1.2 Power Series and the Elementary Functions
Power series and their discs of convergence, and the functions they define: the exponential, the trigonometric and hyperbolic functions, the logarithm with its branches, and complex powers — together with the identities that survive in $\mathbb{C}$ and the ones that do not.
In real analysis $e^x$, $\sin x$ and $\log x$ are introduced in different ways — as the inverse of an integral, through geometry, through a limit — and their Taylor series come afterwards. In complex analysis the order is reversed: the series come first. A power series $\sum a_nz^n$ converges on a disc, the functions $e^z$, $\sin z$, $\cos z$ are defined by their series, and every property — the addition formulas, periodicity, Euler's formula $e^{i\theta}=\cos\theta+i\sin\theta$ — is proved from the series. The payoff is that these functions now live on the whole plane, where they reveal facts invisible on the real line: $e^z$ is periodic with the imaginary period $2\pi i$, $\sin z$ is unbounded, $\cos z=2$ has solutions, and the logarithm is not a single function at all but a family of branches.
The second half of the section is about exactly that multivaluedness. Because $e^z$ is periodic it has no global inverse; choosing an inverse means choosing a branch, and with branches the familiar identities $\log(zw)=\log z+\log w$ and $(z^a)^b=z^{ab}$ can fail. CSIR NET asks about radii of convergence (often with a $\limsup$ trap), behaviour on the boundary circle, solutions of $\sin z=w$ or $e^z=w$, and the values of expressions such as $i^i$ — and Part C options are full of identities that look right but are false for principal values.
Sequences and Series of Complex Numbers
Convergence in $\mathbb{C}$ is convergence in the plane with the distance $|z-w|$. All of the real theory (Unit 1, Chapter 2) carries over, because a complex sequence converges exactly when its real and imaginary parts do.
A sequence $(z_n)$ in $\mathbb{C}$ converges to $z$ if for every $\varepsilon\gt0$ there is $N$ with $|z_n-z|\lt\varepsilon$ for all $n\ge N$. It is a Cauchy sequence if for every $\varepsilon\gt0$ there is $N$ with $|z_m-z_n|\lt\varepsilon$ for all $m,n\ge N$.
A series $\sum_{n\ge0}z_n$ converges to $S$ if its partial sums $S_N=\sum_{n=0}^{N}z_n$ converge to $S$; it converges absolutely if $\sum|z_n|$ converges. A convergent series that does not converge absolutely converges conditionally.
- $z_n\to z$ iff $\operatorname{Re}z_n\to\operatorname{Re}z$ and $\operatorname{Im}z_n\to\operatorname{Im}z$.
- $\mathbb{C}$ is complete: every Cauchy sequence converges.
- An absolutely convergent series converges, and $\left|\sum z_n\right|\le\sum|z_n|$.
- If $\sum z_n$ converges then $z_n\to0$.
- The geometric series $\sum_{n\ge0}z^n$ converges iff $|z|\lt1$, and then its sum is $\dfrac1{1-z}$.
Proof. (1) follows from the inequalities $$\max(|\operatorname{Re}w|,|\operatorname{Im}w|)\le|w|\le|\operatorname{Re}w|+|\operatorname{Im}w|,$$ applied to $w=z_n-z$. (2) The same inequalities show that $(\operatorname{Re}z_n)$ and $(\operatorname{Im}z_n)$ are real Cauchy sequences; by completeness of $\mathbb{R}$ they converge, and (1) finishes. (3) For $M\gt N$, $|S_M-S_N|=\left|\sum_{n=N+1}^Mz_n\right|\le\sum_{n=N+1}^M|z_n|$, which is small when $\sum|z_n|$ converges (Cauchy criterion for the real series). So $(S_N)$ is Cauchy and converges by (2); the inequality follows by letting $N\to\infty$ in $|S_N|\le\sum_{n\le N}|z_n|$. (4) $z_N=S_N-S_{N-1}\to S-S=0$. (5) $S_N=\dfrac{1-z^{N+1}}{1-z}$ for $z\ne1$. If $|z|\lt1$, $|z|^{N+1}\to0$ and $S_N\to\frac1{1-z}$. If $|z|\ge1$, $|z^n|\ge1$ does not tend to $0$, so the series diverges by (4). ∎
Proof. (a) If $L\lt1$ choose $q$ with $L\lt q\lt1$; by definition of $\limsup$, $|z_n|^{1/n}\lt q$ for all large $n$, so $|z_n|\lt q^n$ and $\sum|z_n|$ converges by comparison with a geometric series. If $L\gt1$, then $|z_n|^{1/n}\gt1$ for infinitely many $n$, so $|z_n|\gt1$ infinitely often, $z_n\not\to0$, and the series diverges. (b) If $|z_{n+1}/z_n|\to\ell$ then $|z_n|^{1/n}\to\ell$ (by the standard inequalities $\liminf\left|\tfrac{z_{n+1}}{z_n}\right|\le\liminf|z_n|^{1/n}$ and $\limsup|z_n|^{1/n}\le\limsup\left|\tfrac{z_{n+1}}{z_n}\right|$ from Unit 1), so $L=\ell$. For $L=1$: $\sum\frac1n$ diverges and $\sum\frac1{n^2}$ converges. ∎
The root test is stronger than the ratio test: whenever the ratio limit exists the root limit exists and agrees, but not conversely. For conditionally convergent series — which is what happens on the boundary of a disc of convergence — we need a finer tool.
