5

Starting from the definition of the Gamma function as expressed by

$$\Gamma(z)=\int_0^\infty x^{z-1}e^{-x}\,dx\tag1$$

we can show that the derivative of $\Gamma(z)$ evaluated at $z=1/2$ is given by

$$\Gamma'(1/2)=-\sqrt{\pi} \left(\gamma+\log(4)\right)\tag2$$



Proof of $(2)$: Here, I present for completeness the approach that I took. One can skip this part without losing context.

Differentiating $(1)$ and setting $z=1/2$ reveals

$$\Gamma'(1/2)=\int_0^\infty \frac{e^{-x}}{\sqrt{x}}\log(x)\,dx\tag3$$

Next, we represent the logarithm function in $(3)$ by a Frullani integral to find that

$$\begin{align} \Gamma'(1/2)&=\int_0^\infty \frac{e^{-x}}{\sqrt{x}}\int_0^\infty \frac{e^{-y}-e^{-xy}}{y}\,dy\,dx\\\\ &=\int_0^\infty \frac1y\int_0^\infty \frac{e^{-x}e^{-y}-e^{-(y+1)x}}{\sqrt{x}}\,dx\,dy\\\\ &=\sqrt\pi\int_0^\infty \frac1y \left(e^{-y}-\frac1{\sqrt{y+1}}\right)\,dy\tag4 \end{align}$$

Integrating by parts the integral on the right-hand side of $(4)$, we obtain

$$\Gamma'(1/2)=-\sqrt\pi(\gamma+\log(4))\tag5$$

as was to be shown.



QUESTION: So, what are alternative approaches to evaluating $\Gamma'(1/2)$ if we begin with $(1)$?

Mark Viola
  • 179,405
  • Are you asking for a different approach to integrate line 3? If you start with using the integral definition for the gamma function then you have to eventually get to the integral in line 3, no? – Ty. Jun 27 '20 at 19:35
  • @Ty. If you like, then you may begin with $(3)$. Otherwise, you could augment $(1)$ into an alternative representation, differentiate, let $z=1/2$ and proceed. – Mark Viola Jun 27 '20 at 19:37
  • @Ty. I posted an alternative approach that begins with $(1)$, develops a limit definition of $\Gamma(z)$ therefrom, and then derives a series formula for its logarithmic derivative (aka $\psi(z)$. – Mark Viola Jun 30 '20 at 14:15
  • +1 there. Added a bookmark for this thread containing amazing approaches (both in question as well as answers). – Paramanand Singh Sep 23 '21 at 15:50
  • @ParamanandSingh Thank you my friend! Much appreciate your note. – Mark Viola Sep 23 '21 at 20:06
  • Would the down voter care to give a reason for the unfounded down vote? Incredulous behavior. – Mark Viola Jun 11 '22 at 15:04

3 Answers3

7

Since $\Gamma'(x)=\Gamma(x)\psi(x)$ the determination of $\Gamma'(1/2)$ immediately boils down to the determination of $\psi(1/2)$. Since $$ \sum_{n\geq 0}\left(\frac{1}{n+a}-\frac{1}{n+b}\right)=\psi(a)-\psi(b)$$ and $\psi(1)=-\gamma$ by the Weierstrass product for the $\Gamma$ function, we may just pick $a=\frac{1}{2}$, $b=1$ and compute $$ \psi(1/2)+\gamma=\sum_{n\geq 0}\left(\frac{2}{2n+1}-\frac{2}{2n+2}\right)=2\sum_{m\geq 1}\frac{(-1)^{m+1}}{m}=-2\log 2 $$ to deduce $$ \Gamma'(1/2) = \Gamma(1/2)\psi(1/2) = \sqrt{\pi}\psi(1/2) = -\sqrt{\pi}(\gamma+\log 4)$$ without invoking Frullani.


