Chapter 5

Change of Variables
Formula, Improper
Multiple Integrals

In this chapter we discuss the change of variables formula for multi-

ple integrals, improper multiple integrals, functions defined by integrals,
Weierstrass’ approximation theorem, and the Fourier transform.

5.1 Change of variables formula

The “change of variable” formula, also called the “u-substitution rule”
in one variable is known from elementary calculus and tells us that; if
f : [a, b] → R is continuous on [a, b] and ϕ : [a, b] → R is a one-to-one
function of class C 1 with ϕ0 (x) > 0 for x ∈ (a, b), then with u = ϕ(x),
Z b Z ϕ(b)
f (ϕ(x))ϕ0 (x)dx = f (u)du. (5.1)
a ϕ(a)

To prove this, first we find a differentiable function F such that F 0 = f ,

that is, an antiderivative F of f (this is possible by the Fundamental
Theorem of Calculus). Then
Z ϕ(b)
f (u)du = F (ϕ(b)) − F (ϕ(a)).
ϕ(a)

293
294 Chapter 5 Change of Variables Formula

On the other hand, by the Chain Rule (F ◦ ϕ)0 (x) = f (ϕ(x))ϕ0 (x).
Hence, again by the Fundamental theorem,
Z b
f (ϕ(x))ϕ0 (x)dx = (F ◦ ϕ)(b) − (F ◦ ϕ)(a) = F (ϕ(b)) − F (ϕ(a)).
a

However, one has to be careful here, for (5.1) is fine as it stands when
ϕ is increasing. If ϕ were decreasing, (i.e., ϕ0 (x) < 0) the right side of
(5.1) is
Z ϕ(a) Z ϕ(b)
f (u)du = − f (u)du.
ϕ(b) ϕ(a)

This is corrected by rewriting (5.1) as

Z b Z ϕ(b)
0
f (ϕ(x))|ϕ (x)|dx = f (u)du.
a ϕ(a)

Setting Ω = [a, b], so that ϕ(Ω) is the interval with endpoints ϕ(a) and
ϕ(b) we write
Z Z
f (u)du = f (ϕ(x))|ϕ0 (x)|dx. (5.2)
ϕ(Ω) Ω

We remark that this result is also valid if f is not continuous, but merely
integrable, because sets of measure zero contribute nothing to either
integral.

We shall now generalize (5.2) for multiple integrals. First we define

what is meant by a “change of variables” in an n-dimensional integral.

Definition 5.1.1. Let U be an open set in Rn . Let ϕ : U → Rn be

a one-to-one function of class C 1 (U ), such that the Jacobian Jϕ (x) =
det Dϕ (x) 6= 0 for all x ∈ U . Then ϕ is called a change of variables in
Rn .

Note our hypothesis is slightly redundant, for if ϕ : U → ϕ(U ) is

a one-to-one function such that both ϕ and ϕ−1 are of class C 1 , that
is, if ϕ is a diffeomorphism, then the Chain Rule implies that Dϕ is
nonsingular, so that det Dϕ (x) 6= 0 and hence ϕ is a change of variables.
5.1 Change of variables formula 295

On the other hand, if ϕ is a change of variables in Rn , then the Inverse

Function theorem (Theorem 3.8.2) and Corollary 3.8.3 tell us ϕ(U ) is
open in Rn and the function ϕ−1 : ϕ(U ) → U is of class C 1 . In other
words, a change of variables in Rn is just a diffeomorphism in Rn .

The basic issue for changing variables in multiple integrals is as fol-

lows: Let U be open in Rn and ϕ : U → Rn a diffeomorphism. Suppose
Ω is a bounded simple set (i.e., its boundary has n-dimensional volume
zero) with Ω ⊂ U . Given R an integrable function f : ϕ(Ω) → R we want
to change the integral ϕ(Ω) f (y)dy into an appropriate integral over
Ω (which we hope to be easier to compute). In fact, we shall prove the
following n-dimensional analogue of (5.2)
Z Z
f (y)dy = f (ϕ(x))|Jϕ (x)|dx. (5.3)
ϕ(Ω) Ω

Since the sets over which we integrate are bounded simple sets it is
natural to ask, whether a diffeomorphism maps (bounded) simple sets
to (bounded) simple sets.

Lemma 5.1.2. Let U be open in Rn , ϕ : U → Rn a diffeomorphism

whose resrtiction on Ω◦ is a diffeomorphism and Ω a bounded set with
Ω ⊂ U . Then
∂(ϕ(Ω)) = ϕ(∂(Ω)).

Proof. Since ϕ is continuous and Ω is compact (by the Heine-Borel

theorem), it follows that ϕ(Ω) is also compact and so ϕ(Ω) is closed and
bounded. Hence, ϕ(Ω) = ϕ(Ω). At the same time, since ϕ ∈ C 1 (Ω)
and Jϕ (x) 6= 0 for all x ∈ Ω, by Corollary 3.8.3, ϕ maps open sets onto
open sets. In particular, ϕ(Ω◦ ) is open. Thus, points of ∂(ϕ(Ω)) can
not be images of points of Ω◦ = Ω\∂Ω, that is, ∂(ϕ(Ω)) ⊆ ϕ(∂(Ω)).
On the other hand, let x ∈ ∂(Ω). Then there are sequences {xk } in
Ω and {yk } in U \Ω such that xk → x and yk → x. The continuity of
ϕ implies ϕ(xk ) → ϕ(x) and ϕ(yk ) → ϕ(x). Since ϕ is one-to-one on
U , it follows that ϕ(yk ) ∈
/ ϕ(Ω) and hence ϕ(x) ∈ ∂(ϕ(Ω)). Therefore
ϕ(∂(Ω)) ⊆ ∂(ϕ(Ω)). Thus, ∂(ϕ(Ω)) = ϕ(∂(Ω)).
296 Chapter 5 Change of Variables Formula

Remark 5.1.3. In developing the theory of integration one uses n-

dimensional rectangles in a number of places. However one could use
n-dimensional cubes instead. An n-dimensional open cube centered at
a ∈ Rn of side length 2r is the set

Cr (a) = {x ∈ Rn : ||a − x||∞ < r} ,

where the norm is the“max-norm” (or “box-norm”)

||x||∞ = max {|x1 |, ..., |xn |} .

As we have proved in Theorem 1.7.2 all norms in Rn are equivalent, in

√
particular ||x||∞ ≤ ||x|| ≤ n||x||∞ . So that a cube C in Rn contains
a ball B in Rn and vice versa. Furthermore, knowing beforehand that
a function f is integrable over a bounded simple set Ω contained in a
cube C, by using a partition P = {C1 , ..., Ck } of C involving subcubes
of equal side length 2r, (so that ||P|| → 0 if and only if r → 0 if and
only if k → ∞), one obtains approximating Riemann sums of f that
have substantial computational advantages. In the sequel we shall use
n-dimensional cubes as our basic sets rather than rectangles in Rn .
Proposition 5.1.4. Let S ⊂ Rn be a set of measure zero and suppose
that ϕ : S → Rn satisfies a Lipschitz condition. Then ϕ(S) has measure
zero.
Proof. Since ϕ satisfies Lipschitz condition, there is M > 0 such that,

||ϕ(x) − ϕ(x0 )|| ≤ M ||x − x0 ||,

for all x, x0 ∈ S. Because S has measure S zero, given P> 0, there is a

sequence of closed cubes Cj such that S ⊂ ∞ C and ∞ j=1 ν(Cj ) < .
√j=1 j
Let the side length of Cj be rj and dj = nrj its diameter. Then for
x, x0 ∈ S ∩ Cj we have
√
||ϕ(x) − ϕ(x0 )||∞ ≤ ||ϕ(x) − ϕ(x0 )|| ≤ M ||x − x0 || ≤ M n||x − x0 ||∞ .
√
Therefore ||ϕ(x) − ϕ(x0 )||∞ ≤ M ndj , and so ϕ(S ∩ Cj ) is contained in
√
the cube Kj with side length 2M ndj . The cubes Kj cover ϕ(S) and
∞
X ∞
X
ν(Kj ) = (M n)n ν(Cj ) ≤ (M n)n .
j=1 j=1
5.1 Change of variables formula 297

Since (M n)n is constant and  > 0 is arbitrary, ϕ(S) has measure zero.

Lemma 5.1.5. Let U ⊆ Rn be open and ϕ : U → Rn be C 1 (U ). Suppose

S is a compact set in U . If S has volume zero, then ϕ(S) also has volume
zero.

Proof. Since U is open, for each x ∈ S we can choose an open ball

Bx = Bδx (x) ⊆ U . The collection {Bx : x ∈ S} is an open cover for
S. Since
S S is compact a finite number of these balls cover S. That is,
S⊆ N j=1 Bxj ⊆ U .
Let Sj = S ∩ Bxj . Since Sj ⊂ S and ν(S) = 0 we have ν(Sj ) ≤
ν(S) = 0. Since ϕ ∈ C 1 (U ), Corollary 3.4.8 tells us that ϕ satisfies a
Lipschitz condition on Bxj ans so on Sj . Proposition 5.1.4 then implies
that ν(ϕ(Sj ) = 0 for all j = 1, ..., N (here because S is compact the
notions of volume
S zero and measure SNzero coincide).
Now S = N
SN
j=1 [S ∩ B xj ] = j=1 Sj and so ϕ(S) = j=1 ϕ(Sj ).
Therefore
 
N
[ XN
ν(ϕ(S)) = ν  ϕ(Sj ) ≤ ν(ϕ(Sj )) = 0.
j=1 j=1

Proposition 5.1.6. Let U be open in Rn and ϕ : U → Rn a diffeo-

morphism. If Ω is a bounded simple set with Ω ⊂ U , then ϕ(Ω) is also
bounded and simple.

Proof. First note that since ϕ(Ω) ⊆ ϕ(Ω) and ϕ(Ω) is compact (i.e.,
closed and bounded), the set ϕ(Ω) is also bounded. At the same time,
since Ω is simple, ν(∂(Ω)) = 0. It follows from Lemma 5.1.5 that
ν(ϕ(∂(Ω))) = 0. Now since ϕ is a diffeomorphism, Lemma 5.1.2 tells us
∂(ϕ(Ω)) = ϕ(∂(Ω)). Hence ν(∂(ϕ(Ω))) = ν(ϕ(∂(Ω))) = 0, and ϕ(Ω) is
simple.