(b) Dirichlet's test. If the partial sums $A_n$ of $\sum a_n$ (complex) are bounded, and $(b_n)$ is a real sequence decreasing to $0$, then $\sum a_nb_n$ converges.
Proof. (a) Put $A_{-1}=0$ and write $a_n=A_n-A_{n-1}$: $\sum_{n=0}^Na_nb_n=\sum_{n=0}^NA_nb_n-\sum_{n=0}^{N-1}A_nb_{n+1}$, which rearranges to the formula. (b) Let $|A_n|\le M$. Then $|A_Nb_N|\le Mb_N\to0$, and $\sum_n|A_n(b_{n+1}-b_n)|\le M\sum_n(b_n-b_{n+1})$, a telescoping sum equal to $Mb_0\lt\infty$, so the series in (a) converges absolutely. Hence $\sum_{n\le N}a_nb_n$ has a limit. ∎
Power Series and the Radius of Convergence
Proof. Apply the root test (Theorem 1.19) to $z_n=a_n(z-z_0)^n$: $\limsup|z_n|^{1/n}=|z-z_0|\cdot\limsup|a_n|^{1/n}$, which is $|z-z_0|/R'$, where $R'$ is defined by the formula. So the series converges absolutely when $|z-z_0|\lt R'$ and diverges when $|z-z_0|\gt R'$. Divergence means $a_n(z-z_0)^n\not\to0$; convergence gives bounded terms. Comparing with Definition 1.9, $R=R'$. For uniformity, let $r\lt\rho\lt R$. Since $\sum|a_n|\rho^n$ converges, its terms are bounded, say $|a_n|\rho^n\le M$; then for $|z-z_0|\le r$, $|a_n(z-z_0)^n|\le M(r/\rho)^n$, and the Weierstrass M-test with the convergent geometric series $\sum M(r/\rho)^n$ gives uniform convergence. ∎
Proof. If $|a_{n+1}/a_n|\to1/\lambda$ then $|a_n|^{1/n}\to1/\lambda$ by the ratio–root inequality used in Theorem 1.19(b), and Theorem 1.21 gives $R=\lambda$. ∎
Proof. Take $z_0=0$. Since $n^{1/n}\to1$, $\limsup|na_n|^{1/n}=\limsup|a_n|^{1/n}$, so the radii agree. Fix $|z|\lt r\lt R$ and $0\lt|h|\lt r-|z|$, and let $g(z)=\sum na_nz^{n-1}$. Then $$\frac{f(z+h)-f(z)}{h}-g(z)=\sum_{n\ge2}a_n\left(\frac{(z+h)^n-z^n}{h}-nz^{n-1}\right).$$ Using $\frac{u^n-v^n}{u-v}=\sum_{k=0}^{n-1}u^kv^{n-1-k}$ twice, the bracket equals $h\sum_{k=1}^{n-1}k\,(z+h)^{n-1-k}z^{k-1}$, whose modulus is at most $|h|\binom n2r^{n-2}$, since $|z|,|z+h|\lt r$. Hence the difference is at most $|h|\sum_{n\ge2}|a_n|\binom n2r^{n-2}$; this series converges (twice-derived series, radius $R\gt r$), so the difference tends to $0$ as $h\to0$. Repeating, $f^{(k)}(0)=k!\,a_k$. ∎
So the sum of a power series is complex-differentiable — analytic — in its disc of convergence (§2.1). The converse, that every analytic function is locally the sum of its Taylor series, is one of the deep theorems of Chapter 3; it explains why $R$ is the distance to the nearest singularity. That is already visible here: the real function $\dfrac1{1+x^2}$ is smooth on all of $\mathbb{R}$, yet its Maclaurin series $\sum(-1)^nx^{2n}$ has radius $1$ — because $\dfrac1{1+z^2}$ blows up at $z=\pm i$, at distance $1$ from $0$.
When a series does converge at a boundary point, the sum there is what one expects — the limit of the sum from inside along the radius. This is Abel's theorem, and it is how one evaluates sums such as $1-\frac12+\frac13-\cdots=\log2$.
