1

The Laurent series about $0$ of $1/{z(e^z-1)}$ is

$$\frac{1}{z(e^z-1)}=\frac{1}{z^2}-\frac{1}{2z}+\sum_{n=2}^\infty\frac{B_n}{n!}z^{n-2}$$

When we integrate this we have

$$\int\frac{1}{z(e^z-1)}dz=-\frac{1}{z}-\frac{\ln{|z|}}{2}+\sum_{n=2}^\infty\frac{B_n}{n!(n-1)}z^{n-1}+C$$

Something interesting happens when we subtract the expansion from the definite integral form. It equals a constant.

$$C_0=\int_\infty^x\frac{1}{z(e^z-1)}dz-\left(-\frac{1}{x}-\frac{\ln{|x|}}{2}+\sum_{n=2}^\infty\frac{B_n}{n!(n-1)}x^{n-1}\right)$$

Unfortunately the series is only defined on $0\lt x\lt 2\pi$.

What is the value of the constant $C_0$?

I thought about seeing if I can do the expansion about some other point $a$, so here's what I did

$$\begin{align} \frac{1}{z(e^z-1)}&=\frac{1}{a(z/a)(e^{a(z/a)}-1)}\\ &=\sum_{n=0}^\infty\frac{B_n}{n!}a^{n-2}\left(\frac{z}{a}\right)^{n-2}\\ &=\sum_{n=0}^\infty\frac{B_n}{n!}a^{n-k-2}\sum_{k=0}^\infty{n-2\choose{k}}\left(z-a\right)^k\\ &=\sum_{k=0}^\infty c_k\left(z-a\right)^k \end{align}$$

Where

$$c_k=\sum_{n=0}^\infty\frac{B_n}{n!}{n-2\choose{k}}a^{n-k-2}$$

Using the Euler-Maclaurin summation formula with $f(x)=\frac{(-1)^k}{k!}\Gamma(k+1,-x)$ we get

$$c_k=\frac{\left(-1\right)^k}{a^{k+1}k!}\sum_{n=1}^\infty \Gamma\left(k+1,na\right)$$

So by integration we have

$$\int\frac{1}{z(e^z-1)}dz=\sum_{k=0}^\infty \frac{c_k}{k+1}\left(z-a\right)^{k+1}+C$$

Then

$$\int_\infty^x\frac{1}{z(e^z-1)}dz=\lim\limits_{a\to\infty}\sum_{k=0}^\infty \frac{c_k}{k+1}\left(x-a\right)^{k+1}$$

Then we can try subtracting the original series expression that is defined on $(0,2\pi)$ from this expression that is defined on $(0,\infty)$ for some $x\in (0,2\pi)$.

tyobrien
  • 3,667
  • 1
    This is an indefinite integral, so there is no unique choice for $C_0$ – Ninad Munshi Sep 02 '19 at 21:11
  • Thank you, I was in a rush and didn’t write what I meant. It is corrected. – tyobrien Sep 02 '19 at 21:23
  • That's not a Taylor series, it's a Laurent series. – zhw. Sep 02 '19 at 21:30
  • It isn't so surprising that subtracting the indefinite from the definite integral gets you a constant, that is the fundamental theorem of calculus $\int_a^x f'(x)dx - \int f'(x)dx = f(x) - f(a) - f(x) + C = C - f(a) = C$. What constant it equals is the value of the antiderivative (with + C = 0) at that $a$. Unfortunately since you've set the value to infinity, the easy solution to this problem (plug in infinity into the antiderivative) won't work because it's not in the domain of the expansion. – Ninad Munshi Sep 02 '19 at 21:30
  • Things seem a little confused here. Aren't you integrating over $z$ in the definite integral to get a function of $x$? If so, then subtracting a function of $z$ from that function of $x$ certainly won't yield a constant. – Charles Hudgins Sep 08 '19 at 05:28

1 Answers1

5

The answer is $\color{blue}{C_0=(\ln2\pi-\gamma)/2}$. We rewrite the definition as \begin{align*} C_0&=\lim_{x\to0^+}\left[\frac1x+\frac{\ln x}2-\int_x^\infty\frac{dz}{z(e^z-1)}\right] \\&=\lim_{x\to0^+}\left[\frac{\ln x}2+\int_x^\infty\left(\frac1z-\frac1{e^z-1}\right)\frac{dz}z\right] \end{align*} and use $\lim\limits_{x\to0^+}\left(\ln x+\int_x^\infty(e^{-z}/z)\,dz\right)=-\gamma$ (see e.g. here) to get $$ C_0=-\frac\gamma2-\color{blue}{\int_0^\infty\left(\frac{e^{-z}}2-\frac1z+\frac1{e^z-1}\right)\frac{dz}z}. $$ The integral looks close to Binet's log-gamma integrals, and can be found the way these are obtained.

Another approach is a link to Riemann's $\zeta$; consider, for $\Re a>0$ and $\Re s>0$, $$ I(s,a)=\int_0^\infty\left(\frac{e^{-z}}2-\frac{e^{-az}}{z}+\frac1{e^z-1}\right)z^{s-1}\,dz. $$ For $\Re s>1$, this is easily integrated termwise, giving $$ I(s,a)=\Gamma(s)\left(\frac12-\frac{a^{1-s}}{s-1}+\zeta(s)\right), $$ which then holds for $\Re s>0$ by analytic continuation.

Now, for $0<\Re s<1$, we can take $a\to0^+$, which gives $$ \int_0^\infty\left(\frac{e^{-z}}2-\frac1z+\frac1{e^z-1}\right)z^{s-1}\,dz=\Gamma(s)\left(\frac12+\zeta(s)\right) $$ (in fact, this holds for $-1<\Re s<1$, again by a.c.).

The integral we're looking for is obtained at $s\to 0$, and equals $\zeta'(0)=-\frac12\ln2\pi$.

metamorphy
  • 43,591