Differential Calculus of Multivariable Functions 1314A

1 Introduction
Quantities of interest in real problems usually depend on more than one variable.

Examples

(a) The height of the ground at map coordinates  x, y  might be expressed mathematically as

Height  h  x, y  .

(b) The temperature in a lecture room varies with position and time, Temperature = T  x, y, z, t  .

(c) The volume of a rectangular box of length x, width y and height z is V  x, y, z   xyz .

(d) The magnitude of the electric potential u  x, y, z  due to a uniformly charged sphere with total

Q
charge Q is u  x, y, z   , x 2  y 2  z 2  a 2 , where a is the radius of the sphere.
4 0  x  y  z 
2 2 2

These are all real-valued functions of real variables. In general we have z  f  x1 , x2 ,  ,

, xn   f x
T

where x1, x2 , , xn are the independent variables and z is the dependent variable. f x  
T
is a scalar field.

It is possible to visualize a function of two variables (a surface in 3-D space) by drawing its graph. One

way is to draw a contour map. A level curve of f  x, y  is a set of points  x, y  for which f  x, y   c ,

where c is a constant.
A collection of level curves then forms a contour map (as in the familiar contour maps produced by map
makers).

Examples
(a)

1
Level curves of a function are defined in the same manner as one obtains contour lines for a topographical
map.

Contour map

(b)

Contour map

Same level curves for the function f  x, y   x 2  y 2

2
 Level curves shown on the graph of f  x, y   x 2  y 2

(c)

Level curves for the function f  x, y   x 2  y 2

3
 Some level curves on the graph of f  x, y   x 2  y 2

Computer generated graphs usually show lines for which x is constant and lines for which y is constant in
order to give a picture of the surface.

Examples

4
Part (a) shows a computer-generated graph of z  x 2  y 2 . Part (b) shows this graph with level curves
lifted to it.

2 Limit, Continuity and Partial Derivative

The concepts of limit, continuity and derivative naturally generalize from the one variable situation to two
or more variables.

A function f  x, y  is said to possess a limit f 0 at the point  x0 , y0  if for any   0 it is possible to

find a   0 such that whenever 0  x  x0 2   y  y0 2   , that is,  x, y  is in disc of radius 

centered at  x0 , y0  , we have f  x, y   f 0   . We write lim

 x , y  x0 , y0 
f  x, y   f 0 or

lim f  x, y   f 0 .
x  x0
y  y0

Intuitively, the above says that when  x, y  is approaching  x0 , y0  , f  x, y  is approaching f 0 then

the function f  x, y  is said to possess a limit at the point  x0 , y0  .

Note that the limit exists only if, f  x, y  is approaching f 0 independently of the manner in which

5
 x, y  is approaching  x0 , y0  .

Example
lim
( x , y )(1, 2 )
x 2
 3 y 2  xy   12  3  2 2  1  2  15 .

Example
x2
Show that f  x, y   has no limit at the origin.
x2  y2
Solution:
x2
As  x, y    0,0 along the line y  0 , f  x, 0   2  1 and the limiting value is 1.
x

As  x, y    0,0 along the line x  0, f  0, y  and the limiting value is 0.

x2
Hence lim does not exist.
 x , y  0,0  x 2  y 2

x2
z , 1  x  1,  1  y  1
x2  y 2

Example
x2
Show lim 0
 x , y  0,0  1

x 2
y 
2 2

Solution:
Now,

6
 x2  y2
 x 2  y 2 2 .
x2 1
0 
x  y 2 2 x  y 2 2
1 1
2 2

1
x2
We observe that lim
 x , y  0,0
x 2
y 
2 2
 0 . Therefore, lim
 x , y  0,0 1
0
x 2
y 
2 2

A function f  x, y  is said to be continuous at the point  x0 , y0  if and only if it possesses a limit at

( x0 , y0 ) and lim f  x, y   f ( x0 , y0 ) .
 x , y ( x0 , y0 )

Examples

(a) f  x, y   x 2  3 y 2  xy is continuous everywhere.

Solution:
lim
 x , y  x0 , y0 
f  x, y   lim
 x , y  x0 , y0 
x 2
 3 y 2  xy   x0 2  3 y0 2  x0 y0 .

x2
(b) f  x, y   is continuous everywhere except at the origin where f  x, y  is not defined.
x2  y 2
Solution:
x0 2
If  x0 , y0    0,0 then f  x0 , y0  is defined at  x0 , y0  and f  x0 , y0   2 . In addition,
x0  y02
x2 x0 2
lim f  x, y   lim  2  f  x0 , y0  .
 x , y  x0 , y0   x , y  x0 , y0  x 2  y 2 x0  y02

 x2
 1  x, y    0,0 
(c) f  x, y     x 2  y 2  2 is continuous everywhere iff A  0 .

 A  x, y    0,0
Solution:
 x2
x2  1  x, y    0,0 
Since lim  0 , f  x, y     x 2  y 2  2 is continuous everywhere iff
 x , y  0,0 1

x 2
y 
2 2 
 A  x, y    0,0
A0.

7
Example

Give an example of a function f  x, y  such that f  x, y  has the same limit as f  0,0 when

 x, y  approaches  0,0 along any straight line passing through the origin, however, the function is not

continuous at  0,0 .
Solution:
Consider the function:

 2 xy 2
 if  x, y    0, 0 
f  x, y    x 2  y 4 .
 0 if  x, y    0, 0 


Suppose  x, y  is approaching  0,0  along the line x  t cos  , y  t sin  where  is an angle made

counterclockwise from the positive direction of x-axis to the line x  t cos  , y  t sin  and t is any real
number.

Then if cos   0
2t 3 cos  sin 2 
2t cos  sin  3 2
lim f  x, y   lim f  t cos  , t sin    lim 2  lim t2
( x , y ) (0,0) t 0 t 0 t cos 2   t 4 sin 4  t 0 t 2 cos 2   t 4 sin 4 
x t cos
y t sin  t2
2t cos  sin 2  0
 lim   0  f  0, 0 
t 0 cos   t sin  cos 2 
2 2 4

If cos   0 , then this is the case that  x, y  is approaching  0, 0  along y-axis.

It is obvious that lim f  x, y   lim f  0, y   lim 0  0  f  0, 0  .

( x , y )  0,0  y 0 y 0
x 0

However, if  x, y  is approaching  0,0 along the curve x  t 2 , y  t where t is any real number, we

2t 4
have lim f  x, y   lim f  t , t   lim 4 4  1  0  f  0, 0  .
2
 x , y (0,0) t 0 t 0 t  t
x t 2
y t

3 Partial Differentiation

8
Recall that for a function f  x  of one variable

f ( x0  h )  f ( x0 ) f  x   f  x0 
df
x0   lim or
df
 x0   lim
dx h0 h dx x  x0 x  x0

Now consider the function f  x, y  of two variables. If y is held constant, say y  y 0 , then

f ( x, y0 ) becomes a function of one variable, x and we may define its derivative at x 0 , the partial

derivative of f with respect to x at the point  x0 , y0  as

f f  x0  h, y0   f  x0 , y0  f f  x, y0   f  x0 , y0 
 x0 , y0   lim or  x0 , y0   xlim
x h 0 h x  x0 x  x0

Similarly, the partial derivative of f with respect to y at the point  x0 , y0  as

f f  x0 , y0  k   f  x0 , y0  f f  x0 , y   f  x0 , y0 
 x0 , y0   lim or  x0 , y0   ylim
y k 0 k y  y0 y  y0

AB – curve of f  x, y0 

9
 CD – curve of f  x0 , y 

Note:
(a) A partial derivative is just an ordinary derivative with respect to the variable which is not held
constant.
(b) For z  f ( x, y ) , alternative notations are:
f z
 x, y   f x  x , y    x , y   z x  x , y 
x x
f z
and similarly for  x, y   f y  x, y    x, y   z y  x, y  .
y y
(c) The definition of a partial derivative only requires the idea of a limit of a function of a single
variable.

Example

Find the first partial derivatives of f  x, y   x 2 y  y 3 sin x .