5.1.1 Change of variables; linear case

In this subsection we prove the linear change of variables formula. Here
ϕ = T , where T is a linear map on Rn , in which case JT = det T is a
298 Chapter 5 Change of Variables Formula

4 4

3 3

2 2
T
c T(c)
1 1

-1 1 2 3 4 -1 1 2 3 4

-1 -1

Figure 5.1: Linear Transformation

constant. The proof does not use elementary transformations, rather it

uses the polar decomposition.

Proposition 5.1.7. Let T : Rn → Rn be a linear map, and Ω ⊂ Rn a

bounded simple set. Then

ν(T (Ω)) = | det T |ν(Ω). (5.4)

Proof. We identify T with its standard matrix representation. If T

is singular (i.e., det T = 0), then T (Rn ) is a proper subspace of Rn
of dimension k < n. By Corollary 4.1.25, ν(T (Ω)) = 0, and so the
statement holds trivially.
Now suppose T is invertible (i.e., det T 6= 0). Then T is an iso-
morphism on Rn . Proposition 5.1.6 tells us that T (Ω) is a bounded
simple set. By the Polar Decomposition theorem (Theorem 2.4.11),
T can be written as T = OP , where O is an orthogonal matrix
and P a positive definite symmetric matrix. Therefore P is orthog-
onally conjugate to a diagonal matrix D with positive entries. That
is, P = O1 DO1−1 where O1 and O1−1 are orthogonal matrices. Since
| det(AB)| = | det(A)|| det(B)| and if A is orthogonal | det A| = 1,
(see, the discussion following Definition 2.4.9)) we see that | det T | =
| det((OO1 )DO1−1 )| = det D.
First we shall prove (5.4) for a cube C = [−s, s]n in Rn . Since any
orthogonal transformation O is an isometry (Theorem 2.4.10), it leaves
5.1 Change of variables formula 299

distances fixed. Hence shapes are fixed and so the geometric effect of
the application of O leaves volumes unchanged. Thus it is enough to
see how D affects the volume of C. Since D is diagonal with positive
diagonal entries λ1 , ..., λn we have

D(C) = [−λ1 s, λ1 s] × ... × [−λn s, λn s].

Therefore

ν(D(C)) = (2λ1 s) · · · (2λn s) = λ1 · · · λn (2s)n = (det D)ν(C).

Hence
ν(T (C)) = | det T |ν(C).
Now, let C be a cube containing Ω and partition C into subcubes of
equal side length 2r. Since Ω is simple its characteristic function χΩ is
integrable over C, that is, ν∗ (Ω) = ν(Ω) = ν ∗ (Ω) (see Remark 4.1.41).
Let Ci be the subcubes for which Ci ⊂ Ω with i = 1, ...p and Sp Cj be
thoseSsubcubes for which S Cj ∩ Ω 6= ∅ with j = 1,S..., q. Then i=1 Ci ⊂
Ω ⊂ qj=1 Cj . Therefore pi=1 T (Ci ) ⊂ T (Ω) ⊂ qj=1 T (Cj ). Since the
result holds for cubes, it follows that
p
X q
X
| det T | ν(Ci ) ≤ ν(T (Ω)) ≤ | det T | ν(Cj ).
i=1 j=1

Letting r → 0 we get

| det T |ν(Ω) ≤ ν(T (Ω)) ≤ | det T |ν(Ω).

Thus again, ν(T (Ω)) = | det T |ν(Ω).

Corollary 5.1.8. (The volume of a ball in Rn ). Let B n (r) =

{x ∈ Rn : ||x|| ≤ r} be the ball of radius r > 0 in Rn . Then

ν(B n (r)) = cn r n ,

where cn = ν(B n ) is the volume of the n-dimensional unit ball B n .1

1
Explicit expressions for cn are given in Examples 5.1.33 and 5.3.27.
300 Chapter 5 Change of Variables Formula

Proof. Since ||rx|| = r||x||, the ball B n (r) is the image of B n under
the linear map T (x) = rx on Rn . Furthermore, since det T = r n ,
Proposition 5.1.7 gives ν(B n (r)) = ν(T (B n )) = r n cn . For example, for
n = 3, Example 4.3.19 tells us c3 = 34 π, and of course c2 = π.

Theorem 5.1.9. Let Ω ⊂ Rn be a bounded simple set and let T : Rn →

Rn be a nonsingular linear map. If f : T (Ω) → R is an integrable
function, then (f ◦ T )| det T | is integrable over Ω and
Z Z
f (y)dy = f (T (x))| det T |dx. (5.5)
T (Ω) Ω

Proof. For x ∈ Ω let F (x) = f (T (x)) and set y = T (x). Let C be a cube
containing Ω and partition C into subcubes P = {Cj : j = 1, ..., m} of
equal side length 2r. Then for each j, the sets {T (Cj )} form a partition
of T (Ω). From Proposition 5.1.7, ν(T (Cj )) = | det T |ν(Cj ).

Since f is integrable over T (Ω), by the definition of the integral

over a set, f χT (Ω) is integrable over (any) cube C 0 containing T (Ω).
Moreover, since χΩ = χT (Ω) ◦ T , for any yj = T (xj ) ∈ T (Cj ) we can
write
m
X m
X
f χT (Ω) (yj )ν(T (Cj )) = F χΩ (xj )| det T |ν(Cj ) = | det T |SP (F χΩ ).
j=1 j=1
(5.6)
Let d = maxj=1,...,m {d(T (Cj )}. The continuity of T implies that as
r → 0 then also d → 0.
RNow passing to the limit in (5.6) as r → 0 the first sum converges
to T (Ω) f (y)dy, which automatically implies the existence of the limit
of the second sum (i.e., the integrability of F = f ◦ T on Ω ) and the
equality
Z Z Z
f (y)dy = | det T | f (T (x))dx = f (T (x))| det T |dx.
T (Ω) Ω Ω
5.1 Change of variables formula 301

The meaning of the determinant

Definition 5.1.10. Let {v1 , ..., vk } be linearly independent vectors

in Rn (k ≤ n). We define the k-dimensional parallelepiped Π =
Π(v1 , ..., vk ) with adjacents sides the vectors v1 ,...,vk (also called the
edges of Π) to be the set of all x ∈ Rn such that

x = c1 v1 + ... + ck vk ,

where cj are scalars with 0 ≤ cj ≤ 1, for j = 1, ..., k.

A 2-dimensional parallelepiped is a parallelogram, and a higher di-

mensional analogue is called a parallelepiped. The 3-dimensional version
of the next result was obtained in Proposition 1.3.35.

Proposition 5.1.11. Let v1 , ..., vn be n linearly independent vectors in

Rn and let Π be the parallelepiped Π(v1 , ..., vn ). If A = [v1 ...vn ] is the
n × n matrix with columns the vectors v1 , ..., vn , then

ν(Π) = | det A|.

Proof. Let T be the linear transformation T : Rn → Rn given by T (x) =

Ax. Then T (ei ) = vi , where {ei : i = 1, ..n} are the standard (unit)
basic vectors in Rn . Therefore, since T is linear it maps the unit cube
C n = [0, 1]n onto the parallelepiped Π = Π(v1 , ..., vn ). Proposition 5.1.7
implies
ν(Π) = | det T |ν(C n ) = | det T |.

5.1.2 Change of variables; the general case

Here we shall extend Theorem 5.1.9 for nonlinear change of variables.
The principal idea is based on the local approximation of a C 1 mapping
ϕ : U ⊂ Rn → Rn at a point a ∈ U by its differential Dϕ (a). The local
replacement of a nonlinear relation by a linear one is a basic idea of
mathematics. Recall from the Linear Approximation theorem (Theorem
3.1.3) that Dϕ (a) : Rn → Rn is such that

Dϕ (a)(h) ≈ ϕ(a + h) − ϕ(a)

302 Chapter 5 Change of Variables Formula

in a neighborhood of the point a ∈ U , and from Theorem 3.2.7, that

∂ϕ
the standard matrix of Dϕ (a) is the n × n matrix Dϕ (a) = ( ∂xji (a)),
where ϕj are the component functions of ϕ and xi the components of
x for i, j = 1, ..., n. Using the Linear Approximation theorem, we will
show that for a sufficiently small cube C centered at a ∈ U

ν(ϕ(C)) ≈ | det Dϕ (a)|ν(C).

Lemma 5.1.12. Let U ⊂ Rn be open, and let ϕ : U → ϕ(U ) be a

diffeomorphism in Rn and a ∈ U be fixed. Let C ⊂ U be a cube centered
at a. Then
ν(ϕ(C))
lim = | det Dϕ (a)|,
C↓a ν(C)

where C ↓ a means that C shrinks to its center a ∈ C.

Here the C can be either a cube or a ball in U centered at a. It is

sufficient to consider only balls Br (a), for replacing the norm || · || by
the norm || · ||∞ the same type of argument works also for cubes. Thus
what we are trying to prove is

ν(ϕ(Br (a)))
lim = | det Dϕ (a)|.
r→0 ν(Br (a))

Proof. We begin with the Linear Approximation theorem:

ϕ(x) = ϕ(a) + Dϕ (a)(x − a) + (x)||x − a||,

where (x) tends to 0 as x → a. Since ϕ is a C 1 change of variable,

det Dϕ (a) 6= 0 and so Dϕ (a)−1 exists. Applying this linear transforma-
tion to the equation above yields,

Dϕ (a)−1 ϕ(x) = Dϕ (a)−1 ϕ(a) + (x − a) + ||x − a||Dϕ (a)−1 (x),

Hence by the triangle inequality2

||Dϕ (a)−1 φ(x) − Dφ (a)−1 ϕ(a)|| ≤ ||x − a|| + ||x − a|| ||Dϕ (a)−1 (x)||.
2
For the case of the cube, here use the triangle inequality for the box-norm || · ||∞
√
and the estimate ||x − a|| ≤ n||x − a||∞ .
5.1 Change of variables formula 303

Finally, estimating the effect of the linear transformation Dϕ (a)−1 there

is a positive constant β so that

||Dϕ (a)−1 (x)|| ≤ β||(x)||.

