Sie sind auf Seite 1von 12

Evaluating integrals using residues.

As many things in mathematics, using residues to evaluate integrals gets better with experience. The integrals
that can be evaluated fall roughly into the following categories.
1. Integrals Z 2π
R(cos θ, sin θ) dθ
0
where R is a rational function. The trick is to transform them into an integral over the unit circle using
z2 + 1
 
1 1
cos θ = z+ = ,
2 z 2z
z2 − 1
 
1 1
sin θ = z− = ,
2i z 2iz
dz
dθ = .
iz
For example: Compute

1
Z
dθ, (a > 0).
0 a + cos2 θ
With the indicated changes,

1 1 dz
Z Z
dθ =
a + cos2 θ z 2 +1 2 iz

0 |z|=1 a+ 2z
4 z dz
Z
=
i |z|=1 z 4 + (4a + 2)z 2 + 1
The function being integrated is easily seen to have exactly four poles. The poles are the zeroes of the
denominator, which is a quadratic polynomial in z 2 . The zeroes satisfy
p
2 −4a − 2 ± (4a + 2)2 − 4 p
z = = −(2a + 1) ± 2 a2 + a
2
These are two distinct non-zero squares that will yield four distinct square roots. It is always best to leave
all computations until they become necessary or inevitable, in the hope that they won’t have to be done; so
√ √ √ √
we won’t attempt to compute these square roots for now. Let us call them r1 , − r1 , r2 , − r2 , where
p p
r1 = −(2a + 1) − 2 a2 + a, r2 = −(2a + 1) + 2 a2 + a.

The poles that matter are those within the circle of radius 1; r1 < −1 hence ± r1 will be outside of the

circle. On the other hand, −1 < r2 < 0 so that ± r2 have to be considered. By the Theorem of Residues∗
4 z dz 4 z dz
Z Z
4 2
= √ √ √ √
i |z|=1 z + (4a + 2)z + 1 i |z|=1 (z − r1 )(z + r1 )(z − r2 )(z + r2 )

z √
= 8π Res( √ √ √ √ , − r2 )
(z − r1 )(z + r1 )(z − r2 )(z + r2 )

z √
+ Res( √ √ √ √ , r2 )
(z − r1 )(z + r1 )(z − r2 )(z + r2 )
!
z z
= 8π √ √ √ + √ √ √
(z − r1 )(z + r1 )(z − r2 ) z=−√r1 (z − r1 )(z + r1 )(z + r2 ) z=√r1
4π 4π
= √ √ √ √ =
( r2 + r1 )( r2 − r1 ) r2 − r1
π
=√ .
2
a +a
∗ In these notes we are using the notation Res(f (z), z0 ) or Res(f, z0 ) to denote the residue of f at z0 .
Thus

1 π
Z
2
dθ = √ (a > 0).
0 a + cos θ 2
a +a
Some integrals over intervals that are a fraction of an interval of length 2π can be transformed into an
integral over an interval of length 2π by symmetry considerations, and thus integrated by residues. For
example, compute
Z π/2
1

dθ,
0 a + cos
still with a > 0. Because cos is even (and its square is even more even), we have
π/2 π/2
1 1 1
Z Z
dθ = dθ.
0 a + cos2 θ 2 −π/2 a + cos2 θ

The substitution t = π − θ (and the fact that cos(π − t) = − cos t) shows that
π/2 π
1 1
Z Z
dθ = dt,
0 a + cos2 θ π/2 a + cos2 t

thus
π/2 π
1 1 1
Z Z
2
dθ = dt.
0 a + cos θ 2 0 a + cos2 t
Similarly
0 0
1 1 1
Z Z
dθ = dt,
−π/2 a + cos2 θ 2 −π a + cos2 t
so that finally
π/2 π
1 1 1 π
Z Z
dθ = dθ = √ .
0 a + cos2 θ 4 −π a + cos2 θ 4 a2 + a
2. Integrals of the form Z ∞
f (x) dx
−∞

where f is a function that can be extended as a meromorphic function to either the upper half plane or to
the lower half plane and is analytic at all points of the real line (though first order poles on the real line can
be allowed). Moreover, and that is important, the extension to the upper half plane (resp. the extension
to the lower half plane) should verify that |f (z)| → 0 as |z| → ∞ at a sufficient speed. What constitutes
sufficient speed depends a bit on the function, and circumstances but it basically reduces to: There should
+
exist a sequence {Rn }, 0 < R1 < R2 < . . . and limn→∞ Rn = ∞, such that if CR denotes the upper half
circle of radius R, parameterized by
+
CR (t) = Reit , 0 ≤ t ≤ π,

then if we work with an extension to the upper half plane we should have
Z
lim f (z) dz = 0.
n→∞ +
CR n

