Change of Variable Formula for Multiple Integrals

S. Kumaresan and G. Santhanam

Let f : [a, b] R be a continuous function and : [c, d] [a, b] be a C
1
-function. Then
the change of variable formula for one variable integrals says that
_
t
c
f((x))

(x)dx =
_
(t)
(a)
f(y)dy. (1)
This formula is very easily established via the fundamental theorem of integral calculus.
For multiple integrals the change of variable formula is
Theorem 1. Let U and V be open sub sets of R
n
and : U V be a C
1
dieomorphism.
Then, given a continuous function f : V R with compact support, we have
_
U
f((x))J((x))dx =
_
V
f(y)dy, (2)
where J((x)) stands for the determinant of the Jacobian matrix of .
The usual proof of this theorem is got by approximating the integral by nite sums and
this involves carefully estimating the volumes of images of small cubes under the map
and other small details.
Recently P.D. Lax[1] found a new approach and proved the following variant of the
change of variable formula.
Theorem 2. Let : R
n
R
n
be smooth such that (x) = x for all x with suciently
large norm. Then for any continuous function f : R
n
R with compact support, we have
_
f((x))J((x)) dx =
_
f(y) dy. (3)
The aim of this article is to deduce the standard version of the change of variable
formula 1. To make the article self-contained, we include the proof of Theorem 2 towards
the end of this article.
The proof of the form of the change of variable formula as in Theorem 2 is simple and
highly original. The proof uses only the change of variable formula for one variable. Its
only shortcoming is that it is not in the standard form. To remedy this, we deduced the
standard version of the formula from the version due to Lax as soon as we learnt of the
proof by Lax. Our proof was so simple that we did not think of writing it up. However,
1
in 2002, we came across another paper by Lax [2] in which he uses his version to deduce
other variants of the change of variable of formula but not the standard version. The
proofs of other versions are also rather long and it spoils the simplicity and originality of
the rst result, namely 2. It is worth mentioning that Michael Taylor [3] also proves the
version due to P.D. Lax in the language of dierential forms and he too does not prove the
standard version. These facts convinced us that though our proof is simple and appears
to be natural, it may not be so. Hence we decided to write it up.
Proof of Theorem 1:
Let : U V be a smooth dieomorphism of an open set U R
n
to V . Let f C

(V )
be of compact support. Let S :=
1
(Supp f). Then S is compact and hence contained in
a cube of suciently large side or any ball of suciently large radius. Consider the open
cover {U
1
:= U, U
2
:= R
n
\ S} of R
n
. Let {f
j
} be the partition of unity subordinate to this
cover. We dene
(x) = f
1
(x)(x) + f
2
(x)x, x R
n
.
Since the support of f
2
is a subset of R
n
\ S and hence is disjoint from S, we see that
f
2
(x) = 0 for x S. It follows that for x S, f
1
(x) = 1 and hence (x) = (x). On the
other hand, if x U
2
, then x / S. Since U
1
U
2
= U
1
\ S, the support of f
1
is contained
in a compact neighbourhood of S in U
1
= U. Thus, if x U
2
but not in the support of f
1
,
then f
1
(x) = 0 so that (x) = x for x outside a compact set containing S. Therefore, we
can now apply the version due to Lax, viz., theorem 2 and get
_
f((x))J((x)) dx =
_
f(y) dy. (4)
To deduce the standard version, we need only observe that the LHS of Equation (4) is
nothing but
_
f((x))J((x)) dx.
Remark 1. We observe that we needed only the fact that is a proper map in our proof.
Now we prove theorem 2. Before we start with the proof of theorem 2, we x the follow-
ing notations: We write
x
i
instead of

x
i
,
y
i
instead of

y
i
and we write (x
1
, x
2
, . . . , x
n
) =
(
1
(x
1
, . . . , x
n
),
2
(x
1
, . . . , x
n
), . . . ,
n
(x
1
, . . . , x
n
)).
Proof of theorem2:
We prove the result only for dierentiable functions.
Let f be a dierentiable function with compact support and g : R
n
R be the function
dened by
g(y
1
, y
2
, , y
n
) :=
_
y
1

f(z, y
2
, y
3
, , y
n
)dz. (5)
Let us now observe that
y
1
g = f. Since the function f has compact support, there exists
a real number c such that Support(f) is contained in the n-cube dened by the equations
|y
i
| c for all i = 1, 2, . . . , n. It follows from equation (5) that g(y
1
, y
2
, , y
n
) = 0
when |y
j
| c for some j 2 or when y
1
c.
2
We assume that (x) = x for all x such that x r. We may take c large so that the
c-cube contains the ball of radius r in R
n
. Since is identity outside the ball of radius r,
it follows that f((x)) = 0 outside the c-cube. So in the integrals in equation (4) we can
restrict the integration to the c-cube.
In the left hand side of the equation (4) we substitute f =
y
1
g and get the expression
_

y
1
g((x))J((x))dx. (6)
Then
x
j
(g ) =

n
i=1
(
y
i
g)(
x
j

i
) for j = 1, 2, . . . , n and the gradient
D(g ) = (
x
1
(g ), . . . ,
xn
(g ))
=
_
n

