23
$\begingroup$

I'm curious, how one can prove the following integral $$ \int_0^{\pi/2}{\frac{1+2\cos 2x\cdot\ln\tan x}{1+\tan^{2\sqrt{2}} x}}\tan^{1/\sqrt{2}} x~dx=0 $$ Here is the Wolfram Alpha computation which shows that it is correct to at least 45 digits.

My attempt: I knew the integral $$ \int_0^{\pi/2}\frac{1}{1+\tan^\alpha\phi}d\phi=\int_0^{\pi/4}\frac{1}{1+\tan^\alpha\phi}d\phi+\int_0^{\pi/4}\frac{\tan^\alpha\phi}{1+\tan^\alpha\phi}d\phi=\frac{\pi}{4} $$ which can be calculated for all values of $\alpha$. I tried to find an analogous symmetry that will allow me to cancel all the terms also in this case, but so far no luck. I also suspect that this integral might be related to derivative of Herglotz integral. Herglotz showed that $$ \int_{0}^{1} \frac{\ln\left(1 + t^{\,{\large\alpha}}\right)}{1 + t}\,{\rm d}t $$ can be computed for some algebraic values of $\alpha$, e.g. $\alpha=4+\sqrt{5}$. If we take derivative of this integral with respect to $\alpha$ then we get $$ \int_{0}^{1} \frac{t^\alpha\ln t}{(1 + t)(1+t^\alpha)}\,{\rm d}t $$ Change of variables $t=\tan^2\phi$ gives $$ 4\int_{0}^{\pi/4} \frac{\tan^{2\alpha+1}\phi\cdot\ln \tan\phi}{1+\tan^{2\alpha}\phi}\,{\rm d}\phi $$ which looks quite similar to the integral under consideration.

$\endgroup$
  • 3
    $\begingroup$ By the way, how did you find this identity? $\endgroup$ – Yuriy S Jan 11 '17 at 19:19
  • 2
    $\begingroup$ An equivalent question is, if this equation is right: $\displaystyle \int\limits_0^\infty \frac{\cosh(x)} { \cosh(2x) \cosh(x\sqrt{2})}dx = 2\sqrt{2}\int\limits_0^\infty \frac{x\cosh(x)\sinh(x\sqrt{2})} {\cosh(2x)\cosh^2(x\sqrt{2})} dx$ $\endgroup$ – user90369 Jan 12 '17 at 10:43
  • 2
    $\begingroup$ @user90369, still - your integrals will work nicely for the numerical experiments I've been doing. Thanks. I could investigate the integral: $$Y(A,B)=\int_0^\infty \frac{\cosh (Ax)}{\cosh (Bx)} \frac{\cosh x-2x \sinh x}{\cosh^2 x} dx$$ For this question $$Y(\sqrt{2}/2,\sqrt{2})=0$$ $\endgroup$ – Yuriy S Jan 12 '17 at 11:00
  • 1
    $\begingroup$ @YuriyS : Using your integral it would be nice to know how the function $Y(A):=Y(A,2A)$ look likes. :-) Is there something special with this function ? $\endgroup$ – user90369 Jan 12 '17 at 12:33
  • 1
    $\begingroup$ MAPLE says that it is correct to at least 100 digits. $\endgroup$ – Han de Bruijn Jan 12 '17 at 20:03
5
+250
$\begingroup$

This answer builds on observations made by user90369 and Yuriy S. They define the functions $$ Y_1(A,P):=\int_0^\infty \frac{\cosh Ax}{\cosh 2Ax} \frac{1}{\cosh P x} dx,\tag{1} $$ $$ Y_2(A,P):=-2\int_0^\infty \frac{\cosh Ax}{\cosh 2Ax} \frac{x \sinh P x}{\cosh^2 P x} dx=2\frac{\partial}{\partial P} Y_1(A,P),\tag{2} $$ and then note that OPs conjecture is equivalent to $Y_1(1/\sqrt2,1)+Y_2(1/\sqrt2,1)=0$.

