D.S. Morgan Mathematics

dsmorgan.com

Complex Analysis

Series, singularities and residues, worked through as examples.

Contents

Chapter 1: Cauchy’s Integral Formula and Liouville’s Theorem

An analytic function is pinned down by its values on a closed curve: Cauchy’s integral formula recovers \(f\) and all its derivatives inside from an integral around the edge. Bounding that integral gives Cauchy’s estimates, and from them Liouville’s theorem and the polynomial growth results.

1 Cauchy’s theorem

Theorem 1.1 (\(ML\)-inequality). If \(f\) is continuous on a contour \(\Gamma \) of length \(L\) and \(\abs {f(z)} \leq M\) for all \(z\) on \(\Gamma \), then

\[ \absStar{\int _\Gamma f(z) \dd z} \leq ML. \]

Source: [1, ch. 4]

Theorem page

Theorem 1.2 (Cauchy’s integral theorem). If \(f\) is analytic in a simply connected domain \(D\) and \(\Gamma \) is any closed contour in \(D\), then

\[ \oint _\Gamma f(z) \dd z = 0. \]

Source: [1, ch. 4]

Theorem page

Theorem 1.3 (Deformation of contours). Let \(C\) be a simple closed positively oriented contour, and let \(C_1, \dots , C_n\) be simple closed positively oriented contours inside \(C\) whose interiors are pairwise disjoint. If \(f\) is analytic on a domain containing all these contours and the region inside \(C\) but outside every \(C_k\), then

\[ \oint _C f(z) \dd z = \sum _{k=1}^{n} \oint _{C_k} f(z) \dd z. \]

Source: [1, ch. 4]

Theorem page

2 Cauchy’s integral formula

Theorem 2.1 (Cauchy’s integral formula). Let \(f\) be analytic in a simply connected domain \(D\), and let \(\Gamma \) be a simple closed positively oriented contour in \(D\). For every \(z_0\) inside \(\Gamma \),

\[ f(z_0) = \frac {1}{2\pi i} \oint _\Gamma \frac {f(z)}{z - z_0} \dd z. \]

Source: [1, ch. 4]

Proof. Let \(C_r\) be the circle \(\abs {z - z_0} = r\), with \(r\) small enough that \(C_r\) lies inside \(\Gamma \). The integrand is analytic everywhere except at \(z_0\), so by Theorem 1.3

\[ \oint _\Gamma \frac {f(z)}{z - z_0} \dd z = \oint _{C_r} \frac {f(z)}{z - z_0} \dd z = f(z_0) \oint _{C_r} \frac {\dd z}{z - z_0} + \oint _{C_r} \frac {f(z) - f(z_0)}{z - z_0} \dd z. \]

Parametrising \(z = z_0 + re^{it}\), \(0 \leq t \leq 2\pi \), the first integral is \(\int _0^{2\pi } i \dd t = 2\pi i\). Let \(E_r\) be the second. On \(C_r\) its integrand is at most \(\max _{C_r} \abs {f(z) - f(z_0)} / r\) in modulus, so by Theorem 1.1

\[ \abs {E_r} \leq 2\pi \max _{\abs {z - z_0} = r} \abs {f(z) - f(z_0)} \longrightarrow 0 \quad \text {as } r \to 0, \]

by continuity of \(f\) at \(z_0\). But \(E_r = \oint _\Gamma \frac {f(z)}{z - z_0} \dd z - 2\pi i f(z_0)\) does not depend on \(r\), so \(E_r = 0\).

Theorem page

Theorem 2.2 (Cauchy’s integral formula for derivatives). Under the hypotheses of Theorem 2.1, \(f\) has derivatives of every order in \(D\), each of them analytic in \(D\), and for every \(z_0\) inside \(\Gamma \)

\[ f^{(n)}(z_0) = \frac {n!}{2\pi i} \oint _\Gamma \frac {f(z)}{(z - z_0)^{n+1}} \dd z \qquad (n = 0, 1, 2, \dots ). \]

Source: [1, ch. 4]

Theorem page

Exercise 2.3. Let \(f(z) = \dfrac {\cos z}{(z^2 + 1)^2}\).

(a)
Find the singularities of \(f\).
(b)
Show that the integral of \(f\) around \(\abs {z} = 3\) is the sum of its integrals around small circles centred at the singularities.
(c)
Evaluate \(\oint _{\abs {z} = 3} f(z) \dd z\).

Solution. Singularities. \(\cos z = \frac 12(e^{iz} + e^{-iz})\) is entire and \((z^2 + 1)^2\) is a polynomial, so \(f\) is analytic except where \((z^2 + 1)^2 = 0\), that is, at \(z = \pm i\).

Decomposition. Let \(C_1\) and \(C_2\) be the positively oriented circles \(\abs {z - i} = \frac 12\) and \(\abs {z + i} = \frac 12\). They lie inside \(\abs {z} = 3\), their interiors are disjoint, and \(f\) is analytic everywhere else. By Theorem 1.3,

\[ \oint _{\abs {z} = 3} f(z) \dd z = \oint _{C_1} f(z) \dd z + \oint _{C_2} f(z) \dd z. \]

Around \(i\). Write \(f(z) = g_1(z) / (z - i)^2\) with \(g_1(z) = \cos z / (z + i)^2\), which is analytic on and inside \(C_1\). By Theorem 2.2 with \(n = 1\), the integral is \(2\pi i \, g_1'(i)\). Now

\[ g_1'(z) = \frac {-(z + i) \sin z - 2\cos z}{(z + i)^3}, \]

and \(\sin i = i \sinh 1\), \(\cos i = \cosh 1\), so

\[ g_1'(i) = \frac {-2i \cdot i \sinh 1 - 2\cosh 1}{(2i)^3} = \frac {2(\sinh 1 - \cosh 1)}{-8i} = \frac {-2e^{-1}}{-8i} = \frac {1}{4ie}. \]

Hence \(\oint _{C_1} f(z) \dd z = 2\pi i \cdot \frac {1}{4ie} = \frac {\pi }{2e}\).

