.

There are many such operational methods but the Laplace transformation
is a particularly useful mathematical tool for solving engineering problems
because the boundary conditions are introduced into the equation prior to
its

solution.

The apphcation of operational methods for the solution of engineering

problems was first made by Heaviside (1850-1925), and although his methods
were largely intuitive they were successful. This stimulated the interest of
other mathematicians who developed the theoretical basis of Heaviside's
methods and extended them into what is known as the Laplace transformation. The essential features of the method and its apphcation to the solution of chemical engineering problems involving ordinary differential equations are presented in the following sections, although the methods of
Chapter 2 are normally used for solving these problems. The most useful
appHcations of the Laplace transformation, to partial differential equations
and difference-differential equations, will be given in Chapters 8 and 9
respectively.
6.2.

The Laplace Transform

continuous function of an independent variable t for all values

of t greater than zero, then the integral with respect to / of the product of
/(O with e"^' between the limits and oo is defined as the Laplace transform
of/(0. I.e.
If /(O is

a'

00

Je-/(Orff

if[/(0]

(6.1)

For the present, the only conditions that will be placed on the integral in
equation (6.1) are that the parameter s must be large enough to make the
integral convergent at the upper hmit, and t must be positive.
The Laplace transform of a function can be denoted by placing "^"
before the function as shown in equation (6.1). To be exact, this signifies that
163

MATHEMATICAL METHODS

164

CHEMICAL ENGINEERING

IN

the operation of the Laplace transformation has been carried out on the funcand the modern practice is to write J{s) for the transform of the

tion f(t)

function to imply the connection between f{t) and

its

transform

J{s).

^[/(0]=/(5)

To

will

now

(6.2)

method, the Laplace transforms of

Thus if/(0 = 1, then

familiarize the reader with the

some simple functions

be derived.

=
=

From
stant

(6.3)

JO
or f{s)

Thus

lis.

equation (6.3)

can be seen that the Laplace transform of a con-

it

is

^[K] = K^li] =

KIs

(6.4)

since the constant can be inside or outside the integral in equation (6.3).
This simple result follows from the fact that the Laplace transformation is

a linear operation.

To

illustrate the

required.

procedure further,

let

the Laplace transform of

/"

be

Thus
00

^\_t"']

The

=j

t"e-''dt

(6.5)

integration variable in equation (6.5) can be changed by putting

.'.

St

dz

sdt
00

^lt"]

.-.

s-"-^

z"e-'dz

(6.6)

Comparing equation (6.6) with the definition of the

equation (5.12), gives
^lf} = r{n + l)ls"-'^
If

is

a positive integer, the property of the

gamma

gamma

function,

(6.7)

function given by

equation (5.15) shows that

^\_t"]
Finally, let/(0

e^;

(6.8)

and the Laplace transform of this function

00

J(s)= (e"'e-''dt=
i

n\ls"^^

00

( e-^''''^'dt

is

=
s

(6.9)

way the Laplace transforms of other simple functions can be

Some simple transforms are shown in Table 6.1, and the reader

In the same

obtained.

encouraged to verify each of the examples given. A more complete table

of Laplace transforms is given at the end of the book.

is

6.

THE LAPLACE TRANSFORMATION

Table

6.1.

165

Elementary Transforms

fit)

f(s)

K/s
nl

/"
-^^:j:Y

(n

an integer)

not an integer)

p/\
t"

-^p^
5*^
1

gat

a
sin

(xt

cos at

V7
\/ S^
1

V7

6.2.1.

\J

~S

Laplace Transforms of Derivatives

The method given in Section 6.2 for determining the Laplace transform
of a function can be appUed to the differential coefficient of any continuous
function. Thus let the derivative of a function be df(t)/dt, then the Laplace
transform of this derivative

is

j~e-^-dt =

[/(r)e-]o"

and since the Laplace transform of/(0

is

sjme-^t

f(s),

(6.10)

equation (6.10) becomes

^\^] = sKs)-m

(6.11)

Similarly, the second derivative of a function d^f(t)/dt^ can be converted to

its

Laplace transform by the same procedure to give

where /(O) and /'(O) are the values of the function and
the independent variable

is

its

derivative

when

zero.

Finally the integration processes described above can be extended to

finding the Laplace transform of an nth order derivative of a function.
result

is

The

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

166

^ [^]

s"J(s)-ls'^-'m + s"-'f'{0)...+sf"-'\0)+f"-'\0)-]

(6.13)

where /'(O), /^(O)

P"~^\0) are the values of the first, second, up to the
{nl)ih derivative of the function when the independent variable is zero.

The Laplace transforms of the different order derivatives presented

above are only vahd for continuous functions. If the function undergoes a
step-change in the range in which the independent variable is being considered, each of the equations (6.11) to (6.13) have to be modified. This will
be considered later, in Section 6.5.
Equations (6.11) to (6.13) introduce the fundamental operational proThese are the properties that introperties of the Laplace transformation.
duce the boundary conditions /(O), /'(O), etc. into the problem before
solution of the equation.
At the same time, the differential equation is
reduced to an algebraic equation in terms of the operator s. However, it
should be noted that the operation described is only appHcable to "initial
value" problems. That is, the value of the function and its derivatives must
be known when the independent variable is zero. This becomes apparent
by inspection of the transforms of the derivatives.
Finally, equation (6.13) can be utilized to find the Laplace transform of
a function without carrying out the integration procedure, and to illustrate
the method, the Laplace transform of sin at will be evaluated.
Let

f(t)

=
=

sincct

Then

f'{t)

and

f"{t)= -a^sinar

Also

/(O

and

=o

/'(O|t = o

^[r(0] =

Since
.-.

=
=

acosar

sinar ,=o

acosar ,=o

s^/(s)-s/(0)-/'(0)

- Se[(x^ sin at\ =

s^ ^[sin

<xt\

(6.12)

if[siaO=^3^^.

or

Equation (6.14) gives the expression for the transform that

Table 6.1.
6.2.2.

(6.14)
is

reported in

The Shifting Theorem

This theorem is most valuable in obtaining the transform of an exponential type of function and also in carrying out the inverse transformation
which will be described later. The shifting theorem may be enunciated as
follows

Laplace transform of f{t) is J{s), then the Laplace transform of

f{s + (x), where a is a finite constant. That is, the multiplication
of/(0 by e"*- in effect shifts the origin of s to (s + a).
If the

e~'^ f{t)

is

THE LAPLACE TRANSFORMATION

6.

167

This theorem can be proved by carrying out the integration procedure

it will be more illuminating to verify it by
on
applying the theorem to the following example.
the general function /(t), but

Example. Determine the Laplace transform of

In Table 6.1, if/(/)

cos pt, then/(5)

s/is^'

e''"'

cos ^t.

+ p^).

Therefore, by the shifting theorem

which
Thus

is

the desired result.

However,

this will
CO

00

be verified by integration.

e-''cospte-''dt

^je-''e'^'e-''dt

II

00

^-(a + s-ifi)t

"100

^[l/(a + 5-ii5)]

Using the division rule for complex numbers, and discarding the imaginary
part confirms the result given in equation I.
6.3.

The Inverse Transformation

In order that the operational method of the Laplace transformation can

be exploited for the solution of engineering problems, it must be convenient
to convert the transform back to a function of the independent variable.
That is, it is necessary to perform the operation
[

^-'im=m

(6.15)

On occasion this can be

done by inspection, by looking up the function corresponding to the transform in a table of Laplace transforms, such as the one reproduced at the end

in order to complete the solution of the problem.

of this book.
illustrated

Some manipulation of

the constants

is

often necessary as

by the following example.

Example

1.

Find the inverse transform of

5/(5^

25 + 5).

Completing

the square of the denominator in an attempt to find factors gives

Applying the shifting theorem,

MATHEMATICAL METHODS

168

But the right hand side of equation

and 16 at the end of the book.

the inverse transform

6.3.1.

CHEMICAL ENGINEERING
the

II is

sum

of transforms

+ ism2t
:.
f(t) = e'(cos2t + ism2t)
of f(s) = s/(s^ 2s + 5).

.-.

is

IN

e~'f(t)

number

15

oos2t

III

Inverting the Transform Using Partial Fractions

The Laplace transform of a function is an expression involving the parameter s which may be very complex. When the expression is of the form

and 0(5) are polynomials in s, and (/)(5) is of higher degree than

can be expanded into its partial fractions. If the complex roots of
0(5) are also resolved, then all terms in the resolution must be of the type
\l(s ay with n an integer and a a real or complex constant. The inversion
of f{s) is then simply a case of repeated appHcation of transform number

where

6{s)

6(s),f{s)

10 in the tables, viz.

For use with

is well worth remembering.

transform number 3 can be memorized and the shifting
theorem applied to it. The following example illustrates this technique.

partial fractions, this transform pair

Alternatively,

Example

2.

Expanding by

Determine /(r)

if

/W =

l/is

+ a)(s-\-by.

partial fractions gives

1
^ A
+ a)(s + by ~ s + a

B
+ b~^

C
+ b)^
A(s + by + B(s + a)(s + b)-\- C(s + a) = l
putting s = b gives
C = ll{a-b)
putting s = a gives
A = ll(a-by

^^^

(s

"^

(s

.-.

Evaluating

Evaluating

Evaluating

by

A by

by equating

III

IV

coefficients of s^ in equation II gives

B=-ll(a-by
,.

II

(,-6)7(s)

+ 7^^2
^-4t
+
+
+

fl

(s

b)

VI

6.

Inverting the right

number

THE LAPLACE TRANSFORMATION

hand

side of equation

10 (or special cases 8

and

(a-b)V(O =

m=^

169

IV term by term using transform

9) gives

e-'''-e-'' + (a-b)te-^'

VII

{s

+ a)(s + by

VIII

{a-bf

The expansion of the transform by the use of partial fractions is quickly

and conveniently accomphshed when the denominator contains only simple
roots; but when the denominator contains a repeated root of high order,
a lengthy, time-consuming process. Howby the theory of residues is much more
This
straightforward and is recommended under such circumstances.
method is discussed in Section 6.8 where it is shown to be the most powerful
method of obtaining the inverse transform.
resolution of the coefficients

is

ever, the evaluation of coefficients

6.3.2.

Application to the Solution of Ordinary Differential Equations

The primary appHcation of the Laplace transformation is the provision

of a powerful operational method for removing derivatives from an equation. Therefore, before proceeding further with the properties of the transformation, the operational characteristics will be appUed to the solution of
ordinary differential equations in order to illustrate the technique.
The
method depends on the fact that an ordinary differential equation can be
converted into an equivalent algebraic equation called the 'subsidiary
equation" in which the Laplace transform operator s is the independent
variable. The procedure is illustrated by the following example.
*

Example

3.

Solve -ry

Using equation

4>^

3,

where y{0)

/(O)

1.

(6.12) to transform the derivative gives

and transforming the constant using

.^[3]

tables gives

3/5

Substituting these transforms into the differential equation gives the subsidiary equation

s'Ks)-5-l + 4Ks) =

III

3/s

Solving the algebraic equation III for y(s) gives

v(s)

Inverting equation

-^ +

IV

IV term by term, using transforms number

16, 15,

and

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

170

19 respectively

from the table

y

which

will

back of the book gives

at the

coslt

isinlt

isin2f

|(1

cos

20

simpHfy to

6.4.

icos2r

VI

Properties of the Laplace Transformation

Before further appHcation of the Laplace transformation to the solution

of engineering problems is made, some of its most important properties will
be discussed, so that this operational method can be applied to the solution
of more complicated problems.
6.4.1.

Differentiation of the Transform with Respect to the Operator s

If /(O is a continuous function of the independent variable /, and its

Laplace transform is f(s), then differentiation of the transform with respect
to s corresponds to multiplying the function by t.
This property can be proved as follows.

Ks)

by

definition,

and

jf(t)e-^^dt

(6.17)

differentiating equation (6.17) with respect to s gives

Yl(s)-]==l-tf(t)e-^^dt
ds

But the integral in equation (6.18)

definition (6.1).
6.4.2.

(6.18)

I
is

Hence the property

the Laplace transform of //(/) by the

is

proved.

The Laplace Transform of the Integral of a Function

lff(t) is an integrable function of the independent variable / and its Laplace

transform is f(s), then division of f(s) by s corresponds to integration of /(O
between the limits
and t.
The above property can be verified most easily by showing that the
converse is true. Thus

^ jmdt =j

jf(t)dt

The

right

hand

side

can be integrated by parts by

f

and then

jf(t)dt

du=f(t)dt
dv

so that

e-''dt

Lo

v=

e-''dt

-e-^'ls

letting

(6.19)

THE LAPLACE TRANSFORMATION

6.

171

Substitution of these parts into equation (6.19) gives

/
LO

\mdt

j{i)dt

+^J/(Oe-*'Jr

(6.20)

The

first term is zero at the upper limit because of the exponential term, and
zero at the lower limit because of the range of integration. Therefore

-Ks)

dx

(6.21)

.0

The results of Sections 6.4.1 and 6.4.2 are of great value in the solution of
engineering problems. They find extensive use in the solution of differential
equations with variable coefficients and the method of solution will be
by the next example.
Example. Using the data given below obtain a preliminary estimate of
the diameter of the tubes to be installed in a fixed bed catalytic reactor which
is to be used for the synthesis of vinyl chloride from acetylene and hydrogen chloride.
The tubes are to contain the mercuric chloride catalyst
deposited on 0-1 inch particles of carbon and the heat of the reaction is to
be employed to generate steam at 250F for the remainder of the process.
To do this, the temperature of the inside surface of the tubes should be
illustrated

constant at 300F.
Effective thermal conductivity of

Heat of reaction

at

(fc^)

bed temperature (AH)

Bulk density of bed

The

bed

(p)

Btu
4-0

=
=

It

^ oi^/rx

r/it

46,200 Btu/lb molef

18 Ib/ft^

a function of temperature, concentration and the

various adsorption coefficients J, but for the preliminary estimate assume
that the rate of reaction can be expressed as
rate of reaction

is

To

(l+AT) lb moles/h

lb catalyst.

where Kq = 0-12, ^ = 0-024, and Tis the temperature in degrees Fahrenheit

above a datum of 200F.
The maximum allowable catalyst temperature to ensure a satisfactory
life is 485F (that is, T = 285).
Solution

Consider Fig.

6.1,

which

illustrates the

following additional symbols

describing the process.

G =
R

Flow
Tube

rate in Ib/h ft^ of reactor section.

radius in

X = Radial

ft.

coordinate.

from heats of formation, etc. (negative sign omitted for convenience).

t Wesselhoft, R. D., Woods, J. M. and Smith, J. M. AJ.Ch.EJ. 5, 361 (1959).

t Calculated

MATHEMATICAL METHODS

172
z

C-

CHEMICAL ENGINEERING

IN

= Axial coordinate from inlet.

= Heat capacity of the gas in Btu/lb

F.

\aP

6z

r
G

Fig. 6.1.

Fixed bed reactor

Taking a heat balance over the element

illustrated in Fig. 6.1 gives

= Inx kr-Sz + Inxdx GCi,T + InxSxSz pAHr

Input

dx

Output =

2nx krir-oz

-^

ox

^
ox

dT

2nx /c - Sz

dx

ox

+ 2nxdxGC,
Accumulation

0,

because the process

is

(-s^-)

at steady state.

Input Output = Accumulation

But,

II

(1.25)

and dividing throughout by In k^ xdxdz gives

d^T

dx-

Equation

III is

a well

IdT
dx

known

=0

kr

dz

III

kr

partial differential equation expressing the

relationship between the temperature and the dimensions of the reactor.

If for the preliminary estimate, it is assumed that the temperature of the bed

reaches a maximum at all radii, at the same distance z from the reactor
entrance; then at this radial section of the bed dT/dz will be zero even
though the actual temperatures will vary with x. The temperature will be a

maximum
axis

dT/dx

than zero.

and this must not exceed 485F. Hence at the tube

be zero also, but will have a finite value at all radii greater
Hence the partial differential equation III becomes an ordinary

at the tube axis

will

6.

differential

THE LAPLACE TRANSFORMATION

equation at this particular value of

d^T

IdT

dx^

X dx

pAHvo

z.

That

+ AT) =

(1

173
is

IV

Equation IV can be easily converted to a Bessel equation of zero order

and solved by the method given in Chapter 3, but on this occasion it will be
solved using the Laplace transformation. The equation can be rewritten

Using equations

d^T

dT

dx^

dx

(6.12)

and

pAHro

(l+^r);c

(6.18)

j^r

= --[s^r(5)-5r(0)-r'(0)]

if

dx^

= -s'r'(s)+2sT(s)-r(0)
and using equation

VI

(6.11)

dT
^ rdT
[d^

VII

sT(s)-T(0)

Also,

pAHvo
VIII

and using equation

k^s^

^E

(6.18) again,

^[pAHroA

xT

Substituting the transforms VI, VII, VIII,

in equation

and simpHfying

ds

IX

and IX for the equivalent terms

gives

~
dT

Q = pAH Vq Ajk^,

/Cf

T= -

s\s' + Q)
+Q
=
and P
pAH VqIU^. Equation X
s^

where
can be solved
by using an integrating factor as shown in Section 2.3.4 giving the result

P_

XI

This can be inverted using the tables, numbers 55 and

;.

where

is

T=

KJo(xs/q)

1.

-\

XII

the arbitrary constant of integration which can be evaluated as

follows.

When

0,

^(xVQ) =1,
.-.

285

and

X-1M

T=

285
XIII

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

174

A=
lM =

but
.-.

Q=

Now
.-.

Ve =

0-024
41-7,

and

K=

326-7

18 X 46 200 X 0-12 X 0-024

599

4-0
24-5

and equation XII becomes

At the boundary,

x =

R,
.-.

or

Use of a

T = 326-7Jo(24-5x)-41-7
T = 100;
100 = 326-7yo(24-5K)-41-7
Jo(24-5R) = 0-4337

XIV

table of Bessel functions gives

=
R=

24-5K
or

1-64

0-80 inches

diameter of the tubes that would be installed in the reactor

than 1 -60 in, as for this diameter the cooling rate would be
just sufficient to prevent the temperature of the catalyst at the centre of the
tube exceeding 485F. Hence tubes of diameter 1-50 in would be recom-

For

safety, the

should be

less

mended.

The Step Functions

6.5.

and equipment to undergo sudden changes in their operating conditions; for example the steam
supply could suddenly change, or a vacuum pump or compressor could
fail.
It is desirable to be able to predict what effects these sudden changes
will have on the process and quality of the product; and this can be accomplished by means of the step functions which will be considered below.
It is

6.5.1.

possible for chemical engineering processes

The Unit Step Function

This

is

illustrated in Fig. 6.2,

which shows that the value of the function

Sb(t)

<

b
Fig. 6.2.

The

unit step function

zero for values of the independent variable less than b, and unity for
values of the independent variable greater than b. The Laplace transform
of the unit step function can be obtained in the usual way by integration.

is

6.

THE LAPLACE TRANSFORMATION

175

Thus
^lS,{t)]

jOe-^'dt

jl.e-''dt

e-''ls

(6.22)

If the value of b is zero, the transform would reduce to l/s which is the
transform of unity. Thus multiplication of the transform of unity by e~^^
to ^ and unity for
indicates that the function is zero over the interval
values of / greater than b. This concept can be extended to any other functions which are zero for specific intervals or which undergo sudden changes
as illustrated by the unit impulse function below.
6.5.2.

The Unit Impulse Function

The

unit step function

was characterized by having zero value for all

and a value of unity for

values of the independent variable less than b,

If the value of this function is suddenly

t between b and infinity.
reduced to zero when the value of the independent variable is (b + c) the
function will appear as an impulse when plotted as shown in Fig. 6.3.

values of

i(t-b)
1
i

'1

1
L^
'b'
The

Fig. 6.3.

"c

unit impulse function

Furthermore, if the height of the impulse between b and (b + c) is 1/c, the

area beneath the curve will be unity. As c->0, the area will remain constant
at unity,

and

this limit defines the unit

lit-b)

impulse function as

lim[S,(0-S,^,(0]/c

(6.23)

c-*0

Taking the Laplace transform of both

^[I(t

b)']

lim

sides,

using equation (6.22) gives

(6.24)

\-

cs

By

L'Hopital's rule, Section 3.3.8, this hmit becomes

^U(t-b)']

e-''

(6.25)

Frequently, the unit impulse function is called the "Dirac delta function'*
and is written S(t b). It expresses the rate of change of the unit step
function.
6.5.3.

The Staircase Function

is shown in Fig. 6.4 where it can be seen that the function is formed
by the successive addition of unit step functions at 0, b, 2b, 3b, .... etc.

This

MATHEMATICAL METHODS

176

CHEMICAL ENGINEERING

IN

S(b.t)

32
1

2b

1b

Fig. 6.4.

The

3b

staircase function

Hence the Laplace transform of the staircase function

together the transforms of each step function. Thus
2bs

bs

The
be

infinite series in

summed

(l

will

be

...

nbs

+ e-^' + e-^^' +

equation (6.26)

is

+ ...)ls

(6.26)

a geometrical progression which can

to give the result

^[5(^0] =
Use

obtained by adding

nbs

+
=

is

made of

(6.27)

^^^::^

these functions later in the text but the use of the unit

step function will be illustrated

by the following example.

Example. A control valve of the type shown in Fig. 6.5 is actuated by

ranging from 3 to 15 Ib/in^ gauge operating on a 16-in diameter
diaphragm whose effective area is equivalent to 100 square inches. The
air pressure

Air

^^

Spring

Diaphragm

Becking

Valve stem

plote

Fig. 6.5.

Control valve

dead weight of the moving parts of the valve,' allowing for the
is estimated to be 300 lb weight, the stiffness of
the spring is 600 lb/in and the damping constant is estimated to be 17-0 lb

effective

friction in the gland, etc.,

If the total

sec/in.

valve

if

lift

of the valve

is

2-0

in,

predict the response of the

the controlling air pressure suddenly changes

12 Ib/in^ gauge.

from 6

Ib/in^

to

6.

THE LAPLACE TRANSFORMATION

177

Solution

force balance

Input force

= AP A

on the valve

is

So(t)

Output force

,2

= m -rj.

movable parts

(a) to

overcome

(b) to

overcome resistance of spring

kx.

to

overcome damping resistance

^~T'

(c)

inertia of

dx

Therefore at equilibrium

dx

d^x

m -^ +

di^^'dt

Let a

c/2mco where

Then equation

co

/ex

= APASo(t)

s/k/m

becomes

^ + 2aco + co^x =
dt
The Laplace transform of equation
(s^

sx(0)

ms

In order to ease the calculations,

air pressure

let

x(0)

+ 2acox(0) + x'(0)

the origin of

=
=

of 6 Ib/in^ corresponds to x
.-.

11

II is

APA

+ loiws + a)^)x =

So(0

dt

x'(0)

111

x be changed so that an

0,

and therefore

APA
ms(s^ + 2acos + co^)
The

inverse transform of equation

APA

where

tan(/)

X can now be evaluated

problem. Thus

APAlk =

m=

"

a/Vl

as a function of

partial fractions as
1

a^

VI

by using the data given in the

300 lb dead weight

V
"

IV can be obtained by

e~*"^

IV

2mco

"

300

32x12

0-781 lb sec^/in

0-'

2 X 0-781 X 27-72

"

'^^^

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

178

a^

taiKp

0-920
0-393

0-427

0-920
.*.

(j)

0-404 radians

V gives Table 6.2 and hence Fig. 6.6.

measured from the top of the valve, Fig. 6.6 has been drawn to
show valve opening X.
Substituting these values into equation

Because x

is

Table

6.2.

Evaluation of Equation

UJt

0-92

2
3

4
5

6
7
8

9
10

0-9196
0-8698

0870

0-7337
0-4953
0-3343
0-2257
0-1523
0-1028
0-0694
0-0469
0-0316
0-0214

01346
-0-7075
-0-9910
-0-4939

1-237

0-3929
0-9697
0-7819

0-960
0-933
0-963

-0-0220
-0-5883

1-013

0-6-

(Ot

Fig. 6.6.

0-000
0-362
0-933

Control valve response

1-224
1-075

1007

yx

1-500
1-138

0-567
0-263
0-276
0-425
0-540
0-567
0-537
0-493
0-487

6.

THE LAPLACE TRANSFORMATION

179

Therefore, the response of the valve is lOjll'll = 0-36 sec. That is, in
0-36 sec the position of the valve alters by one inch and has stabilized at its

new

position.
6.6.

It

Convolution

frequently happens in solving problems by the Laplace transformation

that the final transform of the equation is a product of two easily identifiable
transforms, but these are difficult to resolve in the form of a summation.
When this arises, the inverse transformation can be accomplished by the

method of "convolution".
Space does not permit the derivation of the convolution integral, and
its use will be explained.
Any reader

therefore only the mechanics of

interested in the derivation

Thus

if

factors

advised to consult

is

the transform of the equation

g{s)

and

h{s)

whose inverse

Thomson!

or Churchill^.

composed of the
transforms are recognizable from

is

and

J{s)

this is

Laplace transform tables, the inverse transform of J{s)

is

obtained as

follows.

Let the inverse transform of g{s) be

be h{t).

g{t),

and the inverse transform of

h{s)

Then

since

r/ x

/(s)

-/ n r/ ^
g{s) h(s)

(6.28)

fit)

= ^-\g{s)Ks)\ =jgix)Kt-T)dT

=jh(z)git-x)dr

(6.29)

Equation (6.29) gives the convolution integral of /(r) and imphes its symmetry. The convolution integral is frequently given the shorthand form
f(t)

The use of the convolution

= g(t)h*(t)

(6.30)

an inverse transform

integral to find

will

be

illustrated in the next example.

+ ly

Example. Find the inverse transform of 7^

775.

(s^

Transform number 16

in the tables

m=
then

g(t)

shows that

nis)

^^

h(t)

cost

if

Therefore, by the convolution integral (6.29),

t

f(t)

cos T COS (t x)dx

J

=i
t
X

Thomson, W.

T.
Churchill, R. V.

[cos

COS (2t - 0] dx

"Laplace Transformation." Longmans Green & Co.

"Operational Mathematics." McGraw Hill Book Co.

II

MATHEMATICAL METHODS

180

/(O

IN

CHEMICAL ENGINEERING

= ilrcost + i sin (2t - 0]o

= i(rcosr + isin/ + isinO
= i(rcosr + sinO

III

IV

The answer given by equation IV checks with the answer given

number 23.

directly

in the tables

6.7.

Further Elementary Methods of Inversion

Three further methods for finding the inverse Laplace transform will be
given in this section. These methods will often yield answers more quickly
than either partial fractions or convolution, and they can also be used
when previous methods fail.
6.7.1.

Using the Properties of the Transformation

was shown in Section 6.4.2 that dividing the transform of a function

and
/(/) by s was equivalent to integrating the function between the Hmits
/.
Thus the power of s in the denominator of f{s) can be increased in an
integral manner by successive applications of equation (6.21) as illustrated
in the next example.
When this process is combined with the shifting
theorem (Section 6.2.2), a powerful method of inversion results.
It

Example 1. 'md f(t) if J{s) = \ls\s+a).

Transform number 8 shows that if Jy{s)

\l{s-\-a),

then f^{t)

e~\

Application of equation (6.21) gives

s(s-\-a)

s
t

:.

Mt) =

je-'"dt

1-eII

A further application

of equation (6.21) to this result gives

^n
s

'^'^'s^s + a)

J
1
t

(at

Therefore,

if j(i)

a Jo

+ e-'^'-l)

Jls^{s + a), then

m=

(e-''

+ at-l)la

III

6.

Expansion

6.7.2.

in

THE LAPLACE TRANSFORMATION

181

a Descending Power Series

be noted in the tables that as |5|^cx), then/(5)->0 for every transalthough an apphcation of L'Hopital's rule is often necessary to
show this. The first seven transforms in the table also give an inversion for
any negative power of 5; and hence, if/(5) can be expanded in a series containing only negative powers of s, these first few transforms will determine
a series form for/(0.
An ascending power series will always contain positive powers of the
variable after a certain stage, so that to restrict the series to negative powers,
Inspection of transform
a descending power series of s must be used.
numbers 1 to 7 shows that the termwise inversion will lead to an ascending
power series in t. Referring back to Section 3.3, if this series has a finite
radius of convergence, then the result will only be vaHd for small values of t.
Hence, the initial behaviour of a system can be determined by inverting the
The method of obtaining
first few terms of such a power series expansion.
the series will be illustrated by the following examples.
It will

form

pair,

Example

2.

Determine /(O

The logarithmic expansion

ing powers of

\ff{s)
(3.8)

In [1

appHed

+ (ajsY].

to f{s) gives a series of descend-

s. Viz.

Transforming each term separately using number

{atl_(atl

?[
The

series in

3,

(at)^
(atl

gives

III

equation III can be recognized (equation (3.11)) giving the

result

= -(cosat-l)

f(t)

which confirms transform number 105 in the

Example

3,

Determine /(O

if /(^)

IV

tables.

l/s^(s+a^).

In order to obtain a descending power series, s must be considered to be

must be rearranged in the following manner
before the binomial theorem can be appUed. Thus
large so that the term (s + a^)

a^

/,

o*

a^

a^
S^i
M.M.C.E.

'

s'i

S*i

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

182

Applying transform number 6 to each term of equation

2t^

2~t'^

2't'^

x^Ti

1.3V7r

1.3.5V7r

2^

2^t'^
1.

3.5.7^7:

(20-

_(201
1.3.5

1.3

II gives

III

1.3.5.7

Equation III is likely to be a more useful inversion formula for the given
transform than that shown in the tables as number 41.
Other Series Expansions

6.7.3.

more convenient on occasions

to expand f{s) in descending powers

apply the shifting theorem; and many other expansions are
also possible in terms of a wide variety of functions. The only restriction on
any series expansion is that it should be convergent for sufficiently large
values of s, and the following example illustrates such an expansion in terms
of exponential functions.
It is

of

(s

+ a) and

Example

4.

In a note published by T.

Woodt,

the following transform

requires inversion
/(^)

= s + alpe-

Following the method given in the pubhcation, equation

'

can be rearranged

as
7(s)

=
s

Provided /?e"'^^<

+a

+
s

II
oc

+a

III

can be expanded using the binomial theorem. But for sufficiently large values of i", equation III will be satisfied, therefore the expansion
will be valid. Thus
equation

II

Ks)

E + cy^'
,%{s

Inverting term by term using transform

nas

number 63 and

IV
the shifting theorem

gives

where

6.8.

= t-na

Inversion of the Laplace Transform by

Contour Integration
In the above theory of the Laplace transformation, the various methods
given for the inversion of the transform to regenerate a function of the
t

Wood,

T.

Nature, 191, 589 (1961).

6.

THE LAPLACE TRANSFORMATION

183

independent variable depended on the resolution of the transform into

partial fractions, or modification of the transforms appearing in the tables

by using convolution or the shifting theorem. These methods are usually

adequate for solving problems involving ordinary differential equations,
but often fail when applied to complex transforms resulting from the soluThese complex transforms can be
tion of partial differential equations.
inverted by means of a contour integral, and the following section will explain
this

method

in preparation for

differential equations in

application to the solution of partial

its

Chapter

8.

The basis of the inversion of a transform by contour integration is the

Mellin-Fourier theorem, which may be formally stated by the following
expressions without placing any restrictions on the functions. Thus if
J(s)

= \e-^^mdt
a^ip

then

/(0

= T^lim

(6.31)

( e''J(s)ds
J

2nip-yoo

a-ip

where a is a constant greater than the real part of any singularity in the
transform f(s).
The Mellin-Fourier theorem is frequently referred to simply as the
inversion theorem and it can be deduced directly from Fourier's integral
theorem which is comprehensively dealt with in most university texts in
mathematics. Thus Fourier's integral theorem can be expressed

fit)

where /(O

is

any function of

( e-'''f{t)dtd},

e'''

00

(6.32)

00

periodic or otherwise.

/,

/ cannot be less than zero and therefore

which case equation (6.32) becomes

In applied mathematics,
f(t)

for r<0, in

00

m=^ J

let

00

e'"j e-'"f{i)dtdX

(6.33)

-00

Put

iX

= 5 a,

where a

is

chosen to ensure convergence of the

OCT-

too

first

integral

OO

(6.34)

7(0 =

00

00

~j
a

which

is

00

^'

^e-" [e'7(0] dt ds

(6.35)

- i 00

equivalent to equation (6.31) applied to

e""^

f{t).

72

MATHEMATICAL METHODS

184

IN

CHEMICAL ENGINEERING

Therefore, the MelHn-Fourier integral theorem provides a method of

obtaining the inverse transform directly by integration in the complex plane
along a Hne parallel to the imaginary axis, and at distance a along the positive real axis from the origin. The value of this integral may be obtained as

Consider the contour shown in Fig. 6.7 where the arc of the circle
R encloses all the singularities. The integral around this contour
is the sum of the integrals along the straight path AB from if] to + //i,
and around the arc BCA. The integration path from A to B as p->oo is
known as the "first Bromwich path". Expressed symbolically,
follows.

of radius

e''j\s)ds=

ABCA
If

+ ip

e''j\s)ds+

('

- ip

e''jls)ds

(6.36)

BCA

equation (6.36) is to be of any value, the last integral must vanish as

condition for this will now be established.
On the arc BDCEA, l^l = i? and it will be assumed that \f(s)\ <e for all

R^co. The
points

on the

arc.

e''J{s)ds<&

.-.

Because
right

e^'

hand

between

is

(6.31)

DC

analytic everywhere except at infinity, the integral

side of equation (6.37) will

e'^ds

DC

iR and s

have the same value for

paths

iR.
iR

J
DC

on the

all

lis

e''j(s)ds<e
^

sini^r

(6.38)

iR

Along the arc BD, |/(5)|<, {e'^l^e''^ and

length of the integration path is less than 2a.

it

is

certainly true that the

\e''j(s)ds<8e'''2oc

.-.

(6.39)

BD
Similarly, the integral along the arc

Adding these

EA

can be shown to be

less

than

2a^e''^

results together,

e^7(s) ds

< s U^ ^.r ^

EGA

^ ^-^ J^^
t

(5 4Q)

As R-^co,

/?->oo, but sin Rt^\.

Therefore the term in the brackets in
equation (6.40) remains finite as R^oo. If e-^-0 as i^^oo, then

e''J(s)ds-^0

asR->cx)

(6.41)

BCA
Therefore, provided \f(s)\-^0 as |5|->oo in the left hand side half plane, the
last integral in equation (6.36) can be neglected and this gives
a

/(')

= r^
2ni

f
J

73

e^'Ks)ds

= ~2ni

{ e^-J(s)ds

(6.42)

ABCA

The evaluation of
singularity of /(j).

the integrals in equation (6.42) depends

on the type of

THE LAPLACE TRANSFORMATION

6.

6.8.1.

Inversion

when the

185

Singularities are Poles

In this case, the last integral in equation (6.42) can be evaluated by the
theory of residues given in Section 4.16. Thus

(0

= r^

f e''J{s)ds
2ni J

=Y

f e''Ks)ds
J

2ni

(residues of e''J{s))

(6.43)

ABCA

Bri

25+1
Example

Find the inverse transform of

by contour integration.

+ 1)
at 5 =

5(5^

The given function has three simple poles,

0, /, and
equation (4.110) to evaluate each residue separately gives:
Residue at 5 = 0,
r

Residue at

Using

^,

-2

ls{s-i){s + i)i

II

P-i

6.8.2.

+ l)5 ]

/.

/,

,^i

Residue at

(25

III

-/(e'^-e-'O

IV

i-2i _,
(25+i)(5+o ] ,
e'' =
-e
-2
,__iLs(s-0(s + OJ
lim

f(t)

Po

f(t)

+ Pi + P-i = l-

cos

2 sin

Inversion when the Singularity

is

a Branch Point

was shown that the path of a contour integral cannot

cross a "barrier" erected from a branch point, and the integration path must
In Section 4.14.3

it

consequently be modified to exclude this point. If the Laplace transform

branch point at the origin, and the branch cut follows
the negative real axis, then the integration path of Fig. 6.7 must be modified
as shown in Fig. 6.8. On the assumption that this transform does not contain

f{s) contains such a

First

Bromwich
path

Fig.

6.7.

Contour integration of
equation (6.31)

Fig. 6.8.

Contour integration along

second Bromwich path

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

186

Other singularities, the value of the integral around the closed contour

ABCPQRDA is zero by Cauchy's theorem.

Furthermore, as j?-> X), the value

is zero.
The value of the
of the integral around the arcs BCP and
integral along the first Bromwich path is thus equivalent to the value along
the path RQP which is called the "second Bromwich path", or

RDA

= r^

/(0

jf

271/

e''Ks)ds

= -~

( e^'J(s)ds
J

ZTil

RQP

Example

Br2

Evaluate the inverse transform of l/s^/s+l.

2.

In Section 6.4.2

ponded

(6.44)

was shown that division of a transform by s corresand /. Therefore let

it

to integrating the function betw een the limits

1

dt

L\5+l

The

inversion of equation I can be done directly from the tables, but here it
be inverted by contour integration. Since the function has no poles but
a branch point at 5 =
1, the origin will be moved to the branch point
by putting ). 5+1, and equation (6.44) will be applied. Thus

will

Ini J

dt

Br 2 for

this

function

along the arcs

is

y'/v

as illustrated in Fig. 6.8

is

The

zero.

and the value of the

RQP

integral along

will

now

integral

be evaluated in

three parts as follows.

RQ. The

i.
Integration along
varying from oo to 0.

.*.

X e'""

d?.=

.'.

RQ

line

is

A"(cos

defined by

/.

.y

e""" with

= x

sin n)

III

-dx

VX=x"e-"'^'= -/Vx

/.

IV
00

/^(A-i)f

^dA=
J

C e~^^

\/a

RQ
ii.

Integration around the circle

Let
:.
.-.

.-.

f^dx =

-ie-'

?.

e'^

x-^e-''dx

ie-'

\/x

of radius ^

where

varies

from

nXon.

dA^iSe^'dO

^7 =

(^^r=^dA =

VI
VII

c5^e^''

lim ie-'3^

e'^'''' ^ isine), +

nejQ ^ q

VIII

6.

iii.

THE LAPLACE TRANSFORMATION

QP.

Integration along

value of this integral

By

the

same procedure

187
as in part

(i),

the

is
00

IX

-^e-'^'dx
QP

Adding together equations V, VIII and IX and

substituting into equation II

gives

dm

lie-' r

-^e-'^'dx

dt

The

integral in equation

X can

be evaluated by putting xt

a^.

"
2e-'

df(t)

doL

XI

dt

=
by equation

(5.2).

e-'jyjnt

Integrating gives

XII

m = ^-'
The

last

few stages of

function from Chapter

6.8.3.

Inversion

When
tion (6.44)

er fV7

XIII

.s\/s+L

this calculation involve the properties

of the error

5.

when there are Both Poles and a Branch Point

there are pole singularities in addition to the branch point, equais still

appHcable but the contour Br2 must be modified to exclude

Qd

^>
fQ
Fig. 6.9.

Modification of Brg for poles and a branch point

of the singularities. This is done by drawing a circle around each singuand connecting it to the main contour with a pair of parallel coincident
straight lines. Figure 6.9 shows the case of a branch point at the origin and
all

larity

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

188

two poles which are self-conjugate. Bvi

Fourier integral

is

is

the curve

GFEDC and the Mellin-

evaluated as follows.

By Cauchy's theorem, the integral of e^^ f(s) round the closed curve
The parts of this contour which have been drawn
is zero.
with dotted lines make no contribution to the integral; the sections EC and
GA because of equation (6.41), and the pairs of parallel Hnes because they
are coincident and are traversed in opposite directions. Thus,

ABCDEFGA

je''Ks)ds=

e'^f(s)ds

GFEDC

Bri

D and F are both described in the positive

and make a contribution 2ni (residue at the pole)
towards the integral along Br 2.
Hence, the transform can be inverted by adding 2ni times the sum of

The

circles enclosing the poles at

(anticlockwise) direction

all of the poles of e'^ f(s), to the piecewise integration along

an ordinary branch cut without poles. That is,

the residues at

GEC

as for

fit)

-^ f

e''f{s)

Br2

ds=

e''f(s)

Js

+X

(I'esidues

of

e''f(s) )

(6.45)

GEC

Equation (6.45) summarizes the method of inversion of the Laplace

transform by contour integration. To use it, if f{s) does not contain a
branch point, the integral along GEC will be zero and equation (6.45)
simplifies to equation (6.43), but in general the full equation (6.45) is needed.
The integral along GEC can always be performed in a piecewise manner as
in Example 2 above, and the residues of /(5) e^^ can always be found by the
methods given in Section 4.16. Thus any transform can be inverted provided
the conditions for equation (6.41) are satisfied.
That is, \f(s)\-^0 as
l^l-j'oo. As mentioned at the start of Section 6.7.2 this condition is invariably satisfied, and it is true to say that any transform arising in connection
with a physically reasonable problem will satisfy this condition so that the
solution can be found in principle by the method of residues.

6.9.

Application of the Laplace Transform to

Automatic Control Theory

The study or design of automatic control systems for chemical processes
requires setting up differential equations that will predict the dynamic
behaviour of a piece of equipment, and in particular the response of the
equipment to sudden changes. The example in Section 6.5 predicting the
response of the control valve to a sudden change in air pressure illustrates
how these problems can be solved by means of the Laplace transformation.
However, let it be assumed that a simple automatic control system is
described by the linear differential equation

^J + fif + C>-=/(0

(6.46)

6.

THE LAPLACE TRANSFORMATION

The Laplace transform of

-,^
y^'^

where

y{Qi)

and

this

equation

/(^)

is

(As + B)y(Q) + Ay'iO)

= J^^^Br+C^-

time

0.

y{s)

is

^^-^'^

As^ + Bs + C

y'{0) are the values of the

ture, pressure, at a

189

measured

variable,

i.e.

tempera-

called the "response transform".

The

transform on the right of equation (6.47) is the transform of the steady

state solution after the change, and the last term on the right is the transform
of the transient solution. That is, the first term on the right is characteristic
of the control system and the last term is characteristic of the initial disturbance. Furthermore the denominator of the terms on the right hand side
of equation (6.47) determines the way in which the system reaches its new
first

equihbrium position and therefore

and

in automatic control theory

The

transfer function

is

is

is

the important group

called the "transfer function".

defined as the response of a control system to a

Therefore
disturbance of unit impulse from the steady state condition.
referring to equation (6.47) it will be appreciated that the origin of jf can be
chosen so that y{P) and y'{0) are both zero when r = 0. The transfer function

is

now

characterized by

W) = As^ + Bs + C
which defines

it

^^'^^^

as the ratio of the Laplace transform of the response variable

to the Laplace transform of the disturbing variable.

In addition, since this

equal to the reciprocal of the transform of the differential equation
describing the system, it can be evaluated only for equations that can be
However, this need not be a serious restriction because in
transformed.
many cases an approximate subsidiary equation can be found without undue
simplification; on other occasions it may be necessary to obtain the transfer
function experimentally because of the complexity of the control system.
The transfer function describing a part of, or the complete control
system is usually inserted into a "block diagram". Such diagrams are used
extensively in automatic control theory to depict the action of the variables
involved. They are similar to energy flow diagrams but the arrowed fines to
and from the different blocks indicate input and output signals.
typical
block diagram of an automatic control system is illustrated in Fig. 6.10b.
The corresponding process diagram depicting an electrical heating coil
installed in a stirred tank through which a liquid to be heated is passed, is
ratio

is

shown

in Fig. 6.10a.

In Fig. 6.10b the circle represents an algebraic operation and the two
plus signs imply that the output signal /(O is obtained by summing the outputs from the two transfer functions. Each transfer function is enclosed in

a rectangle, and the transfer function together with its block is called a
dynamic function because it indicates a rate of change. In the particular
example it indicates a rate of heat input.

MATHEMATICAL METHODS

190

IN

CHEMICAL ENGINEERING

Electrical

heater

Liquid

In

Liquid

out

.-

As +

JL

B(As +

The only

rules to be

f(t)

(b)

(a)

Fig. 6.10.

1)

Block diagram of

remembered

electrical heating process

in the construction of block

diagrams

are
(a)

Only one hne may enter, and only one line may leave a block repredynamic function.
Only two lines may enter and one line leave a circle representing an

senting a
(b)

algebraic operation.
(c) If

a control system

equations,

is

described by one or more linear differential

and algebraic operations can be re-

the dynamic functions

arranged without altering the controlability of the process.

For further details of these points the reader is advised to consult
"Automatic Process Control", by Eckmanf, as the present text is only
concerned with the mathematical background. However, it is possible to
rearrange, or combine the transfer functions of a series of controlled elements in such a way that the whole process may be represented by blocks
whose transfer functions have poles or zeros that are easily identified. This
is very important in the analysis of systems containing a large number of
variables because the roots of the transformed equation in the denominator
of the transfer function may not be easily obtained if its degree is high.

6.9.1.

Control Systems

Chemical processes are controlled by either the "open loop" or "closed

loop" control system, and in all but the most simple cases, closed loop control
is employed.
Thus in the open loop system the composition of a stream
may be controlled as shown in Fig. 6.11a by proportioning the flow of A
by means of the flow meter in the line through which the liquid B passes,
and the flow controller is only able to take a preset action according to the
signal it receives from this meter. Any other changes in the process conditions, such as the concentration of A or B prior to mixing, will have no
effect on the controller, and will lead to deviations from the chosen position.
If, however, a detecting device such as a pH meter is placed in the product
line as shown in Fig. 6.11b so that the concentration of A in B can be
t

Eckman, D.

P.

"Automatic Process Control." Wiley

&

Sons,

New York

(1958).

THE LAPLACE TRANSFORMATION

6.

191

Liquid

Liquid

rC^C]

Solution of

AinB
Solution of

AinB

(b)

(a)

Open and

Fig. 6.11.

closed loops

measured, and a signal representing the measured concentration changes

fed back to the controller, the possible deviation can be corrected. That is
^'feedback" is an essential feature of a closed loop system.
A block diagram of a closed loop system is illustrated in Fig. 6.12 in
which the transfer functions of the different elements depicting the controlling means have been grouped together in the combined transfer function
Gi(5), whilst the transfer function of the feedback element (i.e. the measuring

-^

f,(s)

fo(5)

fA(s)

G,(s)

TmCs)

C2(s)

Fio. 6.12.

element)

is

shown

Typical closed loop block diagram

in the block

Thus, making use of the rules for dynadifferent parts of

jC-^)-

mic and algebraic functions, the subsidiary equations for

this control process

can be written as

=
Us) =

iMis)

o(s)

Lis)

G,is)Us)

(6.49)

-Ms) = Ms) - G2(s)fo(s)

= G.(s)L(5) = G,{s)lUs) - G,(5)/o(5)]

(6.50)
(6.51)

G,{s)Us)

=
l

(6.52)

+ Gi(s)G,(s)

Inverting equation (6.52) gives the value of the output from the controlled
process as a function of time. That is,

/o(0

J_

Gi(s)J
r G,(s)/,(s)e"rfs

Ini J

+ Gi {s)G,(s)

(6.53)

Br,

Laplace transform of the function describing the control system does

not contain any singularities other than poles, the value of the integral can
If the

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

192

be obtained from equation (6.43) as

o(0

G,(s)/,(5)e'

residues of
l

(6.54)

+ Gi(s)G.(s).

and the behaviour of the control system,

after being subjected to a dis-

turbance, can be predicted.

Control Sytem Stability

6.9.2.

To be useful, a control system must be stable and some

may be obtained by inspection of the residues in equation

indication of

(6.54). Thus
any of the poles of the numerator or zeros of the denominator have
positive real parts, then /o(0 will contain a term e^^ which will increase to
infinity as the time / increases and hence lead to instability. Consequently,
all of the relevant poles and zeros must have negative real parts so that any
disturbance will decay exponentially with time and the control system will

this

if

be stable.

This implies that

must be

(6.53)

of the poles of the integrand of equation

all

in the left half of the s plane for the system to be stable.

If in addition there are poles

on the imaginary

system
with other

axis, the control

will fluctuate indefinitely, whilst a pole at the origin, associated

poles in the negative half plane, indicates that the system will fluctuate

about a new position after a disturbance.

system will experience an "off'set".

This means that the control

Stability Criteria

6.9.3.

The

transfer functions describing a control system

cated, containing

numerous poles and

may

be very compli-

zeros, with the result that the procedure

of evaluating the residues is very lengthy and time-consuming.

Consequently some "short cut" methods, indicating whether or not a control
system will be stable, have been devised. These will not reveal the extent of

Neither will
is not necessary.
of a control system to a disturbance;
for a control system could be mathematically stable but functionally useless
because the time required to attain steady state could be very large.
Two methods extensively used to test the stability of a control system
are those of Routh and Hurwitz, and Nyquist.
Essentially the RouthHurwitz criteria of stability has its basis in matrix theory which will be treated
in Chapter 12; here it will suffice to explain the mechanics of the method so
that the section will be complete in itself.
Thus let the denominator be
expressed in the form of a polynomial, viz.

the instability but for obvious reasons this

these

methods indicate the

The

stability

sensitivity

+ Gi(5)G,(5) =

pos" + Pis"-^

+ ... + A, =

of the control system will depend upon the roots of the suband it can be assessed as follows. The system will be

sidiary equation (6.55)

unstable

if

(a) the coefficients in

(b)

some

(6.55)

equation (6.55) vary in sign,

coefficients are zero.

6.

THE LAPLACE TRANSFORMATION

193

Both of these conditions can be assessed by inspection. However, they must

be further tested by "Routh's Rules" for the assessment of the number of
roots of an equation that have positive real parts. Routh proposed that the
coefficients of terms in an equation be arranged into test functions R^ as
follows. Thus in relation to equation (6.55)

Ro = Po

Ri=Pi
P0P3

D
R2=P2

Pi

R^

= PiPiPs-PoPI-pIPa + PoPiPs
P1P2-P0P3

etc.

Routh's test functions can be conveniently expressed in terms of "test

determinants" which are more rational expressions for the test functions.

They

are

Ai

Po

A.

Pi

Po

P3

Pi

Pi

Ps

Pa

P3

A2

A4

Pi

Po

P3

Pi

Pi

Po

P3

Pi

Pi

Po

Ps

Pa.

P3

Pi

Pi
Pe
Ps
Pa
be noticed that the order of the determinant is
the same as the suffix of A with the exception that Aq is of first order. In
addition, the elements of each determinant are arranged in ascending order
down the diagonals, and in descending order horizontally from left to right.
The determinants are related to the test functions as follows.
In these determinants

it

will

= Ao
Ri = ^i
Ri = A2/A1
R3 = A3/A2
'"Rn = KIK-l
R = AJA.^ =
i^o

In fact,

The necessary and

is

that

all

p,

sufficient condition for the control

test

system to be stable
determinants shall be

The denominator of the complete

transfer function of a

of the

test functions,

or

all

of the

positive.

Example

1.

control system can be represented by

5s^ + 3(a-l)s^

+ 2as^ + (a-2)s-(a-3) =

Discuss the stability of the system.

MATHEMATICAL METHODS

194

Inspection of equation
negative;

and

if

is

IN

CHEMICAL ENGINEERING

shows that

greater than

if

then p^

3,

may

be stable if a has values between 2 and

Routh's test functions which are:

is

less

than

negative.

is

2,

Hence

then

is

pj,

the system

This will be confirmed using

3.

i^i=3(a-l)

_3( a-l)(2a)(a-2)-5(a-2)^ + 9(a-l)V^-3)

^

and

i?4

6a(a-l)-5(a-2)

p4

= 3

These can be simplified to give

Ro =

II

= 3(a-1)

Ki

3(a

1)

15a^-68a2 + 95a-47

_
~

^^

III

6?-lla + 10
= 3-a

i?4

VI

The quadratic in the numerator of R2 and the denominator of R^, has

complex roots and is therefore positive for all real values of a. Hence,
provided 1 < a < 3, Rq, R^, R2, and R4. must be positive. It is now necessary
to determine if there are

numerator o[ R^

positive.
.-.

To form an

any values of oc within this range that make the

Denote this numerator by /(a).

/(a)= 15a^-68a^ + 95a-47

idea of the shape of this function

it

VII

can be differentiated,

viz.

^ = 45a^-136a + 95

VIII

Equation VIII has zeros at a = 1-10 and a = 1-93. The first corresponds to
a maximum and the second to a minimum. Substitution into equation VII
shows that
if
IX
/(a) =-4-8
a =1-10,
Since this

is

maximum

of /(a), /(a) must be negative for

available range of values for

oc

is

now

approximately and choosing a few values in

a

a
a

=
=
=

all

a<

this

range gives

/(7)=-9-0

2-5

/(c()= -0-125

f(cc)=

31-0

The
<a<3

1-93.

restricted to the original range 2

6.

THE LAPLACE TRANSFORMATION

195

Therefore a = 2-5 is an approximate root of equation VII with /(a) = 0.

This value can be improved by the method to be given in Chapter 11, thus
proving that the system described by equation I is stable for
2-503
6.9.4.

<

<

3-0

The Nyquist Diagram

of stability makes use of the complex character of

and the fact that the denominator of an unstable
control loop transfer function will have roots with positive real parts situated
Hence if the contour of integration of
in the right half of the s plane.
equation (6.54) is drawn in this part of the s plane as shown in Fig. 6.13, it

The Nyquist

criteria

the Laplace parameter

s,

Nyquist contour

Fig. 6.13.

of the zeros and poles of 1 + ^{s) 2{s) with positive real

Furthermore, it can be shown that within any closed contour

will enclose all

parts.

[Number!
of zeros J

TNumberl _ J_
[ of poles J

2;!

Change

in arg of

function around contour]

(6.56)

Nyquist stated that [Arg/(5)/27r] was equal to the number of times the
contour encircled the origin of /(5)^. If this is A^, the number of zeros is Z,
and the number of poles is P, then equation (6.56) becomes

Z-P = N
In these equations, an nth order zero or pole
poles of first order.

(6.57)
is

considered to be n zeros or

Nyquist used equation (6.57) to restate the stability requirement of

equation (6.54). A_ control system is stable if there are no zeros of the
function [1 + 0^(5) ^2(5)] in the positive half plane, and this means that if
the sum of the number of encirclements of the origin of this function and the
number of poles enclosed by the contour is zero then the control system is
stable. _ The Nyquist diagram consists of plotting the ''transfer locus"of
[G 1(5) 02(5)] as^ 5 traverses the contour of Fig. 6.13. The origin of the
function [l + G^(s) 62(5)] is (-1,0) and provided \Gi(s)G2(s)\-*0_ sls
|5|->oo the infinite semicircle can be ignored. The poles of Gj^(s) and (72(5)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

196

have to be located in any case thus determining P, and the number of times
and the
(-1,0) is enclosed by the transfer locus determines
Nyquist criteria can be apphed. Usually it is only necessary to plot the
< co < oo because the locus between and oo
transfer locus of ^ = /w for

the point

is

just

its

reflection in the real axis.

The Nyquist criteria can be applied by plotting the function [G^(s) ^2(5)]
from arbitrarily chosen values of co between and 00, or by plotting directly
experimental phase and magnitude data. Since it is not essential to factorize
the open loop transfer function in order to obtain the number of poles, the
stability of the system can be assessed by using the value of A^ obtained from
the transfer locus and the value of

P+N =

then the control system

be illustrated by an example.
0,

P
is

obtained by inspection.

Unless

These principles

unstable.

will

Z=
now

Example 2. Estimate the stability of an automatic control system whose

combined transfer function is
G,(s) 62(5)

Put

(1-10/co)-

(10/co

+ l)^(l-10/co)^

then

+ iy

ioj

9
(10/co

9/(105+1)^

- 300co^)
(l + lOOco^)^

9(1

II

r9(30co
L

(1

- lOOOco^)
III

>2\3
+ 100(X>')

its real and imaginary

These are the coordinates of the transfer function in the Nyquist
diagram and Table 6.3 shows the evaluation of them.
The final two columns of Table 6.3 are plotted in Fig. 6.14 to form the

In equation III the function has been separated into

parts.

^-8
Fig. 6.14.

Stability

diagram for the transfer function

(10.y

ly

6.

THE LAPLACE TRANSFORMATION

197

0'Or^porp^oooor<ifS^'7ipp
I

Ooooo^nO>/^oooooow^Oa^rNa^'-'
O'>s^opp-7lr^lfsoor^^or;-'^pp
I

m
o
OS
psOrnoocNcnfMOspP'T'
f*^

w-)

r*-

00

r^

^m

oo <N \o
Tt
r-

o
m

On

"^

rr,

-h

<;j

Cj

^
5^

m o_ oo
(NfOTtooTfrNooovr^

o
o

<N o

en

^-^

f<-i

Tt fN

'*-'

s:

Omrsir-cojorvjro
^
pr^qswpvo(Nr:>(NfO^^^Ofqcrs
,-^

ryy

aNoor^voTtr40'^oo-iir^ON^^-^r1

r^
cn \n
ooooooooiri^ooinTt
O O O -^ ^ -H
oxxxxxxxxppppp'o
^OOr^TtmvOfOfNOOOOOO
<N ^ rN ^ '^
I

T^

Tj-

Tt

Tt

Trt-

fNl

en 0

rf

Tt

Tj-

ooooooooSInoJn
I

1-1

(S

'^ vo

oooooo

pppppppp'T'-Tir^fS'^oo

198

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

Nyquist diagram for the transfer function 9/(105+1)^. This function does
not have any poles in the right half plane as can be seen by inspection; however, the transfer locus encircles the point (1,0) twice.
.-.

iV

IV

Therefore, by equation (6.57)

Z=N+P=2
and the control system is unstable.
In the example given, the Laplace transform could have been inverted
and the stability checked as a function of time just as rapidly as Table 6.3
and Fig. 6.14 were prepared. However, it must be remembered that with an
actual control system, the transfer function would be very complicated, and
it would then be advantageous to prepare the transfer locus and use the
Nyquist criteria to test stability.

Chapter 7

VECTOR ANALYSIS
Introduction

7.1.

it was shown that a complex number consisted of two independent parts, a real part and an imaginary part. The one symbol z was
used to represent a combination of two symbols x and y so that manipulations with a single symbol were much quicker than the corresponding
elementary operations on the separate variables. It is natural to enquire
whether this principle can be extended so that more than two variables can
be combined into a single symbol thus shortening the work m multi-variable
systems. The Argand diagram is a plane, two-dimensional representation of
the two parts of a complex number which he along two orthogonal axes,
and a natural extension is to represent three parts of a "number" along three

In Chapter 4

orthogonal axes in three-dimensional space. This new type of "number" is

a vector.
Further extensions of this idea cannot be visuaHzed, but any number of
variables can be grouped into a single symbol in two ways the first, by means
of matrices, will be discussed in Chapter 12, and the second using tensors
will be described briefly here. Tensors are introduced to indicate a general
type of number of which vectors are a special case. The two variables which
are combined in a complex number are distinguished as a real part and an
imaginary part, and the principal difference between tensors and matrices
is the labelHng and ordering of the many distinct parts.
The ordinary
operations of arithmetic have to be restricted in the study of complex
numbers by defining principal values, giving a special rule for division, etc.
With vectors, the restrictions are more severe as will be shown in the
following sections of this chapter.
The symbohsm and operations of vector analysis constitute a powerful
shorthand method for describing complex problems and expressing equations in general terms. For example, in Section 1.8, the steady state temperature distribution in a cylindrical conductor was shown to satisfy the equation
;

IdT
2+-
=
dr

d^T
dr^
It is

show that the corresponding temperature distribution

conductor with insulated edges satisfies the equation

a simple matter to

in a flat

(7.1)

d^T
199

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

200

Both of the above equations relate to the one process, heat transfer, but to
systems having different geometrical shapes. The utihty of vector analysis
is that the following equation in vector form will represent all steady state
heat conduction processes with constant thermal conductivity in any
number of dimensions and for any shape of solid. Thus

W^T =
The

(7.3)

rules of vector analysis are explained in the earlier sections of this

chapter, so that the second half of the chapter can be devoted to the use

which

interpretation of Hamilton's operator (V)

is

and

the fundamental symbol

of this compact notation.

Vector equations are usually used to describe general processes such as
the above, but when a particular example is considered, the vector form
(7.3) is expanded into the appropriate particular form (7.1) or (7.2) and the
completion of the problem is outside the domain of vector analysis. It is
thus difficult to find chemical engineering examples which use vector
methods exclusively, and this chapter is primarily a service chapter for the
later parts of the book.
7.2.

Tensors

The notation of tensors can be introduced by referring to complex

The complex number z can be written in the alternative ways

numbers.

= x + iy =

Zj^

+ iz2

(7.4)

first rank because one suffix m is needed

understood that the subscript m takes either of two values,
1 for the real part and 2 for the imaginary part.
A more general tensor z,
can have more than two parts if w is allowed to have more than two values.
The resulting parts z^, Zj, Z3, .... etc., are all independent of one another
just as Zj and Z2 are independent in equation (7.4).
By reference to the
Argand diagram this can be interpreted as extending the number of dimensions of the system. If m is restricted to three values, the system is threedimensional and can be illustrated diagrammatically, whereas if m takes
values greater than three, systems of higher dimensions can be described
but not visualized pictorially.
The notion of a tensor can be further generalized by using more than one
subscript, thus z^ is a tensor of second rank in which both m and n can take
integer values within the dimensions of the system. For instance, in three
dimensions the tensor z^ represents the nine quantities z^, Z12, ^13, ^2i>

where z

is

to specify

referred to as a tensor of

it.

It is

^23 ^31 J ^32> ^335 whilst iu four dimcusious, z^ represents sixteen

quantities. For a clearer distinction between the first and second subscript,

^22

the one index

sometimes raised to the position of a superscript, thus

is

The symboHsm

for the general tensor

is

now

obvious,

it

consists of a

z,"

main

symbol such as z with any number of associated indices which may be subEach index is allowed to take any integer value up

scripts or superscripts.

to the chosen dimensions of the system.

with the tensor

is

called the

The number of

"rank" of the tensor.

indices associated

7.

VECTOR ANALYSIS

201

Tensors of Zero Rank, Scalar

7.2.1.

If the tensor has no index, then it only consists of one quantity independent of the number of dimensions of the system. The value of this
quantity is independent of the complexity of the system, possesses magnitude
only and is called a "scalar". No direction can be associated with a scalar
since direction can only be specified in a system of known dimensionaUty.
All physical properties are examples of scalars. For example, density, mass,
specific heat, coefficient of viscosity, diffusion coefficient, thermal conductivity, etc. Other examples are the scalar point functions such as temperature, concentration, and pressure which are all signified by a number which
may vary with position but not depend upon direction. Energy and time

are also scalars.

coefficient of viscosity for a non-newtonian
be defined as a vector quantity if its value at a point is related to
a specified direction. Normally, however, the coefficient depends upon the
state of the fluid at the point rather than the point itself.
It is

fluid

worth noting that the

may

7.2.2.

Tensors of First Rank, Vectors

When
the

the tensor has a single subscript

it

consists of as

number of dimensions of the system. For

many

elements as

practical purposes this

number

and the tensor has three elements which are normally called components. The tensor of first rank is alternatively named a "vector". Vectors
have both magnitude and direction and can be represented in three dimensions
by a straight Hne in space. The most common examples of vectors are force,
velocity, momentum, angular velocity, weight, and area.
The above classification of scalars and vectors should help to clear some
common misconceptions about the nature of the properties Hsted. For
instance, mass is a scalar and depends only upon the substance involved
and not upon its environment, whereas weight is a force resulting from the
action of gravity upon the mass. Weight is thus a vector since it acts in a
direction chosen by the gravitational field in which the scalar mass is situated. Newton's second law of motion is a vector equation relating the two
vectors force and acceleration by means of a proportionaHty constant, mass.
Pressure is a scalar quantity because it acts equally in all directions and
thus has no special direction. Scalar pressure becomes vector force when a
is

three

surface

is defined to sustain the pressure; the direction of the force vector is

associated with the orientation of the surface and thus surface area is a
vector quantity.

The

results of various

products between vectors and scalars will be better

when multipHcation of vectors has been
defined, especially the reasons why momentum is a vector, whereas energy

understood
is

later in this chapter

a scalar.

Tensors of Second Rank

The one tensor of second rank which occurs frequently

7.2.3.

the stress tensor.

in engineering is
This of course has a double subscript and has both a

202

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

magnitude and two directions associated with it. In three dimensions the
stress tensor consists of nine quantities which can be arranged in a matrix
form thus,
7^11

T,2

T,,

21

22

23

31

32

33.

(7.5)

Take three cartesian coordinate

The
axes X, y, z,^and define an elementary rectangular box of dimensions Sx, dy,
bz. There are three typical faces to the box each having a normal parallel
to one of the coordinate axes. Each face of this box has a force acting upon
it, and this force can be resolved into three components parallel to the three
axes. This is illustrated in Fig. 7.1 for two of the faces where the normal
physical interpretation

is

as follows.

zL

'iy

/]

/-r^y

/^

y/^

'I?

Xq

x'

-I

C
The

Fig. 7.1.

stress tensor

components have been labelled with p and the shear components with t.
To distinguish between these various forces it is necessary to specify both the
direction of the plane and the line of action. The nine elements of the stress
tensor (7.5) can be associated with these individual forces by using the first
subscript to denote the plane and the second subscript to denote the direction of the force. Thus
xy

T =

'yx

T.,

where the subscripts

yy

(7.6)

yz

'zy

x, y, z to restrict the tensor

read as "the shear force on the x facing plane
acting in the y direction". It can be seen that one component of the force
on a plane acts in the normal direction whilst the other two are shear forces
along the surface. The forces illustrated in the diagram are the forces acting
on the element, and the reaction forces exerted by the element on neigh-

to three dimensions.

1, 2,

have been replaced by

Xj,y is

bouring elements are obviously

all

reversed in direction.

For

this reason,

7.

the

components of

ABED)

(e.g.

all

VECTOR ANALYSIS

stress acting

on a plane facing

203
in a positive axis direction

act in the positive direction of the corresponding axis ; whereas

the components acting on a plane facing in a negative direction (e.g. OABC)

all act in the negative direction of the corresponding axis. It is to be noted
that the repeated suffix gives the normal forces, and the unlike suffices

denote the shear forces. Also, the normal forces include the pressure, but
with reversed sign because the force due to pressure acts inwards towards
the element.

Tensor calculus, which

is

a speciaHzed field of mathematics, has

little

direct application to chemical engineering at present since chemical engineers

confine their attention to three dimensions.

special kinds of tensor (scalar

Consequently, only the above

and vector) which have physical

tions will be considered in this chapter.

interpreta-

If the reader is interested in the

manipulation of tensors he is referred to Craig or Sokolnikoff'tIn the subsequent discussion, all quantities will be referred to as either
vectors or scalars and all further remarks will apply to three dimensions.
The subscripts will be labelled by the coordinate variables rather than by
numbers for extra clarity. Bold face type will be used for vectors, and ordinary type for scalars.

7.3.

Addition and Subtraction of Vectors

Any

vector can be represented geometrically by a straight line with an

arrow, the length of the line representing the magnitude of the vector and
the direction of the line as indicated by the arrow representing the direction
of the vector. The addition of two vectors is defined geometrically by the

well-known "triangle of forces". The start of the line representing the

second vector is superimposed on the end of the line representing the first

A
Fig. 7.2.

vector.
start

Addition of vectors

The sum of

of the

first

the two vectors is represented by the line joining the

vector to the end of the second vector as illustrated in Fig. 7.2.

The product of two scalars is obtained according to the rules of ordinary

and to multiply a vector by a scalar, the magnitude of the vector
is multiplied by the scalar and the direction remains unchanged. Multiplication of a vector by a negative number also involves reversing the direction
arithmetic,

H. V. " Vector and Tensor Analysis." McGraw-HUl (1943).

Sokolnikoff, I. S. " Tensor Analysis." Wiley and Sons (1951).

t Craig,

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

204

of the vector. Thus to subtract one vector from another

to reverse the direction of the vector to be subtracted
it

it is

only necessary

and add the other to

as illustrated in Fig. 7.3.

A-B

Subtraction of vectors

Fig. 7.3.

7.3.1.

Components

The process of addition described above can be reversed allowing a

vector to be resolved into equivalent constituent parts called "components".

components of a vector are added together using the above rule, the
is returned. A vector can be resolved into components in an
infinite variety of ways but the resolution into components in three specified
non-coplanar directions is unique. This fundamental property is ultilized by
If all

original vector

defining a coordinate system referred to three mutually orthogonal axes.

Define the three cartesian coordinate axes x, y, z as shown in Fig. 7.4, and

Aj+A
Ax
Fig. 7.4.

Resolving into components

represent the vector /4 by a line starting from the origin.

Complete the
rectangular box which has edges parallel to the axes, and the vector A as
diagonal, and denote the edges of the box by the vectors A^, Ay, A^. The

and it is obvious that all

have the same length and are denoted by the same vector.
Applying the rule of addition twice it can be seen that
subscripts denote the directions of the vectors

parallel edges

Ay^-\-

Ay-T A^

(7.7)

Since vectors are frequently resolved into components along the coordinate axes, it is convenient to define a set of unit vectors having unit

magnitudes and the directions of the coordinate axes. Any vector can then
be expressed as the sum of scalar multiples of these unit vectors. The
standard set of unit vectors which define the cartesian coordinate system is

VECTOR ANALYSIS

7.

1,

7,

in the x, y, z directions respectively.

205

Equation

(7.7)

can

now be

re-

written as

A = AJ + Ayj + A,k

(7.8)

where A^, Ay, A^ are the scalar lengths of the edges of the box, and the unit
vectors i, j, k define the directions. Two applications of the theorem of
Pythagoras in Fig. 7.4 shows that the magnitude of A, which is denoted by
1^1 or A by analogy with complex numbers, is given by

= ^(Al + A'y+A',)

\A\

(7.9)

or the squared magnitude of a vector is equal to the

its three mutually orthogonal components.
7.3.2.

sum

of the squares of

Position Vectors

The

position of a point in space can be specified relative to an origin by

Such a vector

is called a
independent of any
coordinate system and only requires the definition of an origin. An analogy
which may be helpful in grasping this point is the reply to a stranger who

defining the vector joining the origin to the point.

The above

"position vector".

definition of position

asks for directions to a particular destination.

500 yards in a

NNE

is

If the reply given is "travel

direction", then a complete coordinate system

is

given in the form of an

arm pointing out the direction with the instruction "travel 500 yards over
there", the arm is a vector indicating direction and the spoken instruction
imphcitly involved; whereas

gives the magnitude.

The

if

first

the information

reply presupposes that both parties can locate

the exact northerly direction, which

much more

direct

and

is

is

unlikely, while the second reply

is

useful.

A vector defining the position of a point is usually given the symbol r

which can be resolved into components in the same way as any other vector
once a coordinate system has been defined. Thus
r=x + y + z
or

The point can

xi

-h

yj -\-

(7.10)

zk

(7.11)

therefore be specified by

its coordinates either by equation

by equation (7.11) in a scalar form. The latter is
more usual and it is imphed that x is measured in a direction defined by /
and similarly for y and z. Thus when a point is specified by (x, y, z) it is

(7.10) in a vector form, or

important that the order of the coordinates should not be disturbed.

Returning to the addition of two vectors A and B, and referring to Fig.
7.5, it can be seen that the sum of the vectors can be obtained by summing
the separate components. Thus if
vitally

and
then
also

A = AJ + Ayj + A,k

(7.12)

B=:BJ + Byj + B,k

(7.13)

A+B = (A, + B,)i + {Ay + By)j + {A, + B,)k

A-B = {A,-B,)i + {Ay-By)j + {A,-B,)k

(7.14)
(7.15)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

206

For two vectors

to be equal, their difference

A can only
A^-B^ = 0,

equal

equation (7.15),

Thus a

Ay-By =

zero,

and from

0,

andA,-B, =

(7.16)

single vector equation represents three simultaneous scalar equations.

Ax
Fig. 7.5.

7.3.3.

must be

B if

Addition of vectors by components

Properties of Addition

Both the associative law and the commutative law of algebra are valid
That is,

for the addition of vectors.

A + (B + C) = {A + B) + C
A +B = B+A

and

The

distributive

law

vectors by a scalar.

is

(lAl)
(7.18)

also valid for the multiplication of the

sum of two

Thus

m(A + B) =
There

is

7.3.4.

Geometrical Applications

mA + niB

no divergence therefore from the normal

addition and subtraction of vectors is considered.

(7.19)

rules of algebra

when

In general, two vectors are sufficient to define a plane, and if a linear

relationship exists between three vectors then they must be coplanar. This
is obvious from the rule of addition because when two of the vectors have

been added together, they must have the same direction as the third vector.
1. If ^ and B are two position vectors, find the equation of the
Hne passing through the end points of A and B.
Referring to Fig. 7.6 where the origin is O and C is the position vector of
any other point on the Hne through A and B, the vector joining A to B is

Example

straight

B-A.

Similarly, the vector joining B to C is C-^. If ABC is to be a straight

then these two vectors have the same direction and can only differ in
magnitude. Therefore

line,

C-B = miB-A)

VECTOR ANALYSIS

7.

where

w is

a variable scalar and

C is

207

the variable vector.

Equation

can be

rearranged to give

C = {m + \)B-mA
m^-(m + l)B + C =

or

Equation

II is the

and

III

standard equation of a straight hne with

Equation of a straight

Fig. 7.6.

variable,

II

m the independent

line

Equation III is in the form of a

between three vectors, the general form of which is

the dependent variable.

linear relationship

pA-^qB + rC =

IV

Comparison of equations III and IV shows that equation III is the special
case in which p+q-\-r = 0. Thus the general result that if three position
vectors are related by a linear equation then they are coplanar. If the sum
of the three coefficients
coHnear.

Example

2.

is

also zero, then the

end points of the vectors are

Prove that the medians of a triangle are concurrent.

Figure 7.7 illustrates the problem with reference to the triangle

are the mid-points of the sides. It is assumed that
AL intersect at P and that
does not necessarily pass through P.

where L, M,

ABC,

BM

and

CN

Medians of a

Fig. 7.7.

triangle

Taking C as origin, and denoting the position vectors of A and B by ^4

and B, then the equations of the lines BM and AL can be written as follows
using the results of Example 1
Thus the equation of the line through B
and
is
.

and as the parameter

(s+l)B-s(i/l)

s varies, the vector

a determines the various points

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

208

BM.

along the line

Similarly, the equation of

a'

Since
is

is

the point of intersection of

given by a

a',

AL

is

= (r+lM-Ki^)

AL

II

and BM, the position vector of P

or
(s
.-.

+ 1)B - is A =
(t + l + is)A =

+ 1)A -itB
{s + \ + it)B

III

{t

IV

Equation IV relates two vectors which have different directions and the
equahty can only be satisfied by both vectors having zero magnitude.
/.

The

+ 1 + 15 =

solution of equations

ands + l + ir =

is

t=-i

and the vector CP becomes J (B+A) from equation I.

The vector CN is the sum of the vectors along CB and BN; and thus the
position vector of N is given by

B + i(A-B) = i(A + B)
CN and CP have the same

Therefore the vectors

must pass through P

direction

(A+B) and

thus

CN

Example 3. Generalized Vector Method for Stagewise Processes. It has

been shown by Lemlich and Leonard! that many of the existing difference
point constructions for determining the number of theoretical steps in a
stagewise process are variations of a more general vector method. They
have also shown that the comprehensive vector method suggests many
hitherto untried constructions and this example from their work derives
the standard rectangular diagram from the general vector diagram.
In any stagewise process there is more than one property to be conserved
and for the purpose of this example it will be assumed that the three properties, enthalpy (//), total mass flow (M), and mass flow of one component
(C) are conserved. Instead of considering three separate scalar balances,
one vector balance can be taken by using a set of cartesian coordinates in
the following manner. Using x to measure M, y to measure H, and z to
measure C, then any process stream can be represented by a vector thus,

OM = Mii + //J + Cife

OM

where
is the vector
can be represented by

illustrated in Fig. 7.8.

Similarly, a second stream

ON = M2i + H2J + C2k

II

and using vector addition,

OR =
Thus

OM + ON = (Ml + M2)i + (H^ + ^2)7 + {, + C2)k

OR

III

which represents the sum of the two streams, must be a constant

vector for the three properties to be conserved within the system. Hence,
t Lemlich, R.

and Leonard, R. A. A.lCh.E.J.

8,

214 (1962)

7.

VECTOR ANALYSIS

209

OM

ON

or
is deterwhen either of the streams
mined, the other is obtained by subtraction from the constant OR. It will
now be shown that the intersection of these vectors with the plane x = I,
as illustrated in Fig. 7.8, gives the usual enthalpy-concentration diagram
for the Ponchon-Savarit method.
to perform a calculation,

M/

^^

/
^
Fig. 7.8.

Plane t = 1

Representation of a stagewise process

The constant Hne OR will cross the plane x = 1 at point P which will be
and ON will also cross the plane at
The variable vectors
and B respectively, and since OMRN is a parallelogram, APB will be a

OM

a fixed point.

straight Hne.

Let

AP_b

IV

'PB~~a
as illustrated in Fig. 7.8.

aOA

Then by

the triangle rule of addition

ah

aOP + aPA = aOP +

tBA

a-\-b

ab

bOB = bOP + bPB = bOP +

and

Adding the extreme

parts of equations

aOA + bOB =

and VI

(a

+b

AB

gives

+ b)OP

VII

Because equation IV only defines a ratio between a and

a can now be chosen so that

aOA =
Since

OR,

OB is

in the

same

direction as

VI

b,

OM

ON and OP

the magnitude of

VIII
is

in the

same direction

as

can be proved in a similar manner to that used in Example 2 that

equations III and VII are now identical, i.e.

and

it

bOB = ON
(a + b)OP = OR

IX

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

210

Using similar triangles and equations I and VIII, the x component of

can be written in two equivalent ways.

Similarly,

and substituting

into equation

IV

= M,

OM
XI
XII

gives

AP
PB

XIII

Ml

the normal lever law.

Using equations
points A, B, and

I,

II, III,

VIII, IX, X, XI,

P can be determined

pomtAis

point

^1,

Hi

Ci

^^,

^^

H1 + //2
is

1,

and XII, the coordinates of

as

M1 + M2

Ci

+C

M1+M2

Comparing these with the original interpretation of the coordinate axes, the
appropriate two-dimensional coordinates are the enthalpy, and the mass
fraction of the important component.
Thus the normal constructions in the enthalpy-concentration diagram
are vaHd as a special case of a more general vector method. By taking the
intersections of the vectors with different planes, different two-dimensional
diagrams can be obtained. If the chosen plane is symmetrically inchned to
all axes, then the familiar triangular diagram is obtained.
The work of
Lemlich and Leonard is extended to these cases and also to four dimensions.
Suggested extension to higher orders is difficult to visuaHze and matrix
methods may be more appropriate. The example is, however, sufficient
indication of the generality of the vectorial approach which gives a clearer
insight into the connection between related problems.

7.4.

Multiplication of Vectors

first serious divergence from normal algebra arises

necessary to define two distinct kinds of multiplication. Two
physical situations in which the interplay of two vectors gives rise to the

In multiplication, the

because

it is

work done by a force during a displacement when

both the force and the displacement are vectors; and secondly, the moment
of a force about an origin, when both the force and the position vector of its
point of apphcation must be specified.
These two interactions between
vectors are fundamentally different as will be shown in the present section.
result, are firstly, the

7.

7.4.1.

VECTOR ANALYSIS

211

Scalar or Dot Product

The scalar product of two vectors is defined as the product of the magnitude of the one vector with the magnitude of the component of the other
vector resolved along it. This product is signified by placing a dot between
the vectors to be multiphed together. Symbolically,

A.B =
where 6

is

\A\\B\cose

= B.A

(1.20)

the angle between the vectors as shown in Fig. 7.9. The result of
is a scalar quantity, hence the name for the product.

the operation

7
J

A'
|A

/
s n

-/

-/.e

0\^

/
/

A cos 6
Fig. 7.9.

Scalar and vector products

One physical interpretation of this product is the calculation giving the

work done by a force during a displacement. The fundamental definition
of the work done is the magnitude of the force multiphed by the distance
moved by its point of apphcation in the direction of the force. Alternatively,
the work done is the displacement multiplied by the component of the force
causing the displacement.
That these two alternatives are equivalent is
obvious from the definition, equation (7.20). Work and hence energy are
thus scalar quantities which arise from the multiplication of two vectors.

When a vector is multiphed by

be written in the equivalent forms

itself in this fashion,

the scalar result can

A,A = A'-^A'The

(7.21)

inverse process to multiplication, division,

for the following reason.

If

is

not uniquely defined

an equation such as

^.B =
is satisfied,

(a)

(b)
(c)

The
It is

(7.22)

then any of the following three conclusions can be drawn.

The vector A is zero.

The vector B is zero.
Q = 90, and A and B

are mutually perpendicular.

third possibihty arises in equation (7.20) because cos Q

therefore possible for the scalar product of

two vectors

may

be zero.

to be zero

when

both vectors are finite. Division thus has no meaning, for if equation (7.22)
could be divided by a non-zero A, the inevitable conclusion would be that
B is zero which is not necessarily true.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

212

Vector or Cross Product

7.4.2.

Referring to Fig. 7.9 again in which 9 is the angle between the two
A and B, then the vector product oi A and B is defined by

vectors

A/\B = \A\\B\smQn

{122>)

where the symbol ( a ) is used to denote a vector product, /i is a unit vector

along the normal to the plane containing A and B and its positive direction
The right hand screw rule is applied to the sense
is determined as follows.
of rotation from ^ to 5 in measuring the angle 0, thus defining the positive
direction of n. Referring to Fig. 7.9, this rule indicates that n acts into the
also a vector into the plane of the paper.
plane of the paper, thus A ^
This definition is unique, because if Q were described in the opposite sense
as indicated by G\ the positive direction of n would be reversed, the sign of
sin Q would also be reversed, and the vector A /\ B would still act into the
plane of the paper. However, the vector B a A has the same magnitude
as ^ A ^ but the direction of n is reversed due to describing 6 by moving
from B to A. Hence

Bh

AaB=-BaA

(7.24)

and the order of terms in the product is important. The magnitude of the
A and B is equal to the area of the parallelogram formed
by A and B. This is also illustrated in Fig. 7.9 which shows the base of the
parallelogram of length B, and height A sin 6, giving an area of AB sin 9
vector product of

or 1^1

\B\ sin 9.

The moment of a
manner.
relative to

about

force

is

If there is a force

is

an origin O as
defined by

defined by a vector product in the following

acting at a point

P with

position vector r

illustrated in Fig. 7.10, then the

L = fAF

Or-

moment

of

(7.25)

^v7p
/
\

rsin

Fig. 7.10.

Moments

of a force

w, the vectors r and F must have a

taken as this origin and the right hand screw rule
applied to equation (7.25) then n is seen to act out of the plane of the paper.
Using the fundamental definition of L as the product of the force and the
perpendicular distance of the origin from the line of action of the force gives

In order to define the positive normal

common

origin.

If

is

\L\

iFlrsin^

7.

213

VECTOR ANALYSIS

which checks with equation (7.25) and the definition (7.23). The direction
of the vector L which is outwards from the plane of the paper also
corresponds with the fundamental definition of a moment.
Again, vector division

is

impossible, since

if

AaB =

(7.26)

then either

The vector A is zero

The vector B is zero
9 is zero, and the vectors A and

(a)

(b)

or

(c)

7.4.3.

B are

parallel.

Properties of Multiplication

Although the

associative, distributive,

and commutative laws are vahd

need to be conhas already been shown

in equation (7.20) that the commutative law is vahd for the scalar product,
whereas equation (7,24) shows that the commutative law is not valid for the
vector product. That is,
for the addition

sidered

more

and subtraction of

vectors, these rules

carefully with respect to products.

It

A.B = B.A

(7.20)

AaB= -BaA

but

The order of terms

in a vector product

(7.24)

can only be altered

if

the sign

is

reversed.

The

distributive law

is

unreservedly vahd for

A.(B + C)

Aa{B + C)

all

vector products.