In the most common cases one has


Z
(1) lim f (z) dz = 0.
R→∞ +
CR

One can then integrate over the path γR consisting of the line segment from −R to R followed by the half
+
circle CR . As R increases more and more of the poles of f are in the set bounded by γR . In the case of
a finite number of poles one can start already with R large enough so all poles are included. If there is an
infinite number of poles, one will have to go to a sequence Rn → ∞ so that the half-circles avoid the poles.
In both cases Z ∞ Z X
f (x) dx = lim f (z) dz = 2πi Res(f, z).
−∞ R→∞ γR ℑz>0

A similar situation is encountered if we opt (or are forced to opt) for an extension into the lower half plane,
+ −
with CR being replaced by CR , the lower half of the unit circle; parameterized by

CR (t) = Re−it , 0 ≤ t ≤ π,

One thing to notice is that now the segment from −R to R followed by CR is negatively oriented, thus

(assuming the integral of f over CR goes to 0)
Z ∞ X
f (x) dx = −2πi Res(f, z).
−∞ ℑz>0

Typical situations are f (z) = P (z)/Q(z) where P, Q are polynomials and the degree of Q minus the degree
of P is at least 2, or f (z) = (P (z)/Q(z))eaz where a ∈ R, P, Q are polynomials and the degree of Q strictly
larger than the degree of P . In either case Q should have no real zeros; however, there is an interpretation of
the integral if Q has at most simple real zeros.
If P, Q are polynomials of degrees m, n respectively, and n > m, then (as is quite easy to prove) there is
R0 > 0, M > 0 such that for |z| > R0 we have

P (z) m−n
(2) Q(z) ≤ M |z|
.

Thus Z
P (z) +
dz ≤ M Rm−n Λ(CR ) = πM Rm−n+1

CR+ Q(z)

for R > R0 . If m − n + 1 < 0; i.e., if n ≥ m + 2, then this proves that (1) holds. We thus have: Assume P, Q
are polynomials of degrees m, n respectively; assume n ≥ m + 2 and Q has no real zeros. Then
Z ∞
P (x) X P
dx = 2πi Res( , z).
−∞ Q(x) Q
ℑz>0

Here are a few examples.


Example 1. Compute

dx
Z
.
−∞ x4 + 1
The integrand has four simple poles, at the points

eiπ/4 , e3iπ/4 , −eiπ/4 , −e3iπ/4 .

The first two are in the upper half-plane. Thus


Z ∞  
dx 1 iπ/4 1 3iπ/4
4
= 2πi Res( 4 ,e ) + Res( 4 ,e )
−∞ x + 1 z +1 z +1
 
1 1
= 2πi + 2
(z + eiπ/4 )(z 2 + i) z=eiπ/4 (z − i)(z + e3iπ/4 ) z=e3iπ/4
 
1 1
= 2πi iπ/4
− 3iπ/4
4ie 4ie
π −iπ/4 −3π/4 π
= (e −e )= √ .
2 2
Example 2. Compute

dx
Z
.
−∞ (x2 + 1)2
This time the integrand has double poles; a double pole at i in the upper half-plane, one at −i in the lower
half-plane. We have
1 d (z − i)2
Res( , i) = lim
(x2 + 1)2 z→i dz (z 2 + 1)2

d 1
= lim
z→i dz (z + i)2
1
= lim −2(z + i)−3 = − i
z→i 4

Thus ∞
dx 1 π
Z
= (2πi)(− i) = .
−∞ (x2 + 1)2 4 2
We turn now to something a bit more complicated and evaluate
Z ∞
eiax
2
dx
−∞ x + 1

where a ∈ R. We will assume a 6= 0. The integrand has simple poles at ±i and we could suspect that
Z ∞
eiax eiaz
2
dx = 2πiRes( , i),
−∞ x + 1 z2 + 1
but we have to be careful. We need to have
eiaz
lim dz = 0.
+
CR z2 +1