Conversely,

$$\begin{eqnarray*} \gamma=\lim_{n\to +\infty}(H_n-\log n) &=& \sum_{n\geq 1}\left(\frac{1}{n}-\log\left(1+\frac{1}{n}\right)\right)\\&\stackrel{\text{Frullani}}{=}&\sum_{n\geq 1}\int_{0}^{+\infty}e^{-nx}-\frac{e^{-nx}-e^{-(n+1)x}}{x}\,dx\\&=&\int_{0}^{+\infty}\left(\frac{1}{e^x-1}-\frac{1}{x e^x}\right)\,dx\\&\stackrel{\color{red}{\text{Devil}}}{=}&-\int_{0}^{+\infty}e^{-x}\log(x)\,dx=-\Gamma'(1)\end{eqnarray*} $$ where the marked equality is justified by this:

$$ \int_{0}^{M}\left(\frac{1}{e^x-1}-\frac{1}{x}\right)\,dx = \log(1-e^{-M})-\log M$$ $$ \int_{0}^{M}\frac{1-e^{-x}}{x}\,dx\stackrel{\text{IBP}}{=}(1-e^{-M})\log M-\int_{0}^{M}e^{-x}\log(x)\,dx. $$

At this point we have

$$ \mathcal{L}\log(x) = -\frac{\gamma+\log(s)}{s},\qquad \mathcal{L}^{-1}\frac{1}{\sqrt{x}}=\frac{1}{\sqrt{\pi s}}$$ hence by the self-adjointness of the Laplace transform

$$ \Gamma'(1/2)=\int_{0}^{+\infty}e^{-x}\log(x)\frac{dx}{\sqrt{x}}=-\frac{1}{\sqrt{\pi}}\int_{0}^{+\infty}\frac{\gamma+\log(s+1)}{(s+1)\sqrt{s}}\,ds $$

where

$$ \int_{0}^{+\infty}\frac{ds}{(s+1)\sqrt{s}}=2\int_{0}^{+\infty}\frac{ds}{s^2+1}=\pi $$ and $$ \int_{0}^{+\infty}\frac{\log(s+1)}{(s+1)\sqrt{s}}\,ds = 2\int_{0}^{+\infty}\frac{\log(1+s^2)}{1+s^2}\,ds = -4\int_{0}^{\pi/2}\log\cos\theta\,d\theta =\pi\log 4.$$

Jack D'Aurizio
  • 353,855
  • Hi Jack. And how does one start with $(1)$ and deduce the series representation of $\psi(z)$? – Mark Viola Jun 27 '20 at 23:50
  • 1
    @MarkViola: I guess the simplest way is not to start from $(1)$ but just apply $\frac{d}{dx}\log(\cdot)$ to the Weierstrass product for the $\Gamma$ function. Euler's product works equally well. – Jack D'Aurizio Jun 28 '20 at 00:02
  • 1
    Yes Jack. And that's fine so (+1). But I'm more interested in beginning with $(1)$. – Mark Viola Jun 28 '20 at 00:26
  • @MarkViola: please see the update. – Jack D'Aurizio Jun 28 '20 at 01:27
  • Hi Jack. I'm not sure how the development of the EM constant relates directly to evalauting $\Gamma'(1/2)$ from $(1)$ in the OP. – Mark Viola Jun 28 '20 at 03:00
  • @MarkViola: I hope I made it clear now. The equality of the $\color{red}{\text{Devil}}$ gives the Laplace transform of $\log(x)$, the remaining part is the explicit computation of $\Gamma'(1/2)$ by my usual trick. – Jack D'Aurizio Jun 28 '20 at 03:32
  • 1
    Jack, the old LT trick comes to the rescue. – Mark Viola Jun 28 '20 at 03:56
  • 1
    Jack, I've posted a solution that develops a standard limit definition of the Gamma function from which its logarithmic derivative is expressed in terms of the series representation of $\psi(z)$ as given in your solution. This provides a step-by-step solution to the question at hand. I hope you enjoy it. ;-) Stay safe and healthy my friend. – Mark Viola Jun 28 '20 at 05:08
  • +1 for the first approach which looks more elementary to me. – Paramanand Singh Aug 26 '20 at 15:11
  • @JackD'Aurizio Apology for taking so long to accept your solution. And I hope that you're doing well and staying safe and healthy my friend. – Mark Viola Sep 11 '20 at 21:29
4

We begin with the integral representation of the Gamma function as given by

$$\Gamma(z)=\int_0^\infty x^{z-1}e^{-x}\,dx\tag1$$

for $z>0$.

