Skip to main content

Math 382-0: Kursobjekt

Section 1.18 Analyticity of holomorphic functions

We now prove the converse of (1.68).

Proof.

Given \(w\in B_R(z_0)\) we can find an \(R'> 0\) such that \(w\in \overline{B}_{R'}(z_0)\subseteq B_R(z_0)\text{.}\) (See Figure 1.18.2.) Since \(f\) is analytic on \(B_R(z_0)\text{,}\) we have
\begin{equation*} f(w)=\frac{1}{2\pi i}\int_{\gamma_{R'}}\frac{f(z)}{z-w}\, dz\text{,} \end{equation*}
where \(\gamma_{R'}=w+R'e^{it}\text{,}\) \(t\in [0,2\pi]\text{.}\) Since \(R'=\abs{z-z_0}< \abs{w-z_0}\) for all \(z\in \gamma_{R'}\text{,}\) we have \(\abs{(w-z_0)/(z-z_0)}< 1\text{,}\) and hence
\begin{align*} \frac{1}{z-w} \amp =\frac{1}{z-z_0-(w-z_0}\\ \amp =\frac{1}{z-z_0}\cdot \frac{1}{1-(w-z_0)/(z-z_0)}\\ \amp = \frac{1}{z-z_0}\sum_{n=0}^\infty ((w-z_0)/(z-z_0))^n\\ \amp = \sum_{n=0}^\infty \frac{1}{(z-z_0)^{n+1}}(w-z_0)^n \end{align*}
for all \(z\in \gamma_{R'}\text{.}\) We then have
\begin{align*} f(w)\amp =\frac{1}{2\pi i}\int_{\gamma_{R'}}\frac{f(z)}{z-w}\, dz \\ \amp = \frac{1}{2\pi i}\int_{\gamma_{R'}} \left(\sum_{n=0}^\infty \frac{f(z)}{(z-z_0)^{n+1}}(w-z_0)^n\right) \, dz \\ \amp =\frac{1}{2\pi i}\sum_{n=0}^\infty \int_{\gamma_{R'}}\frac{1}{(z-z_0)^{n+1}}(w-z_0)^n\, dz \\ \amp = \frac{1}{2\pi i}\sum_{n=0}^\infty \left(\int_{\gamma_{R'}}\frac{1}{(z-z_0)^{n+1}}\, dz\right)(w-z_0)^n\\ \amp = \sum_{n=0}^\infty a_n(w-z_0)^n\text{,} \end{align*}
where
\begin{equation*} a_n=\frac{1}{2\pi i}\int_{\gamma_{R'}}\frac{1}{(z-z_0)^{n+1}}\, dz\text{.} \end{equation*}
Of course, there is one subtle detail to consider here: namely, why are we justified in swapping the order of the integral and infinite sum? The answer, as usual, hinges on a uniformly convergent sequence. In this case, letting
\begin{equation*} f_k(z)=\frac{f(z)}{(z-z_0)}g_k(z), g_k(z)=\sum_{n=0}^k\left((w-z_0)/(z-z_0)\right)^n\text{.} \end{equation*}
We claim that
\begin{equation*} f_k\to \frac{f(z)}{w-z_0}=\frac{f(z)}{z-z_0}\cdot \frac{1}{1-(w-z_0)/(z-z_0)} \end{equation*}
uniformly on \(\gamma_{R'}\text{.}\) Indeed, since \(g(z)=\frac{f(z)}{(z-z_0)}\) is continuous on \(\gamma_{R'}\text{,}\) we have (using the extreme value theorem) \(\abs{g(z)}\leq M\) for some \(M\geq 0\text{.}\) Furthermore, since power series satisfy a uniform convergence property, we know that \(g_k\to \frac{1}{1-(w-z_0)/(z-z_0)}\) uniformly. Given any \(\epsilon > 0\text{,}\) we can thus pick an integer \(N\geq 0\) such that \(\abs{g_k-\tfrac{1}{1-(w-z_0)/(z-z_0)}}< \tfrac{\epsilon}{M}\) for all \(k> N\) and \(z\in \gamma_{R'}\text{.}\) It follows that
\begin{align*} \abs{f_k(z)- \frac{f(z)}{w-z_0}} \amp = \abs{f_k(z)- \frac{f(z)}{z-z_0}\cdot \frac{1}{1-(w-z_0)/(z-z_0)}}\\ \amp =\abs{\frac{f(z)}{z-z_0}}\abs{g_k(z)-\frac{1}{1-(w-z_0)/(z-z_0)}}\\ \amp < M\cdot \epsilon/M\\ \amp = \epsilon \end{align*}
for all \(k> N\) and \(z\in \gamma_{R'}\text{.}\) This proves our uniform convergence claim, and thus justifies our bringing the integral into the infinite sum.
Lastly, we observe that the derivative formula for the power series coefficients \(a_n\) follows from Corollary 1.17.15. As for the stated integral formula for \(a_n\text{,}\) by the deformation of paths prinicple, we have for any \(r\in (0,R)\)
\begin{equation*} a_n=\frac{1}{2\pi i}\int_{\gamma_{R'}}\frac{f(z)}{z-z_0}\, dz=\frac{1}{2\pi i}\int_{\gamma}\frac{f(z)}{z-z_0}\, dz\text{,} \end{equation*}
where \(\gamma(t)=z_0+re^{it}\text{,}\) \(t\in [0,2\pi]\text{.}\) This completes the proof.
Visual idea behind proof of theorem
Figure 1.18.2. Visual idea behind proof of Theorem 1.18.1
As a consequence of Theorem 1.18.1, we can now add \(\exp\text{,}\) \(\cos\text{,}\) and \(\sin\) to our list of familiar functions with power series representations centered at \(z_0=0\text{.}\) Observe that since these three functions are holomorphic on all of \(\C\text{,}\) and since \(B_R(0)\subseteq \C\) for all \(R> 0\text{,}\) the radius of convergence of their \(z_0=0\)-centered power series is \(\infty\text{.}\) The coefficients of the power series for \(f(z)=e^z\) are easily computed using the fact that
\begin{equation*} f^{(n)}(z)=e^z \end{equation*}
for all integers \(n\geq 0\text{.}\) The power series for \(\cos\) and \(\sin\) can be derived from that of \(\exp\) using
\begin{align*} \cos z \amp = \frac{1}{2}(e^{iz}+e^{-iz})\\ \sin z \amp = \frac{1}{2i}(e^{-iz}-e^{-iz})\text{.} \end{align*}