= A.B + A.C
= AaB + AaC

That is,
(7.27)

(7.28)

The associative law is simplified by the convention that two vectors

without an operating sign between them have no meaning, thus reducing
the need for brackets. For example, the following terms can only be
evaluated in the one particular order indicated by brackets on the right hand
side of each equation.

A.BC.D = (A.B)(C.D)
A.BC = {A.B)C
AaB.C = (AaB).C

(7.29)
(7.30)
(7.31)

In equations (7.29) and (7.30) each dot product yields a scalar which can be
treated as any other scalar, so that equation (7.29) is a product of two
scalars, and in equation (7.30) the magnitude of the vector Cis increased by
the scalar multipher (A.B). In equation (7.31) the vector product must be
taken first because a vector product cannot be formed between the vector
A and the scalar (B.C). Brackets are usually essential when more than one
vector product is involved, and then the associative law is not vahd because

^ A (J5 A
as will be

shown

in Section 7.4.6.

C)

7^

(^ A B) A

MATHEMATICAL METHODS

214

CHEMICAL ENGINEERING

IN

Unit Vector Relationships

7.4.4.

It is

frequently useful to resolve vectors into components along the axial

All operations are then

/, j and k.
performed on the unit vectors and the results of products between them are
all standard.
Since by definition, i, j and k are mutually orthogonal, the
or njl and
angle Q in the definitions of the two products must be either
directions in terms of the unit vectors

because

sinO

and

cos7r/2

the following results are true.

ij=jk =
and

Because

/,

Ai

k.i==0

(7.32)

=j aj = kAk =

(7.33)

j and k have unit magnitude,

ii=jj^k.k=
By

(7.34)

the definition of the vector product in Section 7.4.2,

normal

a 7

is

a vector

of/ and 7 and hence must lie in the k direction. A connow needed which determines whether 1 a j is in the positive or

to the plane

vention is
negative k direction.
If i a j = k, the set of axes is said to be "right
handed", since k is defined by the right hand screw rule from the vector
product of / andy; whereas if i a j = k, the set of axes is left handed. It
is normal to use a right handed set of axes and it can easily be verified that

Ak = kAi=j
JAi=k, kAJ=i, iAk=j
i

and

The use of

AJ

k,

these relationships

is

Thus,

and

(7.36)

by the straightforward applicatwo vectors

illustrated

tion to the simple products of the

and

(7.35)

i,

A = AJ + Ayj + A,k
B = BJ + Byj + B,k
A.B = (A J + A J + A,k). (BJ + BJ + B, k)

AaB

(7.12)

(7.13)

= A,B, + AyBy + A,B,

= (A J + A J + A^k)A (BJ + Byj + B, k)
= A,(Byk-BJ) + Ay(BJ-B,k) + AXBJ-Byi)

using equations (7.32) to (7.36).

Rearrangement of the

last

(7.37)

line

AAB = i{AyB,-A,By)+j(A,B,-A,B;) + k(A,By-AyB,)

Equation (7.38) can be written

in a

more

useful

gives
(7.38)

and compact form by

expressing the right hand side as a determinant. The evaluation of determinants is considered in elementary mathematical texts, thus

AaB =

(7.39)

VECTOR ANALYSIS

7.

215

Scalar Triple Product

7.4.5.

.4 a 5 is a vector, further products can be taken with a third

The scalar product of ^ a ^ with C is best considered geometrically
is the usual angle between the two vectors
as shown in Fig. 7.11, where
{A A B) and
By definition,

Because

vector.

^aB.C
a scalar whose magnitude is the magnitude of {A a B) multipUed by the
component of the vector C resolved along {A a B). But the magnitude of
{A A B) is the area of the parallelogram formed on A and B, and since the
is