Proof. Replacing $a_0$ by $a_0-A$ we may assume $A=0$. Let $A_n=\sum_{k\le n}a_k\to0$. For $0\lt r\lt1$, summation by parts (Theorem 1.20(a)) and $A_Nr^N\to0$ give $\sum a_nr^n=(1-r)\sum_{n\ge0}A_nr^n$. Given $\varepsilon\gt0$ choose $N$ with $|A_n|\lt\varepsilon$ for $n\gt N$. Then $$\left|\sum a_nr^n\right|\le(1-r)\sum_{n\le N}|A_n|+(1-r)\,\varepsilon\sum_{n\gt N}r^n\le(1-r)\sum_{n\le N}|A_n|+\varepsilon,$$ and the first term tends to $0$ as $r\to1^-$. So $\limsup_{r\to1^-}|\sum a_nr^n|\le\varepsilon$ for every $\varepsilon$. ∎
For example $\sum_{n\ge1}\frac{(-1)^{n+1}}{n}x^n=\log(1+x)$ for $|x|\lt1$, and the series converges at $x=1$, so its sum there is $\lim_{x\to1^-}\log(1+x)=\log2$. The converse of Abel's theorem is false: $\sum(-1)^nr^n=\frac1{1+r}\to\frac12$, yet $\sum(-1)^n$ diverges.
The Exponential, Trigonometric and Hyperbolic Functions
Each series has radius $\infty$ (ratio test), so these are entire functions — analytic on all of $\mathbb{C}$ (Theorem 1.23). For real $z$ they are the familiar real functions, since the series are their Maclaurin series. As usual $\tan z=\sin z/\cos z$ where $\cos z\ne0$.
- Addition law: $e^{z+w}=e^ze^w$ for all $z,w$. Hence $e^z\ne0$ and $1/e^z=e^{-z}$.
- Euler's formula: $e^{iy}=\cos y+i\sin y$ for real $y$, and so $e^{x+iy}=e^x(\cos y+i\sin y)$.
- $|e^z|=e^{\operatorname{Re}z}$, $\ \operatorname{Im}z$ is an argument of $e^z$, and $\overline{e^z}=e^{\bar z}$; $\ (e^z)'=e^z$.
- $e^z=1$ iff $z=2\pi ik$ for some $k\in\mathbb{Z}$. Hence $e^z=e^w$ iff $z-w\in2\pi i\mathbb{Z}$: $e^z$ is periodic, and its periods are exactly the numbers $2\pi ik$.
- $e^z$ takes every value $w\ne0$, infinitely often, and never the value $0$.
Proof. (1) Both series converge absolutely, so their product may be computed as the Cauchy product (for absolutely convergent series every arrangement of the double series $\sum_{m,k}u_mv_k$ has the same sum, and grouping by $m+k=n$ is one of them): $$e^ze^w=\sum_{n\ge0}\sum_{k=0}^n\frac{z^k}{k!}\frac{w^{n-k}}{(n-k)!}=\sum_{n\ge0}\frac1{n!}\sum_{k=0}^n\binom nkz^kw^{n-k}=\sum_{n\ge0}\frac{(z+w)^n}{n!}=e^{z+w}.$$ Then $e^ze^{-z}=e^0=1$.
(2) Split the series for $e^{iy}$ into even and odd powers, using $i^{2m}=(-1)^m$: $$e^{iy}=\sum\frac{(-1)^my^{2m}}{(2m)!}+i\sum\frac{(-1)^my^{2m+1}}{(2m+1)!},$$ which is $\cos y+i\sin y$. Combine with (1). (This justifies the notation $e^{i\theta}$ of §1.1.)
(3) $|e^{x+iy}|=e^x|\cos y+i\sin y|=e^x$, and (2) is the polar form with argument $y$. Conjugating the series term by term gives $\overline{e^z}=e^{\bar z}$. The derivative follows from Theorem 1.23: $\sum\frac{nz^{n-1}}{n!}=\sum\frac{z^{n-1}}{(n-1)!}$.
(4) $e^{x+iy}=1$ iff $e^x=1$ and $y$ is an argument of $1$, i.e. iff $x=0$ and $y\in2\pi\mathbb{Z}$. Then $e^z=e^w\iff e^{z-w}=1$ by (1).
(5) Given $w\ne0$, the numbers $z=\log|w|+i(\operatorname{Arg}w+2\pi k)$, $k\in\mathbb{Z}$, satisfy $e^z=|w|e^{i\operatorname{Arg}w}=w$ by (2); they are distinct. And $e^z\ne0$ by (1). ∎
Geometrically, $z=x+iy\mapsto e^z$ sends the horizontal line $\operatorname{Im}z=c$ onto the open ray from $0$ at angle $c$, and the vertical segment $\{x\}\times(-\pi,\pi]$ once around the circle of radius $e^x$. Any horizontal strip $a\lt\operatorname{Im}z\le a+2\pi$ is mapped bijectively onto $\mathbb{C}\setminus\{0\}$ (Interactive 2).
- $\cos z=\dfrac{e^{iz}+e^{-iz}}{2}$, $\ \sin z=\dfrac{e^{iz}-e^{-iz}}{2i}$, $\ \cosh z=\dfrac{e^z+e^{-z}}2=\cos(iz)$, $\ \sinh z=\dfrac{e^z-e^{-z}}2=-i\sin(iz)$; $\ e^{iz}=\cos z+i\sin z$.
