I would like to prove that:

$$ \int_{0}^{\infty} \frac{\ln(t)}{\sqrt{t}}e^{-t} \mathrm dt=-\sqrt{\pi}(\gamma+\ln{4})$$

I tried to use the integral $$\int_{0}^{n} \frac{\ln(t)}{\sqrt{t}}\left(1-\frac{t}{n}\right)^n \mathrm dt$$

$$\int_{0}^{n} \frac{\ln(t)}{\sqrt{t}}\left(1-\frac{t}{n}\right)^n \mathrm dt \;{\underset{\small n\to\infty}{\longrightarrow}}\; \int_{0}^{\infty} \frac{\ln(t)}{\sqrt{t}}e^{-t} \mathrm dt$$ (dominated convergence theorem)

Using the substitution $t\to\frac{t}{n}$, I get:

$$ \int_{0}^{n} \frac{\ln(t)}{\sqrt{t}}\left(1-\frac{t}{n}\right)^n \mathrm dt=\sqrt{n}\left(\ln(n)\int_{0}^{1} \frac{(1-t)^n}{\sqrt{t}} \mathrm dt+\int_{0}^{1} \frac{\ln(t)(1-t)^n}{\sqrt{t}} \mathrm dt\right) $$

However I don't know if I am on the right track for these new integrals look quite tricky.

  • $\begingroup$ More problematic, the integrals don't seem to be convergent. $\endgroup$ – Fabian Jan 19 '12 at 21:36

Consider integral representation for the Euler $\Gamma$-function: $$ \Gamma(s) = \int_0^\infty t^{s-1} \mathrm{e}^{-t} \mathrm{d} t $$ Differentiate with respect to $s$: $$ \Gamma(s) \psi(s) = \int_0^\infty t^{s-1} \ln(t) \mathrm{e}^{-t} \mathrm{d} t $$ where $\psi(s)$ is the digamma function. Now substitute $s=\frac{1}{2}$. So $$ \int_0^\infty \frac{ \ln(t)}{\sqrt{t}} \mathrm{e}^{-t} \mathrm{d} t = \Gamma\left( \frac{1}{2} \right) \psi\left( \frac{1}{2} \right) $$ Now use duplication formula: $$ \Gamma(2s) = \Gamma(s) \Gamma(s+1/2) \frac{2^{2s-1}}{\sqrt{\pi}} $$ Differentiating this with respect to $s$ gives the duplication formula for $\psi(s)$, and substitution of $s=1/2$ gives $\Gamma(1/2) = \sqrt{\pi}$. $$ \psi(2s) = \frac{1}{2}\psi(s) + \frac{1}{2} \psi(s+1/2) + \log(2) $$ Substitute $s=\frac{1}{2}$ and use $\psi(1) = -\gamma$ to arrive at the result.

| cite | improve this answer | |
  • 1
    $\begingroup$ It may be helpful to state that $\Gamma$ is the Gamma-function (which I suspect is well known by most readers), and that $\psi$ is the digamma function (which I suspect is less well-known). $\endgroup$ – JavaMan Jan 19 '12 at 21:58
  • $\begingroup$ @JavaMan Thanks. Heeding your suggestion I have added wiki links for both. $\endgroup$ – Sasha Jan 19 '12 at 22:10
  • $\begingroup$ and that $\psi = [\log \Gamma]'$ $\endgroup$ – Fabian Jan 19 '12 at 22:10
  • 1
    $\begingroup$ @Fabian This apparent from the second line, no ? $\endgroup$ – Sasha Jan 19 '12 at 22:12
  • $\begingroup$ Great! Thank you! $\endgroup$ – Chon Jan 19 '12 at 22:21

I don't have enough time to write a full solution (I will when I find the time), but hopefully this skeletal answer is sufficiently illustrative.

The idea is to use the Frullani integral

$$\log\frac{b}{a}=\int_0^\infty \frac{\exp(-at)-\exp(-bt)}{t}\mathrm dt$$

and substitute this into the original integral:

$$\int_0^{\infty} \frac{\exp(-t)}{\sqrt{t}} \log\,t\;\mathrm dt=\int_0^{\infty} \frac{\exp(-t)}{\sqrt{t}} \left(\int_0^\infty \frac{\exp(-u)-\exp(-tu)}{u}\mathrm du\right)\mathrm dt$$


$$\int_0^{\infty}\frac1{u}\int_0^\infty \frac{\exp(-t-u)-\exp(-t(1+u))}{\sqrt{t}}\mathrm dt\mathrm du$$

which expands to

$$\int_0^{\infty}\frac1{u}\left(\exp(-u)\int_0^\infty \frac{\exp(-t)}{\sqrt{t}}\mathrm dt-\int_0^\infty \frac{\exp(-t(1+u))}{\sqrt{t}}\mathrm dt\right)\mathrm du$$

There is the Laplace transform relation

$$\int_0^\infty \frac{\exp(-cv)}{\sqrt v}\mathrm dv=\sqrt{\frac{\pi}{c}}$$

which we can use here to yield

$$\sqrt{\pi}\int_0^{\infty}\frac1{u}\left(\exp(-u)-\frac1{\sqrt{1+u}}\right)\mathrm du$$

Integration by parts yields the equivalent

$$-\sqrt{\pi}\left(\frac12\int_0^{\infty}\frac{\log\,u}{(1+u)^\frac32}\mathrm du\color{blue}{-\int_0^{\infty}\exp(-u)\log\,u\mathrm du}\right)$$

and the portion highlighted in blue is recognized to be the Euler-Mascheroni constant (see this as well):

$$-\sqrt{\pi}\left(\frac12\int_0^{\infty}\frac{\log\,u}{(1+u)^\frac32}\mathrm du+\color{blue}{\gamma}\right)$$

Still another application of integration by parts yields

$$-\sqrt{\pi}\left(\frac12\left.{\left(\frac{2\log\,u}{\sqrt{u+1}}-2\log \left(1+\frac{2\left(1-\sqrt{1+u}\right)}{u}\right)\right)}\right|_0^\infty+\gamma\right)$$

and then one can use e.g. l'Hôpital's rule to evaluate the limit implicit in the expression above (for $u\to 0$ that is; for $u\to\infty$, no special treatment is needed and the logarithmic expression evaluates to zero) to finally yield the expected $-\sqrt{\pi}(\log\,4+\gamma)$.

| cite | improve this answer | |

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.