Solution:
f f
f  x, y   x 2 y  y 3 sin x   2 xy  y 3 cos x,  x 2  3 y 2 sin x
x y

Example
f
Evaluate 1, 2  for f  x, y   x2 y  y3 sin x .
x
Solution:
Method 1:
f
1,2    2 x 2  8sin x 
d
Consider f  x,2   2 x 2  8sin x . Then x 1   4 x  8cos x  x 1  4  8cos1 .
x dx
Method 2:

f f
f  x, y   x 2 y  y 3 sin x   2 xy  y 3 cos x . Then 1, 2    2 xy  y 3 cos x  x 1  4  8cos1 .
x x y 2

Example

Find the first partial derivatives of f  x, y, z   e2 z cos xy .

Solution:
f f f
f  x, y, z   e2 z cos xy   e2 z y sin xy,  e 2 z x sin xy,  2e 2 z cos xy .
x y z

10
We may differentiate the first partial derivatives of a function again to obtain second partial derivatives.
2 f   f  2 f   f  2 f   f  2 f   f 
f xx     , f yy     , f xy     , f yx    .
xx x  x  yy y  y  yx y  x  xy x  y 

Higher order partial derivatives may be similarly defined.

Example
1
Let f  x, y   log  x 2  y 2  2 , find its partial derivatives up to order two.

Solution:
1
f f
f ( x, y )  log  x  y  x y
2 2 2
 fx   2 , fy   2
x x  y 2
y x  y 2
2 f   f    x  1 2 x2 y2  x2
 f xx  2      2   
x x  x  x  x  y 2  x 2  y 2  x 2  y 2 2  x 2  y 2 2

2 f   f  x2  y2   f    x  2 xy
, f yy      , f xy      2   ,
y 2
y  y   x 2  y 2  2
y  x  y  x  y   x 2  y 2 2
2

  f    y  2 xy
f yx      2 2 

x  y  x  x  y   x 2  y 2 2

Note that in the above example f xy  f yx , which is the case for most functions that are useful in

engineering and science. However, it is not always true.

Example


 xy x 2  y 2

 if  x, y   (0, 0)
f  x, y    x 2  y 2

 0 if  x, y   (0, 0)

(a) If  x, y    0,0 , find f xy  x, y  , f yx  x, y  and verify that they are equal.

(b) Show that f x  0,0   f y  0,0  .

(c) Show that f xy  0,0  , f yx  0,0  exist but they are not equal.

Slolution:
(a)

11
There is a theorem saying that if f xy  x, y  & f yx  x, y  exist at  x0 , y0  and nearby and

f xy  x, y  & f yx  x, y  both are continuous at  x0 , y0  then f xy  x0 , y0   f yx  x0 , y0  .

Now, if  x, y    0,0

    xy  x  y      4 x 2 y 3  x 4 y  y 5   y 6  9 x 2 y 4  9 x 4 y 2  x 6
2 2

f xy  x, y       
y  x  x 2  y 2   y   x 2  y 2 2   x2  y2 
3
    

  4 y 2 x 3  y 4 x  x 5   y 6  9 x 2 y 4  9 x 4 y 2  x 6
f yx  x, y   
x 
  x 2
 y 
2 2 
  x 2
 y 
2 3

So f xy  x, y   f yx  x, y 

(b)
f  x, 0   f  0, 0  00 f  0, y   f  0, 0  00
f x  0, 0   lim  lim  0 , f y  0, 0   lim  lim 0
x 0 x x 0 x y 0 y y 0 y

(c)
xy  x 2  y 2 
0
f  x, y   f  0, y  x2  y2 y  x2  y 2 
If y  0 , f x  0, y   lim  lim  lim  y
x 0 x x 0 x x 0 x2  y2
 y if y  0
Thus, f x (0, y )     y, y
 0 if y  0
f x  0, y   f x  0, 0  y  0
f xy  0, 0   lim  lim  1
y 0 y y 0 y
Similarly,
xy  x 2  y 2 
0
f  x, y   f  x , 0  x2  y 2 x  x2  y 2 
If x  0 , f y  x, 0   lim  lim  lim x
y 0 y y 0 y y 0 x2  y 2
Thus, f Y ( x,0)  x

fY  x, 0   fY  0, 0  x0
f yx  0, 0   lim  lim 1
x 0 x x 0 x

We find that f xy  0,0   f yx  0,0  . Why does this happen?

12
   y6  9 x2 y 4  9 x4 y 2  x6
 if  x, y    0, 0 
For f xy  x, y     x2  y 2 
3


 1 if  x, y    0, 0 

,we observe that along y  0 , lim f xy ( x, y)  lim f xy  x, 0   1 .

( x , y ) (0,0) x 0

However, along x  0 , lim f xy  x, y   lim f xy  0, y   1 .

 x , y  0,0 y 0

It concludes that f xy  x, y  is not continuous at  0, 0  .

4 The Chain Rule

dy df dx
For functions of one variable, if y  f  x  and x  x  t  , then y  f  x  t   and  .
dt dx dt
f df u f df u
If f  f  u  and u  u  x, y  then  and  .
x du x y du y

Example
1
f f
If f  f  u   log u and u   x 2  y 2  2 , evaluate , .
x y
Solution:
f df u 1 1 2 1 1
   x  y   2x    x  y   2x  2
2 2 1 1 2 2 2 x
 .
x du x u 2 1
2 x  y2
(x  y ) 2 2 2

f df u y
And   2 .
y du y x  y 2

For a function of two variables we have

Theorem
dz f dx f dy
If z  f  x, y  and x  x  t  , y  y  t  then   .
dt x dt y dt
Proof:

13
 z t   t   z t  f  x  t   t  , y  t   t    f  x  t  , y  t 
 lim 
dz
 lim .
dt  t0 t  t 0 t

Now as t change to t   t , x and y change by  x and  y respectively, where  x  x  t   t   x  t  ,

 y  y  t   t   y  t  . Thus the change in f may be expressed as

f  x  t   t  , y  t   t    f  x  t  , y  t    f  x   x, y   y   f  x, y 
 AC  AR  RC  PB  RC
  f  x   x, y   f  x, y     f  x   x, y   y   f  x   x , y  

, and

14
 z t   t   z t  f  x  t   t  , y  t   t    f  x  t  , y  t  
 lim 
dz
 lim
dt  t 0 t  t 0 t
f  x   x, y   f  x , y   f  x   x , y   y   f  x   x , y 
 lim
 t 0 t
 f  x   x, y   f  x , y   x f  x   x , y   x   f  x   x , y   y 
 lim   
 t 0
 x t y t 
f  x   x, y   f  x , y  x f  x   x, y   x   f  x   x , y  y
 lim lim  lim lim
 x 0 x  t  0 t  y  0 y  t  0 t
f dx f dy
 
x dt y dt

Example
dz
Let z  f  x, y   x 2  2 xy and x  sin t and y  cos t , find .
dt
Solution:
dz f dx f dy
    2 x  2 y  cos t  2 x   sin t    2sin t  2cos t  cos t  2sin t   sin t 
dt x dt y dt
 sin 2t  2cos 2t

This may be confirmed by direct substitution:

dz
z  f  sin t , cos t   sin 2 t  2sin t cos t   2sin t cos t  2sin 2 t  2cos 2 t  sin 2t  2cos 2t
dt

A special case is when z  f  x, y  and y  y  x  , that is, x  x and hence z is a function of x only.

dz f dx f dy f f dy
Then     .
dx x dx y dx x y dx

Example
x dz
z  f  x, y   tan 1   , y  sin x , find .
 y dx
Solution:
dz f dx f dy f f dy y x sin x x cos x sin x  x cos x
     2  2 cos x  2  2  2 .
dx x dx y dx x y dx x  y 2
x y 2
x  sin x x  sin x
2 2
x  sin 2 x

Example (Chain rule involving second derivatives)

15
 d 2z
Let z = f  x, y  , x = x  t  , y = y t  , evaluate .
dt 2
Solution:
dz f dx f dy
  .
dt x dt y dt
d z  d  dz   d  f dx  f dy   f d x  d  f  dx  f d y  d  f  dy .
2 2 2

   
dt
2
dt  dt  dt  x dt y dt  x dt 2 dt  x  dt y dt 2 dt  y  dt
f
We observe that is depending on x and y directly, depending indirectly on t through x and y
x
d  f 
respectively. To evaluate   we need the Chain Rule and we have
dt  x 
d  f    f  dx   f  dy  2 f dx  2 f dy
         .
dt  x  x  x  dt y  x  dt x 2 dt yx dt
d  f    f  dx   f  dy  2 f dx  2 f dy
   
dt  y  x  y  dt y  y  dt xy dt y 2 dt
For the same reason, .