- All the real identities persist: $\sin^2z+\cos^2z=1$, $\ \cosh^2z-\sinh^2z=1$, $\ \sin(z+w)=\sin z\cos w+\cos z\sin w$, $\ \cos(z+w)=\cos z\cos w-\sin z\sin w$, $\ (\sin z)'=\cos z$, $\ (\cos z)'=-\sin z$.
- $\sin z=\sin x\cosh y+i\cos x\sinh y$ and $\cos z=\cos x\cosh y-i\sin x\sinh y$; hence $$|\sin z|^2=\sin^2x+\sinh^2y,\qquad|\cos z|^2=\cos^2x+\sinh^2y.$$
- Zeros: $\sin z=0\iff z=k\pi$ and $\cos z=0\iff z=\tfrac\pi2+k\pi$ ($k\in\mathbb{Z}$) — only the real zeros. $\cosh z=0\iff z=i\left(\tfrac\pi2+k\pi\right)$ and $\sinh z=0\iff z=k\pi i$.
- Periods: $\sin$ and $\cos$ have exactly the periods $2k\pi$; $\sinh$ and $\cosh$ exactly the periods $2k\pi i$.
- Unboundedness: $|\sin(iy)|=|\sinh y|\to\infty$ as $|y|\to\infty$; in fact $\sin$ and $\cos$ take every complex value.
Proof. (1) Add and subtract the series for $e^{iz}$ and $e^{-iz}$, as in the proof of Euler's formula; the hyperbolic formulas come from $e^{\pm z}$ in the same way, and substituting $iz$ gives $\cosh z=\cos(iz)$, $\sinh z=-i\sin(iz)$. (2) From (1), $4\cos^2z=(e^{iz}+e^{-iz})^2$ and $4\sin^2z=-(e^{iz}-e^{-iz})^2$; adding, $4(\cos^2z+\sin^2z)=4e^{iz}e^{-iz}=4$. For the addition formula expand $\sin z\cos w+\cos z\sin w$ with (1): the products $e^{\pm iz}e^{\pm iw}=e^{\pm i(z\pm w)}$ (Theorem 1.25(1)) combine to $\frac{e^{i(z+w)}-e^{-i(z+w)}}{2i}$. The others are alike; the derivatives follow from $(e^{iz})'=ie^{iz}$.
(3) The addition formula with $w=iy$ gives $\sin(x+iy)=\sin x\cos(iy)+\cos x\sin(iy)$, and $\cos(iy)=\cosh y$, $\sin(iy)=i\sinh y$ by (1). Then $|\sin z|^2=\sin^2x\cosh^2y+\cos^2x\sinh^2y$; writing $\cosh^2y=1+\sinh^2y$ this is $\sin^2x+(\sin^2x+\cos^2x)\sinh^2y$, that is, $\sin^2x+\sinh^2y$. Similarly for $\cos$.
(4) $|\sin z|^2=0\iff\sin x=0$ and $\sinh y=0\iff x=k\pi$, $y=0$. Similarly for $\cos$. For the hyperbolic functions use $\cosh z=\cos(iz)$: $\cos(iz)=0\iff iz=\tfrac\pi2+k\pi$, i.e. $z=-i(\tfrac\pi2+k\pi)$, which is the same set. (5) $\sin(z+p)=\sin z$ for all $z$ gives, at $z=0$ and $z=\tfrac\pi2$, $\sin p=0$ and $\cos p=1$; by (4) $p=k\pi$ with $\cos k\pi=1$, so $p\in2\pi\mathbb{Z}$. The others follow by the substitution $z\mapsto iz$.
(6) Given $c\in\mathbb{C}$, put $u=e^{iz}$: $\sin z=c\iff u-u^{-1}=2ic$, i.e. $u^2-2icu-1=0$. This quadratic has roots $u_1,u_2$ with $u_1u_2=-1$, so both are nonzero, and $e^{iz}=u_1$ has solutions by Theorem 1.25(5). Similarly for $\cos$. ∎
The Complex Logarithm, Branches and Complex Powers
Since $e^z$ takes every nonzero value infinitely often, the equation $e^w=z$ has infinitely many solutions $w$, spaced $2\pi i$ apart. The logarithm is therefore not a function but a choice.
For $z\ne0$, the set of logarithms of $z$ is $$\log z=\{w:e^w=z\}=\{\ln|z|+i\theta:\theta\in\arg z\},$$ where $\ln$ is the real natural logarithm. The principal logarithm is $\operatorname{Log}z=\ln|z|+i\operatorname{Arg}z$, with imaginary part in $(-\pi,\pi]$.
A branch of the logarithm on an open set $D\subseteq\mathbb{C}\setminus\{0\}$ is a continuous function $L:D\to\mathbb{C}$ with $e^{L(z)}=z$ for all $z\in D$. The principal branch is $\operatorname{Log}$ restricted to $\mathbb{C}\setminus(-\infty,0]$, the plane cut along the non-positive real axis (the branch cut).
