Solutions Manual

to accompany

Applied Numerical Methods

With MATLAB for Engineers and Scientists

Steven C. Chapra
Tufts University
CHAPTER 1
1.1 You are given the following differential equation with the initial condition, v(t = 0) = 0,

dv c
= g − d v2
dt m

Multiply both sides by m/cd

m dv m
= g − v2
c d dt c d

Define a = mg / c d

m dv
= a2 − v2
c d dt

Integrate by separation of variables,

dv cd
∫ a 2 − v 2 = ∫ m dt
A table of integrals can be consulted to find that

dx 1 x
∫a 2
−x 2
= tanh −1
a a

Therefore, the integration yields

1 v c
tanh −1 = d t + C
a a m

If v = 0 at t = 0, then because tanh–1(0) = 0, the constant of integration C = 0 and the solution

is

1 v c
tanh −1 = d t
a a m

This result can then be rearranged to yield

gm ⎛ gc d ⎞
v= tanh ⎜ t⎟
cd ⎜ m ⎟
⎝ ⎠

1.2 This is a transient computation. For the period from ending June 1:

1
 Balance = Previous Balance + Deposits – Withdrawals

Balance = 1512.33 + 220.13 – 327.26 = 1405.20

The balances for the remainder of the periods can be computed in a similar fashion as
tabulated below:

Date Deposit Withdrawal Balance

1-May $ 1512.33
$ 220.13 $ 327.26
1-Jun $ 1405.20
$ 216.80 $ 378.61
1-Jul $ 1243.39
$ 350.25 $ 106.80
1-Aug $ 1586.84
$ 127.31 $ 450.61
1-Sep $ 1363.54

1.3 At t = 12 s, the analytical solution is 50.6175 (Example 1.1). The numerical results are:

absolute
step v(12) relative error
2 51.6008 1.94%
1 51.2008 1.15%
0.5 50.9259 0.61%

where the relative error is calculated with

analytical − numerical
absolute relative error = × 100%
analytical

The error versus step size can be plotted as

2.0%

1.0%

relative error
0.0%
0 0.5 1 1.5 2 2.5

Thus, halving the step size approximately halves the error.

1.4 (a) The force balance is

2
dv c'
=g− v
dt m

Applying Laplace transforms,

g c'
sV − v(0) = − V
s m

Solve for

g v ( 0)
V= + (1)
s ( s + c ' / m) s + c ' / m

The first term to the right of the equal sign can be evaluated by a partial fraction expansion,

g A B
= + (2)
s ( s + c ' / m) s s + c ' / m

g A( s + c' / m) + Bs
=
s ( s + c ' / m) s ( s + c ' / m)

Equating like terms in the numerators yields

A+ B=0

c'
g= A
m

Therefore,

mg mg
A= B=−
c' c'

These results can be substituted into Eq. (2), and the result can be substituted back into Eq.
(1) to give

mg / c' mg / c' v ( 0)
V= − +
s s + c' / m s + c' / m

Applying inverse Laplace transforms yields

mg mg −( c '/ m )t
v= − e + v ( 0) e − ( c ' / m ) t
c' c'

or

3
v = v(0)e −( c '/ m )t +
mg
c'
(
1 − e −( c '/ m )t )
where the first term to the right of the equal sign is the general solution and the second is the
particular solution. For our case, v(0) = 0, so the final solution is

v=
mg
c'
(
1 − e −( c '/ m )t )
(b) The numerical solution can be implemented as

⎡ 12.5 ⎤
v(2) = 0 + ⎢9.81 − (0) 2 = 19.62
⎣ 68.1 ⎥⎦

⎡ 12.5 ⎤
v(4) = 19.62 + ⎢9.81 − (19.62)⎥ 2 = 6.2087
⎣ 68 .1 ⎦

The computation can be continued and the results summarized and plotted as:

t v dv/dt
0 0 9.81
2 19.6200 6.2087
4 32.0374 3.9294
6 39.8962 2.4869
8 44.8700 1.5739
10 48.0179 0.9961
12 50.0102 0.6304

60

40

20

0
0 4 8 12

Note that the analytical solution is included on the plot for comparison.

4
1.5 (a) The first two steps are

c(0.1) = 10 − 0.2(10)0.1 = 9.8 Bq/L

c(0.2) = 9.8 − 0.2(9.8)0.1 = 9.604 Bq/L

The process can be continued to yield

t c dc/dt
0 10.0000 -2.0000
0.1 9.8000 -1.9600
0.2 9.6040 -1.9208
0.3 9.4119 -1.8824
0.4 9.2237 -1.8447
0.5 9.0392 -1.8078
0.6 8.8584 -1.7717
0.7 8.6813 -1.7363
0.8 8.5076 -1.7015
0.9 8.3375 -1.6675
1 8.1707 -1.6341

(b) The results when plotted on a semi-log plot yields a straight line

2.4

2.3

2.2

2.1

2
0 0.2 0.4 0.6 0.8 1

The slope of this line can be estimated as

ln(8.1707) − ln(10)
= −0.20203
1

Thus, the slope is approximately equal to the negative of the decay rate.

1.6 The first two steps yield

⎡ 400 400 ⎤
y (0.5) = 0 + ⎢3 sin 2 (0) − 0.5 = 0 + [0 − 0.33333] 0.5 = −0.16667
⎣ 1200 1200 ⎥⎦

[ ]
y (1) = −0.16667 + sin 2 (0.5) − 0.333333 0.5 = −0.21841

5
 The process can be continued to give

t y
0 0
0.5 -0.16667
1 -0.21841
1.5 -0.03104
2 0.299793
2.5 0.546537
3 0.558955
3.5 0.402245
4 0.297103
4.5 0.416811
5 0.727927

0.8
y

0.4

0
0 1 2 3 4 5

-0.4

⎛c⎞
gm −⎜ ⎟ t
1.7 v(t ) = (1 − e ⎝ m ⎠ )
c

⎛ 12.5 ⎞
9.8(68.1) −⎜ ⎟ 10
jumper #1: v(t ) = (1 − e ⎝ 68.1 ⎠ ) = 44.87 m / s
12.5
⎛ 14 ⎞
9.8(75) −⎜ ⎟ t
jumper #2: 44.87 = (1 − e ⎝ 75 ⎠ )
14
44.87 = 52.5 − 52.5e −0.18666 t

0.14533 = e −0.18666 t
ln 0.14533 = ln e −0.18666 t

t = 10.33 sec

1.8 Qin = Qout

Q1 = Q2 + Q3

6
 30 = 20 + vA3

10 = 5 A3

A3 = 2 m2

1.9 ∑ M in − ∑ M out = 0

[1000 + 1200 + MP + 50] − [400 + 200 + 1400 + 200 + 350] = 0

Metabolic production = 300 grams

1.10 ∑ % body weight = 60

4.5 + 4.5 + 12 + 4.5 + 1.5 + IW = 60

% Intracellular water body weight = 33 %

4.5 + 4.5 + 12 + 4.5 + 1.5 + IW = 60

∑ % body water = 100

7.5 + 7.5 + 20 + 7.5 + 55 + TW = 100

% Transcellular water of body water = 2.5 %

7
CHAPTER 2
2.1

>> q0 = 10;R = 50;L = 5;C = 1e-4;

>> t = linspace(0,.5);
>> q = q0*exp(-R*t/(2*L)).*cos(sqrt(1/(L*C)-(R/(2*L))^2)*t);
>> plot(t,q)

2.2

>> z = linspace(-3,3);
>> f = 1/sqrt(2*pi)*exp(-z.^2/2);
>> plot(z,f)
>> xlabel('z')
>> ylabel('frequency')

2.3 (a)
>> t = linspace(5,30,6)

8
 t =
5 10 15 20 25 30

(b)
>> x = linspace(-3,3,7)

x =
-3 -2 -1 0 1 2 3

2.4 (a)
>> v = -2:.75:1

v =
-2.0000 -1.2500 -0.5000 0.2500 1.0000

(b)
>> r = 6:-1:0

r =
6 5 4 3 2 1 0

2.5
>> F = [10 12 15 9 12 16];
>> x = [0.013 0.020 0.009 0.010 0.012 0.010];
>> k = F./x

k =
1.0e+003 *
0.7692 0.6000 1.6667 0.9000 1.0000 1.6000

>> U = .5*k.*x.^2

U =
0.0650 0.1200 0.0675 0.0450 0.0720 0.0800

>> max(U)

ans =
0.1200

2.6
>> TF = 32:3.6:93.2;
>> TC = 5/9*(TF-32);
>> rho = 5.5289e-8*TC.^3-8.5016e-6*TC.^2+6.5622e-5*TC+0.99987;
>> plot(TC,rho)

9
2.7

>> A = [.035 .0001 10 2;

.02 .0002 8 1;
.015 .001 20 1.5;
.03 .0007 24 3;
.022 .0003 15 2.5]

A =
0.0350 0.0001 10.0000 2.0000
0.0200 0.0002 8.0000 1.0000
0.0150 0.0010 20.0000 1.5000
0.0300 0.0007 24.0000 3.0000
0.0220 0.0003 15.0000 2.5000

>> U = sqrt(A(:,2))./A(:,1).*(A(:,3).*A(:,4)./(A(:,3)+2*A(:,4))).^(2/3)

U =
0.3624
0.6094
2.5167
1.5809
1.1971

2.8

>> t = 10:10:60;
>> c = [3.4 2.6 1.6 1.3 1.0 0.5];
>> tf = 0:70;
>> cf = 4.84*exp(-0.034*tf);
>> plot(t,c,'s',tf,cf,'--')

10
2.9

>> t = 10:10:60;
>> c = [3.4 2.6 1.6 1.3 1.0 0.5];
>> tf = 0:70;
>> cf = 4.84*exp(-0.034*tf);
>> semilogy(t,c,'s',tf,cf,'--')

2.10

>> v = 10:10:80;
>> F = [25 70 380 550 610 1220 830 1450];
>> vf = 0:100;
>> Ff = 0.2741*vf.^1.9842;
>> plot(v,F,'d',vf,Ff,':')

11
2.11

>> v = 10:10:80;
>> F = [25 70 380 550 610 1220 830 1450];
>> vf = 0:100;
>> Ff = 0.2741*vf.^1.9842;
>> loglog(v,F,'d',vf,Ff,':')

2.12

>> x = linspace(0,3*pi/2);
>> c = cos(x);
>> cf = 1-x.^2/2+x.^4/factorial(4)-x.^6/factorial(6);
>> plot(x,c,x,cf,'--')

12
13
CHAPTER 3
3.1 The M-file can be written as

function sincomp(x,n)
i = 1;
tru = sin(x);
ser = 0;
fprintf('\n');
fprintf('order true value approximation error\n');
while (1)
if i > n, break, end
ser = ser + (-1)^(i - 1) * x^(2*i-1) / factorial(2*i-1);
er = (tru - ser) / tru * 100;
fprintf('%3d %14.10f %14.10f %12.8f\n',i,tru,ser,er);
i = i + 1;
end

This function can be used to evaluate the test case,

>> sincomp(1.5,8)

order true value approximation error

1 0.9974949866 1.5000000000 -50.37669564
2 0.9974949866 0.9375000000 6.01456523
3 0.9974949866 1.0007812500 -0.32945162
4 0.9974949866 0.9973911830 0.01040643
5 0.9974949866 0.9974971226 -0.00021414
6 0.9974949866 0.9974949557 0.00000310
7 0.9974949866 0.9974949869 -0.00000003
8 0.9974949866 0.9974949866 0.00000000

3.2 The M-file can be written as

function futureworth(P, i, n)
nn = 0:n;
F = P*(1+i).^nn;
y = [nn;F];
fprintf('\n year future worth\n');
fprintf('%5d %14.2f\n',y);

This function can be used to evaluate the test case,

>> futureworth(100000,0.08,8)

year future worth

0 100000.00
1 108000.00
2 116640.00
3 125971.20
4 136048.90
5 146932.81
6 158687.43
7 171382.43
8 185093.02

14
3.3 The M-file can be written as

function annualpayment(P, i, n)
nn = 1:n;
A = P*i*(1+i).^nn./((1+i).^nn-1);
y = [nn;A];
fprintf('\n year annualpayment\n');
fprintf('%5d %14.2f\n',y);

This function can be used to evaluate the test case,

>> annualpayment(35000,.076,5)

year annualpayment
1 37660.00
2 19519.34
3 13483.26
4 10473.30
5 8673.76

3.4 The M-file can be written as

function Tavg = avgtemp(Tmean, Tpeak, tstart, tend)

omega = 2*pi/365;
t = tstart:tend;
Te = Tmean + (Tpeak-Tmean)*cos(omega*(t-205));
Tavg = mean(Te);

This function can be used to evaluate the test cases,

>> avgtemp(5.2,22.1,0,59)

ans =
-10.8418

>> avgtemp(23.1,33.6,180,242)

ans =
33.0398

3.5 The M-file can be written as

function vol = tankvol(R, d)

if d < R
vol = pi * d ^ 3 / 3;
elseif d <= 3 * R
v1 = pi * R ^ 3 / 3;
v2 = pi * R ^ 2 * (d - R);
vol = v1 + v2;
else
error('overtop')
end

This function can be used to evaluate the test cases,

15
 >> tankvol(1,0.5)
ans =
0.1309

>> tankvol(1,1.2)
ans =
1.6755

>> tankvol(1,3.0)
ans =
7.3304

>> tankvol(1,3.1)
??? Error using ==> tankvol
overtop

3.6 The M-file can be written as

function [r, th] = polar(x, y)

r = sqrt(x .^ 2 + y .^ 2);
if x < 0
if y > 0
th = atan(y / x) + pi;
elseif y < 0
th = atan(y / x) - pi;
else
th = pi;
end
else
if y > 0
th = pi / 2;
elseif y < 0
th = -pi / 2;
else
th = 0;
end
end
th = th * 180 / pi;

This function can be used to evaluate the test cases. For example, for the first case,

>> [r,th]=polar(1,1)

r =
1.4142

th =
90

The remaining cases are

16
 x y r θ
1 1 1.4142 90
1 −1 1.4142 −90
1 0 1.0000 0
−1 1 1.4142 135
−1 −1 1.4142 −135
−1 0 1.0000 180
0 1 1.0000 90
0 −1 1.0000 −90
0 0 0.0000 0

3.7 The M-file can be written as

function polar2(x, y)
r = sqrt(x .^ 2 + y .^ 2);
n = length(x);
for i = 1:n
if x(i) < 0
if y(i) > 0
th(i) = atan(y(i) / x(i)) + pi;
elseif y(i) < 0
th(i) = atan(y(i) / x(i)) - pi;
else
th(i) = pi;
end
else
if y(i) > 0
th(i) = pi / 2;
elseif y(i) < 0
th(i) = -pi / 2;
else
th(i) = 0;
end
end
th(i) = th(i) * 180 / pi;
end
ou = [x;y;r;th];
fprintf('\n x y radius angle\n');
fprintf('%8.2f %8.2f %10.4f %10.4f\n',ou);

This function can be used to evaluate the test cases and display the results in tabular form,

>> polar2(x,y)

x y radius angle
1.00 1.00 1.4142 90.0000
1.00 -1.00 1.4142 -90.0000
1.00 0.00 1.0000 0.0000
-1.00 1.00 1.4142 135.0000
-1.00 -1.00 1.4142 -135.0000
-1.00 0.00 1.0000 180.0000
0.00 1.00 1.0000 90.0000
0.00 -1.00 1.0000 -90.0000
0.00 0.00 0.0000 0.0000

17
3.8 The M-file can be written as

function grade = lettergrade(score)

if score >= 90
grade = 'A';
elseif score >= 80
grade = 'B';
elseif score >= 70
grade = 'C';
elseif score >= 60
grade = 'D';
else
grade = 'F';
end

This function can be tested with a few cases,

>> lettergrade(95)
ans =
A

>> lettergrade(45)
ans =
F

>> lettergrade(80)
ans =
B

3.9 The M-file can be written as

function Manning(A)
A(:,5) = sqrt(A(:,2))./A(:,1).*(A(:,3).*A(:,4)./(A(:,3)+2*A(:,4))).^(2/3);
fprintf('\n n S B H U\n');
fprintf('%8.3f %8.4f %10.2f %10.2f %10.4f\n',A');

This function can be run to create the table,

>> Manning(A)

n S B H U
0.035 0.0001 10.00 2.00 0.3624
0.020 0.0002 8.00 1.00 0.6094
0.015 0.0010 20.00 1.50 2.5167
0.030 0.0007 24.00 3.00 1.5809
0.022 0.0003 15.00 2.50 1.1971

18
3.10 The M-file can be written as

function beam(x)
xx = linspace(0,x);
n=length(xx);
for i=1:n
uy(i) = -5/6.*(sing(xx(i),0,4)-sing(xx(i),5,4));
uy(i) = uy(i) + 15/6.*sing(xx(i),8,3) + 75*sing(xx(i),7,2);
uy(i) = uy(i) + 57/6.*xx(i)^3 - 238.25.*xx(i);
end
plot(xx,uy)

function s = sing(xxx,a,n)
if xxx > a
s = (xxx - a).^n;
else
s=0;
end

This function can be run to create the plot,

>> beam(10)

3.11 The M-file can be written as

function cylinder(r, L)
h = linspace(0,2*r);
V = (r^2*acos((r-h)./r)-(r-h).*sqrt(2*r*h-h.^2))*L;
plot(h, V)

This function can be run to the plot,

>> cylinder(2,5)

19
20
CHAPTER 4
4.1 The true value can be computed as

6(0.577)
f ' (1.22) = = 2,352,911
(1 − 3 × 0.577 2 ) 2

Using 3-digits with chopping

6 x = 6(0.577) = 3.462 ⎯⎯ ⎯⎯→ 3.46

chopping

x = 0.577
x 2 = 0.332929 ⎯⎯ ⎯⎯→ 0.332
chopping

3x 2 = 0.996
1 − 3x 2 = 0.004

3.46 3.46
f ' (0.577) = = = 216,250
(1 − 0.996) 2
0.004 2

This represents a percent relative error of

2,352,911 − 216,250
εt = = 90.8%
2,352,911

Using 4-digits with chopping

6 x = 6(0.577) = 3.462 ⎯⎯ ⎯⎯→ 3.462

chopping

x = 0.577
x 2 = 0.332929 ⎯⎯ ⎯⎯→ 0.3329
chopping

3x = 0.9987
2

1 − 3x 2 = 0.0013

3.462 3.462
f ' (0.577) = = = 2,048,521
(1 − 0.9987) 2
0.0013 2

This represents a percent relative error of

2,352,911 − 2,048,521
εt = = 12.9%
2,352,911

Although using more significant digits improves the estimate, the error is still considerable.
The problem stems primarily from the fact that we are subtracting two nearly equal numbers
in the denominator. Such subtractive cancellation is worsened by the fact that the
denominator is squared.

4.2 First, the correct result can be calculated as

y = 1.37 3 − 7(1.37) 2 + 8(1.37) − 0.35 = 0.043053

21
 (a) Using 3-digits with chopping

1.373 → 2.571353 → 2.57

–7(1.37)2 → –7(1.87) → –13.0
8(1.37) → 10.96 → 10.9
– 0.35
–0.12

This represents an error of

0.043053 − 0.12
εt = = 178.7%
0.043053

(b) Using 3-digits with chopping

y = ((1.37 − 7)1.37 + 8)1.37 − 0.35

y = (−5.63 × 1.37 + 8)1.37 − 0.35

y = (−7.71 + 8)1.37 − 0.35

y = 0.29 × 1.37 − 0.35

y = 0.397 − 0.35

y = 0.047

This represents an error of

0.043053 − 0.47
εt = = 9.2%
0.043053

Hence, the second form is superior because it tends to minimize round-off error.

4.3 (a) For this case, xi = 0 and h = x. Thus, the Taylor series is

f " ( 0) 2 f ( 3 ) ( 0) 3
f ( x) = f (0) + f ' (0) x + x + x + ⋅⋅⋅
2! 3!

For the exponential function,

f (0) = f ' (0) = f " (0) = f ( 3)

(0) = 1

Substituting these values yields,

22
 1 2 1 3
f ( x) = 1 + x + x + x + ⋅⋅⋅
2! 3!

which is the Maclaurin series expansion.

(b) The true value is e–1 = 0.367879 and the step size is h = xi+1 – xi = 1 – 0.25 = 0.75. The
complete Taylor series to the third-order term is

h2 h3
f ( xi +1 ) = e − xi − e − xi h + e − xi − e − xi
2 3!

Zero-order approximation:

f (1) = e −0.25 = 0.778801

0.367879 − 0.778801
εt = 100% = 111.7%
0.367879

First-order approximation:

f (1) = 0.778801 − 0.778801(0.75) = 0.1947

0.367879 − 0.1947
εt = 100% = 47.1%
0.367879

Second-order approximation:

0.75 2
f (1) = 0.778801 − 0.778801(0.75) + 0.778801 = 0.413738
2

0.367879 − 0.413738
εt = 100% = 12.5%
0.367879

Third-order approximation:

0.75 2 0.75 3
f (1) = 0.778801 − 0.778801(0.75) + 0.778801 − 0.778801 = 0.358978
2 6

0.367879 − 0.358978
εt = 100% = 2.42%
0.367879

4.4 Use εs = 0.5×102–2 = 0.5%. The true value = cos(π/4) = 0.707107…

zero-order:

23
 ⎛π ⎞
cos⎜ ⎟ ≅ 1
⎝4⎠

0.707107 − 1
εt = 100% = 41.42%
0.707107

first-order:

⎛π ⎞ (π / 4) 2
cos⎜ ⎟ ≅ 1 − = 0.691575
⎝4⎠ 2

0.707107 − 0.691575
εt = 100% = 2.19%
0.707107

0.691575 − 1
εa = 100% = 44.6%
0.691575

second-order:

⎛π ⎞ (π / 4) 4
cos⎜ ⎟ ≅ 0.691575 + = 0.707429
⎝4⎠ 24

0.707107 − 0.707429
εt = 100% = 0.456%
0.707107

0.707429 − 0.691575
εa = 100% = 2.24%
0.707429

third-order:

⎛π ⎞ (π / 4) 6
cos⎜ ⎟ ≅ 0.707429 − = 0.707103
⎝4⎠ 720

0.707107 − 0.707103
εt = 100% = 0.0005%
0.707107

0.707103 − 0.707429
εa = 100% = 0.046%
0.707103

Because εa < 0.5%, we can terminate the computation.

4.5 Use εs = 0.5×102–2 = 0.5%. The true value = sin(π/4) = 0.707107…

zero-order:

24
 ⎛π ⎞
sin ⎜ ⎟ ≅ 0.785398
⎝4⎠

0.707107 − 0.785398
εt = 100% = 11.1%
0.707107

first-order:

⎛π ⎞ (π / 4) 3
sin ⎜ ⎟ ≅ 0.785398 − = 0.704653
⎝4⎠ 6

0.707107 − 0.704653
εt = 100% = 0.347%
0.707107

0.704653 − 0.785398
εa = 100% = 11.46%
0.704653

second-order:

⎛π ⎞ (π / 4) 5
sin ⎜ ⎟ ≅ 0.704653 + = 0.707143
⎝4⎠ 120

0.707107 − 0.707143
εt = 100% = 0.0051%
0.707107

0.707143 − 0.704653
εa = 100% = 0.352%
0.707143

Because εa < 0.5%, we can terminate the computation.

4.6 The true value is f(2) = 102.

zero order:

102 − (−62)
f ( 2) = f (1) = −62 εt = 100% = 160.8%
102

first order:

f ' (1) = 75(1) 2 − 12(1) + 7 = 70

102 − 8
f ( 2) = −62 + 70(1) = 8 εt = 100% = 92.1%
102

second order:

25
 f " (1) = 150(1) − 12 = 138

138 2 102 − 77
f ( 2) = 8 + (1) = 77 εt = 100% = 24.5%
2 102

third order:

f ( 3)
(1) = 150

150 3 102 − 102

f ( 2) = 77 + (1) = 102 εt = 100% = 0.0%
6 102

Because we are working with a third-order polynomial, the error is zero. This is due to the
fact that cubics have zero fourth and higher derivatives.

4.7 The true value is ln(3) = 1.098612

zero order:

1.098612 − 0
f (3) = f (1) = 0 εt = 100% = 100%
1.098612

first order:

1
f ' ( x) = f ' (1) = 1
x
1.098612 − 2
f (3) = 0 + 1(2) = 2 εt = 100% = 82.05%
1.098612

second order:

1
f " ( x) = − f " (1) = −1
x2
22 1.098612 − 0
f (3) = 2 − 1 =0 εt = 100% = 100%
2 1.098612

third order:

2
f ( 3)
( x) = f " (1) = 2
x3
23 1.098612 − 2.66667
f (3) = 0 + 2 = 2.66667 εt = 100% = 142.7%
6 1.098612

fourth order:

26
 6
f ( 4)
( x) = − f ( 4)
(1) = −6
x4
24 1.098612 − (−1.33333)
f (3) = 2.66666 − 6 = −1.33333 εt = 100% = 221.4%
24 1.098612

The series is diverging. A smaller step size is required to obtain convergence.

4.8 The first derivative of the function at x = 2 can be evaluated as

f ' (2) = 75(2) 2 − 12(2) + 7 = 283

The points needed to form the finite divided differences can be computed as

xi–1 = 1.75 f(xi–1) = 39.85938

xi = 2.0 f(xi) = 102
xi+1 = 2.25 f(xi+1) = 182.1406

forward:

182.1406 − 102
f ' ( 2) = = 320.5625 E t = 283 − 320.5625 = 37.5625
0.25

backward:

102 − 39.85938
f ' ( 2) = = 248.5625 E t = 283 − 248.5625 = 34.4375
0.25

centered:

182.1406 − 39.85938
f ' ( 2) = = 284.5625 E t = 283 − 284.5625 = −1.5625
0 .5

Both the forward and backward differences should have errors approximately equal to

f " ( xi )
Et ≈ h
2

The second derivative can be evaluated as

f " (2) = 150(2) − 12 = 288

Therefore,

288
Et ≈ 0.25 = 36
2

which is similar in magnitude to the computed errors.

27
 For the central difference,

( 3)
f ( xi ) 2
Et ≈ − h
6

The third derivative of the function is 150 and

150
Et ≈ − (0.25) 2 = −1.5625
6

which is exact. This occurs because the underlying function is a cubic equation that has zero
fourth and higher derivatives.

4.9 The second derivative of the function at x = 2 can be evaluated as

f ' (2) = 150( 2) − 12 = 288

For h = 0.2,

164.56 − 2(102) + 50.96

f " ( 2) = = 288
(0.2) 2

For h = 0.1,

131.765 − 2(102) + 75.115

f " ( 2) = = 288
(0.1) 2

Both are exact because the errors are a function of fourth and higher derivatives which are
zero for a 3rd-order polynomial.

4.10 Use εs = 0.5×102–2 = 0.5%. The true value = 1/(1 – 0.1) = 1.11111…

zero-order:

1
≅1
1 − 0.1

1.11111 − 1
εt = 100% = 10%
1.11111

first-order:

1
≅ 1 + 0 .1 = 1 .1
1 − 0.1

28
 1.11111 − 1.1
εt = 100% = 1%
1.11111

1 .1 − 1
εa = 100% = 9.0909%
1 .1

second-order:

1
≅ 1 + 0.1 + 0.01 = 1.11
1 − 0.1

1.11111 − 1.11
εt = 100% = 0.1%
1.11111

1.11 − 1.1
εa = 100% = 0.9009%
1.11

third-order:

1
≅ 1 + 0.1 + 0.01 + 0.001 = 1.111
1 − 0.1

1.11111 − 1.111
εt = 100% = 0.01%
1.11111

1.111 − 1.11
εa = 100% = 0.090009%
1.111

The approximate error has fallen below 0.5% so the computation can be terminated.

4.11 Here are the function and its derivatives

1
f ( x) = x − 1 − sin x
2
1
f ' ( x) = 1 − cos x
2
1
f " ( x) = sin x
2
1
f ( 3)
( x) = cos x
2
1
f ( 4)
( x) = − sin x
2

29
Using the Taylor Series expansion, we obtain the following 1st, 2nd, 3rd, and 4th order Taylor
Series functions shown below in the MATLAB program−f1, f2, and f4. Note the 2nd and
3rd order Taylor Series functions are the same.

From the plots below, we see that the answer is the 4th Order Taylor Series expansion.

x=0:0.001:3.2;
f=x-1-0.5*sin(x);
subplot(2,2,1);
plot(x,f);grid;title('f(x)=x-1-0.5*sin(x)');hold on

f1=x-1.5;
e1=abs(f-f1); %Calculates the absolute value of the
difference/error
subplot(2,2,2);
plot(x,e1);grid;title('1st Order Taylor Series Error');

f2=x-1.5+0.25.*((x-0.5*pi).^2);
e2=abs(f-f2);
subplot(2,2,3);
plot(x,e2);grid;title('2nd/3rd Order Taylor Series Error');

f4=x-1.5+0.25.*((x-0.5*pi).^2)-(1/48)*((x-0.5*pi).^4);
e4=abs(f4-f);
subplot(2,2,4);
plot(x,e4);grid;title('4th Order Taylor Series Error');hold off

f(x)=x-1-0.5*sin(x) 1st Order Taylor Series Error

3 0.8

2 0.6

1 0.4

0 0.2

-1 0
0 1 2 3 4 0 1 2 3 4

2nd/3rd Order Taylor Series Error 4th Order Taylor Series Error
0.2 0.015

0.15
0.01

0.1
0.005
0.05

0 0
0 1 2 3 4 0 1 2 3 4

30
4.12
x f(x) f(x-1) f(x+1) f'(x)-Theory f'(x)-Back f'(x)-Cent f'(x)-Forw
-2.000 0.000 -2.891 2.141 10.000 11.563 10.063 8.563
-1.750 2.141 0.000 3.625 7.188 8.563 7.250 5.938
-1.500 3.625 2.141 4.547 4.750 5.938 4.813 3.688
-1.250 4.547 3.625 5.000 2.688 3.688 2.750 1.813
-1.000 5.000 4.547 5.078 1.000 1.813 1.063 0.313
-0.750 5.078 5.000 4.875 -0.313 0.313 -0.250 -0.813
-0.500 4.875 5.078 4.484 -1.250 -0.813 -1.188 -1.563
-0.250 4.484 4.875 4.000 -1.813 -1.563 -1.750 -1.938
0.000 4.000 4.484 3.516 -2.000 -1.938 -1.938 -1.938
0.250 3.516 4.000 3.125 -1.813 -1.938 -1.750 -1.563
0.500 3.125 3.516 2.922 -1.250 -1.563 -1.188 -0.813
0.750 2.922 3.125 3.000 -0.313 -0.813 -0.250 0.313
1.000 3.000 2.922 3.453 1.000 0.313 1.063 1.813
1.250 3.453 3.000 4.375 2.688 1.813 2.750 3.688
1.500 4.375 3.453 5.859 4.750 3.688 4.813 5.938
1.750 5.859 4.375 8.000 7.188 5.938 7.250 8.563
2.000 8.000 5.859 10.891 10.000 8.563 10.063 11.563

First Derivative Approximations Compared to Theoretical

14.0

12.0

10.0

8.0

6.0 Theoretical
Backward
f'(x)

Centered
4.0
Forward

2.0

0.0
-2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5
-2.0

-4.0
x-values

x f(x) f(x-1) f(x+1) f(x-2) f(x+2) f''(x)- f''(x)- f''(x)-Cent f''(x)-

Theory Back Forw
-2.000 0.000 -2.891 2.141 3.625 3.625 -12.000 150.500 -12.000 -10.500
-1.750 2.141 0.000 3.625 -2.891 4.547 -10.500 -12.000 -10.500 -9.000
-1.500 3.625 2.141 4.547 0.000 5.000 -9.000 -10.500 -9.000 -7.500
-1.250 4.547 3.625 5.000 2.141 5.078 -7.500 -9.000 -7.500 -6.000
-1.000 5.000 4.547 5.078 3.625 4.875 -6.000 -7.500 -6.000 -4.500
-0.750 5.078 5.000 4.875 4.547 4.484 -4.500 -6.000 -4.500 -3.000
-0.500 4.875 5.078 4.484 5.000 4.000 -3.000 -4.500 -3.000 -1.500
-0.250 4.484 4.875 4.000 5.078 3.516 -1.500 -3.000 -1.500 0.000
0.000 4.000 4.484 3.516 4.875 3.125 0.000 -1.500 0.000 1.500
0.250 3.516 4.000 3.125 4.484 2.922 1.500 0.000 1.500 3.000

31
 0.500 3.125 3.516 2.922 4.000 3.000 3.000 1.500 3.000 4.500
0.750 2.922 3.125 3.000 3.516 3.453 4.500 3.000 4.500 6.000
1.000 3.000 2.922 3.453 3.125 4.375 6.000 4.500 6.000 7.500
1.250 3.453 3.000 4.375 2.922 5.859 7.500 6.000 7.500 9.000
1.500 4.375 3.453 5.859 3.000 8.000 9.000 7.500 9.000 10.500
1.750 5.859 4.375 8.000 3.453 10.891 10.500 9.000 10.500 12.000
2.000 8.000 5.859 10.891 4.375 14.625 12.000 10.500 12.000 13.500

Approximations of the 2nd Derivative

15.0

10.0

5.0

f''(x)-Theory
f''(x)-Backward
f''(x)

0.0
f''(x)-Centered
-2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5
f''(x)-Forward

-5.0

-10.0

-15.0
x-values

4.13

function eps = macheps

% determines the machine epsilon
e = 1;
while e+1>1
e = e/2;
end
eps = 2*e;

>> macheps

ans =
2.2204e-016

>> eps

ans =
2.2204e-016

32
CHAPTER 5
5.1 The function to evaluate is

gm ⎛ gc d ⎞
f (c d ) = tanh⎜ t ⎟ − v(t )
cd ⎜ m ⎟
⎝ ⎠

or substituting the given values

9.81(80) ⎛ 9.81c d ⎞
f (c d ) = tanh⎜ 4 ⎟ − 36
cd ⎜ 80 ⎟
⎝ ⎠

The first iteration is

0 .1 + 0 .2
xr = = 0.15
2

f (0.1) f (0.15) = 0.860291(−0.204516) = −0.175944

Therefore, the root is in the first interval and the upper guess is redefined as xu = 0.15. The
second iteration is

0.1 + 0.15
xr = = 0.125
2

0.125 − 0.15
εa = 100% = 20%
0.125

f (0.1) f (0.125) = 0.860291(0.318407) = 0.273923

Therefore, the root is in the second interval and the lower guess is redefined as xu = 0.125.
The remainder of the iterations are displayed in the following table:

i xl f(xl) xu f(xu) xr f(xr) |εa|

1 0.1 0.86029 0.2 −1.19738 0.15 −0.20452
2 0.1 0.86029 0.15 −0.20452 0.125 0.31841 20.00%
3 0.125 0.31841 0.15 −0.20452 0.1375 0.05464 9.09%
4 0.1375 0.05464 0.15 −0.20452 0.14375 −0.07551 4.35%
5 0.1375 0.05464 0.14375 −0.07551 0.140625 −0.01058 2.22%
6 0.1375 0.05464 0.140625 −0.01058 0.1390625 0.02199 1.12%

Thus, after six iterations, we obtain a root estimate of 0.1390625 with an approximate error of
1.12%.

5.2
function root = bisectnew(func,xl,xu,Ead)
% bisectnew(xl,xu,es,maxit):

33
% uses bisection method to find the root of a function
% with a fixed number of iterations to attain
% a prespecified tolerance
% input:
% func = name of function
% xl, xu = lower and upper guesses
% Ead = (optional) desired tolerance (default = 0.000001)
% output:
% root = real root

if func(xl)*func(xu)>0 %if guesses do not bracket a sign change

error('no bracket') %display an error message and terminate
end
% if necessary, assign default values
if nargin<4, Ead = 0.000001; end %if Ead blank set to 0.000001
% bisection
xr = xl;
% compute n and round up to next highest integer
n = round(1 + log2((xu - xl)/Ead) + 0.5);
for i = 1:n
xrold = xr;
xr = (xl + xu)/2;
if xr ~= 0, ea = abs((xr - xrold)/xr) * 100; end
test = func(xl)*func(xr);
if test < 0
xu = xr;
elseif test > 0
xl = xr;
else
ea = 0;
end
end
root = xr;

The following is a MATLAB session that uses the function to solve Prob. 5.1 with Ea,d =
0.0001.

>> fcd = inline('sqrt(9.81*80/cd)*tanh(sqrt(9.81*cd/80)*4)-36','cd')

fcd =
Inline function:
fcd(cd) = sqrt(9.81*80/cd)*tanh(sqrt(9.81*cd/80)*4)-36

>> format long

>> bisectnew(fcd,0.1,0.2,0.0001)

ans =
0.14008789062500

5.3 The function to evaluate is

9.81(80) ⎛ 9.81c d ⎞
f (c d ) = tanh ⎜ 4 ⎟ − 36
cd ⎜ 80 ⎟
⎝ ⎠

34
 The first iteration is

− 1.19738(0.1 − 0.2)
x r = 0.2 − = 0.141809
0.86029 − (−1.19738)

f (0.1) f (0.141809) = 0.860291( −0.03521) = −0.030292

Therefore, the root is in the first interval and the upper guess is redefined as xu = 0.141809.
The second iteration is

− 0.03521(0.1 − 0.141809)
x r = 0.141809 − = 0.140165
0.86029 − (−0.03521)

0.140165 − 0.141809
εa = 100% = 1.17%
0.140165

Therefore, after only two iterations we obtain a root estimate of 0.140165 with an
approximate error of 1.17% which is below the stopping criterion of 2%.

5.4
function root = falsepos(func,xl,xu,es,maxit)
% falsepos(xl,xu,es,maxit):
% uses the false position method to find the root
% of the function func
% input:
% func = name of function
% xl, xu = lower and upper guesses
% es = (optional) stopping criterion (%) (default = 0.001)
% maxit = (optional) maximum allowable iterations (default = 50)
% output:
% root = real root
if func(xl)*func(xu)>0 %if guesses do not bracket a sign change
error('no bracket') %display an error message and terminate
end
% default values
if nargin<5, maxit=50; end
if nargin<4, es=0.001; end
% false position
iter = 0;
xr = xl;
while (1)
xrold = xr;
xr = xu - func(xu)*(xl - xu)/(func(xl) - func(xu));
iter = iter + 1;
if xr ~= 0, ea = abs((xr - xrold)/xr) * 100; end
test = func(xl)*func(xr);
if test < 0
xu = xr;
elseif test > 0
xl = xr;
else
ea = 0;

35
 end
if ea <= es | iter >= maxit, break, end
end
root = xr;

The following is a MATLAB session that uses the function to solve Prob. 5.1:

>> fcd = inline('sqrt(9.81*80/cd)*tanh(sqrt(9.81*cd/80)*4)-36','cd')

fcd =
Inline function:
fcd(cd) = sqrt(9.81*80/cd)*tanh(sqrt(9.81*cd/80)*4)-36

>> format long

>> falsepos(fcd,0.1,0.2,2)

ans =
0.14016503741282

5.5 Solve for the reactions:

R1 = 265 lbs. R2 = 285 lbs.

