2

This is a problem from a previous complex analysis qualifying exam that I'm working through to study for my own upcoming qual. I thought this one would be fairly straightforward, but since it's a qual question... of course it isn't. Below is the question and what I've tried.

Problem:

Consider the function $$ f(z) = \frac{z}{e^z-1}. $$

(a) Find the first four terms of the Taylor expansion about $z=0$ and the radius of convergence.
(b) Find all coefficients expressed through recurrent relation (this can count as first part of a solution of part (a)).

What I've considered:

First, I believe that a Taylor series about zero for this function would have a radius of convergence up to the first non-analytic point (aside from zero). This would occur where $e^z = 1$, thus at $z=2\pi i$.

Now, I've tried starting with the Taylor series definition $$ f(a) = \sum_{k=0}^\infty \frac{f^{(k)}(a)}{k!}(z-a)^k, $$

For $f(0)$ have that $f(0) = \frac{0}{0}$, so I instead take the limit $$ f(0) = \lim_{z\to 0} \frac{z}{e^z-1} = \lim_{z\to 0} \frac{1}{e^z} = 1. $$

My first coefficient in a Taylor series $f(a) = \sum_{k=0}^\infty a_kz^k$ is thus $a_0 = 1$. Beyond this, I run into trouble. The derivatives of $f(z)$ all seem to evaluate to infinity when $z=0$. So I tried something else.

Consider $f(z) = -z\cdot \frac{1}{1-Q(z)}$, where $Q(z) = 1 + \frac{z^2}{2!} + \frac{z^3}{3!} + \frac{z^4}{4!} + \dots $. Then using the geometric series, for $|Q(z)|<1$, we have $$ f(z) = -z \sum_{k=0}^\infty Q(z)^k. $$

The thought here was to work out the first four (or more if needed?) terms algebraically and then see if I can find a relationship such that I can specify all coefficients. However, the first term of $1$ in $Q(z)$ seems to be problematic, in that I seem to end up without a finite value for each term... every expanded term has a $z$ term, for example.

Finally, I considered the fact that $e^z-1$ seems to remove said problematic $1$. One can also cancel the $z$ in the numerator, such that we'd have $$ f(z) = \frac{z}{z + \frac{z^2}{2!} + \frac{z^3}{3!} + \frac{z^4}{4!} + \dots} = \frac{1}{1 + \frac{z}{2!} + \frac{z^2}{3!} + \frac{z^3}{4!} + \dots}, $$

but I can't see anywhere useful to go from here. If someone can please walk me through how to do this problem, I would very much appreciate it. I've been fighting with it for a while.

jjagmath
  • 18,214
Serafina
  • 470
  • 1
    Long division. See https://math.stackexchange.com/a/342389/442 or https://math.stackexchange.com/a/167172/442 – GEdgar Jul 17 '22 at 00:30
  • I suspect that the coefficient of $z^n$ in the Taylor series is $-\zeta(1 - n) / (n-1)!$. Using Wolfram Alpha, I checked that this works for $n = 1, 2,$ and $3$. (It's undefined for $n = 0$.)

    To derive this formula, expand $$-z\sum_{k=1}^\infty Q(z)^k = -z\sum_{k=1}^\infty e^{zk} = -z\sum_{k = 1}^\infty \left(1 + kz + \frac{k^2}{2!}z^2 + \cdots\right),$$ and group by coefficient of $z^n$.

    – Joseph Camacho Jul 17 '22 at 00:32
  • @GEdgar Thanks! This was actually really useful. I don't think I could have come up with a recursion formula for all coefficients this way, but it certainly works for getting the first several coefficients... which would have been at least partial credit if this problem showed up and I didn't remember Bernoulli numbers. – Serafina Jul 19 '22 at 21:45

3 Answers3

3

You can find this question in page 97 of stein's fourier book (the 4th problem of the third chapter ) , I give you that question and my solution as follows:

4.In this problem, we find the formula for the sum of the series $$\sum\limits_{0}^{\infty}\dfrac{1}{n^k}$$ where k is any even integer.These sums are expressed in terms of the Bernoulli numbers;the related Bernoulli polynomials are discussed in the next problem.

Define the Bernoulli numbers $B_n$ by the formula $$ \dfrac{z}{e^z-1}=\sum\limits_{n=0}^{\infty}\dfrac{B_n}{n!}z^n, $$ $(a)$ Show that $B_0=1,B_1=-\dfrac{1}{2},B_2=\dfrac{1}{6},B_3=0,B_4=-\dfrac{1}{30}$,and $ B_5=0$.

$(b)$ Show that for $n \geq 1$ we have $$ B_n=\dfrac{1}{n+1}\sum\limits_{k=0}^{n-1}\binom{n+1}{k}B_k. $$

My solution is as follows:

$(a)$ $1=\dfrac{z}{e^z-1}\dfrac{e^z-1}{z}=(\sum\limits_{n=0}^{\infty}\dfrac{B_n}{n!}z^n)(\sum\limits_{n=0}^{\infty}\dfrac{z^n}{(n+1)!})=\sum\limits_{n=0}^{\infty}(\sum\limits_{k=0}^{n}\dfrac{B_k}{k!(n+1-k)!})z^n$

we can get $\sum\limits_{k=0}^{n}\dfrac{B_k}{k!(n+1-k)!}=0$,and it's easy to see $B_0=1$,we could use the formula on the left to get our conclusion.

$(b)$ We know $\sum\limits_{k=0}^{n}\dfrac{B_k}{k!(n+1-k)!}=0$ from (a),and we could get $ B_n=\dfrac{1}{n+1}\sum\limits_{k=0}^{n-1}\binom{n+1}{k}B_k. $

qiujie qiao
  • 121
  • 3
1

Part a and b are both immediately solved using Dividing an infinite power series by another infinite power series The above method shows that in general, if you know the power series for $f(z)$ and $g(z)$ and you know that $g(0) \neq 0$, then the power series coefficients of $h(z) = f(z)/g(z)$ are immediately obtained as a recursive formula in terms of the coefficients of $f$ and $g$.

Mason
  • 10,415
0

The coefficients are called the Bernoulli numbers. My guess is that people who wrote the question expect that you are familiar with them.

jjagmath
  • 18,214
  • I now see from another answer and from doing the long division approach that the coefficients I'm actually looking for here are the Bernoulli numbers divided by $n!$. Thanks for your response! – Serafina Jul 19 '22 at 21:43