Sei sulla pagina 1di 14

COMPLEX ANALYSIS: SOLUTIONS 5

1. Find the poles and residues of the following functions


1 1 π cot(πz) 1
, , , (m, n ∈ Z>0 )
z 4 + 5z 2 + 6 (z 2 − 1)2 z2 z m (1 − z)n

Solution: Throughout we use the following formula for calculating residues: If f (z)
has a pole of order k at z = z0 then
1 dk−1  k

res(f, z0 ) = (z − z ) f (z) .

k−1 0
(k − 1)! dz

z=z0

In particular, if f (z) has a simple pole at z0 then the residue is given by simply
evaluating the non-polar part: (z − z0 )f (z), at z = z0 (or by taking a limit if we have
an indeterminate form).

Let
1 1 1
f (z) := = 2 = √ √ √ √ .
z4 2
+ 5z + 6 2
(z + 2)(z + 3) (z + i 2)(z − i 2)(z + i 3)(z − i 3)
√ √
This has simple poles at z = ±i 2, ±i 3 with residues

√ √ 1 1 1
res(f, i 2) = (z − i 2) = √ = √ ,

√ √
z 4 + 5z 2 + 6 z=i 2 (z + i 2)(z 2 + 3) z=i 2 2i 2

√ √ 1 1
res(f, −i 2) = (z + i 2) √ = − √ ,

4 2
z + 5z + 6 z=−i 2

2i 2

√ 1
res(f, i 3) = − √ ,
2i 3

√ 1
res(f, −i 3) = √ .
2i 3
1
COMPLEX ANALYSIS: SOLUTIONS 5 2

For the second one let


1 1
f (z) = = .
(z 2 − 1)2 (z + 1) (z − 1)2
2

This has double poles at ±1. From the formula we get


d 1
res(f, 1) = = −1/4,

dz (z + 1)2 z=1

d 1
res(f, −1) = = 1/4.

2
dz (z − 1) z=−1

For the third let


π cot(πz)
f (z) = .
z2
Now, cot(πz) has poles wherever sin(πz) = 0, so at z = n ∈ Z. About these points
we have
(z − n)2 (z − n)3
sin(πz) = sin(πn) + π cos(πn)(z − n) − π 2 sin(πn) − π 3 cos(πn) + ···
2! 3!
2
n 2 (z − n)
+ O((z − n)4 )
 
=(−1) π(z − n) 1 − π
3!
and
(z − n)2 (z − n)3
cos(πz) = cos(πn) − π sin(πn)(z − n) − π 2 cos(πn) + π 3 sin(πn) + ···
2! 3!
(z − n)2
=(−1)n 1 − π 2 + O((z − n)4 ) .
 
2!
Hence, for z close to n ∈ Z, we have
1 − π 2 (z − n)2 /2 + O((z − n)4 )
cot(πz) =  
π(z − n) 1 − π 2 (z − n)2 /6 + O((z − n)4 )
1 − π 2 (z − n)2 /2 + O((z − n)4 ) 
1 + π 2 (z − n)2 /6 + O((z − n)4 )

=
π(z − n)
1 π
= − (z − n) + O((z − n)3 )
π(z − n) 3
Therefore, f (z) = π cot(πz)/z 2 has simple poles at z = n 6= 0 and a triple pole at
z = 0. For the simple poles we have

π cot(πz) 1
res(f, n) = (z − n) 2
= 2.
z
z=n n
COMPLEX ANALYSIS: SOLUTIONS 5 3

For the triple pole at at z = 0 we have


1 π2 1
f (z) = − + O(z)
z3 3 z
so the residue is −π 2 /3.

Finally, the function


1
f (z) =
z m (1
− z)n
has a pole of order m at z = 0 and a pole of order n at z = 1. From the formula for
residues we have
dm−1
 
1 1
res(f, 0) = m−1 n

(m − 1)! dz (1 − z) z=0
n(n + 1) · · · (n + m − 2)
=
(m − 1)!
(n + m − 2)!
=
(n − 1)!(m − 1)!
and
dn−1 (−1)n
 
1
res(f, 1) =
(n − 1)! dz n−1 zm
z=1
−m(m + 1) · · · (m + n − 2)
=
(n − 1)!
(m + n − 2)!
=−
(m − 1)!(n − 1)!
= − res(f, 0).

