In a Physics class, I encountered the following integral: $$\int_{0}^{\infty} \left(\frac{x^2}{\exp\left(x-a\right)+1}-\frac{x^2}{\exp\left(x+a\right)+1}\right)\,\mathrm{d}x\text{,}$$

where $a$ is a constant. It is stated to yield: $$\int_{0}^{\infty}\left(\frac{x^2}{\exp\left(x-a\right)+1}-\frac{x^2}{\exp\left(x+a\right)+1}\right)\,\mathrm{d}x=\frac{1}{3}\left[\pi^2a+a^3\right]$$

I first introduced a change of variables: $$y:=x-a$$ $$z:=x^\prime+a$$ This gives: $$\int_{0}^{\infty}\frac{x^2}{\exp\left(x-a\right)+1}\,\mathrm{d}x-\int_{0}^{\infty}\frac{\left(x^\prime\right)^2}{\exp\left(x^\prime+a\right)+1}\,\mathrm{d}x^\prime=\int_{-a}^{\infty}\frac{\left(y+a\right)^2}{e^y + 1}\,\mathrm{d}y-\int_{a}^{\infty}\frac{\left(z-a\right)^2}{e^z+1}\,\mathrm{d}z$$ However, when introduced to a computer, the definite integrals are stated to have no solution. I actually expected to get integrals of the form: $$\int_{0}^{\infty}\frac{y}{e^y+1}\,\mathrm{d}y=\frac{\pi^2}{12}$$ What am I missing here?

  • $\begingroup$ If we observe that the integrand in the original integral is an even function and then integrate from $-\infty$ to $+\infty$ then at least Maple and Wolfram Alpha deliver the the value immediately. $\endgroup$
    – JanG
    Mar 7 at 17:10

You are on the right track. Continue by renaming $z$ to $y$ and writing \begin{align} \int \limits_0^\infty x^2 \left(\frac{1}{\mathrm{e}^{x-a}+1} - \frac{1}{\mathrm{e}^{x+a}+1}\right) \, \mathrm{d} x &= \int \limits_{-a}^\infty \frac{(y+a)^2}{\mathrm{e}^y + 1} \, \mathrm{d} y - \int \limits_a^\infty \frac{(z-a)^2}{\mathrm{e}^z + 1} \, \mathrm{d} z \\ &= \int \limits_{-a}^a \frac{(y+a)^2}{\mathrm{e}^y + 1} \, \mathrm{d} y + \int \limits_a^\infty \frac{(y+a)^2 - (y-a)^2}{\mathrm{e}^{y} + 1} \, \mathrm{d} y \\ &= \int \limits_{-a}^0 \frac{(y+a)^2}{\mathrm{e}^y + 1} \, \mathrm{d} y + \int \limits_0^a \frac{(y+a)^2 - 4 a y}{\mathrm{e}^y + 1} \, \mathrm{d} y + \int \limits_0^\infty \frac{4ay}{\mathrm{e}^y + 1} \, \mathrm{d} y \\ &= \int \limits_{-a}^0 \frac{(a + y)^2}{\mathrm{e}^y + 1} \, \mathrm{d} y + \int \limits_0^a \frac{(a - y)^2}{\mathrm{e}^y + 1} \, \mathrm{d} y + \int \limits_0^\infty \frac{4ay}{\mathrm{e}^y + 1} \, \mathrm{d} y \, . \end{align} Now substitute $y \to - y$ in the first integral and add it to the second integral. Since you know the value of the third integral, this yields \begin{align} \int \limits_0^\infty x^2 \left(\frac{1}{\mathrm{e}^{x-a}+1} - \frac{1}{\mathrm{e}^{x+a}+1}\right) \, \mathrm{d} x &= \int \limits_0^a (a - y)^2 \left[\frac{1}{\mathrm{e}^{-y} + 1} + \frac{1}{\mathrm{e}^y + 1}\right] \, \mathrm{d} y + 4 a \int \limits_0^\infty \frac{y}{\mathrm{e}^y + 1} \, \mathrm{d} y \\ &= \int \limits_0^a (a-y)^2 \, \mathrm{d} y + 4 a \frac{\pi^2}{12} \\ &= \frac{a}{3}(\pi^2 + a^2) \end{align} for $a \in \mathbb{R}$ as desired. Note that we had to combine the two integrals over $y$ and $z$ in a particular way to obtain this result, which is why some computer programs may be unable to evaluate them individually.


If you encountered this integral in a statistical mechanics class, perhaps they wanted you to use the integral of the Fermi-Dirac distribution and properties of the polylogarithm function.

Since $$-\operatorname{Li}_{3}(-x) = \frac{1}{2} \int_{0}^{\infty} \frac{t^{2}}{e^{t}/x+1} \, \mathrm dt, \, \quad x \ge -1,$$ and the inversion formula for the trilogarithm function states that $$\operatorname{Li}_{3}(-x) - \operatorname{Li}_{3} \left(- \frac{1}{x} \right) = - \frac{1}{6} \, \ln^{3}(x) - \frac{\pi^{2}}{6} \ln x \, , \quad x \notin (-1,0], \tag{1}$$

we have $$ \begin{align} \int_{0}^{\infty} \left(\frac{t^2}{e^{t-a}+1}-\frac{t^2}{e^{t+a}+1}\right)\,\mathrm dt &= - 2\operatorname{Li}_{3}(-e^{a})+ 2 \operatorname{Li}_{3} \left(-\frac{1}{e^{a}}\right) \\ &= 2 \left( \frac{1}{6} \, \ln^{3}(e^{a}) + \frac{\pi^{2}}{6} \ln(e^{a}) \right) \\ &= 2 \left(\frac{a^{3}}{6}+ \frac{ \pi^{2}a }{6} \right) \\ &= \frac{a}{3} \left(a^{2} + \pi^{2} \right). \end{align}$$

$(1)$ The inversion formula for the trilogarithm can be derived from the inversion formula of the dilogarithm by dividing both sides of the equation by $x$ and integrating.

  • $\begingroup$ +1 for the extra solution! Actually, when I plugged the non-definite integral into computer software, the answer was given in terms of polylogarithms (which are not familiar to me); thanks to your answer I will be able to understand it! $\endgroup$
    – JD_PM
    Mar 6 at 8:25

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.