# Computing $\int_0^\infty \frac{\log x}{\exp x} \ dx$ [duplicate]

I know that $$\int_0^\infty \frac{\log x}{\exp x} = -\gamma$$ where $\gamma$ is the Euler-Mascheroni constant, but I have no idea how to prove this.

The series definition of $\gamma$ leads me to believe that I should break the integrand into a series and interchange the summation and integration, but I can't think of a good series. The Maclaurin series of $\ln x$ isn't applicable as the domain of $x$ is not correct and I can't seem to manipulate the integrand so that such a Maclaurin series will work.

Another thing I thought of was using $x \mapsto \log u$ to get $\int\limits_{-\infty}^\infty \frac{\log \log u}{u^2} \ du$ and use some sort of contour integration, but I can't see how that would work out either.

• we easy show that $$\int_{0}^{\infty}\ln{x}e^{-x}dx=-\gamma$$ Jun 13, 2013 at 5:32
• But for the sign, isn't this precisely what the OP's asking @math110? Jun 13, 2013 at 10:43
• That was a typo, my mistake. Jun 13, 2013 at 16:29
• If you don't know how to prove it, you can believe it, but you can't know it.
– robjohn
Mar 9, 2015 at 9:06

lemma1: $$\int_{0}^{1}\dfrac{1-e^{-x}}{x}dx-\int_{1}^{\infty}\dfrac{e^{-x}}{x}dx=\gamma$$

pf:use \begin{align}\sum_{i=1}^{n}\dfrac{1}{i}&=\int_{0}^{1}\dfrac{1-t^n}{1-t}dt\\ &=\int_{0}^{n}\dfrac{1-(1-\frac{x}{n})^n}{x}dx \end{align} and $$\ln{n}=\int_{1}^{n}\dfrac{1}{x}dx$$ so $$\gamma=\lim_{n\to\infty}(\sum_{i=1}^{n}\dfrac{1}{i}-\ln{n})=\lim_{n\to\infty}\left(\int_{0}^{1}\dfrac{1-(1-x/n)^n}{x}dx-\int_{1}^{n}\dfrac{(1-x/n)^n}{x}dx\right)$$

so $$\gamma=\int_{0}^{1}\dfrac{1-e^{-x}}{x}dx-\int_{1}^{\infty}\dfrac{e^{-x}}{x}dx=\int_{0}^{1}(1-e^{-x})d(\ln{x})-\int_{1}^{\infty}e^{-x}d(\ln{x})=\cdots=-\int_{0}^{\infty}e^{-x}\ln{x}dx$$

• For those who are wondering, I chose this one as it has the most elementary solution, relying only on the definition of $\gamma$ as opposed to falling back on special functions. Jun 13, 2013 at 16:30
• How has $\int_{1}^{n}\frac{(1-x/n)^n}{x}dx$ appeared? And how was its limit for $n\to\infty$ evaluated?
– user
Mar 12, 2019 at 22:59
• The last step is unclear to me. Can someone explain what happens in the dotdotdots? Dec 11, 2020 at 23:46
• The dots represent integration by parts (boundary terms disappear) Dec 27, 2021 at 19:57
• The passage to the limit is easily justified by dominated convergence, since for $0\le x\le n$ we have $0\le 1-(1-\frac{x}{n})^n\le x$ (by Bernoulli's inequality) and $(1-\frac{x}{n})^n\le e^{-x}$ (thanks to $1-t\le e^{-t}$) Dec 27, 2021 at 20:02

Here is one way, just derived it (by the way, the integral is equal to $-\gamma$). We know from Euler that you can write the logarithm as the following limit

$$\mathrm{ln}(x)=\lim_{n \to 0}\frac{x^n-1}{n}$$

The integral is now

$$\int_0^\infty \frac{\mathrm{ln} x}{\exp x}dx=\lim_{n\to 0}\frac{1}{n}\int_{0}^\infty\frac{x^n-1}{e^x}dx$$

This is just the Euler integral of second kind, thus you get

$$\int_0^\infty \frac{\mathrm{ln} x}{\exp x}dx=\lim_{n\to 0}\frac{\Gamma(n +1)-1}{n}=-\gamma$$

• +1. By the way, if you put your $\TeX$ in double dollar signs ($\$\$$.... stuff .... \\$$), then the $\TeX$ gets centered and renders a bit better. Jun 13, 2013 at 5:41

Recalling the Mellin transform of a function $f$

$$F(s)=\int_{0}^{\infty}x^{s-1}f(x)dx \implies F'(s) = \int_{0}^{\infty} x^{s-1}\ln(x)f(x) dx.$$

So, taking $f(x)=e^{-x}$ and finding its Mellin transform

$$F(s)=\Gamma(s) \implies F'(s)=\Gamma'(s)$$

Taking the limit as $s\to 1$ yields the desired resuly

$$\lim_{s\to 1}F'(s)=-\gamma.$$

Note: You can use the identity to find the limit

$$\psi(x)=\frac{d}{dx}\ln\Gamma(x)=\frac{\Gamma'(x)}{\Gamma(x)}\implies \Gamma'(x)=\Gamma(x)\psi(x),$$

where $\psi(x)$ is the digamma function and $\psi(1)=-\gamma$.

