10

The question came into my mind that how far we can go with the power $n $ in $\int_0^{\infty}\left(\frac{x}{e^{x^2}+e^{-x^2}}\right)^n \mathrm d x $ after I had found the exact values for $n=1$ and $n=2$, say

$$I_1= \frac{\pi}{8},$$ $$I_2= \frac{\sqrt{\pi}-\sqrt{2 \pi}}{8 \sqrt{2}} \zeta\left(\frac{1}{2}\right) $$ Proof: $$ \begin{aligned} I_1= & \int_0^{\infty} \frac{x}{e^{x^2}+e^{-x^2}} \textrm{ d} x \\ = & \frac{1}{2} \int_0^{\infty} \frac{1}{e^x+e^{-x}}\textrm{ d}x, \text { where } x^2 \mapsto x \\ =&\frac{1}{4} \int_0^{\infty}\frac{ 1}{\operatorname{cosh }x} \textrm{ d} x \\ =&\frac{1}{4} \int_0^{\infty} \frac{\operatorname{\cosh} x}{1+\sinh ^2 x} \textrm{ d} x \\ =&\frac{1}{4}\left[\tan ^{-1}(\sinh x)\right]_0^{\infty} \\ =&\frac{\pi}{8} \end{aligned} $$


Using the expansion for $|x|<1$, $$ \frac{1}{(1+x)^2}=\sum_{n=1}^{\infty}(-1)^{n-1} n x^{n-1}, $$ we have $$ \begin{aligned} I_2=& \int_0^{\infty}\left(\frac{x}{e^{x^2}+e^{-x^2}}\right)^2 \textrm{d}x \\ = & \frac{1}{2} \int_0^{\infty} \frac{\sqrt{x}}{\left(e^x+e^{-x}\right)^2} \textrm{d} x, \text { via } x^2 \rightarrow x\\ = & \frac{1}{2} \int_0^{\infty} \frac{e^{-2 x} \sqrt{x}}{\left(1+e^{-2 x}\right)^2} \textrm{d}x \\ = & \frac{1}{2} \sum_{n=1}^{\infty}(-1)^{n-1} n \int_0^{\infty} x^{\frac{1}{2}} e^{-2 n x} \textrm{d}x \end{aligned} $$ Utilising the formula: $\int_0^{\infty} t^{z-1} e^{-s t} d t=\frac{\Gamma(z)}{s^z}$ yields $$ \begin{aligned} I_2 & = \frac{1}{2} \sum_{n=1}^{\infty}\left((-1)^{n-1} n \cdot \frac{\Gamma\left(\frac{3}{2}\right)}{(2 n)^{\frac{3}{2}}}\right)\\ & =\frac{1}{4 \sqrt{2}} \sum_{n=1}^\infty \frac{(-1)^{n-1}}{\sqrt n} \\ & =\frac{\pi}{8 \sqrt{2}}\left(\sum_{n=1}^{\infty} \frac{1}{\sqrt n}-2 \sum_{n=1}^\infty \frac{1}{\sqrt{2 n}}\right) \\ & =\frac{\sqrt{\pi}-\sqrt{2 \pi}}{8 \sqrt{2}} \zeta\left(\frac{1}{2}\right) \end{aligned} $$


Similarly, using the expansion for $|x|<1$ : $\frac{1}{(1+x)^3}=\frac{1}{2} \sum_{n=1}^{\infty}(-1)^{n-1}(n+1) n x^{n-1}$ yields $$ \begin{aligned} I_3 & =\int_0^{\infty}\left(\frac{x}{e^{x^2}+e^{-x^2}}\right)^3 \textrm{d} x \\ & =\frac{1}{2} \int_0^{\infty} \frac{x}{\left(e^x+e^{-x}\right)^3} \textrm{d} x, \text { via } x^2 \rightarrow x\\ & =\frac{1}{2} \int_0^{\infty} \frac{e^{-3 x} x}{\left(1+e^{-2 x}\right)^3} \textrm{d} x \\ & =\frac{1}{2} \sum_{n=1}^{\infty}(-1)^{n-1}(n+1) n \int_0^{\infty} x e^{-(2 n+1) x} \textrm{d} x \\ & =\frac{1}{2} \sum_{n=1}^{\infty}\left((-1)^{n-1}(n+1) n \cdot\frac{\Gamma(2)}{(2 n+1)^2}\right) \\ & =\frac{1}{2} \sum_{n=1}^{\infty} \frac{(-1)^{n-1}(n+1) n}{(2 n+1)^2} \end{aligned} $$ which is divergent and contradictory to the answer $0.0259978$ found in WA.