Thus,

||Dϕ (a)−1 ϕ(x)−Dϕ (a)−1 ϕ(a)|| ≤ ||x−a||+β||(x)|| ||x−a|| < (1+β||(x)||)r.

Since ||x − a|| < r, ||(x)|| → 0 as r → 0. Now let 0 < t < 1 and choose
r > 0 small enough so that 1 + β||(x)|| lies strictly between 1 − t and
1 + t. This means Dϕ (a)−1 (ϕ(Br (a))) is contained in a ball centered at
Dϕ (a)−1 ϕ(a) of radius (1 + t)r. If for some x, Dϕ (a)−1 ϕ(x) lies on the
boundary of this ball, then

||Dϕ (a)−1 ϕ(x) − Dϕ (a)−1 ϕ(a)|| = (1 + t)r > (1 − t)r,

so that Dϕ (a)−1 ϕ(x) lies outside this smaller ball of radius (1 −

t)r. Since, Dϕ (a)−1 is a homeomorphism in a neighborhood of a,
Dϕ (a)−1 (ϕ(Br (a))) also contains this smaller ball. In particular, by
Corollary 5.1.8 we have

cn (1 − t)n r n ≤ ν(Dϕ (a)−1 (ϕ(Br (a))) ≤ cn (1 + t)n r n .

That is,
ν(Dϕ (a)−1 (ϕ(Br (a)))
(1 − t)n ≤ ≤ (1 + t)n .
ν(Br (a))
But since Dφ (a)−1 is linear, Proposition 5.1.7 tells us

ν(Dϕ (a)−1 (ϕ(Br (a))) = | det Dϕ (a)−1 |ν(ϕ(Br (a))).

Thus,
ν(ϕ(Br (a)))
(1 − t)n ≤ ≤ (1 + t)n . (5.7)
| det Dϕ (a)|ν(Br (a))
Letting t → 0 (so that r → 0) yields the conclusion.

Lemma 5.1.13. Let U ⊂ Rn be open and let Ω be a bounded simple SN set

with Ω ⊂ U . Then there exist a compact set S of the form S = i=1 Ci ,
where {C1 , ..., CN } are essential disjoint closed cubes, such that Ω ⊂
S ⊂ U.
304 Chapter 5 Change of Variables Formula

Proof. Since Ω is compact and U is open we may cover Ω by open

cubes lying completely inside U . By the compactness of Ω a finite
number of these cubes Crj (xj ) : xj ∈ Ω, j = 1, ..., p of sides 2rj re-
spectively,
 also cover Ω. Let δ = min {r1 , ..., rp }. Then the open cubes
Cδ (x) : x ∈ Ω cover Ω and by compactness so does a finite number of

them Cδ (xk ) : xk ∈ Ω, k = 1, ..., q and each closed cube Cδ (xk ) ⊂ U .
In general these closed cubes overlap. Partitioning each of these to
smaller closed subcubes and counting the overlaping subcubes once,
we obtained the required finite collection of essential disjoint closed
(sub)cubes {Ci : i = 1, ..., N } whose union S is contained in U .

Next we prove the principal result3 of this section.

Theorem 5.1.14. (Change of variables formula). Let U be an open set

in Rn and let ϕ : U → Rn be a diffeomorphism. Let Ω be a bounded
simple set with Ω ⊂ U . If f : ϕ(Ω) → R is integrable on ϕ(Ω), then
(f ◦ ϕ)| det Dϕ | is integrable over Ω and
Z Z
f (y)dy = f (ϕ(x))| det Dϕ (x)|dx. (5.8)
ϕ(Ω) Ω

Proof. Set Jϕ = det Dϕ . We show that (f ◦ ϕ)|Jϕ | is integrable over

Ω. We shall use Theorem 4.1.42 which characterizes integrable func-
tions in terms of their points of discontinuity. Let D be the set of
discontinuities of f in ϕ(Ω). Then E = ϕ−1 (D) is the set of discon-
tinuities of f ◦ ϕ in Ω. Since f is integrable on ϕ(Ω) the set D has
measure zero. As D ⊂ ϕ(Ω) ⊂ ϕ(Ω) ⊂ ϕ(U ) and ϕ(Ω) is compact
contained in the open set ϕ(U ), we can find a finite number of closed
cubes {Ci : i = 1, ..., N1 } such that D ⊂ ∪N
j=1 Ci ⊂ ϕ(U ) (as in Lemma
1

N1
5.1.13). Set K = ∪i=1 Ci , then K is compact and D ⊂ K ⊂ ϕ(U ). As
3
The Change of Variables formula was first proposed by Euler when he studied
double integrals in 1769, and it was generalized to triple integrals by Lagrange in
1773. Although it was used by Legendre, Laplace, Gauss, and first generalized to
n variables by Mikhail Ostrogradski in 1836, it resisted a fully rigorous proof for
a surprisingly long time. The theorem was first completely proved 125 years later,
by Elie Cartan in a series of papers beginning in the mid-1890s. A popular proof
adapted by many authors is the one given by J. Schwartz (1954) in [29]. In the proof
given here effort has been made in avoiding as many technicalities as possible. A
quite different approach to the problem can be found in P. Lax [20] and [21].
5.1 Change of variables formula 305
S∞
in the proof of Theorem 4.1.30 we may write D = k=1 D k1 with each
D 1 compact. Since ϕ−1 ∈ C 1 (ϕ(U )) and each D 1 has measure zero,
k k
Lemma 5.1.5 and Proposition 4.1.21 tell us that E = ϕ−1 (D) has also
measure zero. Hence f ◦ ϕ is integrable. Since |Jϕ | is continuous on
Ω, the set of points of discontinuity of (f ◦ ϕ)|Jϕ | is the set E and so
(f ◦ ϕ)|Jϕ | is integrable on Ω.

We first prove (5.8) for a closed cube C ⊂ U . Let P = {Cj : j = 1, ..., m}

be a partition of C into subcubes of equal side length 2r centered at
Sm j. In addition {ϕ(Cj )}
xj . Then ϕ(Cj ) is a bounded simple set for each
are mutually essentially disjoint and ϕ(C) = j=1 ϕ(Cj ). From (5.7) of
Lemma 5.1.12 we have

(1 − t)n |Jϕ (xj )|ν(Cj ) ≤ ν(ϕ(Cj )) ≤ |Jϕ (xj )|ν(Cj )(1 + t)n . (5.9)

Let yj = ϕ(xj ). We set F = f ◦ ϕ and suppose f ≥ 0. From (5.9) we

have
m
X m
X
(1 − t)n F (xj )|Jϕ (xj )|ν(Cj ) ≤ f (yj )ν(ϕ(Cj )) ≤
j=1 j=1

m
X
≤ (1 + t)n F (xj )|Jϕ (xj )|ν(Cj ).
j=1

Letting t → 0 yields
m
X m
X m
X
F (xj )|Jϕ (xj )|ν(Cj ) ≤ f (yj )ν(ϕ(Cj )) ≤ F (xj )|Jϕ (xj )|ν(Cj ).
j=1 j=1 j=1
(5.10)
The far left and far right are each a Riemann sum SP (F |Jϕ |) for the
function F |Jϕ |. Let d = maxj=1,...,m {d(ϕ(Cj ))}. Since ϕ is (uniformly)
continuous on C, if r → 0 then also d → 0. Now since the functions f
and F |Jϕ | are integrable on ϕ(C) and C respectively, letting r → 0 in
(5.10) we get
Z Z
f (y)dy = f (ϕ(x))|Jϕ (x)|dx. (5.11)
ϕ(C) C
306 Chapter 5 Change of Variables Formula

To remove the assumption that f ≥ 0, we write f = (f + c) − c where

the constant c ≥ 0 is sufficiently large that f + c ≥ 0 on Ω, for example
c = sup |f (x)|. The argument just given applies to f + c and to the
constant function c. By the linearity of the integral (Theorem 4.2.1)
subtracting the results we get (5.11).

Next we prove (5.8) for a finite union S = N

S
i=1 Ci of essential dis-
joint closed cubes with Ω ⊂ S ⊂ U . That such a set S exists is guar-
anteed from Lemma 5.1.13. Now the finite additivity of the integral
(Corollary 4.2.5) yields
Z N Z
X N Z
X
f (y)dy = f (y)dy = (f (ϕ(x))|Jϕ (x)|dx
ϕ(S) i=1 ϕ(Ci ) i=1 Ci
Z
= (f (ϕ(x))|Jϕ (x)|dx.
S

Finally, we obtain (5.8) for Ω. Let y = ϕ(x) with x ∈ S, where

Ω ⊂ S ⊂ U . Since χΩ = χϕ(Ω) ◦ ϕ, where χΩ is the characteristic
function of Ω, the definition of the integral over a set gives
Z Z Z
f (y)dy = (f χϕ(Ω) )(y)dy = [(f χϕ(Ω) ) ◦ ϕ|Jϕ |](x)dx
ϕ(Ω) ϕ(S) S
Z Z Z
= [(f ◦ϕ)|Jϕ |χΩ ](x)dx = ((f ◦ϕ)|Jϕ |)(x)dx = f (ϕ(x))|Jϕ (x)|dx.
S Ω Ω

Taking f ≡ 1 which is integrable everywhere we get

Corollary 5.1.15. Let U be open in Rn and ϕ : U → Rn a diffeomor-

phism. Let Ω be a bounded simple set with Ω ⊂ U . Then
Z
ν(ϕ(Ω)) = | det Dϕ (x)|dx.
Ω

In certain circumstances a direct application of the Change of Vari-

ables formula (5.8)) is not valid, mainly because ϕ fails to be one-to-one
on the boundary of Ω. The following slight improvement of Theorem
5.1.14 is then useful.
5.1 Change of variables formula 307

Corollary 5.1.16. Let U be open in Rn and ϕ : U → Rn be a C 1 map.