In the upper half-plane z = z + iy where y > 0 and then |eiaz | = e−ay . If a > 0, then we can estimate
+
|eiaz | = e−ay ≤ 1 and the limit of the integral over CR is 0 for the same reason that (1) holds. But if a < 0,
+
then |eiaz | is pretty large and there is little if any hope that the integral over CR will go to 0 and, in fact, it

doesn’t. But now the integral over CR will go to 0. We will have

eiaz

Z ∞
e iax  2πiRes(
 , i), a>0
2
z +1
dx = iaz
 −2πiRes( e
2
−∞ x + 1 
, −i), a < 0.
z2 + 1
We have
eiaz eiaz e−a

Res( 2 , i) = =
z +1 z + i z=i 2i
eiaz iaz
ea

e
Res( , −i) = = −
z2 + 1 z − i z=−i 2i
thus ∞
eiax
Z
dx = πe−|a| .
−∞ x2 + 1
We assumed a 6= 0. This was not quite necessary; if a = 0 one can use either half-plane to get that the
integral evaluates to π. Or one can let a → 0 in the current computation. Notice also that the case a < 0
can be derived from the case a > 0 by the following trick. Suppose we have verified that
Z ∞
eiax
2
dx = πe−a
−∞ x + 1

when a > 0. Assume now a < 0. Then −a > 0 and


Z ∞ Z ∞ −iax
eiax e
2+1
dx = 2+1
dx = πe−(−a) = πea .
−∞ x −∞ x
+
We saw that if a > 0, then the integral over CR of eiaz P (z)/Q(z) goes to 0 as R → ∞, assuming that the
degree of Q is greater than or equal to the degree of P plus 2. It is interesting that the same result is also
true also if the degree of Q is just greater than the degree of P plus 1. That is known as Jordan’s lemma.

Lemma 1 Let f be continuous for |z| > R0 , some R0 > 0, and assume there exists a constant M such that
|f (z)| ≤ M |z|−1 for |z| > R0 . Then, if a > 0,
Z
lim f (z)eiaz dz = 0.
R→∞ +
CR


A similar result is, of course, true for CR , with eiaz replaced by e−iaz .
Proof. We simply have to bound |eiaz | for z = x + iy, y > 0, |z| = R a bit better than before, where we
merely used |eiaz | ≤ 1. If |z| = R, ℑz > 0, we can write z = Reiθ , with 0 < θ < π/2 and

|eiaz | = e−aR sin θ .

Let ǫ > 0 be given, ǫ < π. For ǫ < θ < π − ǫ, sin θ > sin ǫ > 0. Letting δ = a sin ǫ. Then δ > 0 and

e−aR sin θ < e−δR for ǫ < θ < π − ǫ.

For 0 ≤ θ ≤ ǫ and for π − ǫ ≤ θ ≤ π, we estimate e−aR sin θ ≤ 1. Thus


Z
M M π −aR sin θ
Z Z Z
iaz
f (z)e dz ≤ |f (z)||eiaz | |dz| ≤ |eiaz ||dz| = e 6 R dθ

R CR+ 6R 0

CR+ +
CR
Z ǫ Z π−ǫ Z π 
= M e−aR sin θ dθ + e−aR sin θ dθ + e−aR sin θ dθ
0 π π−ǫ
Z ǫ Z π−ǫ Z π 
≤ M dθ + e−δR dθ + dθ
0 π π−ǫ
= M (2ǫ + e−δR ).

Since e−δR → 0 as R → ∞ and ǫ > 0 is arbitrary, the result follows.