- $e^{\operatorname{Log}z}=z$ for all $z\ne0$, but $\operatorname{Log}(e^w)=w$ only when $-\pi\lt\operatorname{Im}w\le\pi$; in general $\operatorname{Log}(e^w)=w-2\pi ik$ for the integer $k$ that brings the imaginary part into $(-\pi,\pi]$.
- $\operatorname{Log}$ is continuous on $\mathbb{C}\setminus(-\infty,0]$ and discontinuous at every point of $(-\infty,0)$: for $x\lt0$, $\operatorname{Log}z\to\ln|x|\pm i\pi$ as $z\to x$ from the upper / lower half-plane. On the cut plane it is analytic with $(\operatorname{Log}z)'=1/z$.
- Any two branches of the logarithm on a connected open set differ by a constant $2\pi ik$.
- There is no branch of the logarithm on $\mathbb{C}\setminus\{0\}$, nor on any open set containing a circle $|z|=r$.
Proof. (1) $e^{\operatorname{Log}z}=e^{\ln|z|}e^{i\operatorname{Arg}z}=z$. Conversely $\operatorname{Log}(e^w)$ has imaginary part $\operatorname{Arg}e^w\in(-\pi,\pi]$, and it differs from $w$ by an element of $2\pi i\mathbb{Z}$ (both are logarithms of $e^w$), which gives the statement.
(2) $\ln|z|$ is continuous on $\mathbb{C}\setminus\{0\}$. On the cut plane, $\operatorname{Arg}z=2\arctan\dfrac{y}{|z|+x}$ (the half-angle formula: $\tan\tfrac\theta2=\tfrac{\sin\theta}{1+\cos\theta}$, valid for $|\theta|\lt\pi$), and $|z|+x\gt0$ there, so $\operatorname{Arg}$ is continuous. At $x\lt0$, points $x+i\varepsilon$ with $\varepsilon\downarrow0$ have $\operatorname{Arg}\to\pi$ and points $x-i\varepsilon$ have $\operatorname{Arg}\to-\pi$: a jump of $2\pi$ in the imaginary part. Analyticity: if $w=\operatorname{Log}z$ and $w_0=\operatorname{Log}z_0$ on the cut plane, then $w\to w_0$ as $z\to z_0$ (continuity) and $$\frac{\operatorname{Log}z-\operatorname{Log}z_0}{z-z_0}=\frac{w-w_0}{e^w-e^{w_0}}\to\frac1{e^{w_0}}=\frac1{z_0},$$ because $(e^w)'=e^w$.
(3) If $L_1,L_2$ are branches on a connected $D$, then $e^{L_1-L_2}=1$, so $h=(L_1-L_2)/(2\pi i)$ is a continuous integer-valued function on $D$. A continuous function into $\mathbb{Z}$ on a connected set is constant (the set where it equals a given integer is open and closed).
(4) Suppose $L$ is a branch on an open set containing the circle $|z|=r$. Define $g(\theta)=\dfrac{L(re^{i\theta})-\ln r-i\theta}{2\pi i}$ for $\theta\in[0,2\pi]$. It is continuous, and $e^{2\pi ig(\theta)}=\dfrac{re^{i\theta}}{re^{i\theta}}=1$, so $g$ takes integer values; by the intermediate value theorem it is constant. But $re^{i\cdot2\pi}=re^{i\cdot0}$, so $g(2\pi)=g(0)-1$. Contradiction. ∎
Proof. $e^{a}=z$ and $e^{b}=w$ imply $e^{a+b}=zw$, and conversely every logarithm of $zw$ is $a+b$ plus some $2\pi ik$, which can be absorbed into $b$. For principal values, the real parts agree ($\ln|zw|=\ln|z|+\ln|w|$) and the imaginary parts $\operatorname{Arg}(zw)$ and $\operatorname{Arg}z+\operatorname{Arg}w$ are both arguments of $zw$, so they differ by $2\pi N$. Since $\operatorname{Arg}z+\operatorname{Arg}w\in(-2\pi,2\pi]$ and $\operatorname{Arg}(zw)\in(-\pi,\pi]$, $|2\pi N|\lt3\pi$, so $N\in\{-1,0,1\}$, and $N=0$ exactly when the sum already lies in $(-\pi,\pi]$. If both real parts are positive, both arguments lie in $(-\tfrac\pi2,\tfrac\pi2)$. ∎
Proof. The values are $e^{a(\operatorname{Log}z+2\pi ik)}$. Two of them, for $k$ and $m$, coincide iff $e^{2\pi i(k-m)a}=1$ iff $(k-m)a\in\mathbb{Z}$ (Theorem 1.25(4)). If $a\in\mathbb{Z}$ this always holds. If $a=p/q$ in lowest terms, $(k-m)p/q\in\mathbb{Z}\iff q\mid(k-m)$, so there are exactly $q$ distinct values ($k=0,\dots,q-1$); each satisfies $w^q=e^{p(\operatorname{Log}z+2\pi ik)}=z^p$, and $z^p$ has exactly $q$ $q$-th roots (Theorem 1.5). If $a\notin\mathbb{Q}$, then $(k-m)a\in\mathbb{Z}$ forces $k=m$, so all values are distinct. ∎
Inverse trigonometric functions. Solving $\sin w=z$ as in Theorem 1.26(6) — a quadratic in $e^{iw}$ — and taking logarithms gives the multivalued formulas $$\sin^{-1}z=-i\log\left(iz+(1-z^2)^{1/2}\right),\qquad\cos^{-1}z=-i\log\left(z+i(1-z^2)^{1/2}\right),$$ $$\tan^{-1}z=\frac1{2i}\log\frac{1+iz}{1-iz}\qquad(z\ne\pm i),$$ where both values of the square root and all values of the logarithm are allowed. For $\tan^{-1}$: $\tan w=z\iff e^{2iw}-1=iz\left(e^{2iw}+1\right)$, i.e. $e^{2iw}(1-iz)=1+iz$. Since $e^{2iw}\ne0$, this has no solution when $z=i$ (it would give $2e^{2iw}=0$) and none when $z=-i$ (it would give $0=2$): $\tan w$ never takes the values $\pm i$. Principal branches are obtained by using $\operatorname{Log}$ and the principal square root, with suitable cuts.