2. Use the substitution eiθ = z along with the residue theorem to show that
Z 2π
dθ 2π
=√ .
0 2 + cos θ 3

Solution: As suggested we let eiθ = z so that dθ = dz/(iz) and the integral becomes
Z Z
1 dz 2 dz
−1
= 2
.
i |z|=1 z(2 + (z + z )/2) i |z|=1 z + 4z + 1

Now z 2 + 4z + 1 has zeros of order 1 at z = z± = −2 ± 3 and so the integrand has
simple poles at z+ and z− . Only z+ lies in the unit disk and therefore by the residue
COMPLEX ANALYSIS: SOLUTIONS 5 4

theorem
Z
2 dz 2  1 
= × 2πi × res , z +
i |z|=1 z 2 + 4z + 1 i z 2 + 4z + 1

1
=4π(z − z+ ) 2
z + 4z + 1 z=z+

1
=4π(z − z+ )
(z − z+ )(z − z− ) z=z+
1
=4π
z+ − z−
1
=4π √
2 3

=√ .
3
3. Evaluate the following integrals via residues. Show all estimates.
(i)
Z ∞
x2
dx
0 x4 + 5x2 + 6
(ii)
Z ∞
x sin x
dx, a real
0 x 2 + a2
(iii)
Z ∞
log x
dx
0 1 + x2

Solution: (i) Since the integrand is an even function the integral in question is equal
to I/2 where
Z ∞
x2
I= 4 2
dx.
−∞ x + 5x + 6
√ √
As a function of a complex variable, the integrand has simple poles at ±i 2, ±i 3.
We will be considering a semicircular
√ √ contour in the upper half plane so we only need
calculate the residues at z = i 2, i 3. A slight modification of the first calculation
in question 1 gives
√ √
z2 √ (i 2)2
 
i 2
res 4 ,i 2 = √ =
z + 5z 2 + 6 2i 2 2
COMPLEX ANALYSIS: SOLUTIONS 5 5

and √ √
z2 √ (i 3)2
 
i 3
res 4 ,i 3 = − √ = − .
z + 5z 2 + 6 2i 3 2
Now, Consider the semicircular contour ΓR , which starts at R, traces a semicircle
in the upper half plane to −R and then travels back to R along the real axis. Then,
on taking R large enough, by the residue theorem
√ √
Z
z2 X i 2 i 3 √ √
4 2
dz = 2πi× residues inside ΓR = 2πi( − ) = π( 3− 2).
ΓR z + 5z + 6 2 2
On the other hand
z2 z2
Z Z
lim dz = I + lim dz
R→∞ ΓR z 4 + 5z 2 + 6 R→∞ γR z 4 + 5z 2 + 6
where γR is the semicircle in the upper half plane. But by the Estimation Lemma
Z z2 z2 1
dz 6 πR max 4  →0

4 2 2
γR z + 5z + 6 R z + 5z + 6 R

z∈γ
√ √
as R → ∞. Hence, I = π( 3 − 2) and so
π √ √
Z ∞
x2
dx = ( 3 − 2).
0 x4 + 5x2 + 6 2
The last estimate in the inequality for the integral over γR should be clear since the
dominant term is the z 4 term in the denominator, but for completeness:
z2 π 1
πR max 4 = max

2
z∈γR z + 5z + 6 −2
R z∈γR 1 + 5z + 6z −4

π 1
6 max
R z∈γR 1 − 5|z| − 6|z|−4
−2

π 1
=
R 1 − 5R − 6R−4
−2

where we have used |z + w| > |z| − |w| in the second line. On taking R > 5, say, this
is 6 2π/R and the constant 2π is absorbed into the  symbol.

