Complexanalysis 4 Applications of Cauchy
Uploaded by
iantyler329Complexanalysis 4 Applications of Cauchy
Uploaded by
iantyler329Complex Analysis (part 4): Applications of Cauchy's Integral Formula (by Evan Dummit, 2022, v. 1.
00)
Contents
4 Applications of Cauchy's Integral Formula 1
4.1 Estimates for Holomorphic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
4.1.1 The Cauchy Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
4.1.2 Entire Functions, Liouville's Theorem, and the Fundamental Theorem of Algebra . . . . . . . 2
4.1.3 The Maximum Modulus Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
4.2 Laurent Series and Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4.2.1 Laurent Series Expansions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
4.2.2 Zeroes of Holomorphic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.2.3 Removable Singularities, Poles, and Essential Singularities . . . . . . . . . . . . . . . . . . . . 10
4.3 Residues and Residue Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.3.1 Residues, The Residue Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.3.2 Calculating De nite Integrals via Residue Calculus: Circular Contours . . . . . . . . . . . . . 19
4.3.3 Calculating De nite Integrals via Residue Calculus: Circular Contours With Detours . . . . . 21
4.3.4 Calculating De nite Integrals via Residue Calculus: Other Contours . . . . . . . . . . . . . . 25
4 Applications of Cauchy's Integral Formula
´ f (z)
In this chapter, our goal is to use Cauchy's integral formula f (z0 ) = γ
dz to study the behavior of holomor-
z − z0
phic functions in a variety of ways. We begin by giving estimates on the growth rate of holomorphic functions and
then studying the various behaviors near zeroes and singularities that holomorphic functions may have. We then
give a lengthy discussion of residue calculus and how to evaluate a wide range of otherwise very di cult-to-evaluate
integrals on the real line.
4.1 Estimates for Holomorphic Functions
• In this section we will give various estimates resulting from Cauchy's integral formula that will help us
characterize useful properties of holomorphic functions.
4.1.1 The Cauchy Estimates
• Our rst goal is to use Cauchy's integral formula to provide estimates on the growth rate for the value of a
holomorphic function.
◦ Recall that if f is holomorphic on the interior of a simply-connected region with counterclockwise bound-
n! ´ f (z)
ary γ, then for each z0 in the region and each n≥0 we have f (n) (z0 ) = γ
dz .
2πi (z − z0 )n+1
• Theorem (Cauchy Estimates): Suppose R is an open region and f is holomorphic on R.
1
1. Suppose z0 ∈ R and that the closed disc D : |z − z0 | ≤ r is contained in R. If |f (z)| ≤ M on D, then
M · n!
f (n) (z0 ) ≤ .
rn
◦ Proof: If we take γ to be the counterclockwise boundary of the disc, then by our usual arclength
(n) n! ´ f (z) n! M M · n!
estimate, we have f (z0 ) = dz ≤ · 2πr · n+1 = .
2πi γ (z − z0 )n+1 2π r rn
P∞
2. Suppose that f is holomorphic on the closed disc |z − z0 | ≤ r . Then the power series f (z) = n=0 an (z −
z0 )n for f centered at z = z0 has radius of convergence at least r.
P∞
◦ Proof: Since holomorphic functions are analytic, we may write f (z) = n=0 an (z−z0 )n for constants
f (n) (z0 ) M
an = , and applying (1) yields |an | ≤ .
n! rn
1/n
◦ Then lim supn→∞ |an | ≤ 1/r, meaning the radius of convergence of the power series is at least r,
as claimed.
3. Suppose that f is holomorphic on the open disc |z − z0 | < r but on no larger open disc. Then the power
P ∞ n
series f (z) = n=0 an (z − z0 ) for f centered at z = z0 has radius of convergence equal to r .
◦ Proof: By applying (2) to the closed discs |z − z0 | ≤ s for 0 < s < r we see that the radius of
convergence is at least s for all s < r, so the radius of convergence must be at least r.
◦ On the other hand, if the radius were greater than r, then since analytic functions are holomorphic,
f would be holomorphic on a disc
• Part (3) of the theorem provides, in many cases, an easy way to identify the radius of convergence of the
power series of a function whose holomorphic behavior is already known, such as a rational function:
1
• Example: Find the radius of convergence of the power series centered at z=0 for f (z) = .
1 − 2z
◦ The function f is holomorphic for z 6= 1/2, so by (3) of the theorem above, the function is holomorphic
for |z| < 1/2 but not for |z| ≤ 1/2 and therefore not for any |z| < r with any r > 1/2.
◦ Thus, the radius of convergence must equal 1/2 .
P∞ n
◦ We can con rm this fact by computing the power series itself, which is n=0 (2z) so that an = 2n : then
1/n
indeed lim supn→∞ |an | =2 so the radius is 1/2.
1
• Example: Find the radius of convergence of the power series centered at z=0 for f (z) = .
1 + z2
◦ The function f is holomorphic for z 6= ±i, so by (3) of the theorem above, the function is holomorphic
for |z| < 1 but not for |z| ≤ 1 and therefore not for any |z| < r with any r > 1.
◦ Thus, the radius of convergence must equal 1 .
P∞ n 2n
◦ We can con rm this fact by computing the power series itself, which is n=0 (−1) z : then indeed
1/n
lim supn→∞ |an | =1 so the radius is 1.
◦ Interestingly, if we consider this function's power series only on the real line, it still has radius of conver-
gence 1, yet the function is in nitely dierentiable on all of R.
◦ The reason for this otherwise mysterious lack of convergence for this power series on the entire real line,
however, becomes quite obvious when we extend our view to the complex plane: as we saw above, the
roots z = ±i of the denominator prevent this function from being holomorphic beyond |z| < 1.
4.1.2 Entire Functions, Liouville's Theorem, and the Fundamental Theorem of Algebra
• The Cauchy estimates, in particular, place very strong restrictions on the behavior of functions that are
holomorphic on the entire complex plane. Such functions are given a special name:
• De nition: A function f (z) holomorphic on the entire complex plane is called an entire function (or just
entire).
2
◦ Examples: Any polynomial in z is entire, as are ez , sin z , cos z , sinh z , and cosh z .
◦ Non-Examples: The functions 1/z , Log(z), and csc z are not entire, since none of them are holomorphic
P∞
at z = 0. The function tan z is not entire since it is not holomorphic at z = π/2. The function n=0 zn
is not entire since it is not holomorphic (or even de ned) for |z| > 1.
• A fundamental result of Liouville is that the only bounded entire functions are constants:
• Theorem (Growth Estimates): Suppose f (z) is an entire function.
1. (Liouville's Theorem) A bounded entire function is constant.
◦ Proof: Suppose f (z) is entire and assume that |f (z)| ≤ M on C.
M
◦ Applying the Cauchy estimate to f0 on the circle of radius r centered at z0 yields |f 0 (z0 )| ≤ .
r
Since this bound holds for all r > 0, taking r→∞ yields |f 0 (z0 )| = 0 hence f 0 (z0 ) = 0.
0
◦ This applies for all z0 , so f is identically zero, and so f must be constant.
n
2. If there exist constants A, B , and n such that |f (z)| ≤ A + B |z| for all z, then f (z) is a polynomial of
degree at most n.
◦ Notice that Liouville's theorem is the special case n = 0 of this result.
n
◦ Proof: Suppose f (z) is entire and assume that |f (z)| ≤ A + B |z| on C.
(n+1) (n+1)
◦ Applying the Cauchy estimate to f on the circle of radius r centered at z0 yields f (z0 ) ≤
n
n!(A + Br )
= n!(A/rn+1 + B/r). Since this bound holds for all r > 0, taking r → ∞ yields
rn+1
f (n+1) (z0 ) = 0 hence f (n+1) (z0 ) = 0.
◦ This applies for all z0 , so f (n+1) is identically zero. Taking the antiderivative n+1 times shows that
f must be a polynomial of degree at most n, as claimed.
• By using Liouville's theorem, we can give a very quick proof of the fundamental theorem of algebra. We rst
make some preliminary remarks.
◦ First, recall the remainder/factor theorem : if
1 p(z) is a polynomial and r ∈ C, then the remainder upon
dividing p(z) by z−r is p(r). In particular, z − r divides p(z) if and only if p(r) = 0.
◦ The usual statement of the fundamental theorem of algebra over C is that any polynomial p(z) of degree
d can be factored into the form p(z) = a(z − r1 )(z − r2 ) · · · (z − rd ) for some (not necessarily distinct)
complex numbers ri .
◦ In order to prove this fact, it su ces to show that every nonconstant polynomial p(z) has a root in C:
then an easy induction using the remainder/factor theorem to remove a linear factor z−r establishes
the general form above.
◦ We now establish that every nonconstant complex polynomial has a root in C.
• Theorem (Fundamental Theorem of Algebra): Every nonconstant complex polynomial has a root in C.
◦ p(z) is a nonconstant complex polynomial with no root in C. Then 1/p(z) is entire hence
Proof: Suppose
in particular continuous on
C.
Pd
◦ From writing p(z) = n=0 an z n = ad z d (1 + ad−1 /z + · · · + a0 /z d ) we see limz→∞ p(z)/z d = ad , and since
d > 0 this means lim|z|→∞ |p(z)| = ∞.
◦ Taking the reciprocal yields lim|z|→∞ |1/p(z)| = 0, so in particular there exists some R such that
|1/p(z)| ≤ 1 for |z| ≥ R.
◦ Additionally, for |z| ≤ R the function |1/p(z)| is continuous on a closed bounded region, hence it is
bounded above, say with |1/p(z)| ≤ M .
◦ Then |1/p(z)| ≤ max(1, M ) for all z ∈ C, meaning that 1/p(z) is a bounded entire function. It is therefore
constant by Liouville's theorem. But this is a contradiction if p(z) is not constant. We conclude that
p(z) must have a complex root, as desired.
1 Purelyfor completeness, if p(z) = a0 + a1 z + · · · + an z n , then p(z) − p(r) =
Pn
ak (z k − rk ), and since z − r divides each term
k=0
in the sum, it divides p(z) − p(r).
3
4.1.3 The Maximum Modulus Principle
• Cauchy's integral formula, as we have previously mentioned, can also be interpreted as an averaging result.
◦ Speci cally, if f (z) is holomorphic for |z − z0 | ≤ r, then the value f (z0 ) is equal to the average value of
f (z) on the circle |z − z0 | = r.
◦ In particular, then, we see that f (z0 ) cannot be strictly larger in absolute value than all of the values
f (z) on the circle |z − z0 | = r (otherwise, we would have a contradiction to the triangle inequality). Since
this holds for each possible radius, we see that f (z0 ) cannot be strictly larger in absolute value than the
value of f (z) for any z in the disc |z − z0 | ≤ r .
◦ By applying the argument to smaller discs centered at other points, we can see that no interior point of
the disc can be a point where f takes its maximum modulus.
• We may state this principle more formally:
• Theorem (Maximum Modulus Principle): Suppose f (z) is holomorphic on a connected bounded region R.
1. If |f (z)| is constant on R, f (z) is constant on R.
then
∂ h i ∂f ∂f ∂f
◦ Proof: We have 0= f (z)f (z) = f 0 (z)f (z) + f (z) = f 0 (z)f (z) since = =0 since f is
∂z ∂z ∂z ∂z
holomorphic.
◦ f 0 (z) be zero or f (z) (hence f (z)) be zero along a sequence of points converging to
This either requires
0
a limit in R, which by the uniqueness result for power series implies f (z) or f (z) must be identically
zero on R.
◦ In either case we see that f (z) is constant on R, as claimed.
2. If R is closed and the maximum value of |f (z)| occurs at a point z0 in the interior of R, then f is constant
on R.
◦ Proof: Suppose |f (z)| is maximized at z = z0 in the interior of R. By de nition there exists some
r>0 such that the disc |z − z0 | ≤ r is contained in R.
◦ By Cauchy's integral formula using the parametrization of the contour γr winding once counterclock-
1 ´ 2π
wise around |z − z0 | = r, we see f (z0 ) = f (z0 + reiθ ) dθ, meaning that f (z0 ) is the average
2π 0
value of f (z) on the circle.
1 ´ 2π 1 ´ 2π
◦ By the triangle inequality we then have |f (z0 )| = 0
f (z0 + reiθ ) dθ ≤ f (z0 + reiθ ) dθ ≤
2π 2π 0
1 ´ 2π
|f (z0 )| dθ = |f (z0 )|, which means we have equality everywhere. This means |f (z)| is constant
2π 0
on the circle |z − z0 | = r hence by (1), f (z) is constant on the circle.
◦ Since f (z) is constant on a curve, by uniqueness of series expansions this means f is constant on all
of R, as claimed.
3. (Maximum Modulus Principle) The maximum value of |f (z)| on R occurs on the boundary of R.
◦ Proof: This follows immediately by noting the result is trivial if f is constant, and then by taking
the contrapositive of (2) if f is not constant.
• We can apply the maximum modulus principle to simplify certain kinds of optimization problems.
• Example: Find the maximum value of z 2 + 4z − 2 for |z| ≤ 1.
◦ A natural rst attempt is to use the triangle inequality to write z 2 + 4z − 2 ≤ z 2 + |4z| + |−2| ≤ 7.
However, this calculation cannot be sharp, because for |z| = 1 we can only have |4z − 2| = 6 for z = −1,
but then z 2 + 4z − 2 = 5 rather than 7.
2 2
◦ A direct approach would be to set z = reiθ and then compute z 2 + 4z − 2 = r2 e2iθ + 4reiθ − 2 =
2
(r2 cos 2θ + 4r cos θ − 2) + (r2 sin 2θ + 4r sin θ)i = (r2 cos 2θ + 4r cos θ − 2)2 + (r2 sin 2θ + 4r sin θ)2 but
this ends up being rather unpleasant to maximize for 0 ≤ r ≤ 1 and 0 ≤ θ ≤ 2π , as one may quickly
discover by attempting to take partial derivatives.
