78 2.

THE LEBESGUE INTEGRATION THEORY

11. Improper Riemann integrals

11.1. Preliminaries. The Riemann integral is defined for a bounded
function f and a bounded region Ω. Can the Riemann integral be
defined for f that is not bounded in a neighborhood of some point
of Ω and (or) not bounded Ω? For example, a continuous function
f(x) = e−x can be integrated over an unbounded interval [0, ∞) using
the rule
Z ∞ Z b
−x
e dx = lim e−x dx = lim (1 − e−b ) = 1
0 b→∞ 0 b→∞

Note f is integrable on every [0, b] because f is continuous. So, the

rule makes sense. Similarly, the function f(x) = x−1/2 is not bounded
on [0, 1], but it is continuous on every [a, 1] so it makes sense to define
Z 1 Z 1
dx dx √
√ = lim √ = lim (2 − 2 a) = 2
0 x a→0+ a x a→0+
The above Riemann integrals are known as improper Riemann integral.
So, a general idea is to reduce the integration domain to a bounded
region so that the function in question is integrable on it, and after
calculation of the integral over the reduced domain one has to take an
appropriate limit with respect to parameters of the reduced domain in
which the later becomes the region in question. The procedure is also
often called a regularization of the improper Riemann integral.
Does the value of the improper integral depend on the regulariza-
tion? The answer is not straightforward, especially for higher dimen-
sional integrals. Consider the function of two variables:
x2 − y 2
f(x, y) =
(x2 + y 2 )2
Evidently it is defined everywhere except the origin. One can choose
f(0, 0) to be any number. Regardless of the choice, f is not bounded
in any neighborhood of the origin. Suppose one wants to integrate this
function over
Ω = {(x, y) | x2 + y 2 ≤ 1 , x ≥ 0 , y ≥ 0}
which is the part of the unit disk that lies in the positive quadrant.
So, f is not bounded on Ω. In attempt to mimic a one-dimensional
improper integral, let us take a subregion Ωa ⊂ Ω
Ωa = {(x, y) | a2 ≤ x2 + y 2 ≤ 1 , x ≥ 0 , y ≥ 0}
 11. IMPROPER RIEMANN INTEGRALS 79

so that Ωa → Ω as a → 0. Using the polar coordinates it is not difficult

to show that ZZ
f(x, y) dxdy = 0
Ωa
Alternatively, the result follows from the symmetry argument. The
region Ωa is symmetric under the reflection about the line y = x:
(x, y) → (y, x), whereas the integrand is skew-symmetric, f(x, y) =
−f(y, x). Can one conclude that the improper Riemann integral of f
over Ω exists and is equal to zero?
It is obvious that Ω can be obtained in many ways as the limit of
subregions. For example, consider a collection of expanding subregions
Ωk ⊂ Ω which are defined in polar coordinates
x = r cos(θ) , y = r sin(θ)
by the conditions
Ωk = {(r, θ) | ak ≤ r ≤ 1 , βk ≤ θ ≤ π/2} , k = 1, 2, ...
where {ak } and {βk } are positive sequences that converge to 0 mono-
tonically. These regions coincide with Ωa with a = ak in which a small
sector with the angle βk is removed. So,
∞
[
Ωk ⊂ Ωk+1 ⊂ Ω , Ωk = Ω
k=1

The latter union is a proper mathematical way of saying that Ωk “ap-

proaches Ω and coincides with Ω in the limit k → ∞”. Using polar
coordinates
Z 1 Z π/2 2
r cos(2θ) 1
ZZ
f(x, y) dxdy = 4
rdrdθ = sin(2βk ) ln(ak )
Ωk a k βk r 2
The left side is an indeterminate form “0 × ∞”. Its limit may not
even exist and, even if it exists, it may have any value! Indeed, take
ak = e−c/βk where c > 0 so that ak → 0 monotonically if βk → 0
monotonically. Then
c sin(2βk )
ZZ
lim f(x, y) dxdy = − lim = −c
k→∞ Ωk k→∞ 2βk
2
If ak = βk , then the limit is 0 and, if ak = e−c/βk , c > 0, then the limit
is −∞. The reader is asked to verify that if the range of the polar angle
in Ωk is restricted to the interval π/2 − βk ≤ θ ≤ π/2, then the limit
can be made arbitrary positive number or +∞ by a suitable choice
of ak . So, the answer really depends on how the improper Riemann
integral is regularized! In fact, a similar result can be established for the
80 2. THE LEBESGUE INTEGRATION THEORY