Then it follows that

d z f d x d  f  dx f d y d  f  dy
2 2 2
     
dt
2
x dt 2 dt  x  dt y dt 2 dt  y  dt
f d 2 x    f  dx   f  dy  dx f d 2 y    f  dx   f  dy  dy
          
x dt 2  x  x  dt y  x  dt  dt y dt 2  x  y  dt y  y  dt  dt
f d 2 x  2 f  dx   2 f dy dx f d 2 y  2 f dx dy  2 f  dy 
2 2

         
x dt 2 x 2  dt  yx dt dt y dt 2 xy dt dt y 2  dt 

Example

z  f  x, y  , x  t 2 , y  t . Suppose

f f
1,1  1,1   2 1,1   2 1,1  e,  1,1   1,1  2e . Evaluate d 2 for t  1
2 2 2 2 2
f f f f z
x y x y xy yx dt
Solution:
dx d 2x dy d2y
x  t2   2t , 2  2 , y  t   1, 2  0 . If t  1 , then x  1, y  1 .
dt dt dt dt
So
f 2 f 2 f
2
d z
2
d 2x  dx  dy dx
t 1  1,1 2 t 1  2 
1,1   t 1  1,1 t 1 t 1
dt
2
x dt x  dt  yx dt dt
f 2 f 2 f
2

1,1 t 1 t 1  2 1,1  
d2y dx dy dy
 1,1 2 t 1  t 1
y dt xy dt dt y  dt 
 e  2  e  4  2e  1  2  e  0  2e  2  1  e  1  15e

16
Theorem (Change of Variable)
z z
If z  f  x, y  and x  x  s, t  , y  y  s, t  , find , .
s t
Solution:
z f x f y z f x f y
  ,   .
s x s y s t x t y t

Example
z z
If z  f  x, y   e2 xy and there is the change of variables x  r cos  , y  r sin  , find , .
r 
Solution:
z f x f y
   2 ye2 xy cos  2 xe2 xy sin   2rer sin 2 sin 2 and
2

r x r y r
z f x f y
   2 ye2 xy  r sin    2 xe2 xy r cos 
 x  y 
.
sin 2
 2rer cos 2
2

x  r cos
y  r sin 

Again these results may be checked by substitution and direct partial differentiation.

Example (Chain rule involving second partial derivatives)

2 z
Let z = f  x, y  , x = x  s, t  , y = y  s, t  , evaluate .
t 2
Solution:
z f x f y
  .
t x t y t

 2 z   z    f x f y    f x    f y 
      
t 2 t  t  t  x t y t  t  x t  t  y t 
  f  x f   x    f  y f   y 
        
t  x  t x t  t  t  y  t y t  t 

f f
We observe that both , are depending on x and y directly, depending indirectly on t through x and
x y
  f    f 
 ,
t  x  t  y 
y respectively. To evaluate we need the Chain Rule and we have

17
  f    f  x   f  y  2 f x  2 f y
        
t  x  x  x  t y  x  t x 2 t yx t
  f    f  x   f  y  2 f x  2 f y
   
t  y  x  y  t y  y  t xy t y 2 t
It follows that

 z    f  x  f   x     f  y  f   y 
2

     
t
2
t  x  t x t  t  t  y  t y t  t 
  2 f x  2 f y  x f  2 x   2 f x  2 f y  y f  2 y
 2      2  
 x t yx t  t x t  xy t y t  t y t
2 2

 2 f  x   2 f y x f  2 x  2 f x y  2 f  y  f  2 y
2 2

 2        
x  t  yx t t x t 2 xy t t y 2  t  y t 2
 2 f  x   2 f x y  2 f  y  f  2 x f  2 y
2 2

 2   2     
x  t  xy t t y 2  t  x t 2 y t 2

Example

Consider z  f  x, y  , suppose x  r cos  , y  r sin  . Show that

2 z 2 f 2 f 2  f
2
 cos 2
  2sin  cos   sin  .
r 2 x 2 xy y 2
Solution:
z f x f y f f
   cos  sin  .
r x r y r x y

 2 z   z    f f    f    f 
     cos   sin    cos     sin   
r 2
r  r  r  x y  r  x  r  y 
  2 f x  2 f y    2 f x  2 f y 
 cos   2    sin    2 
 x r yx r   xy r y r 
2 2 2
 cos2  2 f ( x, y )  2sin  cos  f ( x, y )  sin 2 2 f ( x, y )
x xy y

Example
 2u  2u
Show that in polar coordinates  r ,  in R 2 , Laplace's equation  2u    0 takes the
x 2 y 2
 2u 1  2u 1 u
form 2  2   0.
r r  2 r r
Solution:

18
u u x u y
 
r x r y r
u u x u y
 
 x  y 
x y x y
x  r cos  , y  r sin    cos  ,  sin  ,  r sin  ,  r cos 
r r  
Thus,
u u u
 cos   sin 
r x y
u u u
 (r sin  )  r cos 
 x y
Therefore,
 u   u   u   u 
sin    x   x   cos 
1
 r   cos  sin    r 
      
 u    r sin  r cos    u   u    r sin  r cos    u 
 y   y   
         
 u   sin    u 
 x   cos  
r   r 
   
 u   sin  cos    u 
 y  
   r    
u u sin  u u u cos u
So  cos  (*),  sin   (**)
x r r  y r r 
 f    f  cos   f 
From (**), we learn that  sin   . Notice that the f in the left side of (**) is a
y r r 
function of x, y , however, the f in the right side of (**) after x, y being replaced by r cos and r sin  ,
u
respectively, is a function of r,  . Let  f , therefore, we have
y
u u cos u
In addition, we notice that  sin  
y r r 
Therefore,
 2u   u    u cos u  cos   u cos u 
    sin   sin     sin   
y 2
y  y  r  r r   r   r r  
  2u cos  2 u cos u  cos  u  2 u sin  u cos  2u 
 sin   sin  2   2   cos   sin    
 r r r r   r  r r r r r r 
(1)
Similarly,

19
 2u   u    u  sin    u 
    cos     
x 2
x  x  r  x  r   x 
  u sin  u  sin    u sin  u 
 cos  cos    cos  
r  r r   r   r r  
  2u sin  u sin   2 u  sin    2u u sin   2u cos u 
 cos  cos 2  2    cos   sin    
 r r  r r  r  r r r  2 r  
(2)
 2u  2u  2u 1  2u 1 u
Finally, (1)  (2) we have      0.
x 2 y 2 r 2 r 2  2 r r

Let f : R 2  R . We say f  x, y  is differentiable at  x0 , y0  if and only if there exists a 1 2 matrix

 x  x0 
f  x, y   f  x0 , y0   T  
T such that lim  y  y0   0 .
 x , y  x0 , y0 
 x  x 0    y  y0 
2 2

Observe that if T   a, b  then

 x  x0 
f  x, y   f  x0 , y0   T  
 y  y0   0  lim f  x, y   f  x0 , y0   a  x  x 0   b  y  y0 
lim 0
 x , y  x0 , y0   x , y  x0 , y0 
 x  x 0    y  y0   x  x 0    y  y0 
2 2 2 2

Theorem
f f
Suppose f : R 2  R is differentiable at  x0 , y0  , then the partial derivatives and exist at
x y
 f f 
 x0 , y0  and the 1 2 matrix T is given by T   ,  .
 x y 

Theorem

Suppose f : R 2  R is differentiable at  x0 , y0  , then f : R 2  R is continuous at  x0 , y0  .

Example

 xy 2
 2 if  x, y    0, 0 
Consider the function f  x, y    x  y
2

 0 if  x, y    0, 0 


20
Show that
f f
(a)  0, 0  ,  0, 0  exist .
x y
Proof:
f f  x, 0   f  0, 0  00 f f  0, y   f  0, 0  00
 0, 0  lim  lim 0 ;  0, 0   lim  lim 0
x x 0 x0 x 0 x  0 y y 0 y 0 y 0 y  0

(b)

Show that f  x, y  is not differentiable at  0, 0  .