In the next section, we show that $\Gamma(z)$ as expressed by $(1)$ can be represented by the limit

$$\Gamma(z)= \lim_{n\to\infty}\frac{n^z\,n!}{z(z+1)(z+2)\cdots (z+n)}$$



Limit Definition of Gamma

Let $G_n(z)$ be the sequence of functions given by

$$G_n(z)=\int_0^n x^{z-1}\left(1-\frac{x}{n}\right)^n\,dx$$

I showed in THIS ANSWER, using only Bernoulli's Inequality, that the sequence $\left(1-\frac{x}{n}\right)^n$ monotonically increases for $x\le n$. Therefore, $\left|x^{z-1} \left(1-\frac{x}{n}\right)^n\right|\le x^{z-1}e^{-x}$ for $x\le n$. The Dominated Convergence Theorem guarantees that we can write

$$\begin{align} \lim_{n\to \infty} G_n(z)=&\lim_{n\to \infty}\int_0^n x^{z-1}\left(1-\frac{x}{n}\right)^n\,dx\\\\ &=\lim_{n\to \infty}\int_0^\infty \xi_{[0,n]}\,s^{x-1}\left(1-\frac{s}{n}\right)^n\,ds\\\\ &=\int_0^\infty \lim_{n\to \infty} \left(\xi_{[0,n]}\,\left(1-\frac{x}{n}\right)^n\right)\,x^{z-1}\,\,dx\\\\ &=\int_0^\infty x^{z-1}e^{-x}\,dx\\\\ &=\Gamma(z) \end{align}$$


ALTERNATIVE PROOF: Limit Definition of Gamma

If one is unfamiliar with the Dominated Convergence Theorem, then we can simply show that

$$\lim_{n\to \infty}\int_0^n x^{z-1}e^{-x}\left(1-e^x\left(1-\frac{x}{n}\right)^n\right)=0$$

To do this, we appeal again to the analysis in THIS ANSWER. Proceeding, we have

$$\begin{align} 1-e^x\left(1-\frac{x}{n}\right)^n &\le 1-\left(1+\frac{x}{n}\right)^n\left(1-\frac{x}{n}\right)^n\\\\ &=1-\left(1-\frac{x^2}{n^2}\right)^n\\\\ &\le 1-\left(1-\frac{x^2}{n}\right)\\\\ &=\frac{x^2}{n} \end{align}$$

where Bernoulli's Inequality was used to arrive at the last inequality. Similarly, we see that

$$\begin{align} 1-e^x\left(1-\frac{x}{n}\right)^n &\ge 1-e^xe^{-x}\\\\ &=0 \end{align}$$

Therefore, applying the squeeze theorem yields to coveted limit

$$\lim_{n\to \infty}\int_0^n x^{z-1}e^{-x}\left(1-e^x\left(1-\frac{x}{n}\right)^n\right)=0$$

which implies $\lim_{n\to \infty}G_n(z)=\Gamma(z)$.


Integrating by parts repeatedly the integral representation of $G_n(z)$ reveals

$$G_n(z)=\frac{n^z\,n!}{z(z+1)(z+2)\cdots (z+n)}$$

so that

$$\bbox[5px,border:2px solid #C0A000]{\Gamma(z)=\lim_{n\to \infty}\frac{n^z\,n!}{z(z+1)(z+2)\cdots (z+n)}}\tag2$$



Now, we use $(2)$ to find a limit representation of the derivative of $\Gamma(z)$. To facilitate analysis, we use $(2)$ to find the logarithm of $\Gamma(z)$. Proceeding we have

$$\log\left(\Gamma(z)\right)=\lim_{n\to \infty}\left(z\log(n)+\log(n!)-\sum_{k=0}^n \log(z+k)\right)\tag3$$

Differentiating $(3)$ reveals

$$\begin{align} \frac{\Gamma'(z)}{\Gamma(z)}&=\lim_{n\to\infty}\left(\log(n)-\sum_{k=0}^n \frac1{z+k}\right)\\\\ &=\lim_{n\to\infty}\left(\log(n)-\sum_{k=0}^n\frac1{k+1}-\sum_{k=0}^n \left(\frac1{z+k}-\frac1{k+1}\right)\right)\\\\ &=-\gamma-\sum_{k=0}^\infty \left(\frac1{z+k}-\frac1{k+1}\right)\tag4 \end{align}$$

