8

Using the Weierstrass definition for $\Gamma(x)$ and $\Gamma\Big(x + \frac12\Big)$, how can I prove the duplication formula? This is problem $10.7.3$ in the book Irresistible Integrals, by Boros and Moll.

Any help is highly appreciated.

Lucian
  • 48,334
  • 2
  • 83
  • 154
TaurusLagn
  • 81
  • 1
  • 1
  • 2

2 Answers2

7

The duplication formula can be written as

$$\frac{\Gamma(x)\Gamma(x+\frac1{2})}{\Gamma(2x)}= \frac{\Gamma(\frac1{2})}{2^{2x-1}}= \frac{\sqrt{\pi}}{2^{2x-1}}.$$

We want to derive this formula using the Weierstrass definition for the gamma function,

$$\frac1{\Gamma(x)}=xe^{\gamma x}\prod_{k=1}^{\infty}\left(1+\frac{x}{k}\right)e^{-x/k}.$$

We have

$$\frac{\Gamma(x)\Gamma(x+\frac1{2})}{\Gamma(2x)}=\frac{2xe^{2\gamma x}}{xe^{\gamma x}(x+\frac1{2})e^{\gamma x}e^{\gamma/2}}\frac{\prod_{k=1}^{\infty}\left(1+\frac{2x}{k}\right)e^{-2x/k}}{\prod_{k=1}^{\infty}\left(1+\frac{x}{k}\right)e^{-x/k}\prod_{k=1}^{\infty}\left(1+\frac{x}{k}+\frac{1}{2k}\right)e^{-x/k}e^{-1/2k}}\\ =\frac{1}{e^{\gamma/2}}\lim_{n \rightarrow \infty}\frac{2x\prod_{k=1}^{2n}\left(1+\frac{2x}{k}\right)}{x(x+\frac1{2})\prod_{k=1}^{n}\left(1+\frac{x}{k}\right)\prod_{k=1}^{n}\left(1+\frac{x}{k}+\frac{1}{2k}\right)}\frac{\prod_{k=1}^{2n}e^{-2x/k}}{(\prod_{k=1}^{n}e^{-x/k})^2\prod_{k=1}^{n}e^{-1/2k}}\\ =\frac{1}{e^{\gamma/2}}\lim_{n \rightarrow \infty}P_n(x)Q_n(x).$$

First simplify $P_n(x)$ as follows:

$$P_n(x)=\frac{2x\prod_{k=1}^{2n}\left(1+\frac{2x}{k}\right)}{x(x+\frac1{2})\prod_{k=1}^{n}\left(1+\frac{x}{k}\right)\prod_{k=1}^{n}\left(1+\frac{x}{k}+\frac{1}{2k}\right)}\\=\frac{(n!)^2}{(2n)!\left(x+n+\frac1{2}\right)}\frac{\prod_{k=0}^{n}\left(2x+2k\right)\prod_{k=0}^{n-1}\left(2x+2k+1\right)}{\prod_{k=0}^{n}\left(x+k\right)\prod_{k=0}^{n-1}\left(x+k+\frac1{2}\right)}\\=\frac{(n!)^22^{2n+1}}{(2n)!\left(x+n+\frac1{2}\right)}$$

Next consider $Q_n(x)$:

$$Q_n(x)=\frac{\prod_{k=1}^{2n}e^{-2x/k}}{(\prod_{k=1}^{n}e^{-x/k})^2\prod_{k=1}^{n}e^{-1/2k}}\\=\frac{n^{1/2}}{2^{2x}}\frac{(2n)^{2x}\prod_{k=1}^{2n}e^{-2x/k}}{(n^x\prod_{k=1}^{n}e^{-x/k})^2n^{1/2}\prod_{k=1}^{n}e^{-1/2k}}$$

Reassembling we get

$$\frac{\Gamma(x)\Gamma(x+\frac1{2})}{\Gamma(2x)}=\frac{1}{e^{\gamma/2}}\lim_{n \rightarrow \infty}\frac{(n!)^22^{2n+1}}{(2n)!\left(x+n+\frac1{2}\right)}\frac{n^{1/2}}{2^{2x}}\frac{(2n)^{2x}\prod_{k=1}^{2n}e^{-2x/k}}{(n^x\prod_{k=1}^{n}e^{-x/k})^2n^{1/2}\prod_{k=1}^{n}e^{-1/2k}}\\=\frac{1}{2^{2x-1}}\lim_{n \rightarrow \infty}\frac{n}{\left(x+n+\frac1{2}\right)}\frac{(n!)^22^{2n}}{(2n)!n^{1/2}}\frac{(2n)^{2x}\prod_{k=1}^{2n}e^{-2x/k}}{e^{\gamma/2}(n^x\prod_{k=1}^{n}e^{-x/k})^2n^{1/2}\prod_{k=1}^{n}e^{-1/2k}}.$$

We can evaluate the limit in three parts.

