Unit 6 Higher Order Partial Derivatives
Unit 6 Higher Order Partial Derivatives
Unit 6 Higher Order Partial Derivatives
DERIVATIVES
Structure
6.1 Introduction
Objectives
6.2 Higher Order Partial Derivatives
6.3 Equality of Mixed Partial Derivatives
6.4 Summary
6.5 Solutions and Answers
6.1 INTRODUCTION
In the last unit you studied partial derivatives of first order and differentiability of
functions of several variables. You must have seen often, that partial derivatives of
first order again define functions. For instance, if f(x,y) = 3x3+ 2xy2+ 5y2+ 6,
then fx (x,y) = 9 ~ ~ and + f,2 (x,y)
~ =~ 4xy+ l0y are again real-valued functions
of two variables with the domain R ~ Thus. we can talk of first order partial
derivatives ~f these new functions. If we consider a fuliction of two variables,
there are two first order partial derivatives, which may give rise to four more
partial derivatives, which might again turn out to be functions. If this chain
continues, then we obtain higher order partial derivatives which constitute the
subject matter of this unit. We shall be using these partial derivatives in the next
block. In this unit you will study Euler's, Schwarz's and Young's theorems, which
give some sets of conditions under which the mixed partial derivatives become
equal.
Objectives
After studying this unit, you should be able to
0 define and evduate higher order partial derivatives,
0 state and prove Euler's theorem,
state Schwarz's and Young's theorems,
0 decide about the commutativity of the operations of taking partial,derivatives
with respect to different variables.
In the introduction you have seen that the partiid derivative fx of the function
f (x, y) = 3x3 + 2xg2 + 5y2 + 6 is again a fundtion .pf x and y. In gemral, let
D C R~ aM' let f : D -R bave a. first &&rp&tial derivative fx at every point of
D. Then we get a new fu&on, say.g = fi i&cii is defined on D. This new
function g'niay or may not posseis.first ortier ~artiDI1derivative?. In case it dots,
then dx ad gy'are c&d the m n d order partial.der~ativesof f and are ¬ed
by f& and fxy, respectively. Sirnihrly? if the fmction f has a figst oider partjal
derivative fy at every point of D, then fy .defines.aaew function. And if this new
function has first order partid derivatives, @en we get .two.mwe secqnd order
partial derivatives, namely, f, and f,. Thus, if f(x, y) is a red-valued function
defined in a neighbourhood of (a, b) having both the partial derivatives at all the
points of the neighbourhood, then
fx (a+ h, b) - fx (a, b)
f, (a, b) =. lim
h-0 h
fxy(a, b = lim
fx (a, b + k ) - fx (a,%)
) k-0 k
fy (a+h, b) - fy (a, b)
f, (a, b) = lim -
h-0 h
fy (a, b + k ) - fy (a, b)
f, (a, b) = lim
k-0 k
If we want to indihte the particular point at dhich the second order partial ',
derivatives are taken, then we write
a2f (a, b) a2f
(9) (a, b) ax' 9 fn (8, b), ( ax ay ) (a. b)
a2f (a, b)
, fxy(a, b), and so on.
ax ay
In a similar manner partial derivatives of order higher than two are defined. For
example,
a3f a2f
i.e., stands for the partial derivative of -with respect to x and
ax ax ay ax a~
is written as
a .
-
a2x ay
Similarly, we can extend the idea of partial derivatives of higher orders to
functions of more than two variables. In general, if f is a function of n variables
a2f
XI,x2, ..., X, defined on D c Rn, then - denotes the second order partial
axl h j
derivative of f with respect to xi and xj, obtained by differentiating partially the
af
partial derivative - with respect to xi; Further,' will demote the third
ax, ak, axj ax,
order partial derivative of f with respect to the variables q, xi and xk, obtained by
In the following examples, we sh0.w how t o dcnlate these higher order partial
derivatives.
Example 1 : Let us find all the second ord& partial defiv'ativ& qb $he following
functions:
i) U (x, y) =- x3+ y3 + 3.=y, a is codstant,
ii)U (x,.y, z) = x2 + yz + xi3.
