Sie sind auf Seite 1von 7

0.

1 Practical Guide - Improper Integrals


We should start with some argument about improper integrals. Why do we consider such a problem. Quite simple
question.
Consider a Riemann integrable function ) : (c, ,) R , actually integrable on any subinterval [a, /] (c, ,).
So it makes sense to compute (measure)
b
_
o
)(r)dr
no|
= H(a, /)
The notation is of no importance, just shows the integral depends on a and / as two parameters.
Then ask what happens if a c and / ,, that is
lim
o!o,b!o
b
_
o
)(r)dr
no|
=
o
_
o
)(r)dr
A natural question comes out, why bother about taking limits instead of directly computing the integral
o
_
o
)(r)dr
Simply because :
1) ) may not be Riemann integrable on (c, ,) or
2) c = or , = + (quite unusual situation ! ) or
3) ) is Riemann integrable on (c, ,) , but the integral is hard to compute
We will consider only a very simple case: ) is continuous on (c, ,)
Therefore ) has antiderivatives, say 1 : (c, ,) R , derivable and 1
0
(r) = )(r)
Consequently
b
_
o
)(r)dr = 1(/) 1(a)
and
o
_
o
)(r)dr
no|
= lim
o!o,b!o
b
_
o
)(r)dr = lim
o!o,b!o
[1(/) 1(a)] = lim
b%o
1(/) lim
o&o
1(a)
It is quite easy to realize that such an integral makes sense only if both limits exist and are nite. We state
now a denition.
Denition. Let ) : (c, ,) R , be a continuous function on (c, ,) and 1
0
= ) some antiderivative. Call
"improper" integral the following
o
_
o
)(r)dr
no|
= lim
o!o,b!o
b
_
o
)(r)dr = lim
b%o
1(/) lim
o&o
1(a)
The improper integral is called
i) convergent if both limits exist and are nite
ii) divergent otherwise
we may use the simplied notation
o
_
o
)(r)dr = 1(r)[
o
o
= lim
r%o
1(r) lim
r&o
1(r)
1
Comment. This seems to be just a simple problem of taking limits and has little to do with integration.
Good remark indeed !
All we need to do is compute the antiderivative and take the limits. This should not be so hard.
Examples.
+1
_
0
c
r
dr ,
1
_
1
c
r
dr ,
+1
_
1
1
r
2
dr ,
+1
_
1
1
r
dr ,
1
_
0
1
_
r
dr ,
1
_
0
1
r
2
dr
+1
_
1
1
1 +r
2
dr ,
+1
_
1
1
1 +r
2
dr ,
+1
_
1
1
_
r
dr ,
+1
_
0
1
_
r
dr ,
+1
_
0
1
r
2
dr
Solutions.
+1
_
0
c
r
dr =
_
c
r
_

+1
0
= lim
r!+1
_
c
r
_
lim
r&0
_
c
r
_
=
1
c
1
+ 1 = 1 , = convergent
1
_
1
c
r
dr = (c
r
)[
0
1
= lim
r%0
(c
r
) lim
r!1
(c
r
) = 1
1
c
+1
= 1 , = convergent
+1
_
1
1
r
2
dr =
1
r

+1
1
= lim
r!+1
(
1
r
) (1) = 1
1
+
= 1 , = convergent
+1
_
1
1
r
dr = lnr[
+1
1
= lim
r!+1
lnr ln1 = + , = divergent
1
_
0
1
_
r
dr = 2
_
r

1
0
= 2 0 = 2 , = convergent
1
_
0
1
r
2
dr =
1
r

1
0
= 1 lim
r&0
(
1
r
) = 1 +
1
+0
= + , = divergent
+1
_
1
1
_
r
dr = 2
_
r

+1
1
= lim
r!+1
2
_
r 2 = + , = divergent
+1
_
1
1
1 +r
2
dr = arctan[
+1
1
= lim
r!+1
arctanr arctan1 =

2


4
=

4
, = convergent
+1
_
1
1
1 +r
2
dr = arctan[
+1
1
= lim
r!+1
arctanr lim
r!1
arctanr =

2
(

2
) = , = convergent
+1
_
0
1
r
2
dr =
1
_
0
1
r
2
dr +
+1
_
1
1
r
2
dr = divergent + convergent = divergent
+1
_
0
1
_
r
dr =
1
_
0
1
_
r
dr +
+1
_
1
1
_
r
dr = convergent + divergent = divergent