4
◦ Using the maximum modulus principle, however, we can reduce to having a single parameter: since the
function f (z) = z 2 + 4z − 2 is holomorphic, its maximum modulus on the closed disc |z| ≤ 1 necessarily
occurs on the boundary |z| = 1.
2
◦ Then we have the simpler optimization = (e2iθ +4eiθ −2)(e−2iθ +4e−iθ −2) = −8 cos θ−4 cos 2θ+
f (eiθ )
21, which necessarily takes its extreme values when the derivative 8 sin θ + 8 sin 2θ = 8 sin θ(1 + 2 cos θ)
is zero. By basic trigonometry this occurs for θ = 0, π, 2π/3, 4π/3, 2π , and one may then check that
√
√ −1 ± i 3
θ = 2π/3 and 4π/3 yield maxima of z 2 + 4z − 2 = 27 occurring for z = .
2
• We will also remark that the maximum modulus principle does not hold on unbounded domains (at least,
without modi cation).
z
◦ For a speci c example, consider the function f (z) = ee on the region −π/2 ≤ Im(z) ≤ π/2.
x
◦ On the boundary of this region, with z = x ± iπ/2, we have
z
e = ±ie x
and so |f (z)| = e±ie = 1, so f
is bounded on the boundary of the region.
◦ However, on the real axis, we see that f is unbounded. Thus, f does not attain its maximum modulus
on the boundary of the unbounded region R.
◦ The main issue is that the function f (z) grows extremely quickly on the real axis. If one imposes suitable
restrictions on the growth rate of f inside R, one may recover versions of the maximum modulus principle.
• One result of this nature is due to Lindelöf:
• Theorem (Lindelöf Principle): Let R be a half-strip region of the form x1 ≤ Re(z) ≤ x2 and Im(z) ≥ y1 for
some real x1 , x2 , y1 . If f is holomorphic on R |f (x + iy)| ≤ Ay n
and for some constants A and n on R, then
the maximum modulus of f occurs on the boundary of R.
◦ This result can be substantially generalized into a result known as the PhragménLindelöf principle. We
will not give the details here.
◦ If f is unbounded on the boundary of R then the result is trivial. So assume that |f (z)| ≤ M on the
boundary of R. The rest of the argument then follows by making an estimation
2 relying on the maximum
modulus principle applied to the function f (z)/(z + it)n+1 on a suitable rectangle x1 ≤ Re(z) ≤ x2 ,
y1 ≤ Im(z) ≤ y2 .
4.2 Laurent Series and Singularities
• As we have seen in the motivation for Cauchy's integral formula, the series expansion for a holomorphic
function around a given point z = z0 carries a tremendous amount of information.
◦ We used power series expansions to de ne and analyze various elementary functions such as the exponen-
tial, sine, and cosine, and we also showed during our discussion of formal power series that any rational
function can be expanded as a Laurent series with nitely many negative terms.
◦ More generally, as also follows from our discussion of formal series and the fact that holomorphic functions
are analytic, the quotient of any two holomorphic functions can be expressed as a convergent Laurent
P∞
series n=−k an (z − z0 )n .
◦ We would now like to expand our discussion to cover local expansions of all holomorphic functions at
arbitrary points.
2 For Ay2n M
completeness: take z0 = x0 + iy0 in R, choose any t > −y1 and select y2 > y0 such that ≤
(y2 + t)n+1 (y1 + t)n+1
f (z)
(possible since the left-hand side tends to zero as y2 → ∞). Then applying the maximum modulus principle to g(z) =
(z + it)n+1
M
on the rectangle x1 ≤ Re(z) ≤ x2 and y1 ≤ Im(z) ≤ y2 shows that |g(z0 )| ≤ |g(iy1 + it)| = . In terms of f this yields
" #n (y1 + t)n+1
|z0 + it|
|f (z0 )| ≤ M . Taking the limit as t → ∞ yields |f (z0 )| ≤ M , as desired.
y1 + t
5
4.2.1 Laurent Series Expansions
• There are still examples of holomorphic functions that cannot everywhere be expressed as a Laurent series
P∞
of the form n=−k an (z − z0 )n , where necessarily the center z0 must be a point where the function is not
holomorphic.
◦ A simple example is f (z) = e1/z , z 6= 0.
which is holomorphic for
n
P∞
◦ This function cannot have a Laurent expansion n=−k an z centered at z = 0: if it did, multiplying
k
by z and then taking the limit as z → 0 would yield a nite number a−k , but one can check (e.g., via
k 1/z
L'Hôpital's rule) that limz→0 z e does not exist for any integer k .
◦ On the other hand, we can certainly write down a convergent series expansion for f (z): simply plug in
P∞ (1/z)n P0 1 n
1/z to the power series for ez to see that e1/z = n=0 = n=−∞ z , valid for all z 6= 0.
n! |n|!
◦ This calculation suggests P∞that we should broaden our focus to consider doubly in nite Laurent series
n
expansions of the form n=−∞ an (z − z0 ) .
P∞
• De nition: A Laurent series centered at z = z0 is a series of the form n=−∞ an (z − z0 )n for an ∈ C. We say
P∞ n
P−1 n
that the series converges at a value z if the two tails n=0 an (z − z0 ) and n=−∞ an (z − z0 ) both converge
at z , and we say the series converges absolutely when both tails converge absolutely.
◦ When the series converges, we de ne the value f (z) of the Laurent series to be the sum of the two tail
series.
n
P∞
◦ By our previous analysis of the convergence behavior of power series, we know that if n=0 an (z − z0 )
converges for z = z̃ where |z̃ − z0 | = R, then it converges absolutely for all z with |z − z0 | < R.
P−1 n
◦ By replacing z − z0 with its reciprocal, we can see in the same way that if n=−∞ an (z − z0 ) converges
for z = z̃ where |z̃ − z0 | = r , then it converges absolutely for all z with |z − z0 | > r .
◦ As a consequence, the region of absolute convergence of a general Laurent series is an annulus of the
form r < |z − z0 | < R for some r and R.
• Our main result is that a function that is holomorphic on an annulus can be expressed as a convergent Laurent
series on that annulus:
• Theorem (Laurent Expansions): Suppose f (z) is holomorphic on the annulus r ≤ |z − z0 | ≤ R.
1. For any
P∞ r < s < S < R, the function f (z) has an absolutely convergent Laurent series expansion
n
n=−∞ n (z − z0 ) that converges absolutely and uniformly to f on the region s ≤ |z − z0 | ≤ S .
a
´ f (ζ) ´ f (ζ)
Moreover, the coe cients are given by an =
γR (ζ − z )n+1
dz for n ≥ 0 and γr dz for
0 (ζ − z0 )n+1
n < 0, where γR and γr are the counterclockwise circles |z − z0 | = R and |z − z0 | = r respectively.
◦ The proof of this result is essentially the same as the one we used to show that holomorphic functions
are analytic: we simply expand out 1/(ζ − z) as a geometric series and then integrate term by term.
◦ Proof: Consider the contour γ that follows the circle of radius R once counterclockwise, traverses a
line to the circle of radius r and then follows this circle once clockwise, and then returns along the
same line to its starting point.
◦ The integrals along the line segment cancel one another, so the integral of any function around this
contour γ is the same as the integral around the counterclockwise circle γR of radius R minus the
integral around the counterclockwise circle γr of radius r.
◦ Since the winding number of γ around any point in the annulus is 1, Cauchy's integral formula
´ f (ζ) ´ f (ζ)
implies that for any z in the annulus, we have f (z) = γR
dζ − γr dζ .
ζ −z ζ −z
1
◦ Now we expand both integrals as convergent geometric series: for r < s ≤ |z − z0 | we have =
ζ −z
1 1 P∞ 1 1
· = (z − z0 )n , and for |z − z0 | ≤ S ≤ R we have =
ζ − z0 1 − z − z0 n=0
(ζ − z0 )n+1 ζ −z
ζ − z0
1 1 P∞ 1 P−1 1
− · = − n=0 (ζ − z0 )n = n=−∞ (z − z0 )n .
z − z0 ζ − z0 (z − z0 )n+1 (ζ − z0 )n+1
1−
z − z0
6
z − z0 S
◦ On γR with |ζ − z0 | = R the rst series has common ratio ≤ while on γr with |ζ − z0 | = r
ζ − z0 R
ζ − z0 r
the second has common ratio ≤ . Thus, both geometric series converge absolutely and
z − z0 s
uniformly for s ≤ |z − z0 | ≤ S .
P∞ f (ζ)
◦ Therefore, because f (ζ) is bounded on γ since it is continuous, the partial sums of n=0 (z−
(ζ − z0 )n+1
P−1 f (ζ)
z0 )n and of n=−∞ (z − z0 )n converge absolutely and uniformly for s ≤ |z − z0 | ≤ S to
(ζ − z0 )n+1
f (ζ)
on γR and γr respectively.
ζ −z
◦ Hence by our results on uniform convergence and integrals, we may switch the order of the sums
and integral to see
ˆ ˆ ˆ ∞ ˆ −1
f (ζ) f (ζ) X f (ζ) n
X f (ζ)
f (z) = dζ − dζ = n+1
(z − z0 ) dζ − (−1) (z − z0 )n dζ
γR ζ −z γr ζ −r γR n=0 (ζ − z0 ) γr n=−∞ (ζ − z0 )n+1
∞ ˆ −1 ˆ
X f (ζ) n
X f (ζ)
= n+1
(z − z0 ) dζ + n+1
(z − z0 )n dζ
n=0 γR
(ζ − z0 ) n=−∞ γr
(ζ − z0 )
P∞ ´ f (ζ)
which is of the form f (z) = n=−∞ an (z − z0 )n for an = γR
dζ for n ≥ 0 and
(ζ − z0 )n+1
´ f (ζ)
γr
dz for n < 0. This is the desired Laurent series expansion for f with coe cients
(ζ − z0 )n+1
as claimed.
P∞
2. The coe cients an in a Laurent expansion f (z) = n=−∞ an (z − z0 )n are unique and are also given by
1 ´ f (z)
an = dz where γ is any contour inside the annulus r < |z − z0 | < R with winding
2πi γ (z − z0 )n+1
number 1 around z0 .
◦ Proof: By our observations on the convergence of Laurent expansions, the Laurent expansion con-
verges uniformly to f (z) on γ since γ is in the interior of the annulus.
1 ´ f (z) 1 ´ P∞ n−d−1
◦ Interchanging the sum and integral yields
γ d+1
dz = γ n=−∞ an (z − z0 ) dz
2πi (z − z0 ) 2πi
1 P∞ h´ i
n−d−1
= n=−∞ an γ
(z − z0 ) dz = ad where the last step follows by our usual observation
´
2πi
n−d−1
that
γ
(z − z0 ) dz is 2πi for n − d − 1 = −1 (i.e., for n = d) and is 0 for all other n.
P∞ n
3. If f has a Laurent expansion f (z) =
0 0
P<∞|z − z0 | < R, then
n=−∞ an (z−z0 ) for r the Laurent expansion
n−1
for f can be obtained by dierentiating termwise: f (z) = n=−∞ na n (z − z0 ) .
◦ Proof: Since the Laurent expansion converges absolutely and uniformly for s ≤ |z − z0 | ≤ S for
r < s < S < R, by our results on uniform convergence we may dierentiate termwise to obtain the
0
expansion for f (z). Taking s → r and S → R yields that on the full annulus r < |z − z0 | < R.
◦ Finally, (2) ensures that the Laurent expansion for f 0 (z) is unique.
• In principle, for any holomorphic function f (z) on an annulus r ≤ |z − z0 | ≤ R we may compute the coe cients
in its Laurent expansion using the integral formulas in the theorem above.
◦ However, in practice (e.g., for rational functions) it is often easier to manipulate known series, such as
geometric series, to compute Laurent expansions.
• Example: Find the Laurent expansion of f (z) = 1/(1 − z) on the region |z| > 1.
1 P∞
◦ We have repeatedly worked out the expansion = n=0 z n = 1 + z + z 2 + z 3 + · · · for |z| < 1. We
1−z
can use the same idea to obtain the expansion for |z| > 1, using 1/z in place of z .
7
−1
1 1 1 P∞ X
◦ Explicitly, we have f (z) = =− · = −z −1 n=0 z −n = −z n = · · · − z −4 − z −3 −
1−z z 1 − 1/z n=−∞
z −2 − z −1 .
1
• Example: Find the terms from z −3 through z3 in the Laurent expansion of f (z) = on the
(z − 1)(z − 2)
region |z| < 1, on the region 1 < |z| < 2, and on the region |z| > 2.
1 1
◦ Using a partial fraction decomposition, we have f (z) = − so it su ces to expand each of
z−2 z−1
these series on the given region.
1 1 P∞ 1 1 1
◦ For |z| < 1 we have =− = n=0 (−1)z n , whereas for |z| > 1 we have = · =
z−1 1−z z−1 z 1 − 1/z
1 P∞ −n P∞ −1−n
z = n=0 z .
z n=0
1 1 1 1 P∞ P∞ −1
◦ For |z| < 2 we have =− · =− (z/2)n = n=0 n+1 z n , whereas for |z| > 2 we
z−2 2 1 − z/2 2 n=0 2
1 1 1 1 P∞ ∞
(2/z)n = n=0 2n z −1−n .
P
have = · =
z−2 z 1 − 2/z z n=0
P∞ −1 n P∞ P∞ 1 1 3 7 15
◦ So for |z| < 1 we get f (z) = n=0 n+1
z − n=0 (−1)z n = n=0 (1− n+1 )z n = + z + z2 + z3 + · · · .