integral of f over an unbounded region x2 + y 2 ≥ 1, x ≥ 0, y ≥ 0 (see

Exercises). Naturally, one wants a definitive (or unique) value of the
integral, and, for this reason, the definition of the improper Riemann
integral requires some amendments in order to be applicable for higher
dimensional integrals.

11.2. Improper Riemann integrals. Let Ω ⊂ RN bounded or unbounded.

An exhaustion of Ω is a sequence of regions {Ωk }∞
1 such that
• each Ωk is bounded, closed, and contained in Ω;
• Ωk+1 contains Ωk ;
• the union of all Ωk coincides with Ω except possibly a set of
measure zero.
An example of an exhaustion was given in the previous section. If a
bounded function f is Riemann integrable on a closed bounded region
Ω, then it can be proved that
Z Z
N
lim f(x) d x = f(x) dN x
k→∞ Ωk Ω
which is nothing but the continuity property of the Riemann integral.
If f is not bounded and/or Ω is not bounded, the improper integral is
defined by this property.
Definition 11.1. Let {Ωk }∞1 be an exhaustion of Ω. Suppose that
a function f on Ω is Riemann integrable on each Ωk . Then the function
f is said to be Riemann integrable on Ω if the limit
Z Z
N
lim f(x) d x = f(x) dN x
k→∞ Ωk Ω

exists and is independent of the choice of Ωk . In this case, the value of

the limit is called an improper Riemann integral of f over Ω.

11.3. Improper integral of non-negative functions. It is not difficult to

find a particular exhaustion and investigate the limit of the sequence
of the integrals. Even if the limit is found to exist, one still has to show
that it is independent of the choice of the exhaustion in order to prove
that the found limit is the value of the improper integral. It is not
possible to do so by constructing all possible exhaustions. How then
can this be done?
Suppose that f(x) is non-negative on Ω. For any exhaustion, the
sequence of integrals is monotonically increasing
Z Z
N
0≤ f(x) d x ≤ f(x) dN x
Ωk Ωk+1
 11. IMPROPER RIEMANN INTEGRALS 81

by the positivity property of the Riemann integral and that Ωk ⊂ Ωk+1 .

The sequence converges only if it is bounded. So, there are only two
possibilities: either the limit is a number
Z
lim f(x) dN x = If < ∞
k→∞ Ωk

or it is infinite, If = ∞. Suppose that this limit exists, If < ∞. Let

{Ω0k } be another exhaustion of Ω. Then the sequence of the integrals
is bounded: Z
f(x) dN x ≤ If
Ω0k

because f(x) ≥ 0 and Ω0k ⊂ Ω. Since the sequence is also monotonic,

it converges Z
lim f(x) dN x = If0 ≤ If
k→∞ Ω0k

and its limit cannot exceed If . On the other hand, one can swap the
roles of the exhaustions and use the same argument:
Z
|f(x)| dN x ≤ If0 ⇒ If ≤ If0
Ωk

because Ωk ⊂ Ω and f(x) ≥ 0. Therefore If = If0 and the value of the

limit does not depend on the choice of the exhaustion.
Theorem 11.1. (Improper integral for non-negative functions)
Suppose that
(i) f(x) ≥ 0 , ∀x ∈ Ω;
(ii) {Ωn } and {Ω0n } are exhaustions of Ω;
(iii) f is Riemann integrable on each Ωn and Ω0n
Then Z Z
lim f(x) dN x = lim f(x) dN x
n→∞ Ωn n→∞ Ω0n

where the limit can also be +∞.