I happened to notice that $(1)$ is an integral of a product of two self-reciprocal functions $\sqrt{\frac{2}{\pi}}\int_0^\infty f(x)\cos ax dx=f(a)$ (functions 3 and 4 in this list): $$ f_1(x)=\frac{1}{\cosh\sqrt{\frac{\pi}{2}}x},\quad f_2(x)=\frac{\cosh \frac{\sqrt{\pi}x}{2}}{\cosh \sqrt{\pi}x}.\tag{3} $$

For any such two functions Ramanujan shows that (simple proof is outlined in this answer) $$ \int_0^\infty f_1(x)f_2(\alpha x) dx= \frac{1}{\alpha}\int_0^\infty f_1(x)f_2(x/\alpha) dx.\tag{4} $$

Substituting $(3)$ in $(4)$ one obtains $$ \int_0^\infty \frac{1}{\cosh\sqrt{\frac{\pi}{2}}\alpha x}\frac{\cosh \frac{\sqrt{\pi}x}{2}}{\cosh \sqrt{\pi}x} dx= \frac{1}{\alpha}\int_0^\infty \frac{1}{\cosh\sqrt{\frac{\pi}{2}}x/\alpha}\frac{\cosh \frac{\sqrt{\pi}x}{2}}{\cosh \sqrt{\pi}x} dx, $$ and after trivial algebra $$ \int_0^\infty \frac{1}{\cosh\alpha x}\frac{\cosh \frac{x}{\sqrt{2}}}{\cosh \sqrt{2}x} dx= \frac{1}{\alpha}\int_0^\infty \frac{1}{\cosh x/\alpha}\frac{\cosh \frac{x}{\sqrt{2}}}{\cosh \sqrt{2}x} dx. $$ This gives the functional equation $$Y_1(1/\sqrt{2},P)=\frac{1}{P}Y_1(1/\sqrt{2},1/P).$$ Differentiating this functional equation with respect to $P$ one obtains $$ \frac12 Y_2(1/\sqrt{2},P)=-\frac{1}{P^2}Y_1(1/\sqrt{2},1/P)-\frac{1}{2P^3}Y_2(1/\sqrt{2},1/P), $$ which gives at $P=1$ the required relation $$ Y_1(1/\sqrt2,1)+Y_2(1/\sqrt2,1)=0. $$

$\endgroup$
  • 1
    $\begingroup$ The links you gave seems to allow us to construct dozens of such curious integrals as in the OP $\endgroup$ – Yuriy S Jan 17 '17 at 9:44
  • $\begingroup$ @YuriyS credits to user90369 are added. $\endgroup$ – Nemo Jan 17 '17 at 10:19
  • $\begingroup$ Could you please point a functional equation which works for "Self-reciprocal Fourier functions of the second kind" (paragrath 2. in the same answer)? Is it the same? As far as I see from your proof, it should be the same $\endgroup$ – Yuriy S Jan 17 '17 at 10:56
  • $\begingroup$ @YuriyS , yes, they are the same. $\endgroup$ – Nemo Jan 17 '17 at 11:02
9
$\begingroup$

Disclaimer. This is not an answer! But I hope the following transformations and numerical experiments can be useful in finding the solution.

First, we will get rid of trigonometric functions. Setting $t=\tan \phi$ we obtain:

$$\int_0^\infty \frac{t^a [1+t^2+2(1-t^2) \ln t]}{(1+t^b)(1+t^2)^2}dt$$

Here $a=\frac{1}{2}\sqrt{2}$, $b=2\sqrt{2}$. It makes sense to pose a more general problem with $a=\frac{1}{2}c$, $b=2c$ for some real parameter $c$.

Thus we have:

$$I(c)=\int_0^\infty \frac{t^{c/2} [1+t^2+2(1-t^2) \ln t]}{(1+t^{2c})(1+t^2)^2}dt$$

Here is a plot of $I(c)$. There appear to be only two roots $c= \pm \sqrt{2}$ - the function appears to be even.

enter image description here

It's easy to prove that $I(c)=I(-c)$. Let's set $t=\frac{1}{u}$:

$$I(c)=-\int_\infty^0 \frac{u^{-c/2} [1+u^{-2}-2(1-u^{-2}) \ln u]}{u^2(1+u^{-2c})(1+u^{-2})^2}du$$

Multiplying both sides by $u^2$ we obtain:

$$I(c)=\int_0^\infty \frac{u^{-c/2} [1+u^2+2(1-u^2) \ln u]}{(1+u^{-2c})(1+u^2)^2}du=I(-c)$$


Let's separate the integral into to parts: $$I=\int_0^\infty=\int_0^1+\int_1^\infty$$

