Integration and Polar Coordinates

S. Kesavan
The Institute of Mathematical Sciences,
CIT Campus, Taramani,
Chennai - 600 113.

email: [email protected]

Abstract: A simple proof of the formula for the integration of radial func-
tions on RN , N ≥ 2, is given. As an application, the volume of the N -
dimensional unit ball is computed.

1
1 Introduction
The method of substitution, also known as change of variables, is a useful tool
in integration. One of the high points in a course on multivariate calculus is
the change of variable formula which states that if f = (f1 , f2 , · · · , fN ) : E ⊂
RN → RN is a bijection which is smooth both ways, and if g is an integrable
function on f (E), then
Z Z
g(x) dx = g(f (x))|det(f 0 (x))| dx
f (E) E

where x = (x1 , x2 , · · · , xN ), f 0 (x) is the Jacobian matrix at x whose ij-th

entry is given by
∂fi
(x)
∂xj
and dx indicates that we are performing an N -dimensional volume integra-
tion, i.e., we have an N -fold integral with respect to dx1 dx2 · · · dxN .
The most familiar application of this is the case of polar coordinates in
the plane given by x = r cos θ and y = r sin θ and the integral
Z Z
g(x, y) dxdy

transforms as Z Z
ge(r, θ) rdrdθ

with appropriate limits of integration (here g(x, y) = ge(r, θ)). Similarly, in

R3 , we have the spherical polar coordinates x = r sin θ cos ϕ, y = r sin θ sin ϕ, z =
r cos θ and the integral
Z Z Z
g(x, y, z) dxdydz

transforms as Z Z Z
ge(r, θ, ϕ) r2 sin θdrdθdϕ,

again, with appropriate limits of integration, where g(x, y, z) = ge(r, θ, ϕ).

Polar coordinates are particularly useful when the function to be inte-
grated is a radial function, i.e. it depends only on
N
! 21
X
r = |x| = x2i .
i=1

2
In such a case, the integral reduces to one involving just a single variable.
For instance, if we are integrating over the entire space, the integral reduces
to Z ∞
2π ge(r)r dr
0
when N = 2 and to Z ∞
4π ge(r)r2 dr
0
when N = 3. The most familiar example of this is the evaluation of the
Gaussian integral, ubiquitous in probability theory,
Z ∞
2
e−x dx.
−∞

If we set this as I, then,

Z ∞Z ∞ Z ∞
−(x2 +y 2 ) 2
2
I = e dxdy = 2π e−r r dr
−∞ −∞ 0
√
which yields I = π.
When the dimension exceeds three, it is difficult to explicitly write down
the spherical coordinates and even if we succeed, the ensuing calculations in-
volving the N -dimensional Jacobian will certainly be horrendous. We would
like to have a simple formula, at least for the integration of radial functions,
and this is what we will set out to establish in this article.
Looking at the cases N = 2 and N = 3, we observe that 2πr is the
perimeter of the circle of radius r and 4πr2 is the surface area of the sphere
of radius r in these respective dimensions and so we are tempted to conjecture
that the integral of a radial function would transform as the integral with
respect to r after multiplication by the (N − 1)-dimensional surface measure
of the sphere of radius r in RN . Is this correct, and, if so, what is the surface
measure of the sphere in RN ?
Let us denote by ωN the volume of the unit ball in RN and by αN ,
the (N − 1)-dimensional surface measure of the unit sphere. The change of
variables x 7→ x/R maps the ball of radius R onto the unit ball and the
Jacobian turns out to be a scalar matrix and so we easily conclude that the
volume of the ball of radius R is ωN RN from the change of variable formula
mentioned at the beginning. What about the surface measure of the sphere
of radius R? For this we have a very nice application of Gauss’ divergence

3
theorem: if Ω ⊂ RN is a domain, ∂Ω its boundary on which we have a well-
defined outward unit normal ν, then for a smooth vector valued function
v : Ω → RN , we have
Z Z
div(v(x)) dx = v(x).ν(x) dS(x)
Ω ∂Ω

where the ‘dot’ denotes the usual scalar product in RN and dS(x) indicates
that we are performing a surface integral with respect to the appropriate
induced surface measure.
Setting v(x) = x, we get div(v(x)) = N and so the left-hand side just
gives N times the volume of the domain. If Ω is the ball of radius R, then
it is N ωN RN . The unit outer normal at a point x on the boundary ∂Ω of
the ball Ω of radius R and centre at the origin is just x/R and so v(x).ν(x)
reduces to R and so the right-hand side is just R times the surface measure
of ∂Ω. Hence we conclude that the surface measure of the sphere of radius
R is N ωN RN −1 , which when R = 1 yields

αN = N ω N

.
In the sequel, we will prove that the integral of a radial function f (x) =
f (r) over a region which is spherically symmetric about the origin is indeed
e
given by Z Z
−1
N ωN fe(r)r N
dr = αN fe(r)rN −1 dr

with appropriate limits of integration. We will also determine ωN .

2 Integration of radial functions

Let Ω be the ball of radius R > 0 with centre at the origin. Consider a
continuous real valued function, f , on the closed ball which is radial, i.e.
f (x) = fe(|x|) where fe : [0, R] → R is a continuous function. Consider a
partition
P : 0 = r0 < r1 < r2 < · · · < rn = R
of the interval [0, R]. Let Ωi , 1 ≤ i ≤ n, denote the annular region

{x ∈ RN | ri−1 ≤ |x| < ri }

4
so that Ω = ∪ni=1 Ωi . By the mean value theorem, there exists ξi ∈ (ri−1 , ri )
such that
riN − ri−1
N
= N ξiN −1 (ri − ri−1 ).
Choose yi ∈ Ωi such that |yi | = ξi . Define

fP (x) = f (yi ), x ∈ Ωi , 1 ≤ i ≤ n.