(ii). We have
Z ∞ ∞ Z ∞
xeix
Z 
x sin x 1 x sin x 1
dx = dx = = dx .
0 x 2 + a2 2 −∞ x 2 + a2 2 2
−∞ x + a
2

Denote this last integral by J. Again, we will consider J as the horizontal section of
the contour ΓR from part (i).
COMPLEX ANALYSIS: SOLUTIONS 5 6

In the upper half plane the integrand has a simple pole at z = ia with residue
 zeiz zeiz iae−a e−a

res 2 , ia = (z − ia) = = .
z + a2 z 2 + a2 z=ia 2ia 2
Hence, by the residue theorem
zeiz zeiz
Z Z
−a
πie = lim dz = J + lim dz.
R→∞ ΓR z 2 + a2 R→∞ γR z 2 + a2
Thus it remains to show that this last integral vanishes in the limit. This is similar
to question 7 (ii) of Problems 3; a trivial estimate of the integrand is  1/R which is
not enough for the Estimation Lemma. Instead we apply integration by parts which
is probably the quickest way (see Problems 3). Integrating eiz and differentiating the
rest gives
−R
zeiz zeiz 2z 2
Z Z  
1 1
2 + a2
dz = 2 + a2 )
− 2 + a2
− 2 2 )2
eiz dz.
γR z i(z R i γR z (z + a
The first term on the right is −2R cos R/i(R2 + a2 )  1/R. For the integrals we use
the Estimation Lemma to give
eiz eiz
Z
1 1

2 2
dz 6 πR max 2

2
6 πR
2 2
 ,

γR z + a z∈γR z + a R −a R

z 2 eiz z 2 eiz
Z

6 πR3 1 1
dz 6 πR max  ,

γR (z 2 + a2 )2 z∈γR (z 2 + a2 )2
(R2 − a2 )2 R
as R → ∞. Hence, J = πie−a and so
Z ∞
x sin x 1 π
2 2
dx = =(J) = e−a .
0 x +a 2 2

(iii) This is quite hard and, as I discovered recently, the solution is in Conway’s
book anyway (pg. 117–118). My bad. It’s still instructive to attempt this before
reading Conway though.
We see that, as a function of a complex variable, the integrand has a branch cut
and simple poles at z = i, −i. Taking the branch of the log with −π < arg(z) < π,
we would like to choose a contour which lies just above and below the cut and that
also picks up the residues at i, −i. A natural choice is the contour we used in the
lectures. This consisted of a small circle about z = 0, horizontal lines just above and
below the negative real axis, and a large circle completing the contour.
COMPLEX ANALYSIS: SOLUTIONS 5 7

Unfortunately, with this choice the integrals over the horizontal lines L1 , L2 are
given by
Z Z
log z log z
2
dz + 2
dz
L1 1 + z L2 1 + z
Z 0 Z −∞
log(x + iδ) log(x − iδ)
= 2
dx + dx
−∞ 1 + (x + iδ) 0 1 + (x − iδ)2
Z 0 Z −∞
log |x| + πi log |x| − πi
→ 2
dx + dx
−∞ 1+x 0 1 + x2
Z 0
1
=2πi 2
dx
−∞ 1 + x

and the integral we’re after has disappeared.


This motivates the choice of a new contour: we want something with a horizontal
line over the whole real axis, since then the integral over this line is given by
Z ∞ Z ∞
log |x| log x
2
dx = 2 dx
−∞ 1 + x 0 1 + x2
and we avoid the problem of cancellation. Consequently, we choose a branch of log z
with a branch cut along the negative imaginary axis. To avoid z = 0 and the branch
cut we indent our contour with a small circle in the upper half plane. We complete
the contour with a large circle.
Thus, let Γr,R be the contour consisting of a line from r to R, a semicircle in the
upper half plane traced from R to −R, a line from −R to −r, and a semicircle in
the upper half plane traced from −r to r. Follwing the details in Conway we find
Z ∞
log x
dx = 0.
0 1 + x2

4. Let ΓN be the square which crosses the real axis at ±(N + 1/2) with N ∈ N.
(i) Show that cot(πz) is bounded on ΓN and hence show that
Z
π cot(πz)
lim dz = 0.
N →∞ Γ
N
z2
(ii) For a given N compute the above integral via residues. Conclude something
interesting.