^-^ -^

^ '' /

^ "^

y ^^-^

/^

AaB

q/
/
m^^--

~~

~~~-~--^
/

Lj

A
Fig. 7.11.

**

^^^

^ -~

^^
^-^

^'^'^"^-"-ii.^^

Sy
/^^/^"^'^

--"^

/
/

~"~^

^^

*^''
Scalar triple product

a B) must be perpendicular to the plane of A and B, then the

C along {A a B) must equal the height of the parallelepiped
as shown in Fig. 7.11. The magnitude of ^ a B.C is thus the volume of the
parallelepiped with edges parallel \o A, B, and C. The result is positive or
negative according as d is acute or obtuse. Since the volume of a parallelvector {A

resolved part of

epiped can be expressed as the product of any base with the corresponding
height, and the order of vectors in a dot product is irrelevant, the scalar
triple product can be written in six equivalent ways, thus

^aB.C = ^.BaC = BaC.^ = B.Ca^ = Ca^.B = C.^aB

= [^,B,C]

(7.40)

where the three vectors remain in the cyclic order A, B, C and the positions
of the dot and cross are arbitrary. The other six ways of writing this triple
product have the same numerical value as the expressions in (7.40) but have the
opposite sign due to reversing the order of terms in the vector product. Thus

[^,B,C]=-[^,C,B]
7.4.6.

(7.41)

Vector Triple Product

B with a third vector can be taken, but in

order of multipHcation and position of brackets is of vital
importance. Referring to Fig. 7.11 again, it can be seen that the vector
A A B is perpendicular to the plane of A and B. When the further vector
product with C is taken, the resulting vector must be perpendicular to
and hence in the plane of A and B. The resulting vector triple product can therefore be resolved into components along A and B, thus
The vector product of A a

this case the

AaB

MATHEMATICAL METHODS

216

CHEMICAL ENGINEERING

IN

(AAB)AC = mA + nB

(7.42)

m and n are scalar constants to be determined.

out equation (7.42) by C using a scalar product gives

where

Multiplying through-

C.(AaB)aC = mC.A + nC.B

A a B

Considering
(7.43)

the

is

combined

o.

side

mC.A + nC.B =

m = aC.B

or
is

zero.

is

.-.

which

hand side of equation

appearing twice. Hence

vector, the left

a scalar triple product with the vector

hand

left

a.s

(7.43)

and

n=(xC. A

(7.44)

Substituting equations (7.44) into (7.42) gives

equivalent.

(AaB)aC =

- (C

a[(C B)A
.

A)B']

(7.45)

Since each term in equation (7.45) contains the three vectors, a must be just
a number. Also, because equation (7.45) is valid for any vectors A, B, and

C,

it

A =

must be

valid for a particular set of values for A, B,

B = C = j,

i,

and using equations

(7.35)

and

and C. Thus

letting

(7.36) gives

(AAB)AC = (iAJ)AJ
= kAJ
= i
But using equations

(7.32)

and

(7.34),

C.B=jJ =

C.A=j.i =
Using these

results in

equation (7.45) gives

=
i

.-.

(xi

a=-l

(AaB)aC = (A.C)B-(B.C)A

and
Similarly

it

can be shown that

Aa(BaC) = (A.C)B-(A. B)C

In order to

(7.46)

remember equations

(7.46)

and

(7.47)

(7.47) the following rule

is

helpful.

The terms appearing outside the brackets on the right-hand side are the
terms appearing inside the brackets on the left-hand side. The positive
term is the dot product of the extreme vectors multiplied by the central
vector.

7.5.

/ (say time), then when /

vary by an increment dr. Just as in ordinary
a variable associated with r but it need not have either the same

If a vector r

varies

is

a function of a scalar variable

by an increment

calculus, 5r

is

Differentiation of Vectors

dt, r will

VECTOR ANALYSIS

7.

magnitude or direction as

defines the

first

217

If dr tends to zero as St tends to zero,

r.

hm - =

5r

dr

dt^o ot

at

(7.48)

derivative of r with respect to

/.

If r is interpreted as a position vector resolved into its

==xi + yj-\-zk

and

dr

(7.11)

Sz
Sy
= Sx
1 + J + k
,

St

St

dr

dx

St

St

as St-^0,

dy

dz

Differentiation of a vector with respect to a scalar

differentiation

components, then

Sxi + Syj + Szk

dr
.*.

and taking the hmit

then

and the

rules as

apphed

is

thus similar to ordinary

to products of vectors are unchanged,

viz.

As

end point of the position vector r will trace out a curve

measuring length along this curve, the
differentiation process can be performed with respect to s thus,
t

varies, the

in space.

Taking

5 as a variable

dr

dy

dx,

+
^=-r'
ds
ds
Applying equation

v-7
ds-^

dz

^^ ^^^
(7.52)

+ -r^
ds

(7.9),

dr
ds

^/{dxy+(dyy+(dzy
ds
But,

(dsf
Therefore, dr/ds

is

(dx)^

+ (dyy + (dzy

a unit vector in the direction of the tangent to the curve.

further useful result can be obtained as follows.

Ua^) = (A.A) = 2a/-^

dt

dt

dt

MATHEMATICAL METHODS

218

IN

CHEMICAL ENGINEERING

^ is a vector of constant magnitude but variable direction, then A^ will

be a constant.
If

and the vector dA/dt is perpendicular to A. An example of this is uniform

motion in a circle when the acceleration is perpendicular to the instantaneous velocity.
It has been shown above that dy/ds

is a unit vector and hence differentiand using equation (7.53), then d^rjds^ must be
perpendicular to the tangent drjds. The direction of d^r/ds^ is the normal
to the curve, and the two vectors defined as the tangent and normal define
what is called the "osculating plane" of the curve.

ating again with respect to s

7.5.1.

Partial Differentiation of Vectors

upon
and a fourth independent variable time.
An increment 3T can arise from an increment in any of the four independent
variables, or from a combination of such increments.
The increment 3T
arising from a single space increment, say Sx, can be expressed as a ratio to
Sx and in the hmit as Sx^O, the ratio becomes a derivative. This particular
derivative is denoted by dT/dx which implies that the other independent
variables remain constant during the limiting process, and in this respect,
dT/cx is a "partial derivative". cT/dx is the temperature gradient in the x
direction and is a vector quantity; similarly, dT/dy and dT/dz are also vector
gradients. As will be shown in the next section, these three vectors can be
Temperature

is

a scalar quantity which can depend in general

three coordinates defining position

compounded

The other partial

a scalar rate of change.
A dependent variable such as temperature, having the properties
described above, is called a "scalar point function" and the system of variables is frequently called a "scalar field". There are many scalar point functions and amongst those of interest to chemical engineers are temperature,
concentration, and pressure.
In chemical engineering, there are other
dependent variables which are vectorial in nature, and vary with position.
into a single generalized vector gradient.

derivative with respect to time, dT/dt

Examples are

is

and mass transfer rate, which are

and they constitute "vector fields". A

velocity, heat flow rate,

called "vector point functions"

familiar illustration of a tw^o-dimensional vector field is the disposition of

iron filings showing the field strength of a magnet. The differentiation of
vector point functions will be discussed in the next section.

7.6.

Hamilton's Operator

Returning to the consideration of temperature as a function of the three

space variables, it was shown in Section 7.5.1 that the three partial derivatives of the temperature were vector gradients. If these three vector components are added together, there results a single vector gradient

7.

219

VECTOR ANALYSIS

OX

cy

oz

which defines the operator V for determining the complete vector gradient
of a scalar point function. The operator V is pronounced "del" or "nabla",
and the vector VT is often written "grad T" for obvious reasons.
V can operate upon any scalar quantity as above and yield a vector
gradient. Written out in an abstract manner,

vr can be further illustrated by taking a scalar product of

an infinitesimal increment dr in the position vector r. Thus

The nature of

Vr with

iV+i-^'+^^i
oz
dx
cy

(7-56)

and using the properties of products of unit vectors

(7.32)

and

(7.34), this

simplifies to

^T'

r^T

r^T'

dr.WT = --dx +
ox

dy + dz
cy

Comparing equation (7.57) with equation

can be seen that

(8.9)

dr.VT = dT
If vector division

were possible (which

(7.57)

oz

from the next chapter,

it

(7.58)

not) then equation (7.58) could

the ratio of dT to df, which is the generalit is

be rearranged to show that VT is

ized first derivative of a scalar variable with respect to an independent vector
variable. The expression dT/dr does not exist to represent this quantity, but
Vr is the vector equivalent of the generalized gradient.
The physical interpretation of WT can be found by taking a variable
position vector r to describe an isothermal surface.

T(x,y,z)

With

=C

of r, for any increment dr, the corresponding increment

^rwill be zero because dr Hes on the isothermal surface. But equation (7.58)
this definition

shows that

dr.VT = dT =
and thus

vr must

be perpendicular to dr. Since the vector dr can lie in any

direction in the tangent plane to the isothermal at r, and VT can only have
one value at a point r, then ^T must be perpendicular to the tangent plane
at r. This direction is the fine of most rapid change of T. Thus VT is a
vector in the direction of the most rapid change of T, and its magnitude is
equal to this rate of change.
7.6.1.

Divergence of a Vector, V.A

Because the operator V is of vector form, a scalar product can be

obtained between V and any vector A, with the interpretation that V is to be

'

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

220

treated as a vector as far as the product with

differential operator

upon everything which follows

is
it.

concerned, and as a
Thus, if A is resolved

into components, then

V.A = li\

ex

+j + k-]
cy

ozj

{AJ + AJ + A,k)

Since 7,y and k are constant both in magnitude and direction, the differential
Using the unit vector
operators only take effect upon A^, Ay, and A^.
relationships (7.32) and (7.34), therefore

dA^
dA^
dA,
^
V.^ = -^ +
+ -
^

ox

cy

(7.59)

oz

Equation (7.59) has been obtained by abstract manipulations, but the

operation (V.) on a vector has an important physical interpretation.
The continuity equation expressing the law of conservation of material
within a flowing fluid will be derived in Chapter 8 as

-d (pu) + -d (pv) + -d (pvv) + dp

^=
cy

dx

where p

is

the density,

the velocity.

(7.60)

ct

and
u

is

cz

ui

+ vj-\-wk

(7.61)

Multiplying equation (7.61) by p and applying the operator

equation (7.59) gives the first three terms of equation

(V.) to the result using

(7.60).

Thus

V.pM +

=
^
dt

(7.62)

the vector form of the continuity equation. The physical interpretation of

equation (7.62) is as follows, dpfdl is the rate of change of mass contained
in unit volume, or the rate of accumulation of mass per unit volume. Using
the general conservation law (1.25), then V.pii must equal the output less the
input, which is the net rate of mass flow from unit volume.
In general,
V.A is the net flux of A per unit volume at the point considered, counting
vectors into the volume as negative, and vectors out of the volume as positive.
If a tubular element of volume is taken such that the vectors of the field
A lie in the curved surface of the tubular element, then the only flux of A
within the element must enter and leave at the ends of the tube. If V.A is
positive, then the flux leaving the one end must exceed the flux entering at
the other end. In this respect the tubular element is divergent in the direction
of flow. For this reason, the operation V. is frequently called the "divergence" and is written

is

V.A = d\wA
7.6.2.

Curl or Rotation of a Vector,

(7.63)

Va^.

further differential function can be obtained using Hamilton's operator

by taking a vector product between the operator and another vector.

Ex-

VECTOR ANALYSIS

7.

221

panding into components and using equation (7.39) to express the vector
product gives
k
i
J

VaA =

curl

dx

The

rot

(7.64)

a ) can be explained by referassumed that the z component of

physical interpretation of the operator (V

For simplicity, it
zero and that the flow

ring to Fig. 7.12.

velocity
all

A =

cz

dy

{\v) is

is
is

two-dimensional. This implies that

and equation (7.64) simpHfies

partial derivatives with respect to z are zero,

to

Vam

'dv

du\

dx

dy)

(7.65)

dx

dy

In Fig. 7.12, an element of fluid is defined from an origin O by taking increments 3x and Sy along the axes. Using Taylor's theorem, the ;; component

of velocity at

and the x component of the

velocity at

will

be as shown.

u-^^^y
B

6y
,

Fig. 7.12.

'

d^i

Curl of a velocity vector

OA

of the fluid element is rotating in addition to its lateral and

The side
ongitudinal motion, and its angular velocity is dv/dx as may be seen from
the diagram. This rotation is in an anti-clockwise direction and by the right-

hand screw
of

OA

rule acts in the positive

direction.

Thus the angular

velocity

dv
is

IT-.

dx

Similarly, the angular velocity of

The angular

OB

du

is

k^.

velocity of the fluid element

and denoting the angular velocity by

co/r,

is

the average of these

two

values,

then

"-4 I'M)'

(7.66)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

222

Comparing equations

(7.65)

and

(7,66)

VAii
which defines

C,

shows that

2cofc

(7.67)

the "vorticity" of the fluid element.

The

vorticity

twice the angular velocity of the fluid element and this explains
is

is

thus

why (V a )

called the "rotation" or "curl" operation.

7.7.

Integration of Vectors and Scalars

In ordinary integration with respect to a single independent variable,

upon the Hmits of integration only,

because only one path of integration is possible. When alternative paths are
available in two or more dimensions, the value of an integral depends upon
both the hmits and the path. It was shown in Section 4.16.1 that for an
analytic function of a complex variable, the path of integration was irreleThese properties will now be generaHzed for three dimensions by
vant.
using vectors.
the value of an integral depends

Line Integrals

7.7.1.

An

arbitrary path of integration can be specified by defining a variable

position vector r such that

its

as

shown

in Fig. 7.13.

end point sweeps out the curve between P and

A can be integrated between two fixed

vector

Fig. 7.13.

Line integrals

P and Q along the curve r by summing the components of

along the curve. This integral can be written
points

resolved

^A.dr= \{A^dx + Aydy +

A, dz)

(7.68)

by the definition of scalar products and resolution into components.

If the integral in equation (7.68) is to depend on P and Q but not upon
the path r, then the integrand must be the diff'erential of some function 0,
i.e.

But using equation

A.dr=d(j)

(7.69)

= d(j)
(A-W(l)).di' =

(IJO)

(7.58),

dr.V(i)

.-.

(7.71)

7.

223

VECTOR ANALYSIS

by subtraction of equations (7.69) and (7.70). Since the result is to be independent of the path of integration, dr can have any direction at a point, and
{A V(j)) cannot be perpendicular to all directions. Hence,
/4-V(/)

(7.72)

This means that if a vector field A can be expressed as the gradient of a

scalar field (/>, then the fine integral of the vector A between any two points
P and Q is independent of the path taken.
If
is a single valued function, then the order of differentiation of (j)
with respect to any two variables is immaterial. Thus

d,,.

d (d<i>\

But the k component of V a

^ from

d (d<i>\

equation (7.64)

is

^(dAy_dA\
dy)

\dx
which

is

are zero

zero by equation (1.13).

if

equation (7.72)

Similarly, all three

components of V a ^

is satisfied.

Hence the general result that the line integral of equation (7.68) is independent of the path of integration if V a ^ = 0. Also, if this is true, then
A can be represented as the gradient of some scalar point function (j).
An example of the use of line integrals is the following. If the vector
field ^ is a force field so that a particle at a point r experiences a force A,
then the work done in moving the particle a distance dr from r is defined as
the displacement times the component of force opposing the displacement.
Therefore, the incremental work done (S W) is given by

dW = A.5r

{1.1 A)

Referring to Fig. 7.13, the total work done in moving the particle from P to
Q is the sum of the increments along the path. As the size of the increments
tends to zero, the sum becomes an integral, or
Q

= JA.dr

(7.75)

When

work done

independent of the path of the particle from P to Q,

and the results just enunciated
are vahd. Such a force field can be represented by the gradient of a scalar
function. When a scalar point function is used to represent a vector field,
it is called a "potential" function.
Thus, the gravitational force field is
represented by a gravitational potential function (potential energy), the
electrostatic force field by an electric potential, and a magnetic force field by
a magnetic potential. A further useful potential function will be introduced
this

the force field

is

in Section 7.10.2.

is

said to be "conservative",

MATHEMATICAL METHODS

224

IN

CHEMICAL ENGINEERING

Vector Area and Surface Integrals

7.7.2.

An element of surface has its area defined as a vector by reference to its

boundary. The magnitude of the area is defined as the maximum projected
area of the element, and the direction of the vector is normal to this plane of
projection. The direction of the positive normal is determined by the righthand screw rule applied to the sense of describing the boundary of the
element as illustrated in Fig. 7.14.

or

dS

Fig. 7.14.

When two

Vector area

adjacent elements of area are added together, their separate

vector areas are defined so that the directions of tracing the

the boundary are opposite, so that the

common

ing the

combined area

is

common

part of

traced out by ignor-

boundary. Since areas are thus defined as vectors, the vector

must be applied. Hence it can be seen that the vector area of

any closed figure must be zero, because the projected area on any plane is
doubly described, one clockwise and the other anticlockwise leading to
rule of addition

complete cancellation.

When

discussing line integrals above, the infinitesimal dr was a vector

and the type of product with the vector A had

vector over a surface,

it is

to be specified.

To

integrate a

also necessary to specify the type of product,

and

although the dot product is used again, the physical interpretation is different.
If there is a vector field denoted by A and the element of surface is
denoted by dS = n dS where n is the unit vector normal to the surface
element, then the surface integral is usually defined as

JA.dS = JA.ndS

When ^

is

a force

field,

(1.16)

the above integral gives the total force acting

surface, so that if the surface

is

on the

closed, the integral gives the resultant force

to the integrated effect over the surface. If the vector A is

replaced by the velocity vector u, then the surface integral defined by equa-

on the body due

tion (7.76) gives the net volumetric flow across the surface.

possible to integrate a cross product between

It is

very

little

7.7.3.

practical value

and

is

dS, but this has

Volume Integrals

The element of volume dx dy dz

shape

A and

usually ignored.

is

usually denoted by da so that

its

not restricted in any way. Although both the elements of length

(dr) and surface (dS) are vectors, the element of volume (da) is a scalar
quantity, and thus there is no cause to specify any type of multiplication.
is

7.

VECTOR ANALYSIS

225

There are relationships between these three types of integration and the
following sub-sections will be devoted to them. If the three integrals are
considered in terms of cartesian coordinates, it can be seen that a surface

two

integration involves

linear integrations, whilst a

involves three hnear integrations.

It

volume integration

seems obvious therefore that any

relationship between the different types of integration will involve a

number

of hnear differentiations to match the orders. If these general ideas can be

assimilated, the following relationships may appear to be more logical.
7.7.4.

Stokes' Theorem

Consider any continuous surface S having elements dS defined in the

way and denote by C the curve describing the boundary of the surface
S. The positive direction of the curve C is related to the positive direction of
the normal vector n as given in the definition of vector area in Section 7.7.2.
Stokes' theorem states that if there is a vector field A then the Hne integral
of A taken round C is equal to the surface integral of V a ^4 taken over S.
.
.
.
SymboHcally,
=
=
Va/1.(/5
(7.77)
J.Va^JS
J^.(/f
J
usual

can now be seen that the result of Section 7.7.1 is a special case of the
above theorem. That is, if V a ^ = 0, then the line integral of A around
any closed curve is zero. Equation (7.77) is much more general than this
previous result, and is frequently used in vector analysis. Similar results are
available for other Hne and surface integrals but they are of rarer occurrence.
The formulae are:
^
It

/.

Udr = jttAVUdS

(7.78)

JA.dr = jnAV.AdS

(7.79)

- JAAdr = j(nAV)AAdS

(7.80)

are taken along the curve C which defines the boundary

over which the surface integrals are taken. The negative
sign is necessary in equation (7.80) because the vector A has been moved
from the first to the last term of a vector product so that it follows the
operator V.
Stokes' theorem can be applied to a two-dimensional system by putting

where

all line integrals

of the surface

A=Pi + Q
where

P and Q

are functions of

dr

The normal

to the

xy plane

is

x and y only and

dxi

-\-

dyj

in the z direction,

ndS = dxdyk
and using equation

(7.64) gives

fdQ
~''\dx

Va^ _

the elementary length

dP\

dy)

hence

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

226

Putting these various forms into equation (7.77) and simplifying gives

jiP,.^Q,y)=.jj(^-')d.dy
c

which was used

(7.81)

in Section 4.15 to

prove Cauchy's theorem.

Gauss' Divergence Theorem

7.7.5.

When
shown

the continuity equation

that W.pu

is

was considered in Section 7.6.1 it was

from unit volume. It was also

the net rate of mass flow

stated in Section 7.7.2 that the surface integral of the velocity vector u gives

the net volumetric flow across the surface.

If these

two ideas are now com-

bined by applying them to a closed surface S enclosing a volume o-, the

following result is obtained by equating the two expressions for the mass
flow rate.

j pu.dS
Equation (7.82)

is

valid for

= jv.puda

any vector

(7.82)

giving

JA.dS = jW.Ad(7
5

(7.83)

which

is the "divergence theorem".

Again, this equation can be extended to other products giving the set of
formulae

jnUdS = jVUda

(7.84)

jn.AdS = jV.Ada

(7.85)

\nAAdS = jVAAda

(7.86)

where the surface integrals are taken over the surface S enclosing the volume
(7 over which the volume integrals are taken.
7.7.6.

Green's Theorem

and V are two continuous, diff"erentiable scalar point functions,

then an application of Gauss' divergence theorem (7.83) to the function
If

f/VK

gives

UVV.dS =

The right-hand

side

V.(L/VK)t/c7

can be expanded as the derivative of a product, thus

V.(C/

V)

= {VU).{VV)+UV^V

Substituting equation (7.88) into equation (7.87)

Interchanging

(7.87)

and rearranging

^VU.VVdG = UVV.dS - j UW^Vdcj

J
U and V throughout the above argument gives
jyU.WVd(j = j VVU.dS - VSJ^Uda
J

(7.88)

gives
(7.89)

(7.90)

7.

227

VECTOR ANALYSIS

Equating equations (7.89) and (7.90) and rearranging the integrals gives

j(UWV-VVU).dS = j{UW^V-VV^U)d(T

(7.91)

Equations (7.89, 90, 91) are known as various forms of "Green's theorem".
will be appreciated in the next chapter, the differential equation

As

V2f = V.VF =

(7.92)

of frequent occurrence, and equation (7.91) can be very useful for dealing
with such systems. It is of even more use in the theory of electricity and
magnetism because both the electric and magnetic potentials satisfy equation

is

(7.92).
7.8.

Standard

Identities

Any

of the following equations can be verified quite easily by using the

formulae given in the earher parts of this chapter, but it is useful to have the
following hst available for reference purposes. This first group of equations
arises from differentiating various products.

V.UA = UV.A + A.VU

VaUA = UVaA-AaWU
V.AaB = B.VaA-A.WaB
Va(AaB) = B.VA+AV.B-A.VB-BW.A

(7.93)
(7.94)
(7.95)
(7.96)

In the derivation of the next useful formula, the dual nature of V causes
some difficulty in the notation, but if it is understood that when no symbol
follov/s V, the

symbol immediately preceding

it is

to be differentiated, then

the intermediate steps can be followed.

^ A (V A B) = ABV-A^B
Ba(WaA) = B.AV-B.VA
VA.B = B.AV + A.BV

and

(7.97)
(7.98)
(7.99)

EUminating the two expressions on the right-hand side of equation (7.99)

and rearranging gives the required formula; the notation now being normal.

VA.B = A.WB + B.VA+Aa(WaB) + Ba(VaA)

If

now B =

(7.100)

A, equation (7.100) gives

iWA^ = A.VA + AA{VAA)

The next group of formulae

is

(7.101)

intended to simplify expressions involving

second derivatives.

Va(Va^) = VV.^-V^^
VaV.^ = V.Va^ =0
VaVL/ =

(7.102)
(7.103)
(7.104)

These last two equations are only valid when the order of differentiation
not important in the second mixed derivative.

is

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

228

Curvilinear Coordinate Systems

7.9.

So far in this chapter, vector equations have been developed which are
independent of the coordinate system used. Position vectors only require
an origin to be defined as stated in Section 7.3.2 and thus the vector equations
are very general. On occasions, a vector relationship has been illustrated by
resolving the vectors into a cartesian coordinate system by using the unit
vectors /, y, and k. This particular set consists of unit vectors which remain
constant in both magnitude and direction, but in many problems the cartesian set of coordinates is not the most convenient. Both cylindrical polar
coordinates and spherical polar coordinates are frequently used in chemical
engineering problems and it is therefore necessary to be able to express a
vector relationship in terms of other coordinate systems. Books devoted to
vector analysis! invariably derive the necessary equations for a general coordinate system and the reader is referred to these, but all of the necessary
equations will be developed here for spherical polar coordinates, and quoted
for cylindrical polar coordinates.
One important point which must be
accepted here is that in any coordinate system other than cartesian, the
unit vectors are variable in direction. For example, when a radial coordinate
is used, the value of the variable increases in opposite directions at the
opposite ends of a diameter, and comparing the two positions the unit radial
vector is also reversed. Hence extra terms arise as a result of differentiating
the unit vectors.

7.9.1.

Spherical Polar Coordinates

These can be defined

illustrated in Fig. 7.15.

relative to

The

the cartesian coordinate system as

radial distance of the point

Fig. 7.15.

P from

the origin

Spherical polar coordinates

denoted by
the polar angle of OP relative to the z axis is denoted by 0,
and the azimuthal angle of the plane containing OP and the z axis relative
to the xz plane is denoted by cj). Care must be taken to ensure that (r, 6, (p)
is

is

/*,

a right-handed set of coordinates as further illustrated in Fig. 7.16.

t

Weatherburn, C. E. " Advanced Vector Analysis." Bell

&

Sons,

London

The

(1951).

VECTOR ANALYSIS

7.

volume element

is

229

defined by taking increments Sr, SO, and

d(j)

in the coordi-

nates as before, but now, the edges of the element are in general curved
their lengths are not equal to the simple increments.

r sin 66(f)

<5

^^v

and

If a set of unit vectors

soM^XXl'
Fig. 7.16.
a, b, c are

Volume element

in spherical polar coordinates

defined at the point P, then the incremental position vector dr

describing a general displacement from

dr

now

3ra

P can be

written

rS9b-\- rsmed(l)c

(7.105)

expand Hamilton's operator (V) in terms of the new

coordinate system. Equation (7.58) gives the general result that
It is

possible to

dr.V

(7.106)

Equation (7.105) gives an expression for dr in the limit as Sr-^Q, and the
expression for (V) which satisfies the identity (7.106) is
d

or

h d
r

00

(7.107)

sm

dcj)

The various differential operations, grad, div, and curl, can now be
expressed in terms of spherical polar coordinates by using this identity. The
gradient of a scalar point function U, is simply obtained by placing U after
each term in equation (7.107). Thus

Vl/

dU bdU
= a + - +
dr

cU

dU
-^-
c

sm 9

(7.108)

d(p

Before developing expressions for the divergence and curl of a general

it is necessary to determine them for the unit vectors a, b, and c.
Putting U = O'm equation (7.108) gives

vector,

V0 =
because

r,

and

(/>

are independent.

= VAV0 =

-(VAfc)

= -(VAfe)
r

bir

(7.109)

Substituting into equation (7.104) gives

VA(6/r)

fcAV
.

Aa
5

dr

(7.110)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

230

where equation (7.108) has been used again in the last line. Provided (r, 6, (j))
and hence (a, b, c) is a right-handed set of axes, equation (7.110) can be
simplified and rearranged as follows.

= -(VA6) + c(-l/r2)

VA6 =

.-.

Similarly,

it

c/r

(7.111)

can be shown that

VAa =
VAC = (acote-b)lr
in

(7.112)

(7.113)

To find the divergence of the unit vectors, the above expressions are used
conjunction with equation (7.95). Thus

= V.(6ac)
= c.Va 6. Vac

V.a

.-.

Similarly,

=-+

V.a

2/r

(7.114)

V.ft

(cot0)/r

(7.115)

V.c

(7.116)

Assuming that the vector A can be resolved into components

b, and c as
A = A,a + Aeb + A^c

terms of

in

fl,

(7.117)

and using the above seven equations, therefore

,

V.A =

dA,
dA.
IdAe
A + Aocote

+ +
+
rsinO

dr

dO

d(j)

(^sin0)

and

WaA = -Aq +
,

acotO

dAr

dO

dA^

rsinQ

d(j)

(7.118)

dA,

-\

rsmO

dAc

dr

d(f)

dA,

rsmO

VaA

dr

dcj)

dO

(A^smO)
rsinO

dO

dA,
rsinO

d(t)

^ 5

dr

(rAe)

dr

dO]
(7.119)

7.

By applying equation

VECTOR ANALYSIS

(7.118) to the expansion of

(7.108), the following expression for

d /

V^U

dr

231

.dU\

d /

Vt/ as given by equation

V^C^can be obtained.
.
.

ac/^
^dU\

d^U

(-f)

^^^^^^

sin^ e 50'

The expansion of V^^

more complicated than equation

is

(7.120)

and

is

not

usually given as a standard formula. It can be obtained however by expand-

ing equation (7.102) by two appHcations of equation (7.119) and one application of each of equations (7.118) and (7.108).
7.9.2.

Cylindrical Polar Coordinates

Following the same procedure as above, the following formulae can be

derived for cyHndrical polar coordinates

Fig. 7.17.

illustrated in Fig. 7.17.

same azimuthal angle

dr

is

(co,

</),

z)

which are defined as

Cylindrical polar coordinates

one of the cartesian coordinates, and

the

(p is

as for spherical polar coordinates.

+ coS(j)b + dzc
dU b dU
dU

dcoa

(7.121)

(7.122)
OCO

oz

CO C(j)

(7.123)
CO ceo

Va^ =a

CO

oz

d(p

l^dA,

+
do)

CO dcj)

m dco\
Again,

OCO/

'

7.10.

engineering

[i^-it] "-'
(7.125)

dcp-'

V^A must be evaluated by

One of

oz

using equations (7.102, 122, 123, 124).

The Equations of Fluid Flow

the most important applications of vector analysis in chemical

is

to the study of fluid flow.

vector quantity, and in a

moving

As

stated previously, velocity

is

fluid the velocity is a vector point function.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

232

Ignoring the heating effect due to the friction in a moving fluid, and assuming
no mass transfer is taking place, there are nine scalar variables related
by the fluid flow equations. There are three coordinates and time which are
the four independent variables; and in addition to the three velocity compothat

making five dependent variables. It will

be assumed that the fluid is newtonian so that the coefficient of viscosity
remains constant. Five equations are needed because there are five dependent
variables, and one of these, the continuity equation (7.62) has already been
mentioned. A second equation is the equation of state relating the density
to the pressure which leaves three further equations to be supplied. Instead
of treating the three velocity components separately, the use of a single
vector velocity variable will reduce the requirement to one vector equation
and this is how vector analysis effects a simplification. This equation is
obtained by applying Newton's second law of motion to a fluid element.
nents there are pressure and density

That is, the external forces, the pressure forces, and the viscous forces are
combined and equated to the product of the mass of the element and its
acceleration. The details of the derivation are given in many books devoted
to fluid flowt and in any text book devoted to theoretical hydrodynamics.
For an incompressible fluid, the equation of motion can be written
-^

+ u.Vu= -

ot

-Vi? + vV^ii + F

(7.126)

which is known as the "Navier-Stokes equation". The advantage of the

above vector form is that the five terms can be easily identified as: time
dependent, inertial, pressure, viscous, and external force terms respectively.
The continuity equation for such an incompressible fluid is the foUowing
simplified version of equation (7.62).

V.M

F representing

the external forces

is

usually due to gravity

(7.127)

and

gives rise to

the hydrostatic pressure in the system, and if this is the case, F can be
expressed as the gradient of a scalar point function and combined with the
pressure term. It wiU therefore be ignored in the remainder of this work.

There

no general method of solution for the system of equations (7.126)

and most attempts have centred on either of two simplifying
assumptions. These theories will now be considered separately.

and

is

(7.127)

7.10.1.

Stokes' Approximation

If it is assumed that the fluid velocity is everywhere small, the inertia

term in equation (7.126) can be omitted. Without this term, the system of
equations is linear and can therefore be solved. To find out under what
conditions the approximation is justified, it is convenient to express the
equations in dimensionless form by taking the ratio of all lengths to some
t For example, Knudsen, J. G. and Katz, D. L. " Fluid Dynamics and Heat Transfer."
McGraw-Hill (1958). Lamb, H. "Hydrodynamics." Cambridge University Press (1895).
Milne-Thompson, L. M. " Theoretical Hydrodynamics " (2nd edition). Macmillan (1949).

7.

233

VECTOR ANALYSIS

and the ratio of all velocities to some characteristic

Denoting the dimensionless variables by ('), equation (7.126)

characteristic length (L)

velocity (U).

becomes

In this dimensionless equation, three dimensionless groups have arisen:

Ut/L, which is the dimensionless time, 2p/pU^, the dimensionless pressure
coefficient, and LU/v, the Reynolds number. It is conventional to introduce
a factor of 2 into the pressure coefficient and sometimes into the Reynolds
number, although in the latter case a convenient characteristic length is often
defined to include this factor.

The approximation of neglecting the

inertia term in comparison with the

be interpreted as assuming that the Reynolds number
is small compared with unity. Returning to the dimensional equation (7.126)
and making the approximation gives

viscous term can

now

^=

_lvp + vV'ii

ct

(7.129)

Without vectors, the system of equations (7.127) and (7.129) would contain
four dependent variables, and a lengthy piece of algebra would be required
to find a single differential equation involving only one dependent variable.
However, a simple vector method is available. Operating on equation (7.129)
with (V.) gives
r)

-(V.ii)= --V^/j + vV^V.ii

ct

(7.130)

But the continuity equation (7.127) shows that V.u

0.

V^p =

.-.

(7.131)

and the pressure distribution can be found in general terms. Unfortunately,

the boundary conditions usually govern the velocity and not the pressure,
and thus equation (7.131) is not really useful. Hence, operating upon equation (7.129) with (V A) gives
d

ot

-(Vau)= --VaVp + vV^Vaii

Now

equation (7.67) defines the vorticity of the fluid by

Vaii
and equation

(7.104)

is

(7.67)

shows that the pressure term

I
which

vV^f

is

identically zero.

Hence
(7.132)

analogous to the heat conduction or mass transfer equation, except

is a vector equation whereas both of the others are

that equation (7.132)

scalar equations.


MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

234

Equation (7.132)

will

now

be put into a simpler form for the special

case of fluid flow with an axis of symmetry by introducing the stream function ip so that the continuity equation (7.127) is automatically satisfied.
Taking spherical polar coordinates and assuming that the problem has an

symmetry (the z axis) then from any point P, curves Q^ and Q2 can
be drawn to a point on the z axis as illustrated in Fig. 7.18. By rotating the
plane of the diagram about the z axis, the curves 2i and Q2 become surfaces.
Now the total volume of fluid crossing the surface Q^ from right to left
must equal the volume crossing Q2 in the same direction. Also, because
there is no flow across the axis of symmetry, there will be an equal volumetric flow across any other surface such as Q^. This volumetric flow is deThis definition shows that
is Stokes' stream function.
noted by 271 \l/ where
= at all points on the z axis, and elsewhere it is a scalar point function.
axis of

ij/

i//

Stokes stream function, axi-symmetrical flow

Fig. 7.18.

should be noted that the above definition of

is related to a direction
and in a generahzed non-symmetrical system, a vector stream
function can be defined by
It

ij/

in Fig. 7.18

Va^

(7.133)

but the specialized scalar stream function will be used here.

Taking a smaU displacement from P in each of the coordinate directions,
it can be shown that
u

-,

^ ^^T-,

--

(7.134)

where u and

v are respectively the r and

components of the velocity u.
Putting equations (7.134) into equation (7.118) shows that the continuity
equation is satisfied. The vorticity can be expressed in terms of the stream

function by substituting equations (7.134) into equation (7.119).

^^
c

deWsinOdO

r[dr\smedr)

raV

rsin^L^''^
.-.

sin^ a /
'

r^

#Y

^^ Vsin " ""

Crsin0

Thus

= cV

'

(7.136)

where E^ is the differential operator in the brackets of equation (7.135).

Equation (7.136) shows that the vorticity has a component in the (p direction
only and the scalar symbol C can be used to represent this component.

235

VECTOR ANALYSIS

7.

Applying equation (7.119) to the vorticity gives

VAC = -^-^(Csin0)--|-(Cr)
r sin 6

dO

(7.137)

or

Using equation (7.119) again gives

(7.138)
}:

Combining equations
steady

(i.e. d(,ldt

(7.127, 102, 132, 138)

and assuming that the flow

is

0) gives

dr^

dO [sm 9 dO

which can be rearranged to give

E^

iCr sin 6)

V =

or

(7.139)

which

is the fourth order linear scalar partial differential equation describing

the steady axi-symmetrical flow of an incompressible fluid at low Reynolds

numbers.
The above equation
past a sohd sphere.
7. 10.2.

It is

will

be solved in Section 8.5.4 for the case of flow

The Ideal Fluid Approximation

assumed

the Reynolds

in this case that the viscosity of the fluid

number

is

large.

is

small,

i.e.

that

Thus, neglecting the viscous term in equa-

tion (7.126) gives

+ ii.Vw=--Vp
^
dt
p
Using equation (7.101) to

alter the

(7.140)

form of the middle term and rearranging

gives

V^(i"'

+ ^) =

A (V

A II)

(7.141)

If the flow is steady, and the vorticity is zero, then the right-hand side of
equation (7.141) vanishes and the equation reduces to BernoulU's theorem.

p + ^u^

From

the above analysis

strictly true

when

it

const.

can be seen that Bernoulh's theorem

the laminar flow

is

(7.142)
is

steady, incompressible, inviscid,

only

and

irrotational.

Applying the operator (V a) to equation (7.141) and using the property

(7.104) to simplify the left-hand side gives

= Va(iiaC)-^

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

236

and expanding the vector

product and remembering that

triple

V.;

= V.Vam =

+ ii.VC =
^
ot

gives

(7.143)

Equation (7.143) shows that the substantive derivative

(see Section 8.2.3) of

This means that the vorticity of any fluid element
remains constant, and if the fluid motion has been started from rest, then
the vorticity will be zero everywhere for all time. Thus, for a fluid motion
started from rest, the vorticity is zero and the flow is irrotational.

the vorticity

is

zero.

VA

/.

11

(7.144)

However, it was shown in Section 7.7.1 that if the curl of a vector is zero,
then the vector itself can be expressed as the gradient of a scalar point
function called a potential function. Thus, for an inviscid irrotational fluid,
the velocity can be expressed as the gradient of the ''velocity potential" </>.

Thus

u=

Combining equation

,^

Ars

(7.145)

-Vcf)

(7.145) with the continuity equation (7.127)

shows

that the velocity potential satisfies Laplace's equation

V^0 =

(7.146)

This second order linear partial differential equation will also be solved in
Section 8.5.4 for the case of ideal fluid flow past a sohd sphere.
It is possible to express ideal fluid flow in terms of the stream function
i/^, and in the axi-symmetrical case equation (7.136) becomes

V =

(7.147)

because the flow is irrotational. Although the stream function representation is vaHd for both Stokes' flow and ideal fluid flow, the velocity potential
can only be used in ideal fluid flow. In viscous flow, equation (7.144) is not
valid, the system is not conservative, and energy is dissipated by the action
of viscosity. Therefore, for comparison of the two solutions, both must be
expressed in terms of the stream function.

7.11.

The

Transport of Heat, Mass, and Momentum

partial diff*erential equation

derived from

first

will be
be derived

which governs heat transfer

principles in the next chapter, but here

it

will

by a vector method for comparison. The rate of flow of heat per unit area at
any point is proportional to the temperature gradient at that point; the
constant of proportionality being the thermal conductivity.
Using the
generalized gradient from Section 7.6 gives

^=-/cVr

(7.148)

ot

The
is

definition of the divergence operator in Section 7.6.1 shows that V.Q

the net flow of heat from unit volume. However, the total heat content of

VECTOR ANALYSIS

7.

unit

volume

is

pCpT where Cp

states that heat

is

can only leave

Thus

237

and the conservation law

expense of the remaining contents.

the specific heat,

at the

--(pC,T) = -(V.Q)

(7.149)

Since space derivatives and time derivatives are independent, they are
Hence, operating upon equation (7.148) with (V.), assuming
that the physical properties remain constant, and substituting into equation

commutable.
(7.149) gives

^T

dt

Cpp

.^

= aV'r

or

(7.150)

ct

where a

the "thermal diffusivity".

is

In the transport of mass by molecular diffusion the equation analogous

to (7.148)

is

dN

=-DVc

(7.151)

dt

where A^

is

the molar flux density,

concentration.

shown

is

the diffusivity, and c

is

the molar

Following the same argument as for heat transfer,

it

can be

that

^ = DVh

(7.152)

Equations (7.150) and (7.152) have been derived for systems without
bulk motion, if a bulk flow is included in the derivation the time derivatives
have to be replaced by the substantive time derivatives (see Section 8.2.3)
and collecting the equations (7.126, 150, 152) together, the equations governing the transport processes are

+ M.Vr = aV^r

(7.153)

~ + u.Vc = DVh

(7.154)

-?
dt

+ M.Vw = vV'i/--Vp + F

(7.155)

These equations are seen to be very similar in vector form, yet equation
(7.155) is quite different, having two extra terms, and being the only vector
equation. The analogy between heat and mass transfer however is complete
if a and D have the same value. These points are discussed more fully in the
literature, t
t Klinkenberg, A. and Mooy, H. H. Chem. Engng Prog., 44, 17 (1948). Garner, F. H.,
Jenson, V. G. and Keey, R. B. Trans. Instn Chem. Engrs, 37, 191 (1959).

Chapter 8

PARTIAL DIFFERENTIATION

AND

PARTIAL

DIFFERENTIAL EQUATIONS

8.1.

Introduction

After showing how conservation laws and

rate equations could be used to

derive algebraic equations in Chapter

method was extended

1,

the

to systems

Thus
described by an independent variable, either a distance or time.
ordinary differential equations were derived by considering the equihbrium
of an infinitesimal element and using Taylor's theorem to introduce the
In this chapter, problems requiring the specification of more
than one independent variable will be considered. An example of such a
problem is the change of the temperature distribution within a system as the
steady state condition is approached; when the temperature is a function
both of position and time. The diflferentiation process can be performed
relative to an incremental change in the space variable giving a temperature
gradient, or with respect to an increment of the time variable giving a rate of
temperature rise. These two first derivatives of the temperature have to be
distinguished from each other, and the first part of this chapter clarifies the
meaning of a derivative in a multi-variable system.
The formulation of partial differential equations follows the same rules
as given in Chapter 1, but the formulation of boundary conditions requires
more attention. The second half of this chapter is devoted to the solution of
partial differential equations by three methods.
Studying the boundary
conditions will often suggest a special kind of solution and this is the first
method. The more fundamental methods of separation of variables and
operators form the major part of this chapter and in chemical engineering
there are probably more examples of the use of these methods than any
other mathematical method.
In difficult problems, such as fluid flow, it is often advisable to formulate
the problem in cartesian coordinates or find the vector form of the differential equation directly. The vector equation can then be expanded into a coordinate system which better suits the shape of the boundaries of the problem.
For example, it is easier to specify a spherical surface by r = a in spherical
polar coordinates than by x^+y^ + z^ = a^ in cartesian coordinates. Thus
it can be seen that in problems involving more than one independent variable, the boundary conditions play an important part in determining the
coordinate system for the problem.
It will be shown later that the boundary conditions also influence the
derivative.

238

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

239

method of solution. Therefore the boundary conditions are more

important in the study of partial differential equations than ordinary differen-

choice of

equations.

tial

8.2.

Interpretation of Partial Derivatives

Figure 8.1 shows curves of the system

u=f(x,y)

(8.1)

At each point in the xy plane, a value of u

(8.1) and each curve is obtained by conwith equal values of u. Thus Fig. 8.1 can be interpreted as a

plotted with w as a parameter.

can be calculated from equation

necting points

Fig. 8.1.

First partial derivatives of

= /(jc, y)

map where

u represents height measured perpendicularly to the xy

x is allowed to vary whilst y remains constant, then in general u will
vary and the derivative of u with respect to x will be the rate of change

contour

plane. If
also

of u relative to x, or the gradient in the chosen direction.

This derivative

is

written

fX
where y

placed outside the brackets to imply that y must be held constant

while the derivative is being evaluated.
Similarly, the gradient along a hne parallel to the y axis (x held constant)

is

is

written

du\
(8.3)

dyjx

In Chapter 7 it was shown that V is a vector along the Hne of greatest

slope and has a numerical value equal to that slope. This vector is also
illustrated in Fig. 8.1.

In the present case, z

is

not a relevant variable and

MATHEMATICAL METHODS

240

CHEMICAL ENGINEERING

IN

equation (7.54) gives

Vw =
the gradient in the

du

du

ex

oy

i +7 =

gradu

(8.4)

x direction can be reclaimed from equation

the scalar product (Section 7.4.1) with the unit vector

(8.4)

in the

by taking
direction.

Thus
i.^u

du\dx

(8.5)

Using the appropriate unit vector the gradient can be found in any direction.
If br is an incremental displacement from P so that

br=^ibx^-jby

(8.6)

then u will change by (duldx)Sx due to the change in x, and by (du/dy)by

due to the change in y provided bx and by are small.

^"

= (r:)^-(|)*'

and in the hmit as bx-^dx and by-^dy, bu^du

where
du
is

0-0

= {--]dx +

]dy

(8.8)

called the "total differential" of u.

In general,

if

is

a function of

du\

= f -
which

is

Jxi

many

variables Xj, ^2, X3,

du\
-c/x2

x,

then

/ cu

+...+

the fundamental equation of partial differentiation.

be noted that while a partial differential coefficient such as (cu/dxi)

being evaluated, all other independent variables, X2, x^,
x must be
held constant. These other variables should be placed outside the brackets
enclosing the derivative but in most cases it is obvious which variables must
be held constant and they are omitted for brevity.
It will

is

8.2.1.

Relationships between Partial Derivatives

Returning to functions of one independent variable

y=f(x)
then dy, the infinitesimal change in y,

''

is

related to

[i)'^

(8.10)

dx by
(8.11)

Sometimes, y is not given as an explicit function of x such as equation (8.10)

but as an implicit function thus
g(x,y)

(8.12)

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

241

Considering g as a third variable temporarily, equation (8.9) can be used to

give

where dg must equal zero to satisfy equation (8.12). By rearranging equation

an expression for dy can be obtained and compared with equation

(8.13)

(8.11) to give

d_f^_{^gldx)^dy
dx

dx

(dg/dy)

'

One important

fact concerning partial derivatives is evident from equation

symbol ''dg'' cannot be cancelled out because at least a sign
mistake would be caused. Thus the two parts of the ratio defining a partial
derivative can never be separated and considered alone. This is in marked
contrast to ordinary derivatives where the infinitesimals dx, dy can be treated
(8.14); the

separately.

To demonstrate

the truth of equation (8.14), consider the fol-

lowing example.

Example

1.

If

x^-hy^

a^, find dy/dx.

The above equation can be solved

y

for

giving

(a^-x^)^

Differentiating in the usual way,

dy

^ K-2x) ^

-X

dx

(a^-x^f

(fl2-x')^~

Alternatively, the original equation can be written

^(x,>;)

x2

+ /-fl2 =

III

Differentiating equation Til partially with respect to

x keeping y constant

gives

2x

IV

2y

8x
dg
Similarly,

dy

Substituting the values given by equations

IV and

into equation (8.14)

gives

dy

_
dx"
and eliminating y by using equation

Thus, equations
(8.14).

II

and VI are

2y

I,

dy_
dx

2x

(a-

'-x^f

identical, illustrating the validity

VI
of equation

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

242

The above

ideas can be readily extended to functions of

two independent

variables by generalizing equation (8.12) to

z

9(x,y)

(8.15)

and hence

Equation (8.15) can also be put into the implicit form

F(x,y,z)

(8.17)

and thus

^^=0^^^ (1)^^+0^^=

(-i^>

Rearranging equation (8.18) gives

(dF/dx)

and comparing equations

(8.16)

and

(dFldy)

(8.19) gives

dg

dF/dx

dg

df/dy

dx^ ~

dF/dz'

dy^ "

df/dz

Considering the manner in v^hich a partial derivative

(8.20)

is

obtained,

it is

clear

that

0(1)='
and ehminating (df/dz) from the two equations

dxj \dgj
If z

is

(8.20) gives

(8.22)

\dxj \dF

held constant in equation (8.18), then dz

d_y__ (dF/dx)
dx~ ~(dFWy)

and
(8.23)

Equating (8.22) and (8.23) and rearranging gives

or using the symmetrical version of equation (8.15), viz. (8.17), equation

can be written

(8.24)

(i)(i)(l)=-'
giving further indication that partial derivative signs cannot be cancelled out.

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

8.2.2.

243

Changing the Independent Variables

Referring to equation (8.15) where x and y are the two independent

it may be desired to change to new independent variables u and v
where the old and new variables are related by

variables,

Applying the general formula

<t>{u,v)

(8.9) to

dx

(8.26)

equations (8.26) gives

= z-du + ^dv
du

dv
(8.27)

dy
^

du

-r-

+ - dv

du

dv

which can be substituted into equation

(8.16), thus

\dx/

\dy/

The

dv

\_du

partial derivative of z with respect to u

stant,

and hence dv

is

^Jw + ^jJ
du

dv

(8.28)
^

obtained by keeping v con-

in equation (8.28).

dz

dg

du

dx du

dg

d(j)

d\j/

dy du

or using the alternative nomenclature to clarify the equation,

dz
= dz dx + dz dy

c-

du

dx du

dy du

^^

dz dx

dzdy

(8.29)

, ^x

Equations (8.29) and (8.30) can be extended to any number of independent

variables.

Thus

if

Z =/(Xi,X2,X3,

...

xj

Xi=gi(Ui,U2,U2,...uJ
^2

^2("l.2.W3.---0
etc.

dz

dz dx.

dz

dx-y

dz dx

cui

oxiOUi

0X2 du^

dxdui

The number of

variables

variables n, but

m=

the usual case.

8.2.3.

On

is

need not be the same as the original number of

Independent Variables not Truly Independent

occasions there are restricting relationships between the basic indeFor instance, the composition of the vapour leaving a

pendent variables.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

244

binary liquid is a function of temperature, pressure, and liquid composition;

which are three independent variables. However, a restriction that the liquid
is boiling is frequently apphed and this gives a further equation relating
temperature to pressure and liquid composition. This case can be stated
symbohcally thus,

Applying equation

(8.9) to

y=f(x,ZP)

(8.32)

T=

(8.33)

g(x,P)

both of these gives

df

dT =

and

-^ dx

ox
Substituting for

dT from

df

df

-:^dP

(8.35)

oP

equation (8.35)

into

(8.34)

and

rearranging

gives

and here it is not advisable to replace (dg/dx) with (dT/dx) because the latter
term is ambiguous. It could be obtained from equation (8.32) at constant
y and P, or from equation (8.33) at constant P, yielding different results. If
the data is collected at constant pressure, dP = 0, and equation (8.36)
becomes
\Sx)
or

where the derivatives with one variable held constant are obtained from
equation (8.33), and the others from equation (8.32).
The above system is sometimes studied at constant temperature which
precludes the ehmination of dT between equations (8.34) and (8.35) because
dT = 0. However, a similar method will show that
'dy\

FxjT

A
perty

time

If a

fdy\

/dy\
.

fdP\

C4,r-(ik.e-:).

is when a fluid protemperature T) is a function of three coordinates (x, y, z) and

The temperature increment determined from equation (8.9) is

further case of interest in chemical engineering

(e.g.

(/).

path

is

traversed across the system so that x, y, and z

become functions

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

245

of time only, equation (8.40) can be divided by dt to give

dT

/dT\dx

/dT\dz

fdT\dy

/dT\

^^

^^'

etc., are the velocity components which determine the progress

along the path and dT/dt is the "total time derivative" of the temperature.
A special case of equation (8.41) is when the path of a fluid element is

where dx/dt,

traversed so that dx/dt

of fluid velocity.

D/Dt named

The

u,

dy/dt

v,

and dz/dt

total time derivative

the "substantive derivative".

is

w, are the

components

then given the special symbol

Thus

DT _ /dT
~Dt

\dx^
='-'-(f)-(g)-f

<-'

or in vector form,

DT dT
+ .Vr
-=^

(8.43)

The distinction between DT/Dt and dT/dt can be clarified by considering

At any fixed point in the
a heat exchanger operating at steady state.
exchanger the temperature of the fluid passing the point is constant, thus
dT/dt = 0, since by definition this derivative is taken at fixed values of the
coordinate variables. However, any element of fluid undergoes a temperature rise or fall as it passes through the exchanger, and as far as the fluid
element is concerned DT/Dt ^ 0. Thus it is important to distinguish between the equilibrium of an element of space and the equihbrium of an
element of fluid.
An example of the use of equation (8.41) is the rule for differentiating a
product. Thus if
y
where u and

v are functions of

uv

(8.44)

only, appUcation of equation (8.41) gives

dy

dy du

dy dv

dx

du dx

dv dx

Evaluating the partial derivatives by differentiating equation (8.44),

dy

dx

8.3.

du

dv

dx

dx

Formulating Partial Differential Equations

The method whereby a

physical situation can be represented by a partial

equation was described in Section 1.13 although the details were
only described for ordinary differential equations. The first difliiculty in
describing the model is determining the number of independent variables

diff'erential

needed to specify the system. The normal maximum number is four; three
coordinates defining position, and time. If the problem contains either a

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

246

plane of symmetry or a line of symmetry, one coordinate can be dispensed

with in the appropriate coordinate system. Cartesian or cylindrical polar coordinates are suitable in cases of plane symmetry, whereas either cylindrical
or spherical polar coordinates are suitable for describing axial symmetry.
In Chapter 1 the space element was defined by the two planes at x and
jc + (5a:, whereas in the general three-dimensional case the space element is
defined by taking an increment in each of the coordinate directions. The
relevant conservation law is still applied to this new space element but the

inventory of inputs and outputs must include the behaviour at all six faces
of the space element, and not just two of the faces. The space element in
spherical polar coordinates has already been illustrated in Fig. 7.16 where
it can be seen that each face of the element is defined by keeping one of the
coordinate variables constant, and each edge of the element is defined by
keeping two coordinates constant. Thus, a version of Taylor's theorem can
be applied along each edge of the element, and the value of the dependent
variable can be estabHshed at each corner of the space element in terms of
its value at one corner and the three partial derivatives constituting the
Laplace operator.
These ideas can be better understood from actual examples, and in the
following sub-sections a few of the more common partial differential equations will be derived. The first derivation will be given in full to clarify the
method, but the later derivations will include many of the short cuts anticipating the result.
8.3.1.

Unsteady State Heat Conduction

in

One Dimension

Figure 8.2 illustrates a section of a flat wall of thickness L ft whose height

large compared with L, If the temperature distribution
uniform throughout the wall at zero time and heat is supplied at a fixed

and length are both

is

Fig. 8.2.

One dimensional

heat transfer

one surface, it is required to determine the temperature as a function of position and time.
rate per unit area to the

Since the original temperature is uniform, and every part of each wall
is subjected to the same conditions, no heat will travel parallel to the

surface

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

247

and the temperature will be constant in any plane parallel to the

Thus the temperature distribution can be specified in terms of a
coordinate denoted by x in Fig. 8.2, and the discussion can be restric-

surface

surface.
single

ted to a section of unit area through the wall.

Considering the thermal

equilibrium of a slice of the wall between a plane at distance x from the
heated surface and a parallel plane at x-^Sx from the same surface gives the
following balance, where T is the temperature and k is the thermal conductivity.

dT

Rate of heat input at distance x and time

Rate of heat input

at distance

x and time

k ^

+ ^/

is

ar

dT
d
-/c + -

{<>

dx

t is

dt

This second rate is obtained by applying Taylor's theorem as in Chapter 1

and considering that x is a constant. Similarly, two output rates can be
determined from the corresponding input rates by applying Taylor's theorem
at constant t and variable x, thus
Rate of heat output at distance x + Sx and time t is

Rate of heat output

dT

-k

dx

_ dT

d_

dx

dx

/
^A
+ -(-/c
^

dt\

c
<5r

dx/

dx
d

at time

Heat content of the element

at time

is

dT

d /

dx

dt\

-/c-- + --(

dT\-k)dt

dx

dx)

pCpTdx

+ dtis pCpiT + --dt\dx

is

Taking average input and output

/ is

accumulation of heat in time dt

Heat input

and time t+ St

at distance x-\-dx

Heat content of the element

.'.

(-1) dx

pCp

-z- St

Sx

rates during the time interval St gives

=
,dT

d /

dT\l
St

Heat output

dx

dx

{<)

Sx

dx

dt

Using the general conservation law

(1.25),

dx

iO

Sx styst

and cancelling some terms for

simplicity,

d /

dx

,dT\^

d^

dT\

dT
dt

92

MATHEMATICAL METHODS

248

IN

CHEMICAL ENGINEERING

Dividing throughout by dx5t and assuming that k

is

constant gives
(8.46)

Taking the Hmit of equation (8.46) as 6t^0, the central term vanishes, and
introducing the thermal diffusivity (a) as in equation (7.150) gives

d^T

dT
-

dx-

dt

--^ =
which

is

the required equation.

It

(8.47)

can also be obtained directly from the

vector equation (7.150) by expanding (V^) in cartesian coordinates and deleting all y and z derivatives.

be seen that the same equation (8.47)

rates at time t are used without the
added complication of using average rates. This simpHfication is consistent
with the findings of Chapter 1, but until the reader is confident about which
simplifications are justified and which are not, he is advised to allow for all
possible correction terms as above. Care must be taken in each term to
write down the correct number of infinitesimals {6x, St), since these govern
which terms are rejected in the step from equation (8.46) to (8.47). The
boundary conditions will be derived separately in Section 8.4, and next an
equation relating to unsteady state axi-symmetrical diffusion will be derived.
In the above derivation,

results if the simple input

8.3.2.

Mass

it

will

and output

Transfer with Axial

Symmetry

A spray column is to be used for extracting one component from a binary

mixture which forms the rising continuous phase. In order to estimate the
transfer coefficient it is desired to study the detailed concentration distribution around an individual droplet of the spray.
To confine attention to a single droplet, a spherical polar coordinate

Fig. 8.3.

Mass

transfer with axial

symmetry

system is taken with its origin at the centre of the droplet. During its fall
through the column, the droplet moves into contact with liquid of stronger
composition so that allowance must be made for the time variation of the
system. The concentration will also be a function of both the radial co-

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

249

due
(r) and the angular coordinate (6) but will not depend upon
symmetry provided that the z axis is vertical. Denoting concentration
by c, diffusivity by D and the r and 6 components of the continuous phase
velocity by u and v respectively, a material balance can be taken over the
element ABCD shown in Fig. 8.3. The element is defined in the normal
manner by allowing one independent variable to vary at a time, and the
diagram shows a section through what is in fact a ring-shaped element,
ordinate

(j)

to the

symmetrical about the z

axis.

Area of face

AB

is

2nr^ sin 9 39

(8.48)

Area of face

AD

is

2nr sin 9 Sr

(8.49)

Volume of element
Material

is

2nr^ sin 9 39 3r

(8.50)

transferred across each surface of the element by

is

and molecular

isms, bulk flow

diff"usion.

Thus the

two mechan-

transfer rates at each

surface can be written as follows.

It

Input rate across

AB =

luc

Input rate across

AD =

ivc

D] Inr^ sin 9 39
:j2nrsm93r

should be noted that the concentration gradient in the 9 direction

is

dc

39

- ^7;

from the expansion of Vc in spherical polar coordinates, equation (7.108). The output rates across CD and BC are obtained from the
corresponding input rates by using Taylor's theorem as in Chapter 1.
as obtained

Output rate across

CD =

= (uc-D^j 2nr^ sm939 +

Output

rate across

Ddc\

(vc
\

BC =
27cr

sm93r

-\-

d9J

Accumulation

rate

(uc-D^j 2nr^ sin 9 39\ 3r

d
-- \f
[vc

39 l\

dc\
1
- 2nr sin 93r\39
39
J

= 2'Kr^%m93r39-^
3t

Using the general conservation law (1.25) and cancelling the obvious terms
gives
5

uc

-t

D^]27ir^s'm939\3r-

3r

-3

[/
(

t;c

3c\
1
J27rrsin0(5r 39

2nr^ sin 9 3r 39

dc

MATHEMATICAL METHODS

250

CHEMICAL ENGINEERING

IN

Dividing throughout by the volume of the element (8.50),

dc
dt

dr

Differentiating the products

sin 6

rsmOdeW

and separating the

dOj

velocity terms

(8.51)

from the

diffusion terms gives

dc

V dc
+ -
+ C 11 (r^w) +

dc

rde

dr

dt

'?Jr

r sin

_D^d^

Ode

Vdr
~7dr K
\

(v sin 9)

/ ,dc\

dc\
d
sin
^
r^smOdeX
^0/
2

(8.52)^

But the term in square brackets is the expression for V.i/ given by equation
and is thus zero because of the continuity equation.

(7.118)

dc

V dc

dc

d ( ,dc\

dt'^^d'r'^rde

d /

^dc\

dr

which

is

8.3.3.

The Continuity Equation

the expansion of equation (7.120) in spherical polar coordination.

This important yet simple equation expresses the fact that in a flowing
fluid the mass of the fluid is conserved.
Figure 8.4 shows a rectangular
element of space the lengths of whose edges parallel to the three cartesian

Bl_.

-I
Volume element

Fig. 8.4.

axes are dx, 5y, and dz.

by

w, V

and w

directions of u,

as shown,
v,

and

vv

6x

in cartesian coordinates

The components of the fluid velocity are denoted

and the density of the fluid by p. The positive
must be the same as the positive directions of the

considered to enter the element through faces ABCD,

ABFE, and ADHE, and leave through the other three faces.
The rate at which fluid enters through face
is proportional to the
area of the face {dydz), the fluid density (p), and the velocity component (w)
perpendicular to the face. Strictly, the velocity component only has the value
u at the point A, and the velocity through the face will vary shghtly from u.
This is a small correction which will disappear at the end of the derivation
axes so that fluid

is

ABCD

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

8.

251

If the reader wishes to check the above statement,

components can be considered as values averaged over the
relevant face, and when the limit is taken as the size of the element shrinks
to zero, each average velocity tends to its point value at A. Thus

and
all

it

be ignored.

will

velocity

ABCD =
ADHE =

ABFE =^pwdxdy

Input rate through

Once
rates

pu dy Sz
pv dx Sz

from the corresponding input

again, the output rates are obtained

by applying Taylor's theorem. Thus

Output

rate

d (pu Sy Sz) Sx

BCGF =p..x.z.i(p..x..)..

EFGH = puSySz +

through

dy

d (pw Sx Sy) Sz
CDHG ^ py^dxdy + oz
The mass of fluid contained
.'.

rate of increase of

in the element at time

mass contained

in element

Input Output

/ is

is t-

pSx Sy

(pSx Sy

Sz.

Sz).

Because

.*.

Accumulation

dp

ox

dy

oz

ot

(pu)SxSySz {pv)SxSySz {pw)SxSySz = SxSySz

(8.54)

where the obvious terms have been cancelled.

Dividing by the volume of

the element gives

(P) + |;M+|(p.)-h|=0
Taking the Hmit of equation
leaves

it

(8.55)

(8.55) as the size of the element tends to zero

unaltered, because all surplus terms have been ignored at

an

earlier

stage in anticipation of this result.

Each term of equation (8.55) can be diff*erentiated as a product and

rearranged to give
du

Since p

is

dv

a function of x, y,

dw

z,

dp

and

t,

its

dp

dp

dp

substantive derivative can be taken

as in equation (8.42), thus

Dp

dp

dp

dp

dp

-Dr''Yx'"Yy^''dz^-dt

(''')

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

252

Combining equations

(8.56)
1

and

Dp

--TT
p Dt

which

is

(8.57) gives

du
7-

dx

dw

dv

+ T- =
^
dy
dz

(8.58)
^

the required continuity equation for a compressible fluid.

8.4.

Boundary Conditions

In the last section the differential equation was found by taking a balance
The boundary
over an arbitrary element within the region of interest.
conditions can always be obtained in a similar manner by considering the
equiUbrium of the particular element which contains the boundary. The
variety of possible

boundary conditions

is

quite small

and sub-sections are

devoted to each of four types. These results should be committed to memory,

thus dispensing with the necessity for a detailed derivation on each occasion.
In the study of ordinary differential equations with only one independent
variable, the boundary is defined by one particular value of the independent
variable and the condition is stated in terms of the behaviour of the
dependent variable at the boundary point. In the study of partial differential
equations, each boundary is still defined by giving a particular value to just
The boundary condition must then
one of the independent variables.
specify the behaviour of the dependent variable as a function of all of the
other independent variables. There are three types of boundary condition
arising in practice for which analytical solutions are sometimes possible and
The more compHcated integrothese are considered separately below.
differential type of boundary condition is also derived, but problems in
which they are involved are usually intractable by analytical methods. The
discussion will be confined to heat transfer but the methods are generally
applicable.
8.4.1.

First Type, Function Specified

If values of the

dependent variable

particular boundary, then the condition

itself
is

are given at

said to be of the

all
first

points on a
type.

Con-

sidering time dependent heat transfer in one dimension as in Section 8.3.1,

the temperature will be a function of both

x and

/,

thus the boundaries will

be defined as either fixed values of x or fixed values of

conditions of the first type for this problem are
(i)at
(ii)

at

= 0,
= 0,

t.

Two boundary

r=/(x)

(8.59)

T=

(8.60)

^(0

Equation (8.59) gives the temperature at all values of x for one value of /,
whilst equation (8.60) restricts the behaviour of the temperature at all times
at one value of x.
In the case of steady heat conduction in a cyHndrical conductor of finite
size, it is usual to use cyhndrical polar coordinates so that z = a defines
the end faces and r = Kq defines the curved surface. Hence both boundaries

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

are defined

253

by keeping one of the independent variables constant, and

first type take the form

conditions of the
at

at

fl.

T=f(r,0)

(8.61)

r o

T=

(8.62)

9(z,e)

Due to the axial symmetry in this problem,

temperature will not be a function

of 6 and equations (8.61) and (8.62) can be simplified, but in the corresponddependence and also a t depening unsteady state problem there will be a
The use of cartesian coordinates is not convenient in the above
dence.
problem because the description of the boundaries is not so simple and all
of the coordinates vary on the curved surface.
8.4.2.

Second Type, Derivative Specified

The position of the boundary will still be defined as above by keeping

one independent variable constant, but instead of giving the direct behaviour
of the dependent variable, its derivative is restricted by a boundary condition of the second type. This situation occurs with radiant heating or cooling of a surface, electrical heating of a surface, or the special case of a

thermally insulated surface. It can be seen in these cases that the heat flow
rate is known but not the surface temperature. Because the heat flow rate
is related to the temperature gradient, this boundary condition is of the
second type, and two examples will now be considered.

Example

1.

rectangular block of metal

variation in aU directions, but the surface at

is

subject to temperature

is thermally insulated.
Express this condition by an equation.
It is convenient to use cartesian coordinates in this problem, and the
boundary condition is derived by considering the infinitesimal element

I
Fig. 8.5.

Boundary element

residing at the boundary as shown in Fig. 8.5. Heat can only pass through
the faces of the element by conduction, and the rates can be found for five
of the surfaces using the same method as that used to determine the diff'erential equation for a general volume element.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

254

dT

Input rate through

ADHE = -k SxSz

Output

BCGF = -k

rate

through

dT
dT

dxdz + i-k-SxSzjSy
d /

dT

ABFE = -k--Sxdy

Input rate through

oz

dT

5x ^y

DCGH = -

Output

\
EFGH = -k-Sy3z-\-l-k-dySzjdx

/c

(5x ^>;

dT

rate

through

dT

Output rate through

A:

(5z

dT

d /

Because the face at x

is

ABCD must be zero.

The

rate of accumulation of heat within the element

thermally insulated, the heat input rate through

dT

is

(after

some

SySzSx = pCp^^SxSySz
pC,+ k^^^SySzSx

pCp-w7 SxSySz as before, and completing the balance gives

cancellation)

d^T

-^ Sx

SzSy
-^

dy^

d^T

k -^ SxSySz
dz^

"

dT

+ k^ Sy^ Sz

4-

dx

dT

d^T

Due

to the third term in this equation,

only possible to cancel out dySz

it is

giving
,

Taking the

/d^T

d^T

d^T\

,dT

^ dT

element shrinks to zero, results in all of

tending to zero with one exception. Thus equation I

limit as the size of the

the terms of equation

becomes

^=

atx

III

dx

and

this is the required

Example

2.

boundary condition.

cylindrical furnace

is

Hned with two uniform layers of

What boundary conditions

insulating brick of different physical properties.

should be imposed at the junction between the layers?

Figure 8.6 illustrates the problem, where a denotes the radius of curvature of the boundary between the layers, and the element is considered to lie
across the boundary as shown. The symbols T, k, Cp, and p will be used for
the temperature, conductivity, specific heat, and density of the layers, and
a subscript 1 or 2 will distinguish between the layers. Due to axial sym-

'

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

255

metry, no heat will flow across the faces of the element given by 6 = constant, but allowances will be made for heat flow in the z direction.
One boundary condition, which is of the first type, is that the temperature

must be continuous across the boundary. That

at

is.

Ti

The second boundary condition is that the law of conservation of heat must
be satisfied for the element shown in Fig. 8.6. The rate of flow of heat just

Section through cylindrical element at boundary

Fig. 8.6.

dTy

inside the

boundary of the

first

layer

is

k^aQdz-^, and

using Taylor's

theorem, the rate of flow of heat into the element across the face

which must be evaluated

.'.

Input across

a.i

CD

is

a.

CD =

= -k.aOdz^ +ikiedrdz-(r--^]
dr

dr\

dr /

dr\

drj

Similarly,

Output across

AB
dr

The heat flow

Input at face

rates in the z direction

(z)

can be written:

= -iaedr (k,T, + k2T2)

oz

Output

at face (z

<5z)

= -iaeSr-ikiT, + k,T^) +
Accumulation within the element

= y,C,,ae drSz -^ +

-iaBSr-ik^Tl + k^T^^d:

ip, C,2 a

&r5z

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

256

Completing the heat balance and cancelling the obvious terms gives

-k,a93z-^-[-ik,eSrSz-(r^] + k2aedz-^ +
dr\

dr

or

cr /
d ( dT,\

+ ik2e3rdz-{r^]+iaeSrSz^{k,T, + k2T2)
dr

= iaeSrSz-{pj^CpiT^+P2Cp2T2)

II

Dividing by aOSz gives

aTi

-ki-^ +
^

dr

,,

iki
^

Sr d / arA
8X2

{r--^] + k2^
dr
a dr\ dr

,,
\-

ik2^
^

Sr d / dT2\

r-r^ +
a dr\ dr
[

d^

+ iSr,(k,T, + k2T2)
dz-

= iSr-{p,C,,T,+p2C,2T2)
In the limit as (5r->0, most of the terms in equation

III

III

disappear leaving

which is the second boundary condition.

These examples, particularly the second one, show that most of the
effects which give rise to important terms in the differential equation are
not important in the consideration of boundary conditions. The boundary
condition IV does not include any terms describing either the variation of
temperature with movement in the z direction, or the variation with time,
or even the curvature of the boundary. All of these terms disappeared in the
final hmiting step from equation III.
Incidentally, the differential equation can also be derived from equation
III by moving the element into either layer when all subscripts can be
removed and a must be replaced by r. The two terms (k dT/dr) will cancel
out, followed by cancellation throughout by (dr); the limiting process will
then yield the general differential equation

1^/

dT\

rdr\ dr)

)2>
d^T

dz^

pCpdT

dt

for heat conduction in cylindrical polar coordinates with axial symmetry.

8.4.3.

Third Type,

Mixed

Conditions

This type of boundary condition consists of those in which the derivative

of the dependent variable is related to the boundary value of the dependent
variable by a linear equation. The most usual case of this is when the surface
rate of heat loss is governed by a heat transfer coefllicient. Referring back to
Fig. 8.5, if the surface ABCD at x =
is cooling to surroundings at a

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

8.

257

temperature (Tq) through a gas film giving a resistance which can be expressed
(/z), the boundary condition can be

in terms of a heat transfer coefficient

stated thus.

Rate at which heat is removed from the surface = h{TTQ) per unit area.
Also, the rate at which heat is conducted to the surface internally = kdT/dx
per unit area. These two rates must be equal, and hence
k^-I
ox

= h(T-To)

(8.63)

The above derivation assumes that curvature, time, and other space variations do not affect the balance, thus using the results of the previous section.
Equation

(8.63)

is

true for any surface

which faces in the negative

direc-

but the sign of the left-hand side must be reversed if the

The signs must not be changed
surface faces in the positive direction.
according to whether the surface is being heated or cooled. The only negative
signs arise from the following two considerations,
(i) the heat conduction equation contains a negative sign by definition,
(cf. equation (1.48)).
(ii) the outward normal from the surface may be in the negative coordi-

tion of the

axis,

nate direction.

Equation (8.63) contains both of these negative signs and

Integro- Differential

8.4.4.

is

correct as written.

Boundary Condition

This type of boundary condition arises most frequently in the study of

mass transfer when the material crossing the boundary either enters or leaves
a restricted volume and contributes to a modified driving force. Thus consider the following example.

by

Example 3. A solute is to be leached from a collection of porous spheres

them as a suspension in a solvent. What is the correct boundary

stirring

condition at the surface of one of the spheres ?

The rate at which material diffuses to the surface of a porous sphere of
radius .

is

-4na^D^^\

where

is

an

effective diffusivity,

and c

concentration within the sphere. If V is the volume of solvent and

concentration in the bulk of the solvent, then

dC
V=

-ANna^D

dt

where A^
Sit

a,

is

the

number of

and equation

-,

is

the

is

the

dc

dr

spheres. For continuity of concentration, c = C

must be integrated to determine
Hence the

boundary condition takes the form:

at

4NnDa^
c=-\^^-

dc

a,

dt

II

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

258
8.4.5.

Initial

Value and Boundary Value Problems

the different types of boundary condition have been classified but

the number of conditions required for the solution of a particular problem

So

far,

has not been defined. For ordinary differential equations, it was stated that
number of boundary conditions was equal to the order of the differential
equation, but no corresponding rule has yet been discovered for partial
differential equations. It is possible, however, to give a guiding hint which
will help in the more usual problems.
Considering any one independent variable, say x, and assuming that the
dependent variable is a function of x only so that all derivatives with respect
to the other variables are zero, then the partial differential equation becomes
an ordinary differential equation in terms of x. This simplified problem will
require as many boundary conditions as its order and these must be given
at fixed values of x. This number of boundary conditions is frequently the
same as the number required by the corresponding partial differential
equation at fixed values of x. Treating the other independent variables in
the same manner gives the number of conditions needed at the boundaries
of the other variables. Thus, for the equation of heat conduction in one
dimension derived in Section 8.3.1,

the

d^T _ dT
=
a^
two boundary conditions are needed

(8.47)

at fixed values of
This rule cannot be relied upon in every
case but is meant as a guide for common problems. For example, if it is
applied to equation (8.55) it indicates that one boundary condition is
required at fixed t, two at fixed r, and two at fixed 9, and yet it is not normally necessary to specify any conditions at fixed
except that the solution
should repeat itself at intervals of In.
The boundary conditions lead to an important classification of partial
differential equations as follows. When only one condition is needed in a
particular variable it is specified at one fixed value of that variable, whereas
when two or more conditions are needed, they can all be specified at one
value of the variable, or some can be specified at one value and the rest at
another value. In the latter case, conditions are established at both ends of a
range of values of an independent variable and the range is said to be
"closed" by conditions at the beginning and the end of the range. Alternatively, if only one condition is stated, or all conditions are stated at one fixed
value of the variable, then the behaviour of the dependent variable is
restricted at the beginning of a range but no end is specified and the range
is said to be "open" as far as that independent variable is concerned. If the
range is closed for every independent variable, the problem is alternatively
named a "boundary value" or "jury" problem. If the range of any independent variable is open, the problem is called an "initial value" or "marching" problem. This classification affects the choice of a method of solution.
Thus the method of the Laplace transform, to be given in Section 8.8, can
this rule states that

X,

and one

at a fixed value of

t.

8.

PARTIAL DIFFERENTIATION

AND PARTIAL DIFFERENTIAL EQUATIONS

259

only be applied to initial value problems because the variable removed by

the transformation must have an open range. When numerical methods are
considered in Chapter 11 it will be seen that the calculation techniques are
quite different depending on whether the problem is initial value or boundary
value and that the methods are not interchangeable.

Particular Solutions of Partial Differential Equations

8.5.

The general solution of a partial differential equation is not often possible, but when the boundary conditions are taken into account useful
particular solutions can often be found. The methods given in this section
for finding such particular solutions are the simplest to follow but the most
difficult to initiate. In Section 8.7, the more powerful method of separation
of variables will be described, but again the resultant expression is a collecbecome of general use only when the

tion of particular solutions which

conditions given in Section 8.6 are satisfied.

The other methods using various transforms are treated in Sections 8.8

and 8.9 and they will determine general solutions for particular combinations
of boundary conditions classified in Section 8.4. However, a partial differential equation cannot be reduced to an ordinary differential equation by using
the Laplace transform, which is the most useful of the many transform
methods, unless all variables except one have open ranges. Thus the most
that can be obtained in nearly all cases is a particular solution or a combination of particular solutions, and very rarely can a general solution be
obtained. The particular solutions are complete and valid, and will satisfy
practical requirements but an individual solution has to be obtained for
each problem.
8.5.1.

Compounding

the Independent Variables into

One

Variable

always possible to make a simple change of variables by replacing

one variable with a new variable, but there are exceptional cases in which the
elimination of one variable by substitution results in the disappearance of a
second variable from the problem. The basis of the present method is to
look for this possibihty and hence reduce a partial differential equation in
two independent variables to an ordinary differential equation.
Consider for example the one dimensional heat conduction equation
derived in Section 8.3.1.
It is

d^T _dT
^=
~
dx^

It is

assumed that the solution of equation

T=f(q)
where

(8-47)

dt

xt"

(8.47)

can be written
(8.64)
(8.65)

n is any unknown constant, and q is the new variable. Equation (8.64) can
be differentiated partially with respect to both x and t as follows.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

260

dT

_dfdq_

dx

dq dx

dx^

dT
- =

and
Substituting equations (8.67)

and

dq
(8.67)

T^2
dq^

df dq
=
-f^
dq dt

dt

d,

nxt"-'

df
(8.68)

-f-

dq

(8.68) into equation (8.47),

oit''"^=nxt"-'^
Using equation

(8.65) to eliminate

x from equation

(8.69)

(8.69) gives

.^"0 = ,r^|
The above

steps are valid for

any constant value of n, and

(8.0)
in particular they

are true for

n= -i
With this value of n, equation (8.70) becomes independent of
ordinary differential equation
d'f
df
a^=-i^^
di^^^^'^dq

(8.71)
t

giving the

(8.72)

The dependent
hence

it

variable does not occur explicitly in equation (8.72) and

can be solved by the method given in Section 2.4.1 after one stage of

integration,

df
,

=^e-^'/^

(8.73)

dq
where

is

a constant of integration.

A further integration of equation (8.73)

gives

/=Berf(g/2Va) + C

T = B erf (x/2>/^) + C

or

(8.74)

B and C are arbitrary constants.

The kind of situation of which equation (8.74) is the solution can be
found by putting x =
or r =
and determining the boundary conditions
where

satisfied

by equation

If

and

if

Thus equation

(8.74).

X
/

= 0,
= 0,

T=C
T = B-\-C

(8.74) describes the temperature distribution as a function

of time within a semi-infinite slab of material which is cooled on one surface.

initial temperature throughout the slab is uniform at 5+ C, and C is the
temperature of the exposed face. This is the only situation described by

The

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

261

equation (8.74); any other type of boundary condition is incompatible with

The above method thus determines a particular solution which
will probably not satisfy the required boundary conditions, and therefore
the method is not very valuable.
the equation.

8.5.3.

Superposition of Solutions

It was shown in Section 2.5 that the complementary function of a linear

ordinary differential equation could be added to the particular integral
and the combination still satisfied the differential equation. Thus for any
linear ordinary differential equation in which each term contains the
dependent variable, solutions can be superposed. It will now be shown that
the principle of superposition can be apphed to Hnear partial differential
equations provided that every term contains the dependent variable.
Taking the general Hnear partial differential equation of second order in
two independent variables

/l

2 +/2^^

du

du

+h-. +/5 ^

+/3 T-2

+/6 =

where / are functions of jc and y, and assuming that u = u^ and u

two different particular solutions of equation (8.75), then the sum
W3
will also

M1

+ W2

(8.75)

U2 are

(8.76)

be a solution. Differentiating equation (8.76) gives

du^

dui

du2

dx

dx

dx

dx^

dxdy

dx^

dx^

dxdy

(8.77)

dxdy

etc.

Putting u = Uj, into equation (8.75) by using the set of equations (8.77) it
can be seen that = W3 is also a solution when Wj and W2 ^re solutions. This
is only true when the dependent variable u occurs once and once only in

each term of the partial differential equation, and the above argument can
obviously be extended to any number of independent variables and equations
of any order. Thus the important result emerges, that for a Hnear partial
differential equation, an infinite number of particular solutions can be added
together to give a further solution.
One particular solution of the heat conduction equation was found in
Section 8.5.1, and using the same symbols, a more basic particular solution
can be found by differentiating equation (8.47) with respect to x. Thus

^ dx^\dx)

~ dt\di)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

262

which shows that

can be replaced by dTjdx to give a further particular

Combining equations

solution of equation (8.47).

(8.65, 66, 71, 73) gives

ex
.*.

r=

Ar-^6-^'/*'

(8.78)

is also a solution of equation (8.47).

Considering unit cross-sectional area of a semi-infinite slab of density
specific heat Cp occupying all space for positive x values, then the total
and
p
heat content of the slab at any fixed time t is given by
00

H=\pCpTdx

(8.79)

Using the temperature distribution given by equation

(8.78),

00

Because

is

being evaluated at constant

performed with
that dx

constant.

dzy/octy

t,

all

of the present calculations are

Changing variables by putting x

2z\/(xt so

hence
00

H = IpCpA^foL^ e-'^ dz
The above

integral has

been evaluated in Section

.*.

because

(8.80)

5.2.

H = pCp A\/(xn
a

= kA^/^
= k/pCp

(8.81)

(7.150)

Equation (8.81) shows that the amount of heat in the slab remains constant
for all time.

The

temperature distribution

initial

r->0, of equation (8.78).

is

determined by taking the

limit, as

L'Hopital's rule cannot be used because the expon-

ent cannot be expanded by Taylor's theorem in positive powers of

t.

How-

can be proved that an exponential function tends to its limit more

rapidly than any finite power of the variable tends to its Hmit, but the proof
is not elementary.
for
Equation (8.78) thus shows that when t = 0, T =

ever,

it

values of x except x = 0. At a: = 0, the exponential factor is unity for

time so that as /-O, r->oo. Hence the heat content as given by equation
(8.81) is initially concentrated at the face x = 0, and as time progresses it is
all conducted into the slab according to equation (8.78).
The following
example illustrates the superposition of solutions of this type.
all
all

Example 1 Heat is supplied at a fixed rate

Btu/h ft^ to one face of a
large rectangular slab of density p Ib/ft^, specific heat C- Btu/lb F, and
.

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

263

conductivity A: Btu/h ft^ F ft"\ find the variation of surface temperature

with time during the early stages after exposure.
The physical picture of the above situation is that heat is arriving at the
surface and then being conducted into the slab according to equation (8.78)
so that the temperature distribution at any time t is the accumulated effect
of the heat entering the surface between zero time and time /. Thus let t be a
new time variable within the range < t < r so that during a time increment
(5t, a quantity of heat Qdx will be received by the surface. At time /, this heat
will

will

have been conducted into the slab for

be given by equation (8.78) as

^r =
Using equation

t)

hours and

its

distribution

^(r-T)-*e-"'/''^'-^>

(8.81)

Qdx

kAs/nfa

5T = ^

.-.

Equation

(/

II gives the part

/'(f-T)-^e-^'/'*<'-^><5T

II

n
t which is
and t + ^t.

of the temperature distribution at time

the result of the heat absorbed by the surface between time t

Hence the
all

total temperature distribution can be obtained by adding together

of these temperature increments and taking the limit as 5t->0. Thus

sJ~\{t-T)-^e-^'l^'^'-'Ux

can be integrated in terms of the error function by making the

l/z^, but as the temperature variation at the surface is
that is required by the problem, x can be put equal to zero in equation
giving the surface temperature

Equation

III

substitution
all

III

III

tz=

-mk-

I^

Other problems can be solved by this method, but if g is a function of

it must remain inside the integral in equation III as Q{t).

time

8.5.3.

The Method of Images

The solutions of the conduction equation considered so far have only

applied to a semi-infinite body. It is possible to extend the solutions to finite
media for two types of boundary condition. Firstly, if one plane surface is
insulated or the finite body has a plane of symmetry, and secondly if the
boundary condition

is

of the

first

type.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

264

Example 2. Consider a wall of thickness 2L ft with a uniform initial

temperature throughout, and let both faces be suddenly raised to the same
higher temperature. Find the temperature distribution as a function of time.
Taking an x coordinate measured from one face of the slab, the initial
uniform temperature as the datum and considering unit cross-sectional
area, the differential equation is again

d^T

dT

and the boundary conditions are

r=
T=

r=-.0,

at
at

jc

at

=
=

0,

II

To

III

dT

ox

L,

IV

The last condition takes advantage of the symmetry of the problem so that
no heat crosses the central plane. According to Section 8.5.1, the solution
which satisfies conditions II and III for an infinite slab is
Ti

= To-Toerf(x/2V^)
= Toerfc(x/2Va"0

but as soon as the temperature at

is symmetrical,
the insulated face at x = L is equivalent to another heated face at x = 2L
(this in fact returns to the original problem). But equation V gives the temperature distribution as a function of distance x from any heated face. The
distance of a point in the body from the second heated face is (2L x), and
hence the temperature distribution resulting from the heat entering the face
Sit X = 2L is given by
This

is

satisfactory for small values of

= L starts to

rise,

equation IV

T2

is

/,

violated. Since the system

To erfc [(2L - x)/2 V^]

VI

by analogy with equation V. The combined solution becomes

73

The

To erfc (x/2VaO + To erfc [(2L - x)/2Var]

VII

IV but as t
part of equation VII violates equation III. If a heated face at x = 2L causes
the temperature at x =
to rise at a certain rate, then a correspondingly

increases further, the second

solution Tj satisfies equation

cooled surface at x = 2L will cause a corresponding fall rate at x = 0. The

temperature distribution caused by such a cooled face a.t x = 2L is given by

T^= -To erfc [(2L + x)/2V^]

which can be added to equation VII to
T5

The

satisfy

To erfc (x/2V oa) + Tq erfc [(2L - x)/2 V^]

distribution given

equation

VIII
III again.

Thus

- Tq erfc [(2L + x)|2^/^']

by equation IX now violates equation IV again but a

further reflection at the plane

=L

gives

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

Tg

= - To erfc [(4L - x)/2 Voc7]

265

which can be added to equation IX. This process can be continued indefiand x = L Hke a perfectly reflecting
nitely, by treating each face at x =
mirror so that any source to the left of x = L possesses a corresponding
image source to the right of x L. Also, any source to the right of x =
must have a corresponding image sink to the left of x = 0. Thus the insulated face (or plane of symmetry) acts as a perfect reflector as far as sources
and sinks are concerned, and the constant temperature face acts as a reflector
which interchanges sources and sinks upon reflection. The complete solution
is

thus

T=

To
n

f=

(- l)''{erfc [(x + 2wL)/2V^] + erfc [(2nL + 2L-x)/2V^]}

XI

This method of images was originally developed to solve linear problems

and magnetism, but it can be appHed to any problem involving
the solution of a hnear partial differential equation where superposition of
solutions is permissible and the boundary conditions are of the first type, or
of second type with vanishing derivative.
It is possible to extend this image principle to cylindrical and spherical
boundaries but this is rather more compHcated and so specialized that the
method will not be considered here.
in electricity

8.5.4.

Particular Solution Suggested by the Boundary Conditions

Because all boundary conditions are specified at a fixed value of one

independent variable but over a range of values of the other variables, a
functional form frequently appears in the statement of a boundary condition. If the particular solution required is to satisfy such a condition, it is
reasonable to suppose that the particular solution may have a similar functional form.
This is especially true in classical hydrodynamics and the
following examples illustrate the technique applied to the problem of incompressible fluid flow round a spherical obstacle using each of the approximations to the Navier-Stokes equation (7.126) given in Section 7.10.

Example 3. Ideal Fluid Flow Round a Sphere. In Section 7.10.1 it was

shown that when the viscous terms were neglected in the equation of fluid
motion, a velocity potential (0) could be defined so that

u=-V(j)
and

V^cj)

II

It is convenient to use spherical polar coordinates for the present problem

and there is no dependence on the azimuthal angle and hence no ambiguity
over the meaning of the symbol
Using equation (7.120), equation II can
(j).

be written

r^^ +
-T
^ ^(sin0-^)=O
r^dr\ drj
r^smede\
dOj
,

III

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

266

of the sphere is a, and the fluid flows with velocity Uq along the
= towards the origin at the centre of the sphere as
shown in Fig. 8.7, the boundary conditions can be written:

If the radius

axis of

symmetry

at

as

r-)>oo,

where u and

a,

IV

v are the r

v-*UQsmO

u-^ UqQOsO,

and 6 components of the

velocity u respectively.

Flow round a sphere

Fig. 8.7.

Equation IV prohibits flow through the surface of the sphere, and equation
states that the flow at a large distance from the obstacle, in any direction,
is undisturbed parallel flow.
In order to solve equation III for these boundary conditions, equations V must be expressed in terms of by substituting
into equation I. Thus

<j)

or

Un sin 9

rdO
.'.
(f)

JjQrcosO

as r->>oo,

,'.

(f)-*

UQrcosO

VI

Boundary condition VI suggests trying a particular solution of the form

VII

(l>=f(r) cos 9

'

in the partial differential equation III.

Thus
VIII

dr\

drj

sin 9

39 \

39

Performing the differentiations with respect to 9 and replacing the partial

derivatives with respect to r with ordinary derivatives,

{Ism9cos9) =

COS 9

dr \

drJ
d'f
dr^

Equation IX

is

sin 9

df

+ 2r^
" dr -2/=0

IX

'

independent of 0, and since

it is

both linear and homogeneous

8.

it

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

can be solved by the methods of Sections 2.4 and

2.5.

The

267

result is

f=Ar-\-Blr^
Substituting into equation VII gives the solution
(j)

To

satisfy condition VI,

A =
/.

But w
at

(Ar+Blr^) cos

Uq,
(j)

= (U or +Blr^) cos 9

.*.

Combining equations XI and XII

gives

g)

satisfies

Xn

B = iUoa^

= L7o(r
^o(r + T^
which

XI

= d(l)/dr from equation I, therefore boundary condition IV becomes

r = a,