Let's take these one'hy one.
i) Clearly, for U (x, y) = x3 + y3 + 3axy,
-
au= 3x2 + 3ay and
au
- = 3y2 + 3ax. Therefore,
ax ay
ii) F o r U ( x , y , z ) = "x + xz3 yz
au
- = 2x
au = z and -
+ z3, - au = y + 3x2'. Therefore,
ax ay az
--
ax2 ax ax
-a2u
-
ay ax ' ay
a2u a + 23) = 3z2
-=-
az ax ):(
a~a =
(2x
-a2u
- a (z) = 0
--
ax ay ax
-a2u
- - a (2) = 0
-
ay2 ay
-a2u
- a (z)*=
-- 1
az ay az
a2u *
a ( y + 3x23 = 3z2
-
ax az ax
-a2u
- - - a( y +3xz"= 1
az ay
- a (y +
a2u = - 3x2') = 6x2.
az2 az
Jbmple 2 : I f f (X, y) = x2 tan-' Y - y2 tan-' -,
X
x # 0, y # 0,
X Y
we will prove that
a2f =
- xz - Y2
axay x2+y2
af = x2. 1 .-
1 2~ tan-' -
x;- - y2.
(- F)
1
Here, -
. ay 1 + yZ/~2X Y 1 + x2/y2
X
= x - 2y tan-' -
Y
And therefore,
\ Partial Derivatives In the next example we go a step further and calculate a third order partial
derivative.
Example 3 : If u (x, y, z) = e w , then we can show that
-
a2u = x eXyz+ x 4 z eXYZ,
and
ay az
E l ) Find all the second order partial derivatives of the fouowing functions.
F ;
a) f (x, y) = cos -
X
b) f (x, y) = x5 + Y4 sin x6
c) f (x, y, z) = sin xy + sin yz + cos xz
d) f (x, y, z) = xyz2 + xyz + x3y
E2j If V.(x, y, z) = ( X ~ + ~ ~ + Z ' )show
- " , that
. In Unit 5 y6u have seen that it is not always gossible to find first order partial
derivatives by direct differemtiation (See Examples 5.and 6 of Unit 5). The same is
true for higher order partial derivatives of some functions, This is illustrated by
the folldwing examples.
-
Example 4 : Consider the function
- 0 (x, Y) = (0, 0)
9
We'll now evaluate the second order partial derivatives of f at (0, 0).
f (t, 0) - f (0, 0)
f, (0, 0) = lim -
Higher Order
8-0 Partial Derivatives
= lirn -= 0.
fx (h, 0) = lim
f ( h + t , 0) - f (h,0)
t-0 t
Therefore,
0-0
f, (0, 0) = lim - = 0.
h-0 h
tk ($2 - kZ)
-0
= lirn
t2 + k2
t-0 t
k (tZ- kZ)
= lirn
t-0 t2 + k2
-k-0
Therefore: fv (0, 0) = l h - = -1.
k-0 k
fy (h, 0) = lim
f (h, s) - f 01, 0)
s-0 S
hs (hZ- s 3
-0
= lirn
h2 + s2
s-0 S
h (h2 - s2)
= lim
s-o h2 + s2
h-0
Therefore, f, (0, 0) = lirn - = 1.
h-0 h
Thus, you can see that to evaluate the partial derivatives of this function, we had
to,resort to the definition of partial derivatives, and direct differentiation was not
possible.
In the next example we take up a function which is slightly more complicated.
Example 5 : Let us evaluate fxy(0, 0) and f, (0, O), for the function f given by
fx (0, k) = lim
f (t, k) - f (0, k) .
t -0 t
(t4 + k4) tan-' (k2/t2) - xk4/2
= lim
t-0 t
We hive applied LfHopital's rule By L'Hopital's rule, we have
here, since
(t4+ k4) tan-' (k2/t2) - xit4/2
(-F)
k 1
0
t
is in the - fonn as t-0.
4t3 tan-' - + (k4+t4)
t 1 + (k4/t4)
.
0 fx (0, k) = lim
t -0
1
= lim [4t3 tan-' (k2/t2) - 2k%]
t-0
0-0
Therefore, fxy(0; 0) = lim - = 0
k-0 k
f (0, s) - ((0, 0)
fy (0, 0) = l i i
5-0 S
(a4121 - 0.
= lim
s-0 S
1
4s3 tan-I (s2/h/h4+(h4+s4).