i=1
(
y
i
g)(
x
1

i
), . . . ,
n

i=1
(
y
i
g)(
xn

i
)
_
. (7)
If we let D
j
= (
x
1

j
, . . . ,
xn

j
) denote the gradient of the function
j
: R
n
R, then
we can write the equation (7) as
D(g ) =
n

i=1

y
i
g D
i
. (8)
Therefore
det
_
_
_
_
_
D(g )
D
2
.
.
.
D
n
_
_
_
_
_
= det
_
_
_
_
_
_
_

n
j=1

y
i
g D
i
D
2
D
3
.
.
.
D
n
_
_
_
_
_
_
_
=
n

i=1

y
i
g det
_
_
_
_
_
_
_
D
i
D
2
D
3
.
.
.
D
n
_
_
_
_
_
_
_
= (
y
1
g)(J((x))).
Let us now expand the determinant det(D(g ),
2
, . . . ,
n
) with respect to the rst
row to obtain an expression of the form

i
M
i

x
i
(g ), (9)
where M
i
s are the co-factors of the rst row (
x
1
(g ), . . . ,
xn
(g )) of the Jacobian
3
matrix
_
_
_
_
_
D(g )
D
2
.
.
.
D
n
_
_
_
_
_
. If we now substitute the expression (9) into the integrand in equa-
tion (6), we get
_
(M
1

x
1
(g ) + M
2

x
2
(g ) + + M
n

xn
(g ))dx. (10)
Since the function is twice dierentiable, we can integrate each term by parts over the
c-cube and get

_
(g )(x)
_

x
i
M
i
_
dx + boundary terms. (11)
Now we prove the identity

x
i
M
i
= 0.
This is a classical identity. For the sake of completeness we give a proof of this identity
below. Another proof can also be found in [4].
Proof. Let us recall that
M
i
= (1)
i+1

(i)=i
sgn()

j=i

x
j

(j)
.
Hence

x
i
M
i
= (1)
i+1

(i)=i
sgn()
x
i
_

j=i

x
j

(j)
_
= (1)
i+1

(i)=i)
sgn()
_

x
i

x
1

(1)

x
1

(n)
+
+
x
1

(1)

x
i

xn

(n)
_
= (1)
i+1
n

k=1
m
i,k
,
where m
i,k
denotes the determinant of the matrix
_
_
_

x
1

2

x
i

x
k

2

xn

2
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

x
1

n

x
i

x
k

n

xn

n
_
_
_
for k = i and m
i,i
= 0.
4
Now we shift the k-th column
_
_
_

x
i

x
k

2
.
.
.

x
i

x
k

n
_
_
_
of the matrix m
i,k
to the rst column and
denote the resulting matrix by M
i,k
. Then

x
i
M
i
= (1)
i+1
n

k=1
(i, k)M
i,k
where (i, k) = (1)
k
if k > i and (1)
k1
if k < i. Therefore
n

i=1

x
i
M
i
=
n

i,k=1
(1)
i+1
(i, k)M
i,k
.
Since M
i,k
= M
k,i
, it follows that
n

i=1

x
i
M
i
=
n

i,k=1
(1)
i+1
(i, k)M
i,k
=

k>i
(1)
i+k+1
M
i,k
+

k<i
(1)
i+k
M
i,k
+
= (1)

k>i
(1)
i+k
M
i,k
+

k>i
(1)
i+k
M
k,i
= (1)

k>i
(1)
i+k
M
i,k
+

k>i
(1)
i+k
M
i,k
= 0.
This completes the proof of the identity.
Remark 2. For those who know dierential forms, the proof of this identity is easy. We
need only observe that d(d
2
d
3
. . . d
n
) = (

n
i=1

x
i
M
i
)dx
1
dx
2
. . . , dx
n
and
use the fact d
2
= 0.
We now attend to the boundary integral in equation (11). Since g (x) = g(x), on the
boundary of the c-cube, the only non-zero boundary term will come from the side x
1
= c.
Also, since M
1
= 1, when (x) = x, that boundary term is
_
g(c, x
2
, . . . . . . , x
n
)dx
2
dx
3
dx
n
. (12)
Using the denition of the function g in equation (12) gives
_ _
c

f(z, x
2
, . . . . . . , x
n
)dzdx
2
dx
n
.
This is nothing but the right hand side of the equation (4).
5
References
[1] P.D.Lax, Change of Variables in Multiple Integrals, Amer. Math. Monthly, pp. 497
501, 1999.
[2] P. D. Lax, Change of Variables in Multiple Integrals II, Amer. Math. Monthly,
pp.115-119, 2001.
[3] Michael Taylor, Dierential Forms and the Change of Variable Formula for Multiple
Integrals, Vol. 268, pp.378-383, 2002.
[4] N. Dunford and J. Schwartz, Linear Operators, Part-I, pp. 476, Wiley-Interscience,
New York 1958.
S. Kumaresan,
Department of Mathematics,
Mumbai University,
Mumbai- 400 098, India.
e-mail: kumaresa@sankhya.mu.ac.in
G. Santhanam,
Department of Mathematics,
Indian Institute of Technology,
Kanpur-208016, India
e-mail: santhana@iitk.ac.in
6