Equality of Mixed Partials

2 f 2 f
Theorem. If the partial derivatives xy
and yx
exist and are continuous at (x0 , y0 ), then
2 2
f f
xy (x0 , y0 ) = yx (x0 , y0 )

Proof: Here is an outline of the proof. The details are given as footnotes at the end of the
outline. Fix x0 and y0 and define(1)
1
 
F (h, k) = hk f (x0 + h, y0 + k) f (x0 , y0 + k) f (x0 + h, y0 ) + f (x0 , y0 )
Then, by the mean value theorem,
h i
2
F (h, k) = h1 f
y (x0 + h, y0 + 1 k)
f
y (x0 , y0 + 1 k)
3 f
= x y (x0 + 2 h, y0 + 1 k)
h i
4 1 f f
F (h, k) = k x
(x0 + 3 h, y0 + k) x
(x0 + 3 h, y0 )
5 f
= y x
(x0 + 3 h, y0 + 4 k)
for some 0 < 1 , 2 , 3 , 4 < 1. All of 1 , 2 , 3 , 4 depend on x0 , y0 , h, k. Hence
f f
x y
(x0 + 2 h, y0 + 1 k) = y x
(x0 + 3 h, y0 + 4 k)
for all h and k. Taking the limit (h, k) (0, 0) and using the assumed continuity of both partial
derivatives at (x0 , y0 ) gives
f f
x y
(x0 , y0 ) = y x
(x0 , y0 )

The Details
2 2
f f
(1) We define F (h, k) in this way because both partial derivatives xy (x0 , y0 ) and yx (x0 , y0 )
are defined as limits of F (h, k) as h, k 0. For example,
h i
f 1 f f
y x (x0 , y 0 ) = lim (x0 , y 0 + k) (x 0 , y 0 )
k0 k x x
h i
= lim k1 lim f (x0 +h,y0 +k)f
h
(x0 ,y0 +k)
lim f (x0 +h,y0 )f (x0 ,y0 )
h
k0 h0 h0
f (x0 +h,y0 +k)f (x0 ,y0 +k)f (x0 +h,y0 )+f (x0 ,y0 )
= lim lim hk
k0 h0
= lim lim F (h, k)
k0 h0
Similarly,
h i
f 1 f f
x y
(x0 , y0 ) = lim (x0 + h, y0 ) (x0 , y0 )
h0 h y y
h i
f (x0 +h,y0 +k)f (x0 +h,y0 )
= lim 1
lim lim f (x0 ,y0 +k)f (x0 ,y0 )
h0 h k0 k k0 k

= lim lim f (x0 +h,y0 +k)f (x0 +h,y

hk
0 )f (x0 ,y0 +k)+f (x0 ,y0 )

h0 k0
= lim lim F (h, k)
h0 k0

c Joel Feldman. 2004. All rights reserved. 1
(2) The mean value theorem says that, for any differentiable function (x), the slope of the line



joining the points x0 , (x0 ) and x0 + k, (x0 + k) on the graph of is the same as the
slope of the tangent to the graph at some point between x0 and x0 + k. This is, there is some
0 < 1 < 1 such that
y
y = (x)
(x0 +k)(x0 ) d
k = dx (x0 + 1 k)
x
x0 x0 +1k x0 +k

Applying this with x replaced by y and replaced by G(y) = f (x0 + h, y) f (x0 , y) gives

G(y0 +k)G(y0 ) dG
k
= dy
(y0 + 1 k) for some 0 < 1 < 1
f f
= y (x0 + h, y0 + 1 k) y (x0 , y0 + 1 k)

Hence, for some 0 < 1 < 1,

h i h i
F (h, k) = h1 G(y0 +k)G(y
k
0)
= 1 f
h y
(x0 + h, y0 + 1 k) f
y
(x0 , y0 + 1 k)

f
(3) Define H(x) = y
(x, y0 + 1 k). By the mean value theorem
h i
1
F (h, k) = h
H(x0 + h) H(x0 )
dH
= dx
(x0 + 2 h) for some 0 < 2 < 1
f
= x y (x0 + 2 h, y0 + 1 k)

(4) Define A(x) = f (x, y0 + k) f (x, y0 ). By the mean value theorem

h i
1 A(x0 +h)A(x0 )
F (h, k) = k h
1 dA
= k dx (x0 + 3 h) for some 0 < 3 < 1
h i
1 f f
= k x
(x0 + 3 h, y0 + k) x
(x0 + 3 h, y0 )

f
(5) Define B(y) = x
(x0 + 3 h, y). By the mean value theorem
h i
1
F (h, k) = k B(y0 + k) B(y0 )
dB
= dy (y0 + 4 k) for some 0 < 4 < 1
f
= y x
(x0 + 3 h, y0 + 4 k)

c Joel Feldman. 2004. All rights reserved. 2