+ s4/h4) (2s/h4
= lim
= lim-[4s3tan-' (s2/h2)
s-0
+ 2sh2]
fy (h, 0-0
0) - fy (0, 0) = lim -
Consequently, f, (0, 0) ,= lim = 0.
h-0 h h-0 h
In Unit 5 you have seen yme examples of functions whose partial derivatives fx, fy
do not exist (see Example 6 of Unit 5).
Here we will give you an examplo of a function whose fust order partial
derivatives exist, but higher order ones do not exist. From this example you will
also see that the existma of a partial derivative of a particular order does not
imply the existena of other partial derivatives of the same order.
Exmpk 6 :Let us emnine whether the second order partial derivatives of f at
(0, 0) exist or not, if f : R2 R is defmcd by --
f (x, y) = (m xy' ,xy # O
f (t, 0) - f (0, 0)
Now, fx (0, 0) = lim
0-0
= lim - = 0.
lim
fx (0, k) -
fx (0, 0)
exists or not.
k-0 k
Therefore, let us fmd fx (0, k), for k # 0.
b2
= lim -O
Therefore, f, (0. 0) = lim
fy (h, 0) - fy (0, 0) = lim
0-0
- = 0
h
h-0 h-0 h
f (0,k+s) - f (0,k)
Again, for k # 0, fy (0, k) = lim = h - 0= -o0
s-0 s 8-0 s
E6) Examine the following functions for equality of f, and f, at (0, 0).
The study of the above examples and exercises must have convinced you that we
have to be careful about the order of variables with respect tp which higher order
derivatives are taken. For instance, from Example 4 it is clear that fxyneed not be
equal to f,. Example 6 goes a step further, where f,, exists at (0, 0). while f,
does not, showing that the question of their equality does not arise at all. If you
look at the definitions of f, and f, at a point (a, b) more carefully, you would
see why the expectation of the equality f, (a, b) = f, (a, b) is farfetched. By
definition
fx (a, b + k) - fx (a, b)
fxy,:(a,b) = lim
k-0 k
= lim
k-0
[ lim
h-0
f(a+ h, b i k) - f(a, b + k) - f(a+ h, b) - f(a, b)
hk
Similarly,
f, (a, b) = h
h-0
[ [
lim
k-0
f(a+h, b + k ) - f(a+h, b)-f(a,
hk
b+k)-f(a, b)
and we have already seen in Unit 4 that repeated limits are not equal, in general.
In the next section we will study the conditions under which these mixed partial
derivatives become equal. '
Hlgher Order
6.3 EQUALITY OF MIXED P A ~ T I A LDERIVATIVES Partid Derlvatlves
\
We shall now give a set of sufficient conditions which would ensure that the order
of the variables with respect to which higher order partial derivatives are taken is
immaterial. In other words, if a function f satisfies thkse conditions, then its mixed
partial derivatives will be equal.
Tbeorem 1 : Let f(x, y) be a real-valued function such that the two second order
partial derivatives f,, and f, are continuous at a point (a, b). Then
fxy (a, b) = f, (a, b)
Proof : The continuity of f,, and f, at (a, b) implies that fx, f, , fxy and f, exist
in a neighbourhood, say D of (a, b).
Consider the expression
$ (h, k) = f(a+h, b + k ) - f(a+h, b) - f(a, b + k ) +
f(a, b),
which is defined for all those real numbers h, k for which (a+ h, b + k) E D.
Let Ih denote the closed interval [a, a + h] or [a + h, a] according as h > 0 or
h < 0. Let G(x) be a real-valued function defined on the closed interval Ih by
G(x) = f(x, b + k) - f(x, b)
so that G(a+ h) - G(a) = $ (h, k). Since, for all x in Ih, the points (x, b + k) and
(x, b) belong to D, it follows that f, (x, b + k) and fx (x, b) exist for all x E Ih.
Now we can write
G ' (x) = fx (x, b + k) - fx (x, b).
Therefore, the function G(x) is differentiable on the closed interval I,,. Thus G(x)
satisfies the requirements of Lagrange's mean value theorem, and we get
$ (h, k) = G(a+h) - G(a) =
h G' (a+8h)
= h [ fx(a+8h, b + k ) 2 fx(a+'8h, b)], .............(1)
where 0 .< 8 < 1.
Now we define a function F : Ik R by -
F(t) = fx (a + Oh, t),
where,Ik is the closed interval [b, b + k] or [b+ k, b] according as k> 0 or k < 0.