2
We summarise some of these results.
+1
_
1
1
r
o
dr =
_
convergent for c 1
divergent for c _ 1
1
_
0
1
r
o
dr =
_
divergent for c 1
convergent for c _ 1
Proof.
For c = 1 we have
+1
_
1
1
r
dr = lnr[
+1
0
= lim
r!+1
lnr ln1 = + , = divergent
1
_
0
1
r
dr = lnr[
1
0
= ln1 lim
r&0
lnr = () = + , = divergent
For c 1 we have
+1
_
1
1
r
o
dr =
+1
_
1
r
o
dr =
1
1 c
r
1o

+1
1
=
1
1 c
_
lim
r!+1
1
r
o1
1
_
=
1
1 c
(
1
+
1) =
1
c 1
, convergent
1
_
0
1
r
o
dr =
1
_
0
r
o
dr =
1
1 c
r
1o

1
0
=
1
1 c
_
1 lim
r&0
1
r
o1
_
=
1
1 c
(1
1
+0
) = + , divergent
For c < 1 we have
+1
_
1
1
r
o
dr =
+1
_
1
r
o
dr =
1
1 c
r
1o

+1
1
=
1
1 c
_
lim
r!+1
r
1o
1
_
=
1
1 c
(+1) , divergent
1
_
0
1
r
o
dr =
1
_
0
r
o
dr =
1
1 c
r
1o

1
0
=
1
1 c
_
1 lim
r&0
r
1o
_
=
1
1 c
(1 0) =
1
1 c
, convergent

The hard problem starts if we cannot compute the antiderivative !


How is it posssible to establish if an improper integral is either convergent or divergent ?
Comment.
Well it seems a bit complicated to deal with two limits at a time, so we split the improper integral in two parts
( for any choice c (c, ,) )
o
_
o
)(r)dr =
c
_
o
)(r)dr +
o
_
c
)(r)dr = lim
o!o
c
_
o
)(r)dr + lim
b!o
b
_
c
)(r)
The two improper integrals
c
_
o
)(r)dr ,
o
_
c
)(r)dr
have only one end point that could be called "improper".
It is just easier to deal with such improper integrals.
3
Remark. By splitting an improper integral
o
_
o
)(r)dr =
c
_
o
)(r)dr +
o
_
c
)(r)dr
we get
o
_
o
) is convergent if and only if both
c
_
o
) and
o
_
c
) are convergent
Remember the inequality

b
_
o
)(r)dr

_
b
_
o
[)(r)[ dr
Consequence.
If
o
_
o
[)(r)[ dr is convergent, then
o
_
o
)(r)dr is also convergent
Such an improper integral
o
_
o
[)(r)[ dr convergent is called "absolutely convergent".
The converse is not true.
Example.
+1
_
0
sin r
r
dr is convergent, but
+1
_
1
jsin rj
r
dr is divergent.
Inequality comparison test.
i) if [)(r)[ _ q(r) for all r (c, ,) then
o
_
o
q(r)dr convergent = both
o
_
o
[)(r)[ dr and
o
_
o
)(r)dr are
convergent.
ii) if q(r) _ )(r) for all r (c, ,) then
o
_
o
q(r)dr divergent =
o
_
o
)(r)dr divergent.
Examples.
1. For
+1
_
1
sinr
r
2
dr
we have

sinr
r
2

_
1
r
2
for all r (1, +) , also
+1
_
1
1
r
2
dr is convergent
it follows by inequality comparison test that
both
+1
_
1
[sinr[
r
2
dr and
+1
_
1
sinr
r
2
dr are also convergent.

2. For
+1
_
1
1
_
r +
3
_
r
dr
4
we have
1
_
r +
3
_
r
_
1
3
_
r +
3
_
r
=
1
2
3
_
r
, also
+1
_
1
1
2
3
_
r
dr is divergent
it follows by inequality comparison test that
+1
_
1
1
_
r +
3
_
r
dr is also divergent.