Proof:

We observe that if  x, y  is on the line y  t , x  t , then if t  0

 f   f 
f  x, y   f  0, 0     0, 0    x  0     0, 0    y  0  t3
 x   y  f t, t 
 t t 
2 2 1 t

 x  0   y  0 t2  t2 2 2 t
2 2
2t 2

1 t
Suppose now  x, y  approaches  0, 0  along the straight line y  t , x  t , lim
t 0
does not exist.
2 2 t
It follows that

 f   f 
f  x, y   f  x0 , y0     x0 , y0    x  x 0     x0 , y0    y  y0 
 x   y 
lim does not exist.
 x , y  x0 , y0 
 x  x 0    y  y0 
2 2

So f ( x, y) is not differentiable at  0, 0  .

5 Implicit Functions

Let f  x, y  be a function of two independent variables and suppose f x  x, y  & f y  x, y  exist and are

continuous at  x0 , y0  and nearby, in addition, f  x0 , y0   0 & f y  x0 , y0   0 . Then about  x0 , y0  the

equation f  x, y   0 determines an implicit function y  y  x  about x 0 with y0  y  x0  and

y  y  x  being differentiable at x 0 .

The above is also applicable to system of equations:

21
  f1 x1 ,  , xn , y1 ,  , y m   0

The system of equations   determine locally y1 , , ym as implicit functions
 f x , , x , y , , y   0
 m 1 n 1 m

of x1 , , xn and the partial derivatives of y1 , , ym with respect to xi , 1  i  n can be found.

Example
z
The equation x 3  xy 2  yz 2  z 3  5 determine z as a function of x, y. Find .
x
Solution:
Differentiating x 3  xy 2  yz 2  z 3  5 with respect to x gives
z z z y 2  3x 2
3x 2  y 2  2 zy  3z 2  0 and  if 2 zy  3z 2  0 .
x x x 2 zy  3z 2

Example
dy
Find the gradient of the tangent at a point  x, y  of the conic
dx

f  x, y   ax 2  2hxy  by 2  2 gx  2 fy  c  0

Solution:

y is defined as an implicit function of x by the equation z  f  x, y   ax 2  2hxy  by 2  2 gx  2 fy  c  0

f
f f dy f
  x whenever
dz dy
In this case 0  and we have  0;
dx x y dx dx f y
y
dy 2ax  2hy  2 g
 whenever 2hx  2by  2 f  0 .
dx 2hx  2by  2 f

Example
If z  e x cos y , while x and y are implicit functions of t defined by the equations
dz
x  e  t  t  1, yt  y t  t  y  0 then find
3 x 2 2 2
for t = 0 .
dt
Solution:
dz z dx z dy
Differentiate z = e x cos y with respect to t, we have  
dt x dt y dt

The equation x3 + e x  t 2  t = 1 defines a function x  x  t  implicitly.

22
Differentiate both sides of x3  e x  t 2  t  1 with respect to t.
dx dx dx 2t  1
Then 3x 2  ex  2t  1  0 . And we have  2 , if 3x 2  e x  0 .
dt dt dt 3x  e x

We observe that for x 3 + e x  t 2  t = 1, yt 2 + y 2t  t + y = 0, if t = 0 then x  0 .

dx 2t  1
Therefore, t 0  t 0 1
dt 3x 2  e x x 0

yt  y t  t  y  0 defines a function y  y(t ) implicitly. And we have

2 2
Similarly, The equation

dy dy dy dy 1  y 2  2 yt
2 yt  t 2  2 yt  y 2  1   0 . This implies  .
dt dt dt dt 1  t 2  2 yt

dy 1  y 2  2 yt
For yt 2  y 2t  t  y  0 if t = 0 then y  0 . Therefore, t 0  t 0 1
dt 1  t 2  2 yt y 0

dz z dx z dy dx dy
Also,    e x cos y  e x sin y .
dt x dt y dt dt dt

dz dx dy
Finally, t 0  e x cos y x 0 t 0  e x sin y x 0 t 0 1
dt y 0
dt y 0
dt

Example
 2 1
x  y2  z2
Suppose we solve  2 uniquely for x, y as functions of z near x  1, y  1 . Find
 x  y  z  2

dx dy d 2 x d 2 y
, , , when x  1, y  1, z  2 .
dz dz dz 2 dz 2
Solution:
 dx dy   dx 2 d 2x  dy 
2
d2y
 2        1 (3)
  dz   
1 2 2 x 2 y z (1) 2 2 x 2 2 y
x  y  z  dz
2

2 
dz

dz 2  dz  dz 2

 x  y  z  2  dx  dy  1  0 (2) d x d y
2 2

 dz dz  dz 2  dz 2  0 (4)
Put x  1, y  1, z  2 in (1) & (2), we have
 dx dy
 2 x  2 y  z (1)
 dz dz dx dy dx dy
  4  0, 4  4   0,  1
dx
  dy dz dz dz dz
 1  0 (2)

 dz dz
dx dy d 2x 1 d2y 1
Put x  1, y  1, z  2 and  0,  1 in (3) & (4) , we have   ,  .
dz dz dz 2 4 dz 2 4

23
Example

 xu  yvu 2  2
Suppose that near the point  x , y , u, v  
 1,1,1,1 we can solve  3 uniquely for u and v as
 xu  y 2 4
v  2

 u 
functions u  x, y  , v  x, y  of x and y. Compute   1,1 .
 x 
Solution:

 xu  yvu 2  2
Differentiate all equations with respect to x of  3 . We have
 xu  y 2 4
v  2

 u u 2 v
 u  x  yv 2u  yu 0
x x x

u 3  x 3u 2 u  y 2 4v 3 v  0
 x x
Since x  1, y  1, u  1, v  1 , we have
 u v  u v
1 3  0  4  12  4  0
 x x  x x u u 1
  9  3  1,1  
1  3 u  4 v  0 1  3 u  4 v  0 x x 3

 x x 
 x x

6 Taylor’s Theorem

If a function f  x  of one variable is differentiable through order N in a  l  x  a  l where l  0 ,

then for any real number h such that a  l  a  h  a  l , we have

h h2 h N 1
f a  h   f a   f ' a   f ''  a    f ( N 1)  a   RN  h  where
1! 2!  N  1!
hN ( N )
RN  h   f  a   h  for some 0    1 .
N!
OR
If a function f (x ) of one variable is differentiable through order n  1 in a  l  x  a  l where l  0 ,
then for any real number x such that a  l  x  a  l , we have

f 'a f ''  a  f ( N 1)  a 

f  x  f a   x  a   x  a   x  a   RN  x  where
N 1

2

1! 2!  N  1!

24
 f ( N )  a    x  a 
RN  x    x  a for some 0    1 .
N

N!

If a  0 we have Maclaurin’s Theorem:

h h2 h N 1 hN (N )
f  h   f  0  f ' 0  f ''  0    f ( N 1)  0   RN  h  with RN  h   f  h  for
1! 2!  N  1! N!

some 0    1 .

OR

f '  0 f ''  0  2 f ( N 1)  0  N 1 f ( N )  x  N

f  x   f  0  x x   x  RN  x  with RN  x   x for
1! 2!  N  1! N!

some 0    1 .

Example
1
Express the Maclaurin series of cos x (up to x 4 ) and , hence find the Maclaurin series of
1 y
1
(up to x 4 ).
cos x

Solution:
x2 x4 1
cos x  1    ,  1  y  y2  y3  y4  .
2! 4! 1 y
1 1 1
 
cos x x 2
x 4
 x x42

1   1    
2! 4!  2! 4! 
2 3
 x2 x4   x2 x4   x2 x4 
 1               
 2! 4!   2! 4!   2! 4! 
 x2 x4   x4  x 2 5x 4
 1         1  
 2! 4!   4  2! 24

(up to x 4 ).

Taylor’s Theorem generalizes to a function of two variables f  x, y  which has continuous partial

derivatives of all orders up to N inside the rectangle with vertices  a  p, b  q  ,  a  p, b  q  ,

 a  p, b  q  ,  a  p, b  q  , where p, q  0 .

25
At the point  a  h, b  k  , where a  p  a  h  a  p, b  q  b  k  b  q , we have

1 1
f  a  h, b  k   f  a , b   hf x  a , b   kf y  a , b   h 2 f xx  a , b   2hkf xy  a , b   k 2 f yy  a , b 
1! 2!
1 3
  h f xxx  a, b   3h 2kf xxy  a, b   3hk 2 f xyy  a, b   k 3 f yyy  a, b    RN  h, k 
3!
i
1 
N 1
 
   h  k  f  a, b   RN  h, k 
i 0 i !  x y 
where

1  N  N  i N i  N f 
N
1    
RN  h, k    h  k  f  a   h, b   k      h k  a   h , b   k  
N !  x y  N !  i 0  i  x i y N i 
1  N N  N  1  N  i  1 hi k N i N f 
   a   h, b   k  
N !  i 0 i! xi y N i 
N
h i k N i N f
  a   h, b   k 
i 0 i !( N  i )! x i y N i
for some 0    1 .