First,

$$\lim_{n \rightarrow \infty}\frac{n}{\left(x+n+\frac1{2}\right)}=1.$$

Second, using a well-known identity for the Euler-Mascheroni constant,

$$\lim_{n \rightarrow \infty}\frac{(2n)^{2x}\prod_{k=1}^{2n}e^{-2x/k}}{e^{\gamma/2}(n^x\prod_{k=1}^{n}e^{-x/k})^2n^{1/2}\prod_{k=1}^{n}e^{-1/2k}}=\frac{e^{-2\gamma x}}{(e^{-\gamma x})^2e^{-\gamma /2}e^{\gamma /2}}=1.$$

Third using Stirlings's asymptotic formula $n! \sim \sqrt{2\pi}n^{n+1/2}e^{-n},$

$$\lim_{n \rightarrow \infty}\frac{(n!)^22^{2n}}{(2n)!n^{1/2}}=\sqrt{\pi},$$

and finally we get

$$\frac{\Gamma(x)\Gamma(x+\frac1{2})}{\Gamma(2x)}=\frac{\sqrt{\pi}}{2^{2x-1}}.$$

RRL
  • 90,707
  • Why is it that $\prod_{k=1}^{\infty}\left(1+\frac{2x}{k}\right)e^{-2x/k}=\lim_{n\to\infty}\prod_{k=1}^{\bf{2n}}\left(1+\frac{2x}{k}\right)e^{-2x/k}$. Why does it matter that $k=1,2,\ldots, n$ rather than $k=1,2,\ldots, 2n$? –  Feb 21 '18 at 20:41
  • 1
    A convergent infinite product is the limit of a sequence of partial products $P_n$. Any subsequence such as $P_{2n}$ of a convergent sequence must converge to the same limit. So it is perfectly valid. Why does it matter? Well that is the trick. Since the limit of a ratio of convergent sequences is the ratio of the limits (assuming we don't divide by zero) it reveals some hidden structure. Perhaps you've seen how switching the order of integration unravels a difficult double integral when valid. Similarly switching an integral and an infinite sum. There is a certain art to these manipulations – RRL Feb 22 '18 at 08:39
  • I checked on wolfram and it comes out different when you replace n by 2n – EB97 Mar 29 '23 at 19:49
  • @EB97: It is well known that $\prod_{k=1}^{\infty}\left(1+\frac{x}{k}\right)e^{-x/k} = \frac{e^{-\gamma x}}{x\Gamma(x)}$ and you are telling me that $\lim_{n \to \infty} \prod_{k=1}^{n}\left(1+\frac{x}{k}\right)e^{-x/k} \neq \lim_{n \to \infty} \prod_{k=1}^{2n}\left(1+\frac{x}{k}\right)e^{-x/k}$? – RRL Mar 29 '23 at 21:30
  • Yes you're right, I checked on wolfram with the euler form and changed only the n of the product. – EB97 Mar 30 '23 at 23:02
3

For people looking for an easier way to prove this without using the Weierstrass definition, an alternative is to use Bohr-Mollerup's theorem to the function $F:[0,\infty)\rightarrow [0,\infty)$ $$F(x)=\frac{1}{\sqrt{\pi}}2^{x-1}\Gamma\left(\frac{x}{2}\right)\Gamma\left(\frac{x+1}{2}\right)$$ Noting that this function is log-convex and satisfies $F(1)=1$ and $F(x+1)=xF(x)$ shows that $F(x)=\Gamma(x)$, therefore as analytic functions they must coincide on the complex plane outside of the poles of $\Gamma$, in particular: $\Gamma(2z)=F(2z)$ as desired.

  • I'm curious -- would this generalize to show $\Gamma(nx) = \frac{n^{nx}}{(2\pi)^{(n-1)/2} \sqrt{n}} \Gamma(x) \Gamma(x+1/n) \cdots \Gamma(x + (n-1)/n))$ for integer $n > 0$? (The main thing I would be unsure about would be whether the corresponding $F(x) = C_n n^{x} \Gamma(x/n) \cdots \Gamma ((x+n-1)/n)$ is log-convex.) – Daniel Schepler Aug 25 '22 at 19:51
  • Incidentally, for the doubling formula in particular, I discovered it by playing around with the integral $\int_{-1}^1 (1-t^2)^x dt$. If you use symmetry and make the substitution $t := \sqrt{u}$ you get $B(x+1, \frac{1}{2})$; whereas if you use the substitution $t := 2u-1$ you get $2^{2x+1} B(x+1, x+1)$; and expanding the beta functions on each sides gives something equivalent to the doubling formula. However, that obviously won't generalize as easily to a gamma tripling formula, for example. – Daniel Schepler Aug 25 '22 at 19:55
  • I'm not sure where this formula comes from, but the given F(x) seems log-convex, just note that each component of the product is log-convex and the sum of convex functions is convex. – David Melo Aug 27 '22 at 00:54
  • @DanielSchepler The duplication formula is a special case of the multiplication theorem. So the answer is yes the formula is correct. –  Mar 09 '23 at 16:19