Write beam equations:

x
M + (16.667 x 2 )− 265 x = 0
0<x<3 3
(1) M = 265 − 5.55 x 3

x−3 2
M + 100( x − 3)() + 150( x − (3)) − 265 x = 0
3<x<6 2 3
(2) M = −50 x + 415 x − 150
2

2
M = 150( x − (3)) + 300( x − 4.5) − 265 x
6<x<10 3
(3) M = −185 x + 1650

M + 100(12 − x) = 0
10<x<12
(4) M = 100 x − 1200

Combining Equations:

Because the curve crosses the axis between 6 and 10, use (3).

(3) M = −185 x + 1650

Set x L = 6; xU = 10

36
 M ( x L ) = 540 x L + xU
xr = =8
M ( xU ) = −200 2
M ( x R ) = 170 → replaces x L

M ( x L ) = 170 8 + 10
xr = =9
M ( xU ) = −200 2
M ( x R ) = −15 → replaces xU

M ( x L ) = 170 8+9
xr = = 8.5
M ( xU ) = −15 2
M ( x R ) = 77.5 → replaces x L

M ( x L ) = 77.5 8.5 + 9
xr = = 8.75
M ( xU ) = −15 2
M ( x R ) = 31.25 → replaces x L

M ( x L ) = 31.25 8.75 + 9
xr = = 8.875
M ( xU ) = −15 2
M ( x R ) = 8.125 → replaces x L

M ( x L ) = 8.125 8.875 + 9
xr = = 8.9375
M ( xU ) = −15 2
M ( x R ) = −3.4375 → replaces xU

M ( x L ) = 8.125 8.875 + 8.9375

xr = = 8.90625
M ( xU ) = −3.4375 2
M ( x R ) = 2.34375 → replaces x L

M ( x L ) = 2.34375 8.90625 + 8.9375

xr = = 8.921875
M ( xU ) = −3.4375 2
M ( x R ) = −0.546875 → replaces xU

M ( x L ) = 2.34375 8.90625 + 8.921875

xr = = 8.9140625
M ( xU ) = −0.546875 2

M ( x R ) = 0.8984 Therefore, x = 8.91 feet

5.6 (a) The graph can be generated with MATLAB

>> x=[-1:0.1:6];

37
>> f=-12-21*x+18*x.^2-2.75*x.^3;
>> plot(x,f)
>> grid

This plot indicates that roots are located at about –0.4, 2.25 and 4.7.

(b) Using bisection, the first iteration is

−1+ 0
xr = = −0.5
2

f ( −1) f (−0.5) = 29.75(3.34375) = 99.47656

Therefore, the root is in the second interval and the lower guess is redefined as xl = –0.5. The
second iteration is

− 0.5 + 0
xr = = −0.25
2

− 0.25 − ( −0.5)
εa = 100% = 100%
− 0.25

f ( −0.5) f (−0.25) = 3.34375(−5.5820313) = −18.66492

Therefore, the root is in the first interval and the upper guess is redefined as xu = –0.25. The
remainder of the iterations are displayed in the following table:

i xl f(xl) xu f(xu) xr f(xr) |εa|

1 −1 29.75 0 −12 −0.5 3.34375
2 −0.5 3.34375 0 −12 −0.25 −5.5820313 100.00%
3 −0.5 3.34375 −0.25 −5.5820313 −0.375 −1.4487305 33.33%

38
 4 −0.5 3.34375 −0.375 −1.4487305 −0.4375 0.8630981 14.29%
5 −0.4375 0.863098 −0.375 −1.4487305 −0.40625 −0.3136673 7.69%
6 −0.4375 0.863098 −0.40625 −0.3136673 −0.421875 0.2694712 3.70%
7 −0.42188 0.269471 −0.40625 −0.3136673 −0.414063 −0.0234052 1.89%
8 −0.42188 0.269471 −0.41406 −0.0234052 −0.417969 0.1227057 0.93%

Thus, after eight iterations, we obtain a root estimate of −0.417969 with an approximate error
of 0.93%, which is below the stopping criterion of 1%.

(c) Using false position, the first iteration is

− 12(−1 − 0)
xr = 0 − = −0.287425
29.75 − (−12)

f ( −1) f (−0.287425) = 29.75(−4.4117349) = −131.2491

Therefore, the root is in the first interval and the upper guess is redefined as xu = –0.287425.
The second iteration is

− 4.4117349(−1 − (−0.287425))
x r = −0.287425 − = −0.3794489
29.75 − (−4.4117349)

− 0.3794489 − (−0.2874251)
εa = 100% = 24.25%
− 0.3794489

f ( −1) f (−0.3794489) = 29.75(−1.2896639) = −38.3675

Therefore, the root is in the first interval and the upper guess is redefined as xu = –0.379449.
The remainder of the iterations are displayed in the following table:

i xl f(xl) xu f(xu) xr f(xr) |εa|

1 −1 29.75 0 −12 −0.287425 −4.4117349
2 −1 29.75 −0.28743 −4.4117349 −0.379449 −1.2896639 24.25%
3 −1 29.75 −0.37945 −1.2896639 −0.405232 −0.3512929 6.36%
4 −1 29.75 −0.40523 −0.3512929 −0.412173 −0.0938358 1.68%
5 −1 29.75 −0.41217 −0.0938358 −0.414022 −0.0249338 0.45%

Therefore, after five iterations we obtain a root estimate of –0.414022 with an approximate
error of 0.45%, which is below the stopping criterion of 1%.

5.7 A graph of the function can be generated with MATLAB

>> x=[-0.5:0.1:1.5];
>> f=sin(x)-x.^2;
>> plot(x,f)
>> grid

39
This plot indicates that a nontrivial root (i.e., nonzero) is located at about 0.85.

Using bisection, the first iteration is

0 .5 + 1
xr = = 0.75
2

f (0.5) f (0.75) = 0.229426(0.1191388) = 0.027333

Therefore, the root is in the second interval and the lower guess is redefined as xl = 0.75. The
second iteration is

0.75 + 1
xr = = 0.875
2

0.875 − 0.75
εa = 100% = 14.29%
0.875

f (0.75) f (0.875) = 0.119139(0.0019185) = 0.000229

Because the product is positive, the root is in the second interval and the lower guess is
redefined as xl = 0.875. The remainder of the iterations are displayed in the following table:

i xl f(xl) xu f(xu) xr f(xr) |εa|

1 0.5 0.229426 1 −0.158529 0.75 0.1191388
2 0.75 0.119139 1 −0.158529 0.875 0.0019185 14.29%
3 0.875 0.001919 1 −0.158529 0.9375 −0.0728251 6.67%
4 0.875 0.001919 0.9375 −0.0728251 0.90625 −0.0340924 3.45%
5 0.875 0.001919 0.90625 −0.0340924 0.890625 −0.0157479 1.75%

40
 Therefore, after five iterations we obtain a root estimate of 0.890625 with an approximate
error of 1.75%, which is below the stopping criterion of 2%.

5.8 (a) A graph of the function indicates a positive real root at approximately x = 1.4.

0
-3 -2 -1 0 1 2 3
-2

-4

-6

-8

-10

-12

(b) Using bisection, the first iteration is

0 .5 + 2
xr = = 1.25
2

f (0.5) f (1.25) = −2.08629(−0.2537129) = 0.52932

Therefore, the root is in the second interval and the lower guess is redefined as xl = 1.25. The
second iteration is

1.25 + 2
xr = = 1.625
2

1.625 − 1.25
εa = 100% = 23.08%
1.625

f (1.25) f (1.625) = −0.253713(0.2710156) = −0.06876

Therefore, the root is in the first interval and the upper guess is redefined as xu = 1.625. The
remainder of the iterations are displayed in the following table:

i xl f(xl) xu f(xu) xr f(xr) |εa|

1 0.5 −2.08629 2 0.6862944 1.25 −0.2537129
2 1.25 −0.25371 2 0.6862944 1.625 0.2710156 23.08%
3 1.25 −0.25371 1.625 0.2710156 1.4375 0.025811 13.04%

41
 Thus, after three iterations, we obtain a root estimate of 1.4375 with an approximate error of
13.04%.

(c) Using false position, the first iteration is

0.6862944(0.5 − 2)
xr = 2 − = 1.628707
− 2.086294 − 0.6862944

f (0.5) f (1.628707) = −2.086294(0.2755734) = −0.574927

Therefore, the root is in the first interval and the upper guess is redefined as xu = 1.628707.
The second iteration is

1.4970143(0.5 − 1.628707)
x r = 0.2755734 − = 1.4970143
− 2.086294 − 0.2755734

1.4970143 − 1.6287074
εa = 100% = 8.8%
1.4970143

f (0.5) f (1.4970143) = −2.086294(0.1069453) = −0.223119

Therefore, the root is in the first interval and the upper guess is redefined as xu = 1.497014.
The remainder of the iterations are displayed in the following table:

i xl f(xl) xu f(xu) xr f(xr) |εa|

1 0.5 −2.08629 2 0.6862944 1.6287074 0.2755734
2 0.5 −2.08629 1.628707 0.2755734 1.4970143 0.1069453 8.80%
3 0.5 −2.08629 1.497014 0.1069453 1.4483985 0.040917 3.36%

Therefore, after three iterations we obtain a root estimate of 1.4483985 with an approximate
error of 3.36%.

5.9 (a) Equation (5.6) can be used to determine the number of iterations

⎛ ∆x 0 ⎞
n = 1 + log 2 ⎜ ⎟ = 1 + log 2 ⎛⎜ 35 ⎞⎟ = 10.45121
⎜E ⎟ ⎝ 0.05 ⎠
⎝ a,d ⎠

which can be rounded up to 11 iterations.

(b) Here is an M-file that evaluates the temperature in oC using 11 iterations of bisection
based on a given value of the oxygen saturation concentration in freshwater:

function TC = TempEval(osf)
% function to evaluate the temperature in degrees C based
% on the oxygen saturation concentration in freshwater (osf).
xl = 0 + 273.15;
xu = 35 + 273.15;
if fTa(xl,osf)*fTa(xu,osf)>0 %if guesses do not bracket
error('no bracket') %display an error message and terminate

42
 end
xr = xl;
for i = 1:11
xrold = xr;
xr = (xl + xu)/2;
if xr ~= 0, ea = abs((xr - xrold)/xr) * 100; end
test = fTa(xl,osf)*fTa(xr,osf);
if test < 0
xu = xr;
elseif test > 0
xl = xr;
else
ea = 0;
end
end
TC = xr - 273.15;
end

function f = fTa(Ta, osf)

f = -139.34411 + 1.575701e5/Ta - 6.642308e7/Ta^2;
f = f + 1.2438e10/Ta^3 - 8.621949e11/Ta^4;
f = f - log(osf);

The function can be used to evaluate the test cases:

>> TempEval(8)

ans =
26.7798

>> TempEval(10)

ans =
15.3979

>> TempEval(14)

ans =
1.5552

5.10 (a) The function to be evaluated is

400
f ( y) = 1 − (3 + y )
9.81(3 y + y 2 / 2) 3

A graph of the function indicates a positive real root at approximately 1.5.

43
 10

0
0 0.5 1 1.5 2 2.5

-10

-20

-30

-40

(b) Using bisection, the first iteration is

0 .5 + 2 .5
xr = = 1 .5
2

f (0.5) f (1.5) = −32.2582( −0.030946) = 0.998263

Therefore, the root is in the second interval and the lower guess is redefined as xl = 1.5. The
second iteration is

1 .5 + 2 .5
xr = =2
2

2 − 1.5
εa = 100% = 25%
2

f (1.5) f (2) = −0.030946(0.601809) = −0.018624

Therefore, the root is in the first interval and the upper guess is redefined as xu = 2. The
remainder of the iterations are displayed in the following table:

i xl f(xl) xu f(xu) xr f(xr) |εa|

1 0.5 −32.2582 2.5 0.813032 1.5 −0.030946
2 1.5 −0.03095 2.5 0.813032 2 0.601809 25.00%
3 1.5 −0.03095 2 0.601809 1.75 0.378909 14.29%
4 1.5 −0.03095 1.75 0.378909 1.625 0.206927 7.69%
5 1.5 −0.03095 1.625 0.206927 1.5625 0.097956 4.00%
6 1.5 −0.03095 1.5625 0.097956 1.53125 0.036261 2.04%
7 1.5 −0.03095 1.53125 0.036261 1.515625 0.003383 1.03%
8 1.5 −0.03095 1.515625 0.003383 1.5078125 −0.013595 0.52%

44
After eight iterations, we obtain a root estimate of 1.5078125 with an approximate error of
0.52%.

(c) Using false position, the first iteration is

0.81303(0.5 − 2.5)
x r = 2.5 − = 2.45083
− 32.2582 − 0.81303

f (0.5) f (2.45083) = −32.25821(0.79987) = −25.80248

Therefore, the root is in the first interval and the upper guess is redefined as xu = 2.45083.
The second iteration is

0.79987(0.5 − 2.45083)
x r = 2.45083 − = 2.40363
− 32.25821 − 0.79987

2.40363 − 2.45083
εa = 100% = 1.96%
2.40363

f (0.5) f (2.40363) = −32.2582(0.78612) = −25.35893

The root is in the first interval and the upper guess is redefined as xu = 2.40363. The
remainder of the iterations are displayed in the following table:

i xl f(xl) xu f(xu) xr f(xr) |εa|

1 0.5 −32.2582 2.50000 0.81303 2.45083 0.79987
2 0.5 −32.2582 2.45083 0.79987 2.40363 0.78612 1.96%
3 0.5 −32.2582 2.40363 0.78612 2.35834 0.77179 1.92%
4 0.5 −32.2582 2.35834 0.77179 2.31492 0.75689 1.88%
5 0.5 −32.2582 2.31492 0.75689 2.27331 0.74145 1.83%
6 0.5 −32.2582 2.27331 0.74145 2.23347 0.72547 1.78%
7 0.5 −32.2582 2.23347 0.72547 2.19534 0.70900 1.74%
8 0.5 −32.2582 2.19534 0.70900 2.15888 0.69206 1.69%
9 0.5 −32.2582 2.15888 0.69206 2.12404 0.67469 1.64%
10 0.5 −32.2582 2.12404 0.67469 2.09077 0.65693 1.59%

After ten iterations we obtain a root estimate of 2.09077 with an approximate error of 1.59%.
Thus, after ten iterations, the false position method is converging at a very slow pace and is
still far from the root in the vicinity of 1.5 that we detected graphically.

Discussion: This is a classic example of a case where false position performs poorly and is
inferior to bisection. Insight into these results can be gained by examining the plot that was
developed in part (a). This function violates the premise upon which false position was
based−that is, if f(xu) is much closer to zero than f(xl), then the root is closer to xu than to xl
(recall Fig. 5.8). Because of the shape of the present function, the opposite is true.

45
CHAPTER 6

6.1 The function can be set up for fixed-point iteration by solving it for x

xi +1 = sin (x)
i

Using an initial guess of x0 = 0.5, the first iteration yields

( )
x1 = sin 0.5 = 0.649637

0.649637 − 0.5
εa = × 100% = 23%
0.649637

Second iteration:

(
x 2 = sin 0.649637 = 0.721524 )
0.721524 − 0.649637
εa = × 100% = 9.96%
0.721524

The process can be continued as tabulated below:

iteration xi |εa|
0 0.500000
1 0.649637 23.0339%
2 0.721524 9.9632%
3 0.750901 3.9123%
4 0.762097 1.4691%
5 0.766248 0.5418%
6 0.767772 0.1984%
7 0.768329 0.0725%
8 0.768532 0.0265%
9 0.768606 0.0097%

Thus, after nine iterations, the root is estimated to be 0.768606 with an approximate error
of 0.0097%.

6.2 (a) The function can be set up for fixed-point iteration by solving it for x in two different
ways. First, it can be solved for the linear x,

0.9 x i2 − 2.5
x i +1 =
1 .7

Using an initial guess of 5, the first iteration yields

46
 0.9(5) 2 − 2.5
x1 = = 11.76
1.7

11.76 − 5
εa = × 100% = 57.5%
11.76

Second iteration:

0.9(11.76) 2 − 2.5
x1 = = 71.8
1 .7

71.8 − 11.76
εa = × 100% = 83.6%
71.8

Clearly, this solution is diverging.

An alternative is to solve for the second-order x,

1.7 xi + 2.5
x i +1 =
0.9

Using an initial guess of 5, the first iteration yields

1.7(5) + 2.5
x i +1 = = 3.496
0 .9

3.496 − 5
εa = × 100% = 43.0%
3.496

Second iteration:

1.7(3.496) + 2.5
x i +1 = = 3.0629
0.9

3.0629 − 3.496
εa = × 100% = 14.14%
3.0629

This version is converging. All the iterations can be tabulated as

iteration xi |εa|
0 5.000000
1 3.496029 43.0194%
2 3.062905 14.1410%
3 2.926306 4.6680%
4 2.881882 1.5415%
5 2.867287 0.5090%

47
 6 2.862475 0.1681%
7 2.860887 0.0555%
8 2.860363 0.0183%
9 2.860190 0.0061%

Thus, after 9 iterations, the root estimate is 2.860190 with an approximate error of
0.0061%. The result can be checked by substituting it back into the original function,

f ( 2.860190) = −0.9(2.860190) 2 + 1.7(2.860190) + 2.5 = −0.000294

(b) The formula for Newton-Raphson is

− 0.9 xi2 + 1.7 x i + 2.5

x i +1 = xi −
− 1.8 xi + 1.7

Using an initial guess of 5, the first iteration yields

− 0.9(5) 2 + 1.7(5) + 2.5

x i +1 = 5 − = 3.424658
− 1.8(5) + 1.7

3.424658 − 5
εa = × 100% = 46.0%
3.424658

Second iteration:

− 0.9(3.424658) 2 + 1.7(3.424658) + 2.5

x i +1 = 3.424658 − = 2.924357
− 1.8(3.424658) + 1.7

2.924357 − 3.424658
εa = × 100% = 17.1%
2.924357

The process can be continued as tabulated below:

iteration xi f(xi) f'(xi) |εa|

0 5 −11.5 −7.3
1 3.424658 −2.23353 −4.46438 46.0000%
2 2.924357 −0.22527 −3.56384 17.1081%
3 2.861147 −0.00360 −3.45006 2.2093%
4 2.860105 −9.8E−07 −3.44819 0.0364%
5 2.860104 −7.2E−14 −3.44819 0.0000%

After 5 iterations, the root estimate is 2.860104 with an approximate error of 0.0000%. The
result can be checked by substituting it back into the original function,

f ( 2.860104) = −0.9(2.860104) 2 + 1.7(2.860104) + 2.5 = −7.2 × 10 −14

48
6.3 (a)

>> x = linspace(0,4);
>> y = x.^3-6*x.^2+11*x-6.1;
>> plot(x,y)
>> grid

Estimates are approximately 1.05, 1.9 and 3.05.

(b) The formula for Newton-Raphson is

x i3 − 6 x i2 + 11xi − 6.1
x i +1 = x i −
3 x i2 − 12 xi + 11

Using an initial guess of 3.5, the first iteration yields

(3.5) 3 − 6(3.5) 2 + 11(3.5) − 6.1

x1 = 3.5 − = 3.191304
3(3.5) 2 − 12(3.5) + 11

3.191304 − 3.5
εa = × 100% = 9.673%
3.191304

Second iteration:

(3.191304) 3 − 6(3.191304) 2 + 11(3.191304) − 6.1

x 2 = 3.191304 − = 3.068699
3(3.191304) 2 − 12(3.191304) + 11

3.068699 − 3.191304
εa = × 100% = 3.995%
3.068699

49
Third iteration:

(3.068699) 3 − 6(3.068699) 2 + 11(3.068699) − 6.1

x 3 = 3.068699 − = 3.047317
3(3.068699) 2 − 12(3.068699) + 11

3.047317 − 3.068699
εa = × 100% = 0.702%
3.047317

(c) For the secant method, the first iteration:

x−1 = 2.5 f(x−1) = −0.475

x0 = 3.5 f(x0) = 1.775

1.775(2.5 − 3.5)
x1 = 3.5 − = 2.711111
− 0.475 − 1.775

2.711111 − 3.5
εa = × 100% = 29.098%
2.711111

Second iteration:

x0 = 3.5 f(x0) = 1.775

x1 = 2.711111 f(x1) = −0.45152

− 0.45152(3.5 − 2.711111)
x 2 = 2.711111 − = 2.871091
1.775 − (−0.45152)

2.871091 − 2.711111
εa = × 100% = 5.572%
2.871091

Third iteration:

x1 = 2.711111 f(x1) = −0.45152

x2 = 2.871091 f(x2) = −0.31011

− 0.31011(2.711111 − 2.871091)
x 3 = 2.871091 − = 3.221923
− 0.45152 − (−0.31011)

3.221923 − 2.871091
εa = × 100% = 10.889%
3.221923

(d) For the modified secant method, the first iteration:

x0 = 3.5 f(x0) = 1.775

x0 + δx0 = 3.57 f(x0 + δx0) = 2.199893

50
 0.02(3.5)1.775
x1 = 3.5 − = 3.207573
2.199893 − 1.775

3.207573 − 3.5
εa = × 100% = 9.117%
3.207573

Second iteration:

x1 = 3.207573 f(x1) = 0.453351

x1 + δx1 = 3.271725 f(x1 + δx1) = 0.685016

0.02(3.207573)0.453351
x 2 = 3.207573 − = 3.082034
0.685016 − 0.453351

3.082034 − 3.207573
εa = × 100% = 4.073%
3.082034

Third iteration:

x2 = 3.082034 f(x2) = 0.084809

x2 + δx2 = 3.143675 f(x2 + δx2) = 0.252242

0.02(3.082034)0.084809
x 3 = 3.082034 − = 3.050812
0.252242 − 0.084809

3.050812 − 3.082034
εa = × 100% = 1.023%
3.050812

(e)

>> a = [1 -6 11 -6.1]

a =
1.0000 -6.0000 11.0000 -6.1000

>> roots(a)

ans =
3.0467
1.8990
1.0544

6.4 (a)

>> x = linspace(0,4);
>> y = 7*sin(x).*exp(-x)-1;
>> plot(x,y)
>> grid

51
The lowest positive root seems to be at approximately 0.2.

(b) The formula for Newton-Raphson is

7 sin( xi )e − xi − 1
x i +1 = x i −
7e − xi (cos( x i ) − sin( xi ))

Using an initial guess of 3.5, the first iteration yields

7 sin(0.3)e −0.3 − 1 0.532487

x1 = 0.3 − − 0.3
= 0 .3 − = 0.144376
7e (cos(0.3) − sin(0.3)) 3.421627

0.144376 − 0.3
εa = × 100% = 107.8%
0.144376

Second iteration:

7 sin(0.144376)e −0.144376 − 1 − 0.12827

x 2 = 0.144376 − − 0.144376
= 0.144376 − = 0.169409
7e (cos(0.144376) − sin(0.144376)) 5.124168

0.169409 − 0.144376
εa = × 100% = 14.776%
0.169409

Third iteration:

7 sin(0.169409)e −0.169409 − 1 − 0.00372

x1 = 0.169409 − − 0.169409
= 0.169409 − = 0.170179
7e (cos(0.169409) − sin(0.169409)) 4.828278

52
 0.170179 − 0.169409
εa = × 100% = 0.453%
0.170179

(c) For the secant method, the first iteration:

x−1 = 0.4 f(x−1) = 0.827244

x0 = 0.3 f(x0) = 0.532487

0.532487(0.4 − 0.3)
x1 = 0.3 − = 0.119347
0.827244 − 0.532487

0.119347 − 0.3
εa = × 100% = 151.4%
0.119347

Second iteration:

x0 = 0.3 f(x0) = 0.532487

x1 = 0.119347 f(x1) = −0.26032

− 0.26032(0.3 − 0.119347)
x 2 = 0.119347 − = 0.178664
0.532487 − (−0.26032)

0.178664 − 0.119347
εa = × 100% = 33.2%
0.178664

Third iteration:

x1 = 0.119347 f(x1) = −0.26032

x2 = 0.178664 f(x2) = 0.04047

0.04047(0.119347 − 0.178664)
x 3 = 0.178664 − = 0.170683
− 0.26032 − 0.04047

0.170683 − 0.178664
εa = × 100% = 4.68%
0.170683

(d) For the modified secant method, the first iteration:

x0 = 0.3 f(x0) = 0.532487

x0 + δx0 = 0.303 f(x0 + δx0) = 0.542708

0.01(0.3)0.532487
x1 = 0.3 − = 0.143698
0.542708 − 0.532487

0.143698 − 0.3
εa = × 100% = 108.8%
0.143698

53
 Second iteration:

x1 = 0.143698 f(x1) = −0.13175

x1 + δx1 = 0.145135 f(x1 + δx1) = −0.12439

0.02(0.143698)(−0.13175)
x 2 = 0.143698 − = 0.169412
− 0.12439 − (−0.13175)

0.169412 − 0.143698
εa = × 100% = 15.18%
0.169412

Third iteration:

x2 = 0.169412 f(x2) = −0.00371

x2 + δx2 = 0.171106 f(x2 + δx2) = 0.004456

0.02(0.169412)(−0.00371)
x 3 = 0.169412 − = 0.170181
0.004456 − (−0.00371)

0.170181 − 0.169412
εa = × 100% = 0.452%
0.170181

6.5 (a) The formula for Newton-Raphson is

xi5 − 16.05 xi4 + 88.75 x i3 − 192.0375 x i2 + 116.35 xi + 31.6875

x i +1 = x i −
5 x i4 − 64.2 x i3 + 266.25 xi2 − 384.075 xi + 116.35

Using an initial guess of 0.5825, the first iteration yields

50.06217
x1 = 0.5825 − = 2.300098
− 29.1466

2.300098 − 0.5825
εa = × 100% = 74.675%
2.300098

Second iteration

− 21.546
x1 = 2.300098 − = 90.07506
0.245468

90.07506 − 2.300098
εa = × 100% = 97.446%
90.07506

54
Thus, the result seems to be diverging. However, the computation eventually settles down
and converges (at a very slow rate) on a root at x = 6.5. The iterations can be summarized
as

iteration xi f(xi) f'(xi) |εa|

0 0.582500 50.06217 −29.1466
1 2.300098 −21.546 0.245468 74.675%
2 90.07506 4.94E+09 2.84E+08 97.446%
3 72.71520 1.62E+09 1.16E+08 23.874%
4 58.83059 5.3E+08 47720880 23.601%
5 47.72701 1.74E+08 19552115 23.265%
6 38.84927 56852563 8012160 22.852%
7 31.75349 18616305 3284098 22.346%
8 26.08487 6093455 1346654 21.731%
9 21.55998 1993247 552546.3 20.987%
10 17.95260 651370.2 226941 20.094%
11 15.08238 212524.6 93356.59 19.030%
12 12.80590 69164.94 38502.41 17.777%
13 11.00952 22415.54 15946.36 16.317%
14 9.603832 7213.396 6652.03 14.637%
15 8.519442 2292.246 2810.851 12.728%
16 7.703943 710.9841 1217.675 10.585%
17 7.120057 209.2913 556.1668 8.201%
18 6.743746 54.06896 286.406 5.580%
19 6.554962 9.644695 187.9363 2.880%
20 6.503643 0.597806 164.8912 0.789%
21 6.500017 0.00285 163.32 0.056%
22 6.5 6.58E−08 163.3125 0.000%

(b) For the modified secant method, the first iteration:

x0 = 0.5825 f(x0) = 50.06217

x0 + δx0 = 0.611625 f(x0 + δx0) = 49.15724

0.05(0.5825)50.06217
x1 = 0.5825 − = 2.193735
49.15724 − 50.06217

2.193735 − 0.5825
εa = × 100% = 73.447%
2.193735

Second iteration:

x1 = 2.193735 f(x1) = −21.1969

x1 + δx1 = 2.303422 f(x1 + δx1) = −21.5448

0.05(2.193735)(−21.1969)
x 2 = 2.193735 − = −4.48891
− 21.5448 − (−21.1969)

55
 − 4.48891 − 2.193735
εa = × 100% = 148.87%
− 4.48891

Again, the result seems to be diverging. However, the computation eventually settles down
and converges on a root at x = −0.2. The iterations can be summarized as

iteration xi xi+δxi f(xi) f(xi+ δxi) |εa|

0 0.5825 0.611625 50.06217 49.15724
1 2.193735 2.303422 −21.1969 −21.5448 73.447%
2 −4.48891 −4.71336 −20727.5 −24323.6 148.870%
3 −3.19524 −3.355 −7201.94 −8330.4 40.487%
4 −2.17563 −2.28441 −2452.72 −2793.57 46.865%
5 −1.39285 −1.46249 −808.398 −906.957 56.200%
6 −0.82163 −0.86271 −250.462 −277.968 69.524%
7 −0.44756 −0.46994 −67.4718 −75.4163 83.579%
8 −0.25751 −0.27038 −12.5942 −15.6518 73.806%
9 −0.20447 −0.2147 −0.91903 −3.05726 25.936%
10 −0.20008 −0.21008 −0.01613 −2.08575 2.196%
11 −0.2 −0.21 −0.0002 −2.0686 0.039%
12 −0.2 −0.21 −2.4E−06 −2.06839 0.000%

Explanation of results: The results are explained by looking at a plot of the function. The
guess of 0.5825 is located at a point where the function is relatively flat. Therefore, the first
iteration results in a prediction of 2.3 for Newton-Raphson and 2.193 for the secant method.
At these points the function is very flat and hence, the Newton-Raphson results in a very
high value (90.075), whereas the modified false position goes in the opposite direction to a
negative value (-4.49). Thereafter, the methods slowly converge on the nearest roots.

60

40

20

-2 0 2 4 6 8
-20
Modified Newton
secant Raphson
-40

-60

6.6
function root = secant(func,xrold,xr,es,maxit)
% secant(func,xrold,xr,es,maxit):
% uses secant method to find the root of a function
% input:
% func = name of function
% xrold, xr = initial guesses
% es = (optional) stopping criterion (%)

56
 % maxit = (optional) maximum allowable iterations
% output:
% root = real root

% if necessary, assign default values

if nargin<5, maxit=50; end %if maxit blank set to 50
if nargin<4, es=0.001; end %if es blank set to 0.001

% Secant method
iter = 0;
while (1)
xrn = xr - func(xr)*(xrold - xr)/(func(xrold) - func(xr));
iter = iter + 1;
if xrn ~= 0, ea = abs((xrn - xr)/xrn) * 100; end
if ea <= es | iter >= maxit, break, end
xrold = xr;
xr = xrn;
end
root = xrn;

Test by solving Prob. 6.3:

>> secant(inline('x^3-6*x^2+11*x-6.1'),2.5,3.5)

ans =
3.0467

6.7
function root = modsec(func,xr,delta,es,maxit)
% secant(func,xrold,xr,es,maxit):
% uses the modified secant method
% to find the root of a function
% input:
% func = name of function
% xr = initial guess
% delta = perturbation fraction
% es = (optional) stopping criterion (%)
% maxit = (optional) maximum allowable iterations
% output:
% root = real root

% if necessary, assign default values

if nargin<5, maxit=50; end %if maxit blank set to 50
if nargin<4, es=0.001; end %if es blank set to 0.001
if nargin<3, delta=1E-5; end %if delta blank set to 0.00001

% Secant method
iter = 0;
while (1)
xrold = xr;
xr = xr - delta*xr*func(xr)/(func(xr+delta*xr)-func(xr));
iter = iter + 1;
if xr ~= 0, ea = abs((xr - xrold)/xr) * 100; end
if ea <= es | iter >= maxit, break, end
end
root = xr;

57
 Test by solving Prob. 6.3:

>> modsec(inline('x^3-6*x^2+11*x-6.1'),3.5,0.02)

ans =
3.0467

6.8 The equation to be differentiated is

gm ⎛ gc d ⎞
f ( m) = tanh⎜ t⎟ − v
cd ⎜ m ⎟
⎝ ⎠

Note that

d tanh u du
= sech 2 u
dx dx

Therefore, the derivative can be evaluated as

df (m) gm ⎛ gc d ⎞⎛ 1 m ⎞ cd g ⎛ gc d ⎞ 1 cd g
= sech 2 ⎜ t ⎟⎜ − ⎟t + tanh ⎜ t⎟
dm cd ⎜ m ⎟⎜ 2 c d g ⎟ m2 ⎜ m ⎟2 gm c d
⎝ ⎠⎝ ⎠ ⎝ ⎠

The two terms can be reordered

df (m) 1 cd g ⎛ gc d ⎞ 1 gm m c d g ⎛ gc d ⎞
= tanh⎜ t⎟ − t sech 2 ⎜ t⎟
dm 2 gm c d ⎜ m ⎟ 2 cd cd g m 2 ⎜ m ⎟
⎝ ⎠ ⎝ ⎠

The terms premultiplying the tanh and sech can be simplified to yield the final result

df (m) 1 g ⎛ gc d ⎞ g ⎛ gc d ⎞
= tanh ⎜ t⎟ − t sech 2 ⎜ t⎟
dm 2 mc d ⎜ m ⎟ 2 m ⎜ m ⎟
⎝ ⎠ ⎝ ⎠

6.9 (a) The formula for Newton-Raphson is

− 2 + 6 xi − 4 xi2 + 0.5 x i3
x i +1 = x i −
6 − 8 xi + 1.5 xi2

Using an initial guess of 4.5, the iterations proceed as

iteration xi f(xi) f'(xi) |εa|

0 4.5 −10.4375 0.375
1 32.333330 12911.57 1315.5 86.082%
2 22.518380 3814.08 586.469 43.586%
3 16.014910 1121.912 262.5968 40.609%
4 11.742540 326.4795 118.8906 36.384%

58
 5 8.996489 92.30526 55.43331 30.524%
6 7.331330 24.01802 27.97196 22.713%
7 6.472684 4.842169 17.06199 13.266%
8 6.188886 0.448386 13.94237 4.586%
9 6.156726 0.005448 13.6041 0.522%
10 6.156325 8.39E−07 13.59991 0.007%

Thus, after an initial jump, the computation eventually settles down and converges on a
root at x = 6.156325.

(b) Using an initial guess of 4.43, the iterations proceed as

iteration xi f(xi) f'(xi) |εa|

0 4.43 -10.4504 -0.00265
1 -3939.13 -3.1E+10 23306693 100.112%
2 -2625.2 -9.1E+09 10358532 50.051%
3 -1749.25 -2.7E+09 4603793 50.076%
4 -1165.28 -8E+08 2046132 50.114%
5 -775.964 -2.4E+08 909393.5 50.171%
⋅
⋅
⋅
21 0.325261 -0.45441 3.556607 105.549%
22 0.453025 -0.05629 2.683645 28.203%
23 0.474 -0.00146 2.545015 4.425%
24 0.474572 -1.1E-06 2.541252 0.121%
25 0.474572 -5.9E-13 2.541249 0.000%

This time the solution jumps to an extremely large negative value The computation
eventually converges at a very slow rate on a root at x = 0.474572.

Explanation of results: The results are explained by looking at a plot of the function. Both
guesses are in a region where the function is relatively flat. Because the two guesses are on
opposite sides of a minimum, both are sent to different regions that are far from the initial
guesses. Thereafter, the methods slowly converge on the nearest roots.

59
 4

0
0 2 4 6

-4

-8

-12

6.10 The function to be evaluated is

x= a

This equation can be squared and expressed as a roots problem,

f ( x) = x 2 − a

The derivative of this function is

f ' ( x) = 2 x

These functions can be substituted into the Newton-Raphson equation (Eq. 6.6),

x i2 − a
x i +1 = xi −
2 xi

which can be expressed as

xi + a / xi
x i +1 =
2

6.11 (a) The formula for Newton-Raphson is

x i +1 = x i −
(
tanh x i2 − 9 )
2 x i sech 2
(
xi2 −9 )
Using an initial guess of 3.2, the iterations proceed as

iteration xi f(xi) f'(xi) |εa|

0 3.2 0.845456 1.825311
1 2.736816 −0.906910 0.971640 16.924%

60
 2 3.670197 0.999738 0.003844 25.431%
3 −256.413 101.431%

(b) The solution diverges from its real root of x = 3. Due to the concavity of the slope, the
next iteration will always diverge. The following graph illustrates how the divergence
evolves.

0.5

0
2.6 2.8 3 3.2 3.4

-0.5

-1

6.12 The formula for Newton-Raphson is

0.0074 xi4 − 0.284 xi3 + 3.355 x i2 − 12.183 x i + 5

x i +1 = x i −
0.0296 xi3 − 0.852 x i2 + 6.71xi − 12.1832

Using an initial guess of 16.15, the iterations proceed as

iteration xi f(xi) f'(xi) |εa|

0 16.15 −9.57445 −1.35368
1 9.077102 8.678763 0.662596 77.920%
2 −4.02101 128.6318 −54.864 325.742%
3 −1.67645 36.24995 −25.966 139.852%
4 −0.2804 8.686147 −14.1321 497.887%
5 0.334244 1.292213 −10.0343 183.890%
6 0.463023 0.050416 −9.25584 27.813%
7 0.46847 8.81E−05 −9.22351 1.163%
8 0.46848 2.7E−10 −9.22345 0.002%

As depicted below, the iterations involve regions of the curve that have flat slopes. Hence,
the solution is cast far from the roots in the vicinity of the original guess.

61
 10

0
-5 5 10 15 20
-5

-10

6.13 The solution involves determining the root of

x 6
f ( x) = − 0.05
1− x 2+ x

MATLAB can be used to develop a plot that indicates that a root occurs in the vicinity of x
= 0.03.