Let Ω be a bounded simple set with Ω ⊂ U . Suppose f : ϕ(Ω) → R is
integrable. Then the change of variables formula (5.8) remains valid if
ϕ is only asssumed to be a diffeomorphism on the interior of Ω.

Proof. Let C be a cube containing Ω and partition C into subcubes of

equal side length 2r. Let S
C1 , ..., Cp be the subcubes contained in Ω, and
A = Ω\K, where K = pi=1 Ci . Then K ⊂ Ω and so ν(K) ≤ ν(Ω).
Since Ω is simple, the difference ν(Ω) − ν(K) can be made arbitrarily
small by taking r sufficiently small. That is, ν(A) = 0 ans so also
ν(A) = 0. Therefore by Lemma 5.1.5 (with S = A) it follows that
ν(ϕ(A)) ≤ ν(ϕ(A)) = 0. Theorem 5.1.14 and the finite additvity of the
integral yield
Z Xp Z
f (y)dy = f (y)dy
ϕ(K) i=1 ϕ(Ci )

p Z
X Z
= (f (ϕ(x))|Jϕ (x)|dx = (f (ϕ(x))|Jϕ (x)|dx.
i=1 Ci K

Finally, as Ω = K ∪ A, ϕ(Ω) = ϕ(K) ∪ ϕ(A) and the integral

over
R sets of volume
R zero is zero, the additivity of the integral gives
ϕ(Ω) f (y)dy = Ω (f (ϕ(x))|Jϕ (x)|dx.

More generally a further extension of the Change of Variables for-

mula can be made allowing singularities of ϕ by using a special case of
Sard’s theorem: Let ϕ : U → Rn be of class C 1 on an open set U ⊂ Rn
and
A = {x ∈ U : det Dϕ (x) = 0} .
Then ϕ(A) has measure zero, (see Theorem 4.5.1). Hence the places
where | det Dϕ (x)| ≡ 0 on Ω contribute nothing to the integrals in (5.8).

Next we give an extension of the Change of Variables formula4 for

continuous real valued functions f on Rn with compact support (see,
Section 4.4).
4
Although the integral of f is taken over Rn and such integrals are studied in the
next section, since the integrand f has compact support it is an integral over any
cube containing the support of f .
308 Chapter 5 Change of Variables Formula

Theorem 5.1.17. Let U be an open set in Rn and ϕ be a diffeomor-

phism of U with its image ϕ(U ). Then
Z Z
f (y)dy = f (ϕ(x))| det Dϕ (x)|dx, (5.12)
Rn Rn

for any continuous function f with compact support whose support lies
in ϕ(U ).

Proof. Let y ∈ supp(f ) and Vy be a neighborhood of y contained in the

open set ϕ(U ). Since the {Vy } cover supp(f ) we can find a finite number
of them {Vyi : i = 1, ..., k} which also cover. Let ψi be a partition of
unity subordinate to this cover. By Corollary 4.4.4 f = ki=1 ψi f , where
P
ψi f is a continuous function whose support lies in Vyi . By the Change
of Variables formula (5.8) equation (5.12) holds for each ψi f , i = 1, ..., k
and hence by linearity of the integral also for their sum.

We now apply the Change of Variables formula to a number of

examples.

5.1.3 Applications, polar and spherical coordinates

Let Ω ⊂ Rn be a bounded simple set and f : Ω → R an integrable
function. One of the purposes of the Change of Variables
R formula is to
simplify the computation of the multiple integral Ω f on which either
the integrand f or the region Ω is complicated and for which direct
computation is difficult. Therefore, a C 1 change of variables ϕ : Ω∗ → Ω
is chosen, with Ω = ϕ(Ω∗ ), so that the integral is easier to compute with
the new integrand (f ◦ ϕ)|Jϕ |, or with the new region Ω∗ = ϕ−1 (Ω). In
this notation the Change of Variables formula becomes
Z Z Z
f= f= (f ◦ ϕ)|Jϕ |. (5.13)
Ω ϕ(Ω∗ ) Ω∗

The Change of Variables formula (5.13) for double integrals is

Z Z Z Z
f (x, y)dxdy = f (x(u, v), y(u, v))|Jϕ (u, v)|dudv,
Ω Ω∗
5.1 Change of variables formula 309

where ϕ(u, v) = (x(u, v), y(u, v)) for (u, v) ∈ Ω∗ = ϕ−1 (Ω) and
    ∂x ∂x 
 ∂(x, y)   
|Jϕ (u, v)| = | det Dϕ (u, v)| = 
  = det ∂u
 ∂v 
∂y ∂y  .
∂(u, v) 
∂u ∂u

The Change of Variables formula (5.13) for triple integrals is

Z Z Z
f (x, y, z)dxdydz
Ω
Z Z Z
= f (x(u, v, w), y(u, v, w), z(u, v, w))|Jϕ (u, v, w)|dudvdw,
Ω∗
where ϕ(u, v, w) = (x(u, v, w), y(u, v, w), z(u, v, w)) for (u, v, w) ∈ Ω∗ =
ϕ−1 (Ω) and
  ∂x ∂x ∂x 
   
 ∂(x, y, z)   ∂u ∂v ∂w
∂y ∂y ∂y

|Jϕ (u, v, w)| = | det Dϕ (u, v, w)| = 
  = det  ∂u ∂u ∂w  .

∂(u, v, w) 
 ∂z ∂z ∂z 
∂u ∂v ∂w

Example 5.1.18. Let f (x, y) = xy and let Ω be the parallelogram

bounded by the lines x − y = 0, x − y = 1, x + 2y
R R= 0 and x + 2y = 6.
Using the Change of Variables formula compute Ω f (x, y)dxdy.

Solution. The equations of the bounding lines of Ω suggest the lin-

ear change of variables u = x − y, v = x + 2y. That is, the linear
mapping L : R2 →  R2 given
 by L(x, y) = (x − y, x + 2y). The standard
1 −1
matrix of L is L = , and det L = 3. Hence L is an isomorphism
1 2
on R2 .
The vertices of the parallelogram Ω are (0, 0), (2, 2), ( 38 , 53 ), and
( 23 , − 13 ) and these are mapped by L to (0, 0), (0, 6), (1, 6), and (1, 0),
respectively. Since L is linear, it follows that the image of the parallel-
ogram Ω under L is the rectangle

Ω∗ = {(u, v) : 0 ≤ u ≤ 1, 0 ≤ v ≤ 6} .

In fact, we are actually interested in the inverse mapping T = L−1 from

the uv-plane to the xy-plane defined by
x = 13 (2u + v), y = 31 (v − u).
310 Chapter 5 Change of Variables Formula

That is, the linear transformation T (u, v) = 13 (2u + v, v − u) with Jaco-

bian JT (u, v) = det T = 31 and T (Ω∗ ) = Ω.
Hence the Change of Variables formula yields

1 1
Z Z Z Z
xydxdy = (2u + v)(v − u)| |dudv
Ω 9 Ω∗ 3
1Z 6 6
1 2 1 77
Z Z
= (2u + v)(v − u)dudv = (− + v + v 2 )dv = .
27 0 0 0 3 2 27
Example 5.1.19. Compute the integral
Z Z
y−x
e y+x dxdy,
Ω

where Ω = {(x, y) : x ≥ 0, y ≥ 0, x + y ≤ 1}.

Solution. Here the difficulty in the computation of the integral comes

y−x
from the integrand. Note that the function f (x, y) = e y+x is not de-
fined at (0, 0) and, in fact, does not even have a limit there. However, no
matter how one defines f at (0, 0) the function is integrable by Theorem
4.1.42. The set Ω is a simple region in R2 . Actually, Ω is the triangle
bounded by the line x + y = 1 and the coordinate axes. The integrand
suggests the use of the new variables

u = y − x, v = y + x,

i.e., the linear transformation L(x, y) = (y − x, y + x). L is an isomor-

phism on R2 and its inverse T = L−1 is the linear transformation
v−u v+u
, ).
T (u, v) = (
2 2
 1 1

−2 2
The standard matrix representation is T = 1 1 , and its Jacobian
2 2
JT (u, v) = det T = − 12 . To find the image Ω∗ of Ω under L in the uv-
plane, note that the lines x = 0, y = 0 and x + y = 1 are mapped onto
the lines u = v, u = −v and v = 2, respectively. The points inside Ω
satisfy 0 < x + y < 1 and these are mapped into points of Ω∗ satisfying
0 < v < 1. Hence Ω∗ is the triangle in the uv-plane bounded by the
5.1 Change of variables formula 311

4
4

3
3

2
Y 2
W*
1 W
1

0
0 1 2 3 4 0
0 1 2 3 4

Figure 5.2: ψ(x, y) = (xy, x2 − y 2 )

lines v = u, v = −u and v = 2. Now T (Ω∗ ) = Ω and the Change of

Variables formula yields
1
Z Z Z Z
y−x u
e y+x dxdy = e v dudv
Ω 2 Ω ∗

Z 1 Z v  Z 1
1 u 1 1 1 1
= e v du dv = (e − )vdv = (e − e−1 ) = sinh(1).
2 0 −v 2 0 e 4 2
Example 5.1.20. Let f (x, y) = x2 + y 2 . Evaluate
R R
Ω f (x, y)dxdy,
where Ω is the region in the first quadrant of the xy-plane bounded by
the curves xy = 1, xy = 3, x2 − y 2 = 1, and x2 − y 2 = 4.

