§1.

0 PARTIAL DIFFERENTIATION

1.1 FUNCTIONS OF SEVERAL VARIABLES

Functions of several variables occur frequently in Engineering. For example, if we have

a right circular cone of altitude h and radius of base r , its volume and lateral surface

area are, respectively V  r 2 h and S  r r 2  h 2 .
3
Here V and S are functions of the two variables r, h . Also the law of cosine expresses
the length c of the third side of a triangle as a function of the lengths a, b of the other
two sides and of the included angle  . The formula expressing c as a function of three

variables is c  a 2  b 2  2ab cos   2 .

1

A function of two variables is defined as follows. Let D be a collection of number pairs

( x, y) , and suppose that with each pair ( x, y) is associated a unique number z , thus
giving us a certain collection of a number triples ( x, y, z ) . This collection is called a
function of two variables. If we denote the function by a single letter, say f , then we
write z  f ( x, y) and call z the value of f at ( x, y) . Where x and y are independent
variables and z is called the dependent variable. The collection of all the values of
f is called the range of f , while the collection D of allowable pairs ( x, y) is called
the domain of f .
Functions of three or more variables are defined in a similar manner.

1.2 PARTIAL DERVATIVES

The ideas of calculus apply to functions of several variables as well as to functions of
one variable. Of course, because more variables are involved, the notation and technical
detail are more complicated but the essential ideas are the same.

For function of one variable, f (x) , the derivative is defined by

df lim f lim  f x  x   f x 
   .
dx x  0 x x  0  x

1
and this is a measure of the rate of change of the value of the function f (x) with respect
to its variable x .
For a function of several variables, it is also useful to know how the function changes
when one, some or all of the variables change. To achieve this we define the partial
derivative of a function.

First, we consider a function f ( x, y) of the two variables x and y . The partial

f
derivative, , of f ( x, y) with respect to x is its derivative with respect to x treating
x
the values of y as being constant.

f  df  lim  f x  x, y   f x, y 

Thus,    .
x  dx  y CONSTANT x  0  x 

f
Likewise, the partial derivative, , of f ( x, y) with respect to y is its derivative with
y
respect to y treating the value of x as being constant, so that:

f  df  lim  f x, y  y   f x, y 

  
y  dy  x CONSTANT y  0  y 


The process of obtaining the partial derivatives is called partial differentiation.

Note the use of ‘curly dees’, is to distinguish between partial differentiation and
ordinary differentiation.
A concise notation is sometimes used for partial derivatives; as an alternative to the
f f
‘curly dees’, written as f x  and f y  .
x y
If we write z  f ( x, y) then the partial derivatives may also be written as
z z
, or z x , z y .
x y

Example1:

2
Although, partial derivatives have been introduced here in the context of function of
two variables, the concept may be readily extended to obtain the partial derivatives of
a function of as many variables as we please.
Finding partial derivatives is no more difficult than finding derivatives of functions of
one variable, with the constant multiplication, sum, product and quotient rules having
counterparts for partial derivatives.

Example 2:

1.3 HIGHER-ORDER PARTIAL DERIVATIVE

For function of two variables, there are four different second-order partial derivatives.
f   f  2 f
The partial derivative with respect to x of is   , usually abbreviated as
x x  x  x 2
or f xx . Similarly, taking two successful partial derivatives with respect to y gives us

  f   2 f
   f yy .
y  y  y 2
For mixed second-order partial derivative, one derivative is taken with respect to
  f 
each variable. If the first partial derivative is taken with respect to x , we have  
y  x 

2 f
, abbreviated as or  f x y  f xy . If the first partial derivative is taken with respect
yx

  f  2 f
to y , we have   , abbreviated as , or  f y x  f yx .
x  y  xy

Example 3

Theorem 1
If f xy ( x, y) and f yx ( x, y) are continuous on an open set containing (a, b) , then

f xy (a, b)  f yx (a, b) .

3
Third -, fourth -, or even higher-order partial derivative can be taken. Theorem 1 can
be extended to show that as long as the partial derivative are all continuous in an open
set, the order ( arrangement) of differentiation of higher derivative does not matter.
3 f
With higher-order partial derivative, notation such as becomes quite awkward
xyx
and so, f xyx is usually used instead.