Suppose that a bounded function f is Riemann integrable on a
bounded, closed region Ω, then the non-negative functions
1 
f± (x) = |f(x)| ± f(x) ≥ 0
2
are also integrable (because the absolute value of an integrable func-
tion is integrable). The function f+ (x) coincides with f(x) whenever
f(x) ≥ 0 and vanishes otherwise, whereas f− (x) coincides with −f(x)
82 2. THE LEBESGUE INTEGRATION THEORY

whenever f(x) ≤ 0 and vanishes otherwise. They can alternatively be

written as
f+ (x) = max{f(x), 0} , f− (x) = − min{f(x), 0}
Thus, any Riemann integrable function can be written as the difference
of two non-negative integrable functions:
f(x) = f+ (x) − f− (x) ,
Z Z Z
N
f(x) d x = f+ (x) d x − f− (x) dN x
N
Ω Ω Ω

and vice versa (integrability of f± implies integrability of f and |f| =

f+ + f− by the linearity of the integral).
Using the limit laws, the following theorem can easily be established
from the above representation.
Theorem 11.2. Suppose that the improper integrals of f± over Ω
exist. Then the improper integral of f over Ω exists and can be com-
puted in any exhaustion {Ωn } of Ω.
Indeed, by the limit laws and the existence of the improper integral
of f± ,
Z Z  
N
f(x) d x = lim f+ (x) − f− (x) dN x
Ω n→∞
ZΩn Z
N
= lim f+ (x) d x − lim f− (x) dN x
n→∞ Ω n→∞ Ω
n n
Z Z
= f+ (x) dN x − f− (x) dN x
Ω Ω
and, by Theorem 11.1 the values of the limits in the right side of the
equation do not depend on the choice of the exhaustion (or regulariza-
tion) of the integrals.
Corollary 11.1. Let {Ωn } be an exhaustion of Ω. Suppose that
f and its absolute |f| are integrable on each Ωn and
Z Z
N
lim |f(x)| d x = |f(x)| dN x < ∞
n→∞ Ωn Ω

Then the improper integral of f over Ω exists and

Z Z
N
f(x) d x = lim f(x) dN x
Ω n→∞ Ωn

In other words, if the improper integral of the absolute value of f

converges in any particular regularization, then the improper integral
 11. IMPROPER RIEMANN INTEGRALS 83

of f exists and can be computed in any suitable regularization. Indeed,

since
0 ≤ f± (x) ≤ |f(x)|
It is concluded that monotonic sequences of integrals of f± over Ωn are
bounded:
Z Z Z
N N
0≤ f± (x) d x ≤ |f(x)| d x ≤ |f(x)| dN x < ∞
Ωn Ωn Ω

and, hence, converge. By Theorem 11.1 the limits are independent of

the choice of Ωn . By Theorem 11.2, the improper integral of f over Ω
exists (it is independent of regularization).

11.4. Absolutely and conditionally convergent integrals. If the limit

Z
lim f(x) dN x
n→∞ Ωn

exists for some exhaustion (regularization) {Ωn }, it is called convergent

in this exhaustion (regularization). If the integral of the absolute value
also converges in this exhaustion
Z
lim |f(x)| dN x = If < ∞
n→∞ Ωn

then the integral of f over Ω is said to be absolutely convergent, and if

If = ∞, then the integral of f is said to be conditionally convergent in
the exhaustion {Ωn }. In the previous section, it was shown that
• the absolute convergence of the Riemann integral implies the
existence of the improper Riemann integral (the value of the
integral does not depend on the regularization).
This terminology is due to the analogy with absolutely and condi-
tionally convergent series. The region Ω can be viewed as the union of
shells obtained by carving out Ωk from Ωk+1 :
[∞
Ω= Sk , Sk = Ωk \ Ωk−1
k=1

where Ω0 is an empty set by definition. Note that Ωk contains its

boundary so that Ωk \ Ωk−1 does not contain its inner boundary. The
shell Sk is closed and bounded set (the missing boundary is added by
taking the closure of Ωk \ Ωk−1 ). Suppose that f is integrable on each
Sk . Then
Z ∞ Z
X
N
f(x) d x = f(x) dN x
Ω k=1 Sk
84 2. THE LEBESGUE INTEGRATION THEORY

If the series of integrals of the absolute value converges

Z X∞ Z
N
|f(x)| d x = |f(x)| dN x = If < ∞
Ω k=1 Sk

then the series for the integral of f converges absolutely and, hence,
converges because
Z  Z
N 
|f(x)| dN x
 

 f(x) d x ≤
Sk Sk

Recall that every absolutely convergent series converges.