If RN (h, k )  0 as N   we have the Taylor Series representation of f  a  h, b  k  .

OR
For the case of two variables:
1
f  x, y   f  a , b   ( x  a ) f x  a, b   ( y  b) f y  a, b   
1! 
1
( x  a) 2 f xx  a, b   2( x  a )( y  b) f xy  a, b   ( y  b) 2 f yy  a, b     RN  x , y 
2! 
i
1
N 1
  
   ( x  a)  ( y  b)  f  a, b   RN  x, y 
i 0 i !  x y 
Where
N
1   
RN  x, y     x  a    y  b   f a    x  a  , b    y  b 
N! x y 
1  N N N f 
     x  a   y  b 
N !  i 0  i 
i N i

x i y N i
 a    x  a  , b    y  b   

1  N N  N  1  N  i  1 x  a   y  b  
i N i
N f
 
N !  i 0 x i y N i
 a    x  a  , b    y  b   
i! 
 x  a  y  b
i N i
N
N f

i !( N  i )! xi y N i
 a    x  a  , b    y  b
i 0

If RN ( x, y)  0 as N   we have the Taylor Series representation of f  x, y  .

26
Successive terms given by Taylor’s Theorem give the tangent plane approximation to the surface

z  f  x, y  (Taking the first two terms of the series) , a quadratic surface approximation (the first three

terms), a cubic surface (the first four terms) etc, all meeting z  f  x, y  at the point  a, b  .

Example
Find the tangent plane and quadratic surface approximations given by Taylor’s Theorem at the point
1
1,0  for the cone f ( x, y )   x 2
y 
2 2
. Hence estimate f  0.9,0.1 .

Solution:
x y
f x  x, y   1
, f y  x, y   1
,
x 2
y 
2 2
x 2
y 2 2

y2  xy x2
f xx  x, y   3
, f xy  x, y   3
, f yy  x, y   3
,
x 2
y 
2 2
x 2
y 2 2
 x 2
y 
2 2

, which exist everywhere except at the origin  0,0  .

The tangent plane approximation is:

1
p1  x, y   f 1, 0    x  1 f x 1, 0    y  0  f y 1, 0    p1  x, y   x .
1! 

The quadratic surface approximation is:

1
p2  x, y   f 1, 0    x  1 f x 1, 0    y  0  f y 1, 0  
1! 
1
  x  1 f xx 1, 0   2  x  1 y  0  f xy 1, 0    y  0  f yy 1, 0 
2 2

2! 
1
 p2  x, y   x  y 2
2

Hence, using the tangent plane f  0.9,0.1  p1  0.9,0.1  0.9 . And, using the quadratic surface

f  0.9,0.1  p2  0.9,0.1  0.905 . The correct value is f  0.9,0.1  0.9055 (to 4 s.f)

27
Example

z 3  2 xz  y  0 determines an implicit function z = z  x, y  of x and y about the point 1,1 . Use the

Taylor theorem to expand z = z  x, y  about the point 1,1 up to the linear terms.

Solution:
Note that from z 3  2 xz  y  0 when x  1, y  1 we have z  1 .

z z z 2z z
z 3  2 xz  y  0  3z 2  2z  2x  0   2 & x 1 2.
x x x 3z  2 x x y 1

z z z 1 z
z 3  2 xz  y  0  3z 2  2x 1  0   2 & x 1  1
y y y 3z  2 x y y 1

 z z 
z  x, y   z 1,1   x 1  x  1  x 1  y  1  
 x y 1
y y 1 

So z  x, y   1  2  x  1  1 y  1 

Example

Let f  x  be a smooth function, that is, f  x  has continuous derivatives up to any order.

Then the Taylor series of f  x  about the point a up to order 2 is

f 'a f ''  a 
f  x  f a   x  a   x  a 
2

1! 2!

Hence or otherwise, find the Taylor series of g  x, y, z   cos  x  y  z   cos x cos y cos z about the

  
point  0, ,  up to order 2.
 2 2
Solution:

cos  x  y  z   cos x cos  y  z   sin x sin  y  z 

 cos x  cos y cos z  sin y sin z   sin x  sin y cos z  cos y sin z 
 cos x cos y cos z  cos x sin y sin z  sin x sin y cos z  sin x cos y sin z

 
2
x2  1
Observe that cos x  1   , cos x    x    , sin x  x  , sin x  1   x   
2  2 2 2

28
g  x, y, z   cos  x  y  z   cos x cos y cos z   cos x sin y sin z  sin x sin y cos z  sin x cos y sin z
 x2  1    1  
2 2

  1    1   y     1   z    
 2   2  2   2  2 
 1   
2
      1 
2

x   1   y   
   z       x     y     1  2  z  2   
 2  2    2    2     
 x2 1    1          
2 2
 
  1    y     z       x  z       x  y   
 2 2 2  2 2     2    2 

  1     
2 2
x2 1  
 1    y     z    x  z    x  y   
2 2 2  2 2  2  2

A bound for the error involved in approximating a function near a point  a, b  may be obtained by

summing over estimates of the moduli of the terms in RN  x, y  .

Linear and quadratic approximations to the cone z  x 2  y 2 ,  x, y    0,0

Example (Estimation of Small Changes)

For ( x, y ) near  a, b  we may expect

f  x, y   f  a, b    x  a  f x  a, b    y  b  f y  a, b  f  x, y 

29
Writing f  x, y   f  a, b    f , x  a   x, y  b   y

So from f  x, y   f  a, b    x  a  f x  a, b    y  b  f y  a, b  we have

f f
f   x   y , which gives the approximate change in f due to small changes  x and  y in x
x y
and y respectively.

Example

(a) Use Taylor’s formula to find a linear approximation of f  x, y   cos x cos y near the origin.

(b) Estimate the error in the approximation if x  0.1 and y  0.1 .

Solution:
(a)
By Taylor’s Theorem, we have
1
f  x, y   f  0,0    xf x  0,0   yf y  0,0    x 2 f xx  x, y   2 xyf xy  x, y   y 2 f yy  x, y  , where
2
0    1.

Then the linear approximation of f  x, y   cos x cos y near the origin is

p1  x, y   f  0,0   xf x  0,0   yf y  0,0  .

And f  0,0   cos x cos y x 0  1, f x  0,0    sin x cos y x 0  0, f y  0,0    cos x sin y x 0 0.
y 0 y 0 y 0

We have p1  x, y   1 . If  x, y  is close to  0,0  , f  x, y   cos x cos y  p1  x, y   1 .

(b)
The error in the approximation is

E  x, y   f  x, y   f  0,0    xf x  0,0   yf y  0,0 

1 2
  x f xx  x, y   2 xyf xy  x, y   y 2 f yy  x, y 
2
Observe that f xx  x, y    cos x cos y, f yy  x, y    cos x cos y, f xy  x, y   cos x cos y .

30
E  x, y   f  x, y   f  0,0    xf x  0,0   yf y  0,0 
1 2
  x f xx  x,  y   2 xyf xy  x,  y   y 2 f yy  x ,  y 
2
1
  x 2 f xx  x,  y   2 xyf xy  x,  y   y 2 f yy  x ,  y 
2

E  x, y  
2

1 2
x f xx  x,  y   2 xyf xy  x,  y   y 2 f yy  x,  y  

1 2
2
 x f xx  x,  y   2 x y f xy  x,  y   y f yy  x, y 
2


1 2
2
 x  cos  x cos  y  2 x y cos  x cos  y  y  cos  x cos  y
2


1 2
2
 x cos  x cos  y  2 x y cos  x cos  y  y cos  x cos  y
2

 
1 2
cos  x 1, 2
x 2 x y  y
2

 2  0.1  0.02
x 0.1,
2

cos  y 1 y 0.1

Example

(a) Use Taylor’s formula to find a quadratic approximation of f  x, y   e x sin y at the origin.