2 2 2 4 8 16
−1 n P∞ −1−n
P∞ 1 1 1 1
◦ For 1 < |z| < 2 we get f (z) = n+1
z − n=0 z
n=0 = · · · − z −3 − z −2 − z −1 − − z − z 2 − z 3 − · · · .
2 2 4 8 8
P∞ n −1−n P∞ −1−n P∞
◦ Finally, for |z| > 2 we get f (z) = n=0 2 z − n=0 z = n=1 (2n −1)z −1−n = · · · + 7z + 3z + z −2 .
−4 −3
• Example: Find the terms from z −3 through z3 in the Laurent expansion of f (z) = z −2 ez + ze1/z on the region
1 < |z| < 2.
P∞ 1 1 1 1 1 3 P∞ 1
◦ We have z −1 ez = z −2 n=0 z n = z −2 +z −1 + + z+ z 2 + z +· · · and ze1/z = z n=0 z −n =
n! 2 6 24 120 n!
1 −3 1 −2 1 −1
· · · + z + z + z + 1 + z.
24 6 2
1 7 3 3 7 1 1 3
◦ Adding yields z −2 ez + ze1/z = · · · + z −3 + z −2 + z −1 + + z + z 2 + z + ··· .
24 6 2 2 6 24 120
• Example: Find the terms up through z3 in the Laurent expansion, along with the radius of convergence, for
f (z) = csc(z) centered at z = 0.
1 1 1 1
◦ As a formal series, f (z) = = 3 5 7
= ·
sin(z) z − z /3! + z /5! − z /7! + · · · z 1 − z /3! + z /5! − z 6 /7! + · · ·
2 4
1 7 4 1 7 3
= z −1 (1 + z 2 + z + · · · ) = z −1 + z + z + · · · , using our previous techniques to compute
6 360 6 360
2 4 6
the multiplicative inverse of 1 − z /3! + z /5! − z /7! + · · · .
◦ For the radius of convergence, we note that f (z) only fails to be holomorphic when sin(z) = 0, which
occurs for z = kπ for integers k.
◦ Therefore, by our results on the radius of convergence of a series expansion, the radius of convergence
must be π , the distance from z=0 to the next closest point where f (z) is not holomorphic.
4.2.2 Zeroes of Holomorphic Functions
• The notion of a zero (or root) of a polynomial is quite familiar, as is the notion of the multiplicity of a zero.
◦ Explicitly, if p(z) is a polynomial, we say that r ∈ C is a zero (or root) of p if p(r) = 0. By the
remainder/factor theorem, if r is a root of p(z), then z−r divides p(z), which is to say, p(z)/(z − r) is
equal to a polynomial.
8
◦ By repeatedly taking out factors of z −r until there are none remaining (a process which must terminate
eventually as long as p(z) is not the zero polynomial), when r is a root of p(z) we may write p(z) =
(z − r)d q(z) for some polynomial q(z) of which r is not a root and some unique positive integer d, which
we call the multiplicity of that root.
◦ Example: For p(z) = z 3 − z 2 = z 2 (z − 1), the roots of p(z) are z=0 and z = 1. The root z=0 has
multiplicity 2, while the root z = 1 has multiplicity 1.
◦ We may equivalently identify the multiplicity of r using the derivatives of p(z): it is not hard to verify via
repeated application of the product rule that r has multiplicity d if and only if the rst d − 1 derivatives
of p(z) vanish at r but p(d) (r) = 0.
◦ Example: For p(z) = z 3 − z 2 we have p(0) = p0 (0) = 0 but p00 (0) = −2, so 0 has multiplicity 2 (as seen
above using the factorization).
• The zeroes of holomorphic functions can be classi ed in a similar way.
• De nition: Suppose f (z) is holomorphic on a region R and let z0 ∈ R. We say that z0 is a zero of f if
f (z0 ) = 0, and if f is not identically zero we say the order (or order of vanishing) of f at z0 is the smallest
positive integer d such that p(i) (z0 ) = 0 for i = 1, 2, . . . , d − 1 but p(d) (z0 ) 6= 0.
◦ For completeness, we also de ne the order of vanishing of the identically zero function to be ∞ everywhere.
◦ It is easy to nd the order of vanishing using the power series expansion of
P∞ f (z) around z = z0 : if
n
f (z) = n=0 an (z − z0 ) , then by repeatedly dierentiating and evaluating at z0 , we see that the order
is the smallest d for which ad 6= 0.
◦ This calculation also shows that the order of vanishing is always well de ned when f is not the zero
function (since the power series of a nonzero function is necessarily nonzero by uniqueness of series
expansions) and also agrees with our earlier notion of the order of a power series.
◦ Additionally, by factoring out (z − z0 )d f (z) at z = z0 , we also obtain
from the power series expansion of
an analogous factorization result: if f is a nonzero holomorphic function with a zero of order d at
z = z0 , then we may write f (z) = (z − z0 )d g(z) where g(z) is holomorphic and does not vanish at z0 .
• Example: Find the zeroes of f (z) = sin(z 2 ) and calculate their orders.
◦ Since sin w = 0 if and only if w = kπ for an integer k , the zeroes of f (z) are the square roots of these
√ √
values, which are ± kπ, ±i kπ for nonnegative integers k .
◦ For the orders, we note f 0 (z) = 2z cos(z 2 ), which for z 2 = kπ is simply 2z cos(kπ) = ±2z . This is nonzero
except at z = 0, so all of the zeroes except z = 0 have order 1 .
◦ At z=0 we have f 00 (z) = 2 cos(z 2 ) − 4z 2 sin(z 2 ) so that f 00 (z) = 2. Thus z=0 has order 2
• We will mention now one other very useful analogy between polynomials and holomorphic functions.
◦ It is easy to see (e.g., by induction on the degree) that a nonzero polynomial necessarily has nitely
many roots.
◦ In contrast, holomorphic functions may have in nitely many zeroes: for example, f (z) = sin z has zeroes
at z = kπ for all integers k.
◦ Although there may be in nitely many of them, as we showed using the uniqueness of power series, the
zeroes of analytic (equivalently, holomorphic) functions do retain one very convenient property: they are
isolated, meaning that if z0 is a zero of f and f is not identically zero, then there exists some r>0 such
that f (z) 6= 0 for 0 < |z − z0 | < r.
◦ In other words, each zero of a nonzero holomorphic function f is a positive distance away from all other
zeroes of f.
9
4.2.3 Removable Singularities, Poles, and Essential Singularities
• We have seen various examples of functions that are holomorphic except at some isolated set of points. We
now study these isolated singularities in more detail. First, we give a precise de nition:
• De nition: Suppose R is an open set and z0 ∈ R . We say that z0 is an isolated singularity of f if f is
holomorphic on all of R except at z0 .
1 ez 1/z
◦ Examples: The functions , , and e all have a single isolated singularity at z = 0.
z z
◦ Example: The function csc z has an isolated singularity at z = kπ for each integer k .
1
◦ Example: The function
2
has isolated singularities at z = i and z = −i.
z +1
◦ We can also construct some more arti cial examples by taking a holomorphic function and changing its
value at a point, or merely omitting a point from its domain entirely.
(
z2 for z=6 0, 1
◦ Example: The function f (z) = has isolated singularities at z=0 and z = 1.
100 for z = 0, 1
◦ Example: The function g(z) = z + 5 for z 6= −2 has an isolated singularity at z = −2 (it is not de ned
there).
z2 − 1
◦ Example: The function h(z) = for z 6= 1 has an isolated singularity at z = 1 (the expression
z−1
given is not de ned there).
◦ Of course, for f (z) above it is clear that we have taken the wrong value for f (z) at two points, while for
the function g(z) we should just extend the domain to include z = −2, and for h(z) we should simplify
the expression to h(z) = z + 1 for z 6= 1 and then extend its domain in the same way.
sin z
• A less trivial but similar situation arises for the function f (z) = , which has an isolated singularity at
z
z = 0.
◦ It is natural to try to extend the domain of this function to includez = 0, and the most sensible way
˜ sin z
to do it is to extend the de nition by setting f (0) = limz→0 = cos(0) = 1 (note that the limit is
z
simply the limit de nition of the complex derivative of sin z at z = 0).
sin z
◦ Indeed, with this choice, the function f˜(z) = for z 6= 0 and f˜(0) = 1 is in fact holomorphic at
z
z = 0. This can be shown directly by computing f˜0 (0) using the limit de nition of the derivative, but
we can give a much more conceptually natural approach using power series.
2n+1
sin z 1 P∞ n z
P∞ n z 2n
◦ Explicitly, we have the series expansion = (−1) = (−1) for all
z z n=0 (2n + 1)! n=0
(2n + 1)!
z 6= 0.
P∞ z 2n n
◦ So if we just consider the function g(z) = n=0 (−1)
de ned by this power series, then g(z)
(2n + 1)!
is holomorphic on all of C and agrees with f˜(z) for all z 6= 0 by the series expansion for sine.
◦ But g(z) also agrees with f˜ at z = 0 since g(0) = 1 = f˜(0) also. Therefore g and f˜ are exactly the same
function, so since g is holomorphic at z = 0, so is f˜.
• We see that in some cases, we may remove the isolated singularity by de ning (or rede ning) the value of
f (z0 ) so that the new function is holomorphic at z0 .
1
◦ However, this is not always possible for all singularities. For example, for f (z) = , there is no way to
z
1
assign a value to f (0) that makes the resulting function holomorphic, or even continuous, since limz→0
z
is not de ned. In a sense we will make more precise later, it is reasonable to view this limit as being ∞,
1
since limz→0 = ∞, so the function grows uniformly large in absolute value as z approaches zero, but
z
this still does not allow us to make the function holomorphic at z = 0.
10
◦ The function f (z) = e1/z has a similar issue at z = 0: there is no possible choice for f (0) that makes
f continuous there, let alone holomorphic: the limit limz→0 e1/z is also unde ned. In fact, the behavior
is actually worse than that of 1/z , since even the absolute value limit limz→0 e1/z is unde ned (for
instance, along the positive real axis the limit is ∞ while along the imaginary axis it is 1).
• We may use the Laurent expansion of a function near an isolated singularity to classify dierent types of local
behavior for a holomorphic function.
◦ More speci cally, suppose that z0 is an isolated singularity of f . Then since z0 is isolated, there is a
positive value of R such that f (z) is holomorphic on the punctured disc 0 < |z − z0 | ≤ R.
◦ From our discussion of Laurent expansions, for any
P∞ 0 < r < R the function f (z) has a convergent Laurent
n
expansion f (z) = n=−∞ an (z − z0 ) on the annulus r ≤ |z − z0 | ≤ R.
◦ By uniqueness of Laurent expansions, all of these expansions must agree with one another, so in fact,
this Laurent expansion converges on the full punctured disc 0 < |z − z0 | ≤ R.
◦ We can then identify three dierent classes of behavior depending on how many of the coe cients an
with n<0 are nonzero: none of them, nitely many of them, or in nitely many of them.
• P∞ Suppose that zn0 is an isolated singularity of f and f (z) has a convergent Laurent expansion
De nition:
f (z) = n=−∞ an (z − z0 ) on the punctured disc 0 < |z − z0 | ≤ R. If all of the coe cients an with n < 0
are zero, we say f has a removable singularity at z0 . If some but only nitely many of the coe cients an with
n < 0 are nonzero, we say f has a pole at z0 (the largest k such that a−k 6= 0 is called the order of the pole;
a pole of order 1 is called simple). Finally, if in nitely many of the an with n < 0 are nonzero, we say f has
an essential singularity at z0 .
(
z2 for z 6= 0, 1 z2 − 1
◦ Example: The functions f (z) = , g(z) = z + 5 for z 6= −2, and h(z) = for
100 for z = 0, 1 z−1
z 6= 1, all discussed above, each have removable singularities (for f at z = 0 and z = 1, for g at z = −2,
and for h at z = 1).
sin z
◦ Example: The function has a removable singularity at z = 0, since its Laurent expansion at z = 0
z
1 2 1 4
is 1 − z + z − · · · has no terms with a negative power of z .
6 120
1 z−1 ez
◦ Example: The functions , , and each have a pole at z = 0, since their Laurent expansions at
z z2 z
1
z = 0 are z −1 , −z −2 + z −1 , and z −1 + 1 + z + · · · . The orders of these poles are 1, 2, and 1 respectively.
2
1
◦ Example: The function 2 has a pole of order 2 at z = 0 and a pole of order 1 at z = 1, since the
z (z − 1)
−2
respective Laurent expansions at z = 0 and z = 1 are −z −z −1 −1−· · · and (z −1)−1 −2+3(z −1)+· · · .
◦ Example: The function e1/z has an essential singularity at z = 0, since its Laurent expansion is
P∞ 1 −n
n=0 z which has in nitely many terms with a negative power of z.
n!
◦ Example: The function 1/(ez − 1) has a simple pole at z = 2πik for each integer k, since its Laurent
1 1
expansion is (z − 2πik)−1 − + (z − 2πik) + · · · for each such k .
2 12
• These three types of singularities have very dierent properties, and we can characterize each of them based
on the local behavior of f (z) near the singularity. We begin with removable singularities:
• Theorem (Classi cation of Removable Singularities): Suppose that
P∞ z0 f and f (z)
is an isolated singularity of
n
has a convergent Laurent expansion f (z) = n=−∞ an (z − z0 ) on the punctured disc 0 < |z − z0 | ≤ R.
1. If the function f (z) has a removable singularity at z = z0 , then f (z) is bounded on the
( punctured disc 0 <
f (z) for z 6= z0
|z − z0 | ≤ R. Moreover, the limit limz→z0 f (z) = L exists, and the function f˜(z) = is
L for z = z0
holomorphic at z0 .