Generally speaking, if the degree of Q is just one above the degree of P then (P/Q)eiax will not be integrable
over R, but the integral will exist as an improper integral. As a way of going into improper integrals, let us
compute another integral. Compute Z ∞
sin πx
dx.
−∞ 1 + x3
One could try to apply the same method as above, extending the integrand to the upper half-plane in the
obvious way, as z 7→ sin z/(1 + z 3 ). This won’t work. The problem is that sin, a mild mannered, modest,
bounded function as long as it is restricted to the real line, becomes a monster once it is allowed to roam over
the whole complex plane. For z = iy, y ∈ R, it morphs into sinh y (up to a factor of i) a function that grows
exponentially as |y| → ∞. The rule for integrals in which one has a rational function P/Q, where P, Q are
polynomials with real coefficients, times a sine or a cosine is
Z ∞ Z ∞
P (x)eiax

P (x) cos ax
dx = ℜ dx ,
−∞ Q(x) −∞ Q(x)
Z ∞ Z ∞
P (x)eiax

P (x) sin ax
dx = ℑ dx .
−∞ Q(x) −∞ Q(x)

Thus, to deal with our integral, we have to use


Z ∞ Z ∞ iπx 
sin πx e
3
dx = ℑ 3
dx .
−∞ 1 + x −∞ 1 + x

But a new problem has appeared; now the integrand has a pole on the real axis. Because the pole is simple
one can deal with it; it turns out that the integral exists in what is called a principal value sense. To deal
with the pole we will integrate over a path having the shape depicted in the picture below. It consists of a
line segment starting at −R and going all the way to −1 − ǫ, so we stop before reaching the pole that here
happens to be at -1. We follow with a half-circle of radius ǫ in the upper half-plane, avoiding the pole. This
takes us to −1 + ǫ. We then continue on the real axis to R and follow with the half-circle of radius R. The
path can be described as γR,ǫ = γ1 + γ2 + γ3 + γ4 , where

γ1 (t) = t, −R ≤ t ≤ −1 − ǫ,
γ2 (t) = −1 − ǫe−it , 0 ≤ t ≤ π,
γ3 (t) = t, −1 + ǫ ≤ t ≤ R,
γ4 (t) = Reit , 0 ≤ t ≤ π.

We take R large enough and ǫ small enough so that the pole at (1 + i 3)/2 is included in γR,ǫ .

By the theorem of residues, √


eiπz eiπz 1 3i
Z
dz = 2πiRes( , + )
γR 1 + z3 1 + z3 2 2
My calculations, which could be wrong!, give
√ √
eiπz 1 3i 2ie−π 3/2
Res( , + )= √ ;
1 + z3 2 2 3(−3 + i 3)

thus √ √ √
eiπz 4πe−π 3/2 πe−π 3/2
(3 + i 3)
Z
3
dz = √ =
γR 1+z 3(3 − i 3) 9
We now let R → ∞ and ǫ → 0. By Jordan’s Lemma, or the earlier estimates, the integral over γ4 goes to
0. However, the integral over γ2 does not usually go to 0 as ǫ → 0. It being an integral over a negatively
oriented half-circle, it will go to −πi times the residue. That is, we have:

Lemma 2 Assume f is analytic in D′ (a, r) and has a simple pole at a. Then, if γ1,ǫ , γ2,ǫ are the paths given
by
γ1,ǫ (t) = a + teit , γ1,ǫ (t) = a − te−it , 0 ≤ t ≤ π,
then Z Z
lim f (z) dz = πiRes(f, a), lim f (z) dz = −πiRes(f, a).
ǫ→0+ γ1,ǫ ǫ→0+ γ2,ǫ
Proof. Since the integral over the full circle is 2πiRes(f, a), it suffices to prove that one of the two limits
works out. We have
Z Z π
lim f (z) dz = lim f (a + ǫeit )ǫieit dt
ǫ→0+ γ1,ǫ ǫ→0 0
Z π
lim f (a + ǫeit )ǫeit dt = πiRes(f, a),

=i
0 ǫ→0

where we used that


lim f (a + ǫeit )ǫeit = lim (z − a)f (z) = Res(f, a).

ǫ→0 z→a

Returning to our integral we thus have proved


−1−ǫ ∞
√ √
eiπx eiπx eiπz πe−π 3/2 (3 + i 3)
Z Z 
lim dx + dx − iπRes( 3 , −1) = ;
ǫ→0 −∞ x3 + 1 −1+ǫ x3 + 1 z +1 9

i.e., considering that


eiπz 1
3
Res( , −1) = − ,
z +1 3
−1−ǫ iπx Z ∞ iπx
√ √
πe−π 3/2 (3 + i 3) πi
Z 
e e
lim dx + dx = − .
ǫ→0 −∞ x3 + 1 3
−1+ǫ x + 1 9 3
The limit on the left hand side is not exactly the integral from −∞ to ∞, but what is called the principal
value. We digress here a bit on p.v.’s. Suppose f is defined at all points of an interval (a, b), except perhaps
at c and that f is integrable on (a, c − ǫ), (c + ǫ, b) for all ǫ > 0. Without going too much into integration
theory, let us just say that f is considered integrable on (a, b) if and only if
!
Z b Z Z c−ǫ b
|f (x)| dx = lim |f (x)| dx + |f (x)| dx < ∞.
a ǫ→0+ a c+ǫ