Since fxyexists on D, it follows that the function F is differentiable on Ik.
Therefore, by Lagrange's mean value theorem, we get
F(b + k) - F(b) +
= kF1(b B'k) for some 8', 0 < 8' < 1. This means that
fx (a+8h, b+k) - fx(a+Bh, b) = k fxy(a+8h, b+B1k)
Using Equation (I), we get
and consequently
I
lirn
h-0
@ !d!?-= lim
hk h-0
fxy(a+eh, b + e t k j
= fx, (a, b)
as f,, is given to be continuous at (a, b).
Starting with the function
H(Y) = f(a + h, Y) - f(a, Y)
for y E Ik and proceeding exactly as above we can prove that
lim $(h, k) = f, (a, b)
h-0
k-0
and conclude f,, (a, b) = f, (a, b).
This result was proved by L. Euler around 1734, when he was working on some (L. Euler (1707-1783)
problems in hydrodynamics. Later the German mathematician Hermann Amandus
Schwarz (1843-1921) proved another theorem about the equality of mixed partial 519
derivatives. ThePEL- conditions in Schwarz's theorem are less restrictive than those
in Euler's theorem (Theorem 1). We give only the statement of Schwarz's theorem
here.
Tbeanm 2 :(Scbwarz's Tbeonm) : Let f(x, y) be a real-valued function defined in
a neighbourhood of (a, b) such that
i) fy exists on a certain neighbourhood of (a, b).
ii) fxyis continuous at (a, b).
Then f, exists at (a, b) and f, (a, b) = fxy(a, b).
We now give an example to illustrate this.
Exunple 7 :Let us evaluate f, at a point (x, y) for the function f defined by
f(x, y) = x4+ x2y2 + y6. Then we'll use Schwarz's theorem to evaluate f, at the
point (x, y).
By direct differentiation you can see that
fx (x, y) = 4x3 + 2xy2. Therefore, fxy(x, y) = 4xy.
Since 4xy is a polynomial, f, is a continuous function.
Further, f,, (x, y) = 2 x 4 + 6yS exists. Hence f satisfies the conditions of
Schwarz's theorem and so f, (x, y) = f, (x, y) = 4xy.
Are you ready for an exercise now?
E8) Evaluate fxyat a point (x, y) for each of the following functions.
a) T(x, y) .= x2 + xy + y2
b) f(x, y) = ex cos y - eYsin x -
In Euler's Theorem we assume that both the mixed partial derivatives are
continuous, whereas iq Schwarz's theorem we iksume that only one of them, say
f,, is continuous, and that fy exists. But even though the conditions of Schwarz's
theorem are less strict, these are still not necessary for the equality of mixed partial
deiivatives. In other words, we can have functions whose mixed partial derivatives
at some point are equal, but which do not satisfy the requirements of Sohwarz's
theorem. We give one such function in the following example.
ExAple 8 : Consider the function f defined by
We will show that fxy (0, 0) = f, (0, 0), even though f does not fulfil the
requicements.of SchwarZ's thaoiem. .
Now, fx(0, 0) = lim f0,0) f(0, 0)
h-0 h
0-0
= lim
h-0
-h
= 0.
I .- Also, for' # 0,
fx(0, y) = lim f(h, Y) - f(O, Y)
h-0 h
= lirn -hy2
h-0 h2+y2
fy (x, 0) = lirn
f(x. k) - f(x, 0)
k-0 k
We'll now show that the conditions of Schwarz's theorem are not satisfied. Now,
'
for x # 0, y # 0, we can find the partial derivatives of f at (x, y) by
differentiating directly. Thus,
Further, f, (x, y) = -
8x3y3 8x3y3
Now, lirn does not exist. Put y = rnx in a i d ,take the
(x. Y)-(0, 0) (x2+y2)3 (x2+ y213
limit as x-0. You will find that the limit is different for different values of m.
This means that lim fxy(x, y) does not exist, which implies that fxyis not
I (a, y)-(0, 0)) .
continuous at (0, 0).
There is another criterion which tells us when fxyequals f,, at a particular point.
We state this also withoht proof.
Theorem 3 (Young's Theorem) : Let f(x, y) be a real-valued function defined in a
neighbourhood of a point (a, b) such that both the first qrder partial derivatives f,
and fy are differentiable at (a, b). Then fxy(a, b) = fb' (a, b).