>> f = inline('x./(1-x).*sqrt(6./(2+x))-0.05')

f =
Inline function:
f(x) = x./(1-x).*sqrt(6./(2+x))-0.05

>> x = linspace(0,.2);
>> y = f(x);
>> plot(x,y)

The fzero function can then be used to find the root

62
 >> format long
>> fzero(f,0.03)

ans =
0.02824944114847

6.14 The coefficient, a and b, can be evaluated as

>> format long

>> R = 0.518;pc = 4600;Tc = 191;
>> a = 0.427*R^2*Tc^2.5/pc

a =
12.55778319740302

>> b = 0.0866*R*Tc/pc

b =
0.00186261539130

The solution, therefore, involves determining the root of

0.518(233.15) 12.557783
f (v) = 65,000 − +
v − 0.0018626 v(v + 0.0018626) 233.15

MATLAB can be used to generate a plot of the function and to solve for the root. One way
to do this is to develop an M-file for the function,

function y = fvol(v)
R = 0.518;pc = 4600;Tc = 191;
a = 0.427*R^2*Tc^2.5/pc;
b = 0.0866*R*Tc/pc;
T = 273.15-40;p = 65000;
y = p - R*T./(v-b)+a./(v.*(v+b)*sqrt(T));

This function is saved as fvol.m. It can then be used to generate a plot

>> v = linspace(0.002,0.004);
>> fv = fvol(v);
>> plot(v,fv)
>> grid

63
 Thus, a root is located at about 0.0028. The fzero function can be used to refine this
estimate,

>> vroot = fzero('fvol',0.0028)

vroot =
0.00280840865703

The mass of methane contained in the tank can be computed as

V 3
mass = = = 1068.317 m 3
v 0.0028084

6.15 The function to be evaluated is

⎡ ⎛r − h⎞ 2 ⎤
f ( h) = V − ⎢r 2 cos −1 ⎜ ⎟ − (r − h) 2rh − h ⎥ L
⎣ ⎝ r ⎠ ⎦

To use MATLAB to obtain a solution, the function can be written as an M-file

function y = fh(h,r,L,V)
y = V - (r^2*acos((r-h)/r)-(r-h)*sqrt(2*r*h-h^2))*L;

The fzero function can be used to determine the root as

>> format long

>> r = 2;L = 5;V = 8;
>> h = fzero('fh',0.5,[],r,L,V)

h =
0.74001521805594

64
6.16 (a) The function to be evaluated is

TA ⎛ 500 ⎞ T A
f (T A ) = 10 − cosh⎜⎜ ⎟⎟ +
10 ⎝ T A ⎠ 10

The solution can be obtained with the fzero function as

>> format long

>> TA = fzero(inline('10-x/10*cosh(500/x)+x/10'),1000)

TA =
1.266324360399887e+003

(b) A plot of the cable can be generated as

>> x = linspace(-50,100);
>> w = 10;y0 = 5;
>> y = TA/w*cosh(w*x/TA) + y0 - TA/w;
>> plot(x,y),grid

6.17 The function to be evaluated is

f (t ) = 9e −t sin(2πt ) − 3.5

A plot can be generated with MATLAB,

>> t = linspace(0,2);
>> y = 9*exp(-t) .* sin(2*pi*t) - 3.5;
>> plot(t,y),grid

65
 Thus, there appear to be two roots at approximately 0.1 and 0.4. The fzero function can be
used to obtain refined estimates,

>> t = fzero('9*exp(-x)*sin(2*pi*x)-3.5',[0 0.2])

t =
0.06835432096851

>> t = fzero('9*exp(-x)*sin(2*pi*x)-3.5',[0.2 0.8])

t =
0.40134369265980

6.18 The function to be evaluated is

2
1 1 ⎛ 1 ⎞
f (ω ) = − + ⎜ ωC − ⎟
Z R 2
⎝ ωL ⎠

Substituting the parameter values yields

2
1 ⎛ 2⎞
f (ω ) = 0.01 − + ⎜ 0.6 × 10 −6 ω − ⎟
50625 ⎝ ω⎠

The fzero function can be used to determine the root as

>> fzero('0.01-sqrt(1/50625+(.6e-6*x-2./x).^2)',[1 1000])

ans =
220.0202

6.19 The fzero function can be used to determine the root as

66
 >> format long
>> fzero('2*40*x^(5/2)/5+0.5*40000*x^2-95*9.8*x-95*9.8*0.43',1)

ans =
0.16662477900186

6.20 If the height at which the throw leaves the right fielders arm is defined as y = 0, the y at 90
m will be –0.8. Therefore, the function to be evaluated is

⎛ π ⎞ 44.1
f (θ ) = 0.8 + 90 tan ⎜ θ0 ⎟ −
⎝ 180 ⎠ cos (πθ 0 / 180)
2

Note that the angle is expressed in degrees. First, MATLAB can be used to plot this
function versus various angles.

Roots seem to occur at about 40o and 50o. These estimates can be refined with the fzero
function,

>> theta = fzero('0.8+90*tan(pi*x/180)-44.1./cos(pi*x/180).^2',0)

theta =
37.8380

>> theta = fzero('0.8+90*tan(pi*x/180)-44.1./cos(pi*x/180).^2',[40 60])

theta =
51.6527

Thus, the right fielder can throw at two different angles to attain the same result.

6.21 The equation to be solved is

67
 ⎛π ⎞
f ( h) = πRh 2 − ⎜ ⎟h 3 − V
⎝3⎠

Because this equation is easy to differentiate, the Newton-Raphson is the best choice to
achieve results efficiently. It can be formulated as

⎛π ⎞
πRxi2 − ⎜ ⎟ xi3 − V
x i +1 = x i − ⎝3⎠
2πRxi − πxi2

or substituting the parameter values,

⎛π ⎞
π (10) xi2 − ⎜ ⎟ xi3 − 1000
x i +1 = x i − ⎝3⎠
2π (10) xi − πx i2

The iterations can be summarized as

iteration xi f(xi) f'(xi) |εa|

0 10 1094.395 314.1593
1 6.516432 44.26917 276.0353 53.458%
2 6.356057 0.2858 272.4442 2.523%
3 6.355008 1.26E-05 272.4202 0.017%

Thus, after only three iterations, the root is determined to be 6.355008 with an approximate
relative error of 0.017%.

6.22
>> r = [-2 6 1 -4 8];
>> a = poly(r)

a =
1 -9 -20 204 208 -384

>> polyval(a,1)

ans =
0

>> b = poly([-2 6])

b =
1 -4 -12

>> [q,r] = deconv(a,b)

q =
1 -5 -28 32
r =
0 0 0 0 0 0

68
 >> x = roots(q)

x =
8.0000
-4.0000
1.0000

>> a = conv(q,b)

a =
1 -9 -20 204 208 -384

>> x = roots(a)

x =
8.0000
6.0000
-4.0000
-2.0000
1.0000

>> a = poly(x)

a =
1.0000 -9.0000 -20.0000 204.0000 208.0000 -384.0000

6.23
>> a = [1 9 26 24];
>> r = roots(a)

r =
-4.0000
-3.0000
-2.0000

>> a = [1 15 77 153 90];

>> r = roots(a)

r =
-6.0000
-5.0000
-3.0000
-1.0000

Therefore, the transfer function is

( s + 4)( s + 3)( s + 2)
G (s) =
( s + 6)( s + 5)( s + 3)( s + 1)

69
CHAPTER 7
7.1
>> Aug = [A eye(size(A))]

Here’s an example session of how it can be employed.

>> A = rand(3)

A =
0.9501 0.4860 0.4565
0.2311 0.8913 0.0185
0.6068 0.7621 0.8214

>> Aug = [A eye(size(A))]

Aug =

0.9501 0.4860 0.4565 1.0000 0 0

0.2311 0.8913 0.0185 0 1.0000 0
0.6068 0.7621 0.8214 0 0 1.0000

7.2 (a) [A]: 3 × 2 [B]: 3 × 3 {C}: 3 × 1 [D]: 2 × 4

[E]: 3 × 3 [F]: 2 × 3 ⎣G⎦: 1 × 3

(b) square: [B], [E]; column: {C}, row: ⎣G⎦

(c) a12 = 5, b23 = 6, d32 = undefined, e22 = 1, f12 = 0, g12 = 6

(d) MATLAB can be used to perform the operations

⎡5 8 13⎤ ⎡ 3 −2 1 ⎤
(1) [ E ] + [ B ] = ⎢8 3 9 ⎥ (2) [ E ] − [ B] = ⎢− 6 1 3⎥
⎣⎢5 0 9 ⎥⎦ ⎣⎢ − 3 0 − 1⎦⎥

⎡20 15 35⎤
(3) [A] + [F] = undefined (4) 5[ F ] = ⎢ 5 10 30⎥
⎢⎣ 5 0 20⎥⎦

⎡54 68⎤
(5) [A] × [B] = undefined (6) [ B] × [ A] = ⎢36 45⎥
⎢⎣24 29⎥⎦

(7) [G] × [C] = 56 (8) [C ]T = ⎣2 6 1⎦

⎡5 2⎤
⎡4 3 7 ⎤
(9) [ D] = ⎢⎢4
T 1⎥ (10) I × [B ] = ⎢1 2 6⎥
3 7⎥ ⎢⎣1 0 4⎥⎦
⎢⎣6 5⎥⎦

7.3 The terms can be collected to give

70
 ⎡− 7 3 0 ⎤ ⎧⎪ x1 ⎫⎪ ⎧⎪ 10 ⎫⎪
⎢ 0 4 7 ⎥ ⎨ x 2 ⎬ = ⎨− 30⎬
⎣⎢− 4 3 − 7⎦⎥ ⎪⎩ x 3 ⎪⎭ ⎪⎩ 40 ⎪⎭

Here is the MATLAB session:

>> A = [-7 3 0;0 4 7;-4 3 -7];

>> b = [10;-30;40];
>> x = A\b

x =
-1.0811
0.8108
-4.7490

>> AT = A'

AT =

-7 0 -4
3 4 3
0 7 -7

>> AI = inv(A)

AI =
-0.1892 0.0811 0.0811
-0.1081 0.1892 0.1892
0.0618 0.0347 -0.1081

7.4
⎡ 23 − 8⎤
[ X ] × [Y ] = ⎢ 55 56 ⎥
⎢⎣− 17 24 ⎥⎦

⎡ 12 8⎤
[ X ] × [ Z ] = ⎢− 30 52⎥
⎢⎣− 23 2 ⎥⎦

[Y ] × [ Z ] = ⎡ 4 8⎤
⎢⎣− 47 34⎥⎦

[ Z ] × [Y ] = ⎡ 6 16 ⎤
⎢⎣− 20 32⎥⎦

7.5 Terms can be combined to yield

71
 2kx1 − kx 2 = m1 g

− kx1 + 2kx 2 − kx3 = m 2 g

− kx 2 + kx3 = m3 g

Substituting the parameter values

⎡ 20 − 10 0 ⎤ ⎧⎪ x1 ⎫⎪ ⎧⎪ 19.62 ⎫⎪
⎢ − 10 20 − 10⎥ ⎨ x 2 ⎬ = ⎨ 29.43 ⎬
⎢⎣ 0 − 10 10 ⎥⎦ ⎪⎩ x 3 ⎪⎭ ⎪⎩24.525⎪⎭

A MATLAB session can be used to obtain the solution for the displacements

>> K=[20 -10 0;-10 20 -10;0 -10 10];

>> m=[2;3;2.5];
>> mg=m*9.81;
>> x=K\mg

x =
7.3575
12.7530
15.2055

7.6 The mass balances can be written as

(Q15 + Q12 )c1 − Q31c3 = Q01c 01

− Q12 c1 + (Q23 + Q24 + Q25 )c 2 =0

− Q23 c 2 + (Q31 + Q34 )c 3 = Q03 c 03

− Q24 c 2 − Q34 c3 + Q44 c 4 − Q54 c 5 = 0

− Q15 c1 − Q25 c 2 + (Q54 + Q55 )c 5 = 0

The parameters can be substituted and the result written in matrix form as

⎡6 0 − 1 0 0 ⎤ ⎧ c1 ⎫ ⎧ 50 ⎫
⎢− 3 3 0 0 0 ⎥ ⎪⎪c 2 ⎪⎪ ⎪⎪ 0 ⎪⎪
⎢ 0 − 1 9 0 0 ⎥ ⎨c3 ⎬ = ⎨160⎬
⎢ 0 − 1 − 8 11 − 2⎥ ⎪c 4 ⎪ ⎪ 0 ⎪
⎢⎣− 3 − 1 0 0 4 ⎥⎦ ⎪c ⎪ ⎪⎩ 0 ⎪⎭
⎩ 5⎭

MATLAB can then be used to solve for the concentrations

>> Q = [6 0 -1 0 0;
-3 3 0 0 0;
0 -1 9 0 0;
0 -1 -8 11 -2;
-3 -1 0 0 4];
>> Qc = [50;0;160;0;0];

72
 >> c = Q\Qc

c =
11.5094
11.5094
19.0566
16.9983
11.5094

7.7 The problem can be written in matrix form as

⎡ 0.866 0 − 0 .5 0 0 0 ⎤ ⎧ F1 ⎫ ⎧ 0 ⎫
⎢ 0 .5 0 0.866 0 0 0 ⎥ ⎪ F2 ⎪ ⎪− 1000⎪
⎢− 0.866 − 1 0 −1 0 0 ⎥ ⎪⎪ F3 ⎪⎪ = ⎪ 0 ⎪
⎢ − 0 .5 0 0 0 − 1 0 ⎥ ⎨H 2 ⎬ ⎨ 0 ⎬
⎢ 0 1 0 .5 0 0 0 ⎥ ⎪ V2 ⎪ ⎪ 0 ⎪
⎢⎣ 0 0 − 0.866 0 0 − 1⎥⎦ ⎪⎪ V ⎪⎪ ⎪⎩ 0 ⎪⎭
⎩ 3⎭

MATLAB can then be used to solve for the forces and reactions,

>> A = [0.866 0 -0.5 0 0 0;

0.5 0 0.866 0 0 0;
-0.866 -1 0 -1 0 0;
-0.5 0 0 0 -1 0;
0 1 0.5 0 0 0;
0 0 -0.866 0 0 -1]
>> b = [0 -1000 0 0 0 0]';
>> F = A\b

F =
-500.0220
433.0191
-866.0381
-0.0000
250.0110
749.9890

Therefore,

F1 = –500 F2 = 433 F3 = –866

H2 = 0 V2 = 250 V3 = 750

7.8 The problem can be written in matrix form as

⎡1 1 1 0 0 0 ⎤ ⎧i12 ⎫ ⎧ 0 ⎫
⎢0 − 1 0 1 − 1 0 ⎥ ⎪i52 ⎪ ⎪ 0 ⎪
⎢0 0 −1 0 0 1 ⎥ ⎪⎪i32 ⎪⎪ = ⎪ 0 ⎪
⎢0 0 0 0 1 − 1⎥ ⎨i65 ⎬ ⎨ 0 ⎬
⎢0 10 − 10 0 − 15 − 5⎥ ⎪i ⎪ ⎪ 0 ⎪
⎢⎣5 − 10 0 − 20 0 0 ⎥⎦ ⎪⎪i54 ⎪⎪ ⎪⎩200⎪⎭
⎩ 43 ⎭

MATLAB can then be used to solve for the currents,

73
 >> A = [1 1 1 0 0 0 ;
0 -1 0 1 -1 0;
0 0 -1 0 0 1;
0 0 0 0 1 -1;
0 10 -10 0 -15 -5;
5 -10 0 -20 0 0];
>> b = [0 0 0 0 0 200]';
>> i = A\b

i =
6.1538
-4.6154
-1.5385
-6.1538
-1.5385
-1.5385

7.9

>> k1 = 10;k2 = 40;k3 = 40;k4 = 10;

>> m1 = 1;m2 = 1;m3 = 1;
>> km = [(1/m1)*(k2+k1), -(k2/m1),0;
-(k2/m2), (1/m2)*(k2+k3), -(k3/m2);
0, -(k3/m3),(1/m3)*(k3+k4)];
>> x = [0.05;0.04;0.03];
>> kmx = km*x

kmx =
0.9000
0.0000
-0.1000

Therefore, &x&1 = −0.9, &x&2 = 0 , and &x&3 = 0.1 m/s2.

74
CHAPTER 8
8.1 The flop counts for the tridiagonal algorithm in Fig. 8.6 can be summarized as

Mult/Div Add/Subtr Total

Forward elimination 3(n – 1) 2(n – 1) 5(n – 1)
Back substitution 2n – 1 n–1 3n – 2
Total 5n – 4 3n – 3 8n – 7

Thus, as n increases, the effort is much, much less than for a full matrix solved with Gauss
elimination which is proportional to n3.

8.2 The equations can be expressed in a format that is compatible with graphing x2 versus x1:

x 2 = 0.5 x1 + 3

1 34
x2 = − x1 +
6 6

which can be plotted as

10
9
8
7
6
5
4
3
2
1
0
0 2 4 6 8 10 12

Thus, the solution is x1 = 4, x2 = 5. The solution can be checked by substituting it back into
the equations to give

4( 4) − 8(5) = 16 − 40 = −24

4 + 6(5) = 4 + 30 = 34

8.3 (a) The equations can be expressed in a format that is compatible with graphing x2 versus x1:

x 2 = 0.11x1 + 12

x 2 = 0.114943x1 + 10

75
 which can be plotted as

140

120

100

80

60

40

20

0
0 200 400 600 800 1000

Thus, the solution is approximately x1 = 400, x2 = 60. The solution can be checked by
substituting it back into the equations to give

− 1.1(400) + 10(60) = 160 ≈ 120

− 2(400) + 17.4(60) = 244 ≈ 174

Therefore, the graphical solution is not very good.

(b) Because the lines have very similar slopes, you would expect that the system would be
ill-conditioned

(c) The determinant can be computed as

− 1.1 10 = −1.1(17.2) − 10(−2) = −19.14 + 20 = 0.86

− 2 17.4

This result is relatively low suggesting that the solution is ill-conditioned.

8.4 (a) The determinant can be evaluated as

D = 0⎡ 2 − 1⎤ − (−3) ⎡1 − 1⎤ + 7 ⎡1 2 ⎤
⎢⎣− 2 0 ⎥⎦ ⎢⎣5 0 ⎥⎦ ⎢⎣5 − 2⎥⎦

D = 0(−2) + 3(5) + 7(−12) = −69

(b) Cramer’s rule

76
 2 −3 7
3 2 −1
2 −2 0 − 68
x1 = = = 0.9855
− 69 − 69

0 2 7
1 3 −1
5 2 0 − 101
x2 = = = 1.4638
− 69 − 69

0 −3 2
1 2 3
5 −2 2 − 63
x3 = = = 0.9130
− 69 − 69

(c) Pivoting is necessary, so switch the first and third rows,

5 x1 − 2 x 2 =2
x1 + 2 x 2 − x3 = 3
− 3x 2 + 7 x3 = 2

Multiply pivot row 1 by 1/5 and subtract the result from the second row to eliminate the a21
term.

5 x1 − 2 x 2 =2
2.4 x 2 − x 3 = 2.6
− 3x 2 + 7 x3 = 2

Pivoting is necessary so switch the second and third row,

5 x1 − 2 x 2 =2
− 3x 2 + 7 x3 = 2
2.4 x 2 − x 3 = 2.6

Multiply pivot row 2 by 2.4/(–3) and subtract the result from the third row to eliminate the
a32 term.

5 x1 − 2 x 2 =2
− 3 x 2 + 7 x3 = 2
4.6 x3 = 4.2

The solution can then be obtained by back substitution

4 .2
x3 = = 0.913043
4 .6

2 − 7(0.913043)
x2 = = 1.463768
−3

77
 2 + 2(1.463768)
x1 = = 0.985507
5

(d)

− 3(1.463768) + 7(0.913043) = 2
0.985507 + 2(1.463768) − (0.913043) = 3

5(0.985507) − 2(1.463768) = 2

8.5 Prob. 8.3:

>> A=[-1.1 10;-2 17.4];

>> det(A)

ans =
0.8600

Prob. 8.4:

>> A=[0 -3 7;1 2 -1;5 -2 0];

>> det(A)

ans =
-69

8.6 (a) The equations can be expressed in a format that is compatible with graphing x2 versus x1:

x 2 = 0.5 x1 + 9.5

x 2 = 0.51x1 + 9.4

The resulting plot indicates that the intersection of the lines is difficult to detect:

22

20

18

16

14

12

10
5 10 15 20

Only when the plot is zoomed is it at all possible to discern that solution seems to lie at
about x1 = 14.5 and x2 = 10.

78
 14.7
14.65
14.6
14.55

14.5
14.45

14.4
14.35
14.3
9.75 10 10.25

(b) The determinant can be computed as

0.5 − 1 = 0.5(−2) − (−1)(1.02) = 0.02

1.02 − 2

which is close to zero.

(c) Because the lines have very similar slopes and the determinant is so small, you would
expect that the system would be ill-conditioned

(d) Multiply the first equation by 1.02/0.5 and subtract the result from the second equation
to eliminate the x1 term from the second equation,

0.5 x1 − x 2 = −9.5

0.04 x 2 = 0.58

The second equation can be solved for

0.58
x2 = = 14.5
0.04

This result can be substituted into the first equation which can be solved for

− 9.5 + 14.5
x1 = = 10
0 .5

(e) Multiply the first equation by 1.02/0.52 and subtract the result from the second equation
to eliminate the x1 term from the second equation,

0.52 x1 − x 2 = −9.5

− 0.03846 x 2 = −0.16538

79
 The second equation can be solved for

− 0.16538
x2 = = 4.3
− 0.03846

This result can be substituted into the first equation which can be solved for

− 9.5 + 4.3
x1 = = −10
0.52

Interpretation: The fact that a slight change in one of the coefficients results in a radically
different solution illustrates that this system is very ill-conditioned.

8.7 (a) Multiply the first equation by –3/10 and subtract the result from the second equation to
eliminate the x1 term from the second equation. Then, multiply the first equation by 1/10
and subtract the result from the third equation to eliminate the x1 term from the third
equation.

10 x1 + 2 x 2 − x 3 = 27

− 5.4 x 2 + 1.7 x3 = −53.4

0.8 x 2 + 5.1x3 = −24.2

Multiply the second equation by 0.8/(–5.4) and subtract the result from the third equation to
eliminate the x2 term from the third equation,

10 x1 + 2 x 2 − x 3 = 27

− 5.4 x 2 + 1.7 x3 = −53.4

5.351852 x3 = −32.11111

Back substitution can then be used to determine the unknowns

− 32.11111
x3 = = −6
5.351852
(−53.4 − 1.7(−6))
x2 = =8
− 5 .4
( 27 − 6 − 2(8))
x1 = = 0.5
10

(b) Check:

80
 10(0.5) + 2(8) − (−6) = 27

− 3(0.5) − 6(8) + 2(−6) = −61.5

0.5 + 8 + 5(−6) = −21.5

8.8 (a) Pivoting is necessary, so switch the first and third rows,

− 8 x1 + x 2 − 2 x3 = −20

− 3 x1 − x 2 + 7 x3 = −34

2 x1 − 6 x 2 − x3 = −38

Multiply the first equation by –3/(–8) and subtract the result from the second equation to
eliminate the a21 term from the second equation. Then, multiply the first equation by 2/(–8)
and subtract the result from the third equation to eliminate the a31 term from the third
equation.

− 8 x1 + x2 − 2 x3 = −20

− 1.375 x 2 + 7.75 x3 = −26.5

− 5.75 x 2 − 1.5 x 3 = −43

Pivoting is necessary so switch the second and third row,

− 8 x1 + x2 − 2 x3 = −20

− 5.75 x 2 − 1.5 x 3 = −43

− 1.375 x 2 + 7.75 x3 = −26.5

Multiply pivot row 2 by –1.375/(–5.75) and subtract the result from the third row to
eliminate the a32 term.

− 8 x1 + x2 − 2 x 3 = −20

− 5.75 x 2 − 1.5 x3 = −43

8.108696 x3 = −16.21739

The solution can then be obtained by back substitution

− 16.21739
x3 = = −2
8.108696

− 43 + 1.5(−2)
x2 = =8
− 5.75

81
 − 20 + 2(−2) − 1(8)
x1 = =4
−8

(b) Check:

2(4) − 6(8) − (−2) = −38

− 3(4) − (8) + 7(−2) = −34

− 8(4) + (8) − 2(−2) = −20

8.9 Multiply the first equation by –0.4/0.8 and subtract the result from the second equation to
eliminate the x1 term from the second equation.

⎡0.8 − 0.4 ⎤ ⎧ x1 ⎫ ⎧ 41 ⎫
⎢ ⎪ ⎪ ⎪ ⎪
0.6 − 0.4⎥ ⎨ x 2 ⎬ = ⎨45.5⎬
⎢ − 0.4 0.8 ⎥⎦ ⎪⎩ x3 ⎪⎭ ⎪⎩ 105 ⎪⎭
⎣

Multiply pivot row 2 by –0.4/0.6 and subtract the result from the third row to eliminate the
x2 term.

⎡0.8 − 0.4 ⎤ ⎧ x1 ⎫ ⎧ 41 ⎫
⎢ ⎪ ⎪ ⎪ ⎪
0.6 − 0.4 ⎥ ⎨ x 2 ⎬ = ⎨ 45.5 ⎬
⎢ 0.533333⎥⎦ ⎪⎩ x3 ⎪⎭ ⎪⎩135.3333⎪⎭
⎣

The solution can then be obtained by back substitution

135.3333
x3 = = 253.75
0.533333

45.5 − ( −0.4)253.75
x2 = = 245
0 .6

41 − (−0.4)245
x1 = = 173.75
0 .8

(b) Check:

0.8(173.75) − 0.4(245) = 41

− 0.4(173.75) + 0.8(245) − 0.4(253.75) = 25

− 0.4(245) + 0.8(253.75) = 105

8.10 The mass balances can be written as

82
 Q21c 2 + 400 = Q12 c1 + Q13 c1

Q12 c1 = Q21c 2 + Q23 c 2

Q13 c1 + Q23 c 2 = Q33 c 3 + 200

or collecting terms

(Q12 + Q13 )c1 − Q21c 2 = 400

− Q12 c1 + (Q21 + Q23 )c 2 =0

− Q13 c1 − Q23 c 2 + Q33 c3 = 200

Substituting the values for the flows and expressing in matrix form

⎡ 120 − 20 0 ⎤ ⎧⎪ c1 ⎫⎪ ⎧⎪400⎫⎪
⎢ − 80 80 0 ⎥ ⎨c 2 ⎬ = ⎨ 0 ⎬
⎢⎣− 40 − 60 120⎥⎦ ⎪⎩c 3 ⎪⎭ ⎪⎩200⎪⎭

A solution can be obtained with MATLAB as

>> A = [120 -20 0;-80 80 0;-40 -60 120];

>> b = [400 0 200]';
>> c = a\b

c =
4.0000
4.0000
5.0000

8.11 Equations for the amount of sand, fine gravel and coarse gravel can be written as

0.32 x1 + 0.25 x 2 + 0.35 x 3 = 6000

0.30 x1 + 0.40 x 2 + 0.15 x3 = 5000

0.38 x1 + 0.35 x 2 + 0.50 x 3 = 8000

where xi = the amount of gravel taken from pit i. MATLAB can be used to solve this
system of equations for

>> A=[0.32 0.25 0.35;0.3 0.4 0.15;0.38 0.35 0.5];

>> b=[6000;5000;8000];
>> x=A\b

x =
1.0e+003 *
7.0000
4.4000
7.6000

Therefore, we take 7000, 4400 and 7600 m3 from pits 1, 2 and 3 respectively.

83
8.12 Substituting the parameter values the heat-balance equations can be written for the four
nodes as

− 40 + 2.2T1 − T2 = 4
− T1 + 2.2T2 − T3 = 4
− T2 + 2.2T3 − T4 = 4
− T3 + 2.2T4 − 200 = 4

Collecting terms and expressing in matrix form

⎡2.2 − 1 0 0 ⎤ ⎧T1 ⎫ ⎧ 44 ⎫
⎢ − 1 2 . 2 − 1 0 ⎥ ⎪T2 ⎪ ⎪ 4 ⎪
⎢ 0 − 1 2.2 − 1⎥ ⎨T3 ⎬ = ⎨ 4 ⎬
⎢⎣ 0 0 − 1 2.2⎥⎦ ⎪T ⎪ ⎪⎩204⎪⎭
⎩ 4⎭

The solution can be obtained with MATLAB as

>> A=[2.2 -1 0 0;-1 2.2 -1 0;0 -1 2.2 -1;0 0 -1 2.2]

>> b=[44 4 4 204]'
>> T=A\b

T =
50.7866
67.7306
94.2206
135.5548

84
CHAPTER 9
9.1 The flop counts for LU decomposition can be determined in a similar fashion as was done
for Gauss elimination. The major difference is that the elimination is only implemented for
the left-hand side coefficients. Thus, for every iteration of the inner loop, there are n
multiplications/divisions and n – 1 addition/subtractions. The computations can be
summarized as

Outer Loop Inner Loop Addition/Subtraction Multiplication/Division

k i flops flops
1 2, n (n – 1)(n – 1) (n – 1)n
2 3, n (n – 2)(n – 2) (n – 2)(n – 1)
. .
. .
. .
k k + 1, n (n – k)(n – k) (n – k)(n + 1 – k)
. .
. .
. .
n–1 n, n (1)(1) (1)(2)

Therefore, the total addition/subtraction flops for elimination can be computed as

∑ [n ]
n −1 n −1

∑k =1
(n − k )(n − k ) =
k =1
2
− 2nk + k 2

Applying some of the relationships from Eq. (8.14) yields

∑ [n ]
n −1
n3 n2 n
2
− 2nk + k 2 = − +
k =1 3 2 6

A similar analysis for the multiplication/division flops yields

n −1
n3 n
∑k =1
(n − k )(n + 1 − k ) = −
3 3

[n ] [ ]
⎡1 ⎤ n
3
3
+ O ( n 2 ) − n 3 + O ( n) + ⎢ n 3 + O ( n 2 ) ⎥ = + O(n 2 )
⎣3 ⎦ 3

Summing these results gives

2n 3 n 2 n
− −
3 2 6

For forward substitution, the numbers of multiplications and subtractions are the same and
equal to

85
 n −1
(n − 1) n n 2 n
∑i =
i =1 2
= −
2 2

Back substitution is the same as for Gauss elimination: n2/2 – n/2 subtractions and n2/2 +
n/2 multiplications/divisions. The entire number of flops can be summarized as

Mult/Div Add/Subtr Total

Forward elimination n3
n n 3
n2
n 2n3
n2 n
− − + − −
3 3 3 2 6 3 2 6
Forward substitution n2 n n 2
n n2 − n
− −
2 2 2 2
Back substitution n2 n n2 n n2
+ −
2 2 2 2
Total n3 n n3 n2 5n 2n 3 3n 2 7 n
+ n2 − + − + −
3 3 3 2 6 3 2 6

The total number of flops is identical to that obtained with standard Gauss elimination.

9.2 Equation (9.6) is

[ L]{[U ]{x} − {d }} = [ A]{x} − {b} (9.6)

Matrix multiplication is distributive, so the left-hand side can be rewritten as

[ L][U ]{x} − [ L]{d } = [ A]{x} − {b}

Equating the terms that are multiplied by {x} yields,

[ L][U ]{x} = [ A]{x}

and, therefore, Eq. (9.7) follows

[ L][U ] = [ A] (9.7)

Equating the constant terms yields Eq. (9.8)

[ L]{d } = {b} (9.8)

9.3 The matrix to be evaluated is

⎡ 10 2 − 1⎤
⎢− 3 − 6 2 ⎥
⎢⎣ 1 1 5 ⎥⎦

Multiply the first row by f21 = –3/10 = –0.3 and subtract the result from the second row to
eliminate the a21 term. Then, multiply the first row by f31 = 1/10 = 0.1 and subtract the
result from the third row to eliminate the a31 term. The result is

86
 ⎡10 2 − 1⎤
⎢ 0 − 5.4 1.7 ⎥
⎢⎣ 0 0.8 5.1⎥⎦

Multiply the second row by f32 = 0.8/(–5.4) = –0.148148 and subtract the result from the
third row to eliminate the a32 term.

⎡10 2 −1 ⎤
⎢ 0 − 5 .4 1.7 ⎥
⎢⎣ 0 0 5.351852⎥⎦

Therefore, the LU decomposition is

⎡ 1 0 0⎤ ⎡10 2 −1 ⎤
[ L]{U ] = ⎢− 0.3 1 0 ⎥ ⎢ 0 − 5 .4 1.7 ⎥
⎢⎣ 0.1 − 0.148148 1⎥⎦ ⎢⎣ 0 0 5.351852⎥⎦

Multiplying [L] and [U] yields the original matrix as verified by the following MATLAB
session,

>> L = [1 0 0;-0.3 1 0;0.1 -0.148148 1];

>> U = [10 2 -1;0 -5.4 1.7;0 0 5.351852];
>> A = L*U

A =
10.0000 2.0000 -1.0000
-3.0000 -6.0000 2.0000
1.0000 1.0000 5.0000

9.4 The LU decomposition can be computed as

⎡ 1 0 0⎤ ⎡10 2 −1 ⎤
[ L]{U ] = ⎢− 0.3 1 0 ⎥ ⎢ 0 − 5 .4 1.7 ⎥
⎢⎣ 0.1 − 0.148148 1⎥⎦ ⎢⎣ 0 0 5.351852⎥⎦

Forward substitution:

⎡ 1 0 0⎤ ⎧⎪ 27 ⎫⎪
{d } = ⎢− 0.3 1 0⎥ ⎨− 61.5⎬
⎢⎣ 0.1 − 0.148148 1⎥⎦ ⎪⎩− 21.5⎪⎭

d1 = 27

d 2 = −61.5 + 0.3(27) = −53.4

d 3 = −21.5 − 0.1(27) − (−0.148148)(−53.4) = −32.11111

Back substitution:

87
 ⎡10 2 − 1 ⎤ ⎧⎪ x1 ⎫⎪ ⎧⎪ 27 ⎫⎪
{x} = ⎢ 0 − 5.4 1.7 ⎥ ⎨ x 2 ⎬ = ⎨ − 53.5 ⎬
⎢⎣ 0 0 5.351852⎥⎦ ⎪⎩ x3 ⎪⎭ ⎪⎩− 32.11111⎪⎭

− 32.11111
x3 = = −6
5.351852

− 53.4 − 1.7( −6)

x2 = =8
− 5 .4

27 − 2(8) − (−1)(−6)
x1 = = 0.5
10

For the alternative right-hand-side vector, forward substitution is implemented as

⎡ 1 0 0⎤ ⎧⎪ 12 ⎫⎪
{d } = ⎢− 0.3 1 0⎥ ⎨ 18 ⎬
⎣⎢ 0.1 − 0.148148 1⎥⎦ ⎪⎩− 6⎭⎪

d1 = 12

d 2 = 18 + 0.3(12) = 21.6

d 3 = −6 − 0.1(12) − (−0.148148)(18) = −4

Back substitution:

⎡10 2 − 1 ⎤ ⎧⎪ 12 ⎫⎪
{x} = ⎢ 0 − 5.4 1.7 ⎥ ⎨21.6⎬
⎣⎢ 0 0 5.351852⎦⎥ ⎪⎩ − 4 ⎪⎭

−4
x3 = = −0.747405
5.351852

21.6 − 1.7(−0.747405)
x2 = = −4.235294
− 5.4

12 − 2( −4.235294) − (−1)(−0.747405)
x1 = = 1.972318
10

9.5 The system can be written in matrix form as

⎡ 2 − 6 − 1⎤ ⎧⎪ − 38⎫⎪
[ A] = ⎢− 3 − 1 7 ⎥ {b} = ⎨− 34 ⎬
⎢⎣− 8 1 − 2⎥⎦ ⎪⎩− 20⎪⎭

Partial pivot:

88
 ⎡− 8 1 − 2 ⎤ ⎧⎪− 20⎫⎪
[ A] = ⎢− 3 − 1 7 ⎥ {b} = ⎨− 34 ⎬
⎢⎣ 2 − 6 − 1 ⎥⎦ ⎪⎩− 38⎪⎭

Forward eliminate

f21 = −3/(−8) = 0.375 f31 = 2/(−8) = −0.25

⎡− 8 1 −2 ⎤
[ A] = ⎢ 0 − 1.375 7.75 ⎥
⎢⎣ 0 − 5.75 − 1.5 ⎥⎦

Pivot again

⎡− 8 1 −2 ⎤ ⎧⎪− 20⎫⎪
[ A] = ⎢ 0 − 5.75 − 1.5 ⎥ {b} = ⎨− 38⎬
⎢⎣ 0 − 1.375 7.75 ⎥⎦ ⎪⎩− 34 ⎪⎭

f21 = −0.25 f31 = 0.375

Forward eliminate

f32 = −1.375/(−5.75) = 0.23913

⎡− 8 1 −2 ⎤
[ A] = ⎢ 0 − 5.75 − 1.5 ⎥
⎢⎣ 0 0 8.108696 ⎥⎦

Therefore, the LU decomposition is

⎡ 1 0 0⎤ ⎡ − 8 1 −2 ⎤
[ L]{U ] = ⎢− 0.25 1 0⎥ ⎢ 0 − 5.75 − 1 .5 ⎥
⎣⎢ 0.375 0.23913 1⎦⎥ ⎢⎣ 0 0 8.108696⎦⎥

Forward elimination

⎡ 1 0 0⎤ ⎧⎪− 20⎫⎪
{d } = ⎢− 0.25 1 0⎥ ⎨ − 38⎬
⎢⎣ 0.375 0.23913 1⎥⎦ ⎪⎩− 34 ⎪⎭

d1 = −20

d 2 = −38 − (−0.25)(−20) = −43

d 3 = −34 − 0.375(−20) − 0.23913(−43) = −16.21739

Back substitution:

89
 ⎡− 8 1 − 2 ⎤ ⎧⎪ x1 ⎫⎪ ⎧⎪ − 20 ⎫⎪
⎢ 0 − 5.75 − 1.5 ⎥ ⎨ x 2 ⎬ = ⎨ − 43 ⎬
⎣⎢ 0 0 8.108696⎦⎥ ⎪⎩ x3 ⎪⎭ ⎪⎩− 16.21739⎪⎭

− 16.21739
x3 = = −2
8.108696

− 43 − (−1.5)(−2)
x2 = =8
− 5.75

− 20 − 1(8) − (−2)(−2)
x1 = =4
−8

9.6 Here is an M-file to generate the LU decomposition without pivoting

function [L, U] = LUNaive(A)

% LUNaive(A):
% LU decomposition without pivoting.
% input:
% A = coefficient matrix
% output:
% L = lower triangular matrix
% U = upper triangular matrix

[m,n] = size(A);
if m~=n, error('Matrix A must be square'); end
L = eye(n);
U = A;
% forward elimination
for k = 1:n-1
for i = k+1:n
L(i,k) = U(i,k)/U(k,k);
U(i,k) = 0;
U(i,k+1:n) = U(i,k+1:n)-L(i,k)*U(k,k+1:n);
end
end

Test with Prob. 9.3

>> A = [10 2 -1;-3 -6 2;1 1 5];

>> [L,U] = LUnaive(A)

L =
1.0000 0 0
-0.3000 1.0000 0
0.1000 -0.1481 1.0000

U =
10.0000 2.0000 -1.0000
0 -5.4000 1.7000
0 0 5.3519

90
 Verification that [L][U] = [A].

>> L*U

ans =
10.0000 2.0000 -1.0000
-3.0000 -6.0000 2.0000
1.0000 1.0000 5.0000

Check using the lu function,

>> [L,U]=lu(A)

L =
1.0000 0 0
-0.3000 1.0000 0
0.1000 -0.1481 1.0000

U =
10.0000 2.0000 -1.0000
0 -5.4000 1.7000
0 0 5.3519

9.7 The result of Example 9.4 can be substituted into Eq. (9.14) to give

⎡ 2.44949 ⎤ ⎡2.44949 6.123724 22.45366⎤

[ A] = [U ]T [U ] = ⎢6.123724 4.1833 ⎥⎢ 4.1833 20.9165 ⎥
⎢22.45366 20.9165 6.110101⎥ ⎢ .110101⎥⎦
⎣ ⎦⎣ 6

The multiplication can be implemented as in

a11 = 2.44949 2 = 6.000001

a12 = 6.123724 × 2.44949 = 15

a13 = 22.45366 × 2.44949 = 55.00002

a 21 = 2.44949 × 6.123724 = 15

a 22 = 6.123724 2 + 4.1833 2 = 54.99999

a 22 = 22.45366 × 6.123724 2 + 20.9165 × 4.1833 = 225

a 31 = 2.44949 × 22.45366 = 55.00002

a 32 = 6.123724 × 22.45366 + 4.1833 × 20.9165 = 225

a 33 = 22.45366 2 + 20.9165 2 + 6.1101012 = 979.0002

91
9.8 (a) For the first row (i = 1), Eq. (9.15) is employed to compute

u11 = a11 = 8 = 2.828427

Then, Eq. (9.16) can be used to determine

a12 20
u12 = = = 7.071068
u11 2.828427

a13 15
u13 = = = 5.303301
u11 2.828427

For the second row (i = 2),

u 22 = a 22 − u12
2
= 80 − (7.071068) 2 = 5.477226

a 23 − u12 u13 50 − 7.071068(5.303301)

u 23 = = = 2.282177
u 22 5.477226

For the third row (i = 3),

u 33 = a 33 − u13
2
− u 23
2
= 60 − (5.303301) 2 − (2.282177) 2 = 5.163978

Thus, the Cholesky decomposition yields

⎡2.828427 7.071068 5.303301⎤

[U ] = ⎢ 5.477226 2.282177 ⎥
⎢ 5.163978 ⎥⎦
⎣

The validity of this decomposition can be verified by substituting it and its transpose into
Eq. (9.14) to see if their product yields the original matrix [A]. This is left for an exercise.