The comparison test for the absolute convergence. It has been shown
that the absolute convergence of a Riemann integral in a particular
regularization is sufficient to establish the existence of the improper
Riemann integral, just like the absolute convergence of a series implies
the convergence of the series. It is therefore important to establish
some tests for absolute convergence of Riemann integrals.
Let f and g be Riemann integrable on Ω, and
|f(x)| ≤ g(x) , ∀x ∈ Ω
Then Z Z
N
|f(x)| d x ≤ g(x) dN x
Ω Ω
If now f and g are not integrable in the proper sense, then for any
exhaustion {Ωn }
Z Z
N
|f(x)| d x ≤ g(x) dN x
Ωn Ωn

If the improper integral of g converges, then the integral f converges

absolutely because
Z Z
N
g(x) d x ≤ g(x) dN x < ∞
Ωn Ω
Z Z
N
⇒ lim |f(x)| d x ≤ g(x) dN x
n→∞ Ωn Ω

By Theorem 11.2 the limit does not depend on the choice of the ex-
haustion and the improper integral of f exists.
Theorem 11.3. (Comparison test for absolute convergence)
If the absolute value of a function f is bounded on Ω by a function
whose improper Riemann integral over Ω exists, the improper integral
of f over Ω also exists (it converges absolutely).
 11. IMPROPER RIEMANN INTEGRALS 85

It should be pointed out that in the Riemann theory the integrabil-

ity of |f| does not imply integrability of f (e.g., f(x) = 1 if x is rational,
and f(x) = −1 otherwise so that |f(x)| = 1). In the above theorem, it
is assumed that f is integrable on any region of an exhaustion so that
its improper integral can at least be defined.
Suppose Ω = RN and f is a continuous function. Clearly it is
integrable on any ball Ωn = Bn (that is, |x| ≤ n, n = 1, 2, ...). So the
existence of the improper integral would depend on how fast f fall off
as |x| → ∞.
Proposition 11.1. Let f be a continuous function such that
M
|f(x)| ≤ , |x| ≥ R
|x|p
for some constants M and R, and p > N, then the improper integral
of f over the whole space exists and
Z Z
N
f(x) d x = lim f(x) dN x < ∞ .
RN n→∞ |x|≤n

Consider the case N = 2. The integral over the whole plane is split
into the integral over the disk BR and the rest of the plane R2 \ BR .
Since the integral over BR is a regular integral, one has to investigate
the convergence of the improper integral over the rest of the plane.
Since f ≥ 0, if it converges in a particular regularization, then it con-
verges in any other regularization to the same value. Let Ωn be an
annulus R ≤ |x| ≤ n. Then
Z 2π Z n
M 2 M
Z Z
2
|f(x)| d x ≤ p
d x= p
rdrdθ
Ωn Ωn |x| 0 R r
 
2πM 1 1
= −
p − 2 Rp−2 np−2
The right side converges if p > N = 2 when n → ∞. Therefore the
integral of f converges absolutely and, hence, the improper integral of
f exists by the comparison test (it can be computed in any suitable
regularization).
Similarly, suppose f is not bounded in any neighborhood of a par-
ticular point, and it is continuous otherwise. Without loss of generality,
the singular point can be chosen to be the origin x = 0 (values of f(x)
becomes infinitely large as x approaches 0). Then it is integrable on
Ωn = Ω \ Ba where Ba is a ball of radius a, |x| < a.
86 2. THE LEBESGUE INTEGRATION THEORY