As in the case of Schwarz's theorem the conditions stated in Young's theorem are
less strict than in Theorem 1. However, these are not necessary for the equality of
mixed partial derivatives. 'Our next example illustrates this fact.
Example 9 : Consider the function f in Example 8.
2hk
or,
(hZ+k2)2
= h 4(h, k) + k $(h, k).
We'll show that fxy(0, 0, 0) # f, (0, 0, 0), whereas f, (0, 0, 0) = f, (0, 0, 0).
Let us first calculate fxy(0, 0, 0). For this we need to evaluate fx (0, 0, 0) and
fx (0, k, 0). Now
fx (0, 0, 0) = Iim
f(P, 0, 0) - f(0, 0, 0) = lim
0-0
- = 0 and '
P-0 P P-0 p
f(P, k, 0) - f(O, k, 0)
f, (0, k, 0) = lim
p3k pk3 - -0 Hlgber Order
p2+k2 PartiPI Derivatives
= lirn
P-0
P
p2k - k3
= lim = -k.
P-0 P2+k2
= lim
-k-O
-
k-0 k
I = -1.
Next, we'll evaluate f, (0, 0, 0). For this we need-fy(h, 0, 0) and f, (0, 0, 0).
f, (0, 0, 0) = lirn
fy(h. 0, 0) - fJ0, 0. 0)
h-0 h
f, (h, 0, C) = lirn
f01, 0, r) - f(h. 0, 0) = 0
r-0 f
O , otherwise
f(h, 0, 0) - f(0, 0, 0)
Now, fx(0, 0, 0) = lirn = 0, and
h-0 h
,y # 0, z , # 0
0 , otherwise
Show that at the origin fxy,f,, f, and f, all exist, but neither f, nor f,
exists.
6.4 SUMMARY
In this h i t , we have
1) ~ntrducedpartial derivatives of order more than one.
2) Evaluated these higher order partial derivatives for various functions.
3) 'studied examples of functions which show that, in general, the two partial
derivatives of order more than one obtained by changing the order of variables
'are not equal in value, even if both exist.
4) Applied the following three sets of sufficient conditions which ensure the
equality of f, (a,- b) and f, (a, b).
"~uler'stheorem sitye that if f, and f, are both continuous at a point (a, b),
, then
- - -y2 C O S -Y- -
- ',ZY . y
sin -
'
x4 X x3 x
= - s1i n - - Y + -Y
-s- Y
x2 X x3 X
f, = -1~ h Y- - -
x2 X . X
(
a:) (-5)
= -1s i n Y + Y
-COO-_-
Y
x2 X x3- X
1
f y y = - - COO- Y
x2 X
a2f - a2f
So, - -
ax ay ay ax
a2f
ay ax
= 3y%ec"xy3) + 6xy5 sec2 (xy3)tan (xy3) .
a2f
= 3 y ~ e c ~ x y+
3 )6xy5 spc2 (xy3) tan (xy3)
kay
a2f a2f
So, - = -
ayax hay ..
*
E4) xXyy zz = c
Taking logarithms on both'sides, we get
x lnx + y lny + z lnz = lnc.
Differentiating with respact to y, treating z as a function of x and y, we get
= az - lny + 1
" ay lnz + 1
. - =a2z lny + 1 az-
-1- = In ey In ex
" ax@ o n ~ + i ) z~ ' ax z (In e ~ ) ~
Putting x = y = z, we get
-hk-0
= lim (since k is fixed we can suppase
h-O that(h1 < Ik() '
For,'h # 0, fy (h, 0) = lirn f(h, k) - f(h, 0)
k-0 k
hk - 0
= lim
k-o
- k
(since h is fmed we can suppose
that IkJ c IhJ)
= h.
fy (x, Y) = x + 5
f, (x, Y) = 1
Clearly fy exists everywhere and f, is continuous being a constant
function. This shows that f satisfies the requirements of Schwarz's
theorem. Hence, by Schwarz's theorem, f, exists and f, = fxy = 1.
b) f(x, y) = ex cos y - ey sin x
.'. fy (x, y) = - ex sin y - eY sin x
axid fxy (x, y) = - ex sin y - eYcos x
It is easy to sbe that fy exists and fxyis continuous. So, in view of .
Schwarz's theorem, f, exists and