Solution. To simplify the region Ω we make the substitution u = xy,

v = x2 − y 2 . This transformation ψ : R2 → R2

ψ(x, y) = (xy, x2 − y 2 )

is one-to-one on Ω and maps the region Ω in the xy-plane onto the

rectangle Ω∗ in the uv-plane bounded by the lines u = 1, u = 3, v = 1,
and v = 4, ie ψ(Ω) = Ω∗ = [1, 3] × [1, 4]. Its Jacobian is
 
∂(u, v) y x
Jψ (x, y) = = det = −2(x2 + y 2 ) 6= 0
∂(x, y) 2x −2y

for all (x, y) ∈ Ω. As before, we are interested in its inverse ϕ = ψ −1

given by
"
1
#1 " 1
#1 
2 2
(4u + v ) 2 + v 2 2 2
(4u + v ) 2 − v 2 
ϕ(u, v) =  , .
2 2
312 Chapter 5 Change of Variables Formula

The Jacobian of ϕ is

∂(x, y) 1 1 1
Jϕ (u, v) = = =− =− 1 .
∂(u, v) Jψ (x, y) 2(x2 + y 2 ) 2(4u + v 2 ) 2
2

Now Ω = ϕ(Ω∗ ) and the Change of Variables formula gives

Z Z Z Z
f (x, y)dxdy = f (ϕ(u, v))|Jϕ (u, v)|dudv
Ω Ω∗

3Z 4 3Z 4
1 1
Z Z
1
2 2
= (4u + v ) · 2
1 dudv = dudv = 3.
1 1 2(4u2 + v 2 ) 2 2 1 1

Example 5.1.21. (The solid torus). Let 0 < a < b be fixed.

The solid torus in R3 is the region Ω obtained by revolving the disk
(y − b)2 + z 2 ≤ a2 in the yz-plane, about the z-axis. Calculate the
volume of Ω. (See, Figure 4.5)

Solution. Note that the mapping ϕ : R3 → R3 defined by the

equations

x = (b + r cos φ) cos θ
y = (b + r cos φ) sin θ
z = r sin φ,

maps the rectangle R = {(θ, φ, r) : θ, φ ∈ [0, 2π], r ∈ [0, a]} onto the
torus, ie ϕ(R) = Ω. In particular, ϕ is one-to-one in the interior of R.
Its Jacobian is

∂(x, y, z)
Jϕ (θ, φ, r) = = r(b + r cos φ),
∂(θ, φ, r)

which is positive in the interior of R. By Corollary 5.1.16, the Change

of Variables formula applies, and Corollary 5.1.15 gives
Z Z Z
ν(Ω) = ν(ϕ(R)) = r(b + r cos φ)dθdφdr = 2π 2 a2 b.
R
5.1 Change of variables formula 313

Polar coordinates

The map which changes from polar coordinates to rectangular coordi-

nates is
ϕ(r, θ) = (r cos θ, r sin θ) = (x, y).

Its Jacobian is
 
∂(x, y) cos θ −r sin θ
Jφ (r, θ) = = det = r cos2 θ + r sin2 θ = r,
∂(r, θ) sin θ r cos θ

which is zero for r = 0. ϕ maps the set [0, +∞) × [0, 2π) in the rθ-
plane onto R2 . Note that on this set ϕ is not one-to-one, since for
0 ≤ θ < 2π it sends all points (0, θ) into (0, 0). However ϕ restricted to
S = (0, +∞) × (0, 2π) (the interior of [0, +∞) × [0, 2π)) is one-to-one
and Jφ (r, θ) 6= 0 on S. Although ϕ(S) excludes the non-negative x-axis,
this set has 2-dimensional volume zero and therefore contributes nothing
to the value of an integral. Note in particular,that ϕ maps the rect-
angle Rα = [0, α] × [0, 2π] onto the disk Bα = (x, y) : x2 + y 2 ≤ α2 .

Thus by restricting ϕ in the interior of Rα we can apply the Change

of Variables theorem (Corollary 5.1.16) to convert integration over Bα
into integration over Rα .

Example 5.1.22. Compute

Z Z
2 +y 2 )
e−(x dxdy,
Bα

where Bα = (x, y) : x2 + y 2 ≤ α2 .


Solution. Changing to polar coordinates and using (5.13) with Ω = Bα

and Ω∗ = [0, α] × [0, 2π] we get
Z Z Z α Z 2π
2 +y 2 ) 2
e−(x dxdy = e−r rdrdθ
Bα 0 0
Z α Z 2π  Z 0
2 2
= e−r r dθ dr = π eu du = π[1 − e−α ].
0 0 −α2
314 Chapter 5 Change of Variables Formula

In particular, letting α → ∞ we find5

Z Z
2 2
e−(x +y ) dxdy = π.
R2

Example 5.1.23. Compute

Z Z p
sin x2 + y 2 dxdy,
Ω

where Ω = (x, y) : π 2 < x2 + y 2 ≤ 4π 2 .



Solution. Changing to polar coordinates we see that Ω = ϕ(Ω∗ ), where

Ω∗ = (π, 2π] × [0, 2π] and we get
Z Z p Z 2π Z 2π
sin x2 + y 2 dxdy = sin r · rdrdθ
Ω π 0
"
2π Z #
Z 2π  2π
= 2π r sin rdr = 2π −r cos r  + cos rdr
π π π
2π
= −6π + 2π sin r  = −6π 2 .
2

π

Example 5.1.24. Compute

Z Z s 2 2
a b − b2 x2 − a2 y 2
dxdy,
Ω a2 b2 + b2 x2 + a2 y 2
n o
x2 y2
where Ω = (x, y) : a2
+ b2
≤ 1, x > 0, y > 0 .

Solution. Changing to polar coordinates using the transformation

ϕ(r, θ) = (ar cos θ, br sin θ) = (x, y),

we see that Ω∗ = (0, 1] × [0, π2 ) and Jϕ (r, θ) = abr. Now the integral
becomes
2 2
5
e−(x +y ) dxdy = π is improper. See Section 5.2, for a detailed
RR
The integral R2
discussion on improper multiple integrals.
5.1 Change of variables formula 315
π r Z r
1Z
1 − r2 π 1 1 − r2
Z
2
ab rdrdθ = ab rdr
0 0 1 + r2 2 0 1 + r2
Z √2 p
π
= ab 2 − ρ2 dρ,
2 0
√
where ρ2 = 1 + r 2 . Finally, setting ρ = 2 sin t, this latter integral
yields π2 ( π4 − 21 )ab.

Example 5.1.25. Find the area of the region S in R2 bounded by the

circles r = 1 and r = √23 cos θ (outside of the circle r = 1).

Solution. Solving 1 = √2 cos θ we find the (polar) coordinates of the

3
points of intersections of the given circles to be (1, π6 ) and (1, 11π
6 ) (the
reader is recommended to make a drawing; in cartesian coordinates the
circle r = √23 cos θ is (x − √13 )2 + y 2 = 13 and the circle r = 1 is, of
course, x2 + y 2 = 1). By the symmetry of the region the area is
Z Z Z π Z 2 !
√
6 3
A(S) = rdrdθ = 2 cos θrdr dθ
S 0 1

π √
4 3 3−π
Z
6
2
= ( cos θ − 1)dθ = .
0 3 18

Cylindrical coordinates

The cylidrical coordinates are just the polar coordinates in the xy-plane
with the z-coordinate added on, (Figure 5.3),

ϕ(r, θ, z) = (r cos θ, r sin θ, z).

To get a one-to-one mapping we must keep r > 0 and restrict θ, say, in

0 < θ ≤ 2π. The Jacobian of ϕ is easily seen to be Jϕ (r, θ, z) = r, and
the Change of Variables formula (5.13) becomes
Z Z Z Z Z Z
f (x, y, z)dxdydz = f (r cos θ, r sin θ, z)rdrdθdz.
Ω Ω∗
316 Chapter 5 Change of Variables Formula
z

Hr,Θ,zL

Θ y
r

Figure 5.3: Cylindrical coordinates

Example 5.1.26. Find the volume of the region Ω bounded by the

2 2
surfaces ( x a+y
2 )2 + zb = 1, z = 0, where a, b > 0.

Solution. As we know from Definition 4.1.39

Z Z Z Z
ν(Ω) = 1= dxdydz.
Ω Ω
Changing to cylidrical coordinates we get
 4

Z Z Z Z a Z 2π Z b(1− r 4 )
a
ν(Ω) = rdrdθdz =  dz  drdθ
Ω∗ 0 0 0

a
r4
 
2
Z
= 2πb r 1 − 4 dr = πa2 b.
0 a 3
Example 5.1.27. Find the moment of inertia IL of a cylinder x2 +y 2 =
a2 of height h, if its density at each point is propotional to the distance
of this point from the axis of the cylinder, with respect to a line L par-
allel to the axis of the cylinder and at distance b from it.

Solution. Let Ω be the cylinder. The moment of inertia of Ω about

the line L is
Z Z Z
IL = [d(x, y, z)]2 ρ(x, y, z)dxdydz,
Ω
5.1 Change of variables formula 317

where d(x, y, z) is the distance from (x, y, z) to the line L, and ρ(x, y, z)
is the density. Placing the base of the cylinder on the xy-plane with its
center at the origin, the line
pL is described by x = b, y = 0. The density
function is ρ(x, y, z) = k x2 + y 2 and [d(x, y, z)]2 = (x − b)2 + y 2 .
Passing to cylindrical coordinates we get
Z h Z 2π Z a
IL = [(r cos θ − b)2 + (r sin θ)2 ](kr)rdrdθdz
0 0 0
Z h Z 2π Z a
= kr 2 [r 2 + b2 − 2r cos θ]drdθdz
0 0 0
Z h Z 2π Z a
=k r 2 [r 2 + b2 ]drdθdz + 0
0 0 0
2 b2
 
3 a
= 2πka h + .
5 3

Spherical coordinates

Given (x, y, z) ∈ R3 , the spherical coordinates (r, θ, φ) are defined by

Φ(r, θ, φ) = (r cos θ sin φ, r sin θ sin φ, r cos φ) = (x, y, z).

p
Here r = x2 + y 2 + z 2 , θ is the longitude (the angle from the x-axis to
the vector (x, y, 0)), and φ is the co-latitude (the angle from the positive
z-axis to the vector (x, y, z)). (see, Figure 5.4)
It is readily seen that Φ maps therectangle R = [0, α]×[0, 2π]×[0, π]
in the rθφ-space into the ball Bα = (x, y, z) : x2 + y 2 + z 2 ≤ α2 .
Again Φ is not globally invertible; we have Φ(r, θ, φ) = Φ(r, θ +
2kπ, φ + 2mπ) = Φ(r, θ + (2k + 1)π, φ + (2m + 1)π) for each k, m ∈
Z. The map Φ is one-to-one on {(r, θ, φ) : r > 0, 0 ≤ θ < 2π, 0 < φ < π}
and maps this set onto R3 − {(0, 0, z) : z ∈ R}. Since {(0, 0, z) : z ∈ R},
has volume zero does not affect the value of an integral. Now
∂(x, y, z)
JΦ (r, θ, φ) =
∂(r, θ, φ)
 
cos θ sin φ −r sin θ sin φ r cos θ cos φ
= det  sin θ sin φ r cos θ sin φ r sin θ cos φ  = −r 2 sin φ.
cos φ 0 −r sin φ
318 Chapter 5 Change of Variables Formula

Figure 5.4: Spherical coordinates (r, θ, φ)

Hence the formula for integration in spherical coordinates is

Z Z Z
f (x, y, z)dxdydz
Ω
Z Z Z
= f (r cos θ sin φ, r sin θ sin φ, r cos φ)r 2 sin φdrdθdφ.
Ω∗

3
2 +y 2 +z 2 ) 2
e(x
RRR
Example 5.1.28. Compute the integral Ω dxdydz,
where Ω is the unit ball in R3 .