w = -(C/o-2B/r^)cos0 =

COS0

Xm

IV and VI and the differential

The stream function form of equation

the boundary conditions

equation III and

XIII is

is

thus the solution.

\l/

= iUor^(^-^) sin^

XIV

which can be checked by differentiating equations XIII and XFV and comparing the two expressions for each velocity component.
It is worth noting that the 9 component of velocity does not vanish at the
surface of the sphere. That is,

which means that there is slipping at the surface of the sphere. This is not
physically reasonable and is caused by neglecting the viscous term in the
equation of motion. The difficulty can be overcome by assuming that there
is a thin layer of fluid near any boundary in which viscous effects are important, and this is the basis of Prandtl's "boundary layer theory". The reader
is

referred to Goldstein or Schlichtingf for further details.

Example 4. Viscous Flow Round a Sphere, The simplified equation of

motion describing this problem in terms of the stream function is

^^ =
where the operator E^

is

defined by equation (7.135) in spherical polar co-

ordinates as

'H
^^
dr"-

9\x\9
/
c)\lA
sin^a^/_L_#\
r^ d9\sm9d9)
()

" Modern Developments in Fluid Dynamics,'* Vol. I.

Oxford
t Goldstein, S.
University Press (1938). Schlicting, H. " Boundary Layer Theory " (translated from
German), 4th edition. McGraw-Hill.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

268

for large values of r in terms of the stream function

The boundary condition

is:

The boundary conditions

xj/^^Uor^sm^e

r-^oo,

as

at the surface of the sphere of radius

a which

impose conditions of zero shp are

at

a,

or at

a,

Equation

0,

=
^
or

0,

t;

-^=0

IV

cO

III suggests trying the particular solution

\l/=f{r)sm^d
Substituting into equation II gives

f 5sin^^\
Te \si^
de )

sin0 ^ (

^~^

VI

jy ~

2/

Denoting -rj

"2

by F(r), and applying the operator E^ to equation VI,

the steps between equations

replaced by F.

and VI

will

be repeated identically with

Therefore

V =
Substituting for

F in

IF
/d^F
(-^--y)sin20
~
\dr^
r

VII

terms of/ and equating to zero as given by equation

/d^f

d"-

I,

If

dr"- Xdr"-

+
3::^-r234
r^dr^

Equation VIII

is

linear

dr^

3T--4/=o
r
rUr

VIII

and homogeneous and has the solution

f=Ar'^ + Br^ + Cr-{-D/r

IX

Substituting into equation V, therefore

,/^

is

a solution of equation

(^r^ + Br^

I.

and
.-.

il/

To

+ Cr 4- D/r) sin^ 6

satisfy

boundary condition

A=
B = iUo
= {iUor^ + Cr-^Dlr)sm^9

III

XI

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

269

Using boundary conditions IV gives

I
and

iUoa^ + Ca + Dla =

XII

Uoa + C-Dla^ =

XIII

Solving equations XII and XIII for

C and D and substituting into equation XI

gives the final solution

>l'

= iU.r'(^l-l"+\^ymH

XIV

the vorticity distribution can be obtained by substituting equation

XIV

into

equation (7.136). Thus

C

^sine

XV

Comparing equation XIV in Example 3 with equation XIV in Example 4,

can be seen that the two solutions are quite different, and yet they are frequently confused because both are symmetrical about the equatorial plane
of the sphere. In the ideal fluid solution the fluid slips over the surface of
the sphere and the vorticity is zero everywhere, whereas in viscous flow, the
fluid does not slip at the boundary, and a vorticity distribution is set up. It
should be noted that a second order equation was solved in Example 3, but
a fourth order equation arose in Example 4, hence more boundary conditions could be satisfied in the second case. This explains why there is good
agreement between theory and experiment at very low Reynolds numbers
where inertial effects are small; whereas there is only approximate agreement for 9 < 90 between theory and experiment at high Reynolds numbers,
because viscous effects can never be completely ignored.
it

8.6.

Orthogonal Functions

Before the method of separation of variables can be used to solve partial

diff'erential equations, a fundamental relationship between the individual
members of a set of functions must be investigated. The method of separation of variables normally gives the solution as the sum of an infinite set of
related functions,
satisfies

and the method

is

most useful when

this set

of functions

the foUowing definition of orthogonality.

Two functions 0(x) and (l)j(x) are said to be orthogonal with respect to
a weighting function r(x) over the interval from a to Z? if
b

r(x)

(l){x) (/),(x)

dx

(for

^ n)

(8.82)

Thus,

if there is a set of functions </)(x) where is a positive integer, and the

integrated product of any pair of functions with a fixed weighting function
is
is

Um

= n (i.e. the same function

is orthogonal.
taken twice) then the integral in equation (8.82) will not normally vanish.

zero, then the set of functions

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

270

Example 1. Show that the functions sin nx and sin mx are orthogonal
with respect to unity in the range O^x^tt, for integer values of m and n.
from the trigonometrical

Starting

2 sin nx sin

identity

cos (n

mx =

m)x

cos {n + m)x

and integrating over the given range, therefore

2

sin

nx sin

cos (n

mx dx =

m)x dx

[sin (n

cos (n +

- m)x]S
n

+m

m)x dx

[sin (n

II

+ m)x]5

=
because the sine functions vanish at both limits of integration.
the first integral in equation III becomes

If

m=

n,

jdx =

fO

p
\

sin

mxdx =

nx sin

if

which

is

8.6.1.

The Sturm-Liouville Equation

n^m

^^

<
[ 7r/2 if

This

III

71

=m

the complete evaluation of the orthogonality integral.

is

differential

the name which describes any second order linear ordinary

equation which can be written in the form

ix \^^^^ S]
where A

is

that the

first

"^^^^^

^ ^'^^^^^ ^

^^-^^^

a constant, and p, q, and r are functions of x. It should be noted

and second differential terms of a second order equation cannot
always be combined into the form given in equation (8.83).
For any particular set of functions p, q, and r, the solution of equation
(8.83) for y will depend upon X, Hence, if X can take a set of discrete values
A, a corresponding set of solutions y = </>(x) will be obtained. Taking any
two values of ?. (X and X) and the corresponding solutions (y = </)(a:) and
y = (t>m(x) ), then

[pW^] +
and
Multiplying the

^p(x) ^]
first

tracting the second

lq(x) + Xr(x)-]cl>

[q(x) + X^ r(x)](/>

of these equations by 0^, the second by

from the

first

gives

(/>,

and sub-

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

x from aio b

Integrating equation (8.84) with respect to

iX-uJK^)4>A.dx =

[pix/-^] dx

J^|^

-J4>.l [pwf]

dx

gives

271

Evaluating each of the integrals on the right-hand side by parts gives

b

(^-'^jj

r(x)(l)<t>dx

[*.K,i"]:-/,tf-.>- [*..)f ]>/.<) tf.

a

Simplifying by canceUing the two identical integrals and combining the

boundary values gives

a-AjJrW^^dx = [pW{^-,^ - "^".^j]'

(8.85)

Hence the functions (l)(x) and (/>^(x) which correspond to two different
and X) of the constant, will be orthogonal if the right-hand side
of equation (8.85) is zero. It is possible for the function on the right-hand
side of (8.85) to have the same value at both Umits and thus cancel itself out,
but it is more usual for the term in brackets to vanish independently at both
limits. The term is zero when p(x) =
at the Hmit or when the boundary
conditions are suitable. When the same boundary conditions apply to every
values (X

equation of the set, independent of the value of

boundary condition which are suitable.
(i)

at

A
(ii)

(iii)

at

<l>n

= a.

dx

at

a,

X,

there are three types of

=
= (t>m =
y

^
dx

a.

dcj>m

dx

P<t>m

if both boundary conditions are of the above three types (with

duplication allowed), then the set of functions (j){x) will be orthogonal with
respect to the weighting function r{x) which occurs in equation (8.83) in

Therefore,

conjuction with

X.

MATHEMATICAL METHODS

272

IN

CHEMICAL ENGINEERING

It should be noted that the conditions given above for the solutions to be
orthogonal are more restrictive than the three types of boundary conditions
Condition (i) states that the boundary value of the
listed in Section 8.4.
dependent variable must be zero and cannot be a function of another independent variable. This condition can be relaxed a little to allow the dependent variable to have a constant finite value at the boundary by moving the
origin so that the new dependent variable will be zero at the boundary. If
condition (i) is used at both boundaries, the origin cannot be moved to two
different places, but one method of overcoming this difficulty will be described
in Example 2 of Section 8.7. Thus most type 1 boundary conditions will give
rise to an orthogonal set of solutions to a Sturm-Liouville equation, provided
that there is no dependence upon another independent variable.
There is no easy way of extending condition (ii) to include a constant
gradient at the boundary, hence only a vanishing derivative of type 2 boundary conditions will give orthogonal solutions. The same remarks apply to
condition (iii) as to condition (i). That is, any type 3 boundary condition is
suitable provided that only one movement of the origin is required to bring
the condition at both boundaries to one of the forms (i), (ii), or (iii). Condition (iii) does include conditions (i) and (ii) as the special cases ^ = oo, and

respectively.

the integro-differential boundary condition given in Section

not suitable for obtaining orthogonal solutions to a Sturm-Liouville

Finally,
8.4.4

is

equation.

Method of Separation

8.7.

of Variables

When

a dependent variable, say T, is related to two independent variables X and / by means of a partial differential equation, the solution will be

of the form

T = F(x,0
The method of separation of variables looks

(8.86)

for solutions of the

form

T=/(x).g(0
where the functional dependence upon x

upon

/.

(8.47)

is

(8.87)

separated from the dependence

glance at the particular solutions (8.74) or (8.78) of equation

sufficient to

show

that they cannot be expressed in the

and thus any solution which

variables will be a

The procedure

new

is

is

form

(8.87),

obtained by the method of separation of

one.

method is to substitute equation (8.87)

and separate the terms dependent upon x from

to follow in this

into the differential equation

upon / by the equality sign. The equation thus takes the

form of a function of x only, being equal to a function of t only, which can
only be satisfied by both sides of the equation having the same constant
value.
Thus, two ordinary differential equations, each involving an unknown parameter are derived from one partial differential equation. The
procedure can be better understood in reference to particular examples, and
special applications of the time dependent Laplace equation will be used in
those dependent

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

273

it occurs so frequently in chemical engincoordinate systems will be used and the

description will be related to simplified forms of the heat conduction equation

the following sub-sections because

eering.

The

three

most

common

aV^T =

(8.88)

dt

With a suitable interchange of symbols and interpretation

of boundary conditions, the solutions could apply equally well to the mass
diffusion equation
dc

for convenience.

DV'c =

(8.89)

or any other equation of the same form.

8.7.1.

Unsteady State Linear Heat Conduction

Equation (8.88)

in cartesian coordinates takes the familar

d^T

form

_dT

0^2=-^
dx^

Assuming

that the solution can be expressed

then

and

(8-47)

dt

by

T=f(x).g{t)

(8.87)

^^l=f(x).gV)

(8.90)

:^2=rM-9(t)

(8.91)

dx'

where dashes denote differentiation with respect to the argument of the

function. Since f(x) is a function of x only, the differentiation can only be
performed with respect to x; and similarly, g(t) can only be differentiated
with respect to t. It is customary to drop the arguments at this stage and refer
back to equation (8.87) if necessary. Substituting equations (8.90) and (8.91)
into (8.47) gives
(8.92)

af'g =fg'

The

variable dependence can be separated by dividing

-V

The left-hand side of equation

must therefore be constant for

(8.93)
all

by

ocfg,

thus

is

(8-93)

independent of

time.

and the

ratio /"//

Similarly, the ratio g'/ocg

must be

constant at all positions because it is independent of x. Hence each half of

equation (8.93) must be equal to the same constant denoted by A, thus

and

r + A/-=0

(8.94)

+ Xocg =

(8.95)

g'

MATHEMATICAL METHODS

274

IN

CHEMICAL ENGINEERING

Equations (8.94) and (8.95) are both linear ordinary differential equations
with constant coefficients and the solutions are

and
if

(8.97)

/=.4o + BoX

(8.98)

g
is

(8.96)

not zero; or

is

and
if

f=Asin \IJx + B cos yJ'Xx

g = Ce-''''

zero.

CQ

(8.99)

Putting equations (8.96) to (8.99) into (8.87) gives the particular

solutions

T = Ao^-BqX
T=

and

(^^sin

VXx + B^cos^l x)e-^'''

(8.100)
(8.101)

where the arbitrary constant Cq has been "absorbed" into Aq and ^o^ ^^^
the subscripts X denote that the arbitrary constants in equation (8.101) will
depend upon the choice of X.
Ax, Bx, and X must be chosen to satisfy the boundary conditions
simplest assumption will be made that

at

and

To

at
satisfy

gives

Aq =

= 0,
= L,

T=
T=

and the
(8.102)
(8.103)

condition (8.102), Bx must be zero, and condition (8.103) then

and

=
which means that Ax

Ax{sm^/JL)e-^''

unless

y/XL = nn

(8.104)

Hence, the only solutions of equation (8.47) having the form (8.101) which
also satisfy the

boundary conditions (8.102) and (8.103) are given by

T=

/lsin^e-"'''/^'

(8.105)

where n

is an integer and the subscript has been altered for convenience.

Thus the most general solution of this type can be found by superposing the

particular solutions (8.105).

r= f A^sin'^e-"'^'''^'^'
L
n=l

(8.106)

The final boundary condition which is usually imposed on solutions of

equation (8.47) is one specifying the initial temperature distribution when
t = 0.
Thus, let
and

T=/o(x)

when^ =

substitute into equation (8.106).

Therefore

fo(x)=

f^sin^

(8.107)

(8.108)

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

275

satisfy equation (8.108) is accomplished by

using the properties of orthogonal functions. Comparing equation (8.94)
with the general Sturm-Liouville equation (8.83) shows ihsiip(x) = r(x) = 1,
q(x) = 0, so that equation (8.94) is of Sturm-Liouville type. The boundary
conditions (8.102) and (8.103) are of the suitable form (i) and thus the
to
functions sin (nnx/L) are orthogonal with respect to unity in the range

Choosing the constants A to

This property is utiHzed by multiplying both sides of equation (8.108) by

any other function from the set [sm(mnx/L)] and integrating the result
from to L. Thus

L.

/o(x)

sm

-^-dx= L

^n sin

sm

dx

L
1^

sin^j
by using the orthogonaUty property

(8.82).

dx

(8.109)

Evaluating the final integral in

equation (8.109) gives

L

A=-jfo(x)sm

dx

(8.110)

which can be evaluated from the known initial distribution /(a:). Substitution of the value given by equation (8.110) into equation (8.106) gives the
final solution

r=

IJsin^ J/o(x)sin^Jxe-"^"'/^^

(8.111)

convenient to introduce the nomenclature of the method in relationThe method relies upon the feasibility of
separating the variable dependence as in equation (8.93) so that a "separation constant" (-A) can be introduced. This constant can be zero, positive,
negative or even complex, and the choice is usually made on a physical basis.
The A used in the above calculation has to be real and positive to make the
solution of equation (8.95) physically reasonable.
Thus equation (8.97)
gives the time dependence of the particular solution as a decaying exponential for positive values of L If A were complex, its real part would have to
be positive to control the temperature at large times, and its imaginary part
would introduce hyperboHc functions into equation (8.101). The hyperbolic
cosine term would be removed by condition (8.102) and the hyperbolic sine
term by condition (8.103). Hence only real positive values of A are permitted
by the boundary conditions. It is not always possible to substitute the right
form for the constant initially, but the right form is usually apparent at a
later stage. In most of the illustrations of the method in this book, the most
It is

ship to the above calculation.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

276

convenient form will be chosen at the start for ease of presentation, although
in an actual problem this form would not be known.
Any other constant (such as a in the above problem) is normally transferred to the simpler side of the separated equation (8.93) for algebraic convenience. When the equivalent of equation (8.104) is reached, the separation
constant is usually restricted to an infinite number of discrete values which
can be labelled by an integer suffix. Thus
A

(nnlL)'

(8.112)

called the "eigen- values" of the problem because they are the
only permitted values of the separation constant. The functions arising from

and these are

the eigen-values are called the *'eigen-functions".

By comparing

the ordinary differential equation and its boundary condiwhich determine the eigen-values with the conditions given in Section
8.6.1, the orthogonaHty of the eigen-functions can be proved without the
necessity of performing the integrations. Hence, the last set of arbitrary
constants is determined by the orthogonaHty property.
tions

8.7.2.

Steady State Heat Transfer with Axial Symmetry

Laplace's equation in spherical polar coordinates with axial symmetry

was derived

analogous process of mass transfer in

was allowed. Putting u = v =
removing the time deriving from equation (8.52), and replacing c by T

the
0,

in Section 8.3.2 for the

more general

case in which fluid velocity

gives

d I

,dT\

,a^T

dT

d /

dT\

d^T

dT

'^^^'Tr^W^''''^ = '

(8-113)

Looking for solutions of the form

T=f{r)g{e)

(8.114)

by substituting into equation (8.113) gives

r'f"9

+ 2rf'g+fg" + cotefg' =

(8.115)

Dividing by^^ and separating the variables gives

.jriM'=_^_:^' =

,(,^i)

(8.116)

The conventional choice of separation constant

in this case

is

/(/+!) for

convenience of presentation. A can be chosen, but the algebra becomes

formidable and the above form of separation constant is then derived in the
argument leading to equation (8.132). The first part of equation (8.116) is

''t^ +

2'-t-

-'('+!)/=

(8-117)

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

which

is

Hnear and homogeneous with a solution of the form

f=Ar"

(8.118)

n{n-l)Ar" + 2nAr"-l{l+l)Ar"

.-.

:.
.-.

+ n-l^-l =
(n-/)(n + /+l) =

.-.

which

277

n^

f=Ar' + Br-^-'

(8.119)

partially explains the choice of separation constant.

The second

part of equation (8.116) can be written

g + cot0^ + + % =
/(/

(8.120)

Because of the product sin 6 dO in the denominator of the second term,

convenient to change the dependent variable by putting

m=

and

QOsQ

dm = - sin 9 dO

it is

(8.121)

Hence,

(l-m^)^ - 2m^ +

l{l+\)g

(8.122)

which is known as "Legendre's equation of order /". This has to be solved

by the method of Frobenius given in Chapter 3. Thus let

.-.

{\-m^)Y,(n + c){n + c-\)a^nf^'-^

-2mf;( + c)flm''-'^-^ + /(/ + l)f;am"-'^ =

The

indicial equation

is

found by equating
c(c
.-.

Equating

1)^0 =
c =
or

coefficients of

(8.123)

nf~^, hence

(8.124)

coefficients of m"""^ gives

(c
.-.

The recurrence
(s

relation obtained

+ l)cai=0
=
or fli =
by equating

+ c + 2)(5 + c + l)a,+2 =

(s

(8.125)

coefficients o{

nf^^

+ c-/)(s + c + + l)a,
/

is

(8.162)

The

roots of the indicial equation differ by an integer, but since a^ is indeterminate when c = 0, the solution falls into case Illb and the method
yields

two normal ascending power series solutions. As stated in Section

complete solution can be obtained in this case by putting c =

3.5.5, the
M.M.C.E.

10

MATHEMATICAL METHODS

278

IN

CHEMICAL ENGINEERING

and leaving Gq and Oj as the arbitrary constants.

relation (8.126) becomes
(5

+ 2)(5 + lK+2 =

(s-/)(s +

Thus, the recurrence

+ lK

(8.127)

2p will give the recurrence relation governing the series multipUed by Qq, and putting s = 2/7+1 v^ill give the other series multiplied by
a^. Thus
(8.128)
(2p + 2)(2;7 + l)a2p+2 = (2p-/)(2p + / + l)2p
Putting s

(2p + 3)(2;7 + 2)a2p+3

and

= (2p-/ + l)(2p + + 2)a2p+i

/

(8.129)

be useful, they must be convergent at all points in

= 0, i.e.
and in particular the series must be convergent when
m = \. The ratio of successive terms of the series defined by equation (8.128)
when m = \ is given by
If the solutions are to

space,

(2n-lH2n + l+l)

u^,

"'

where the symbolism of Chapter 3 has been introduced and the terms have
been renumbered. After some algebra, equation (8.130) can be put in a form
suitable for Gauss' test (Section 3.2.5) thus

4n^ + 2n-l^-l

'n+l

where

As

f(n)

==

n^

4n^^2n-l^-l

^'''''^

Hence the series is divergent by

is finite.
can be proved that the series defined by equation
(8.129) is divergent and thus both solutions are physically unreasonable for
general values of /. However, there are special values of / which are exceptions to the above analysis, and these provide the solutions.
Suppose / = 1. Then equation (8.129) with p =
gives
->> 00,

Gauss'

/()-* J/(/+l) which

test.

Similarly,

it

= (0)(3)a,
^3 =
= a-j = ag =
=

(3)(2)a,
i.e.
.*.

a^

...

Hence, when / = 1, g = a^m is a solution of equation (8.122). This can

easily be verified by substitution, and the solution is finite when m = I. The
solution

is

thus physically reasonable.

can readily be seen that if / is an odd integer, there will be a value of p

making the right-hand side of equation (8.129) zero and thus giving a polynomial of finite degree. Similarly, if / is any even integer, there will be a value
of/? making the right-hand side of equation (8.128) zero with similar effect.
These solutions are called the "Legendre polynomials" and are given the
It

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

symbol Pi{m). The solution of equation (8.122)

is

thus

g=^APim) + BQim)
where QJim) denotes

the infinite series.

solution exists mathematically, but

grounds and

it is

279

(8.132)

remember that this

invariably unreasonable on physical
It is as

well to

be rejected at this stage.

Combining equations (8.119) and (8.132) according to equation (8.114)
it

will

gives the particular solution

T=

(>l,r'

+ B,r-^-i)P,(cos0)

(8.133)

Again, any number of these solutions can be superposed for integer values
of / giving
7

=
1

{Ay + Bir-'-'^)Picose)
f
=

(8.134)

Equation (8.122) can be written in the form

d_

dm <'-''S
which shows that

it is

/(/

+ 1)^ =

(8.135)

of Sturm-Liouville type with unit weighting function.

The function /7(w) = \m^is> zero at both limits of integration (1 and + 1)

and thus the solutions in this range will be orthogonal for any boundary
conditions as shown by equation (8.85). Hence the Legendre polynomials
are orthogonal with respect to unity in the range from 1 to + 1
A convenient way of deriving the polynomials is given by Rodrigue's
formula
Pn{x)

= ^.^ni^'-^r

(8.136)

2"n\dx'

which gives
Po(x)

P2W = 3x2-l)

P,{x)

P,{x)

= \{5x'-3x)

^^-^^^^
\

etc.

for the

first

few polynomials.

Example 1. Two concentric spherical metalhc shells of radii a and b

(where b > a), are separated by a solid thermal insulator. The inner shell is
kept at a constant uniform temperature T^ by a steady distributed source of
heat. The temperature on the outer shell is found experimentally to vary
with angular distribution from a fixed point on the surface. The distribution
can be specified as

T = r2+r3cos^ +

T^^cos'^e

How

should Ti, T2, T^, and T^ be plotted against heat input so that the
thermal conductivity of the insulator can be found from a straight line plot
of results for different rates of heat input ?

The relevant differential equation is Laplace's equation with axial

symmetry in spherical polar coordinates (8.113). This equation has just
102

MATHEMATICAL METHODS

280

IN

CHEMICAL ENGINEERING

been solved in terms of Legendre polynomials to give the solution

7=

(Atr'

The boundary conditions given

at

a,

T=

Tj^

at

b,

T=

T2

+ Bir-^-')Pi{cosO)

in the

problem

(8.134)

are:

T^PoicosO)

+ T2Cose + T^cos^e

= 174 P2(cos 9) + Ta Pi(cos e) + {T2 + iT^)Po{cos 0)

by using equations

(8.137).

II

Putting these boundary conditions into equation

(8.134) gives

f=

f= o (Aia'-^Bia-'-')Pi(cose) =

T^PoicosO)

III

(Aib' + Bib-'-')Pi(cose)

(T2

+ \T^)Po(cose) + T2Pi(cose) + iT^P2(cose)

IV

Multiplying both sides of equations III and IV by P(cos 6)d(cos 6), inte0^9 ^n, and using the orthogonahty property of the
Legendre polynomials amounts to the same thing as equating coefficients of
the same polynomials in equations III and IV. For example, multiplying by
Pi (cos 9)d{cos 9) and integrating would remove all terms from both sides
grating over the range

except those which already contained Pi(cos

Ao + Boa-' =
Aia' + Bia-^-^

Hence,

9).

T,

(for

^0 + ^0^"' =

7^2

A,b + B,b-^ =

7^

>

VI

0)

+ 17;

VII
VIII

A2b^ + B2b-^=iT^
Aib' + Bib-^-^

IX

(for/>2)

Although there are an infinite number of equations and unknowns in the

above set, they can be solved in pairs. Thus equations V and VII can be
solved for Aq and Bq, and equation VI with / = 1 and equation VIII can be
solved for A^ and B^. Rather than complete the algebra at this stage, it is
more efficient to find the required answer in terms of Ai and Bi, then solve
sufficient

The
a

is

of equations

rate at

which the

to

for the constants required.

solid conducts heat

away from

the sphere of radius

given by
jt

Q=^ -kilna^'f^)

sin 9

d9

XI

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

and substituting into

Differentiating equation (8.134) with respect to r

equation

XI

gives
i^

Q=

281

00

llAia'-'-(l + l)Bia-'-^}Pi(cose)smede

-27r/cfl'
I

= -2nka^

[lAia'-'-(l + l)Bia-'-^]

Pi(m)dm

XII

But the integral occurring in equation XII is the orthogonality integral if

Pq(w) = 1 is inserted into it. Therefore, all terms after the first are zero in
the series in equation XII.

Q=

.-.

47ik

XIII

Bo

and hence only Bq need be determined from the

Thus

Ao + Boa~^ =

of equations

set

to X.

Ti

Ao + Bob-'=:T2+iT4
4
_

Bo(a-'-b-')=T,-T,-\T^

Bo=~(T,-T^-\T^)

:.

XIV

Substituting into equation XIII gives

e = ^(r,-r,-iT,)
ba
Therefore,

if

is

plotted as ordinate

XV

and {Ti-T2-^ T^)

as abscissa, for a

variety of heat supply rates, the gradient will be 4nkab/(b a)

from which

the thermal conductivity (k) can be evaluated.

8.7.3.

Equations Involving Three Independent Variables

The steady state flow of heat in a cyHnder

equation in cylindrical polar coordinates (7.125)

d^T
2

do)^

J_^ +
+ 0)^
Z2i^
0) do
d9^

is

]^dT

djj

T:i7:

dz

^=

governed by Laplace's

(8-138)

where there are three independent variables a>, 6, and z. This equation has
to be solved in stages and solutions are first sought in the form
T=f(co,e).g(z)
Substituting this

form into equation

.25

(8.139)

(8.138) gives

+ -5r,s + -i^9+/9" =

Dividing by fg and separating the terms dependent upon z

ir^ + i^ + ^]=Z^=_,2

(8140)

MATHEMATICAL METHODS

282

CHEMICAL ENGINEERING

IN

where -v^ is the chosen separation constant. Each side of equation (8.140)
must still be constant because a function of co and 9 is equated to a function
of z only.
The second half of equation (8.140) gives

^2-^9 =

(8.141)

(8.142)

which has the solution

g

The other

A,e'' + B,e-'"

half of equation (8.140) gives the partial differential equation

.2^

co^T-^

^-^

+ co-^+
,

,..

^ + co^vy=0

^ ^

_2,.2

(8.143)

which has only two independent variables. Applying the rr.-thod of separation of variables again, by looking for solutions of the type

f(co,e)
.-.

Dividing by

FG

F(co).G(e)

(o^F"G + coFG + FG"-\-coh^FG

and separating the

(8.144)

(8.145)

variables gives

(o^v^

^ = k^

(8.146)

Solving the second part as before gives

G^Ak cos kd + Bk sin kO

The

first

(8.

147)

part of equation (8.146) gives

(o^F" + coF'

+ ((o^v^-k^)F =

Changing the variable by substituting

cov

.'.

x^^ + x^
+ (x^-k'')F =
dx

(8.148)

(8.149)

<ix^

which

is the standard form of Bessel's equation (3.48).

The most convenient form has been chosen for both separation constants, and k must be an integer for the trigonometrical functions in equation
(8.147) to be cyclic with period 2n. The boundary conditions usually make

V real as

chosen.

Equation (8.149) has the solution

F=

AJ^(x) + BY,,(x)

(8.150)

but at the axis of the cylinder, Yj^x) = Yk(va)) is infinite and thus B =
on
physical grounds. Combining equations (8.142, 147, 150) gives the particular
solution

T=

(A,e''

+ B,e-'')cos{ke + )J^(va))

(8.151)

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

Solutions of this type can be superposed for integer values of k,

283

and

values of v permitted by the boundary conditions, which involves a double

summation.
BesseFs equation (8.149) can be rewritten as

d f dF
(x^-{x-k''x-^)F =
dx\
which is of Sturm-Liouville type with a weighting function {Ijx). The range
of integration is dependent upon v because of equation (8.148) and it is not
However, for suitable
useful to proceed further with the general case.
boundary conditions, the eigen-functions will be orthogonal because equation
(8.149) is of Sturm-Liouville type.
8.7.4.

General Use of the Orthogonality Property

If the separated partial differential equation is

of Sturm-Liouville type

and the boundary conditions are suitable, the solution of the complete
partial differential equation takes the form
(8.83)

T = aofomoM + aj,m,{x) +
where

t is

. .

the other independent variable

+ aJMU^) +

and

<t>(x)

other boundary condition will be given at a fixed value of

T = f(x)). Therefore
/(x)

(say

= Ao0oW + ^0i(x)+...+^(^(x)+...

Multiplying equation (8.153) by

orthogonality integrals gives

r(x)<t)ix)

ir(x)f{x)<l>Mdx

But all terms

Hence

(8.152)

in the infinite

sum

The

are orthogonal.
f

0,

(8.153)

and integrating to generate the

Act>(x)r(x)(l>Mdx

are zero except the term

(8.154)

when n = m.

jr(x)f(x)cl>Mdx

= JArix)l<l>M?dx

or rearranging,
b

!r{x)f(x)^Jx)dx

A =

I
8.7.5.

(8.155)

'-,

r(x)l4>{x)ydx

Eigen-functions not Orthogonal

Sometimes the variables

will separate in the partial differential

equation

to give an equation of Sturm-Liouville type, but the

boundary conditions are

eigen-functions are not orthogonal. If an attempt is

such that the resulting

made to use the orthogonality relationships, the general equation (8.154)

is

MATHEMATICAL METHODS

284

CHEMICAL ENGINEERING

IN

obtained, but the terms in the infinite sum are not zero. Instead of each
equation (8.154) being solvable for A, each equation contains all of the
^^s and thus an infinite set of equations, each containing an infinite number
of unknowns is obtained. Such a system of equations is normally insoluble
and an approximate solution has to be found by assuming that A-j, A^, ....
etc. are all zero and solving the first six equations for the first six coefficients. The accuracy can be improved by taking an increasing number of
equations and coefficients, but the effort is rarely worth while and six equations

is

the normal practical limit. There are however

ing an exact solution

Example

2.

and these are

A solid rectangular

and maintained

two ways of

find-

slab at a uniform temperature Tq has

The temperature of one exposed

four edges thermally insulated.

raised to

still

illustrated below.

face

its
is

temperature of the other exposed

the temperature distribution varies

at Tj, whilst the

Find how
is held constant at Tq.
with time.
Because the edges are insulated, heat will only flow in one direction and
equation (8.47) will describe the process. The solution follows the method
given in Section 8.7.1 identically until the particular solutions (8.100) and
(8.101) arise. The boundary conditions can be written:
face

= 0,
= L,

at

and

at

These do not

satisfy the

r=
T=

To

T^

II

conditons of orthogonality given in Section

Substituting equations

and

8.6.

II into the particular solution (8.100) gives

To

= Ao

III

T,=Ao + BoL
Hence, the particular solution

IV

is

T=^To + (T,-To)xlL
and

this should be recognized as the steady state solution.

Equation V is now used to define a new variable Z thus,

Z=

VI

r-To-(Ti-ro)x/L

Substituting into the differential equation (8.47) gives the

dx^

dt

which is the same as equation (8.47) with

conditions I and II become

= 0,
x = L,

at

and

at

The equation and boundary conditions

new equation

T replaced

by Z.

The boundary

Z=
Z=
in

terms of

VIII

IX

are

now

identical with the corresponding equations in Section 8.7.1.

solution

is

given by equation (8.111) with

T replaced

by Z.

completely
Hence, the

In the present

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

8.

problem, the

final

boundary condition that the temperature

is

285

Tq at

all

points initially becomes

Z = -(Ti

X with

Comparing equation

when

Tq)xIL

equation (8.107) shows that

/oW=

XI

-{T,-To)xlL

L
[
J

nnx
nnx
T^-Tol
r
X cos
/oW sin L dx = nn
L
.

cos
J

1^
nnx

dx

L
JQ

Zii::Z2r(_i)"L]

XII

nn
by integrating by parts and simplifying.

Substituting equation XII into the

solution (8.111) gives

Z=

y
=

{-mT,-To)sm'^e-"'^'^'^'^'

XIII

nn

Returning to the original variables by using equation VI,

^^:^ =
Ti-Tq

-+

-(-l)"sin
E= onn
L

e-"^'^^-/^^

XIV

The above method, which should be adopted when

the boundary condiand II will not give an orthogonal set of eigen-functions, involves
finding any particular solution (such as V) which satisfies the conditions and
defining a new variable (Z) so that the new solution consists of an orthogonal
set of eigen-functions. In this example, the first eigen-function is given by
equation V and provided it is removed from the others an orthogonal set
remains and the method can be pursued as if it were normal.
tions I

Example 3. In the study of flow distribution in a column packed with

woven wire gauze,t it has been observed that the hquid tends to aggregate at
the walls. If the column is a cyhnder of radius b ft and the feed to the column
distributed within a central core of radius a ft with velocity Uq ft/sec,
determine the fractional amount of liquid on the walls as a function of distance from the inlet in terms of the parameters of the system.
is

The problem is illustrated in Fig. 8.8. where the vertical velocity C/ ft/sec
considered to be a function of the radial coordinate r and the vertical coordinate z measured from the inlet. It is assumed that a normal diffusion
equation
is

dU
V=-D
dr
appHes to the velocity distribution.
fluid velocity
ties.

will

t Porter,

and

D ft is

F ft/sec

is

the horizontal

component of

the coefficient governing the equalization of veloci-

be a property of the packing and of the particular fluid used.

K. E. and Jones, M. C.

Trans. Instn.

Chem.

Engrs., 41, 240 (1963).

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

286

Taking a material balance over the element shown

Input
^

Output

in Fig. 8.8 gives

dU

= InrU dr D

--Inrbz
dr

= 2nrUdr +

dU
d (InrU Sr)dz '-D-2nrSz
oz

Accumulation

dr

[D

dU

InrSz]
\

=
d / dU
\
- -(D-2nrdz]dr
^Q

{InrU dr)dz

or\dr
dU ^d / dU\
'Tz=''drV^)

dz

"

dz

The boundary conditions

z

II

III

[dr^ '^r'di^J

Flow

Fig. 8.8.

At

d /

dr\dr

in a

packed column

are as follows.
0,

<

if

if

r>a,

a,

U = Uq

IV
and

at

0,

U=

must be

finite

boundary condition at r = b must allow fluid to accumulate on

and a reasonable assumption is that there is a capacity 2nbk ft^/
unit length at r = b. It is further assumed that fluid which has diff*used to
the wall is equally distributed in this capacity, so that the same flow rate
exists in this fictitious layer on the wall as in the packing in contact with the

The

final

the walls

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

8.

287

This condition can be stated symbolically thus,

wall.

at

dU

dU

dr

oz

-D-2nb = 2nbk--

b,

dU
dU
= k-^D
oz
or

i.e.

VI

The problem is thus to solve equation III, satisfy the initial condition IV,
and satisfy the boundary conditions V and VI. Trying the method of separation of variables by putting

VII

U=f{r)g{z)
.-.

= D(f"g+f'glr)

fg'

Dividing by Dfg gives

^J.1PL=_,2
Dg
where -a^

is

VIII

f
The

the chosen separation constant.

first

part of the equation

gives

^ + a'Dg =

dz

which has the solution

is real and has been chosen

The second part of equation VIII gives

indicating that a

.2*^ J

^^J

+
X

IX

A'e-''"^'

r-f

correctly.

22
+ a^rY=0

Equation X can be transformed to the standard form of Bessel's equation

(3.48) by putting x = ar. The solution is thus

=
a

0,

ifa

0.

if

AJo(ar) + Byo(ar)

XI

or

/=^o + ^olnr
Since Yq{0)

Bq

= B=

0.

= oo

The

and InO

= oo,

XII

boundary condition

V asserts

that

solutions of equation III are thus

if

oi

if

= 0,
# 0,

U = Ao
"'^'^^
1/ = AJoio^r) e

XIII

XIV

Boundary condition VI is satisfied by solution XIII, and putting equation

into equation VI gives

XIV

D^a7i(a6)e-"'^^
by using equations
Simplifying equation

(3.88)

and

-kA^oi''DJo((xb)e-

''"'''

XV

(3.97) to differentiate the Bessel function.

XV gives
Ji(a b) + kaJo(oi b)

XVI

MATHEMATICAL METHODS

288

IN

CHEMICAL ENGINEERING

which determines the eigen-values a. There are an infinite number of

solutions of equation XVI, the first few of which have been tabulated.! The
most general solution of equation III is therefore obtained by adding all

Thus

permissible particular solutions together.

U = Ao+

XVII

Y. AJo{ocr)e-'^'^'

where the eigen-functions are zero order Bessel functions of the first kind.
It can be shown by rearranging equation X that it is of Sturm-Liouville
type with a weighting function r. Boundary condition VI is not one of the
types Hsted in Section 8.6 and hence the eigen-functions will not be orthogonal. Nevertheless, the usual method can be followed by putting z =
and multiplying equation XVII by r Jo(ocr) and integrating. This gives
b

Ul^orJo(^r.r)dr

''

00

AorJo(cc^r)dr

+
"

X
"

^n
^

Jo(^

r)Jo(of,, r)r

It is

simpler to evaluate the terms of equation XVIII separately.

first

term becomes
b

^r

XVIII

Thus

the

^\z = ofJo{^mr)dr=

U qvJ ^i^x^r) dr

=
by using equation (3.88) with k

Similarly, the next

\.

XIX

U,aJ,{oi^a)loi,

term gives

= AqB

AorJo(<xj)dr

J^(cc^b)lcc^

= AokbJo(oc^b)
Taking a general term with n
equation (3.92) gives
r

An

JoKr)Joi<^mr)rdr

9^

from the

infinite

XX
sum and

using

Ah
-Y^[ocJo(cc^b)J^{ccb)-cc,Jo((xb)Ji(ci^b)]

^"
^

an

-am

- koi'Jo(o^nb)Jo(o^m b) + k^lj ,{a,b)J ^{oi^ b)-\

because a must satisfy equation XVI. Therefore

h

>lJjo(ar)7o(a.r)r<fr= -/c/l b J ^{oib)h(oiM

For the special term

Hence

in the series

when

XXI

m, equation (3.93) must be

used.

Crank,

J.

Carslaw, H.

S.

Press (1947).

" The Mathematics of Diffusion."

Oxford University Press (1956).
and Jaeger, J. C. " Conduction of Heat in Solids." Oxford University

PARTIAL DIFFEREISTTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

8.

AjlJo(v)yrdr =

289

ifoM[J?(afo) + J?(a^6)]

XXII
by using equations (3.97) and XVI.
Combining equations XIX, XX,

Ji(o:^a)

XXI and XXII

into equation XVIII,

=-Aokb Jo(oiM + ib'A^(l + k'o^'JJl(^M

am

Completing the

infinite series

XXIII

kAbJo{oi,b)Jo(oi^b)

by inserting the terms k

A^b JliccJ))

twice

with opposite signs, therefore

Ji(a.)

= -A,b

k Jo{a^b) + iA^{b' + 2kb + b'k'ai)Jl(oib)

am

XXIV

-kbJoioi^b)tAJo(o^nb)
But when z

and

b,

equations IV and

= ^0+

XVII

give

XXV

t^nJoio^nb)
n=l

Because

all

of the terms in equation

XXV

occur in equation

XXIV

with a

factor khJoia^b), then

Una
-Jii^mO) = ^AJb^ + 2bk + b^eal)Jl{oiM

both fortunate and unusual that equation XXV

A from equation XXIV. The solution by
from equation XXVI into equation XVII is thus

which determines A^.

eUminates
putting Afn

XXVI

It is

all coefficients

U = Ao+ Y
^,

except

IUqU
a

Ji(cca) Jo(ccr)

{b'

,-alDz

+ 2bk + bVoi')Jl(ab)

Taking an overall material balance

XXVII

at large values of z,

na^Uo = {nb^ + 2nbk)Ao

:.

a^Uo

b{b

+ 2k)Ao

Which determines Aq, the equilibrium velocity

The fraction of total fluid in the wall layer

F=

XXVIII

in the packing.
is

given by

27ibkU

noHTo
2k
6

+ 2/c^t^i

Abk
aa {b^

J,(afl)g-''^"

+ 2bk + b^k^al)Jo{^b)

XXIX

mathematical methods

290

in

chemical engineering

The Laplace Transform Method

8.8.

was shown in Chapter 6 that a Laplace transformation would remove

It is thus fairly
the derivatives from an ordinary differential equation.
obvious that the same technique can be used to remove all derivatives with
respect to one independent variable from a partial differential equation.
However, a particular independent variable can only be removed if it has
an open range, because the Laplace transformation involves integrating with
respect to the variable from zero to infinity. Since the boundary conditions
can only be used at the extremities of the range of integration, they must all
be imposed at zero. Hence the variable to be transformed must have an
open range as described in Section 8.4.5.
If the partial differential equation has two independent variables the
removal of one of them yields an ordinary differential equation which can
be solved by the methods of Chapter 2. An added compHcation in the study
It

of partial differential equations

is

the occurrence of arbitrary constants in

the solution of the transformed equation. These constants

before the solution

is

must be evaluated

inverted because they are normally functions of the

transform parameter as will be seen in the examples. Hence the boundary

conditions which are not used to transform the equation must themselves
be transformed.
These ideas are better illustrated by solving specific equations, and in
the first three examples below, the one dimensional heat conduction equation will be solved again for the same selection of boundary conditions as
used in the previous sections of this chapter. This duphcate derivation
emphasizes the versatihty of the Laplace transformation by determining all
of the solutions by one technique instead of using a different method for
each specific example. The later examples illustrate the solutions of different
partial differential equations.

Example
infinite

medium.

Solutions of the linear heat conduction equation in a semiTaking the now familiar form of the conduction equation

(8.47),

and using the

initial

at

condition that
r

0,

T=

the Laplace transformation can be used to

Equation (6.11) from Chapter 6 gives

Because

;c

and

t
are independent variables, then

L^x'J

~ d

II

To

remove the variable

as follows.

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

Hence the transform of equation

I is

a3 = sT-To
Equation

III is

291,

ffl

a second order linear ordinary differential equation having

the solution

T=
q^

where

/le-^ + Be*+ro/s

IV

s/a

has been introduced for simplicity of presentation.

(a) The first problem, solved in Section 8.5.1, involved maintaining the
exposed face of the slab at a fixed temperature, i.e.

at

0,

r=Ti

0,

T = TJs

VI

Transforming equation VI gives:

at

As X approaches
finite.

Therefore

T must remain finite, and hence T must remain

and condition VII substituted into equation IV gives

infinity,

B=

EUminating

VII

from equation FV

gives

r = (ri-To)(e-^5) + To/s
Inverting by using transforms 83

and

VIII

in the table gives

T = (Ti- To) erfc (x/2 V^) + To

which can be compared with equation (8.74).
(b) In Section 8.5.2 it was shown that equation
expressing the

movement of a

(8.78)

IX
was the solution
body

fixed quantity of heat within a solid

the heat is initially concentrated at the surface. This solution will now
be derived from equation IV.
If all of the heat is at the surface initially, then Tq = 0; and since B =
to satisfy conditions at infinity, equation IV becomes

when

r = ^e-^
If the total heat content

(H) of the body

H=

'

is

X
constant,

pC.Tdx

(8.79)

Transforming equation (8.79) gives

XI

MATHEMATICAL METHODS

292

Using equation

for

T and

integrating, therefore

HA

7r-

pCpS
Cs
Eliminating

A from

CHEMICAL ENGINEERING

IN

equation

= -=
a
q

la
Aj\

XII

gives

XIII
pCpsjoL

^i

\ls

This can be inverted by using transform 84 to give

r=

XIV

r^e-"'/^-^

pCpVaTT

which

given by equation (8.81).

from Section 8.5.2, heat was supphed at a fixed rate to
the exposed face and this boundary condition can be expressed by the
(c)

is

identical with equation (8.78) with

In Example

equation

at

dT

Q=-i^

0,

XV

ox

Taking the uniform initial temperature as the datum and applying the
physical Hmitation on the temperature at infinite depth, equation IV simphfies

to

r = ^e-*

XVI

Transforming the boundary condition XV,

at

Substituting equation

XVI

ks^ = -ox

0,

XVII

gives

-Aqe-^''=

-Aq

into equation

-Q
-^=

XVII

ks
:.

A=

XVIII

Qlkqs

j^Q:l^^-3/2^-^Jiix

XIX

Because integrating a function with respect to t is equivalent to dividing its

transform by s, as shown by equation (6.21), equation XIX can be inverted
by integrating the function given in transform 84 in the tables. Thus

which

is

the

same

J Vtct

result as that given

by equation

III in the original

example.

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

At the surface

(x

0),

XX gives

equation

2Q

lou
'at

k
as in

Example

1,

293

XXI

Section 8.5.2, equation IV.

Example 2. Heat conduction between parallel planes. In Section 8.7.5,

the second example considered the flow of heat between parallel planes
maintained at different temperatures. The boundary conditions were:
at

at

at

Because condition
(8.47)

is

as above,

=
=
=

T=
T=
T=

0,
0,

L,

To

11

T,

III

the same, the Laplace transformation of equation

I is

and the general solution

f=
q^ =

where

at

at

is

^e-^ + 5e^^ +

IV

ro/s

s/cc

Transforming the boundary conditions

=
=

II

and

III gives

T = To/s
T = TJs

0,

A and

Using these conditions, solving for

IV

To

B,

VI
VII

and substituting into equation

gives

T=

'

'

VIII

first term on the right-hand side of equation VIII must be inverted by

method of residues (see Section 6.8) as follows. The denominator is
zero when either
IX
5 =

The
the

qL

or

inn

(i.e. s

ccn^n^/l})

and all of the above poles are simple poles.

The residue at 5 = is

which can be evaluated by L'Hopitars rule to give

residue at

The

simplest

is

(Ti

method of evaluating

tion VIII in terms of

Section 4.16.1.

Tq)xIL

XI

the other residues

is

to express equa-

hyperboHc functions and use equation (4.113) from

Thus

f^

T,-To sinh(Vs/ax)
s

'

s'mh(yJsl(xL)

Tp
^

^^

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

294

The

residue at s

is

cosh(\/s/aL)

Sn

or using equation X,

inn

Adding
1

cosh(i7t)
all

mtc

of the residues together, and inverting the

last

term by transform

in the tables, gives

=7+1 -(-D-sin^e-"^"^-/^^
J^
L
Ti-To L %nn
as in

Example

Example
example

3.

2,

XIII

Section 8.7.5, equation XIV.

Symmetrical heat conduction between parallel planes. The

Unear heat conduction equation

in Section 8.5.3 again involved the

with the boundary conditions

at

0,

T=

at

0,

T=

at

L,

dT
- =

II

To

dx
Solving as in Example

above gives

T=
where

q^

/le-* + Be^*

0,

s/cc

The transformed boundary conditions

at

IV

are

T=

VI

at

dT
L,

ox

Using conditions VI and VII to determine

A and B

VII
in equation

IV

gives the

transformed solution

This could be inverted by the method of residues as in the

to obtain the error function form given in Section 8.5.3,
expand equation VIII in terms of exponential functions.
similar to that used in Section 6.7.3 and proceeds thus.

last
it is

example, but
necessary to

The method

is

8.

PARTIAL DIFFERENTIATIONS AND PARTIAL DIFFERENTIAL EQUATIONS 295

lqL_

-^^'^){\-e-

2qL_

4qL^-l

4,L_^^-8.L^

e-^"(l-e--2^^)(l4-e-

= ^-^1-^'-2qL_^^-4gr-_g-6L_^g-8qL_
00

= Z(-i)"'^-(2nL + *)
n

Similarly,

it

can be shown that

y
n

Substituting equations

IX and

by using transform 83 in the

r=

IX

/_l\"g-(2nL+2L-x)q

X into

VIII, using

and inverting each term

tables, gives

f= - 1)" {erfc [(x + 2^zL)/2V^] + erfc [(2nL + 2L - x)/2 V^O

To

XI

which agrees with its counterpart in Example 2, Section 8.5.3.

The above solutions of the linear heat conduction equation and the
analogous linear diffusion equation are all contained in Crank or Carslaw
and Jaegert, together with many other solutions for different combinations
of the four types of boundary condition in the three common coordinate
systems. As pointed out by those authors, the error function type of solution arising from the Laplace transform is useful for determining the behaviour of the system in the early stages, but the summation of eigen-functions
is more useful for determining the approach to equilibrium in the later stages
of the process.

Example 4. Exploitation of an oilfield. An extensive shallow oilfield is to

be exploited by removing product at a constant rate from one well. How
will the pressure distribution in the formation vary with time ?
Taking a radial coordinate r measured from the base of the well system
it is known that the pressure (/?) obeys the normal diffusion equation

^+1^ = 1^
dr^

where
oil

rj

is

dr

,.

the permeability, h

is

is

assumed to be uniform

at Pq.

If

Inrkh dp

\im
r->o

is

dt

the hydrauhc diffusivity.

The initial pressure in the formation

is removed at a constant rate q, then

where k

rj

/^

II

^r

the thickness of the formation,

and

/i is

the

coefficient of viscosity.
t Crank,

J.

Carslaw, H.

S.

Press (1947).

"The Mathematics of Diffusion." Oxford University Press (1956).

and Jaeger, J. C. " Conduction of Heat in Solids." Oxford University

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

296

Transforming equation

by the same procedure as given in Example

dp
_
po
p=
2+7^--p=-r dr

ni

rV^

IV

d^p

4-

dr^

rj

t]

Putting

X
x^

which

is

- x^p=
x
dx

^ +
dx^

the modified form of Bessel's equation (3.64) having the solution

p==AIo(x) + BKo(x) + pols

VI

Equation (3.82) shows that as x->oo, Iq(x)-^ co and hence

^ =

p=^BKo(x) + pols
Transforming boundary condition

Using equation (3.71) that

x->0 gives

VII

II gives

= lin,2_^l^

r-o

/o(0)

and

1,

VIII

dr

A^

and taking the

limit of equation (3.68)

as

IX

Ko(x)^-\nx
Therefore, near the base of the well, equation VII becomes

p^ -B\nx =

-Blnir^Js/rj)

Substituting into equation VIII, therefore

q
-

InrkhB

,.

lim

InkhB

5^^ifi^_
2nkhs

^^
XI

XII

Hence the transformed solution VII becomes

p

= a Ko(r^/slri)

Po
1

Since integration of a function with respect to

transform by

s,

as

shown by equation

Xlll

(6.21),

t is

equivalent to dividing

its

transform number 117 in the

tables enables equation XIII to be inverted, thus

t

P--JI-^e-^'"''dt + p,
Putting

and hence

dt

XIV

XV

r^l4rjz

4r]z^

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

297

20

XVI

dz + pi

where Zq

r^jArjt.

(5.44), the solution

Comparing the integral in equation XVI with equation

can be written in terms of a tabulated function. Thus
P

Example

=
4nkh

EK-r^l4r]t) + Po

XVII

Gas absorption accompanied by a first order reaction in a

wall column is to be used for the absorption of a gas
(A). The gas is consumed in the liquid phase by a pseudo first order reaction
Develop expressions giving the point
in terms of the gas concentration.
absorption rate and effective penetration depth as a function of distance
from the hquid inlet.
Figure 8.9 shows a section through the liquid film where z is the distance
from the liquid inlet, and x is a coordinate measured inwards from the

falling film.

5.

A wetted

Velocity distribution in a falling film

Fig. 8.9.

surface of the film which

it is

is

assumed to be of uniform thickness L. Although

a poor assumption which depends

velocity distribution v{x) will be

upon

inlet conditions, the

assumed to

parabolic

exist right at the inlet

and

remain unchanged throughout the length of the column.

The gas is assumed pure so that there will be no gas phase resistance and
the surface concentration of A in the liquid will be constant at Cq, For any
element of the hquid (3x Sz) to be in equilibrium, the amount of A carried in
the z direction by bulk flow, plus the amount diffusing in the x direction,
must be sufficient to supply the chemical reaction proceeding within the
element.

Expressing this fact symbolically.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

298

dc

Input rate

vcSx Dzr^^
ox

Output

vcdx

rate

Consumption
where

is

rate

dc
d
~-(vcSx)5z D^-dz
ox
dz

d /

t-

dx\

dc

dx

Ddz]Sx

kc dx Sz

the concentration of

the specific reaction rate constant.

D is its diffusivity, and k is

Completing the balance and cancelling

in the liquid,

appropriate terms gives

dz

Transforming equation

dx^

by the method previously explained gives

svc + D^ =

kc

dx^

D-7^-(/c + st?)c =

II

dx

But V = V(lx^/L^) where V is the interfacial velocity, and equation II

would have to be integrated using the method of Frobenius to obtain an
accurate solution. However, the penetration of A into the liquid will be
small for small values of z and as a first approximation it will be assumed
that x^L so that v can be replaced by Fin equation 11. Thus

d^

D-j-:2-(k + sV)c

III

dx'

which is a linear second order differential equation with constant coefficients

having the solution
c

As x
A'

X'exp(xy^) + Bexp(-xy^)

increases, the concentration of

0.

At

:c

0,

must decrease

the boundary condition gives c

to zero,

Cq

Cq/s

c,

Using transform 82 and the shifting theorem (Section

6.2.2),

IV
and hence

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

299

Applying the property given in equation (6.21) to equation VI and inverting

equation

gives

l~V~

fx

x^V

Putting

a^

= x^F/4D

and

p^

kz\

^^

klV

VIII

to simplify the algebra, gives

z

It is diflScult to

o'^

Co

evaluate this integral, but

and hence

r^

its

form can be changed by putting

+ p^z-2aP
z

dr=(^-^^-^)dz

and
into equation

IX

to give

00

/.

Coe-^'^eifcf-^

py/zj

CqP

-p=exp( -

- - P^z\dz

XI

of P were reversed in equation IX, the integral would not be

any way. Hence the sign of P may be reversed in equation XI

If the sign

altered in

without altering the

result.

Thus
z

Coe^-'erfc^-^

V^sT^ +

Adding equations XI and XII

Co;8

-^exp(--^ - jJ^z) dz

together, dividing the result

XII

by 2 and returning

300

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

to the original

symbols through equation VIII yields the

+
The point absorption

iexp(xy^)erfcg7^Vy)

rate per unit area

R=
Transforming equation

final result

XIV

"I

is

given by

XIV
x=

gives

R= -D
dx

and substituting

for c

XIII

from equation

Dco

XV
x=

gives

k\

IV (

XVI
Inverting equation XVI by using transform 38
(Section 6.2.2), therefore

R=
:.

Cos/oVe'

kz/V

and the

shifting

theorem

'k

V
Vt.^JP'"'^^^'^"']
R = Co^/kD [erfV/cz/F + V^/^exp (- fcz/F)]

XVII

In order to determine the depth of penetration of the solute, some

must be set for its concentration to be negligible. This hmit

arbitrary limit

Fig. 8.10.

chosen as c/cq

Approximation for

erfc

and equation XIII gives an implicit expression for

x as a function of film length z. llie relationship is
simple enough to solve by trial and error except for small values of z but an
approximate solution can be found for this region as follows.

is

002,

the penetration depth

Figure 8.10 shows {2l\ln)e~^^ plotted against y so that erfc>' is the

If y is large it is reasonable to suppose that erfc y is some

shaded area.

8.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

301

constant multiple of the area of the triangle formed as shown by the tangent
to the curve. Putting

h=-^e-'"
dh

7'-

dy~

XVIII

-y'

2yh

but from the triangle,

dh

_ -h

dy
a

:.

XIX

iy

The approximation can be written

crfc y

where

K is

XX

= iKha

some proportionality constant greater than unity.

and a from equations XVIII and XIX gives

Substituting for h

erfc3;

Evaluating

K=

1-71.

K by putting y
Hence

= ^e"^'
on both

1-5

XXI

sides of equation

erfcy=^(0-484/3;)e-^'

XXI

gives

XXII

Ify = 2-5, the error in equation XXII is less than J%, and if >^ = 1, erfc y =
0-178 whereas (0-484/;;)e->'' = 0-157. Hence equation XXII should only be
used for >^>1.
Introducing dimensionless coordinates into equation XIII by putting

X = xyjklb
c

and Z^

kzjV

XXIII

ie-erfc(|-z)+i.^erfcg + z)

Co

Using equation XXII gives

c

Co

0-242

2Z

0-242

2Z

0-242Z

is

vahd

+z

(-.?-)

X^

expi-7^-Z'j

(|i-)
which

exp

4Z

for

^>1+Z

XXIV

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

302

The penetration depth

0-002

Solving by

Z^

for

trial

is

thus given by

(^ - ^^) = ^'^^^^ ^^P ( ~ ^ ~ ^0

and error using equation

XXV

for

Z<

^^^

and equation XIII

gives

0-1

0-5

1-0

1-5

2-0

3-0

0-435

2-130

4-00

5-33

6-00

6-215

which

is

plotted in Fig. 8.11.

AQ

Fig. 8.11.

Relation between penetration of solute and length of wetted wall

8.9.

Other Transforms

In the last section, the Laplace transform was

shown

to be quite versa-

and although the transform involved introducing an exponential function, this did not influence the type of solution obtained. The chief restriction on its use is that the problem must be of initial value type, but provided
the dependent variable and its derivative remain finitet as the transformed
variable tends to infinity, no analytical difficulties arise.
The inversion
process is often tedious if the method of residues has to be adopted, but in
tile,

principle the inversion process can always be completed.

t

The method

infinite to

values so there

is

applicable when the dependent variable or its derivative become

exponential order, but physical variables never actually take infinite

is still

finite

no

restriction.

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

8.

303

No

reason has been given for using the function e"*' as the "kernel" of
and attempts have been made to use other functions
as kernels thus giving rise to other transforms. The purpose of the present
section is to give the names and formulae for these other transforms and
indicate their range of applicabihty.
the Laplace transform

8.9.1.

Finite Fourier Transforms

finite Fourier sine transform can be derived from the general formula
developed in Section 8.7.1 for linear heat conduction. Replacing the boundary condition (8.107) into the general solution (8.111) and dropping the
suffix from/o(x) gives the identity

The

which

is

and no longer related to the specific problem of

obtain the conventional notation, put z = nxIL,

universally true

To

Section 8.7.1.

.*.

The

finite

/W = - E

Fourier sine transform

/()

sin

/(z) sin nz

dz

(8.157)

defined as

is

nz

J/(z) sin

MzJz

(8.158)

Equation (8.157) provides the means of inverting the transform. Thus

/(^)

which

is

transform

very

much

= -Z/Wsinnz

(8.159)

simpler than the inversion formula for the Laplace

(6.31).

The most important properties of the finite Fourier sine transform can
be determined by transforming a second derivative. Thus,

}d^v

-T-jSm wzdz
Provided the

first

practical cases)

[dv
= \-sm nz

derivative

and

is finite

at

-cos nzJz y

rev

both ends of the range

(8.160)

(as

it

will

be in

since sin nz vanishes at both ends of the range, then

71

7t

)2

^r-T

J az^

sm nz dz = n

^- cos nz dz
J dz

= n\Vcosnz +

V sin nzdzl

= -n^V+nlVo-i-iyV^-]

(8.161)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

304

Thus the value of

restrictions

S(a)

S(b)

must be known at both ends of the

needed to show the following severe

the dependent variable

Little further investigation

range.

on the use of the

is

Fourier sine transform.

finite

The variable to be transformed must have a closed range.

Type 1 boundary conditions must be given at both ends of the

S(c) All derivatives to be

range.

transformed must be of even order.

Similarly, a finite Fourier cosine transform can be defined

J(n)

/(z)

= ^/(O) + ^

jf(z) cos

nzdz

by
(8.162)

and inverted by

71

f Kn) cos nz

71..

(8.163)

but again the following restrictions apply.

The variable to be transformed must have a closed range.

C(b) Type 2 boundary conditions must be given at both ends of the range.
C(c) All derivatives to be transformed must be of even order.
C(a)

8.9.2.

Fourier Transforms

Equation (8.156) can be written

L

Kx) =

2 sin

jfix) sin

dx 5

g)

(8.164)

where 5n = \. Putting/? = nnIL and taking the limit as L->oo, then S(nK/L)
-^dp and the sum becomes an integral in terms of the "continuous" variable
p.

Thus
00

00

fix)

The Fourier

=-

sin

sine transform

is

px

f(x) sin

px dx dp

(8.165)

defined by
00

Jip)=\ fix) sin pxdx

and equation

(8.166)

(8.165) gives the inversion formula

00

/(x)

it.

(8.167)

mathematically elegant because the same formula

used both for forming the transform and for
Although the Fourier sine transform is less restricted than the

This transform

is

(except for a constant

inverting

= ^f/(p)sinpxdp

2/7i) is

PARTIAL DIFFERENTIATION AND PARTIAL DIFFERENTIAL EQUATIONS

8.

305

If an attempt is
it does not overcome all of the difficulties.
equation
with the upper
in
derivative
as
second
(8,160)
remove
a
made
limit changed, the term (dVjdz) sin pz oscillates boundedly as z increases
indefinitely unless {dVldz)^0 as z-*oo. Hence the Fourier sine transform
finite

transform,
to

has the following limitations.

(i)
(ii)
(iii)

(iv)

must have an open range,

boundary condition must be given at the lower limit,
The dependent variable and its derivative must vanish as z-^-oo.
All derivatives to be transformed must be of even order.

The

variable to be transformed

A type

Similarly, a Fourier cosine transform

can be defined by

00

Kp)

= \mcospxdx

(8.168)

and inverted by

f(x)

=-

f(p) COS

px dp

(8.169)

but the same restrictions as given above for the Fourier sine transform
apply, except that the boundary condition at the lower limit must be type 2.
The more general complex Fourier transform is defined by

Kp)=

j f(x)e'^^dx

(8.170)

= l^JKp)e''^dp

(8.171)

and inverted by

f(x)

This transform is not so severely restricted in appHcation because derivaof any order can be removed by its use, but it does not allow boundary
conditions to be used at all! It is thus useful for general problems, but for

tives

specific practical

8.9.3.

problems

it is

virtually useless.

Hankel and Mellin Transforms

When the next two transforms are used, it is usually to remove the radial
coordinate from a system of cylindrical polar coordinates, and because the
integrals range from zero to infinity, they can only be applied to axisymmetrical disturbances in infinite media.
The Hankel transform

is

defined by
00

Kp) = \Kx)xJ{px)dx

(8.172)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

306

and inverted by the symmetrical formula

00

fM
where J{px)

is

= jJip)pJ(px)dp

the Bessel function of the

The MeUin transform

is

first

(8.173)

kind of order

n.

defined by

Kp) = JKx)x''-Ux

(8.174)

and inverted by
y

+ ioo

m ^j
=

Jip)x-^dp

(8.175)

y ico

which has certain

similarities

with the inversion formula for the Laplace

transform.
8.10.

Conclusions

The methods given

lar solutions

in Section 8.5 are useful for finding isolated particuof partial differential equations, but these solutions are unHkely

boundary conditions.

If the boundary conditions are

then the method of images can be
very useful for generating a more general solution.
The most powerful
methods for finding solutions are undoubtedly the separation of variables
(Section 8.7) and the Laplace transform (Section 8.8). Due to its versatility,
the Laplace transform should be tried whenever a variable has an open
range, and the method of separation of variables should be used in all other
cases, including those in which the Laplace transform has been tried and
found to be inconvenient. There are many partial differential equations
which cannot be solved by either method, and the numerical methods
described in Chapter 1 1 are recommended for these.
The other transform methods of Section 8.9 are speciaHzed in that the
four real Fourier transforms are only capable of removing derivatives of
even order, and cannot solve any problem in which there is either a mixture
of type 1 and type 2 boundary conditions, or a type 3 condition. The complex
Fourier transform can be taken without reference to boundary conditions
and is thus only suitable for problems in which a variable can alter throughout its entire range from oo to + oo without crossing a boundary. The
use of other transforms such as the Hankel or MelHn transform is only
advantageous when cyhndrical polar coordinates would normally be used

to satisfy the desired

of type

1,

or vanishing derivative type

2,

It is not efficient for an engineer to try and learn

the characteristics of all of these transforms when most problems requiring
their use can be solved conveniently by using the Laplace transform. The
reader is therefore recommended to famiharize himself with the Laplace

for stating the problem.

transform, and turn to the other transforms later

if

he so desires.

Chapter 9

FINITE DIFFERENCES

9.1.

Introduction

and chemical reacand these processes are

often analysed by stage to stage calculations or in the simpler examples by
graphical methods. Thus distillation, gas absorption and extraction operations performed in plate columns involve stepwise changes in the concentration of each phase as it passes from plate to plate. In particular, in a distillation column, the vapours leaving any plate differ in composition from the
vapours entering the same plate by a finite amount, thus an abrupt change
in the vapour composition occurs on the plate. Therefore, in contrast with
a packed distillation column where the vapour composition changes continuously throughout the packed height and where the height of the packing
is estimated from the solution of a differential equation, a plate column

Chemical processes involving

separation, purification

tions are frequently carried out in a series of stages,

should be analysed in such a way that these abrupt finite changes enter into
This is possible by finite difference calculus and in particular
by the solution of finite difference equations; as will be shown in this chapter.

the analysis.

9.2.

The Difference Operator, A

Considering an independent variable x which can take a number of discrete values Xq; (xo + h); (xq + IH); etc., then any function y = f(x) will take
a corresponding set of values

From this series of values of the dependent variable

can be constructed as follows.

y^

a series of differences

yi-yoiyz-yiiys-yii-'-yn-yn-i
This series can be written in the form

Ayo;Ayi;Ay2i'--^yn-i
where

Ay

is

Ajo

>' i

- }^o = /(^o + h) -/(xq) =

A/(xo)

(9. 1)

>^ with respect to h.

The symbol "A" is
and has meaning only when applied to a

called the "first difference" of

called the difference operator,

function of a given independent variable with reference to a specific increment h. In fact writing A before a variable or function takes a lot on trust.
A more precise representation would be A^ implying that the differencing
x,h

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

308

operation has been applied once to a function of an independent variable x

through an increment h. However, it is seldom necessary to make these
additions to A as both x and h would be cleariy understood in a particular
problem. Since it is possible to repeat the differencing operation, the number
just above and on the right of A should be inserted for all differences greater
than the first. Thus A^ would be the second difference of a variable or
function, and A" would be the th difference. These will be discussed more
below.

fully

Properties of the Difference Operator

9.2.1.

Let there be another function of the independent variable x that has

+ K); etc., and let this function be expressed as

discrete values at Xq\ {xq

Then

the series of

first

(^(x)

(9.2)

differences of z can be written

or in difference form

Azq; Azi; Az2; AZ3

Az^

where

Adding equations
A>^o

or

+ Azo

more

(9.1)

and

Zi

-Zq =

is,

If

Az_i
(9.3)

(9.3) gives

(ji-yo) + (zi-Zo)

(yi

+ Zi)-{yo + Zo) =

AC^'o

+ Zo)

(9.4)

generally
A(>'

That

(p{xo+h)-(l)(xo)

A: is

+ z) =

+ Az

A>^

(9.5)

obeys the "distributive law of algebra".

a finite constant
A{ky)

= ky^,-ky =

kAy

(9.6)

A can be interchanged without altering the result. That

obeys the "commutative law of algebra" with regard to constants.
Finally, A obeys the "index law of algebra". That is
so that k and

AP+^

A^A^

AA^

is,

(9.7)

This will be further explained below.

9.2.2.

Differences of Second

and Higher Orders

In Section 9.2 a sequence of

was constructed thus

first

differences of the function

(yi-yo)'Ay2-yi)i---'Ayn-yn-i)

The

Ayo',Ayii"-iAyn-i

y = f(x)
(9.8)

differencing process described by equation (9.8) can be repeated to give

Ay,-Ayo = A'yo
But from the relationship of Ay to y

AVo =

A>'i-A3;o

it

(9.9)

can be seen that

(}^2->^i)-(>'i->^o)

Second difference relationships of the type given

>'2-2>'i+>'o

(9.10)

in equation (9.10) are

309

FINITE DIFFERENCES

9.

derivatives in differential equations by finite

method of solution. This will be
numerical
differences prior to using a

employed to replace second

illustrated later in Section 11.4.1 et seq.

The procedure developed

in the

above paragraph to obtain an expression

y can be extended to differences of

for the second difference of a variable

higher order.

For example the

A'>'o

and the fourth

difference

third difference of yQ

is

>^3-3y2 + 3>^i-yo

is

AVo =

>'4-4j3

+ 6>^2-4yi + >^o

In general terms for any variable y in a sequence, the

^""yn^ ym^n-

(9.11)

{Vjym^n-l+

(9.12)

wth

difference

is

(9.

[^

13)

In equation (9.13) the coefficients of the expansion on the right of the

equation are the same as those obtainable from a binomial expansion of

{\-xT.
9.2.3.

Difference Tables

The most convenient way of expressing the first, second and higher order
differences of a function is by means of difference tables. Such a table is
illustrated below for a function y = f{x).
Table
ariable

>'o

xi

yi

xi

yi

Function

xq

9.

The Difference Table

A2

A3

A4

^...Ayo

t^yi

A4;;0
i^yi

XI

>'3

^
X4

"

"

A>'3

yA

In the above table, the second column gives the value of the function at
specified values of the variable x. The third column having the heading A
is constructed from the values of y in the second column by subtracting y^
from 3^1, y^ from y2, and so on to give the column of first differences. The
fourth column of second differences is obtained from the first differences in
the third column by subtracting the entry above from the one below. The
difference table is completed by repeating the procedure outlined above.
This will be illustrated by the following example.
M.M.C.E.

11

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

310

Example. Construct a difference table of >^ = f{x) from the following

data and state what conclusions can be drawn from the result.

1-0

7-567

M
12-159

1-2

he difference table of these data

10

7-567

11

12-159

1-3

17-399

23-339

1-4

1-5

30-031

37-527

1-6

45-879

is

Ai

A2

A3

A4

4-592
0-648

0-052

5-240
1-2

17-399

1-3

23-339

1-4

30-031

1-5

37-527

1-6

45-879

0-700

0052

5-940
0-752

0-052

6-692

0-804
0-052

7-496
0-856
8-352

Inspection of this difference table shows that all the 3rd differences have the
same value which from the elementary properties of a polynomial suggests
that the data represents a function of the third degree.

Difference tables are very useful for assessing scientific data whether it
be interpolation, extrapolation or fitting an equation to a curve of experimetal data. These appUcations will be illustrated in Section 9.4 after more
of the background has been established. However, certain properties of
difference tables will be stated at this point. These are
(i) If the value of a function is known at equally spaced values of the
independent variable of the function, or if the difference between successive
values of the function is known, all the columns of higher differences can be
constructed.
(ii) If one complete column of differences of any order, and one item of
each column on the left of the complete column is known, all the columns
of the lower differences and the values of the function itself can be determined within the range of the independent variable.
(iii) If the members of the column of nih. differences have the same
constant value, the data from which the difference table was constructed is
derived from a polynomial of degree n.
These rules can be tested by considering the example given above. Thus
A^jK = 0-052 and the last member of the A^ column is 0-856. Therefore the
last but one member will be (0-856-0-052) = 0-804. In a similar manner
each item in the A^ column can be inserted and thereafter the A^ column can
be completed if one of its members is given.

9.

9.2.4.

311

FINITE DIFFERENCES

Finite Difference of a Product or Quotient

On

occasion it is necessary to find the difference of a product of two

Thus, if two functions of an independent variable , which can
have integer values only, are written as/() and </>(), then the first difference
of the product of these two functions is
functions.

=f(n + l)(t>(n + l)-f(n)ct>(n)

= /( + !) l(t>(n + 1) - (Pin)-] + ct>{n) [/( + 1) -/()]
= f{n + l)A0(n) + <t>{n)^f(n)

A[/(n)(/>(n)]

(9.14)

or grouping the terms differently,

A[/(n)(/)(n)]

(i>{n

+ \)^f{n)+f{n)^4>{n)

(9.15)

which demonstrates the similarities between finite difference calculus

and

infinitesimal calculus in respect of the differential coeflficient of a product of

two functions.
If the first difference of the quotient of /()

and

</>() is

desired, a similar

procedure gives,

Kn + l) f(n)
L^WJ cPin + 1) <l>(n]
l)j)(n)-(/)(n + l)/(n)
^ /(n +
(/)( + l)0(n)
^ </>() U{n + 1) -/(n)] -/(n)
4)(n + \)(i>(n)
\fin)}

\_ct>{n

+ 1) - 4>{n)\

(l>(nW(n)-f(n)A^(n)
(/>(

+ l)0(n)

which again demonstrates the similarity between finite difference calculus and
infinitesimal calculus. However, it should be observed that the denominator
of equation (9.16) does not contain a squared term and in this respect differs
shghtly from the differential coefficient of a quotient.

9.3.

The

Other Difference Operators

above sections are, to be precise, called

"forward differences" and are the ones most generally used for the solution
of engineering problems. Consequently they are simply referred to as the
first, second or th differences and the prefix "forward" is omitted when it is
clearly understood which type of difference is meant. However, in certain
interpolation formulae the terms "backward difference" and "central
difference" are employed and therefore their meanings must be understood.
In actual fact there is no basic difference between a forward difference and a
backward difference since both express the difference between the higher and
next lower value of the function, and the significance between these terms is
differences presented in the

11-2

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

312

best understood by reference to Table 9.1.

In that table

it

will

be seen that

column refers to yg and the various differences are

obtained from successive members of the series starting from yQ. These are
the forward differences and they are shown by the sohd arrow. On the other
hand, the last entry in each column is the first, second and third forward
difference of each successively lower member of the series as shown by the
dotted arrow. These entries in the table can be relabelled as backward
the

first

entry in each

differences by using the following definition.

^yn

The second backward

difference

V^y,
It is

in

now clear that

Table

all

= yn-yn-i =
is

^yn-i

(9.17)

written similarly thus,

= yn-2y.,+y.2 =

^^yn-i

(9.18)

of the backward differences shown by the dotted arrow

9.1 refer to y4..

third type of difference

known

as the "central difference"

is

symbol-

ized by prefixing d to the function.

Thus
<5y

(9.19)

y+i->'n-i

Central differences would be represented by a horizontal line in Table 9.1.

To sum up, forward differences should be used when data from the
initial parts of a table are to be assessed and backward differences used when

data from the end of a table are to be considered. This is particularly true
and extrapolation purposes. Finally, for interpolation near
the centre of a collection of data central differences should be employed.
There are two further important difference operators which will be
described in the following sub-sections.
for interpolation

The operator

9.3.1.

The operation of changing the value of a function to correspond with a

change of one increment in the independent variable is denoted by E.
That

is

In a similar

EyQ

y^

E(Ey,.2)

E'y.2

(9.20)

manner
y

Ey_,

= >o

(9.21)

Equation (9.21) shows that E" operates on yQ n times to increase the value of
the function corresponding to an increase of /7 increments in the independent
variable.

The exponent n
Thus

in

E" can take any positive or negative integer

E-'KxQ)=KxQ-h)
where the exponent (-1)

signifies that the

function

lower value of the independent variable.

The operator E has significance only when

value.

(9.22)
is

reduced to the next

it is placed before a function

or variable, and in this respect it is similar to other operators. Also, E can
be treated hke an algebraic symbol in that it obeys the distributive, commu-

9.

tative,

313

FINITE DIFFERENCES

The reader

and index laws of algebra.

is

asked to verify these for

himself.

Since E and A are both finite difference operators that obey the same laws
of algebra, it would be expected that they must be inter-related. The relationship between them can be shown as follows.

yn^,=Ey,
yn^,-yn

(9.23)
(9.24)

^yn

Substituting equation (9.23) into (9.24) gives

{E-\)y

^y

(9.25)

which can be written as an identity between operators thus,

= 1+A

(9.26)

This simple relationship is most important because it often allows algebraic

expressions in terms of the one operator to be simplified in terms of the
other by using ordinary algebraic manipulations. Thus, if x is the independent variable that is capable of taking the values Xq {xq + h) {xq + 2h)
;

'^''*'''

/(xo

By

+ ft) =

(9.27)

/(xo)

Taylor's theorem (see Section 3.3.7)

EKx,)
where

= /(xo + h)= /(xo) + hDfixo) +

is

^' D'fixo)

the differential operator d/dx.

^ D'f(xo) +

(9.28)

From equation (9.28) it can be seen

that
2

/i,n\3

/(xo)

The terms

in the square bracket

/(xo)

(l

+ A)/(xo)

(9.29)

form the expansion of the exponential

function e^^, and therefore equation (9.29), provides a relationship between

the differential operator and the finite difference operators. That is

E=\ + ^ =
The

(9.30)

e''^

three symbols in equation (9.30) are all operators and they can be treated
In the mathematical relations developed to show

like algebraic quantities.

the relationship between them, the analysis has been confined to the first
order of the operator. This was done for convenience only, and for any
value of
.
^
,,
,^^
^ ^..

E"'

{l

+ ^Y =

e'"^^

(9.31)

In addition, any polynomial of these symbols having constant coefficients

represents an operation on a function. Thus

7(^o)

= (l+A)7(xo)
= /(xo) +
A/(xo) +
A^/(^o) +
(7)
(2)

[i

^'Kxo) +
(3)

+ (7)a + (-)a^ + (-)a3+...]/(xo)

...

(9.32)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

314

Equation (9.32) is known as "Newton's forward difference formula". The

polynomial within the square bracket is a polynomial operator and m is
This polynomial operator only has
any number, integer or otherwise.
significance when it operates on a function such as/(x) in equation (9.32).
This result finds extensive use in interpolation (Section 9.4).
In a similar manner it is possible to produce an expansion of "" in terms
of the operator V. Thus by equation (9.26)

=G-^)'=
and if the last expression
on /(xo) the result is
7(^o)

in

(l-Afi-')-"

equation (9.33)

is

(9.33)

expanded and made to operate

= (l-A-')"7(xo)

+ ...{^^''^~^'^K'f{xo-rh) + ...
Using equations
-/(>.)

(9.17, 18) to introduce

backward

(9.34)

differences,

("^') v/w+^^^ ("*;"')s'-/(>.)+

-/(.)+ (7) v/(o)
+

.(7)v.(-')v=.(-^)v ^+
Equation (9.35)

is

known

as

/(^o)

...

(9.35)

"Newton's backward difference formula" and

Again m can be any number,

equation (9.32) is used for interpolation.

integer or otherwise.
like

The Difference Quotient

9.3.2.

The difference quotient was introduced by Norlund (1920) in order to

show the similarity between finite difference calculus and infinitesimal
calculus. The difference quotient is defined by the equation

^ /(xo + ft)-/(xo)
On

occasion

it is

(9 3^^

referred to as "Norlund's Operator".

Similarly

A would

A/(x

=A

be the second difference quotient and

p+
o

^)-/fa) l

is

written

_ fixo + 2h)-2f(xo + h)+f(xo)

^^^^^

These concepts can be extended to the mth difference quotient, thus

/(^o)

\f(xo

+ mh) + (7)/(^o + '^^^h) +

(2

)/(-^o

+ '^^^^h) +... +/(xo)]

(9.38)

9.

315

FINITE DIFFERENCES

Equation (9.38) shows the relationship between the difference operator and
which may be summarized as

difference quotient

AJ(xo)

h^'Afixo)

(9.39)

Thereafter the relationship of the difference quotient with the other finite
difference operators follows from the above sections which may be summarized thus
A"'

= (E-IT =

(e""^

- IT =

h'"A

(9.40)

Finally, it can be shown that the difference quotient obeys the distributive
law for addition, the index law for positive integers and the commutative
law for constants. It is left for the reader to verify this statement.

9.4.

Interpolation

It frequently happens that the engineer is confronted with a collection of

data relating two variables, such as the pressure inside a vessel corresponding
The pressure is thus some unknown
to the temperature of its contents.
function of the temperature, and he is required to predict the pressure at
some temperature in between or outside the limits of the data he possesses.
If the pressure is required within the limits of the data, the mathematical
procedure is called "interpolation"; whereas if the point lies outside the range
of the data, the mathematical process is called "extrapolation".
Interpolation or extrapolation depends on the assumption that the
functional relationship between the two variables is continuous over the
range of the independent variable being considered. Therefore in order to
be able to interpolate or extrapolate, the engineer must obtain this functional
relationship or some other functional relationship that approximates to it.
For this purpose the trends in the data must be considered from the point of
view of the experimental results and the mathematical analysis. For example,
the chemical engineer would be famihar with the general shape of a vapour
pressure-temperature curve and its position on a graph from the experimental data. The mathematician is able to assign an approximate equation
to this curve and thereafter interpolation is straightforward.
Usually the experimental data is approximated by a polynomial and the
degree of the polynomial can often be estimated by preparing a difference
table from the collected data. The difference column that gives an approximately constant value, gives the degree of the polynomial which can be
fitted to the data by selection of the constants.

9.4.1.

Newton's Formulae

When

the data points are available at equal intervals of the independent

Newton's difference formulae can be employed and the procedure
best illustrated by the following examples.

variable,
is

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

316

Example

Estimate the value o^ Jq{z) at z

Jq{z) given in the following table.
1.

Jo(z)

The following

difference table

is

4-0

3-8

-0-4026

-0-3918

-0-3643

-0-3202

3-479 from the values of

3-6

3-4

3-2

-0-3971

constructed from the data supplied.

^Mz)

.ov^y

A3/o(r)

^'-Mz)

2
3

From the above table it can be seen that Jo{z) can be approximated by a
polynomial of second degree within the range of data supplied.
(a) Using forward differences

and

Then by Newton's forward

+ mh) =

ZQ

Zq

3-2

m=

.-.

Jo(zo

+ mh
and h

0-2

1-395

difference equation (9.32)

Joizo)

[^j

AJo(zo)

+ (^jA^Joizoj

Substituting from the difference table gives

Jq(z)

-0-3202 + (l-395x -00441)

1-395 X 0-395

x 0-0166

= -0-3771
(b)

For

Using backward

differences.

this interpolation, Zq will

be selected at 3-8 because AVo(z)

will give

a value derived from Jq(3-4). Consequently the interpolation formula will

include differences obtained from the neighbourhood of /o(3*479) as shown

by the

Thus
Then

lines in the difference table.

Zo

3-8

3-479

3-8

let

.-.

Inserting these figures into

m=

and

+m

0-2

x 0-2

-1-605

Newton's backward difference formula

(9.35)

9.

+ m/i) =

o(zo

317

FINITE DIFFERENCES

+ (7) V^(^o) + ("^J^) VVo(zo;

Jo(^o)

1-605

-0-4026

x-O-OlOS]
1

+
=

1-605 X -0-605 X 0-0167'

[-

-0-3772

/o(3 '479) predicted by the forward diflference formula agrees

very well with the result obtained using the backward difference formula.
The result estimated from Chistovaf for /o(3-479) is -0-3772 which agrees

The value of

with both

results.

Either of the difference formulae can be used for interpolation, but for

extrapolation the forward difference formula must be used.

Example

2.

Using the data given in the table below, estimate the vapour
ammonia vapour at 167F. Latent heat of ammonia

pressure and density of

544 Btu/lb.

Temperature

Pressure Ib/in^

From

70

80

90

128-8

153-0

180-6

100
211-9

110
247-0

the above data the following difference table

^p

70

128-8

80

1530

is

120
286-4

130
330-3

140
379-1

prepared
bJ-p

ts}p

lA'l
3-4

27-6

90

180-6

100

211-9

110

247-0

120

286-4

130

330-3

140

379-1

0-3

3-7

31-3

01
3-8

351

0-5
4-3

39-4

0-2
4-5

43-9

0-4
4.9

48-8

The fourth

differences t^^p have not

been calculated because A^;? are very

contribution to the estimated pressure. Inspection of the difference table shows that the last entry in the A^/? column
(i.e. 0-4) is derived from p = 247-0 Ib/in^ corresponding to 110F.
Theresmall and will

make

very

little

fore 110F will be taken as the base temperature.

p =f(t) =f(to + mh)

Then
t Chistova

E.

A. " Tables of Bessel Functions." Pergamon Press, London (1959).

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

318

where

167,

167

.-.

^o

110 +

10m

Newton's forward difference formula

P

10
5-7

(9.32) gives

(T)a.o.0AVo + 0A'Po

Po

^.^r.
247-0

5-7

.^,

X 39-4

- x ^^
x ,, + 5-7x4-7x3-7
1x2x3

5-7x4-7

247-0 + 224-58

0-4

4-5

1x2

=
=

=
orm =

and h

110

+ 60-28 + 6-61

538-5 Ib/in^

The value extrapolated from Perry f

538.0 Ib/in^.

is

In order to find the density of the

ammonia vapour

it is

convenient to

use the Clausius Clapeyron equation J

dp

SH

dT

nv,-vd

where Tis the temperature in degrees Rankine. Assuming that the volume of
the liquid is negUgible compared with the volume of the vapour gives
p
^

=
1

V,

T dp
SH dT

III

dp/dT can be evaluated from the difference table by taking the logarithm of
equation (9.30). Thus
/i)

Interpreting

as

ln(l

+ A) = A-iA2+iA^-...

d/dT and substituting equation IV into

IV
III gives

= ;J^(Ap-iA^P+iA'p-...)

The values of the differences in the bracket of equation V must be determined at the required temperature of 167F. This can be done by extrapolating each column of differences in the table using the same method as
the extrapolation for/? at 167F.
at

167F,

p =

Substituting into equation

The

538-5, A/?

results are

70-4,

10 X 544
.-.

This value
t Perry,

is

J.

more
H.

reliable

(Ed.).

6-8,

A^ =

6-8

0-4\

l.r.
___(70.4-+ -)

0-0099

Ib/ft^

than that obtained from the ideal gas laws.

" Chemical Engineers Handbook," 3rd edition.

(1950).
X

Bransom,

S.

0-4

gives

627
'

A^ =

H. " Applied Thermodynamics." Van Nostrand (1960).

McGraw

Hill

9.

9.4.2.

319

FINITE DIFFERENCES

BesseVs Interpolation Formula

This formula

is

for use with central differences along a horizontal line in

can be seen in Table 9.1 that values are only available

are filled by taking an arithmetical average
of the two values adjacent to the one required in the same column and
denoting the value by fxd, jxd^, .... etc., depending upon the order of the difference. It is customary to choose a fractional value of m which is less than
unity and use the formula

the difference table.

It

in alternate columns.

The gaps

EJ(xo)=f(xo) + mdf(xo + ih)

(m + l)m(m-l)(m-2)

^^^^

[fiS'f(xo

m-i

,,^
+ "^-^S'Kxo + ih)^
+ ih)
,

4!

+
9.4.3.

(9.41)

...etc.

Lagrange's Interpolation Formula

Quite often, the data are available at unequal intervals of the independent
variable and this precludes the use of a normal difference table.f Lagrange's
method fits a polynomial to the data but otherwise has no connection with
finite difference calculus: it is included in this chapter merely for completeness within this section concerning interpolation.
The fourth degree polynomial which passes through the four points
(xq, yo), (^1, yi), (X2, y2)^ (x^, y^) can be written

(x-Xi)(x-X2)(x-X3)
^(^o-^i) (^0-^2) (^0-^3)

(x-Xo)(x-X2)(x-X^)
_^

\xi-Xo)(Xi-X2)(Xi-X^)
(x-Xo)(x-Xi)(x-X3)

^(X2-Xo)(X2-Xi)(X2-X3)

(^3-^0) (^3-^1) (^3 -^2)

and

this

form

is

known

as "Lagrange's interpolation formula".

more than four points

The

exten-

obvious the numerator and denominator of each fraction is extended to include the further values of x and the
complete equation is extended by adding further terms to include all values
of >^. It is to be noted that the term corresponding to the appropriate y is
sion to

t It
is

is

is

fairly

possible to calculate in terms of " divided differences " where the first difference
ratio of By to bx, but the method used herein is considered to be more

found from the

satisfactory.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

320

omitted from each numerator and its counterpart from the denominator.
The use of many more than four points is not recommended due to the

computational

effort.

Example 3. Find the thermal conductivity of propane at 100 atm. and

210F from the following data given by Leng and Comings.
rCF)

P(atm)

^(Btu/h ft2.F

96-7

0-0490

131-5

00518

88-9

0-0440
0-0466

154

189

131-8

140-9

0-0402
0-0435

95-3

0-0353

123-0

00376

96-7

222

284

ft-i)

From the graph given in the paper, other results show that in this region the
conductivity varies approximately linearly with pressure so that a linear
interpolation is adequate in this direction.

A set of values for conductivity and temperature at a pressure of 100 atm.

can be found from the above values as follows. At 154F and 100 atm.,
k

100-0 -96-7

0-0490

0-0493

Hence the following

(0-0518-0-0490)

table can be compiled.

154

189

222

284

0493

0-0447

0-0404

0-0357

Substituting these values into equation (9.42) gives the thermal conductivity
of propane at 100 atm. and 210F. Thus

(210-189)(210-222)(210-284)
0-0493

(154-189)(154-222)(154-284)

(210-154)(210-222)(210-284)
0-0447
(189 -154) (189 -222) (189 -284)

t Leng, D. E. and Comings, E.

W.

I.E.C. 49

(ii),

2042, 1957.

9.

/c

(210-154)(210-189)(210-284)

0-0404

0-0357

(222 - 154) (222 - 189) (222 - 284)

(210-154)(210-189)(210-222)
(284 -154) (284 -189) (284 -222)

= 0-0419 Btu/hft^F ft"

difference equation

finite

Finite Difference Equations

9.5.

321

FINITE DIFFERENCES

is

a relationship between an independent

and sucThus an equation of the type

variable that can take discrete values only, a dependent variable,

cessive differences of the dependent variable.

+ l>^l + 2>'2 + ---+n>'n = </>W

equation if y = f{x) and x can take the

'30>'0

(9-43)

values Xq, Xq + H,
a finite difference
XQ + nh only. In equation (9.43), Oq, a^, aj, .... a can be constants or
functions of the independent variable, (pix) could be a constant or a function
of X. Equation (9.43) can therefore be written
is

....

+ aJ{xo + nh) = 0(x)

(aE" + a.,E"-' + ... + ao)f(xo) = (P(x)
/(xq + rh) = EJ(xq). But = 1 + A,

o/(^'o)

or

because
.-.

+ i/(^o + /i) +

(pA" + ;7_iA"-^

+ ...+;.o)/K) =

0W

where the coefficients p^ are obtained from a^. Equations

different forms of the same finite difference equation.

(9.44)

(9.45)

(9.43) to (9.45) are

Finite difference equations are similar to differential equations in that they

can be classified by "order" and "degree". Thus the order of a finite difference equation is determined by the order of the highest difference. For
example, the equation
(9.46)
yn^i-ciyn = ^
is

of order

because

it

involves only the

first

difference,

(-a)j = [A-(a-l)]>'

On

the other

(9.47)

hand the equation

yn-ciy,-,

is

i.e.

+ by,.2 =

of order 2 because the highest difference

[A2

is

of order

(9.48)
2,

thus

+ (2-)A + (l-a + 6)]y_2 =

(9.49)

should be noted from equations (9.47) and (9.49) that it is the difference
between the highest and lowest members of the dependent variable that

It

determines the order of the equation, and not the highest

indicated by the suffix to the dependent variable.

member

as

MATHEMATICAL METHODS

322

CHEMICAL ENGINEERING

IN

The degree of a finite difiference equation is the highest degree of the

dependent variable, or any of its differences in the equation. Equations
containing y, Ay and A"y are of the first degree and are said to be Hnear,
whereas equations containing products of the type >'>'_ i or their differences
are of degree 2. Linear difference equations are by far the most common,
and they will be treated initially.
Linear Finite Difference Equations

9.6.

Equation

(9.43)

coefficients Qq, a^,

is
....

a linear
a

may

finite difference equation of order n and the

be functions of the independent variable or

constants.

The solution of this type of equation is similar to a Hnear differential

equation in that the complete solution is made up of a complementary
solution and a particular solution. Proof of this can be obtained in a manner
very similar to that

shown

and
However, each part of the complete solution

in Section 2.5.1 for differential equations,

therefore will not be repeated.

will
9.6.

be discussed.

L The Complementary

Solution

For ease of manipulation, consider the second order

linear finite difference

equation with constant coefficients.

yn^2-Ay^, + By =
In the same

way

(t>(n)

as for differential equations, the

the solution of equation (9.50)

when

(j)(n) is

zero,

is

i.e.

yn^2-Ay^,+By =
But equation

(9.50)

complementary solution

(9.51)

(9.51) can be written in the form

(E^-AE + B)y =

(9.52)

Since the polynomial within the brackets can be treated as an algebraic

quantity,

it

can be factorized.
.-.

{E-p,)(E-p,)y=^0

(9.53)

The two factors could be in either order since the operator E will commute
with constants such as pj and P2', hence there are two independent solutions
of equation (9.51) satisfying

or
It

(0

{E-p,)y

(ii)

(JE:-P2)>'

=
=

can be seen by substitution that the solution of

and the solution of

Hence

(i) is

yn

A'pl

(9.54)

>

B'pl

(9.55)

(ii) is

the general solution of equation (9.51)

y

A'p\

is

+ B'pl

given by
(9.56)

9.

323

FINITE DIFFERENCES

where A' and B' are two arbitrary constants that would be evaluated by the
boundary conditions of the problem.
It will be noticed that equation (9.56) is the complementary solution of
the original second order equation and that it contains two arbitrary constants. In a similar manner it can be shown that an nth order equation will
provide a complementary solution containing n arbitrary constants; which
again shows the similarity to the complementary function of a Unear differential

equation.

Equation (9.52)

is

known

difference equation (9.50).

plex.

as the "characteristic equation" of the finite

roots

Its

may be

different or equal, real or

com-

pi and p2 are real,

the final complementary solution. If the roots are complex,

The case of

different roots

is

given above, and

if

equation (9.56) is
they will occur in conjugate pairs thus,
p^

= a+

ij?

p2

re'"^;

(x-ip

re'"'^

(9.57)

In this case, the solution (9.56) can be written in the form

y^

r\A' cos

n<t>

by using De Moivre's theorem (Section

solution

when

B' sin n0)

4.7).

This

(9.58)
is

the complementary

the roots of the characteristic equation are complex

and

is

similar to the solution of the corresponding differential equation.

Finally,

when

the roots of the characteristic equation are equal, the

complementary solution takes the form

y

{A'

+ B'n)p"

(9.59)

again Hke the comparable form of the differential equation (Section 2.5.2b).
9.6.2.

The Particular Solution

This can invariably be found either by the method of undetermined coor the method of inverse operators both of which are very similar
to the corresponding method for Hnear differential equations. Since these
methods were described in Section 2.5, they will only be repeated briefly here
with reference to the second order finite difference equation
efficients

{E^-AE + B)y =

A particular

solution of this equation

(9.50)

<t>{n)

is

by the method of inverse operators. The operator can be factorized as

the above sub-section, and separated into partial fractions. Thus

E'-AE + B ~ {E-p,){E-P2) ~ {E-pd ~

where a

l/ipi-Pz)-

Each

partial fraction

(E-p^)

on the right-hand

in

^^'^^^

side of

MATHEMATICAL METHODS

324

IN

CHEMICAL ENGINEERING

equation (9.61) can be expanded as follows

-1
1 -f

-Pi

i-p,/

(1-Pi)
a

+ --,-...

(i-Pi)

1-Pi

and

E-P2

(1-P2)L

(9.62)

(1-Pi)

A^
A
1-^
+ 7^V2--I
I-P2 (I-P2)

(9-63)

of the operators are most convenient for finding the

a polynomial in , because only a finite number
of terms will be needed in the expansions.
When (j){n) = ka", an alternative procedure is advantageous. Thus,

The forms

(9.62, 63)

particular solution if

(/)(a?) is

and

Ea"

E^'a"

a"^^

a.a"

a""^"

a'".a"

In general,

f(E)a"=f(a)a"

(9.64)

provided /() can be expanded as a polynomial in E. Equation (9.64) is the

key to a particular solution when (/)() = ka". The usual difficulties can arise
when/(fl) is infinite, but these can be overcome by techniques similar to those
given in Section 2.5.5.
It should be noted, however, that the equation
analogous to equation (2.79) takes the form
f(E)(a"y)

a"f(aE)y

(9.65)

where a and E appear as a product instead of a sum on the right-hand side.

Equation (9.65) is of Httle practical value since it does not lead to any
simplification in the form of the operator.
Hence, rewriting equation (9.50) with (/>() = ka",

(E^-AE + B)y =
and using equation

a'^

(9.66)

(9.64), the particular solution is

yn

provided

ka"

Aa + B ^

0.

2
.+ Bo
a^-Aa

(9-67)

This exceptional case can be treated in a similar

manner to the corresponding differential equation by using equation (9.65)

and introducing the difference operator A using equation (9.26).
In concluding this sub-section on the determination of particular solutions of finite difference equations,

should be pointed out that a particular

when </)(/?) takes the form of a polynomial or power by the method of undetermined coefficients. Here there is