(b)
>> A = [8 20 15;20 80 50;15 50 60];
>> U = chol(A)

U =
2.8284 7.0711 5.3033
0 5.4772 2.2822
0 0 5.1640

(c) The solution can be obtained by hand or by MATLAB. Using MATLAB:

>> b = [50;250;100];
>> d=U'\b

d =

92
 17.6777
22.8218
-8.8756

>> x=U\d

x =
-2.7344
4.8828
-1.7187

9.9 Here is an M-file to generate the Cholesky decomposition without pivoting

function U = cholesky(A)
% cholesky(A):
% cholesky decomposition without pivoting.
% input:
% A = coefficient matrix
% output:
% U = upper triangular matrix
[m,n] = size(A);
if m~=n, error('Matrix A must be square'); end
for i = 1:n
s = 0;
for k = 1:i-1
s = s + U(k, i) ^ 2;
end
U(i, i) = sqrt(A(i, i) - s);
for j = i + 1:n
s = 0;
for k = 1:i-1
s = s + U(k, i) * U(k, j);
end
U(i, j) = (A(i, j) - s) / U(i, i);
end
end

Test with Prob. 9.8

>> A = [8 20 15;20 80 50;15 50 60];

>> cholesky(A)

ans =
2.8284 7.0711 5.3033
0 5.4772 2.2822
0 0 5.1640

Check with the chol function

>> U = chol(A)

U =
2.8284 7.0711 5.3033
0 5.4772 2.2822
0 0 5.1640

93
CHAPTER 10
10.1 First, compute the LU decomposition The matrix to be evaluated is

⎡ 10 2 − 1⎤
⎢− 3 − 6 2 ⎥
⎢⎣ 1 1 5 ⎥⎦

Multiply the first row by f21 = –3/10 = –0.3 and subtract the result from the second row to
eliminate the a21 term. Then, multiply the first row by f31 = 1/10 = 0.1 and subtract the
result from the third row to eliminate the a31 term. The result is

⎡10 2 − 1⎤
⎢ 0 − 5.4 1.7 ⎥
⎢⎣ 0 0.8 5.1⎥⎦

Multiply the second row by f32 = 0.8/(–5.4) = –0.148148 and subtract the result from the
third row to eliminate the a32 term.

⎡10 2 −1 ⎤
⎢ 0 − 5 .4 1.7 ⎥
⎢⎣ 0 0 5.351852⎥⎦

Therefore, the LU decomposition is

⎡ 1 0 0⎤ ⎡10 2 −1 ⎤
[ L]{U ] = ⎢− 0.3 1 0 ⎥ ⎢ 0 − 5 .4 1.7 ⎥
⎢⎣ 0.1 − 0.148148 1⎥⎦ ⎢⎣ 0 0 5.351852⎥⎦

The first column of the matrix inverse can be determined by performing the forward-
substitution solution procedure with a unit vector (with 1 in the first row) as the right-hand-
side vector. Thus, the lower-triangular system, can be set up as,

⎡ 1 0 0⎤ ⎧⎪ d1 ⎫⎪ ⎧⎪1⎫⎪
⎢− 0.3 1 0⎥ ⎨d 2 ⎬ = ⎨0⎬
⎢⎣ 0.1 − 0.148148 1⎥⎦ ⎪⎩ d 3 ⎪⎭ ⎩⎪0⎪⎭

and solved with forward substitution for {d}T = ⎣1 0.3 − 0.055556⎦ . This vector can then
be used as the right-hand side of the upper triangular system,

⎡10 2 − 1 ⎤ ⎧⎪ x1 ⎫⎪ ⎧⎪ 1 ⎫⎪
⎢0 − 5 .4 1.7 ⎥ ⎨ x 2 ⎬ = ⎨ 0 .3 ⎬
⎢⎣ 0 0 5.351852⎥⎦ ⎪⎩ x3 ⎪⎭ ⎪⎩− 0.055556⎪⎭

which can be solved by back substitution for the first column of the matrix inverse,

94
 ⎡ 0.110727 0 0⎤
[ A] −1 = ⎢− 0.058824 0 0⎥
⎢⎣ − 0.010381 0 0⎥⎦

To determine the second column, Eq. (9.8) is formulated as

⎡ 1 0 0⎤ ⎧⎪ d1 ⎫⎪ ⎧⎪0⎫⎪
⎢− 0.3 1 0⎥ ⎨d 2 ⎬ = ⎨1⎬
⎢⎣ 0.1 − 0.148148 1⎥⎦ ⎪⎩ d 3 ⎪⎭ ⎪⎩0⎪⎭

This can be solved with forward substitution for {d}T = ⎣0 1 0.148148⎦ , and the results
are used with [U] to determine {x} by back substitution to generate the second column of
the matrix inverse,

⎡ 0.110727 0.038062 0⎤
[ A] −1 = ⎢− 0.058824 − 0.176471 0⎥
⎢⎣ − 0.010381 0.027682 0⎥⎦

Finally, the same procedures can be implemented with {b}T = ⎣0 0 1⎦ to solve for {d}T =
⎣0 0 1⎦ , and the results are used with [U] to determine {x} by back substitution to
generate the third column of the matrix inverse,

⎡ 0.110727 0.038062 0.00692 ⎤

[ A] −1 = ⎢− 0.058824 − 0.176471 0.058824⎥
⎣⎢ − 0.010381 0.027682 0.186851⎦⎥

This result can be checked by multiplying it times the original matrix to give the identity
matrix. The following MATLAB session can be used to implement this check,

>> A = [10 2 -1;-3 -6 2;1 1 5];

>> AI = [0.110727 0.038062 0.00692;
-0.058824 -0.176471 0.058824;
-0.010381 0.027682 0.186851];
>> A*AI

ans =
1.0000 -0.0000 -0.0000
0.0000 1.0000 -0.0000
-0.0000 0.0000 1.0000

10.2 The system can be written in matrix form as

⎡− 8 1 − 2 ⎤ ⎧⎪ − 38⎫⎪
[ A] = ⎢ 2 − 6 − 1 ⎥ {b} = ⎨− 34 ⎬
⎢⎣− 3 − 1 7 ⎥⎦ ⎪⎩− 20⎪⎭

Forward eliminate

f21 = 2/(−8) = −0.25 f31 = −3/(−8) = 0.375

95
 ⎡− 8 1 −2 ⎤
[ A] = ⎢ 0 − 5.75 − 1.5 ⎥
⎢⎣ 0 − 1.375 7.75 ⎥⎦

Forward eliminate

f32 = −1.375/(−5.75) = 0.23913

⎡− 8 1 −2 ⎤
[ A] = ⎢ 0 − 5.75 − 1.5 ⎥
⎢⎣ 0 0 8.108696 ⎥⎦

Therefore, the LU decomposition is

⎡ 1 0 0⎤ ⎡ − 8 1 −2 ⎤
[ L]{U ] = ⎢− 0.25 1 0⎥ ⎢ 0 − 5.75 − 1.5 ⎥
⎢⎣ 0.375 0.23913 1⎥⎦ ⎢⎣ 0 0 8.108696⎥⎦

The first column of the matrix inverse can be determined by performing the forward-
substitution solution procedure with a unit vector (with 1 in the first row) as the right-hand-
side vector. Thus, the lower-triangular system, can be set up as,

⎡ 1 0 0⎤ ⎧⎪ d 1 ⎫⎪ ⎧⎪1⎫⎪
⎢− 0.25 1 0⎥ ⎨d 2 ⎬ = ⎨0⎬
⎢⎣ 0.375 0.23913 1⎥⎦ ⎪⎩ d 3 ⎪⎭ ⎪⎩0⎭⎪

and solved with forward substitution for {d}T = ⎣1 0.25 − 0.434783⎦ . This vector can then
be used as the right-hand side of the upper triangular system,

⎡− 8 1 − 2 ⎤ ⎧⎪ x1 ⎫⎪ ⎧⎪ 1 ⎫⎪
⎢ 0 − 5.75 − 1.5 ⎥ ⎨ x 2 ⎬ = ⎨ 0.25 ⎬
⎢⎣ 0 0 8.108696⎥⎦ ⎪⎩ x3 ⎪⎭ ⎪⎩− 0.434783⎪⎭

which can be solved by back substitution for the first column of the matrix inverse,

⎡ - 0.115282 0 0⎤
[ A] −1 = ⎢ − 0.029491 0 0⎥
⎢⎣− 0.053619 0 0⎥⎦

To determine the second column, Eq. (9.8) is formulated as

⎡ 1 0 0⎤ ⎧⎪ d 1 ⎫⎪ ⎧⎪0⎫⎪
⎢− 0.25 1 0⎥ ⎨d 2 ⎬ = ⎨1⎬
⎢⎣ 0.375 0.23913 1⎥⎦ ⎪⎩ d 3 ⎪⎭ ⎪⎩0⎪⎭

This can be solved with forward substitution for {d}T = ⎣0 1 − 0.23913⎦ , and the results
are used with [U] to determine {x} by back substitution to generate the second column of
the matrix inverse,

96
 ⎡ - 0.115282 − 0.013405 0⎤
[ A] −1 = ⎢ − 0.029491 − 0.16622 0⎥
⎢⎣− 0.053619 − 0.029491 0⎥⎦

Finally, the same procedures can be implemented with {b}T = ⎣0 0 1⎦ to solve for {d}T =
⎣0 0 1⎦ , and the results are used with [U] to determine {x} by back substitution to
generate the third column of the matrix inverse,

⎡ - 0.115282 − 0.013405 − 0.034853⎤

[ A] −1 = ⎢ − 0.029491 − 0.16622 − 0.032172⎥
⎣⎢− 0.053619 − 0.029491 0.123324 ⎦⎥

10.3 The following solution is generated with MATLAB.

(a)
>> A = [15 -3 -1;-3 18 -6;-4 -1 12];
>> format long
>> AI = inv(A)

AI =
0.07253886010363 0.01278065630397 0.01243523316062
0.02072538860104 0.06079447322971 0.03212435233161
0.02590673575130 0.00932642487047 0.09015544041451

(b)
>> b = [3800 1200 2350]';
>> format short
>> c = AI*b

c =
320.2073
227.2021
321.5026

(c) The impact of a load to reactor 3 on the concentration of reactor 1 is specified by the
−1
element a13 = 0.0124352. Therefore, the increase in the mass input to reactor 3 needed to
induce a 10 g/m3 rise in the concentration of reactor 1 can be computed as

10 g
∆b3 = = 804.1667
0.0124352 d

(d) The decrease in the concentration of the third reactor will be

g
∆c 3 = 0.0259067(500) + 0.009326(250) = 12.9534 + 2.3316 = 15.285
m3

10.4 The mass balances can be written and the result written in matrix form as

97
 ⎡6 0 − 1 0 0 ⎤ ⎧ c1 ⎫ ⎧Q01c 01 ⎫
⎢− 3 3 0 0 0 ⎥ ⎪⎪c 2 ⎪⎪ ⎪⎪ 0 ⎪⎪
⎢ 0 − 1 9 0 0 ⎥ ⎨c 3 ⎬ = ⎨Q03 c 03 ⎬
⎢ 0 − 1 − 8 11 − 2⎥ ⎪c 4 ⎪ ⎪ 0 ⎪
⎢⎣− 3 − 1 0 0 4 ⎥⎦ ⎪c ⎪ ⎪ 0 ⎪
⎩ 5⎭ ⎩ ⎭

MATLAB can then be used to determine the matrix inverse

>> Q = [6 0 -1 0 0;-3 3 0 0 0;0 -1 9 0 0;0 -1 -8 11 -2;-3 -1 0 0 4];

>> inv(Q)

ans =
0.1698 0.0063 0.0189 0 0
0.1698 0.3396 0.0189 0 0
0.0189 0.0377 0.1132 0 0
0.0600 0.0746 0.0875 0.0909 0.0455
0.1698 0.0896 0.0189 0 0.2500

The concentration in reactor 5 can be computed using the elements of the matrix inverse as
in,

−1 −1
c5 = a 51 Q01c 01 + a 53 Q03 c 03 = 0.1698(5)20 + 0.0189(8)50 = 16.981 + 7.547 = 24.528

10.5 The problem can be written in matrix form as

⎧ F1, h ⎫
⎡ 0.866 − 0.5 0 ⎤ ⎧ F1 ⎫ ⎪
F1,v ⎪
0 0 0
⎢ 0 .5 0 ⎥ F2⎪ ⎪
0 0.866 0 0 ⎪ ⎪
⎢− 0.866 − 1 0 −1 0 0 ⎥ ⎪⎪ F3 ⎪⎪ = ⎪ F2, h ⎪
⎢ − 0 .5 0 0 0 − 1 0 ⎥ ⎨ H 2 ⎬ ⎨ F2,v ⎬
⎢ 0 1 0.5 0 0 0 ⎥⎪ V ⎪ ⎪F ⎪
⎢⎣ 0 0 − 0.866 0 0 − 1⎥⎦ ⎪⎪ V2 ⎪⎪ ⎪ 3,h ⎪
⎩ 3 ⎭ ⎪⎩ F3,v ⎪⎭

MATLAB can then be used to solve for the matrix inverse,

>> A = [0.866 0 -0.5 0 0 0;

0.5 0 0.866 0 0 0;
-0.866 -1 0 -1 0 0;
-0.5 0 0 0 -1 0;
0 1 0.5 0 0 0;
0 0 -0.866 0 0 -1];
>> AI = inv(A)

AI =
0.8660 0.5000 0 0 0 0
0.2500 -0.4330 0 0 1.0000 0
-0.5000 0.8660 0 0 0 0
-1.0000 0.0000 -1.0000 0 -1.0000 0
-0.4330 -0.2500 0 -1.0000 0 0
0.4330 -0.7500 0 0 0 -1.0000

98
 The forces in the members resulting from the two forces can be computed using the
elements of the matrix inverse as in,

−1 −1
F1 = a12 F1,v + a15 F3,h = 0.5(−2000) + 0(−500) = −1000 + 0 = −1000

−1 −1
F2 = a 22 F1,v + a 25 F3,h = −0.433(−2000) + 1(−500) = 866 − 500 = 366

−1 −1
F3 = a 32 F1,v + a 35 F3, h = 0.866(−2000) + 0(−500) = −1732 + 0 = −1732

10.6 The matrix can be scaled by dividing each row by the element with the largest absolute
value

>> A = [8/(-10) 2/(-10) 1;1 1/(-9) 3/(-9);1 -1/15 6/15]

A =

-0.8000 -0.2000 1.0000

1.0000 -0.1111 -0.3333
1.0000 -0.0667 0.4000

MATLAB can then be used to determine each of the norms,

>> norm(A,'fro')

ans =
1.9920

>> norm(A,1)

ans =
2.8000

>> norm(A,inf)

ans =
2

10.7 Prob. 10.2:

>> A = [-8 1 -2;2 -6 -1;-3 -1 7];

>> norm(A,'fro')

ans =
13

>> norm(A,inf)

ans =
11

Prob. 10.3:

>> A = [15 -3 -1;-3 18 -6;-4 -1 12]

99
 >> norm(A,'fro')

ans =
27.6586

>> norm(A,inf)

ans =
27

10.8 (a) Spectral norm

>> A = [1 4 9 16;4 9 16 25;9 16 25 36;16 25 36 49];

>> cond(A)

ans =
8.8963e+016

(b) Row-sum norm

>> cond(A,inf)

Warning: Matrix is close to singular or badly scaled.

Results may be inaccurate. RCOND = 3.037487e-019.
(Type "warning off MATLAB:nearlySingularMatrix" to suppress this
warning.)
> In cond at 45

ans =
3.2922e+018

10.9 (a) The matrix to be evaluated is

⎡16 4 1⎤
⎢ 4 2 1⎥
⎢⎣49 7 1⎥⎦

The row-sum norm of this matrix is 49 + 7 + 1 = 57. The inverse is

⎡− 0.1667 0.1 0.0667 ⎤

⎢ 1.5 − 1 .1 − 0 .4 ⎥
⎢⎣ − 2.3333 2.8 0.5333⎥⎦

The row-sum norm of the inverse is ⎪-2.3333⎪ + 2.8 + 0.5333 = 5.6667. Therefore, the
condition number is

Cond[A] = 57(5.6667) = 323

This can be verified with MATLAB,

>> A = [16 4 1;4 2 1;49 7 1];

>> cond(A,inf)

100
 ans =
323.0000

(b) Spectral norm:

>> A = [16 4 1;4 2 1;49 7 1];

>> cond(A)

ans =
216.1294

Frobenius norm:

>> cond(A,'fro')

ans =
217.4843

10.10 The spectral condition number can be evaluated as

>> A = hilb(10);
>> N = cond(A)

N =
1.6025e+013

The digits of precision that could be lost due to ill-conditioning can be calculated as

>> c = log10(N)

c =
13.2048

Thus, about 13 digits could be suspect. A right-hand side vector can be developed
corresponding to a solution of ones:

>> b=[sum(A(1,:)); sum(A(2,:)); sum(A(3,:)); sum(A(4,:)); sum(A(5,:));

sum(A(6,:)); sum(A(7,:)); sum(A(8,:)); sum(A(9,:)); sum(A(10,:))]

b =
2.9290
2.0199
1.6032
1.3468
1.1682
1.0349
0.9307
0.8467
0.7773
0.7188

The solution can then be generated by left division

>> x = A\b

101
 x =
1.0000
1.0000
1.0000
1.0000
0.9999
1.0003
0.9995
1.0005
0.9997
1.0001

The maximum and mean errors can be computed as

>> e=max(abs(x-1))

e =
5.3822e-004

>> e=mean(abs(x-1))

e =
1.8662e-004

Thus, some of the results are accurate to only about 3 to 4 significant digits. Because
MATLAB represents numbers to 15 significant digits, this means that about 11 to 12 digits
are suspect.

10.11 First, the Vandermonde matrix can be set up

>> x1 = 4;x2=2;x3=7;x4=10;x5=3;x6=5;
>> A = [x1^5 x1^4 x1^3 x1^2 x1 1;x2^5 x2^4 x2^3 x2^2 x2 1;x3^5 x3^4
x3^3 x3^2 x3 1;x4^5 x4^4 x4^3 x4^2 x4 1;x5^5 x5^4 x5^3 x5^2 x5 1;x6^5
x6^4 x6^3 x6^2 x6 1]

A =
1024 256 64 16 4 1
32 16 8 4 2 1
16807 2401 343 49 7 1
100000 10000 1000 100 10 1
243 81 27 9 3 1
3125 625 125 25 5 1

The spectral condition number can be evaluated as

>> N = cond(A)

N =
1.4492e+007

The digits of precision that could be lost due to ill-conditioning can be calculated as

>> c = log10(N)

c =
7.1611

102
Thus, about 7 digits might be suspect. A right-hand side vector can be developed
corresponding to a solution of ones:

>> b=[sum(A(1,:));sum(A(2,:));sum(A(3,:));sum(A(4,:));sum(A(5,:));
sum(A(6,:))]

b =
1365
63
19608
111111
364
3906

The solution can then be generated by left division

>> format long

>> x=A\b

x =
1.00000000000000
0.99999999999991
1.00000000000075
0.99999999999703
1.00000000000542
0.99999999999630

The maximum and mean errors can be computed as

>> e = max(abs(x-1))

e =
5.420774940034789e-012

>> e = mean(abs(x-1))

e =
2.154110223528960e-012

Some of the results are accurate to about 12 significant digits. Because MATLAB
represents numbers to about 15 significant digits, this means that about 3 digits are suspect.
Thus, for this case, the condition number tends to exaggerate the impact of ill-conditioning.

103
CHAPTER 11
11.1 (a) The first iteration can be implemented as

41 + 0.4 x 2 41 + 0.4(0)
x1 = = = 51.25
0 .8 0 .8

25 + 0.4 x1 + 0.4 x3 25 + 0.4(51.25) + 0.4(0)

x2 = = = 56.875
0.8 0.8

105 + 0.4 x 2 105 + 0.4(56.875)

x3 = = = 159.6875
0 .8 0.8

Second iteration:

41 + 0.4(56.875)
x1 = = 79.6875
0.8

25 + 0.4(79.6875) + 0.4(159.6875)
x2 = = 150.9375
0.8

105 + 0.4(150.9375)
x3 = = 206.7188
0.8

The error estimates can be computed as

79.6875 − 51.25
ε a ,1 = × 100% = 35.69%
79.6875

150.9375 − 56.875
ε a,2 = × 100% = 62.32%
150.9375

206.7188 − 159.6875
ε a ,3 = × 100% = 22.75%
206.7188

The remainder of the calculation proceeds until all the errors fall below the stopping
criterion of 5%. The entire computation can be summarized as

iteration unknown value εa maximum εa

1 x1 51.25 100.00%
x2 56.875 100.00%
x3 159.6875 100.00% 100.00%
2 x1 79.6875 35.69%
x2 150.9375 62.32%
x3 206.7188 22.75% 62.32%
3 x1 126.7188 37.11%
x2 197.9688 23.76%

104
 x3 230.2344 10.21% 37.11%
4 x1 150.2344 15.65%
x2 221.4844 10.62%
x3 241.9922 4.86% 15.65%
5 x1 161.9922 7.26%
x2 233.2422 5.04%
x3 247.8711 2.37% 7.26%
6 x1 167.8711 3.50%
x2 239.1211 2.46%
x3 250.8105 1.17% 3.50%

Thus, after 6 iterations, the maximum error is 3.5% and we arrive at the result: x1 =
167.8711, x2 = 239.1211 and x3 = 250.8105.

(b) The same computation can be developed with relaxation where λ = 1.2.

First iteration:

41 + 0.4 x 2 41 + 0.4(0)
x1 = = = 51.25
0 .8 0 .8

Relaxation yields: x1 = 1.2(51.25) − 0.2(0) = 61.5

25 + 0.4 x1 + 0.4 x 3 25 + 0.4(61.5) + 0.4(0)

x2 = = = 62
0 .8 0.8

Relaxation yields: x 2 = 1.2(62) − 0.2(0) = 74.4

105 + 0.4 x 2 105 + 0.4(62)

x3 = = = 168.45
0 .8 0 .8

Relaxation yields: x 3 = 1.2(168.45) − 0.2(0) = 202.14

Second iteration:

41 + 0.4(62)
x1 = = 88.45
0.8

Relaxation yields: x1 = 1.2(88.45) − 0.2(61.5) = 93.84

25 + 0.4(93.84) + 0.4(202.14)
x2 = = 179.24
0 .8

Relaxation yields: x 2 = 1.2(179.24) − 0.2(74.4) = 200.208

105 + 0.4(200.208)
x3 = = 231.354
0 .8

105
 Relaxation yields: x 3 = 1.2( 231.354) − 0.2(202.14) = 237.1968

The error estimates can be computed as

93.84 − 61.5
ε a ,1 = × 100% = 34.46%
93.84

200.208 − 74.4
ε a,2 = × 100% = 62.84%
200.208

237.1968 − 202.14
ε a ,3 = × 100% = 14.78%
237.1968

The remainder of the calculation proceeds until all the errors fall below the stopping
criterion of 5%. The entire computation can be summarized as

iteration unknown value relaxation εa maximum εa

1 x1 51.25 61.5 100.00%
x2 62 74.4 100.00%
x3 168.45 202.14 100.00% 100.000%
2 x1 88.45 93.84 34.46%
x2 179.24 200.208 62.84%
x3 231.354 237.1968 14.78% 62.839%
3 x1 151.354 162.8568 42.38%
x2 231.2768 237.49056 15.70%
x3 249.99528 252.55498 6.08% 42.379%
4 x1 169.99528 171.42298 5.00%
x2 243.23898 244.38866 2.82%
x3 253.44433 253.6222 0.42% 4.997%

Thus, relaxation speeds up convergence. After 6 iterations, the maximum error is 4.997%
and we arrive at the result: x1 = 171.423, x2 = 244.389 and x3 = 253.622.

11.2 The first iteration can be implemented as

27 − 2 x 2 + x3 27 − 2(0) + 0
x1 = = = 2.7
10 10

− 61.5 + 3x1 − 2 x 3 − 61.5 + 3(2.7) − 2(0)

x2 = = = 8 .9
−6 −6

− 21.5 − x1 − x 2 − 21.5 − (2.7) − 8.9

x3 = = = −6.62
5 5

Second iteration:

106
 27 − 2(8.9) − 6.62
x1 = = 0.258
10

− 61.5 + 3(0.258) − 2(−6.62)

x2 = = 7.914333
−6

− 21.5 − (0.258) − 7.914333

x3 = = −5.934467
5

The error estimates can be computed as

0.258 − 2.7
ε a ,1 = × 100% = 947%
0.258

7.914333 − 8.9
ε a,2 = × 100% = 12.45%
7.914333

− 5.934467 − (−6.62)
ε a ,3 = × 100% = 11.55%
− 5.934467

The remainder of the calculation proceeds until all the errors fall below the stopping
criterion of 5%. The entire computation can be summarized as

iteration unknown value εa maximum εa

1 x1 2.7 100.00%
x2 8.9 100.00%
x3 -6.62 100.00% 100%
2 x1 0.258 946.51%
x2 7.914333 12.45%
x3 -5.93447 11.55% 946%
3 x1 0.523687 50.73%
x2 8.010001 1.19%
x3 -6.00674 1.20% 50.73%
4 x1 0.497326 5.30%
x2 7.999091 0.14%
x3 -5.99928 0.12% 5.30%
5 x1 0.500253 0.59%
x2 8.000112 0.01%
x3 -6.00007 0.01% 0.59%

Thus, after 5 iterations, the maximum error is 0.59% and we arrive at the result: x1 =
0.500253, x2 = 8.000112 and x3 = −6.00007.

11.3 The first iteration can be implemented as

27 − 2 x 2 + x3 27 − 2(0) + 0
x1 = = = 2.7
10 10

107
 − 61.5 + 3x1 − 2 x 3 − 61.5 + 3(0) − 2(0)
x2 = = = 10.25
−6 −6

− 21.5 − x1 − x 2 − 21.5 − 0 − 0
x3 = = = −4.3
5 5

Second iteration:

27 − 2(10.25) − 4.3
x1 = = 0.22
10

− 61.5 + 3(2.7) − 2(−4.3)

x2 = = 7.466667
−6

− 21.5 − ( 2.7) − 10.25

x3 = = −6.89
5

The error estimates can be computed as

0.22 − 2.7
ε a ,1 = × 100% = 1127%
0.258

7.466667 − 10.25
ε a,2 = × 100% = 37.28%
7.466667

− 6.89 − ( −4.3)
ε a ,3 = × 100% = 37.59%
− 6.89

The remainder of the calculation proceeds until all the errors fall below the stopping
criterion of 5%. The entire computation can be summarized as

iteration unknown value εa maximum εa

1 x1 2.7 100.00%
x2 10.25 100.00%
x3 -4.3 100.00% 100.00%
2 x1 0.22 1127.27%
x2 7.466667 37.28%
x3 -6.89 37.59% 1127.27%
3 x1 0.517667 57.50%
x2 7.843333 4.80%
x3 -5.83733 18.03% 57.50%
4 x1 0.5476 5.47%
x2 8.045389 2.51%
x3 -5.9722 2.26% 5.47%
5 x1 0.493702 10.92%
x2 7.985467 0.75%

108
 x3 -6.0186 0.77% 10.92%
6 x1 0.501047 1.47%
x2 7.99695 0.14%
x3 -5.99583 0.38% 1.47%

Thus, after 6 iterations, the maximum error is 1.47% and we arrive at the result: x1 =
0.501047, x2 = 7.99695 and x3 = −5.99583.

11.4 The first iteration can be implemented as

3800 + 3c 2 + c3 3800 + 3(0) + 0

c1 = = = 253.3333
15 15

1200 + 3c1 + 6c 3 1200 + 3(253.3333) + 6(0)

c2 = = = 108.8889
18 18

2350 + 4c1 + c 2 2350 + 4(253.3333) + 108.8889

c3 = = = 289.3519
12 12

Second iteration:

3800 + 3(108.889) + 289.3519

c1 = = 294.4012
15

1200 + 3( 294.4012) + 6(289.3519)

c2 = = 212.1842
18

2350 + 4(294.4012) + 212.1842

c3 = = 311.6491
12

The error estimates can be computed as

294.4012 − 253.3333
ε a ,1 = × 100% = 13.95%
294.4012

212.1842 − 108.8889
ε a,2 = × 100% = 48.68%
212.1842

311.6491 − 289.3519
ε a ,3 = × 100% = 7.15%
311.6491

The remainder of the calculation can be summarized as

iteration unknown value εa maximum εa

1 x1 253.3333 100.00%
x2 108.8889 100.00%
x3 289.3519 100.00% 100.00%

109
 2 x1 294.4012 13.95%
x2 212.1842 48.68%
x3 311.6491 7.15% 48.68%
3 x1 316.5468 7.00%
x2 223.3075 4.98%
x3 319.9579 2.60% 7.00%
4 x1 319.3254 0.87%
x2 226.5402 1.43%
x3 321.1535 0.37% 1.43%
5 x1 320.0516 0.23%
x2 227.0598 0.23%
x3 321.4388 0.09% 0.23%

Note that after several more iterations, we arrive at the result: x1 = 320.2073, x2 = 227.2021
and x3 = 321.5026.

11.5 The equations must first be rearranged so that they are diagonally dominant

− 8 x1 + x 2 − 2 x3 = −20

2 x1 − 6 x 2 − x3 = −38

− 3 x1 − x 2 + 7 x3 = −34

(a) The first iteration can be implemented as

− 20 − x 2 + 2 x 3 − 20 − 0 + 2(0)
x1 = = = 2 .5
−8 −8

− 38 − 2 x1 + x3 − 38 − 2( 2.5) + 0
x2 = = = 7.166667
−6 −6

− 34 + 3 x1 + x 2 − 34 + 3(2.5) + 7.166667
x3 = = = −2.761905
7 7

Second iteration:

− 20 − 7.166667 + 2(−2.761905)
x1 = = 4.08631
−8

− 38 − 2 x1 + x3 − 38 − 2(4.08631) + ( −2.761905)
x2 = = = 8.155754
−6 −6

− 34 + 3x1 + x 2 − 34 + 3( 4.08631) + 8.155754

x3 = = = −1.94076
7 7

The error estimates can be computed as

110
 4.08631 − 2.5
ε a ,1 = × 100% = 38.82%
4.08631

8.155754 − 7.166667
ε a,2 = × 100% = 12.13%
8.155754

− 1.94076 − (−2.761905)
ε a ,3 = × 100% = 42.31%
− 1.94076

The remainder of the calculation proceeds until all the errors fall below the stopping
criterion of 5%. The entire computation can be summarized as

iteration unknown value εa maximum εa

0 x1 0
x2 0
x3 0
1 x1 2.5 100.00%
x2 7.166667 100.00%
x3 -2.7619 100.00% 100.00%
2 x1 4.08631 38.82%
x2 8.155754 12.13%
x3 -1.94076 42.31% 42.31%
3 x1 4.004659 2.04%
x2 7.99168 2.05%
x3 -1.99919 2.92% 2.92%

Thus, after 3 iterations, the maximum error is 2.92% and we arrive at the result: x1 =
4.004659, x2 = 7.99168 and x3 = −1.99919.

(b) The same computation can be developed with relaxation where λ = 1.2.

First iteration:

− 20 − x 2 + 2 x 3 − 20 − 0 + 2(0)
x1 = = = 2 .5
−8 −8

Relaxation yields: x1 = 1.2(2.5) − 0.2(0) = 3

− 38 − 2 x1 + x3 − 38 − 2(3) + 0
x2 = = = 7.333333
−6 −6

Relaxation yields: x 2 = 1.2(7.333333) − 0.2(0) = 8.8

− 34 + 3 x1 + x 2 − 34 + 3(3) + 8.8
x3 = = = −2.3142857
7 7

Relaxation yields: x 3 = 1.2( −2.3142857) − 0.2(0) = −2.7771429

111
Second iteration:

− 20 − x 2 + 2 x3 − 20 − 8.8 + 2(−2.7771429)
x1 = = = 4.2942857
−8 −8

Relaxation yields: x1 = 1.2(4.2942857) − 0.2(3) = 4.5531429

− 38 − 2 x1 + x3 − 38 − 2(4.5531429) − 2.7771429
x2 = = = 8.3139048
−6 −6

Relaxation yields: x 2 = 1.2(8.3139048) − 0.2(8.8) = 8.2166857

− 34 + 3 x1 + x 2 − 34 + 3(4.5531429) + 8.2166857
x3 = = = −1.7319837
7 7

Relaxation yields: x 3 = 1.2( −1.7319837) − 0.2(−2.7771429) = −1.5229518

The error estimates can be computed as

4.5531429 − 3
ε a ,1 = × 100% = 34.11%
4.5531429

8.2166857 − 8.8
ε a,2 = × 100% = 7.1%
8.2166857

− 1.5229518 − (−2.7771429)
ε a ,3 = × 100% = 82.35%
− 1.5229518

The remainder of the calculation proceeds until all the errors fall below the stopping
criterion of 5%. The entire computation can be summarized as

iteration unknown value relaxation εa maximum εa

1 x1 2.5 3 100.00%
x2 7.3333333 8.8 100.00%
x3 -2.314286 -2.777143 100.00% 100.000%
2 x1 4.2942857 4.5531429 34.11%
x2 8.3139048 8.2166857 7.10%
x3 -1.731984 -1.522952 82.35% 82.353%
3 x1 3.9078237 3.7787598 20.49%
x2 7.8467453 7.7727572 5.71%
x3 -2.12728 -2.248146 32.26% 32.257%
4 x1 4.0336312 4.0846055 7.49%
x2 8.0695595 8.12892 4.38%
x3 -1.945323 -1.884759 19.28% 19.280%
5 x1 3.9873047 3.9678445 2.94%
x2 7.9700747 7.9383056 2.40%

112
 x3 -2.022594 -2.050162 8.07% 8.068%
6 x1 4.0048286 4.0122254 1.11%
x2 8.0124354 8.0272613 1.11%
x3 -1.990866 -1.979007 3.60% 3.595%

Thus, relaxation actually seems to retard convergence. After 6 iterations, the maximum
error is 3.595% and we arrive at the result: x1 = 4.0122254, x2 = 8.0272613 and x3 =
−1.979007.

11.6 As ordered, none of the sets will converge. However, if Set 1 and 3 are reordered so that
they are diagonally dominant, they will converge on the solution of (1, 1, 1).

Set 1: 8x + 3y + z = 12
2x + 4y – z = 5
−6x +7z = 1

Set 3: 3x + y − z = 3
x + 4y – z = 4
x + y +5z =7

Because it is not diagonally dominant, Set 2 will not converge on the correct solution of (1,
1, 1). However, it will also not diverge. Rather, it will oscillate. The way that this occurs
depends on how the equations are ordered. For example, if they can be ordered as

−2x + 4y − 5z = −3
2y – z = 1
−x + 3y + 5z = 7

For this case, Gauss-Seidel iterations yields

iteration unknown value εa maximum εa

1 x1 1.5 100.00%
x2 0.5 100.00%
x3 1.4 100.00% 100.00%
2 x1 -1 250.00%
x2 1.2 58.33%
x3 0.48 191.67% 250.00%
3 x1 2.7 137.04%
x2 0.74 62.16%
x3 1.496 67.91% 137.04%
4 x1 -0.76 455.26%
x2 1.248 40.71%
x3 0.4992 199.68% 455.26%
5 x1 2.748 127.66%
x2 0.7496 66.49%
x3 1.49984 66.72% 127.66%
6 x1 -0.7504 466.20%
x2 1.24992 40.03%
x3 0.499968 199.99% 466.20%
7 x1 2.74992 127.29%

113
 x2 0.749984 66.66%
x3 1.499994 66.67% 127.29%
8 x1 -0.75002 466.65%
x2 1.249997 40.00%
x3 0.499999 200.00% 466.65%

Alternatively, they can be ordered as

−x + 3y + 5z = 7
2y – z = 1
−2x + 4y − 5z = −3

For this case, Gauss-Seidel iterations yields

iteration unknown value εa maximum εa

1 x1 -7 100.00%
x2 0.5 100.00%
x3 3.8 100.00% 100.00%
2 x1 13.5 151.85%
x2 2.4 79.17%
x3 -2.88 231.94% 231.94%
3 x1 -14.2 195.07%
x2 -0.94 355.32%
x3 5.528 152.10% 355.32%
4 x1 17.82 179.69%
x2 3.264 128.80%
x3 -3.9168 241.14% 241.14%
5 x1 -16.792 206.12%
x2 -1.4584 323.81%
x3 6.15008 163.69% 323.81%
6 x1 19.3752 186.67%
x2 3.57504 140.79%
x3 -4.29005 243.36% 243.36%
7 x1 -17.7251 209.31%
x2 -1.64502 317.32%
x3 6.374029 167.31% 317.32%
8 x1 19.93507 188.91%
x2 3.687014 144.62%
x3 -4.42442 244.06% 244.06%

11.7 The equations to be solved are

f 1 ( x , y ) = − x 2 + x + 0 .5 − y

f 2 ( x, y ) = x 2 − y − 5 xy

The partial derivatives can be computed and evaluated at the initial guesses

114
 ∂f 1, 0 ∂f 1, 0
= −2 x + 1 = −2(1.2) + 1 = −1.4 = −1
∂x ∂y
∂f 2,0 ∂f 2,0
= 2 x − 5 y = 2(1.2) − 5(1.2) = −3.6 = −1 − 5 x = −1 − 5(1.2) = −7
∂x ∂y

They can then be used to compute the determinant of the Jacobian for the first iteration is

− 1.4(−7) − (−1)(−3.6) = 6.2

The values of the functions can be evaluated at the initial guesses as

f 1,0 = −1.2 2 + 1.2 + 0.5 − 1.2 = −0.94

f 2,0 = 1.2 2 − 5(1.2)(1.2) − 1.2 = −6.96

These values can be substituted into Eq. (11.12) to give

− 0.94(−3.6) − (−6.96)(−1)
x1 = 1.2 − = 1.26129
6.2
− 6.96(−1.4) − (−0.94)(−3.6)
x 2 = 1.2 − = 0.174194
6.2

The computation can be repeated until an acceptable accuracy is obtained. The results are
summarized as

iteration x y εa1 εa2

0 1.2 1.2
1 1.26129 0.174194 4.859% 588.889%
2 1.234243 0.211619 2.191% 17.685%
3 1.233319 0.212245 0.075% 0.295%
4 1.233318 0.212245 0.000% 0.000%

11.8 (a) The equations can be set up in a form amenable to plotting as

y = x2 −1

y = 5 − x2

These can be plotted as

115
 5

0
0 0.5 1 1.5 2 2.5
-1

-2

Thus, a solution seems to lie at about x = y = 1.6.