Solution. Changing to spherical coordinates Ω = Φ(Ω∗ ), with Ω∗ =

[0, 1] × [0, 2π] × [0, π]. Now the formula of integration in spherical coor-
dinates gives
Z Z Z 3
Z Z Z
2 2 2 3
e(x +y +z ) 2 dxdydz = er r 2 sin φdrdθdφ
Ω Ω∗
Z 1Z π  Z 2π  Z 1 Z π 
2 r3 2 r3
= r e sin φ dθ dφ = 2π r e sin φdφ dr
0 0 0 0 0
5.1 Change of variables formula 319

1 1
4 4π(e − 1)
Z Z
2 r3
= 4π r e dr = π et dt = ,
0 3 0 3

where we set t = r 3 .
 
√ 1 1
RRR
Example 5.1.29. Compute dxdydz, where Ω is Ω + z
x2 +y 2
p
the region in R3 bounded below by the cone z = x2 + y 2 and above
by the sphere x2 + y 2 + z 2 = 1, and z > 0.

Solution. The region of integration Ω represents an “ice cream cone”.

Note that Ω is an elementary region6 in R3 . However, both the region
Ω and the integrand suggest a change to spherical coordinates. Then
Ω = Φ(Ω∗ ), where
n πo
Ω∗ = (r, θ, φ) : 0 < r < 1, 0 ≤ θ < 2π, 0 < φ < .
4

Now the Change of Variables formula yields

!
1 1
Z Z Z
p + dxdydz
Ω x2 + y 2 z

 
1 1
Z Z Z
= + r 2 sin φdrdθdφ
Ω∗ r sin φ r cos φ
π
Z
4
 Z 1 Z 2π  
= (1 + tan φ) r dθ dr dφ
0 0 0

π
Z
4
 Z 1 
= 2π (1 + tan φ) rdr dφ
0 0

π   √ 
π 2
Z π
4
=π (1 + tan φ)dφ = π[(φ − log(cos φ))|0 ] = π − log 4
.
0 4 2
q q
6
Ω = {(x, y, z) : −√12 ≤ x ≤ 1 1 1
p
√
2
,− 2
−x2 ≤ y ≤ 2
−x2 , x2 +y 2 ≤ z ≤
p
1−x2 −y 2 }. Note that working with rectangular coordinates the integral becomes
quite tedious.
320 Chapter 5 Change of Variables Formula

Example 5.1.30. Find the volume of the region Ω in R3 above the

cone z 2 = x2 + y 2 and inside the sphere x2 + y 2 + z 2 = z.

Solution. Here Ω = Φ(Ω∗ ), where

n πo
Ω∗ = (r, θ, φ) : 0 ≤ r < cos φ, 0 ≤ θ < 2π, 0 < φ < .
4
Hence the volume ν(Ω) is equal to
Z Z Z Z π Z cos φ Z 2π  
4
2
1dxdydz = dθ r sin φdr dφ
Ω 0 0 0
π π
 cos φ 
2π
Z Z Z
4 4
2
= 2π sin φ r dr dφ = cos3 φ sin φdφ
0 0 3 0
1
2π 2π 3 π
Z
= √ t3 dt = · = ,
3 2 3 16 8
2

where we set t = cos φ.

Example 5.1.31. Find the centroid of a ball of radius a. Find the mo-
ment of inertia of a ball of constant density with respect to a diameter.

Solution. Place the ball B with its center at the origin, so that B
is described by x2 + y 2 + z 2 ≤ a2 . From Example 4.3.19 the volume of
the ball is ν(B) = 43 πa3 . Using spherical coordinates
Z a Z 2πZ π
1 1
Z Z Z
x= xdxdydz = (r cos φ)r 2 sin φdφdθdr = 0.
ν(B) B ν(B) 0 0 0
By symmetry x = y = z. Hence its centroid is its center (0, 0, 0).

Let the density of the ball be ρ(x, y, z) = k. By symmetry Ix = Iy =

Iz . Since Ix + IyR +
R RIz =22I0 , 2where I0 is the moment of inertia about
the origin, I0 = (x + y + z 2 )kdxdydz, we have
B

2 2 a π 2π 4 8πka5
Z Z Z
Ix = Iy = Iz = I0 = kr sin φdθdφdr = .
3 3 0 0 0 15
4πka3
Since the mass of the ball is m = 3 , the answer can be expressed
2 2
5 a m.
5.1 Change of variables formula 321

Exercise 5.1.32. Show that the centroid of the portion of the ball of
radius a in the first octant is the point 3a
2 (1, 1, 1).

Spherical coordinates in Rn

This is the higher dimensional analogue of the spherical p coordinates in

R3 . Given (x1 , ..., xn ) ∈ Rn we let r = ||x|| = x21 + ... + x2n , and
Θ = (θ, φ1 , ..., φn−2 ), a generic point on the unit sphere, S n−1 .
For n ≥ 2 the n-dimensional spherical mapping Φn : Rn → Rn written
as
Φn (r, Θ) = Φn (r, θ, φ1 , ..., φn−2 ) = (x1 , ..., xn )
is defined by

x1 = r cos φ1 ,

x2 = r sin φ1 cos φ2 ,

x3 = r sin φ1 sin φ2 cos φ3 ,

..........................................

xn−1 = r sin φ1 · · · sin φn−2 cos θ,

xn = r sin φ1 · · · sin φn−2 sin θ.

Φn maps the rectangle

Rα = [0, α] × [0, 2π] × [0, π] × · · · × [0, π]

= {(r, θ, φ1 , ..., φn−2 ) ∈ Rn : r ∈ [0, α], θ ∈ [0, 2π], φi ∈ [0, π], i = 1, ..., n−2}
onto the ball Bα = {x ∈ Rn : ||x|| ≤ α}.
Thus for n = 2

Φ2 (r, Θ) = Φ2 (r, θ) = (r cos θ, r sin θ),

which are just the polar coordinates in R2 , while for n = 3

Φ3 (r, Θ) = Φ3 (r, θ, φ1 ) = (r cos φ1 , r sin φ1 cos θ, r sin φ1 sin θ),

322 Chapter 5 Change of Variables Formula

which are the spherical coordinates in R3 (viewing the x1 -axis as the

z-axis and φ1 as φ).
The Jacobian of Φn can be calculated by induction and is

∂(x1 , x2 , x3 , ..., xn )
JΦn (r, Θ) = JΦn (r, θ, φ1 , ..., φn−2 ) =
∂(r, θ, φ1 , ..., φn−2 )

n−2
Y
= r n−1 sinn−2 φ1 sinn−3 φ2 · · · sin2 φn−3 sin φn−2 = r n−1 sinn−k−1 φk .
k=1

Clearly, Φn is not one-to-one on Rα , however Φn restricted to the

interior of Rα it is one-to-one with JΦn (r, Θ) > 0 and the Change of
Variables formula applies.

Example 5.1.33. (The volume of the unit ball in Rn ). Find the volume
of the n-dimensional unit ball,

B n = {x ∈ Rn : ||x|| ≤ 1} .

Solution. In Corollary 5.1.8 we denoted the volume ν(B n ) = cn . Now

Z Z
cn = 1= JΦn (r, Θ)
Bn R1

Z π π 
2π
Z
n−2 n−3
= ··· sin φ1 sin φ2 · · · sin φn−2 dφ1 · · · dφn−2
n 0 0

n−2 Z π  n−2
2π Y 2π Y 2π
= sink φdφ = Ik = [I1 I2 · · · In−2 ] ,
n 0 n n
k=1 k=1
Rπ
where Ik = 0 sink φdφ. From elementary calculus we know that for
k = 1, 2, ...
π
(2k)!π
Z
I2k = sin2k φdφ = ,
0 22k (k!)2
π
22k−1 ((k − 1)!)2
Z
I2k−1 = sin2k−1 φdφ = .
0 (2k − 1)!
5.1 Change of variables formula 323

Figure 5.5: Ellipsoid: x2 + 2y 2 + 3z 2 = 1

2π
Now, since Ik−1 Ik = k ,when n = 2m (m = 1, 2, ...), we have
 m−1 
2π π πm
c2m = = ,
2m (m − 1)! m!
while when n = 2m + 1 (m = 0, 1, ...),
 m−1
2m+1 π m

2π π
c2m+1 = · I2m−1 = .
2m + 1 (m − 1)! 1 · 3 · 5 · · · (2m + 1)
Thus,
πm 2m+1 π m
c2m = , c2m+1 = . (5.14)
m! 1 · 3 · 5 · · · (2m + 1)
4π π2 8π 2 π3 16π 3
Hence c1 = 2, c2 = π, c3 = 3 , c4 = 2 , c5 = 15 , c6 = 6 , c7 = 105 ,
and so on.

Example 5.1.34. Find the volume of the n-dimensional solid ellipsoid

2 x2n
 
n n x1
E = x ∈ R : 2 + ... + 2 ≤ 1 .
a1 an
Solution. Consider the linear transformation T : Rn → Rn given by

T (x1 , ..., xn ) = (a1 x1 , ..., an xn ).