Making the substitution $t=\frac{1}{u}$ we obtain:

$$\int_1^\infty \frac{t^{c/2} [1+t^2+2(1-t^2) \ln t]}{(1+t^{2c})(1+t^2)^2}dt=\int_0^1 \frac{u^{3c/2} [1+u^2+2(1-u^2) \ln u]}{(1+u^{2c})(1+u^2)^2}du$$

Thus, we can introduce another form of the integral with finite limits:

$$I(c)=\int_0^1 \frac{t^{c/2}(1+t^c) [1+t^2+2(1-t^2) \ln t]}{(1+t^{2c})(1+t^2)^2}dt$$

The plot of the integrand is presented below:

enter image description here

Because now $0<t<1$ we can expand a part of the integrand into a series:

$$I(c)=\sum_{k=0}^\infty (-1)^k \int_0^1 \frac{t^{(2k+1/2)c}(1+t^c) [1+t^2+2(1-t^2) \ln t]}{(1+t^2)^2}dt$$

Expanding the numerator we obtain $8$ integrals in the form:

$$\int_0^1 \frac{t^\alpha}{(1+t^2)^2} dt \quad \text{and} \quad \int_0^1 \frac{t^\alpha \ln t}{(1+t^2)^2} dt$$

The first integral can be evaluated using the digamma function while the second integral can be obtained by differentiating the first w.r.t. $\alpha$.

Thus, the integral is equal to the following series:

$$I(c)=\frac{1}{16} \sum_{k=0}^\infty (-1)^k \times$$

$$ \times \left[ 4 \left(\psi\left(\frac{c (4 k+1)+2}{8}\right)-\psi \left(\frac{c (4 k+1)+2}{8}+\frac{1}{2}\right)\right)+ \\ +4 \left(\psi \left(\frac{c (4 k+3)+2}{8}\right)-\psi \left(\frac{c (4 k+3)+2}{8}+\frac{1}{2}\right)\right)+ \\ +c (4 k+1) \left(\psi ^{(1)}\left(\frac{c (4 k+1)+2}{8}\right)-\psi ^{(1)}\left(\frac{c (4 k+1)+2}{8}+\frac{1}{2}\right)\right)+ \\ +c (4 k+3) \left(\psi ^{(1)}\left(\frac{c (4 k+3)+2}{8}\right)-\psi ^{(1)}\left(\frac{c (4 k+3)+2}{8}+\frac{1}{2}\right)\right) \right]$$

This seems too complicated, but I have hope that some properties of the digamma function can help in showing that $I(\sqrt{2})=0$.


A little more generalization. Consider the integral with two parameters:

$$I(a,b)=\int_0^\infty \frac{t^a [1+t^2+2(1-t^2) \ln t]}{(1+t^b)(1+t^2)^2}dt=$$

$$=\int_0^1 \frac{(t^a+t^{b-a}) [1+t^2+2(1-t^2) \ln t]}{(1+t^b)(1+t^2)^2}dt$$

$$I(a,b)=I(b-a,b)$$

The curve $I(a,b)=0$ is represented below:

enter image description here

We see it has two branches. So far I haven't been able to find other algebraic pairs on it except for:

$$I \left(\frac{1}{2} \sqrt{2},2\sqrt{2} \right)=0 \\ I \left(\frac{3}{2} \sqrt{2},2\sqrt{2} \right)=0$$

And the trivial one:

$$I(1,2)=0$$


Update:

The integral can be simplified in the following way:

$$\int_0^\infty \frac{t^a [1+t^2+2(1-t^2) \ln t]}{(1+t^b)(1+t^2)^2}dt=\frac{1}{2} \int_0^\infty \frac{t^{a-1} [1+t^2+2(1-t^2) \ln t]}{(1+t^b)(1+t^2)^2}dt^2$$

Introducing $v=t^2$, $\alpha=\frac{a-1}{2}$ and $\beta=\frac{b}{2}$ we need to investigate:

$$J(\alpha,\beta)=\int_0^\infty \frac{v^{\alpha} [1+v+(1-v) \ln v]}{(1+v^\beta)(1+v)^2}dv=\int_0^1 \frac{(v^{\alpha}+v^{\beta-\alpha-1}) [1+v+(1-v) \ln v]}{(1+v^\beta)(1+v)^2}dv$$

$$J(\alpha,\beta)=J(\beta-\alpha-1,\beta)$$

This integral seems much better. We now need to prove that:

$$J\left(\frac{\sqrt{2}-2}{4},\sqrt{2} \right)=0$$ or $$J\left(\frac{3\sqrt{2}-2}{4},\sqrt{2} \right)=0$$