(b) The equations can be solved in a number of different ways. For example, the first
equation can be solved for x and the second solved for y. For this case, successive
substitution does not work

First iteration:
x = 5 − y 2 = 5 − (1.5) 2 = 1.658312
y = (1.658312) 2 − 1 = 1.75

Second iteration:
x = 5 − (1.75) 2 = 1.391941
y = (1.391941) 2 − 1 = 0.9375

Third iteration:
x = 5 − (0.9375) 2 = 2.030048
y = (2.030048) 2 − 1 = 3.12094

Thus, the solution is moving away from the solution that lies at approximately x = y = 1.6.

An alternative solution involves solving the second equation for x and the first for y.
For this case, successive substitution does work

First iteration:
x = y + 1 = 1.5 + 1 = 1.581139
y = 5 − x 2 = 5 − (1.581139) 2 = 1.581139

Second iteration:
x = 1.581139 = 1.606592
y = 5 − (1.606592) 2 = 1.555269

Third iteration:

116
x = 5 − (1.555269) 2 = 1.598521
y = (1.598521) 2 − 1 = 1.563564

After several more iterations, the calculation converges on the solution of x = 1.600485 and
y = 1.561553.

(c) The equations to be solved are

f 1 ( x, y ) = x 2 − y − 1

f 2 ( x, y ) = 5 − y 2 − x 2

The partial derivatives can be computed and evaluated at the initial guesses

∂f 1, 0 ∂f 1,0
= 2x = −1
∂x ∂y
∂f 2,0 ∂f 2, 0
= −2 x = −2 y
∂x ∂y

They can then be used to compute the determinant of the Jacobian for the first iteration is

− 1.4(−7) − (−1)(−3.6) = 6.2

The values of the functions can be evaluated at the initial guesses as

f 1,0 = −1.2 2 + 1.2 + 0.5 − 1.2 = −0.94

f 2,0 = 1.2 2 − 5(1.2)(1.2) − 1.2 = −6.96

These values can be substituted into Eq. (11.12) to give

− 0.94(−3.6) − (−6.96)(−1)
x1 = 1.2 − = 1.26129
6.2
− 6.96(−1.4) − (−0.94)(−3.6)
x 2 = 1.2 − = 0.174194
6.2

The computation can be repeated until an acceptable accuracy is obtained. The results are
summarized as

iteration ξ ψ εa1 εa2

0 1.5 1.5
1 1.604167 1.5625 6.494% 4.000%
2 1.600489 1.561553 0.230% 0.061%
3 1.600485 1.561553 0.000% 0.000%

117
CHAPTER 12
12.1 The data can be tabulated as

i y (yi – y )2
1 8.8 0.725904
2 9.4 0.063504
3 10 0.121104
4 9.8 0.021904
5 10.1 0.200704
6 9.5 0.023104
7 10.1 0.200704
8 10.4 0.559504
9 9.5 0.023104
10 9.5 0.023104
11 9.8 0.021904
12 9.2 0.204304
13 7.9 3.069504
14 8.9 0.565504
15 9.6 0.002704
16 9.4 0.063504
17 11.3 2.715904
18 10.4 0.559504
19 8.8 0.725904
20 10.2 0.300304
21 10 0.121104
22 9.4 0.063504
23 9.8 0.021904
24 10.6 0.898704
25 8.9 0.565504
Σ 241.3 11.8624

241.3
y= = 9.652
25

11.8624
sy = = 0.703041
25 − 1

s y2 = 0.7030412 = 0.494267

0.703041
c.v. = × 100% = 7.28%
9.652

12.2 The data can be sorted and then grouped. We assume that if a number falls on the border
between bins, it is placed in the lower bin.

lower upper Frequency

7.5 8 1

118
 8 8.5 0
8.5 9 4
9 9.5 7
9.5 10 6
10 10.5 5
10.5 11 1
11 11.5 1

The histogram can then be constructed as

8
7
6
Frequency

5
4
3
2
1
0
7 8 9 10 11 12
Bin

12.3 The data can be tabulated as

i y (yi – y )2
1 28.65 0.390625
2 28.65 0.390625
3 27.65 0.140625
4 29.25 1.500625
5 26.55 2.175625
6 29.65 2.640625
7 28.45 0.180625
8 27.65 0.140625
9 26.65 1.890625
10 27.85 0.030625
11 28.65 0.390625
12 28.65 0.390625
13 27.65 0.140625
14 27.05 0.950625
15 28.45 0.180625
16 27.65 0.140625
17 27.35 0.455625
18 28.25 0.050625
19 31.65 13.14063
20 28.55 0.275625
21 28.35 0.105625
22 28.85 0.680625
23 26.35 2.805625
24 27.65 0.140625

119
 25 26.85 1.380625
26 26.75 1.625625
27 27.75 0.075625
28 27.25 0.600625
Σ 784.7 33.0125

784.7
(a) y = = 28.025
28

33.0125
(b) s y = = 1.105751
28 − 1

(c) s y2 = 1.1057512 = 1.222685

1.105751
(d) c.v. = × 100% = 3.95%
28.025

(e) The data can be sorted and grouped.

Lower Upper Frequency

26 26.5 1
26.5 27 4
27 27.5 3
27.5 28 7
28 28.5 4
28.5 29 6
29 29.5 1
29.5 30 1
30 30.5 0
30.5 31 0
31 31.5 0
31.5 32 1

The histogram can then be constructed as

8
7
6
Frequency

5
4
3
2
1
0
26 27 28 29 30 31 32
Bin

120
 (f) 68% of the readings should fall between y − s y and y + s y . That is, between 28.025 –
1.10575096 = 26.919249 and 28.025 + 1.10575096 = 29.130751. Twenty values fall
between these bounds which is equal to 20/28 = 71.4% of the values which is not that
far from 68%.

12.4 The sum of the squares of the residuals for this case can be written as

n
Sr = ∑ (yi =1
i − a1 x i )
2

The partial derivative of this function with respect to the single parameter a1 can be
determined as

∂S r
∂a1
= −2 ∑ [( y i − a1 x i ) xi ]

Setting the derivative equal to zero and evaluating the summations gives

∑y i − a1 ∑x i

which can be solved for

a1 =
∑y i

∑x i

So the slope that minimizes the sum of the squares of the residuals for a straight line with a
zero intercept is merely the ratio of the sum of the dependent variables (y) over the sum of
the independent variables (x).

12.5
i xi yi x i2 x iy i
1 0 9.8100 0 0
2 20000 9.7487 4.0E+08 194974
3 40000 9.6879 1.6E+09 387516
4 60000 9.6278 3.6E+09 577668
5 80000 9.5682 6.4E+09 765456
Σ 200000 48.4426 1.2E+10 1925614

5(1,925,614) − 200,000(48.4426)
a1 = = −3.0225 × 10 −6
5(1.2 × 10 ) − 200,000
10 2

48.4426 200,000
a0 = − 3.0225 × 10 −6 = 9.80942
5 5

Therefore, the line of best fit is (using the nomenclature of the problem)

121
 g = 9.80942 − 3.0225 × 10 −6 y

The value at 55,000 m can therefore be computed as

g = 9.80942 − 3.0225 × 10 −6 (55,000) = 9.6431825

12.6 Regression gives

p = 8100.47 + 30.3164T (r2 = 0.999)

12000

10000

8000

6000
-50 0 50 100 150

⎛ p⎞V
R=⎜ ⎟
⎝T ⎠ n

p
= 30.3164
T

1 kg
n=
28 g/mole

⎛ 10 ⎞
R = 30.3164⎜ 3 ⎟ = 8.487
⎝ 10 / 28 ⎠

This is close to the standard value of 8.314 J/gmole.

12.7 Linear regression gives

0.6 y = 0.0454x + 0.1077

2
R = 0.999
0.4

0.2

0
0 2 4 6 8 10

122
 Forcing a zero intercept yields

0.6 y = 0.061x
2
R = 0.8387
0.4

0.2

0
0 2 4 6 8 10

One alternative that would force a zero intercept is a power fit

0.6 0.4069
y = 0.1827x
2
R = 0.9024
0.4

0.2

0
0 2 4 6 8 10

However, this seems to represent a poor compromise since it misses the linear trend in the
data. An alternative approach would to assume that the physically-unrealistic non-zero
intercept is an artifact of the measurement method. Therefore, if the linear slope is valid,
we might try y = 0.0454x.

12.8 The function can be linearized by dividing it by x and taking the natural logarithm to yield

ln( y / x) = ln α 4 + β 4 x

Therefore, if the model holds, a plot of ln(y/x) versus x should yield a straight line with an
intercept of lnα4 and an intercept of β4.

x y ln(y/x)
0.1 0.75 2.014903
0.2 1.25 1.832581
0.4 1.45 1.287854
0.6 1.25 0.733969
0.9 0.85 -0.05716
1.3 0.55 -0.8602
1.5 0.35 -1.45529
1.7 0.28 -1.80359
1.8 0.18 -2.30259

123
 3 y = -2.4733x + 2.2682
2
2 R = 0.9974
1
0
-1
-2
-3
0 0.5 1 1.5 2

Therefore, β4 = −2.4733 and α4 = e2.2682 = 9.661786, and the fit is

y = 9.661786 xe −2.4733 x

This equation can be plotted together with the data:

0
0 0.5 1 1.5 2

12.9 The data can be transformed, plotted and fit with a straight line

v, m/s F, N ln v ln F
10 25 2.302585 3.218876
20 70 2.995732 4.248495
30 380 3.401197 5.940171
40 550 3.688879 6.309918
50 610 3.912023 6.413459
60 1220 4.094345 7.106606
70 830 4.248495 6.721426
80 1450 4.382027 7.279319

8 y = 1.9842x - 1.2941
2
7 R = 0.9481
6
5
4
3
2
2 2.5 3 3.5 4 4.5

124
 The least-squares fit is

ln y = 1.9842 ln x − 1.2941

The exponent is 1.9842 and the leading coefficient is e−1.2941 = 0.274137. Therefore, the
result is the same as when we used common or base-10 logarithms:

y = 0.274137 x 1.9842

12.10 (a) The data can be plotted

2000
1600
1200
800
400
0
0 10 20 30

The plot indicates that the data is somewhat curvilinear. An exponential model (i.e., a semi-
log plot) is the best choice to linearize the data. This conclusion is based on

• A power model does not result in a linear plot

• Bacterial decay is known to follow an exponential model
• The exponential model by definition will not produce negative values.

The exponential fit can be determined as

t (hrs) c (CFU/100 mL) ln c

4 1590 7.371489
8 1320 7.185387
12 1000 6.907755
16 900 6.802395
20 650 6.476972
24 560 6.327937

125
 y = -0.0532x + 7.5902
7.6 2
R = 0.9887
7.2
6.8
6.4
6
0 10 20 30

Therefore, the coefficient of the exponent (β1) is −0.0532 and the lead coefficient (α1) is
e7.5902 = 1978.63, and the fit is

c = 1978.63e −0.0532t

Consequently the concentration at t = 0 is 1978.63 CFU/100 ml. Here is a plot of the fit
along with the original data:

2400
2000
1600
1200
800
400
0
0 10 20 30

(b) The time at which the concentration will reach 200 CFU/100 mL can be computed as

200 = 1978.63e −0.0532t

⎛ 200 ⎞
ln⎜ ⎟ = −0.0532t
⎝ 1978.63 ⎠

⎛ 200 ⎞
ln⎜ ⎟
t= ⎝
1978.63 ⎠
= 43.08 d
− 0.0532

12.11 (a) The exponential fit can be determined with the base-10 logarithm as

t (hrs) c (CFU/100 mL) log c

4 1590 3.201397
8 1320 3.120574
12 1000 3
16 900 2.954243

126
 20 650 2.812913
24 560 2.748188

3.3 y = -0.0231x + 3.2964

2
3.2 R = 0.9887
3.1
3
2.9
2.8
2.7
0 10 20 30

Therefore, the coefficient of the exponent (β5) is −0.0231 and the lead coefficient (α5) is
103.2964 = 1978.63, and the fit is

c = 1978.63(10) −0.0231t

Consequently the concentration at t = 0 is 1978.63 CFU/100 ml.

(b) The time at which the concentration will reach 200 CFU/100 mL can be computed as

200 = 1978.63(10) −0.0231t

⎛ 200 ⎞
log10 ⎜ ⎟ = −0.0231t
⎝ 1978.63 ⎠

⎛ 200 ⎞
log10 ⎜ ⎟
t= ⎝ 1978.63 ⎠
= 43.08 d
− 0.0231

Thus, the results are identical to those obtained with the base-e model.

The relationship between β1 and β5 can be developed as in

e −α1t = 10 −α 5t

Take the natural log of this equation to yield

− α 1t = −α 5 t ln 10

or

α 1 = 2.302585α 5

127
12.12 The power fit can be determined as

W (kg) A (m2) log W log A

70 2.1 1.845098 0.322219
75 2.12 1.875061 0.326336
77 2.15 1.886491 0.332438
80 2.2 1.90309 0.342423
82 2.22 1.913814 0.346353
84 2.23 1.924279 0.348305
87 2.26 1.939519 0.354108
90 2.3 1.954243 0.361728

0.37
logA = 0.3799logW - 0.3821
0.36 2
R = 0.9711
0.35

0.34

0.33

0.32

0.31
1.8 1.84 1.88 1.92 1.96

Therefore, the power is b = 0.3799 and the lead coefficient is a = 10−0.3821 = 0.4149, and the
fit is

A = 0.4149W 0.3799

Here is a plot of the fit along with the original data:

2.35
2.3
2.25
2.2
2.15
2.1
2.05
70 75 80 85 90

The value of the surface area for a 95-kg person can be estimated as

A = 0.4149(95) 0.3799 = 2.34 m 2

128
12.13 The power fit can be determined as

Mass Metabolism
(kg) (kCal/day) log Mass log Met
300 5600 2.477121 3.748188
70 1700 1.845098 3.230449
60 1100 1.778151 3.041393
2 100 0.30103 2
0.3 30 -0.52288 1.477121

2
logMet = 0.7497logMass + 1.818
1 2
R = 0.9935

0
-1 0 1 2 3

Therefore, the power is b = 0.7497 and the lead coefficient is a = 101.818 = 65.768, and the
fit is

Metabolism = 65.768Mass 0.7497

Here is a plot of the fit along with the original data:

8000

6000
4000

2000

0
0 100 200 300 400

12.14 Linear regression of the log transformed data yields

logε& = −5.41 log B + 2.6363 log σ (r2 = 0.9997)

129
 0.7 0.8 0.9 1 1.1 1.2
-2.4

-2.8

-3.2

-3.6

Therefore,

B = 10 -5.41 = 3.88975 × 10 -6

m = 2.6363

and the untransformed model is

ε& = 3.88975 × 10 -6 σ 2.6363

A plot of the data and the model can be developed as

0.005
0.004
0.003
0.002
0.001
0
0 5 10 15

12.15 Linear regression of the data yields

τ = 2.779 + 0.685γ& (r2 = 0.977121)

8
6
4
2
0
0 2 4 6

Therefore, µ = 0.685 and τy = 2.779 N/m2.

130
12.16 The data can be transformed

strain stress log(strain) log(stress)

50 5.99 1.69897 0.777427
70 7.45 1.845098 0.872156
90 8.56 1.954243 0.932474
110 9.09 2.041393 0.958564
130 10.25 2.113943 1.010724

Linear regression of the transformed data yields

log τ = −0.13808 + 0.54298 log γ& (r2 = 0.989118)

1.1
1
0.9
0.8
0.7
0.6
1.6 1.8 2 2.2

Therefore, µ = 10–0.54298 = 0.72765 and n = 0.54298. The power model is therefore,

τ = 0.72765γ& 0.54298

A plot of the power model along with the data can be created as

12

0
0 50 100 150

131
CHAPTER 13
13.1 The data can be tabulated and the sums computed as

i x y x2 x3 x4 xy x2y
1 10 25 100 1000 10000 250 2500
2 20 70 400 8000 160000 1400 28000
3 30 380 900 27000 810000 11400 342000
4 40 550 1600 64000 2560000 22000 880000
5 50 610 2500 125000 6250000 30500 1525000
6 60 1220 3600 216000 12960000 73200 4392000
7 70 830 4900 343000 24010000 58100 4067000
8 80 1450 6400 512000 40960000 116000 9280000
Σ 360 5135 20400 1296000 87720000 312850 20516500

Normal equations:

⎡ 8 360 20400 ⎤ ⎧⎪a 0 ⎫⎪ ⎧⎪ 5135 ⎫⎪

⎢ 360 20400 1296000 ⎥ ⎨ a1 ⎬ = ⎨ 312850 ⎬
⎢⎣20400 1296000 87720000⎥⎦ ⎪⎩a 2 ⎪⎭ ⎪⎩20516500⎪⎭

which can be solved for the coefficients yielding the following best-fit polynomial

F = −178.4821 + 16.12202v + 0.037202v 2

Here is the resulting fit:

2000

1500

1000

500

0
0 20 40 60 80 100
-500

The predicted values can be used to determined the sum of the squares. Note that the mean
of the y values is 641.875.

i x y ypred ( yi − y) 2 (y − ypred)2
1 10 25 -13.5417 380535 1485
2 20 70 158.8393 327041 7892
3 30 380 338.6607 68579 1709

132
 4 40 550 525.9226 8441 580
5 50 610 720.625 1016 12238
6 60 1220 922.7679 334229 88347
7 70 830 1132.351 35391 91416
8 80 1450 1349.375 653066 10125
Σ 1808297 213793

The coefficient of determination can be computed as

1808297 − 213793
r2 = = 0.88177
1808297

The model fits the trend of the data nicely, but it has the deficiency that it yields physically
unrealistic negative forces at low velocities.

13.2 The sum of the squares of the residuals for this case can be written as

∑ (y )
n
2
Sr = i − a1 x i − a 2 x i2
i =1

The partial derivatives of this function with respect to the unknown parameters can be
determined as

∑ [( y ]
∂S r
= −2 − a1 x i − a 2 xi2 ) x i
∂a1
i

∑ [( y ]
∂S r
= −2 − a1 x i − a 2 x i2 ) xi2
∂a 2
i

Setting the derivative equal to zero and evaluating the summations gives

(∑ x )a + (∑ x )a = ∑ x y
2
i 1
3
i 2 i i

(∑ x )a + (∑ x )a = ∑ x
3
i 1
4
i 2
2
i yi

which can be solved for

a1 =
∑x y ∑x −∑x y ∑x
i i
4
i
2
i i
3
i

∑ x ∑ x − (∑ x ) 2
i
4
i
3 2
i

a2 =
∑x ∑x y −∑x y ∑x
2
i
2
i i i i
3
i

∑ x ∑ x − (∑ x ) 2
i
4
i
3 2
i

The model can be tested for the data from Table 12.1.

133
 x y x2 x3 x4 xy x2y
10 25 100 1000 10000 250 2500
20 70 400 8000 160000 1400 28000
30 380 900 27000 810000 11400 342000
40 550 1600 64000 2560000 22000 880000
50 610 2500 125000 6250000 30500 1525000
60 1220 3600 216000 12960000 73200 4392000
70 830 4900 343000 24010000 58100 4067000
80 1450 6400 512000 40960000 116000 9280000
Σ 20400 1296000 87720000 312850 20516500

312850(87720000) − 20516500(1296000)
a1 = = 7.771024
20400(87720000) − (1296000) 2

20400(20516500) − 312850(1296000)
a2 = = 0.119075
20400(87720000) − (1296000)

Therefore, the best-fit model is

y = 7.771024 x + 0.119075 x 2

The fit, along with the original data can be plotted as

2500
2000
1500
1000
500
0
0 20 40 60 80 100

13.3 The data can be tabulated and the sums computed as

i x y x2 x3 x4 x5 x6 xy x2y x3y
1 3 1.6 9 27 81 243 729 4.8 14.4 43.2
2 4 3.6 16 64 256 1024 4096 14.4 57.6 230.4
3 5 4.4 25 125 625 3125 15625 22 110 550
4 7 3.4 49 343 2401 16807 117649 23.8 166.6 1166.2
5 8 2.2 64 512 4096 32768 262144 17.6 140.8 1126.4
6 9 2.8 81 729 6561 59049 531441 25.2 226.8 2041.2
7 11 3.8 121 1331 14641 161051 1771561 41.8 459.8 5057.8
8 12 4.6 144 1728 20736 248832 2985984 55.2 662.4 7948.8
Σ 59 26.4 509 4859 49397 522899 5689229 204.8 1838.4 18164

Normal equations:

134
 ⎡ 8 59 509 4859 ⎤ ⎧a 0 ⎫ ⎧ 26.4 ⎫
⎢ 59 509 4859 49397 ⎥ ⎪ a1 ⎪ = ⎪ 204.8 ⎪
⎢ 509 4859 49397 522899 ⎥ ⎨a 2 ⎬ ⎨1838.4⎬
⎢⎣4859 49397 522899 5689229⎥⎦ ⎪a ⎪ ⎪⎩18164 ⎪⎭
⎩ 3⎭

which can be solved for the coefficients yielding the following best-fit polynomial

y = −11.4887 + 7.143817 x − 1.04121x 2 + 0.046676 x 3

Here is the resulting fit:

0
0 5 10 15
-2

The predicted values can be used to determined the sum of the squares. Note that the mean
of the y values is 3.3.

i x y ypred ( yi − y) 2 (y − ypred)2
1 3 1.6 1.83213 2.8900 0.0539
2 4 3.6 3.41452 0.0900 0.0344
3 5 4.4 4.03471 1.2100 0.1334
4 7 3.4 3.50875 0.0100 0.0118
5 8 2.2 2.92271 1.2100 0.5223
6 9 2.8 2.4947 0.2500 0.0932
7 11 3.8 3.23302 0.2500 0.3215
8 12 4.6 4.95946 1.6900 0.1292
Σ 7.6000 1.2997

The coefficient of determination can be computed as

7.6 − 1.2997
r2 = = 0.829
7 .6

13.4
function p = polyreg(x,y,m)
% polyreg(x,y,m):
% Polynomial regression.

135
 % input:
% x = independent variable
% y = dependent variable
% m = order of polynomial
% output:
% p = vector of coefficients

n = length(x);
if length(y)~=n, error('x and y must be same length'); end
for i = 1:m+1
for j = 1:i
k = i+j-2;
s = 0;
for l = 1:n
s = s + x(l)^k;
end
A(i,j) = s;
A(j,i) = s;
end
s = 0;
for l = 1:n
s = s + y(l)*x(l)^(i-1);
end
b(i) = s;
end
p = A\b';

Test solving Prob. 13.3:

>> x = [3 4 5 7 8 9 11 12];
>> y = [1.6 3.6 4.4 3.4 2.2 2.8 3.8 4.6];
>> polyreg(x,y,3)

ans =

-11.4887
7.1438
-1.0412
0.0467

13.5 Because the data is curved, a linear regression will undoubtedly have too much error.
Therefore, as a first try, fit a parabola,

>> format long

>> T = [0 5 10 15 20 25 30];
>> c = [14.6 12.8 11.3 10.1 9.09 8.26 7.56];
>> p = polyfit(T,c,2)

p =
0.00439523809524 -0.36335714285714 14.55190476190477

Thus, the best-fit parabola would be

c = 14.55190476 − 0.36335714T + 0.0043952381T 2

136
 16

12

0
0 5 10 15 20 25 30

We can use this equation to generate predictions corresponding to the data. When these
values are rounded to the same number of significant digits the results are

T c-data c-pred rounded

0 14.6 14.55190 14.6
5 12.8 12.84500 12.8
10 11.3 11.35786 11.4
15 10.1 10.09048 10.1
20 9.09 9.04286 9.04
25 8.26 8.21500 8.22
30 7.56 7.60690 7.61

Thus, although the plot looks good, discrepancies occur in the third significant digit.

We can, therefore, fit a third-order polynomial

>> p = polyfit(T,c,3)

p =
-0.00006444444444 0.00729523809524 -0.39557936507936 14.60023809523810

Thus, the best-fit cubic would be

c = 14.600238095 − 0.395579365T + 0.007295238T 2 − 0.000064444T 3

We can use this equation to generate predictions corresponding to the data. When these
values are rounded to the same number of significant digits the results are

T c-data c-pred rounded

0 14.6 14.60020 14.6
5 12.8 12.79663 12.8
10 11.3 11.30949 11.3
15 10.1 10.09044 10.1
20 9.09 9.09116 9.09
25 8.26 8.26331 8.26
30 7.56 7.55855 7.56

Thus, the predictions and data agree to three significant digits.

137
13.6 The multiple linear regression model to evaluate is

o = a 0 + a1T + a 2 c

The [Z] and y matrices can be set up using MATLAB commands in a fashion similar to
Example 13.4,

>> format long

>> t = [0 5 10 15 20 25 30];
>> T = [t t t]';
>> c = [zeros(size(x)) 10*ones(size(x)) 20*ones(size(x))]';
>> Z = [ones(size(T)) T c];
>> y = [14.6 12.8 11.3 10.1 9.09 8.26 7.56 12.9 11.3 10.1 9.03 8.17
7.46 6.85 11.4 10.3 8.96 8.08 7.35 6.73 6.2]';

The coefficients can be evaluated as

>> a = Z\y

a =
13.52214285714286
-0.20123809523810
-0.10492857142857

Thus, the best-fit multiple regression model is

o = 13.52214285714286 − 0.20123809523810T − 0.10492857142857c

We can evaluate the prediction at T = 12 and c = 15 and evaluate the percent relative error
as

>> cp = a(1)+a(2)*12+a(3)*15

cp =
9.53335714285714

>> ea = abs((9.09-cp)/9.09)*100

ea =
4.87741631305987

Thus, the error is considerable. This can be seen even better by generating predictions for
all the data and then generating a plot of the predictions versus the data. A one-to-one line
is included to show how the predictions diverge from a perfect fit.

138
 16

12

4
4 8 12 16

The cause for the discrepancy is because the dependence of oxygen concentration on the
unknowns is significantly nonlinear. It should be noted that this is particularly the case for
the dependency on temperature.

13.7 The multiple linear regression model to evaluate is

y = a 0 + a1T + a 2 T 2 + a 3T 3 + a 4 c

The [Z] matrix can be set up as in

>> T = 0:5:30;
>> T = [T T T]';
>> c = [0 0 0 0 0 0 0 10 10 10 10 10 10 10 20 20 20 20 20 20 20]';
>> o = [1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1]';
>> y = [14.6 12.8 11.3 10.1 9.09 8.26 7.56 12.9 11.3 10.1 9.03 8.17
7.46 6.85 11.4 10.3 8.96 8.08 7.35 6.73 6.2]';
>> Z = [o T T.^2 T.^3 c];

Then, the coefficients can be generated by solving Eq.(13.10)

>> format long

>> a = (Z'*Z)\[Z'*y]

a =
14.02714285714287
-0.33642328042328
0.00574444444444
-0.00004370370370
-0.10492857142857

Thus, the least-squares fit is

y = 14.027143 − 0.336423T + 0.00574444T 2 − 0.000043704T 3 − 0.10492857c

The model can then be used to predict values of oxygen at the same values as the data.
These predictions can be plotted against the data to depict the goodness of fit.

139
 >> yp = Z*a
>> plot(y,yp,'o')

Finally, the prediction can be made at T = 12 and c = 15,

>> a(1)+a(2)*12+a(3)*12^2+a(4)*12^3+a(5)*15
ans =
9.16781492063485

which compares favorably with the true value of 9.09 mg/L.

13.8 The multiple linear regression model to evaluate is

y = a 0 + a1 x1 + a 2 x 2

The [Z] matrix can be set up as in

>> x1 = [0 1 1 2 2 3 3 4 4]';
>> x2 = [0 1 2 1 2 1 2 1 2]';
>> y = [15.1 17.9 12.7 25.6 20.5 35.1 29.7 45.4 40.2]';
>> o = [1 1 1 1 1 1 1 1 1]';
>> Z = [o x1 x2 y];

Then, the coefficients can be generated by solving Eq.(13.10)

>> a = (Z'*Z)\[Z'*y]

a =
14.4609
9.0252
-5.7043

Thus, the least-squares fit is

y = 14.4609 + 9.0252 x1 − 5.7043x 2

140
 The model can then be used to predict values of the unknown at the same values as the
data. These predictions can be used to determine the correlation coefficient and the standard
error of the estimate.

>> yp = Z*a

>> SSR = sum((yp - y).^2)

SSR =
4.7397

>> SST = sum((y - mean(y)).^2)

SST =
1.0587e+003

>> r2 = (SST - SSR)/SST

r2 =
0.9955

>> r = sqrt(r2)
r =
0.9978

>> syx = sqrt(SSR/(length(y)-3))

syx =
0.8888

13.9 The multiple linear regression model to evaluate is

log Q = log α 0 + α 1 log( D) + α 2 log(S )

The [Z] matrix can be set up as in

>> D = [.3 .6 .9 .3 .6 .9 .3 .6 .9]';

>> S = [.001 .001 .001 .01 .01 .01 .05 .05 .05]';
>> Q = [.04 .24 .69 .13 .82 2.38 .31 1.95 5.66]';
>> o = [1 1 1 1 1 1 1 1 1]';
>> Z = [o log10(D) log10(S)]

Then, the coefficients can be generated by solving Eq.(13.10)

>> a = (Z'*Z)\[Z'*log10(Q)]
a =
1.5609
2.6279
0.5320

Thus, the least-squares fit is

log Q = 1.5609 + 2.6279 log( D) + 0.5320 log(S )

Taking the inverse logarithm gives

141
 Q = 101.5609 D 2.6279 S 0.5320 = 36.3813D 2.6279 S 0.5320

13.10 The linear regression model to evaluate is

p (t ) = Ae −1.5t + Be −0.3t + Ce −0.05t

The unknowns can be entered and the [Z] matrix can be set up as in

>> p = [7 5.2 3.8 3.2 2.5 2.1 1.8 1.5 1.2 1.1]';
>> t = [0.5 1 2 3 4 5 6 7 8 9]';
>> Z = [exp(-1.5*t) exp(-0.3*t) exp(-0.05*t)];

Then, the coefficients can be generated by solving Eq.(13.10)

>> Z = [exp(-1.5*t) exp(-0.3*t) exp(-0.05*t)];

>> a = (Z'*Z)\[Z'*p]
a =
3.7778
4.3872
1.3775

Thus, the least-squares fit is

p (t ) = 3.7778e −1.5t + 4.3872e −0.3t + 1.3775e −0.05t

The fit and the data can be plotted as

>> pp = Z*a
>> plot(t,p,'o',t,pp)

13.11 First, an M-file function must be created to compute the sum of the squares,

function f = fSSR(a,Im,Pm)
Pp = a(1)*Im/a(2).*exp(-Im/a(2)+1);
f = sum((Pm-Pp).^2);

142
 The data can then be entered as

>> I = [50 80 130 200 250 350 450 550 700];

>> P = [99 177 202 248 229 219 173 142 72];

The minimization of the function is then implemented by

>> a = fminsearch(@fSSR, [200, 200], [], I, P)

a =
238.7124 221.8239

The best-fit model is therefore

I
I − +1
P = 238.7124 e 221.8239
221.8239

The fit along with the data can be displayed graphically.

>> Pp = a(1)*I/a(2).*exp(-I/a(2)+1);
>> plot(I,P,'o',I,Pp)

13.12 First, an M-file function must be created to compute the sum of the squares,

function f = fSSR(a,xm,ym)
yp = a(1)*xm.*exp(a(2)*xm);
f = sum((ym-yp).^2);

The data can then be entered as

>> x = [.1 .2 .4 .6 .9 1.3 1.5 1.7 1.9];

>> y = [0.75 1.25 1.45 1.25 0.85 0.55 0.35 0.28 0.18];

143
 The minimization of the function is then implemented by

>> a = fminsearch(@fSSR, [1, 1], [], x, y)

a =
9.8545 -2.5217

The best-fit model is therefore

y = 9.8545 xe −2.5217 x

The fit along with the data can be displayed graphically.

>> yp = a(1)*x.*exp(a(2)*x);
>> plot(x,y,'o',x,yp)

13.13 (a) The model can be linearized by inverting it,

1 K 1 1
= 3
+
v0 k m [S ] km

If this model is valid, a plot of 1/v0 versus 1/[S]3 should yield a straight line with a slope of
K/km and an intercept of 1/km. The slope and intercept can be implemented in MATLAB
using the M-file function linregr (Fig. 12.12),

>> S = [.01 .05 .1 .5 1 5 10 50 100];

>> v0 = [6.078e-11 7.595e-9 6.063e-8 5.788e-6 1.737e-5 2.423e-5
2.43e-5 2.431e-5 2.431e-5];
>> a = linregr(1./S.^3,1./v0)

a =
1.0e+004 *
1.64527391375701 4.13997346408367

144
These results can then be used to compute km and K,

>> km=1/a(2)
km =
2.415474419523452e-005

>> K=km*a(1)
K =
0.39741170517893

Thus, the best-fit model is

2.415474 × 10 −5 [ S ]3
v0 =
0.39741 + [ S ]3

The fit along with the data can be displayed graphically. We will use a log-log plot because
of the wide variation of the magnitudes of the values being displayed,

>> v0p = km*S.^3./(K+S.^3);

>> loglog(S,v0,'o',S,v0p)

(b) An M-file function must be created to compute the sum of the squares,

function f = fSSR(a,Sm,v0m)
v0p = a(1)*Sm.^3./(a(2)+Sm.^3);
f = sum((v0m-v0p).^2);

The data can then be entered as

>> S = [.01 .05 .1 .5 1 5 10 50 100];

>> v0 = [6.078e-11 7.595e-9 6.063e-8 5.788e-6 1.737e-5 2.423e-5
2.43e-5 2.431e-5 2.431e-5];

145
The minimization of the function is then implemented by

>> format long

>> a = fminsearch(@fSSR, [2e-5, 1], [], S, v0)

a =
0.00002430998303 0.39976314533880

The best-fit model is therefore

2.431 × 10 −5 [ S ]3
v0 =
0.399763 + [ S ]3

The fit along with the data can be displayed graphically. We will use a log-log plot because
of the wide variation of the magnitudes of the values being displayed,

>> v0p = a(1)*S.^3./(a(2)+S.^3);

>> loglog(S,v0,'o',S,v0p)

146
CHAPTER 14
14.1 (a) Newton’s polynomial. Ordering of points:

x1 = 3 f(x1) = 6.5
x2 = 4 f(x2) = 2
x3 = 2.5 f(x3) = 7
x4 = 5 f(x4) = 0

Note that based purely on the distance from the unknown, the fourth point would be (2, 5).
However, because it provides better balance and is located only a little bit farther from the
unknown, the point at (5, 0) is chosen.

First order:

2 − 6.5
f 1 (3.4) = 6.5 + (3.4 − 3) = 6.5 + ( −4.5)(3.4 − 3) = 4.7
4−3

Second order:

7−2
− (−4.5)
f 2 (3.4) = 4.7 + 2.5 − 4 (3.4 − 3)(3.4 − 4)
2.5 − 3
− 3.333333 − (−4.5)
= 4.7 + (3.4 − 3)(3.4 − 4)
2 .5 − 3
= 4.7 + (−2.333333)(3.4 − 3)(3.4 − 4) = 5.259887

Third order:

0−7
− (−3.333333)
5 − 2 .5 − (−2.333333)
f 3 (3.4) = 5.259887 + 5−4 (3.4 − 3)(3.4 − 4)(3.4 − 2.5)
5−3
− 2.8 − (−3.333333)
− (−2.333333)
= 5.259887 + 5−4 (3.4 − 3)(3.4 − 4)(3.4 − 2.5)
5−3
0.5333333 − (−2.333333)
= 5.259887 + (3.4 − 3)(3.4 − 4)(3.4 − 2.5) = 4.95152
5−3

(b) Lagrange polynomial.

First order:

3 .4 − 4 3 .4 − 3
f 1 (3.4) = 6 .5 + 2 = 4.7
3−4 4−3

147
 Second order:

(3.4 − 4)(3.4 − 2.5) (3.4 − 3)(3.4 − 2.5) (3.4 − 3)(3.4 − 4)

f 2 (3.4) = 6.5 + 2+ 7 = 5.259887
(3 − 4)(3 − 2.5) (4 − 3)(4 − 2.5) (2.5 − 3)(2.5 − 4)

Third order:

(3.4 − 4)(3.4 − 2.5)(3.4 − 5) (3.4 − 3)(3.4 − 2.5)(3.4 − 5)

f 3 (3.4) = 6.5 + 2
(3 − 4)(3 − 2.5)(3 − 5) ( 4 − 3)(4 − 2.5)(4 − 5)
(3.4 − 3)(3.4 − 4)(3.4 − 5) (3.4 − 3)(3.4 − 4)(3.4 − 2.5)
+ 7+ 0 = 4.95152
(2.5 − 3)(2.5 − 4)(2.5 − 5) (5 − 3)(5 − 4)(5 − 2.5)

14.2 The points can be ordered so that they are close to and centered around the unknown. A
divided-difference table can then be developed as

x f(x) First Second Third Fourth

3 5.25 7.25 2 0.25 0
5 19.75 5.25 2.75 0.25
2 4 8 1.75
6 36 6.25
1 4.75

Note that the fact that the fourth divided difference is zero means that the data was
generated with a third-order polynomial.