11
◦ Proof: By hypothesis, all of the terms an with n < 0Therefore, if we de ne f˜(z) =
are zero.
P∞ n
n=0 an (z − z0 ) , then f˜(z) = f (z) 0 < |z − z0 | ≤ R.
on the punctured disc
◦ In particular, the radius of convergence of the series for f˜ is at least R. So f˜ is holomorphic on the
entire disc |z − z0 | ≤ R, hence is continuous and thus bounded there. Since f agrees with f˜ on the
punctured disc, that means f is bounded as well.
◦ Likewise, L = limz→z0 f˜(z) = f˜(z0 ) exists, and indeed f˜(z) as described is holomorphic at z0 .
2. Conversely, suppose f (z) is bounded on the punctured disc 0 < |z − z0 | ≤ R. Then f has a removable
singularity at z = z0 .
◦ This result is often called Riemann's removable singularities theorem.
◦ Proof: We must show that all of the coe cients an with n<0 in the Laurent expansion of f (z) are
zero.
◦ By hypothesis, there exists some M |f (z)| ≤ M on the disc. Let 0 < r < R and take γr
such that
to be the counterclockwise circle of radius r z0 .
centered at
1 ´
◦ By our results on Laurent coe cients, we have a−n = (z −z0 )n−1 f (z) dz , so since on γ we have
2πi γ
1 ´
(z − z0 )n−1 f (z) ≤ rn−1 M , applying the arclength bound yields |a−n | = (z − z0 )n−1 f (z) dz ≤
2π γ
1
· 2πr · rn−1 M = rn M . Since n ≥ 1, taking r → 0 shows that a−n = 0.
2π
◦ Thus, all coe cients an with n < 0 in the Laurent expansion of f (z) are zero, so f has a removable
singularity at z0 .
• From the results (1) and (2) together, we can see that removable singularities are characterized by having f
remain bounded as we approach the singularity.
• Next, we study the behavior of poles:
• Theorem (Classi cation of Poles): Suppose that
P∞ z0 is an isolated singularity of f and f (z) has a convergent
n
Laurent expansion f (z) = n=−∞ an (z − z0 ) on the punctured disc 0 < |z − z0 | ≤ R.
1. If f has a pole at z = z0 , then f (z) is unbounded on the punctured disc 0 < |z − z0 | ≤ R and in fact
limz→z0 |f (z)| = ∞. Moreover, the pole has order k if and only if limz→z0 (z − z0 )k f (z) exists and is
nonzero.
◦ limz→z0 |f (z)| = ∞ when f has a pole at z = z0 is the origin of the name pole, since
The fact that
if we plot the surfacez 0 = |f (x0 + iy 0 )| in a 3-dimensional coordinate system, the graph stretches up
0 0
to +∞ as x + iy → z0 (i.e., the surface looks like it has a pole holding it up).
◦ Proof: By hypothesis, only nitely many of the terms an with n < 0 are nonzero. P∞ Therefore, assuming
n
the pole order is k , then the Laurent expansion for f is of the form f (z) = n=−k an (z − z0 ) where
a−k 6= 0.
P∞
◦ Then for 0 < P |z − z0 | ≤ R, we see that f (z) = (z − z0 )−k [ n=0 an−k (z − z0 )n ] = (z − z0 )−k g(z)
∞ n
where g(z) = n=0 an−k (z − z0 ) .
◦ Note that g(z) has a removable singularity at z = z0 by (1), hence is holomorphic for |z − z0 | ≤ R.
Note also that g(z0 ) = a−k is nonzero by assumption.
−k −k
◦ Thenlimz→z0 |f (z)| = limz→z0 |z − z0 | |g(z)| = |g(z0 )| limz→z0 |z − z0 | = ∞ by the continuity
of g at z = z0 , that |g(z0 )| is positive, and that the quantity |z − z0 | is a positive real number that
tends to 0 as z → z0 .
◦ For the last statement, we have limz→z0 (z − z0 )k f (z) = limz→z0 g(z) = g(z0 ) = a−k which is
d
nonzero by hypothesis. On the other hand, from this calculation we see limz→z0 (z − z0 ) f (z) =
d−k
a−k limz→z0 (z − z0 ) , and for d − k < 0 the limit does not exist while for d − k > 0 the limit is
d
zero. Thus limz→z0 (z − z0 ) f (z) only converges to a nonzero value when d = k is the pole order of
f.
2. The function f has a pole of order k at z = z0 if and only if there exists a holomorphic function g(z) on
g(z)
the disc |z − z0 | ≤ R with g(z0 ) 6= 0 and f (z) = for all 0 < |z − z0 | ≤ R.
(z − z0 )k
12
P∞
◦ Proof: If f has a pole of order k , then the Laurent expansion for f is of the form f (z) = n=−k an (z−
z0 )n where a−k 6= 0.
P∞
◦ If as in (2) we then take g(z) = n=0 an−k (z − z0 )n , then g is holomorphic on the disc, g(z0 ) =
g(z)
a−k 6= 0, and f (z) = .
(z − z0 )k
g(z)
◦ Conversely, if f (z) = for a holomorphic g , then g has a power series expansion g(z) =
P∞ (z − z0 )k
n
n=0 bn (z − z0 ) where by hypothesis b0 = g(z0 ) 6= 0.
g(z) P∞
◦ Then we have the Laurent expansion f (z) = = n=−k bn+k (z − z0 )n = b0 (z − z0 )−k +
(z − z0 )k
b1 (z − z0 )1−k + · · · , which since b0 6= 0, means f has a pole of order k .
3. If f is holomorphic and has a zero of order k at z = z0 , then 1/f has a pole of order k at z0 .
◦ Proof: As noted in our discussion of zeroes of holomorphic functions, if f has a zero of order k at z0
then f (z) = (z − z0 )k g(z) h(z) with h(z0 ) 6= 0.
for a holomorphic
◦ From our results on analytic functions, since h(z0 ) 6= 0, the formal power series inverse g(z) = h(z)−1
is analytic (hence holomorphic) on a disc of positive radius centered at z0 .
1 1/h(z) g(z)
◦ Then = k
= , so by (3), 1/f has a pole of order k at z0 .
f (z) (z − z0 ) (z − z0 )k
4. Conversely, if f has a pole of order k at z = z0 then 1/f has a removable singularity at z0 , and removing
the singularity yields a function that has a zero of order k at z0 .
g(z)
◦ Proof: Suppose f has a pole of order k at z = z0 . Then by (3), we have f (z) = for a
(z − z0 )k
holomorphic function g(z) with g(z0 ) 6= 0. As in (4), since g(z0 ) 6= 0 it has a holomorphic inverse
h(z) on a disc of positive radius centered at z0 .
1
◦ Then = (z − z0 )k g(z)−1 = (z − z0 )k h(z), and so since 1/f is the product of two holomorphic
f (z)
functions for z 6= z0 , it has a removable singularity, and by our characterization of the order of a
zero, we see that 1/f has a zero of order k at z0 .
• These results show that zeroes and poles are two sides of the same proverbial coin, and are related simply by
taking reciprocals.
◦ Additionally, we can see that as we approach a pole of f, the value of |f (z)| tends uniformly to ∞, in
contrast to the behavior when approaching a removable singularity where f remains bounded.
◦ From our characterization of removable singularities, this should not be surprising, since f must be
unbounded when approaching a pole or essential singularity.
• One might imagine, then, that the behavior near an essential singularity might be similar to that of a pole
(e.g., that |f | will tend to ∞ while approaching the singularity).
◦ That is, however, not at all the case, as can be seen by the example of f (z) = e1/z .
◦ Indeed, solving e1/z = reiθ produces 1/z = ln r + iθ + 2πki so that z = 1/(ln r + iθ + 2πki). In particular,
as k→∞ these values approach zero.
◦ Thus, on any punctured disc around 0, no matter how small the radius, the values taken by f (z) include
every complex number of the form reiθ for r 6= 0: in other words, every nonzero complex number!
◦ It turns out that this behavior is typical of essential singularities:
• Theorem (Casorati-Weierstrass): Suppose that f is holomorphic for 0 < |z − z0 | ≤ R and z0 is an essential
singularity of f. Then for any 0 < r < R, the values of f taken on the punctured disc 0 < |z − z0 | < r are
dense in C.
◦ A dense set S is one with the property that for any z ∈ C there exist zi ∈ S such that limn→∞ zn = z :
in other words, every point in C is a limit point of the set S . Equivalently, for any z ∈ C and any > 0,
there exists a point z0 ∈ S such that |z − z 0 | < . Equivalently, the closure of the set S is all of C.
13
◦ For example, the points x + iy with x, y rational are a dense subset of C.
◦ An equivalent way of posing the result of Casorati-Weierstrass is that for any c ∈ C, there exists a
sequence {zn }n≥1 such that zn → z0 and f (zn ) → f (c) as n → ∞.
◦ Proof: Suppose otherwise, so that the values off on 0 < |z − z0 | < r are not dense in C. This means there
ζ ∈ C and some > 0 such that |f (z) − ζ| ≥ for all z in the punctured disc 0 < |z − z0 | < r.
exists some
1
◦ So now consider the function g(z) = : it is holomorphic on 0 < |z − z0 | < r since the denominator
f (z) − ζ
is never zero.
1 1
◦ Also, since |g(z)| = ≤ on the punctured disc, we see g is bounded hence has a removable
f (z) − ζ
singularity at z0 . Therefore, g(z) is actually holomorphic for |z − z0 | ≤ r . If it is nonzero at z0 , then
1 1
is holomorphic at z0 , while if g is zero at z0 then has a pole at z0 by our characterization of
g(z) g(z)
poles.
1
◦ But then f (z) = ζ + either has a removable singularity at z0 or a pole at z0 . But this is a
g(z)
contradiction since z0 is an essential singularity of f.
• In fact, the result of Casorati-Weierstrass can be very substantially strengthened:
• Theorem (Picard's Big Theorem): Suppose that f is holomorphic for 0 < |z − z0 | ≤ R and z0 is an essential
singularity of f. Then for any 0 < r < R, the values of f taken on the punctured disc 0 < |z − z0 | < r include
every complex number, with at most one exception.
◦ This theorem is quite di cult and we do not include a proof. We will, however, mention that this result
is connected with a number of other results, some connections among which we can describe.
• To motivate the rst, we observe that the image of a nonconstant entire function must be dense in C.
◦ To see this, if the image is not dense, then there exists some ζ ∈ C and some > 0 such that |f (z) − ζ| ≥
for all z ∈ C.
1
◦ Then by the same argument as in the proof above, the function g(z) = is entire and bounded
f (z) − ζ
above in absolute value by 1/, so by Liouville's theorem it is constant.
◦ Another result of Picard generalizes this result, in the same manner as Casorati-Weierstrass:
• Theorem (Picard's Little Theorem): Suppose f is a nonconstant entire function. Then the image of f includes
every complex number, with at most one exception.
◦ Equivalently, if f is an entire function that omits two or more values from its image, then f is constant.
◦ The allowance of one exception is certainly necessary, since f (z) = ez never takes the value 0.
◦ There are several approaches to proving Picard's little theorem, but Picard's original approach was rst
to construct the elliptic modular function λ, which is essentially an explicit covering map of C\{0, 1} by
the unit disc D.
◦ Then if f is entire and omits at least two values from its image, by rescaling and translating we can
assume the two omitted values are 0 and 1. The next part (requiring all of the hard work!) is to show
that if f : C → C\{0, 1} is entire, then f is the composition of a function g:C→D and the modular
function λ : D → C\{0, 1}; intuitively, the reason that this composition exists is that λ is a covering
map.
◦ But then g is entire and its image is bounded (since it is contained in the unit disc), so by Liouville's
theorem, g is constant, and thus f is also constant.
• We can see that essential singularities tend to have rather odd analytic properties, and in particular, cannot
merely be obtained as a quotient of holomorphic functions.
◦ For this reason, we often restrict attention to functions that are more well-behaved near their singularities.
14
◦ Since removable singularities can be ignored by rede ning the function properly, we are then mostly
interested in poles.
• De nition: Let R be a region. We say a function f (z) is meromorphic on R if f is holomorphic on R except
for a set of isolated singularities all of which are poles (or removable).
f (z)
◦ If f (z) and g(z) are both holomorphic on R and g is not identically zero, then the quotient is
g(z)
meromorphic on R: this quotient can only fail to be holomorphic when g(z) = 0 and this only occurs at
an isolated set of points by our results about zeroes of holomorphic functions.
ez ez + sin z sinh(z)
◦ So, for example, the functions , , and are all meromorphic on C.
z sin z + 2z z 3 tan z
◦ In fact, the converse of this observation holds as well: a meromorphic function on R is necessarily the
quotient of two holomorphic functions on R.
◦ It is clear that this statement is true locally: if h(z) is meromorphic and z0 is a pole, then as we showed,
f (z)
we can write h(z) = for some holomorphic function f. If h(z) has only nitely many poles,
(z − z0 )k
then by taking an appropriate common denominator, one obtains a quotient of holomorphic functions
representing h(z) on its entire domain.
◦ In the case h(z) has in nitely many poles, more work is required to construct an appropriate denominator
function, but it can be done with a suitable in nite product and a result known as the Weierstrass
factorization theorem.
4.3 Residues and Residue Calculus
• Now that we have established various useful preliminaries, we turn our attention to using these results to
evaluate integrals.
◦ The general idea is to combine all of the facts we have accumulated about Cauchy's integral formula,
winding numbers, and meromorphic functions to give a general integration formula.
4.3.1 Residues, The Residue Theorem
• So suppose that f is meromorphic on a simply connected region
´ R and γ is a closed contour: we would like
to give as simple a formula for the value of
γ
f (z) dz as we can.