Solution: (i) We have


eπiz + e−πiz e2πiz + 1 e2πix e−2πy + 1
cot(πz) = i πiz = i 2πiz = i 2πix −2πy .
e − e−πiz e −1 e e −1
COMPLEX ANALYSIS: SOLUTIONS 5 8

Thus, if y > 1
1 + e−2πy 1 + e−2π
| cot(πz)| 6 6 .
1 − e−2πy 1 − e−2π
Since cot(πz) = cot(πz) we obtain the same bound for y 6 −1. It remains to show
that cot(πz) is bounded for z = ±(N + 1/2) + iy with |y| < 1. But in this case
±2πi(N +1/2) −2πy
| − e−2πy + 1| e2π − 1

e e + 1
| cot(πz)| = ±2πi(N +1/2) −2πy = < =: C.
e e − 1 | − e−2πy − 1| e−2π + 1
Since C is bigger than the bound for |y| > 1, we may say that | cot(πz)| < C on all
of ΓN .

(ii) By the estimation lemma


Z
π cot(πz) cot(πz) C 8πC

2
dz 6 8π(N +1/2) max
2
< 8π(N +1/2)
2
=

ΓN z z∈ΓN z (N + 1/2) N + 1/2
and this tends to zero as N → ∞.
On the other hand, for a fixed N this integral is given by 2πi × the sum of residues
inside ΓN . Using the results from question 1 then gives
π2 
Z
π cot(πz)  X 1
dz = 2πi − .
ΓN z2 −N 6n6N
n 2 3
n6=0

Therefore, on letting N → ∞ we have



X 1 π2 X 1 π2
0= − = 2 −
−∞6n6∞
n2 3 n=1
n2 3
n6=0

and so

X 1 π2
ζ(2) = = ,
n=1
n2 6
which is interesting.
5. (i) Let f (z) = z 6 + cos z. Find the change in argument of f (z) as z travels once
around the circle of radius 2, center zero, in the positive direction.
(ii) How many solutions does 3ez − z = 0 have in the disk |z| 6 1?
(iii) Use Rouche’s Theorem to prove that a polynomial of degree n has n roots in C.
COMPLEX ANALYSIS: SOLUTIONS 5 9

Solution: (i) By the argument principle, the change in argument of f (z) as z travels
around the circle is equal to (2π ×) the number of zeros minus poles of f (z) inside
the circle, so just the number of zeros then.
To find the number of zeros of f (z) inside the circle we compare f (z) with its
dominant term z 6 and apply Rouche’s Theorem. So let g(z) = z 6 . On the circle we
have
|g(z)| = |z|6 = 26 = 64

and
|eiz | + |e−iz |
|f (z) − g(z)| = | cos(z)| 6 6 e|z| = e2 < 64.
2
Thus, Rouche’s Theorem applies and f (z) has the same number of zeros inside the
circle as z 6 , which is 6. Hence, the change in argument is 2π × 6 = 12π.

(ii). Rephrasing the question we ask for the number of zeros of f (z) = 3ez − z in
the closed unit disk. We apply Rouche’s Theorem again. This time the dominant
term is g(z) = 3ez . On the unit circle we have

|g(z)| = 3e<(z) > 3e−1 > 1

and
|f (z) − g(z)| = |z| = 1.

By Rouche’s Theorem 3ez − z has the same number of zeros in the unit disk as 3ez ,
which is none. So the answer is no solutions.

(iii) In the lectures we showed that the polynomial z 6 + z + 2 has 6 zeros in the
disk |z| 6 2 by comparing it with z 6 and applying Rouche’s Theorem. We generalise
this by comparing a general polynomial

f (z) = an z n + an−1 z n−1 + · · · + a0 , an 6= 0

with its leading term


g(z) = an z n

on arbitrarily large circles. On the circle |z| = R > 1 we have

|g(z)| = |an |Rn


COMPLEX ANALYSIS: SOLUTIONS 5 10

and
|f (z) − g(z)| =|an−1 z n−1 + an−2 z n−2 + · · · + a0 |
6|an−1 |Rn−1 + |an−2 |Rn−2 + · · · + |a0 |
<n max |aj |Rn−1
06j6n−1
 n maxn−1 |aj | 
j=0
= |an |Rn
|an |R
Therefore, on a circle of radius R > n maxn−1
j=0 |aj |/|an | we have |f (z) − g(z)| < |g(z)|.
By Rouche’s Theorem f (z) has the same number of zeros inside the circle as g(z),
which is n.

6. Prove that the function


X 1
f (z) =
n∈Z
(n + z)2
is meromorphic on C.