In this case both limits Z c−ǫ Z b


lim f (x) dx, lim f (x) dx
ǫ→0+ a ǫ→0+ c+ǫ

exist separately and one writes (or has)


!
Z b Z c−ǫ Z b
f (x) dx = lim f (x) dx + f (x) dx .
a ǫ→0+ a c+ǫ

But it is possible that the limit in the last expression exists without the function being integrable. An obvious
example is f (x) = 1/x in the interval (−1, 1). One has
Z −ǫ 1 
1 1 1
Z
lim dx + dx = lim 2 log = ∞
ǫ→0+ −1 |x| ǫ |x| ǫ→0 ǫ

but, because x 7→ 1/x is odd,


Z −ǫ 1 
1 1
Z
lim dx + dx = 0.
ǫ→0+ −1 x ǫ x
We write
1
1
Z
p.v. dx = 0.
−1 x
The existence of this limit is due to compensation of positive and negative values of the integrand; it is
something like a conditional integral.
One has a similar situation at ∞ and at −∞. Let a ∈ R and suppose f is integrable in (a, b) for all b > a. If
Z b
lim |f (x)| dx < ∞,
b→∞ a

then we say f is integrable on (a, ∞) and define


Z ∞ Z b
f (x) dx = lim f (x) dx,
a b→∞ a

the last limit being guaranteed to exist. But due to compensations, this limit may exist even if the function
is not integrable; then the limit is once more called the principal value integral of f . A similar consideration
R∞ R∞ R0
is valid at −∞. Concerning an integral of the form −∞ , one defines it as the sum of 0 plus −∞ ; the
function being integrated has to be integrable over both (0, ∞) and (−∞, 0) to be considered integrable over
(−∞, ∞). Before we return to our integral we state a Lemma; its proof should be immediate to anybody who
followed all the calculations we did with the integral we have not yet quite finished to compute.

Lemma 3 Assume P, Q are polynomials of degrees m, n respectively and n ≥ m + 1. Assume a ∈ R and


a > 0. Then
Z ∞
P (x)eiax X X
p.v. dx = 2πi Res(P (z)eiaz /Q(z), z) + πi Res(P (x)eiax /Q(x), x).
−∞ Q(x)
ℑz>0 x∈R

If Q(x) 6= 0 for x ∈ R and if n ≥ m + 2, then the integral exists in the normal (Lebesgue) sense.

Back to our integral. We now just take imaginary parts; the imaginary part of the integrand has a removable
singularity at −1, so the integral exists without need of bringing in principal values. The end result is


sin πx π(e−π 3/2 − 3)
Z
3
dx = .
−∞ 1+x 9

For a final example of this type of integral, let us calculate


Z ∞
sin x
dx.
0 x
This integral will only exist as a principal value integral. Moreover, it is from 0 to ∞, not from −∞ to ∞. If
f is an even function, then one can calculate
Z ∞
1 ∞
Z
f (x) dx = f (x) dx.
0 2 −∞

If f is not even, then one is out of luck. But x 7→ sin x/x is even, and we can calculate (using Lemma 3 and
the fact that eiz /z has no poles away from the real axis)
Z ∞  Z ∞ ix 
sin x 1 e
p.v. dx = ℑ p.v. dx
0 x 2 −∞ x
eiz
 
1 1
= ℑ πiRes( , 0) = ℑ(πi)
2 z 2
π
= .
2

3. Integrals of the form Z ∞


f (x)R(x) dx
0
where R is a rational function and f is a function that extends analytically to all the plane minus a line; a
function having a branch point at 0, such as xb with b ∈
/ Z or such as log x.
Example: Compute


Z
dx,
0 1 + x2
where 0 < α < 1. One approach is to work with a determination of log analytic except on the line of
integration; in this case in {reiθ : r > 0, 0 < θ < 2π}. In other words,

log z = log |z| + iθ, 0 < θ < 2π.