(b) Show that in the approximation the error E  x, y   1.474 103 if x  0.1 , y  0.1 and

e0.1  1.1052 .
(c) Find an approximation of e0.09 sin 0.06 with an error less than 1.474 103 .
Solution:
(a)
By Taylor’s Theorem, we have
1
f  x, y   f  0, 0    xf x  0, 0   yf y  0, 0     x 2 f xx  0, 0   2 xyf xy  0, 0   y 2 f yy  0, 0  
2
, where 0    1 .
1 3
  x f xxx  x, y   3x yf xxy  x, y   3xy f xyy  x, y   y f yyy  x, y  
2 2 3

Then the quadratic approximation of f  x, y   e x sin y at the origin is:

1
p2  x, y   f  0, 0    xf x  0, 0   yf y  0, 0    x 2 f xx  0, 0   2 xyf xy  0, 0   y 2 f yy  0, 0  ,
2
and

31
 f  0, 0   e x sin y x 0  0, f x  0, 0   e x sin y x 0  0, f y  0, 0   e x cos y x 0 1
y 0 y 0 y 0
.
f xx  0, 0   e sin y
x
x 0  0, f xy  0, 0   e cos y
x
x 0  1, f yy  0, 0   e sin y
x
x 0 0
y 0 y 0 y 0

1
Hence, p2  x, y   f  0, 0    xf x  0, 0   yf y  0, 0    x 2 f xx  0, 0   2 xyf xy  0, 0   y 2 f yy  0, 0   y  xy .
2

If  x, y  is close to  0,0  , f  x, y   e x sin y  p2  x, y   y  xy .

(b)
The error in the approximation is
1 3
E  x, y   f  x, y   p2  x, y    x f xxx  x, y   3x 2 yf xxy  x,  y   3xy 2 f xyy  x,  y   y 3 f yyy  x,  y  .
6

Observe that

f xx  x, y   e x sin y, f xy  x, y   e x cos y, f yy  x, y   e x sin y

 f xxx  x, y   e x sin y, f xxy  x, y   e x cos y, f xyy  x, y   e x sin y, f yyy  x, y   e x cos y

1 3
E  x , y   f  x , y   p2  x , y    x f xxx  x,  y   3x 2 yf xxy  x,  y   3xy 2 f xyy  x,  y   y 3 f yyy  x,  y  
6
1 3
 x f xxx  x, y   3 x y f xxy  x, y   3 x y f xyy  x, y   y f yyy  x, y  
2 2 3

6  
1 3
  x e x sin  y  3 x y e x cos  y  3 x y e x sin  y  y e x cos  y 
2 2 3

6 
1  3 x
 x e sin  y  3 x y e x cos  y  3 x y e x sin  y  y e x cos  y 
2 2 3

6 

0    1,  0.1  x  0.1  0.1   x  0.1   x  0.1 .

Observe that sin  y , cos  y  1 , e x  e x  e  x  e0.1 .

Thus,
1  3 x
E  x, y   x e sin  y  3 x y e x cos  y  3 x y e x sin  y  y e x cos  y 
2 2 3

6  
1 0.008e0.1
 0.1 e0.1  3  0.1  0.1 e0.1  3  0.1 0.1 e0.1   0.1 e0.1    1.474 10 3
3 2 2 3

sin  y , cos y 1 6   6
x , y  0.1
e x  e0.1

32
(c)

f  0.09, 0.06   e0.09 sin 0.06  p2  0.09, 0.06    y  xy  x 0.09  0.06  0.09  0.06
y  0.06 .
 0.06  0.91  0.0546

7 Maxima ad Minima

A function f  x, y  has a local maximum at a point  a, b  if for all points  x, y  close to  a, b  , we

f  x, y   f  a, b  and the point  a, b  is called a local maximum point .

A function f  x, y  has a local minimum at a point  a, b  if for all points  x, y  close to  a, b  , we

have f  x, y   f  a, b  and the point  a, b  is called a local minimum point .

In a region the global (absolute) maximum (minimum) of a function is either the greatest (least) of the
local maxima (minima) or occurs on the boundary of the region.

z  f  x, y    x 2  3 y 2  e1 x  y2
2

For a local maximum or minimum at  a, b  the curves lying in the two planes x  a and y  b must

f f
also have a maximum or minimum at  a, b  . Hence we must have  a, b    a, b   0
x y

If this condition holds we say  a, b  is a stationary point.

33
It is possible to have a stationary point which is neither a local maximum nor a local minimum. If along

one path through  a, b  the point is a local minimum, whilst along a different path it is a local

maximum, we say that the point  a, b  is a saddle point. There is no equivalent point for functions of

one variable.

Theorem

Assume f  x, y  has the partial derivative up to the order we need and all partial derivatives are

continuous, also f x  a, b   f y  a, b   0 . If f xx  a, b  0, f xx  a, b  f yy  a, b   f xy2  a, b   0 for f  x, y  at

 a, b  then f  x, y  has a local minimum there. If f xx  a, b   0, f xx a, b  f yy a ,b   f xy2 a ,b   0 for

f  x, y  at  a, b  then f  x, y  has a local maximum there.

If f xx  a, b  f yy  a, b   f xy2  a, b   0 , then f  x, y  has a saddle point at  a, b  .

Examples

(a) Determine the nature of the stationary points of f  x, y   x 2  2 y 2 .

Solution:

 f x  x, y   2 x

f  x, y   x 2  2 y 2  
 f y  x, y   4 y


So f x  x, y   2 x  0 & f y  x, y   4 y  0   x, y    0,0  is the only stationary point.

f xx  x, y   2, f yy  x, y   4, f xy  x, y   0 , f xx 0,0  2  0 and f xx  0,0 f yy  0,0  f xy2  0,0  8  0 .

f  0,0  0 is a local minimum and  0,0  is a local minimum point and clearly f  x, y   0 hence

f  0,0  0 is a global minimum.

34
(b) Determine the nature of the stationary points of f  x, y   2 x  x 2  2 y 2 .

Solution:

f x  x, y   2  2 x, f y  x, y   4 y

f x  x, y   2  2 x  0 & f y  x, y   4 y  0   x, y   1,0  , which is the only stationary point.

f xx  x, y   2, f yy  x, y   4, f xy  x, y   0 , f xx 1,0  2  0 and f xx 1,0  f yy 1,0   f xy2 1,0   8  0 .

f (1,0)  1 is a local maximum. In addition,

f  x, y   f 1,0  2 x  x 2  2 y 2  1    x  1  2 y 2  0
2

Thus f  x, y   f 1,0 and 1,0  is a global maximum point.

35
(c) Determine the nature of the stationary points of f  x, y   x 2  2 y 2 .

Solution:

f x  x, y   2 x, f y  x, y   4 y .

f x  x, y   2 x  0, f y  x, y   4 y  0   x, y    0,0 , which is the only stationary point.

f xx  x, y   2, f yy  x, y   4, f xy  x, y   0

Since f xx  0,0  f yy  0,0   f xy2  0,0   8 ,  0, 0  is a saddle point.

 0, 0  being a saddle point can also be observed by noting that f  x, y  may be either positive or

negative depending on the relative magnitudes of x and y .

If y  0 ,x  0, f  x,0   f  0,0   x 2  0 . If x  0, y  0, f  0, y   f  0,0   2 y 2  0 .

Thus  0,0 is a saddle point

36
Example

Determine the nature of the stationary points of f  x, y   x 3  y 3  3  x  y  .

Solution:
f f
f  x, y   x 3  y 3  3  x  y    3x 2  3,  3y2  3 .
x y
 f
 x  3x  3  0  x  1
2

 f 
  3 y 2  3  0  y  1
 y

Hence there are 4 stationary points 1,1 , 1, 1 ,  1,1 ,  1, 1 . In addition,

2 f 2 f 2 f
 6 x ,  6 y , 0
x 2 y 2 xy
2
2 f 2 f 2 f 2 f 2 f  2 f 
Stationary Point   Conclusion
x 2 xy y 2 x 2 y 2  xy 
1,1 6 0 6 36 Local Minimum Point
 1, 1 6 0 6 36 Local Maximum Point
1, 1 6 0 6 36 Saddle Point
 1,1 6 0 6 36 Saddle Point

Example

Determine the nature of the stationary points of f  x, y   x 4  y 4 .