Solution: Clearly, f (z) has poles only at the integers so we need to show that f (z) is
analytic on C\Z. By the results given in the lectures, this will follow if we can show
that the series is uniformly convergent on compact subsets of said region. Since
1 X 1
f (z) = 2 +
z n6=0
(n + z)2

it suffices to show the uniform convergence of this last sum.


So let |z| 6 R, let z be fixed away from integers: |z + n| > ρ >P 0 ∀n ∈ Z, and let
−M be the nearest integer to z. Note that the series looks like 2 n6=0 n−2 which is
convergent. So we first pull out the convergent part by writing
X 1
X
1
(n + z)2 = .

n 2 |1 + z/n|2
6=0 n6=0

Then if |n| 6 |M |
|z + n| > |z + m| =⇒ |1 + z/n| > (|M |/|n|)|1 + z/m| > |1 + z/M | > ρ/|M |.
If |n| > |M | + 1
|<(z)| |M | + 1 1
|1 + z/n| > |1 + <(z)/n| > 1 − >1− = .
|n| |M | + 2 |M | + 2
COMPLEX ANALYSIS: SOLUTIONS 5 11

Therefore,
|M |2

X 1 X 1 X 1 1
(n + z)2 6 ρ2

2
+ (|M | + 2)2 2
+
n6=0
n n |z ± (M ± 1)|2
06=|n|6|M | |n|>|M |+1
2
(R + 1) 2 4
62ζ(2) + 2ζ(2)(R + 3) +
ρ2 ρ2
<∞.

Hence, by the Weierstrass M test with Mn = n−2 max((R + 3)2 , (R + 1)2 /ρ2 ) the
series is uniformly convergent on compact subsets of C\Z.
7. Let ϕ : R → R be a smooth function whose derivatives ϕ(k) (t), k > 0, are of rapid
decay at ∞ i.e.
lim tA ϕ(k) (t) = 0
t→∞

for all A ∈ R and all k > 0. The Mellin transform of ϕ is defined as


Z ∞
ϕ̃(z) = ϕ(t)tz−1 dt.
0

(i) Prove that ϕ̃(z) is analytic in the region <(z) > 0.


(ii) Use integration by parts to show that ϕ̃(z) can be analytically continued to
C\Z60 , with possible simple poles at z = −n.
(iii) Find the residues at these poles and compute the formal sum

X
res(ϕ̃(z)t−z , −n).
n=0

What does this look like?


(iv) Given part (iii), suggest an argument that would prove
Z
1
ϕ̃(z)t−z dz = ϕ(t)
2πi <(z)=c

where the integral is over the vertical line from c − i∞ to c + i∞ with c > 0.

Solution: (i). Since ϕ is smooth on R it is bounded in a neighbourhood of t = 0:


|ϕ(t)| 6 C for t ∈ [0, t0 ], say. Then
Z t0 Z t0
<(z)
z−1
|ϕ(t)t |dt 6 C t<(z)−1 dt = Ct0 /<(z) < ∞.
0 0
COMPLEX ANALYSIS: SOLUTIONS 5 12

Also, since ϕ is of rapid decay, for any z there exists a t1 = t1 (<(z)) such that
|ϕ(t)tz−1 | 6 Dt−2 for all t > t1 . Then
Z ∞
|ϕ(t)tz−1 |dt 6 D/t1 < ∞.
t1
Rt
Since the remaining integral t01 |ϕ(t)tz−1 |dt is finite the integral is absolutely con-
vergent in the region <(z) > 0.
Let γ be a closed curve in the region <(z) > 0. By absolute convergence and
Fubini’s Theorem
Z Z Z ∞ Z ∞ Z
z−1 −1
ϕ̃(z)dz = ϕ(t)t dtdz = ϕ(t)t tz dzdt.
γ γ 0 0 γ

But since tz is analytic in the region <(z) > 0 for all t ∈ [0, ∞) this last inner integral
is zero by Cauchy’s Theorem. Hence,
Z
ϕ̃(z)dz = 0.
γ

The continuity of ϕ̃(z) follows exactly as it does for the Gamma function, hence it
is analytic by Morera’s Theorem.