First order:

f 1 (4) = 5.25 + 7.25(4 − 3) = 12.5

Second order:

f 2 ( 4) = 12.5 + (4 − 3)(4 − 5)2 = 10.5

Third order:

f 3 (4) = 10.5 + (4 − 3)(4 − 5)(4 − 2)0.25 = 10

Fourth order:

f 3 (4) = 10.5 + (4 − 3)(4 − 5)(4 − 2)(4 − 6)0 = 10

14.3 Lagrange polynomial.

First order:

4−5 4−3
f 1 ( 4) = 5.25 + 19.75 = 12.5
3−5 5−3

148
 Second order:

(4 − 5)(4 − 2) (4 − 3)(4 − 2) (4 − 3)(4 − 5)

f 2 ( 4) = 5.25 + 19.75 + 4 = 10.5
(3 − 5)(3 − 2) (5 − 3)(5 − 2) (2 − 3)(2 − 5)

Third order:

(4 − 5)(4 − 2)(4 − 6) (4 − 3)(4 − 2)(4 − 6)

f 3 ( 4) = 5.25 + 19.75
(3 − 5)(3 − 2)(3 − 6) (5 − 3)(5 − 2)(5 − 6)
(4 − 3)(4 − 5)(4 − 6) ( 4 − 3)(4 − 5)(4 − 2)
+ 4+ 36 = 10
(2 − 3)(2 − 5)(2 − 6) (6 − 3)(6 − 5)(6 − 2)

14.4 (a) The points can be ordered so that they are close to and centered around the unknown. A
divided-difference table can then be developed as

T, oC c = 10 g/L first second third

10 10.1 -0.214 0.0026 0.000107
15 9.03 -0.227 0.003667
5 11.3 -0.20867
20 8.17

Second order:

f 2 ( 4) = 10.1 − 0.214(12 − 10) + 0.0026(12 − 10)(12 − 15) = 9.6564

Third order:

f 3 (4) = 9.6564 + 0.000107(12 − 10)(12 − 15)(12 − 5) = 9.65192

(b) First, linear interpolation can be used to generate values for T = 10 and 15 at c = 15,

8.96 − 10.1
f 1 (T = 10, c = 15) = 10.1 + (15 − 10) = 9.53
20 − 10

8.08 − 9.03
f 1 (T = 15, c = 15) = 9.03 + (15 − 10) = 8.555
20 − 10

These values can then be used to determine the result at T = 12,

8.555 − 9.53
f 1 (T = 12, c = 15) = 9.53 + (12 − 10) = 9.14
15 − 10

(c) First, quadratic interpolation can be used to generate values for T = 5, 10 and 15 at c =
15,

f 2 (T = 5, c = 15) = 12.8 − 0.15(15 − 0) + 0.0025(15 − 0)(15 − 10) = 10.7375

149
 f 2 (T = 10, c = 15) = 11.3 − 0.12(15 − 0) + 0.0003(15 − 0)(15 − 10) = 9.5225

f 2 (T = 15, c = 15) = 10.1 − 0.107(15 − 0) + 0.0006(15 − 0)(15 − 10) = 8.54

These values can then be used to determine the result at T = 12,

f 2 (T = 12, c = 15) = 10.7375 − 0.243(12 − 5) + 0.00465(12 − 5)(12 − 10) = 9.1016

14.5 MATLAB can be used to generate a cubic polynomial through the first 4 points in the table,

>> x = [1 2 3 4];
>> fx = [3.6 1.8 1.2 0.9];
>> p = polyfit(x,fx,3)
p =
-0.1500 1.5000 -5.2500 7.5000

Therefore, the roots problem to be solved is

1.6 = −0.15 x 3 +1.5 x 2 −5.25 x + 7.5

or

f ( x) = −0.15 x 3 +1.5 x 2 −5.25 x + 5.9 = 0

Bisection can be employed the root of this polynomial. Using initial guesses of xl = 2 and xu
= 3, a value of 2.2156 is obtained with εa = 0.00069% after 16 iterations.

14.6 (a) Analytical:

x2
0.93 =
1 + x2

0.93 + 0.93 x 2 = x 2

0.07 x 2 = 0.93

0.93
x= = 3.644957
0.07

(b) A quadratic interpolating polynomial can be fit to the last three points using the
polyfit function,

>> format long

>> x = [3 4 5];
>> y = x.^2./(1+x.^2);
>> p = polyfit(x,y,2)
p =
-0.01040723981900 0.11402714932127 0.65158371040724

150
 Thus, the best fit quadratic is

f 2 ( x) = −0.01040724 x 2 + 0.11402715 x + 0.6515837

We must therefore find the root of

0.93 = −0.01040724 x 2 + 0.11402715 x + 0.6515837

or

f ( x) = −0.01040724 x 2 + 0.11402715 x − 0.2784163

The quadratic formula yields

− 0.11402715 ± (0.11402715) 2 − 4(−0.01040724)(−0.2784163)

x= = 7.2835775
2(0.11402715) 3.6729442

Thus, the estimate is 3.67294421.

(c) A cubic interpolating polynomial can be fit to the last four points using the polyfit
function,

>> format long

>> x = [2 3 4 5];
>> y=x.^2./(1+x.^2)
>> p = polyfit(x,y,3)
p =
0.00633484162896 -0.08642533936652 0.41176470588235 0.27149321266968

Thus, the best fit cubic is

f 3 ( x) = 0.006334842 x 3 − 0.08642534 x 2 + 0.4117647 x + 0.2714932

We must therefore find the root of

0.93 = 0.006334842 x 3 − 0.08642534 x 2 + 0.4117647 x + 0.2714932

or

f ( x) = 0.006334842 x 3 − 0.08642534 x 2 + 0.4117647 x − 0.6585068

Bisection can be employed the root of this polynomial. Using initial guesses of xl = 3 and xu
= 4, a value of 3.61883 is obtained.

14.7 (a) Because they bracket the unknown, the two last points are used for linear interpolation,

151
 6.7664 − 6.5453
f 1 (0.118) = 6.5453 + (0.118 − 0.11144) = 6.6487
0.12547 − 0.11144

(b) The quadratic interpolation can be implemented easily in MATLAB,

>> v = [0.10377 0.1144 0.12547];

>> s = [6.4147 6.5453 6.7664];
>> p = polyfit(v,s,2)
p =
354.2358 -64.9976 9.3450
>> polyval(p,0.118)
ans =
6.6077

Therefore, to the level of significance reported in the table the estimated entropy is 6.6077

(c) The inverse interpolation can be implemented in MATLAB. First, as in part (b), we can
fit a quadratic polynomial to the data to yield,

p =
354.2358 -64.9976 9.3450

We must therefore find the root of

6.45 = 354.2358 x 2 − 64.9976 x + 9.3450

or

6.45 = 354.2358 x 2 − 64.9976 x + 2.8950

In MATLAB, we can generate this polynomial by subtracting 6.45 from the constant
coefficient of the polynomial

>> p(3)=p(3)-6.45
p =
354.2358 -64.9976 2.8950

Then, we can use the roots function to determine the solution,

>> roots(p)
ans =
0.1074
0.0761

Thus, the value of the specific volume corresponding to an entropy of 6.45 is 0.1074.

14.8 This problem is nicely suited for the Newton interpolating polynomial. First, we can order
the data so that the points are closest to and centered around the unknown,

T D
300 1.139

152
 350 0.967
400 0.854
250 1.367
450 0.759
200 1.708

Then we can generate the divided difference table,

T D first second third fourth fifth

300 1.139 -0.003440 1.18000E-05 4.00000E-09 -2.93333E-10 -2.77333E-12
350 0.967 -0.002260 1.16000E-05 -4.00000E-08 -1.60000E-11
400 0.854 -0.003420 7.60000E-06 -3.76000E-08
250 1.367 -0.003040 1.51200E-05
450 0.759 -0.003796
200 1.708

First-order (linear) fit:

f 1 (330) = 1.139 − 0.00344(330 − 300) = 1.0358

Thus, the linear estimate is 1.036 to the level of significant digits provided in the original
data.

Second-order (quadratic) fit:

f 2 (330) = 1.0358 + 1.18 × 10 −5 (330 − 300)(330 − 350) = 1.0287

The quadratic estimate is 1.029 to the level of significant digits provided in the original
data.

Third-order (cubic) fit:

f 3 (330) = 1.0287 + 4 × 10 −9 (330 − 300)(330 − 350)(330 − 400) = 1.028888

The cubic estimate is also 1.029.

Fourth-order (quartic) fit:

f 4 (330) = 1.0289 − 2.93333 −10 (330 − 300)(330 − 350)(330 − 400)(330 − 250) = 1.0279

The quartic estimate now seems to be diverging slightly by moving to a value of 1.028.
This may be an initial indication that the higher-order terms are beginning to induce slight
oscillations.

Fifth-order (quintic) fit:

f 2 (330) = 1.0279 − 2.77333 −12 (330 − 300)(330 − 350)(330 − 400)(330 − 250)(330 − 450) = 1.02902

153
Oscillations are now evidently occurring as the fifth-order estimate now jumps back up to
slightly above a value of 1.029.

On the basis of the foregoing, I would conclude that the cubic equation provides the best
approximation and that a value of 1.029 is a sound estimate to the level of significant digits
provided in the original data.

Inverse interpolation can be now used to determine the temperature corresponding to the
value of density of 1.029. First, MATLAB can be used to fit a cubic polynomial through
the four points that bracket this value. Interestingly, because of the large values of the
temperatures, we get an error message,

>> T = [250 300 350 400];

>> D =[1.3670 1.139 0.967 0.854];
>> p = polyfit(T,D,3)
Warning: Polynomial is badly conditioned. Remove repeated data points
or try centering and scaling as described in HELP POLYFIT.
(Type "warning off MATLAB:polyfit:RepeatedPointsOrRescale" to suppress
this warning.)
> In polyfit at 78

p =
0.0000 0.0000 -0.0097 3.2420

Let’s disregard this warn and proceed to adjust the polynomial so that it can be used to
solve the inverse interpolation problem. To do this, we subtract the specified value of the
density from the polynomial’s constant coefficient

>> p(4)=p(4)-1.029

p =
0.0000 0.0000 -0.0097 2.2130

Then we can use the roots function to determine the temperature that corresponds to this
value

>> roots(p)
ans =
1.0e+003 *
-2.8237
0.5938
0.3300

Thus, even though the polynomial is badly conditioned one of the roots corresponds to T =
330 as expected.

Now let’s perform the inverse interpolation, but with scaling. To do this, we will merely
subtract the value at the midpoint of the temperature range (325) from all the temperatures.
This acts to both reduce the magnitudes of the temperatures and centers them on zero,

>> format long

>> D = [1.3670 1.139 0.967 0.854];

154
 >> T = [250 300 350 400];
>> T = T - 325;

Then, the cubic fit can be generated with no error message,

>> p = polyfit(T,D,3)
p =
0.00000000400000 0.00001150000000 -0.00344250000000 1.04581250000000

We can set up the roots problem

>> p(4)=p(4)-1.029
p =
0.00000000400000 0.00001150000000 -0.00344250000000 0.01681250000000

We can then use the roots function to determine the temperature that corresponds to the
given density

>> r = roots(p)
ans =
1.0e+003 *
-3.14874694489127
0.26878060289231
0.00496634199799

By adding back the offset of 325, we arrive at the expected result of 330,

>> Tinv = r(3)+325

Tinv =
3.299663419979927e+002

14.9 A MATLAB session provides a handy way to solve this problem

>> i = [-1 -0.5 -0.25 0.25 0.5 1];

>> V = [-193 -41 -13.5625 13.5625 41 193];
>> p = polyfit(i,V,5)
p =
0.0000 -0.0000 148.0000 -0.0000 45.0000 0.0000

The interpolating polynomial is therefore

V = 148i 3 + 45i

The polyval function can be used to determine the interpolation at i = 0.1,

>> polyval(p,0.10)
ans =
4.6480

14.10 Third-order case: The MATLAB polyfit function can be used to generate the cubic
polynomial and perform the interpolation,

>> x = [1 1.5 2 2.5];

>> J = [0.765198 0.511828 0.223891 -0.048384];

155
 >> p = polyfit(x,J,3)
p =
0.0670 -0.3705 0.1014 0.9673
>> Jpred = polyval(p,1.82)
Jpred =
0.3284

The built-in function besselj can be used to determine the true value which can then be
used to determine the percent relative error

>> Jtrue = besselj(0,1.82)

Jtrue =
0.3284
>> ea = abs((Jtrue-Jpred)/Jtrue)*100
ea =
0.0043

Fourth-order case:

>> x = [1 1.5 2 2.5 3];

>> J = [0.765198 0.511828 0.223891 -0.048384 -0.260052];
>> p = polyfit(x,J,4)
p =
-0.0035 0.0916 -0.4330 0.1692 0.9409
>> Jpred = polyval(p,1.82)
Jpred =
0.3283
>> Jtrue = besselj(0,1.82);
>> ea = abs((Jtrue-Jpred)/Jtrue)*100
ea =
0.0302

Fifth-order case:

>> x = [1 1.5 2 2.5 3 0.5];

>> J = [0.765198 0.511828 0.223891 -0.048384 -0.260052 0.938470];
>> p = polyfit(x,J,5)
p =
-0.0027 0.0231 -0.0115 -0.2400 -0.0045 1.0008
>> Jpred = polyval(p,1.82)
Jpred =
0.3284
>> Jtrue = besselj(0,1.82);
>> ea = abs((Jtrue-Jpred)/Jtrue)*100
ea =
5.2461e-004

14.11 In the same fashion as Example 14.6, MATLAB can be used to evaluate each of the cases,

First order:

>> t = [1990 1980];

>> pop = [249.46 227.23];
>> ts = (t - 1955)/35;
>> p = polyfit(ts,pop,1);

156
>> polyval(p,(2000-1955)/35)
ans =
271.6900

Second order:

>> t = [t 1970];
>> pop = [pop 205.05];
>> ts = (t - 1955)/35;
>> p = polyfit(ts,pop,2);
>> polyval(p,(2000-1955)/35)
ans =
271.7400

Third order:

>> t = [t 1960];
>> pop = [pop 180.67];
>> ts = (t - 1955)/35;
>> p = polyfit(ts,pop,3);
>> polyval(p,(2000-1955)/35)
ans =
273.9900

Fourth order:

>> t = [t 1950];
>> pop = [pop 152.27];
>> ts = (t - 1955)/35;
>> p = polyfit(ts,pop,4);
>> polyval(p,(2000-1955)/35)
ans =
274.4200

Although the improvement is not great, the addition of each term causes the prediction for
2000 to increase. Thus, using higher-order approximations is moving the prediction closer
to the actual value of 281.42 that occurred in 2000.

157
CHAPTER 15
15.1 (a) The simultaneous equations for the natural spline can be set up as

⎡1 ⎤ ⎧ c1 ⎫ ⎧ 0 ⎫
⎢1 3 0.5 ⎥ ⎪c ⎪ ⎪ 0 ⎪
⎢ ⎥⎪ 2 ⎪ ⎪ ⎪
⎢ 0.5 2 0.5 ⎥ ⎪⎪c3 ⎪⎪ ⎪⎪ − 6 ⎪⎪
⎢ ⎥⎨ ⎬ = ⎨ ⎬
⎢ 0.5 3 1 ⎥ ⎪c 4 ⎪ ⎪− 24⎪
⎢ 1 4 1⎥ ⎪c5 ⎪ ⎪ 15 ⎪
⎢ ⎥⎪ ⎪ ⎪ ⎪
⎣⎢ 1⎦⎥ ⎪⎩c6 ⎪⎭ ⎩⎪ 0 ⎭⎪

These equations can be solved for the c’s and then Eqs. (15.21) and (15.18) can be used to
solve for the b’s and the d’s. The coefficients for the intervals can be summarized as

interval a b c d
1 1 3.970954 0 0.029046
2 5 4.058091 0.087137 -0.40664
3 7 3.840249 -0.52282 -6.31535
4 8 -1.41909 -9.99585 5.414938
5 2 -5.16598 6.248963 -2.08299

These can be used to generate the following plot of the natural spline:

10

0
0 2 4 6

(b) The not-a-knot spline and its plot can be generated with MATLAB as

>> x = [1 2 2.5 3 4 5];

>> y = [1 5 7 8 2 1];
>> xx = linspace(1,5);
>> yy = spline(x,y,xx);
>> plot(x,y,'o',xx,yy)

158
 Notice how the not-a-knot version exhibits much more curvature, particularly between the
last points.

(c) The piecewise cubic Hermite polynomial and its plot can be generated with MATLAB
as

>> x = [1 2 2.5 3 4 5];

>> y = [1 5 7 8 2 1];
>> xx = linspace(1,5);
>> yy = interp1(x,y,xx,'pchip');
>> plot(x,y,'o',xx,yy)

15.2 The simultaneous equations for the clamped spline with zero end slopes can be set up as

159
⎡ 1 0 .5 ⎤ ⎧ c1 ⎫ ⎧ 0 ⎫
⎢0.5 2 0.5 ⎥ ⎪c ⎪ ⎪ − 90 ⎪
⎢ ⎥⎪ 2 ⎪ ⎪ ⎪
⎢ 0.5 2 0.5 ⎥ ⎪c3 ⎪ ⎪− 108⎪
⎢ ⎥⎪ ⎪ ⎪ ⎪
⎢ 0.5 2 0.5 ⎥ ⎨c 4 ⎬ = ⎨ 144 ⎬
⎢ 0.5 2 0.5 ⎥ ⎪c5 ⎪ ⎪ 36 ⎪
⎢ ⎥⎪ ⎪ ⎪ ⎪
⎢ 0. 5 2 0 . 5 ⎪
⎥ 6c ⎪ ⎪ 18 ⎪
⎢ ⎥ ⎪ ⎪ ⎪ ⎪
⎣ 0.5 1 ⎦ ⎩c7 ⎭ ⎩ ⎭

These equations can be solved for the c’s and then Eqs. (15.21) and (15.18) can be used to
solve for the b’s and the d’s. The coefficients for the intervals can be summarized as

interval a b c d
1 70 0 15.87692 -31.7538
2 70 -7.93846 -31.7538 -24.7385
3 55 -58.2462 -68.8615 106.7077
4 22 -47.0769 91.2 -66.0923
5 13 -5.44615 -7.93846 13.66154
6 10 -3.13846 12.55385 -12.5538

The fit can be displayed in graphical form. Note that we are plotting the points as depth
versus temperature so that the graph depicts how the temperature changes down through the
tank.

0 50 100
0

Inspection of the plot indicates that the inflection point occurs in the 3rd interval. The cubic
equation for this interval is

T3 ( x) = 55 − 58.2462(d − 1) − 68.8615(d − 1) 2 + 106.7077( d − 1) 3

where T = temperature and d = depth. This equation can be differentiated twice to yield the
second derivative

d 2T3 ( x)
= −137.729 + 640.2462(d − 1)
dx 2

160
 This can be set equal to zero and solved for the depth of the thermocline as d = 1.21511 m.

15.3 (a) The not-a-knot fit can be set up in MATLAB as

>> x = linspace(0,1,11);
>> y = 1./((x-0.3).^2+0.01)+1./((x-0.9).^2+0.04)-6;
>> xx = linspace(0,1);
>> yy = spline(x,y,xx);
>> yh = 1./((xx-0.3).^2+0.01)+1./((xx-0.9).^2+0.04)-6;
>> plot(x,y,'o',xx,yy,xx,yh,'--')

(b) The piecewise cubic Hermite polynomial fit can be set up in MATLAB as

>> x = linspace(0,1,11);
>> y = 1./((x-0.3).^2+0.01)+1./((x-0.9).^2+0.04)-6;
>> xx = linspace(0,1);
>> yy = interp1(x,y,xx,'pchip');
>> yh = 1./((xx-0.3).^2+0.01)+1./((xx-0.9).^2+0.04)-6;
>> plot(x,y,'o',xx,yy,xx,yh,'--')

161
15.4 The simultaneous equations for the clamped spline with zero end slopes can be set up as

⎡ 1 ⎤ ⎧ c1 ⎫ ⎧ 0 ⎫
⎢100 400 100 ⎥ ⎪c ⎪ ⎪− 0.01946⎪
⎢ ⎥⎪ 2 ⎪ ⎪ ⎪
⎢ 100 600 200 ⎥ ⎪c3 ⎪ ⎪− 0.00923⎪
⎢ ⎥⎪ ⎪ ⎪ ⎪
⎢ 200 800 200 ⎥ ⎨c 4 ⎬ = ⎨− 0.00098⎬
⎢ 200 800 200 ⎥ ⎪c5 ⎪ ⎪ 0.001843 ⎪
⎢ ⎥⎪ ⎪ ⎪ ⎪
⎢ 200 800 200⎥ ⎪c6 ⎪ ⎪ 0.001489 ⎪
⎢ ⎪ ⎪ ⎪ ⎪
⎣ 1 ⎥⎦ ⎩c7 ⎭ ⎩ 0 ⎭

These equations can be solved for the c’s and then Eqs. (15.21) and (15.18) can be used to
solve for the b’s and the d’s. The coefficients for the intervals can be summarized as

interval a b c d
1 0 0.009801 0 -1.6E-07
2 0.824361 0.005128 -4.7E-05 1.3E-07
3 1 -0.00031 -7.7E-06 1.31E-08
4 0.735759 -0.0018 2.13E-07 2.82E-09
5 0.406006 -0.00138 1.9E-06 -8.7E-10
6 0.199148 -0.00072 1.39E-06 -2.3E-09

The fit can be displayed in graphical form as

1.2

0.8

0.6

0.4

0.2

0
0 500 1000

(b) The not-a-knot fit can be set up in MATLAB as

>> x = [0 100 200 400 600 800 1000];

>> y = x/200.*exp(-x/200+1);
>> xx = linspace(0,1000);
>> yc = xx/200.*exp(-xx/200+1);
>> yy = spline(x,y,xx);
>> plot(x,y,'o',xx,yy,xx,yc,'--')

162
 (c) The piecewise cubic Hermite polynomial fit can be set up in MATLAB as

>> x = [0 100 200 400 600 800 1000];

>> y = x/200.*exp(-x/200+1);
>> xx = linspace(0,1000);
>> yc = xx/200.*exp(-xx/200+1);
>> yy = interp1(x,y,xx,'pchip');
>> plot(x,y,'o',xx,yy,xx,yc,'--')

Summary: For this case, the not-a-knot fit is the best.

15.5 (a) The not-a-knot fit can be set up in MATLAB as

>> x = [-1 -0.6 -0.2 0.2 0.6 1];

>> y = [0 0 0 1 1 1];
>> xx = linspace(-1,1);
>> yy = spline(x,y,xx);
>> plot(x,y,'o',xx,yy)

163
(b) The clamped spline with zero end slopes can be set up in MATLAB as

>> x = [-1 -0.6 -0.2 0.2 0.6 1];

>> y = [0 0 0 1 1 1];
>> ys = [0 y 0];
>> xx = linspace(-1,1);
>> yy = spline(x,ys,xx);
>> plot(x,y,'o',xx,yy)

(c) The piecewise cubic Hermite polynomial fit can be set up in MATLAB as

>> x = [-1 -0.6 -0.2 0.2 0.6 1];

>> y = [0 0 0 1 1 1];
>> xx = linspace(-1,1);
>> yy = interp1(x,y,xx,'pchip');
>> plot(x,y,'o',xx,yy)

164
15.6 An M-file function to implement the natural spline can be written as

function yy = natspline(x,y,xx)
% natspline(x,y,xx):
% uses a natural cubic spline interpolation to find yy, the values
% of the underlying function y at the points in the vector xx.
% The vector x specifies the points at which the data y is given.

n = length(x);
m = length(xx);
aa(1,1) = 1; aa(n,n) = 1;
bb(1) = 0; bb(n) = 0;
for i = 2:n-1
aa(i,i-1) = h(x, i - 1);
aa(i,i) = 2 * (h(x, i - 1) + h(x, i));
aa(i,i+1) = h(x, i);
bb(i) = 3 * (fd(i + 1, i, x, y) - fd(i, i - 1, x, y));
end
c = aa\bb';
for i = 1:n - 1
a(i) = y(i);
b(i) = fd(i + 1, i, x, y) - h(x, i) / 3 * (2 * c(i) + c(i + 1));
d(i) = (c(i + 1) - c(i)) / 3 / h(x, i);
end
for i = 1:m
yy(i) = SplineInterp(x, n, a, b, c, d, xx(i));
end

function hh = h(x, i)
hh = x(i + 1) - x(i);

function fdd = fd(i, j, x, y)

fdd = (y(i) - y(j)) / (x(i) - x(j));

function yyy = SplineInterp(x, n, a, b, c, d, xi)

for ii = 1:n - 1
if xi >= x(ii) - 0.000001 & xi <= x(ii + 1) + 0.000001
yyy=a(ii)+b(ii)*(xi-x(ii))+c(ii)*(xi-x(ii))^2+d(ii)*(xi-x(ii))^3;
break
end
end

165
 The program can be used to duplicate Example 15.3:

>> x = [3 4.5 7 9];

>> y = [2.5 1 2.5 .5];
>> xx = linspace(3,9);
>> yy = natspline(x,y,xx);
>> plot(x,y,'o',xx,yy)

15.7 (a) The not-a-knot fit can be set up in MATLAB as

>> x = [1 3 5 6 7 9];
>> y = 0.0185*x.^5-0.444*x.^4+3.9125*x.^3-15.456*x.^2+27.069*x-14.1;
>> xx = linspace(1,9);
>> yy = spline(x,y,xx);
>> yc = 0.0185*xx.^5-0.444*xx.^4+3.9125*xx.^3-15.456*xx.^2+27.069*xx-14.1;
>> plot(x,y,'o',xx,yy,xx,yc,'--')

(b) The function can be differentiated to give

f ' ( x) = 0.0925 x 4 − 1.776 x 3 + 11.7375 x 2 − 30.912 x + 27.069

This function can be evaluated at the end nodes to give f'(1) = 6.211 and f'(9) = 11.787.
These values can then be added to the y vector and the spline function invoked to develop
the clamped fit:

166
>> yd = [6.211 y 11.787];
>> yy = spline(x,yd,xx);
>> plot(x,y,'o',xx,yy,xx,yc,'--')

167
CHAPTER 16
16.1 A table of integrals can be consulted to determine

1
∫ tanh dx = a ln cosh ax
Therefore,

t
t gm ⎛ gc d ⎞ gm m ⎡ ⎛ gc d ⎞⎤
∫ 0 cd
tanh ⎜
⎜ m ⎟
⎝
t ⎟ dt =
⎠ cd
⎢ln cosh⎜
gc d ⎣⎢ ⎜ m ⎟⎥
⎝
t ⎟⎥
⎠⎦ 0

gm 2 ⎡ ⎛ gc d ⎞ ⎤
⎢ln cosh⎜ t ⎟ − ln cosh(0)⎥
gc d2 ⎢⎣ ⎜ m ⎟ ⎥⎦
⎝ ⎠

Since cosh(0) = 1 and ln(1) = 0, this reduces to

m ⎛ gc d ⎞
ln cosh⎜ t⎟
cd ⎜ m ⎟
⎝ ⎠

16.2 (a) The analytical solution can be evaluated as

∫
4

0
[
(1 − e − 2 x ) dx = x + 0.5e − 2 x ]
4
0 = 4 + 0.5e − 2 ( 4 ) − 0 − 0.5e − 2 ( 0 ) = 3.500167731

(b) single application of the trapezoidal rule

0 + 0.999665
(4 − 0) = 1.99329 (ε t = 42.88%)
2

(c) composite trapezoidal rule

n = 2:

0 + 2(0.981684) + 0.999665
(4 − 0) = 2.96303 (ε t = 15.35%)
4

n = 4:

0 + 2(0.86466 + 0.981684 + 0.99752) + 0.999665

(4 − 0) = 3.3437 (ε t = 4.47%)
8

(d) single application of Simpson’s 1/3 rule

168
 0 + 4(0.981684) + 0.999665
(4 − 0) = 3.28427 (ε t = 6.17%)
6

(e) composite Simpson’s 1/3 rule (n = 4)

0 + 4(0.86466 + 0.99752) + 2(0.981684) + 0.999665

(4 − 0) = 3.47059 (ε t = 0.84%)
12

(f) Simpson’s 3/8 rule.

0 + 3(0.930517 + 0.995172) + 0.999665

( 4 − 0) = 3.388365 (ε t = 3.19%)
8

16.3 (a) The analytical solution can be evaluated as

π /2
∫ (6 + 3 cos x) dx = [6 x + 3 sin x ]0
π /2
= 6(π / 2) + 3 sin(π / 2) − 6(0) − 3 sin(0) = 12.424778
0

(b) single application of the trapezoidal rule

⎛π ⎞9+6
⎜ − 0⎟ = 11.78097 (ε t = 5.18%)
⎝2 ⎠ 2

(c) composite trapezoidal rule

n = 2:

⎛π ⎞ 9 + 2(8.12132) + 6
⎜ − 0⎟ = 12.26896 (ε t = 1.25%)
⎝2 ⎠ 4

n = 4:

⎛π ⎞ 9 + 2(8.77164 + 8.12132 + 7.14805) + 6

⎜ − 0⎟ = 12.386125 (ε t = 0.3111%)
⎝2 ⎠ 8

(d) single application of Simpson’s 1/3 rule

⎛π ⎞ 9 + 4(8.12132) + 6
⎜ − 0⎟ = 12.4316 (ε t = 0.0550%)
⎝2 ⎠ 6

(e) composite Simpson’s 1/3 rule (n = 4)

⎛π ⎞ 9 + 4(8.7716 + 7.14805) + 2(8.12132) + 6

⎜ − 0⎟ = 12.42518 (ε t = 0.0032%)
⎝2 ⎠ 12

(f) Simpson’s 3/8 rule.

169
 ⎛π ⎞ 9 + 3(8.59808 + 7.5) + 6
⎜ − 0⎟ = 12.42779 (ε t = 0.0243%)
⎝2 ⎠ 8

16.4 (a) The analytical solution can be evaluated as

4
4 ⎡ x2 x6 ⎤
∫ (1 − x − 4 x + 2 x ) dx = ⎢ x − − x4 + ⎥
3 5
−2
⎣ 2 3 ⎦ −2

42 46 (−2) 2 ( −2) 6
=4− − 44 + − (−2) + + (−2) 4 − = 1104
2 3 2 3

(b) single application of the trapezoidal rule

− 29 + 1789
(4 − (−2)) = 5280 (ε t = 378.3%)
2

(c) composite trapezoidal rule

n = 2:

− 29 + 2(−2) + 1789
(4 − (−2)) = 2634 (ε t = 138.6%)
4

n = 4:

− 29 + 2(1.9375 + (−2) + 131.3125) + 1789

(4 − (−2)) = 1516.875 (ε t = 37.4%)
8

(d) single application of Simpson’s 1/3 rule

− 29 + 4(−2) + 1789
(4 − (−2)) = 1752 (ε t = 58.7%)
6

(e) composite Simpson’s 1/3 rule (n = 4)

− 29 + 4(1.9375 + 131.3125) + 2( −2) + 1789

(4 − (−2)) = 1144.5 (ε t = 3.6685%)
12

(f) Simpson’s 3/8 rule.

− 29 + 3(1 + 31) + 1789

(4 − (−2)) = 1392 (ε t = 26.09%)
8

(g) Boole’s rule.

7(−29) + 32(1.9375) + 12(−2) + 32(131.3125) + 7(1789)

(4 − (−2)) = 1104 (ε t = 0%)
90

170
16.5 (a) The analytical solution can be evaluated as

∫
1.2

0
[
e − x dx = − e − x ]
1.2
0 = −e −1.2 − (−e 0 ) = 0.69880579

(b) Trapezoidal rule

1 + 0.90484 0.90484 + 0.74082 0.74082 + 0.60653

(0.1 − 0) + (0.3 − 0.1) + (0.5 − 0.3)
2 2 2
0.60653 + 0.49659 0.49659 + 0.38674 0.38674 + 0.30119
+ (0.7 − 0.5) + (0.95 − 0.7) + (1.2 − 0.957)
2 2 2
= 0.09524 + 0.164566 + 0.134735 + 0.110312 + 0.110416 + 0.085992 = 0.70126 (ε t = 0.35%)

(c) Trapezoidal and Simpson’s Rules

1 + 0.90484 0.90484 + 3(0.74082 + 0.60653) + 0.49659

(0.1 − 0) + (0.7 − 0.1)
2 8
0.49659 + 4(0.38674) + 0.30119
+ (1.2 − 0.7) = 0.09524 + 0.40826 + 0.195395 = 0.698897 (ε t = 0.0131%)
6
16.6 (a) The integral can be evaluated analytically as,

4
⎡ x3
2
3 x ⎤
2

∫ − +
2
⎢
−2 3
3 y x y ⎥ dy
⎣ 2 ⎦0

( 4) 3
2 ( 4) 2
∫ −2 3
− 3 y 2 ( 4) + y 3
2
dy

2
∫ −2
21.33333 − 12 y 2 + 8 y 3 dy

[21.33333 y − 4 y 3
+ 2y4 ]
2
-2

21.33333( 2) − 4(2) 3 + 2(2) 4 − 21.33333( −2) + 4(−2) 3 − 2(−2) 4 = 21.33333

(b) The composite trapezoidal rule with n = 2 can be used the evaluate the inner integral at
the three equispaced values of y,

− 12 + 2(−24) − 28
y = −2: (4 − 0) = −88
4

0 + 2(4) + 16
y = 0: (4 − 0) = 24
4

171
 − 12 + 2(8) + 36
y = 2: (4 − 0) = 40
4

These results can then be integrated in y to yield

− 88 + 2(24) + 40
(2 − (−2)) =0
4

which represents a percent relative error of

21.33333 − 0
εt = × 100% = 100%
21.33333

which is not very good.

(c) Single applications of Simpson’s 1/3 rule can be used the evaluate the inner integral at
the three equispaced values of y,

− 12 + 4(−24) − 28
y = −2: ( 4 − 0) = −90.66667
6

0 + 4(4) + 16
y = 0: (4 − 0) = 21.33333
6

− 12 + 4(8) + 36
y = 2: (4 − 0) = 37.33333
6

These results can then be integrated in y to yield

− 90.66667 + 4(21.33333) + 37.33333

(2 − (−2)) = 21.33333
6

which represents a percent relative error of

21.33333 − 21.33333
εt = × 100% = 0%
21.33333

which is perfect

16.7 (a) The integral can be evaluated analytically as,

3
4 6 ⎡ x4 ⎤ 4 6
∫ ∫
−4 0
⎢ − 2 yzx ⎥ dy dz =
⎣ 4 ⎦ −1
∫ ∫
−4 0
20 − 8 yz dy dz

∫ [20 y − 4 zy ]
4 6 4 4
∫ ∫ ∫
2 6
20 − 8 yz dy dz = 0 dz = 120 − 144 z dz
−4 0 −4 −4

172
∫
4
[
120 − 144 z dz = 120 z − 72 z 2
−4
]
4
−4 = 120(4) − 72(4) 2 − 120(−4) + 72(−4) 2 = 960

(b) Single applications of Simpson’s 1/3 rule can be used the evaluate the inner integral at
the three equispaced values of y for each value of z,

z = −4:

− 1 + 4(1) + 27
y = 0: (3 − (−1)) = 20
6

23 + 4(25) + 51
y = 3: (3 − (−1)) = 116
6

47 + 4(49) + 75
y = 6: (3 − (−1)) = 212
6

These results can then be integrated in y to yield

20 + 4(116) + 212
( 6 − 0) = 696
6

z = 0:

− 1 + 4(1) + 27
y = 0: (3 − (−1)) = 20
6

− 1 + 4(1) + 27
y = 3: (3 − (−1)) = 20
6

− 1 + 4(1) + 27
y = 6: (3 − (−1)) = 20
6

These results can then be integrated in y to yield

20 + 4( 20) + 20
( 6 − 0) = 120
6

z = 4:

− 1 + 4(1) + 27
y = 0: (3 − (−1)) = 20
6

− 25 + 4(−23) + 3
y = 3: (3 − (−1)) = −76
6

173
 − 49 + 4(−47) − 21
y = 6: (3 − (−1)) = −172
6

These results can then be integrated in y to yield

20 + 4(−76) − 172
( 6 − 0) = −456
6

The results of the integrations in y can then be integrated in z to yield

696 + 4(120) − 456

( 6 − 0) = 960
6

which represents a percent relative error of

960 − 960
εt = × 100% = 0%
960

16.8 (a) The trapezoidal rule can be implemented as,

5+6 6 + 5.5 5.5 + 7 7 + 8 .5

d = (2 − 1) + (3.25 − 2) + (4.5 − 3.25) + (6 − 4.5)
2 2 2 2
8 .5 + 6 6+6 6+7 7+7 7+5
+ (7 − 6) + (8 − 7) + (8.5 − 8) + (9.3 − 8.5) + (10 − 9.3) = 58.425
2 2 2 2 2

(b) The polynomial can be fit as,

>> format long

>> t = [1 2 3.25 4.5 6 7 8 8.5 9.3 10];
>> v = [5 6 5.5 7 8.5 6 6 7 7 5];
>> p = polyfit(t,v,3)

p =
-0.00657842294444 0.01874733808337 0.56859435273356
4.46645555949356

The cubic can be plotted along with the data,

>> tt = linspace(1,10);
>> vv = polyval(p,tt);
>> plot(tt,vv,t,v,'o')

174
 The cubic can then be integrated to estimate the distance traveled,

10
d= ∫
1
− 0.006578t 3 + 0.018747t 2 + 0.568594t + 4.46646 dt

[ ]10
= − 0.001645t 4 + 0.006249t 3 + 0.284297t 2 + 4.46646t 1 = 58.14199

16.9

z w(z) ρgw(z)(60 – z) ρgzw(z)(60 – z)

60 200 0 0
50 190 1.8639E+07 9.3195E+08
40 175 3.4335E+07 1.3734E+09
30 160 4.7088E+07 1.4126E+09
20 135 5.2974E+07 1.0595E+09
10 130 6.3765E+07 6.3765E+08
0 122 7.1809E+07 0

7.1809 + 4(6.3765 + 4.7088 + 1.8639) + 2(5.2974 + 3.4335) + 0

f t = 60 × 10 7 = 2.5480 × 10 9
3(6)

D 0 + 4(0.63765 + 1.4126 + 0.93195) + 2(1.0595 + 1.3734) + 0

∫ 0
ρgzw( z )( D − z ) dz = 60
3(6)
× 10 9 = 5.5982 × 1010

5.5982 × 1010
d= = 21.971
2.5480 × 10 9

16.10 (a) Trapezoidal rule:

0 + 2(54.937 + 51.129 + 36.069 + 27.982 + 19.455) + 13.311

f = 30 = 996.1363
2(6)

175
 0 + 2( 274.684 + 511.292 + 586.033 + 559.631 + 486.385) + 399.332
30
2(6) 13088.45
f = = = 13.139 m
996.1363 996.1363

(b) Simpson’s 1/3 rule:

0 + 4(54.937 + 36.069 + 19.455) + 2(51.129 + 27.982) + 13.311

f = 30 = 1042.294
3(6)

0 + 4( 274.684 + 586.033 + 486.385) + 2(511.292 + 559.631) + 399.332

30
3(6) 13215.97
f = = = 12.6797 m
1042.294 1042.294

176
CHAPTER 17
17.1 The integral can be evaluated analytically as,

2
2 ⎛ 3⎞ 2
I= ∫ 1
⎜ 2 x + ⎟ dx =
⎝ x⎠ ∫1
4 x 2 + 12 + 9 x − 2 dx

2
⎡ 4x 3 9⎤ 4( 2) 3 9 4(1) 3 9
I =⎢ + 12 x − ⎥ = + 12 ( 2) − − − 12(1) + = 25.8333
⎣ 3 x ⎦1 3 2 3 1

The tableau depicting the implementation of Romberg integration to εs = 0.5% is

iteration → 1 2 3
εt → 6.9355% 0.1613% 0.0048%
εa → 1.6908% 0.0098%
1 27.62500000 25.87500000 25.83456463
2 26.31250000 25.83709184
4 25.95594388

Thus, the result is 25.83456.