Around \(-i\). Likewise, with \(g_2(z) = \cos z / (z - i)^2\),

\[ g_2'(-i) = \frac {-(-2i)(-i \sinh 1) - 2\cosh 1}{(-2i)^3} = \frac {2(\sinh 1 - \cosh 1)}{8i} = -\frac {1}{4ie}, \]

so \(\oint _{C_2} f(z) \dd z = -\frac {\pi }{2e}\).

Total.

\[ \oint _{\abs {z} = 3} \frac {\cos z}{(z^2 + 1)^2} \dd z = \frac {\pi }{2e} - \frac {\pi }{2e} = 0. \qedhere \]
\(\relax \vartriangleleft \)

Remark 2.4. The answer \(0\) can be predicted without computing. \(f\) is even, and the substitution \(z = -w\) maps the circle \(\abs {z} = 3\) onto itself with the same orientation, so \(\oint f(z) \dd z = \oint f(-w)(-\dd w) = -\oint f(w) \dd w\).

Exercise 2.5. Let \(\Gamma \) be a simple closed positively oriented contour, and let \(f\) be analytic inside and on \(\Gamma \). For \(z_0\) not on \(\Gamma \):

(a)
evaluate \(\oint _\Gamma \frac {f'(z)}{z - z_0} \dd z\) by Cauchy’s integral formula;
(b)
evaluate \(\oint _\Gamma \frac {f(z)}{(z - z_0)^2} \dd z\) by the formula for derivatives;
(c)
compare the two;
(d)
say what goes wrong when \(z_0\) lies on \(\Gamma \).

Solution. By Theorem 2.2, \(f'\) is analytic wherever \(f\) is.

\(z_0\) inside \(\Gamma \). Theorem 2.1 applied to \(f'\), and Theorem 2.2 with \(n = 1\) applied to \(f\), give

\[ \oint _\Gamma \frac {f'(z)}{z - z_0} \dd z = 2\pi i f'(z_0) = \oint _\Gamma \frac {f(z)}{(z - z_0)^2} \dd z. \]

\(z_0\) outside \(\Gamma \). Both integrands are analytic on a simply connected domain containing \(\Gamma \) and its inside but not \(z_0\), so both integrals are \(0\) by Theorem 1.2.

Comparison. In either case the two integrals are equal.

\(z_0\) on \(\Gamma \). Both integrands are unbounded near \(z_0\), so they are not continuous on \(\Gamma \). Neither formula applies, and the integrals are not even defined as ordinary contour integrals; they would have to be treated as improper integrals. \(\relax \vartriangleleft \)

Remark 2.6. The equality has a one-line explanation. Off \(z_0\),

\[ \frac {\dd }{\dd z} \frac {f(z)}{z - z_0} = \frac {f'(z)}{z - z_0} - \frac {f(z)}{(z - z_0)^2}, \]

and the integral of a derivative around a closed contour is \(0\).

3 Cauchy’s estimates and Liouville’s theorem

Theorem 3.1 (Cauchy’s estimates). Let \(f\) be analytic inside and on the circle \(C_R \colon \abs {z - z_0} = R\), with \(\abs {f(z)} \leq M\) on \(C_R\). Then

\[ \abs {f^{(n)}(z_0)} \leq \frac {n! \, M}{R^n} \qquad (n = 0, 1, 2, \dots ). \]

Source: [1, ch. 4]

Proof. On \(C_R\) the integrand in Theorem 2.2 satisfies \(\abs {f(z) / (z - z_0)^{n+1}} \leq M / R^{n+1}\), and \(C_R\) has length \(2\pi R\). By Theorem 1.1,

\[ \abs {f^{(n)}(z_0)} \leq \frac {n!}{2\pi } \cdot \frac {M}{R^{n+1}} \cdot 2\pi R = \frac {n! \, M}{R^n}. \qedhere \]

Theorem page

Theorem 3.2 (Liouville’s theorem). A bounded entire function is constant.

Source: [1, ch. 4]

Proof. Suppose \(\abs {f} \leq M\) on \(\C \). Fix \(z_0\). For every \(R > 0\), Theorem 3.1 with \(n = 1\) gives \(\abs {f'(z_0)} \leq M / R\). Letting \(R \to \infty \) gives \(f'(z_0) = 0\). So \(f' \equiv 0\) on the connected set \(\C \), and \(f\) is constant.

Theorem page

Exercise 3.3. Let \(f\) be entire with \(A \leq \operatorname {Im} f(z) \leq B\) for all \(z\). Show that \(e^{if}\) is bounded and entire, and deduce that \(f\) is constant.

Proof. Write \(f = u + iv\), so \(if = -v + iu\) and

\[ \abs {e^{if}} = \abs {e^{-v} e^{iu}} = e^{-v}. \]

Since \(A \leq v \leq B\), this gives \(e^{-B} \leq \abs {e^{if}} \leq e^{-A}\), so \(e^{if}\) is bounded. It is entire because it is a composition of the entire functions \(if\) and \(\exp \), with \(\frac {\dd }{\dd z} e^{if} = if' e^{if}\) by the chain rule.

By Theorem 3.2, \(e^{if} \equiv c\) for some constant \(c\), and \(c \neq 0\) because the exponential never vanishes. Differentiating, \(if'(z) e^{if(z)} = 0\), so \(f' \equiv 0\) and \(f\) is constant.

Exercise 3.4. Let \(f\) be entire with \(f^{(7)}\) bounded on \(\C \). What kind of function is \(f\)? Generalise, and decide whether the same holds for real functions.

Solution. By Theorem 2.2, \(f^{(7)}\) is entire, and it is bounded, so it is a constant \(c\) by Theorem 3.2. Hence \(f^{(n)} \equiv 0\) for \(n \geq 8\). By Taylor’s theorem (Theorem 8.2) the series of \(f\) about \(0\) converges to \(f\) on all of \(\C \), and its coefficients \(f^{(n)}(0) / n!\) vanish for \(n \geq 8\). So

\[ f(z) = \sum _{n=0}^{7} \frac {f^{(n)}(0)}{n!} z^n, \]

a polynomial of degree at most \(7\), and exactly \(7\) when \(c \neq 0\).

Generalisation. The same argument shows that if \(f\) is entire and \(f^{(k)}\) is bounded, then \(f\) is a polynomial of degree at most \(k\).