Exercises

1.4 TOTAL DIFFERENTIAL

Consider a function u  f ( x, y) of two variables x and y . Then the corresponding

increment in u is given by u  f ( x  x, y  y)  f ( x, y) .
This can be rewritten as:
u   f ( x  x, y  y)  f ( x, y  y)   f ( x, y  y)  f ( x, y) .
Dividing the first bracketed term by x and the second by y , gives:
f ( x  x, y  y)  f ( x, y  y) f ( x, y  y )  f ( x, y)
x  y
x y
From the definition of the partial derivative we may approximate this expression by
f f
u  x  y .
x y
This could be written as
f f
du  dx  dy .
x y
We see that the differential du is an approximation to the change u in u  f ( x, y)
resulting from small changes x and y in the independent variables x and y, that is
f f f f
u  du  dx  dy  x  y
x y x y
This extends to functions of as many variables as we please, provided that the partial
derivative exists. For example, for a function of three variables x, y, z defined by
u  f ( x, y, z) we have

4
 f f f
u  du  dx  dy  dz
x y z
f f f
 x  y  z
x y z
The differential of a function of several variables is often called a total differential,
emphasizing that it shows the variation of the function with respect to small changes in
all independent variables.

1.5 CHAIN RULE

The rules and results of ordinary differentiation are carried over to partial
differentiation. In particular, the composite-function rule still hold, but in a modified
form.
Consider the two-variable case where z  f ( x, y) and x and y are themselves
function of two independent variables, s and t . Then z itself is also a function of s
and t say F ( s, t ) , and we can find its derivative using a composite-function rule that
gives the rate of change of z with respect to s and t in terms of the rate of change of
z with respect to x and y and the rate of change of x and y with respect to s and t
. Thus
z z x z y z z x z y
  and  
s x s y s t x t y t

Example 5’s:

5
1.6 IMPLICIT DIFFERENTIATION
For a function of single variable, sometimes f is defined as a function of x , not by
giving the value of y explicitly in terms of x , but by giving an equation in x and y .
Such an equation may not determine y uniquely in terms of x , but in the situations we
commonly meet, the equation determines one or more distinct functions.

For instance, the equation x 2  y 2  16  0 determines two functions of x :

y  16  x 2 and y   16  x 2

Also, the equation 5x 2  6 xy  5 y 2  128 determine two function of x , which we find

by solving for y by the quadratic formula:

y

6 x  36 x 2  45 5 x 2  128 
10

or y 
1
5

3x  4 40  x 2  and y
1
5
3x  4 40  x 2 
When an equation in x and y determines y as a function of x (or as one of several
functions of x ), but when we do not have a direct explicit formula of y in terms of x
we say that y is defined implicitly as a function of x by the equation.

When y is a differentiable function of x which is defined implicitly by an equation of

dy
a suitable type, we can calculate the derivative directly from the equation without
dx
solving for y explicitly. The derivative will usually be expressed in terms of both x
dy
and y rather than in terms of x alone. In finding we use the composite function
dx
(chain) rule and regard y as a function of x whenever it appears.
For instance
d 2
dx
 
y  2y
dy
dx
d 2 3
dx
 
x y  x2 3y 2
dy
dx
 2 xy 3

6
Examples

Now, we desire to obtain a general formula for obtaining the differentials of implicit
function. Consider the implicit function of two variables. The relation f ( x, y, z ) may
define any one of the variables as an implicit function of the other two. Let z be the
dependent variable. Then
z z
dz  dx  dy -------------------------------------------- (1)
x y

f f f
df  dx  dy  dz  0 ------------------------------------------- (2)
x y z

Substituting dz in equation (1) into equation (2) gives

 f f z   f f z 
  dx    dy  0 ----------------------------------------- (3)
 x z x   y z y 

Inasmuch as x and y are independent and equation (3) must hold for all values of dx
and dy , it follows that the coefficient (in parentheses) of dx and dy must individually
f
be zero. If  0 , this condition gives
z
df
z dx Fx
 - =
x df Fz
dz
df
z dy Fy
Similarly, = - =
y df Fz
dz

Example

Note that, much like implicit differentiation with two variables, implicit differentiation
with three variables yields expression for the derivatives that depend on all three
variables.