The solid ellipsoid E n is the image of the unit ball B n under T . By

Proposition 5.1.7

ν(E n ) = ν(T (B n )) = | det T |ν(B b ) = a1 a2 · · · an ν(B n ).

324 Chapter 5 Change of Variables Formula

2
x2 2
In particular, the volume of the 3-dimensional ellipsoid a2
+ yb2 + zc2 ≤ 1
is 43 πabc.
Example 5.1.35. (Invariant integrals on groups)(*).
We now give an application of the Change of Variables formula to calcu-
lating Invariant integrals on groups . Rather than deal with generalities,
we shall do this with an example which has enough complexity to illus-
trate the general situation well.
Consider the set G of all matrices of the following form.
   
uv
G= x= : u 6= 0, v ∈ R .
01

which evidently can be identified with the (open) left and right half
planes. We write x = (u, v). As the reader can easily check, G is
a group under matrix multiplication (in particular G is closed under
multiplication). G is called the affine group of the real line R. We
define left and right translation (Lg and Rg respectively) in G as follows:
Lg (x) = gx, Rg (x) = xg. The reader can easily check that each Lg
and Rg is a diffeomorphism of G. Now as such they also operate on
functions. Namely if f ∈ Cc (G), the continuous functions on G with
compact support, we write Lg (f )(x) = f (gx) and Rg (f )(x) = f (xg).
Notice that Lgh = Lg Lh and Rgh = Rg Rh for each g and h ∈ G.
We say an integral over G is left invariant if
Z Z
Lg (f ) = f
G G

for every g ∈ G and f ∈ Cc (G). Similarly, an integral over G is called

right invariant if Z Z
Rg (f ) = f
G G
for every g ∈ G and f ∈ Cc (G). We now calculate such left or right
invariant integrals with the help of the Change of Variables formula.
To do this we first find the derivatives and Jacobians of the Lg ’s and
Rg ’s. A direct calculation, which we leave to the reader, shows that
if g = (u, v), then D(Lg ) = uI and hence | det D(Lg )| = u2 . Similarly
| det D(Rg )| = |u|. The first observation we make is that these Jacobians
are independant of x (as well as v) and only depend on g. They are also
5.1 Change of variables formula 325

nowhere zero. (This latter point is obvious in our case, but is true in
general since Lg and Rg are diffeomorphisms).
f
Now consider f ∈ Cc (G). Then | det D(L g )|
is also ∈ Cc (G) and in
f
fact for each g, as f varies over Cc (G) so does | det D(Lg )| . Similarly for
f
each g, as f varies over Cc (G) so does Now let φ ∈ Cc (G)
| det D(Rg )| .
and Lg be a change of variables. The Change of Variables formula (5.8)
tells us Z Z
φ(gx)| det D(Lg )|dudv = φ(x)dudv.
G G
f
Since this holds for all φ and we can take φ = | det D(Lg )| , we get

f (gx) f (x)
Z Z
| det D(Lg )|dudv = dudv.
G | det D(Lg )(gx)| G | det D(Lg )(x)|

After taking into account that Lgh = Lg Lh , the Chain Rule and the
fact that | det(AB)| = | det(A)|| det(B)| we get

dudv dudv
Z Z
f (gx) = f (x) ,
G | det D(Lg )| G | det D(Lg )|

for every f ∈ Cc (G). Thus

dudv dudv
=
| det D(Lg )| u2

is a left invariant integral for G. Similaly, dudv

|u| is a right invariant inte-
gral for G. Notice that here left invariant and right invariant integrals
are not the same.

EXERCISES

1. Evaluate the integral

√
Z Z
x + ydxdy,
Ω

where Ω is the parallelogram bounded by the lines x + y = 0,

x + y = 1, 2x − 3y = 0 and 2x − 3y = 4.
326 Chapter 5 Change of Variables Formula

2. Let n > 0. Compute the integral

Z Z
(x + y)n dxdy,
Ω

where Ω = {(x, y) : x ≥ 0, y ≥ 0, x + y ≤ 1}.

3. Evaluate the integral
Z 1Z xp
x2 + y 2 dydx,
0 0

using the transformation x = u, y = uv.

4. Evaluate the integral
3y
Z Z
p dxdy,
Ω 1 + (x + y)3
where Ω = {(x, y) : x + y < a, x > 0, y > 0}.
Hint. Set u = x + y, v = x − y.
5. Evaluate
x−y
Z Z
sin( )dxdy,
Ω x+y
where Ω = {(x, y) : x ≥ 0, y ≥ 0, x + y ≤ 1}.
6. Evaluate Z Z
sin x sin y sin(x + y)dxdy,
Ω
where Ω = (x, y) : x ≥ 0, y ≥ 0, x + y ≤ π2 .


7. Evaluate Z Z p
x2 + y 2 dxdy,
Ω
where Ω is the region by the circles x2 + y 2 = 4, x2 + y 2 = 9.
8. Evaluate Z Z y 
tan−1 dxdy,
Ω x
n √ o
where Ω = (x, y) : x2 +y 2 ≥ 1, x2 +y 2 ≤ 9, y ≥ √x3 , y ≤ x 3 .
Hint. Change to polar coordinates.
5.1 Change of variables formula 327

9. Evaluate Z Z
(x − y)4 ex+y dxdy,
Ω

where Ω is the square with vertices (1, 0), (2, 1), (1, 2), and (0, 1).

10. Evaluate Z 1 Z 1−x

y log(1 − x − y)dydx,
0 0
using the transformation x = u − uv, y = uv.

11. Show that if Ω = (x, y) : y ≥ 0, x2 + y 2 ≤ 1 ,



√
(x + y)2 2− 2
Z Z
p dxdy = π.
Ω 1 + x2 + y 2 3

12. Find the volume of the region in R3 which is above the xy-plane,
under the paraboloid z = x2 + y 2 , and inside the elliptic cylinder
x2 y2
9 + 4 = 1. Hint. Use elliptical coordinates x = 3r cos θ, y =
2r sin θ.

13. Use the transformation in R2 given by x = r cos3 θ,

y = r sin3 θ to prove that
n o the volume of the set Ω =
2 2 2
(x, y, z) : x + y + z ≤ 1 is 4π
3 3 3
35 .

14. Let Ω be the region in the octant with x, y, z ≥ 0, which is bounded

by the plane x = y + z = 1. Use the change of variables x =
u(1 − v), y = uv(1 − w) z = uvw to compute the integral

1
Z Z Z
dxdydz.
Ω y+z

15. Evaluate the integral

z
Z Z Z
dxdydz,
Ω 1 + x2 + y 2

where Ω = (x, y, z) : 1 ≤ x2 + y 2 ≤ 3, x ≥ 0, x ≤ y, 1 ≤ z ≤ 5 .


Hint. Use cylindrical coordinates.

328 Chapter 5 Change of Variables Formula

16. Integrate the function f (x, y, z) = z 4 over the ball in R3 centered

at the origin with radius a > 0.

17. Evaluate
1
Z Z Z
p dxdydz,
Ω x2 + y2 + (z − 2)2
where Ω is the unit ball x2 + y 2 + z 2 ≤ 1.

18. Evaluate
yz
Z Z Z
dxdydz,
Ω 1+x
where Ω is the portion of the closed unit ball in R3 which lies in
the positive octant x, y, z ≥ 0.

19. Evaluate
1
Z Z Z
3 dxdydz,
Ω (x2 + y 2 + z 2 ) 2
where Ω is the solid bounded by the two spheres x2 + y 2 + z 2 = a2
and x2 + y 2 + z 2 = b2 , where 0 < b < a.

Answers to selected Exercises

8 1 1
√ √
1. 15 . 2. 6(n+4) . 3. 6[ 2 + log(1 + 2)]).

π 38π π2 1 3 11
5. 0. 6. 16 . 7. 3 . 8. 6 . 9. 5 (e − e). 10. − 36 .

1 3 4πa7 2π 19−24 log 2

14. 2. 15. 2 π log 2. 16. 35 . 17. 3 . 18. 36 .

19. 4π log( ab ).

5.2 Improper multiple integrals

The situation we encounter here is as follows: we are given a function
f defined on a set Ω ⊆ Rn where f may not be integrable on Ω accord-
ing to Definition 4.1.37, either because Ω is unbounded or because f is
unbounded on Ω. Integrals over unbounded regions or integrals of un-
bounded functions are refered to as improper integrals. It is useful and
5.2 Improper multiple integrals 329

often necessary to be able to integrate over unbounded regions in Rn or

to integrate unbounded functions. As might be expected, by analogy
with improper integrals of functions of a single variable, this involves a
process of taking a limit of a (proper) multiple integral. However the
process of defining improper integrals in dimension n > 1 is trickier than
in dimension n = 1, (this is due to the great variety of ways in which a
limit can be formed in Rn ). In this connection we have the following.

Definition 5.2.1. Let Ω be a set in Rn (possibly unbounded). An

exhaustion of Ω is a sequence of bounded simple sets {Ωk } such that
Ω1 ⊂ Ω2 ⊂ Ω3 ⊂ ... ⊂ Ω with Ω = ∞
S
Ω
k=1 k .