Could you help me by giving opinions or an alternative? Your help is highly appreciated.

Integreek
  • 8,530
Lai
  • 31,615
  • 3
    You need to justify swapping integration and infinite summation. Commonly, we'd use Fubini to do this, but it doesn't apply here, since the series you found is not absolutely convergent, which is part of what Fubini requires to be applied. (I also think you've dropped a scalar multiple, but given the convergence issue, that's minor) – Brian Moehring May 21 '25 at 15:51
  • Surprisingly, the sum obtained in the last step is consistent with the equation (1) in the post – Lai May 21 '25 at 15:53

2 Answers2

7

We consider the below integral: $$ I(n) = \int_0^{\infty} \left( \frac{x}{e^{x^2} + e^{-x^2}} \right)^n \, dx $$

For background, recall that the Gamma function is defined by $$ \Gamma(z) = \int_0^\infty t^{z-1} e^{-t} \, dt, \quad \Re(z) > 0, $$ so that in particular $$ \int_0^\infty t^{z-1} e^{-st} \, dt = \frac{\Gamma(z)}{s^z} $$

Also recall the identity: $$ (1+u)^{-n} = \sum_{m=0}^\infty \binom{n+m-1}{m} (-u)^m $$

The main result is stated below: \begin{align} I(n) = \frac{\Gamma\!\left(\tfrac{n+1}{2}\right)}{2} \sum_{m=0}^\infty (-1)^m \binom{n+m-1}{m} (n+2m)^{-\frac{n+1}{2}} \end{align}

Substitution $t = x^2$. Let $t = x^2$, so that $dx = \frac{dt}{2\sqrt{t}}$, and $x = \sqrt{t}$. Then, \begin{align} I(n) &= \int_0^\infty \frac{x^n}{(e^{x^2} + e^{-x^2})^n} \, dx \\ &= \frac{1}{2} \int_0^\infty \frac{t^{n/2}}{(e^t + e^{-t})^n} t^{-1/2} \, dt \\ &= \frac{1}{2} \int_0^\infty t^{\frac{n-1}{2}} \frac{dt}{(e^t + e^{-t})^n} \end{align}

Hence, $$ I(n) = \frac{1}{2} \int_0^\infty t^{\frac{n-1}{2}} \frac{1}{(e^t + e^{-t})^n} \, dt $$

Rewriting the denominator. Observe that $$ e^t + e^{-t} = e^{-t}(1 + e^{2t}) = e^t(1 + e^{-2t}) $$ We use the form: $$ (e^t + e^{-t})^{-n} = e^{-nt}(1 + e^{2t})^{-n} = e^{nt}(1 + e^{-2t})^{-n} $$

Inserting this into the integral gives: $$ I(n) = \frac{1}{2} \int_0^\infty t^{\frac{n-1}{2}} e^{-nt} (1 + e^{-2t})^{-n} \, dt $$

Now apply the binomial series: $$ (1 + e^{-2t})^{-n} = \sum_{m=0}^\infty \binom{n+m-1}{m} (-e^{-2t})^m = \sum_{m=0}^\infty \binom{n+m-1}{m} (-1)^m e^{-2mt} $$

Substituting this into the integral: $$ I(n) = \frac{1}{2} \int_0^\infty t^{\frac{n-1}{2}} e^{-nt} \sum_{m=0}^\infty \binom{n+m-1}{m} (-1)^m e^{-2mt} \, dt $$

Assuming term-by-term integration is justified: \begin{align} I(n) = \frac{1}{2} \sum_{m=0}^\infty (-1)^m \binom{n+m-1}{m} \int_0^\infty t^{\frac{n-1}{2}} e^{-(n+2m)t} \, dt \end{align}

Using the Gamma function: $$ \int_0^\infty t^{\frac{n-1}{2}} e^{-(n+2m)t} \, dt = \frac{\Gamma\!\left( \frac{n+1}{2} \right)}{(n + 2m)^{\frac{n+1}{2}}} $$

Therefore: $$ I(n) = \frac{1}{2} \sum_{m=0}^\infty (-1)^m \binom{n+m-1}{m} \frac{\Gamma\!\left( \frac{n+1}{2} \right)}{(n + 2m)^{\frac{n+1}{2}}} $$