no difference in technique between finite difference equations and differential
equations; the trial solutions take exactly the same form in the corresponding
it

solution can be obtained in most cases

cases.

9.

FINITE DIFFERENCES

325

Finally the complete solution is the sum of the complementary solution

and a particular solution in a manner similar to differential equations. If
possible the operational method should be used to evaluate both parts of
the solution, but on occasion the method of undetermined coefficients may
be more suitable.

Example 1. G^+i lb moles/h of a wet gas containing T^ + i lb moles/mole

of solute are fed into the base of a plate absorption column where the solute
is to be stripped from the gas by absorption in Lq lb moles/h of lean oil
which is fed into the top of the column. If the solute in the entering oil is
Xq lb moles/mole of lean oil and the solute in the exit gas is Y^ lb moles
of solute per mole of wet gas, show that the performance of the absorber
can be expressed in terms of the absorption factor A = (LJKGi^ + i) ^^^ the
number of ideal stages by the Kremser-Brown equation

Y
y
where K

is

>4^+l_ A

Y~

A^'^'''

the equiUbrium constant.

That

is

= K^X^

ym

where y and x^ are the mole fractions of the solute

in the gas

and

liquid

phases respectively.
Y,

Gi

-o/o

V
m-1
(Y

X_',

I'Cu-l

^m

A^
N

1
^N^1 Yn +

Ln X^

Fig. 9.1

Considering Fig.

9.1

and taking a material balance over

plate

m gives

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

326

But

yrrr

G^=YGs +

and x L

= X^ Lq

= -^"l {^^
''-m{hrya:

or

K.

Substituting equation III into equation I to eliminate

y _

II

Ill

X^ and X^.^

gives

^m+l+(^ff,-i/-Km-iG^-i)y^_^
l

+ (LJiCGJ

But (L_i/Ar_iG;_i) = ^;-i, the absorption factor for plate m\\ and
^, the absorption factor for plate m. Therefore equation IV
can be written as

(LJKJj^ =

Taking a mean value of the absorption factor as suggested by Edmistert or

Horton and Franklin| equation V becomes

y+i-(i+^)y^+^y^_i = o

vi

is the mean value of the absorption factor between the bottom and
top of the column.
Putting equation VI into operational form and factorizing gives

where

[_E^-{l

+ A)E + A-\Y^ = {E-\){E-A)Y^ =

VII

which has the solution

y^

Ci

+ C2.4'"

VIII

where C^ and C2 are arbitrary constants.

Now
since

Yq

is

IX

assumed constant; and equation VIII gives

+ C2

+ C2^^"'^
y^+i-yo = C2(A^^^-l)

XI

Yo

y^+j

Similarly,
.-.

and

= KXo

=
=

C,
Ci

Y^+^-Y^

also

=^

C2(A^''^-A)

XII
XIII

Dividing equation XIII by equation XII gives

y
Y
which

is

the well

known Kremser

AN+i_A
AN+l_i

equation.

W. C. I.E.C. 35, 837 (1943).

G. and Franklin, W. B. I.E.C. 32, 1384 (1940).

t Edmister,
t Horton,

_y
-Y

XIV

9.

327

FINITE DIFFERENCES

Simultaneous Linear Difference Equations

9.6.3.

The

analysis presented in the sections above has been confined to one

dependent variable. However, when two or more dependent variables are
present in a system of difference equations it is usually possible to ehminate
all of these variables except one by the normal rules of algebra. Thus consider the following examples.

Example

2. In a proposed chemical process 10 000 Ib/h of a pure liquid

be fed continuously to the first of a battery of two equal sized stirred
tank reactors operating in series. If both vessels are to be maintained at the
same constant temperature so that the chemical reaction

A is to

kz

ki

A->B-C
of the vessels that will give the

will take place, estimate the size

yield of the product B.

and k2

0-05

min"^

The

specific reaction rate constant k^

at the temperature of the reactors,

and the

maximum
0-1

min~^

fluid density

constant at 60 Ib/ft^.
Let the concentration of
leaving any vessel n be C^ , the concentration of B be Cfi , and of C be Cc^n- Then a material balance over stage n
is

gives for:

component A,
where 6

is

the nominal holding time in the reactor;

component

B,

^B,n-l~^B,nk2CB^rfi k^C^nd
Solution of equation II is obtained in the usual way. Thus
k2e = ^. Then
/.
x^
^
(l + a)C^.-C^.-i =

+ a)-l]Q, =
C^,n = K,p'[
constant and pi =

III
let

where K^ is an arbitrary
Cyin from equation VI into equation
[(l

The complementary

The

is

VI
1/(1+ a).

and rearranging

Substituting for
gives

+ i5)-l]Q_i=aXip"i

solution of equation

Cs,n

where K2

III

the arbitrary constant

VII

VU is

K2P"2

and P2

a and

IV

[(l

or

k^O

VIII

1/(1+ fi).

particular solution of equation VII can be written

oiKipl

"(l + Pi-l

-T^-

IX

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

328

by using equation

(9.64).

The complete
CB,n

Since the

initial

feed to the reactors

K2 =

pure A, at

is

0,

C^ = C^

X gives
XIII

-7r^(p"l-p"2)

must be a maximum for = 2 and

maximum by varying n at fixed 9,

to find the

than by varying 6 at fixed

Hence, by differentiating equation XIII,

n.

dCB

ccC.Q

dn

jS-a

C^

Pi

Inpi

Pi

lnp2

Pi
1

+a

+P

p2

XV

+ 0-1^

+ 0-05^

and

,llO:05gyjn(l+0.05^)

.hen

XIV

Since

(Pl\np,-pl\np2)

^nd

XII

The condition required is

some value of 6. It is easier

that

XI

^C^,o

Substituting these values into equation

CB,n

thus

= C^

Kj^

and

is

K2p"2+^p\

.'.

solution of equation VII

\1 + 0'W)

In (1

^^^

+ 0-10)

Following the methods of Section 11.5 to solve

this

by successive approxima-

tion gives the result

.'.

XVII

0-456

Reactor volume

60x60y

Volumetric feed rate

10,000

Optimum volume of each

reactor

^ ,^,

0-456
is

l-27ft^.

Example 3. 8540 Ib/h of an animal fat are to be hydrolyzed and extracted

column using 3760 Ib/h of water.f If the column is to operate

in a spray

under counter current flow conditions, the percentage hydrolyzable glycerine

is 8-53% by weight, and the glycerine in the fatty acid leaving the
tower is 0-24% by weight, calculate the number of theoretical stages in the
column.
in the fat

t Jefferys, G. V., Jenson, V.

G. and Miles,

F. R.

Trans. Instn.

Chem. Eng.

39, 389 (1961).

9.

FINITE DIFFERENCES

329

A mass balance will give a glycerine concentration in the Sweetwater of

18-8%, and the total weight of fat phase held up in the column is 12 200 lb.
The distribution ratio of glycerine between water and fat is 10-32, and the
reaction rate constant is 10-2 h~^.
Solution.

plate

column

In order to determine N, the number of theoretical stages, the

illustrated in Fig. 9.2 is considered and the following symbols

are used.

L
G

Ib/h fat phase rising through the column.

lb fat phase held up per stage.

weight fraction of glycerine in raflfinate.
weight fraction of glycerine in extract.
weight fraction of unreacted fat in rafRnate.
lb fat required to produce 1 -0 lb of glycerine.
pseudo first order reaction rate constant expressed in terms of fat
concentration (h~^).

X
y
z

w
k

Ib/h water phase descending through the column.

Cj.

^
Ar

n-1

1
FK5. 9.2

glycerine balance over stage n gives

Gj_i

and an equivalent
column gives

+ Lx + +
i

kH

Gy^ + Lx^

glycerine balance between plate n

-Z;v +

+ Gy_i =

G};^

+ Lfx +

and the base of the

II

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

330

The equilibrium

relation

is

yn

where

is

= ^^n

Substitution of {zjw) from equation

of X by the use of equation III gives

mG^

mG/L =

Equation

a,

kH/L =

a linear

is

into equation

mG

kH fm

Letting

III

the equilibrium ratio.

subsidiary equation

[E^-(a + P+l)E + oc(l + p)']y =

and the complementary solution
y

The

particular solution

gives

The

equation.

finite difference

and elimination

mG

and rearranging IV

p,

II,

is

VI

is

A(x"

+ B(l + py

VII

is

(E-cc){E-l-P)

C=

where

- ^z^+i

j?^a>;^

Because

is

independent of n and

= + A,
1

"

can be replaced by unity.

Hence equation VIII becomes

-C

a-1
The complete

solution of equation

V is

then

with the boundary conditions that at n = 0, ^ = 0; and at n = A'+l,

X = 0. Substituting these boundary conditions into equation
gives

and

The

a-1

;[''+ '-(l

+ ^Sf+'J

B = - /?0VZ!!1^>) [..r'^'"\..1
a-1
J[a''*'-(,l+pf*'\
V
final solution after simplifying is

mzjv^i

^"-K^TTDr

(a-l)a^-)?(l+^f ]
(l-a+^)(l + ^rJ

9.

Substituting the data supplied into equation

10-32x3760

_kH _

10-2x12200

where 12 200/

gives
^

^,
4'54

_
"

14-6

8540

^~~L~
A^^

XI

mG =

331

FINITE DIFFERENCES

8540

H, the hold-up per

xN

stage.

Then from equation XI

/10-32xO-0853\

^.oo

4-54^

3-54 X 4-54^ - (14-6/Ar) (1

+ 14-61 Nf

XII

(^'--)(-W
gives

A^

9.7.

2-8 (by trial

and

error)

Non-Linear Finite Difference Equations

These types of equation occasionally arise in the solution of engineering

problems and they are difficult and sometimes impossible to solve. However, first order non-Unear equations can be solved graphically, and some
second order equations can be solved by special substitutions. These methods
will be discussed separately below.
9.7.1.

Graphical Solution

Consider the

first

order non-linear difference equation

y^,-yi + Ay+B =
where

and

B are

constants,

ing to their subscripts.

and rearrange

it

(9.68)

to separate the terms accord-

Thus

y^,==y'-Ay-B

(9.69)

By

selecting arbitrarily a suitable set of values for y, a corresponding set of

values of >^+i can be calculated from equation (9.69). Hence a table of

values of ^'^ and y+i can be compiled and the results plotted on rectangular
coordinates in the form yvsy+i.
The curve AB in Fig. 9.3 illustrates
equation (9.69). In order to find the solution, the diagonal jv = y+i shown
by the line PQ is constructed. Starting at the boundary condition yQ at

(of coordinates

>'o,

yi) and drawing the ordinate, the point

the diagonal where yo = >'i. From C, a horizontal line

at the point D. The coordinates of
meet the curve

AB

is

is

located

on

drawn back to
are (y^, y2) by

equation (9.66) and hence y2 is evaluated. By continuing this stepwise procedure to A^ steps, the value of y^+i can be obtained. Thus the value of >^
corresponding to any value of n can be obtained.
This method will work for all first order equations provided that they
can be separated into a simple form such as equation (9.69). It is usually
easier to choose values for the variable on the more complicated side of the

MATHEMATICAL METHODS

332

IN

CHEMICAL ENGINEERING

h +

Graphical solution of a

Fig. 9.3.

finite difference

equation

equation and solve for the other variable. Thus it is simpler to choose a set
of values for y and solve equation (9.69) for y+i, than assume values for
y + and solve the quadratic for y.
i^

Example 1. 1000 Ib/h of ethyl alcohol is to be esterified by reacting with

850 Ib/h of acetic acid in a battery of continuous stirred tank reactors, each
of 30 ft^ capacity and maintained at 100C. If the equilibrium relation is
such that 75-2% of the acid will be esterified, estimate the number of reactors
required for 60
conversion.
At 100C the specific reaction rate for esterification is 4-76x10"'^ 1/g
mole min and that for the hydrolysis of the ester is 1 -63 x 10""* 1/g mole min.
The density of the reaction mixture can be assumed constant at 54 Ib/ft^.
Solution.
In a battery of A^ stirred tank reactors, consider a material
balance on reactor m. That is,

is the concentration of reactant A leaving vessel /w, r is the rate

of reaction, q is the volumetric flow rate, and V is the volume of one reactor.
For a second order reaction

where C^,

A + B^C + D
the rate of chemical reaction
r

and

if

is

'A,

^A,m

{^l^A^B^^l^C^Wnfi

IV

the nominal holding time in the vessel.

concentration of reactant B exceeds that of reactant A by an
c initially, this difference will be maintained throughout the system.
is

If the

amount

III

in the effluent

the vessel.

where

is

= ki^CACB-k2CcCD

assumed to be perfect, the concentration of the comfrom the vessel will be the same as the contents of
Therefore equation I becomes

the agitation

ponents

II

9.

That

is,

in

any reactor

HNITE DIFFERENCES

333

the concentration of reactant

is

(C^^ + c).

By

the stoichiometry of the reaction the concentration of each product

Substituting these concentrations of components B, C, and
equation IV and rearranging gives

is

into

V is a first order non-linear finite difference equation which will be

solved graphically for the total number of reactors required to convert
60% of the acetic acid. Thus assuming the feed is uniformly mixed, the
Equation

initial

concentration of acetic acid

850x54
^-'^

^^'

C^^o
c

.-.

^^ ^ lb

= "185^ =

Similarly,

is

_ 24-8x454 ^ ^e moles
p^ ^^^^^^ ^ 6-63

10*18 g moles/1.

10-18-6-63

3-55 g moles/1.

_^

30-0x54x60

'=

=^^-^""'

1850

Substituting these values into equation

C^,.-i

V gives

= Q.. + [4-76xlO-^QJC^., + 3-55)-l-63x

xlO->63-C^.J^]52-6
= 0-0164Ci,^+ 1-202 C^,^-0-376
VI

Arbitrary values of C^ given in the table were selected and substituted

into equation VI to give the corresponding values of C^ ,_i.

^A,m-t

^A,m

^A,m-i

is

6-6

8-27

60

7-43

5-0
4-0

30

6 05
4-70
3-38

2-0

2 09

plotted against

C^ ^

in Fig. 9.4;

and

starting with the feed

of 6.63 g moles/1, the number of reactors

The required conversion is 60%,

sition

.*.

acid in final effluent

.".

number of

0-40x6-63

is

2-65 g moles/1;

reactors to produce this effluent

Tillert extended the graphical procedure presented

t Tiller, F.

M. Chem.

Eng. Prog. 44, 299 (1948).

compo-

''stepped off" as shown.

is 7.

above to second order

334

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

equations.

Thus consider the second order non-linear

finite

difference

equation

yn^2-^Pyn^i
where

and

and

are functions of >'

+ Qyn = R
P can be a

8
Acetic acid

in

feed

A/

/h

<

//

6-

1'
3

7-

'^

(9.70)

function of >'+i.

1
1

'

Fig. 9.4.

Graphical reactor analysis

The graphical procedure is as follows. On rectangular coordinates plot

{RQy^v^y^ as represented by the curve AB in Fig. 9.5. Similarly plot
[/? (2+^)>^ + i] vs>'+i on the same scales and let this curve be represented by AC. Finally, plot the diagonal y^ vs >^, represented by OD. Two

R-(P+0)y

R-Oy

Fig. 9.5.

Graphical solution of a nonlinear second order difference equation

9.

boundary conditions

will

be

335

FINITE DIFFERENCES

known

for a second order equation; yo will be

one, and y^ can be evaluated by using an overall material or energy balance

for the problem. Mark off yo and yi on the abscissa and draw ordinates

from each to the curve AC. The vertical difference between the curves at yi
corresponds to -Py^ and if this is added to (RQyo) at L to give the point
M, the vertical distance from the abscissa to
corresponds to y+2 ^Y
equation (9.70). Drawing a horizontal line to the diagonal at T and a vertical
line to the axis locates y2 in its correct position. It cuts the curves AB and
AC to give Py2 which on adding to the AB curve at y^ gives the value of
y^ which is located in its correct position relative to y2 by drawing a horizontal line from V to the diagonal. This establishes the stepwise procedure and
can be continued to locate y. In Fig. 9.5 the stepwise procedure is indicated
by the arrows.
The graphical procedure given above for second order non-linear difference equations is of limited application. Thus it cannot be applied to the
solution of equations in which product terms appear from more than one
level of >'. For instance, in Example 3, Section 3.3.9, equation I could not
be solved by this graphical method because on rearrangement to the form

y^,-{A-Cy.,+Ey)y + {B-Dy.,)y.,=0

(9.71)

P = A-Cy.,+Ey

(9.72)

the function

be noticed that P contains terms involving y and y-i. The term

Cynyn-i cannot be transferred to the other term without upsetting the
condition to be satisfied by Q in equation (9.70). Hence equation (9.71)
cannot be solved by the graphical method proposed by Tiller; the numerical
method to be presented in Section 11.4.1 must be used.
It will

9.7.2.

Analytical Solution

At the present time only a hmited number of non-linear finite difference

equations can be solved analytically and these are solvable because they can
be transformed into linear equations. Thus consider the following example.
Example

2.

Solve the equation

Converting equation

to logarithmic form,

log>'+2

logy

II

21og);+i

Letting
"n

equation

II

logy

becomes the second order

n +

linear equation

2-2Mn+l+Wn =

Solution of equation III by the methods already explained

Wn

or

logy

=
=

+ C2W
Ci + C2n

Ci

IH
is

IV

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

336

Now

Ci=logX andC2 = logB

y^ = AB"

let

then

VI

The solution of the above equation depended on the linearization of equation

I and unless such a transformation can be found it is not possible to solve
the equation.

non-linear finite difference equation of second order that frequently

is the Riccati finite difference equation.

appears in engineering problems

yn+iyn + Ay^,+By, + C

where A, B and C are constants.

by the following method.
Let

and

It is

yn

+d

(9.74)

and rearrange thus

+ S){u + S) + A(u^,+S) + B(u + d) + C =

+ (B + S)u + lS'' + (A + B)S-\-C']=0

u+,u + (A + d)u + ,
If S is

(9.73)

converted to a Hnear difference equation

substitute equation (9.74) into (9.73)

(u^,

(9.75)

chosen so that
S^-\-(A

and equation

(9.75)

is

+ B)S-\-C =

(9.76)

divided by (u+^u), then

(B +

S) + (A + d)- +1 =

Putting

lK =

ll(y-3)

(9.77)

(9.78)

then equation (9.77) becomes

x^,+Px, + Q =
which

is

first

order linear

finite difference

P = ^^

B+S

(9.79)

equation with

J-ande=
^ B+S

Equation (9.79) can be solved by the methods already given, to yield the
final solution

y-3
where the constant
problem. This

is

is

A+S
{- B + 3
L

.A

A + B + 23

(9.80)

from the boundary conditions of the

example below.

to be evaluated

illustrated in the

Example 3. A benzene-toluene feed containing 60 mole per cent of

benzene is fed continuously to a distillation column. If there are 9 plates
between the reboiler and the feed plate, and the top product contains 98 mole
per cent benzene whilst the liquid leaving the base of the column contains
2 mole per cent benzene, estimate the overall plate efficiency of the column.
The feed enters the column at its boihng point and the relative volatility of
benzene to toluene can be considered to be constant at 2-3. The reflux ratio
is

3-0.

9.

Consider 100 lb moles feed and

and

let

D=

W
D+W

=
=

iOO

.'.

337

FINITE DIFFERENCES

moles

distillate;

moles residue.
I

+ 0-02(100-D)
and W = 39-6 lb moles.

60

II

0-98Z)

from which D = 60-4 lb moles

mass balance between the bottom of the column and some plate n

stripping section

Lx, + ,
where

L
G

is
is

= Gy-^Wx^

III

the molar reflux down the column and

the molar vapour rate up the column.

Since the relative volatility

X and y on any

plate

is

constant, the equilibrium relationship between

is

Substituting equation

IV

"^"

y^

IV

into equation III gives

aGx
_
Wx
"^^"l+(a-l)x/

-"--" +

,-n-[

L = F + RD =

Now
and

in the

is

L(a-l)

(100 + 3 X 60-4)

^"-L(a-1) =

V
^

VI

281-2 lb moles/h

letting

^=

^~
=
C=

^=-^ =

0-769

1-3

aG + (a-l)Wx^ _

(2-3

x 4 x 60-4) + (l-3 x 39-6 x 0-02)

L(a-l)

281-2 X 1-3

1-523

Wx^ = 39-6x0-02 ^^^^^

^^
= 0-0022
281-2x1-3
L(a-l)

equation VI becomes the Riccati equation*

x^,x + Ax^,-Bx-C

The

solution of equation VII

in the

above

text.

Thus

is

obtained by following the procedure set out

let

VIII

X-d
where 3

is

obtained from the equation

S^

or

VII

+ (A-B)d-C =
(5^-0-754(5 =

0-0022

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

338

The

roots of this equation are

5

The complete

or -0-003

0-757

solution of equation VII

is

A+d
Selecting the root S

when =

0, x

= 0-003 and

XI

A-B + 26

1>

inserting the

boundary condition that

0-02 gives

K = 42-2
The value of corresponding

to x

0-60

is

given by

0-7661"
1-66

42-2

1-32

XII

1-526
as

7-0 ideal stages.

Subtracting the reboiler from this

overall plate efficiency as 6-0/9

9.8.

number

gives the

67%.

Differential-Difference Equations

It has been shown in the above sections that chemical engineering equipment consisting of a number of stages can be analysed by means of finite
difference equations when the operating conditions are steady.
However,

when

of equipment is subjected to a step change in operating

started up, or is being shut down, the compositions of the
streams passing through these stages change with time. This results in the
presence of differential terms in addition to the finite difference terms in what
is known as a "differential-difference" equation.
This type of equation will
be treated in this section, but it should be pointed out that on many occasions
the final equation describing the process is too complex for analytical solution, and it is necessary to make a "stage-to-stage" analysis with the aid of
a computer. The solution of such problems will be treated in Chapter 1 1
The analytical solution of differential-difference equations is accomplished
by converting this type of equation into a finite difference form by means of
the Laplace transformation.
The transformed equation is solved and
inverted to show how the conditions of the process streams passing through
the equipment vary with time during the transient period. The technique to
apply, where possible, is illustrated by the following examples.
this type

conditions, or

is

Example 1. A system consists of A'^ stirred tanks each of volume v ft^

arranged in cascade. If each tank initially contains pure water and a salt
stream of concentration ^o Ib/ft^ is fed to the first tank at a rate R ft^/h,
calculate what the output concentration from the last tank should be as a
function of time if the stirring is 100% efficient.
Use the result to compare the transient behaviour of all systems of total
volume Nv

= V (constant).

Solution.

339

FINITE DIFFERENCES

9.

With the normal convention regarding

subscripts, a salt balance

over the th stage gives

Rx.,-Rx =

v^

Taking the Laplace transformation of equation I by using equation (6.11) to

remove the derivative, and remembering that the system initially contains

no

salt,

Rx.i-Rx =

.-.

Solving this Unear

finite difference

vsx

II

equation by the method of Section 9.6

gives
III

\R + vs/
Because the feed composition

is

constant,

Xq

IV

XqIs

s=^f_^y

'"-7(f5)"
From

transform number 10 in the table at the end of the book,

[(_!)!
Hence, using equation

(s

+ af

(6.21),
t

Integrating the right-hand side of equation VII by parts

[s{s

+ ar\

(N-i)\[

Inverting equation

'

_^N-l^-at

iN-2^-at

fl(N-l)!

a\N-2y.

'

times gives

^1

}'

-t^-^e'"^

+ a(N-2)\J
a{N-l)\

VI by using equation

*''-^o[l

'

(N 1)

t^'-^e-^'dt

^-at
'"

+
a""

VIII,

'

(jv-1)!

-.
a^

VIII

MATHEMATICAL METHODS IN CHEMICAL ENGI^fEERING

340

^S

^0 ""^0

Rtiv

Rt

,N-1

{Rt/vy

{RtIvY

IX

which is the solution to the first part of the problem.

For the second part, if K = Nv, equation IX becomes

1 H

In particular,

if

A'^

NRt {NRtVf
^^^ +
+
V
2!

if

A'^

{NRtVf-^
^
{N-\)\

'n

and

...

^o~"^o

Rt/V

XI

4,

>

4Rt

^^^/^ \^

32/Rt\

^/RtY

XII

0-5
h"1h

Fig. 9.6.

It is difficult

Nv

Output from an

to evaluate equation

into equation

VI

A'^

tank system of total volume

in the limit as

N^oo,

but putting

V=

gives

which can be expanded by the binomial theorem thus,

N + l/Ksy
(iV +l)(iV + 2) /VsV
Vs
/VsV _ {N+1){N
3!iV2
r'^ 2N [rJ
\r)
3!
yj^j

XIII

9.

If

is

341

FINITE DIFFERENCES

large, the expression in the

square brackets can be approximated to

give
Vs

/VsV

/VsV3

= ^^-K./K

XIV

The

inversion of equation

Hence

of the book.

as

XIV

is

number

given by

61 in the table at the

end

N^ oo
XV

x^-^XoS,(t)

k=

where

V/R

Equations X, XI, XII, and XIV have been plotted in Fig. 9.6 to show
the system responds for different values of A''. The important result
emerges that a large number of small stirred tanks in series has approximately the same response to a step change as a single tube of the same volume
with plug flow. The converse is also true, that flow in a tube with a small
amount of back-mixing can be represented by a chain of stirred tanks of the
same total volume.

how

Example 2. Acetic anhydride is to be hydrolyzed in a laboratory stirred

tank reactor battery consisting of three vessels of equal size. 1800 ml of
anhydride solution of concentration 0-21 g moles/1 is charged into each
vessel at 40C, and 600 ml/min of a solution containing 0-137 g moles/1 is
continuously fed to the first reactor. Estimate the time required for the
reactor system to settle down to steady state operation.

The

specific reaction rate constant for the hydrolysis at

40C is 0-38 min~ ^.

and consider

Solution. Initially let there be a battery of n reactors in series

a material balance over reactor m.

Let q be the volumetric flow rate.

V be the volume of each reaction vessel.
C be the concentration of anhydride.
of the liquid and the temperature of each reaction vessel
considered to be constant, then
If the density

Now

V/q

9 the holding time,

dC

and r^

dt

Since

all

the vessels are to be of the

is

kC^.

-cCil")
""'

same

size

'

and the battery

is

to be operated

isothermally,

- =Q(constant)
M.M.C.E.

III
12

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

342

Thus carrying out the Laplace transformation on equation

II,

by making

use of

^[^]=5C(s)-C,(0)
from Section

6.2.1, gives

5C(s)-C,(0) + eC,(s)

^C,-i(s)

IV

But the vessels all start with the same composition in them, hence C{0) =
Equation IV can be put into the standard form for a first order differ-

0-21.

ence equation thus,

0(s

Solving equation

+ Q)C^is)-C^.,(s) =

0'2ie

by the methods of Section 9.6 gives the general solution

0*21 9

where

an arbitrary constant.

is

m=

When

Co

Co(s)

0-137

0-210

0-137

= ^ + e-(^^ePl

which determines A. Substituting into equation VI and rearranging gives

0-210

0-137
'"^^^

"

sO'^is

+ QT

0-137

'^

0{s

+ Q)-\[

^...r

0-210

0"(5 +

+ ^.

Qt\

...

VIII

by algebraic division.
Each term in the square brackets can be inverted directly by using the
tables at the end of the book (transform number 10), and the first term can
be inverted by using the property of the transform given by equation (6.21).
Hence the solution is

C.=

0-137

0-137

r,

{Qtf

(eO""M

nt

Equation IX is a general expression giving the concentration of the

from a stirred tank reactor battery containing n vessels of equal
size, each starting from the same initial concentration.
Expressions of this

effluent

343

FINITE DIFFERENCES

9.

type have been derived for different kinds of reaction by Mason and Piretj
and the reader is advised to consult these authors for a more extensive
analysis.

In the special example considered here, the last vessel to reach steady
state will be the third,

and since

1800

,
= - = t;^=3-0^
600

^^ =

2=

^ + 0-38

0-713

OQ .-.

C3

C3

2-14

0-014-0-014(l+0-713r + 0-254r^)e--'^^^

+ 0-21(1 + 0-333r + 0-056r^)e-''^'^'

.-.

0-014 + (0-196 + 0-060r + 0-0082f2)e-'''^'

which gives C3, the composition of the effluent from the

a function of time, has been plotted in Fig. 9.7. This shows
that the laboratory reactor battery has reached steady state conditions
within the limits of experimental detection in about 15 minutes.
Equation

final reactor as

I-'

0-20^

0-15

O'lO

-\

0-05 -

^^^-o.^.___
1

?~

?^

10

12

14

16

Time (min)

Fig. 9.7.

Time

to reach steady state

Example

3.
1000 Ib/h of a solution containing 0-01 lb nicotine per lb of
being extracted by 1200 Ib/h of kerosine in a counter current
extractor containing 8 stages.
If the concentration of the feed hquor is
suddenly changed to 0-02 lb nicotine per lb of water, how long will it take
for the extractor to settle down to steady state operation under the new
conditions ?

water

is

The equiUbrium

relation

is

y = 0-86X
where

and

The hold up per

Y is

X is lb

of kerosine

nicotine per lb of water.

stage can be taken to be constant at

400
t

lb nicotine per lb

Mason, D. R. and

lb

water and 200 lb kerosine.

Piret. E. L.

r.E.C. 42, 817 (1950).

MATHEMATICAL METHODS

344

Y.

"

Yn

CHEMICAL ENGINEERING

IN

Y...

r^

<

>

>

Yn+1

>

^1

Fig. 9.8.

K-

Stagewise counterflow system

Solution

The system
meanings.

is

and the symbols have the following

illustrated in Fig. 9.8

R Flow

rate of water

Flow

rate of kerosine phase (1200 Ib/h)

Total number of stages

phase (1000 Ib/h)

X Nicotine concentration

(8)

in water phase in th stage

Y Nicotine concentration in kerosine phase in nth stage

Distribution ratio (0-86)

Hold up of kerosine per stage (200 lb)

Hold up of water per stage (400 lb)

Xi Original feed concentration

Xp New

(0-01 lb/lb)

feed concentration (0-02 lb/lb)

nicotine balance over the nth stage gives

RX., + SY^,-RX,-SY =

dX.

h-^ + H

dY
^^

The equilibrium

relationship gives
Yn

Eliminating Y, Y +

= mX

between equations

II

and

II

and rearranging,
dX."

aZ^i-(a + l)X + Z_i=/?

where

The plant

= mSIR,

a
is initially

(h

+ mH)IR

III

IV

operating at steady state, and the concentration in

first.
Thus, putting

each stage must be determined for these conditions

dXJdt = 0, equation III becomes

aX^,-{a+V,X + X,n-l
The method of Section 9-6

gives the general solution

X
The boundary conditions
at

0,

iV

VI

A^-B(x-''

for this part of the

problem are

^0 = ^0

VII
at

+ l,

Ys^,^mX^^,^0

9.

^=

Hence,

345

FINITE DIFFERENCES

^^

Therefore, equation VI gives the

^=

a^

initial state

^"'

of the system where

A and B

are determined by equations VIII.

For the unsteady state part of the problem,

Laplace transformation of equation III. Thus

it

necessary to take the

is

ocX^,-(x + l)X + X.,=P(sX-A-Bcc-")

where equation VI has been used for the

IX
Rearranging,

conditions.

initial

or in operator form,

[a:^-(a+l+Si5):+l]X_i

= -Ap-B^o^-"

Equation X is hnear so that its solution can be expressed as the sum of a

complementary function and a particular solution. The complementary
function is given by

X., = Cpl-'+Dpr'

P1+P2 =

where

and

P1P2

(a

XI

+ l + si5)/a

XII
XIII

1/of

The

particular solutions for the separate terms on the right hand side of
be found by the inverse operator method described in Section 9.6.2.
Thus for the first term

X can

a-a-l-sj5 + l

and for the second term,

-BfaCombining equations XI, XIV, and

equation X; which is

X = Cp\ + Dpi

or

XV

gives the complete solution of

-4-

Fir/~^

XVI

The boundary conditions can

at
at

= N+l,

also be transformed thus,

Xo = X,,

0,

X;v+i=0,

.'.

.'.

Xo =

X^^,=0

555

Substituting these boundary values into equation

VIII to eliminate

X,ls

A and
Xp

XVI and

]
>

XVII

using equations

B, gives

^ ^ + A+B = ^
-=C
C+D +
+D

X'o

XVIII

MATHEMATICAL METHODS

346

and

CHEMICAL ENGINEERING

IN

cpr^+^pr^+

Cp^,^'+Dp''2-''

^^^'

XIX

Solving equations XVIII and XIX for

equation XVI gives the solution

and D, and substituting

N+\
Pi

Pi

into

_ PiN+l Pin

XX

p^'-pV'

XX it must be remembered that pj and p2

by equations XII and XIII as functions of s. Therefore, the
method of residues must be used to invert the last term in equation XX.
In order to invert equation

are determined

To

simplify the nomenclature, put

The numerator on the right-hand side of equation XXI could become

infinite if IP2I became infinite.
But if \p2\ is large, equation XIII ensures
that IpiI is small. Hence it can be shown that as IP2I -> 00, J{s) -> 0; and
the only singularities off{s) arise from the zeros of the denominator. These
zeros occur at 5 =
and when Pi^"^^ = Pi^^^- The second solution shows
that IpiI = IP2I, and since Pi and P2 are the roots of a quadratic equation
with real coeflScients, they must be equal, or mutual complex conjugates.
Therefore, putting
p^

re-'^,

p^+i_p^+i ^

=
Hence f(s) has

singularities

when

P2

XXII

re''

J.N+l^^iN + l)i^_^-iN+l)ie-^

2ir^'-^sm(N + i)e

0,

XXIII

and when

(N+l)e = kn
where k

is

an

XXIV

integer.

Putting equations

XXII

into equations XIII

r^

2oircose

and XII

gives

XXV

1/a

= + l + sp

XXVI

(x

a + l-2Vacos[/c7r/(iV + l)]

^^^^^

(N+ 1),
XXVII gives a distinct set of values of 5 for /: = 0, 1,
but repeated values for integer values of k exceeding {N+ 1). When k = 0,
or (A^+ 1), equations XXII show that pi = P2, and an application of
L'Hopital's rule to equation XXI shows that/(5) remains finite as P2
PiTherefore /(5) has (A^'+l) simple poles:
Equation

one
and

at 5
A'^

0,

given by equation

XXVII

with k

2,

A^.

9.

347

FINITE DIFFERENCES

Using equation (4.113) to find the residue of f(s)

at 5

a"^

0,

Pi

P2

1,

0, gives at

XXVIII

^0= ^-N-i_i = ^^.1^1

Substituting equations

XXII

into equation
r"

^^^>

=7

-^^^

XXI

gives

sin(iV+l-n)e
^'^^^
sin(iV+l)e

sin(iV + l)0

Remembering that is a function of s given by equation

equation (4.113) to find the residue of e'^fis) gives
r"
s

sin

~s(iV+l)

_
~
*

(N + 1 - n)e 2ar sin 6 e^

cos(iV + l)0

XXV, and XXVII

Combining the

l)(a +

to simplify

[yv
.N+l-n
+ l-n

XXXI,

-(a + l-2Vacos0)n

^""^

l-2Vacos0)

residues as given

remainder of equation

XXXI

2(-l)^a-^"-^>/^sin0sin(iV+l-n)0
(7V

using

sm{N + l-n)e
e''
(N + l)cos(N + l)e{deids)

- r"
Using equations XXIV,

XXVI, and

by equation

(6.43)

and inverting the

XX,

r^N+l-n
+ l-n
r^/V

-i

-i

-]

^^^

2(-l)^a-("-^>/^sin^sin(iV+l~n)g

+ (Ap AoJL^

0/ z.
Jk=l

(iV

+ l)(a + l-2Vacos0)

^^

(iV+l)(a + l-27acos0)

Equation XXXII is the complete solution to the problem where 6 is given

by equation XXIV, and s by equation XXVII.
Since the final stage will be the last to reach the new steady state conditions, a graph should be plotted showing Xg as a function of time. From
equation XXXII,

+ -L^l .(X.-X,)L^z|^i^..
it

9(a

2yja cos 6)

XXXIII

^
MATHEMATICAL METHODS

348

IN

CHEMICAL ENGINEERING

Using the numerical values given in the problem,

=
=

knl9

a^

20
40
60

2
3

10159

1-868

3-087
4-384
5-603
6-596
7-245

80

0-3528

100
120
140
160

-0-3528
-1-0159
-1-5564
-1-9092

7
8

a-^/2^ 0-8956
2a-7/2 sin2^
sin2 d

9(a

0-2257
0-8743

0-544

1-9092
1-5564

4
6

+ mH
=
R

1-3278,

2 Va cos 6

20/c

mSi'R == 1-032

P
.-.

+ l-2Vacos

01170

01896

0-4132
0-7500
0-9699
0-9699
0-7500
0-4132

0-1729
0-1469

01170

00059

9)

01149
0809
0-0489
0-0230

= 0-195-0-1896 e^'' + 0-1729e^^'-0-1469e^^^ +

+ 0- 1 149 e''' - 0-0809 e''' + 0-0489 e''' - 0-0230 e''' + 0-0059 e'''

lOOXg

Equation

XXXIV

XXXIV

has been plotted in Fig. 9.9 from the following cal-

culated values.

^8

2
3

4
6
8

0972

10

0-0975
0-1012
0-1107
0-1232

12

14
16
18

01470
01640

20
24

0-20

x%

0-1752
0-1824

01870
01899
0-1917
0-1929
0-1942
<i>-^

,^-c^

^-'^'''^''^

0-18
<o

o
O

/^

0-16 r

rf

U-14

012
0-10,

r^\
2

10

12

14

16

18.

20 22 24

Time

Fig. 9.9.
It

can be seen that

the

new

conditions.

it

Nicotine concentration in extract

takes about 24 hours for the system to settle

down

to

Chapter 10

TREATMENT OF EXPERIMENTAL RESULTS

10.1.

Introduction

analysis of many chemical engineering problems is concerned with using

experimental results to verify theories or develop empirical formulae. Some
of the data points are often inaccurate and methods must be found for
This subject falls
eliciting rehable information with reasonable certainty.
into the realm of statistics, but the mathematical approach given in this

The

chapter yields vaHd conclusions.

great deal can be learnt by presenting the results graphically with
suitably chosen coordinates and inspecting them visually; and the use of
various kinds of available graph paper is described in the first part of the
chapter. With the exception of Section 10.4, where a contour representation
of three variables is described, all discussion is confined to systems of two
variables it being assumed that all other variables can be kept constant and
treated as parameters.
Experimental results can also be used to determine an average value of a
variable, or a set of data can be integrated to find some total value. For
example, the determination of an average velocity across the cross-section
of a pipe or piece of equipment involves averaging data points; and the
prediction of an accumulated composition from a set of values of point composition necessitates numerical or graphical integration. Methods for performing these calculations are given in the final part of this chapter.

10.2.

Graph Paper

Results can be presented in the form of a table or as a graph. Although

the tabular form is capable of showing greater accuracy, the graphical
presentation is more readily assimilated, because it gives a clearer picture

of the behaviour of the variables. The graphical method is usually sufficiently

accurate for the presentation of experimental points, but for theoretical
points calculated from a numerical
is

method

(see

Chapter

11) a tabular

form

preferred.

For a particular piece of equipment where the number of variable

is small, a direct plot of the variables is the most useful, but if
comparison with other work on similar equipment is envisaged, the appropriate dimensionless groups should be used along the axes. If the dimensionless groups are chosen correctly, each group should only contain one
principal variable and the rest of its constituent parts should be parameters.
parameters

349

MATHEMATICAL METHODS

350

IN

CHEMICAL ENGINEERING

No purpose would be served by plotting a dimensionless group against a

dimensional variable and this form of presentation should be avoided. Any
mathematical equation must be dimensionally consistent and hence if one
term is dimensionless, so must all other terms be dimensionless. Disregard
of this important fact can lead to apparent anomalies, particularly when a
logarithmic plot

is

used.

There are many types of graph paper which are described below and the
" The most desirable
correct type should be chosen for each problem.
shape for a curve is a straight line," and the scales should be chosen in such
a way that the major part of the curve is incHned at 45 to each axis. This
ensures equal accuracy in all regions of the graph both in plotting the curve

and reading

results

from

it.

Linear Graph Paper

10.2.1.

The most common and

generally used graph paper has its axes divided

any range of numerical values of a variable can
be accommodated by choosing a suitable scale. Most normal types of graph
paper are subdivided into ten divisions, and it is customary to count in a
decimal system; hence factors of three should be avoided when choosing the
scales, so that each subdivision will represent an easily written number.
into equal intervals so that

The equation of a

straight line

y
where
to

m is

0.

is

usually written

= mx + c

(10.1)

the gradient, and c (the intercept) is the value of y corresponding

Because c is defined in this manner, the origin should not be

suppressed.

For a general curve on

linear scales, the equation

y=/(x)
is

satisfied.

By

(10.2)

differentiating equation (10.2) the gradient of the tangent at a

Conversely, if a tangent is drawn to the curve then a

numerical value can be found for the derivative at that point. The most
accurate way to differentiate a curve graphically is to choose a gradient and
construct a set of parallel hnes with that gradient which intersect the curve
point can be found.

Fig. 10.1.

Graphical differentiation

TREATMENT OF EXPERIMENTAL RESULTS

10.

351

If each chord so constructed is bisected, and the

as shown in Fig. 10.1.
midpoints are joined with a smooth curve, this curve must intersect the
original curve orthogonally at the point where the tangent has the chosen

slope.

Integration of equation (10.2) with respect to x between the fixed hmits

is represented graphically by the area bounded by the
curve y = /(x), the ordinates x = a and x = b, and the x axis. This can

X = a and x = b

readily be visualized by dividing the area into vertical strips of equal width

sum of the areas of these

This area can be evaluated by
counting the squares on the graph paper which form the desired area, or by
using a numerical method from Section 10.7.
and considering the

Semi-Logarithmic Graph Paper

10.2.2.

In

integral as the limit of the

as their width approaches zero.

strips

many

practical cases, particularly those involving the

approach to

steady state conditions of a process, the dependent variable is a decaying

exponential function of the independent variable. For example, in Section
1.6,

equation (1.29)

is

of the form

y^Ae-'""

(10.3)

an experiment is designed to determine the

coefficients A and b from a series of measurements of x and y, it is more
convenient to use the logarithmic form of equation (10.3) which is
after a

change of

origin.

If

my = \nA-bx
Defining a

new dependent

variable

(10.4)

by

y=

ln>;

(10.5)

equation (10.4) becomes

Y=-bx -{-In A

(10.6)

and (10.6) it can be seen that b is the

gradient of the curve of Y against x, and In A is the intercept. Because such
plots are frequently needed, semi-logarithmic graph paper is available which
has one scale divided logarithmically, as on a sHde rule. Hence, the values
of y can be used directly instead of consulting log tables to evaluate F, the
logarithms being intrinsically taken on the graph paper itself. It must be
remembered however, that the gradient {b) must be determined from the
linear measurement of Y and not from reading the scales which give values
of y. The intercept can be measured as a Hnear value of In A when x = 0,
or read directly from the scale of >^ as A, which is simpler and more usual.
It should be noted that the logarithms are all taken to the base e and the
gradient must be corrected for the scales of the axes. To do thfs, the measured
gradient must be multiplied by the length of the unit on the linear scale and
divided by the length of the unit on the logarithmic scale. The logarithmic
unit is the distance between 1 and 2-718 on the scale of y. Graph paper is
usually printed with these two units approximately equal, but a 5%
discrepancy is quite common and should be corrected.

By comparison of equations

(10.1)

MATHEMATICAL METHODS

352
10.2.3.

IN

CHEMICAL ENGINEERING

Logarithmic Graph Paper

Dimensional analysis frequently indicates an empirical equation of the

form
y = Cx"
(10.7)
where x and y are dimensionless groups and C and n are constants to be
determined experimentally. Because the groups are usually made dimensionless by using the physical properties of the materials used, the values of
X and y are likely to vary over very wide ranges. To combine many results
into a single correlation apphcable to such a wide range, a logarithmic scale
is superior to a Hnear scale. Also, the logarithmic form of equation (10.7) is

Y=nX + c
where

Y=

X=

and

In C;

(10.8)

and

equation of a
straight Hne. Therefore, for plotting such equations as (10.8), graph paper
with both axes subdivided logarithmically is available. It is normal for the
logarithmic unit to be the same along each axis so that the gradient n can be
found by direct linear measurement. The intercept c is the value of Y when
J^ = 0. This can be found more easily as the value y = C when x = I.
The origin of linear scales can never be plotted on logarithmic graph
paper. If any point is referred to as the origin, the point x = I, y = 1,
corresponding to J\f = 0, 7 = 0, is usually meant. The intercept as described above is determined relative to axes through this origin.
Inspection of equation (10.7) shows that if n is positive, then the curve
will pass through the origin of Unear scales. If n is negative, the curve will
In y,

In x,

ji

Fig. 10.2.

this is the

Equivalent curves on log-log and linear graph paper

be asymptotic to both axes. It is therefore useless trying to fit an equation

of the form (10.7) to any set of points which obviously intersect either axis,
unless the point of intersection can be transferred to the origin. The three
types of curve which give rise to straight line plots on log-log paper are
illustrated in Fig. 10.2.
10.2.4.

Triangular Graph Paper

In the study of liquid-liquid extraction systems where three components

are present in two phases, a convenient graphical representation of the

10.

TREATMENT OF EXPERIMENTAL RESULTS

353

composition of a phase is needed. The data can be presented on a triangular

diagram by using the geometrical property of a triangle which is proved
below.
Figure 10.3 shows any triangle
lines

AP, BP, CP

The

Fig. 10.3.

PG

is

lie

drawn
on BC.

Triangles

parallel to

BPH, BEC

AB, and

Similarly, triangles

CPG, CFB
.-.

PGH, ABC

triangular diagram

is

drawn

parallel to

AC so that G and

PE:BE::HC:BC

C10.9)

are similar.

PF:CF::BG:BC

(10.10)

are similar
.-.

Adding equations

PH

are similar.
.-.

Also, triangles

and any interior point P. The

BC, AC, AB at D, E, F respectively.

ABC

are produced to meet

PD:AD::GH:BC

(10.11)

(10.9, 10, 11) together gives

Dropping perpendiculars

PE PF PD
= 1
+
+
BE CF AD
from P and B on to AC

BE

,^^,^,

(10.12)
^

at

and

shows that

BR

Hence, if hnes are constructed parallel to AC and subdividing BR into

equal intervals, the concentration of one component of the ternary mixture
can be read against the scale as the ratio PQ/BR which is the same as
PE/BE. Similarly, the other two components can be represented by the ratios
PD/AD and PF/CF. Equation (10.12) then shows that the three separate
concentrations add up to the whole mixture, for any point P inside any
triangle

ABC.

Although the above proof is valid for any triangle, it is normal practice
to use an equilateral triangle for a symmetrical representation of the three
components. If printed triangular paper is not available, however, it is more
convenient to use rectangular graph paper and construct diagonals so that
an isosceles right-angle triangle can be used.

MATHEMATICAL METHODS

354
10.2.5.

IN

CHEMICAL ENGINEERING

Special Graph Papers

There are many other types of graph paper available for special applicaFor example, the axes can be subdivided to a quadratic scale or a

tions.

square root scale for orifice calibrations. Alternatively, a reciprocal scale

can be used to automatically change a flow rate into a residence time. Other
special graph papers are provided for thermometer calibrations, and then the
borderline between special graph papers and charts for chart recorders is
rather ill-defined. Any of the above graph papers can aid the understanding
of a phenomenon by providing a pictorial representation, and anything
which does this can be loosely described as a graph.

10.3.

Whenever a

Theoretical Properties

of experimental data is plotted against linear scales, a

decision has to be taken regarding the type of curve which best summarizes
the results. Experimental data are always subject to error and the assessment of these errors will be discussed in Section 10.5. Hence any fundamental
properties which the curve must possess on theoretical grounds are invaluable
aids to the correct choice of representative curve. There are three points
which are always worth considering and these will be discussed separately.
10.3.1.
It is

set

The Origin
often possible to state with certainty the value of the dependent

when

the independent variable is zero. For example, if the pressure

drop across a length of pipe is being determined for various flow rates, it is
certainly true that the pressure drop is zero for a zero flow rate. Although
all other points on the graph are subject to experimental error and the curve
need only pass close to them, the origin is one point which the curve must
pass through. Hence the origin is more important than any single experimental point.
When time is the independent variable, the intercept at zero time is
frequently set as a boundary condition instead of being a measured value,
and this point is thus more accurately determined than any other. If a
chemical reaction is being studied by plotting the concentration of one of the
products as a function of time, the origin must be a point on the curve.
There are many other examples, and a few moments' thought should
always indicate whether or not a curve should pass through the origin.
variable

10.3.2.

Starting Gradient

In most problems involving axial symmetry, the gradient of the curve

tends to zero as the axis

is

approached.

Examples of

profiles for flow inside a cylindrical pipe,

this are the velocity

temperature profiles within a

cylindrical conductor, or the variation of local

mass

transfer coeflficient

near the front pole of a dissolving sphere. Not all dependent variables have
a vanishing derivative on the axis of a symmetrical system as may be seen

10.

355

TREATMENT OF EXPERIMENTAL RESULTS

from equation XV, example

4,

Section

distribution for slow flow past a sphere.

which gives the

8.5.4,

When

0, C

0,

but

vorticity

dC^jdd

0.

of the final product of two or more consecutive

chemical reactions is being plotted against time from mixing the reactants,
the curve must pass through the origin, and have a vanishing derivative
when it does so. This is because the rate of formation of the product is
proportional to the concentration of the intermediate product which is
If the concentration

initially zero.

10.3.3.

Efficiencies

lot depends upon the context, but most efficiency curves either approach
a Hmit asymptotically, or pass through a maximum value before decaying to
zero. One thing is certain, however; an efficiency curve cannot be a straight
hne over an appreciable range. Any straight line, except a horizontal one,

must exceed 100% and fall below zero for some values of the independent
variable, and both types of behaviour are inconsistent with a properly
defined efficiency.
10.4.
It is

Contour Plots

often difficult to isolate the independent variables in a system so

that one variable can be altered through a range of values whilst all other

independent variables are held constant.

When only two independent
variables are involved, it is possible to estabhsh a family of curves by presenting the results in a manner analogous to a contour map which shows
height as a function of two independent coordinates. This method has been

up rate

250lb/hr

Feed composition
Fig.

10.4.

Contour plot of three variables

used to illustrate the dependence of start-up time on boil-up rate and feed
composition for a distillation column.j The results are shown in Fig. 10.4
in the form of start-up time varying with feed composition, using the boil-up
rate as a parameter. It will be seen that very few experiments were performed
t Barker, P. E., Jenson, V.

G. and Rustin, A.

Inst. Pet. J.

(To be published).

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

356

at the exact boil-up rate, the actual rates being indicated adjacent to the

However, the three curves shown are almost as accurately determined

would have been from the same number of experiments performed
at the three exact boil-up rates. There is no reason why the positions of the
three variables should not be interchanged, but it is customary to plot one

points.

as they

of the independent variables as abscissa.

Propagation of Errors

10.5.

Any

experimentally determined quantity is subject to error, and hence

result which is based on experimental evidence is also limited
The determination of the derived error from the observed
in accuracy.
error is a calculation of some importance since it indicates the position where

any calculated

The way in which errors

experimental technique should be improved.
accumulate is governed by different rules depending upon the type of calculation performed

(i.e.

addition,

multiplication,

subtraction,

etc.),

and

these will be considered separately.

10.5.1.

Propagation through Addition

If a quantity z

is

calculated

from
z

x-hy

(10.14)

where x and y are two measured variables, then any error in determining
X 01 y will result in an error in z. Equation (10.14) is assumed to be exactly
true and satisfied identically by the exact values of x, y, and z. But any
experimental reading {x') for the variable {x) will only approximate to the
true value and this is usually stated in the form

x-dx <x' <x + dx

i.e.

the error in the determination of

is

(10.15)

given by

\x'-x\<5x

(10.16)

dx is called the " absolute error " in x, has the same dimensions as x, and
is usually assigned by the observer with due regard to the manner in which

measurement was made.

argument leads to the
Thus
the

A similar

definition of dy, the absolute error in y.

y-dy<y' <y + dy
The inequahties

(10.15)

and

(10.17)

(10.17) can be written in the alternative

forms

x'-6x<x<x'^-dx
y'-dy<y<y' + dy
Adding y
x'

to each term in inequality (10.18)

+ y'-dx-dy < x'-dx + y < x + y <

and

x'

(10.18)
(10.19)

using (10.19) gives

+ dx + y <

x'

+ y' + dx + dy
(10.20)

TREATMENT OF EXPERIMENTAL RESULTS

10.

Remembering
y\ 6x and

that

357

x and y cannot be determined except by means of x\

can be combined with equation (10.14)

dy, inequality (10.20)

to give
x'

But

z' is

+ y'-dx-5y <

z'

Comparing

the

+ y' + 6x + dy

is

the absolute error in

and

(10.22)

it

(10.22)

can be seen that

5x + 5y

(10.23)

Hence, the absolute error

z.

in the result

the absolute errors in the constituent parts.

Propagation through Subtraction

Assuming

that

= x-y

(10.24)

and using the same terminology as above, the following

can be obtained from (10.18) and (10.19).
x'

(10.21)

- {dx -Vdy)<z<z' + {dx + dy)

the inequalities (10.18)

sum of

10.5.2.

x'

and hence

dz

is

<

the result obtained by putting the experimental values {x' and y')

into equation (10.14),

where dz

dx {y' -\-dy) <x' dx y <x y <x' -\-dx y <

set

of inequalities

x' -\-dx

{y' dy)
(10.25)

But

z'
:.

z'-{6x + 6y)
.-.

dz

x'-y'

<z< z'-\-{dx^-dy)

(10.26)

(10.27)

dx + dy

Hence, the absolute error in the difference of two quantities

absolute errors in those quantities.

The above two

results

is

the

sum of the

can be generahzed to any number of operations

involving addition and subtraction only. Thus, the absolute error in the result

of any calculation involving only addition and subtraction is the sum of the
absolute errors in the constituent parts.
It should be noted that each operation increases the absolute error, but
any subtraction reduces the result. Hence the " relative error", which is the
ratio of the absolute error to the result, will always increase during subtraction.

The

relative error

is

usually expressed as a percentage

and

is

dimensionless.
10.5.3.

Propagation through Multiplication and Division

Using the same terminology as before but

z

letting z

be determined from

= xy

(10.28)

the inequality
(x'
is

- dx) {y' -5y)<z<

valid provided that x'

and

y' are

{x'

+ dx) {y' + by)

both positive.

(10.29

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

358

Defining the relative errors by

Xf

and

Sx/x'

y,

Sy/y'

inequality (10.29) can be rewritten

x'y'(l-x,)(l-y,)

<

< xy(l + x,)(l + jv)

(10.30)

Provided the errors are small, products of two errors can be neglected by
comparison with the error itself. Since z' is the value calculated by substituting x' and y' into equation (10.28), (10.30) can be written in the form

z\l-x,-y,)<z<z\l + x,-\-y,)
which shows that

x,

+ yr

Thus, the relative error in a product

is

the

z,

(10.31)

sum of

the relative errors in the

constituent parts.
Starting with the inequaUty

x'-Sx
y
it

+3y

+ 5x

x'

y -oy

can be shown that the relative error in a quotient

,.^^^x

is

also the

sum of

the

relative errors in the constituent parts.

10.5.4.

Propagation through a General Functional Relationship

can be seen from the above calculations that the error in the calculated
is greatest when the individual readings are at the ends of the error
ranges. The terminology can be simpHfied by writing
It

result

x'

= x + dx

(10.33)

with the interpretation that x' is the poorest acceptable reading obtained
unknown true value x; and 3x can be either positive or negative. If
z is to be determined from the general expression
for the

z=f{x,y)

(10.34)

by using the values

x' and y' for x and y^ then the value obtained (z') will
be in error as follows. Applying equation (8.9) to equation (10.34) gives

dz

= ^dx + ^dy
dx

(10.35)

dy

Provided the errors Sx and Sy are small, dx and dy can be replaced by them
in equation (10.35). Thus
dz

= ^Sx + ^Sy
ox

(10.36)

dy

Equation (10.36) can also be obtained from the generaHzed form of Taylor's
theorem by neglecting products of errors in comparison with the errors
themselves. The equation can be apphed to functions of any number of
variables and is the general law for propagation of errors.

10.

TREATMENT OF EXPERIMENTAL RESULTS

359

Example 1
If z = xy" with n known and x and y determined experimentally, determine the relative error in z in terms of the relative errors in
.

X and

y.
(x,

>')

df
J-

df
/=/
dx

.-.

xy"

nxy"-'

II

oy

Substituting into equation (10.36),

Sz

y"Sx-{- nxy"~'^Sy

(bx

z,

.'.

The

special case n

Sx and Sy can have

n Sy\

x,-hnyr

III

reduces to equation (10.31) for the product. Because

is negative equation III should be

either sign, if n

written
z,

Hence,

if

= x,-ny,

IV

the relative error in a quotient

is

the

sum of

the relative

errors in the constituent parts.

Example

2.

It

was shown

in the first

problem

in Section 2.5.7 that if a

chemical reaction

A->B
has a first order reaction rate constant /: hr~ \ the concentration of A leaving
a tubular reactor of length L ft with velocity u ft/h is given by

where Cq is the initial concentration of A, diffusion has been neglected and

plug flow has been assumed. How accurately must k be known and the
flow rate be steady for it to be possible to design a reactor to give 94J-95 J
completion ?
When the reactor is designed, L will be known very accurately; and (c/cq)
can be considered as a single variable (/) since the limit is on completion.

Applying equation

(10.36),
.-.

5f= --fSk +

^fSu

which can be rearranged using equation

Since

/<

the

left

hand

to give

-df _du

5k

f\nf~V

side will be positive.

"
Th

errors can be of either

MATHEMATICAL METHODS

360
sign

IN

CHEMICAL ENGINEERING

and the worst case must be considered, therefore

-df

Su

Sk

/In/

Using the numerical values given,

0-005

Equation IV shows that if the relative error in k exceeds 10% it will be

impossible to design the reactor to contain the completion within the given
Hmits.
If the flow rate is Hkely to surge by 5 %, then k must be known to
an accuracy of 5
otherwise the design is impossible.

Sources of Error

10.5.5.

There are many sources of error which should be assessed before quoting
They can be classified as follows.

the accuracy of any determination.

(i)
(ii)
(iii)

Errors of measurement,
Precision errors,
Errors of method.

The

first type of error is due to physical limitations of reading a scale.

Without a vernier attachment a length of 2 cm could easily be in error by
Precision errors are the "built-in" errors of the
J mm which is 2J%.
apparatus.
For example, the scales on a mercury-in-glass thermometer
assume that the bore is uniform, so that a reading of 28C could easily be
rC in error due to an uncahbrated scale. Errors of method include such
faults as neglecting heat losses, assuming constant molal overflow, or

neglecting back-mixing in a tubular reactor.

All of the above errors can only be estimated and rarely measured,

consequently an error

'*

is

significant figure "

usually given to only one or two significant figures.

any digit, except zero when it is positioning the

is

decimal point. It is usual to imply the accuracy of a reading by the number

of significant figures quoted. That is, if a reading is known to be in error

by about 2%, three

significant figures are barely justified; whilst an error

only justify two significant figures. To quote too many significant
figures is misleading since it implies a false sense of accuracy; whilst quoting
too few squanders the accuracy which has been dearly bought. One extra
significant figure should always be carried through a calculation to smother
the calculation errors due to rounding-off".

of

10%

will

Curve Fitting

10.6.

If a set

of experimental data is to be represented by an empirical equation,

the equation must possess the correct theoretical properties as discussed
in Section 10.3. Assuming that an equation of the form

+ bx''

(10.37)

10.

known

TREATMENT OF EXPERIMENTAL RESULTS

361

a set of results for x and y, values are required for a, b, and

graphically is to estimate a value for a and
plot {y a) against x on log-log paper. If the value of a is correct, the result
will be a straight line of slope n and intercept In b as described in Section
Too large a value for a will give a curve on
10.2.3, but this is unUkely.
is

The

to

best

fit

way of finding them

is convex upwards, and too small a value for a gives a

concave upwards as illustrated in Fig. 10.5.

log-log paper which

curve which

is

In I
Fig. 10.5.

Log-log plot of >'

bx"

The approach to curve fitting in the earlier sections of this chapter has
been of a visual or graphical nature, but there are two analytical methods
available. However, it is still necessary to choose the appropriate type of
curve, but instead of determining the parameters graphically, they can be
found analytically by the method of averages, or the method of least squares.

10.6. 1

Method of Averages

Having chosen the type of curve which it is desired to fit to the experimental data, the problem becomes one of determining the values of certain
parameters in the equation so that the " best fit " is obtained. In graphical
methods, the best fit is a matter of opinion, but in the present method a set
of average points is defined and the best curve is the one passing through
the average points.
Suppose that eight experimental values of a variable y are available at
eight different known values of x, and the best curve of the type

y
is

= ^ + Bsinx + Ccosx

to be chosen to represent them.

(10.38)

Because the values of y are experimental,

unlikely that any pair of values (x, y) will satisfy the best fit curve
Therefore, the experimental results are substituted into
(10.38) exactly.

it

is

equation (10.38) and an extra term

is

included for the

unknown

error.

Thus


MATHEMATICAL METHODS

362

CHEMICAL ENGINEERING

IN

y^

A Bsmxi - Ccosx^ = R^

y2

A Bsinx2 CCOSX2 = R2

y^-A Bs'mx^ CCOSX3 = R3

>'8

^sin X8

C cos Xg =

(10.39)

>

Ri

where R are the error terms.

and C analytically, three equations are

and these can be obtained from the set of eight equations (10.39)
by assuming relationships between the error terms. Assuming that
In order to determine A, B,

required,

333
Ri+i^2 + ^3

and adding together the

first

(10.40)

three of equations (10.39) gives

^ sinx-C ^ cosx =
X y-3A-B n=l
n=l

(10.41)

n=l

Dividing by the number of points used

Vi
_i

Vy
fi

sinxi +
+ y^
1 _ ^ _ ^

sinxj

(3) gives

sinx^
e

_ c_ cosx, +

C0SX2

cosx^

(10.42)

Equation (10.42) can be obtained directly from equation (10.38) by using

an *' average point " to represent the three actual points. It should be noted
that the average point is determined in a special manner which depends
upon the functions arising in the type of equation (10.38) chosen.
Two further equations, similar to (10.42) can be obtained by arranging
the remaining points in two groups.

Thus
y^ + ys

>^6

+ >'7 + >^8
3

_^_ ^ sinx4 + sinx5 _ ^ cosx4 + cosx5 ^ ^

AB
.

sin

Xg

sin X7

sin

cosxg

Xg

C0SX7

cosxg

3
(10.44)

Because every symbol with a suffix is a known experimental result, only

A, B, and C are unknown in the three equations (10.42, 43, 44) and hence
they can be determined.
In applications of the method of averages, the following points must be
observed.
(i)

(ii)
(iii)

(iv)

(v)

Points must be arranged in ascending values of x.

The number of groups must equal the number of unknown parameters,

Groups should contain approximately equal numbers of points,
Each experimental point should be used once only,
The appropriate average must be taken.

10.

TREATMENT OF EXPERIMENTAL RESULTS

363

Example 1. The thermal conductivity of graphite varies with temperature

according to the equation
k

ko-(xT

Experimentally, it is only possible to obtain a mean conductivity over a

temperature range. It is required to find the point conductivity from the
mean conductivity given below.

is

TCQ

390

500

1000

1500

km

1-41

1-38

119

115

T and

determined between

25C

in kilo-erg/cm^ secC

cm

^.

Solution
It is first necessary to determine the relationship between k and the
point values of k. Considering unit cross-sectional area of a conductor of
length L and end temperatures T^ and 72, an elementary heat balance yields
the equation

which defines the constant K, Substituting equation

I into

equation

II gives

-(kQ-ciT)dT = Kdx
-koT + i(xT^ = Kx + p

:.

III

Using the boundary conditions that

x

at

=
=

T=Ti

0,

TV

X
L,
T = ^2
-koT,+iaT^ = P
-koT2 + iocTi = p + KL

at
.-.

and

Subtracting equation VII from

VI

where T

is

= kJT2-T,)

IX
the end points.

There are four experimental points and two parameters

Therefore, the points must be spht into two groups of two, thus

/co-207-5a-l-41
/co-262-5a-l-38

2/co-470a
t I.C.T. 5, 86.

McGraw-Hill,

New York

VIII

ko-oiT

mean temperature between

the arithmetic

VI
VII

gives

koiT2-T,)-\a{T2-T,)(T2 + T,)=^KL
.-.

-2-79
(1937).

=
=
=

A:

and

a.

Ri
i?2

MATHEMATICAL METHODS

364

IN

CHEMICAL ENGINEERING

/co-512-5a-M9 =
/co-762-5a-M5 =

and

2/co- 1275a -2-34

Solving equations

and XI for ko and a

a
/Co

hence,

/c

i?3

/?4

=~~

XI

gives

= 000056

1-526

l-526-0-00056r

Substituting the experimental temperatures

into

XII
equation IX gives the

following comparison.

TCQ

390

500

1000

1500

km (experimental)

1-41

1-38

119

115

km (equation IX)

1-41

1-38

1-24

MO

Equation XII can be expressed in terms of engineering units by multiplying

the right hand side by 57-8, multiplying a by 5/9, and moving the origin
from 32F to 0F. The result is

= 88-9-0-0180r

/c

where T is measured in ^F, and k

example 4, Section 2.4.3.

in

Btu/h

ft^

XIII

F ft~^ This result was used

in

Method of Least Squares

often than not, this method is used to fit the best straight line to a
set of data. Hence the formulae will be derived determining the values of m
and c which give the best least mean squares fit of the equation
10.6.2.

More

y
to

N pairs

of results for x and

= mx-^c

(10.1)

y.

Following the method of the previous section, each pair of results is

substituted into equation (10.1) and an extra term is introduced to allow
for the

unknown

error.

y,-mx,-c =

(10.45)

R may be either positive or negative, therefore, to obtain a positive representation for the errors, equation (10.45) is squared thus,

yl^m^xl + c^'-lmx.y.-lcy. + lcmx, = Rl

(1

.46)

10.

365

TREATMENT OF EXPERIMENTAL RESULTS

Equations (10.46) for the individual experimental points are

now added

together to give

n=l

n=l

n=l

n=l

n=l

n=l
(10.47)

Equation (10.47) gives an expression for the sum of the squares of the error

The " method of least squares "

terms.

defines the best straight line as the

is a minimum.
be shown in Section 13.3 that the left-hand side of equation (10.47)
will have a turning value when its partial derivatives with respect to both

one for which the sum of the squares of the error terms
It will

and

Thus

c are zero.

2m

xl-2

n=l

xy+2c

2cJV - 2

(10.48)

n=

n=l
y + 2m

(10.49)

n=l

n=l

Solving equations (10.48) and (10.49) gives the result

NY.xy-Y,xY,yn
(10.50)

sums must be taken over all experimental points.

way to use the above formula is to construct a table with four
columns headed x, y, xl, xy and complete the table for the experimental
results. The sums of the columns will then give all of the values required
by equations (10.50). If a desk calculating machine is available, the four
sums can be obtained from two cumulative calculations without con-

where

all

The

best

structing the table.

The following example shows how

the

method can be extended

to a third

variable by fitting the best plane to the three dimensional plot.

Example

2. It

has been proposed! that the second order chemical reaction

CO + Cl2-^COCl2
proceeds on the surface of an activated carbon catalyst after adsorption of
the two reactants. Each of the three substances are adsorbed to a different
extent, but the number of sites occupied by carbon monoxide is small
compared with sites otherwise occupied. Assuming that the process is
controlled by the surface reaction, which

is

irreversible, find the best values

of the adsorption coefficients from the following experimental results.

t Potter, C. and Barron, S.

Chem. Eng. Prog. 47, 473

(1951).

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

366

Reaction Rate

Partial Pressures

CO

Cl2

COCI2

0-406
0-396
0-310
0-287
0-253
0-610

0-352
0-363
0-320
0-333
0-218
0-118
0-608

0-226

0179

000414

0-231

00440

0-356
0-367
0-522
0-231
0-206

000241
0-00245

000157
0-00390
0-00200

Solution

The following nomenclature is

C number of sites in the

used.
state specified

adsorption coefficient
k specific reaction rate constant
p partial pressure
r rate of reaction
Subscripts denote the following:

a carbon monoxide
b chlorine
c phosgene
V vacant site
total number of sites
/

The equilibrium of the three components between the catalyst and the
vapour phase can be expressed thus,

The

reaction rate

is

Ca

= KaPaC

C,

= K,p,C,

III

= kCXy

IV

given by
r

The

total

sites in

number of

each

state.

active sites

must be equal

to the

sum of the number of

Therefore

Q = Q + C, + Q + C,

is known to be small compared with

ignored in equation V. Using equations II and
from equation V,

But

C,

Ehminating

V
and Q, hence

III to

ehminate

it

= CXl + K,p, + K,p,)

Q and Q from equation IV by using equations

r

= kK,K,Clp,p,

can be

Q and Q
VI

and

II,

VII

Eliminating

between equations VI and VII and rearranging gives

+K,p, + K,p, = C.yJkK^K, y/^Jjr

Equation VIII

367

TREATMENT OF EXPERIMENTAL RESULTS

10.

is

VIII

of the form
l

+ ccx + Py-yz =

IX

where x, y, and z are the experimental quantities pi p^, and s/PaPj^- , P,

and y are the constants to be determined, and R is the experimental error.
Squaring equation IX and summing over the A^ experimental points
gives

N + oi'Z^i + P'Y.yn-^y'l^n+2oi^x + 2Pl^y-2y^z +

X

Differentiating equation
partially with respect to
equating to zero gives the three equations

a, P,

and

y in turn,

+ lx + PY.^y-yl^nZ =
Ply'n + i:yn + oil^xy-yY^yz =
-yT.z'+Y^n + OcYXnZn + Plyn^n =

XI

Z^n'

XII
XIII

and XIII for a, P, and y it

sums over the given experimental results.

In order to solve equations XI, XII

to evaluate the nine

and

is

necessary
Inspection

of the given data indicates that there is a much higher level of inert gas in
is considered unHkely, and the result should be rejected.
However, the data is sparse and it has been assumed that an error has been
made which did not affect the proportions of the components present. This
result has therefore been scaled to the same order of inerts as the other
the sixth result. This

readings.

The preliminary

calculations are

shown

in Table 10.1.

Equations XI,

XII and XIII now become

0-9008a+ 0-6673j?- 14-5957+ 2-315

0-6673a+ 0-7370^5- 12-8907+ 2-145
14-595a +12-890^5 -256-37
Solving for

a,

j5,

and

+41-99

=
=

=0

by successive elimination, or by matrix inversion

(see

Section 12.10) gives

2-61,

i5=l-60,

Returning to the original symbols,

Xft

2-61,

X,=

this

means

l-60,

0-393

that

qV/^' =

0-243

Putting these values into equation VIII, the final column in Table 10.1 can
be calculated to show the agreement between the experimental results and
the empirical equation.
It

10%

can be seen that the greatest error in z

error in predicting the rate of reaction.

is

which corresponds to a

368

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

o
oo
r>

OS
op
uo

(N

f^

s
6

uo

vb

vb

i
(N

fsj

VO

uo
oo
'"'

fs

OO

OS
CO

6
VO

i.

CO

r~-

CO
wo

oo

wo
ON
WO

rr

wo

o
OS

WO

OO
(N

CO

OS
'"^

(N

VO

CO

o
fs

CO

"^

OO
CO

p
6>
CO

i
6

'"^

CO
wo

CO

9^

CO
vb
wo

CO
wo
CO

d^
wo

wo

VD
WO

VO

o
6

p
6

as

00

CO

o
6

?
=

o
6

VO
CO

^
-^

CO

Co
OO

wo

3
6

2
6

CO
oo
wo

un

\b

CO
OS
WO

wo
CO

VO
CO

wo

2
CO
6

CO
ro
CO

oo

CO

CO
wo

VO

oo

g
6

wo

?
en

6
in

6
s
'^
6
Tf
Tf

wo

CO

CO

?
"^

wo

rg

fS

00
CO

r-

8
6

'"^

?
^
wo

(N

oo

wo

CO
fs

cs

VO

en

<N

10.

TREATMENT OF EXPERIMENTAL RESULTS

The methods of averages and

definitions of the best curve

which

fits

least

369

squares are based on different

the data. If a desk calculating machine

available, the method of least squares should always be used since it is

more soundly based. Without the aid of a machine, however, the calculation is rather tedious and Hable to arithmetical error and the method of
is

averages, which

is

shorter, should be used to give

10.7.

an adequate

result.

Numerical Integration

sometimes necessary to perform a calculation which involves

For example, the volumetric flow rate of a gas through a duct
can be determined from the linear velocity distribution by evaluating a
suitable integral. If an average value is required for a particular quantity,
For instance, an average surface
it is frequently obtained by integration.
temperature can be determined by integrating the measured temperature
distribution over the surface and dividing by the area. Also, if a representative sample of a process stream is required during a steady state investigation,
it is much more rehable to use a composite sample than a single sample.
The intervals at which the parts of the composite sample are taken, and the
proportions in which they should be mixed, will be determined theoretically
in the following work. Occasionally, a difficult integral arises in a theoretical
prediction, and this has to be evaluated numerically for each individual case.
Many of the problems mentioned above are treated graphically, by
plotting the variables against one another and evaluating the area enclosed
by two ordinates and the parts of the abscissa and curve which join the
ordinates. In this section, numerical methods for calculating this area will
be presented as an alternative to the graphical method.
One way of integrating a set of data is by fitting an empirical equation
to the points, using one of the methods from the previous section, and then
integrating the equation analytically. Unless the empirical equation is needed
for some other purpose, this procedure is not usually worth the effort. All of
the methods to be described involve fitting a polynomial to a set of points
so that the polynomial passes through the points. Since the points may be
subject to experimental error, the curve obtained in this way will not be a
It

is

integration.

Fig. 10.6.

Polynomial

fit

for integration

MATHEMATICAL METHODS

370

good representation of the

IN

CHEMICAL ENGINEERING

and would

data,

certainly give false results for a

method is quite satisfactory.

The above statement is illustrated in Fig. 10.6 where a cubic has been fitted
to four points. For general purposes, the straight line is as good a fit to the
four points as is justified by their accuracy; and if the slope of the curve is
required, the cubic fit would give misleading results. It is quite clear, however,
derivative; but for purposes of integration the

approximately the same. Although the

fit because it passes through
the points. Thus, polynomials are used to fit data for integration purposes.

that the area beneath both curves

cubic

is

10.7.1.

The

the

more complex curve,

is

the easier to

it is

The Trapezium Rule

simplest polynomial, consisting of two terms,

is

the linear equation

ao + a^x

(10.51)

only contains two parameters (^o ^^^ ^i) it gives the equation of the
straight hne joining two points. Taking two points (a'i, y^) and (xj, yi),
Gq and a^ can be evaluated as
Since

it

an

= X2yi-xiy2

Integrating equation (10.51) with respect to

yi-yi
(10.52)

x between x

x^ and

X2

gives

= jydx =
=

[aoX + ia,x^Y,]

ao{x2-Xi) + ia^(x2-Xi)(x2

Using the values for Aq ^^d a^ given

I

+ x^)

(10.53)

in equations (10.52) gives

= X2yi-xiy2+i(x2y2-x2yi+xiy2-xiyi)

= i(.x2y2+x2yi-xiy2-xiyi)
= i(>'2 + 3'i)x(^2-^i)
= average height x width

(10.54)

The value of
the

the integral is thus proved to be the arithmetic average of

two ordinates multiplied by the distance between them. Figure 10.7

illustrates

a repeated use of equation (10.54).

It

can be seen that

Ax

Fig. 10.7.

The trapezium

rule

if

the curve

10.

TREATMENT OF EXPERIMENTAL RESULTS

371

convex upwards, the trapezium rule under-estimates the area by neglecting

Conversely, if the curve is convex downwards, the
trapezium rule over-estimates the area. The error can be reduced by subis

the segments shown.

dividing the intervals

still

further.

If the intervals Ajc are all equal,

and there are four of them, the

integral

can be written
I

= jydx = 2Ax(y,+2y2 + 2y^ + 2y^+ys)

Xl

= i(yi+2y2 + 2y^ + 2y^ + ys)x(xs-x,)

= average height x width

(10.55)

In this combined formula, the average height is no longer a plain arithmetic

average, but twice as much weight is given to the interior points as to the end
points.

10.7.2.

Simpson's Rule

The next polynomial which contains

y

and

this

can be

three terms

is

the quadratic

ao + aiX-{-a2X^

fitted to three points.

However,

(10.56)
it

will

now

be shown that

all

cubic equations of the type

ao + a^x + a2X^ + a3X^

(10.57)

which pass through any three chosen points at equally spaced values of x,
all have the same area beneath them between the end ordinates. The three
ordinates and the area required are illustrated in Fig. 10.8, where h is the
increment of x.

Yn

y,

Fig. 10.8.

The algebra

Simpson's rule

in the following derivation is simpler if the variable

is

changed to a new variable z according to the formula

z

(x-X2)lh

(10.58)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

372

which transforms the required integral as follows

*3

ydx =

j ydz

(10.59)

-1

Xl

yi, y2, and y^ are now at z = 1,0, and 1.

The polynomial (10.57) can be transformed using (10.58) to a polynomial
of the same order in z, thus

and the three ordinates

where the

bs are functions

bo

+ b^z + b2Z^ + b2Z^

of the

as.

(10.60)

Following the same method as for

the trapezium rule, the three values of z are substituted into equation (10.60)
with the corresponding ordinates. Thus

y,

= bo-b, + b2-b2

(10.61)

yi

bo

(10.62)

y^

bo

+ b, + b2 + b^

(10.63)

Substituting equation (10.60) into equation (10.59) and evaluating the integral
gives
1

(bo

+ b,z + b2Z^ + b^z^)dz

-1

==h[boZ + ib,z'+ib2z'-hib,Z^]l,

=
Adding equations

h(2bo + ib2)

(10.61)

and

(10.64)

(10.63) together gives

y,+y3 = 2bo + 2b2

(10.65)

and using equation (10.62) shows that

yi-2y2 + y3 = 2b2

(10.66)

Substituting equations (10.62) and (10.66) into (10.64) gives

I

h[2y2

+ i(y,-2y2 + y,)']

= i(^i+4j2 + y3)x2/i
= average height x width

(10.67)

The average height

in this case is found by adding the end values to four

times the central value, and dividing the result by six.
The same result is obtained for all cubics through the three points because
the step from equations (10.61) and (10.63) to equation (10.65) eliminates

two parameters

(b^ and b^) instead of one, which is normally the case.

range of integration is subdivided into equal intervals by using any
odd number of ordinates (say 7), Simpson's rule can be applied to each
group of three points and the result of all integrations added together.

If the

TREATMENT OF EXPERIMENTAL RESULTS

10.

373

Thus
X7

= jydx =

j'^(y,+4y2 + 2y2 + 4y^ + 2y,

+ 4ye + y,)x(x,-x,)

(10.68)

average height x width

The average height in equation (10.68) is found by adding four times the
even ordinates, and twice the interior odd ordinates, to the end ordinates;
the sum being divided by the total number of times an ordinate has been
used.

Formula (10.68) gives twice as much weight to some of the interior points
than to others, and is not really satisfactory for this reason. It imphes that
some of the interior points should be measured more accurately than others.
Nevertheless, the results obtained from the use of the formula are quite
accurate

if sufficient

sub-intervals are used.

The methods used above can be extended by fitting polynomials of higher

degree to a greater number of points, but this is inconvenient. The increased
complexity arising from the greater number of points is not justified by a
significant improvement in accuracy. A slightly different approach, due to
Gauss, enables a polynomial of degree (2nl) to be fitted to n points,
resulting in a much more efficient process.
10.7.3.

Gauss'

Method

The derivation of

the formula for fitting a quintic equation to three

points will be described below in order to illustrate the method.

always possible to transform an integral over a finite range to an

between the limits 1 and 1, as shown in the derivation of Simpson's
In this derivation, it will be assumed that this change of variable has

It is

integral
rule.

been made by considering the integral

1

I=jydx = 2y

(10.69)

-1

which defines y^, the average height.

Assuming

that the quintic

ao + l^ + ^2^^

+ 3^^ + ^4^'^ + 5^^

(10.70)

adequately represents the shape of the curve y ws x between x = I and

1, an expression for y^ can be obtained by substituting equation (10.70)
into equation (10.69) and integrating. Thus

X =

or

ym

cio

+ ia2 +

(10.71)

Referring back to the derivation of Simpson's rule, equations (10.61,

63) contained four unknowns and could not be solved completely.
However, the coefficients bo and Z>2 could be evaluated due to b^^ and b^

62,

M.M.C.E.

13

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

374

being eliminated together.

It will

now

be shown that in Gauss' method, the

three coefficients Qq, 02, a^ occurring in equation (10.71) can be evaluated

from only three ordinates, if three special values of the abscissa are chosen.

Denoting these three values by x^, ^2, X3, and the corresponding ordinates
by y^, y2, y-^, each pair of values must satisfy equation (10.70). Hence

=
=
=

^1

yi
yi
It is

now assumed

in the

+ ^l^l + ^2^1 + ^3^1+^4^1+^5^1

(10.72)

+ <3l^2 + ^2^2 + ^3^2 + ^4^2 + ^5^2

0 + ^1^3 + ^2^3 + '^3^3 + ^4^3 + ^5^3

(10.73)

flfo

ao

that the above three equations can be

proportions K^

eliminated,

and

K2

(10.74)

added together

K^ so

that a^, a^, ^5, are simultaneously

the coefficients of ^o, 02, a^ correspond with equation
:

(10.71).

Thus,
J^i

3^1

+ ^2 J2 + ^3 J^3 =

(^1

+ ^2 + K3K +
+ (XiXi+ 1^2X2+ 1^3X3)^1
+ {K^xl + K2xl + K:,xl)a2
+ {K^xl-\-K2xl + K^xl)a^

H^^A-^K2xt + K^xt)a^
+ {K^A + K2xl-K^xl)a,

(10.75)

and hence

K.-^X-^

-\-

Ki+X2 + K3 =
K.2X2 4" -^^3 X3 =

KiXi + K2X2 + K2X;^ = ^

iCiXi-\- K.2X2

(10.76)

+ TC3 X3 =

K1X1+K2X2 + K2X2 = J
Kixl + K2xl + K^xl=0
(10.76)

is

a set of

six

equations in the six unknowns x^, ^2, X3, Kj^, K2, K2,
is tedious and proceeds according to

and can thus be solved. The solution

the following pattern.

(iii)

EUminate K^ by multiplying each equation by x^ and subtracting

it from the following equation.
EHminate K2 from the new set of five equations by multiplying each
by X2 and subtracting from the following equation.
Eliminate ^^3 in a similar manner.

(iv)

The

(i)

(ii)

three remaining equations show that Xj, X2, X3 can be interpreted as the roots of the cubic equation

5x3-3x =

(10.77

TREATMENT OF EXPERIMENTAL RESULTS

10.

= - Xj = V

Therefore X3

(v)

and X2

/5,

375

0.

Substituting these values back into various parts of the calculation

(vi)

it

can be shown that K2

4/9,

and K^

= K^ =

5/18.

Equations (10.71) and (10.75) are identical, and hence

1

5>;i

ydx =

+ 8>;2
+ 5>;3

(10.78)

where ji, J2, ys are evaluated at x = 0-7746, 0, and 0-7746 respectively.

Gauss' two-point method is seldom used since it shows Httle advantage
over Simpson's rule, but Gauss' four-point method should be used for
improved accuracy. The formulae are
1

i j ydx

= iiyi+y2)

(10.79)

-1

where y^ and yj are evaluated

at

0-5773; and

ydx = 0-1139(y,+yd + 0-326\(y2 + y3)

(10.80)

-1

where y^, y2, yz,

are evaluated at

ya.

x = -0-8611, 0-3400, 0-3400 and

0-8611 respectively.
4

Example, Evaluate

(1

+x^)~^ dx by using

the following methods.

(a)

Analytical

(b)

Trapezium
Trapezium

(c)

(d)
(e)
(f)

(g)

rule (3 points)
rule (9 points)

Simpson's rule
Simpson's rule

(3 points)

(9 points)

Gauss 3 point.
Gauss 4 point.

Compare

all other results with the analytical solution to determine their

accuracy in representing the above function.

Solution
(a)

Put X

sinh

z,

and hence dx

4
.-.

/=

f/.

\{l
J

cosh z dz.
a

coshzJz

2x-i
^dx= r\-j- 2-;
+ x^)
JV(H-smh2z)
T

= [z]S
= sinh-M
= 2-0947
132

MATHEMATICAL METHODS

376

(b)

(c)

(d)

From

From

1+^2

00

100

0-5

1-25

10

200

1-5

3-25

20

2-5

7-25

3
3-5

10 00
13-25

40

17 00

CHEMICAL ENGINEERING

>'=:(l+x2)-i

00

00000

0-89445
0-70711
0-55475
0-44722
0-37138
0-31623
0-27473
0-24254

the trapezium rule for three points,

1-00000 + 2(0-44722) + 0-24254

2-1369

the trapezium rule for nine points,

i(l'00000 + 0-24254) + i(0-89445 + 0-70711

2-0936

+ 0-55475 +
+ 0-44722 + 0-37138 + 0-31623 + 0-274

3)

Simpson's rule for three points (10.67) gives

_
~
=
(e)

IN

2
3

(1-00000+ 1-78888 + 0-24254)

2-0209

Simpson's rule for nine points gives

iVO'i

i(l -00000 + 0-24254) + KO-7071 1

2-0941

+ 4>'2 + 2>^3 + 4>^4 + 2>^5 + 4j^6 + 2>'7 + 4>'8 + >'9) X 4

+ 0-44722 + 0-3 1623) +
+ 1(0-89445 + 0-55475 + 0-37138 + 0-27473)

In Gauss' three point method (10.78), if the range of integration is

and 0-7746. Therefore, if
1, then the ordinates are required at
the range of integration is
to 4, then ordinates are required at x = 2 and
X = 22 (0-7746) = 3.5492 and 0-4508.
(f)

-1

to

=
=

A[5(0-91 165 + 0-271 18) + 8(0-44722)]

2-1093

10.

(g)

TREATMENT OF EXPERIMENTAL RESULTS

377

In Gauss' four point method, the ordinates will be required at x =

and 3-7222. Equation (10.80) then gives the average

0-2778,, 1-3200, 2-6800,

height within the range of integration.

/

=
=

4[0-1739(0-96351

+ 0-25945) + 0-3261(0-60386 + 0-34959)]

2-0944

The above

results

can be compared in Table

Table

10,2.

Comparison of Integration Formulae

Method
Trapezium

Result

rule (3 pts)

Simpson's

(9 pts)
(3 pts)

,,

(9 pts)

Gauss 3 point
Gauss 4 point

It

Absolute

Relative
error

error

-00011

2-0209
2-0941

-00006

(%)

21
005

00422

21369
20936

00738

3-5

21093

00146

0-03
0-7

2 0944

-00003

0-01

can be seen that Gauss' method

that Gauss' four point

10.2.

method

is

is

the superior one for three points, and

better than either of the other nine point

methods. The reason why Simpson's rule is so poor for three points is illustrated in Fig. 10.9, where the true curve and the parabolic fit can be compared-

FiG. 10.9.

Parabolic approximation for Simpson's rule

The cause of the bad fit is the point of inflection

be represented by a quadratic equation.
10.7.4.

at

0-7071 which cannot

Comparison of Methods

In most cases, the four-point method of Gauss gives an accuracy equivaof Simpson's rule with nine points equally spaced across the
same range. Gauss' method is thus more economical in the use of data, but
lent to the use

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

378

the calculations are a

little

more complicated because of

the

more

difficult

weighting factors.
Gauss' method should have many applications to the investigation of
the behaviour of full scale production equipment, where it is necessary to
compress all readings into as short a time as possible due to plant fluctuareduction in the number of readings required without loss of
tions.
accuracy is an obvious advantage. Another example is obtaining a representative sample of a steady state process stream. It is much more efficient

Gauss intervals, mix them in the Gauss proportions and analyze the composite sample; than to take twice as many
samples at equal intervals and analyze each one separately or mix them all
to take a set of samples at the

together in equal proportions.

An

interesting point has

been noted by McDermottj concerning the

integration of experimental results. Figure 10.10 illustrates the true relation-

ship between

two variables x and y as the

Fig. 10.10.

straight

Comparison of the trapezium

rule

Hne AB. The two groups

and Simpson's

rule

of three equally spaced experimental points (a), (b) have been chosen so
that the total error in each group is zero, and these are the only two possible
groups which satisfy this condition (except for their mirror images in the
line AB). It is obvious from the group (a) that the parabolic fit seriously
under-estimates the area, whilst the two straight fines joining the points
give a closer approximation. For the group (b), the difference between the
straight fine area and the parabolic area is smafi, but does show a marginal
advantage in favour of the trapezium rule. It is thus apparent that the
trapezium rule is more satisfactory than Simpson's rule. This conclusion
is in direct contrast to the result demonstrated in Table 10.2, where the
higher polynomials gave the better results.
These two conflicting conclusions can be reconciled as foUows. For
evaluating analytical integrals or integrating accurate experimental data,
the methods of Gauss and Simpson's rule should be used; but if the data are
t McDermott, C.
Birmingham (1962).

M.Sc. Thesis, Chemical Engineering Department, University of

10.

TREATMENT OF EXPERIMENTAL RESULTS

379

The question of
accurate the data must be to justify the use of Gauss' method or
Simpson's rule is a very difficult one, and the answer must depend upon the
For linear relationships the trapezium rule
type of curve being fitted.
inherently inaccurate, the trapezium rule should be used.

how

appears to be the best for all experimental data, and the simpHcity of the
However, this statement does not
is a further recommendation.
detract from the value of Gauss' method or Simpson's rule for integrating
accurate experimental or theoretical data; nor does it invahdate a planned
experiment to sample a process at the Gauss intervals.

method

Chapter

NUMERICAL METHODS
Introduction

11.1.

There are many problems in mathematics for which no analytical solution

is known. There are also others, for which the analytical solution is tedious
and the answer may be in the form of an infinite series that can only be
interpreted after much computational effort. A numerical method may be
the only one which will yield a solution to the first kind of problem, and may
be the most efficient method of solving the second kind of problem. Mistakes
are difficult to locate in some analytical solutions and the self-checking
features of numerical methods based on successive approximation eliminate
this difficulty.

Simpson's rule and Gauss' method, which were described in Chapter 10,
can be interpreted as numerical solutions of the first order differential
equation

'=m

(11.1)

These solutions are valuable if /(x) is not a simple function.

The purpose of this present chapter is to present numerical solutions for
more compHcated types of ordinary differential equations, to locate roots of
algebraic and transcendental equations, and solve partial differential equations
numerically.

Order Ordinary Differential Equations

many first order differential equations were solved, but

First

11.2.

In Section 2.3

if

not hnear and the variables will not separate, none of the
methods given can be applied. The problem is thus to solve
the equation

is

(11-2)

j^=fi^,y)
ax

There are two methods available, one due to Picard which

the other due to Runge and Kutta which is numerical.
11.2.1.

Picard' s

It is first

at
is

is

algebraic,

and

Method

assumed that a boundary condition of the type

x

y=^b

a,

available to particularize the solution of equation (11.2).

380

(11.3)

As a

first

11.

NUMERICAL METHODS

381

approximation, in the neighbourhood of the starting point, y can be replaced

by b on the right-hand side of equation (11.2). Thus

-j--=f(x,b)
Equation

(11.4)

can

now be

(11.4)

integrated with respect to

/'^

x giving

jf(x,b)dx

which determines y^^ as a function of

(11.5)

now substituted into the

right-hand side of equation (11.2), a further solution y^^^ can be obtained.
If y^^^ is

x.

Thus
jfix,/'')dx

(11.6)

This process can be continued indefinitely, by putting the latest version of

y"^ in the right-hand side of equation (11.2) to determine y^"'^^\ Thus

/"+!)

J/(x,y"))Jx

(11.7)

After each integration (11.7), equation (11.3) must be used to evaluate the
constant of integration.

Example

1.

If

>^

when x =

dy

find the value of

is satisfied,

Putting

>'

and the equation

1,

x^-y
J

dx

y when x

in the right-hand side

2.

of equation

gives

d_y^_:^-\
dx
/i)

Integrating,

Ix

\dx

= ix^-lnx +
But y^^

when x =

Ci

II

= i + i^'-lnx

III

1,

Substituting y^^^ into the

.-.

C,

.'.

y^^

right-hand side of equation

dj^ _

dx

J.

2x

Inx

Inx

2x

I gives

Integrating,
^(2)

^i^^2_^xnx + Kin xf + C2

IV

382

MATHEMATICAL METHODS N CHEMICAL ENGINEERING

But /^^ =

when X =

I,

y^^^

:.

i-\-ix'

-iInx +

i(l

nx^

further cycle of calculation gives

/2>
Putting

X =

= | + |x2-Jln x +

i(lnx)'--idnxf

VI

2 in each approximation gives

/!)

1-807

1-644
1-670
It is

apparent from these figures that a few further cycles of calculation are
lies within the range l-644-^ 1-670; nearer

necessary, but that the true answer

to the latter figure.

An
is

analytical solution

linear.

so that at

available for this

is

problem because equation

Solving by the method of Section 2.3.4 gives

X =

= ix^+ix-^

VII

2,

1-667

VIII

can be demonstrated that Picard's method is converging towards the

analytical solution in this particular case by expanding equation VII into
a logarithmic series. If x = e\
It

x-'

e-'

= '-'" 21-

SI-"'"

^(Inx)^

= l-lnx + -^^
^

(Inx)^^

^+

...

Using equation VII, therefore

y

Comparing

ix' + i-i^rix

i(\nx)^-}(\nx)'

+ ...

IX

V, and VI with the corresponding

can be seen that the set of approximate
solutions y"^ approaches the true solution as n increases.
Picard's method has two disadvantages.
Firstly, as successive substitutions are made the amount of work is increased and the integrations
become more complex; and secondly, integrals frequently arise which cannot
be evaluated analytically and these have to be solved by a numerical method.
It is reasonable therefore to use a numerical method from the very beginning.
coefficients in equations III,

coefficients in equation IX,

11.2.2.

it

The Runge-Kutta Method

The method

to be described is a composite one, using the principles

discovered separately by Runge and by Kutta. The analytical details of the
modem method differ from the original development and the algebra is

II.

NUMERICAL METHODS

383

rather tedious; however, the general principles can be outlined by reference

Assuming that equation (11.2) is satisfied, and an initial conyo 2it X = Xq) is given, it is desired to find the value of y when
XQ + h where h is some given constant.

to Fig. 11.1.
dition (y

iQ+h

-I
Fig. 11.1.

Third order Runge-Kutta process

The general solution of equation (11.2) will consist of a family of curves,

each curve having a particular value of the constant of integration. Three
specially chosen curves of this family (a, b, c) are illustrated in Fig. 11.1;
the curve b passing through the point (xq, yo). The gradient of the tangent
to the curve at (xq, yo) can be calculated from equation (11.2) thus determining the length Ar^ by
k,

hf(xo.yo)

(11.8)

yo + k^ is obviously a poor approximation to the desired answer unless the

curve happens to be a straight line, and a better approximation can be

obtained from the following reasoning. The gradient of the chord to the
curve between Xq and Xq + H should be approximately equal to the gradient
of the tangent to the curve at Xq + p. Unfortunately, at this stage of the
calculation, the position of the curve at Xo + ih is not known; however, an
approximate value for y at x = Xq + ^H is given by y = >^o + 2^i* This
point (xq + P, ^^0 + 2^i) liss on a different curve of the family as illustrated
by c in Fig. 11.1. The gradient of the tangent to curve c at (xq + ^h, j'o + i^i)
can be calculated from equation (11.2) and the length k2 constructed by
drawing a line parallel to this tangent through (xq, yo). Thus^
k2

hf(xo + iKyo

+ ik,)

(11.9)

For the type of curve illustrated, k2 is also an underestimate. The method

of obtaining an even closer estimate for the end ordinate is to find some
average gradient of three tangents, one at the start of the interval, one in

384

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

the centre,

and one

All three tangents should be taken to the

not possible. The tangents so far, are to the correct

at the end.

correct curve b but this

is

curve b, and to the curve c which lies below b. To compensate for the effect
of the latter tangent, the final tangent must be taken to a curve lying above
b. On the ordinate at ;c = Xq + H, y = ;^o + ^2 is approximately on curve b,

>'o

+ ^i

is

below

definitely

j
should be above

b.

b,

hence

>'o

+ /c2 + (/c2-/ci)

Finding the gradient of the tangent to the curve a which

passes through this point from equation (11.2) gives

^3

= V(^o + ^,yo + 2/c2-/ci)

can be shown analytically that the ordinate dX x

through (xo, yo) is given by
It

>^o

(11.10)

XQ

+h

to the curve

+ i(/ci+4/c2 + /c3)

(11.11)

where k^, k2, k^ are given by the equations:

k,

hf(xo.yo)

k2

hf{xo

+ ih,yo + ik,)

k3

hf(xo

+ h,yo+2k2-k,)

(11.8)
(11.9)

(11.10)

"
This formula (11.11) is known as the " third order Runge-Kutta formula
because the term corresponding to the third derivative term in the Taylor
series for

To

y expanded about

illustrate the use

given in example

Example

is

2.

(xq, yo) is correct.

it will

of this method,

be applied to the problem

1.

If

>^

when x =

and the equation

\,

dy

x^ y

dx

value of y when x = 2.
Using the above symbols, h = \; hence equation

satisfied, find the

/ci=/(l,l)

Equation (11.9)

now

(11.8) gives
II

gives

fc2=/(li.l)

(9/4) -1

3/2

-^
~ 6
Putting equations

II

and

III

III into (11.10) gives

4-2i

IV

NUMERICAL METHODS

11.

Substituting equations

II, III

and IV into

385

(11.11) gives the

answer

= l + i(0 + 3Kf)
= 1J

which is identical with the analytical solution given by equation VIII in

example 1.
If the range of integration is large, it can be subdivided just as for
Simpson's rule, with consequent increase of accuracy. This will be illustrated
by an example in the next sub-section.
The method of Picard, and the original formulae of Runge and Kutta
are given by Piaggio,t together with the theoretical justification of each
method.
11.3.

Higher Order Differential Equations

(Initial Value Type)

It is a simple matter to convert an nth order differential equation to n

simultaneous first order equations. It is only necessary to define ( 1) new

variables

by
^

dy

"^

^2

^x
...2

(11.12)

etc.

W._i

dx n-1

and remove all derivatives from the original differential equation except
which is replaced by (dw_ildx).
The solution of the set of equations (11.12) with the original differential

(d"yldx")

equation is now very similar to the solution of a single first order equation,
provided that the boundary conditions are of initial value type (Section
8.4.5). In this case, at some value x = Xq, values for y, w^, W2
w_i,