- The ratio test can fail when the root test works. For $\sum(3+(-1)^n)^nz^n$ the ratios have no limit; $\limsup|a_n|^{1/n}=4$ gives $R=\tfrac14$. Series with gaps ($\sum z^{n^2}$, $\sum z^{2n}$) need the root test or a substitution.
- $\limsup$, not $\liminf$. $1/R=\limsup|a_n|^{1/n}$; with $\liminf$ you get a wrong (too large) radius.
- The boundary is a separate question. $\sum z^n$ diverges at every boundary point, $\sum z^n/n$ converges at every boundary point but one, $\sum z^n/n^2$ at all of them. Convergence at one boundary point says nothing about another: $\sum(-1)^nz^n/n$ converges at $z=1$ and diverges at $z=-1$.
- Absolute convergence inside does not mean uniform convergence inside. $\sum z^n$ converges on $|z|\lt1$ but not uniformly there (its sum is unbounded); uniformity holds on each smaller closed disc.
- Real smoothness does not give a big radius. $\frac1{1+x^2}$ is smooth on $\mathbb{R}$ but its Maclaurin series has $R=1$, because of the complex singularities $\pm i$.
- $\sin$ and $\cos$ are unbounded in $\mathbb{C}$, and take every value: $\sin z=2$ and $\cos z=2$ have solutions. Only their zeros stay real.
- $\cosh$ and $\sinh$ do have non-real zeros: $\cosh\left(\tfrac{i\pi}2\right)=\cos\tfrac\pi2=0$. "$\cosh z\ge1$" is a real-variable fact.
- $e^z$ is not injective (period $2\pi i$); it omits exactly one value, $0$. It is injective on every horizontal strip $a\lt\operatorname{Im}z\le a+2\pi$, and on nothing taller.
- $\operatorname{Log}(zw)\ne\operatorname{Log}z+\operatorname{Log}w$ and $\operatorname{Log}(e^w)\ne w$ in general; $\operatorname{Log}(z^2)=2\operatorname{Log}z$ holds exactly when $\operatorname{Arg}z\in(-\tfrac\pi2,\tfrac\pi2]$ (Example 7).
- No logarithm (and no square root) is continuous around $0$. Every branch needs a cut from $0$ to $\infty$ (Theorem 1.27(4)).
- Complex powers are multivalued: $i^i$ is a set of infinitely many real numbers; $1^{\sqrt2}$ has infinitely many values; the principal value of $(-8)^{1/3}$ is $1+\sqrt3\,i$, not $-2$; $(z^a)^b\ne z^{ab}$ (e.g. $((-1)^2)^{1/2}=1\ne-1$).
- $\tan z$ never equals $\pm i$, so $\tan:\mathbb{C}\setminus\{\tfrac\pi2+k\pi\}\to\mathbb{C}$ is not onto.
Exercises
- 1 Find the radius of convergence of (a) $\sum\dfrac{n!}{n^n}z^n$; (b) $\sum\left(1+\tfrac1n\right)^{n^2}z^n$; (c) $\sum3^nz^{2n}$; (d) $\sum\dfrac{(2+i)^n}{n}z^n$; (e) $\sum z^{n!}$.
- 2 Determine exactly where on $|z|=1$ the series $\sum_{n\ge1}\dfrac{(-1)^n}{n}z^n$ converges.
- 3 Show that $\sum_{n\ge1}n^2z^n=\dfrac{z(1+z)}{(1-z)^3}$ for $|z|\lt1$.
- 4 Find the sum of $\sum_{n\ge0}\dfrac{n\,z^n}{n!}$.
- 5 Show that $\sum_{n\ge1}z^n/n$ does not converge uniformly on the open disc $|z|\lt1$.