(ii). Integrating by parts once gives



ϕ(t)tz 1 ∞ 0 1 ∞ 0
Z Z
z
ϕ̃(z) = − ϕ (t)t dt = − ϕ (t)tz dt.
z t=0 z 0
z 0
R∞ 0
By the same reasoning used in part (i), the integral 0 ϕ (t)tz dt is analytic in the
region <(z) > −1. Hence, the above expression provides a meromorphic continuation
of ϕ̃(z) to said region. At z = 0 we have a residue of
1 ∞ 0
Z Z ∞
ϕ0 (t)dt = ϕ(0).
z

z·− ϕ (t)t dt =−
z 0 z=0 0

In particular, if ϕ(0) = 0 then there is no pole at z = 0.


More generally, integrating by parts n times gives
Z ∞
(−1)n
(1) ϕ̃(z) = ϕ(n) (t)tz+n−1 dt.
z(z + 1) · · · (z + n − 1) 0
Again, this expression is seen to be analytic in the region <(z) > −n, with the
exception of possible simple poles at z = 0, −1, . . . , −n + 1. Since n is arbitrary,
we’re done.
COMPLEX ANALYSIS: SOLUTIONS 5 13

(iii). From equation (1) we see that


Z ∞
(−1)n+1

(n+1)

z+n
res(ϕ̃(z), z = −n) =(z + n) ϕ (t)t dt
z(z + 1) · · · (z + n) 0 z=−n
n+1 Z ∞
(−1)
= ϕ(n+1) (t)dt
(−n)(−n + 1) · · · (−1) 0
1 ∞ (n+1)
Z
=− ϕ (t)dt
n! 0
1
= ϕ(n) (0).
n!
In particular, if ϕ(t) is of rapid decay at zero as well as at infinity, then ϕ̃(z) is entire.
Since all the poles are simple, an extra factor of t−z in the above residue calculations
gives a factor of t−z |z=−n = tn and hence
∞ ∞
X
−z
X 1 (n)
res(ϕ̃(z)t , −n) = ϕ (0)tn .
n=0 n=0
n!

This, of course, looks like ϕ(t) in the form of a Taylor expansion about t = 0.

(iv). Assuming that the above series converges and equals ϕ(t) (so ϕ is analytic
in the real analysis sense) then we would like a contour integral which captures all
these residues. As suggested, we should look at
Z
1
ϕ̃(z)t−z dz.
2πi <(z)=c
We would like to truncate this integral at heights z = c ± iT and then consider
the truncated integral as part of a rectangular contour which encloses some of the
residues. To estimate these integrals we need bounds on ϕ̃(z).
From equation (1) we see that for fixed <(z), ϕ̃(z) → 0 as |=(z)| → ∞. More
precisely, as =(z) → ∞ equation (1) gives
Z ∞
1 C<(z),n
|ϕ̃(z)| 6 |ϕ(n) (t)|t<(z)+n−1 dt 6 ,
|z(z + 1) · · · (z + n − 1)| 0 |=(z)|n
for some constant C<(z),n . Consequently,
1h ∞
Z Z c+iT Z Z −T i
1 −z 1 −z
ϕ̃(z)t dz = ϕ̃(z)t dz + + ϕ̃(c + iy)t−c−iy dy
2πi <(z)=c 2πi c−iT 2π T −∞
Z c+iT
1
= ϕ̃(z)t−z dz + O(T −n+1 )
2πi c−iT
COMPLEX ANALYSIS: SOLUTIONS 5 14

We now consider this last integral as part of a rectangular contour Γ whose left edge
crosses the real axis at −N − 1/2. Then,
Z c+iT Z Z
1 −z 1 −z 1
ϕ̃(z)t dz = ϕ̃(z)t dz − ϕ̃(z)t−z dz
2πi c−iT 2πi Γ 2πi other edges of Γ
N Z
X 1 (n) 1
= n
ϕ (0)t − ϕ̃(z)t−z dz
n=0
n! 2πi other edges of Γ
by the residue theorem. Using our bound on |ϕ̃(z)| we can estimate the remaining
integrals. The largest contribution comes from the integral over the left edge of Γ,
and this is seen to be O(tM +1/2 ). The other integrals are small in size. On letting
T → ∞ we get
Z N
1 −z
X 1 (n)
ϕ̃(z)t dz = ϕ (0)tn + O(tN +1/2 ) = ϕ(t).
2πi <(z)=c n=0
n!