This example comes from the book by Elias Wegert: Visual Complex Functions. Consider the function $f(z):=z^{1/2}$ for $z \in \mathbb{C}$ with $|z - 1| < 1.$ For these $z$, the function can be represented by a convergent power series

$$f(z) = \sum_{k=0}^{\infty} {1/2 \choose k} (z-1)^{k},$$

which means that $z^{1/2}$ is analytic in $D(1,1).$ Now, Theorem 3.3.1 in that book says that, given

$$f(z) = \sum_{k=0}^{\infty} a_k(z-z_0)^k, \quad \forall z \in D(z_0,r),$$

you can "rearrange" the latter series into

$$f(z) = \sum_{k=0}^{\infty} b_k (z-z_1)^k, \quad \forall z \in D(z_1,r_1),$$

where the $b_{k}$'s are given by $\sum_{n=k}^{\infty} {n \choose k} a_n (z_1-z_0)^{n-k},$ $k = 0, 1, \ldots, $ provided only that $|z_1 - z_0| < r$ and $r_1 := r - |z_1 - z_0|.$ Noting that $\rm{e}^{i \pi/4} \in D(1,1)$ we can try to apply the Theorem to the series defining $z^{1/2}$ with $z_1 = \rm{e}^{i \pi/4}$. This should give

$$f(z) = \sum_{k=0}^{\infty} \left(\sum_{n=k}^{\infty} {n \choose k} {1/2 \choose n} (\rm{e}^{i \pi/4} - 1)^{n-k} \right) (z-z_1)^k.$$

This in turn should be equal to

$$f(z) = \rm{e}^{i \pi/8} \sum_{k=0}^{\infty}\rm{e}^{-i k\pi/4} {1/2 \choose k} (z-z_1)^k,$$

the point being here that while the original series diverges at, say, $z = i,$ the last one is already convergent there (...). The only thing that completely eludes me is why (as it seems)

$$\sum_{n=k}^{\infty} {n \choose k} {1/2 \choose n} (\rm{e}^{i \pi/4} - 1)^{n-k} = \rm{e}^{i \pi/8}\rm{e}^{-i k\pi/4} {1/2 \choose k},$$

for $k = 0, 1 \ldots.$ I thought about induction on $k$ but was only able to do the base step $k=0.$ Could anybody help me? Thanks.


Remember that the coefficients of the Taylor series of $f$ about $z_1$ are $\dfrac{f^{(k)}(z_1)}{k!}$. Writing $z^{1/2}$ for the square root, we have

$$z_1^{1/2} = \sum_{n=0}^\infty \binom{1/2}{n}(z_1-1)^n,$$

which is the $k= 0$ case. Now differentiate the identity with respect to $z_1$,

$$\frac{1}{2}z_1^{1/2}z_1^{-1} = \sum_{n=1}^\infty \binom{n}{1}\binom{1/2}{n}(z_1-1)^{n-1}$$

is the case $k = 1$. Generally,

$$\frac{d}{dz_1} z_1^{1/2-k}\binom{1/2}{k} = (k+1)\cdot z_1^{1/2-(k+1)}\binom{1/2}{k+1},$$


$$\begin{align} \frac{d}{dz_1}\sum_{n=k}^\infty \binom{n}{k}\binom{1/2}{n}(z_1-1)^{n-k} &= \sum_{n=k+1}^\infty (n-k)\binom{n}{k}\binom{1/2}{n}(z_1-1)^{n-(k+1)}\\ &= \sum_{n=k+1}^\infty (k+1)\binom{n}{k+1}\binom{1/2}{n}(z_1-1)^{n-(k+1)}. \end{align}$$

Divide by $k+1$ to obtain the $k+1$ case from the $k$ case.

Nicer, in my opinion, is a direct exploitation of the series centered at $1$: Near $z_1$, where $\lvert z_1-1\rvert < 1$, we have

$$\begin{align} \sqrt{z} &= \sqrt{z_1}\cdot \sqrt{\frac{z}{z_1}}\\ &= \sqrt{z_1}\sqrt{1 + \left(\frac{z}{z_1}-1\right)}\\ &= \sqrt{z_1}\sum_{k=0}^\infty \binom{1/2}{k}\left(\frac{z}{z_1}-1\right)^k\\ &= \sqrt{z_1}\sum_{k=0}^\infty z_1^{-k}\binom{1/2}{k}(z-z_1)^k. \end{align}$$

  • $\begingroup$ Well, that helped a lot, thank You. If I am right, you mean it like this: ok, we know that (under appropriate conditions) $f(z) = \sum_{n=0}^{\infty} b_n (z - z_1)^n,$ where the numbers $b_n$ are to be determined. Now, on one hand, $f^{(k)}(z_1) = k! \cdot b_k.$ On the other hand $f^{(k)}(z_1) = 1/2 \cdot (1/2-1) \cdot \ldots \cdot (1/2-k+1) z_1^{1/2-k},$ frome where $b_k = {1/2 \choose k} z_1^{1/2-k}.$ The other approach with $z = z_1 \cdot \frac{z}{z_1}$ seems to be, indeed, even much nicer (and shorter of course). $\endgroup$ – Jorge.Squared Jan 27 '14 at 16:29
  • $\begingroup$ Yes, that's the idea. And from writing $(z-1)^n = ((z-z_1)+(z_1-1))^n$, we have got the other representation $$b_k = \sum_{n=k}^\infty \binom{n}{k}\binom{1/2}{n}(z_1-1)^{n-k}.$$ Since the coefficients must be the same, the identity follows. $\endgroup$ – Daniel Fischer Jan 27 '14 at 16:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.