◦ First, since γ is closed, it contains only nitely many singularities of f, which we may freely assume are
poles.
P∞
◦ Locally around each singularity z0 , we may express f as a convergent Laurent series f (z) = n=−k an (z−
z0 )n . As we have seen, the integral of f on a suitable contour (inside the region of convergence) then
depends only on the winding number of γ around z0 and the coe cient a−1 .
◦ In essence, therefore, up to needing to be careful about deforming γ so that it lies inside the region of
convergence, we should be able to express the integral of f on γ in terms of the winding number of γ
around each singularity along with the coe cient a−1 at each singularity.
• To phrase all of this more conveniently, we introduce notation for this coe cient a−1 :
• De nition: Suppose f (z)
P∞ is meromorphic on an open region R and z0 ∈ R. If f (z) has a local Laurent
expansion f (z) = n=−k an (z − z0 )n at z0 , we de ne the residue of f at z0 , denoted Resf (z0 ), to be the
coe cient a−1 .
1 ´ f (z)
◦ From Cauchy's integral formula, if f is holomorphic on S\{z0 }, then we have Resf (z0 ) = dz
2πi γ z − z0
where γ is any contour inside S with winding number 1 around z0 .
◦ Trivially, if f is holomorphic at z0 , then the residue is zero: the only points where f can have a nonzero
residue are the poles of f.
15
1 1 ez
◦ Example: The residues of , , and at z = 0 are 1, 0, and 1 respectively.
z z2 z
1 1 ez 1
◦ Example: The residues of , , and at z = 1 are 1, , and e respectively.
z − 1 sin(πz) z−1 π
P∞
◦ In the particular case where f hasPa∞simple pole atn+1 z0 , we have f (z) = a−1 (z − z0 )−1 + n=0 an (z − z0 )n ,
and thus (z − z0 )f (z) = a−1 + n=0 an (z − z0 ) . As z → z0 all of the terms in the sum approach
zero, yielding Resf (z0 ) = a−1 = limz→z0 (z − z0 )f (z).
◦ Indeed, since (z − z0 )f (z)
has a removable singularity at z = z0 , we can really just view this limit as
being the natural value of (z − z0 )f (z) evaluated at z = z0 .
1
◦ Equivalently, if we write f (z) = then h has a simple zero at z = z0 , in which case limz→z0 (z −
h(z)
z − z0 1
z0 )f (z) = limz→z0 = 0 by L'Hôpital's rule.
h(z) h (z0 )
g(z)
◦ More generally, if f (z) = where g(z0 ) 6= 0 and h has a simple zero at z = z0 , then the residue of f
h(z)
g(z0 )
at z0 is by the same calculation.
h0 (z0 )
• Example: Find the residue of f (z) = ez / sin(z) at z = 0.
◦ sin(z) has a simple zero at z = 0 and e0 = 1, f (z) has a simple pole at z = 0. Using the formula
Since
z 0 0
above with g(z) = e and h(z) = sin(z) we see the residue is g(0)/h (0) = e / cos 0 = 1 .
2 1
◦ Indeed, the Laurent series for f (z) at z = 0 is z −1 + 1 + z + z 2 + · · · , and the coe cient of z −1 is
3 3
indeed 1.
• We can give a similar residue formula for poles of higher order:
• Proposition (Residue Calculations): Suppose f (z) is meromorphic on an open regionR and f has a pole of
1 dk−1
order k at z0 ∈ R. Then the residue of f at z0 is given by Resf (z0 ) = limz→z0 k−1 (z − z0 )k f (z) .
(k − 1)! dz
P∞ n
◦ Proof: By hypothesis f has a Laurent expansion of the form f (z) = n=−k an (z − z0 ) .
◦ Then (z − z0 ) f (z) = a−k + a1−k (z − z0 ) + · · · + a−1 (z − z0 )k−1 + a0 (z − z0 )k + · · ·
k
is holomorphic at z0
after removing the singularity.
dk−1
◦ Dierentiating term by term k−1 times yields
k−1
(z − z0 )k f (z) = (k − 1)!a−1 + k!a0 (z − z0 ) + · · · ,
dz
and now taking the limit as z → z0 (which here amounts merely to setting z = z0 ) yields (k − 1)!a−1 .
1
• Example: Find the residue of f (z) = at z=0 and at z = 3.
z 2 (z − 3)3
◦ From the expression given we see f (z) has a pole of order 2 at z = 0.
1 1 d 2 −3
◦ Applying the residue calculation formula we have z 2 f (z) = so that [z f (z)] = .
(z − 3)3 1! dz (z − 3)4
1
Then setting z=0 yields Resf (0) = − .
27
◦ Likewise, f has a pole of order 3 at z = 3.
1 1 d2 3
◦ Applying the residue calculation formula yields (z − 3)3 f (z) = , so that [(z − 3)3 f (z)] = 4 .
z2 2! dz 2 z
1
Setting z=3 yields Resf (3) = .
27
• Theorem (Residue Theorem): Suppose that R is a bounded simply connected region and f is a meromorphic
function on
´R with polesz1 ,P
z2 , . . . , zn in R. Then for any closed contour γ not passing through any of the
n
zi , we have
γ
f (z) dz = 2πi k=1 Wγ (zk )Resf (zk ).
16
◦ The idea of the proof is simply to subtract o all of the negative-power terms from the local Laurent
expansions at each pole, resulting in a holomorphic function that necessarily integrates to zero around
γ. Then each of the remaining nite series can just be integrated directly.
◦ Proof: For each k with 1 ≤ k ≤ n, let sk (z) be the negative-power part of the Laurent series expansion
for f centered at z = zk .
◦ Then each sk is holomorphic for all z 6= zk , and f (z) − sk (z) has a removable singularity at zk (since we
have removed all of the negative-power terms from that local Laurent series).
Pn
◦ This means f (z) − k=1 sk (z) has removable singularities at all of the zk , so after removing the singu-
larities, it is holomorphic on R.
´ Pn ´
◦ Then by Cauchy's integral theorem, we have [f (z)− k=1 sk (z)] dz = 0, so rearranging yields f (z) dz =
Pn ´ γ γ
k=1 s (z) dz .
γ k
◦ Then, as we have previously shown by direct integration and the de nition of the winding number, for
P−1 ´
sk (z) = n=−d an (z − zk )n we have γ sk (z) dz = 2πi · Wγ (zk ) · a−1 = 2πi · Wγ (zk ) · Resf (zk ).
´
◦ Plugging in for each s (z) dz and summing immediately yields the desired formula.
γ k
• In the particular special case where γ is the counterclockwise boundary of the simply connected region R
(which in practice is the situation we are usually concerned with), the residue theorem has the following form:
• Corollary (Cauchy's Residue Theorem): Suppose that R is a bounded simply connected region with coun-
terclockwise boundary
´ Pn γ , and f is a meromorphic function on R with poles z1 , z2 , . . . , zn in R. Then
γ
f (z) dz = 2πi k=1 Res f (zk ).
◦ We emphasize here that the only residues that contribute to the sum are those in R, namely, on the
interior of γ.
• The most direct application of the residue theorem is to compute contour integrals of meromorphic functions.
´ 2e2z − z
• Example: Evaluate
γ
dz where γ is the counterclockwise circle |z| = 3.
(z − 2)(z − 4)
◦ The integrand f (z) is meromorphic with simple poles at z=2 and z = 4.
◦ ´However, only the rst of these poles lies inside the circle |z| = 3, so by the residue theorem, we see
γ
f (z) dz = 2πi · Resf (2).
2e2z − z
◦ To compute the residue at 2 we compute (z − 2)f (z) = and then evaluate at z=2 to obtain
z−4
2
2e − 2
Resf (2) = = 1 − e2 . Thus, the value of the integral is 2πi(1 − e2 ) .
−2
´ ez
• Example: Evaluate
γ
dz where γ is the counterclockwise circle |z| = 6 followed by the boundary of
(z 2 + 1)2
the counterclockwise upper semicircle.
◦ The integrand f (z) is meromorphic with poles of order 2 at z=i and z = −i.
´
◦ Since the contour winds around z = i twice and z = −i once, by the residue theorem we have γ
f (z) dz =
2πi · [2Resf (i) + Resf (−i)].
d d 1 −2
◦ To compute the residue at i we compute [(z − i)2 f (z)] = [ 2
]= and then evaluate
dz dz (z + i) (z + i)3
−2 i
at z=i to obtain
3
=− .
Resf (i) =
(2i) 4
d d 1 −2
◦ Likewise, for the residue at −i, we have [(z + i)2 f (z)] = [ ]= and then evaluate at
dz dz (z − i)2 (z − i)3
−2 i
z = −i to obtain Resf (i) = = .
(−2i)3 4
◦ Thus, the value of the integral is 2πi · [2(−i/4) + i/4] = π/2 .
17
´ 1
• Example: Evaluate
γ
dz where γ is the counterclockwise boundary of the rectangle with vertices
ez −1
±1 − iπ and ±1 + 5iπ .
◦ The integrand f (z) is meromorphic with poles at z = 2πik for each integer k.
◦ The poles with
´k = 0, 1, and 2 lie inside the rectangle while the others lie outside, so by the residue
theorem we see
γ
f (z) dz = 2πi · [Resf (0) + Resf (2πi) + Resf (4πi)].
z − 2πik
◦ For each residue, we compute (z − 2πik)f (z) = and then take the limit as z → 2πik to obtain
ez − 1
z − 2πik 1
Resf (2πik) = limz→2πik = limz→2πik = 1 by L'Hôpital's rule (or equivalently, the reciprocal
ez − 1 ez
of the de nition of the derivative).
´ 1
◦ Since each residue is 1, we see
γ
dz = 6πi .
ez −1
• As an application of the residue theorem, we can give a procedure for evaluating trigonometric integrals
3 of
´ 2π
the form
0
r(cos θ, sin θ) dθ where r is a rational function.
eiθ + e−iθ eiθ − e−iθ iθ
◦ Explicitly, since cos θ = and sin θ =
, if we instead substitute z = e then the
2 2i
´ 2π ´ z+z −1
z−z −1
1
integral
0
r(cos θ, sin θ) dθ becomes γ r( , ) · dz where γ is the unit circle.
2 2i iz
z + z −1 z − z −1 1
◦ Thus, setting f (z) to be the rational function f (z) = r( , ) · , we wish to calculate the
2 2i iz
integral of f (z) around the unit circle. As an immediate application of the residue theorem, we see that
this integral equals 2πi times the sum of the residues of f (z) at each of its poles inside the unit circle.
´ 2π 1
• Example: Evaluate
0
dθ.
2 + cos θ
z + z −1 z − z −1 1 −2i
◦ Using the method described above we calculate , )·
f (z) = r(= 2 which
√ 2 2i iz z + 4z
√+ 1
has simple poles at z = −2 ± 3. The only one of these inside the unit circle is z = −2 + 3, and the
√
−2i i
residue of f there is limz→(−2+ 3)
√
= −√ .
2z + 4|z=−2+ 3 3
´ 2π 1 ´ i 2π
◦ Hence by the residue theorem we see that
0
dθ = γ f (z) dz = 2πi · (− √ ) = √ .
2 + cos θ 3 3
´ 2π 1
• Example: Evaluate dθ.
0
1 + 4 sin2 θ
z + z −1 z − z −1 1 iz
◦ Using the method described above we calculate f (z) = r( , )· = 2 4
which
√ 2 2i iz 1 − 3z
√ +z
±1 ± 5 1− 5
has simple poles at z = . Two of these lie inside the unit circle: z0 = ± , at each of
2 2
iz0 i i
which the residue is
3 = 2
=− √ .
−6z0 + 4z0 −6 + 4z 2 5
´ 2π 1 ´ i 2π
◦ Hence by the residue theorem we see that
0
dθ = γ f (z) dz = 2πi · 2(− √ ) = √ .
2 + cos θ 2 5 5
3 We 2 dt
also remark that there is a standard substitution t = tan(θ/2) often called the Weierstrass substitution, which has dθ = ,
1 + t2
1 − t2 2t
cos θ = ,
and sin θ = that allows inde nite integrals of this form to be evaluated by converting them into rational functions
1 + t2 1 + t2
of t. The reader may nd it enlightening to consider the similarities and dierences between the Weierstrass substitution and our
method using residue calculus.
18
4.3.2 Calculating De nite Integrals via Residue Calculus: Circular Contours
• One of the most classical and (perhaps) unexpected applications of residue calculus is its use in evaluating
de nite integrals on the real line that resist other techniques.
◦ Rather than diving immediately into interesting examples, we will illustrate the ideas with an integral
that can be easily computed using standard calculus methods.
´∞ 1
• Example: Evaluate
−∞
dx.
x2 + 1
´ 1
◦ By de nition, this integral is the limit limR→∞ γ1
dz where γ1 is the line segment along the real
z2 + 1
axis from −R to R.
◦ In order to apply the residue theorem we need a closed contour, so we close this contour by adding in
the contour γ2 traversing the upper half of the circle |z| = R counterclockwise from R to −R:
1
◦ By the residue theorem applied to f (z) = on this closed contour γ = γ1 ∪ γ2 , since f (z) has
z2 +1
single simple poles at z=i R) and z = −i (outside the contour), we have
(inside the contour for large
´ z+i 1
γ
f (z) dz = 2πi · Resf (i) = 2πi · limz→i 2 = 2πi · = π.
z +1 2i
´ ´
◦ Thus, we see limR→∞ [ γ1 f (z) dz + γ2 f (z) dz] = π , so to evaluate the integral we are after, we just need
´
to nd limR→∞ f (z) dz .
γ2
1 1
◦ On γ2 we have |f (z)| = ≤ 2 by the triangle inequality, so applying the arclength estimate
|z 2 + 1| R −1
´ πR
to γ2 we have
γ2
f (z) dz ≤ 2 → 0 as R → ∞.
R −1
´ ´∞ 1 ´ 1
◦ Thus, limR→∞ γ2 f (z) dz = 0, and so we obtain −∞ 2 dx = limR→∞ γ1 2 dz = π .
x +1 z +1
◦ Remark: As we should expect, this calculation agrees with the result obtained directly via the funda-
´∞ 1
mental theorem of calculus:
−∞ x2 + 1
dx = tan−1 (x)|∞
x=−∞ = (π/2) − (−π/2) = π .
• In general, the technique consists of complexifying the integrand in some manner, closing the contour,
making a residue calculation, and then estimating or otherwise dealing with the integral along the added
pieces of the contour.
´∞ 1
• Example: Evaluate
−∞
dx.
(x2 + 1)5
´ 1
◦ This integral is limR→∞ γ1
dz where γ1 is the line segment along the real axis from −R to R.
(z 2 + 1)5
As above, we close this contour by adding in the contour γ2 traversing the upper half of the circle |z| = R
counterclockwise from R to −R:
19
1
◦ By the residue theorem applied to f (z) = on this closed contour γ = γ1 ∪ γ2 , since f (z) has
(z 2 + 1)5
poles of order 5 at
´ z = i (inside the contour for large R) and z = −i (outside the contour), we have
γ
f (z) dz = 2πi · Res f (i).
1 d4 1 1 5·6·7·8 35i
◦ The residue is given by limz→i 4 [(z − i)5 2 5
] = limz→i 9
=− .
4! dz (z + 1) 4! (z + i) 256
1 1
◦ On γ2 we have |f (z)| = 5 ≤ (R2 − 1)5 by the triangle inequality, so the arclength estimate yields
2
|z + 1|
´ πR
γ2
f (z) dz ≤ → 0 as R → ∞.
(R2 − 1)5
´ ´∞ 1 35i 35π
◦ Thus, limR→∞ γ2
f (z) dz = 0, and so we obtain
−∞
dx = 2πi · (− )= .
(x2 + 1) 5 256 128
◦ Remark: One may compute this integral directly using standard real-valued calculus techniques (namely,
´ π/2
substituting x = tan u to obtain −π/2 cos8 u du and then reducing using double-angle identities, which
does indeed yield 35π/128) but it is a far messier computation.
• In both of the examples so far, the integral along a component of the contour tends to zero as the contour
grows large.
◦ In general, we can see that ifp(z) = ad z d + · · · + a0 is any polynomial of degree d ≥ 2 and γ2 is the upper
p(z)
semicircle from R to −R, then since lim|z|→∞ = |ad |, for su ciently large R with z on γ2 we see
zd
|p(z)| ≥ cRd for any constant c with 0 < c < |ad |.
´ 1 2π 1−d
◦ Therefore, we obtain an arclength estimate γ2 p(z) dz ≤ 2πR · d
= R → 0 as R → ∞.
cR c
◦ More generally, if f (z) is any function such that there exist constants c > 0 and > 0 such that
|f (z)| ≤ cR−1− for |z| = R (equivalently, if |f (z)| = O(R−1− ) in big-oh notation), we obtain a similar
´
arclength estimate
γ2
f (z) dz → 0 as R → ∞ along any portion γ2 of the circle |z| = R.
´∞ 1
• Example: Evaluate
−∞
dx.
x4 +4
´ 1
◦ This integral is limR→∞ γ1
dz where γ1 is the line segment along the real axis from −R to R. As
z4 +4
above we take γ2 to be the upper semicircle from −R to R and γ = γ1 ∪ γ2 as R → ∞.
4
◦ The integrand has simple poles at the four values of z with z + 4 = 0, which are z = ±1 ± i, and only
z = 1 − i and z = 1 + i lie inside γ . Since both poles are simple, with g(z) = 1/f (z) = 4 + z 4 , the residues
1 −1 − i 1 1−i
are Resf (1 + i) = = and Resf (−1 + i) = = .
0
g (1 + i) 16 0
g (−1 + i) 16
´
◦ By our estimates, since 4 + z 4 has degree 4, on γ2 we have γ2 f (z) dz → 0 as R → ∞. Thus by the
´∞ 1 ´ 1 π
residue theorem, we see
−∞ x4 + 4
dx = γ 4 dz = 2πi[Resf (1 + i) + Resf (1 − i)] = .
z +4 8
1 (−x + 2)/8 (x + 2)/8
◦ Remark: Using partial fractions one may decompose = 2 + 2 and then
x4 + 4 x − 2x + 2 x + 2x + 2
evaluate the inde nite integral: but again, this computation is quite messy.
20
• In some cases, we cannot obtain a suitable estimate directly, and must resort to changing the function.
´∞ cos x
• Example: Evaluate
−∞
dx.
x2 + 1
´ cos z
◦ If we apply the method used so far, we write limR→∞ γ1
dz where γ1 is the line segment along
x2 + 1
the real axis from −R to R and then take γ2 to be the upper semicircle from −R to R and γ = γ1 ∪ γ2
as R → ∞.
cos z
◦ However, this time, we run into a problem: the function on the semicircle γ2 no longer has a
z2 + 1
ix −y −ix y
cos z e e +e e ix −y
convenient estimate, since in general = for z = x + iy . We have e e =
z2 + 1 z2 + 1
−y −ix y y R
e ≤ 1, but the other term e e = e can be very large (up to e ) on γ2 : this is problematic since
´ cos z 1 + eR
it gives an arclength estimate
γ1 x 2 + 1
dz ≤ πR which does not go to zero as R → ∞.
R2 − 1
cos x eix ´ ∞ cos x
◦ What we can do instead is observe that
2
= Re[ 2 ] when x is real, and so −∞ 2 dx =
x +1 x +1 x +1
´∞ e ix
Re[ −∞ 2 dx].
x +1
eiz
◦ Therefore, we try taking take the function f (z) = and integrating around γ = γ1 ∪ γ2 . This
z2 + 1
function has simple poles at z = −i and at z = i, only the latter of which lies inside γ . The residue at
eiz i
z = i is limz→i (z − i)f (z) = limz→i =− .
z+i 2e
eiz 1 ´
◦ Since on γ2 we have |f (z)| =
2
≤ 2 , our estimates imply
γ2
f (z) dz → 0 as R → ∞. Thus
|z + 1| R −1
´∞ eix ´ eiz i π
by the residue theorem, we see
−∞ x2 + 1
dx = γ z2 + 1
dz = 2πi[Resf (i)] = 2πi · (− ) = .
2e e
´ ∞ cos x π
◦ Taking the real part then shows
−∞ x2 + 1
dx = .
e
´ cos x
◦ Remark: Unlike the previous examples, in this case the inde nite integral dx is non-elementary,
x2 + 1
and thus cannot be evaluated using typical calculus techniques.
• In general, using a semicircular contour like the ones in the examples above will be eective for any meromor-
phic function f (z) that has only nitely many poles and decreases su ciently rapidly as |z| → ∞.
1+
◦ More precisely, if there exist positive constants A and such that |f (z)| ≤ A/ |z| for su ciently large
|z|, then the integral of f (z) on the semicircle γ2 will tend to zero as R → ∞.
◦ Then, assuming f has only nitely many poles
´∞ Pn z1 , . . . , zk in the upper half-plane, the residue theorem
yields immediately that
−∞
f (x) dx = 2πi k=1 Resf (zk ).
4.3.3 Calculating De nite Integrals via Residue Calculus: Circular Contours With Detours
• In many other cases, we must resort to contours more complicated than a semicircle. In such situations often
we will end up with several components that may all contribute to the value of the integral we are seeking.
◦ The computations are often motivated by trying to evaluate the integral of a function similar to the one
we seek to integrate on the real line, but nding the right choice of contour and function is as much an
art form as a science.
´∞ 1
• Example: Evaluate
0
dx.
x3 + 1
1
◦ Here, it is sensible to try f (z) = , but we cannot use the upper semicircle as our contour because
z3 + 1
we only want to integrate from 0 to ∞ on the real axis.
21
◦ Instead, we can exploit the rotational symmetry of the function (namely, that f (e2πi/3 z) = f (z)) by
using a smaller portion of the circle.
◦ Explicitly, take γ = γ1 ∪ γ2 ∪ γ3 where γ1 is the segment from 0 to R, γ2 is the counterclockwise arc of
|z| = R from R toRe2πi/3 , and γ3 is the segment from Re2πi/3 back to 0:
´R
1 ´∞ 1 ´
◦ Letting IR = 0
dx and I = 0 dx, by parametrizing the segments we see γ1 f (z) dz =
x3 + 1 x3 + 1
´R 1 ´ ´R 1
0 t3 + 1
dt = IR and γ3 f (z) dz = − 0 2πi/3
· e2πi/3 dt = −e2πi/3 IR .
(te )3 + 1
´
◦ Additionally, since f (z) is a polynomial of degree 3, we see that γ2 f (z) dz → 0 as R → ∞.
◦ The function f (z) z = eiπ/3 , eiπ , e5iπ/3 , and only the rst one lies inside γ , with
has simple poles at
z − eiπ/3 1
residue given by limz→eiπ/3
3
= 2πi/3 = e−2πi/3 /3 via L'Hôpital's rule.
z +1 3e
´ √
◦ Then by the residue theorem, we have γ f (z) dz = 2πi · Resf (eiπ/3 ) = π(−3i 3 − 3i). Taking R → ∞
2iπ/3
´ ´ ´ ´
produces 2πi · 3e = γ f (z) dz = γ1 f (z) dz + γ2 f (z) dz + γ3 f (z) dz = I + 0 + (−e2πi/3 )I .
2πi · e−2πi/3 /3 2π
◦ Solving for I yields I= 2πi/3
= √ .
1−e 3 3
´ ∞ sin x
• Example: Evaluate the sinc integral
0
dx.
x
sin z
◦ First we remark that this integral converges near 0 since has a removable singularity at 0 as we have
z
´ sin x cos x ´ cos x
previously seen. It also converges as x→∞ via integration by parts: dx = − 2 − dx
x x x2
and the latter integral is absolutely convergent as x → ∞.
◦ In order to use a circular contour, we must change the function, since as we saw above, sin z does not
obey the necessary bound on |z| = R.
eiz eix sin x
◦ So we will again try using the function f (z) = , since Im[ ]= for real x.
z x x
◦ However, f (z) has a pole at z = 0, so we cannot take a contour that passes through 0. So instead, we
will take a contour that detours a small amount around zero: explicitly, take γ = γ1 ∪ γ2 ∪ γ3 ∪ γ4
where γ1 is the segment from 1/R to R, γ2 is the upper semicircular arc of radius R from R to −R, γ3
is the segment from −R to −1/R, and γ4 is the upper semicircular arc from −1/R to 1/R:
´ ´ ´ ´
◦ Then since f has no poles inside γ we see γ f (z) dz = 0 = γ f (z) dz = γ1 f (z) dz + γ2 f (z) dz +
´ ´
γ3
f (z) dz + γ4 f (z) dz . Now we calculate the integral on each piece.
22
´ ´ R eit
◦ On γ1 we have
γ1
1/R t
dt.
f (z) dz =
´ ´ π eiR(cos t+i sin t) ´π
◦ On γ2 we have γ2 f (z) dz = 0 · iReit dt = i 0 eiR cos t−R sin t dt which is bounded above
´ π −R sin t ´Re
π
it
in absolute value by
0
e dt ≤ 0 e−Rt dt = (1 − e−πR )/R, which goes to zero as R → ∞.
´ ´ −1/R eit ´ R e−it
◦ On γ3 we have
γ3
f (z) dz = −R
dt = − 1/R dt upon negating t.
t t
◦ Finally, on γ4 (noting that the path is traversed clockwise, so we must negate the integral), we have
´ ´ π ei(cos t+i sin t)/R i it ´π
γ4
f (z) dz = −
0
· e dt = −i 0 e(− sin t+i cos t)/R dt. As R → ∞ the integrand tends
eit /R R
0
uniformly to e = 1, so the integral approaches −iπ .
´ ´ ´ R eit ´ R e−it ´ R 2i sin t
◦ In particular we can compute γ1 f (z) dz + γ3 f (z) dz = 1/R dt − 1/R dt = 1/R dt, and
t t t
´ ∞ sin t
so as R → ∞ this sum approaches 2i dt.
0 t
´ ´ ´ ´ ´ ∞ sin t
◦ Thus, taking R → ∞ in γ1 f (z) dz + γ2 f (z) dz + γ3 f (z) dz + γ4 f (z) dz = 0 produces 2i 0 dt +
t
´ ∞ sin t π
0 + (−iπ) = 0, from which we conclude 0 dt = .
t 2
◦ Remark: Using the entire semicircle, which at rst seems like the natural approach for evaluating
´ ∞ sin x
−∞ x
dx (which is twice our desired integral by symmetry of the integrand), in fact allows us to
eiz
nd a convenient cancellation when integrating along the two components of γ on the real axis.
z
Our selection of a symmetric contour is precisely what allows us to sum the contributions and thereby
eix
cancel the pole at z = 0 (which only causes divergence in the integral for the real part of as x → 0).
x
• In the example above, we used a contour that took a small detour around one of the poles of the function,
and obtained an integral along a semicircular arc around the pole whose value conveniently came out to be
−iπ .
◦ Roughly speaking, the contour is half of a full path enclosing the pole (the other half being the bottom
semicircle), and the corresponding integral comes out to be half of 2πi times the value of the winding
number around the pole (−1, since the circle was oriented clockwise).
◦ In fact, whenever we have a contour that detours along a circular arc around a simple pole, we obtain a
similar result:
• Lemma (Fractional Residues): Suppose f is meromorphic with a ´simple pole at z0 and γr is the circular arc
parametrized by γr (t) = z0 + reit for θ1 ≤ t ≤ θ2 . Then limr→0+ γr f (z) dz = i(θ2 − θ1 )Resf (c).
1
◦ The reason for the name of the lemma is that the circular-arc integral equals
2π (θ2 − θ1 ) times the
value obtained by integrating around a full circle: in other words, it behaves as if we are computing the
appropriate fractional residue based on what fraction of the circle we are including.
P∞
◦ Proof: By hypothesis, the local Laurent expansion for f (z) is f (z) = a−1 (z − z0 )−1 + n=0 an (z − z0 )n
0 < |z − z0 | ≤ R.
for
P∞ an P∞
◦ Then g(z) = n=0 an (z −z0 )n is holomorphic hence has an antiderivative (z −z0 )n+1
G(z) = n=0
´ n +1
so by the fundamental theorem of line integrals we see g(z) dz = G(γr (θ2 )) − G(γr (θ1 )). Then
´ γr
limr→0+ γr g(z) dz = limr→0 [G(γr (θ2 )) − G(γr (θ1 ))] = G(z0 ) − G(z0 ) = 0 since γr shrinks towards z0
and G is continuous.
´ ´ ´ ´ ´θ 1
◦ Now γr f (z) dz = γr a−1 (z −z0 )−1 dz + γr g(z) dz , and since γr a−1 (z −z0 )−1 dz = a−1 θ12 it ·ireit =
re
i(θ2 − θ1 )a−1 , taking r → 0+ yields the desired result.
´∞ sin 2x
• Example: Evaluate
−∞
dx.
x(x2 + 1)
23
◦ First we remark that this integral converges absolutely, since the integrand has a removable singularity
1
at 0 and is bounded by
3 for large |x|.
|x|
◦ As in the last examples, the sine function is not well behaved on the circle |z| = R so we instead work
e2iz eix sin x
with f (z) = and observe that Im[ ] = for real x.
z(z 2 + 1) x x
◦ As before we take a contour that detours around the pole at z = 0:
e2iz
◦ Then γ encloses the single simple pole at z = i of f (z) = at which the residue equals
z(z 2 + 1)
z−i e 2iz
1 e 1 −2 ´
limz→i 2
= · = − e−2 , so by the residue theorem we see
γ
f (z) dz = −πie−2 .
z +1 z 2i i 2
◦ Now we calculate the integral on each piece.
´ ´R e2it e2it ´∞
◦ On γ1 we have
γ1
f (z) dz = 1/R
dt dt as R → ∞.
which tends
0
t(t2 + 1) t(t2 + 1)
e2iR(cos t+i sin t) e−2R sin t 1
◦ On γ2 we have |f (z)| = it 2 2it
≤ 2 − 1)
≤ 2 − 1)
so since the arclength of γ2 is
´ (Re )(R e + 1) R(R R(R
πR we see that γ2 f (z) dz = O(R−2 ) → 0 as R → ∞.
´ ´R e−2it ´∞ e−2it
◦ On γ3 we have
γ3
f (z) dz = − 1/R
dt which tends − 0
dt as R → ∞.
t(t2 + 1) t(t2 + 1)
◦ On γ4 , since it is a clockwise semicircle tending to zero around the simple pole at z = 0 of f (z) at
which the residue equals limz→0 zf (z) = 1, by the fractional residues lemma the integral tends to −πi
as R → ∞.
´∞ e2it ´ ∞ e−2it
◦ So taking R → ∞ and putting all of this together yields −πie−2 = 0 2
dt+0− 0 dt−
t(t + 1) t(t2 + 1)
´ ∞ 2 sin 2t ´∞ sin 2t πi(1 − e−2 )
πi which simpli es to πi(1 − e−2 ) = 0 t(t2 + 1)
dt so that
0 2
dt = .
t(t + 1) 2
√
´∞ 5
x
• Example: Evaluate
0
dx.
x2 + 1
z 1/5 eLog(z)/5
◦ The most natural integrand is f (z) = = . However, this presents an obvious di culty:
z2 + 1 z2 + 1
namely, that Log(z) is not holomorphic on the interval [0, ∞), which is precisely the path we want to
integrate along!
◦ But we can easily rectify this issue by selecting a dierent branch cut of the logarithm whose branch
cut is not along the real axis, but elsewhere, such as along the negative imaginary axis: namely, to take
log(z) with imaginary part in [−π/2, 3π/2) rather than [0, 2π).
log(z)/5
e
◦ So, with f (z) = , we see that f (z) has a pole at z=i and singularities along [0, −i∞): thus our
z2 + 1
contour must cut around zero. As above, we try a semicircular detour:
24
elog(z)/5
◦ Then γ encloses the single simple pole at z = i of f (z) = at which the residue equals
z2 + 1
z − i log(z)/5 e log(i)/5
e iπ/10 ´ eiπ/10
limz→i e = = , so by the residue theorem we see
γ
f (z) dz = 2πi · .
z2 + 1 2i 2i 2i
◦ Now we calculate the integral on each piece.
√
´ ´ R elog(t)/5 ´R 5
t
◦ On γ1 we have
γ1
f (z) dz = 1/R t2 + 1
dt = 1/R t2 + 1
dt which tends to our desired integral I as
R → ∞.
it
elog(Re )/5 R1/5
◦ On γ2 we have |f (z)| = ≤ = O(R−9/5 ) so since the arclength of γ2 is πR we see
(Reit )2 + 1 R2 − 1
´
that
γ2
f (z) dz = O(R−4/5 ) → 0 as R → ∞.
√
´ ´R elog(−t)/5 ´ R eiπ/5 5 t
◦ On γ3 we have
γ3
f (z) dz = 1/R
dt = 1/R 2 dt which tends to eiπ/5 I as R → ∞.
(−t)2 + 1 t +1
it
elog(e /R)/5 R−1/5
◦ On γ4 we have |f (z)| = it 2
≤ = O(R−1/5 ) so since the arclength of γ4 is π/R we
(e /R) + 1 1 − 1/R2
´
see that
γ4
f (z) dz = O(R−6/5 ) → 0 as R → ∞.
◦ So taking R → ∞ and putting all of this together yields πeiπ/10 = I + 0 − eiπ/5 I + 0 so that I =
iπ/10
πe π π
= −iπ/10 = .
1 + eiπ/5 e + eiπ/10 2 cos(π/10)
4.3.4 Calculating De nite Integrals via Residue Calculus: Other Contours
• There are many other integrals we can evaluate with a more inspired selection of contour. We illustrate with
some examples.
´∞ eax
• Example: For 0 < a < 1, evaluate
−∞
dx.
ex+1
◦ First, note that this integral converges (absolutely): as x → −∞ the integrand is asymptotic to e−a|x|
(a−1)x
while as x → +∞ the integrand is asymptotic to e , whose integrals both converge in the given
limit.
eaz
◦ A sensible choice of integrand is f (z) = . This function has simple poles at z = πi(1 + 2k) for
ez + 1
integers k , which will cause di culties if we integrate over a large semicircular contour since we will have
to sum the residues over all of the poles. This is not necessarily a problem, but a more serious issue is
that we do not get a good estimate for |f (z)| on the circle |z| = R, since the denominator can take very
small values (or even the value zero!) depending on the value of R.
◦ We can avoid these issues if instead we pick a rectangular contour enclosing just the pole at z = πi
that has one side extending far along the real axis. Furthermore, since the denominator is periodic with
period 2πi, while the numerator changes by a factor of e2πia 6= 1 upon increasing z by 2πi, if we select
the height of the rectangle to be 2πi, then the parallel side of the contour will be expressible in terms of
the original integral.
◦ So we take the contour γ = γ1 ∪ γ2 ∪ γ3 ∪ γ4 where γ1 is the segment from −R to R, γ2 is the segment
from R to R + 2πi, γ3 is the segment from R + 2πi to −R + 2πi, and γ4 is the segment from −R + 2πi
to R:
25
eaz
◦ Then γ encloses the single simple pole at z = πi of f (z) = at which the residue equals
ez + 1
z − πi az e aπi ´
limz→πi z e = d z = −eaπi , so by the residue theorem we see γ f (z) dz = 2πi(−eaπi ) =
e +1 dz (e ´ + 1)|z=πi
´ ´ ´ ´
γ
f (z) dz = γ1
f (z) dz + γ2 f (z) dz + γ3 f (z) dz + γ4 f (z) dz .
◦ Now we calculate the integral on each piece.
´ ´R eat
◦ On γ1 we have
γ1
f (z) dz = −R
dt which tends to our desired integral I as R → ∞.
et +1
´ ´ 2π ea(R+it) ea(R+it) eaR
◦ On γ2 we have
γ2
f (z) dz = 0 R+it
idt, and since R+it ≤ R → 0 as R → ∞ the
e +1 e +1 e −1
integral tends to 0.
´ ´R ea(t+2πi) ´ R eat
◦ On γ3 , noting the reversed orientation, we have γ3
f (z) dz = − −R
dt = −e2πia −R t dt,
e(t+2πi)+1 e +1
which tends to e−2πai I as R → ∞.
´ ´ 2π ea(−R+it) ea(−R+it) e−aR
◦ On γ4 we have
γ2
f (z) dz = 0 e−R+it + 1
idt, and since ≤ →0 as R →∞ the
e−R+it + 1 2
integral tends to 0.
−2πieaπi
◦ So taking R → ∞ and putting all of this together yields −eaπi = I +0−e2πia I +0 so that I = =
1 − e2aπi
−2πi −2πi π
= = .
e−aπi − eaπi −2i sin(aπ) sin(aπ)
´∞ 1
• Example: Evaluate
0
dx.
x2 + 3x + 2
1
◦ It seems obvious that we would want to select the function f (z) = and then include the real
z 2 + 3z + 2
interval [0, R] as a portion of our contour.
◦ However, unlike the examples above where we exploited a convenient property of the integrand (e.g.,
rotational symmetry, periodicity) to convert the integral along a segment returning to the origin to
something again involving the original integral I , here there is no obvious way to do that: neither a
rotation (reiθ )2 + 3(reiθ ) + 2 nor a translation (r + it)2 + 3(r + it) + 2 is expressible as a nice multiple
of r2 + 3r + 2.
Log(z)
◦ What we will do instead is to work with the function f (z) = . The main idea is that f (z) has
z 2 + 3z + 2
a jump discontinuity across the positive real axis that causes the values of f (z) to dier by 2πi across
the discontinuity, so if we integrate along a contour that passes just below the axis, its value will be 2πi
greater than the corresponding integral just above the axis.
◦ We also want to have a component of the contour along the circle of radius R (so that the integral on
that component tends to zero by the arclength estimate) and we must also avoid z = 0. Fitting all of
these pieces together eventually leads to a contour often called the keyhole contour due to its visual
similarity to a keyhole:
26
◦ Explicitly, we take γ = γ1 ∪ γ2 ∪ γ3 ∪ γ4 where γ1 is the segment from 1/R0 + i to R00 + i, γ2 is the
00 00 00 0
counterclockwise arc along |z| = R from R + i to R − i, γ3 is the segment from R − i to 1/R − i,
0 0 0
p
and γ4 is the counterclockwise arc along |z| = 1/R from 1/R − i to 1/R + i, where R =
√ 1/R − 2 ,
2
R00 = R2 − 2 , and is small relative to 1/R (e.g., = 1/R2 ).
Log(z)
◦ Then γ encloses the two simple poles at z = −1 and z = −2 of f (z) =
whose residues are
+ 3z + 2 z2
z+1 z+1
Resf (−1) = limz→−1 2 Log(z) = iπ and Resf (−2) = limz→−2 2 Log(z) = − ln 2 − iπ ,
z + 3z + 2 ´ z + 3z + 2
so by the residue theorem we see
γ
f (z) dz = 2πi[iπ − ln 2 − iπ] = −2πi ln 2.
◦ Now we calculate the integral on each piece.
´ ´ R00 Log(t + i) ´∞ ln(t)
◦ On γ1 we have
γ1
f (z) dz = 1/R0
dt, which tends to
0
dt as R→
(t + i)2 + 3(t + i) + 2 t2 + 3t + 2
∞.
log(Reit ) ln(R)
◦ On γ2 we have |f (z)| = ≤ 2 = O(R−2 ln R) so since the arclength of
´(Reit )2 + 3(Re it ) + 2 R − 3R − 2
γ2 is < 2πR we see that
γ2
f (z) dz = O(R−1 ln R) → 0 as R → ∞.
´ ´ R00 Log(t − i)
◦ On γ3 , noting the reversed orientation, we have
γ3
f (z) dz = − 1/R0
dt, which
(t − i)2 + 3(t − i) + 2
´ ∞ ln(t) + 2πi
tends to − 0 t2 + 3t + 2
dt as R → ∞.
log(eit /R) ln(R)
◦ On γ4 we have |f (z)| = it 2 it
≤ = O(ln R) so since the arclength of γ2 is
´ (e /R) + 3(e /R) + 2 2
< 2π/R we see that
γ2
f (z) dz = O(R−1 ln R) → 0 as R → ∞.
´∞ ln(t) ´ ∞ ln(t) + 2πi
◦ So taking R → ∞ and putting all of this together yields −2πi ln 2 = 0
dt− 0 2 dt =
t2 + 3t + 2 t + 3t + 2
´∞ 1 ´∞ 1
−2πi 0 2
dt so that 0 2 dt = ln 2 .
t + 3t + 2 t + 3t + 2
1
◦ Remark: Of course, this integral is very easy to evaluate using partial fraction decomposition: =
t2 + 3t + 2
1 1 ´∞ 1 t+1 ∞
− , so
0 2
dt = ln( ) = ln 2. But this technique of using the keyhole contour
t+1 t+2 t + 3t + 2 t+2 t=0
is applicable to a wide range of other integrals that are not so easy to compute in the same manner.