Real functions. This fails on \(\R \). For \(g(x) = \sin x\), every derivative is \(\pm \sin x\) or \(\pm \cos x\), so \(g^{(7)}(x) = -\cos x\) is bounded, yet \(\sin x\) is not a polynomial. \(\relax \vartriangleleft \)

Remark 3.5. Applying Theorem 3.1 directly to \(g = f^{(7)}\) gives \(\abs {f^{(8)}(z_0)} \leq M / R\) for every \(R\), which is the same argument that proves Liouville’s theorem.

4 The maximum modulus principle

Theorem 4.1 (Maximum modulus principle). Let \(D\) be a bounded domain, and let \(f\) be continuous on the closure of \(D\) and analytic in \(D\). Then \(\abs {f}\) attains its maximum over the closure on the boundary of \(D\). If \(\abs {f}\) also attains it at a point of \(D\), then \(f\) is constant.

Source: [1, ch. 4]

Theorem page

Exercise 4.2. Let \(f(z) = \left (z - \frac 12\right )\left (z - \frac 14\right )\) on the closed unit disc. Explain why \(\max _{\abs {z} \leq 1} \abs {f(z)}\) is attained on \(\abs {z} = 1\), compute \(\abs {f}\) there, and find the maximum.

Solution. Where. \(f\) is a polynomial, so it is analytic on the open unit disc and continuous on its closure. By Theorem 4.1, the maximum of \(\abs {f}\) is attained on \(\abs {z} = 1\).

On the circle. Write \(z = e^{i\theta }\) and \(a = \cos \theta \). For real \(c\), \(\abs {z - c}^2 = (z - c)(\bar z - c) = 1 - 2ca + c^2\), so

\[ \abs {f(z)}^2 = \left (\tfrac 54 - a\right )\left (\tfrac {17}{16} - \tfrac 12 a\right ) = \tfrac 12 a^2 - \tfrac {27}{16} a + \tfrac {85}{64}. \]

Maximum. The derivative in \(a\) is \(a - \frac {27}{16} < 0\) on \([-1, 1]\), so \(\abs {f}^2\) is largest at \(a = -1\), that is, at \(z = -1\), where it equals \(\frac {225}{64}\). Hence

\[ \max _{\abs {z} \leq 1} \abs {f(z)} = \frac {15}{8}. \qedhere \]
\(\relax \vartriangleleft \)

Remark 4.3. The triangle inequality gets there faster: for \(\abs {z} \leq 1\), \(\abs {f(z)} \leq \left (1 + \frac 12\right )\left (1 + \frac 14\right ) = \frac {15}{8}\), with equality at \(z = -1\).

Chapter 2: Power and Taylor Series

Every function analytic on a disc is the sum of its Taylor series there, and power series can be differentiated, multiplied and compared term by term. Each section states one of these facts and then puts it to work.

5 Radius of convergence

Theorem 5.1 (Cauchy–Hadamard). Let \(\sum a_n (z - z_0)^n\) be a power series and define \(R \in [0, \infty ]\) by

\[ \frac {1}{R} = \limsup _{n \to \infty } \abs {a_n}^{1/n}, \]

with \(1/0 = \infty \) and \(1/\infty = 0\). The series converges absolutely for \(\abs {z - z_0} < R\) and diverges for \(\abs {z - z_0} > R\). We call \(R\) the radius of convergence.

Source: [1, ch. 5]

Theorem page

Theorem 5.2 (Ratio test). If \(a_n \neq 0\) for all large \(n\) and \(L = \lim _{n \to \infty } \abs {a_{n+1}/a_n}\) exists in \([0, \infty ]\), then the radius of convergence of \(\sum a_n (z - z_0)^n\) is \(R = 1/L\).

Source: [1, ch. 5]

Theorem page

6 Termwise differentiation

Theorem 6.1 (Termwise differentiation). Suppose \(f(z) = \sum _{n=0}^{\infty } a_n (z - z_0)^n\) has radius of convergence \(R > 0\). Then \(f\) is analytic on \(\abs {z - z_0} < R\), and

\[ f'(z) = \sum _{n=1}^{\infty } n a_n (z - z_0)^{n-1}, \]

where the differentiated series also has radius of convergence \(R\).

Source: [1, ch. 5]

Proof. Translating, we may take \(z_0 = 0\).

Same radius. Since \(n^{1/n} \to 1\),

\[ \limsup _{n \to \infty } \abs {n a_n}^{1/n} = \limsup _{n \to \infty } \abs {a_n}^{1/n}, \]

so \(\sum n a_n z^{n-1}\) has radius \(R\) by Theorem 5.1. Similarly, \(\sum n(n-1) \abs {a_n} r^{n-2}\) converges for every \(0 \leq r < R\).

The derivative. Fix \(\abs {z} < R\), choose \(r\) with \(\abs {z} < r < R\), and let \(g(z) = \sum n a_n z^{n-1}\). For \(\abs {w} < r\) with \(w \neq z\),

\[ \frac {f(w) - f(z)}{w - z} - g(z) = \sum _{n=2}^{\infty } a_n \left ( \frac {w^n - z^n}{w - z} - n z^{n-1} \right ). \]

Expanding \(\frac {w^n - z^n}{w - z} = \sum _{j=0}^{n-1} w^j z^{n-1-j}\), the bracket equals \(\sum _{j=1}^{n-1} z^{n-1-j} (w^j - z^j)\). Since \(\abs {w^j - z^j} \leq j r^{j-1} \abs {w - z}\), the bracket is at most

\[ \abs {w - z} \, r^{n-2} \sum _{j=1}^{n-1} j = \abs {w - z} \, \frac {n(n-1)}{2} \, r^{n-2} \]

in modulus. Hence

\[ \absStar{\frac {f(w) - f(z)}{w - z} - g(z)} \leq \frac {\abs {w - z}}{2} \sum _{n=2}^{\infty } n(n-1) \abs {a_n} r^{n-2} \longrightarrow 0 \quad \text {as } w \to z. \]

So \(f'(z) = g(z)\) exists at every point of the disc. That is, \(f\) is analytic there.

Theorem page

Exercise 6.2. Find the radius of convergence and a closed form for \(f(z) = \sum _{n=1}^{\infty } n^3 z^n\). Use it to evaluate \(\sum _{n=1}^{\infty } n^3 / 3^n\) and \(\sum _{n=3}^{\infty } n^3 / 2^n\).

Solution. Radius. With \(a_n = n^3\),

\[ \absStar{\frac {a_{n+1}}{a_n}} = \left (1 + \frac {1}{n}\right )^3 \to 1, \]

so \(R = 1\) by Theorem 5.2.

Closed form. Start from the geometric series

\[ \frac {1}{1 - z} = \sum _{n=0}^{\infty } z^n, \qquad \abs {z} < 1. \]

Differentiating term by term (Theorem 6.1) and multiplying by \(z\) replaces each coefficient \(a_n\) by \(n a_n\) without changing the disc. Doing this three times gives

\begin{align*} \frac {z}{(1-z)^2} &= \sum _{n=1}^{\infty } n z^n, \\ \frac {z(1+z)}{(1-z)^3} &= \sum _{n=1}^{\infty } n^2 z^n, \\ \frac {z(1 + 4z + z^2)}{(1-z)^4} &= \sum _{n=1}^{\infty } n^3 z^n. \end{align*}

For instance, the last line comes from the quotient rule:

\[ \frac {\dd }{\dd z} \frac {z + z^2}{(1-z)^3} = \frac {(1 + 2z)(1 - z) + 3(z + z^2)}{(1-z)^4} = \frac {1 + 4z + z^2}{(1-z)^4}. \]

The sums. Both points lie inside \(\abs {z} < 1\). At \(z = 1/3\),

\[ \sum _{n=1}^{\infty } \frac {n^3}{3^n} = \frac {\frac 13 \cdot \frac {22}{9}}{\left (\frac 23\right )^4} = \frac {22}{27} \cdot \frac {81}{16} = \frac {33}{8}. \]

At \(z = 1/2\) the full sum is \(\frac {\frac 12 \cdot \frac {13}{4}}{\left (\frac 12\right )^4} = 26\), and removing the terms \(n = 1, 2\), which are \(\frac 12\) and \(2\), leaves

\[ \sum _{n=3}^{\infty } \frac {n^3}{2^n} = 26 - \frac 52 = \frac {47}{2}. \qedhere \]
\(\relax \vartriangleleft \)

Remark 6.3. The operator \(\vartheta = z \frac {\dd }{\dd z}\) sends \(z^n\) to \(n z^n\), so \(f = \vartheta ^3 \bigl (\frac {1}{1-z}\bigr )\). The same trick sums \(\sum p(n) z^n\) for any polynomial \(p\).

Exercise 6.4. Let

\[ f(z) = \sum _{n=0}^{\infty } \frac {(-1)^n z^{2n}}{4^n (n!)^2}. \]

Show that \(f\) is entire and that \(f''(z) + \frac {1}{z} f'(z) + f(z) = 0\) for \(z \neq 0\).

Proof. \(f\) is entire. Fix \(z \neq 0\) and let \(c_n\) be the \(n\)th term. Then

\[ \absStar{\frac {c_{n+1}}{c_n}} = \frac {\abs {z}^2}{4(n+1)^2} \to 0, \]

so the series converges absolutely for every \(z\). Its radius of convergence is \(\infty \), and \(f\) is entire by Theorem 6.1.

Derivatives. By Theorem 6.1, twice,

\[ f'(z) = \sum _{n=1}^{\infty } \frac {(-1)^n \, 2n \, z^{2n-1}}{4^n (n!)^2}, \qquad f''(z) = \sum _{n=1}^{\infty } \frac {(-1)^n \, 2n(2n-1) \, z^{2n-2}}{4^n (n!)^2}. \]

Every term of \(f'\) contains a factor \(z\), so

\[ \frac {1}{z} f'(z) = \sum _{n=1}^{\infty } \frac {(-1)^n \, 2n \, z^{2n-2}}{4^n (n!)^2} \]

is again a convergent power series. In fact it is entire, so the singularity of \(f'(z)/z\) at the origin is only apparent.

The equation. Adding the last two series term by term,

\[ f''(z) + \frac {1}{z} f'(z) = \sum _{n=1}^{\infty } \frac {(-1)^n \, 4n^2 \, z^{2n-2}}{4^n (n!)^2} = \sum _{n=1}^{\infty } \frac {(-1)^n z^{2n-2}}{4^{n-1} ((n-1)!)^2}, \]

since \(4^n = 4 \cdot 4^{n-1}\) and \((n!)^2 = n^2 ((n-1)!)^2\). Put \(k = n - 1\):

\[ f''(z) + \frac {1}{z} f'(z) = \sum _{k=0}^{\infty } \frac {(-1)^{k+1} z^{2k}}{4^k (k!)^2} = -f(z). \qedhere \]

Remark 6.5. This \(f\) is the Bessel function \(J_0\). Multiplying through by \(z\) gives Bessel’s equation of order zero, \(z f'' + f' + z f = 0\), which then holds at \(z = 0\) as well.

7 Uniqueness of coefficients

Theorem 7.1 (Uniqueness of power series). If \(\sum a_n (z - z_0)^n\) converges to \(f(z)\) on some disc \(\abs {z - z_0} < r\), then

\[ a_n = \frac {f^{(n)}(z_0)}{n!} \qquad (n \geq 0). \]

In particular, two power series with the same sum near \(z_0\) have the same coefficients.

Source: [1, ch. 5]

Proof. Applying Theorem 6.1 \(k\) times,

\[ f^{(k)}(z) = \sum _{n=k}^{\infty } \frac {n!}{(n-k)!} \, a_n (z - z_0)^{n-k}, \qquad \abs {z - z_0} < r. \]

Putting \(z = z_0\) kills every term except \(n = k\), so \(f^{(k)}(z_0) = k! \, a_k\). If two series have the same sum \(f\) near \(z_0\), both sets of coefficients equal \(f^{(n)}(z_0)/n!\).

Theorem page

Exercise 7.2. Let \(f(z) = \sum _{n=0}^{\infty } z^{2^n} = z + z^2 + z^4 + z^8 + \dotsb \). Show that \(f(z) = z + f(z^2)\), and find every power series \(g\) that satisfies \(g(z) = z + g(z^2)\) near the origin.

Solution. \(f\) satisfies the equation. The coefficients are \(0\) or \(1\), with infinitely many \(1\)s, so the radius of convergence is \(1\). For \(\abs {z} < 1\),

\[ z + f(z^2) = z + \sum _{n=0}^{\infty } z^{2^{n+1}} = z^{2^0} + \sum _{n=1}^{\infty } z^{2^n} = f(z). \]

All solutions. Suppose \(g(z) = \sum a_n z^n\) converges near \(0\) and \(g(z) = z + g(z^2)\). Then

\[ a_0 + a_1 z + a_2 z^2 + a_3 z^3 + a_4 z^4 + \dotsb = a_0 + z + a_1 z^2 + 0 z^3 + a_2 z^4 + \dotsb , \]

and by Theorem 7.1

\[ a_0 \text { is free}, \qquad a_1 = 1, \qquad a_{2k} = a_k \ (k \geq 1), \qquad a_{2k+1} = 0 \ (k \geq 1). \]

Write \(n \geq 1\) as \(n = 2^m q\) with \(q\) odd. Halving \(m\) times gives \(a_n = a_q\), which is \(1\) if \(q = 1\) and \(0\) otherwise. So \(a_n = 1\) exactly when \(n\) is a power of \(2\), and

\[ g(z) = c + f(z), \qquad c \in \C . \]

Conversely, each such \(g\) works, because the constant appears on both sides of \(g(z) = z + g(z^2)\). The solutions are the translates of \(f\) by constants. \(\relax \vartriangleleft \)

Remark 7.3. The gaps in the exponents \(2^n\) make \(f\) impossible to continue past the unit circle: every point of \(\abs {z} = 1\) is a singularity. This is a natural boundary, proved as a separate graduate proof.

8 Products and Taylor’s theorem

Theorem 8.1 (Cauchy product). If \(f(z) = \sum a_n (z - z_0)^n\) and \(g(z) = \sum b_n (z - z_0)^n\) both converge on \(\abs {z - z_0} < r\), then on the same disc

\[ f(z) g(z) = \sum _{n=0}^{\infty } c_n (z - z_0)^n, \qquad c_n = \sum _{k=0}^{n} a_k b_{n-k}. \]

Source: [1, ch. 5]

Theorem page

Theorem 8.2 (Taylor’s theorem). If \(f\) is analytic on \(\abs {z - z_0} < \rho \), then

\[ f(z) = \sum _{n=0}^{\infty } \frac {f^{(n)}(z_0)}{n!} (z - z_0)^n \]

for every \(z\) in that disc. In particular, the radius of convergence of the Taylor series is at least the distance from \(z_0\) to the nearest point where \(f\) fails to be analytic.

Source: [1, ch. 5]

Theorem page

Exercise 8.3. Find the first four nonzero terms of the Maclaurin series of \(1/\cos z\). Justify each step.

Solution. A series exists. \(\cos z\) is entire and \(\cos 0 = 1 \neq 0\), so \(1/\cos z\) is analytic near \(0\). By Taylor’s theorem it has a Maclaurin series \(\sum b_n z^n\).

Multiply out. Write \(\cos z = 1 - \frac {z^2}{2} + \frac {z^4}{24} - \frac {z^6}{720} + \dotsb \). Since \(\cos z \cdot \frac {1}{\cos z} = 1\), Theorem 8.1 gives

\[ \Bigl (1 - \tfrac 12 z^2 + \tfrac 1{24} z^4 - \tfrac 1{720} z^6 + \dotsb \Bigr ) \bigl (b_0 + b_1 z + b_2 z^2 + \dotsb \bigr ) = 1 + 0z + 0z^2 + \dotsb . \]

Compare coefficients. By Theorem 7.1 the coefficients of each power of \(z\) agree on both sides:

\begin{alignat*}{3} & z^0\colon \quad & b_0 &= 1, \\ & z^2\colon & b_2 - \tfrac 12 b_0 &= 0, &\qquad & b_2 = \tfrac 12, \\ & z^4\colon & b_4 - \tfrac 12 b_2 + \tfrac 1{24} b_0 &= 0, && b_4 = \tfrac {5}{24}, \\ & z^6\colon & b_6 - \tfrac 12 b_4 + \tfrac 1{24} b_2 - \tfrac 1{720} b_0 &= 0, && b_6 = \tfrac {61}{720}. \end{alignat*}

The odd equations read \(b_1 = 0\), \(b_3 = \frac 12 b_1 = 0\), and so on, as they must, since \(1/\cos z\) is even. Hence

\[ \frac {1}{\cos z} = 1 + \frac {z^2}{2} + \frac {5 z^4}{24} + \frac {61 z^6}{720} + \dotsb . \qedhere \]
\(\relax \vartriangleleft \)

Remark 8.4. The zeros of \(\cos z\) nearest the origin are \(\pm \pi /2\), so this series converges exactly on \(\abs {z} < \pi /2\). The numerators \(1, 1, 5, 61\) are the Euler (secant) numbers.

Exercise 8.5. Expand \(p(z) = z^4\) in powers of \(z - 1\).

Solution. \(p\) is entire, so by Theorem 8.2 its Taylor series about \(1\) converges to it everywhere. The derivatives at \(1\) are \(p(1) = 1\), \(p'(1) = 4\), \(p''(1) = 12\), \(p'''(1) = 24\), \(p^{(4)}(1) = 24\), and \(p^{(j)} \equiv 0\) for \(j \geq 5\). Dividing by \(j!\),

\[ z^4 = 1 + 4(z - 1) + 6(z - 1)^2 + 4(z - 1)^3 + (z - 1)^4. \qedhere \]
\(\relax \vartriangleleft \)

Remark 8.6. The coefficients are the binomial coefficients \(\binom {4}{j}\), as they must be: \(z^4 = \bigl (1 + (z - 1)\bigr )^4\).

Chapter 3: Singularities, Laurent Series and Residues

Near an isolated singularity a function is no longer a power series, but it is still a series in both positive and negative powers. The negative part classifies the singularity, and a single coefficient, the residue, is all that survives when we integrate around it.

9 Laurent series

Theorem 9.1 (Laurent’s theorem). If \(f\) is analytic in the annulus \(r < \abs {z - z_0} < R\), then

\[ f(z) = \sum _{n=-\infty }^{\infty } a_n (z - z_0)^n \]

there, where both halves of the series converge absolutely in the annulus and uniformly on every closed subannulus. The coefficients are unique, and

\[ a_n = \frac {1}{2\pi i} \oint _C \frac {f(\zeta )}{(\zeta - z_0)^{n+1}} \dd \zeta \]

for any positively oriented circle \(C \colon \abs {\zeta - z_0} = \rho \) with \(r < \rho < R\).

Source: [1, ch. 5]

Theorem page

Definition 9.2. Let \(f\) be analytic in \(0 < \abs {z - z_0} < R\), with Laurent series \(\sum _{n=-\infty }^{\infty } a_n (z - z_0)^n\) there. The isolated singularity \(z_0\) is

  • removable if \(a_n = 0\) for every \(n < 0\);
  • a pole of order \(m\) if \(a_{-m} \neq 0\) and \(a_n = 0\) for every \(n < -m\);
  • essential if \(a_n \neq 0\) for infinitely many \(n < 0\).

The residue of \(f\) at \(z_0\) is \(\Res (f; z_0) = a_{-1}\).

Exercise 9.3. Find the Laurent series of \(f(z) = e^{1/(2z^2)}\) about \(0\), classify the singularity, and find \(\Res (f; 0)\).

Solution. The series \(e^w = \sum _{j=0}^{\infty } w^j / j!\) converges for every \(w\), because \(e^w\) is entire. Putting \(w = 1/(2z^2)\) gives, for every \(z \neq 0\),

\[ e^{1/(2z^2)} = \sum _{j=0}^{\infty } \frac {1}{j! \, 2^j z^{2j}} = 1 + \frac {1}{2z^2} + \frac {1}{8z^4} + \frac {1}{48z^6} + \dotsb . \]

By uniqueness of Laurent coefficients (Theorem 9.1), this is the Laurent series on \(0 < \abs {z} < \infty \). It has infinitely many negative powers, so \(0\) is an essential singularity. Only even powers occur, so \(a_{-1} = 0\) and \(\Res (f; 0) = 0\). \(\relax \vartriangleleft \)

Remark 9.4. A zero residue does not mean a mild singularity: here it sits beside an essential singularity. It only says that \(\oint f = 0\) around small circles.

10 Removable singularities

Theorem 10.1 (Removable singularities). Let \(f\) have an isolated singularity at \(z_0\). The following are equivalent:

(i)
\(z_0\) is a removable singularity;
(ii)
\(\lim _{z \to z_0} f(z)\) exists and is finite;
(iii)
\(f\) is bounded in some punctured disc \(0 < \abs {z - z_0} < \delta \).

In that case, setting \(f(z_0) = \lim _{z \to z_0} f(z)\) makes \(f\) analytic at \(z_0\).

Source: [1, ch. 5]

Theorem page

Exercise 10.2. Show that \(f(z) = \dfrac {1 - \cos z}{z^2}\) has a removable singularity at \(0\), and find the value of \(f(0)\) that makes \(f\) analytic there.

Solution. For every \(z\), \(\cos z = 1 - \frac {z^2}{2!} + \frac {z^4}{4!} - \frac {z^6}{6!} + \dotsb \), so for \(z \neq 0\)

\[ f(z) = \frac {1}{2!} - \frac {z^2}{4!} + \frac {z^4}{6!} - \dotsb = \frac 12 - \frac {z^2}{24} + \frac {z^4}{720} - \dotsb . \]

The Laurent series has no negative powers, so the singularity is removable. Equivalently, \(\lim _{z \to 0} f(z) = \frac 12\) (Theorem 10.1). Setting \(f(0) = \frac 12\) makes \(f\) the sum of an everywhere convergent power series, so \(f\) is entire. \(\relax \vartriangleleft \)

11 Zeros and poles

Theorem 11.1 (Zeros and poles). Let \(m \geq 1\).

(i)
\(f\) has a pole of order \(m\) at \(z_0\) if and only if \(f(z) = g(z) / (z - z_0)^m\) near \(z_0\), with \(g\) analytic at \(z_0\) and \(g(z_0) \neq 0\).
(ii)
\(g\) has a zero of order \(m\) at \(z_0\) if and only if \(g(z) = (z - z_0)^m h(z)\) with \(h\) analytic at \(z_0\) and \(h(z_0) \neq 0\).
(iii)
If \(g\) has a zero of order \(m\) at \(z_0\), then \(1/g\) has a pole of order \(m\) there.

Source: [1, ch. 5]

Theorem page

Exercise 11.2. Find the poles of \(f(z) = \dfrac {e^z}{z^2 (z - 2)^2}\) and their orders.

Solution. \(e^z\) is entire, so \(f\) is analytic except at the zeros \(0\) and \(2\) of the denominator.

At \(0\). \(f(z) = g(z) / z^2\) with \(g(z) = e^z / (z - 2)^2\), which is analytic at \(0\) with \(g(0) = \frac 14 \neq 0\). By Theorem 11.1, \(0\) is a pole of order \(2\).

At \(2\). \(f(z) = g(z) / (z - 2)^2\) with \(g(z) = e^z / z^2\), analytic at \(2\) with \(g(2) = e^2 / 4 \neq 0\). So \(2\) is also a pole of order \(2\). \(\relax \vartriangleleft \)

12 Residues

Theorem 12.1 (Cauchy’s residue theorem). Let \(\Gamma \) be a simple closed positively oriented contour, and let \(f\) be analytic inside and on \(\Gamma \) except at finitely many points \(z_1, \dots , z_n\) inside \(\Gamma \). Then

\[ \oint _\Gamma f(z) \dd z = 2\pi i \sum _{k=1}^{n} \Res (f; z_k). \]

Source: [1, ch. 6]

Proof. Surround each \(z_k\) by a small positively oriented circle \(C_k\), with the discs disjoint and inside \(\Gamma \). By Theorem 1.3, \(\oint _\Gamma f = \sum _k \oint _{C_k} f\). On \(C_k\), \(f\) equals its Laurent series about \(z_k\), which converges uniformly there (Theorem 9.1), so it may be integrated term by term. Since

\[ \oint _{C_k} (z - z_k)^n \dd z = \begin {cases} 2\pi i, & n = -1, \\ 0, & n \neq -1, \end {cases} \]

only the \(a_{-1}\) term survives, and \(\oint _{C_k} f = 2\pi i \Res (f; z_k)\).