For $x\le n$, $\left(1-\frac{x}{n}\right)^n$ increases with $n$. By Bernoulli's Inequality, \begin{align} \frac{\left(1-\frac{x}{n+1}\right)^{n+1}}{\left(1-\frac{x}{n}\right)^n} &=\left(\frac{(n+1-x)\,n}{(n+1)(n-x)}\right)^{n+1}\left(1-\frac{x}{n}\right)\\ &=\left(1+\frac{x}{(n+1)(n-x)}\right)^{n+1}\left(1-\frac{x}{n}\right)\\[6pt] &\ge\left(1+\frac{x}{n-x}\right)\left(1-\frac{x}{n}\right)\\[9pt] &=\frac{n}{n-x}\frac{n-x}{n}\\[12pt] &=1 \end{align} Therefore, by Monotone Convergence, \begin{align} \int_0^\infty\log(x)e^{-x}\mathrm{d}x &=\lim_{n\to\infty}\int_0^n\log(x)\left(1-\frac{x}{n}\right)^n\mathrm{d}x\\ &=\lim_{n\to\infty}n\int_0^1(\log(x)+\log(n))\,(1-x)^n\,\mathrm{d}x\\ &=\lim_{n\to\infty}n\int_0^1(\log(1-x)+\log(n))\,x^n\,\mathrm{d}x\\ &=\lim_{n\to\infty}\frac{n}{n+1}\log(n)-n\int_0^1\sum_{k=1}^\infty\frac{x^k}{k}x^n\,\mathrm{d}x\\ &=\lim_{n\to\infty}\frac{n}{n+1}\log(n)-n\sum_{k=1}^\infty\frac1{k(n+k+1)}\\ &=\lim_{n\to\infty}\frac{n}{n+1}\log(n)-\frac{n}{n+1}\sum_{k=1}^\infty\left(\frac1k-\frac1{n+k+1}\right)\\ &=\lim_{n\to\infty}\frac{n}{n+1}\left(\log(n)-\sum_{k=1}^{n+1}\frac1k\right)\\[6pt] &=-\gamma \end{align}

This is just the definition of the gamma function: $$\Gamma(s)=\int_0^{\infty}x^{s-1}e^{-x}dx.$$ Differentiating it with respect to $s$ and then setting $s=1$, we get $$-\gamma=\psi(1)=\left[\frac{d}{ds}\ln\Gamma(s)\right]_{s=1}=\int_0^{\infty}\ln x\,e^{-x}dx.$$

• It's not completely clear to me how you get the log under the integral. Jul 11, 2013 at 20:19
• @NickKidman $$\frac{d\Gamma(s)}{ds}=\frac{d}{ds}\int_0^{\infty }x^{s-1}e^{-x}dx=\int_0^{\infty}x^{s-1}\ln x\,e^{-x}dx.$$ Then it suffices to set $s=1$ in the final expression. Jul 11, 2013 at 20:32
• Ah okay, because the denominator in $\Gamma'(s)/\Gamma(s)$ becomes $1!$. Jul 11, 2013 at 20:35
• @NickKidman Yes, exactly. Jul 11, 2013 at 20:37

We are all familiar with the famous limit $~\displaystyle\lim_{h\to0}\frac{a^h-1}h~=~\ln a,~$ which, by the simple

substitution $h~=~\dfrac1n,~$ becomes $~\displaystyle\lim_{n\to\infty}~n\Big(\sqrt[^n]a-1\Big)~=~\ln a.~$ At the same time, using the

Taylor series expansion of $~\ln(1-t)~=-\displaystyle\sum_{k~=~1}^\infty\frac{t^k}k,~$ we can show, by switching the order

of summation and integration, and appealing to telescoping series, that $$\int_0^1\ln\Big(1-\sqrt[^n]x\Big)~dx~=-H_n,$$ the $n^{th}$ harmonic number. By combining the two formulas, we have \begin{align}\gamma&~=~\lim_{n\to\infty}\bigg[H_n-\ln n\bigg]~=~\lim_{n\to\infty}\bigg[-\int_0^1\ln\Big(1-\sqrt[^n]x\Big)~dx-\int_0^1\ln n~dx\bigg]~=~\\\\&~=-\lim_{n\to\infty}\int_0^1\ln\bigg[n\Big(1-\sqrt[^n]x\Big)\bigg]~dx~=-\int_0^1\ln\big(-\ln x\big)~dx~=-\Gamma'(1), \end{align} where the last identity has been obtained by differentiating under the integral sign Euler's

first historical expression for the $\Gamma$ function, $~n!~=~\displaystyle\int_0^1\big(-\ln x\big)^n~dx,~$ at $n~=~0$. But the

same function can also be written as $~n!~=~\displaystyle\int_0^\infty\frac{x^n}{e^x}~dx,~$ which, when differentiated under

the integral sign
at $n~=~0$, yields the original integral. QED.