7
§2.0 SERIES REPRESENTATION OF FUNCTIONS

2.1 POWER SERIES

The basic mathematics involved in power series is a natural extension of the series of

the type: a
n 0
n x n  a0  a1 x  a 2 x 2  a3 x 3  ...  a n x n  ...

Where a0 , a1 , a2 ,... are independent of x , is called a power series. We refer to the

constants, ak (k  0,1,2,...) as the coefficient of the series.

Convergence of power series

Power series will, in general converge for certain values of x and diverge elsewhere.
When the sum S n of a series of n terms tends to a limit as n   , the series is
convergent. Applying d’Alembert’s ratio test to the above series, we see that it is
lim a n 1 x n 1
absolutely convergent when: 1
n   an x n

Thus the series converges if

lim an 1
x < 1
n   an

lim an 1
That is, if x <
n   an

lim an
Denoting by r, we see that the series is absolutely convergent for
n   an 1
 r < x < r and divergent for x < r and x > r . The limit r is called the radius of
convergent of the series.

NOTE: D’Alembert’s ratio test.

 lim u n 1
Suppose we have a series of positive terms  u k , and also  l exists.
k o n   un

Then the series is convergent if l < 1 and divergent if l > 1 . If l  1 , we are not able to
decide, using this test, whether the series converges or diverges.

2.2 REPRESENTATION OF FUNCTIONS USING SPECIAL POWER SERIES

8
The primary reason for studying series is that we can use them to represent functions.
This opens up all kinds of possibilities for us, from approximating the values of
transcendental functions to calculating derivatives and integrals of such functions, to
studying differential equations.

Power series may be added, multiplied and divided within their common domains of
convergence (provided the denominator is nonzero within the common domain) to give
power series that are convergent, and the properties are often exploited to express a
given power series in terms of standard series band to obtain power series expansion of
complicated functions.
Four elementary power series that are of wide spread use are:
(a) Geometric series
1
 1  x  x 2  x 3  ....................  (1) n x n  ........................
1 x
 1  x  1
(b) The binomial Series
r r r r n
(1  x) r  1    x    x 2    x 3  ..........    x  ...........
1  2 1 n
 r  r (r  1)......................(r  n  1)
Where   
n 1  2  3......................n
is the binomial coefficient.

r is any real number. When r is a positive integer, say N, the series terminates at the
term x N . When r is not a positive integer the series does not terminate.
Setting r  -1 gives
1 (1) (1) 2 2 (1)(2)(3) 3
(1  x) 1   1 x x  x  .......
1 x 1 1 2 1 2  3
which simplifies to the geometric series
1
 1  x  x 2  x 3  .............
1 x
Similarly,
(2) (2) 3 2 (2)(3)(4) 3
(1  x) 2  1  x x  x  .................
1 1 2 1 2  3
which simplifies to the arithmetic-geometric series

9
 1
 1  2 x  3x 2  4 x 3  .................
(1  x) 2

So the geometric series may be thought of as a special case of the binomial series.

(c) The exponential series

x x2 x3 xn
ex  1    ...........   ...... (all x )
1! 2! 3! n!
The number e is defined by
n

lim x
e  1  
n  n

Similarly, the function e x is defined by

n
 1
lim
e  1  
n  n

Using the binomial expansion, we have


 n  x  n(n - 1)  x 
lim
2
n(n - 1)(n - 2)  x 
3


e 
x
1           ......................
n 
 1n 1 2  n  1 2  3 n 


 1

lim 1- 1
n

1 - 1 (1 - 2
n
     
n X 3  ......................
e  1  X  X 
2

x

n  1 1 2 1 2  3 
 

x x 2 x3
 1    ...........
1! 2! 3!

So we see the connection between binomial and exponential series.

(d) The logarithmic series

n 1
x2 x3 x4 n x
ln(1  x)  x     ..............  (1)  .............. (-1< x <1)
2 3 4 n 1

Example

Exercise

10
using Maclaurin series obtain the four elementary power series above.

2.3 TAYLOR SERIES

Taylor series is a method of constructing polynomial approximations to functions that

are not polynomials. The higher derivatives of all orders will play a pivotal role. They
are essential means for exploring and computing with transcendental functions (e.g.
sin x , cos x , ln x , e x )


Suppose that the power series a x
n 0
n
n
has radius of convergence r  0 . This means

that the series converges absolutely to some function f on the interval (r , r ) . We have


f (x) = a
n 0
n x n  a0  a1 x  a 2 x 2  a3 x 3  a 4 x 4 .............. for each (r , r )

Differentiating term by term, we get


f ( x)   an nx n 1  a1  2a2 x  3a3 x 2  4a4 x 3  ...... , again for each x  (-r, r) .
n 0

Likewise, we get


f ( x)   a n n(n  1) x n2  2a 2  3  2a3 x  4  3a 4 x 2  ...... and
n 0


f ( x)   a n n(n  1)(n  2) x n3  3  2a3  4  3  2a 4 x  ......
n 0

and so on (all valid for - r,  x  r ).

Notice that if we substitute x  0 in each of the above derivatives all the terms of the
series drop out, except one. We get

f (0)  a0 ,

f (0)  a1 ,

11
 f (0)
f (0)  2a2 = 2!a2  a2 
2!

f (0)
f (0)  3  2a3 = 3!a3  a3 
3!

And so on. Observe that in general, we have

f n (0)  n!an , solving for an we have

f n (0)
an  for n  0,1,2,3,........
n!


Thus, we found that if a x
n 0
n
n
is a convergence power series with radius of

convergence r  0 , then the series converges to some function f that we can write as

 
f n (0) n f (0) 2
f (x) =  an x , =
n 0
n

n 0 n!
x = f (0)  f (0) x 
2!
x  ....... for x  (-r,  r)

Instead of starting with a series, suppose that you start with an infinitely differentiable
function, f (i.e. f can be differentiated infinitely often). Then we can construct the
series


f n (0) n

n0 n!
x

called a TAYLOR SERIES expansion of f about x  0 .

We have constructed polynomials to approximate a given function near 0 by taking the

functions and their derivatives at x  0 . This is also called Maclaurin series only when
the expansion is about x  0 .

But we can do the same thing near any number a . Instead of expressing the
approximating polynomial in terms of powers of x , it will be convenient to write it in
terms of x  a . To find the formula for these polynomials, we will imitate for a what
we did at 0. That is for the power series

12
 
f (x) =  a ( x  a)
n 0
n
n
= a0  a1 ( x  a)  a2 ( x  a) 2  a3 ( x  a)3  a4 ( x  a) 4  ........

for each x  (a  r , a  r ) . Obtaining f (a) and differentiating term by term, at x  a , we

get for the function and derivatives


f (a)   an ( x  a) n  a0 = 0! at xa
n 0


f (a) =  a n( x  a )
n 0
n
n 1
= a1 = 1! a1 at xa


f (a) =  an n(n  1)( x  a) n 2 = 2a2 = 2! a2 at xa
n 0


f (a) =  a n(n  1)(n  2)( x  a)
n 0
n
n 3
= 3  2a3 = 3!a3 at xa

and so on. Thus in general, we have

f n (a)  n!an

solving for an , we have

f n (a)
an  , for n  0,1,2,........
n!

Hence, in general if the function f has derivatives through order n at a , then the nth-
order Taylor polynomial of f at a is defined as

f (a) f n (a)
pn ( x; a)  f (a)  f (a)( x  a)  ( x  a) 2  .........  ( x  a) n
2! n!

A Taylor polynomial for f at a is a partial sum of the Taylor series generally defined
as


f n ( a ) x  a  f (a)
n
f n (a)

n 0 n!
 f (a)  f (a)( x  a) 
2!
( x  a) 2  ......... 
n!
( x  a) n  ..... .

13
This series is called the Taylor series at x  a associated with the function f (x) when
a  0 , the series is also called Maclaurin series associated with f (x)

Example:

2.4 THE ERROR IN TAYLOR SERIES (TAYLOR THEOREM)

The Remainder Rn ( x; a)

Definition: The remainder (error) Rn ( x; a)

Let f be a function and let Pn ( x; a) be the associated nth  order Taylor polynomial at

a . The number Rn ( x; a) defined by the equation

f (x) = Pn ( x; a) + Rn ( x; a) That is,

Rn ( x; a) = f (x) - Pn ( x; a)

is called the remainder or error in using the Taylor polynomial Pn ( x; a) to

approximate f (x)

If the remainder Rn ( x; a) does approach 0 as n   , then the function f (x) is

represented by its Taylor series. The following theorem expresses Rn ( x; a) in terms of

a derivative (this is called “derivative” or “Lagrange” form of the remainder, it is also
known as Taylor’s theorem). This formula enables us to show that for f (x) = e x , sin x
, or cos x , Rn ( x; a) indeed approaches 0 as n   . It also enables us to obtain an
estimate of the error in using a Taylor polynomial to approximate a given function.

Theorem: Lagrange’s form of remainder (or Taylor’s theorem)

Assume that a function f (x) has continuous derivatives of orders through n  1 in an

interval that includes the numbers a and x . Let Pn ( x; a) be the n th Taylor polynomial

associated with f (x) in powers of x  a . Then there is a number Cn between a and x

such that

14
 f n 1(Cn )
Rn ( x; a) = ( x  a) n 1
(n  1)!

§3.0 ETREMUM PROBLEMS

In many industrial situations the role of management is to make decisions that will lead
to the most effective use of the resources available. Effective management seeks to
optimize the constituents parts of the whole operation. A wide variety of mathematical

15
techniques is used to solve such optimization problems, including some based on the
use of calculus, which we shall consider here.

3.1 Optimal values:

Figure.3.1

A basic idea is that the optimal value of a differentiable function f (x) (that is, its
maximum or minimum value) generally occurs where its derivative is zero; that is,
where f (x) = 0.
As can be seen from figure 1, this is a necessary condition since at a maximum or
minimum value of the function its graph has a horizontal tangent. Figure 1 does,
however, show that these extremal values are generally only local maximum or
minimum values, corresponding to turning points on the graph. Hence, in seeking the
extremal values of a function it is also necessary to check the end points, (if any) of the
domain of the function.
Having determined the critical or stationary point where f (x) = 0, we need to be
able to determine their character or nature; that is , whether they correspond to a local
minimum, a local maximum or a point of inflection of the function f (x) . We can do
this by examining values of f (x) close to and on either side of the critical point. We
see that from figure 2.

16
Figure3.2.

 If the value of f (x) , the slope of the tangent, changes from positive to negative
as we pass from left to right through a stationary point then the later correspond
to a local maximum
 If the value of f (x) changes from negative to positive as we pass from left to
right through a stationary point then the later correspond to a local minimum.
 If f (x) does not change sign as we pass through a stationary point then the
later corresponds to a point of inflection.

An alternative approach to determining the nature of a stationary point is to

calculate the value of the second derivative f (x) at the point. Note that f (x)
determines the rate of change of f (x) . Suppose that f (x) has a stationary point at
x  a , so that f (a)  0 . Then provided f (a) is defined, either f (a)  0 ,
f (a)  0 or f (a)  0 .

If f (a)  0 then f (x) is decreasing at x  a , and since f (a)  0 , it follows that

f (x)  0 for values of x just greater than a . We therefore conclude that x  a
corresponds to a local maximum. Since the graph is concave down at a local
maximum, f (a)  0 .

Similarly, we can argue that if f (a)  0 , then the stationary point x  a

corresponds to a local minimum.

Summarizing, we have
 The function f (x) has a local maximum at x  a provided f (a)  0 and
f (a)  0

17
  The function f (x) has a local minimum at x  a provided f (a)  0 and
f (a)  0

If f (a)  0 , we cannot assume that x  a correspond to a point of inflection, and we

must revert to considering the sign of x  f (x) on either side of the stationary point.
Example.

3.2 EXTREMA OF FUNCTIONS OF SEVERAL VARIABLES

Definition:
For a function of two variables, we call f (a, b) a local maximum of f if there is
an open disk R centred at (a, b) , for which f (a, b)  f ( x, y) for all ( x, y)  R .
Similarly, f (a, b) is called a local minimum of f if there is an open disk R
centred at (a, b) , for which f (a, b)  f ( x, y) for all ( x, y)  R . In either case
f (a, b) is called a local extremum of ** sometimes referred to as relative
extremum of f

The idea here is the same as it was in function of single variable. That is, if
f (a, b)  f ( x, y) for all ( x, y) “near” (a, b) , we call f (a, b) a local maximum.
Much as with function of one variable, it turns out that local extrema can occur
where the first (partial) derivatives are zero or do not exist. However, having zero
first partial derivatives at a point does not by itself guarantee a local extremum.

Definition:
The point (a, b) is a critical point of the function f ( x, y) if (a, b) is in the domain
f f f f
of f and either ( a, b) = (a, b) =0 or one or both of and do not exist
x y x y

at (a, b) .

18
If (a, b) is a local extremum (local maximum or local minimum), then (a, b) must
be a critical point of f . Although local extrema can only occur at critical points,
every critical point need not correspond to a local extremum.

Proof:
Suppose that f ( x, y) has a local maximum at (a, b) . Holding y constant at y  b ,
notice that the function g ( x)  f ( x, b) has a local maximum at x  a . This implies
f
either g (a)  0 or g (a) doesn’t exist. Note that g (a) = (a, b) . Likewise,
x
holding x constant at x  a , observe that the function h( y)  f (a, y) has a local

maximum at y  b . It follows that h(b)  0 or h(b) doesn’t exist. Note that h(b)
f f
= (a, b) . Combining these two observations, we have that each of (a, b) and
y x
f
(a, b) equals zero or doesn’t exist. We can then conclude that (a, b) must be a
y
critical point of f . A nearly identical argument shows that if f ( x, y) has a local
minimum at (a, b) then (a, b) must be a critical point of f .

When looking for local extrema, you must first find all critical points to determine
if it is the location of a local maximum, local minimum or neither, since local
extrema can only occur at critical points. To determine the nature of the critical
point of function of several variables, we use a generalisation of the second
derivative test for function of a single variable.

Theorem: (Second derivative test)

Suppose that f  x, y  has continuous second-order partial derivatives in some open

disk containing the point  a, b  and that f x  a, b   f y  a, b   0 . We define the

discriminate D for the point  a, b  by:

D  a, b   f xx  a, b  f yy  a, b    f xy  a, b  .
2

19
(i) If D  a, b   0 and f xx  a, b   0 or f yy  a, b   0 , then f has a local

minimum at  a, b  .

(ii) If D  a, b   0 and f xx  a, b   0 or f yy  a, b   0 , then f has a local

maximum at  a, b 

(iii) If D  a, b   0 then f has a saddle point at  a, b 

(iv) If D  a, b   0 , then no conclusion can be drawn. The point may be a

minimum, maximum or saddle point. Further investigation is required.

When a function along some path through  a, b  has a maximum value while along

other paths through the same point has a minimum value. Such a point is called
saddle point.

Figure3.3

Definition:
The point P  a, b, f  a, b   is a saddle point of Z  f  x, y  if  a, b  is a critical

point of f and if every open disk centred at  a, b  contains points  x, y  in the

domain of f for which f  x, y   f  a, b  and points  x, y  in the domain of f for

which f  x, y   f  a, b  .

20
Note that to have D  a, b   0 , we must have both f xx  a, b   0 and f yy  a, b   0

or both f xx  a, b   0 and f yy  a, b   0 .

3.3 Absolute (or Global) Extrema

Theorem:
Suppose that f  x, y  is continuous on the closed and bounded region R C R 2 .

Then f has both an absolute (global) maximum and an absolute (global) minimum
on R . Further the absolute extrema must occur at either a critical point in R or on
the boundary of R

Now, we shall explain what is meant by a bounded and closed set of points in the
xy -plane. A set of points (i.e, a collection of points) is said to be bounded if it is
entirely contained within some square in the plane. For example, the set of all points

 x, y  such that x 2   y  2   16 is bounded. The set of points  x, y  such that

2

1  x  4 , with no restriction on y , is not bounded. A set S of points is said to be

closed if S contains all points Q such that points of S can be found as close as

one pleases to Q . For example, the set S of points  x, y  such that 0  x  1 and

0  y  1 is not closed, whereas the set of points ** such that ** is closed. Again

the set of points  x, y  such that 0  x 2  y 2  1 is bounded but not closed.

Example

3.4 Optimization of constrained function (Lagrange multiplier)

There are frequent situations in engineering applications when we wish to obtain

the stationary values of functions of more than one variable and for which the
variables themselves are subject to one or more constraint conditions. Here we will
show the technique for solving such problems.

Example: Obtain the extremum value of the function f  x, y   2 x 2  3 y 2

subject to the constraint 2 x  y  1

21
Solution
In this particular example it is easy to eliminate one of the two variables x and y .

Eliminating y , we can write f  x, y  as

f  x, y   f  x   2 x 2  31  2 x 
2

 14 x2  12 x  3
we can now apply the techniques used for functions of one variable to obtain the
extremum value. Differentiating gives:
f   x   28x  12 and f   28

An extremal value occurs when f   x   0 ; that is, x  3 7 , y  1 7 and since

f   3 7   0 , this correspond to a minimum value. Thus the extremum is a

minimum, at x  3 7 , y  1 7 .

In example above we were fortunate in being able to use the constraint equation to
eliminate one of the variables. In practise, however, it is often difficult, or even
impossible to do this, and we have to retain all the original variables.

Let us consider the general problems of obtaining the stationary points of f  x, y 

subject to the constraint g  x, y   k . We shall refer to such point as conditional

stationary points.
This method of solution is a more versatile technique called the method of
Langrange multiplier, in which the introduction of a third variable (multiplier)
enables one to solve constrained optimization problems without first solving the
constraint equation for one of the variables.

More specifically, the method of Langrange multiplier uses the fact that any relative
extremum of the function f  x, y  subject to the constraint g  x, y   k must occur

at a critical point  a, b  of the function:

F  x, y   f  x, y     g  x, y   k 

where  is a new variable called Langrange multiplier. To find the critical point
of F , compute the partial derivatives:

22
Fx  f x   g x Fy  f y   g y F    g  k 

And solve the equations Fx  0 , Fy  0 and F  0 simultaneously as follows

Fx  f x   g x  0 or fx   gx

Fy  f y   g y  0 or fy  gy

F    g  k   0 or g k

Finally evaluate f  a, b  at each critical point  a, b  of F . Then comparing

f  a, b  at the critical points, if f has a constrained maximum (minimum), it will

be given by the largest (smallest) of the critical values  a, b  .

Proof
Although a rigorous explanation of the proof of Lagrange multiplier is beyond the
scope of this course, there is a rather simple geometric argument that is quite
convincing. This argument depends on the fact that for the level curve F  x, y   C

, the slope at each point (using implicit function of several variable method)  x, y 

is given by
dy F
 x
dx Fy

Now, considering the constrained optimization problem:

Maximize f  x, y  subject to g  x, y   k . Geometrically, this means you must find

the highest level curve of f that intersects the constraint curve g  x, y   k . This

critical intersection will occur at a point where the constraint curve is tangent to a
level curve; that is where the slope of the constraint curve g  x, y   k is equal to

the slope of a level curve f  x, y   C . This implies:

Slope of constraint curve = Slope of level curve

gx f
  x
gy fy

or equivalently,

23
 fx f
  y
gx gy

If you let  denote the common ratio, then

fx fy
 and 
gx gy

From which you get the first two Lagrange equations

fx   gx and fy  gy

The third Lagrange equation g  x, y   k is simply a statement of the fact that the

point of tangency actually lies on the constraint curve

24