Theorem page

Theorem 12.2 (Residue at a pole). If \(f\) has a pole of order \(m\) at \(z_0\), then

\[ \Res (f; z_0) = \frac {1}{(m-1)!} \lim _{z \to z_0} \frac {\dd ^{m-1}}{\dd z^{m-1}} \bigl [ (z - z_0)^m f(z) \bigr ]. \]

For a simple pole, \(\Res (f; z_0) = \lim _{z \to z_0} (z - z_0) f(z)\).

Source: [1, ch. 6]

Proof. Near \(z_0\),

\[ (z - z_0)^m f(z) = a_{-m} + a_{-m+1}(z - z_0) + \dots + a_{-1}(z - z_0)^{m-1} + \dotsb , \]

a power series once its removable singularity at \(z_0\) is filled in. By Theorem 7.1, its \((m-1)\)st derivative at \(z_0\) is \((m-1)! \, a_{-1}\), and this derivative is continuous, so it equals the limit.

Theorem page

Exercise 12.3. Let \(f(z) = \dfrac {1}{z^2 \sin z}\). Find \(\Res (f; 0)\) in two ways, first by Theorem 12.2 and then from the Laurent series, and use it to evaluate \(\oint _{\abs {z} = 1} f(z) \dd z\).

Solution. Order of the pole. From the series for \(\sin z\),

\[ z^2 \sin z = z^3 h(z), \qquad h(z) = 1 - \frac {z^2}{6} + \frac {z^4}{120} - \dotsb , \]

with \(h(0) = 1 \neq 0\). So \(z^2 \sin z\) has a zero of order \(3\) at \(0\), and \(f\) has a pole of order \(3\) there (Theorem 11.1).

A series for \(z / \sin z\). This is \(1/h(z)\), which is analytic near \(0\). Write \(h = 1 - c\) with \(c = \frac {z^2}{6} - \frac {z^4}{120} + \dotsb \). For small \(z\), \(\abs {c} < 1\), and the geometric series gives

\[ \frac {z}{\sin z} = 1 + c + c^2 + \dotsb = 1 + \frac {z^2}{6} + \left (\frac {1}{36} - \frac {1}{120}\right ) z^4 + \dotsb = 1 + \frac {z^2}{6} + \frac {7 z^4}{360} + \dotsb . \]

By the formula. With \(m = 3\),

\[ \Res (f; 0) = \frac {1}{2!} \lim _{z \to 0} \frac {\dd ^2}{\dd z^2} \left [ \frac {z}{\sin z} \right ] = \frac 12 \lim _{z \to 0} \left ( \frac 13 + \frac {7 z^2}{30} + \dotsb \right ) = \frac 16, \]

differentiating term by term (Theorem 6.1).

From the Laurent series. Dividing by \(z^3\),

\[ f(z) = \frac {1}{z^3} \cdot \frac {z}{\sin z} = \frac {1}{z^3} + \frac {1}{6z} + \frac {7z}{360} + \dotsb , \]

so again \(\Res (f; 0) = a_{-1} = \frac 16\).

The integral. The singularities of \(f\) are the zeros \(z = k\pi \) of \(z^2 \sin z\), and only \(z = 0\) lies inside \(\abs {z} = 1\). By Theorem 12.1,

\[ \oint _{\abs {z} = 1} \frac {\dd z}{z^2 \sin z} = 2\pi i \cdot \frac 16 = \frac {\pi i}{3}. \qedhere \]
\(\relax \vartriangleleft \)

Exercise 12.4. Find the singularities and the residues there of

\[ \text {(a)}\ f(z) = \frac {\sin z}{z(z + 1)}, \qquad \text {(b)}\ g(z) = \frac {\cos z}{z (z - 1)^2}. \]

Solution. (a) At \(0\). Since \(\frac {\sin z}{z} = 1 - \frac {z^2}{3!} + \dotsb \), \(f(z) \to 1\) as \(z \to 0\). The singularity is removable (Theorem 10.1), so there are no negative powers and \(\Res (f; 0) = 0\).

(a) At \(-1\). \(f(z) = \frac {\sin z / z}{z + 1}\), and \(\frac {\sin (-1)}{-1} = \sin 1 \neq 0\), so \(-1\) is a simple pole and

\[ \Res (f; -1) = \lim _{z \to -1} \frac {\sin z}{z} = \sin 1. \]

(b) At \(0\). \(g(z) = \frac {\cos z / (z - 1)^2}{z}\), with numerator \(1 \neq 0\) at \(z = 0\). This is a simple pole, and

\[ \Res (g; 0) = \lim _{z \to 0} \frac {\cos z}{(z - 1)^2} = 1. \]

(b) At \(1\). \(g(z) = \frac {\cos z / z}{(z - 1)^2}\), with numerator \(\cos 1 \neq 0\) at \(z = 1\), so this is a pole of order \(2\). By Theorem 12.2 with \(m = 2\),

\[ \Res (g; 1) = \lim _{z \to 1} \frac {\dd }{\dd z} \frac {\cos z}{z} = \lim _{z \to 1} \frac {-z \sin z - \cos z}{z^2} = -\sin 1 - \cos 1. \qedhere \]
\(\relax \vartriangleleft \)

Chapter 4: Real Integrals by Residues

Some real integrals that resist every calculus technique become routine once they are read as contour integrals. The simplest case is an integral of a rational function of \(\cos \theta \) and \(\sin \theta \) over a full period, which becomes an integral around the unit circle.

13 Integrals over a period

Theorem 13.1 (Trigonometric integrals). Let \(R(x, y)\) be a rational function with no poles on the circle \(x^2 + y^2 = 1\). Then

\[ \int _0^{2\pi } R(\cos \theta , \sin \theta ) \dd \theta = \oint _{\abs {z} = 1} R\left ( \frac {z + z^{-1}}{2}, \frac {z - z^{-1}}{2i} \right ) \frac {\dd z}{iz}, \]

and the right-hand side is \(2\pi i\) times the sum of the residues of its integrand inside \(\abs {z} = 1\).