Update 2:

Some further work on the integral. From the comment of @user90369 we introduce a new form of the integral:

$$Y(A):=\int_0^\infty \frac{\cosh Ax}{\cosh 2Ax} \frac{\cosh x-2x \sinh x}{\cosh^2 x} dx$$

Setting $t=e^x$ we obtain:

$$Y(A)=2 I(2A)$$

Separating the integral into two parts and introducing a new parameter $P$ we have:

$$Y_1(A,P):=\int_0^\infty \frac{\cosh Ax}{\cosh 2Ax} \frac{1}{\cosh P x} dx$$

$$Y_2(A,P):=-2\int_0^\infty \frac{\cosh Ax}{\cosh 2Ax} \frac{x \sinh P x}{\cosh^2 P x} dx=2\frac{\partial}{\partial P} Y_1(A,P)$$

Thus, we can work with $Y_1$ first:

$$Y_1(A,P)=\frac{1}{A} \int_0^\infty \frac{\cosh y}{\cosh 2y} \frac{1}{\cosh p y} dy$$

Here $p=P/A$. Setting $e^y=z$ we obtain:

$$Y_1(p):=\int_0^\infty \frac{\cosh y}{\cosh 2y} \frac{1}{\cosh p y} dy=2 \int_0^1 \frac{1+z^2}{1+z^4} \frac{z^p}{1+z^{2p}} dz=\int_0^\infty \frac{1+z^2}{1+z^4} \frac{z^p}{1+z^{2p}} dz$$

Despite @Lucian's comment, I can't represent $Y_1(p)$ in terms of Beta function in closed from, only as a series, but it might be possible. I would wager a closed form in terms of hypergeometric functions.

For particular values of $p$ there are closed forms. So far I've found closed forms for $p=(0,1/4,1/3,1/2,2/3,1,2,3,4,5)$. They are all algebraic multiples of $\pi$. But to evaluate the original integral, we need the closed form for general $p$. Seethis question.


Update 3:

If we allow double sums we can express the problem in a compact and almost symmetrical way:

Prove that for $B=\sqrt{2}$: $$\sum_{k,n=0}^\infty (-1)^{k+n} \frac{4k+1-(2n+1)B}{(4k+1+(2n+1)B)^2}=-\sum_{k,n=0}^\infty (-1)^{k+n} \frac{4k+3-(2n+1)B}{(4k+3+(2n+1)B)^2}$$

$\endgroup$
  • $\begingroup$ Good work! Thanks! (+1) $\endgroup$ – Tyrell Jan 11 '17 at 6:45
  • $\begingroup$ @Tyrell, well, I haven't done anything to answer the question yet - that's why I offered a bounty in hopes that someone more knowledgeable comes along $\endgroup$ – Yuriy S Jan 11 '17 at 6:59
  • $\begingroup$ This type of integral has beta function written all over it. $\endgroup$ – Lucian Jan 12 '17 at 18:19
  • $\begingroup$ @Lucian, not as any finite combination of beta functions, I'm afraid $\endgroup$ – Yuriy S Jan 14 '17 at 21:47
0
$\begingroup$

This is not a complete answer, but I think this is a possible solution. If you apply the change of variable $t=\tan(x)$, the original integral can be writen as

$$\int_{0}^{\infty}\frac{1 + 2\left(\frac{1-t^{2}}{1+t^{2}}\right)\log(t)}{1+t^{2\sqrt{2}}}t^{1/\sqrt{2}}\frac{dt}{t^{2} + 1} = \int_{0}^{\infty}\frac{t^{1/\sqrt{2}}}{(t^{2} + 1)(1+t^{2\sqrt{2}})}dt + \int_{0}^{\infty}\frac{2(1-t^{2})\log(t)}{(t^{2} + 1)^{2}(1+t^{2\sqrt{2}})}t^{1/\sqrt{2}}dt,$$

and this integral would be solved by the method of residues for Mellin Transformations.

Checks Pags. 60-62 of http://jacobi.fis.ucm.es/artemio/Notas%20de%20curso/VC.pdf (I couldn't find an english version)

$\endgroup$
  • $\begingroup$ Can you complete the calculations and prove the statement? $\endgroup$ – Tyrell Jan 13 '17 at 8:30

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.