- 6 Let $\sum a_nz^n$ have radius $R\in(0,\infty)$. Find the radii of $\sum a_nz^{2n}$ and of $\sum a_n^2z^n$.
- 7 Give a power series with radius $1$ that converges at $z=1$ but diverges at $z=-1$.
- 8 Use Abel's theorem to find $\sum_{n\ge0}\dfrac{(-1)^n}{2n+1}$ from the series of $\tan^{-1}x$.
- 1 Write $e^{2+i\pi/3}$ in the form $a+ib$.
- 2 Find all $z$ with $e^z=-2$.
- 3 Find the real and imaginary parts and the modulus of $\sin(1+2i)$.
- 4 Solve $\sin z=i$.
- 5 Find all zeros of $\cosh z$ and of $1+e^z$.
- 6 Prove that $|\sinh y|\le|\sin(x+iy)|\le\cosh y$.
- 7 Prove that $|\cos z|^2+|\sin z|^2=\cosh(2\operatorname{Im}z)$.
- 8 Prove that $|e^{z^2}|\le e^{|z|^2}$, and find when equality holds.
- 9 Show that $\overline{\sin z}=\sin\bar z$, and find all $z$ for which $\sin z$ is real.
- 1 Find $\operatorname{Log}(-1-i)$ and the set $\log(-1-i)$.
- 2 Find the principal values of $(-1)^i$ and $2^i$.
- 3 Find all values of $i^{1/3}$.
- 4 Compute the one-sided limits of $\operatorname{Log}z$ as $z\to-1$ from the upper and from the lower half-plane.
- 5 Find $z,w$ with $\operatorname{Log}(zw)\ne\operatorname{Log}z+\operatorname{Log}w$, and compute the integer $N$ of Theorem 1.28 for them.
- 6 Solve $\tan z=2i$.
- 7 Show that the principal square root $z^{1/2}=e^{\frac12\operatorname{Log}z}$ has positive real part for every $z\in\mathbb{C}\setminus(-\infty,0]$.
- 8 Show that $((-1)^2)^{1/2}\ne(-1)^{2\cdot\frac12}$ for principal values, and explain which rule fails.
- 1 Let $f(z)=\sum a_nz^n$ have radius $R\gt0$ and suppose $f(1/k)=0$ for all large $k\in\mathbb{N}$. Prove that $a_n=0$ for all $n$.
- 2 Let $f(z)=\sum_{n\ge0}z^{2^n}$. Show that $R=1$ and that $|f(r\zeta)|\to\infty$ as $r\to1^-$ whenever $\zeta^{2^m}=1$ for some $m$. Conclude that $f$ cannot be continued continuously to any arc of the unit circle.
- 3 Prove that there is no continuous $g$ on the unit circle with $g(z)^2=z$ for all $|z|=1$.
- 4 Prove that $e^z$ is injective on the strip $\{-\pi\lt\operatorname{Im}z\le\pi\}$ and maps it onto $\mathbb{C}\setminus\{0\}$.
- 5 Prove that the only periods of $e^z$, i.e. the $p$ with $e^{z+p}=e^z$ for all $z$, are $p=2\pi ik$.
- 6 Prove that $\cos$ is not injective on any open set containing $0$, and find the largest $\delta$ such that $\sin$ is injective on the disc $|z|\lt\delta$.
E1. CSIR Part B GATE MA The radius of convergence of $\sum_{n\ge1}\left(3+(-1)^n\right)^nz^n$ is (A) $1$; (B) $\tfrac12$; (C) $\tfrac13$; (D) $\tfrac14$.
E2. CSIR Part B The number of solutions of $e^z=1+i$ in the strip $\{z:0\le\operatorname{Im}z\lt10\pi\}$ is (A) $1$; (B) $4$; (C) $5$; (D) infinitely many.
E3. CSIR Part B The image of the strip $\{z:0\lt\operatorname{Im}z\lt\pi\}$ under $w=e^z$ is (A) the upper half-plane $\operatorname{Im}w\gt0$; (B) the right half-plane $\operatorname{Re}w\gt0$; (C) the punctured unit disc $0\lt|w|\lt1$; (D) $\mathbb{C}\setminus[0,\infty)$.
E4. CSIR Part C Let $\sum_{n\ge0}a_nz^n$ have radius of convergence $1$. Which of the following are necessarily true? (A) $\sum na_nz^n$ has radius of convergence $1$. (B) $\sum a_nz^n$ converges at some point of the circle $|z|=1$. (C) $\sum a_nz^{2n}$ has radius of convergence $1$. (D) If $\sum a_nz^n$ converges at $z=1$, then it converges at $z=-1$.
E5. CSIR Part C GATE MA Which of the following statements are true? (A) $|\sin z|\le1$ for every $z$ with $|\operatorname{Im}z|\le1$. (B) Every zero of $\cos z$ is real. (C) $e^z=e^w$ if and only if $z-w\in2\pi i\mathbb{Z}$. (D) $\sin z$ is bounded on the strip $|\operatorname{Im}z|\le1$.