Source: [1, ch. 6]

Proof. Parametrise the circle by \(z = e^{i\theta }\), \(0 \leq \theta \leq 2\pi \). Then \(\dd z = iz \dd \theta \), \(\cos \theta = \frac 12(z + z^{-1})\) and \(\sin \theta = \frac {1}{2i}(z - z^{-1})\), which turns the contour integral into the real one. The integrand on the right is rational in \(z\), so it has finitely many poles, and Theorem 12.1 applies.

Theorem page

Exercise 13.2. Evaluate \(\displaystyle \int _0^{2\pi } \frac {\dd \theta }{1 + \frac 14 \cos \theta }\).

Solution. To the circle. By Theorem 13.1, with \(C\) the circle \(\abs {z} = 1\),

\[ I = \oint _C \frac {1}{1 + \frac 18\left (z + z^{-1}\right )} \frac {\dd z}{iz} = \oint _C \frac {8z}{z^2 + 8z + 1} \frac {\dd z}{iz} = \frac {8}{i} \oint _C \frac {\dd z}{z^2 + 8z + 1}. \]

Poles. The roots of \(z^2 + 8z + 1\) are \(z_{1,2} = -4 \pm \sqrt {15}\), both simple. Since \(\sqrt {15} \approx 3.87\), \(z_1 = -4 + \sqrt {15} \approx -0.13\) lies inside \(C\) and \(z_2 = -4 - \sqrt {15} \approx -7.87\) lies outside.

Residue. By Theorem 12.2,

\[ \Res \left ( \frac {1}{(z - z_1)(z - z_2)}; z_1 \right ) = \frac {1}{z_1 - z_2} = \frac {1}{2\sqrt {15}}. \]

Total.

\[ I = \frac {8}{i} \cdot 2\pi i \cdot \frac {1}{2\sqrt {15}} = \frac {8\pi }{\sqrt {15}}. \qedhere \]
\(\relax \vartriangleleft \)

Exercise 13.3. Show that for real \(a\) with \(0 < \abs {a} < 1\),

\[ \int _0^{2\pi } \frac {\dd \theta }{1 + a\cos \theta } = \frac {2\pi }{\sqrt {1 - a^2}}. \]

Solution. As above,

\[ \int _0^{2\pi } \frac {\dd \theta }{1 + a\cos \theta } = \frac {2}{ia} \oint _{\abs {z} = 1} \frac {\dd z}{z^2 + \frac {2}{a} z + 1}. \]

The roots \(z_{1,2} = \bigl (-1 \pm \sqrt {1 - a^2}\bigr )/a\) are real with product \(1\) and \(\abs {z_1} < \abs {z_2}\), so exactly one of them, \(z_1\), lies inside the circle. Its residue is \(1/(z_1 - z_2) = a / \bigl (2\sqrt {1 - a^2}\bigr )\), and so

\[ \int _0^{2\pi } \frac {\dd \theta }{1 + a\cos \theta } = \frac {2}{ia} \cdot 2\pi i \cdot \frac {a}{2\sqrt {1 - a^2}} = \frac {2\pi }{\sqrt {1 - a^2}}. \]

With \(a = \frac 14\) this gives \(2\pi / \sqrt {15/16} = 8\pi / \sqrt {15}\), agreeing with Exercise 13.2. \(\relax \vartriangleleft \)

Remark 13.4. As \(a \to \pm 1\) the integral blows up: one pole moves onto the unit circle, where \(1 + a\cos \theta \) vanishes.

Chapter 5: Real and Complex Differentiability

Complex differentiability looks like the real definition with \(x\) replaced by \(z\), but it is far more rigid. Here are two places where the theories part ways.

14 One derivative gives all of them

On \(\R \), a function can be differentiable once and no more. On \(\C \), an analytic function has derivatives of every order, all analytic (Theorem 2.2).

Exercise 14.1. Show that \(f(x) = x^{5/3}\) is differentiable on \(\R \), but \(f'\) is not differentiable at \(0\).

Solution. For \(x \neq 0\), \(f'(x) = \frac 53 x^{2/3}\). At \(0\),

\[ f'(0) = \lim _{h \to 0} \frac {h^{5/3}}{h} = \lim _{h \to 0} h^{2/3} = 0, \]

which agrees with \(\frac 53 x^{2/3}\) at \(x = 0\). But

\[ \frac {f'(h) - f'(0)}{h} = \frac 53 h^{-1/3} \]

is unbounded as \(h \to 0\), so \(f''(0)\) does not exist. \(\relax \vartriangleleft \)

Remark 14.2. No analytic function behaves like this: once \(f'\) exists on an open set, so do \(f''\), \(f'''\), and the rest.

15 No mean value theorem

For a real function differentiable on \([a, b]\), some \(c\) between \(a\) and \(b\) has \(f'(c) = \frac {f(b) - f(a)}{b - a}\). The special case \(f(a) = f(b)\) is Rolle’s theorem. Neither survives in \(\C \).

Exercise 15.1. Let \(g(z) = z^3\) and \(\omega = -\frac 12 + \frac {\sqrt 3}{2} i\). Show that \(g(1) = g(\omega )\), but \(g'\) does not vanish anywhere on the segment from \(1\) to \(\omega \).

Solution. \(\omega = e^{2\pi i/3}\), so \(g(\omega ) = e^{2\pi i} = 1 = g(1)\). If the mean value theorem held, some \(c\) on the segment would have \(g'(c) = 3c^2 = 0\), that is, \(c = 0\). But the segment consists of the points \(1 + t(\omega - 1)\), \(0 \leq t \leq 1\), with imaginary part \(\frac {\sqrt 3}{2} t\), which is zero only at \(t = 0\), the point \(1\). So \(0\) is not on the segment. \(\relax \vartriangleleft \)

Remark 15.2. The exponential gives another example: \(e^0 = e^{2\pi i}\), yet \(e^z\) is never zero.

References

[1]
E. B. Saff and A. D. Snider, Fundamentals of Complex Analysis with Applications to Engineering and Science, 3rd ed., Prentice Hall, Upper Saddle River, NJ, 2003.