Proposition 11.2. Suppose that f is not bounded in any ball Ba

and continuous on Ω \ Ba where Ω is a bounded region with a piecewise
smooth boundary. If
M
|f(x)| ≤ p , |x| ≤ R
|x|
for some constants M and R, and p < N, then the improper integral
of f exists and
Z Z
N
f(x) d x = lim+ f(x) dN x .
Ω a→0 Ω\Ba

A proof of this assertion can also be done by using spherical coor-

dinates in RN . For example, if N = 2, the using the polar coordinates
Z 2π Z R
M 2 M
Z Z
2
|f(x)| d x ≤ p
d x= rdrdθ
|x| 0 a rp
a≤|x|≤R a≤|x|≤R
2πM  2−p 2−p

= R −a
2−p
The right side converges in the limit a → 0+ if p < 2 = N and so
does the left side. Therefore the integral of f converges absolutely and,
hence, the improper integral of f exists by the comparison test.

Remark. In the Lebesgue integration theory, all integrals (proper and

improper) are treated in the same footing, and the integrability of the
absolute value of a function implies integrability of the function. In
this sense, the Lebesgue integrals are often called absolutely convergent
integrals.

11.5. Conditionally convergent integrals. Let the integral of f over Ω be

conditionally convergent, that is, it converges in some regularization,
but the integral of the absolute value diverges. In this case, the integrals
of f± must diverge, and the value of a conditionally convergent integral
is an indeterminate form “∞ − ∞” which happens to be a number in
a particular regularization:
Z Z Z 
N N N
lim f(x) d x = lim f+ (x) d x − f− (x) d x
n→∞ Ωn n→∞ Ωn Ωn

Note that the divergence of the integral of |f| = f+ + f− implies that

either the integral of f+ , or f− , or both diverge because f± ≥ 0. The
conditional convergence of the integral of f (the existence of the limit
in the left side) is only possible when integrals of f± diverge.
 11. IMPROPER RIEMANN INTEGRALS 87

The integrals of f± resemble the (divergent) series of positive and

negative terms of a conditionally convergent series. The sum of such
a series depends on the arrangement of terms (the order in which the
terms are added). In the case of conditionally convergent integrals, the
value depends on the choice of the exhaustion (or regularization). In
other words, by choosing a suitable exhaustion one can always make
the difference of the integrals of f+ and f− over Ωn to be convergent to
any desired number even though both the sequences diverges to +∞.
This is illustrated with the following example.
Consider the improper integral
Z ∞ Z bn
sin(x) sin(x)
dx = lim dx
0 x n→∞ 0 x
where {bn } is positive, monotonically increasing, unbounded sequence.
Here the integrand extended to x = 0 by continuity (the integrand
approaches 1 as x → 0+ ). In particular, let us take
bn = πn , n = 1, 2, ...
This regularization corresponds to the exhaustion:
Ωn = [0, πn]
The corresponding shells are intervals
Sn = [π(n − 1), πn] , Ωn = Ωn−1 ∪ Sn
If the limit exists, then is equal to the sum of the series
Z ∞ ∞ Z πn
sin(x) X sin(x)
dx = dx
0 x n=1 π(n−1)
x
This is an alternating series because the integrand is positive on S2k−1
and negative on S2k . It follows from the inequality
| sin(x)| | sin(x)| | sin(x)|
≤ ≤ , n>1
πn x π(n − 1)
that Z πn
2 2 | sin(x)|
≤ an ≤ , an = dx > 0
πn π(n − 1) π(n−1) x
and Z ∞ ∞
sin(x) X
dx = (−1)n+1 an
0 x n=1
The sequence {an } is positive and converges to 0 monotonically because
2
an+1 ≤ ≤ an
πn
88 2. THE LEBESGUE INTEGRATION THEORY

By the alternating series test, the series converges.