will be available and an extension of the Runge-Kutta process can be used.
The formulae for the second order differential equation
.

g..(,2)
are as follows.

Putting

dy
dx
equation (11.13) becomes

^ = F{x,y,w)
t Piaggio,

H. T. H. " Differential Equations

".

G. Bell

(11.15)

&

Sons Ltd., London (1928).

MATHEMATICAL METHODS

386

IN

CHEMICAL ENGINEERING

From the initial values (xq, yQ, Wq) and the interval
be defined as before. Thus
k,

Xi =

(h) in x, k^

and K^ can

/zwo

(11.16)

/iF(xo,yo,Wo)

(11.17)

where k^ is the first approximation to the change in y and K^ is the corresponding change in w. Proceeding in a similar manner to the first order case,
k2

h{wo + \K,)

(11.18)

K2 = hF(xo + ih,yo + ikWo + iK,)

k,

X3 =

(11.19)

h(wo + 2K2-K,)

(11.20)

hF{xo + h,yo + 2k2-k,,Wo + 2K2-K,)

(11.21)

The values of y and w

at the

end of the interval are given by

y = yo + i(ki+4k2 + k,)
w = Wo+iiK,+4K2 + K^)

Example

and the

initial

value of jv
Solution.

1.

Given the

diiferential

(11.23)

equation

x =

conditions that at

when x =

(11.22)

y =

0,

and dyldx

0,

find the

1.

Putting

dy
I

= \, Xq = 0, yg
used in succession. Thus
Since h

fci

x^vv 2>;^

Jx

II

Wq

0,

equations (11.16) to (11 .21 ) can be

^1 =
kz

1,

^w

III

l(3^)=

-0.2500

= 1(0-0-1250)=

K2 _

-0-1250

IV

r(0-5)^(-0-125)-2(l)^-|
8

= -0-25391
2k,- k. = -0-2500
2K,-K, = -0-25782
fca = -0-25782

VI
VII
VIII

IX

NUMERICAL METHODS

11.

i^3

(1)2(- 0-25782) -2(0-7500)2

-0-17285

Equation (11.22)

now

-,

gives the solution

y
.-.

387

= 1- -1(0 + 0-50000 + 0-25782)

= 0-8737

XI

between k^, k2, and k^ in this example

There is
and the above solution may not be very accurate. To check this, the integration can be performed in two stages by choosing h = ^ and applying the
formulae (11.16) to (11.23) twice. The intermediate and final results are
a large relative variation

at

and

at

=
=

0-5,

1-0,

>;

= 0-9687,
= 0-8779

w = -0-1230

XU

The answers given by equations XI and XII agree to better than

and the solution can be written
y

%,

0-878

with reasonable certainty that the third significant figure

of the correct solution.

is

within two units

Direct Use of Taylor's Theorem

11.3.1.

The following method has been rendered almost obsolete by the

increased availability of digital computers, but for manual calculations with
a desk machine it can still give accurate answers quite efficiently.
Taylor's theorem (Section 3.3.7) states that

f(x,

+ h)

=f(xo) + hfXxo) + ihY(xo)+

...

^^/"\xo) +

...

(11.24)

can be differentiated indefinitely. For a second order differential

equation of the form (11.13), with the initial values of >; and dy/dx given at
X = Xq, all terms in equation (11.24) can be found as follows.
The boundary conditions specify /(xq) and f'(xo), and the differential
equation (11.13) can be used to calculate /"(-x^o)- By differentiating equation
(11.13) with respect to x, an equation can be developed giving /^^^(xq) in
terms of Xq, /(xq), /'(^o) and /'(xq).
Hence P"\xo) can be found by
differentiating equation (11.13) n-2 times.
Equation (11.24) thus gives a
series solution to the non-linear differential equation (11.13). The method
will be illustrated by repeating example 1.

if fix)

Example

and the
of y

2.

initial

when x =

Given the

differential

conditions that at
1.

equation

0,

>^

and dyjdx

0, find the

value

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

388
Solution

At X

The

given that

0, it is

differential

y'

II

equation can be rewritten in the form

8/'

Using equations

and

II,

III

therefore

/'= -0-2500

IV

^y^^'^^x^f + lxy'-Ayy'

at

= x'>;'-2/

0,

Differentiating equation III gives

atx =

.-.

Differentiating equation
8y(^)

8/5)

=
.-.

gives

+ Axy" + 2y' - Ayy" - A{y'f

atx = 0,
/^> = 0-1250

+ 6x/^^ + 6/-4;;/^>-12>;V'
/^>= -0-1875
atx = 0,

;cV'^^

^ xV^> + 8x/''> + 12/3>-4>'/''>-16//^^-12(/)'

/^^= -0-1562
atx = 0,
.-.

Substituting into equation (11.24) with h

y
.-.

VI

VII
VIII

manner,

in this

8^6)

x^y^^^
.-.

Continuing

/2^

0,

IX

X
XI
XII

gives

= 1-0000-01250 + 0-0052-0-0016-0-0002
= 0-8784

XIII

The direct use of Taylor's theorem is apparently simpler than the RungeKutta method in this application, because both calculations have been done
manually. However, the successive differentiations performed above cannot
be done by a digital computer and a more complicated problem would favour
the use of a computer programmed for the Runge-Kutta method using

many

sub-intervals of integration.

11.4.

Higher Order Differential Equations

(Boundary Value Type)

With second and higher order differential equations, the boundary

may be specified at two different values of the independent
variable.
In such cases, there is insufficient data available at x = Xq, for
either of the above methods to be used directly. However, a trial and error
method based on either of the above methods is practicable for second
conditions

order equations.

Thus, consider the equation

NUMERICAL METHODS

11.

again, with the

boundary conditions:

at

and

389

at

= Xo,
= x^,

yo

(11.25)

yN

(11.26)

N integration

has been assumed that

steps are necessary to cover

a value is assumed for /(xo),
equation (11.13) can be solved as an initial value problem, thus generating
a value for y at x = x^^. This value is unhkely to equal yff, therefore a
second choice is made for j'(^o) resulting in a second value for y slI x = x^^.
The value of y at x = x^^ can thus be evaluated as a function of y'(xQ), and
the correct value of yXxo) can be found by using an interpolation formula
from Chapter 10. The value of >^'(^o) obtained in this way can then be
checked by a final solution of an initial value problem.
If this method is adopted, the early trials are usually made with a small
value of A^ and widely spaced values of >''(^o)- As the solution is refined, the
value of iVis increased and the assumed values of >^'(-^o) ^re contained within
a narrower band. The calculated internal point values then give the shape
of the solution curve {y vs x).

where

it

the distance between the boundaries.

If

Example 1. In example 4, Section 2.4.3, the problem of cooHng a graphite

was considered, and the problem was stated in mathematical
terms as follows. Find the rate of flow of heat into the water cooler at
X = 1 ft when the temperature distribution satisfies
electrode

where Tq

70,

/cq

88-9, a

The boundary conditions

0-0180,

and p

2-40.

are

at

0,

T=

2700

II

at

l,

T=

300

III

and the

rate of flow of heat into the cooler

H=

is

given by

dT

-16-4-

ax

IV

Solution

Since dT/dx is required at x

dent variable by putting

it is

convenient to change the indepen-

= l-x

(^o-aT)^-a^^y-^(T-To) =

VI

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

390

and the boundary conditions become,

z

at

and

at

=
=

0,

r=

300

VII

1,

T=

2700

VIII

From

equations VII and VIII the average temperature gradient is 2400, but
the gradient at x =
should be smaller than this because less heat reaches
the water cooler than leaves the furnace, and the conductivity is higher at
the lower temperature. Therefore, as a first approximation, put

r = 2000
7"
dz

Putting

atz

IX

>v

equation VI becomes

dw

oiw^

+ P(T-To)

XI

Uq-olT

dz

Taking one interval and using the method of Section

values can be calculated.

11.3, the following

= 2000
= 2434

Xi =
^2 =

1673

2A:2-/ci=2868

1K2-K^=

2477

11 543

k^
A:2

= 4477

^3
.-.

T=

.-.

w=

1^3

868-9

300 +-^(2000 + 9736+ 4477)

2000 + i(869 + 6692 + 11 543)

=3002

5184

XII
XIII

The answer given by equation XIII is suspect because K^ is so different from

Ki and K2* It appears that r'(0) has been chosen 10% too large by comparing equations VIII and XII. Therefore T\0) is reduced to 1800 for the
second trial, and the interval is halved to reduce the discrepancy between
A'3

and K2.

The

calculation for the

.-.

/.

For the second

first

interval gives

k^= 900
iCi = 352-5
k2= 988
1^2 = 476-9
k^ = 1201
X3 = 834-1
=1309
T(i)= 300 + ^(900+3952 + 1201)
w(i) = 1800 + -1(352-5 + 1907-6+ 834-1) = 2316

XIV

XV

interval,
/ci

/C2

k^
.-.

T(l)

.-.

w(l)

=
=
=

1158

Ki= 761-5

1348

K2 =
X3 =

2009

1232

3948

= 1309 + ^(1158 + 5392 + 2009) = 2735

= 2316 + 1(762+4928 + 3948) =3922

XVI
XVII

NUMERICAL METHODS

11.

391

XVI shows that the temperature drop across the electrode is

2435 compared with the true value 2400. Hence, the assumed value
too high. Therefore, putting dTjdx = - 1770
r(0) = 1800 is probably
at ;c = 1 into equation IV gives the result
Equation

now

U%

if

29000Btu/h

XVIII

considered to be sufficiently accurate for the purpose of

could be improved further by taking more intervals and
choosing T'(0) more accurately, but this is not justified by the accuracy of

This answer

the problem.

is

It

the original data.

Finite Difference

11.4.1

Methods

solving higher order ordinary differential

An
equations of boundary value type, is to replace the differential equation with
an equivalent finite difference equation. This technique is most useful for
second order equations, or higher order equations replaced by simultaneous
second order equations.
There are many bases for the method, but the derivation from Taylor's
theorem is probably the easiest to follow. The closed range of the independent variable a ^ x ^ b is divided into equal intervals and the points are
labelled x, where Xq = a, and ;c^ = b. The dependent variable (y) can be
expressed in the neighbourhood of any point (x) in terms of x and its
derivatives at x by means of Taylor's theorem (Section 3.3.7), and if h
denotes the increment of x between neighbouring points,
alternative

method of

yn^i =/(x+i)

=fM + hfXx)+ih'r(x) +

...

yn=f(x)
yn-i

(11.27)

(11.28)

=/fc-i) =f(x)-hnx) + ih'fXx)-...

(11.29)

terms in h^ and higher degree can be neglected

relate neighbouring point
values to the derivatives at the central point (x); and they can be solved
to determine the derivatives in terms of the point values. Thus
If /z is

chosen

sufficiently small,

These three equations

in equations (11.27, 29).

fM = yn
/X^n)

(11.30)

^"^'^""/""'

^"--'^;;-^^"=
rfa) =

(11.31)

%:l

(11.32)

These equations illustrate the use of Norlund's operator, as may be seen by

comparing equations (11.32) and (9.37), and they allow any second order
differential equation to be replaced by an equivalent second order difference
equation.

The

calculation details

and treatment of boundary conditions

easily explained with reference to

a particular example.

is

more

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

392

Example 2. The cylindrical combustion chamber of a rocket motor has

coolant ducts of lenticular cross-section at frequent evenly spaced intervals
within the cyhndrical wall. The dimensions of a typical duct are illustrated
in Fig. 11.2. Radiant heat is received uniformly by the internal surface of

0-3251

Fig. 11.2.

Rocket motor cooling duct

the combustion chamber at a rate Q = 51 cals/sec cm^ and this heat is

conducted into the liquid coolant at a bulk temperature T^ = 130C according to a variable heat transfer coefficient h cals/sec cm^C. h can be determined
from the Dittus-Boelter equation expressed for gases as /z = aw*^, where
u is the local centre Une velocity and a = 0-00626 for the coolant used.
Spikinsf has determined the velocity distribution as

"-h(i^)1
the coordinate measured as shown in Fig. 11.2, and Uq, the axial
If the metal conductivity k = 0-107
has a value 100 cm/sec.
cals/sec cm^C cm~^, find the temperature distribution within the chamber

where z

is

velocity,

wall.

Solution

The shape of

the duct is rather compHcated, so it was idealized as shown

There are Hnes of symmetry at AE, DG, and JK, and a typical
section has been straightened out by distorting the junction CKJBF. In the
ideahzed situation, radiant heat enters through AB and BC which is chosen
to equal BJ, negligible heat passes through the external surface CD, and the
coolant flows over EF and F'G. All areas and heat paths are correct in this
model, except for the cjurved path from BF to CF which is only approximately
correct.
Neglecting the temperature drop through the metal wall with
respect to the temperature drop within the coolant film, reduces the problem
to the determination of the metal temperature {T) as a function of a single
in Fig. 11.3.

coordinate

(x).

t Spikins, D. J., Ph.D. Thesis, Chemical Engineering Department, University of

Birmingham, 1958.

NUMERICAL METHODS

1.

393

I
Idealized metal section

Fig. 11.3.

Taking a heat balance over an element of metal

in the section

ABFE

gives

d^T

h ,_

dx^

ks

_.

^^--(T-T,) +

Q
I

ks

or inserting the data

d^T

0-8

0-00626 X 100^-^

0107

dx^

To simpUfy

X 0-1219

[i'llj

the arithmetic,

it is

57

II

0-107 X 0-1219

convenient to put

I'llX

III

e=T-Ts

IV

- 30-8 (l-XT'^ + 7048 =

dX
section BCFT, where no coolant flows

and

:,

Similarly, in the

(T-Ts) +

but heat

is

still

absorbed in a thicker section,

d^e

dX'

Finally, in the section 1-1

<:

4228

VI

2-1,

d'-e

18-5[l-(2-l-Xrp-'0

dX'
In the

finite diff'erence

BC is

ten equal intervals,

VII

AB and CD are each divided into

and the point temperatures are labelled
difference form of equation V can be found

representation,

one

interval,

0 where n = lOX. The finite

from equations (11.30, 31, 32) as

0+i-[2+O-3O8(l-X3)-]0 + 0_i
which

is

valid for

9.

The

finite difference

+ 7O-5 = O

form of equation VII

^^i-[2 + 0-185{l-(2-l-X)T']^n + ^-i =

is vaUd for 12 <
The four points 0, 10,

which

in the following

manner.

assumed that a point n

:^

VIII
is

IX

20.

11, 21 are influenced

Point

= -\

is

exists

by the boundary conditions

symmetry, and it can be

in a plane of

where 0_i

O^.

This ensures that the

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

394
gradient

is

zero at the boundary and the point

point of equation VIII.

Similarly, 622

becomes an ordinary

^20 ^^^ equation

IX

is

valid for

21.

The conditions
to first principles,

at points 10 and 1 1 are rather more

and using equation (11.31),

^9
6
V21SX

6in
10

^11-^10

difficult,

the temperature gradient at n

9i

the temperature gradient at

lOJ

is

is

l-21dX

Heat balance

Fig. 11.4.

at

Taking a heat balance over an element of width

but returning

BF

SX =

0-1 astride point 10

as illustrated in Fig. 11.4 gives

-0-1219/c,,
1-21

,,

-0-2032/c,,

...^.

SX

V21dX
+\{e,o+e,)\'2i8xhL=,.,s

which

simplifies to

O-366609-(?io

similar balance at point

1 1

+ O-6213^ii+26-3 =

XI

gives

0-4976010-^11 +0-4927^12

+ 10-5 =

XII

There are thus ten equations of type VIII, ten of type IX, plus equations

XI and XII which are to be solved for 0 where ^ ^ 21. The simplest
method of solution is to estimate values of 0 for odd values of n and solve
the equations directly for 0 at even values of n.
improved odd values can be calculated.

values,

From
By

the calculated even

repetition, the solution

can be obtained as accurately as required.

Equations VIII and IX are first solved for 0 and the
0+i, 0_i evaluated at each point as shown in Table 11.1.

coefficients of

11.

Table

11.1.

_
4
6
8

10
12
14
16
18

20

On-l

0-8665
0-4336
0-4363
0-4438
0-4601
0-6213
0-4842
0-4690
0-4616
0-4585
0-4577

0-4336
0-4363
0-4438
0-4601
0-3666
0-4842
0-4690
0-4616
0-4585
0-4577

coefficient of

plus

^n+l

On-X

395

Coefficients in the Difference Equations

coefficient of

NUMERICAL METHODS

30-5

30-6
30-8
31-3
32-4
26-3

9
11

13

15

17
19
21

plus

^n+l

0-4333
0-4345
0-4392
0-4504
0-4743
0-4927
0-4752
0-4646
0-4597
0-4579

0-4333
0-4345
0-4392
0-4504
0-4743
0-4976
0-4752
0-4646
0-4597
0-4579
0-9154

30-5
30-6

31-0
31-8
33-4
10-5

shown in Table 11.2, values must be estimated

Neglecting the change of heat transfer coefficient with
velocity, and assuming that 0^ is constant for ^ 10 gives 0 = 227 by
equation VIII; hence 0^ = 250 will be a fair approximation for odd values
and 10. An approximate solution of equation VII by the
of n between
gives
methods of Chapter 2 after neglecting the term (2-1

To

start the calculation

for 6 with n odd.

^^

= 250e-

4-25(X-l-l)

XIII

Evaluating equation XIII at the chosen points completes the estimate of the
temperature distribution.
The equations given in the left-hand half of Table 11.1 can be used to
calculate the values in the second row of Table 11.2. The calculation is thus
started and alternate use of the left- and right-hand halves of Table 11.1
gives the solution.
quicker way to the solution is by making a fresh estimate
at any stage, and this has been done on two rows of Table 1 1.2. The solution
given in the last row was obtained from eight further rows of calculation.
This solution will be fairly accurate in the metal immediately surrounding
the duct, but not very good in the weld section CKJBF. Taking a heat balance
over an element within this section gives

XIV

d^
or

1640

XV

dz'

Solving equation

XV and remembering that z =

-397

is

a plane of symmetry

gives

T = A-820(1 -397-2)2

XVI

'

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

396

-^

vn

r>

v^

*r)

IT)

wo

VO VO

wo

r-

(N

S
On
^^

00

oo

t^

S
o

CO
CO

^
oo

jq

OO
ON
(N

VO

O
ro

m
OO

S
<N

in

00

00
CO

<

Tf

CO

VO
WO

wo

VO
oo

CO

CO

?;

CO

CO
^5
<N
oo

ON

u-> CO
CO CO
<N rN

VO
CO
<N

VO
CO

>n
CO

w-^

wo

wo

wo

CO
CO

wo

s
ON

w^
CO

w-^ VO
CO CO
N

Tf
CO
<N

(N
CO

w-^

ON

00
CO

r^
ro

CO

wo
5
<N 3

5
(N

o
oo
oo

s
(N

?i

CO

00
00

VO
00

s
(N

VO
<N rN

CO

SS 8
CO CO

CO
CO

o
CO

P5
CO

CO

W-i

i
CO
CO
CO

ON <N
eN CO
CO CO

S
CO

!
(N CO

VO
<N

t^
CO
fN

r-

CO

ON

O
>n

as
CO

Tt

S
fN
^

S
<N

VO
<N

a\

CO

<N

?i

r(N
CO

r^i

t^

Tj-

VO

CO CO

CO

VO
VO

o
in

^
CO
co

en
vo

00

CO

oo

VO

VO
oo

00
r-

^
CO

S
CO
o
<N

p-

CO
oo

r-

en

VO

ON
W^

ON
OO

CO
oo

g
o Tt
^S

00

>o

VO
(N

SO

-CI

Si

00

wo
(N

wo

VO
00

w->

VO

ON
CO

CO
t^
wo wo

wo

,__

'^
r--

^
<N

Tt Tt
<N <N

r--

wo
(N

CO

uo

I
a\

^
r4

ON
'^

VO

CO

wo
oo

r-

00

^H

^^
VO

CO

^
00

^^ 1

Tf
<N

n- VO
wo w-^

CO
lo

oo

as

"^
r) tN

CO

o
oo

as

00

^
CO

r4
CO

CO
<N
wo

r~-

o
^

as as

(N

r-^

ON

9.

<N

-H

vo

as

OO

r-

CO
CO

00

JS

<N

?3

as

VO

VO

Tf

(S
CO

^
m

CO

VO

vo
OS

as

CO

vo

>r)

00

VO

VO

wo
ro

WO
ro

CO

"

NUMERICAL METHODS

11.

But the temperature

F (z = 1 -270) is
Ts + e = 130 + K309 + 256) = 412i

at point

T=
A=

.-.

397

XVII

426

have been plotted in Fig. 11.5, which shows that the hottest
from the corner of the coolant channel
is about 1-5
along the heated wall, and not at the centre of the uncooled weld. Because
the temperature drop across the metal wall has been neglected, the results

These

results

mm

part of the duct

500

400

Weld

L__<T---^
Exposed

wall

300

Rear

200

100

wall

^^--^"^
1

0-508

0-254
Fig. 11.5.

6
7
0-792

10

016

Temperature distribution

II

1-270

in cooling

duct

presented should approximate to the average temperature within the metal

at the point in question, but the exposed surface temperature could be about
35C higher than the figure given.
It is possible to modify the calculation procedure with advantage in more
compHcated problems, by working in terms of temperature changes instead
of temperatures. To do this, the first row of estimated temperatures is subtracted from the first row of calculated temperatures at the same points. All
subsequent calculations are then performed to give correction terms, and
the final temperature at each point becomes the sum of the values in that
column. The result obtained in this way should then be checked by one cycle

of the basic calculation in terms of the temperatures.

11.5.

In
is

many problems

it is

readily accomplished

Roots of
are

Algebraic Equations

necessary to solve a quadratic equation, and this

by means of the formula

ax^-{-bx

+ c=:0

_ - feV(fe'-4ac)
""""

(11.33)
(11.34)

2a
Less frequently, a cubic equation needs to be solved, and this can also be

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

398

done

analytically without too

quartic equation

much

difficulty.

An

analytical solution of a

also feasible but not very profitable in

is

most

practical

cases; but polynomial equations of higher degree are insoluble analytically,

hence numerical or graphical methods must be used.

equations such as

tanx
are also insoluble analytically

Transcendental
(11.35)

/cx

and methods must be established

for solving

such equations in particular numerical cases.

Analytical Solution of the Cubic Equation

11.5.1.

A method due to

Tartagha (1545)
ax^

have three
by putting
will

roots, the

the following.

is

The cubic equation

+ 3hx^-V2>cx-\-d =

(11.36)

sum of which will be (3bla), Changing the variable

z

=X+

(11.37)

a
gives the equation

z^

+ 3Hz + G =

(11.38)

which has zero for the sum of its roots, where

a^H = ac-b^
a^G = a^d-3abc + 2b^

and

To

solve equation (11.38), a further substitution

where

t;

is

a function of u which
u^

Choosing

= Hju

is

made. That

is,

= u+v

is still

to be determined.

(11.39)

Hence

+ v^ + 3(uv + H)(u + v) + G =

will simplify

(11.40

equation (11.40) to

M^-H^ + Gw' =
Equation (11.41)

is

(11.41)

a quadratic in u^ which has the solution

u^

= -iGy(G' + 4H^)

(11.42)

= H,

six values
Equation (11.42) determines six values for u and since uv
of V are determined. Hence equation (11.39) appears to give six roots to the
cubic (11.38). However, there are only three distinct values for z because
each pair of values for u and v occurs twice, once a.s u + v and again as
v + u. Equation (11.37) transforms the values of z into the corresponding
values of a: which satisfy equation (11.36).

Example

1.

Solve the equation

x^-4x^ + lx-5 =
Removing

the second term by putting

z

= x-ii

II

NUMERICAL METHODS

399

27z^+45z-ll=0
z = u+v
27w^ + 27i;^ + (8lMi; + 45)(M + t;)-ll =

III

11.

gives

Putting

IV

Assuming that
8lMi;

+ 45 =

-=-'

9w

729m^--297m^-125

.-.

The solution of equation VI

3

"

.-.

From

equation

is

VI

obtained from (11.34) as

297 + 7(88 209 + 364500 )

~
1458
= 0-6652
= 0-8729

V
t;= -0-6364

Hence, from equations IV and

VIII

II,

X
is

VII

one root of the cubic equation

IX

1-5698

I.

The simplest way of finding the other roots

(x 1-5698) algebraically to obtain

is

to divide equation I

by

x2-2-4302x + 3-1851=0
which has the roots

1-2151

+ 2-61431

XI

In general, if the cubic equation (1 1.36) has three real roots, G^ + 4H^<0
and equation (1 1.41) has complex roots. The cube root of equation (1 1.42) is
best evaluated in the

Argand diagram

one real root, as in example

Hence the
cubic equation.
roots.

11.5.2.

method

is

in this case.

When

the cubic has only

the quadratic (11.41) will always have real

most useful for finding a single real root of a
1,

Graphical Location of Roots

all numerical methods it is useful to know an approximate answer to

problem this can then be successively improved. This was the basic technique in example 2 of Section 1 1 .4. 1 where a very approximate temperature
distribution was assumed. It can be seen that a very good first approximation
will yield an accurate solution quickly, but even a poor first approximation
will not extend the subsequent calculation unduly. Graphical methods are
very useful for determining a first approximation. If the graph is drawn
accurately, the graphical solution may be acceptable as the final solution;
but even a rough sketch will give the approximate location of the root of
an equation. Thus consider the following example.

In

the

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

400

Example

2.

Find the approximate location of the roots of the equation

tan X = kx
where k is a. constant.
Equation I can be resolved into two equations

and

>'

tan X

kx

II

III

which must be satisfied simultaneously. Taking a set of values for x, y can

be evaluated from each of equations II and III, and both equations can be
plotted graphically as

shown

in Fig. 11.6.

"*y=tan x

Fig. 11.6.

Graphical solution of tan x

kx

= ^71 and
then repeated indefinitely at intervals
of n along the x axis. Equation III is just the equation of a straight line and
the solutions of equation I are the points of intersection of the two curves.
The gradient of equation II at the origin is unity and therefore two possibilities arise. If k = ki<\, there is only one root on the principal branch;
but if k = k2>l, there are three roots on the principal branch. Altogether
there are an infinite number of roots which occur approximately at odd
multiples of in. As already mentioned, the first few roots near the origin
are a fair distance from the asymptotes and there may be one or three roots
near the origin, depending upon the value of k.
The

principal branch of

j^

tan x has asymptotes at x

passes through the origin. This shape

11.5.3.

is

Improvement of Roots by Successive Approximation

The equation to be solved can always be rearranged, as in the above

example, to give a simple function on at least one side of the equation. After
a graphical investigation to find an approximate root, the value obtained
can be substituted into one side of the equation, and a second value obtained
from the other side of the equation. This process of substituting into one side
and estimating a new value from the other, can be repeated until two

II.

NUMERICAL METHODS

401

consecutive results agree. Only one side of the equation can be used for the
evaluation step because the one side will give a convergent solution whereas

The method is illustrated

the other side gives a divergent solution.
reference to the problem arising in Section 6.9.3., equation VII.
Example

3.

Find the root of equation

15a^-68a^ + 95a-47
which

close to a

is

by

2-5.

when solving cubic equations to eliminate the second

It is
term by using the substitution (11.37). Thus putting
always useful

z
into equation I

4 = a-l-5111

II

and rearranging,
z^

0-5170z + 0-4641

III

side of equation III has been plotted in Fig. 11.7 to show the location
of the root. The mechanism of the numerical method is also illustrated by

Each

0-5I7Z + 0-464I

Fig. 11.7.

the stepped path

ABCD.

Solution of a cubic equation

Starting

from the approximate value z

z^, the

right-hand side of equation III is evaluated to give the ordinate of A. The

point where the curve y = z^ has the same ordinate value at B, determines a
new approximation for the root (z = Z2). Repeating the process gives the
points

and

and a

closer approximation to the root.

It is

obvious from

would
and the subsequent evaluation of z from the right-hand
side of III would be further from the root than the original estimate.
The determination of which side of an equation to substitute into,
depends upon the nature of the two functions in the vicinity of the root.

the diagram that substituting into the left-hand side of equation III
give the point A',

Thus, the left-hand side of equation III is very sensitive to small changes in
z whereas the right-hand side is relatively insensitive. Hence, the approximate value is always put into the insensitive side, and the new root evaluated

from

the sensitive side.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

402

The approximate root a = 2-5 corresponds

Hence the calculation proceeds,

0-5170zi +0-4641
Z2

.-.

0-5170Z2

to z

0-9811

1-0

from equation

II.

zf

0-9937

+ 0-4641 = 0-9778 =
Z3 = 0-9926

zl

.-.

0-5170Z3

+ 0-4641 = 0-9773 =
Z4 = 0-9924

zl

.-.

0-5170Z4 + 0-4641
Z5

.-.

0-5170Z5

0-97717

zl

0-99233

+ 0-4641 = 0-97713 =
zg = 0-99232

zl

.-.

Since Z5 and Zg are approximately equal, z^

Hence, by equation

is

close

enough to the true

root.

II,

2-5034

The idea in the above method is to divide the equation into two parts,
one very sensitive, and the other insensitive. This is why the second term
was removed from the original equation; otherwise it would have increased
the sensitivity of the right-hand side of equation

11.5.4.

Newton's Method

In this method, Taylor's theorem

If a root of the equation

is

/(x)
is

III.

required,

used to improve an approximate root.

(11.43)

and x = x^ is an approximate root, the function

expanded in the neighbourhood of Xj thus,

in

equation

11.43) can be

f(x,

+ 3x)=f(x,) + 5xfXx,)

(11.44)

where the higher order terms have been neglected. If X2 = Xi-\-dx is to be

a better approximation to the solution of equation (11.43), then (11.44) gives
/(^i) + (^2-^i)/'(^i)

or

X2

=
(11.45)

x,-f(x,)ir(x,)

Further approximations can be found by repeated apphcation of equation

(11.45).

Thus
^3

= ^2-/fe)//'feX

etc.

This method can be employed to solve non-linear simultaneous equations

and the following is an example of such use.

NUMERICAL METHODS

n.

403

Example 4. In a study of the settling of suspensions of spherical

Ohverf has suggested the correlation

particles,

V,=^Vo{l-kc){i-Kc^l^)
where V^

is

the setthng velocity of the suspension,

Vq

is

the Stokes' velocity of a single particle,

is

the

A:,

From

volume concentration of the sohds.

K are empirical constants.

the following data given in the paper, find the best values for k, K,

and Vq by the method of

000333
00666

0010
0020
050

least squares.

F,(cm/sec)

K,(cm/sec)

00943
00815
00761
00740
00688
00612

010
015

00492
00397
00316
00234
00169
00125

0-20
0-25
0-30
0-35

Solution

Rearranging equation
experimental error,

and introducing an error tenn to accommodate the

+ pc- ajSc^/^ = R
= K, and j5 = Vok

Vs-Vo +
where

Vo ac^/3

II
III

Following the method described in Section 10.6.2 by squaring equation II

and equating to zero the partial derivatives of R^ with respect to Kq, a, and

gives

Fo(iV + a2^c2/3-2aXc^/3)

a(2/?FoZc^/'-^'Zc/^-Fo^Ec^/')

AZc2 + a'Ic/'-2a^c^/3)
= aXnc^/3 + FoSc-2aFoZc^/' + aVoZc^/'-Z^sC
where N = 12 is the number of experimental points, and each sum
over

all

These sums, which are evaluated in Table

points.

substituted into equations IV, V,

and VI

11.3,

VI
taken
can be

is

to give:

Ko(12 + 2-4949a2-9'5544a)

=
t Oliver,

62-92- 18'093a + 1-440^5- l-7516ai? + 0-5488a2i?

D. R. Chem. Eng.

Sci. 15, 230, 1961.

II

404

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

(Njr-~ovooo/-^oooo

OpPPP'TiCSfncncpfOfp

66666666666

OfS"^Ooo<N(N'7'Op'^T-iCiO

^-H^^(S(S(N^'^'^6

QoorsivocnTtoo
ppppp6

,>-sQoOOO^<Nt1-VO

666666666
,

oo.F-(ONVoascom<Nfo

QOoo-^r-^mONO^
oopppppppppp
6666666666

'-^^OOOOOOOOO
OO'iTtoo>/-^o>no/^

0000<NO<NO<NOr^
ooooO'^rsTtvoosrN

pppOpppOpp'TH

X)

ooOT-Hvo^HfsooON"^m

Oppppppppp^'T'

ir>fOr4TtrfrJ-r^O/^oovo

Oopppppp'7''7ir^(N

for--ooooooooo
mvoooooooooo
C?opppP'7i'7'<NeNfpf^

^
^

&'

f~\

'-^

1
^
d)

<--'opoO'-^<NrN<^mTt'^

ON
-^
ON
tT

11.

NUMERICAL METHODS

405

a(l-0976Foi5-0-14854^' -2-4949 Fo')

+ l-7516Foi5-0-2270/?2
_
^(0-35055 + 0- 14854a2 0.4540a)
= 2-1103a-3-8445 + l-440Fo-l-7516aFo + 0-5488a2Fo
which are to be solved for

The

Vlll

18-093Fo-2-1103i5-4-7772Ko^

a, p,

results given in the

and

IX

Vq.

paper were obtained graphically, and they are

0-75,

)S

18-98,

Fq

9-35

Using the method of Section 11.5.3 to improve these values by substituting

into equation VII to evaluate Vq, then using this
a and ^ from equations
in equation VIII to evaluate a etc., the
value of Vq with j5 from equation
following results were obtained.

a
Unfortunately, there

is

0-747,

i5

18-67,

Fo

XI

9-21

no guarantee that the correct equation is being used

and the results tended to oscillate in successive
Hence the values in equations XI can be checked by

for each determination

stages of calculation.

Newton's method as follows.

Each value is assumed to be
That is,
a

0-747

+ x,

by a small amount

in error

i5=18-67 + >',

Fo

9-21

(x, y,

or

+z

z).

XII

The higher order functions can be linearized by neglecting products of the

small quantities x, y, and z. Thus

=
p^ =
Vq^ =
oip =
(xVo =
pVo =
a^p =
(x^Vq =
olPVq =
aP^ =
aFo^ =
a^

+ 1-494X
348-57 + 37-37>^

0-55801

84-824+ 18-42Z
13-946+ 18-67x + 0-747>^
6-880 + 9-21x + 0-747z
171-95 + 9-21>'+18-67z
iO-418 + 27-89x + 0-558;^

5-1393 + 13-76x+0-558z
128-45 + 172-0x + 6-880>'+13-95z

260-4 + 348-6x+27-92j;
63-36 + 84-82x+13-76z

Putting the above linearized expressions into equations VII, VIII, and
simplifying, therefore

IX and

+ 0-438>^-6-255z =
74-61x-65-82>'-18-19z =
2-138x + 0-0944>;-0-438z =
18-19x

M.M.C.E.

0-029

XIII

0-15

XIV

XV

0-0029
14

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

406

Solving these equations by successive elimination or by matrices gives the

solutions

00014,

Hence, the values of

a

a, ^,

-0-0006,

= -00006

XVI

and Vq are

0-748,

i?=

18-67,

Kq

9-21

XVll

III,

and remembering

from equations XII

Returning to the original constants by equations

the factor (100) introduced in Table 11.3, gives the final result

=
k =
K=

Vq

0-0921 cm/sec

2-027
0-748

which confirms the graphical values given

11.5.5.

Comparison of Methods

The method of
repeated
easiest

in the original paper.

many

method

successive approximation consists of a simple operation

times,

and

if

the process

is

convergent, this

to use because any calculation errors

more

show up

is

probably the

in the following

but involves a fair amount of

is usually obtained after only
two cycles of calculation, but if a second cycle were needed in the above
example, a new set of equations XII would have to be substituted into
equations VII, VIII, and IX and the complete calculation repeated.
Thus Newton's method is more efficient and reliable, but the selfchecking feature and simpHcity of the method of successive approximation,
restore the balance of advantage and disadvantage between the two methods.
cycle.

Newton's method

algebraic manipulation.

11.6.

is

An

efficient

accurate solution

Difference-Differential Equations

The numerical solution of this type of equation follows directly from the
method given in Section 11.3. Considering the general form of equation
which

arises in

chemical engineering,

%"=f(yn-uyn,yn^i)

(11-46)

where the values of y are given at some value of / (say / = 0) and taking
an interval (h) in t, then the modified Runge-Kutta process can be performed
on equation (11.46) for each value of n. Thus, evaluating

K,t
for each value of ,

K,2 =
and

hf(y.y,y^,)

(11.47)

and then evaluating

hf(y.,+iK.i,,,y + ik^,.y^,+iK^u,)

(11.48)

finally

K3 = hf(yn-i+^K-i,2-K-i.uyn+^K,2-K,uyn+i+^K+i,2-K+i.i)
(11.49)

11.

NUMERICAL METHODS

the value for y at the end of the increment

yn(h)

The whole process

(11.23) except that

Equations
to k,

which

yn+i(k.,i+^k^2

similar to that used

is

is

407

given by

+ K,3)

(11.50)

from equation

many more dependent

(11.15) to equation

variables are involved.

48, 49) can be used to evaluate the three approximations

the change in y during the time interval, at all values of n

(1 1.47,

is

except the boundary values at n = 0, N. These boundary conditions take

precedence over equations (11.47, 48, 49) and usually effect a simplification
of the calculation in the region of the boundary.

The method

just described

undoubtedly the most accurate way of

is

solving an equation of the type (11.46) numerically, but the arithemetical

complicated and confusing, leading to many errors by

It is usually used in digital computer programs and
is normally used for manual computation.

details are rather

manual computation.
the following method
11.6.1.

Step-by 'Step

Method

The increment

in y corresponding to an increment in / in equation

equation (11.50) as a weighted average of k^i, A:^2> ^^^
k^2- Provided the increment in t is small enough, the three approximations
to k will not differ greatly from one another and /: i can be accepted as a
first approximation.
This dispenses with the need to calculate k^2 ^^^
k^2 thus simplifying the arithmetic considerably, at the expense of accuracy.
The loss of accuracy can be reduced in most cases by taking a larger number
of smaller increments and hence the size of increment can usually be chosen
to give an adequate solution. The following simplified example is chosen
to illustrate the procedure.
(11.46), is given in

Example L
gas absorption column contains six theoretical plates each
having a liquid hold-up of 10 lb. An inert gas passes through the absorber
between periods of continuous operation and the pure solvent is fed at a
reduced rate of 200 Ib/h. When the main process is restarted, gas containing
5 % by weight of the soluble component enters the base of the absorber at
1000 Ib/h. If the absorption coefficient is given by
y
what time
keep the

= 0-40x

interval can elapse before the liquor rate

exit gas

concentration below

must be increased to

0-2%?

Solution

Taking a material balance over the th plate as

Gy_,+Lx^,-Gy-Lx =

illustrated in Fig. 11.8,

dx

H-^

II

at

Because the boundary conditions and concentration hmit are expressed in

terms of gas composition, it is convenient to use equation I in the form
y

kx

III
14-2

MATHEMATICAL METHODS

408

Flow through

Fig. 11.8.

to eliminate x

CHEMICAL ENGINEERING

IN

plate of a gas absorber

/7th

and x+i from equation

II.

Thus

(^yn-l+yn+l-OCyn-yn
where

Hdy
L dt

IV

= kGIL

Taking a time interval At and denoting the new gas composition above the
nth plate by y*, the derivative in equation IV can be replaced by a finite
difference form. Thus
TT

ccy-i+yn + i-{GC+l)y

= -(y*-y)
LAt

VI

Solving equation VI for the unknown:

LAr

LAt

VII

On

new

physical considerations, the

value y^*

is

expected to depend

the old value y on the same plate. Also, a hypothetical increase in

the value of >^ at fixed values of ;^_i and y+i should result in an increase
in the value of y^*. Hence to keep the problem physically reasonable. At

upon

must be chosen to

satisfy

LAt

l-^(a + l)>0

VIII

At<HlL(oi + l)

or

IX

Inserting the numerical data given in the problem,

0-40x1000

^^

2-0

200

H
L(a +
Hence, At

is

1)

chosen to be 1/120 to

10

200x3-0

60

satisfy IX.

Equation VII now takes the

form
y:

= U2y-i+3yn+yn^i)

NUMERICAL METHODS

11.

The boundary conditions

are:

= 0,
= 0,
M = 7,

at

=
>'o =
^7 =

at
at

third condition results

XI

The

409

0-05

XII
XIII

from the pure solvent feed

to the top of the

absorber (i.e. x-j 0).

Table 11.4 can now be constructed to receive the results, and the
boundary conditions can be inserted in the first row and the end columns
{yQ and y-T). The index (m) has been used to count the time increments.
There is a contradiction between conditions XI and XII regarding the
composition yQ zX t = 0, because of the step change in feed composition.
Experience has shown that the best solution usually results from taking a
value for yQ equal to the arithmetic average of the two conflicting requirements.
Table

yo

2
3

4
5

6
7
8

9
10
11

2-500
5-000
5-000
5-000
5-000
5-000
5-000
5-000
5-000
5-000
5-000
5-000

11.4.

Gas Compositions

y^

yi

0-833
2-083
2-755
3-183
3-483
3-707
3-880
4-018
4-131
4-225
4-304

0-278
0-833
1-350
1-790
2-159
2-470
2-733
2-958
3-152
3-321

yz

0-093
0-324
0-617
0-926
1-227
1-512
1-776
2-018
2-240

in

Absorption Column (%)

y^.

y^

y^

0-031
0-123
0-269
0-451
0-653
0-865
1-080
1-291

0-010
0-046
0-113
0-210
0-330
0-469
0-620

0-003
0-017
0-046
0-093
0-156
0-234

7?

now be built up, one row at a time, by successive

This shows that the exit gas concentration {y^) exceeds
the limit of 0-2% after 11 time increments. Since Ar = 1/120 h = Jmin,
the Hquor rate must be increased 5 min after the rich gas has entered the
absorber.
The

table of results can

use of equation X.

11.7.

Partial Differential Equations

In Sections 11.3 and 11.4, it was shown that some methods of solving
ordinary dififerential equations were unsuitable for boundary value problems,
and others were unsuitable for initial value problems. It is again necessary
to present independent methods of solution for the same classification of
problems. Initial value problems were defined in Section 8.4.5 as those

problems in which at least one independent variable has an open range;

whereas boundary value problems have closed ranges for all independent

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

410

variables. Time is the open range variable of most frequent occurrence in

chemical engineering, and coordinates describing a finite space are the most
usual closed range variables. Occasionally, time is disguised as the ratio
of a coordinate to a velocity, as in the faUing film of example 5 in Section 8.8.
Two methods will be presented for solving initial value problems, followed
by two methods for boundary value problems. Whereas the latter two
methods are equally useful and accurate, the iteration method is to be
preferred for most initial value problems.
The discussion will be confined to equations containing only two
independent variables; the extension to three variables presents no new
ideas but leads to difficulties in organizing the calculation on a two dimensional piece of paper.

11.7.1.

Method for

Step-by-step

Initial

Value Problems

The simplest initial value partial diff'erential equation is

the time dependent

linear heat conduction equation

dt

a^

(11.51)

dx'

few analytical solutions of equation (11.51) have already been presented

Chapter 8, but if the boundary conditions are functions of time, a
numerical solution is often more easily obtained. The basis of the numerical
method is to replace the space derivative in equation (11.51) by a finite
Thus
diff*erence representation, using equation (11.32).

in

Comparisons of equations

(11.46, 52)

shows that the methods suggested

for diff'erence-differential equations can

(11.52).

In particular, the step-by-step

now be

used to solve equation

method can be used by replacing

the

time derivative by an equivalent finite difference expression. To simplify the

symbolism, it is convenient to introduce a second suffix {n) to count the time
increments. Hence by using equation (11.31), equation (11.52) becomes
-*

m,n-f

^m,n

2Ar

__i__(T
IT _L T"
^
" ^^^y^^m^Un-^^m,n^
^m-l.n)

or

By using equations (11.31, 32) to replace the derivatives in equation (11.51),

equation (11.53) is accurate to second order diff'erences. Unfortunately,
equation (11.53) is unstable, giving a solution showing an unbounded
oscillation of temperature as time progresses. The reason for this is that the
new point temperature {T+ 1) is calculated from four known temperatures,
one of which (T ) has a negative coefficient. This means that a small positive

11.

NUMERICAL METHODS

error in r^, will lead to a negative error in

411

r +i. On the next step in the

augment the positive error in

calculation,' this negative error in T^,+i v^ill

T and
shows

generate an increased positive error in T^^^2' Equation (11.53)

n is increased by unity. As a general
all known point values are put on to the same side of an equation

this quite clearly if the subscript

result, if

coefficient is negative, then the solution will be unstable.

Since equation (1 1 .53) is unstable, an alternative finite difference expression
must be used for the time derivative in equation (1 1.52). An expression which

and any

is

only correct to

first

differences

^ ^m,n

is
-*m, n

+1

^m,n

^aa ca\

At

dt

which leads to the equation

T,.^,=MT,^i. + (l-2M)T,, + MT_i.

(11.55)

as a finite difference representation of equation (11.51).

M = a(A0/(Ax)2

(11.56)

termed the " modulus " of the equation, and its value determines the
" stabiHty " and " convergence " of the method.
Formulae do exist to
determine which methods are stable and convergent, but the statement above
that all coefficients in equation (1 1.55) should be positive for a stable solution
is both simple and effective. Hence
is

M^i
If the

temperature

is

known

(11.57)

at all values of

(or

m)

for a value of

(or ), then repeated appHcation of equation (11.55) will determine the comBefore the method
plete temperature distribution at time t + At (or
l).

can be started, values must be selected for Ax and At such that the condition
(11.57) is satisfied. The most convenient procedure is to choose Ax to be a
suitable fraction of the total range of x, and then choose
to satisfy (11.57)
and give the simplest coefficients in equation (11.55).
The simplest choice of value for Af is J so that equation (11.55) becomes

This value of
is on the limit of stabihty and equation (11.58) cannot be
expected to give answers of great accuracy, it does however have the merit
of simplicity and will give a good approximation to the answer if A;c is chosen
sufficiently small. The first known use of equation (11.58) was by Schmidtf
who used it as the basis of a graphical construction.
Boundary conditions were classified in Section 8.4. The first type will
simply fix the value of T^ at some value of
for all values of n. The
other two types both involve a derivative which must be replaced by a
finite difference expression. Equation (1 1.31) is the best one for this purpose,
but an equation analogous to (11.54) must frequently be used for speed of
calculation. Thus accuracy is sacrificed by using an expression correct to

t Schmidt, E.

'*

Foeppls Festschrift

",

Springer-Verlag

OHG,

Berlin (1924).

MATHEMATICAL METHODS

412
first

Because

differences only.

made

to avoid the unstable

missible to

make

IN

this type

CHEMICAL ENGINEERING
of approximation has already been

form of the general equation

(11.53),

it is

per-

a similar approximation in the representation of boundary

conditions.

These remarks, and a few others are best illustrated by reference to an

example. The following type of problem is of frequent occurrence in reactor
design, and involves the simultaneous solution of two inter-connected initial
value problems. No new difficulties arise as a result of the inclusion of a
second dependent variable, but the calculation is of course about twice as
long as a similar calculation with one dependent variable.

Example 1. The dehydrogenation of ethyl benzene has been studied by

Wenner and Dybdalf by passing a gaseous mixture of ethyl benzene and steam
through a tubular catalytic reactor. They have shown that the rate of
reaction can be adequately expressed by

'c

^ (pe

^^) lb moles/h lb catalys

where Pe

is

the partial pressure of ethyl benzene (atm)

Ps

is

the partial pressure of styrene (atm)

Pfj is the partial pressure

The reaction

rate constant
;^

is

of hydrogen (atm)

given by

= 12600exp(-11000/r)

II

and the equilibrium constant by

K = 0-027 exp[0-021(r- 773)]

III

T is the temperature in K.
each reaction tube has an internal diameter (2a = 4") and is to be
suppHed with 0-152 lb moles/h ethyl benzene and 1-52 lb moles/h steam at
600C, the total mass velocity will be 500 Ib/h ft^. The tube is heated by
flue gas which flows countercurrent to the reaction mixture at a rate (R =
287 Ib/h) and leaves at a temperature of 620C. The following data are also
where
If

supplied.

Bulk density of catalyst

(p)

Operating pressure (P)

Heat of reaction of ethyl benzene (A//)

Eff'ective

thermal conductivity of bed

Ratio of effective

diff*usivity to

(Arf)

-2

Ib/ft^

atm

60 000 Btu/lb mole

0-26 Btu/h

=0-52

(Cp)

Estimate what length of reactor tube

of ethyl benzene.
t

90

hnear velocity (De/u)

= 0-0014

Specific heat of reaction mixture (Cp)

Specific heat of flue gas

=
=
=
=

is

ft"

ft

Btu/lb F

0-24 Btu/lb F

required to achieve 45

Wenner, R. R. and Dybdal, E. C. Chem. Eng. Prog.

ft^

44, 275. 1948.

conversion

NUMERICAL METHODS

413

Solution

In the element of volume

shown

in Fig. 11.9, ethyl

benzene

is

consumed

at a rate

InrSrSz pr^
and heat

is

lb

moles/h

absorbed at a rate

InrSrSz pr, AH Biulh

6z

4u,c,T

c
Fig. 11.9.

Element of reactor tube

Taking a balance of the mass flow rates of ethyl benzene through the element,
accounting for bulk flow in the axial direction, mass diffusion in the radial
direction, and losses by reaction, gives

InrSruc lnrSz

De

d(uc)

dr

2nr5ruc

is

(2nrSruc)5z

oz

+ 2nrSz
where

De

d(uc)

dr

InrSz
dr\
u
d

dr

Inrdrdzpr^

dr

and u is the
After cancellation of terms, this equation simplifies

the concentration of ethyl benzene (lb moles/ft^)

linear gas velocity (ft/h).

to

DE(d\uc)

d{uc)

Similarly, a heat balance for the

dz
It is

more convenient

volume element

GcXdr""
to

work

\d{uc)\

in

dr)

gives

GC^

terms of the fraction of ethyl benzene

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

414

converted which

denoted by

is

UqCq

where the subscript

denotes entry conditions. Differentiating,

(q)

d(uc)

.'.

= UoCQdx

Hence equation IV becomes

Dr /d^x

dx

dx\

or.

-[-^'r--]--^^-^ =

oz

u \dr

VII

UqCq

r or I

V and VII can be evaluated by using the given

problem it is also convenient to check that the
units used are consistent.
Here for instance, two temperature scales are
being used and a correction must be made. Thus
The

coefficients in equations

data.

At

this stage in the

/c

0-26

GC^

500 X 0-52

00010

ft

^^^^
^ = 60000x90r,
^ 20
500x0-52

Ai/pr,

^,^

800r, F/ft

GCj,

= 00014

'

'

11600r, X/ft

ft

Hence equations

pr^

90r^X7r

UqCq

0-152x36

^^_^

and VII become

dT

/d^T

- 00010 Id'-T
y +

dT\
+
rdr)

VIII

+ -'^l-51.7r. =
g- 0-0014^^
\d?^~rY)

IX

Yz

^^

^
Sl'lrAl

\dr^

/d^x

11 600r,

dx\

After the reaction mixture has passed a distance z through the bed, a fraction
{x) of the ethyl benzene will have been converted. Thus the reaction mixture
will consist of:

10

\x
X
X

ll+x

moles
moles
moles
moles

steam
ethyl benzene

styrene

hydrogen

total

Since the total pressure

is

1-2 atmospheres, the partial pressures

reactants are:
1-2(1

-x)

l-2x

of the

Substituting equations II

415

NUMERICAL METHODS

II.

and

into equation I gives

15100exp(-11000/r)[i:i^^-^i^]

K is

where

Two

given by equation III.

of the boundary conditions for this problem are,

when

0,

T=

and when

0,

or

The

XI

XII

873

XIII

third condition expresses the fact that the flue gas loses heat to the

reaction mixture through the wall of the tube.

an element of the tube of length

dz RC'

Taking a heat balance over

dz,

-- = 2na bz kv -r
dr l^

dz

InakE dT

dT

XIV

or

dr

RC'p

XIV becomes

Inserting the numerical values, equation

dT

dT

XV

= 0-0040-
dr

dz

The corresponding boundary conditions

mass transfer equation IX

for the

are,

when

0,

when

0.

XVI
XVII

dr

and when

XVIII
dr

This completes the mathematical description of the problem and to start

and (11.32) are used to replace all derivatives
with respect to r in equations VIII and IX whilst an equation of the type
(11.54) is used to replace the derivatives with respect to z. Thus

its

solution, equations (11.31)

m,n +

Az

^-

9T

-4-

0-0010

2m(Ajr?

+ 11600r, =

XIX

and
'm,n +

-X,

Az

0-0014

m-

l,n

^T i+l,fi~^m-l,n

2m(Ar)2
-51-7r,

XX

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

416

XIX and XX

Equations

can be rearranged to the forms

Mn- jT^_i,-

11

600A2r,

XXI

and
m,

n+

M '(l+^)^.+

l.n

+ (l-2M'K. +

+ MYl-^^x,_i, + 51-7Azr,

M = 0-0010 Az/(Ar)^

where

M' =

XXII
XXIII

XXIV

0-0014 Az/(Ar)2

and M\ it is necessary to inspect equations

Thus it is seen that when m = 0, four of the coefficients
become infinite, and different equations must be found to represent VIII
and IX along the axis of the tube. Boundary conditions XIII and XVII
show that an indeterminate fraction arises in each of VIII and IX and
L'Hopital's rule must be used to resolve these terms. Thus
Before values are chosen for

XXI and

XXII.

hm^ = -^
The

finite difference

ro.+i

Zo.+i

equations for use on the axis are thus

= 2MT,^ + {l-4M)To, + 2MT.,^ - 11 600 Azr,

= 2M'xi, + {l-4M')xo,n + 2M'x_i, + 51-7 Azr,

Equations XIII and XVII expressed in

(11.31)

XXV

finite difference

XXVI
XXVII

form by using equation

become
andxi.

7^i.n=7^-i.n

x_i,

XXVIII

and M' must be chosen such that no

For a stable calculation process,
negative coefficients arise in any of equations XXI, XXII, XVI, or XXVII;
and the

largest permissible consistent values are

M = 0-179,

M' =

0-250

XXIX

In order to obtain an approximate answer quickly, four radial increments are

chosen. Hence Ar = 1/24, and equation XXIV gives

Az = 0-310
Equations XXI, XXII,
T^.+i

0-179 (l

XXVI, and XXVII now

+ J-]

T+,

XXX

ft

take the form

+ 0-642r,. +

+ 0-179 ('-.y
l---r_i -3600r,

XXXI

11.

,+i

417

NUMERICAL METHODS

+ ^)x,+i. + 0-50x,, +

0-25 (l

+ 0-25(1 -^)x^.,, + 16'0r,

XXXII
XXXIII

To.+i =0-716ri, + 0-284ro.-3600r.

^o.n +

XXXIV
T^^q = 873,

=^i.n + 16-0r^

Boundary conditions XII and XVI state quite simply that

and x^^o = 0; but the other two conditions XV and XVIII involve
derivatives. The simplest yet accurate way of representing equation XVIII
" fictitious points " at r = + Ar by imagining that the
is to use a row of
tube is one increment larger in radius than it actually is. In this example,
flf

the tube wall

XVIII

is

at

4,

but with a

fictitious

row of points

at

5,

equation

gives

X^
5,n

XXXV

Xn3,n^

and ^4 +i can be calculated from the standard formula XXXII.

If the same idea of fictitious points is used in equation XV, difficulties
arise due to a first order representation (1 1.54) being used for the z derivative,
and the second order representation (11.31) being used for the r derivative.
This inconsistency cannot be resolved at present by taking a second order
representation for the z derivative without generating an oscillation a first
order representation must therefore be used for the r derivative. Thus
:

Ar

Az
or

T^,n^i

XXXVI

T4,+o-030(r4.-r3.)

and the reader is

an improved method.)
The equations are now in a suitable form for calculation, and the next
step is to evaluate the coefficients in equations XXXI and XXXII so that they
(This

is

the least accurate equation in the present example,

referred to the next sub-section 11.7.2 for

can be presented as shown in Table

Table

11.5.

coefficient

11.5.

Coefficients in Reactor Design Equations

of

coefficient

of

m
Tm-\,T\

2
3

0089
0134
0149

^ m,n

Tm+l,n

0-284
0-642
0-642
0-642

0-716
0-269
0-224
0-209

Xm-l,n

Xm,n

Xm+l,n

0125

0-500
0-500
0-500
0-500

0-375
0-313
0-292

1000
0-187
0-208
0-500

418

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

Next, the sheets on which the calculations are to be performed must be

double fold of foolscap is suitable in the present case, and this
The initial conis ruled into two adjacent tables as shown in Table 11.6.
ditions of 873K for the temperature and zero for the conversion are filled in
on the first row, except at Tq^4, where the average of the reactant temperature
and flue gas temperature is used, as in the example of Section 11.6.1.
at each point in the temperature
Equation III is then used to calculate
enclosed in square brackets in equation XI
table. The function of x and
Since equation XXXII
can now be evaluated in the conversion table.
involves 16-0/,, it is convenient to tabulate 16-0 x 15 100 exp (-11 000/T)
prepared.

Fig. 11.10.

Conversion

profiles at different reactor depths

The product of corresponding items as given by

equation XI results in the evaluation of 16-Or^. The evaluation of the term
3600r, in equation XXXI is now simple. Equations XXXI, XXXII, XXXIII,
XXXIV, and XXXVI can now be used to find T^^^ and Xi,, with the
modification of adding the other 10A^ to Tj 4 to complete the initial step
change.
The remainder of the results are summarized in Table 11.6 until the last
step from n = 10 to = 11, which is presented in full detail. The conversion
for use in equation XI.

profiles are illustrated in Fig. 11.10.

11.

NUMERICAL METHODS

419

oooovomoor<^r~0"^r^ON

66666666666
OOOO'-^VOTfrNOOTj-r^iO
Tj-Tt-Hoor^-H^vooNOsrr--r^moomoo<Nnc'iTf

m
6

O "O
66

ON
o

66666666666

m
6

o
On

000\n-^VO<NO\<NO\00
TrTi-vooNONoo\oor-m
r^r^r^^o-^r^-HTj-vooN

OO
n
ro

<NvO

66666666666

66

o\

on

CT\

O Tt
66
^fcjo
tri-^

oooNONNor--ooo"^ooo
Tj-Tj-Nor-^r-HONOfNr-o

On

r-~r.tNNOOrovooo^^r-iNO

ro

ONr^^m
^hoo

66666666666

66

OOONONO^ONOOO\rfO\
Tj-Ttvor-^osTtONmr-ON

0
r-

Pp^'7'N<N<N<Nrpcnrp

p
6

66666666666

Om
66

^
vo

B*

.6

s,l

r^

|cn">^o^obo\'^fn'<i-vo
1
^
O 'onOnOnOnOnOnOOOO
OOOOOOOOOOOOOnONOnOn 6
ON

1-293

907-9

11-2

NO
rj

oo^oo<^Oooooooo\0<STf

O^HOOOOOOOOOOOOOOOOOOOO
^H vo

His

Hf^HN

oo'^"tr^<Noovo>o<ow-^NO

O^oooooooooooooooooooo

o
^
O

-*

-^

tNl

pNb:^

O'-HOOOOOOOOOOOOOOOOOOOO
NO

o O
OO
00

o
o Om

OO

i ^^2
O o n 00

Hn

00-^^000<Nr^Tj-fSOO\
o
so o '^ fn rg rq -^ ^ ^ ^ o
O ^
O^-HOOOOOOOOOOOOOOOOOOOO o
t

"O 00

<N
v^

^(.^tltff,^t,^t,^blK^tltl

0-300

2-5 809

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

420

The average conversion of

the material leaving the reactor

/
X

is

given by

rxdr

XXXVII

^^

rdr

I
the numerator of which can be evaluated by Simpson's rule (Section 10.7.2).

Thus

12

i(^l. + ^2.n + 3X3, + X4,)

XXXVIII

Inserting the values of ;c^ when = U gives

X = 0-4573
which is above 45 % conversion. Whenw = 10, x =

XXXIX

.'.

of the reactor tube required

0-4324.

Hence the length

is

L=

10-7Az

XL

3ft4in.

The above method of designing a tubular reactor is superior to the

method ignoring radial temperature and conversion variations, and
the numerical work required about six hours with an electric desk calculating
simpler

machine. In order to improve the accuracy of the solution, attention must be

paid to the following points.
(i) The conversion profiles near the tube walls exhibit a point of inflexion
which cannot be represented adequately by a formula which is only correct
to second differences. This error can be reduced by choosing a smaller value
for Ar.
(ii)

The

partial derivatives with respect to z

have been represented by a

difference equation correct to first order terms whereas the derivatives with

respect to r are represented correctly to second order diff'erences.

As noted

XV

previously, boundary condition

is only represented
order differences.
In view of the approximations inherent in the expression for the reaction
rate and distribution coefficient, and the low accuracy to which the eff'ective
thermal conductivity and mass diff'usivity can be determined, the foregoing
method is considered sufficiently accurate. However, an improved repre"
sentation of this problem will be derived to illustrate the ** iteration
(iii)

correctly to

first

technique.
11.7.2.

Iteration

This is the correct name for an organized method of trial and error
which leads to the solution by obtaining successively improved values.
To improve the ordinary step-by-step method, all derivatives need to be
replaced by finite differences correct to second order terms. That is, equations

11.

(11.31, 32)

NUMERICAL METHODS

must be used. Referring

421

to equation (11.51)

and

its

stable finite

difference equation (11.55), Fig. 11.11 shows diagrammatically, the layout

of the points used in the calculation. Equation (11.55) represents equation
m+i,

m,n

-1, n

"E

Fig. 11.11.

(11.51) at the point

P
m,n+1

m-1,n+l

m+1,n+1

Point values used in iteration method

and involves the temperatures

at the points B, C,

D.

Viz.

r,.,i

=Mr,^i.+(i-2M)r,.+Mr^.i.

M = (x{At)l(Axy

where

(11.55)
(11.56)

Equation (11.54) which was used to replace the time derivative is only correct
A, but is an adjusted form of equation
(11.31) correct to second differences at point E, mid-way between A and D.
Thus, if the differential equation is replaced by a difference equation vahd
at point E instead of point A, then an improvement should be apparent.
Using equation (11.32) at point D gives
to first order differences at point

d^T
T"^2

7T~T2(-^m+l,n+l~2T; +!

Taking an arithmetic mean of the

finite difference

+ r^_i

+i)

(11.59)

expressions for the second

space derivative at points

and
gives the following expression correct to
second differences for the space derivative at E. Thus

dx^

-2

E-2(Axy

v-'m+l.n'^ 'w-l,n+ -'m+l,n+l+ 'm-l.n+l

^Tm,n

^Tjn,n+l)

(11.60)

By

substituting equations (11.54, 60) into (11.51)

difference equation correct to second differences

is

and rearranging, a

finite

obtained.

+(i-M)r,,+iM(r,^i,+r,_i.)

(11.61

This is the form in which the equation is used, but there are two unknown
terms on the right-hand side within the first bracket. The use of equation
(1 1.61) is very similar to the method described in Section 1 1.5.3 for improving
a root of an algebraic equation, except that here, there is a set of simultaneous
equations to be solved. Therefore, a first approximation is guessed for
r^^n+i for all values of m. The right-hand side of equation (11.61) is next
evaluated to provide a better approximation for r + j. These new quantities

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

422

are then used

on the right-hand

side of equation (11.61) to determine a

Repetition of this process will give successively
improved values. The above technique applied to equations of the type
(11.51) is known as the " Crank-Nicolson " method.
further advantage of equation (11.61) over (11.55) is its greater

further approximation.

Equation (11.61) is actually stable for all positive values of M,

but the use of too large a value leads to inaccurate results, and very slow
convergence. The effect can be described in relationship to Fig. 1 1 .7. The
speed of approach to the true solution depends upon the relative gradients
of the two curves enclosing the stepwise construction; and as
is increased,
the two gradients approach one another and more steps are needed to find
the solution. Because a large value of
impUes a large value of Ar, the
finite difference representation correct to second order may not be good
enough. Hence, a large value of
will squander the inherent accuracy of
the method without accelerating the solution. The authors' recommended
choice for
is the same value as would be used to make the ordinary
stabiHty.

method stable.
The treatment of boundary conditions

step-by-step

is little

altered in this

new method,

but equation (11.31) should always be used to replace any derivatives in the
boundary conditions, so that the whole process is correct to second order
differences. The details are best illustrated by an example, and the reactor
of the previous section will now be redesigned.

Example 2 (continuation of example 1). The problem and its treatment

XVIII is reached, and replacing the finite
difference representations XIX and XX by the more accurate second difference
are identical until equation

formulae gives
'm.n+l

0-0010

T^T

-*m-l,n

^-'m.n' -'m l,n ^^ -*m+l,n

-'m+l,n

Az
,

-^m+l.n-H ~^-^m,n+l

-*m-l,n+l

-^m+l,n+l~-^m-l,n+l

^
,

2m(Ar)2

(Ar)2

+ 5800[r> + l) + r,(w)] =

XLI

and

Az

2
.

^m +

2m{^rf

(Ar)^

~ ^^m, n-H+^m-l.n+l

^m+l.f+l~^m-l,n+l

2m(Ar)^

(Ar)2

-25-85[r,(n

+ l) + r]=0

and M' as defined by equations XXIII and

ing the axis points as before gives the set of equations,
Introducing

XLII

XXIV, and

treat-

11.

NUMERICAL METHODS

423

(l+M)T,.+i=iM

+ (1 - M)T^^ - 5800[r,(n + 1) + r,(n)] Az

XLIII

+ U-^j(^m-l,n + ^m-l.n+l) +
+ (1 - M'K, + 25-85[r,(n + 1) + r,()] Az XLIV
(l+2M)ro.+i = 2M(Ti.+ri, + i) + (l-2M)ro,- 5800[r,(n + 1) + r,(n)] Az
XLV
(l+2M')xo. + = 2M'(xi, + Xi, + i) + (l-2M')xo. +
XLVI
+ 25-85[r,(n + 1) + rXn)] Az
Choosing M and M' to have the same values as given by equations
XXIX following the recommendation above, and selecting six radial
i

increments to improve the accuracy gives

=
Az =
Ar

and hence

XLVI

Equations XLIII to
'/;.+!

=0-0758

:^

thus

1/36

become

(^m+l,n+ 'm+l,n+l) +

+0-697r,-678[r> +
.Vn-Kl= 0-100

XLVII

0-138

)(^m+l,fi

l)

+ r]

XLVIII

+ ^m+l,n+l) +

XLIX
L

+ 0-600x^, + 2-85[r,(n + 1) + r,(n)]

ro,^i=0-263(ri. + Ti,^0 + 0-474To,-589[r,(M + l) + r,(n)]
Xo.+i=0-333(xi, + Xi, + i) + 0-334xo, + 2-37[r> + l) + r,()]

LI

Boundary conditions XII, XVI, and XVIII are treated as before the last
by means of fictitious points. However, an equation correct to second
differences can now be used for equation XV as follows. The tube wall is
given by w = 6, hence, using equation (11.54),
;

LIT

dz

Az

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

424

Averaging the radial derivatives, as given by equation

'N'T-'

(11.31), gives

-I

LIII

-^=^a7.+7^7.-fl-7^5.n-7^5.n+l)
Substituting equations LII

and LIII

10^
^7,n +

XV and rearranging gives

into equation

^5,n+^5,+l~^7,n

Ar

(^6, n+

~ ^6,n)

LIV

coefficient of the last term would necessitate calculating T^ and

^6,n+i to two decimal places to determine T-j^n+i to the nearest unit. Since
this is inconvenient, 7^7 +i must be eUminated algebraically from equations
LIV and XLIII. This simplifies to

The high

/lO^ArM

OTTe,^,

/lO^ArM

+ 5800 Az [r,(n + 1) + r,(n)]

= l-0212r6.-0-0106(r5, + r5.+i) + 47-5[r,(/i + l) + r]

LV
LVI

Table 1 1.7 can now be constructed to present the coefficients of equations

XLVIII, XLIX, L, LI, and LVI. The actual calculation is performed on a
double fold of foolscap again, ruled out in the same form as Table 11.6
except that there are more columns to accommodate
^ m ^ 6. The
headings in the left-hand column are exactly the same and the figures are

compounded according
Table

to the appropriate difference equation.

Coefficients in the Iteration

11.7.

Method

coefficient of

Tm-\,n

rp

Tm+i,n
lm+i,n+i

0-4740
0-6968
0-6968
0-6968
0-6968
0-6968
1-0212

0-2630
0-1137
0-0947
0-0884
0-0853
0-0834

m-i,n+i

0-0379

0569
0-0632
0-0663
0-0682

4
5

6 -0-0106

coefficient of

Xm-i,n
Xm-i,n+i

-589
-678
-678
-678
-678
-678
47-5

0-050
0-075
0-0833
0-0875
0-0900
0-2000

^m+i,n
^m+i,n+i

0-334
0-600
0-600
0-600
0-600
0-600
0-600

0-333
0-150
0-125
0-1167
0-1125
0-1100

2-37
2-85
2-85
2-85
2-85
2-85
2-85

inserting 7"^ q = ^73 for m ^ 5,

should be noted that in the iteration process,
the full value of the step change is used at T^^q because this figure is used in a
difference equation valid at = J which is above the step change. In the
ordinary step-by-step method, only half of the step change is included
initially because the difference equation applies to ==
which coincides
with the step.

The

Tg

calculation

893,

and xq

is

commenced by
0.

It

11.

NUMERICAL METHODS

Following the method of constructing Table

15 100 exp

(-11 000/r);

^^
11+x K{ii+xy
^J'^"^

^,

425
11.6,

the functions K;

lOOr,; 678r,; and2-85r, are

(The last two terms are modified at the boundaries

calculated in turn.
first approximation must now
according to the table of coefficients 11.7.)
be estimated for T^^i and jc^^ for all values of m. The above functions are
again calculated at = 1 so that all terms needed in the equations are

available for calculating T^^ i and x^^ more accurately. These values are
then used in a repeat calculation of the above coefficients, and hence further
values are obtained for T^^i and x^i.
shortened version of the
The results are presented in Table 11.8.
iterations leading to the first step are given in the first nine Hnes and the
solution only is given thereafter. It can be seen that T^^ and x^^ were
each calculated three times due to a poor initial estimate, but in all other
cases only two iterations were ever necessary because the correct answer
could frequently be predicted by using an extrapolation formula from
Section 9.4.
The numerical work on this calculation took considerably
longer, but the increased accuracy and reUability is well worth the extra
effort. The amount of desk calculation required in the solution of this type
of problem is sometimes prohibitive, but the increased availabihty of digital
computers has ehminated this physical restriction. This subject will be
considered further in Chapter 14.
The required reactor length is predicted to be

L=

23-8Az

3ft4in

LVII

Although the agreement between this figure and that obtained previously
by the step-by-step method is perfect, it is considered to be fortuitous. A
more detailed comparison of the two solutions reveals a 4 discrepancy in
the centre line exit temperature and more seriously a factor of two discrepancy between the two quantities of heat taken up from the flue gas.
Nine increments in the z direction in the iterative solution are equivalent
to four increments in the step-by-step method, and Fig. 11.12 gives a
pictorial comparison of the two versions of the situation at directly comparable stages along the length of the reactor. The iterative solution gives
smaller conversions initially, but due to higher temperatures being maintained,
the rate of conversion does not fall away so rapidly. Hence, at 45
con-

two

have cancelled to predict the same reactor length.

There is a further calculation procedure of an iterative nature which is
often used to solve problems of the above type. The method will be described
by reference to equation (11.51) and Fig. 11.11.
Replacing the space derivative by its finite difference expression as in

version, the

effects

Section 11.7.1 gives

dT.m,n
8t

(Ax)^(^'"+".-2r'"."+^"-i.-)

Using the known temperatures at A, B, and

C in

(11-62)

Fig. 11.11 to evaluate the

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

00 tNONr^OO-^vOr-iOO
pocpOOOppO'7^*7''7-7-r^fSfSfNrnfnrnrnrp-rf'^^

426

0666666666660066666660600666606
r^

-^ r-

rsj

rN>y-^fNiovovovovooor^ONO\<NoooN>OTfONOr--ONr^

\D

voosfN-^r^otN-^vooootNTj-ior-aNO
^ -^(NtNCNCNr-immmr'-imTiOOOOOOOOOO-

io vo r- c
"^ '^ "^ -^

ir->

6666666666666666666666666666666

<SOoof*^voovorNim'rf^r>oofNooOTfr^u->r^Tt'Or^Tt^^TtmooONr~--^
,mo^HO^H
<^^T-,osoom'or-r^r-vOTtr^ON'OfNoomoorvi'sOONfNr^oo

'pppppOOppp-7-7^'7|'7l'7fNfS(Nf^^<Nfn^pfnrpcpfprnr^^^T^

6666666666666666666666666666666
rNooooNTi-voo tnvo'^
no-^oscNvooNrvjior-^cs-^

rn

-- voooosOfNf'^'^^r^oooN
'OOOOOOOOO"P'Ti^'Ti'Ti'^f^rNlfSfSrMfSrSrnrncnrpfnrnfnfn'^
6666666666666666666666666666666
OiO'ic>ovooonaN'OvooeN

oor<-j\or--

roor-fNcxsrsiNOOt^r^OJNior^aNOfNmTf
'(N'^<N^rn-Hromooo<NTtvor-a\orsimiovor^a\0-^tNmiovot^oo
pppppppppP'7i'7'7^'7i'7i'7ifNfS<Nr^CNfSrSrnrpfnrpfp<^f<irp

6666666666666666666666666666666
(NOt~^osvDooooNOr^aNr->a\oo\oor^mTr-^
Tt -^ 0
^O"^t^oor^m^
mu^
^Ov^O^^Ooo<Nmr*^^ooT}o^m^~Om"|^-w^
'VI ^ -,>4^i^>^^^ _
O m in
-^N-^(Nf^HfnnooorqTfior-ooo^roTf>ovor^ONO
<NmTtmvo
pppppppppp-T'-T-i'Ti'T^-Ti'^CNlcSfSfSfNfSfSrsJf^cprnrpf^r-ifp

6666666666666666666666666666666

OOoooNV>r-^o\OesooovomosmTrr^ma\
-^ir)ONOCT\OONOoo(Nmfn-Hoomoo<Nmooo
"-,W^W,W^W^^^W>~V>,^,^.".^v.>^v,-,000NTt^O^-0^0'^(NNfn^*^
-Hr40(NOmOmoooorNi'^ior^ooo-^(NTt
-Hr40(NOmOmoooorNi'^ior^ooo-^(NTt^^r^ooON-^rsjmTf>y->>o

6666666666666666666666666666666
s.

s.

s.

s.

><

>i

H H H H

8868SSS88S

>i

H H H H H

0<^-7"f^p^n-yirn--rs)Tt^aNrNjuooo'yi'^oo<s>^op-^rt-r^prnr^pTj-r^^-i
On

On
00

On
00

On
00

-^-^^^''^^"<N<NfNrv|
OnOnOnOnOvOOOOOOOO'
oocooooooo^O^O^ONO^O^O^ONO^ONONONO\O^ONO^ONO^ON

O^OONtNO\00\or-~OooONTrmmvoootNOoor^vovor-ooorNjTtr^O

p-y^fnoot^oo^oo^'-(q^wo^^wooo^^^<^lnpr^lo^oO'^uppt^T^
rnfnt~^(NvbfS^fS^-^vorn6cbvbun'^fnrnr^fslf^
NO
NO
NO
NOVOOlOl^TtTtTt'<:fTtTfTtTt'^TfTtTfT}-rt'<^'<^TtTtTt
r^
ooooooooooooooooooooooooooocoooooooooooooooooooo
00
00
00
00

p<poo'^oorsit^fNjroNpr^'^vo-;^pfpoovor~p0'^qNqNpcsiuoqN^
r*^rnt^fM^fSvbfS^6'o6vbrn6obNb'<^fNj'^666NONOOOOONO^
NO
00

r--

00

O
00

VO
00

voNOvou^TtTt-Tt-rnfnrornrnrnrn(N<N(N(N(N(N<N(Nrnro
OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOCOO

p-T'poo'^oorsr^fSfSoopr^qNvovoqNVONOoofsqNopqNr^r^'^fS'Ti-^fN

(NvbfSvOfS^6'^6o^obo(N6obNbofn<N'^-^6666666
00 00 00

poori-oofsr--r^<Nooo\"ovO'^ooqNrsqNO^^"^qsr^r^qsf^
^^fnr^fSbr^|Nbrsvb6'^d^wo^cbr^'^6^^r^^*^'^6^obr^
NO
NO
NONO>orJ-Tj-Tfmr-)r<^(NfN(N(NrN|''' -^-^'^''1-^^^
r~sO
00
00
00
000000000000000000000000000000000000000000000000
00

pr7-p'^^"^rsr^fNfSooq^wpNppr^r-pwprn'<^r;>fSop^;oop"<^^qNN^
HONNOTfcSOONt~^NOo/^rfr^rnfnc^
rOfSr^CSVOfSvbfS^6'^ONn-^00Tf
NO
NO
vo
NONOu^-^TtTfm<^mr^(NrNfsi<N '-H-^ ^
r^
,

00

^-

00

00

K. NO

-;

00

oo

NO

^-

b-^

00

oo
NO

S ^2
r-s

000000000000000000000000000000000000000000000000

5"

s'

S*

fc

s"

11.

NUMERICAL METHODS

427

right-hand side of equation (11.62) gives a value for dT^Jdt at the point A.
If equation (11.54) is now used to estimate r,+i at point D, the value
obtained will be identical with that from the ordinary step-by-step method.

This value

is

now

treated as a

improved as follows.

set

of

approximation to T,+i which can be

approximations for the temperatures at

10

10

2-0
-^

Fig. 11.12.

first
first

Radius

Comparison of

in

step

2-0

inches

by step and

iteration solution

and G are now available for calculating dT^+Jdt from a

formula analogous to (11.62). By taking the average of this rate of change
of temperature at D with the previously calculated rate at A gives a good
points D, F,

value for the rate at E.

I.e.

5T
dt

\e

2\

dt

(11.63)
dt

L)

Using this value of the rate at E and the known temperature at A, an

improved value can be obtained for the temperature at D. This leads to an
improved time derivative at D and hence to an improved value for the rate
of change of temperature at E through equation (11.63).
Both of the iteration methods should give results of equal accuracy for
equal values of corresponding increments, but the method described in
detail and used in example 2 is far more efficient for the following reasons.
The first method, which involves a little more algebra initially, and uses a
more comphcated difference equation, has much greater stability and needs

MATHEMATICAL METHODS

428

CHEMICAL ENGINEERING

IN

example considered, only two iterations

were needed at each step which means that the volume of calculations is
only double that of the ordinary step-by-step method with the same
far fewer iterations at each step. In the

The other type of iteration frequently requires four or five

on each step which doubles the work again without improving the

increments.
iterations

accuracy.

The method described

recommended.
11.7.3.

in

detail

and used

in

example 2

is

therefore

Liebmann's Method or Method of Squares for Boundary Value

Problems

This method is a natural extension of that used in Section 11.4.1 to

an ordinary differential equation of boundary value type.
The
appHcation of the method to partial differential equations is explained by
Thom and Apeltt who have suggested many improvements on the basic
Essentially, the calculation can be described as alternating between
idea.
points situated in the centres of black and white squares on a chequered
board, instead of between odd and even points along a line.
Considering the two dimensional version of Laplace's equation in
cartesian coordinates
solve

dx^

and using equation

^=

(11.64)

dy

(11.32) to replace

both derivatives by

finite difference

expressions, gives

{^xy

{^yy

where
and n count the x and y increments respectively. Rearrangement of
equation (11.65) gives

where

= r,^i,+r_i,+/^2(T,^i + T,._i)
R = ^x|^y

network as shown
problem so that Ax and

l^y are constant,

2(i+i^2^r^,

(11.66)
(11.67)

superposed on the region of the

then the points of intersection can
be designated with a circle or a cross depending upon the sum of the subscripts being odd or even. If the values are estimated at all points marked
with a cross, equation (11.66) can be used to calculate values at all points
marked with a circle. The roles of the cross and circle can then be interchanged
to enable new values to be calculated to replace the original estimates.
"
Alternate calculations between the " cross points " and " circle points
If a

in Fig.

11.13

is

Difficulties can arise at the boundaries if they are not

rectangular but they can be overcome by the use of '' fictitious points " and
linear interpolation formulae. The best solution is to use the appropriate

will give the solution.

fThom, A. and Apelt, C.

Van Nostrand, London (1961).

J.

"Field Computations

in

Engineering and Physics".

( )

) i

( )

i>

()

{)

(J

()

ik

JE

i)

m-2 m-1
Fig. 11.13.

429

NUMERICAL METHODS

11.

n-1

m +2

m+1

Classification of points in

n+1

Liebmann's method

coordinate system to suit the boundary shapes so that most if not all of the
boundaries correspond to Unes of the lattice.
An alternative procedure for difficult boundary points is to use " irregular
stars " and derive special formulae for particular boundary points. Thus,
equations (11.31, 32) were derived from Taylor's theorem on the assumption
of equal intervals; without this assumption the following equations are
approximately true.

Kxo + a) =fixo) + afXxoHWr(xo)

(11.68)

f(xo-b) =f(xo)-bf'(xo) + ibY(xo)

(11.69)

Solving equations (11.68) and (11.69) for the derivatives gives

fXxo)

J"(Xo)

b'f(xo + a) - a'fixo -b)- (b'

bf(xo

(11.70)

+ a) + af(xo -b)-(a + b)f(xo)

iab(a + b)

which can be used instead of equations

11.7.4.

- a')f(xo)

ab(a + b)

(11.31, 32)

(11.71)

where appropriate.

Relaxation Methods for Boundary Value Problems

This method which is fully described by Southwellf takes the algebra a

than Liebmann's method and gives greater freedom to the
calculator to exercise his discretion and make use of previous calculating
experience. The partial derivatives are replaced as before to give a finite
difference equation of the form
Httle further

^(^m-l,n Tln+l.n* '^m,n-l T,n+U ^m.n)

(11.72)

Equation (11.72) will be satisfied at each point in the network when the
solution has been obtained; but originally, values T* etc. can only be
" Relaxation Methods in Engineering Science ". Oxford University
t Southwell, R. V.
Press (1940). Southwell, R. V. " Relaxation Methods in Theoretical Physics '\ Oxford
University Press (1946).

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

430

estimated, and substitution into equation (11.72) will leave an error term
/?^

''

called the

^m,n

F(T^-l,ni '^m+l,n '^m,n-l ^m,n+l5 ^m, n)

The object of the method

values

r*

If

Thus

residual ".

r*

is

to " relax " the residuals to zero

is

adjusted, the alteration of

will

(11.73)

by adjusting the
be given by

dF

= --^dT:^

SR^^

(11.74)

dTlm,n

Equation (11.73) shows that

R^

is

influenced by changes in five point tem-

peratures; conversely, the adjustment of one temperature will influence the

value of five residuals according to equations of the type (11.74).

These
equations are therefore called the " influence equations ". Invariably there
is

one

coefficient

dominating the others

in the set of five influence equations

and this indicates which temperature needs to be altered by the smallest

amount to relax a particular residual The mechanics of the method can best
be iUustrated by the following example.

Example

ratio of peak velocity to average velocity

an incompressible fluid through a straight
of the lengths of the two sides of the rectangle

Determine the

3.

for the steady laminar flow of

rectangular duct.
is

The

ratio

2:1.

Solution

The Navier-Stokes equation (7.126) governs the fluid velocity, and

taking cartesian coordinates x and y measured along the edges of the
rectangular cross section, and z along the axis of the duct, the fluid velocity
will only have a z component denoted by u. The velocity will be a function
of X and y only, hence

no external
becomes
are

all

terms arising from u. yu will be zero. Since there

and the fluid flow is steady, equation (7.126)

forces acting,

--Vp + vV^u =

Because u only has a z component, the x and y components of equation

become

=
^
oy

r=o

OX

whereas the z component of equation

,-^

where a

is

II

gives

+ ^. +

III

a constant given by
1

dp

/i

oz

a=---/

IV

1.

NUMERICAL METHODS

431

Equation III is " Poisson's equation". Using equation (11.32) to replace

the derivatives in equation III as in Section 11.7.3 gives
im+l,n-'^^m,n + H m-l,n

" m,n +

i-2w + M m,n- 1
{Ayy

(Axy

Because the ratio of the lengths of the sides of the duct are in simple proHence equation V becomes
it is convenient to choose Ay = Ax,

portion,

+ w.
+ t/, +l
^m+l,n^^m-l,n"T-M^
,

where

i?^ is the residual

+ M _1 4w^ + a(Ax)

i?

VI

which has to be relaxed to zero. If the length of

is L, the largest possible choice for Ax is

JL, giving three intersections of the lattice within the rectangle as shown in

the shorter side of the rectangle

Fig. 11.14.
2

f^2,1

f^u ^.1

Fig. 11.14.

The

velocity

Primary

must be zero on

"m,0

^3.1

lattice across

duct

boundaries; hence

all

Wm.2

\l

^2,1

"0.1

=0

="4.1

VII

and equation VI becomes

Wm+ 1. 1 +m- 1. 1 -4m^. 1+iccL^

= K,

VIII

To avoid decimals and yet obtain accurate values for the velocities, it is
convenient to choose a large value for ocL^. Dimensions have not been conand putting a numerical value for ocL^ amounts to using an
arbitrary set of units for velocity, length, etc.
This does not cause any
difficulty in the interpretation of the result as will be seen later. Therefore,
sidered here,

put
ocL?

The

set

of three equations from equation VIII

3.i

i^i^i

+ Wi.i-4w2.i + 2000 =

i?2.i

+ 2000 =

i?3,i

2.i-4wi,i
not worth using the

is

2,i-4wi,i+2000

It is

IX

8000

full

relaxation

method at this

stage because equations

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

432

are quite easily solved algebraically with R^

R2

/?3

The

0.

result is

"1,1

"3.1

714

^2^1

857

XI

These three point values are obviously not sufficient to represent the
Hence, Ax is put equal to \L so that
velocity distribution adequately.
equation VI becomes

Because both central planes through the duct are planes of symmetry, it is
only necessary to calculate the velocity distribution in a quarter of the duct.
The calculation is shown in full in Table 11.9.

Table

= Ay = iQ

Velocity Distribution in

Duct (Ax

(398)

(578)

(655)

400
430

580

655

675
670

570

645

em

11.9.

(677)

20
5

40

-5

515

-20
-40

UO

XI

760
715

-J0
-130

(398)

400
430
II

30
50

UO

-20
-60

840

60

-5
-5

(867)

10

(897)

(867)

895

865
$60
840

(759)

(514)

-20

10

35
25

(578)

580
570

10

-5
-10
-20

865

IV

III

IX

X
5

(677)
(655)

20

(655)

675

20

vn

890
860

10

655

40

670

655

-10

645

10

6m

645

VI

V
VIII

4 m-

The planes of symmetry across which the velocity gradient must vanish
= 2 and m = 4. It is therefore convenient to include an
extra row of points in each direction as shown. Firstly, the velocities 715
are defined by n

1.

433

and 860 are inserted at the appropriate points as given by the eariier solution
(equation XI). The velocities at the other nodes are estimated as follows.
If the axes of x and y were rotated through 45 and a lattice spacing
A;c = Aj' = Ljl was chosen, an equation identical to VI (with a different
Using this formula with
interpretation of m and n) would be satisfied.

aijUixf-

1000 gives the value

i.i=K0 + 0+0 + 715 + 1000)

XIII

428J

estimate for Wj^i is 430.

Hence
= 2 to give jL,3 = 430. Similarly,
the

first

NUMERICAL METHODS

3.1

This value

is

reflected in the plane

= i(0+0 + 715 + 860 + 1000)

= 643J

XIV

Using both planes of symmetry.

5.1

3,1

Equation XII which

XV

645

appropriate to the lattice in Table 11.9 is used to

These velocities are put into Table 11.9
in the positions indicated in Fig. 11.14; that is, in the lower left-hand corner
of the square on the upper right-hand side of the point. The residuals
R n are also calculated from equation XII and inserted as shown in Table
is

estimate the remaining velocities.

11.9

and

Fig. 11.14.

By

partially differentiating equation

XII the influence

equations can be determined. Thus

if^.,n

l,

5R.n

5R.-Un =

5m+l,

5..-l

dR^\ =

Fig. 11.15.

Equations

shown

XVI

in Fig.

= -4

XVI

Basic relaxation pattern

are usually presented as the basic relaxation pattern as

This pattern was used in Table 11.9 to relax the

11.15.

residuals in the order

shown by

the

be noticed that the largest residual

roman numerals near

is

the points.

It will

usually relaxed at each step by altering

MATHEMATICAL METHODS

434

same

the velocity at the

point.

IN

CHEMICAL ENGINEERING

Unless the residuals at the neighbouring

points are of opposite sign to the one being relaxed, it shortens the calculation to " over-relax " the residuals by altering the velocity so much that
the sign of the residual
7?2.2

making

140

^1/2.2

is

reversed.

is

and

the largest residual initially,

45.

The

it

is

over-relaxed by

residuals are then altered according to equations

XVI

or Fig. 11.15 as shown. There is no need to calculate a residual at the

point (2, 3) since this is a repetition of the residual at (2, 1). The method is
straightforward until point (3, 2) is relaxed at the IVth step. Here, the velocity
at (5, 2) must also be adjusted by the same amount as at the point (3, 2).
This results in the residual at (4, 2) being altered by a double increment since
it is influenced by both velocity changes.
At the Vlth step, points (3, 3) and (5, 1) must both be adjusted when the
velocity at (3, 1) is altered. The results given in Table 11.9 are correct to the
nearest 5 units in velocity. The calculation was continued to the nearest
unit and this figure is shown in brackets.
full presentation of the calculation of residuals in the latter stages would only confuse Table 11.9.
The whole calculation can now be repeated after further subdivision of
the intervals.
Taking Ax = JL, the velocities given in Table 11.10 were

obtained.

Table

11.10.

Velocity Distribution in

Duct (Ax

Ay

|L)

155

246
294
309

250
408
496
524

310
516
634

349
587
725
770

374
633
785
835

389
660
822
876

398
677
843
898

401
683

671

850
905

After all residuals have been reduced to zero with a nominal tolerance,
a technique called " residual balancing " should be used. This involves
all residuals and relaxing still further until this residual sum is
within the nominal tolerance of zero. The accuracy of the result is greatly
increased by this final step.

summing

1 % between the last two

and since all velocities have moved in the same direction, the
ratio of peak to average velocity will have altered by much less than 1 %.
The change of values can be interpreted as giving a more accurate result for
the pressure drop incorporated in the value of a. Hence the figures given in

The

centre line velocity has altered by about

calculations,

Table 11.10 are

To

sufficiently accurate.

obtain the average velocity, Simpson's rule (Section 10.7.2) can be

used since it is basically more accurate than the relaxation method and thus
will not reduce the overall accuracy of the result.

NUMERICAL METHODS

11.

The average

velocity across

"m

and the average

u

any

-hi^^m,

line at

is

given by

+ 2U, 2 + 4; 3. + . 4)

velocity across the duct

2i:(4wi

constant

435

is

+ 2w2 + 4w3 + 2w4 + 4m5H-2w6 + 4w7 + M8)

Hence using the values given

in

XVII

XVIII

Table 11.10 to evaluate equations XVII and

XVIII,
u

XIX

453-4

Therefore, to the accuracy of the calculation, the ratio of the peak velocity
to the average velocity is 2.00.
In order to calculate the pressure drop along the duct, dimensions must
be considered. Equation VIII shows that aL^ must have the same dimen-

any velocity u (say u). Because equation IX selects a numerical

value for aZ^, the unit of velocity is imphcitly determined. There is no need
to derive this unit because both equations IX and XIX must be expressed
in terms of this unit. Dividing these two equations by one another will give
a dimensionless equation which is independent of the choice made in

sions as

equation IX. Thus

^_4534
aL'

"

8000

Using equation IV to eliminate a gives

dp

_
dz"
A-r-=

or

-8000/iti

453-4L2

XXI

35-3 fiu

dz

where

A is the cross sectional area of the duct and w is the superficial velocity.

The corresponding formula

for laminar flow through a cyUndrical pipe

= Snuu = 25'lfiu
A^
dz

is

XXII

Hence, the pressure drop per unit length of the rectangular duct of side
ratio 2:1 is 40 % greater than that along a cyUndrical duct of the same cross
sectional area with the same superficial velocity.
The solutions of Laplace's or Poisson's equations by relaxation methods
or Liebmann's method are comparatively simple because the coefficients in
the influence equations depicted in Fig. 11.15 are simple integers.

In more

compHcated problems, such as the solution of the Navier-Stokes equation

(7.126) for laminar flow around a sphere,t the influence equations are much
more compHcated and even vary as the calculation proceeds due to the nonlinearity of the
t Jenson, V.

equation of motion.

G. Proc. Roy, Soc. A249, 346 (1959).

436

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

The treatment of curved boundaries is similar to Liebmann's method

and either fictitious points or irregular stars can be used. The reader is
referred to the specialized books already mentioned by Thom and Apelt or
Southwell for further details of both methods.
The numerical work involved in solving partial differential equations by
the methods described in this chapter is considerable, but the advent of the
digital computer has made their use quite attractive. This development will
be discussed in Chapter 14, where it will be shown that a series of similar
numerical calculations favour the use of a computer, whereas an isolated
calculation is best performed on a desk machine.

Chapter 12

MATRICES

12.1.

Introduction

has been shown that two variables can be combined and

complex variable, and also that three variables can be
treated as components of a single vector variable. Each of these techniques

In earlier chapters

it

treated as a single

further
has resulted in the simplification of special types of problems.
generalization of this idea leads to the consideration of greater numbers of
There is frequently a
variables as a single quantity called a " matrix ".

fundamental relationship between the various constituent parts of a matrix

which leads to the variables being arranged as a rectangular array to draw
attention to these finks.

Thus

in Section 7.2.3, the stress tensor

Pxx

^xy

^xz

hx

Pyy

hz

.^zx

Tz,

Pzz

was written

(7.6)

Elements in a particular row refer to forces acting on one plane whereas

elements in a particular column refer to forces acting in one direction. In
general, the nine quantities are independent.
Students in engineering and science are introduced to the properties of
determinants in their initial courses in mathematics, and it will be recalled
that the determinant is a square array of numbers or symbols that has a
specific value. Consideration of the above stress tensor should estabfish the
fundamental difference between a matrix and the determinant which has the
same elements in corresponding positions. It is that a matrix cannot be
reduced to a specific value by multiplications between its constituent
elements, whereas a determinant is evaluated in this manner.
The rules for addition and multiphcation of matrices will be enunciated
in this chapter and the subject will be developed to include matrix algebra.
In addition the rules of matrix calculus will be introduced, and the types of
chemical engineering problems for which matrix methods are suitable will be
indicated.

M.M.C.E.

437

15

mathematical methods in chemical engineering

438

The Matrix

12.2.

The array of symbols

a 11

a 12

a 13

21

22

23

'ml

'ml

...

a In
2n

called a matrix. Each symbol such as ^, j, fif2, or (3^ is called an element

of the matrix, and the square brackets [ ] enclosing the array signifies that
it is a matrix, just as the two straight lines
enclosing a square array signify
that to be a determinant. The square brackets are thus a special part of the
notation. For operations concerning the matrix rather than its elements,
the above array can also be written in the short forms
is

A=
where the

[a.v]

of the array has to be separately indicated.

size

The matrix represented above contains

rows and n columns.

It is

therefore called a "

by matrix ", or alternatively the matrix is said to
x ". It should be noted that the matrix is always described
be of order "
in terms of " rows by columns ". That is, the number of rows is always

stated

first

and

this is

^2

61

array

is

followed by the number of columns. For example, the

a 2 by 3 matrix implying that there are two rows and

"

The committing to memory of the phrase " rows by columns

be of great help later on when matrices have to be manipulated in the

three columns.
will

solution of engineering problems.

The Row Matrix

12.2.1.

matrix having n elements arranged in a single row is a (1 x 77) matrix,

'*
row matrix " or " row vector
Thus
'inJ IS a
12 "13
11
row matrix or row vector.
The position vector of a point (Section 7.3.2) is usually described in
terms of its coordinates as [x y z] which is a particular (1x3) row matrix.
called a

12.2.2.

A
'*

The Column Matrix

matrix having

elements arranged in a single column is called a

". To adhere to convention it would

column matrix " or " column vector

be written

12.

439

MATRICES

21

but because a vertical array of elements would occupy a larger space, such
a matrix would be written {^n ^21
^mi)- The brackets { } signify
that the elements are to be arranged in a vertical column.

12.2.3.

The Unit Matrix

A matrix in which the number of rows of elements is equal to the number

of columns of elements is called a *' square matrix " and if all the elements
except those in the diagonal from the top left-hand corner to the bottom
right-hand corner are zero, the matrix is called a " diagonal matrix ".
diagonal matrix in which the diagonal elements are all unity is called the
" unit matrix ". Thus

0"

1_

the unit matrix of order three. Furthermore the non-zero elements are
situated on the " principal diagonal ". The significance of the unit matrix

is

will

be discussed later in Section 12.3.4.

12.2.4.

The Null Matrix

matrix whose elements are

denoted by 0.
12.3.

all

zero

is

called a

null matrix "

and

is

Matrix Algebra

In order to be able to use matrices in the solution of engineering problems,

manner in which the operations of addition, subtraction, multipHcation,

and division are performed must be defined. These mathematical processes
the

will

be considered in the following sub-sections.

Matrix Addition

12.3.1

This operation can only be carried out on matrices of the same order.
Thus the sum of two matrices is obtained by adding together the corresponding elements of each matrix. For example, the sum of the matrices
"1

31

and

6'

is:

15-2

MATHEMATICAL METHODS

440

Similarly, subtraction

is

IN

CHEMICAL ENGINEERING
"2

13

_8

(12.1)

Thus
lie

negative addition.

2
1

-3

-1

(12.2)

These operations can be extended to any number of matrices of the same

order, irrespective of whether the elements consist of real or complex
numbers or symbols. The sum or difference does not exist if the matrices
are of different order.
12.3.2 Scalar Multiplication
If any matrix is multiplied by a number, the elements of the matrix so
formed are the products of the number and the corresponding elements of

the original matrix.

"1

Thus
2"

"3

+ 3m

2k + 7 m

4k + 5m

6k 2m

7"
/c

+m
6.

.5

-2

(12.3)

Both k and m are scalar quantities and each operates on a matrix. For this
reason such an operation is called '' scalar multiplication ". It should be
noted from the above illustration that the coefficients k and m multiply
each element within their respective matrix, and in this respect matrices
differ from determinants.
12.3.3.

Matrix Multiplication

The product of two matrices will exist only when the number of columns
of the first matrix is equal to the number of rows in the second matrix. When
"
this condition is satisfied the two matrices are said to be " conformable
and they yield a product. Thus if A is a matrix of order (m x n) and B is
a matrix of order (n xp) the product AB will exist and will be a matrix of
order (mxp). However the product BA will not exist since (nxp) is not
conformable with (m x n). The following example will make this point
clear.

Example

1.

Evaluate the product

^3

-4

2"
3

"1

2"
8

7_

_7

9_

12.

-4

-1

2"

"1

7_
_7

441

MATRICES

2"

(3xl) + (-4x4) + (-lx0) + (2x7)

= (1x1)+

(2x4)+

(OxO) + (3x7)

(6x1)+

(5x4)+

(6x0) + (7x7)

9_

(3x2) + (-4x8) + (-lx4) + (2x9)"

(1x2)+

(2x8)+

(0x4) + (3x9)

(6x2)+

(5x8)+

(6x4) + (7x9X

-12^

30

45

75

139

II

In the above example it will be seen that the first element in the final
matrix is obtained by multiplying corresponding elements from the first
row of the first matrix and the first column of the second matrix and summing
the products thus
;

-4

[3

-1

2]

{1

7}

= 3-16-0+14 =

(12.4)

This product of a row and a column is called an " inner product ".
It can also be seen that the first matrix in the above example is of order
(3 X 4) whereas the second matrix is of order (4 x 2) and the product matrix
is of order (3 x 2).
That is

(3xj4).(4ix2)

(3x2)

(12.5)

No

product would exist if the (4 x 2) matrix was multiplied by the (3 x 2)

matrix since they are not conformable. Hence the order of terms in a matrix
product is important.
In certain cases the product of two matrices will exist irrespective of the
order in which the multiplication is carried out. For example, if the matrix
A is of order (m x n) and the matrix B is of order (n x m) the product C of

AB = C
is

a matrix of order

product

(mx\n).(n\xm)

(12.6)

= (mxm). On

of

BA = D
is

the other hand, the

a matrix of order (nx\m).(fn\xn)

{nx

(12.7)
n).

In particular, this condition

square matrices, but even with these, the product will not be
the same in general. It is thus necessary to use terminology which specifies
will exist for

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

442

Equation (12.6) can alternatively be described

the order of multiplication.

as:
(i)

(ii)

C
C

is
is

obtained by pre-multiplying B by A.
obtained by post-multiplying A by B.

when

In the exceptional case

equal,

the products of two square matrices are

i.e.

AB = BA

(12.8)

commute, or be commutable.
In Section 12.2.3 the unit matrix was shown to be a diagonal matrix in
which the elements were unity. The significance of unity is that when it is
multiplied by a number or function, the value of the number or function
remains unchanged. Therefore the definition of the unit matrix must satisfy
this condition. Thus consider the unit matrix of order three to operate on a
third order column vector
the matrices are said to

0"

(Ixxi) + (0XX2) + (0XX3)'

~^r

= (0xXi) + (lXX2) + (0xX3)

X2

(0

_^3_

1_

X Xi) + (0 X X2) + (1 X

(12.9)

JC3)_

L-^aJ

Hence the unit matrix conforms to a fundamental law of mathematics.

The unit matrix is usually given the symbol I and can be written into an
equation as such, thereby conserving space. For example equation (12.9)
would be written:
Ix = xl = X
(12.10)

The reader can check

that the matrices in equation (12.10) are commutable.

Further evidence that the unit matrix conforms to the fundamental laws of
mathematics is that

P=I
and again the reader
It is

is

(12.11)

asked to check equation (12.11).

also important to note that the product of

two matrices can be the

null matrix without either of the original matrices being null.

Example

2.

Find the two results of multiplying

1

A = -2

51

-1

-3

BA =

0_

The above property

gives rise to the

22"

18

-1

-9

same

together where

6^
3

2
but

0'

AB

and

and

11

11_

difficulties

when

division process as were encountered with vectors in Chapter

division by a square matrix

is

possible, as will be

shown

7.

defining a

However,

in Section 12.7.

MATRICES

12.

443

Determinants of Square Matrices and Matrix Products

12.4.

an array of elements which are

both
magnitude and position. The determinant of a square matrix A would be
written |A|. Obviously matrices that are not square do not possess a corresponding determinant.
It has been shown in texts on determinants that the product of two
determinants |A| and |B| of the same order (nxn) is also a determinant of