´∞ ln x
• Example: Evaluate
0
dx.
x2 + 3x + 2
Log(z)2
◦ We take f (z) = and integrate around the keyhole contour γ = γ1 ∪ γ2 ∪ γ3 ∪ γ4 used above:
z 2 + 3z + 2
Log(z)2
◦ As before γ encloses the two simple poles at z = −1 and z = −2 of f (z) =
whose residues
+ 3z + 2 z2
z+1 z+1
are Resf (−1) = limz→−1 Log(z)2 = (iπ)2 and Resf (−2) = limz→−2 2 Log(z)2 =
z 2 + 3z + 2 ´ z + 3z + 2
−(ln 2+iπ)2 , so by the residue theorem we see γ f (z) dz = 2πi[(iπ)2 −(ln 2+iπ)2 ] = 4π 2 ln 2−2πi(ln 2)2 .
27
◦ Now we calculate the integral on each piece.
´ ´ R00 Log(t + i)2 ´∞ ln(t)2
◦ On γ1 we have
γ1
f (z) dz = 1/R0
dt, which tends to
0
dt as R→
(t + i)2 + 3(t + i) + 2 t2 + 3t + 2
∞.
log(Reit )2 ln(R)2
◦ On γ2 we have |f (z)| = ≤ 2 = O(R−2 ln2 R) so since the arclength
(Reit )2
´ + 3(Re it ) + 2 R − 3R − 2
of γ2 is < 2πR we see that
γ2
f (z) dz = O(R−1 ln2 R) → 0 as R → ∞.
´ ´ R00 Log(t − i)2
◦ On γ3 , noting the reversed orientation, we have
γ3
f (z) dz = − 1/R0
dt, which
(t − i)2 + 3(t − i) + 2
´ ∞ (ln(t) + 2πi)2
tends to − 0
dt as R → ∞.
t2 + 3t + 2
log(eit /R)2 ln(R)2
◦ On γ4 we have |f (z)| = it 2 it
≤ = O(ln2 R) so since the arclength of γ2 is
´ (e /R) + 3(e /R) + 2 2
< 2π/R we see that
γ2
f (z) dz = O(R−1 ln2 R) → 0 as R → ∞.
´∞ ln(t)2
◦ So taking R → ∞ and putting all of this together yields 4π 2 ln 2 − 2πi(ln 2)2 = 0
dt −
t2 + 3t + 2
´ ∞ (ln(t) + 2πi)2 ´∞ 4π 2 ´ ∞ −2πi ln t
0 2
dt = 0 2 dt + 0 2 dt. So comparing the imaginary parts yields
t + 3t + 2 t + 3t + 2 t + 3t + 2
´∞ ln t
the desired
0 t2 + 3t + 2
dt = (ln 2)2 .
´∞ 1
◦ Remark: Notice that comparing the real parts yields 0 2 dt = ln 2, as we calculated above.
t + 3t + 2
• We make some remarks about the eectiveness of using various dierent contours for evaluating integrals
based on the shape and behavior of the function. For example, suppose we wish to evaluate an integral of the
´∞ p(x)
form
0
xa dx where p(x) and q(x) are polynomials and 0 < a < 1. (Such integrals arise often as Mellin
q(x)
transforms.)
◦ In order for this integral to converge, q(x) must have no roots in (0, ∞) so that there are no singularities
on the real axis, and deg(p) − deg(q) + a must be less than −1 so that the integral converges as x → ∞.
p(z)
◦ Then the keyhole contour will be eective in evaluating this integral using integrand f (z) = ea·Log(z) ,
q(z)
since the contributions on the small and large circles will both tend to zero, while the contributions on
the line segments running parallel to the real axis will dier by a multiplicative factor of e2πia 6= 1, so
they will not cancel.
◦ If instead we want to integrate a rational function directly, then including an extra factor of Log(z) as
we illustrated in examples above, will prevent the two integrals along the real axis from cancelling one
´∞ p(x)
another. A similar approach works for evaluating integrals of the form
0
(ln x)k dx. Depending on
q(x)
the precise nature of the cancellation that occurs, it may also be necessary to increase the power of ln x
by 1.
´ ∞ x1/2 ln x
• Example: Evaluate
0
dx.
x2 + 1
eLog(z)/2 Log(z)
◦ We take f (z) = and integrate around the keyhole contour γ = γ1 ∪ γ2 ∪ γ3 ∪ γ4 .
z2 + 1
eiπ/4 (iπ/2) π
◦ There are simple poles at z = ±i; the residue at z = i is = eiπ/4 while the residue at
2i 4
3iπ/4
e (3iπ/2) 3π
z = −i is = − e3iπ/4 , so by the residue theorem the integral around the full contour is
−2i 4
2
π
√ (1 + 2i).
2
28
◦ For large |z| we see the integral on |z| = R is O(R−1/2 ln R) → 0 and for small |z| the integral on z = 1/R
−3/2
is O(R ln R) → 0 also.
´ ∞ eln x/2 ln x ´ ∞ x1/2 ln x
◦ On the segment above the real axis the integral tends to
0
dx = 0
dx as R →
x2 + 1 x2 + 1
´∞ e (ln x+2πi)/2
(ln x + 2πi)
∞, while on the segment below the real axis the integral tends to −
0 2
dx =
x +1
´ ∞ x1/2 ln x ´ ∞ x1/2
0
dx + 2πi 0 x2 + 1
dx.
x2 + 1
π2 ´ ∞ x1/2 ln x ´ ∞ x1/2
◦ Comparing yields √ (1 + 2i) = 2 0 2
dx + 2πi 0 dx, so extracting real parts yields
2 x +1 x2 + 1
´ ∞ x1/2 ln x π2
0 2
dx = √ .
x +1 2 2
´∞
• As another class of examples, suppose we wish to evaluate a Fourier transform
−∞
fˆ(a) =
f (x)eiax dx of
some real-valued function f (x). By taking real and imaginary parts, we equivalently obtain the values of the
´∞ ´∞
Fourier coe cients
−∞
f (x) cos(ax) dx and −∞ f (x) sin(ax) dx.
◦ For integrals like these, using a rectangular contour with vertices ±R and ±R + iR will be eective as
long as there exists a positive constant A|f (z)| ≤ A/ |z| for all su ciently large |z|, since (as
such that
iaz
one may check using estimates like the ones we gave earlier) the integrals of g(z) = f (z)e on the other
three sides of the rectangle will then tend to zero.
◦ Then as long as g(z) (equivalently, f (z))´has only nitely many P poles z1 , z2 , . . . , zn in the upper half-plane,
∞ iax n Pn iazn
we obtain a general integration formula
−∞
f (x)e dx = 2πi k=1 Resg (zk ) = 2πi k=1 e Resg (zk ).
◦ This method can also be adjusted to work with a semicircular contour, and as we have seen it can also
be used in the situation where the function f (z) has a pole on the real axis: we simply have the contour
take a semicircular detour around the singularity and use the fractional residue lemma to account for
the associated contribution.
´∞ x sin 2x
• Example: Evaluate
−∞
dx.
(x2 + 1)2
ze2iz
◦ We take f (z) = and integrate around the rectangle with vertices ±R and ±R + iR.
(z 2 + 1)2
◦ There are poles at z = ±i, but only the pole at z = i lies inside the contour. The residue at z = i is
d d 2iz 1
e (z + i)−2 + 2ize2iz (z + i)−2 − 2ze2iz (z + i)−3 = e−2 , so by
limz→i [(z − i)2 f (z)] = limz→i
dz dz 2
−2
the residue theorem the integral on the contour is iπe .
◦ For z = ±R + it iR + t with −R ≤ t ≤ R, we see that |f (z)| = O(R−3 ) so the integral on each
and
component tends to 0 as R → ∞.
´∞ xe2ix
◦ On the real axis the integral tends to −∞ 2 dx as R → ∞.
(x + 1)2
´∞ xe2ix ´ ∞ x sin 2x
◦ Comparing yields −∞ 2 2
dx = iπe−2 , and extracting imaginary parts yields −∞ 2 dx =
(x + 1) (x + 1)2
π
.
e2
Well, you're at the end of my handout. Hope it was helpful.
Copyright notice: This material is copyright Evan Dummit, 2022. You may not reproduce or distribute this material
without my express permission.
29
You might also like
- The Case Against The Case For Christ - Robert M. Price56% (27)The Case Against The Case For Christ - Robert M. Price308 pages
- Academic Writing Genres - Essays, Reports & Other Genres (EAP Foundation Book 2)No ratings yetAcademic Writing Genres - Essays, Reports & Other Genres (EAP Foundation Book 2)535 pages
- Cheat-Sheet - Boas Mathematical Methods in The Physical Sciences67% (3)Cheat-Sheet - Boas Mathematical Methods in The Physical Sciences4 pages
- Stein and Shak Arch I Complex Analysis So LN100% (8)Stein and Shak Arch I Complex Analysis So LN52 pages
- Pool Activity Level (PAL) Instrument For Occupational ProfilingNo ratings yetPool Activity Level (PAL) Instrument For Occupational Profiling35 pages
- How To Be A Network Engineer in A Programmable Age PDF100% (1)How To Be A Network Engineer in A Programmable Age PDF38 pages
- Student Council Vice President: Zanna McclearyNo ratings yetStudent Council Vice President: Zanna Mccleary8 pages
- S.D. Jain Modern School: Vesu Char Rasta, U.M.Road, Surat 395007No ratings yetS.D. Jain Modern School: Vesu Char Rasta, U.M.Road, Surat 3950071 page
- Hydrological Modeling and Climate Change Impact Assessment Using HBVNo ratings yetHydrological Modeling and Climate Change Impact Assessment Using HBV13 pages
- MA 530 Complex Analysis: Review: Yingwei Wang100% (1)MA 530 Complex Analysis: Review: Yingwei Wang36 pages
- Section 43 (A) in A Vacuum: Cleaning Up False Advertising in An Unfair Competition MessNo ratings yetSection 43 (A) in A Vacuum: Cleaning Up False Advertising in An Unfair Competition Mess14 pages
- In Any Lighting Conditions: Easy-to-Read DisplayNo ratings yetIn Any Lighting Conditions: Easy-to-Read Display2 pages
- Saluting On The March: Name: Parth Anand Lalit Class: FybscNo ratings yetSaluting On The March: Name: Parth Anand Lalit Class: Fybsc10 pages
- 5 Applications of Cauchy's Integral Theorem(s) : For Continuous Functions F: D C We DefineNo ratings yet5 Applications of Cauchy's Integral Theorem(s) : For Continuous Functions F: D C We Define11 pages
- Tristan Needham - Visual Complex AnalysisNo ratings yetTristan Needham - Visual Complex Analysis15 pages
- Exercise Solutions To Complex Analysis: Thanks To The William's Work!No ratings yetExercise Solutions To Complex Analysis: Thanks To The William's Work!20 pages
- Complex Analysis: 1 The Field of Complex NumbersNo ratings yetComplex Analysis: 1 The Field of Complex Numbers3 pages
- Mathematics 113, Problem Set 3 Solutions: Alexander P. EllisNo ratings yetMathematics 113, Problem Set 3 Solutions: Alexander P. Ellis3 pages
- Cockroaches: Pictorial Key To Some Common Species: Harry D. PrattNo ratings yetCockroaches: Pictorial Key To Some Common Species: Harry D. Pratt8 pages
- The Hubble Constant Tension: Eleonora Di Valentino Dillon BroutNo ratings yetThe Hubble Constant Tension: Eleonora Di Valentino Dillon Brout674 pages
- Applications of Cauchy's Integral Theorem: and F A Function Holomorphic in B (ANo ratings yetApplications of Cauchy's Integral Theorem: and F A Function Holomorphic in B (A8 pages
- General Agreement On Trade and Tariff (GATT) /WTONo ratings yetGeneral Agreement On Trade and Tariff (GATT) /WTO21 pages
- Complex Differentiation: Umberto L. Hryniewicz April 4, 2023No ratings yetComplex Differentiation: Umberto L. Hryniewicz April 4, 20233 pages
- Holomorphic and Analytic Functions: December 18, 2014 9:14 9" X 6"No ratings yetHolomorphic and Analytic Functions: December 18, 2014 9:14 9" X 6"12 pages
- Lecture Notes in Physics: Founding Editors100% (3)Lecture Notes in Physics: Founding Editors402 pages
- Interlayer Valley Excitons in Heterobilayers of Transition Metal DichalcogenidesNo ratings yetInterlayer Valley Excitons in Heterobilayers of Transition Metal Dichalcogenides12 pages
- Exciton Dynamics in Suspended Monolayer and Few-Layer Mos 2D CrystalsNo ratings yetExciton Dynamics in Suspended Monolayer and Few-Layer Mos 2D Crystals9 pages
- Stein-Shakarchi Complex Analysis Solution Chapter 3 Meromorphic Functions and The LogarithmNo ratings yetStein-Shakarchi Complex Analysis Solution Chapter 3 Meromorphic Functions and The Logarithm7 pages
- Department of Electrical & Electronics EngineeringNo ratings yetDepartment of Electrical & Electronics Engineering25 pages
- Spongebob Squarepants (Theme Song) (Arr. Paul Lavender) - Snare Drum by Paul Lavender - Marching Band - Digital Sheet Music SHNo ratings yetSpongebob Squarepants (Theme Song) (Arr. Paul Lavender) - Snare Drum by Paul Lavender - Marching Band - Digital Sheet Music SH1 page
- 5 Cauchy Integral Formula Power and Laurent Series ExpansionNo ratings yet5 Cauchy Integral Formula Power and Laurent Series Expansion9 pages