We select ǫ, R, 0 < ǫ < R and let γǫ,R be the path of the picture.

The little half-circle about the origin has radius ǫ, the big almost full circle has radius R. If we do the obvious
parameterizations we have by the Theorem of residues, assuming R > 1 so that the closed path contains the
poles of the integrand:
R
(t + iǫ)α zα
Z Z
dt + dz
0 1 + (t + iǫ)2 γR,ǫ 1 + z2
R α
(t − iǫ) zα
Z Z
− 2
dt + 2
dz
0 1 + (t − iǫ) δǫ 1 + z
zα zα
 
= 2πi Res( , i) + Res( , −i) .
1 + z2 1 + z2

Here γR,ǫ is the arc of circle joining in a counterclockwise direction R + iǫ to R − iǫ; δe the half-circle going
clockwise from −ǫi to ǫi. We let ǫ → 0 and R → ∞. Both the integral over γR,ǫ and the integral over δǫ are
easily seen to go to 0. Concerning the two other integrals we have if z = |z|eiθ with 0 < θ < 2π, then

z α = eα log z = |z|α eiθ/2 .

The main thing to realize is that for t > 0,

lim (t + iǫ)α = tα , lim (t − iǫ)α = tα e2πiα .


ǫ→0+ ǫ→0+

Thus !
R R R
(t + iǫ)α (t − iǫ)α tα
Z Z Z
lim dt − dt = (1 − e2πiα ) dt.
ǫ→0+ 0 1 + (t + iǫ)2 0 1 + (t + iǫ)2 0 1 + t2
Thus, considering that
1 3
iα = e 2 πiα , (−i)α = e 2 πiα ,
we get for R > 1 (after letting ǫ → 0+)
Z R α
zα zα
 
t
(1 − e2πiα ) dt = 2πi Res( , i) + Res( , −i)
0 1 + t2 1 + z2 1 + z2
!
z α z α

= 2πi +
z + i z=i z − i z=−i
1 3
!
e 2 πiα e 2 πiα
= 2πi +
2i (−2i)
 1 3
 1
= π e 2 πiα − e 2 πiα = πe 2 πiα (1 − eπiα ).

Dividing by 1 − e2πiα and letting R → ∞, we get


Z ∞
tα 1 1 − eπiα
dt = πe 2 πiα .
0 1 + t2 1 − e2πiα
The right hand side does not look very real, yet it should be real. To see it is, we can work on it a bit.
1 1 1 1 1
1 1 − eπiα 1 e 2 πiα (e− 2 πiα − e 2 πiα ) e− 2 πiα − e 2 πiα sin πα π
πe 2 πiα = πe 2 πiα = π = π 2
= .
1 − e2πiα eπiα (e−iπiα − eπiα ) e−iπiα − eπiα sin πα 2 cos πα
2

That is,

xα π
Z
dx = .
0 1 + x2 2 cos πα
2
There is another way of computing integrals such as this one in which we use the principal determination of
the logarithm: log z = ln |z| + i arg z and | arg z| < π. We integrate over the path γ of the picture;

γ4

γ2
γ1 γ3
−R −ǫ ǫ R
We have γ = γ1 + γ2 + γ3 + γ4 , with γ1 th line segment from −R to −ǫ, γ2 the half circle of radius ǫ around
the origin, γ3 the line segment from ǫ to R and finally γ4 the half circle of radius R around 0. This method
is based on the fact that with the determination of the logarithm mentioned above, if t is a negative real
number, then arg t = π and
tα = eα log t = eα(ln |t|+iπ) = |t|α eπiα
so that
−ǫ −ǫ R
zα tα |t|α tα
Z Z Z Z
dz = dt = eπiα dt == eπiα dt,
γ1 1 + z2 −R 1 + t2 −R 1 + t2 ǫ 1 + t2
so that on the one hand
Z R
zα tα zα zα
Z Z Z
πiα
2
dz = (1 + e ) 2
dt + 2
dz + 2
dz.
γ 1+z 0 1+t γ2 1 + z γ4 1 + z
on the other hand, if 0 < ǫ| < 1 < R, the theorem of residues implies
1