Remark 1.18.4. Radius of convergence, revealed.

When a power series is known to be the power series expansion of a familiar function, we can often use Theorem 1.18.1 to give a lower bound on the radius of convergence of the power series. Furthermore, further investigation of the function can sometimes reveal exactly what the radius of convergence is. For example consider the power series expansion
\begin{equation} \frac{1}{1+z^2}=\sum_{n=0}^\infty (-1)^{n}z^{2n}\text{.}\tag{1.74} \end{equation}
Since \(f(z)=\frac{1}{1+z^2}\) is holomorphic on \(B_1(0)\text{,}\) we know from Theorem 1.18.1 that the series above converges on \(B_1(0)\text{.}\) Is it possible that the series converges on \(B_S(0)\) for some \(S> 1\text{?}\) No, as an investigation of the behavior of \(f\) near \(z=i\) reveals. Indeed, if the series converged on \(B_S(0)\) for some \(S> 1\text{,}\) then it would define a continuous function \(g\) on \(B_S(0)\) that satisfies \(g(z)=f(z)\) for all \(z\in B_1(0)\text{.}\) But then we would have
\begin{equation*} g(i)=\lim\limits_{z\to i}g(z)=\lim\limits_{n \to\infty }f((1-1/n)i)\text{,} \end{equation*}
a contradiction since \(\lim\limits_{z\to i}f(z)=\infty\text{.}\) Thus we see that it is the poor behavior of \(f\) at \(i\) (and \(-i\)) that is responsible for the radius of convergence being \(1\text{.}\) Observe that this simple obstruction at \(i\) remains undetected when investigating the power series representation (1.74) solely within the realm of the reals.

Remark 1.18.5. Radius of convergence, tempered.

Just to temper your excitement about the method outlined above, consider the function \(\Log\text{,}\) with domain \(U=\C-R_{\pi}\text{,}\) and its power series expansion around \(z_0=-1+i\text{.}\) The usual manipulations show that we have
\begin{equation} \frac{1}{z}=\frac{1}{z_0}\frac{1}{1+(z-z_0)/z_0}=\sum_{n=0}^\infty \frac{(-1)^{n}}{z_0^{n+1}}(z-z_0)^n \text{,}\tag{1.75} \end{equation}
and hence, taking antiderivatives,
\begin{align*} \Log(z)\amp =\Log(z_0)+\sum_{n=1}^\infty\frac{(-1)^{n-1}}{nz_0^{n}}(z-z_0)^{n} \\ \amp = \ln\sqrt{2}+\frac{3\pi i}{4}+\sum_{n=1}^\infty\frac{(-1)^{n-1}}{nz_0^{n}}(z-z_0)^{n}\text{.} \end{align*}
What is the radius of convergence of this power series? Since \(B_1(z_0)=B_1(-1+i)\subseteq U\text{,}\) we know that the radius of convergence is at least \(1\text{,}\) and this seems imposed upon us by the ray \(R_\pi\) marking an end to the domain of \(\Log\text{.}\) And yet, the radius of convergence of this power series is \(\abs{z_0}=\sqrt{2}\text{,}\) as we see by observing that the series (1.75) derived from the geometric series has radius of convergence \(\sqrt{2}\text{,}\) and thus the series for \(\Log z\text{,}\) obtained by taking the formal antiderivative, also has radius of convergence \(\sqrt{2}\text{.}\)
Without saying too much, what is at issue here is the notion of analytic continuation. We content ourselves by remarking that if, instead of taking \(\Log\) and its domain \(U=\C-R_\pi\) we took instead \(\Log_0\) and its domain \(V=\C-R_0\text{,}\) then an analysis as in the previous remark would have sufficed to show that the radius of convergence is \(\sqrt{2}\text{.}\)
We end this section by collecting a list of equivalent formulations of being holomorphic. With the exception of the last statement, all of these equivalences now follow from previous results. For example, that (1) is equivalent with (2) follows from Theorem 1.17.9 and Theorem 1.18.1. That the last statement is also equivalent to being holomorphic follows from the proof of Theorem 1.18.1, as the only fact we used to show \(f\) is analytic on each \(B_R(z_0)\subseteq U\) is the fact that \(f\) is continuous and satisfies the Cauchy integral formula for all \(w\in B_R(z_0)\text{.}\)