Let $G(z)$ be the Barnes G-function.

I want to use the infinite product representation

$$ G(z+1)=(2\pi)^{z/2}\text{exp}\left(-\frac{z(z+1)}{2}- \frac{\gamma z^{2}}{2}\right)\, \prod_{k=1}^{\infty}\left(1+\frac{z}{k}\right)^{k}\text{exp}\left(\frac{z^2}{2k}-z\right)$$

to show that the Barnes G-function satisfies the functional equation

$$G(z+1) = \Gamma(z) G(z). $$

Specifically, I want to show that $$\frac{G(z+1)}{G(z)} = \frac{e^{-\gamma z}}{z} \prod_{k=1}^{\infty} \frac{e^{\frac{z}{k}}}{1+\frac{z}{k}}$$ where the right side of the equation is the Weierstrass infinite product representation of the gamma function.


$$ \begin{align} \frac{G(z+1)}{G(z)} &= \frac{(2\pi)^{z/2}\text{exp}\left(-\frac{z(z+1)}{2}- \frac{\gamma z^{2}}{2}\right)\, \prod_{k=1}^{\infty}\left(1+\frac{z}{k}\right)^{k}\text{exp}\left(\frac{z^2}{2k}-z\right)}{(2\pi)^{(z-1)/2}\text{exp}\left(-\frac{(z-1)z}{2}- \frac{\gamma (z-1)^{2}}{2}\right)\, \prod_{k=1}^{\infty}\left(1+\frac{z-1}{k}\right)^{k}\text{exp}\left(\frac{(z-1)^2}{2k}-(z-1)\right)} \\ &= \sqrt{2 \pi} \exp \left(-z - \gamma z + \frac{\gamma}{2} \right) \prod_{k=1}^{\infty} \left(\frac{k+z}{k+z-1} \right)^{k} \exp \left( \frac{2z-1-2k}{2k} \right). \end{align}$$

But it's not clear to me what to do next.

  • $\begingroup$ Functional equation only does not fix the Barnes function. For example, $G(z)\sin2\pi z$ will satisfy the same functional equation. $\endgroup$ – Start wearing purple Oct 1 '13 at 12:32
  • $\begingroup$ I actually knew that. The question was badly worded. I just want to show that it satisfies the functional equation. But that alone seems to be very difficult to show. After doing a bunch of searching, it seems to be shown nowhere except in Barnes original papers. $\endgroup$ – Random Variable Oct 1 '13 at 15:53
  • $\begingroup$ Concerning your last question: it depends on how do you define $G(z)$ itself. I never liked this convexity thing. What is really important about $G(z)$ is not the convexity, but the set of poles/zeros which is, in a sense, minimal. Another way to introduce $G(z)$ is the integral representation. Yet another thing is that $G(z)$ is a limiting case of a much nicer and rigid function $\Gamma_q(z)$ in the case where the two periods of the latter coincide. $\endgroup$ – Start wearing purple Oct 4 '13 at 19:00
  • $\begingroup$ Are you referring to the above integral representation or some other integral representation? I derived that by using the infinite product representation. So in a sense I feel like I'm going in circles. $\endgroup$ – Random Variable Oct 4 '13 at 19:38
  • $\begingroup$ No, for me the defining integral representation of the Barnes $G$-function would be $$ G(1+z)= \left(2\pi\right)^{\frac{z}{2}} \exp \int_0^{\infty} \frac{ds}{s} \Biggl[ \frac{1-e^{-zs}}{ 4\sinh^2 \frac{s}{2}} - \frac{z}{s}+ \frac{z^2}{2} \,e^{-s} \Biggr]$$ $\endgroup$ – Start wearing purple Oct 5 '13 at 0:00

I will start from your last expression. The idea is, roughly, to write $(k+z-1)^k$ in the denominator as $(k+z-1)^{k-1}\cdot (k+z-1)$, then to replace $k-1$ by $k'$ and observe that the product telescopically simplifies. Also, to avoid mistakes, it is better to start with a finite product.

We have \begin{align} \prod_{k=1}^N\left(\frac{k+z}{k-1+z}\right)^k \exp\frac{2z-1-2k}{2k}=\frac{\prod_{k=1}^N(k+z)^k\exp\frac{2z-1-2k}{2k}}{\prod_{k=1}^N(k-1+z)^{k-1}(k-1+z)}=\\ =\frac{\prod_{k=1}^{N}(k+z)^k\exp\frac{2z-1-2k}{2k}}{\prod_{k=0}^{N-1}(k+z)^{k}\prod_{k=0}^{N-1}(k+z)}=\\ =\frac{(N+z)^N\prod_{k=1}^{N}\exp\frac{2z-1-2k}{2k}}{z\prod_{k=1}^{N-1}(k+z)}=\\ =\frac{N^N\left(1+\frac{z}{N}\right)^N\prod_{k=1}^{N}\exp\frac{2z-1-2k}{2k}}{\frac{z}{N+z}\prod_{k=1}^{N}k(1+\frac{z}{k})}=\\ =\frac{(N/e)^N}{N!}\times\left(1+\frac{z}{N}\right)^N\times\frac{N+z}{z}\times\prod_{k=1}^{N}\frac{e^{z/k}}{1+z/k}\times \exp\left\{-\frac12\sum_{k=1}^N\frac1k\right\}.\tag{1} \end{align} Now consider the asymptotics of different factors in the last expression as $N\rightarrow\infty$:

  • From Stirling's formula we get $$\frac{(N/e)^N}{N!}=\frac{1}{\sqrt{2\pi N}}+O\left(N^{-3/2}\right).$$

  • $\left(1+\frac{z}{N}\right)^N$ tends to $e^z$.

  • Using product representation of the gamma function, in the limit we can replace $\prod_{k=1}^{N}\frac{e^{z/k}}{1+z/k}$ by $ze^{\gamma z}\Gamma(z)$.

  • Harmonic series is known to behave as $$\sum_{k=1}^N\frac1k=\ln N+\gamma+O\left(N^{-1}\right)$$

Therefore, the limit of (1) gives $$\frac{1}{\sqrt{2\pi N}}\times e^z\times N\,e^{\gamma z}\Gamma(z)\times \frac{e^{-\gamma/2}}{\sqrt{N}}\sim\frac{\exp\left\{z(\gamma+1)-\frac{\gamma}{2}\right\}}{\sqrt{2\pi}}\Gamma(z).$$ This is equivalent to the functional relation you want to prove.


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.