Limit comparison test. Suppose


o
_
o
)(r)dr has just one end point that may be called "improper", say , and
)(r) _ 0 for all r (c, ,) Compare this to another improper integral
o
_
o
q(r)dr , q(r) 0 for all r (c, ,) .
Assume the limit exists
lim
r%o
)(r)
q(r)
Then
i) if the limit is nite and non zero the improper integrals
o
_
o
) and
o
_
o
q have the "same nature" (that is they
are either both convergent, either both divergent )
ii) if the limit is zero and
o
_
o
q is convergent, then
o
_
o
) is also convergent.
Practical advice.
A frequently used "tool" is try to compare to integrals like
i)
+1
_
1
1
r
n
dr or ii)
1
_
0
1
r
n
dr or iii)
1
_
0
1
(1 r)
n
dr ,
b
_
o
1
(/ r)
n
dr
The problem is :
" nd some value for : R such that the limit is nite and non zero "
lim
r%o
)(r)
r
n
(if it is possible, that is if such value exists, it is not possible for any function)
Examples.
1. For
+1
_
1
1
r
4
+r
3
+
_
r + 2
dr
compare to
+1
_
1
1
r
n
dr
take the limit
lim
r!+1
1
r
4
+r
3
+
p
r+2
1
r
m
= lim
r!+1
r
n
r
4
+r
3
+
_
r + 2
= lim
r!+1
r
n
r
4
(1 +
1
r
+
_
1
r
7
+
2
r
8
= 1
this limit is nite and nonzero only for : = 4.
5
So we actually compare to
+1
_
1
1
r
4
dr , which is convergent, and therefore by limit comparison test
+1
_
1
1
r
4
+r
3
+
p
r+2
dr
is also convergent.
2. For
3
_
2
1
rsin(3 r)
dr
compare to
3
_
2
1
(3 r)
n
dr
take the limit
lim
r%3
1
r sin t(3r)
1
(3r)
m
= lim
r%3
(3 r)
n
rsin(3 r)
=

3
this limit is nite and nonzero only for : = 1.
So actually we compare to
3
_
2
1
(3r)
dr which is divergent, and therefore by limit comparison test it follows
that
3
_
2
1
r sin t(3r)
dr is also divergent.
It is also possible to use directly integration by parts or a change of variable for improper integrals.
Integration by parts. For improper integrals
o
_
o
)(r)q(r)dr
. .
.
=
__
)(r
_
q(r)

o
o

o
_
o
__
)(r)
_
q
0
(r)dr
. .
1
where
__
)(r
_
q(r)

o
o
= lim
r%o
__
)(r
_
q(r) lim
r&o
__
)(r
_
q(r)
and the formula makes sense only if both limits are nite.
Then the improper integrals () and (1) have the "same nature", that is they are either both convergent ,
either both divergent.
You may prefer let 1(r) =
_
)(r)dr = 1
0
(r) = )(r) and write the formula as
o
_
o
1
0
(r)q(r)dr
. .
.
= 1(r)q(r)[
o
o

o
_
o
1(r)q
0
(r)dr
. .
1
Example.
For
+1
_
1
sinr
r
dr
we integrate by parts as follows
+1
_
1
sinr
r
dr =
__
sinr
_
1
r

+1
1

+1
_
1
__
sinr
_ _
1
r
_
0
dr =
6
= lim
r!+1
cos r
r
lim
r&1
cos r
r

+1
_
1
cos r
r
2
dr = cos 1
+1
_
1
cos r
r
2
dr
+1
_
1
sinr
r
dr = cos 1
+1
_
1
cos r
r
2
dr
On the other hand we have
[cos r[
r
2
_
1
r
2
for all r (1, +)
since
+1
_
1
1
r
2
dr is convergent it follows by inequality comparison test that
+1
_
1
cos r
r
2
dr is also convergent, and so is
+1
_
1
sin r
r
dr .
Change of variable. For improper integrals.
o
_
o
)(r)dr =
J
_
c
)(n(t))n
0
(t)dt
where r = n(t) , n : (c, ,) (c, d) is bijective and we have r c =t c , r , =t d
Example.
For
1
_
0
1
_
r(1 r)
dr
proceed as follows.
Try a change of variable to get rid of the radical. So let r = a
2
and (1 r) = /
2
.
By adding there equalities we get 1 = a
2
+/
2
.
Now a simple model for such a case is a = sint and / = cos t (or conversely)
Therefore consider the change of variable r = sin
2
t ,
then we have r 0 =t 0 and r 1 =t
t
2
, and nally
1
_
0
1
_
r(1 r)
dr =
t/2
_
0
1
_
sin
2
t(1 sin
2
t)
(sin
2
t)
0
dt =
=
t/2
_
0
1
sint cos t
2 sint cos tdt =
t/2
_
0
2dt =
Consequently the improper integral is convergent and
1
_
0
1
_
r(1 r)
dr =

Das könnte Ihnen auch gefallen