Finally, factor out the constant: $$ I(n) = \frac{\Gamma\!\left( \frac{n+1}{2} \right)}{2} \sum_{m=0}^\infty (-1)^m \binom{n+m-1}{m} (n + 2m)^{-\frac{n+1}{2}} $$

Numerical verification for the closed form and the integral

\begin{array}{c|c|c} n & \text{closed form } & \text{Integral Expression} \\ \hline 1 & 0.39269908169872415481 & 0.39269908169872415482 \\ 2 & 0.094766002680412 & 0.094766002680412133579 \\ 3 & 0.0259978496360762 & 0.025997849636077179039 \\ 4 & 0.0075402944966602 & 0.007540294496602175356 \\ 5 & 0.002256598131146 & 0.0022565981311454201286 \\ 6 & 0.000689062333003 & 0.00068906233300314074189 \\ 7 & 0.000213374690745 & 0.00021337469074537602378 \\ 8 & 0.0000667573925 & 0.00006675739251729180495 \\ 9 & 0.000021051436954106155279 & 0.000021051436954106155279 \\ 10 & 6.679926256 \times 10^{-6} & 6.6799262559104934239 \times 10^{-6} \\ \end{array}

  • How did you get the numerical values for the "closed form"? Since $$\binom{n+m-1}{m} = \binom{m+(n-1)}{n-1} \sim \frac{m^{n-1}}{(n-1)!}$$ the terms of the series (without the sign) are $\sim \frac{m^{(n-3)/2}}{2^{(n+1)/2}(n-1)!}$, so they don't tend to $0$ for $n\geqslant 3$, and for $n > 3$ they even tend to $+\infty$. – Dermot Craddock May 21 '25 at 18:16
  • @DermotCraddock I wrote Mathematica code for the closed form expression and evaluated it with Mathematica and it worked – Martin.s May 21 '25 at 18:28
  • @DermotCraddock expr[n_] := (Gamma[(n + 1)/2]/2) * Sum[(-1)^m * Binomial[n + m - 1, m] * (n + 2 m)^(-(n + 1)/2), {m, 0, Infinity}]

    Table[N[expr[n], 20], {n, 1, 10}]

    – Martin.s May 21 '25 at 18:29
  • 1
    Then I guess that Mathematica silently employed some summation method for divergent series. Nothing wrong with using such methods, but it should be stated if done. – Dermot Craddock May 21 '25 at 18:31
  • 1
    Following Martin.s's (this) answer, I did some algebra to obtain a "more legit" closed form: $$I_n=\frac{\Gamma\left(\frac{n+1}{2}\right)(-1)^{\lceil n/2\rceil}}{2^{\frac{n+3}{2}}(n-1)!}\sum_{k=0}^{n-1}\left([x^k]\prod_{a=-\frac{n-2}{2}}^{\frac{n-2}{2}}(x+a)\right)\sum_{m=1}^{\infty}\frac{(-1)^m}{\left(m+\frac{n}{2}-\lceil \frac{n}{2}\rceil\right)^{\frac{n+1}{2}-k}}$$ – Quý Nhân May 22 '25 at 00:40
  • @Martin.s, thank you very much for your solution in series. However, I wonder about the convergence of the infinite series, for example, $I_3=\frac{1}{2} \sum_{n=1}^{\infty} \frac{(-1)^{n-1}(n+1) n}{(2 n+1)^2}$ is divergent. – Lai May 22 '25 at 00:57
  • @Lai That sum-integral interchanging you and Martin.s did is invalid. But if anyone can prove that it works under regularization of Dirichlet series, it will open up a way to evaluate these types of integrals without worrying about convergence, I think. – Quý Nhân May 22 '25 at 03:18
7

\begin{align} &\int_0^{\infty}\bigg(\frac{x}{e^{x^2}+e^{-x^2}}\bigg)^3dx \>\>\>\>\>x^2\to x \\ =& \ \frac1{16}\int_0^\infty \frac x{\cosh^3x}dx = \frac1{32}\int_0^\infty \frac x{\sinh x}d(\tanh^2x)\\ \overset{ibp}= &\ \frac1{32}\int_0^\infty \frac x{\cosh x}-\frac{\sinh x}{\cosh^2 x}\ dx= \ \frac1{32} (2G-1) \end{align}

where $\int_0^\infty \frac x{\cosh x}dx=2G$.

Quanto
  • 120,125