order (n x n). Furthermore the product

The determinant of a square matrix

is

identical with the corresponding elements of the matrix with respect to

|A|x|B|

|B|x|A|

say.

|C|

(12.12)

From

equation (12.12) it can be seen that the elements in |C| will be identical
with the corresponding elements in the matrix product AB. Hence the determinant of the product of two square matrices is equal to the product of their
determinants. That is,
|A| |B|

|AB|
|AB|

so that

and

|B| [A]

|BA|

(12.13)

|BA|

even though in general AB 7^ BA.

Finally if A is a square matrix and
" singular matrix". On the other hand

(12.14)

=0,

|A|
if

the matrix

|A| 5^ 0, the

is

matrix

called a
is

" non-

singular ".
12.5.

The Transpose of a Matrix

rows of a ( x m) matrix are written in the form of columns, a new

matrix of order {m x n) will be formed.
This new matrix is called the
" transpose " of the original matrix. If the original matrix is denoted by
A, the transpose will be denoted by A', the accent in the top right-hand
corner signifying that A has been transposed. Thus if
If the

0'
"2

-1

A'

4'

(12.15)

-1
6
In a similar

column

A
and

manner

it

can be seen that the transpose of a row vector

is

vector.

matrix and

its

transpose are always conformable for multiplication

if

AA' =
the matrix
12.5.1

is

(12.16)

said to be " orthogonal ".

The Transpose of the Product of Matrices

Let A be a matrix of order (m x n) and

The product of these two matrices will be

AB = C

let

be a matrix of order

{n xp).

(12.17)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

444

where C is a matrix of order (mxp) whose typical element in the

andyth column will be
Cij

= Z
k=

but the typical element in the transposed matrix will be

^ij

^Ji

^jkbki

Therefore extending equation (12.19) to

(AB)' it follows that

That

^kjf^'ik

k=l

(AB)'

clj

where

k=l

bik<j

(12.19)

k=l
all

the elements in the product

B'A'

(12.20)

two matrices

the transpose of the product of

row

(12.18)

<^ik^kj

/th

equal to the product

of their transposes taken in the reverse order. This idea can be extended to
any number of matrices that are conformable. Thus
is,

(ABC)'

12.6.

is

C'B'A'

(12.21)

Adjoint Matrices

any element of a square matrix of order (nxn), and the cofactor

ofaij in the determinant |A| is Aij, the transpose of the matrix whose elements
are made up of all the ^,yS is called the " adjoint of A
and is written
If aij is

A=

adj

(12.22)

[Aj,-]

Example. Evaluate the adjoint of the matrix

3

3'

A = -2
L

adj

A=

14

12.7.

Reciprocal of a Square Matrix

Consider a square matrix A to be non-singular and of order

The product of this matrix with its adjoint is

adj

Since the properties of determinants

Z
show

^ir^rj

(n

n).

(12.23)

that
(12.24)

r=X

12.

it

follows that

on

all

445

MATRICES

the elements in the matrix product

adj

except those

the principal diagonal will be zero, hence

1

adj

= A

or
Since

is

adj

|A|

...

A=

(12.25)

|A|l

non-singular, equation (12.25) can be divided by

is defined by

and

if

new matrix A~^

A-^ =

adj

A
(12.26)

equation (12.25) becomes

-1
AA-^=I

A "Ms

(12.27)

thus seen to be the reciprocal of the non-singular square matrix A.

Each element of

this reciprocal

matrix

is

obtained thus

Aji
(12.28)
''^=IA|

Hence matrix

accomplished by multiplying the one matrix by

As with all multiplications, the order of terms
but any matrix will commute with its own reciprocal.

division

is

the reciprocal of the other.

must be

specified,

AA-'=I = A-'A
12.7.1.

(12.29)

The Rec iprocal of the Product of Square Matrices

Consider A and B to be two square non-singular matrices of the same

Their reciprocals A~^ and B"^ will exist and thus the product of
each is

order.

(AB)(B-^A-^) = AIA-'
Therefore

B"^A~^

is

the reciprocal of the product

(AB)-'
That

is,

=AA-^ =1

(12.30)

AB, or

=B-'A-i

(12.31)

two non-singular conformable

equal to the product of their reciprocals in the reverse

the reciprocal of the product of

square matrices

is

order.

This concept can be extended to any

and in general

number of conformable square

matrices,

(ABC.XYZ)-^ =Z-'Y-'X-^..C-'B"'A-'

(12.32)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

446

The Reciprocal of a Transposed Matrix

12.7.2.

In a

manner very

above it can be shown that the

equal to the transpose of its reciprocal;

similar to that given

reciprocal of a transpose of a matrix

is

or

{k-')'
It is left

= (kr'

(12.33)

to the reader to verify this.

12.8.

The Rank and Degeneracy of a Matrix

These two important terms are employed extensively in matrix calculus

and algebra, and they must be understood in order to make use of matrices
However, the reader must be
in the solution of simultaneous equations.
reminded of other definitions, which arise elsewhere in elementary mathematics, in order to appreciate the significance of" degeneracy " and " rank ".
These will be considered first.
12.8.1.

Linear Dependence

Let Xj, ^2, X3

equation

x be

initially

a set of scalar variables that satisfy the

;'l^l+P2^2+P3^3 + ---+Pn^n

(12.34)

Pn ^re constants, some but not all of which may be zero.

Since these scalar variables satisfy equation (12.34) they are linearly dependent.

where Pi, P2

This idea can be extended to matrices in the following way. Thus let
be a set of vectors of the same order, and the zero on the
right of equation (12.34) be a null matrix of order corresponding to the
vectors. Since Pi, Pz, P3^ etc. are not all zero the vectors are hnearly dependent. However, if equation (12.34) was satisfied only when the constants
were zero the vectors would be " linearly independent ".
The vectors x^, X2, etc. could be the n columns of a (mxn) matrix X,
and if there exists a set of n scalar constants p^, pj, etc. not all zero so that
equation (12.34) is satisfied, the columns of the matrix are hnearly dependent.
In particular if the matrix X is square, its columns will be linearly dependent
only if the determinant 1X| =0. Furthermore, if |X| =
Xi, X2, X3, etc.

|X|

|X'|

(12.35)

so that the rows of X will also be linearly dependent. To sum up, if the
rows or columns of a matrix are Hnearly dependent the matrix is singular.
12.8.2.

The Minor of a Matrix

Let A be a matrix of order (m x n) in which m is less than n. The determinants formed from the m rows and any m of the n columns are called
'*
minors " of the matrix A. Furthermore if r is an integer which is less than
and all the minors of order r are zero, it follows that all the minors of order
greater than r will also be zero. Thus the value of the determinant of any

MATRICES

12.

447

minor of order (r+1) is the sum of the products of each element in the
(r+ l)th column and its cofactor. Since these cofactors are minors of order
r which are all zero, the minors of order (/+ 1) must all be zero. By repeating
this argument to the minors of order m it can be seen that all the minors of
order greater than r will be zero.
Minors of a matrix are often referred to as the first minor, second minor,
and so on. By this is meant that the minor of highest order is the first minor,
the next highest minor is the second minor, and so on to the minor that will
contain a single element only. For a square matrix of order {n x n) the first
minor will be the determinant of the matrix and of order n. The second
minor will be of order ( 1), and so on to the minor of order 1.
12.8.3.

Degeneracy

matrix of order (m x n) is said to have a degeneracy of r when at least

one of its (r+ l)th minors does not vanish, whereas all of its rth minors are
equal to zero.
For a square matrix of order {n x n) and degeneracy r this means that
there are r linear relationships between the n columns of the matrix. Thus
the columns of a singular matrix are related by a single equation of the
type (12.34), where Xj, X2, etc. are the columns of the matrix and such a
matrix is " singly degenerate ", or has a degeneracy of one.

Example.

Show

that the matrix below has a degeneracy of one.

"2

-2'
4

.5

The determinant of

this

matrix

2.

is

4-2

Hence the matrix is singular. However, all the minors of order two are
non-vanishing so that the degeneracy is one, and therefore the columns of
the matrix are related by the single relationship
8{2

Similarly the rows are related

1[2
12.8.4.

4}-6{-2

5}-7{4
by a

-2] + l[3

2}

II

2]

III

single relationship

4]-l[5

The Rank of a Matrix

The rank of a matrix

vanishing minor.

is equal to the order of the highest order nonIn the above example, the rank of the matrix

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

448

-2'

.5
is

two.

to (n

2_

For a square matrix of order n and degeneracy

the rank

r,

is

equal

r).

From

the above definition

it

will

be apparent that

the null matrix have a rank equal to or greater than one.

all

matrices except

Thus the matrix

is

of order

4,

degeneracy

3,

and rank

1.

The Sub-matrix

12.9.

It is frequently convenient in matrix calculations to divide a matrix of

high order by dotted vertical or horizontal lines to form what are termed
" sub-matrices ". Thus the matrix of order 4 below is divided into four
second order matrices.

11

12

13

14

21

22

23

24

'Ai

A2"

31

32

33

34

A3

A4.

41

42

^43

44_

where

Ai

11

A,
21

The dotted

lines are said to

^22]
*'

13

14

L"23

^24

etc.

partition " the matrix.

This technique is especially useful for handling large matrices in digital

computers, because it allows the matrix to be considered as a combination of
smaller matrices.
12.10.

Solution of Linear Algebraic Equations

Matrices are often used to express a system of linear algebraic equations

in concise form. Thus consider the set of linear algebraic equations

z.Xj

"T"

2xi

3x2 2x4

3x2

+ 2x3+

2x-y

4x-i + 2xA

X2

^^3

X4
(12.36)

12.

449

MATRICES

This set can be written in matrix notation thus

-2"

-3

or

more

X2

-4

^3

-3

-1_

.^4_

_6_

concisely

Ax =
where

"8"

"^r

A represents

vectors of order 4.

(12.37)

a square matrix of order 4 whilst

The

x and b

solution of equation (12.37)

are each

column

is

= A-^b

(12.38)

which is possible only if the reciprocal of the matrix A can be evaluated.

Consequently in this text the method of solution is confined to square matrices,
and in the particular case of equations (12.36)
"8"
"
-0-15'
-0-10
0-11
0-39

"^r
X2

-0-11

0-06

-0-10

0-04

X3

012

0-22

-0-02

-0-24

_^4.

0-50

-0-02

-0-02

-0-21_

(12.39)

_6_

whic h
Xi

l-93;

X2

0-56;

X3

0-58;

X4

1-0

It should be pointed out that a solution of equations (12.36) will be

obtained only if A and Ab are of the same rank. When this is so the set of
equations is said to be consistent. A is called the matrix of the equation, and
Ab the " augmented matrix " of the equation.

12.11.

Matrix

Series

Matrices can form series in much the same way as scalar variables, with
each term in the series containing powers of a matrix. Hence before discussing series it is proposed to consider the properties of powers of matrices.
12.11.1.

Powers of Matrices

a matrix of order {n x m), the product AA

components of the product are conformable. That is
If

is

will
if

only exist

m=

n.

continued product of a matrix will exist only if the matrix is square.

this is the case the continued product can be written in the form

AiA2A3A4...A

= A"

if

the

Thus the

When
(12.40)

Raising a matrix to a power n signifies that the matrix is to be multipHed by

itself times and not that each element in the matrix is raised to the power n.

MATHEMATICAL METHODS

450

IN

CHEMICAL ENGINEERING

For example:
'1

^3"

a^

4.

.^2

4.

'i

^3'

.^2

4.

al

+ a^a
2"3

ayQi

l3

+ Cli^A.

+ ^34

^2^3

+ ^4
(12.41)

The usual

algebraic rules for integer indices apply to matrices.

A"xA" =
(A")'' =

and

exponent of a matrix

If the

matrix

be

(12.42)

A"^

(12.43)

fractional implying that the th root of a

is

commutable with the parent. However,

may

A"^'"

sought, the resulting matrix will be of the

is

Thus

it is

same order as, and

number of roots

possible that the

but this will have to be established for each particular matrix

infinite,

whose roots are required.

12.11.2

Matrix Polynomials

Matrix

and these
example

series
series

can be formed from different powers of a square matrix,

have similar properties to the famihar scalar series. For

floX"

X + flJ =

+ aiX"-' + fl2X""^+

(12.44)

a square matrix and the coefficients ^o* ^i> ^2>

^ are scalar constants. It will be noticed that the coefficient a^ has been
multiplied by the unit matrix of the same order as the matrix X so that all
terms in equation (12.44) are conformable for addition.
.

In this series

typical.

is

is

Series of the type expressed

by equation (12.44) can contain a

finite

or

number of terms and the summation and tests for convergence

of such series are best made by the use of the Cayley-Hamilton theorem and
an

infinite

Sylvester's

However
It

is

theorem which

will

be treated in Sections 12.14.2 and 12.15.

certain properties will be dealt with here.

same manner as
by the following example.

possible to factorize matrix polynomials in the

scalar polynomials as illustrated

Example. Evaluate the matrix polynomial

X^-9X^ + 26X-24I

X=

where

The polynomial

X^-9X^ + 26X-24I

can be factorized:

(X-2I)(X-3I)(X-4I)
The

first

factor

II

is

"2

r
4

-2

0"

"0

1"

2_

^
1

III

MATRICES

12.

451

and the second and third factors are

'2

.3

0"

"3

-1

1"

4_

.0

3_

1.

"2

"4

0"

"-2

.3

4_

0.

and

IV
4

Hence the value of the polynomial

'0

r
2_

.3

Finally

it is

-1

1"

-2

is

1"

-3

3"

=
3

1,

0.

-1

i_

3.

possible for matrices to be exponents of the scalar quantity

or function, and the well

known

exponential function

X^

is

written

X'
(12.45)

where

is

a square matrix.

The matrix exponent X in e^ behaves in a similar manner to a scalar

exponent. Thus if Y is a second square matrix of the same order as X

e\-^=l

Similarly

where the unit matrix

12.12.

(12.47)

must be of the same order as X.

Differentiation

and Integration of Matrices

The elements of a matrix may be functions of an independent

variable,

or derivatives. In both cases the matrix can be differentiated with respect to

the independent variable and integrated. Each of these operations will now

be discussed.
12.12.1.

Differentiation

Consider a square matrix Y whose elements are functions of an independent variable x. A small change in the value of x will bring about a corresponding change in the value of the matrix Y, and the limit of the change in
Y for an infinitesimal change in x is the derivative of the matrix with respect
to X. This is written in the normal way (dY/dx).
The differential coefficient of the product of two or more matrices with
respect to an independent variable is obtained in a manner similar to that of
the differential coefficient of the product of scalar functions, but in the case
of the derivative of a matrix product the order of the matrices in the product
must be maintained. Thus

(XYZ) =
dx

^YZ4-X^Z+XYf
dx
dx
dx

(12.48)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

452

Finally the differentiation of the exponential function of a matrix

similar to the differentiation of a scalar exponential function.

iie--)
as

X+^+^ +

That

...=Xe'^'

is

is,

(12.49)

would be expected.

12.12.2. Integration

Integration of a matrix whose elements are derivatives or functions of an

independent variable is accomplished by integrating each element with
respect to the independent variable. Thus if the typical element yij{x) of the
matrix Y is a function of x, then the new element in the integrated matrix
will be

Jy
An

Jx= [J^oW^x]

(12.50)

integration series that arises frequently in the solution of systems of

linear differential equations is

as " the matrizant "

known

which by

definition

is

MS(A) =

1 A(xi) dx^

Xl

JA(xi) J A(X2) Jx^ dx^

X

X2

Xl

+ J A(xi) J

A(X2)

A(X3) dx^ dx2dx^...

(12.51)

The form of the matrizant given

of httle value in calculations because of its complexity.
However, if the number of integration terms in equation (12.51) is large, the
value of the matrizant over specified hmits approximates to

where MJ(A)

is

the matrizant of the matrix A.

in equation (12.51)

is

M5(A)= limn[l + A(x,_i)n^oo

where n

is

the matrix

the

number of integrals

in equation (12.51).

are constants the matrizant

M5(A) =

If all the elements in

becomes

^^^^ + ^'(^-''oy +
12.13.

(12.52)

n_

^A(.-x)

(12.53)

Lambda-Matrices

Matrices of this type arise in many fields of applied mathematics where

systems of linear differential equations are to be solved. Therefore they
deserve consideration. A A-matrix is a square matrix in which the elements
are functions of a scalar parameter /. Such a matrix is completely defined
by its order, rank, and degree. Its order depends on the number of rows or
columns as described above. Its rank is the order of the highest nonvanishing minor irrespective of the value assigned to X, and finally the degree
of a 2-matrix is the highest degree of A in any of the elements of such a matrix.

What

Example.

is

and degree of the A-matrix:

the order, rank

+A

A^

1X^-IX^\

matrix

is

this

/I

of rank

is

in

this

2.

any element

matrix

-1

2-3"

+A

-)}

2+

A^-3A+1

1-2

)}

A^+l

'1

)}

is

2 and therefore the >l-matrix

-1

is

"

0"

2^

-1

_-3

1_

"1

o_

_1

other matrices. However, the reciprocal of a 2-matrix

of 2 and is not therefore a 2-matrix.

A
F(2)

-3"

-1

The rank of the 2-matrix can be easily verified from equation

The addition, multiplication and division of 2-matrices

12.13.1.

is

2.

The expansion of

1-/1

The highest degree of X

A'

A+1

series.

therefore the A-matrix

of degree

-)?

matrix into a matrix

the determinant of the matrix A(A) is zero, and hence this
However, its minors of order two are non-vanishing,
singular.

and expand

When

453

MATRICES

12.

is

0_
I.

is

similar to

normally independent

The Determinantal Equation

2-matrix is usually represented by the symbol f(2), its adjoint by

its determinant by A(2) or det(2). These are related by the equation

and

f(2)F(2)

F(2)f(2)

A(2)I

(12.54)

Equation (12.54) can be differentiated with respect to 2 to give

[f(2)F(2)]

=-

[F(2)f(2)]

=-[A(2)]

dX

dk

(12.55)

dX

and the differentiation can be repeated until the derivatives vanish.

The equation formed by equating (12.54) to zero, that is
A(2)
is

called the " determinantal equation "

(12.56)

and

its

roots which

may be

distinct

Thus let 2^ be a
matrix formed by substitution
will be singular.
However, if

or multiple, are of great importance in matrix analysis.

root of the determinantal equation; then the

of Xp into the original matrix is f(2p) and it
2

is

a multiple root

f(2

need not have a degeneracy greater than one.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

454

Thus the A-matrix

(X-a)

'

(X-a)
(X-b)_
has a root X = b, and a double root X = a, but is of degeneracy one. When
of degeneracy one the adjoint F(Xp) is of unit rank.
When {(Xp) has a degeneracy of q, at least q roots of the determinantal
equation are equal. Furthermore the adjoint matrix and its derivatives up
to and including F^"^(Ap) are all null.

{(Xp) is

12.14.

let

The Characterisfic Equation

Let A be a square matrix of order n whose elements are all constants, and
X be a column vector whose elements are related by the equation

Ax = Xx
where A is a
form thus

scalar coefficient.

is

and only

written in the matrix

a square A-matrix of order

(12.58) will be non-trivial if

may be

Equation (12.57)
(A-;iI)x

where (A XT)

(12.57)

(12.58)

The

n.

solution to equation

if

|A-aI|=0

(12.59)

Hence X can only take values that will satisfy equation (12.59) and for this
reason equation (12.59) is called the " characteristic equation " of the matrix
A. Its roots Xp are called the " latent roots " or the " eigenvalues " (see
Section 8.7.1) of A.

When A- All
I

both

12.14.1.

and

its

is

zero, |(A AI)'|

is

zero,

and

since

(A XT)'

= A'-AI,

transpose have the same latent roots.

Diagonal Canonical Form

For each root of equation (12.59) there will be a corresponding set of

values for the elements in the column vector x, and each set of values constitute
a vector of the latent root. For this reason the vector corresponding to the
latent root is called a *' latent column vector " and there will be a latent
column vector for each latent root. Furthermore, if the latent roots are
distinct, the latent column vectors are linearly independent.
Therefore
consider a non-singular square matrix A of order n, and let the latent roots
of A be distinct. In addition let kj, k2, kj,
k be the latent column
vectors corresponding to the latent roots of A, and K the square matrix
comprising all the column vectors. K will be of order n and non-singular.
.

Also

Aki =
and

;iiki;

Ak2 = A,k2;

AK = KA

...;

AK = KK

(12.60)
(12.61)

12.

MATRICES

455

where
...

^1

k^

...

A=

(12.62)

A.

...

K-^AK = K-^KA = A

Therefore

(12.63)

showing that the matrix A can be reduced to what is known

canonical form ".
In a similar manner, latent row vectors can be determined which in turn
can be used to reduce A to the diagonal canonical form.
as " diagonal

Example

Evaluate the latent roots and reduce the following matrix to

diagonal form.

The

"2-3

characteristic equation

r
3

-4_

can be expressed in the form

|I>1-A|=0
from

in order to eliminate the negative signs

Thus

in matrix

in front of the latent roots.

form

-r

1-2

-3

A-1

-3

-2

X+A

expanding the determinant of equation

II

II

8]=0
A3 = -2

(;i-l)[(>l-2)(A + 4) +

from which

When

When

>li

Ai

^2

0,

-10

-10

-3

-3

11

11

1,

the adjoint

-6

^2

0;

the adjoint

-13

1;

is

-10"
-101

-3 =

"-10"
[-101

[1

1]

-3

IV

11_

11_
is

"-r

- 9"

=
13

III

9_

i_

[6

13

9]

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

456

and when I

From

A3

= - 2,

the adjoint

-12

-9

21

-1

-3

-9 =

[3

3]

VI

7_

matrix

is

-4"

-3

21_

andK"^ = ~

-3

11

"-4"

-12"

equations IV, V, and VI the

-10

is

3'

-12

-26

-18

-3

-1

-3

VII

6
'J

so that

2-3

1'

-5

-10

-4'

-3

-3

11

~ ~6

12.14.2.

01
1

-12

-26

-3

-2J

-1

3"

-18

VIII

-3

The Cayley-Hamilton Theorem

This theorem can be stated as follows: Any square matrix satisfies its
own characteristic equation.
The proof of this theorem will not be given however, any interested reader
is recommended to see " Elementary Matrices " by Fraser, Duncan, &
;

Collar.!

Here

it

will suffice to illustrate its application to the

above third

order matrix.

Example

satisfies its

The

2.

own

Show

that the square matrix

2-3

-5

-4

characteristic equation.

characteristic equation of the given matrix

|i;i-A|

By

is

=X{k-\){X + 2) =

the Cayley-Hamilton theorem

A(A-I)(A-f2I)
t Fraser, R. A.,

Duncan, W.

University Press (1960).

J.

and Collar, A. R.

=
:

II

Elementary Matrices

",

Cambridge

MATRICES

12.

where

in equation II

2-3

is

That

the null matrix.

-3

457
is,

-3

313 303 333

-5

-4J \_-5

-5J [-5

The Cayley-Hamilton theorem can be used to

Thus since a square matrix satisfies its own

ro

[0

calculate

Example

The

power of the matrix

to give the desired

Calculate A"^

3.

if

characteristic equation

-1
.-2

III

0_

powers of matrices.

characteristic equation, the

characteristic equation containing the matrix in place of

and rearranged

o"

000

-2J

>l

can be expanded

as illustrated below.

3"

is

\U-A\ = X^-3X + 2 =
Replacing 1 by

and rearranging

gives

A2 = 3A-2I
Multiplying by

II

gives

A^ = 3A2-2A =

9A-6I-2A

= 7A-6I

III

Repeating the process of multiplying by

and using equation

II gives

A^ = 15A-14I
Hence using equation

IV

I,

-29

45"

-30

46

This process can be repeated to any positive or negative integer power of

a square matrix.

12.15.

From

the above section

A"
where

Theorem

Sylvester's
it

will

be seen that in general

.rt-2
= p,A"-'+p,A"-'+p,A"-'
+ ...+pJ

(12.64)

a square matrix and Pu P2,

p are scalar coefficients. Let the
series on the right of equation (12.64) be represented by P(A), then applying
Lagrange's interpolation formula (Section 9.4.3)
is

P(A)

/A-lA,)(A-U3)(...)(A-I^)
'

ai-A2)(Ai-A3)(...)(Ai-A)
p..
^

'^

(A-I^)(A-U3)(...)(A-U)
(^2-^l)(A2-^3)(..-)(^2-^)

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

458

ri(A-iA,)
Pi^iY^

(12.65)

where Aj, A2,

^ can be chosen as the roots of the characteristic equation
which are assumed to be distinct in the case under consideration. Equation
(12.65) is an expression of " Sylvester's theorem " but the extended products
in the numerator and denominator are sometimes inconvenient computationally. However, they can be simplified as follows.
Denoting the A-matrix
A-I/l = f(A)
(12.66)

equation (12.54) shows that

(A - UiM^i)
since Xi

is

one

A(Xi)l

(12.67)

of the distinct roots of the characteristic equation.

Further-

more,

A-Ui^O

andF(/l.)#0

(12.68)

Similarly,

(A-Ud n (A-Uj) =

where

it

(12.69)

n(A-IA,.)#0

(12.70)

f\{A-Uj) = CF(Xd

(12.71)

but

Hence

n (A-Uj) =

can be shown that

can be any diagonal matrix with constant elements. The matrices

and (12.69) are commutative in each case; therefore C

in equations (12.67)

(a) times the unit matrix. By selecting square matrices

of different orders in which all elements are unity it can be shown that
a = (-1)"-^ Thus

must be some constant

n (A-lXj) = (-iy-'F(Xd = n
where n

is

(12.72)

the order of the square matrix.

The denominator of equation

A(A)

fl

;=1

(12.65) can be simplified as follows.

a- A) = (-1)" J=l
fl a-

A,.)

~ [AW] |,= = (- D" fl i^i-^j) = A'W

but

Hence

f(Xj)

Sylvester's

theorem becomes

{Ai)

(12.73)

(12.74)

12.

MATRICES

459

f(>^) is defined as lA A instead of by equation (12.66) the only effect

change some of the signs. The negative sign in front of equation (12.75)
has to be removed, but equations (12.65) and (12.72) are not changed.

If

is

to

Finally,

when

the characteristic equation has multiple roots Sylvester's

theorem takes a much more compHcated form due to A'(>^,) vanishing for
the repeated roots. Equations (12.65) and (12.75) can be modified by using
L'Hopital's rule until the indeterminate fractions are resolved, but the work
involved shows no significant advantage over evaluating the original
polynomial (12.64) directly. However, this form is known as the " confluent
form " and is derived by Fraser, Duncan, & Collar.f

12.16.

Transformation of Matrices

In many calculations involving matrices it is advantageous to rearrange

the matrix into an " equivalent form " having the same rank, degeneracy,

and

latent roots as the original matrix, but being

particular problem.
(i)
(ii)

(iii)

more

suitable for the

This can be accomplished by

Interchange of any two columns or rows,

Addition of any multiple of any column or row to any other column
or row.
Multiplication of any column or row by a scalar.

12.16.1. Interchange

of Columns or Rows

Consider the post-multiplication of a third order matrix

modified unit matrix

by the

0"

"1

_o

0_

columns interchange
11

12

13

'21

22

23

32

33j

0"

0_

"ii

13

12

21

23

22

_31

33

(12.76)

a 32J

In equation (12.76) it can be seen that the second and third columns of the
matrix A have been interchanged. It will also be readily seen that if the second
and third rows of the unit matrix had been interchanged and the matrix A premultiplied by this modified unit matrix the result would have been the
interchange of the second and third rows of A.
t Fraser, R. A.,

Duncan, W.

University Press (1960).

J.

and CoUar, A. R. " Elementary Matrices

",

Cambridge

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

460

It should be observed that the modified unit matrix

transpose and that

01

01

is

the

same

as

its

o^

(12.77)

and therefore the modified unit matrix has the properties of

unity.

It is

equivalent to the unit matrix.

These operations can be extended to matrices of any order.

12.16.2.

Addition of any Column or Row, or a Multiple of any Column or

Ron\ to any other Column or Row

Consider the product

11

12

13

21

22

23

32

33.

0"

''a 11

(/^11+12)

13

!2i

(^21+22)

23

!31

(^5^31

+ ^32)

(12.78)

a 33.

on the left of
equation (12.78) had the effect of adding p times the first column to the
second column. Pre-multiplication of A by this matrix would have brought
about the addition of p times the first row to the second row.
The matrix which operates on A is a unit matrix to which has been
added the element ^ in its first row and second column. Its determinant is
Post-multiplication of the matrix

by the second matrix

unity.
12.16.3.

Multiplication of any

Column

or

Row

by a Scalar

Consider the product

11

L"31

12

13

^22

23

32

33.

0'

y
1

'11

ycii2

13

2l

yaii

23

.31

732

33

(12.79)
J

second column of A has been multiby the scalar y. This was accomplished by exchanging the central unit
element for the scalar y. If the matrix A had been pre-multiplied by this
matrix the effect would have been to multiply the second row by the scalar y.
The determinant of the converted unit matrix is y.

The

result of this operation is that the

plied

12.16.4.

Equivalent Matrices

matrices are said to be " equivalent " to pne another if it is possible

to pass from one to the other by a series of transformations of the types
presented above.

Two

The above definition of equivalent matrices includes A-matrices. However,

with this latter type of matrix the matrices that bring about the trans-

461

MATRICES

12.

formations must have determinants that are independent of X but

finite. Thus if fi(/l) and fiW are two equivalent ;i-matrices, then if
i^(A)

and

will

[^1

Afi(A)B

(12.80)

be 2-matrices whose determinants will each be constant.

Quadratic Form

12.17.

An

still

expression of the type

'2

^3

^nj

11

12

fll3

21

22

23

31

32

33

-^1

^3

3n

(12.81)

called a " quadratic

n3

n2

!nl

a homogeneous function of the second

degree. The coefficient of XiXj in the quadratic form where / =^ j is (a,7 + ji).
For example, the conic
is

form

".

It is

\a
a

ax^ + 2hxy-\-by^

[x

h'\
h

\x

y]

(12.82)

can be written more concisely in the quadratic form.

x'Ax

(12.83)

whilst the equation

aj^xl

+ a2xl + a2xl + ,..ax^ =

(12.84)

would be written
L^l

^2

^3

^nj

fll

...

^2
^3

xT^]x

0'

...

X2

...

X3
(12.85)

...

where the symbol [A] indicates that the matrix

is

a diagonal matrix.

Example. Determine the nature of the quadric

6x^

+ 3y^-2z^ + 4xy =

and evaluate the lengths of the semi axes and equations of the principal
planes.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

462

The equation

can be written in matrix form

"6

z]

0'

latent roots of

14

= x'Ax

If

-2
The

from

are obtained

6-X

3-X

'=0

III

-(2 + X)
or

+ 2)(/1-2)(A-7) =

(a

IV

latent roots are X^ = 2; ^2 = 2; X^

vector kj corresponding to the root A^ = 2 is

and the

AKj

= -2

/C21

-2_

Ai.
6kii+2k2i = -2/cii
or

2/C11

The

latent

column

/cn"

"^n"

1.

= '^'^l

Aj[ Kj^

0"

2
or

+ 3/C21 =

-2i;:2i

VI

^21
.^31_

VII

>

2/C31 = 2/C31
from which /c^j = /:2i = 0, and /tji is arbitrary. It is normal to choose orthogonal column vectors which satisfy equation (12. 16), and thus /csj = 1. Therefore, the latent column vector corresponding to the latent root X^ = 2 is
{0

1}.

In a similar manner, the latent

root

X2

is

column vector corresponding

-2/V5

{l/\/5

0},

and

corresponding

to

to the latent

X^

is

{2/V5' I/V5 0}.

Then by equation (12.63)

r
1/V5

-2/V5

2/V5

i/Vs

o_

"6

0'

"0

I/V5

-2/V5

-2

_1

"-2

2/V5-

i/Vs

0'

2
_

7_

VI II

12.

463

MATRICES

= KX
=
X {X Y Z}
K KX = X
K'x

Applying the orthogonal transformation x

then

or

i/Vs

-2/V5

2/V5

i/Vs

X.

[X~

2x
Y'

XI

z\

-7-

^
=Z

-/-

XII

II

(KX)'A(KX)

and by Section

^-^=

Furthermore, from equation

IX

"xl

or

where

XIII

14

12.5.1

or

X'K'AKX = X'AX =

14

XIV

-2X^ + 2Y^ + 1Z^ =

14

XV

or in standard form

X^

Y^

Z^

XVI

a hyperboloid of one sheet with semiaxes \l 1 '.\l 1 \\J 1.

The principal planes are A^ = 0, Y = 0, and Z = when referred to the
new axes. With respect to the original axes, the principal planes are

which

is

x-2y =
12.18.

0',

2x + j

0;

Application to the Solution of Differential Equations

Matrices can be applied to the solution of systems of differential

equations, and in this respect they are of great utiHty in the solution of

Hence the essential features of the application technique will be presented in this section.
Consider a system of linear differential equations expressing relationships
between a number of dependent variables y^, y2, yz,
y and an independent variable x. These equations can be written in conventional form

engineering problems.

/ii(^)yi+/i2W>'2 + .-.+/iWj

f2i(D)y,+f22(D)y2 + ...+f2niD)y

= Mx)

fni(D)y, +f2(D)y2

...

+fnn{D)y

(i>i(x)

(12.86)

= Ux)

or written in condensed matrix form

fWW = [/oW]W = {0W}

(12.87)

MATHEMATICAL METHODS

464

CHEMICAL ENGINEERING

IN

In equations (12.86) and (12.87) /,y(D) is a polynomial of the differential

operator having constant coefficients. The matrix of differential operators
{(D) is analogous to the >^.-matrix and in the determination of the complementary function the technique depends upon the substitution of ). for
D. However, certain nomenclature are required and in addition to the terms
for 2-matrices given in Section 12.13 the following are necessary.

{(D)

is

called the i)-matrix.

|f(Z))| is

called the Z)-determinant

The highest degree of

order of the system.
12.18.1.

and

is

frequently expressed A(D).

any of the elements of f(D)

in

is

said to be the

Solution by Conversion to an Equivalent System

In some systems of differential equations it is possible to transform

form that will permit ready solution of the differenThis is accompHshed by pre-multiplying the Z)-matrix by
tial equations.
another non-singular Z)-matrix of the same order whose jD-determinant is
free from the differential operator D.
The following example will clarify
the D-msLtrix into such a

the method.

Example. Solve the system of equations

D^y^

+ {D^ + 3D-2)y2 =
D^y^ Dyj =

These equations can be put

D^

(D'

-D

[y,

Pre-multiplying both sides of equation

sinx

form thus

in matrix

+ 3Z)-2)1

P^

sm X

by

-D

-D

D^ + 2

II

gives
1

-D

D^

-D

D^ + 2

D^

(D'

+ 3D

2)'

>i

-D

-D

D^ + 2

III

2D'

-D

.y2

.)
(2D2 + 3D-2)

'yi'

-(2D^ + 3D')).

.yi.

{2D^ + ZD-2)y2

D sin X
IV

-Dx + (D^ + 2)smx

- cosx

2D^yi-2D^y2-3D^y2 = -1 + sinx
Solving equation

y2

smx

VI

by the methods of Chapter 2 gives

Ae~^''-hBe^''-^x-l + ^(4cosx

3sinx)

VII

12.

Integrating equation

465

MATRICES

VI twice and then

for yj

from equation substituting

VII gives
y^

C-\-Ex ix'

where A, B, C, and

E are

Ae'^'' + 2Be^'' + -^{3cosx-2ismx)

VIII

the four arbitrary constants required by this third

order system of equations.

In the above example the matrix
1

-D

-D

D^ + 2

but guided by the aim that all but one of the dependent
variables were to be eliminated from one of the equations. This enabled
that variable to be evaluated and inserted into the other equation which was
subsequently solved by conventional methods. When there are a large number
of dependent variables it is necessary to convert the D-matrix into triangular
form by pre-multiphcation by another D-matrix whose determinant is a
constant. The transformed system must be of the type

was obtained by

7ii(o)

trial,

/12(D)

/13(D)

fln(D)

/22(D)

/23(D)

f2n(D)

/33(D)

yi
yz
(12.88)

it can be seen that the last row is a function of ;^ only,

row above is a function of >^ and >^_i only. Hence the equation
of the last row will give an expression for y^ which will be inserted into the
equation of the row above to give an expression for j_i and so on until

In equation (12.88)
whilst the

the system has been solved.

12.19.

Solutions of Systems of Linear Differential Equations

Systems of linear differential equations can be solved in a similar manner

to single linear differential equations by establishing the complementary
function and the particular integral. However, with systems of equations
the process will be much more complicated because of the inter-relationship
between the dependent variables and the effect of one solution on the particular integrals of the others. Each part of the complete solution will now
be considered separately.
12.19.1.

in

The Complementary Function

A system of linear differential equations has been written in matrix form

equation (12.87) and the complementary function of this equation will be

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

466

obtained by solving

when {^(x)}

is

The

That

zero.

f(D){y}

is

by solving the equation

(12.89)

solution of equation (12.89) will depend

D-matrix, and each latent root

is

=
on the

latent roots of the

"
said to contribute a " constituent solution

complementary function. Hence the solution of equation (12.89)

depends upon whether the latent roots of f(Z)) are distinct or multiple. These
to the

be considered separately.
Distinct latent roots. Let X^ be any distinct root of
by substituting X for D in the D-matrix. Then since

will

(i)

f(D) e^'-* F(A,)

e^'^"

f(A,)F(A,)

^{Xr) is a solution of equation (12.89), but as

can be written in the form
e^*"*

F(A,)

A(>1)

obtained

(12.90)

X^ is

a simple root F(A^)

{MM

(12.91)

so that the constituent solution corresponding to the distinct root X^

arbitrary multiple of
(ii)

Then

an

Let

Multiple latent roots.

the matrices

X^

be any one of s repeated latent roots.

if

Wo(x,4) =

^'^"F(A,)

WJx,A,) = ;g.e^-F(A.)]=e-(A^,yp(A.)

and
it

is

e'^'"^k^.

can be shown by a similar procedure to that given for distinct roots that

f(D){y}=f(D)W,(x,AJ

Wp(x, X^ all satisfy the system of differential

{s\) for obvious reasons.
convenient to write the matrix Wp(x, AJ in the form e^'''Zp{x, X^

so that Wo(x, AJ; Wi(x, X^\

equations. In all the above/?
.

It is

(12.92)

where

P^^ '-\K) Kp-1)^'f

^-U)
^^-^ +
+ ^^^ ^^:;
U^As) = F^(A,) + ^^^--^
1!
2!

...

x^F(A.)

(12.93)

and there will be s matrices of Z corresponding to the set of repeated roots.

Hence since each matrix will have n columns (the order of i{D)) there will
be ns constituent solutions, but of these only s solutions will be hnearly
independent.
Hence for a root X^ from s repeated roots the constituent
solution will be
{y}

and the

= {^i.W

/c2sW

U^)} e'^'

(12.94)

columns of the type represented by equation (12.94) corresponding

may be chosen to be proportional to any s linearly
independent columns of the matrices
s

to the s equal roots

Zi(x,A,)

=
=

FW +

Zp(x, X,)

'(X,)

Zo(x,^.)

F(A,)

xF(A,)

^^^''

'

x'F(X,)

467

MATRICES

12.

Since p = sl the elements in the column vector of equation (12.94) will

be polynomials in x of degree (s-l). These column vectors are called
" modal columns " and the matrix formed from aU the n roots of f(D) is

/C12W
k22{x)

/C21W

k2n(x)

K(x)

and

is

known

(12.95)

as the "

modal matrix

Those columns obtained from

".

distinct

roots will have constant elements.

Finally if the exponential functions e^"^ are collected into a diagonal
matrix M{x) where
?^^*

...

,X2X

Mx
M(x) =

(12.96)

fA.fiX

and {C}

is a column of n arbitrary constants, the complementary function

of the system is

{y}

K(x) Mix) {C}

Example. Solve the following system of

(12.97)

differential equations.

(D^ + 4D + 3)y2-{D^ + D + 2)y,

The above equations can be written

(D'-3D + 2)

(D'-l)

_-{D^ + D + 2)
The

>l-matrix obtained

rom which
The

in matrix

(D2

+ 4D + 3)J

from the Z)-matrix

form thus

[yr

is

-(A-1)(A-2)

(A-l)(A + l)j

.-(A2 + A + 2)

(A+1)(A + 3)J

A(A)

(A

[^2.

- 1)^ (A + 1) (A + 2)

II

latent roots are

= -1;

A2=-2;

^4

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

468

The

respective adjoints are for

>?.i

and X^
'0'

0"

F(/i)

[1

III
.2

6,

_2_

-1

-3'

12.

"-r

[1

3]

IV
and the adjoints

for

F(A3)

and A4 are

y^.3

"8

"2"

0"

[4

=
_4

F'(^)

.1_

2A- 3_

-2"

'1

.3

-1.

.1.

[3

-1]

=
k4(x) =

k3(x)

{2

1}

{2

l}

and the complementary function

1

-1

2(1

(1

.2

;.= i

and A4 the modal columns are

for the latent roots ^3

2.

'6

VI
_2A +

and

0]

-2/'

"22 + 4

VII

l}x

{2

VIII

is

+ x)-

0'

-X

^-2x

+ ^).

rc.1
C2

IX
C^

e"

e'_

or the conventional result

>^i=2[C3 + C4(l + x)]e^-C,e-^and

>;2

.Q.

is

= 2Cie-" + 4C2e-'" + [C3 + C4(l + x)]e"

X
XI

12.19.2 The Particular Solution

To complete
it is

the solution of the system of Hnear differential equations

necessary to obtain a particular solution of equation (12.87). Essentially

this solution will

be
{y,}

F(D)
f(D)-'{0(x)}

A(D)

{/-W}

(12.98)

Hence it is necessary to be able to evaluate the last term in equation (12.98).

Therefore consider the case i^(.v) = Re"* where R is a column of constants.

Then

(n)
which

is

A((?) i= 0.

r(D)R,x^,.FW,

\{Dy

the particular solution

when ^(x)

(12.99)

A(0)
is

exponential provided that

When

A{9)

469

MATRICES

12.

the following procedure

must be carried

out.

Let the

particular solution be

{y^}

where b

is

constant.

b[F'(^) + xF(0)] e^"

(12.100)

Then
f(D)b[F'(^)

+ xF(^)] e^* =

Re^*

(12.101)

from which
he'''{(D

+ e)['(0) + x(e)'] = Re'"

(12.102)

Using Taylor's theorem to expand f(D-\-6), and operating with the powers
of

upon X

gives

be'"[f(0)F'(^) + f' (^)F(0) + xf(^)F(0)]

=
f \e)F{e) + f{9W(0) =
f(^)F(0)

Since

and

Re'"

(12.103)

A(0)

A'(^)

equation (12.103) gives the value

b

Hence the

particular solution

{yp}

R/A'(0)

(12.104)

is

^^

[F'(0)

+ xF(0)] e'"

(12.105)

Other forms for (p(x) can be treated in a similar manner to the corresponding forms of (l>(x) in the inverse operator method (Section 2.5.5).
Finally, the complete solution is the sum of the complementary function
and the particular integral as for Hnear differential equations.

Example.
battery of A'' stirred tank reactors are arranged to operate
isothermally in series. Each reactor has a volume of V ft^ and is equipped
with a perfect agitator so that the composition of the reactor effluent is the

same as the tank contents. If initially the tanks contain pure solvent only,
and at a time designated ^o ^ ft^/min of a reactant A of concentration
Cq lb moles/ft^ are fed to the first tank, estimate the time required for the

Mh

concentration of
leaving the
tank to be
lb moles/ft^.
The stoichiometry of the reaction is represented by the equation
fci

k2

A:^B^t:C
k[

/C2

first order and the feed to the reactor does not contain
any product B or C, but is assumed to contain a catalyst that initiates the
reaction as soon as the feed enters the first reactor.

All the reactions are

Solution

The concentration of A
The concentration of B
The concentration of C
M.M.C.E.

will
will
will

be written C^
be written C^
be written Q.
16

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

470

Then a mass balance on any tank n

be

will

V^ = qC^.,-qC-rV
where r is the rate of chemical reaction.
For component A this gives
II

component B,

for

^^ =

^Q.-i-(^Q.n+^/c;Q+F/c2Q.-F/c,C^.-F/c,'Cc.)

at

III

and

for

component C,

qCrn-i-(qCrn+Vk',Cc,-Vk,CsJ

IV

dt

Dividing equations
holding time, gives

C^,

dC^,n
dt

dCn
-

II, III,

dt

and IV by

and calhng

{V/q), 9 the

nominal

p-(^+^l)Q.n + ^iQ.

^^ - Q +

k2

+ K^ Cs, + hC^, + k',Cc,

VI

and

dCc

-=

dt

~^" ~\e'^ V ^C.n + kzCB,,

VII

Equations V, VI, and VII can be written in matrix form as follows

'A,n

c A,n-l

CB.n

n-1

lQ.hj

C,

(iie+k,)

-k\

-k,

(iie-\-k2+k[)
-/C2

n-lj

'

-k'2
(l/^

+ /ci).

"Q,'

Q,
Cc.n\
VIII

or

more concisely
dC

C_i

dt

-GC

IX

where C and C_i are column vectors and G is a third order square matrix
Equation IX can be applied to the first reactor thus

with constant elements.

dt

The

solution of equation

X is

C,=^" +
where K^

is

471

MATRICES

12.

XI

e-<='Ki

the constant of integration.

When

Cj

0,

/.

K,= -G-'Col9

/.

C,

XII

=\(l-e-^')G-'Co

XIII

Applying equation IX to the second tank and substituting the expression for
Ci given by equation XIII gives

C2=^^-^^'re-^+e-G'K2
When

0,

C2

XIV

and

K2

= G

C2

= ^(I-e"'^^-Gre-G0G-'Co

Cq/^

XV

Applying equation IX to the third tank, substituting the expression for C2

and evaluating the constant of integration gives
1

^3
Extending

this to

r.

_..

(GO

tank

N by

Sylvester's

XVI

G-^Cn

2!

1!

comparing equations XIII, XV, and

concentration of the effluent from tank A^

By

(GO'

_^^,

,-^

XVI

the

is

theorem (Section 12.15)

e-Gr^_ y e-^^'^

XVIII

on the assumption that all the latent roots of G are distinct, and therefore
equation XVII can be written in the very useful form that

Equation XIX was initially derived by Acrivos and Amundsonf who also
present equations for the transient behaviour of plate extractors, absorbers,

and

distillation equipment. The reader interested in the further application

of matrix analysis to the solution of stagewise processes is recommended to
read this article.
t Acrivos, A.

and Amundson, N. R. I.E.C.

Al, 1533 (1955).

472

mathematical methods
12.20.

in

chemical engineering

Conclusions

This chapter has attempted to present the elements of matrix algebra

and calculus which has had to be concise because of the limited space
available for this topic.
The reader may draw the conclusion that the
arithmetic involved will be exorbitant. This is true, but when one considers
that these methods can be used in conjunction with hign speed digital computers it will be appreciated that the arithmetic need not prevent these
methods being apphed. Equation XIX in the above example looks quite
formidable, but when used in conjunction with a computer the value of C^v
for different process conditions and a large number of tanks is quickly
obtained.

Chapter 13

OPTIMIZATION
13.1.

Introduction

Optimization in the chemical process industries infers the selection of

equipment and operating conditions for the production of a given material
so that the profit will be a maximum. This could be interpreted as meaning

maximum
minimum

output of a particular substance for a given capital outlay, or

investment for a specified production rate. The former is a
mathematical problem of evaluating the appropriate values of a set of
variables to maximize a dependent variable, whereas the latter may be
considered to be one of locating a minimum value. However, in terms of the
profit, both types of problem are maximization problems, and the solution
of both is generally accompHshed by means of an economic balance between
capital and operating costs. Such a balance can be represented as shown in
Fig. 13.1 in which the capital, operating and total costs are plotted against

the

the

f(slze

Economic balance of process

Fig. 13.1.

/ (size

of equipment), where

of equipment)

of equipment) implies some function of

could be the actual size of the equipment;
number of pieces of equipment such as number of stirred tank reactors in a
reactor battery or frames in a filter press or some parameter related to the
size of the equipment such as the reflux ratio in a distillation unit, or the
the size of the equipment.

f (size
It

solvent to feed ratio in a solvent extraction process.

In certain circumstances it is unnecessary to make a complete economic
balance involving all the variables of a process. Thus, in Section 9.6.3,
473

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

474
the

number of

stirred

tank reactors for the

maximum

output of a product

for the chemical reaction

A-^B^C

(13.1)

was evaluated without cost data. This was possible because in the example
quoted the size of the reaction vessels was specified. If vessel size was
included as a parameter in the analysis, an economic balance would have been
necessary. That does not mean to say that in the example quoted a " non
problem " was solved because in the chemical industry it is possible that the
equipment will already exist, and the engineer is expected to find the optimum
operating conditions. Hence problems involving two variables are worth
considering initially, even though in the majority of problems encountered
more than two variables will be required to describe the process, or operation
However, before attempting this, it is necessary to classify
adequately.
diff'erent

types of optimization.

13.2.

Types of Optimization

Optimization in the chemical field can be divided into two classes which
have been defined by Himsworthf as
(i)
(ii)

13.2.1.

Static optimization

Dynamic

optimization.

Static Optimization

is the establishment of the most suitable steady state

operation conditions of the process. These include optimum size of equipment and production levels, in addition to temperatures, pressures and
flow rates.
They would be established by setting up the best possible
mathematical model of the process which is maximized by some suitable
technique to give the most favourable operating conditions. These conditions would be nominal conditions and would not take into account
fluctuations in the process about these nominal conditions.

Static optimization

13.2.2.

Dynamic Optimization

Dynamic optimization

is the establishment of the best procedure for

correcting the fluctuations in the process conditions about the nominal
conditions established in the static optimization analysis. It requires a know-

ledge of the dynamic characteristics of the equipment and also necessitates

predicting the best
corrected.

way

In reality

it

in
is

which a change in the process conditions can be

an extension of the automatic control analysis of

the process.

In addition to the above types of optimization there are the operational

research aspects, but these are common to many industries and will not be
discussed in this text.
t

Himsworth,

F. R.

Trans. Inst.

Chem. Eng. 40, 345

(1962).

13.

13.3.

475

optimization

Analytical Procedures

a process can be described by a small number of variables, it is

possible to fmd the maxima or minima by differentiating and equating the
These are then solved in order to find the
resulting expressions to zero.
optimum conditions. However, as the number of variables increases, the

When

solution of the resulting equations becomes prohibitive. Therefore analytical

optimization procedures have limited applicability, but they deserve brief

consideration.

Problems Involving Two Variables

13.3.1.

On

optimum conditions of a
maxima of an equation containing
one independent and one dependent variable. For example the optimum
number of stirred tanks was evaluated by solution of such an equation in
Section 9.6.3 and has been referred to above. The reader will be famihar
with the technique of differentiating and equating to zero to find the maxima
rare occasions

it is

possible to establish the

process or operation by determining the

or minima and

need not be repeated here. However, it must be emphasized

will in all probabiHty be a " curtailed optimum ",
and should be treated cautiously because one or two parameters have been
that such an

it

optimum

specified arbitrarily.
13.3.2.

Problems Involving Three Variables

Many engineering operations of a simple type can be optimized by solution

of an equation containing two independent variables. Thus
be represented by the equation

let

an operation

z=f{x,y)

(13.2)

where x and y are the two independent variables. Equation (13.2) can be
expressed in three dimensions by a surface in which the maximum value of
z will be represented by the peak of a mountain. Similarly the minimum will
be the lowest point of the bowl in the surface. Mathematically this means
that for z to be a maximum at a point (x, y) it is obvious that the function
f{x, y) must be a maximum for displacements away from the point in any
direction. That is, the conditions
dz

-=

ox

and

dz

(13.3)

oy

must be satisfied simultaneously, which means that the tangent plane to

the surface expressed by equation (13.2) is horizontal. In the case of functions
of one independent variable, the first derivative vanishes at a maximum, a
minimum, or a point of inflexion; these being resolved by reference to the
second derivative. Similarly in the case of two independent variables, there

may be

a true

maximum

in the surface

where

all

sections exhibit a

maximum,

minimum, or the two orthogonal sections may be any combination of

maximum, minimum, and point of inflexion. The special combination of a
maximum with respect to x and a minimum with respect to y is termed a

a true

MATHEMATICAL METHODS

476

Fig. 13.2

IN

CHEMICAL ENGINEERING

Saddle point

" saddle point " and is illustrated in Fig. 13.2. To ensure that a particular
point is actually a maximum or minimum, it is necessary to fulfil the conditions stated below.
Consider Fig. 13.3 in which z is represented as a function of position over
the xy plane. Let P be any point in the plane and let the value of z change

Fig. 13.3.

at different rates as the function

If

PQ

is

Directional derivatives

moves away from P

one direction, the derivative of z

j.^

Q^P

in different directions.

in that direction

may

be defined as

r /(^>>')Q-/(^,y)p 1
I

PQ

(13.4)
J

477

OPTIMIZATION

13.

where PQ is the absolute length between the points P and Q.

Ss and in addition let
f(x.y)Q-f(x,y), = dz

Then

PQ

the derivative of z in the direction

Let this be
(13.5)

is

Hm
dz

Sz

But

dz

^
= Sx + Sy

oy

ox
dz

Q^p

ds

but from

Sz

dz Sx

dz Sy

ds

dx Ss

dy Ss

Fig. 13.3

= cos a

and

-^r-

sin

(13.7)

ds

Ss

dz
dz
dz

= - cosa + ^ sina

_^

hence

ox

^^

(13.8)

ds

oy

where dz/dx and dz/dy are the derivatives in the x and y directions. The
condition for a maximum or minimum is that equation (13.3) is satisfied
irrespective of the value of a. Differentiating equation (13.8) with respect to
s gives

d^z
^ 2
ds

d I

d fdz\
J /
^
ds\ds/

ds\

d^z dx

= cosa 2
dx

dz

sina

ox

d^z

dz

cosa +

oy

dy

dx dy ds

ds

dx dy ds

d (dz\
since ~T\'~r)

d^z dy
+ sina^:;-2

dx

d^z

+ cosa--r- + sina- -

oy

^ ^^

(13.9)

ds

dz
is

the total differential of

(13.7) give the derivatives of

with respect to

-7-

x and y with respect

But equations

s.

to s in terms of a,

therefore

^ =
For any value of
z to be a

of

a.

cos^a:

:^

>

a, d^zjds'^

maximum. That

Thus

2 sin a cos

is

|-

sin

a;-,

dxdy

dx^

ds^

for z to be a

minimum and

d^z/ds^ must have the

(13.10)

dy^

same

d'^zjds'^

<

for

sign for all values

letting

_d^

^""''dxdy'

substituting into equation (13.10)

^=

d^z

_^

_i!i

^"""ax^'

^''~dp

and dividing by

z^^ gives

[(cos a z^^

sin a

z^yf

+ iz^^Zyy-z^/) sin^ a]

(13.11)

'XX

It

can be seen from equation (13.11) that

if z^^Zy,,

> z^/

the terms inside

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

478

the square bracket will be greater than zero for all values of a,
will

have the same sign as

be a maximum.

If z^x is negative, z will

If z^x

is

and d^z/ds^

z^^.

positive, z will be

minimum.

In the special case z^^Zyy

z^^, there will be some values for a which

exhibit a point of inflexion whereas all other values of a give a consistent

maximum

or minimum. If z^^Zyy < z^^, the surface has a saddle point.

The above principles of maxima and minima have been used by Jenson
and Jeffreyst for the evaluation of the optimum number of stages in solvent
extraction processes. The analysis for counter current processes is as follows.
Consider a feed solution containing Xq lb solute per lb of solute-free

of a countercurrent extraction process at the

Let the process contain A'' stages and
let an extracting solvent be fed into stage A^ at the rate of B Ib/h on a solutefree basis. The solvent passes countercurrent to the feed liquor through the
plant, finally emerging as the final extract from stage 1. The spent feed will
be discharged from stage TV with a solute concentration X^. The carrier
liquor of the feed will be assumed to be immiscible in the extracting solvent.
Assume that the feed rate and the solute content of the feed are constant
and let the unit cost of the feed in any system of currency be y per lb of solute.
Let the total hourly cost of each stage including operation, fixed charges,
and the depreciation of capital be a. Finally let the cost of the solvent,
including recovery from the final extract and make-up of spillage losses, be
P per lb of solvent. Then on the basis of one hour's operation:
liquor to be fed to stage
rate of

on a

Ib/h

solute-free basis.

Cost of feed

is

Cost of solvent
Similarly

product

if

the value of

is

-0 lb

is

yCX^

(13.12)

^B

(13.13)

of extract product, the total value of the

is

IC{X^-X^)
and the hourly

profit

from the process

is

(13.14)

given by

P = AC(Xo-X;v)-(aiV + /?B + yCXo)

But X^

is

related to

^N = ^0
[

where S

mB

and

(13.15)

Xq by

is

^J^rf ]

"(see Section 9.6.3)

(13.16)

the distribution ratio.

Substitution of equation (13.16) into (13.15) gives

P = XCX
in

s^-i
U^+i-i_

which the variables are

t Jenson, V. G.

and

N and

Jeffreys,

G. V.

[yCXo + aN + pB]

B.

The maximum value of P

Brit.

Chem. Eng.

6,

676 (1960).

(13.17)

is

obtained

479

OPTIMIZATION

13.

by the procedure presented above. Thus

dP

dN

]'"S- =
= ACJoS^[
(si/l)2

(13.18)

From which

_
icJ^"
g

5^^(5-1)1115
(13.19)

(5'^-'^-l)'

and

dP

^^+1
i\cN-i_/'\r 1^/'c^^_1^c^ /-mS'\
N(S^^'-l)S^-'-(N+l)(S^-l)S"
l
I

,^^

_^
(13.20)

or

= ^

r(5^-'^+N-iV5-5)"
(13.21)

N+l
1\2
(5'"^^-l)

Am An

In equations (13.19) and (13.21) the cost terms have been separated from the
process terms. They are two complex simultaneous algebraic equations that
will be difficult to solve analytically, but since A'' is a whole number, an
approximate but sufficiently accurate solution can be obtained as follows.
The right-hand sides of equations (13.19) and (13.21) can be evaluated
for

any pair of values of 5 and

dimensionless cost groups

5 and

A^ to give a numerical value for each of the

aC^q

and

Hence by taking a

series

of

AtviXq

repeating the calculations, a contour chart can be

Such
group using 5 as ordinate and
as abscissa.
contour charts have been prepared by the above authors for values of A^
ranging from
to 20 and values of S between
and 2; these are shown in
Figs. 13.4 and 13.5.
In a particular example the basic cost terms would be known enabling
the cost groups to be evaluated, and for these values equation (13.19) is
satisfied along one curve in Fig. 13.4 and equation (13.21) satisfied along a
curve in Fig. 13.5. Both equations can only be satisfied simultaneously at
the point of intersection of these two chosen curves. This point is easily
located by superimposing the two contour charts as shown in Fig. 13.6 and

pairs of values of

drawn

for each cost

the coordinates of the point of intersection give the modified solvent to feed

and number of ideal stages for optimum operation.

Figure 13.6 is an accurate version of the graphical solution in which the
values of 5 have been restricted to between
and 1.0 since for conventional
solvent extraction this is considered to be the most applicable range.
The procedure applied to the optimization of counter flow extraction
using Fig. 13.6 is perfectly general and the figure has universal application
for the evaluation of the optimum solvent to feed ratio and number of ideal
stages. The technique employed of using contour charts can be extended
to other fields of chemical engineering.

ratio

MATHEMATICAL METHODS

480

IN

CHEMICAL ENGINEERING

2-On

Fig. 13.4.

5
Fig. 13.5.

iO

15

Contour chart of equation

10

20
(13.19)

20

15

Contour chart of equation

(13.21)

3.

OPTIMIZATION

481

5%

by weight solution of acetaldehyde in

If the annual
capital and operating costs per stage are estimated to be 1600, the cost of
supplying, pumping and regenerating the solvent from the final extract is
estimated to be 4-5 per ton, and the value of the acetaldehyde is taken to
be 155 per ton; what is the optimum number of stages and quantity of
Example.

toluene

is

8000 Ib/h of a

to be treated with water to extract the acetaldehyde.

solvent required for the process.

The distribution ratio of weight of
acetaldehyde in water to acetaldehyde in toluene may be considered constant
at 2-2.

Solution

5%

feed solution gives

CXo = 400
a

4s.

12

Xq

0-053.

Ib/h.

per day: p

0-0402s. per lb:

Fig. 13.6.

Optimization chart

l-384s. per lb.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

482

and

0-0072

XCXo
From

-^ =

0-251

hnXo

Fig. 13.6

N=

10 or 11 stages

S=0-9 =

3840 Ib/h of solvent

Problems Involving Four Variables

13.3.3.

The

presented above can be extended to four

or minimum values of

analytical procedure

variables.

That

is,

maximum

the

w=f(x,y,z)

(13.22)

can be located by differentiating equation (13.22) and equating to zero.

dw =
when

df

^dx +

df
-:^dy

ox

oy

Thus

df

+ -fdz =

(13.23)

oz

the conditions

df

df

/ = 0;

/ = 0;

oy

ox

df

/=

(13.24)

oz

are satisfied.

In problems containing four or more variables, the analytical solution

becomes increasingly complex and impracticable. Therefore, other techniques
of optimization must be applied, and these will be considered in later sections

of

this chapter.

13.3.4. Optimization with

In

a Restrictive Condition

some optimization problems a

relationship

must be

satisfied

by the

In vapour-Hquid equihbria for example, the temperature, pressure, and compositions of liquid and vapour are related by the

independent variables.

One

boiling point condition.

eliminated analytically, but this

alternative

method

is

variable, say temperature, can always be

may

not be algebraically convenient, and an

available.

Thus consider equation

(13.22) with the restrictive condition

g{x,y.z)

(13.25)

Differentiating equation (13.25) gives

^_^rf,

OX

^_^rf,

oy

+ ^?,, =

(13.26)

oz

the optimum point, equations (13.23) and (13.26) must be satisfied for
displacements in any direction, and in particular for dx = 0. Hence the
equations

At

^dy + ^dz =

(13.27)

oy

oz

^Idy +

=
fdz
dz

(13.28)

dy

OPTIMIZATION

13.

must be

consistent.

This

is

only possible

=
f + ,fi
oy

By

relaxing the conditions that

if

and|^ + A^-^=0
oz

oy

it

483

dx

(13.29)

oz

and imposing the condition dy

0,

0,

follows that

|^
OX

+ A^ =

13.30)

ox

Equations (13.29) and (13.30) can be obtained by optimizing the function

W=f(x,y,z) + Xg(x,y,z)
when

the

optimum point

is

(13.31)

given by the solution of the equations

dx

ox

dy

dy

dz

dz

g{x, y,z)

(13.32)

which contain the four unknowns (x, y, z, X), X is known as a " Lagrange
The above method can be extended to two restrictive conditions by introducing two arbitrary multipliers.
multiplier ".

13.4.

The Method of Steepest Ascent

The method of steepest ascent is one of the best known methods of

optimization when the number of variables involved is large. The procedure
was developed by Box and his co-workers. t Essentially the method consists
of defining the criterion to be maximized as the " response " of the process.
This may be the profit, the yield, or the output from a process and will
depend on a number of factors such as temperature, concentration, pressure,
etc. The relationship between the response and these factors can be written
in the

form
;/

which

is

=/(xi,X2,X3,...,x)

called the '"response function

(13.33)

".

The response function can be plotted in terms of two of the factors to

give what is known as the " response surface ".
Such a surface will be
undulating and reaching the optimum will correspond to climbing the
highest hill on the surface. Generally speaking the greatest gradient will be
in the direction of the maximum, and therefore if the steepest ascent can be
t Box,

G. E.

P.

and Wilson, K. B.

/. Statistical

Soc. 1,

B13

(1951).

MATHEMATICAL METHODS

484

IN

CHEMICAL ENGINEERING

XI

Response contour

Fig. 13.7.

located on a response surface,

response.

However

it is

it

map

will lead in the direction

of the

optimum

very difficult to represent a surface on paper, con-

sequently it is advantageous to plot the surface in the form of lines of equal

response on a graph whose coordinates are two of the variables. These lines
are called "response contours " and are shown in Fig. 13.7. The line of
steepest ascent starting from any point is obtained by trial by drawing hnes
perpendicular to the contours. These lines will show the extent to which the
different factors must be varied in order to give a maximum increase in the
response.
This procedure will only direct the way to the optimum conditions since
at the

maximum

is zero, and
becomes more

the gradient

the line of steepest ascent

as the

maximum

0-3

0-6

0-4

o
E

<

0-2

0-2

0-4

NaOCI
Fig. 13.8.

is

difficult to predict.

0-6

0-8

1-0

concentration

Yield contours of chloramine synthesis

approached

Hence

this

13.

OPTIMIZATION

485

method alone will not locate the maximum but will rapidly show the engineer
the region of optimum conditions. This technique was employed by Jeffreys,
EUis and Whartonf in estabhshing the optimum conditions for the preparation of chloramine from sodium hypochlorite and ammonia. Some of
their results are shown in Fig. 13.8 where it can be seen that yield contours
have been drawn in between the experimental results denoted by points. It
can be seen that the

maximum

centration irrespective of the

13.5.

yield

is

ammonia

obtained at low hypochlorite conconcentration.

The Sequential Simplex Method

This method is an extension of the steepest ascent method and was proposed by Himsworth.l It differs from the above method in that no attempt
is made to find the hnes of steepest ascent, but rather that a rapid determination is made of a direction which is steep although not necessarily the steepest,
so that moves are made frequently in a favourable direction.
For a process involving two variables the procedure is as follows. In
Fig. 13.9, three points are selected fairly close together in such a manner

Fig. 13.9.

The

sequential simplex

method of optimization

form the

vertices of an equilateral triangle. Let these be points

Following this, three further points are selected such that each
one forms an equilateral triangle with any two of the first three points. If
the response surface is locally a plane or nearly so, one of the possible new
points will give a higher result than the other two, and furthermore the point
4 will be a " mirror image " to the lowest point in the first triangle. Hence

that they
1, 2,

and

3.

after the first three trials the point giving the lowest results

is

rejected

and

t Jeffreys, G. V., Ellis, S. R. M. and Wharton, T. H. " Rashig Synthesis of Hydrazine ".
LE.C. (Process Design Section) to be published.
X Himsworth, F. R. Trans. Inst. Chem. Eng. 40, 345 (1962).

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

486

replaced by the mirror image point forming the second triangle. This conis repeated until the summit is reached. Thus as long as the

struction process

surface is sloping and reasonably plane over the area of a particular triangle,
each step in the construction leads to a more favourable region and the path
taken from the starting point will zig-zag about the line of steepest ascent.
For a system containing more than two variables, the procedure will be
similar except that (k + l) trial points will be made, where k is the number of
variables. The most favourable of these points is retained to replace the
appropriate member of the original set, thus completing a move towards the
optimum. The (A:+ 1) points correspond to the vertices of a regular simplex
in k dimensions. Hence the name of the optimizing technique.
It is claimed that this method has the following advantages:
(i)

The

are elementary requiring no mathematical or

knowledge,
Each move is determined by the previous result and therefore human
judgement is ehminated.
It is possible to add a variable at any time,
A numerical measure of the response is not required. It is only
necessary to rank the result and discard the lowest.
calculations

statistical

(ii)

(iii)

(iv)

13.6.

Dynamic Programming

This method is probably the most powerful method of optimization and

has been developed by Bellman.f It finds considerable application in the
optimization of multi-stage processes with no feedback such as stirred tank
reactor batteries, crossflow extractors and whole chemical processes. Thus
dynamic programming formulation reduces a classical problem containing
variables to A^ problems containing one variable, and when this technique
is used in conjunction with high speed computers it offers a powerful method
of solving " multi-dimensional " constrained optimization problems encountered in chemical engineering.
In this section it is only possible to introduce the elements of dynamic
programming, and therefore any reader wishing to acquire a greater knowledge of this optimization technique should consult Bellman's text book
"
referred to above, and also " The Optimal Design of Chemical Reactors
by Aris.l However, for an understanding of the basis of this technique
certain terms and definitions must be understood and these will be considered
initially.
Thus consider a multi-stage chemical engineering process, as for
example a cross flow extraction plant into which a feed of known composition
is fed continuously at a constant rate.
This feed enters stage 1 and mass
transfer occurs to produce an extract and raffinate. The ralfinate will pass
to the next stage for further treatment and will subsequently emerge from

t Bellman, R., " Dynamic Programming ". Princetown University Press (1957).
" The Optimal Design of Chemical Reactors ". Academic Press (1960).
t Aris, R.,

13.

487

OPTIMIZATION

Stage A^ with considerably reduced concentration of solute, it is hoped.

Extraction will occur in each stage and the extent of extraction will depend
on the interfacial area (a function of agitation), on the solubihty and the
temperature, etc., which are variables selected by the chemical engineer. That
is

the selection of the process variables calls for decisions to be taken at each

and these decisions affect the change in composition of the raffinate

produced by any one stage. Since the raffinate passes through each stage in
turn these decisions affect the state of the final raffinate from the process.
The set of all decisions taken is called the " operating pohcy " or simply the
pohcy and the set of decisions that optimizes the extraction process is called
stage,

" the optimal policy

".

Bellmant stated that the essential condition of an optimal policy was

that whatever the transformation that took place in the first stage of an
stage process, the remaining stages must make use of an optimal (A^ 1)stage policy with respect to the state of the product from the first stage, if
the whole process is to operate at optimum conditions. Furthermore, by

varying the operating conditions of the first stage systematically and employing the optimal (A^ l)-stage policy for the remaining stages it will be
possible to find the optimum policy for all
By
stages of the process.
applying these concepts to a one-stage process, then a two-stage process and
so on until all the A'' stages have been considered, it is possible to estabhsh
the optimal pohcy for any number of stages in a process provided that there
is no feedback.

The advantage of analyzing an optimization problem

in this

way

is

that

only necessary to vary the operating parameters of one stage at a time.

For example, if there are p operating variables to each stage of a process,

it is

and the

difficulty of evaluating the state of the product from a stage for one
of conditions of the p operating variables is measured by a constant A
the determination of the optimum operating conditions for the whole process
for a given feed state will involve NA^ mathematical operations by dynamic
programming, whereas the classical solution of the same problem would
require A^^ mathematical operations. This constitutes considerable saving
of mathematical effort and computing time.
set

Bellman and KalabaJ

illustrated the principles of dynamic promanner by evaluating the minimum energy required
compression of a gas from Pq to P^f in an A^ stage compressor with
interstage coohng. Thus consider the operation of a gas compressor

Aris,

gramming
for the

perfect

in a simple

containing A'^ stages with perfect intercoohng so that the temperature of the
gas is the same at the entrance to each stage. Let the fixed charges on the

compressor depend on the number of stages, but be independent of the interstage pressures employed. Furthermore let the cost of coohng be a fixed
charge included as part of the stage cost. Then the problem becomes the

t Bellman, R., op. cit.

X Aris, R., Bellman, R.

95 (1960).

and Kalaba, R., " C.E.P. Symposium Series

".

56,

No.

31,

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

488

determination of the interstage pressure ratio for which the energy consumption is a minimum. Thus the total energy consumed by an TV stage

compressor

E,

is

nRTy
y-1 l\Po

(y-i)/y

K=

Let

(y-D/y

--(^)

(j)

nRTy
7-1

+
a

y-1

(13.34)

Pi
r,-

Pi-i

which on substitution into equation (13.34) gives

EN = K(^Y.Ji-N)
Also

(13.35)

ri rj r^

(13.36)

Po

i=l

For one stage

^1,1= PnIPo
and
which

E,
is

= K{rl-l)

(13.37)

the only solution.

For two stages

Pi

Po

and

The only

Pjv

2, 2

2. 1

Pi
(13.38)

-[^(S"variable

is

p^, hence
(13.39)

"

Pi \Pi/

LPi VPo/

^Pi

(13.40)

Po

Pi

policy is to choose rj^i = r2,2 ^iid this gives an optimum energy

consumption denoted by/2(r) where r from equation (13.36) is the pressure
ratio across the two stages.

The optimal

For three stages

r,..-'-^

3.2

Po

The energy consumption

^^3,3

>

in the first stage will

3,1

P2

Pi

be

^^(^5.1-1)

(13.41)

but the optimal policy for the other two stages has already been determined,
and the optimal energy consumption is

{^<t)

13.

The optimal

489

OPTIMIZATION

from

three-stage policy will be obtained

Mr) = Min [X(r3, ?- l)+/2(r/r3.

'3.

where

rj

is

(13.42)

0]

the only variable.

For k stages.
The optimal poHcy for (kl) stages will have been determined, and the
optimal energy consumption for the latter (k 1) stages will be/^_ lirjr^^).
The energy consumed in the first stage will be

E,,,^K{r,^l-\)

(13.43)

hence the optimal k stage pohcy will be determined by

A(r)

Min[X(r,, r-l)+A-i(r/r,. J\

(13.44)

By repeated appHcation of equation (13.44), the A^ dimensional problem

one dimensional
with no feedback has been converted into a sequence of
problems. Generally at this point, solution of the equations would be made
by means of a digital computer, but in this particular problem an analytical

is possible which can be found in most texts on chemical engineering

thermodynamics. t The solution will show that the pressure will be the same
for each stage.
To conclude this introduction to dynamic programming, consider the

solution

optimization of a cross flow extraction process.

Example. In a cross flow extraction process, a solute B is to be extracted

its solution in a parent liquid C by means of a solvent A which is
completely immiscible with C. The feed rate to the process is to be q Ib/h
Ib/h of solvent are to be distributed between the
and
stages of the
process in order to obtain the maximum amount of extract product. The cost
of pumping, regenerating and providing make-up solvent may be taken to be

form

a fraction (A) of the value of the extract.

The equilibrium

B between

relation of the solute

the phases

is

Solution

mass balance over stage n

in the process

^(x_i-x)
Since the phases are immiscible q

Since the cost of pumping,

t

Dodge,

B., "

etc.

W>'

constant; therefore for any stage n

x_i-x

so that

(1944).

is

is

r>;

of the solvent

is

let

III

to be a fraction X of the value

Chemical Engineering Thermodynamics

".

McGraw

Hill,

New York

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

490

of the extract, the problem can be treated as one in which the quantity of
solvent fed to each stage is such that the profit is a maximum. That is the
profit

is

Pn(xq)

IV
1

V
1

VI
N

Now

P^{xo)=^MaxY,v(y-X)

let

Vn

VII

Then the functional equation corresponding

programming is that

to the formalism of

dynamic
VIII

where

^1

and

From

Poixo)

=^0-1^1^1

yi

IX

equation IX,

mvi

+1

and yi
"

mui

Substituting into equation VII gives for stage

Pi(xo)

mX(

"

XI

+l

laxkf^

= Max

XII

Differentiating with respect to Vi gives

dvi

or

XIII

+l
(mi^i + l)^
I
mxo X(rnv^ + 1)^ =

\mvi

XIV

from which, with the aid of equation XI

XV

mX(
Substituting the terms in equations

Pi(xo)

XV into

equation VII with

Xo - 2^l~xJJm

+ Xjm

Using the functional equation VIII on stage 2 gives

N=

gives

XVI

13.

whence by

OPTIMIZATION

differentiating equation

XVII

in a

491

manner

similar to that already

presented,

"^^

1/^,--'
VmX m

and

F2(^o)

The flow

ratio V2

^>=
=

1/'
V m

^0 - 3

>'i

^lx~)?
.

v;;i^

IX

XVIII

XIX

obtained from the functional equation VIII by considering

poUcy with a feed of
Then by repeating the procedure given above, it can be

is

the second stage to be using an optimal one-stage

composition x^.

shown

that

V
That

is,

nrX

for the two-stage policy the ratio of solvent to raffinate rate to each

stages, the optimal

stage must be the same, or extending this concept to
poHcy is obtained by dividing the total quantity of solvent available equally
between the A'' stages.
The above problem was presented by Aris et a!.'\ and extended to more
complex crossflow extractions in which the solvent and the parent liquid
in the feed to the process are partially miscible, and the equilibrium is not
represented by a straight Hne. Solution of such problems can only be solved
by computer and these authors did in fact use a Univac Scientific Computer

Model

1103.

t Aris, R.,

Rudd, D.

F.

and Amundson, N. R. Chem. Eng.

Sci. 12,

88 (1960).

Chapier 14

COMPUTERS
14.1.

Introduction

In engineering mathematics, the need for numerical calculation constantly

arises. For instance in the method of Frobenius and separation of variables,
terms of an infinite series have to be evaluated for many values of the variables.
In addition, the solution of difference equations by graphical methods
(Section 9.7.1) is a less accurate but more convenient way than the corres-

ponding numerical method of successive substitution. Much numerical

work is involved in inverting and multiplying matrices, but the most arduous
Because
computation occurs in the numerical methods of Chapter 11.
arithmetical operations are basically simple, and desk calculating machines
and slide rules are in everyday use, the complete mechanization of arithmetic
has been a natural consequence. The development has taken place along
two channels leading to analogue computers and digital computers. Whereas
the analogue computer is a development of the shde rule principle, the
digital computer is a development of the desk calculating machine. The
slide rule operates by representing each number by a geometrical length, a
particular length on one scale always representing the same series of digits.
The basic scales are logarithmic, but most shde rules have double logarithmic
scales, and scales with varying lengths representing the logarithmic unit
(i.e. squares, cubes). The principle is to replace a number by an equivalent
length, and to calculate in terms of these lengths.
In the general purpose
analogue computer, each number is replaced by an equivalent electrical
potential, and the calculation is then performed electrically in terms of these
potentials.

The digital computer is a development of the desk calculating machine

which is capable of storing its own calculating instructions in the form of a
program.
Speeds have been increased enormously by using electronic
valves and electrical signals instead of revolution counters and mechanical
linkages, and improved electronic techniques are constantly increasing the
speed of calculation

The

still

further.

between the two computers can be illustrated by

The fraction J can in
with the desk machine.
principle be located exactly on a slide rule; only human error hmits the
accuracy. However, the fraction must be approximated by 0-33333333 on
an eight-figure desk machine and there is thus a physical Hmit to the accuracy.
Although the fraction exists on a slide rule but not on a calculating machine,
no-one would suggest that the shde rule was more accurate. This same
essential difference

comparing the

slide rule

492

14.

COMPUTERS

493

argument applies to a comparison of analogue and digital computers the

analogue computer uses a continuous number scale, whereas the digital
computer uses a discrete number scale.
Analogue and digital computers also use different modes of calculation
the analogue calculates in parallel whereas the digital operates in series.
In the analogue method, all dependent variables are continuously available
and can be displayed graphically on a multibeam oscilloscope, but the
digital computer can only perform a set sequence of commands as stored
in its program and can only offer discrete values for one variable at a time.
Attempts are being made to develop parallel digital computers with multiple
arithmetic units, and series analogue computers with memory faciHties.
This is leading naturally to the production of hybrid computers.
The great advantage of automatic calculation is that once an analogue
has been constructed or a program has been written, a whole set of problems
with a range of values for many parameters can be solved very quickly,
enabling the optimum choice of parameters to be made. Also, particular
characteristics of the behaviour of the system can be associated with various
controUing parameters thus helping in the instrumentation and choice of
;

control system for a plant.

Passive

14.2.

Analogue Computers

computer to understand, and its mots

an inter-connected arrangement of electrical impedances.
The differential equation relating the potential at any point in the network
to time is chosen to be of the same form as the differential equation it is
desired to solve. Provided that the boundary conditions are also analogous,
the behaviour of the voltage at a point will be identical to the behaviour of
the dependent variable at the analogous point.
This

is

by

far the simplest type of

common form

is

V\M
R

n+1

r^AAA
R

Typical section of passive analogue circuit

Fig. 14.1.

Consider, for example, the section of an electrical circuit depicted in

14.1.
Denoting the electric potential at point n by e, and using

Fig.

Kirchoff's law

-e
o
^n+1o

K
or

o
e
-1

i^

"-e
o

=C

(14.1)

dt

= ^^(e+i
fe_^,-2e. +
2e

dt

Ao
de

e.^.')
en -i)

(14.2)

494

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

which expresses the rate of change of potential at as a function of the

Comparing equations (14.2) and (11.52),
it can be seen that e will have the same behaviour as r if
potentials at neighbouring points.

RC =

(Ax)2/a

(14.3)

Equation (14.3) is thus the equation linking the one dimensional heat conduction equation to its electrical analogue (Fig. 14.1), and both sides of
the equation have the dimensions of time. It is not essential that equation
(14.3) should be satisfied for the use of the analogue, but if

RC =

2(Ax)^/a

(14.4)

the electrical potentials will alter at half the rate of the corresponding temperatures. In moving from equation (14.3) to (14.4), the time scale of the

analogue has been modified.

This technique enables slow processes to be

and fast processes to be retarded.
Whilst the circuit of Fig. 14.1 is suitable within the range of the problem,
special circuits must be constructed to terminate the repeated circuit in
conformity with the boundary conditions. The terminal circuits for two
boundary conditions are illustrated in Fig. 14.2. The closing of the switch
accelerated for investigation purposes,

R/2

hMArrWA
5-

Fig. 14.2.

Analogue

C/2.

circuit for

boundary conditions

change in the electric potential at the end of the circuit

which corresponds to a sudden change in temperature at one face of a large
slab. The reason for the resistor of value R/l is as follows. In the analogue,
electrical resistance represents reciprocal thermal conductivity, whereas
electrical condensers represent thermal capacity.
R and C thus represent
the thermal resistance and capacity of a space increment.
Considering a
symmetrical system in which both faces of the slab are subject to the same
step change; circuit (a) would terminate both ends of the system. To
maintain the correct total resistance and capacity within the system and
maintain symmetry, the first resistor in the chain must have half of the
resistance of any other member.
Figure 14.2(b) shows the termination of the circuit corresponding to an
insulated face. A similar argument to the above explains the condenser of
in (a) causes a step

half value.

The analogue
the system.

is used by alternately closing the switch and discharging

Using a potentiometer circuit, the potential at any point can be

14.

COMPUTERS

495

determined as a function of time, or an oscilloscope can be used to display

The benefits from using the analogue are the ease and rapidity
and the repeatability of the results.
measurements,
making
of
It will be seen that this type of analogue involves the finite difference
representation of the prototype system, and hence it intrinsically contains
the errors associated with curtailing a Taylor series.
the behaviour.

14.2.1.

Electrolytic

Tanks

The flow of electric current in an extended conductor obeys Laplace's

equation, hence by studying the potential distribution in such a conductor,
any solution of Laplace's equation can be inferred. The most common
example of this technique is the electrolytic trough which consists of a shallow
flat bottom tank containing an electrolyte with submerged electrodes and
insulators to represent the boundaries. By the nature of the system, the tank
has a uniformly distributed resistance, but its electrical capacity is negligible.
It is thus suitable for boundary value problems associated with Laplace's
equation, but not for initial value problems associated with Poisson's
equation,

V2r =

aV
ct

(14.5)

The analogy is obviously restricted to one or two dimensions, although

axi-symmetrical systems can be studied in a tilted tank which is analogous
to a typical wedge section from a cylinder. Boundary conditions of type 1
(dependent variable fixed) are represented by electrodes of the same shape
as the boundary maintained at the analogous constant potential. Lines of
symmetry or other type 2 boundary conditions are represented by insulated
boundaries, or electrodes with fixed current supplies. Type 3 boundary
conditions have to be imposed by trial and error, matching the potential on
the electrode with the current drawn from or supphed by it. The analogue
is used by exploring the tank with a fine probe and plotting the equipotentials.
The mechanical and electrical systems need careful design.
double
pantograph is frequently used so that a recording pen will follow an identical
path to the probe, but the system must be accurately constructed because
both the curvature and spacing of the equipotentials are physically important.
Unless the electrolyte has the same depth at all parts of the tank, the equipotentials will be distorted; hence a well supported glass plate is usually used
for the bottom of the tank.
The electrical system has to be chosen to avoid gas formation on the
electrodes, polarization potentials, and yet be at a safe voltage for operation.
The probe must be submerged sufficiently to mask the effect of the meniscus,
and not draw sufficient current to distort the equipotentials it is trying to
locate. It is normal to use about 20 volts at 1000 c/s to ehminate electrolysis
and polarization effects without introducing serious phase lags due to the
natural capacitance of the system. The probe is often connected to the grid
of a triode valve with cathode bias to avoid drawing current from the tank.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

496

tank is being used to simulate a two dimensional flow field, a double

probe is needed to diff'erentiate the potential field and obtain a velocity
component. It is of vital importance in this connection that neither probe
influences the potential distribution in the neighbourhood of the other probe.
Carefully designed electronic circuits are also needed to detect the small
difference between two larger alternating voltages.
For studying solutions of Laplace's equation with awkward shaped
boundaries, the electrolytic tank is ideal; the alternative solution using
numerical methods (Section 11.7.4) is frequently too large for the rapid
store of a digital computer (see Section 14.4.3) and too laborious for manual
If the

calculation.

14.3.

Active Analogue Computers

An

analogue computer for general calculations, as opposed to the above

wiU be described in this section. The basic unit
of such a computer is the " operational amplifier ", which is capable of
amphfying D.C. voltages with an amphfication factor of 10^-10^. The size of
an analogue computer is usually expressed in terms of the number of
operational ampHfiers it contains. Although the high gain amplifier is the
basic unit of the computer, its accuracy really depends upon the tolerance
of the resistors and condensers in the associated calculating circuits. Three
figure accuracy is all that can be expepted even from good quality components, and hence this is the limit to which this type of computer can work.
special purpose analogues,

14.3.1. Basic

A mpUfier for A ddition

Figure 14.3 illustrates the arrangement of computing elements around

The standard symbol

the operational amplifier during the addition process.

-AAAAr

S.J.

g^o

R
\MA/

e3o

v\^A^^

Fig. 14.3.

Amplifier
'^

Addition

circuit

for the operational amplifier is a triangle, and the diagram illustrates the
amplifier with a feed-back resistor {Rp) connecting its output potential {cq)
to its input at the " summing junction " (S.J.). Three input resistors of equal

value {R) connect the summing junction to three input potentials (^i, ^2, ^3).
Denoting the summing junction potential by e^ and the amplification
factor by /i, the following current balance can be obtained at the summing
junction.

COMPUTERS

14.

497

es^e,-e^^e^^eo-e_^^^
(14.6)

Rf

because the amplifier draws no current, and the amphfying factor equation

Eliminating

e^

eo= -

liCs

(14.6)

and

between equations

(14.7)
(14.7)

.o=-f(^. + ^a + .3)(l +

and rearranging,

~^')

If Rp and R are chosen to have the same value, and ^

equation (14.8) simplifies to

^0=

-(^1

(14.8)

is

sufficiently large,

+ ^2 + ^3)

(14.9)

which shows that the output potential is equal to the sum of the input
potentials but of opposite polarity. Provided ^ > 10^, the error caused by
assuming that /i is infinite is less than one part in 10"".
A typical computer arrangement has a 1 megohm resistor in the feedback circuit, two 1 megohm resistors and two 100 kilohm resistors in the
input circuit. Thus the ratio R^jR = 10 in the latter case, resulting in the
insertion of a scale factor into the addition. Incidentally, this reduces the
accuracy of the addition by a factor of ten and hence ji is usually chosen to

exceed 10^.
The standard range of potentials for computing purposes
100 volts, hence equation (14.7) shows that

e,< 10"^
the

i.e.

summing junction

is

volts

is

usually

(14.10)

virtually at earth potential during a calculation.

fortunate that equation (14.9) gives addition of the inputs with a sign
change, since this faciUtates subtraction. Putting a single input into the
It is

R = Rp, results in a sign change only. Hence to subone potential from another, its sign is changed by an ampUfier, and then
added in a second amplifier.

addition circuit with

tract
it is

14.3.2. Basic Amplifier for Integration

To
back

obtain an integrating circuit,

it is

only necessary to replace the feedThis is shown in Fig. 14.4

resistor {Rp) in Fig. 14.3 with a condenser.

eio

v^
Fig. 14.4.

^^0
SJ.

Integration circuit

where only one input is shown for simpUcity. Assuming that the summing
junction is still at earth potential due to the high amphfication factor, a

MATHEMATICAL METHODS IN CHEMICAL ENGI^fEERING

498

similar current balance gives

i--C^

(14.11)

Integrating equation (14.11) gives

t

^o

Choosing a

megohm

resistor

= ^Je,dt

(14.12)

and a microfarad condenser

gives

eo=-je,dt

(14.13)

in seconds.
The physical interpretation of equation
a constant potential, the output potential (^q) increases
in value by an amount e^ during each second of operation.
By connecting other inputs through resistors (R) to the summing junction

where

(14.13)

many

is

is

measured

that

if ^i is

variables can be simultaneously

added together and integrated. With

input resistors of different values, the addition can be in different proportions.

It should again be noted that equation (14.13) involves a change of polarity
indicated by the negative sign in front of the integral.

14.3.3. Multiplication

by a Constant

simple earthed potentiometer has the effect of multiplying a potential

by any fractional constant. Thus

in Fig. 14.5,

eo

ke,

(14.14)

provided no current is drawn from Cq. The elements in a computer installation are usually chosen so that a much larger current passes down the

(1-k)R

^^

oe

kR

Fig. 14.5.

resistor than

Multiplication by a constant

is likely to be drawn from Cq.

In addition, facihties are usually
provided for setting the potential at Cq at the desired fraction of the reference
voltage temporarily applied at e, with Cq under loaded conditions. Thus
errors due to loading the potentiometer arm can be ehminated.

14.

4.3.4.

COMPUTERS

499

Multiplication by a Variable

There are four kinds of multiplier in general use; the servo-multiplier,

the mark-space multiplier, the quarter-square multiplier,

multipher.

The

first is

and the Hall

effect

electro-mechanical, but the other three are purely

electronic.

The servo-multipHer consists of a small D.C. motor coupled to all of

arms of a gang of about six potentiometers. One variable is fed
to the end of any potentiometer, as e^ in Fig. 14.5, and the other variable is
used to position the armature of the D.C. motor; thus giving an output
potential, which is representative of the product, from the wiper arm of the
potentiometer.
The key to the design of this item of equipment is the
positioning of the armature to represent the variable. The standard reference
voltage (100 V) is appUed to one of the potentiometer gang and the output
from its wiper will be uniquely determined by the position of the D.C. motor
armature. This potential is compared electronically with the value of the
variable, and a signal is generated in proportion to the discrepancy. The
signal is fed to the motor so that its armature rotates in such a direction as
to reduce the difference. Hence it can be seen that a strongly coupled feedthe wiper

back of this nature

cause the armature to follow the changing value of the

due to the inertia of the armature and
coupled wiper arms, but a safe frequency limit is always quoted by the
supplier of the instrument.
variable.There

is

will

inevitably a time lag

The mark-space

wave of fixed
The height {h) is

multiplier operates by generating a square

frequency and wavelength

(A) as illustrated in Fig. 14.6.

Fig. 14.6.

Square wave

in

mark-space multiplier

determined by one variable, and the ratio {ajX) is fixed by the other variable.
Thus the average height of this curve is a measure of the product of the two
variables. Integration of the curve to determine the average height is also
performed electronically, and provided A is small, the output will be smooth

and

faithful.

An
is

algebraic identity

exploited in the quarter-square multiplier.

The addition and two sub-

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

500
tractions

on the

right

hand

side present Httle difficulty,

and two standard

units for generating the square of a variable are the only new equipment
required. It will be shown in Section 14.3.5 that such a function can be

represented as a succession of straight line segments using biased diodes.

circuits for the above operations are usually combined into a single unit,
so that the details of the above circuitry need not be understood.

The

an electrical potential difference is apphed across one pair of faces of a

cubic block of semi-conductor, and a magnetic field is imposed across a
second pair of faces, then an electrical potential difference appears across
the remaining pair of faces. This is known as the " Hall effect " and the
potential difference generated is proportional to both the input potential
difference and the magnetic field strength. The device can be used for multiplication by generating a current which is proportional to the one variable
and passing it through an electromagnet to produce the magnetic field.
If

The other

variable potential

is

applied to the appropriate pair of faces, and

the resultant potential difference will represent the product.

All of the above multipliers have their uses, but the servo-multiplier
restricted in frequency range,

from

and the Hall

effect multiplier

is

must be shielded

stray magnetic fields.

14.3.5. Function Generation

With the exception of the simple trigonometrical and exponential funcwhich can be obtained by solving Hnear differential equations with

tions

constant coefficients, a device is needed for generating arbitrary functions of

a variable. This need arises particularly when experimental data in the form
of a graph is to be presented to the computer. Such a device makes use of
the rectifying characteristic of a diode. Compared with the input resistors
of the operational amplifier, the diode resistance is neghgible to potential
differences of one polarity but infinite to the other. Thus if a diode is inserted
in any line connecting computing components, only signals of one polarity
will be conducted. If the diode is suitably biased using a resistance network,
the cut-off point can be varied from zero to either positive or negative
potentials.

Basic diode circuit

Consider Fig. 14.7 which shows a diode interupting the passage of an

input signal e^ to a summing amplifier. The diode will not conduct until the
armature of the potentiometer picks up a positive potential, and hence the
output potential ^q

will

be zero until e^ reaches a certain positive potential

14.

COMPUTERS

501

determined by both the position of the armature and the fixed negative
(K). The rate of change of Cq with respect to ej is independent
of K, therefore the slope of the output versus input line can be chosen by
positioning the armature of the potentiometer, and the starting point is then
fixed by the value of K. A combination of circuits similar to Fig. 14.7 all
feeding the summing junction of the same amplifier from the same source
ei results in Cq representing the sum of many straight Hne segments which
together will represent a functional relationship between Cq and ej. A unit
which utiHzes these ideas in a more efficient circuit is called a " diode function generator ", and to generate a given function it is only necessary to
represent the curve by a series of straight line segments and then to set two
potentiometers for each segment.
potential

14.3.6. Application to Differential