E6. CSIR Part C Using principal values throughout, which of the following are true? (A) $\operatorname{Log}\bar z=\overline{\operatorname{Log}z}$ for every $z\in\mathbb{C}\setminus(-\infty,0]$. (B) $\operatorname{Log}(z_1z_2)=\operatorname{Log}z_1+\operatorname{Log}z_2$ whenever $\operatorname{Re}z_1\gt0$ and $\operatorname{Re}z_2\gt0$. (C) $i^i$ is a real number. (D) $(z^2)^{1/2}=z$ for every $z\ne0$.
Suggested plan — 3 lectures × 60 min.
- Series and radius. Theorems 1.18–1.20 quickly (they are Unit 1 facts in a new setting), then Cauchy–Hadamard with proof (Theorem 1.21) and Example 1 — let the class find the trap in (d). Theorem 1.23 stated, proof outlined. Example 2 with Interactive 1 projected: put $z$ on the circle at angle $180^\circ$, then move the angle to $0^\circ$ for $\sum z^n/n$.
- The exponential and trigonometric functions. Definition 1.10; prove the addition law via the Cauchy product and Euler's formula; Theorem 1.25(4)–(5). Theorem 1.26 parts (3) and (4) on the board; Example 4(b) as a model for every "solve $\sin z=c$" problem; Example 5. Show $e^z$ and $\sin z$ in Interactive 2.
- Logarithms and powers. Definition 1.11 and Theorem 1.27 with the proof of (4) — the most conceptual argument of the section. Theorems 1.28–1.29, Examples 6–7, inverse trigonometric functions. Interactive 2 with $\operatorname{Log}$ and the size slider at maximum (the cut). Exam corner E4 and E6 live.
Discussion questions. (1) Why is it natural to define $e^z$ by a series rather than as a limit or an inverse function? (2) Is $i^i$ "a number"? Discuss what it means for an expression to be multivalued, and how calculators decide. (3) The real function $\frac{1}{1+x^2}$ is perfectly smooth; why does its Taylor series stop converging at $|x|=1$?
Misconceptions to watch for. Using $\liminf$ or the average in Cauchy–Hadamard; assuming convergence on the whole boundary circle, or nowhere on it; "$|\sin z|\le1$"; "$e^z$ is one-to-one"; treating $\operatorname{Log}$ as a homomorphism; writing $(-8)^{1/3}=-2$ for the principal value.
Suggested homework. I(1, 2, 5, 6, 8), II(2, 4, 6, 9), III(1, 3, 5, 6), IV(1, 3, 4); Exam corner E1, E2, E5.
Research question. The series $\sum z^n/n!$ converges for every $z$. Can a computer simply add up its terms to compute $e^z$ — and if not, how do scientific libraries actually do it?
Why it matters. Every simulation of waves, circuits or quantum systems evaluates $e^{i\omega t}$ millions of times; satellite navigation and audio signal processing depend on fast, accurate complex exponentials. Floating-point arithmetic keeps only about 16 significant digits, and an innocent-looking convergent series can lose all of them.
What to do.
- In double precision, sum $\sum_{n=0}^{N}z^n/n!$ for $z=20i$, $z=40i$ and $z=-30$, with $N$ large enough that the terms are negligible. Compare with the library value of $e^z$; record the absolute error.
- Find the size of the largest term in each case ($|z|^n/n!$ is largest near $n=|z|$) and explain the errors through cancellation: the result is tiny compared with the terms being added.
- Implement two cures: range reduction using periodicity ($e^{x+iy}=e^x(\cos y+i\sin y)$ with $y$ reduced modulo $2\pi$, and $e^{-30}=1/e^{30}$), and scaling and squaring ($e^z=(e^{z/2^k})^{2^k}$). Measure the errors again.
- Estimate how many terms of the series are needed for 15 correct digits when $|z|\le\tfrac12$, using the tail bound $\sum_{n\gt N}\frac{|z|^n}{n!}\le\frac{|z|^{N+1}}{(N+1)!}\cdot\frac{1}{1-|z|/(N+2)}$.
Tools: Python (cmath, or mpmath for high-precision reference values) or any language with complex floating point.
Hand in: A table of errors (naive series, range reduction, scaling and squaring), and a paragraph explaining the cancellation with the size of the largest term.
Going further. Read about the CORDIC algorithm, which computes $e^{i\theta}$ in hardware with only shifts and additions by writing $\theta$ as a sum of the angles $\arctan2^{-k}$; and about how the matrix exponential $e^{A}$ is computed in numerical linear algebra by the same scaling-and-squaring idea.
🧪 Self-check quiz
Q1. The radius of convergence of $\sum_{n\ge1}\dfrac{n^5}{2^n}z^n$ is
Q2. The equation $e^z=0$ has
Q3. For $z=x+iy$, $|e^{z}|$ equals
Q4. The principal value of $(-1)^{1/2}$ is
Q5. On the circle $|z|=1$, the series $\sum_{n\ge1}z^n/n$