17.2 (a) The integral can be evaluated analytically as,

[ ]
I = − 0.01094 x 5 + 0.21615 x 4 − 1.3854 x 3 + 3.14585 x 2 + 2 x 0 = 34.87808
8

(b) The tableau depicting the implementation of Romberg integration to εs = 0.5% is

iteration → 1 2 3 4
εt → 20.1699% 42.8256% 0.0000% 0.0000%
εa → 9.9064% 2.6766% 0.000000%
1 27.84320000 19.94133333 34.87808000 34.87808000
2 21.91680000 33.94453333 34.87808000
4 30.93760000 34.81973333
8 33.84920000

Thus, the result is exact.

(c) The transformations can be computed as

(8 + 0) + (8 − 0) x d 8−0
x= = 4 + 4 xd dx = dx d = 4dx d
2 2

These can be substituted to yield

∫ [− 0.0547(4 + 4 x ]
1
I= d ) 4 + 0.8646(4 + 4 x d ) 3 − 4.1562(4 + 4 x d ) 2 + 6.2917(4 + 4 x d ) + 2 4dx d
−1

177
 The transformed function can be evaluated using the values from Table 17.1

I = 0.5555556 f (−0.774596669) + 0.8888889 f (0) + 0.5555556 f (0.774596669) = 34.87808

which is exact.

(d)
>> format long
>> y = inline('-0.0547*x.^4+0.8646*x.^3-4.1562*x.^2+6.2917*x+2');
>> I = quad(y,0,8)

I =
34.87808000000000

17.3 Although it’s not required, the analytical solution can be evaluated simply as

I= ∫
3

0
[ ]3
xe x dx = e x ( x − 1) 0 = 41.17107385

(a) The tableau depicting the implementation of Romberg integration to εs = 0.5% is

iteration → 1 2 3
εt → 119.5350% 5.8349% 0.1020%
εa → 26.8579% 0.3579%
1 90.38491615 43.57337260 41.21305531
2 55.27625849 41.36057514
4 44.83949598

which represents a percent relative error of 0.102 %.

(b) The transformations can be computed as

(3 + 0) + (3 − 0) x d 3−0
x= = 1.5 + 1.5 x d dx = dx d = 1.5dx d
2 2

These can be substituted to yield

∫ [(1.5 + 1.5x ]
1
I= d )e1.5+1.5 xd 1.5dx d
−1

The transformed function can be evaluated using the values from Table 17.1

I = f (−0.577350269) + f (0.577350269) = 39.6075058

which represents a percent relative error of 3.8 %.

(c) Using MATLAB

>> format long

>> I = quad(inline('x.*exp(x)'),0,3)

178
 I =
41.17107385090233

which represents a percent relative error of 1.1×10−8 %.

>> I = quadl(inline('x.*exp(x)'),0,3)

I =
41.17107466800178

which represents a percent relative error of 2×10−6 %.

17.4 The exact solution can be evaluated simply as

>> format long

>> erf(1.5)

ans =
0.96610514647531

(a) The transformations can be computed as

(1.5 + 0) + (1.5 − 0) x d 1.5 − 0

x= = 0.75 + 0.75 x d dx = dx d = 0.75dx d
2 2

These can be substituted to yield

I=
2
π ∫
1

−1
[e − ( 0.75+ 0.75 xd ) 2
]0.75dx d

The transformed function can be evaluated using the values from Table 17.1

I = f (−0.577350269) + f (0.577350269) = 0.974173129

which represents a percent relative error of 0.835 %.

(b) The transformed function can be evaluated using the values from Table 17.1

I = 0.5555556 f (−0.774596669) + 0.8888889 f (0) + 0.5555556 f (0.774596669) = 0.965502083

which represents a percent relative error of 0.062 %.

17.5 (a) The tableau depicting the implementation of Romberg integration to εs = 0.5% is

iteration → 1 2 3 4
εa → 19.1131% 1.0922% 0.035826%
1 199.66621287 847.93212300 1027.49455856 1051.60670352
2 685.86564547 1016.27190634 1051.22995126
4 933.67034112 1049.04507345
8 1020.20139037

179
 Note that if 8 iterations are implemented, the method converges on a value of
1053.38523686. This result is also obtained if you use the composite Simpson’s 1/3 rule
with 1024 segments.

(b) The transformations can be computed as

(30 + 0) + (30 − 0) x d 30 − 0
x= = 15 + 15 x d dx = dx d = 15dx d
2 2

These can be substituted to yield

1 ⎡ 15 + 15 x d − 2.5(15+15 xd ) / 30 ⎤
I = 200 ∫ ⎢
−1 22 + 15 x
⎣ d
e ⎥ 15dx d
⎦

The transformed function can be evaluated using the values from Table 17.1

I = f (−0.577350269) + f (0.577350269) = 1162.93396

(c) Interestingly, the quad function encounters a problem and exceeds the maximum
number of iterations

>> format long

>> I = quad(inline('200*x/(7+x)*exp(-2.5*x/30)'),0,30)
Warning: Maximum function count exceeded; singularity likely.
(Type "warning off MATLAB:quad:MaxFcnCount" to suppress this
warning.)
> In quad at 88

I =
1.085280043451920e+003

The quadl function converges rapidly, but does not yield a very accurate result:

>> I = quadl(inline('200*x/(7+x)*exp(-2.5*x/30)'),0,30)

I =
1.055900924411335e+003

17.6 The integral to be evaluated is

∫ (10e )
1/ 2 2
−t
I= sin 2πt dt
0

(a) The tableau depicting the implementation of Romberg integration to εs = 0.1% is

iteration → 1 2 3 4
εa → 25.0000% 2.0824% 0.025340%
1 0.00000000 20.21768866 15.16502516 15.41501768
2 15.16326649 15.48081663 15.41111155
4 15.40142910 15.41546811

180
 8 15.41195836

(b) The transformations can be computed as

(0.5 + 0) + (0.5 − 0) x d 0.5 − 0

x= = 0.25 + 0.25 x d dx = dx d = 0.25dx d
2 2

These can be substituted to yield

∫ [10e ]
1 2
− ( 0.25 + 0.25 xd )
I= sin 2π (0.25 + 0.25 x d ) 0.25dx d
−1

For the two-point application, the transformed function can be evaluated using the values
from Table 17.1

I = f (−0.577350269) + f (0.577350269) = 7.684096 + 4.313728 = 11.99782

For the three-point application, the transformed function can be evaluated using the values
from Table 17.1

I = 0.5555556 f (−0.774596669) + 0.8888889 f (0) + 0.5555556 f (0.774596669)

= 0.5555556(1.237449) + 0.8888889(15.16327) + 0.5555556(2.684915) = 15.65755

(c)
>> format long
>> I = quad(inline('(10*exp(-x).*sin(2*pi*x)).^2'),0,0.5)

I =
15.41260804934509

17.7 The integral to be evaluated is

1/ 7
0.75 ⎛ r ⎞
I= ∫ 0
10⎜1 − ⎟
⎝ 0.75 ⎠
2πr dr

(a) The tableau depicting the implementation of Romberg integration to εs = 0.1% is

iteration → 1 2 3 4
εa → 25.0000% 1.0725% 0.098313%
1 0.00000000 10.67030554 12.88063803 13.74550712
2 8.00272915 12.74249225 13.73199355
4 11.55755148 13.67014971
8 13.14200015

(b) The transformations can be computed as

(0.75 + 0) + (0.75 − 0) x d 0.75 − 0

x= = 0.375 + 0.375 x d dx = dx d = 0.375dx d
2 2

181
 These can be substituted to yield

2
⎡ ⎛ 0.375 + 0.375 x ⎞1 / 7
1 ⎤
I= ∫
⎢10⎜1 − ⎟ 2π (0.375 + 0.375 x d )⎥ 0.375dx d
d
−1
⎢⎣ ⎝ 0.75 ⎠ ⎥⎦

For the two-point application, the transformed function can be evaluated using the values
from Table 17.1

I = f (−0.577350269) + f (0.577350269) = 14.77171

(c)
>> format long
>> I = quad(inline('10*(1-x/0.75).^(1/7)*2*pi.*x'),0,0.75)

I =
14.43168560836254

17.8 The integral to be evaluated is

8
I= ∫2
(9 + 4 cos 2 0.4t )(5e −0.5t + 2e 0.15t ) dt

(a) The tableau depicting the implementation of Romberg integration to εs = 0.1% is

iteration → 1 2 3 4
εa → 7.4179% 0.1054% 0.001212%
1 411.26095167 317.15529472 322.59571622 322.34570788
2 340.68170896 322.25568988 322.34961426
4 326.86219465 322.34374398
8 323.47335665

(b)
>> format long
>> y = inline('(9+4*cos(0.4*x).^2).*(5*exp(-0.5*x)+2*exp(0.15*x))')
>> I = quadl(y,2,8)

I =
3.223483672542467e+002

17.9 (a) The integral can be evaluated analytically as,

4
2⎡ x3 x2 ⎤
∫ ⎢
−2 3
⎣
− 3y 2 x + y 3 ⎥ dy
2 ⎦0

2( 4) 3 ( 4) 2
∫ −2 3
− 3 y 2 ( 4) + y 3
2
dy

2
∫ −2
21.33333 − 12 y 2 + 8 y 3 dy

182
[21.33333 y − 4 y 3
+ 2y4 ]
2
-2

21.33333( 2) − 4(2) 3 + 2(2) 4 − 21.33333( −2) + 4(−2) 3 − 2(−2) 4 = 21.33333

(b) The operation of the dblquad function can be understood by invoking help,

>> help dblquad

A session to use the function to perform the double integral can be implemented as,

>> dblquad(inline('x.^2-3*y.^2+x*y.^3'),0,4,-2,2)

ans =
21.3333

183
CHAPTER 18
18.1 (a) The analytical solution can be derived by the separation of variables,

dy
∫ y ∫
= t 3 − 1.5 dt

The integrals can be evaluated to give,

t4
ln y = − 1.5t + C
4

Substituting the initial conditions yields C = 0. Substituting this value and taking the
exponential gives
4
/ 4 −1.5t
y = et

(b) Euler method (h = 0.5):

t y dy/dt
0 1 -1.5
0.5 0.25 -0.34375
1 0.078125 -0.03906
1.5 0.058594 0.109863
2 0.113525

Euler method (h = 0.25):

t y dy/dt
0 1 -1.5
0.25 0.625 -0.92773
0.5 0.393066 -0.54047
0.75 0.25795 -0.2781
1 0.188424 -0.09421
1.25 0.164871 0.074707
1.5 0.183548 0.344153
1.75 0.269586 1.040434
2 0.529695

(c) Midpoint method (h = 0.5)

t y dy/dt tm ym dym/dt
0 1 -1.5 0.25 0.625 -0.92773
0.5 0.536133 -0.73718 0.75 0.351837 -0.37932
1 0.346471 -0.17324 1.25 0.303162 0.13737
1.5 0.415156 0.778417 1.75 0.60976 2.353292
2 1.591802

184
 (d) RK4 (h = 0.5)

t y k1 tm ym k2 tm ym k3 te ye k4 φ
0 1.0000 -1.5000 0.25 0.6250 -0.9277 0.25 0.7681 -1.1401 0.5 0.4300 -0.5912 -1.0378
0.5 0.4811 -0.6615 0.75 0.3157 -0.3404 0.75 0.3960 -0.4269 1 0.2676 -0.1338 -0.3883
1 0.2869 -0.1435 1.25 0.2511 0.1138 1.25 0.3154 0.1429 1.5 0.3584 0.6720 0.1736
1.5 0.3738 0.7008 1.75 0.5489 2.1186 1.75 0.9034 3.4866 2 2.1170 13.7607 4.2786
2 2.5131

All the solutions can be presented graphically as

0
0 0.5 1 1.5 2
Euler (h=0.5) Euler (h=0.25) M idpoint
Analytical RK4

18.2 (a) The analytical solution can be derived by the separation of variables,

dy
∫ y ∫
= 1 + 2 x dx

The integrals can be evaluated to give,

2 y = x + x2 + C

Substituting the initial conditions yields C = 2. Substituting this value and rearranging gives

2
⎛ x2 + x + 2 ⎞
y = ⎜⎜ ⎟
⎟
⎝ 2 ⎠

Some selected value can be computed as

x y
0 1
0.25 1.336914
0.5 1.890625
0.75 2.743164

185
 1 4

(b) Euler’s method:

y (0.25) = y (0) + f (0,1) h

f (0,1) = (1 + 2(0)) 1 = 1

y (0.25) = 1 + 1(0.25) = 1.25

y (0.5) = y (0.25) + f (0.25,1.25)0.25

f (0.25,1.25) = (1 + 2(0.25)) 1.25 = 1.67705

y (0.5) = 1.25 + 1.67705(0.25) = 1.66926

The remaining steps can be implemented and summarized as

x y dy/dx
0 1 1
0.25 1.25 1.67705
0.5 1.66926 2.584
0.75 2.31526 3.804
1 3.26626 5.42184

(c) Heun’s method:

Predictor:

k1 = (1 + 2(0)) 1 = 1

y (0.25) = 1 + 1(0.25) = 1.25

k 2 = (1 + 2(0.25)) 1.25 = 1.6771

Corrector:

1 + 1.6771
y (0.25) = 1 + 0.25 = 1.33463
2

The remaining steps can be implemented and summarized as

x y k1 xe ye k2 dy/dx
0 1 1.0000 0.25 1.25 1.6771 1.3385
0.25 1.33463 1.7329 0.5 1.76785 2.6592 2.1961
0.5 1.88364 2.7449 0.75 2.56987 4.0077 3.3763
0.75 2.72772 4.1290 1 3.75996 5.8172 4.9731
1 3.97099

186
 (d) Ralston’s method:

Predictor:

k1 = (1 + 2(0)) 1 = 1

y (0.1875) = 1 + 1(0.1875) = 1.1875

k 2 = (1 + 2(0.1875)) 1.1875 = 1.49837

Corrector:

1 + 2(1.49837)
y (0.25) = 1 + 0.25 = 1.33306
3

The remaining steps can be implemented and summarized as

x y k1 x + 3/4h y + (3/4)k1h k2 dy/dx

0 1 1 0.1875 1.1875 1.49837 1.3322
0.25 1.33306 1.73187 0.4375 1.65779 2.41416 2.1867
0.5 1.87974 2.74208 0.6875 2.39388 3.67464 3.3638
0.75 2.72069 4.12363 0.9375 3.49387 5.37392 4.9572
1 3.95998

(e) RK4

x y k1 xm ym k2 xm ym k3 xe ye k4 φ
0 1.0000 1 0.125 1.1250 1.32583 0.125 1.1657 1.34961 0.25 1.3374 1.73469 1.3476
0.25 1.3369 1.73436 0.375 1.5537 2.18133 0.375 1.6096 2.2202 0.5 1.8919 2.75096 2.2147
0.5 1.8906 2.74997 0.625 2.2343 3.36322 0.625 2.3110 3.42043 0.75 2.7457 4.14253 3.4100
0.75 2.7431 4.14056 0.875 3.2606 4.96574 0.875 3.3638 5.04368 1 4.0040 6.00299 5.0271
1 3.9998

0
0 0.2 0.4 0.6 0.8 1
Euler Heun Analytical

RK4 Ralston

187
18.3 (a) Heun’s method:

Predictor:

k1 = −2(1) + (0) 2 = −2

y (0.5) = 1 + (−2)(0.5) = 0

k 2 = −2(0) + 0.5 2 = 0.25

Corrector:

− 2 + 0.25
y (0.5) = 1 + 0.5 = 0.5625
2

The remaining steps can be implemented and summarized as

t y k1 xi+1 yi+1 k2 dy/dt

0 1 -2.0000 0.5 0 0.2500 -0.875
0.5 0.5625 -0.8750 1 0.125 0.7500 -0.0625
1 0.53125 -0.0625 1.5 0.5 1.2500 0.59375
1.5 0.82813 0.5938 2 1.125 1.7500 1.17188
2 1.41406 1.1719

(b) As in Part (a), the corrector can be represented as

− 2 + (−2(0) + 0.5 2 )
y i1+1 = 1 + 0.5 = 0.5625
2

The corrector can then be iterated to give

− 2 + (−2(0.5625) + 0.5 2 )
y i2+1 = 1 + 0.5 = 0.28125
2

− 2 + (−2(0.28125) + 0.5 2 )
y i3+1 = 1 + 0.5 = 0.421875
2

The iterations can be continued until the percent relative error falls below 0.1%. This
occurs after 12 iterations with the result that y(0.5) = 0.37491 with εa = 0.073%. The
remaining values can be computed in a like fashion to give

t y
0 1.0000000
0.5 0.3749084
1 0.3334045
1.5 0.6526523

188
 2 1.2594796

(c) Midpoint method

k1 = −2(1) + (0) 2 = −2

y (0.25) = 1 + (−2)(0.25) = 0.5

k 2 = −2(0.5) + 0.25 2 = −0.9375

y (0.5) = 1 + ( −0.9375)0.5 = 0.53125

The remainder of the computations can be implemented in a similar fashion as listed below:

t y dy/dt tm ym dym/dt
0 1 -2.0000 0.25 0.5 -0.9375
0.5 0.53125 -0.8125 0.75 0.328125 -0.0938
1 0.48438 0.0313 1.25 0.492188 0.57813
1.5 0.77344 0.7031 1.75 0.949219 1.16406
2 1.35547

(d) Ralston’s method:

k1 = −2(1) + (0) 2 = −2

y (0.375) = 1 + (−2)(0.375) = 0.25

k 2 = −2(0.25) + 0.375 2 = −0.3594

− 2 + 2( −0.3594)
y (0.25) = 1 + 0.5 = 0.54688
3

The remaining steps can be implemented and summarized as

t y k1 t + 3/4h y + (3/4)k1h k2 dy/dt

0 1 -2.0000 0.375 0.25 -0.3594 -0.9063
0.5 0.54688 -0.8438 0.875 0.230469 0.3047 -0.0781
1 0.50781 -0.0156 1.375 0.501953 0.8867 0.58594
1.5 0.80078 0.6484 1.875 1.043945 1.4277 1.16797
2 1.38477

All the versions can be plotted as:

189
 1.5

0.5

0
0 0.5 1 1.5 2

Heun without corr Ralston

M idpoint Heun with corr

18.4 (a) The solution to the differential equation is

p = p0 e
kgt

Taking the natural log of this equation gives

ln p = ln p 0 + k g t

Therefore, a semi-log plot (ln p versus t) should yield a straight line with a slope of kg. The
plot, along with the linear regression best fit line is shown below. The estimate of the
population growth rate is kg = 0.0178/yr.

8.8 y = 0.0178x - 26.808

2
8.6 R = 0.9976
8.4
8.2
8
7.8
1940 1960 1980 2000

(b) The ODE can be integrated with the fourth-order RK method with the results tabulated
and plotted below:

t p k1 pmid k2 pmid k3 pend k4 φ

1950 2555.00 45.41 2668.53 47.43 2673.58 47.52 2792.60 49.64 47.49
1955 2792.46 49.63 2916.55 51.84 2922.06 51.94 3052.15 54.25 51.91
1960 3051.99 54.25 3187.61 56.66 3193.64 56.76 3335.81 59.29 56.73
1965 3335.64 59.29 3483.87 61.92 3490.45 62.04 3645.84 64.80 62.00
1970 3645.66 64.80 3807.66 67.68 3814.85 67.81 3984.69 70.82 67.77
1975 3984.48 70.82 4161.54 73.97 4169.41 74.11 4355.02 77.41 74.06
1980 4354.80 77.40 4548.31 80.84 4556.91 81.00 4759.78 84.60 80.95

190
 1985 4759.54 84.60 4971.03 88.36 4980.43 88.52 5202.15 92.46 88.47
1990 5201.89 92.46 5433.04 96.57 5443.31 96.75 5685.64 101.06 96.69
1995 5685.35 101.05 5937.98 105.54 5949.21 105.74 6214.06 110.45 105.68
2000 6213.75 110.44 6489.86 115.35 6502.13 115.57 6791.60 120.72 115.50
2005 6791.25 120.71 7093.02 126.07 7106.43 126.31 7422.81 131.93 126.24
2010 7422.43 131.93 7752.25 137.79 7766.90 138.05 8112.68 144.20 137.97
2015 8112.27 144.19 8472.74 150.60 8488.76 150.88 8866.67 157.60 150.79
2020 8866.22 157.59 9260.20 164.59 9277.70 164.90 9690.74 172.25 164.80
2025 9690.24 172.24 10120.84 179.89 10139.97 180.23 10591.40 188.25 180.12
2030 10590.85 188.24 11061.47 196.61 11082.38 196.98 11575.76 205.75 196.86
2035 11575.17 205.74 12089.52 214.88 12112.37 215.29 12651.61 224.87 215.16
2040 12650.96 224.86 13213.11 234.85 13238.09 235.30 13827.45 245.77 235.16
2045 13826.74 245.76 14441.14 256.68 14468.44 257.17 15112.57 268.61 257.01
2050 15111.79

16000

12000

8000

4000

0
1950 1970 1990 2010 2030 2050

18.5 (a) The analytical solution can be used to compute values at times over the range. For
example, the value at t = 1955 can be computed as

12,000
p = 2,555 = 2,826.2
2,555 + (12,000 − 2,555)e −0.026(1955−1950)

Values at the other times can be computed and displayed along with the data in the plot
below.

(b) The ODE can be integrated with the fourth-order RK method with the results tabulated
and plotted below:

t p-rk4 k1 tm ym k2 tm ym k3 te ye k4 φ
1950 2555.0 52.29 1952.5 2685.7 54.20 1952.5 2690.5 54.27 1955.0 2826.3 56.18 54.23
1955 2826.2 56.17 1957.5 2966.6 58.06 1957.5 2971.3 58.13 1960.0 3116.8 59.99 58.09
1960 3116.6 59.99 1962.5 3266.6 61.81 1962.5 3271.1 61.87 1965.0 3425.9 63.64 61.83
1965 3425.8 63.64 1967.5 3584.9 65.36 1967.5 3589.2 65.41 1970.0 3752.8 67.06 65.37
1970 3752.6 67.06 1972.5 3920.3 68.63 1972.5 3924.2 68.66 1975.0 4096.0 70.15 68.63
1975 4095.8 70.14 1977.5 4271.2 71.52 1977.5 4274.6 71.55 1980.0 4453.5 72.82 71.52
1980 4453.4 72.82 1982.5 4635.4 73.97 1982.5 4638.3 73.98 1985.0 4823.3 75.00 73.95
1985 4823.1 75.00 1987.5 5010.6 75.88 1987.5 5012.8 75.89 1990.0 5202.6 76.62 75.86

191
1990 5202.4 76.62 1992.5 5394.0 77.20 1992.5 5395.5 77.21 1995.0 5588.5 77.63 77.18
1995 5588.3 77.63 1997.5 5782.4 77.90 1997.5 5783.1 77.90 2000.0 5977.8 78.00 77.87
2000 5977.7 78.00 2002.5 6172.7 77.94 2002.5 6172.5 77.94 2005.0 6367.4 77.71 77.91
2005 6367.2 77.71 2007.5 6561.5 77.32 2007.5 6560.5 77.32 2010.0 6753.8 76.77 77.29
2010 6753.7 76.77 2012.5 6945.6 76.06 2012.5 6943.9 76.07 2015.0 7134.0 75.21 76.04
2015 7133.9 75.21 2017.5 7321.9 74.21 2017.5 7319.4 74.23 2020.0 7505.0 73.09 74.20
2020 7504.9 73.09 2022.5 7687.6 71.83 2022.5 7684.5 71.85 2025.0 7864.2 70.47 71.82
2025 7864.0 70.47 2027.5 8040.2 68.98 2027.5 8036.5 69.01 2030.0 8209.1 67.43 68.98
2030 8208.9 67.43 2032.5 8377.5 65.75 2032.5 8373.3 65.80 2035.0 8537.9 64.04 65.76
2035 8537.7 64.05 2037.5 8697.8 62.23 2037.5 8693.3 62.28 2040.0 8849.1 60.41 62.25
2040 8849.0 60.41 2042.5 9000.0 58.50 2042.5 8995.2 58.56 2045.0 9141.8 56.61 58.53
2045 9141.6 56.62 2047.5 9283.1 54.65 2047.5 9278.2 54.72 2050.0 9415.2 52.73 54.68
2050 9415.0

10000
8000
6000
4000
2000
0
1950 1970 1990 2010 2030 2050

pdata panal p-rk4

Thus, the RK4 results are so close to the analytical solution that the two results are
indistinguishable graphically.

18.6 We can solve this problem with the M-file Eulode (Fig. 18.3). First, we develop a function
to compute the derivative

function dv = dvdt(t, v)
if t < 10
% chute is unopened
dv = 9.81 - 0.25/80*v^2;
else
% chute is opened
dv = 9.81 - 5/80*v^2;
end

Notice how we have used an If statement to use a higher drag coefficient for times after the
cord is pulled. The Eulode function can then be used to generate results and display them
graphically..

>> [t,v] = Eulode(@dvdt,[0 30],0,0.1);

>> plot(t,v)

192
18.7 (a) Euler’s method:

t y z dy/dt dz/dt
0 2 4 16 -16
0.1 3.6 2.4 3.658049 -10.368
0.2 3.965805 1.3632 -2.35114 -3.68486
0.3 3.730691 0.994714 -3.77687 -1.84568
0.4 3.353004 0.810147 -3.99072 -1.10035

3
y
2 z

0
0 0.1 0.2 0.3 0.4

(b) 4th-order RK method:

k1,1 = f 1 (0,2,4) = −2(2) + 5(4)e −0 = 16

2( 4) 2
k1, 2 = f 2 (0,2,4) = − = −16
2

y (0.05) = 2 + 16(0.05) = 2.8

z (0.05) = 4 − 16(0.05) = 3.2

193
k 2,1 = f 1 (0.05,2.8,3.2) = −2( 2.8) + 5(3.2)e −0.05 = 9.619671

2.8(3.2) 2
k 2, 2 = f 2 (0.05,2.8,3.2) = − = −14.336
2

y (0.05) = 2 + 9.619671(0.05) = 2.480984

z (0.05) = 4 − 14.336(0.05) = 3.2832

k 3,1 = f 1 (0.05,2.480984,3.2832) = −2(2.480984) + 5(3.2832)e −0.05 = 10.65342

2.480984(3.2832) 2
k 3, 2 = f 2 (0.05,2.480984,3.2832) = − = −13.3718
2

y (0.1) = 2 + 10.65342(0.1) = 3.065342

z (0.1) = 4 − 13.3718(0.1) = 2.662824

k 4,1 = f 1 (0.1,3.065342,2.662824) = −2(3.065342) + 5(3.2832)e −0.1 = 5.916431

3.065342(2.662824) 2
k 4, 2 = f 2 (0.1,3.065342,2.662824) = − = −10.8676
2

The k’s can then be used to compute the increment functions,

16 + 2(9.619671 + 10.65342) + 5.916431

φ1 = = 10.41043
6
− 16 + 2(−14.336 − 13.3718) − 10.8676
φ2 = = −13.7139
6

These slope estimates can then be used to make the prediction for the first step

y (0.1) = 2 + 10.41043(0.1) = 3.041043

z (0.1) = 4 − 13.7139(0.1) = 2.628615

The remaining steps can be taken in a similar fashion and the results summarized as

t y z
0 2 4
0.1 3.041043 2.628615
0.2 3.342571 1.845308
0.3 3.301983 1.410581
0.4 3.107758 1.149986

A plot of these values can be developed.

194
 5

3
y
2 z

0
0 0.1 0.2 0.3 0.4

18.8 The second-order van der Pol equation can be reexpressed as a system of 2 first-order
ODEs,

dy
=z
dt
dz
= (1 − y 2 ) z − y
dt

(a) Euler (h = 0.2). Here are the first few steps. The remainder of the computation would be
implemented in a similar fashion and the results displayed in the plot below.

t y (h = 0.2) z (h = 0.2) dy/dt dz/dt

0 1 1 1 -1
0.2 1.2 0.8 0.8 -1.552
0.4 1.36 0.4896 0.4896 -1.77596
0.6 1.45792 0.1344072 0.134407 -1.6092
0.8 1.4848014 -0.187433 -0.18743 -1.25901

(b) Euler (h = 0.1). Here are the first few steps. The remainder of the computation would be
implemented in a similar fashion and the results displayed in the plot below.

t y (h = 0.1) z (h = 0.1) dy/dt dz/dt

0 1 1 1 -1
0.1 1.1 0.9 0.9 -1.289
0.2 1.19 0.7711 0.7711 -1.51085
0.3 1.26711 0.6200145 0.620015 -1.64257
0.4 1.3291115 0.4557574 0.455757 -1.67847

195
 4 y (h = 0.1)
z (h = 0.1)
3
y (h = 0.2)
2 z (h = 0.2)

0
0 2 4 6 8 10
-1

-2

-3

-4

18.9 The second-order equation can be reexpressed as a system of two first-order ODEs,

dy
=z
dt
dz
= −9 y
dt

(a) Euler. Here are the first few steps along with the analytical solution. The remainder of
the computation would be implemented in a similar fashion and the results displayed in the
plot below.

t yEuler zEuler dy/dt dz/dt yanalytical

0 1 0 0 -9 1
0.1 1 -0.9 -0.9 -9 0.955336
0.2 0.91 -1.8 -1.8 -8.19 0.825336
0.3 0.73 -2.619 -2.619 -6.57 0.62161
0.4 0.4681 -3.276 -3.276 -4.2129 0.362358

4 yEuler
3
yanal
2
1
0
-1 0 1 2 3 4
-2
-3
-4
-5

196
(b) RK4. Here are the first few steps along with the analytical solution. The remainder of
the computation would be implemented in a similar fashion and the results displayed in the
plot below.

k1,1 = f 1 (0,1,0) = z = 0

k1, 2 = f 2 (0,1,0) = −9 y = −9(1) = −9

y (0.05) = 1 + 0(0.05) = 1

z (0.05) = 0 − 9(0.05) = −0.45

k 2,1 = f 1 (0.05,1,−0.45) = −0.45

k 2, 2 = f 2 (0.05,1,−0.45) = −9(1) = −9

y (0.05) = 1 + ( −0.45)(0.05) = 0.9775

z (0.05) = 0 − 9(0.05) = −0.45

k 3,1 = f 1 (0.05,0.9775,−0.45) = −0.45

k 3, 2 = f 2 (0.05,0.9775,−0.45) = −9(0.9775) = −8.7975

y (0.1) = 1 + (−0.45)(0.1) = 0.9550

z (0.1) = 0 − 8.7975(0.1) = −0.8798

k 4,1 = f 1 (0.1,0.9550,−0.8798) = −0.8798

k 4, 2 = f 2 (0.1,0.9550,−0.8798) = −9(0.9550) = −8.5950

The k’s can then be used to compute the increment functions,

0 + 2(−0.45 − 0.45) − 0.8798

φ1 = = −0.4466
6
− 9 + 2(−9 − 8.7975) − 8.5950
φ2 = = −8.8650
6

These slope estimates can then be used to make the prediction for the first step

y (0.1) = 1 − 0.4466(0.1) = 0.9553

z (0.1) = 0 − 8.8650(0.1) = −0.8865

The remaining steps can be taken in a similar fashion and the first few results summarized
as

197
 t y z yanal
0 1.0000 0.0000 1.00000
0.1 0.9553 -0.8865 0.95534
0.2 0.8253 -1.6938 0.82534
0.3 0.6216 -2.3498 0.62161
0.4 0.3624 -2.7960 0.36236
0.5 0.0708 -2.9924 0.07074

As can be seen, the results agree with the analytical solution closely. A plot of all the values
can be developed and indicates the same close agreement.

4 yRK4
3
yanal
2
1
0
-1 0 1 2 3 4
-2
-3
-4
-5

18.10 A MATLAB M-file for Heun’s method with iteration can be developed as

function [t,y] = Heun(dydt,tspan,y0,h,es,maxit)

% [t,y] = Heun(dydt,tspan,y0,h):
% uses the midpoint method to integrate an ODE
% input:
% dydt = name of the M-file that evaluates the ODE
% tspan = [ti, tf] where ti and tf = initial and
% final values of independent variable
% y0 = initial value of dependent variable
% h = step size
% es = stopping criterion (%)
% optional (default = 0.001)
% maxit = maximum iterations of corrector
% optional (default = 50)
% es = (optional) stopping criterion (%)
% maxit = (optional) maximum allowable iterations
% output:
% t = vector of independent variable
% y = vector of solution for dependent variable

% if necessary, assign default values

if nargin<6, maxit = 50; end %if maxit blank set to 50
if nargin<5, es = 0.001; end %if es blank set to 0.001
ti = tspan(1);
tf = tspan(2);
t = (ti:h:tf)';
n = length(t);
% if necessary, add an additional value of t

198
% so that range goes from t = ti to tf
if t(n)<tf
t(n+1) = tf;
n = n+1;
end
y = y0*ones(n,1); %preallocate y to improve efficiency
iter = 0;
for i = 1:n-1
hh = t(i+1) - t(i);
k1 = feval(dydt,t(i),y(i));
y(i+1) = y(i) + k1*hh;
while (1)
yold = y(i+1);
k2 = feval(dydt,t(i)+hh,y(i+1));
y(i+1) = y(i) + (k1+k2)/2*hh;
iter = iter + 1;
if y(i+1) ~= 0, ea = abs((y(i+1) - yold)/y(i+1)) * 100; end
if ea <= es | iter >= maxit, break, end
end
end
plot(t,y)

Here is the test of the solution of Prob. 18.5. First, an M-file holding the differential
equation is written as

function dp = dpdt(t, p)
dp = 0.026*(1-p/12000)*p;

Then the M-file can be invoked as in

>> [t,p]=Heun(@dpdt,[1950 2000],2555,5,0.1);

>> disp([t,p])
1.0e+003 *

1.9500 2.5550
1.9550 2.8261
1.9600 3.1165
1.9650 3.4256
1.9700 3.7523
1.9750 4.0953
1.9800 4.4527
1.9850 4.8222
1.9900 5.2012
1.9950 5.5868
2.0000 5.9759

The following plot is generated

199
18.11 A MATLAB M-file for the midpoint method can be developed as

function [t,y] = midpoint(dydt,tspan,y0,h)

% [t,y] = midpoint(dydt,tspan,y0,h):
% uses the midpoint method to integrate an ODE
% input:
% dydt = name of the M-file that evaluates the ODE
% tspan = [ti, tf] where ti and tf = initial and
% final values of independent variable
% y0 = initial value of dependent variable
% h = step size
% output:
% t = vector of independent variable
% y = vector of solution for dependent variable

ti = tspan(1);
tf = tspan(2);
t = (ti:h:tf)';
n = length(t);
% if necessary, add an additional value of t
% so that range goes from t = ti to tf
if t(n)<tf
t(n+1) = tf;
n = n+1;
end
y = y0*ones(n,1); %preallocate y to improve efficiency
for i = 1:n-1
hh = t(i+1) - t(i);
k1 = feval(dydt,t(i),y(i));
ymid = y(i) + k1*hh/2;
k2 = feval(dydt,t(i)+hh/2,ymid);
y(i+1) = y(i) + k2*hh;
end
plot(t,y)

Here is the test of the solution of Prob. 18.5. First, an M-file holding the differential
equation is written as

function dp = dpdt(t, p)
dp = 0.026*(1-p/12000)*p;

200
 Then the M-file can be invoked as in

>> [t,p]=midpoint(@dpdt,[1950 2000],2555,5);

>> disp([t,p])
1.0e+003 *
1.9500 2.5550
1.9550 2.8260
1.9600 3.1163
1.9650 3.4253
1.9700 3.7521
1.9750 4.0953
1.9800 4.4529
1.9850 4.8227
1.9900 5.2021
1.9950 5.5881
2.0000 5.9776

The following plot is generated

18.12 A MATLAB M-file for the fourth-order RK method can be developed as

function [t,y] = rk4(dydt,tspan,y0,h)

% [t,y] = rk4(dydt,tspan,y0,h):
% uses the fourth-order Runge-Kutta method to integrate an ODE
% input:
% dydt = name of the M-file that evaluates the ODE
% tspan = [ti, tf] where ti and tf = initial and
% final values of independent variable
% y0 = initial value of dependent variable
% h = step size
% output:
% t = vector of independent variable
% y = vector of solution for dependent variable

ti = tspan(1);
tf = tspan(2);
t = (ti:h:tf)';
n = length(t);
% if necessary, add an additional value of t

201
% so that range goes from t = ti to tf
if t(n)<tf
t(n+1) = tf;
n = n+1;
end
y = y0*ones(n,1); %preallocate y to improve efficiency
for i = 1:n-1
hh = t(i+1) - t(i);
k1 = feval(dydt,t(i),y(i));
ymid = y(i) + k1*hh/2;
k2 = feval(dydt,t(i)+hh/2,ymid);
ymid = y(i) + k2*hh/2;
k3 = feval(dydt,t(i)+hh/2,ymid);
yend = y(i) + k3*hh;
k4 = feval(dydt,t(i)+hh,yend);
phi = (k1+2*(k2+k3)+k4)/6;
y(i+1) = y(i) + phi*hh;
end
plot(t,y)

Here is the test of the solution of Prob. 18.2. First, an M-file holding the differential
equation is written as

function dy = dydx(x, y)
dy = (1+2*x)*sqrt(y);

Then the M-file can be invoked as in

>> [x,y] = rk4(@dydx,[0 1],1,0.1);

>> disp([x,y])
0 1.0000
0.1000 1.1130
0.2000 1.2544
0.3000 1.4280
0.4000 1.6384
0.5000 1.8906
0.6000 2.1904
0.7000 2.5440
0.8000 2.9584
0.9000 3.4410
1.0000 4.0000

The following plot is generated

202
18.13 Note that students can take two approaches to developing this M-file. The first program
shown below is strictly developed to solve 2 equations.

function [t,y1,y2] = rk42(dy1dt,dy2dt,tspan,y10,y20,h)