However, by the comparison test:
n Z πn n Z ∞
2X1 | sin(x)| X | sin(x)|
≤ = ak ⇒ dx = ∞
π k 0 x 0 x
k=1 k=1
P1
the series and the integral do not converge absolutely because n
=
∞. So, the integral is only conditionally convergent and does not ex-
ist in the sense of Definition 11.1. Its value depends on the choice of
regularization. The latter follows from the Riemann theorem about
alternating series. In particular, the sum can be made equal to any
desired number by a suitable rearrangement. In this example, a re-
arrangement correspond to a different exhaustion made of the same
set of intervals Sn . Consider a rearrangement {Sn0 } of the sequence of
intervals {Sn } and put
Ω01 = S10 , Ω0n+1 = Ω0n ∪ Sn+1
0

So, Ω0n is a collection of any n intervals from {Sn } and Ω0n+1 is obtained
by adding any of remaining intervals in the collection {Sn }. In other
words, the order in which the intervals from {Sn } are added to obtain
an exhaustion is changed, but
[∞ [∞
Ωn = Ω0n = [0, ∞) .
n=1 n=1

The function is still integrable on any finite collection of intervals Ω0n .

Therefore in this exhaustion (regularization)
Z ∞ ∞ Z
sin(x) X sin(x)
dx = dx
0 x n=1 Sn
0 x
The
P series in the left side is a rearrangement of the alternating series
n+1
n (−1) an . Its sum can be any number, or ±∞, or the series may
not even converge.
The above example admits a generalization. Let f(x) be continuous
on Ω. Then let Ω+ be a subset where f(x) ≥ 0 and Ω− be a subset
where f(x) ≤ 0. Then Ω = Ω+ ∪ Ω− . Let {Ω± n } be exhaustions of Ω ,
±

respectively. Consider the following exhaustion of Ω

{Ωn } = {Ω+ − + −
1 , Ω1 , Ω2 , Ω2 , · · · }

Then the limit of integrals of f over Ωn is the sum of an alternating

series
Z X∞ Z Z 
N N N
lim f(x) d x = f(x) d x + f(x) d x
n→∞ Ωn +
Sn −
Sn
n=1
 11. IMPROPER RIEMANN INTEGRALS 89

where S1± = Ω± ± ± ±
1 and Ωn+1 = Ωn ∪ Sn . If the integral does not converge
absolutely, then one can rearrange the series so that it can converge to
any desired number. This rearrangement corresponds to a new exhaus-
tion (a different order of selecting shells Sn± in the new exhaustion).

11.6. Exercises.

1. Consider the improper integral

x2 − y 2
ZZ
dxdy
Ω (x2 + y 2 )2

where
Ω = {(x, y) | x2 + y 2 ≥ 1 , x ≥ 0 , y ≥ 0}

Take an exhaustion {Ωn } which is a rectangle in the polar coordinates

(r, θ) ∈ [1, an ] × [αn , π/2 − βn ] where αn and βn are positive and tend
to 0 monotonically, while 1 < an increases monotonically to infinity, as
n → ∞. Show that there is a choice of αn , βn , and an such that the
sequence of integrals over Ωn can converge to any real number or ±∞.

2. For the integrand in Problem 1, find f± (x, y) and show that for
any exhaustion the improper integrals of f± diverge
ZZ
f± (x, y) dxdy = ∞
Ω

3. Prove Proposition 11.1 for N = 3 using spherical coordinates.

4. Prove Proposition 11.2 for N = 3 using spherical coordinates.

5. Let f(t) be a continuous function and t ≥ 0. Then for any ball

Ba ⊂ RN
a
2π N/2
Z Z Z
N N −1
f(|x|) d x = σN f(r) r dr , σN = dS = ,
Ba 0 |x|=1 Γ(N/2)

where σN is the surface area of the unit sphere |x| = 1. The assertion is
proved by converting the integral to spherical coordinates in RN . Use
this result to prove Propositions 11.1 and 11.2.
90 2. THE LEBESGUE INTEGRATION THEORY

6. Let p and q be positive integers. Do the following improper in-

tegrals exist in the sense of Definition 11.1?
Z ∞ 2
sin (x)
(i) dx
0 xp
Z ∞
cosq (x)
(ii) dx ,
1 xp
Z 1  
p 1
(iii) sin dx
0 x