Setting $z=1/2$ in $(4)$ and using $\Gamma(1/2)=\sqrt \pi$ yields

$$\begin{align} \Gamma'(1/2)&=\sqrt{\pi}\left(-\gamma-\sum_{k=0}^\infty \left(\frac{1}{k+1/2}-\frac1{k+1}\right)\right)\\\\ &=\sqrt{\pi}\left(-\gamma-2\sum_{k=0}^\infty \left(\frac{1}{2k+1}-\frac1{2k+2}\right)\right)\\\\ &=\sqrt{\pi}\left(-\gamma-2\sum_{k=1}^\infty \left(\frac{1}{2k-1}-\frac1{2k}\right)\right)\\\\ &=\sqrt{\pi}\left(-\gamma-2\sum_{k=1}^\infty \frac{(-1)^{k-1}}{k}\right)\\\\ &=-\sqrt\pi\left(\gamma+\log(4)\right) \end{align}$$

as was to be shown!

Mark Viola
  • 179,405
-1

I fear you might not like this "non rigorous" methode, so just to provoke you. We begin with the fact that the gamma function is a product: $$\Gamma'(1/2)=\frac{\prod_{n=1}^{-3/2+h}(1+n)-\prod_{n=1}^{-3/2}(1+n)}{h}$$

$$\prod_{n=1}^{-3/2+h}(1+n)=\prod_{n=1}^{-3/2}(1+n)\prod_{n=-1/2}^{-3/2+h}(1+n)$$ $$\prod_{n=1}^{-3/2}(1+n)=(-1/2)!=(\pi)^{1/2}$$ $$=\prod_{n=-1/2}^{-3/2+h}(1+n)=\prod_{n=1}^{h}(-1/2+n)=\prod_{n=1}^{h}(1/2)\prod_{n=1}^{h}(2n-1)$$ $$\prod_{n=1}^{h}(1/2)=(1/2)^h=1-h\ln(2)$$ $$\prod_{n=1}^{h}(2n-1)=\frac{\prod_{n=1}^{2h}(n)}{\prod_{n=1}^{h}(2n)}$$ We know that $\prod_{n=1}^{h}(n)=1-h\gamma$*

$$\frac{\prod_{n=1}^{2h}(n)}{\prod_{n=1}^{h}(2n)}=(1-h\ln(2))\frac{1-2h\gamma}{1-h\gamma}$$

All together gives: $\prod_{n=1}^{-3/2+h}(1+n)=\pi^{1/2}(1-h\ln(2))^2(1-2h\gamma)(1+h\gamma)=$ $$\pi^{1/2}\big(1-h\big(\gamma+2\ln(2)\big)+O(h^2)$$ Clearly the order h is the derivative

*the fact that it's the Euler–Mascheroni constant is directly visible from either it's polynomal alternatively, it comes instinctively because $\prod_{n=1}^{h}(1+n)$ as h goes to 0, we have it's polynomal form of n^s, because it's the first derivative we need it's order $n^1$. Thuse we have the $\sum_s \sum_n n^s/s!$, times the last (or first, depends on your view) refined stirling number, which is $(-1)^{s+1}(s-1)!$. The order n in $\sum_n n^s=s\zeta(1-s)$, therefor: The regularised sum: $$\prod_{n=1}^{h}(1+n)=1+h\sum_{s=1}^{s\infty} (-1)^{s+1}\zeta(1-s)=1+h(1-\gamma)$$

Gerben
  • 593
  • How on earth are you defining the products with non-integer limits? – Mark Viola May 19 '21 at 23:42
  • And how are you defining a summation with non-integer limits? – Mark Viola May 20 '21 at 00:49
  • The Gamma function, $\Gamma(z)$ and the Harmonic numbers for complex values, $H(z)$, are well-defined extensions of the factorial $n!$ and the Harmonic numbers $H_n$. Inasmuch as you have not given a well-defined definition of the product or sum with non-integer limits, I do not find this post useful. That said, others might find it useful if they understand the development. – Mark Viola May 20 '21 at 02:41