% [t,y1,y2] = rk42(dy1dt,dy2dt,tspan,y10,y20,h):
% uses the fourth-order RK method to integrate a pair of ODEs
% input:
% dy1dt = name of the M-file that evaluates the first ODE
% dy2dt = name of the M-file that evaluates the second ODE
% tspan = [ti, tf] where ti and tf = initial and
% final values of independent variable
% y10 = initial value of first dependent variable
% y20 = initial value of second dependent variable
% h = step size
% output:
% t = vector of independent variable
% y1 = vector of solution for first dependent variable
% y2 = vector of solution for second dependent variable

ti = tspan(1);
tf = tspan(2);
t = (ti:h:tf)';
n = length(t);
% if necessary, add an additional value of t
% so that range goes from t = ti to tf
if t(n)<tf
t(n+1) = tf;
n = n+1;
end
y1 = y10*ones(n,1); %preallocate y's to improve efficiency
y2 = y20*ones(n,1);
for i = 1:n-1
hh = t(i+1) - t(i);
k11 = feval(dy1dt,t(i),y1(i),y2(i));
k12 = feval(dy2dt,t(i),y1(i),y2(i));
ymid1 = y1(i) + k11*hh/2;
ymid2 = y2(i) + k12*hh/2;
k21 = feval(dy1dt,t(i)+hh/2,ymid1,ymid2);
k22 = feval(dy2dt,t(i)+hh/2,ymid1,ymid2);
ymid1 = y1(i) + k21*hh/2;
ymid2 = y2(i) + k22*hh/2;
k31 = feval(dy1dt,t(i)+hh/2,ymid1,ymid2);

203
 k32 = feval(dy2dt,t(i)+hh/2,ymid1,ymid2);
yend1 = y1(i) + k31*hh;
yend2 = y2(i) + k32*hh;
k41 = feval(dy1dt,t(i)+hh,yend1,yend2);
k42 = feval(dy2dt,t(i)+hh,yend1,yend2);
phi1 = (k11+2*(k21+k31)+k41)/6;
phi2 = (k12+2*(k22+k32)+k42)/6;
y1(i+1) = y1(i) + phi1*hh;
y2(i+1) = y2(i) + phi2*hh;
end
plot(t,y1,t,y2,'--')

Here is the test of the solution of Prob. 18.7. First, M-files holding the differential equations
are written as

function dy = dy1dt(t, y1, y2)

dy = -2*y1 + 5*y2*exp(-t);

function dy = dy2dt(t, y1, y2)

dy = -y1*y2^2/2;

Then the M-file can be invoked as in

>> [t,y1,y2]=rk42(@dy1dt,@dy2dt,[0 0.4],2,4,0.1);

>> disp([t,y1,y2])
0 2.0000 4.0000
0.1000 3.0410 2.6286
0.2000 3.3426 1.8453
0.3000 3.3020 1.4106
0.4000 3.1078 1.1500

The following plot is generated

A better approach is to develop an M-file that can be used for any number of simultaneous
first-order ODEs as in the following code:

function [t,y] = rk4sys(dydt,tspan,y0,h)

204
% [t,y] = rk4sys(dydt,tspan,y0,h):
% uses the fourth-order RK method to integrate a pair of ODEs
% input:
% dydt = name of the M-file that evaluates the ODEs
% tspan = [ti, tf] where ti and tf = initial and
% final values of independent variable
% y0 = initial values of dependent variables
% h = step size
% output:
% t = vector of independent variable
% y = vector of solution for dependent variables

ti = tspan(1);
tf = tspan(2);
t = (ti:h:tf)';
n = length(t);
% if necessary, add an additional value of t
% so that range goes from t = ti to tf
if t(n)<tf
t(n+1) = tf;
n = n+1;
end
y(1,:) = y0;
for i = 1:n-1
hh = t(i+1) - t(i);
k1 = feval(dydt,t(i),y(i,:))';
ymid = y(i,:) + k1*hh/2;
k2 = feval(dydt,t(i)+hh/2,ymid)';
ymid = y(i,:) + k2*hh/2;
k3 = feval(dydt,t(i)+hh/2,ymid)';
yend = y(i,:) + k3*hh;
k4 = feval(dydt,t(i)+hh,yend)';
phi = (k1+2*(k2+k3)+k4)/6;
y(i+1,:) = y(i,:) + phi*hh;
end
plot(t,y(:,1),t,y(:,2),'--')

This code solves as many ODEs as are specified. Here is the test of the solution of Prob.
18.7. First, a single M-file holding the differential equations can be written as

function dy = dydtsys(t, y)
dy = [-2*y(1) + 5*y(2)*exp(-t);-y(1)*y(2)^2/2];

Then the M-file can be invoked as in

>> [t,y]=rk4sys(@dydtsys,[0 0.4],[2 4],0.1);

>> disp([t,y])
0 2.0000 4.0000
0.1000 3.0410 2.6286
0.2000 3.3426 1.8453
0.3000 3.3020 1.4106
0.4000 3.1078 1.1500

205
CHAPTER 19
19.1 (a) Euler’s method. Here are the first few steps

t x y dx/dt dy/dt
0 2.0000 1.0000 1.2000 -0.2000
0.1 2.1200 0.9800 1.2974 -0.1607
0.2 2.2497 0.9639 1.3985 -0.1206
0.3 2.3896 0.9519 1.5028 -0.0791
0.4 2.5399 0.9440 1.6093 -0.0359
0.5 2.7008 0.9404 1.7171 0.0096

The computation can be continued and the results plotted versus time:

16
x
y
12

0
0 5 10 15 20 25 30

Notice that the amplitudes of the oscillations are expanding. This is also illustrated by a
state-space plot (y versus x):

16

12

0
0 4 8 12 16

(b) RK4. Here is the first step in detail.

k1,1 = f 1 (0,2,1) = 1.5(2) − 0.7(2)(1) = 1.6

k1, 2 = f 2 (0,2,1) = −0.9(1) + 0.4(2)(1) = −0.1

206
x(0.05) = 2 + 1.6(0.05) = 2.08

y (0.05) = 1 − 0.1(0.05) = 0.995

k 2,1 = f 1 (0.05,2.08,0.995) = 1.67128

k 2, 2 = f 2 (0.05,2.08,0.995) = −0.06766

x(0.05) = 2 + 1.67128(0.05) = 2.083564

y (0.05) = 1 − 9(0.05) = 0.996617

k 3,1 = f 1 (0.05,2.083564,0.996617) = 1.671785

k 3, 2 = f 2 (0.05,2.083564,0.996617) = −0.06635

x(0.1) = 2 + 1.671785(0.1) = 2.167179

y (0.1) = 1 − 0.06635(0.1) = 0.993365

k 4,1 = f 1 (0.1,2.167179,0.993365) = 1.743808

k 4, 2 = f 2 (0.1,2.167179,0.993365) = −0.03291

The k’s can then be used to compute the increment functions,

1.6 + 2(1.67128 + 1.671785) + 1.743808

φ1 = = 1.671656
6
− 0.1 + 2(−0.06766 − 0.06635) − 0.03291
φ2 = = −0.06682
6

These slope estimates can then be used to make the prediction for the first step

x(0.1) = 2 + 1.671656(0.1) = 2.16766

y (0.1) = 1 − 0.06682(0.1) = 0.993318

The remaining steps can be taken in a similar fashion and the first few results summarized
as

t x y
0 2 1
0.1 2.167166 0.993318
0.2 2.348838 0.993588
0.3 2.545029 1.001398
0.4 2.755314 1.017509
0.5 2.978663 1.042891

207
A plot of all the values can be developed. Note that in contrast to Euler’s method, the
cycles do not amplify as time proceeds.

6
x
y
4

0
0 5 10 15 20 25 30

This periodic nature is also evident from the state-space plot. Because this is the expected
behavior we can see that the RK4 is far superior to Euler’s method for this particular
problem.

0
0 1 2 3 4 5

(c) To implement ode45, first a function is developed to evaluate the predator-prey ODEs,

function yp = predprey(t,y)
yp = [1.5*y(1)-0.7*y(1)*y(2);-0.9*y(2)+0.4*y(1)*y(2)];

Then, the solution and plot can be obtained:

>> [t,y] = ode45(@predprey,[0 30],[2 1]);

>> plot(t,y(:,1),t,y(:,2),'--')
>> legend('x(prey)','y(predator)')

208
19.2 (a) Here are the results for the first few steps as computed with the classical RK4 technique

t x y z
0 5 5 5
0.1 9.78147 17.07946 10.43947
0.2 17.70297 20.8741 35.89688
0.3 10.81088 -2.52924 39.30744
0.4 0.549578 -5.54419 28.07462
0.5 -3.1646 -5.84128 22.36888
0.6 -5.57588 -8.42037 19.92312
0.7 -8.88719 -12.6789 22.14148
0.8 -11.9142 -13.43 29.80001
0.9 -10.6668 -7.21784 33.39903
1 -6.84678 -3.43018 29.30717

The results from t = 0 to 20 can be displayed graphically as

50 x y z
40

30

20

10

-10
0 5 10 15 20

-20

-30

209
The solution appears chaotic bouncing around from negative to positive values. Although
the pattern might appear random, an underlying pattern emerges when we look at the state-
space plots. For example, here is the plot of y versus x.

25

15

-20 -10 -5 0 10 20

-15

-25

And here is z versus x,

50

40

30

20

10

0
-20 -10 0 10 20

(b) To implement any of the MATLAB functions, first a function is developed to evaluate
the Lorenz ODEs,

function yp = lorenz(t,y)
yp = [-10*y(1)+10*y(2);28*y(1)-y(2)-y(1)*y(3);-2.666667*y(3)+y(1)*y(2)];

Then, the solution and plots for the ode23 function can be obtained:

>> [t,y] = ode23(@lorenz,[0 20],[5 5 5]);

>> plot(t,y(:,1),t,y(:,2),'--',t,y(:,3),':')
>> legend('x','y','z')
>> plot(y(:,1),y(:,2))

210
Notice how this plot, although qualitatively similar to the constant step RK4 result in (a),
the details are quite different. However, the state-space representation looks much more
consistent.

>> plot(y(:,1),y(:,2))

(c) The ode45 again differs in the details of the time-series plot,

>> [t,y] = ode45(@lorenz,[0 20],[5 5 5]);

>> plot(t,y(:,1),t,y(:,2),'--',t,y(:,3),':')
>> legend('x','y','z')

211
 (d) The ode23tb also differs in the details of the time-series plot,

>> [t,y] = ode23tb(@lorenz,[0 20],[5 5 5]);

>> plot(t,y(:,1),t,y(:,2),'--',t,y(:,3),':')
>> legend('x','y','z')

Close inspection of all the above results indicates that they all yield identical results for a
period of time. Thereafter, they abruptly begin to diverge. The reason for this behavior is
that these equations are highly sensitive to their initial conditions. After a number of steps,
because they all employ different algorithms, they begin to diverge slightly. When the
discrepancy becomes large enough (which for these equations is not that much), the
solution will tend to make a large jump. Thus, after awhile, the various solutions become
uncorrelated. Such solutions are said to be chaotic. It was this characteristic of these
particular equations that led Lorenz to suggest that long-range weather forecasts might not
be possible.

19.3 First step:

212
 Predictor:

y10 = 5.222138+[−0.5(4.143883)+e−2]1 = 3.285532

Corrector:

− 0.5(4.143883) + e −2 − 0.5(3.285532) + e −2.5

y1 = 4.143883 + 0.5 = 3.269562
1

The corrector can be iterated to yield

j yi+1j ε a ,%
1 3.269562
2 3.271558 0.061

Second step:

Predictor:

y20 = 4.143883+[−0.5(3.271558)+e−2.5]1 = 2.590189

Predictor Modifier:

y20 = 2.590189+4/5(3.271558-3.285532) = 2.579010

Corrector:

− 0.5(3.271558) + e −2.5 − 0.5(2.579010) + e −3

y 2 = 3.271558 + 0.5 = 2.573205
1

The corrector can be iterated to yield

j yi+1j ε a ,%
1 2.573205
2 2.573931 0.0282

19.4 Before solving, for comparative purposes, we can develop the analytical solution as

t3
−t
y= e3

Thus, the true values being simulated in this problem are

t y
0 1
0.25 0.782868
0.5 0.632337

213
 The first step is taken with the fourth-order RK:

k1 = f (0,1) = 1(0) 2 − 1 = −1

y (0.125) = 1 − 1(0.125) = 0.875

k 2 = f (0.125,0.875) = −0.861328

y (0.125) = 1 − 0.861328(0.125) = 0.89233

k 3 = f (0.125,0.89233) = −0.87839

y (0.25) = 1 − 0.87839(0.25) = 0.78040

k 4 = f (0.25,0.78040) = −0.73163

− 1 + 2(−0.861328 − 0.87839) − 0.73163

φ= = −0.86851
6

y (0.25) = 1 − 0.86851(0.25) = 0.7828723

This result compares favorably with the analytical solution.

The second step can then be implemented with the non-self-starting Heun method:

Predictor:

y (0.5) = 1 + (0.7828723(0.25) 2 − 0.7828723)0.5 = 0.633028629

Corrector: (First iteration):

− 0.7339 + (0.633028629(0.5) 2 − 0.633028629)

y (0.5) = 0.7828723 + 0.25 = 0.63178298
2

Corrector: (Second iteration):

− 0.7339 + (0.63178298(0.5) 2 − 0.63178298)

y (0.5) = 0.7828723 + 0.25 = 0.63189976
2

The iterative process can be continued with the final result converging on 0.63188975.

19.5 (a) h < 2/100,000 = 2×10−5.

(b) The implicit Euler can be written for this problem as

( )
y i +1 = y i + − 100,000 y i +1 + 99,999e −ti +1 h

214
 which can be solved for

y i + 99,999e − ti +1 h
y i +1 =
1 + 100,000h

The results of applying this formula for the first few steps are shown below. A plot of the
entire solution is also displayed

t y
0 0
0.1 1.904638
0.2 1.818731
0.3 1.740819
0.4 1.67032
0.5 1.606531

0
0 1 2

19.6 The implicit Euler can be written for this problem as

y i +1 = y i + (30(sin t i +1 − y i +1 ) + 3 cos t i +1 )h

which can be solved for

y i + 30 sin t i +1 h + 3 cos t i +1 h
y i +1 =
1 + 30h

The results of applying this formula are tabulated and graphed below.

t y t y t y t y
0 0 1.2 0.952306 2.4 0.622925 3.6 -0.50089
0.4 0.444484 1.6 0.993242 2.8 0.270163 4 -0.79745
0.8 0.760677 2 0.877341 3.2 -0.12525

215
 1.5

0.5

0
0 1 2 3 4
-0.5

-1

19.7 (a) The explicit Euler can be written for this problem as

x1,i +1 = x1,i + (999 x1,i + 1999 x 2,i )h

x 2,i +1 = x 2,i + (− 1000 x1,i − 2000 x 2,i )h

Because the step-size is much too large for the stability requirements, the solution is
unstable,

t x1 x2 dx1/dt dx2/dt
0 1 1 2998 -3000
0.05 150.9 -149 -147102 147100
0.1 -7204.2 7206 7207803 -7207805
0.15 353186 -353184 -3.5E+08 3.53E+08
0.2 -1.7E+07 17305943 1.73E+10 -1.7E+10

(b) The implicit Euler can be written for this problem as

x1,i +1 = x1,i + (999 x1,i +1 + 1999 x 2 ,i +1 )h

x 2 ,i +1 = x 2 ,i + ( − 1000 x1,i +1 − 2000 x 2 ,i +1 )h

or collecting terms

(1 − 999h) x1,i +1 − 1999hx 2 ,i +1 = x1,i

1000hx1,i +1 + (1 + 2000h) x 2 ,i +1 = x 2 ,i

or substituting h = 0.05 and expressing in matrix format

⎡ − 48.95 − 99.95⎤ ⎧ x1,i +1 ⎫ = ⎧ x1,i ⎫

⎣⎢ 50 101 ⎥⎦ ⎨⎩ x 2 ,i +1 ⎬⎭ ⎨⎩ x 2 ,i ⎬⎭

Thus, to solve for the first time step, we substitute the initial conditions for the right-hand
side and solve the 2x2 system of equations. The best way to do this is with LU
decomposition since we will have to solve the system repeatedly. For the present case,
because its easier to display, we will use the matrix inverse to obtain the solution. Thus, if
the matrix is inverted, the solution for the first step amounts to the matrix multiplication,

216
 ⎧ x1,i +1 ⎫ ⎡ 1886088
⎨x
⎩ 2 ,i +1 ⎭ ⎣
. 186648
. ⎤1
⎦ {} {3.752568
⎬ = ⎢ − 0.93371 − 0.9141⎥ 1 = − 184781
. }
For the second step (from x = 0.05 to 0.1),

⎧ x1,i +1 ⎫ ⎡ 1886088
⎨x
⎩ 2 ,i + 1 ⎭ ⎣
. 186648
.
{
⎬ = ⎢ − 0.93371 − 0.9141⎥ − 184781
⎦ . }{
⎤ 3.752568 = 3.62878
− 181472
. }
The remaining steps can be implemented in a similar fashion to give

t x1 x2
0 1 1
0.05 3.752568 -1.84781
0.1 3.62878 -1.81472
0.15 3.457057 -1.72938
0.2 3.292457 -1.64705

The results are plotted below, along with a solution with the explicit Euler using a step of
0.0005.

4 x1

0
0 0.1 0.2
-2 x2
19.8 (a) The exact solution is

y = Ae 5t + t 2 + 0.4t + 0.08

If the initial condition at t = 0 is 0.8, A = 0,

y = t 2 + 0.4t + 0.08

Note that even though the choice of the initial condition removes the positive exponential
terms, it still lurks in the background. Very tiny round off errors in the numerical solutions
bring it to the fore. Hence all of the following solutions eventually diverge from the
analytical solution.

(b) 4th order RK. The plot shows the numerical solution (bold line) along with the exact
solution (fine line).

217
 15

10

-5
0 1 2 3 4
-10

(c)
function yp = dy(t,y)
yp = 5*(y-t^2);

>> tspan = [0,5];

>> y0 = 0.08;
>> [t,y] = ode45(@dy1,tspan,y0);

(d)
>> [t,y] = ode23s(@dy1,tspan,y0);

(e)
>> [t,y] = ode23tb(@dy1,tspan,y0);

30
20
10
0
-10 0 1 2 3 4 5
-20
-30

RK4 Analytical ODE45

ODE23s ODE23tb

19.9 (a) As in Example 17.5, the humps function can be integrated with the quad function as in

>> format long

>> quad(@humps,0,1)

ans =
29.85832612842764

(b) Using ode45 is based on recognizing that the evaluation of the definite integral

218
 b
I= ∫ a
f ( x) dx

is equivalent to solving the differential equation

dy
= f (x)
dx

for y(b) given the initial condition y(a) = 0. Thus, we must solve the following initial-value
problem:

dy 1 1
= + −6
dx ( x − 0.3) + 0.01 ( x − 0.9) 2 + 0.04
2

where y(0) = 0. To do this with ode45, we must first set up an M-file to evaluate the right-
hand side of the differential equation,

function dy = humpsODE(x,y)
dy = 1./((x-0.3).^2 + 0.01) + 1./((x-0.9).^2+0.04) - 6;

Then, the integral can be evaluated as

>> [x,y] = ode45(@humpsODE,[0 0.5 1],0);

>> disp([x,y])
0 0
0.50000000000000 21.78356481821654
1.00000000000000 29.85525185285369

Thus, the integral estimate is within 0.01% of the estimate obtained with the quad function.
Note that a better estimate can be obtained by using the odeset function to set a smaller
relative tolerance as in

>> options = odeset('RelTol',1e-8);

>> [x,y] = ode45(@humpsODE,[0 0.5 1],0,options);
>> disp([x,y])
0 0
0.50000000000000 21.78683736423308
1.00000000000000 29.85832514287622

19.10 The nonlinear model can be expressed as the following set of ODEs,

dθ
=v
dt
dv g
= − sin θ
dt l

where v = the angular velocity. A function can be developed to compute the right-hand-side
of this pair of ODEs for the case where g = 9.81 and l = 0.6 m,

function dy = dpnon(t, y)
dy = [y(2);-9.81/0.6*sin(y(1))];

219
The linear model can be expressed as the following set of ODEs,

dθ
=v
dt
dv g
=− θ
dt l

A function can be developed as,

function dy = dplin(t, y)
dy = [y(2);-9.81/0.6*y(1)];

Then, the solution and plot can be obtained for the case where θ(0) = π/8. Note that we only
depict the displacement (θ or y(1)) in the plot

>> [tn yn] = ode45(@dpnon,[0 10],[pi/8 0]);

>> [tl yl] = ode45(@dplin,[0 10],[pi/8 0]);
>> plot(tn,yn(:,1),tl,yl(:,1),'--')
>> legend('nonlinear','linear')

You should notice two aspects of this plot. First, because the displacement is small, the
linear solution provides a decent approximation of the more physically realistic nonlinear
case. Second, the two solutions diverge as the computation progresses.

For the larger initial displacement (θ(0) = π/8), the solution and plot can be obtained as,

>> [tn yn] = ode45(@dpnon,[0 10],[pi/2 0]);

>> [tl yl] = ode45(@dplin,[0 10],[pi/2 0]);
>> plot(tn,yn(:,1),tl,yl(:,1),'--')
>> legend('nonlinear','linear')

220
 Because the linear approximation is only valid at small displacements, there are now clear
and significant discrepancies between the nonlinear and linear cases that are exacerbated as
the solution progresses.

19.11 A function can be developed to compute the right-hand-side of the ODE,

function yp = dpdt(t, p)
yp = 0.026*(1-p/12000)*p;

The function ode45 can be used to integrate this equation and generate results
corresponding to the dates for the measured population data. A plot can also be generated
of the solution and the data,

>> tspan = 1950:5:2000;

>> pdata = [2555 2780 3040 3346 3708 4087 4454 4850 5276 5686 6079]';
>> [t,p] = ode45(@dpdt,tspan,2555);
>> plot(t,p,t,pdata,'o')

The sum of the squares of the residuals can be computed as

221
>> SSR = sum((p - pdata).^2)

SSR =
4.2365e+004

222
CHAPTER 20
20.1 The matrix inverse can be evaluated and the power method expressed as

⎡0.0375 0.025 0.0125⎤

⎢ 0.025 0.05 0.025 ⎥ − λ[ I ] = 0
⎢⎣0.0125 0.025 0.0375⎥⎦

Iteration 1:

⎡0.0375 0.025 0.0125⎤ ⎧⎪1⎫⎪ ⎧⎪0.075⎫⎪ ⎧⎪0.75⎫⎪

⎢ 0.025 0.05 0.025 ⎥ ⎨1⎬ = ⎨ 0.1 ⎬ = 0.1⎨ 1 ⎬
⎢⎣0.0125 0.025 0.0375⎥⎦ ⎪⎩1⎪⎭ ⎪⎩0.075⎪⎭ ⎪⎩0.75⎪⎭

Iteration 2:

⎡0.0375 0.025 0.0125⎤ ⎧⎪0.75⎫⎪ ⎧⎪0.0625⎫⎪ ⎧⎪0.71428571⎫⎪

⎢ 0.025 0.05 0.025 ⎥ ⎨ 1 ⎬ = ⎨0.0875⎬ = 0.0875⎨ 1 ⎬
⎢⎣0.0125 0.025 0.0375⎥⎦ ⎪⎩0.75⎪⎭ ⎪⎩0.0625⎪⎭ ⎪⎩0.71428571⎪⎭

0.0875 − 0.1
εa = × 100% = 14.29%
0.0875

Iteration 3:

⎡0.0375 0.025 0.0125⎤ ⎧⎪0.71428571⎫⎪ ⎧⎪0.060714⎫⎪ ⎧⎪0.708333⎫⎪

⎢ 0.025 0.05 0.025 ⎥ ⎨ 1 ⎬ = ⎨0.085714⎬ = 0.085714⎨ 1 ⎬
⎢⎣0.0125 0.025 0.0375⎥⎦ ⎪⎩0.71428571⎪⎭ ⎪⎩0.060714⎪⎭ ⎪⎩0.708333⎪⎭

0.085714 − 0.0875
εa = × 100% = 2.08%
0.085714

The iterations can be continued. After 10 iterations, the relative error falls to 0.00000884%
with the result

⎧⎪0.70710678⎫⎪
0.085355⎨ 1 ⎬
⎪⎩0.70710678⎪⎭

Thus, the smallest eigenvalue is 1/0.085355 = 11.71573.

20.2 (a) Minors:

(2 − λ ) 3 −4 λ 4 8 4 8 3 − λ = − λ3 + 10λ2 + 101λ + 18
7 − λ − 2 10 7 − λ + 10 10 4

(b) The eigenvalues can be determined by finding the roots of the characteristic polynomial
determined in (a). This can be done in MATLAB,

223
>> a = [1.0000 -10.0000 -101.0000 -18.0000];
>> roots(a)

ans =
16.2741
-6.0926
-0.1815

(c) The power method for the highest eigenvalue can be implemented with MATLAB
commands,

>> A = [2 2 10;8 3 4;10 4 5];

>> x = [1 1 1]';

First iteration:
>> x = A*x
x =
14
15
19

>> e = max(x)
e =
19

>> x = x/e
x =
0.7368
0.7895
1.0000

Second iteration:
>> x = A*x
x =
13.0526
12.2632
15.5263

>> e = max(x)
e =
15.5263

>> x = x/e
x =
0.8407
0.7898
1.0000

Third iteration:
>> x = A*x
x =
13.2610
13.0949
16.5661

>> e = max(x)
e =

224
 16.5661

>> x = x/e
x =
0.8005
0.7905
1.0000

Fourth iteration:
>> x = A*x
x =
13.1819
12.7753
16.1668

>> e = max(x)
e =
16.1668

>> x = x/e
x =
0.8154
0.7902
1.0000

Thus, after four iterations, the result is converging on a highest eigenvalue of 16.2741 with
a corresponding eigenvector of [0.811 0.790 1].

(d) The power method for the lowest eigenvalue can be implemented with MATLAB
commands,

>> A = [2 2 10;8 3 4;10 4 5];

>> x = [1 1 1]';
>> AI = inv(A)

AI =
-0.0556 1.6667 -1.2222
-0.0000 -5.0000 4.0000
0.1111 0.6667 -0.5556

First iteration:
>> x = AI*x

x =
0.3889
-1.0000
0.2222

>> [e,i] = max(abs(x))

e =
1
i =
2

>> x = x/x(i)
x =
-0.3889
1.0000
-0.2222

225
 Second iteration:
>> x = AI*x
x =
1.9599
-5.8889
0.7469

>> [e,i] = max(abs(x))

e =
5.8889
i =
2

>> x = x/x(i)
x =
-0.3328
1.0000
-0.1268

Third iteration:
>> x = AI*x
x =
1.8402
-5.5073
0.7002

>> [e,i] = max(abs(x))

e =
5.5073
i =
2

>> x = x/x(i)
x =
-0.3341
1.0000
-0.1271

Thus, after three iterations, the estimate of the lowest eigenvalue is converging on the
correct value of 1/(−5.5085) = −0.1815 with an eigenvector of [−0.3341 1 -0.1271].

20.3 MATLAB can be used to solve for the eigenvalues with the polynomial method. First, the
matrix can be put into the proper form for an eigenvalue analysis by bringing all terms to
the left-hand-side of the equation.

⎡ 4 − 9λ 7 3 ⎤ ⎧⎪ x1 ⎫⎪
⎢ 7 8 − 4λ 2 ⎥⎨x2 ⎬ = 0
⎢⎣ 3 2 1 − 2λ ⎥⎦ ⎪⎩ x3 ⎪⎭

Then, each row can be divided by the coefficient of λ in that row.

⎡0.4444 − λ 0.7778 0.3333 ⎤ ⎧⎪ x1 ⎫⎪

⎢ 1.75 2−λ 0 .5 ⎥ ⎨ x 2 ⎬ = 0
⎢⎣ 1 .5 1 0.5 − λ ⎥⎦ ⎪⎩ x3 ⎪⎭

226
 MATLAB can then be used to determine the eigenvalues as the roots of the characteristic
polynomial,

>> A=[4/9 7/9 3/9;7/4 8/4 2/4;3/2 2/2 1/2];

>> p=poly(A)
p =
1.0000 -2.9444 -0.2500 0.2917

>> e=roots(p)
e =
2.9954
-0.3386
0.2876

20.4 (a) MATLAB can be used to solve for the eigenvalues with the polynomial method. First,
the matrix can be put into the proper form by dividing each row by 0.36.

>> A = [2/.36 -1/.36 0 0;-1/.36 2/.36 -1/.36 0;0 -1/.36 2/.36 -

1/.36;0 0 -1/.36 2/.36]

A =
5.5556 -2.7778 0 0
-2.7778 5.5556 -2.7778 0
0 -2.7778 5.5556 -2.7778
0 0 -2.7778 5.5556

Then, the poly function can be used to generate the characteristic polynomial,

>> p = poly(A)
p =
1.0000 -22.2222 162.0370 -428.6694 297.6871

The roots of this equation represent the eigenvalues,

>> e = roots(p)
e =
10.0501
7.2723
3.8388
1.0610

(b) The power method can be used to determine the highest eigenvalue:

>> A = [2/.36 -1/.36 0 0;

-1/.36 2/.36 -1/.36 0;
0 -1/.36 2/.36 -1/.36;
0 0 -1/.36 2/.36];
>> x = [1 1 1 1]';

First iteration:

>> x = A*x
x =

227
 2.7778
0
0
2.7778
>> e = max(x)
e =
2.7778
>> x = x/e
x =
1
0
0
1

Second iteration:

>> x = A*x
x =
5.5556
-2.7778
-2.7778
5.5556
>> e = max(x)
e =
5.5556
>> x = x/e
x =
1.0000
-0.5000
-0.5000
1.0000

Third iteration:

>> x = A*x
x =
6.9444
-4.1667
-4.1667
6.9444
>> e = max(x)
e =
6.9444
>> x = x/e
x =
1.0000
-0.6000
-0.6000
1.0000

The process can be continued. After 9 iterations, the method does not converge on the
highest eigenvalue. Rather, it converges on the second highest eigenvalue of 7.2723 with a
corresponding eigenvector of [1 −0.6180 −0.6180 1].

228
(c) The power method can be used to determine the lowest eigenvalue by first determining
the matrix inverse:

>> A = [2/.36 -1/.36 0 0;-1/.36 2/.36 -1/.36 0;0 -1/.36 2/.36 -

1/.36;0 0 -1/.36 2/.36];
>> AI = inv(A)
AI =
0.2880 0.2160 0.1440 0.0720
0.2160 0.4320 0.2880 0.1440
0.1440 0.2880 0.4320 0.2160
0.0720 0.1440 0.2160 0.2880
>> x = [1 1 1 1]';

First iteration:

>> x = AI*x
x =
0.7200
1.0800
1.0800
0.7200
>> e = max(x)
e =
1.0800
>> x = x/e
x =
0.6667
1.0000
1.0000
0.6667

Second iteration:

>> x = AI*x
x =
0.6000
0.9600
0.9600
0.6000
>> e = max(x)
e =
0.9600
>> x = x/e
x =
0.6250
1.0000
1.0000
0.6250

Third iteration:

>> x = AI*x
x =
0.5850
0.9450

229
 0.9450
0.5850
>> e = max(x)
e =
0.9450
>> x = x/e
x =
0.6190
1.0000
1.0000
0.6190

The process can be continued. After 9 iterations, the method converges on the lowest
eigenvalue of 1/0.9450 = 1.0610 with a corresponding eigenvector of [0.6180 1 1 0.6180].

20.5 The parameters can be substituted into force balance equations to give

(0.45 − ω )X2
− 0.2 X2 =0
( )
1
− 0.24 X 1 + 0.42 − ω 2 X 2 − 0.18 X3 = 0
(
− 0.225 X 2 + 0.225 − ω 2 X 3 = 0 )
A MATLAB session can be conducted to evaluate the eigenvalues and eigenvectors as

>> A = [0.450 -0.200 0.000;-0.240 0.420 -0.180;0.000 -0.225 0.225];

>> [v,d] = eig(A)

v =
-0.5879 -0.6344 0.2913
0.7307 -0.3506 0.5725
-0.3471 0.6890 0.7664
d =
0.6986 0 0
0 0.3395 0
0 0 0.0569

Therefore, the eigenvalues are 0.6986, 0.3395 and 0.0569. The corresponding eigenvectors
are (normalizing so that the amplitude for the third floor is one),

⎧⎪ 1.693748 ⎫⎪ ⎧⎪− 0.92075⎫⎪ ⎧⎪0.380089 ⎫⎪

⎨− 2.10516 ⎬ ⎨− 0.50885⎬ ⎨0.746999 ⎬
⎪⎩ 1 ⎪⎭ ⎪⎩ 1 ⎪⎭ ⎪⎩ 1 ⎪⎭

A graph can be made showing the three modes

230
 -2 -1 0 1 2 -1 0 1 0 1

Mode 1 Mode 2 Mode 3

20.6 As was done in Section 20.2, assume that the solution is ij = Ij sin(ωt). Therefore, the second
derivative is

d 2i j
2
= −ω 2 I j sin(ωt )
dt

Substituting these relationships into the differential equations gives

1
− L1ω 2 I 1 sin(ωt ) + ( I 1 sin(ωt ) − I 2 sin(ωt )) = 0
C1
1 1
− L2ω 2 I 2 sin(ωt ) + ( I 2 sin(ωt ) − I 3 sin(ωt )) − ( I 1 sin(ωt ) − I 2 sin(ωt )) = 0
0.001 0.001
1 1
− L3ω 2 I 3 sin(ωt ) + i3 − ( I 2 sin(ωt ) − I 3 sin(ωt )) = 0
0.001 0.001

All the sin(ωt) terms can be cancelled. In addition, the L’s and C’s are constant. Therefore,
the system simplifies to

⎡1 − λ −1 0 ⎤ ⎧⎪ I 1 ⎫⎪
⎢ −1 2−λ − 1 ⎥ ⎨I 2 ⎬ = 0
⎢⎣ 0 −1 2 − λ ⎥⎦ ⎪⎩ I 3 ⎪⎭

where λ = LCω2. The following MATLAB session can then be used to evaluate the
eigenvalues and eigenvectors

>> a = [1 -1 0;-1 2 -1;0 -1 2]

>> [v,d] = eig(a)
v =
-0.7370 -0.5910 0.3280
-0.5910 0.3280 -0.7370
-0.3280 0.7370 0.5910
d =
0.1981 0 0

231
 0 1.5550 0
0 0 3.2470

The matrix v consists of the system's three eigenvectors (arranged as columns), and d is a
matrix with the corresponding eigenvalues on the diagonal. Thus, MATLAB computes that
the eigenvalues are λ = 0.1981, 1.5550, and 3.2470. These values in turn can be used to
compute the natural frequencies for the system

⎧ 0.4450 ⎫
⎪ ⎪
⎪ LC ⎪
⎪ 1.2470 ⎪
ω =⎨ ⎬
⎪ LC ⎪
⎪ 1.8019 ⎪
⎪ LC ⎪
⎩ ⎭

20.7 The force balances can be written as

⎡m1 0 0 ⎤ ⎧ &x&1 ⎫ ⎡ 2k −k − k ⎤ ⎧⎪ x1 ⎫⎪
⎢0 ⎪ ⎪
m2 0 ⎥ ⎨ &x&2 ⎬ + ⎢− k 2k − k ⎥⎨x2 ⎬ = 0
⎢0 m3 ⎥⎦ ⎪⎩ &x&3 ⎪⎭ ⎢⎣− k −k 2k ⎥⎦ ⎪⎩ x3 ⎪⎭
⎣ 0

Assuming that the solution is xi = Xi sin(ωt), we get the following matrix

⎡2k − m1ω 2 −k −k ⎤⎧ X ⎫
⎢ ⎥⎪ 1 ⎪
⎢ −k 2 k − m 2ω 2 −k ⎥⎨ X 2 ⎬ = 0
2 ⎪X ⎪
⎢⎣ −k −k 2k − m3ω ⎥⎦ ⎩ 3 ⎭

Using MATLAB,

>> k = 1;
>> kmw2 = [2*k,-k,-k;-k,2*k,-k;-k,-k,2*k];
>> [v,d] = eig(kmw2)

v =
0.8034 0.1456 0.5774
-0.2757 -0.7686 0.5774
-0.5278 0.6230 0.5774

d =
3.0000 0 0
0 3.0000 0
0 0 0.0000

Therefore, the eigenvalues are 0, 3, and 3. Setting these eigenvalues equal to mω 2 , the
three frequencies can be obtained.

mω 1 = 0 ⇒ ω 1 = 0 (Hz) 1st mode of oscillation

2

mω 2 = 0 ⇒ ω 2 = 3 (Hz) 2nd mode

2

232
 mω 3 = 0 ⇒ ω 3 = 3 (Hz) 3rd mode
2

20.8 The pair of second-order differential equations can be reexpressed as a system of four first-
order ODE’s,

dx1
= x3
dt
dx 2
= x4
dt
dx 3
= −5 x1 + 5( x 2 − x1 )
dt
dx 4
= −5( x 2 − x1 ) − 5 x 2
dt

An M-file can be set up to evaluate the right-hand side of these ODEs:

function dx = dxdt(t, x)
dx = [x(3);x(4);-5*x(1)+5*(x(2)-x(1));-5*(x(2)-x(1))-5*x(2)];

(a) x1 = x2 = 1

>> tspan = [0,10];

>> y0 = [1,1,0,0];
>> [t,y] = ode45('dxdt',tspan,y0);
>> plot(t,y(:,1),t,y(:,2),'--')
>> legend('x1','x2')

Because we have set the initial conditions consistent with one of the eigenvectors, the two
masses oscillate in unison.

(b) x1 = 1, x2 = –0.6

>> tspan=[0,10];
>> y0=[1,-0.6,0,0];

233
 >> [t,y]=ode45('dxdt',tspan,y0);
>> plot(t,y(:,1),t,y(:,2),'--')
>> legend('x1','x2')

Now, because the initial conditions do not correspond to one of the eigenvectors, the
motion involves the superposition of both modes.

20.9

function [e, v] = powmax(A)

% [e, v] = powmax(A):
% uses the power method to find the highest eigenvalue and
% the corresponding eigenvector
% input:
% A = matrix to be analyzed
% output:
% e = eigenvalue
% v = eigenvector

es = 0.0001;
maxit = 100;
n = size(A);
for i=1:n
v(i)=1;
end
v = v';
e = 1;
iter = 0;
while (1)
eold = e;
x = A*v;
[e,i] = max(abs(x));
e = sign(x(i))*e;
v = x/e;
iter = iter + 1;
ea = abs((e - eold)/e) * 100;
if ea <= es | iter >= maxit, break, end
end

234
 Application to solve Prob. 20.2,

>> A = [2 2 10;8 3 4;10 4 5];

>> [e, v] = powmax(A)
e =
16.2741
v =
0.8113
0.7903
1.0000

20.10

function [e, v] = powmin(A)

% [e, v] = powmin(A):
% uses the power method to find the lowest eigenvalue and
% the corresponding eigenvector
% input:
% A = matrix to be analyzed
% output:
% e = eigenvalue
% v = eigenvector

es = 0.0001;
maxit = 100;
n = size(A);
for i=1:n
v(i)=1;
end
v = v';
e = 1;
Ai = inv(A);
iter = 0;
while (1)
eold = e;
x = Ai*v;
[e,i] = max(abs(x));
e = sign(x(i))*e;
v = x/e;
iter = iter + 1;
ea = abs((e - eold)/e) * 100;
if ea <= es | iter >= maxit, break, end
end
e = 1./e;

Application to solve Prob. 20.2,

>> [e, v] = powmin(A)

e =
-0.1815
v =
-0.3341
1.0000
-0.1271

235