Lemma 5.2.2. Let Ω ⊂ Rn be a bounded simple set and let ν be n-

dimensional volume. If {Ωk } is an exhaustion of Ω, then

lim ν(Ωk ) = ν(Ω).

k→∞

Proof. Clearly Ωk ⊂ Ωk+1 ⊂ Ω implies that ν(Ωk ) ⊂ ν(Ωk+1 ) ⊂ ν(Ω)

and limk→∞ ν(Ωk ) ≤ ν(Ω). To get equality, let  > 0. Note that since
ν(∂(Ω)) = 0, we can cover ∂(Ω) by a finite number S of open rectangles
R1 , ..., RN of total volume less than . Let E = N j=1 Rj . Then the set
n
Ω ∪ E is open in R and by construction it contains Ω. At the same
time, ν(Ω ∪ E) ≤ ν(Ω) + ν(E) < ν(Ω) + .
For each k = 1, 2, ..., applying this construction to each set Ωk of
the exhaustion {Ωk } withSk = 2k , we S∞obtain a sequence of open sets
{Ωk ∪ Ek } such that Ω = ∞ k=1 k Ω ⊂ k=1 (Ωk ∪ Ek ), and ν(Ωk ∪ Ek ) ≤
ν(Ωk ) + ν(Ek ) < ν(Ωk ) + k .
Now the collection of open sets {E, Ωk ∪ Ek : k = 1, 2, ...} is an open
cover of the compact set Ω. So that there is a finite number of these
open sets, say, E, (Ω1 ∪ E1 ), ..., (Ωm ∪ Em ) covering Ω. Since Ω1 ⊂ Ω2 ⊂
· · · ⊂ Ωm , the sets E, E1 , E2 , ..., Em , Ωm also form an open cover for Ω.
Hence,
m
X
ν(Ω) ≤ ν(Ω) ≤ ν(Ωm ) + ν(E) + ν(Ei ) < ν(Ωm ) + 2.
i=1

Therefore, ν(Ω) ≤ limk→∞ ν(Ωk ).

330 Chapter 5 Change of Variables Formula

We begin by defining the improper integral of a non-negative func-

tion. Let Ω be a (possibly unbounded) set in Rn and {Ωk } an exhaus-
tion of Ω. Suppose f : Ω → R is a non-negative (possibly unbounded)
function such that
R f is integrable over the sets Ωk ∈ {Ωk }. Then the
integrals sk = Ωk f exist for all k ∈ N and by the monotonicity of the
integral they form an increasing sequence of real numbers {sk }. If this
sequence is bounded by Theorem 1.1.20 we get a finite limit
Z
lim sk = lim f < ∞,
k→∞ k→∞ Ωk

otherwise it diverges to ∞. Hence the following definition is a natural

consequence.
Definition 5.2.3. Let {Ωk } be an exhaustion of the set Ω ⊆ Rn and
suppose f : Ω → [0, ∞) is integrable over each Ωk . If the limit
Z Z
f = lim f (5.15)
Ω k→∞ Ωk

exists and has a value independent of the choice of the sets in the ex-
haustion of Ω, this limt is called the improper integral of f over Ω.
When this limit is finite, we say that the integral converges and f is
integrable over Ω. If there is no common limit for all exhaustions of Ω
or its value is +∞, we say the integral diverges and f is not integrable
over Ω.
Definition 5.2.3 extends the concept of multiple integral to the case
of an unbounded region of integration or an unbounded integrand. This
calls for the following remark.
Remark 5.2.4. If Ω is a bounded simple set in Rn and f is integrable
over Ω, then the integral over Ω in the sense of Definition
R 5.2.3 converges
and has the same value as the (proper) integral Ω f of Definition 4.1.37.
Indeed, that f |Ωk is integrable over Ωk follows from the Lebesgue
criterion (as in Theorem 4.2.1 (6)). Since f is integrable over Ω, it is
bounded. Hence |f (x)| ≤ M for all x ∈ Ω, for some M > 0. From the
additivity of the integral we have
Z Z  Z 

  
 f− f = f  ≤ M ν(Ω\Ωk ).

(5.16)
  
Ω Ωk Ω\Ωk 
5.2 Improper multiple integrals 331

Now, since ν(Ω\Ωk ) = ν(Ω) − ν(Ωk ), letting k → ∞ in (5.16) and using

Lemma 5.2.2 we get Z Z
f = lim f,
Ω k→∞ Ωk

and the two definitions agree.

The verification that an improper integral converges is simplified by

the following proposition. We prove that for a non-negative function
the existence of the limit (5.15) is independent of the choice of the
exhaustion {Ωk } of Ω.

Proposition 5.2.5. Let Ω be a set in Rn and f : Ω → [0, ∞). Let {Ωk }

and {Ω0m } be exhaustions of Ω. Suppose f is integrable over each Ωk
and each Ω0m . Then
Z Z
lim f = lim f,
k→∞ Ωk m→∞ Ω0m

where the limit may be finite or +∞.

R
Proof. Let s = limk→∞ Ωk f be finite. For each fixed m = 1, 2, ..., the
sets Gkm = Ωk ∩ Ω0m , k = 1, 2, ... form an exhaustion of the set Ω0m .
Since each Ω0m is a bounded simple set and f is integrable over Ω0m it
follows from Remark 5.2.4 that
Z Z Z
f = lim f ≤ lim f = s.
Ω0m k→∞ Gkm k→∞ Ωk

Since f ≥ 0 and Ω0m ⊂ Ω0m+1 ⊂ Ω, it follows that limm→∞

R
Ω0m f exists
and Z
lim f = s0 ≤ s.
m→∞ Ω0
m

Reversing the roles of the exhaustions shows that s ≤ s0 also. Thus

s = s0 . If s = +∞, then arguing as in the above paragraph, we have
s ≤ s0 and so s0 = +∞ also.

Example 5.2.6. 1. If Ω = Rn and f : Rn → [0, ∞) is continuous

(or essentially continuous), we can take for the exhaustion to be

Ωk = Bk = {x ∈ Rn : ||x|| ≤ k} ,
332 Chapter 5 Change of Variables Formula

the ball of radius k centered at the origin. Another choice can be

Ωk = Ck = [−k, k] × [−k, k] × · · · × [−k, k],

the cube of side length 2k centered at the origin.

2. If Ω = B n the unit ball in Rn and f : B n → [0, ∞) is continuous

(or essentially continuous) but f (x) → ∞ as x → 0, we can take
Ωk to be the spherical shells
 
n 1
Ωk = x ∈ B : ≤ ||x|| ≤ 1 .
k
Note here that the union of the Ωk ’s is B n \ {0}, but this is im-
material, since, as we know, omission of a single point or even a
set of volume zero from a region of integration does not effect the
value of an integral.
Our first example of improper double integrals is a classic calculation
that leads to the following important integral.
Example 5.2.7. (Euler-Poisson integral).
Z ∞
2 √
e−x dx = π. (5.17)
−∞

Solution. Exhausting the plane R2 by a sequence of disks

Bk = (x, y) ∈ R2 : x2 + y 2 < k


and recalling Example 5.1.22 we have

Z Z
2 2
e−(x +y ) dxdy = π.
R2

If we now consider the exhaustion of the plane by the squares Ck =

[−k, k] × [−k, k], then by Fubini’s theorem and the fact that ea+b = ea eb
we have
Z Z Z k Z k 
−(x2 +y 2 ) −x2 −y 2
e dxdy = e e dy dx
Ck −k −k
5.2 Improper multiple integrals 333

Z k  Z k  Z k 2
−y 2 −x2 −x2
= e dy e dx = e dx .
−k −k −k

Letting k → ∞, Proposition 5.2.5 tells us that

Z ∞ 2
2
e−x dx = π.
−∞

Since any exponential is positive, taking the positive square root of both
sides yields the desired integral.
In particular,
Z Z Z
2 2 2 n
··· e−(x1 +...+xn) dx1 · · · dxn = e−||x|| dx = (π) 2 .
Rn Rn

The Euler-Poisson integral is inaccessible by one variable calculus

2
(the antiderivative of e−x is not an elementary function). The function
2
e−x is known as the Gaussian 7 function and comes up in many contexts.
Its graph is the “bell-shaped curve”. Rescaling it by the factor √1π , so
that the total area under its graph is 1, gives the normal distribution in
probability and statistics.

Corollary 5.2.8. (Comparison test). Let f and g be functions defined

on the set Ω ⊆ Rn and integrable over exactly the same bounded simple
R Ω. Suppose 0 ≤ f (x) ≤ g(x) for
subsets of R all x ∈ Ω. If the improper
integral Ω g converges, then the integral Ω f also converges.

Proof. Let {Ωk } be an exhaustion of Ω on whose elements both f and

g are integrable.
R By the
R monotonicity of the integral (Theorem 4.2.1
(2)) we have Ωk f ≤ Ωk g for all k = 1, 2, .... Since the functions are
non-negative, letting k → ∞
Z Z
f≤ g < ∞.
Ω Ω

7
C. Gauss (1777-1855). One of the great mathematicians with numerous pioneer-
ing contributions in several fields in mathematics, physics and astronomy.
334 Chapter 5 Change of Variables Formula

The basic properties of multiple integrals stated in Theorem 4.2.1

are readily extended to improper integrals of non-negative functions
using Definition 5.2.3. Thus far we worked with non-negative functions.
We shall now define the improper integral for a function f : Ω ⊆ Rn →
(−∞, ∞). The essential point is that the Rprecceding theory can be
applied to |f |, so that it makes sense to say Ω |f | converges. Before we
go on we need to define the so-called positive and negative parts of f .
Definition 5.2.9. (The functions f + , f − ).
Let f : Ω ⊆ Rn → R be a function. For x ∈ Ω we define the positive
part f + of f by
|f (x)| + f (x)
f + (x) = max {f (x), 0} =
2
and the negative part f − of f by
|f (x)| − f (x)
f − (x) = max {−f (x), 0} = .
2
Lemma 5.2.10. Let f : Ω ⊆ Rn → R. Then f + ≥ 0, f − ≥ 0,
f − = (−f )+ and
f = f + − f −.
Furthermore, |f | = max {f, −f } and

|f | = f + + f − .

The proof of the lemma is left to the reader as an exercise.

Definition 5.2.11. Let Ω be a setRin Rn and f : Ω → R. We say f is
absolutely integrable if the integral Ω |f | converges.
Proposition 5.2.12. Let f : Ω ⊆ Rn → R. If |f | is integrable over Ω,
then f is integrable.
Proof. Since 0R ≤ f + ≤ |fR| and 0 ≤ f − ≤ |f |, theR comparison testR tells
us that both Ω f + and Ω f − converge. Thus Ω f = Ω f + − Ω f −
R

also converges and f is integrable.

Next we prove the change of variable formula for improper integrals.

We begin with a preliminary result, which gives a useful exhaustion of
any open set in Rn in terms of compact simple sets.