Solution:

37
  f
 x  4 x
3

f  x, y   x 4  y 4   . We observe  0, 0  is the only stationary point.

 f  4 y 3
 y

And   0,0   f xx  0,0  f yy  0,0   f xy2  0,0   0 the test inconclusive, but clearly  0, 0  is a local

minimum Point.

Consider a point P  x0 , y0  and a straight line y  ax  b . The value  y0   ax0  b 

2
is called the

square of the vertical displacement of the data point P  x0 , y0  from the line y  ax  b .

Determine the straight line y  ax  b such that the sum S of the squares of the vertical displacements

of the data points P1 1,3.1 , P2  2, 6.1 , P3  3,9.2  from the line y  ax  b is the smallest.

In other words, determine the values of a, b such that S is the smallest where

S   3.1  a  b    6.1  2a  b    9.2  3a  b 

2 2 2

S  a, b    3.1  a  b    6.1  2a  b    9.2  3a  b 

2 2 2

 S
0  a  2  3.1  a  b   2  6.1  2a  b   3  9.2  3a  b  

0  S  2  3.1  a  b    6.1  3a  b    9.2  3a  b  
 b  
1  22  32  a  1  2  3 b  1 3.1  2  6.1  3  9.2

1  2  3 a  3b  3.1  6.1  9.2
 6.1
 a  3.05
14 a  6b  42.9 14 a  6b  42.9  2
  
6a  3b  18.4 12a  6b  36.8 b  18.4  18.3  0.1  1
 3 3 30

2S 2S 2S

Then S  a, b    3.1  a  b    6.1  2a  b    9.2  3a  b   2  28,  6,  12
2 2 2

a b2 ba
2S
Observe that  28  0 and
a 2
 1   1   1   1 
  3.05,   Saa  3.05,  Sbb  3.05,   Sab2  3.05,   28  6 144  168 144  24  0 .
 30   30   30   30 

38
 1
It follows that S  a, b  has a local minimum value at a  3.05, b  .
30

S  a, b    3.1  a  b    6.1  2a  b    9.2  3a  b 

2 2 2
Observe that is a degree 2 polynomial in a, b

 1 
and  3.05,  is the only local minimum point, S  a, b  actually has the global minimum value
 30 
 1  1  1 
S  3.05,   at  3.05,  and this fact can be shown as follows:
 30  600  30 

Due to Taylor’s expansion theorem and noticing that S  a, b  is a degree 2 polynomial in a, b , we have

 1 
S  a, b   S  3.05, 
 30 
1  1  
2
1   1   1   1 
  S aa  3.05,   a  3.05   2 S ab  3.05,   a  3.05   b    Sbb  3.05,  b   
2

2   30   30   30   30  30  

1 1  
2
 1  
  28  a  3.05   24  a  3.05   b    6  b   
2

2S 2S 2S

2   30   30  
x 3.05  28, x 3.05  6, x 3.05 12
a 2 1 b 2 1 ba 1
y y y
30 30 30

  a  3.05  
1  1   28 12   
  a  3.05 b     1 
2  30   12 6   b  
  30  
 28 12 
See that   is a 2  2 real symmetric matrix which eigenvalues are 17  265   0  .
 12 6 
17  265 0 
Then there exist a 2  2 orthogonal matrix P and the diagonal matrix D   
 17  265 
 0

 28 12  17  265 0  1 17  265 0  T

such that    P   P  P  P
 12 6     P 1  PT   
 0 17 265   0 17 265 

 a  3.05 
 y1  T 
Let    P 1  . Then
 y2  
 b 
 30 

39
  a  3.05   T  a  3.05 
 1   28 12     1  17  265 0
 a  3.05 b    12 6   1    a  3.05 b   P  P  1 

 30    b    30   0  
17  265   b  
 30   30 
T
  a  3.05   17  265  T  a  3.05 
0
  PT    P  
  b  1     
17  265   b 
1 

  30    0
 30 
17  265   y1 
   
0
  y1 y2       17  265 y12  17  265 y22  0
 17  265   y2 
 0

  a  3.05  
 1  1  1   28 12   
It follows that S  a, b   S  3.05,    a  3.05 b     1   0 .
 30  2   30  12 6  b  
  30  
 1  1  1 
Therefore, S  a, b  has the global minimum value S  3.05,   at  3.05,  .
 30  600  30 

Example
Find the global (absolute) maximum and global minimum values of the function

f  x, y   x 2  y 2  x  y  1 in the disk defined by x 2  y 2  1 .

Solution:

Remark: Since f  x, y   x 2  y 2  x  y  1 is a continuous function defined on a bounded closed disk

x 2  y 2  1 , according to some theorems not mentioned f  x, y  can attain global maximum and global

minimum at stationary points or at those points where f  x, y  do not have partial derivatives or at the

points on the boundary .

Consider f ( x, y )  x 2  y 2  x  y  1 on x 2  y 2  1 , where x 2  y 2  1 is the boundary of
x 2  y 2  1.
Let x  cos t, y  sin t,0  t  2 , consider

g  t   f  cos t,sin t   cos2 t  sin 2 t  cos t  sin t  1  2  cos t  sin t , where 0  t  2 .

 5
Then g ' (t )  sin t  cos t  0  sin t  cos t  t  or
4 4
   5     5 
g ''  t   cos t  sin t and g ''    0, g ''    0. g  0  g  2   1, g    2  2, g    2  2 .
4  4  4  4 
 5  5
It is obvious that g  t  attains the greatest value g    2  2 at t  and the smallest value
 4  4

40
   
g    2  2 at t 
4 4
5 2 2
When t  ,x   ,y .
4 2 2
 2 2
When t  ,x  ,y 
4 2 2
Consider f ( x, y)  x  y  x  y  1 inside x 2  y 2  1.
2 2

 f
 2x  1  0
 x  x  1/ 2
 f  and f xx  2, f yy  2, f xy  0
  2y 1  0  y  1/ 2
 y

1 1 1 1 1 1 1 1 1 1
  ,   f xx  ,  f yy  ,   f xy2  ,   4 and f xx  ,   2 .
2 2 2 2 2 2 2 2 2 2
1 1 1 1 1
f ( x, y)  x 2  y 2  x  y  1 has a local minimum at  ,  and f  ,  
2 2 2 2 2
 1 1  1 1 1
Compare f  ,   2  2 and f  ,   .
 2 2 2 2 2
1 1
We conclude that f  x, y   x 2  y 2  x  y  1 has a absolute minimum value at  ,  and
2 2
1 1 1
f  ,  .
2 2 2

 2 2
Also f ( x, y)  x 2  y 2  x  y  1 has a absolute maximum value at   ,  and
 2 2 

 2 2
f  ,   2 2 .
 2 2 

Example

Determine whether the function f  x, y, z   xy  yz  xz has any local extrema.

Solution:

 f x ( x, y, z )  y  z  0
 x  y  0
f  x, y, z   xy  yz  xz   f y ( x, y, z )  x  z  0    y 0 x  y  z 0
  y  x  0
 f z ( x, y , z )  y  x  0

 0,0,0  is the only critical point of f  x, y, z   xy  yz  xz and f  0,0,0   0 .

41
 1 1  1 1 
However, consider the points  , ,0  ,  ,  ,0  , we have
n n  n n 
1 1  1 1 1  1
f  , ,0   2  0  f  0,0,0   f  ,  ,0    2 .
n n  n n n  n
1 1  1 1 
And observe that  , ,0  ,  ,  ,0  can be as close to  0,0,0  as we want for sufficiently large
n n  n n 

n  0 . As a result, f  x, y, z   xy  yz  xz has no local extrema.

8 Directional Derivative and Gradient

For the scalar function z  f  x, y  the partial derivatives f x  x, y  and f y  x, y  measure the rates of

change of z  f  x, y  in the x and y directions respectively. We would like to know the rate of change in

any direction.

Consider a point P which lies on some level curve AB and a (variable) point Q on a nearby level curve
CD.

42
The rate of change of f in the direction PQ is known as the directional derivative of f  x, y  in the

df f f f  P   f Q 
direction PQ and is  lim where  , PQ   s
ds  s0  s s PQ
df
Geometrically it is clear that will be a maximum in the direction normal to the level curve at P
ds
df
(since f  P   f Q  is constant and PQ is least in that direction). is also denoted as Dn f where
ds
n is the unit vector in the direction PQ.