 α
e 2 πiα

z
Z
1

2
dz = 2πiRes 2
, i = 2πi = πe 2 πiα .
γ 1 + z 1 + z 2i
Equating both expressions and letting ǫ → 0, R → 0; verifying that then the integrals over γ2 , γ4 go to 0, we
get
Z ∞ 1
xα e 2 πiα 1 π
dx = π = π − 1 πiα = .
0 1+x 2 1+e πiα
e 2
1
+ e2 πiα 2 cos πα
2
This second method seems faster. Probably, because it is faster.
There are, of course, zillions of other integrals that can be computed by residues. The thing is finding an
appropriate contour. That is not always easy.
Many Fourier transforms are computed in closed form using residues. A typical example is the following, based
on the fact that Z ∞
2 √
e−x dx = π.
−∞
Recall first that if f is integrable over the whole real line then the Fourier transform of f is the function
fˆ : R → R defined by Z ∞
1
fˆ(ξ) = √ e−ixξ f (x) dx
2π −∞

for ξ ∈ R. (Definitions vary slightly; some people omit the factor of 1/ 2π, others use e−2πixξ instead of e−ixξ .)
2
Suppose f (x) = e−ax , where a > 0. Compute fˆ.
We have Z ∞ Z ∞
1 2 1 2
fˆ(ξ) = √ e−ixξ e−ax dx = √ e−ax −ixξ dx.
2π −∞ 2π −∞
We can complete the square in the exponent,
ξ2 ξ2
 
ixξ iξ iξ
−ax2 − ixξ = −a(x2 + ) = −a (x + )2 + 2 = −a(x + )2 − ,
a 2a 4a 2a 4a
hence 2 ∞
e−ξ /4a
Z
iξ 2
fˆ(ξ) = √ e−a(x+ 2a ) dx.
2π −∞

In all of this, ξ remains fixed. For R > 0 let γR be the path defined by γR (t) = t + (iξ/2a), −R ≤ t ≤ R;. Then
2
e−ξ /4a
Z
2
(3) fˆ(ξ) = √ lim e−az dz.
2π R→∞ γR
To compute this complex integral, consider the rectangular path ΓR in which we follow γR by the line segment
joining the right endpoint R + (iξ/2a) of γR to R, continue along the line segment from R to −R, finally add the
line segment from −R to −R + (iξ/2a). If ξ > 0 (so that the path γR is in the upper half-plane), then a picture
would look as follows.

−R + (iξ/2a) R + (iξ/2a)

−R R
Since the integrand is analytic everywhere,
Z
2
(4) e−az dz = 0.
ΓR

In our notation,
ΓR = γR + λR+(iξ/2a),R + λR,−R + λ−R,−R+(iξ/2a)
and we see that !
Z Z ξ/2a
−az 2 −aR2 at2 +2iRt
lim e dz = lim −ie e dt = 0.
R→∞ λR+(iξ/2a),R R→∞ 0

In case there is any question, notice that


Z Z
ξ/2a 2 ξ/2a Z ξ/2a
at +2iRt at2 +2iRt 2
e dt ≤ |e | dt = eat dt


0 0 0

which is a finite number, independent of R. Here we assumed implicitly that ξ > 0; if ξ < 0 one should have the
last two integrals going from ξ/2a to 0. If ξ = 0 there is nothing that needs to be done.
Similarly one sees that Z
2
lim e−az dz = 0.
R→∞ λ−R,−R+(iξ/2a)

Since Z Z R
−az 2 2
e dz = − e−at dt
λR,−R −R

and, by the substitution x = a t, √
R aR
1
Z Z
−at2 2
e dt = √ √
e−x dx,
−R a − aR

letting R → ∞ in (4) gives



1
Z Z
2 2
0 = lim e−az dz − √ e−x dx;
R→∞ γR a −∞

that is, √

1 π
Z Z
2
−x2
lim e−az dz = √ e dx = √ .
R→∞ γR a −∞ a
Using this in (3),
2 √
ˆ e−ξ /4a π 1 2
f (ξ) = √ √ = √ e−ξ /4a .
2π a 2a
In particular, the function
2
x 7→ e−x /2

has the interesting property of being a non-zero function that is its own Fourier transform (case a = 1/2).

Das könnte Ihnen auch gefallen