It

was shown

Equations

an amplifier can be used to integrate

is only one time variable (as

in Section 14.3.2 that

a function with respect to time.

Since there

opposed to three coordinates) it is only possible to integrate with respect to

a single variable in an analogue computer. Hence analogue computers are
restricted to solving ordinary differential equations. The numerical methods
of Section 11.7 can be used to transform many partial differential equations
into difference-differential equations and in this form they are amenable to
solution by analogue methods; although much computing equipment is
required.

The apphcation of analogue computers

to ordinary differential equations

further restricted because the independent variable

is represented by time
which is of necessity an open range variable. Hence initial value problems
can be solved directly, but boundar>' value problems have to be solved by
trial and error using the method presented in Section 11.4. The method will
be illustrated by the following example of an initial value problem involving
is

linear differential equations.

Example. Solve the

differential equations

+ 2 + +
T^
dx^
dx
dx

3>;-7z

dz

subject to the initial conditions

at

0,

4,

l,

dy/dx

Solution

A signal representing
representing

its

>'

or z can always be obtained by integrating a signal

In order to solve the given equations, signals

derivative.

representing y, z, dy/dx, dz/dx, and d^y/dx^ must be generated, and the first
three can be obtained by integration from the last two. Hence the differential
M.M.C.E.

j-7

502

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

equations must be rearranged to generate dzjdx and d^y/dx^ algebraically.

Thus

2= -2i^dx

dx^

dz

-3y + lz

dx

_y_52 + 2

II

dx

A simple potentiometer can be used as in

Fig. 14.5 to multiply by a fractional

convenient to represent the independent variable (x) by
the computer time variable (t) using a scale factor (a) to prevent the coefficients in equations I and II exceeding unity. Thus putting

constant, thus

it is

at

III

d^y

dy

dz

dr

dt

dt

-4 = -2a~-a~-3a^y + la^z

IV

dz

= ay 5az + 2a

dt

Choosing the maximum suitable value of a, equations

0-2t

^ .dy
-0-4
-f

X
d^y
2

dt^

III,

IV, and

V become
VI

^dz

^
-0-2-

dt

-0-12>; + 0-28z

VII

dt

= 0-2};-z + 0-4
^
dt

VIII

Equation VII can be simpHfied by eliminating dzjdt using equation VIII.

Thus
2
dt""

= -0-4-^ -

0-16y + 0-48z-008

IX

dt

In most analogue systems, there is a Hmit to the electrical potential

allowed within the computing circuits and 100 volts will be assumed for this
Hmit. The initial values of y and z are 1 and 4 respectively, and assuming a
stable system in which all derivatives vanish as ;c
oo, >^ = If, z = f. The
unit of ^^ and z is therefore chosen as 25 volts.
The circuit shown in Fig. 14.8 can thus be constructed. It is selfexplanatory except for the four feeds from the 100 V supplies. In order to
set the initial conditions, the condenser in the feedback line of the integrating
amplifier must be charged to an appropriate potential. The electronic means
of doing this usually involves a sign change, and the polarities of the supplies
for the initial conditions have been chosen to allow for this. The initial
condition part of the circuit is automatically disconnected as the computation
starts. Two further connections are made to the supply points to represent
the constant terms in equations VIII and IX.

COMPUTERS

14.

0-2

>

Potentiometer

503

I.e.

Initial

Summer

Fig. 14.8.

condition

Integrator

Active analogue computing circuit

A double beam oscilloscope connected to the outputs of the amplifiers

producing ^v and z is the best means of exhibiting the solution, but calculations repeatedly suspended at fixed time intervals and measured by a bridge
most accurate solution.
Assuming an integrating time constant of one second, as in Section
14.3.2, equation VI shows that 5 seconds of computing time correspond to a
unit increase in the value of x. The accuracy of the potentiometer settings
can be checked by allowing the calculation to proceed for a long time and
comparing the potentials with the known values of y and z at steady state,

circuit give the

14.3.7. Application to

Process Control Systems

In the above sections, the operational amplifier has been shunted with a
feedback resistance (Rp) to give an addition operator, or with a condenser
(C) to give an integrating operator; but these two circuits can be combined

shown in Fig. 14.9 to represent a further useful operation.

usual current balance at the summing junction gives

as

eo

R^R

+C

Taking the

dco
(14.15)
dt
17-

MATHEMATICAL METHODS

504

IN

CHEMICAL ENGINEERING

Introducing the differential operator (D) and rearranging,

(D + a)eo= -bei
where

At

i?

this stage,

it

differential operator

l/K^ C

and b

(14.16)

= l/RC

(14.17)

is necessary to establish the equivalence between the

(D) and the Laplace transform parameter (s) when the

AAAAr

'1o

AAAAr

-o^O
S.J.

Fig. 14.9.

initial

Simple transfer function

conditions of the problem are irrelevant.

Putting

/(0)=/'(0)= .... =f"-'\0) =

in

equation (6.13) gives

^[DYCO] =

s"f(s)

(14.18)

F(s)f{s)

(14.19)

which can be generalized to give

nFiD)f(t)]

showing that the operator F(D) transforms to

theorem of Section 6.2.2 is identical to:

f(DXyen = e'y{D + p)y

F{s).

Also, the shifting

(2.79)

Hence, in control theory where it can be assumed

running at steady state, the differential operator
(D) can be replaced by the Laplace transform parameter (s). Thus equation
nan be
V>p written
writt(=n
(14.16) can
with s replaced by D.

that the system

is

initially

(14.20)

and the circuit of Fig. 14.9 represents the transfer function bl(s + a) with
an input signal (ej) and an output signal (eo). The response of the control
system characterized by the above transfer function can therefore be studied
by altering e^ in any desired manner.
Any control system can be simulated by combining circuits of the type
shown in Fig. 14.9, and the stability can be studied quickly on the analogue
computer. If the various resistors have adjustable values, the best settings for
the real control system can be determined by trial and error on the computer
without introducing any hazard into the real process.

14.

14.3.8. Application to

COMPUTERS

505

Simulation

It has been shown above that the analogue computer can simulate the
behaviour of a control system. This idea can be extended to any process
which can be represented in the form of ordinary differential equations.
useful application of this nature is to the study of reaction mechanisms and
If experimental curves are available to
the evaluation of rate constants.
exhibit the concentrations of the initial reactants, the intermediate products,
and the final products any proposed mechanism can be studied on the computer by displaying the various concentrations on an oscilloscope. By adjusting potentiometer ratios, the rate constants can be varied until the computer
gives results of similar appearance to the experimental results. The final
settings of the potentiometers will thus determine the reaction rate constants
Having determined the rate constants, the
in the proposed mechanism.
yield of any product can be maximized by trial and error, by feeding different
reaction mixtures to the system and allowing the reaction to proceed for
different lengths of time. Thus the computer can save time and expense in
the way of chemicals and analysis of samples after the initial investigation
has been performed experimentally.

14.4.

Digital Computers

In digital computation, every calculation is reduced to basic fundamentals.

All calculations are performed in the " arithmetic unit " which can only

perform the four operations of addition, subtraction, multipHcation, and

Even these operations are not fundamental, since the other three
operations are performed by repeated subtractions. Nevertheless any one
of the above four operations can be performed on a pair of numbers entering
the unit.
Other operations, such as extracting a square root, must be
expressed in terms of the four basic operations, but only two numbers and a
result can simultaneously exist in the arithmetic unit. Hence there must be an
orderly flow of numbers through the arithmetic unit. The control of this
flow is governed by the " program " which refers to each quantity in the
calculation by a label called its " address ". The numbers can be considered
to be stored in a set of pigeon-holes, each hole having a unique address. The
calculation proceeds by extracting numbers from the pigeon-holes, processing
division.

them

in the arithmetic unit,

and

re-allocating the results to the pigeon-holes.

One point which must be appreciated

at this stage is that the program itself

must also be stored in a section of the pigeon-holes. A piece of stored
program information consists of an instruction accompanied by a reference
to the address of the next piece of program. The calculation thus proceeds
by the logic of the program controlling the passage of numbers through the
arithmetic unit, and at the same time maintaining its own relentless progress.
The operation of the various parts of the digital computer will be discussed
in the

following sub-sections.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

506
14.4.1.

The Binary System

The decimal system of calculation has little to recommend it except

The structure of a number in decimal notation is exposed in the

famiHarity.

equation:

723

= 7x10^ + 2x10^ + 3x10

(14.21)

This notation removes the necessity of having a different symbol for each
number by repeating the ten basic symbols in various orders, the position
of a digit implying the power of ten by which it is to be multiplied. To
store a decimal digit in a computer requires a device having ten distinct
There are many devices having two distinct
states which is inconvenient.
states such as off/on, or charged/discharged, and it is simpler to use these
devices for storing numbers. Each digit can only have two values in this
system, and the powers of two have to be used instead of the powers of ten.
Thus the quantity given in equation (14.21) is written

lOUOlOOU =

2^ + 2^

+ 2^ + 2^ + 2 + 1

(14.22)

binary system. Strictly, 2 and its powers should also be written in

binary notation, but there is no ambiguity in equation (14.22) as written.
All computers will automatically translate decimal numbers into equivalent
binary numbers, so the translation details need not be considered.
in the

14.4.2. Floating Point Arithmetic

Equation (14.22) shows that a large number of digits are required to

number in normal form. There is a more economical way of
establishing an extensive number scale with a restricted number of digits and
this will be illustrated in the decimal system.
In the normal way, four digits can express any number in the range
-^ 9999; but using three digits for a number and one digit for an exponent, the range covered can be increased to 0-1 -) 10^. Thus
express a

723

0-723 X 10^

(14.23)

With the convention that the number written Hes between OT and 0-999,
and the fourth digit is the exponent, the number in equation (14.23) can be
stored as 7233. A number written in this form is known as its *' floating
point " form. Both the number and the exponent are written in the binary
notation, and a standard number consisting of 40 binary digits is interpreted
as follows. The first digit gives the sign of the number, the next 30 digits
express the number, the next digit gives the sign of the exponent, and the last
8 digits give the magnitude of the exponent of two. Thus any number in the
range

10-^^<|x|
can be stored to nine significant figures

ax
where

<

1,

<

10'^

in the

form

2"

|6|<256(=2')

14.

COMPUTERS

507

In floating point arithmetic, the quantity " zero " cannot exist! Whenever
the number zero arises, the computer represents it as 10"^^. This eUminates
the necessity for special treatment of the division process to reject zero as a
divisor because the case can never arise.
14.4.3.

Storage Echelons

information in the storage space of the computer is to be equally

complete store must be scanned for each item. As the size of
problems and hence stores increases, much valuable time is lost in such
searches. All calculations are therefore broken down into convenient sections
so that one part can be dealt with at a time. Thus, during a section of the
calculation, only a small part of the storage space need be scanned for the
next piece of information. The storage is therefore divided into at least two
sections: an inner " rapid access " store which always contains the current
section of calculation, and the " main " store which may have 16 or 32 times
the capacity of the rapid access store. The complete program is initially
read into the main store (drum), and one complete section is transferred to the
inner store (core). On completion of this section of calculation, the contents
of the core are transferred to the drum and the next section is brought into
the core. It is not necessary to transfer the complete core contents on each
occasion, but it is more economical to transfer large portions than small
If all

accessible, the

portions.
If the main store is not large enough to hold the complete problem, part
of the program or data can be put on to magnetic tape to be read into the
main store as required by the program.
Such a facility is called a
" backing store ".
14.4.4. Input

and Output Devices

As mentioned above, information can be inserted and extracted by using

magnetic tape but this is an unusual method for normal programs. The most
convenient methods of transmitting data are by means of holes punched in
paper tape, or holes punched in cards. Since a hole is either present or absent,
this representation is of a binary nature. The information on the paper tape
can be translated into electrical signals in the computer by a pair of contacts
closing in every hole as it passes. The output tape can similarly be used to
operate a hne printer or teleprinter. A printed output can also be obtained
directly by coupling the line printer to the output channel of the computer.
14.4.5. Sub-routines

In many manual calculations, reference must be made to various tables

of functions such as trigonometric, exponential, or Bessel functions. It is
not economical to occupy large sections of the main store with detailed tables

and interpolation formulae; it is much simpler to program a calculation of

each function. With the trigonometric functions for example, the series
expansions can be used to calculate the function of any angle between
and Jtt, and the result extended to any other angle by recurrence relations.

MATHEMATICAL METHODS

508

IN

CHEMICAL ENGINEERING

Programs for such frequently used functions are usually permanently stored
within the computer as " sub-routines ". Whenever such a function has to
be evaluated, the appropriate sub-routine is called into the core, used to
evaluate the result, and then sent back to
special sub-routines can be written

its

own

storage location.

by the programmer

Other

to insert tables of

physical properties, or vapour-liquid equilibria, etc.

14.4.6.

The

The Program

of instructions which the computer obeys are stored in the

machine code " which the logical and arithmetic units of the
computer understand. Only an expert can write a program in such a language
and each manufacturer designs into his machine a translator which enables the
operator to converse with the computer in a simpler language called an
" autocode " or " user language ".
An example of such a language is
" Fortran ". The idea behind an autocode is to enable the user to write the
instructions in a reasonably intelhgible form which can be punched into
paper tape for example. The computer then accepts this tape as a sequence
of punched holes which the translating program is designed to sort out.
There must be no ambiguity either in the step from a written program to a
punched tape, or from a punched tape to a stored or " compiled " program.
Unfortunately, different manufacturers use different program languages,
and different installations of computers of the same type use different dialects
of the basic language. It is difficult therefore to describe programming in
any detail, since the most efficient program can only be written by someone
who is familiar with the particular computer being used.
All but the simplest programs must be sub-divided into sections, each
of which can be stored in the core of the computer, and this should be
arranged so that there is a minimum number of changes from one section to
another. A computer mentality must also be acquired which distinguishes
between a good manual calculation system and a good computer system.
An example of this is the solution of partial differential equations of boundary
value type.
In Sections 11.7.3 and 11.7.4, Liebmann's method and the
relaxation method were both described.
The most efficient relaxation
technique is usually to relax the largest residual at each stage and to call
upon experience to under- or over-relax. To the computer, the above
description presents two difficulties; firstly, all residuals must be compared
in pairs to locate the largest at each stage, and secondly, it is not easy to
series

particular "

write a simple instruction regarding the policy of under- or over-relaxing.

Liebmann's method on the other hand systematically recalculates half of
the values at each stage

and

is

admirably suited to computer calculations.

A further illustration has already been given in that a computer can calculate
a function by means of a sub-routine

more conveniently than consulting a

table.

Considerable time can be spent in writing a program and running it a

few times to remove the faults, and for isolated calculations the computer
may not save time. For example, the reactor design in Section 11.7 would

14.

COMPUTERS

509

require a fairly lengthy program and the sum of programming time and
computing time would be comparable with the manual calculation time.
The decision is Hkely to rest upon the availabiUty of a suitable computer
and the urgency of the problem; but if the diameter of the reactor tubes is to
be selected by repeat calculations for a range of tube diameters, a computer
program must be written. The programming time is a fixed charge which can
be likened to obtaining a photographic negative, and each calculation of a
similar set can be performed very economically, just as any number of
prints of any size can easily be obtained from a negative.

14.5.

Comparison of the Uses of Analogue and Digital Computers

Before comparing the capabiUties of the two types of general purpose

computer, it is necessary to investigate the circumstances in which the use
of any computer is economically or otherwise justified. In the design of
chemical engineering equipment, laborious trial and error calculations are
frequently encountered, and alternative equipment can be considered. For
example, a reaction can be carried out batchwise, in a stirred tank reactor
system, or a tubular reactor. In each type of equipment there are many
parameters which can be varied, and therefore it is desirable to design the
most efficient reactor of each type in order to compare the optimum designs.
A great deal of repetitive calculation is thus required, and a computer investigation can be profitable in terms of cost and time. Long manual calculations
are very boring, and barren in intellectual content, whereas time spent
designing a computer layout or program stimulates fundamental logical
thought. Once the computer is set up, the location of the optimum design is
relatively simple in most cases, and the end product of the investigation will
be a rehable optimum. In simulation studies, the effect of control system
failures, and operation under foreign conditions can be studied without
risk to any equipment.
The digital computer is most useful for deahng with numerical data and
should be used to apply statistical methods (such as least mean squares) to
experimental results. There is no reasonable limit to the accuracy or complexity of a calculation, and non-linear problems can be solved just as easily
as Unear problems. However, digital computers are usually expensive, and
are made in large complex units of moderate overall reHability. Calculating
by an ordered sequence of single instructions, the time taken is governed
by the length of the sequence and the speed of the machine. Thus the digital
computer cannot work in real time.
An analogue computer imitates the real system and it is a simple matter to
study the development of the solution. Each potentiometer adjustment can
often be associated with variations of a particular physical parameter, and
the operator thus retains the " feel " of the system. Since the calculation is
of a parallel nature, the computer can work in real time and act as a control
system for relatively simple pieces of equipment subject to rapid disturbances.
The mechanism is simple, relatively cheap, and reliable, and small units can

510

MATHEMATICAL METHODS

IN

CHEMICAL ENGINEERING

be purchased and built into a computer of any size. Unfortunately, the

is rarely better than 0-1% and cumulative
errors limit the size of useful equipment. Because special purpose units are
individual accuracy of a unit

used separately for addition, multiplication, integration, etc., many of the

units will be idle in a particular calculation, which is not efficient. Problems
containing many non-Hnearities require a large number of multiplying units
which are expensive, and division can only be performed on a specially connected multiplier. Another important restriction on the analogue computer
is its inability to deal with more than one independent variable, without
borrowing the digital technique of changing a partial differential equation
Overall, the digital computer is
into a difference-differential equation.
much more flexible, but the ability of the analogue computer to work in real
time gives it many useful chemical engineering applications.
In many estabHshments a choice has to be made between purchasing
an analogue or a digital computer. Depending primarily upon the interests
of the department requiring the computer, either type may be obtained.
This computer is then pressed into service for as many applications as
possible because it is available, rather than because it is suitable. There is
little harm in this situation since any computer is usually better than none,
but due to its greater flexibiUty, the digital computer is usually selected.
Currently the analogue computer is being equipped with a memory unit
so that large problems can be solved in sections with a reduced amount of
hardware. This is leading to a reduction of cost and can maintain the speed
advantage over digital computers by retaining parallel operation. Meanwhile digital computers are being developed with multiple arithmetic units to
give a degree of parallel operation. Success in this field will revolutionize
computer programming which at present is of a serial nature.

Computers are growing

in size,

and

digital

machines are becoming ever

of application. The demands

and this will ensure that computer
developments will not be wasted. Computers are now solving problems
which are completely insoluble by other methods, and this creates a new
difficulty; it is impossible to check independently whether the answer is
faster,

but as the machines grow, so do the

made upon computers

fields

are also increasing

correct or not.

Although some of the advances in computer technique are astonishing,

must be realized that the computer is just an instrument which can only
be fully exploited by an expert, or a team, knowing both the capabilities of
the machine and the fundamentals of the problem. The best computer in
the world is of no value to the chemical engineer unless he can understand
his own problem, formulate it, and devise a method of solution.
it

PROBLEMS
1.

It is desired to produce a substance B from a raw material ^ in a continuous stirred tank reactor of effective volume Fft^. If Q ft^/min of a
solution of A of concentration Cq is fed to the empty reactor, and the
chemical reaction of A is represented by

A B-^C
in

which

of

all

the reactions are

in the initial discharge

show

order,

first

from the reactor

that the

number of moles

given by the solution of

is

the differential equation

-IF^^-dT^^^'-^
P=

where

+ /c2 + /C3

/c^

C = QCok,
2.

Find the general solution of the

--

dx
If ^

3.

when x =

7c/3,

differential

2 V tan a:

show

that

equation

= 4 sm x

y has a maximum value of J.

Solve the differential equation

"-'(!)"-'=

--0

with the help of the boundary conditions

At

4.

l,

-1

y =

dx

Solve the equation

dx^

2 +

2>;

dx

*cosx
Birmingham 1962.

511

MATHEMATICAL METHODS

512
5.

IN

CHEMICAL ENGINEERING

Solve the equation

x^

^ +

2x

dx'^

2>'

dx

In

Birmingham 1960.
6.

proposed to produce X, Y, and Z from A and B by chemical reaction

on a semi-batch basis. That is, 20 moles of pure A is charged into the
reactor and B is added at the rate of 0-75 moles/min. If all the reactions
are second order and are represented stoichiometrically by the equations
It is

A + B-^X
X + B-^Y
Y + B-^Z

2
3

prepare a plot of the composition of the reaction mixture as a function

of the number of moles of B reacted.
The specific reaction rate constants are
for reaction

7.

1,

^i

for reaction 2,

/:2

for reaction 3,

/tj

= 1-8 x 10""^
= 0*3xl0"'^
= 0*1 x 10"^

1/mole.min

1/mole.min
1/mole.min

Product from a plant flows at a rate of 2 tons/hr into a storage vessel

holding 20 tons of material and is then pumped to another process. Owing
to a mishap in the plant, the material flowing to storage contains 1 % of
a contaminant.
(a) If the flow from storage is 2 tons/hr and the contamination of
supply lasts for 1 hour, what is the peak contamination in the discharge ?
(b) What is the peak contamination in (a) if the discharge is reduced
to 1 ton/hr?
(c) For how long would the 1
contamination of supply have to
continue before the discharge contamination became 0-8% with flows

as in (a) ?
(d) In (a), if the

supply contamination ceased completely after

hour,

what further time would elapse before the discharge contamination

to

if

(e) In (a), what supply contamination could be tolerated for

discharge contamination could be allowed up to 1 %?

Assume complete mixing

8.

fell

0-01%?

component A

hour

in the storage vessel.

is absorbed from a gas phase by a liquid containing a

with which it reacts chemically by a pseudo first order
reaction. The rate of absorption can be assumed to depend on both the
rate of chemical reaction and the rate of molecular diff'usion through the
laminar film of hquid of thickness d. Show that when the concentration

constituent

PROBLEMS
of /4 at a liquid depth d
film

is

513

zero, the concentration profile of A through the

is

where C^

Show

is

C^.iSinh[a((3-x)]
sinh(a(5)

and a = \/(k/D).
mass transfer through the surface

the concentration at depth x,

also that the rate of

is

known

a dimensionless quantity

as

is

the

Hatta

tanh(a5)J

number.
9.

Two

(a)

thin wall pipes of

in external diameter

have flanges J in thick

on the ends joining them together. If the conductivity

of the flange metal is k Btu/h ft^ F ft"^ and the exposed surfaces of
the flange lose heat to the surroundings which are at TiF by means of
and 4

in diameter

a heat transfer coefficient h Btu/h ft^ F, show that the equation giving
the temperature distribution in the flange is
/

d^T

dT\

6k{rj+-] = hr(T-T,)
V-d?-'Tr)

where

r is the radial distance coordinate in inches.

(b) If, in addition, the circular faces

insulated

(i.e.

0)

and the

if

the pipe temperature

of the flanges are thermally

where
and determine the rim temperature
200F, Ti = 60F, and k = 200.
flanges only lose heat through the rim

20, solve the simplified equation

is

M.Sc, Birmingham
10.

1961.

second order irreversible chemical reaction between two materials

B takes place in a single stirred tank reactor according to

and

A + B-^C + D
feed streams enter the tank each at a rate of R lb mole/h, the one
stream contains ACq mole fraction of A, and the other contains 2Cq
mole fraction of B. Material leaves the tank continuously at a rate
2R lb mole/h so that the tank always contains
lb moles and V ft^ of
material. The specific reaction rate constant is k ft^/lb mole/h.
When the reactor is closed down, both feeds are stopped, the vessel
remains full, and the reaction goes to completion leaving Cq mole
fraction of A unreacted. If the reactor is started from this condition,
by turning on both feed streams simultaneously, perform the following

Two

calculations
(a)

actants

Show

that the diff'erence in concentration between the

A and B in the tank remains

constant during start-up.

two

re-

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

514

Determine how the concentration of

(b)

in the

tank varies with

time during start-up.

In part (b) a quadratic equation should arise which will not
by a and j5 and continue the calculation in

(Note.

factorize; denote the roots

terms of a and

p.)

M.Sc, Birmingham
11.

As

1962.

any plate of a distillation column its composition

changes from the entry concentraction (xq) to the exit concentration
The composition at any point on the plate is influenced by the
(xj).
passage of the stripping gas at a rate G, the bulk flow of the Hquid at a
rate L, and mixing on the plate which can be expressed in terms of an
eddy diff'usivity (De). Assuming a constant Murphree point efficiency
liquid flows across

(x)

given by

and a

straight line

equihbrium curve given by y*

= mx + b show

that

the liquid composition satisfies the equation

d^x

^372

dz^

where

dx

mGV

dz

Lzi

-y-.- -Tr-^Mvi^-^*) =

the distance measured along the plate

is

Zi is the distance

from the

inlet weir,

between the weirs,

is

the linear velocity of the Hquid,

X*

is

the hquid composition in equilibrium with the entering gas

which

is

constant across the plate.

Allowing for diff"usion in the entry section where no gas flows, find the
hquid composition at all points on the plate.

Andrew,

12.

R. C, and Mackay, R. W.,

Chem. Eng. Sci. 17, 541 (1962).

S. P. S., Hutchinson,

ft^ capacity initially contains 10 ft^ of hquid. There

streams to the tank and one outlet stream. The first inlet
stream is steady at 2 ft^/min with a temperature of 100F, whereas the
second inlet stream at a temperature of 200F is subject to surging on a
2 minute cycle.
During the first minute, the flow rate is steady at
2ft^/min; and during the second minute the flow stops completely; it
then continues to start and stop at 1 minute intervals. The outlet stream
is always steady at 3 ft^/min.
If the tank is well stirred and the temperature in the outlet stream is
always the same as that in the tank, over what range does this outlet
temperature vary when the system has settled to its equihbrium cycle?

surge tank of 20

are

two

inlet

M.Sc, Birmingham

1960.

PROBLEMS
13.

515

Two vertical cylindrical tanks of 6ft diameter and 4ft diameter are
joined at the base by a horizontal pipe 2 ft 6 in long and ^ in diameter.
The 6 ft diameter tank also has an outlet at the base consisting of a
horizontal tube 2 ft long and J in in diameter.
If the larger tank is filled with oil to a depth of 12 ft whilst the smaller
tank is empty, and both tubes are open simultaneously, prove that the
maximum depth of oil occurring in the smaller tank will be 4-81 ft.
Assume

that the flow in both tubes

is

laminar and neglect kinetic

losses.

14.

Derive an expression giving the true average temperature difference, in

terms of the terminal temperatures and fluid properties, which is
applicable to a spht flow heat exchanger.
(This problem has been solved

C LeRoy Carpenter,
Chem. Eng. Prog. 47, 211, 1951.)

by D. Q. Kern and

15. Is the series

L3

1.3.5
'^**'

"^2.4"^

2.4.6

convergent or divergent ?

16.

Although the
it

17.

series

by removing

all

y^ -

is

divergent,

show

that the series obtained

terms in which the digit "0" occurs

is

convergent,

from
i.e.

" Log-mean " temperature differences are frequently used and defined by

T2-T,
'""

In

(T^/ro

What is T when T2 = T^l

Show further that the log-mean
value as T2/T1 ->
18.

Given the

value tends to the arithmetic

mean

1.

differential

equation

x^
+ fc^ + y =
dx^
dx
2

/c7^

(a) Use the method of Frobenius to find the roots of the indicial
equation and determine the recurrence relation.
(b) State without further calculation what form the complete solution
takes for all real values of k.
(c)

Complete the solution for the case when k

\.

Birmingham 1960.

MATHEMATICAL METHODS

516
19.

CHEMICAL ENGINEERING

Use the method of Frobenius to solve the equations:

x^ + (l-2x)^-2^ =

(a)

^(1-^)7Z^ + (3-7x)3^

(b)

20.

IN

9);

A constant temperature vessel is

supported on a truncated cone which is

is at constant temperature. The
small end of the support is of radius 1-5 in and is at 130F, the temperature of the vessel; whilst the wide end of the support is of radius 2-0 in
and is at 50F, the temperature of the surface. The vertical height of the
support is 4-0 in and its thermal conductivity is 40 Btu/h ft^ F ft"^
The curved surface of the support loses heat to the surroundings at
a temperature 50F according to a heat transfer coefficient of
2 Btu/h ft^ F.
By assuming that the temperature is constant across any circular
cross-section of the support and by measuring all temperatures relative
to the temperature of the base, calculate the first five terms in the series
giving the temperature distribution within the support.
in turn resting

on another surface which

M.Sc, Birmingham
21. (a)

Use the substitution x

ae^^ to solve

d^
Use the substitution y

(b)

(Py
dx^
22.

Show

V> =
=

\vx~^ to solve

Idy
xdx

_p1

.2

x^

that the nth roots of unity can be expressed as

where z

cos (2nln)

1960.

1, z, z^,

...,

z"~S

sin (In/n).

Hence, or otherwise, prove that

where a
23.

is

any

if

is

not a factor of a

in

if

is

a.

factor of a

integer.

By using the known trigonometrical identities, and equation

(4.59, 60, 61)

express tanh (a + b) in terms of tanh a and tanh b,

(b) express sinh a in terms of cosh 2a,
(a)

(c)

prove that

sinh"^a

+ sinh~*/} =

s\nh~^ (aJi

+ b^ + b\/l-\-a^)

PROBLEMS
24.

By putting

e'^,

show

517

that

/
where the integral is taken around a unit circle centred at the origin,
starting from the point e'*^.
If two circuits of the origin are permitted by allowing 6 to vary from
(/)

to

(t)

+ 4n, show

that the integral

is

zero for

all

(This example illustrates the importance of

by taking a branch cut

at

values of

making z~^

(p.

single valued

cj).)

around any

25. Integrate the following functions

finite

curve enclosing the

origin.
.

26.

By

cosz

sinz

-^

(a)

-^

(b)

around a suitable contour, show that

integrating e''"7(l+z^)

cosmx

+ x^

ax

71

for all positive real values of m.

sm
e'

27. Evaluate the residue

of

-.

e'V~^ around

28. Integrate the function

two

at the origin.

circles centred at the origin

a contour consisting of the parts of

lie in the first quadrant, and the

which

and imaginary axes joining the two arcs. By

allowing the one circle to increase in radius indefinitely, and the other
to shrink to the origin, show that
parts of the positive real

00

x~^ COS

xdx =

r(a) cos ina

provided

29.

< a <

1.

By raising the trigonometrical identity, sin 26 = 2 sin 9 cos 6,

(2 l)th power and integrating with respect to 6 over the range
prove the duplication formula for the
r(2n)

gamma

function.

2'"-^7r-^r(n)r(n + i)

to the
to

Jtc,

MATHEMATICAL METHODS

518
30.

IN

Prove by substitution that the function

I(x)

satisfies Bessel's

Hence prove

cos

Show

x sin 0)^0

((/)

equation of order n provided that n

nJ(x)

cos bx

dx

/
e'" and

(Introduce the function

Show

an integer or zero.

that

cos ax

32.

is

that
I(x)

31.

CHEMICAL ENGINEERING

that the Laplace

use the

Transform
^o(0

(i)

COS

cot

method of Section

5.5.3.)

of:

=
(s^

(ii)

="

+ l)^
s

CO

sin cot

(iii)

{s

33. Invert the following

s-1

(ii)

s(s

34.

Laplace Transforms.

sA + B

....

(i)

+ 2)

+ ay + co^

1
(iii)

(5 + ico) (5 ico)

Prove by integration that

rz is

(s2

+ 4)(52 + l)

the Laplace Transform of

s{s+a)"

35.

The diagram
mechanisms.
pletely filled

a hydrauhc device used in automatic control

operates as follows. The movable cylinder A is comwith oil, and when a sudden change in pressure occurs at
illustrates
It

AP-^c

519

PROBLEMS
the piston B, oil is forced via the Une C through the valve
very narrow aperture so that the flow of oil is restricted.
the cylinder

D which has a
Consequently

A moves against the spring 5, but as the oil passes through the

valve the compressed spring forces the cylinder to return to

its

original

position.

For a pressure change

that

if

and the flow of

K
m

is

laminar, the diff'erential equation

is

\dx

d^x

/ a

dr

\mK)

dt

is

the pressure change-flow proportionahty constant,

is

the eff'ective

mass of the system.

Solve this equation using the

the cyUnder

36.

and corresponding displacement x show

is a, the spring constant is k

through the valve

oil

describing the system

where

AP

the cross-sectional area of the piston

initial

conditions that the displacement of

Xq.

is initially

stirred tank reactor of efl'ective volume V ft^ containing a solution

of a reactant A, of concentration C^^o is heated to such a temperature
that the pseudo first order reaction

A-^B
takes place.

When

concentration

^ ft^/min.

the chemical reaction

If the specific reaction rate

assumed to contain a perfect

concentration of

37.

in

is

initiated a solution of

of

fed continuously to the reactor at the rate of

is

is

k min"^ and the

vessel

is

develop an expression giving the

the effluent as a function of time.
stirrer,

liquid feed stream of heat capacity Cp is passed through a small bore

cyhndrical pipe of radius a ft at the rate of R Ib/h. The pipe wall is
maintained at a temperature of T^ and the entering Hquid at Tq.
Determine the temperature of the hquid as a function of time and
distance from the inlet if the pipe is initially full of hot Hquid at Ty
when the flow is suddenly started.
The heat transfer coefficient is a function of distance x ft along the

pipe from the inlet and

is

expressed as

U=

kx-^

where k is constant.
Conduction within the hquid can be neglected and radial variation of
temperature can be assumed to be zero.
38.

The temperature of a

liquid flowing at a rate q through an agitated

heated vessel of volume V is to be controlled by placing a
temperature detecting element in the liquid leaving the vessel. The detector is connected to a controller which regulates the supply of energy
electrically

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

520
to

an immersion heater

such a way that the heat suppUed

in

T); where r is
heating element, and T is

tional to the temperature difference (r

safe

maximum

temperature of the

ture of the liquid.

The immersion heater

is

propor-

the constant
the tempera-

supplies heat to the hquid at a

Q^ = UcA(Tc T); where A is the heat transfer area, U^ is the overall

heat transfer coefficient and T^ is the surface temperature of the coil.
If the temperature of the entering liquid suddenly falls from Tq to a
new steady state temperature Tj, develop an equation to show the
variation of liquid temperature as a function of time, during the period
rate

of transient behaviour.
The effective mass of the immersion heater can be taken to be m and
the specific heat of the heater material s, the heat capacity of the liquid
Cp and its density p. Agitation in the vessel can be assumed to be perfect.

39,

The denominator of

the complete transfer function of a control system

can be represented by
6s^ + 5(a-l)s3

+ 5as2 + (a^-2a+l)s-(a-6) =

For what range of values of a

40. Plot the

is

the system stable ?

Nyquist diagram and discuss the

^i(s)G,(s)

41.

stability

of the system

(5-l)(4s+l)

Prove the following by vector methods:

(a)

In any skew quadrilateral, the mid-points of the sides are at the

corners of a parallelogram.
(b) A cube has twelve edges, six of which form the corners at the
ends of one diagonal. Prove that the mid-points of the other six edges
form the corners of a regular hexagon.

42. (a)

Find the unit vector perpendicular to the vectors 2/+3/ + 4^,

Also determine the angle between these two vectors.
(b) Prove that the vectors i-\-3j2k, i 5/ + 3Ar, and 2i2j-{-k are

i-\-2j2k.

coplanar.

43. (a)

Show

that

(AaB).{AaC) = A\B.C)-(A.B){A.C)
(b)

\4.

ir{AAB)A{AAD) =

A,

what

is

(AaB)a{Ba D)?

and C are three independent vectors,

and z which make
= xA-\-yB-^zC
If A, B,

MaN

find the values of x, y,

PROBLEMS

Hence show

that

MaN
45. If**

46.

is

AM

A.N

B.M

B.N

CM

C.N

the position vector,

(a) div

I-

(b) curi

I-

521

show

that:
r"r

(d) curl r"r

(c)

[^,B,C]

div

(n-{-3)r"

CSS

Use the theorems connecting Une,

surface,

and volume

integrals to

estabhsh the results

j(AA(l>).dr:= j(V.(t>)A.dS- j(iA.V)<l>.dS

(i)

jU(l).dS

(ii)

S
(iii)

where

jUV.<l>d(7

J^.VA<^^a =

j(<^A^).</S

is

j<l>.VUd(j

<T

+ J<^.VA^J<7
a

and

a constant vector,

are variable vectors

and

t/ is

variable scalar.

Birmingham {Maths) 1953.

47. In a

porous medium, the continuity equation can be written

where

is the

porosity

and the equation of motion (Darcy's law) can be

written

u=
where gravity
(a)

(b)

48.

Show

Show
Show

is

neglected,

that for

is

Vp

the permeabiUty

an incompressible

fluid,

jj.

the viscosity.

0.

that for isothermal flow of a compressible gas.

Isfipo

dp

dt

that the equation of fluid

2^2
v>

motion (7.126) can be expressed

alternative forms

du
(a)

and

V^ =

-maC=
ot

-V/ + vV^m

in the

MATHEMATICAL METHODS

522

^-aC=-Vx-vVaC
ot

(b)

49.

CHEMICAL ENGINEERING

IN

= WP + 2"^ + ^

where

and

F=-VQ

Two

concentric spherical metaUic shells of radii a and b cm (a < b) are

separated by a solid of thermal diffusivity a cm^/sec. The outer surface
of the inner shell is maintained at TqC and the inner surface of the
outer shell at T^^'C, Derive the differential equation governing the un-

steady state temperature distribution in the solid as a function of time

and radial coordinate.

Show

^l|^^%^^+

r=

nnl(b a).

where P

form

that the solution takes the

fi"sinm.-a)}e-^

Show how B can be determined from any

initial

temperature distribution.

50.

cyhndrical block of metal at a uniform temperature roF has its

The temperature of the curved surface is suddenly raised to and maintained at TiF. Prove that the tem-

circular faces thermally insulated.

T of any

perature

where

point in the block at any time

T.-T

__2

Ti-To

r is the radial

thermal

diffusivity,

Jo(ar)

^t'l

ocJi((xa)

coordinate, a

and

is

/ is

given by:

^.,,^.

the radius of the cyHnder,

Jo(aa)

is

of a tracer

is

the

=
M.Sc, Birmingham

51. If a liquid

K is

a are the positive roots of

1962.

flowing through a fixed bed at a rate v cm/sec when a pulse

added, show that the later distribution of the tracer is given

by the solution of

d^C
dx^

where

is

infinite

the mixing coefficient which operates in the axial direction

Removing x by

(x) only.

dC_dC
~^ dx~ dt

the substitution z

= x vt, and

assuming an

bed length, solve the above equation.

Show that the maximum concentration of tracer passing a downstream observation station arrives earlier than the residence time by an
amount

Ejv'^.

{See

J. J.

Carberry and R. H. Bretton, A.I.Ch.E.J.

4, 367, 1958.)

PROBLEMS
52.

523

Considering two dimensionless flow and neglecting the inertia terms in

show that the change of variables given by

the Navier-Stokes equation,

_ _d^_dilj

dx

dy

dy

dx

= Po-9py + PYf~^yf^y

yields the equations

dx^

dy""

p\d^'^ dy^)~~dt
Show

that a solution of these equations takes the

(l)

i/r

= Ae-^'"-"' cos Kx + My
= 5e-'"^+"'sinKx

m = s/K^ + (pnlfi)

if

(This

53.

method has been used to describe flow influidized beds by

W. J. Rice and R. H. Wilhelm, A.I.Ch.EJ. 4, 423, 1958.)

The steady laminar flow of a

liquid through a heated cyUndrical pipe

has a parabolic velocity distribution if natural convection effects and
variation of physical properties with temperature are neglected. If the
fluid entering the heated section is at a uniform temperature (tj), and
the wall is maintained at a constant temperature (/^), develop Graetz's
solution by neglecting the thermal conductivity in the axial direction.

(The solution
Trans.

54.

form

Assuming
tension

Am.

Inst.

that Stokes' approximation

(y) varies

is

is

reported by T. B. Drew,

Chem. Engrs.

26, 26, 1931.)

vaUd and that the

interfacial

according to
y

cLCOsO

+P

determine the fall velocity of a spherical droplet of a material of viscosity

is the polar angle measured
/Zi through a medium of viscosity 112 where
;

from the front of the droplet.

Show that there is only motion inside the droplet

a^g^p
where a

is

the radius of the drop

>

if

3a

and Ap

is

the density difference.

MATHEMATICAL METHODS

524
55.

CHEMICAL ENGINEERING

IN

The sudden closure of a valve generates a pressure wave within the liquid
flowing in the pipe leading to the valve. The passage of this wave causes
compression of the liquid and expansion of the pipe. Show that the
velocity of the liquid and the pressure are related by the equations
dp

p dv

dx

g dt

d'p
,d'p

^ = c^^

and
where
If

c
it

the velocity of propagation of the pressure wave.

can be assumed that a uniform pipe of length L connects a

is

reservoir at

P(xJ)

x =

Po

to the valve at

4cpvo

^
2.

show

L,

.^nx
^sm(2n +
Tsm(2n + 1)

-IT

^.

net

7X

l)

{G, R. Rich, Trans.

56.

that

A.S.M.E. 67, 361, 1945.)

A liquid at a uniform temperature falls as a thin film down a flat vertical

surface.

Determine the velocity distribution and the steady

state thick-

ness of the film.

If a short section of the wall is maintained at a higher temperature
than the Uquid, determine the average heat transfer coefficient in the
heated section. The velocity distribution can be simphfied in the thin
layer where temperature gradients are important.

57.

Given that the values of sinh x for x

are 0-100, 0-150, 0-201, 0-253, 0-305

sinh

58.

X when x

0-1, 0-15, 0-2, 0-25, 0-30 and 0-35

and 0-357; estimate the value of

0-328.

The thermal conductivity of methyl chloride

at the temperatures 50F,

100F, 150F, 200F. 250F, 300F, 350F and 400F is 0-0057, 0-0069,
0-0081, 0-0092, 0-0104, 0-0116, 0-0127 and 0-0139; estimate the thermal

conductivity at 413F.

59. Solve the difference equations

(ii)

ynyn+2

(iii)

>'+4-9yH-3 + 30>;+2-44>'n+i+24y

=
=

yn-^i
4''a

60. Solve the difference equation

given that

^o

0;

>'i

3;

>^2

6;

^3

36

PROBLEMS
61.

525

of a carrier gas containing Yq lb solute A per lb of carrier gas

ideal plates and
fed into the base of a plate column containing
contacted with Llb/h of solute-free liquid containing Xf^+^lb solute A

G Ib/h

is

per lb of solute-free liquor. If the concentration of A in the exit gas and

7^ and X^ respectively, show that the concentration
of A in the gas phase leaving any plate n in the column is given by

liquid streams are

where

is

Henry's law constant expressed by Y^

HXff.

number of plates required to absorb 99% mole ketene

moles/h of gas containing 4-36% ketene by volume using
138 lb moles/h of glacial acetic acid. The plate efficiency, of a 3-5 ft
diameter tower containing 0-7D weirs and therefore a plate hold-up of
2-09 lb moles Hquid, is 40%. The pseudo Ifirst order reaction rate constant for the chemical absorption process may be taken to be 0-075 sec"-^
and the equihbrium relationship

62. Calculate the

from 135

lb

Y*

= 2X

where Y* is the lb moles ketene/lb mole carrier gas in equilibrium with

X lb moles ketene/lb mole of acetic acid in the hquid.

(Problem taken from

Problem of Chemical Engineering Design,
by G. V. Jeffreys; published by Institution of Chemical Engineers,
London, 1962.)

is to be produced in a battery of continuous stirred tank

operating in series by reacting 2000 Ib/h of butanol with
326 Ib/h of glacial acetic acid containing sufficient sulphuric acid to
catalyze the reaction at 100C.
Under these conditions the rate of
reaction can be expressed by the equation

63. Butyl acetate

reactors

where k is 0-28 ft^/lb mole.min

and
C^ is the concentration of acetic acid in lb moles/ft^.
If the effective volume of each tank is 10 ft^ and the density of the
reaction mixture is assumed to be constant at 48 Ib/ft^, estimate the

number of reaction vessels required

in the final discharge

64.

A reactant A

is

if

the concentration of the acetic acid

not to exceed 3

Ib/ft^.

is to be converted into a product 5 in a battery of

continuous stirred tank reactors of total volume Kft^. If the feed rate is
q ft^/min and the concentration of ^ is C^^q show that the production of
5 is a maximum when all the tanks are the same size. The rate of reaction
can be taken to be first order and the battery is to operate isothermally.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

526
65.

8000 Ib/h of a 5

by weight solution of acetaldehyde

in toluene

treated with 5000 Ib/h of water in a 6 theoretical stage extraction

is

to be

column

operating under counterflow conditions. If the concentration of the feed

is suddenly changed to a 3% by weight solution, how long
will it take for the extractor to settle down to steady state operation
under the new conditions?
conditions

The equiUbrium
where

y=
X=

relation

is

F=

2-2A^

lb acetaldehyde per lb of water

\b acetaldehyde per lb of toluene.

The hold up per

stage can be

assumed to be constant

at

750 lb toluene; 300 lb water.

66.

A block of metal

12 in x 12 in x 2 in has

its

rectangular faces thermally

and its vertical square faces losing heat to the surroundings at

60F by means of a heat transfer coefficient. It can be assumed that the
temperature in the block is uniform and that the properties of the
insulated

metal are
specific heat c

The following

density p

0-10,

4201b/ft^

table gives the metal temperature as a function of time

Time(min)
Temperature CF)

200

Determine the heat transfer

10

15

20

25

30

35

40

186

173

161

151

141

133

125

119

coefficient.

In addition to the experimental readings, the given values are subject

to the following errors
c,

Assuming

+ 1%;

p,

0-1%;

all

dimensions,

that the thermocouples are accurate,

heat transfer coefficient determined

yV'

how

accurately

is

the

Birmingham 1961.

67.

Use the method of averages

to find the best curve of the type:

y
which

68.

fits

= Asm(x + B)

the following values.

A and B

are constant.

30

60

90

120

150

0-944

1-242

1-208

0-850

0-264

-0-392

Use the method of

least squares to

Nu =

fit

the best equation of the type

aPr"

to the following data.

527

PROBLEMS
Pr

Nu

Pr

0-46
0-53
0-63
0-74
4-2
5-6

24-8

100

26-5

17-7

3
5

Nu

Pr

Nu

84-5

32

140
127
189
245
315
380

31-6
70-3

58-5

115
115
150
170
165
193

340
590

95

245

55

28-5

18-6

300

25-3

60-3

41

69
58-4
70-7

370

93
185

480
195

{The information above was collected from various sources by

W, L. Friend and A. B. Metzner, A.I.Ch.EJ. 4, 393, 1958.)

69.

leaving a batch distillation column is being continuously

reprocessed so that it is difficult to measure directly the amount produced.
However, the flow rate can be continuously measured and samples are
taken at hourly intervals giving the following results.

The top product

Time

(hours)

More

volatile

90
60

ponent (%)

Flow

rate (Ib/min)

(a)

(b)

70.

91

91

63

62

89
62

87
60

84
59

80
57

75
57

com92
62

What is the total amount produced during the period of 8 hours ?

What proportion of this total is the more volatile component ?

The reaction

rate constant for the

decomposition of a substituted dibasic

acid has been determined at various temperatures as follows.

TCO

500

A:xl04(hr-i)

Use

the

method of

in the

equation

where

T is

701

least squares to

measured

89-4
45-4

7-34

1-08

1010
138

determine the activation energy (E)

in degrees Kelvin.

71. Calculate the first four positive roots

X tan X

of

to four decimal places.

Show graphically why the method

(Section 11.5.3) oscillates about the root
Newton's method

yields the

answer

far

of successive approximation

and converges slowly, whereas

more

efficiently.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

528

number of plates required to absorb 95 % mole of a component A from 500 lb moles/h of a gas containing 5 % of ^ by volume
using 600 lb moles/h of a 5 % mole solution of the absorbent B. A reacts
with B according to a second order irreversible reaction

72. Calculate the

r
If the liquid hold

= kC^Cs

up on each tray

is

4 lb moles and gaseous hold up

is

that the difference equation takes the form (9.71). The

plate efficiency is 15%, k = 2 x 10^ ft^/lb mole.h, the absorbent solution
negligible,

show

has the physical properties of water, and Henry's law constant

l-2x 10^ moles/mole.

73.

is

A mathematical model for the washing of a filter cake has been proposed
by M. T. Kuo (A.I.Ch.EJ. 6, 566, 1960). It is assumed that the wash
Hquor is in plug flow through the pores of the cake, and that a stagnant
film of filtrate remains on the pore walls. A mass transfer coefficient
governs the passage of solute across the boundary between the two Uquid
layers. Show that the process is described by the equations:
A2-^ =

hS(c,-C2)
del

dci

for

<

<

L;

t>xlU.

velocity of the

length of the pore channel.

mass transfer

S
A
A2

transfer area per unit length of pore.

cross sectional area occupied

Ci
C2

concentration of solute in
concentration of solute in

wash

liquor.

coefficient.

by wash hquor.
by filtrate.
wash liquor.

cross sectional area occupied

filtrate.

Calculate the concentration of the exit wash Hquor as a function of time

for the following numerical values.

U=
hS/Aj^

74.

0-00682 sec"^

Initial

concentration of

An

bearing formation

oil

in shape.

L =

0-0030 in/sec

filtrate is

is

hS/A2

in

0-00475 sec"^

0-015 moles/ft^.

assumed

and circular
formation at a point midway

to be of uniform depth,

single well penetrates the

between the centre of the formation and its perimeter. If oil is withdrawn
at constant pressure, determine how the pressure distribution in the
formation changes, and the rate of production of oil as a function of time.
(The governing equation can be checked with Example 4, Section 8.8.)

PROBLEMS
75.

529

immiscible liquids are in laminar jflow through a cyUndrical pipe,

and the flow rates are adjusted so that each hquid occupies a semi-circular
If the viscosity ratio between the two hquids is
section of the pipe.
10
1, find the flow rate ratio.

Two

{Relaxation methods have been used to solve this problem by

A, R. GemmelandN. Epstein; Canad. J. Chem. Eng. 40, 215, 1962.)

If a

matrix

A=

-1

3"

write

A + B;

2_

_4

(iv)

AB"^

(ii)

AB;

(iii)

BA;

Determine the latent roots of each of the following matrices.

"2-3
1

-5

-4J

What

is

4-1"

VI

11

(i)

78.

-3

"1

down
(i)

77.

B=

and a matrix

7-1

and(ii)

-4

L-4

4_

the rank and degeneracy of each of these matrices?

10,000 Ib/h of sulphuric acid (sp. heat 0-36)

is

to be cooled in a three-

stage countercurrent cooler of the following type.

Hot

acid at 174C

No. 1 where it is thoroughly agitated in contact with

coohng coils. The continuous discharge from this tank at 88C flows to
a second stirred tank and leaves at 45C. Thereafter it flows to the third
stirred tank and leaves at 28C. Coohng water at 18C flows into the
coil of tank No. 3 and thence to the coils of tank No. 2 and finally to the
coils of the first tank. The temperature of the water leaving the coils of
the hot acid tank is 88C. To what temperatures would the contents of
is

fed to tank

each tank rise if due to trouble in the supply, the cooling water suddenly
stopped for one hour?
On restoration of the water supply, water is put on the system at the
rate of 12,000 Ib/h. Calculate the acid discharge temperature after one
hour. The capacity of each tank is 8,000 lb of acid and the overall coefficient of heat transfer in tank 1 is 230 C.H.U/h.ft^ C, in tank 2 is
180 C.H.U/h.ft^ C and in tank 3 is 120 C.H.U/h.ft^ C which may be

assumed constant.
79.

A battery
in series.

of

N stirred tank reactors are arranged to operate isothermally

Each reactor has a volume of Fft^ and

is

equipped with a

perfect agitator so that the composition of the reactor effluent

same

as the tank contents.

If initially the tanks contain

is

the

pure solvent

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

530
only,

tion

and at a time designated tQ, q ft^/min of a reactant A of concentraCq lb moles/ft^ are fed to the first tank, estimate the time required

for the concentration of

leaving the

The reaction can be represented

Mh tank to be

It)

moles/ft^.

stoichiometrically by the equation

A B-^C
and
it

all

may

first order. There is no 5 or C in the feed but

be assumed that the feed contains a catalyst that initiates the

the reactions are

reaction as soon as the feed enters the

first

reactor.

of dry coal gas containing 0-2 lb light oil/lb dry gas is to be

wash oil in a 6 plate absorption column to
remove the light oil. If the concentration of light oil in the feed suddenly
changes to 0-3 lb light oil/lb dry gas show that the disturbance in the
absorption column can be described by the equation

80. 4,000 Ib/h

treated with 2,500 Ib/h of

aX

RX + PQ

dt

where

X2

= {X,

^6}

...

06

...

it

c
-/?*!

{^0

R= B

B
B,

A=

and

mH + h

L + mG
mH + h

C=

mG

mH + h

The symbols:
h

on each plate and may be taken to be 75 lb.

vapour hold-up on each plate and may be taken to be 1 -0 lb.
and G are the wash oil and dry gas flow rates and the equihbrium

H=
L

liquid hold-up

PROBLEMS

531

relationship of the light oil between gas

written

where

Y=

is

and

liquid

may be

mX

0-7.

Solve the transient equation and plot the composition of light

plate as a function of time.

oil

on each

81. A'o lb of a solute/lb solute-free liquor is fed to the first stage of a iV stage
Ib/h on a solute-free basis.
crossflow extraction process at the rate of

If

B lb

is fed to each stage and the

m, calculate the optimum number of stages

of extracting solvent free from solute

solute distribution ratio

and solvent to feed

is

ratio using the data given below.

Total cost of operating one stage is a including operation, fixed

charges and depreciation of capital.
Cost of solvent is ^ per lb of solvent.
Cost of feed Hquor is y per lb of solute.
Cost of extract product is X per lb of solute.

{Problem taken from: V, G, Jenson and G. V. Jeffreys;

Brit. Chem. Eng. 6, 676, 1961.)
82.

1,000 ft^/h of an aqueous liquor containing 5 lb of diethylamine per ft^

of solution

is

to be treated with toluene in a countercurrent Uquid

extraction process.

Estimate the optimum nuoiber of stages and the

quantity of solvent required if the capital and operating cost of each stage
including vessels, pumps and labour is estimated to be $7000 per annum.

The cost of regenerating, recycling and adding make-up solvent is

estimated to be $0-015 per ft^ of solvent and the value of the diethylamine
is estimated to be $0-53 per lb. The plant is to be operated 8000 hours
per annum and the distribution ratio of diethylamine between toluene
and water

is

0-86.

In question 81 above

it

will

stages for a crossflow extraction

variables

be found that the optimum number of

is

related to the other cost

and process

by the equation

CP

VmXoJ

'^

N+l^\XmXo)\

Therefore, compare the two methods of operation in carrying out the

extraction of diethylamine.

APPENDIX
Table of Laplace Transforms

As)

fit)

5
1

j(n

\.2,...)

(n-\)\
1

4
\/7rr

S-*

s-^"-'^)(/i

^v^
=

2"t"-+
l,2,...

lx3x5...(2-l)V7i
7

T'-'

t*-'

e"
1
*/j'"
te

{s-af

10

V-a/"

= ''''

{n-\y.
t'-'e'"

11

-^-(e-'-e'")

12*

(s-a){s-b)
*

Here a and b represent

distinct constants.

(From Operational Mathematics (Ed. 2) by R. V. Churchill Copyright

Hill Book Company Inc. Used by permission.)
S32

1958,

McGraw-

APPENDIX

533

Table of Laplace Transforms {continued)

Ks)
1

(ae'^-^e")

13*

(s-a)(s-b)
1

14*

(a-b)(b-c)(c-a)

{s-a){s-b)(s-c)
1

sina^

15
s^

+ a'

s'

+ a'

cosaf

16

-sinh at

17

s^-a^
s

cosh a r

18

s'-a^
1

19
sis^

20

(1

+ a"")
1

sV + a')

a'

(at

21
{s'

+ aY

(s^

+ a'f

at cos at)

sina^

la

s'

23

sin at)

(s^

^(sina^
la^

+ a^
s

22

cosaO

a'

-ism at +

at cos at)

la

s'-a^
24

{s^

+ a'-y
s

25
{s^

+ a^){s^ + b^)

(a'

cos at

cos at
b')

cos bt
2

b^-a

26

e^'sinbt

{s-af + b^-

a
(s-a^2 + l2

e'

27

cos bt

Z?

Here

a, b,

and

(in 14) c represent distinct constants.

18

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

534

Table of Laplace Transforms (continued)

Ks)
3a-

28
s^

+ a^

atyj?>

4a'

29
s^

/-

sin af

+ 4a^

cosh ar

cos at

sinarsinhaf

2a^
31

_1

_J.

^{smh at

smat)

J. jCcoshflt

cos at)

2a
32
s

a 4

2a

33

(1

+ a^/

(s'

+a

r) sin ar

ar cos a

d"

e'

34*

==6*^(1+2^0

35

(s-a)*

y/nt
1

36

\l s

\l s

2V t:^^

(e^'-O

37

/--fle''''erfc(aVO

Vs + a
"T^ + ae'^erfCaVO

38

Vs
39
s

+ a'

slnt

yjn

40
^/'s(s-a'')

-e'^'ediay/l)
a

41

a\ln
* L{t) is the

Laguerre polynomial of degree

n.

flrV3\

smhat

30

v3sin^-

X [^^^-2

APPENDIX

535

Table of Laplace Transforms {continued)

Ks)
2
1
h''-a

e''[fc-aerf(aVO]

42

-te'''erfc(6V0

1__

43

e'^^'erfcCaV?)

\/s(Vs + a)
1

44

.^'gerf(a70-l]

45

Vs(5-fl')(Vs + 6)
46*

+ e^''erfc(feV7)

(1-5)"

nl

= i^2n(V0

(2n)ly/nt
n!

47

^2+l(V^)

-1

48t

4s
49
\/s

+ ay/s-\-b

TW

50
(s

b)t

-,(/c>0)

+ fl)H5 + ^)

.(^'-)
51
(s

te

+ ar(s + b)^

'"-[/.("-i^')

.(^^')]
52

\ls-\-2a

\Js
-e-''h(.at)

y/s

+ 2a+y/s

7'-'"""'.(^'')

(/c>0)

53

(V5 + a + Vs + 6)2*

/f(x)

t /(x)

is

the Hermite polynomial,

/f(;c)

e'*^(e~'*).

= i~VJix\ where / is BesseFs function of the first kind.

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

536

Table of Laplace Transforms {continued)

f(s)

54

(Vs + g + Vs)-''^

(i;>-l)

-e-i"IXiat)

\/s\/s + a
1

55

56

Jo(at)

+ a^-sy

i^s'

{v>-\)

57

'

(/c>0)
r(/c)

58

59

{^|7^-sf{k>0)
{s-y/s^-ay

60

2\k

ka"

Uat)

(v>-l)

(k>0)
Y{k)

{)

^'-(^'
'

< <
when > k
when <t <k

lf-/c

fc

'0
ks

^(.>0)

whent>fc

when

rO

62

<t <k

when

JO
^'^'^-{l

61

63

J.-,iat)

{)

--^^

when

>

r(//)

<t < k
when > k
[0
S(k, t) = n when {n-\)k<t<
(i

64

when

65

s(l-e-*0

coth ^ks

'0
1

66

sie'^'-a)

when

+ a + a^ +

<t <k
g""^
.-. +

when nk
(m

<t<{n + i)k
1,2,...)

= (-ir-'
when 2/c(m-1) <t <2kn

M(2/c,0
67

- tanh ks

nk

(Fig. 6.4)

2s

1,2,...)

APPENDIX

537

Table of Laplace Transforms {continued)

Ks)
1

68
5(1

/(O

iM(^,0 + i =

+^-''0

i-(-ir

l)/c < < nk

when 2n/c <
<{ln + l)k
2k -t when(2n + l)/c
< < (2n + 2)k

when

(n

H{Kt) =

tanh \ks

69

F(t)

70
s

sinh ks

= 2(n-1)
when (2n -3)k<t< (2n - l)/c
(t>0)

M(2k,t + 3k) + l = l + {-l)"

when (2n - 3)/c < < (2n - l)/c

1__

71
s

cosh ks

(r>0)
72

F(0 = 2n-1
when 2k{n-l)

- coth ks
s

k
73
s'

+ k"-

< <2kn
t

ns

coth

\s\nkt\

Ik
sin
1

74

2)n
< < {2n-\)n
when {2n -\)n<t < 2nn

when

(2rt

(s'

+ l)(l-e-'"0

75

Jo(2-Jki)

76

-T= Q0s2\l kt
yint

77

-j=cosh2\Jkt
yjnt

78

Jl^-(*/s)

sin 2\lkt

\Jnk
1

79

sinh 2\/ /c^

/^\ (M-l)/2

80

81

e-^'^/^H/i>0)

/,-i(2V^)

J_^fc/^

e''"{ii>0)
18*

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

538

Table of Laplace Transforms (continued)

lis)

82

e-'^'(k

>

exp

0)

(-S)

Isl nf

83
s

84

85

ae

(k^O)

86

ia

e^'e''

^'erfc(aV7+^)

+ yjs)
+ cdc(-K)
V2V/

(k^O)

87
y/s(a

e''^e"'^erfc(aVr

+ \/s)

^j
<t <
when >

when
88
\/s(s

+ a)

\e-^''Io(ia^/t''-k^)

-k^s2 + a^

\jQ(a\Jt^
k^s2-a2

<

- k^) when >

t

<k

<t<k

when

fO

when

Jo

89

90
\lQ{ayJt'^-k^)

when

>

^-ki^s2+a2-s)

91

(k^O)

Jo(a^/t^

+ 2kt)

when
92

^-ks_^-k^s^ + a2

<t <

ak

J,(a^t'-k')

^t'-t
when

when
93

^-k^s^-a2_^-k5

>

<t <k

ak

^t'-t

I^{a^t^-e)
when

>

APPENDIX

539

Table of Laplace Transforms {continued)

Ks)

94

T. T-n

(i^>

<t<

when

'

fc

-1)

Vs' + a^CVs' + a^ + sV

when
95

r(l)

-logs

log

[r'(l)

>

= - 0-5772]

i3

96

-^logs(fc>0)

97*

s<>>

rrw

logn

e'^'lloga-Eii-at)']

logs
98*
s^

costSi(t)- sinfCi(0

+1

slogs

- sinrSi(0-cosrCi(0

99*
s^

+1

100* -log(l

+ fcs)(fc>0)

S-fl
101

+ kV)

103* -log(s2

104*

|(e''-0

102* -log(l

-K4)

+ a')(a>0)

4log(s^ + a^)(a>0)

-<3
-2Ci(aO

2 log a
2

- [ar log a +

sin ar

ar Ci

(at)'\

^,

-(1

cosaO

^,

- (1

105

log

106

log

107

arctan

cosh a

- sin kt
t

The exponential-integral, cosine-integral, and

and Si(0 are defined in Section 5.5.3.

sine-integral functions

Ei(

/),

Ci(/),

MATHEMATICAL METHODS IN CHEMICAL ENGINEERING

540

Table of Laplace Transforms {continued)

As)

108

- arctan -

Si(fcO

109

e^"'\v{c{ks){k>0)

110

-e^'^'erfcCi^sX/oO)

111

e'''Qifcsfks{k>0)

-(i)
7r\/7(r

+ /c)
when^O<(</c

rO

112

r'

-7-erfc(V/5
\(7rO~^

when

>

113

-7-e''^erfc(V/cs)(;^>0)

Vs
114

115

V7r(r

+ /c)

-sin(2/cV0

J/-""G;)

116

Koiks)

117

Ko(ks/l)

118

-e'"Ki(ks)

JO
\(t2_j^2yi
/

when

0<t <k

whenf>/c

e\

^VK'+2/c)

119

,-K,(k^s)

120

7;--(')

,-Xo(2V2fer)

Vt"
r[l(2fe

121

e\

- 0] ~ * when

ne-^'Ioiks)

when

\.0

''"'

122

e-'^hiks)

< <
> 2k
(

whpnn<f<2fc

7r/cVK2/c-0
(o

2fc

when

>

2k

APPENDIX

541

Table of Laplace Transforms (continued)

lis)

123

(a>0)

-e'Ei{-as)
t^-a
1

-^(a>0)

124
a
125

g-s,.) COS s + Ci s sin 5

{t

+ a)
1

7V\

SUBJECT INDEX
Associative law of Algebra, 51, 52, 206,

213
Autocode, 508
Average point, 362

Absolute convergence, 76
Absolute value of complex numbers, 75,
119
Absorption factor, 325, 326
Acetaldehyde, extraction of, 481-482
Acetic anhydride, hydrolysis of, 341-343
Acrivos and Amundson equation, 471
Addition of matrices, 439-440
Addition of vectors, 203-210
" Address " of numbers, 505
Adsorption coefficients, 365-369
Algebra, complex, 117-148; see also
under complex
matrix, 439-442
three basic laws, 51, 52, 206, 213

Averages, method in curve fitting, 361364

Axial symmetry, 234-236, 248-250, 276,
354-355

B
Benzene, batchwise chlorination
28

Benzene toluene, feed

column, 336-338

in

of,

26-

distillation

Benzoic acid
extraction

from toluene,

1, 2, 3, 5,

Algebraic equations

Bernoulli's theorem, 235

by matrices, 448-449
numerical methods of solving, 397406
Alternating series, 79
Ammonia vapour, pressure and density
of, 317-318
Amplifier
basic for addition, 496-497

Bessel equation, 106-116, 173, 282, 283,

linear, solution

basic for integration,

287, 296, 306

integral properties of,

modified, 110
second kind, 107-109
tables of (Chistova), 317
Bessel's interpolation formula, 319
Beta function, 154-155
argument of, 154
relation with gamma function, 156157

497^98

operational, 496

Amplitude of complex number, 119

Amplitude of elliptic integrals, 157, 158
Analysis of a continuous fat hydrolysing
plant, 62-66, 328-331
Analytic functions of complex variable,

Binary system of digital computers,

506
Binomial power series, 80
Binomial theorem, 80, 143, 181, 182, 340
Blockdiagram, 189, 191

131-148
Arbitrary constant, 18, 42, 45, 290, 323
Argand diagram, 118-1 19, 120, 199, 200,
399
algebraic operations on, 120-123

Boundary conditions,
analogue

Argument of
complex number,

119, 122, 136

gamma function,

152

tion,

circuit for, 494,

502

at fixed temperature, 19

beta function, 154

Ascent, steepest

15

Bessel functions
first kind, 107

method

constant heat flow rate, 19

initial condition, 19
in Laplace transformation, 1 66
in partial differential equations, 252-

of optimiza-

483-485

259
543

SUBJECT INDEX

544

Boundary conditions

Columns

cont.

in partial differential equations

first

type, function specified,

cont.

252-

253
in Sturm-Liouville equation,

271-

272
integro-differential,

third type,

257

mixed conditions, 256-

257

Schliching, 267

Boundary value problems, 258

Liebmann's method for solution

function, 42-45, 465-

Complementary solution, 322-323

Complex conjugate, 118, 123-124
Complex number, 75, 117-128

Argand diagram

of,

428^29
method of solution

of,

429-

436
solution of higher order diff'erential

equations, 388-397
Branchcut, 137, 155, 185
Branchpoints, 134-137
inversion of Laplace transformation,
185,187-188
Branch principal, 137
184, 186

or

468

absolute values of, 75, 119

operations
algebraic
on
diagram, 120-123
amplitude of, 119, 122, 136

Prandtl, 267

Bromwich path,

algebra, 51, 206,

divergent series, 76

Complementary
specified,

thermally insulated, 19
Boundary layer theory

relaxation

Commutative law of

213, 308, 312, 315, 442

Comparison test for convergent

particular solutions suggested by,

265-267
second type, derivative
253-256

addition of, 460

interchange of, 459
multiplication of, 460

Argand

118-119
136
conjugate numbers, 123-124
De Moivre's theorem, 124-125, 323
imaginary part, 117
modulus and argument, 119
nth roots of unity, 125-126
principal values of, 1 19-120

argument

of,

of, 119, 122,

real part, 117

series, 74,

79-82, 126-127

tensors of, 200

trigonometrical, exponential identities,

128

Complex variables, 128-148, 222

analytic functions of, 131-148

Cauchy 's

Cauchy
integral formula, 141, 142

theorem,
226

formula,

integral

141,

142
137,

144,

145,

186,

188,

Cauchy-Riemann conditions for derivatives of complex variables, 131, 132,

135, 138

Cayley-Hamilton theorem, 450, 456

Characteristic equation, 454-457
Chloramine synthesis, optimum conditions for, 485
Circle, uniform motion in, 218

Cauchy's theorem, 137, 144, 145, 188,

226
derivatives of, 130-131
evaluation of residues, 144, 145-147
at multiple poles, 147-148
integration of functions of, 137

Laurent's expansion, 142-145

132-147
branch points, 134-137
essential, 133-134

singularities of,

Clausius-Clapeyron equation, 318

poles or unessential, 133, 137-148

theory of residues, 142-145
Components of vectors, 204-205

Code

Computers, 492-510

Clairaut substitutions, 33

autocode of computer, 508

machine, 508
Column matrix, 438-439

active analogue, 496-505,

application
tions,

to

509-510
equa-

differential

501-503

SUBJECT INDEX

Computers

cont.

active analogue

cont.

application to process control

systems, 503-504

radius of, 80

basic amplifier for addition, 496-497

basic amplifier for integration, 497-

498
function generation, 500-501
multfplication by a constant, 498
multiplication by a variable, 499-500

analogue, 492-505, 509-510

comparison of analogue with

digital,

509-510
505-510
binary system, 506

digital, 492,

floating point arithmetic,

506-507

input and output devices, 507

storage echelons, 507
sub-routines, 507-508
the program, 508-509
electrolytic tanks, 495-496
passive analogue, 493-496

Conductor
transfer
radial
heat
through, 14, 199
temperature distribution in, 199

cylindrical,

flat,

Conformable matrices, 440

Conjugate, complex,

545

Convergence
absolute, 76
of complex series, 126

1 1

Convergent series, 75
comparison test for, 76
Gauss' test for, 78
integral test for, 78

Raabe's

test for,

78

ratio test for, 77

Convolution, 179-180
Coordinate systems, curvilinear, 228-231
Coordinates
cylindrical polar, 231
cylindrical polar with axial symmetry,
256
spherical polar, 228-231, 248, 250
Cosine integral, 158, 539
Countercurrent extractor, 343
Counter flow extraction, optimization
of, 479
Crank-Nicolson method, 422
Cross or vector product, 212-213

Cubic equation, analytical solution

398-399
Curve fitting, 360-369
least square, 364, 369
method of averages, 361-364

Conjugate numbers, 123-124

Conservation law, general, 9, 19, 220,
237, 247, 249
Constant of integration, 34, 35, 151
Constituent solution to complementary
function, 466
Continuity equation, 220, 232, 250-252
Contour charts, 479, 480

Degeneracy of a matrix, 446-448

Degeneracy, single, 447
De Moivre's theorem, 124-125, 323
Density of ammonia vapour, 317-318
Dependence, linear of a matrix, 446

Contour

Derivative

integration, 137-148, 155, 156,

158, 162

partial, 218,

inversion in Laplace transform by,

182-188

Contour

plots,

Control systems
automatic, 188
closed loop, 190

open loop, 190

stability,

192

stability criteria,

192-198

195-198
Control theory, automatic, 188-198
Control valve, response to varying air
pressure, 176-179
stability criteria (Nyquist),

239-242

substantive, 245
total,

355-356

of,

240

Desk calculating machine, 492

Determinantal equation, 453-454
Determinants
/)-, 464
and matrix
of square matrices
products, 443
test, 193,437
Difference(s)

backward, 311, 314, 316

central, 311, 319
first,

307


SUBJECT INDEX

546
DifFerence(s)

Differential equations, partial, 23, 245-

cont.

forward, 311, 316

fourth, 309
mih, 309
second, 308
tables, 309-310, 316, 317, 318
third, 309
Difference equations
characteristic equation, 323

306,

of,

245-252

particular solutions of, 259-267

solution by

method of images, 263-

267
solution by numerical methods, 409-

436

338-348
by numerical methods,
406-^09
finite, 321-338
degree and order of, 321-322
linear finite, 322-331
complementary solution, 322-323
particular solution, 323-326
non-linear finite, 331-338
analytical, 335-338
graphical solution, 331-335
Riccati finite, 336, 337
simultaneous linear, 327-331
Difference formula, Newton's, 314-318
differential,

solution

superposition of solutions, 261-263

Differential equations, second order,
linear,

33,41-66, 86-116

auxiliary equation

complex roots
equal roots

43-45
43-44

to,

to,

unequal roots to, 43

complementary function, 42-45
illustrative problems of, 59-66
particular integrals, 45-59
inverse operators, 50-56
undetermined coefficients, 46-50
variation of parameters, 56-59
particular solution of systems of, 468-

471
solution by numerical methods, 385-

Difference operator
J, 307, 308
,312-314

397
solution by series, 86-116

Difference quotient, 314-315

Differential equations, 23-73, 86-116,
163, 169-179, 245-306, 380-397,
409-436, 463-472, 501-503
application of active analogue to, 501503
Bessel's equation, 106-116

complementary function of system

465-468

409-436

formulation

solutions of systems of,

equations,

Differential

non-linear,

dependent

465^71
second

order,

33^1

variable

explicitly,

not

occuring

34

homogeneous, 35^1
problems of, 59-66

illustrative

of,

independent variable not occurring

order and degree, 23

solution by conversion to equivalent
system, 464-^65

equations, simultaneous,
66-73, 465-472
Differential operator, 50-56
application to exponentials, 52-53
use in eliminating dependent variables

non-linear, 23

solution by matrices, 463-472

standard methods of solving, 24

with variable coefficients, solution by
Laplace transformation, 170-175
Differential equations, first order, 24-33
exact, 24-25
homogeneous, 25-28
linear, 28
use in heat transfer, 30-33
solved by integrating factors, 28-33
solution by numerical methods, 380385

explicitly, 34, 35

Differential

from
simultaneous differential equations,
67,68
Bessel's
of
properties
Differential
equation, 115
Differential total, 240

Differentiation of matrices, 450-451

Differentiation of vectors, 216-218

Differentiation process, interpretation
of, 18,

238-240

Diffusion equation, linear, 295

SUBJECT INDEX

Error function, 58, 149-151, 187, 264

Diode, 500

properties of, 151

Diode function generator, 501

Errors, propagation of, 356-360

Dirac delta function, 175

Distillation apparatus, 6
Distillation

column

feed composition, 355

plate efficiency of, 336-338
Distillation

variation

rate,

with feed

rate, 8

Distributive law of algebra, 51, 206, 213,

308,312,315
Dittus-Boelter equation, 392

Divergence, 220
Divergent series, 75

comparison

test for,

Exponential function, 74, 127

76

test for,

optimization, 486-491

Efficiencies,

355

Eigen-functions, 276, 283, 295

not orthagonal, 283-289
Eigen-values, 276, 454
Electrode, graphite temperature profile,

38^1, 389-391
Electrolytic tanks,

through addition, 356-357

through general functional relationship, 359-360
through multiplication and division,
357-358
through subtraction, 357
Ethyl alcohol, esterification of, 332-333
Ethyl benzene, dehydrogenation of,
412-420,422-428
Euler's constant, 108, 109

78
integral test for, 78
Raabe's test for, 78
ratio test for, 77
Dot or scalar product, 211
Droplet, detailed concentration distribution around, 248
Duct, rectangular, velocity ratio of flow
through, 430-436
Dummy variable, 149, 151, 156
Dynamic function, 1 89
Dynamic programming method of

Gauss'

547

495^96

Elliptic integral

of first kind, 157

of second kind, 158
Elimination, systematic in simultaneous differential equations, 68-72
Enthalpy, concentration diagram, 209,

210
Error
absolute, 356

of measurement, 360
of method, 360
of precision, 360
relative, 357

identities of, 128

transform of, 166

Exponential integral, 159, 539
Exponential power series, 80, 127, 313
Extraction
of benzoic acid, 1-6, 9-1
of nicotine, 343-348
Extraction plant, dynamic programming, 486-491
Extrapolation, 315-318

Factor, integrating, 29
Feedback, 191, 496-498
Fictitious points, 417, 428
Finite differences, 307-348
difference quotient, 314-315
equations of, 321-348
of a product or quotient, 311
operator -J, 307-308
-E, 312-314
-Norlund,314,391
second and higher orders of, 308-309
tables of, 309, 310, 316, 317, 318
Fin, temperature distribution of, 100-103
Fixed bed catalytic reactor, 171-175,
412-420, 422-428
Floating point, 506-507

Fluid flow
equations

of, 231-236
235-236
ideal round sphere, 265-267
in packed column, 285-289
Stokes approximation, 232-235

ideal,

use of continuity equation for compressible fluid, 250-252

SUBJECT INDEX

548
Fluid flow

Gauss'

cont.

round sphere, 267-269

with axis of symmetry, 234

viscous,

Geometrical
206-210

Fluid property (temperature), 244

Force field, 224
conservative, 223
Forces, triangle of, 203
Fortran, 508
Fourier
integral theorem, 183
law, 20

transforms, 303-305
Frobenius method, 90-109, 277, 298, 492
Functions, orthogonal, 269-289
Functions, potential, 223, 236
Functions and definite integrals, 149-162

function
beta, 154-155

convergent or divergent
278
applications of vectors,

test for

series, 78,

Gradient, starting of graph, 354-355

Graphical location of roots of algebraic
equations, 399-400
Graphical solution of non-linear finite
diff'erence equations, 331-335
Graphite
electrode, rate of flow of heat from,
389-391
electrode, temperature profile of, 38-41
thermal conductivity of, 363
Graph paper, 349-354
linear, 350-351
logarithmic, 352
semi-logarithmic, 351
triangular, 352-354
Green's theorem, 227-228

error, 149-151

evaluation of B{\-q,

q),

155

gamma, 151-154
between

relationship

beta

and

gamma, 156-157
functions,

tabulated defined by in-

tegrals,

157-159

integral(s)

evaluation of, 159-162

elliptic of 1st kind, 157
of 2nd kind, 158
Laplace transform of, 170-175
sine, cosine and exponential, 158definite,

Gamma function,

107, 151-154, 162

Laplace transforms, 164

relationship with beta function, 156-

157

Gas absorption
application of series to, 84-86
in falling film,

297-302

Gas absorption column

performance

eff'ect,

500

multiplier, 499

Halogenation of hydrocarbon, 26-28

Hamilton's operator, 200, 218-222, 229

Hankel transform, 305

Heat conduction equation
boundary conditions, 19, 252-257, 258
cylindrical polar coordinates, 256

spherical polar coordinates, 276

steady state, 200
steady state linear, 199

159

in

Hall

of,

325-326

transient operations, 407-409

Gas compressor, minimum consumption of energy of, 487^89

Gas heater, tubular, exist gas temperature, 103-105
Gauss' divergence theorem, 226
Gauss' method of numerical integration,
373-377, 380

unsteady state, 236, 237, 244-245

unsteady state linear, 246-248
Heat conduction solutions
axial symmetry, 14-16, 276-281
rocket motor coolant duct, 392-397
steady state with axial symmetry, 276281

unsteady state linear, 259-265, 273276, 284-285, 290-295, 410-420,

422-428
Heat content of semi-infinite slab, 262
Heat exchanger, 245
Heat flow, sign convention in, 20

Heat loss
from fin, 100-103
in simple water

still,

through pipe flanges,

1 1

SUBJECT INDEX
Heat transfer

coefficient, 19,

549

Interpolation formula

256

319

Heaviside, 163

Bessel's,

Hermite polynomial, 535

Hydrocarbon, halogenation

Lagrange's, 319-321, 457

of,

Newton's, 315-318

26-28

Hydrolysis
of acetic anhydride, 341-343
of animal fat in spray column, 62-66,

32B-331
of tallow in spray column, 6266
Hyperbolic and trigonometric inverse
functions, 81

Hyperbolic functions, 81, 293

for half integer order of Bessel's
identities of, 128
series, 8

Images, method

of,

166, 167-170,
179-188
by contour integration, 182-188
convolution integral, 179
in
expansions
descending
power
series, 181

further elementary

methods

of,

180-

182
other series expansions, 182
solutions

ot

differential

equations,

169-170

equation, 116

Hyperbolic power

Inverse transformation,

in

solving partial

263-267
Index law of algebra, 308, 313, 315
Indicial equation, 91-100
differential equations,

using partial fractions, 168-169

using properties of transformation,
180
when singularities are poles, 185
when singularity is a branch point, 185
when there are both poles and a
branchpoint, 187-188
Isothermal surface, 219
Iteration,

420-428

Bessel's equation, 106

roots different, not integer, 91, 100

roots differing by an integer, 96, 98,
100

Jet laminar,

roots equal, 93, 100

Jury problem, 258

Infinite series,

dynamics

of,

38

74-79

Influence equations, 430

Input and output devices in digital

computers, 507
Integral test for convergent or divergent
series, 78
Integral properties of Bessel's equation,

Kettle, closed, heating of, 16

Kirchoff's Law, 493

Kremser-Brown equation,

325, 326

115
Integrals

area and surface, of vectors, 224

Lagrange multiplier, 483

222
volume, of vectors, 224-225

Lagrange's interpolation formula, 319-

line,

Integration of vectors and scalars, 222-

227
Integration, numerical, 369-379

Integration of matrices, 452

Initial value,

problems,

166,258,302
step-by-step

solution,

method

of

410-428

type equations, solution by numerical

methods, 385-388
Interpolation, 314, 315-321

321,457
Laguerre polynomial, 534
Lambda matrices, 452-454
Laplace equation, 236, 272, 276, 279,
281,428,435,495
Laplace operator, 163-167
Laplace transformation, 142, 157, 163198, 290-302, 338-348
application to
automatic control
theory, 188-198
control systems, 190-198

SUBJECT INDEX

550
Laplace transformation
convolution, 179-180

Mass

cont.

transfer

cont.

diffusion equation, 273

differentiation of, 170, 173

equation, 233

use of to find inverse

transform, 179-180
inverse transforms, 167-170, 180-188,
303
solution of differential equations,

integro-differential

integral,

169-170

boundary

idealization of, 2

using partial fractions, 168-169

method of solving differentialdifference equation, 338-348
method of solving partial differential
equation, 290-302
of derivatives, 165-166
of integral of a function, 1 70

representation of,

Matrices, 199, 210, 437-472

addition, 439-440

444

adjoint,

application to solution of differential

equations, 463-471
augmented, 449

properties of, 170-174

characteristic equations,

166-167
step functions, 174-179
staircase function, 175-176
unit step, 174-175

column, 438
commutable, 442
conformable, 440
D-, 464
degeneracy of, 446, 447

shifting theorem,

unit impulse, 175

use

of,

con-

257
sign convention in, 20
with axial symmetry, 248-250
Mathematical principles, 1-22
dition,

determinantal equation, 453-454

determinants of square matrices and
matrix products, 443
diagonal, 439

176-179

table of, 532-541

transforms, 163-167, 258

Latent roots, 453-459, 462-472
distinct, 453, 458,

454-457

466

differentiation

multiple, 453, 458, 466

Latent column vector, 454

Laurent's expansion, 142-145, 146, 148
Legendre's equation of order " / ", 277
Legendre polynomials, 278-280
L*H6pital's rule, 61, 82, 147, 175, 181,

262,293,416
Liebmann's method for boundary Value

and

integration

of,

450-452
equivalent, 460-461
lambda, 452-454
linear dependence of, 446
matrix algebra, 439-442
matrix series, 449^51
minors of, 446-447
modal, 467

problems, 428-429, 508

Logarithmic power series, 80

multiplication, 440-442

Loop

null,

non-singular, 443

439

closed control system, 190

open control system, 190

orthogonal, 443
polynomial, 450-451

open transfer function, 196

powers

of,

449-450

quadratic form, 461-463

rank

M
Magnet,

field strength

by iron

filings,

218
Many- valued functions, 129, 133, 134,
136
Marching problem, 258

Mass

447^48

446
reciprocal of a transposed matrix, 446

row, 438
singular, 443

solution of linear algebraic equations,

transfer

by molecular

of,

reciprocal of a matrix product, 445

reciprocal of a square matrix, 444-

diffusion,

237

448-449

SUBJECT INDEX
Matrices

cont,

solution of systems of linear differential

equations,

465^71

square, 439

sub-matrix, 448
subtraction, 440
Sylvester's theorem, 457-459
transformation of, 459-461
unit,

439

Matrizant, 452
Medians of triangle, concurrency, 207-

208
Mellin-Fourier theorem,,! 83, 184, 188
Mellin transform, 306
Method of formulation, summary,

Numerical methods, 380-436

algebraic equations, 397-406
analytical
solution
of cubic
equation, 398-399
comparison of methods, 406
graphical location of roots, 399400
improvement of roots by successive
approximation, 400-402
Newton's method, 402-406
difference-differential equations, 406409
step-by-step method, 407^09
first

order ordinary differential equations,

higher

21

Minor of a matrix, 446-447

Modal columns, 467
Modulus of complex number,

Modulus of

finite

differential

equations,

boundary value type, 388-397

119, 122,

difference equation,

411

Mole fractions, 325

Moment

of a force, 212
Multiplier
hall effect,

380-385

order

385-397

134, 157

499

mark space, 499

quarter square, 499
servo, 499

Murphree

551

plate efficiency, 84

Multi- valued functions, 129, 133, 134,

136

N
Navier-Stokes equation, 232-237, 265,
430, 435
Neumann form of Bessel's equation,
108, 109

initial value type, 385-388

use of Taylor's theorem, 387-388

420-428
Liebmann's method for boundary
value problems, 428-429
partial differential equations, 409436
Picard's method, 380-382
relaxation methods for
boundary
value problems, 429-436
Runge-Kutta method, 382-388, 390391, 406-407
step-by-step method, 407-409, 410428
Null matrix, 439
Nyquist criteria of stability in automatic
control systems, 195-198
Nyquist diagram, 196
iteration,

Newton
formulae, 314-318
method of solving algebraic equations,

402-406
second law of motion, 201, 232
Nicotine, extraction of, 343-348
Norlund's operator, 314, 391
Number(s)
absolute value of, 75
" address " of, 505

complex, 75, 117-128

conjugate, 123-124
imaginary, 117

Offset, 192
Oilfield, exploitation of,

295-297

Oliver, settling velocity of suspensions,

403
Operator
diff'erential,

50-56, 504
307-308, 311-315

finite difference,

Laplace, 163-167
Optimization, 473-491
analytical procedures, 475-483
involving 2 variables, 475
involving 3 variables, 475-482

SUBJECT INDEX

552

Optimization cont.
analytical procedures
cont.
involving 4 variables, 482
with are strictive condition, 482^83
dynamic, 474
dynamic programming, 486-491
method of steepest ascent, 483-485
sequential simplex method, 485-486

474
types, 474
Optimal policy, 487

Partial differential equations

cont.

compounding independent

variables

one variable, 259-261

conclusions, 306
continuity equation, 250-252
formulating, 245-252
Fourier transforms, 303-305
into

Hankel and Mellin transforms, 305306

static,

images, method of, 263-265

initial value type, 258-259, 290-302,

Optimum
curtailed, 475

410-428
420-428
Laplace transform, 290-302
step-by-step, 410-420
mass transfer with axial symmetry,
248-250
numerical methods, 409-436
orthogonal functions, 269-272
Sturm-Liouville equation, 270-272
other transforms, 302-306
particular
solution
suggested
by
boundary conditions, 265-267
particular solutions, 259-269
separation of variables, 272-289
iteration,

operation, stages for, 478-482

Optimization chart, 481

Origin in graph, 354
Orthogonal functions, 269-272
Orthogonality property, general use of,
283
Orthogonal, eigen-functions not, 283289
Oscilloscope, double beam, 493, 503, 505
Osculating plane of curve, 218
Over-relaxation, 434

eigen-functions

orthogonal,

not

283-289
Parameters, 17-18
of Beta function, 154
use in evaluation of definite integrals,
159
use in Laplace transformation, 163,

168,195,290,504
use in partial derivatives, 239
use in sine integration, 158
variation of, 56-59
Partial

derivatives,

interpretation

of,

239-245
Partial differential equations, 245-306,

409^36
boundary conditions, 252-259
1st type,

2nd

function specified, 252-253

type, derivative specified,

253-

256

equations involving
variables,

independent

281-283

general use of orthogonality property, 283

steady state heat transfer with axial

symmetry, 276-281
unsteady state linear heat conduction, 246-248, 273-276
superposition of solutions, 261-263
Partial differentiation, 239-245
changing independent variables, 243
between independent
relationships
variables, 243-245
relationships between partial derivatives, 240-242

use in inverting transforms, 168-169

Partial fractions

3rd type, mixed conditions, 256-257

Particular integral, 45-59

integro diff"erential, 257

Particular solution, 323-325, 468-471

boundary value

type, 258-259,

438, 508

Liebmann's method, 428-429

relaxation method, 429-438
solution by computer, 508

428-

Partition of matrix, 448

Picard's

method of solving

differential equations,

1st

380-382

Pipe flanges, heat loss through, 1

Poisson's equation, 43 435, 495
1 ,

order

SUBJECT INDEX
Ratio

Polar coordinates
cylindrical, 231
spherical, 228-231

in

order, 133, 155

complex

variables, 133, 141-148

inversion of Laplace transform, 185,

optimum volume

187-188
second order, 133

diff"usion

and chemical

Recurrence relation, 93-109, 277

Relaxation methods for boundary value
problems, 429^36, 508

series,

Laplace

transformation, 181
Powers of matrices, 449-450
Prandtl's " boundary layer theory ", 267

Residual, 430
Residual balancing, 434
Residues, 144, 156, 293, 302
evaluation of, 145-147

evaluation at multiple poles, 147-148

theory of, 142-148, 169, 182-188

Response

Pre-multiplication, 442

Pressure distribution in

yield,

reaction in 59-61

Potentiometer, 498-505

by

maximum

412^20,422-428
simultaneous

Potential function, 223, 236

79-86
descending, expansion

for

327-328, 474
Reactors, tubular
accuracy of rate of reaction, 359-360
dehydrogenation of ethyl benzene in,

Point conductivity (graphite), 363

Polynomial functions
application of inverse operator, 55
Hermite, 538
Laguerre, 534
matrix, 450-451
Ponchon-Savarit method, 209
Post multiplication, 442

Power

convergent or divergent
77
Reactors, tank
concentration of effluent from, 469-47
esterification of ethyl alcohol, 332-333
hydrolysis of acetic anhydride, 341343
test for

series,

Poles
first

553

oilfield,

295-297

Problems
in chemical engineering,
59-66
selected, 511-531
illustrative

Product, inner, 441

Program of computers, 505, 508-509
Propane, thermal conductivity of, 320321
Pythagoras, theorem of, 205

Quadriatic form of matrices, 461-463

semi-axes and principal planes

contours, 484
function, 483
surface, 483

transform 1 89

Reynolds number, 233, 235, 269

Riccati finite diff"erence equation, 336,

338
Right-hand screw rule, 212, 214, 221,
224
Rocket motor cooling duct, temperature
distribution of, 392-397
Rodrigue's formula, 279
Routh-Hurwitz criteria of control system
stability,

of,

461-463

Routh's

192

rules, 193

Row matrix, 438

Runge-Kutta formula, 384
Runge-Kutta method of solving

Raabe's

convergent or divergent
78
Radial heat transfer through cylindrical
conductor, 14
Radius of convergence, 80, 90
Rank of a matrix, 446
Rank of a tensor, 200
test for

1st

order diff"erential equations, 382388, 390-391, 406-407

series,

Saddle point, 476

Salt accumulation
in tank, 11-13

SUBJECT INDEX

554
Scalars,201,203
field, 218

Singularities

or dot product, 211, 224

point function, 218, 223, 229, 232
triple product, 215

branch points, 134-137, 185

133-134
of complex variables, 132-137
poles, 133, 141-148, 185, 187-188
Slide rule, 492

variables, 232

Solvent extraction, 1-6, 9-11, 343-348,

integration of, 222-227

essential,

Separation constant, 275, 276, 282

Sequential simplex method of optimiza-

485-486
74-116
alternating, 79

478-482
Sphere(s)

boundary condition

tion,

application in chemical engineering,

83-86
Bessel's equation, 106-116, 173, 282,

287
example, heat loss through pipe
flanges. 111

of half-integer order,
116
functions of negative order, 116
modified, 110
properties, 113-116
comparison test, 76
convergent, 75
divergent, 75
Gauss' test, 78
indicial equation, 91-100
functions

infinite,

74-79

properties of, 75-76

integral test, 78

matrix, 449-51

power, 79-86
descending expansion, 181
Raabe's test, 78
ratio test, 77
simple solutions, 86-90
solution by method of Frobenius, 90105
Shifting

theorem,

166-167,

180,

182,

504
Sign conventions, 19-21
in fluid flow,

in heat flow,
in

at

porous surface,

257

Series,

20
20

mass transfer, 20

Significant figure, 360

Simpson's Rule for numerical integration,

371-373, 380, 385, 420, 434
Sine integral, 158,539
Single valued function, 129, 130, 133,
134, 136

ideal fluid flow round, 265-267

laminar flow round, 435
thermal conductivity of insulator
between two concentric metallic,
279-281
viscous flow round, 267-269
Spherical particles, settling velocity of
suspensions of, 403-406
Stagewise counterflow system, 344
Stagewise processes, generalized vector

method for, 208-210

Stars, irregular,

Step-by-step

of

429

method of solution

diff"erence-diff"erential

equations,

407^09
of initial value problems, 410-420
Step functions, 174-179
staircase function, 175-176
unit impulse, 175
unit step, 174-175
use of, 176-179
Stokes
approximation, 232-235
stream function, 234, 236
theorem, 137,225-226
Storage echelons, 507
Store
backing, 507
inner (core), 507
main (drum), 507
rapid access, 507
Stream function
for fluid flow, 236
Stokes', 234
Stress tensor, 201-203, 437
Sturm-Liouville equation, 270-272, 275
279, 283, 288
Sub-routine programs in computer, 507508
Subsidiary equation, 169

SUBJECT INDEX
Substantive time derivatives, 245
Successive approximation, improvement
of roots of algebraic equations by,

400-402
Sulphuric acid, acid discharge tempera-

68-72

ture in cooling,

Summing junction,
Suspensions

of

496, 503
spherical

Sylvester's

particles

403^06

settling velocity of,

Theorem, 450, 457-459, 471

555

Transfer function, 189, 193, 196

in Nyquist diagram, 197
Transformation of matrices, 459-461

Transforms
Fourier, 303-305
Hankel, 305
Laplace, 163-168
Laplace, elementary, 165
Laplace, table of, 532-541
Mellin, 306

response, 189

Transpose
of a matrix, 443-444
Tallow, continuous hydrolysis of, 62-66,
328-331
Tank system, output from, 338-341
Tartaglia's method of solution of cubic
equation, 398
Taylor's theorem,

9, 10, 21, 82, 147,

221,

238, 246, 247, 249, 251, 255, 262,

313, 358, 391, 402, 429, 469, 483,

495
direct use of,

387-388

Temperature
surface variation with time in

rectangular slab, 263, 291

total time derivative,

245

variation in all directions, 253

Temperature distribution
in semi-iniinite slab, 260, 291
in transverse fin,
in wall, 264-265,

100-103
294-295

of rocket motor cooling duct, 392-397

with time in rectangular
slab, 284-285, 293
Tensors, 200-203
first rank (vectors), 201
variation

second rank (stress), 201-203

rank of, 200
zero rank (scalars), 201
Test determinants, 193
Test functions, 193

Thermal conductivity, 236

of graphite, 363
of insulator, 279-281
of propane, 320-321

Thermal

diff'usivity,

237, 248, 249

Time increment
eff'ect

effect

of in salt accumulation, 12
of in unsteady state operation,

matrix, reciprocal of, 446

of product of matrices, 443-444

ofrow vector, 443
Trapezium rule for numerical integration, 370-371
Treatment of experimental results
comparison of methods, 377-379
contour plots, 355-356
curve fitting, 360-369
method of averages, 361-364
method of least squares, 364-369
graph paper, 349-354
linear, 350-351
logarithmic, 352
semi-logarithmic, 351
triangular, 352-354
numerical integration, 369-379
Gauss' method, 373-377
Simpson's rule, 371-373
trapezium rule, 370-371
propogation of errors, 356-360
through addition, 356-357
through general functional relationship, 358-360
through multiplication and division,
357-358
through subtraction, 357
sources of error, 360
theoretical properties, 354-355
efficiencies, 355
starting gradient, 354-355
the origin, 354
Triangle, medians concurrent, 207-208
Triangle of forces, 203
Trigonometric functions
application of inverse operator, 55
extension of exponential operation,
53,110

SUBJECT INDEX

556
Trigonometric functions

cont.

identities, 128
in half-integer Bessel's equations, 116
solution by digital computers, 507
Trigonometric and hyperbolic inverse

functions, 81

Trigonometric power
Triple product
scalar, 215
vector, 215-216

series, 81

u
Unit matrix, 439, 442
Unitvectors, 204, 214,240
divergence of, 230
Unity,

Unsteady

vectors

roots of, 125

/2th

state operation, 8-13, 68,

198, 246, 343, 407,

Vector analysis cont.

Green's theorem, 226
Hamilton's operator, 218-222
ideal fluid approximation, 235-236
scalar triple product, 215
standard identities, 227
Stokes' approximation, 232-235
Stokes' theorem, 225-226
tensors, 200-203
of first rank (vectors), 201
of second rank, 201-203
of zero rank (scalars), 201
transport of heat, mass and momentum, 236-237
vector triple product, 215

192-

469

addition and subtraction

of,

203-210

and scalars integration of, 222-227

and scalars, line integrals, 222-223
components, 204-205
curl or rotation of, 220-222, 229

Vapour-liquid equilibria, 244

Vapour pressure of ammonia, 317-318
Variable(5)

complex, 128-148
dependent, 17-22, 23, 25, 30, 34, 252,
307, 321, 354
elimination from different equations,

66-73

differentiation of,

216-218

divergence of, 219-220, 229

geometric applications, 206-210
multiplication of, 210-216
partial differentiation of, 218
position 205-206
properties of addition, 206
scalar or dot product, 211

dummy,

149,151, 156
independent, 17-22, 23, 25, 30, 34,

unit vector relationships, 214

vector or cross product, 212, 213

volume

252, 307, 321, 354

integrals,

224-225

boundary value or Jury problem,

Velocity potential, 236

258
changing, 243
closed range, 258

Viscosity, coefficient of, 201, 232

Viscous flow round sphere, 267-269
Vinyl chloride, synthesis of, 171-175
Vorticity distribution, 269, 355
Vorticity of fluid element, 222, 233, 234,
235, 236

compounding

into

one

variable,

259-261
elimination

from

differential

equations, 67

equations involving three, 281-283

initial value or marching problem,
258
open range, 258
method of separation of, 272-289
not truly independent, 243-245
Vector analysis, 199-237
curvilinear coordinate systems, 228231
equations of fluid flow, 231-236
Gauss' divergence theorem, 226

w
Water-Still, simple with preheated feed,

Weber form of

Bessel's equation, 108,

109

Work, done by

force during displace-

ment, 211

z-plane, 129, 132, 134, 145, 155

^^M
^mm^^^^^^^^ms^

::TsCTpgy^^c!^^^^^::^gr'Ti]Bi^^
^Ji^^p5^4-^7^^Tgl^^

aiiiiiii

CTi^t^^fflj!

Hi