If the direction PQ is inclined a an angle  to the x-axis we have

 x   s cos ,  y   s sin 

f f  f f f
From Taylor’s Theorem  f   x   y , thus  cos   sin 
x y  s x y
df f f
And as  s  0  Du f  cos   sin  (*) , where u  i cos   j sin 
f f
f  x  y
x y

ds x y
df df
By considering as a function of  it is easy to show that has a maximum value of
ds ds
1  f 
 f  2  f  2  2  y  f f
      when   tan 1   . If we define a vector gradf  f  i  j known as the
 x   y    f  x y
 x 
 
df df
gradient of f, its magnitude is that of the maximum value of and its direction is that in which
ds ds
its maximum value.

Hence f gives us the magnitude and direction (normal to the level curve at P) of the greatest rate of
change of f.

Also, from (*), if n is a unit vector in any other direction PQ, then n   cos   i  sin   j and

43
df
 Dn f  gradf  n  f  n .
ds

The following theorem is a generalization of the above discussion to a function of n independent

variables.

Theorem

Let f : R n  R and suppose f x1  x1 ,..., xn  , , f xn  x1 ,..., xn  exist and continuous everywhere. Show

 f x1  x1 ,..., xn    x1 
   
that the gradient gradf  x1 ,..., xn   f  x1 ,..., xn      0 of f at    points in the
 f  x ,..., x   x 
 xn 1 n   n

directions which f increases most rapidly and f  x1 ,..., xn  is the rate of change of f in that direction.

Proof:

 v1   a1 
Given v    where v  1 , i.e. a unit vector and a    , the straight line passing through a with
 
v  a 
 n  n

 v1   x1   a1   v1 
       
direction v    is         t    , where t is any real number .
v   x  a  v 
 n  n  n  n

The directional derivative Dv f  a1 ,..., an  is defined as:

f  a1  tv1 ,..., an  tvn   f  a1 ,..., an 

Dv f  a1 ,..., an   lim if the limit exists.
t 0 t

Let g  t   f  a1  tv1,..., an  tvn  . By Chain Rule we have

dg f dx1 f dxn
Dv f    ...  , where xi  ai  tvi , 1  i  n .
dt x1 dt xn dt
dxi
Observe that  vi , 1  i  n
dt

44
 dg f dx f dx
Dv f  a1 ,..., an   t 0   a1 ,..., an  1 t 0  ...   a1 ,..., an  n t 0
dt x1 dt xn dt
f f
 Dv f  a1 ,..., an    a1 ,..., an  v1  ...   a1 ,..., an  vn
x1 xn
 f X1  a1 ,..., an    v1 
   
 Dv f  a1 ,..., an         f  a1 ,..., an   v
 f  a ,..., a    v 
 Xn 1 n   n

1  f x  a1 , a2 , a3    1 
   
For example, if v   0  , then Dv f  a1 , a2 , a3    f y  a1 , a2 , a3     0   f x  a1 , a2 , a3 
 0  f z  a1 , a2 , a3    0 
 

 f x1  x1 ,..., xn    x1 
   
Now, we would to show that the gradient f  x1 ,..., xn      0 of f at    points in the
 f  x ,..., x   x 
 xn 1 n   n

directions in which f increases most rapidly.

 v1 
Let v    be a unit vector i.e. v  1 . Then by Cauchy-Schawz inequality, we have
v 
 n

Dv f  a1 , a2 , a3   f  a1 , a2 , a3   v  f  a1 , a2 , a3  v  f  a1 , a2 , a3 

f  a1 ,..., an 
In addition, let w  .
f  a1 ,..., an 

Then,
f  a1 ,..., an 
Dw f  a1 ,..., an   f  a1 ,..., an   w  f  a1 ,..., an  
f  a1 ,..., an 

f  a1 ,..., an 
2

  f  a1 ,..., an 
f  a1 ,..., an 

Example
T
 1 1 
Let f  x, y, z   10  xy  xz  yz and v   , , 0  . Find Dv f 1,1,1 .
 2 2 

45
Solution:

 1 1 
g  t   f 1  t ,1  t ,1 . Then using chain rule, we have
 2 2 
dg f dx f dy f dz
t 0  1,1,1  0  1,1,1  0   1,1,1 0 
dt x dt y dt z dt
 f   dx   1   1 
 x 1,1,1   dt  0    2   2
     y  z    2  
f   1    
1,1,1     0     x  z   
   dy  1  4
  2   2 2
 y dt   x 1 2     2 2
     x  y  y 1   2  
 f 1,1,1    0  
dz  0     0 
 z   dt
z 1
   
    

Example

Suppose that the temperature (in degree Celsius) in the point  x, y , z  is

1
w  f  x, y , z    7400  4 x  9 y  0.03xy   2 z . Suppose a hawk hovering at the point
180

P  200,200,5 above the airport suddenly dives at a speed of 3 km/min in the direction specified by the

 3,4, 12 
T
vector . What instantaneous rate of change of temperature does the bird experience?

Solution:

The unit vector in the direction of the given vector 3i  4 j  12k is

3i  4 j  12k 3 4 12
u  i j k
32  42   12  13 13 13
2

4  0.03 y 9  0.03x
The temperature gradient vector f  x, y, z    i j  2k has the value
180 180
4  0.03  200 9  0.03  200 10 15
f  200,200,5   i j  2k   i j  2k at the initial position
180 180 180 180

P  200,200,5 of the hawk. Therefore the hawk’s initial rate of change of temperature with respect to

distance is:
 10  3 12  0
15 4 C
Du f  200,200,5  f  200,200,5  u    i j  2k    i  j  k   1.808
 180 180   13 13 13  km

46
 ds km
Its speed is 3 , so the time rate of change of temperature experienced by the hawk is
dt min

dw dw ds ds  0
C   km  0
C
  Du w   1.808   3   5.424 .
dt ds dt dt  km   min  mim

Thus the hawk initially gets warmer by almost 5.5 degrees per minute as it dives toward the ground.

Example

Consider f  x, y, z   axy 2  byz  cz 2 x3 .

f
(a) Determine the value of a, b, c such that the value of the partial derivative at P 1, 2, 1 is the
z

greatest among all directional derivatives of f at P 1, 2, 1 .

(b) Let a, b, c be the values found in (i), find the directional derivative Dv f 1, 2, 1 with

v  i  2 j  3k .

Solution:
(a)

f  x, y, z   axy 2  byz  cz 2 x3  f  x, y, z    ay 2  3cz 2 x 2  i   2axy  bz  j  by  2czx 3  k

And f 1, 2, 1   4a  3c  i   4a  b  j   2b  2c  k .

f
In order the value of the partial derivative at P 1, 2, 1 is the greatest among all directional
z
 4a
 c
4a  3c  0 3
  4a 16a
derivatives of f at P 1, 2, 1 we have 4a  b  0  b  4a  4a  0 0a 0
2b  2c  0 b  c  0 3 3
 

 8a  32a
Then f 1, 2, 1   8a   k  k , a  0.
 3  3
(b)

i  2 j  3k 32a i  2 j  3k 32a
Di  2 j 3k f 1, 2, 1  f 1, 2, 1  k  , a  0.
i  2 j  3k 3 14 14

47
Example

 xy 2
 2 if  x, y    0, 0 
Consider the function f  x, y    x  y
2
.
 0 if  x, y    0, 0 


Recall that f  x, y  is not differentiable at  0, 0  .

a
(a) Given v    where a 2  b 2  1 and a  0, b  0 , the straight line passing through the origin
b
a  x a
with direction v    is    t   , where t is any real number. Find the directional derivative
b  y b

f v  0, 0  .

(b) Show that the chain rule is not applicable to find f v  0, 0  .

Solution:
(a)
 ab 2
 t if t  0 ab2
f  at , bt    a 2  b 2  f  at , bt   2 t
 0 a  b 2

 if t  0
ab 2
t
 ab 2
Then f v  0, 0   lim
2 2
a b  2 .
t 0 t a  b2

f  x, y  has directional derivatives in all directions.

(b)
ab2
By (a), f v  0, 0  exists and f v  0, 0    0 where a 2  b 2  1 and a  0, b  0 .
a 2  b2

 a   f  0, 0    a   0   a  ab2
However, f  0, 0       x         
       0 f 0, 0 .
 b   f y  0, 0    b   0   b 
 v
a 2  b2

Remark: This example shows the fact that the chain rule is not applicable if f is not differentiable.

48