Let
µ(P) = max (ri − ri−1 ).
1≤i≤n

If x ∈ Ωi , then

|f (x) − fP (x)| = |fe(|x|) − fe(|yi |)| = |fe(|x|) − fe(ξi )|.

Since fe is uniformly continious, given ε > 0, there exists δ > 0 such that, for
all x ∈ Ω, we have
|f (x) − fP (x)| < ε
for all partitions P such that µ(P) < δ. Thus, fP converges uniformly to f
as µ(P) → 0. Consequently,
Z Z
fP (x) dx → f (x) dx
Ω Ω

as µ(P) → 0. But since fP is constant in each subdomain Ωi and since the

N
volume of Ωi is given by ωN (riN − ri−1 ), we get
Z n
X n
X
fP (x) dx = fe(ξi )ωN (riN − ri−1
N
) = fe(ξi )ωN N ξiN −1 (ri − ri−1 )
Ω i=1 i=1

and the last term converges, as µ(P) → 0, to

Z R
N ωN fe(r)rN −1 dr.
0

Thus, we have shown that if f is a continuous and radial function (i.e.

f (x) = fe(|x|)) on the closed ball of radius R, centered at the origin, then
Z Z R
f (x) dx = N ωN fe(r)rN −1 dr.
Ω 0

5
By passing to the limit as R → ∞, it is now a simple exercise to see that if
f is a continuous and improperly Riemann integrable function on RN which
is radial, then
Z Z ∞
f (x) dx = N ωN fe(r)rN −1 dr.
RN 0

3 Volume of the unit ball

Let us now apply the preceding result to the function
PN
x2i 2
f (x) = e− i=1 = e−r .

On one hand (see Section 1),

Z Z ∞
2 N
N
f (x) dx = Πi=1 e−xi dxi = π 2 .
RN −∞

On the other hand, using polar coordinates,

Z Z ∞
2
f (x) dx = N ωN e−r rN −1 dr.
RN 0

Setting r2 = s, we get
Z ∞
1 ∞ −s N −1
Z  
−r2 N −1 1 N
e r dr = e s 2 ds = Γ
0 2 0 2 2

where Γ denotes the usual Gamma function defined by

Z ∞
Γ(t) = e−s st−1 ds, t > 0.
0

By a simple integration by parts, we see that

Γ(t + 1) = tΓ(t), t > 0 (1)

and so we deduce from the above computations that

N N
π2 π2
ωN = N N = .
2
Γ( 2 ) Γ( N2 + 1)

6
 When t = 1/2, the integral defining the Gamma function reduces to the
Gaussian integral mentioned at the beginning of this article and we have
√
 
1
Γ = π.
2

By repeated use of (1) we see that for any positive integer n, Γ(n + 1) = n!.
Thus we can compute Γ( N2 + 1) by appealing to these facts and repeatedly
using (1).

Examples
π
ω2 = = π.
Γ(2)
3
π2
ω3 = 3 .
Γ( 2 + 1)
Now,
3√
     
3 3 3 31 1
Γ +1 = Γ = Γ = π.
2 2 2 22 2 4
Thus,
4
ω3 = π.
3
In the same way we can show that

π2 8π 2
ω4 = , ω5 =
2 15
and so on.

4 Concluding remarks
The integration formula for continuous radial functions that was proved in
this article is a very special case of a deep result in measure theory called
the coarea formula. It states that if u : RN → R is a Lipschitz continuous
function and is such that for almost every r ∈ R, the level set
def
{u = r} = {x ∈ RN | u(x) = r}

7
is a smooth (N − 1)-dimensional hypersurface in RN , and if f : RN → R is
a continuous and integrable function, then
Z Z ∞ Z 
f (x)|Du(x)| dx = f (x) dS(x) dr.
RN −∞ {u=r}

(By another deep theorem of Radamacher, Lipschitz continuous functions

are differentiable almost everywhere and Du denotes the gradient vector of
u.) In particular, choosing u(x) = |x|, we get the polar coordinate formula
Z Z ∞ Z 
f (x) dx = f (x) dS(x) dr.
RN 0 {|x|=r}

If f is radial, then, following our usual notation, the inner integral just
becomes fe(r) times the surface measure of the sphere of radius r which
gives N ωN rN −1 fe(r). For a detailed treatment of the coarea (pronounced
co-area) formula, the reader is referred to the treatise ‘Measure Theory and
Fine Properties of Functions’ by L. C. Evans and R. F. Gariepy (CRC Press,
1992).
Finally, we can also derive a recursive relation for the volume of the unit
ball. Indeed, we can slice up the N -dimensional hemisphere by sections
parallel to the base. Then at height t, 0 < t < 1, we get an √ infinitesimal
‘cylinder’ with ‘base’ an (N − 1)-dimensional ball of radius 1 − t2 and of
‘height’ dt. Thus,
π
Z 1 Z
2
N −1
2
ωN = 2 ωN −1 (1 − t ) 2 dt = 2ωN −1 cosN θ dθ.
0 0

Recall that
N −1 N −3
· · · 12 π2 , if N is even

π
N N −2
Z 
2
cosN θ dθ =
N −1 N −3
0 · · · 23 , if N is odd.

N N −2

The reader is referred to an article by K. B. Athreya (cf. Resonance, April

2008) where this idea is developed and the asymptotic behaviour of ωN , as
N → ∞ is examined.