4
$\begingroup$

I stumbled upon these 4 limit/integral identities involving Euler's constant aka gamma (~0.5772). They appear to be valid based on inspection but I have no idea how to prove them. In addition, I have no idea how these identities connect to the commonly known identities for gamma, or even how any of them follow from connect to each other. Any insight would be much appreciated!

$$ \gamma = \lim_{x \to \infty } -x+ \frac{1}{\int_{0}^{1}\frac{t^{x-t}}{\Gamma(t)}dt} $$ $$ \gamma = \lim_{x \to \infty } -x+2+ \frac{1}{\int_{1}^{k}\frac{\Gamma(t)}{t^{x-t}}dt} $$ $$ -\gamma = \lim_{x \to \infty } -x-2+ \frac{1}{\int_{0}^{1}\frac{\Gamma(t)}{t^{-x-t}}dt} $$ $$ -\gamma = \lim_{x \to \infty } -x+ \frac{1}{\int_{1}^{k}\frac{t^{-x-t}}{\Gamma(t)}dt}. $$

$$k\in \mathbb{R}, k>1$$

$\endgroup$

3 Answers 3

5
$\begingroup$

$\newcommand\ga\gamma\newcommand\Ga\Gamma$Concerning the first question, we have to show that $$g(x):=-x+1/I(x)\to\ga \tag{1}\label{1}$$ (as $x\to\infty$), where $$I(x):=I_1(h,x)+I_2(h,x),$$ $$I_1(h,x):=\int_0^{1-h} dt\,\frac{t^{x-1}}{t^{t-1}\Ga(t)},\quad I_2(h,x):=\int_{1-h}^1 dt\,\frac{t^{x-1}}{t^{t-1}\Ga(t)},$$ $0<h<1$. Since $t^{t-1}\Ga(t)\ge c$ for some $c>0$ and all $t\in(0,1)$, we have $$I_1(h,x)=O\big(\int_0^{1-h} dt\,t^{x-1}\Big)=O((1-h)^x).$$

To estimate $I_2(h,x)$, note that $\frac1{t^{t-1}\Ga(t)}=1-(\ga+o(1))(1-t)$ as $t\uparrow1$. So, if $h\downarrow0$ so that $(1-h)^x=o(x^{-2})$ (say, if $h=1/\sqrt x$ for $x>1$), then $$I_2(h,x)=\int_{1-h}^1 dt\,t^{x-1}(1-(\ga+o(1))(1-t)) =\frac1x\,\Big(1-\frac{\ga+o(1)}x\Big)$$ and hence $$I(x)=\frac1x\,\Big(1-\frac{\ga+o(1)}x\Big)$$ and $$\frac1{I(x)}=x\,\Big(1+\frac{\ga+o(1)}x\Big)=x+\ga+o(1),$$ so that \eqref{1} follows.

The other formulas should follow similarly. (There should be only one question in one post.)

$\endgroup$
2
  • $\begingroup$ Thank you. Can the o(1) term be specified further? $\endgroup$
    – Mitch
    Commented Jan 2 at 3:31
  • 2
    $\begingroup$ @Mitch : Of course. This can be done by using more terms of the asymptotic expansion of $\frac1{t^{t-1}\Gamma(t)}$ and then following the lines of the proof. For instance, taking one more term of the expansion, we have $\frac1{t^{t-1}\Gamma(t)}=1-\gamma (1-t)+\left(-1+\frac{\gamma ^2}{2}-\frac{\pi^2}{12}\right) (1-t)^2+O\left((1-t)^3\right)$. $\endgroup$ Commented Jan 2 at 3:40
8
$\begingroup$

For a sufficiently regular function (such as $\Gamma(t+1)$ or its inverse),

$$\int_0^1 t^xf(t)\,dt=\dfrac{f(1)}{x+1}-\dfrac{f'(1)}{(x+1)(x+2)}+\cdots\;,$$ so your formulas follow from $\Gamma'(1)=-\gamma$

$\endgroup$
2
  • $\begingroup$ Thank you! Is there a comparable formula to the one above for $\int_{0}^{1} t^{-x}f(t)dt$ ? $\endgroup$
    – Mitch
    Commented Jan 5 at 19:55
  • 1
    $\begingroup$ When $x\to+\infty$ this integral diverges at $0$ except if $f(t)$ is very special, such as $e^{-1/t^2}$. $\endgroup$ Commented Jan 5 at 20:12
2
$\begingroup$

Let's look at the first identity, $$\gamma = \lim_{x \to \infty }\left( -x+ \frac{1}{\int_{0}^{1}\frac{t^{x-t}}{\Gamma(t)}dt}\right).$$ The integrand $t^{x-t}/\Gamma(t)$ is sharply peaked at $t=1$ for large $x$. I expand $$\frac{t^{-t}}{\Gamma(t)}=1+(t-1)(\gamma-1)+{\cal O}(t-1)^2,$$ and evaluate $$ \lim_{x \to \infty }\left( -x+ \frac{1}{\int_{0}^{1}\frac{t^{x-t}}{\Gamma(t)}dt}\right)=\lim_{x\to\infty}\left(-x+\frac{1}{\int_0^1 t^x[1+(t-1)(\gamma-1)]\,dt}\right)$$ $$\qquad=\lim_{x\to\infty}\frac{\gamma x+2}{x-\gamma +3}=\gamma.$$

$\endgroup$
1
  • 2
    $\begingroup$ (i) Why can you truncate the asymptotic expansion by removing the $O(\cdot)$ term? (ii) How exactly do you use the qualitative description "sharply peaked" to use the truncated asymptotic expansion on the entire interval $(0,1)$, rather than just near $1$? $\endgroup$ Commented Jan 1 at 22:18

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .