86

I discovered the following conjectured identity numerically while studying a family of related integrals.

Let's set $$ R^{+}:= \frac{2}{\pi}\int\limits_{0}^{\pi/2}\sqrt[\normalsize{8}]{x^2 + \ln^2\!\cos x} \sqrt{ \frac{1}{2}+\frac{1}{2}\sqrt{ \frac{1}{2}+ \frac{1}{2} \sqrt{ \frac{\ln^{2}\!\cos x}{ x^2 + \ln^2\! \cos x}}}}\,\mathrm{d}x, \tag1 $$

$$ R^{-}:= \frac{2}{\pi}\int\limits_{0}^{\pi/2}\frac{1}{\sqrt[\normalsize{8}]{x^2 + \ln^2\!\cos x}} \sqrt{ \frac{1}{2}+\frac{1}{2}\sqrt{ \frac{1}{2}+ \frac{1}{2} \sqrt{ \frac{\ln^{2}\!\cos x}{ x^2 + \ln^2\! \cos x}}}}\,\mathrm{d}x. \tag2 $$

We may numerically observe with at least 500 digits of precision that

$$ \begin{align} & R^{+}R^{-} \stackrel{?}{=}1 \tag 3 \\\\ & R^{+} \stackrel{?}{=} \sqrt[\normalsize{4}]{\ln 2} \tag4 \\\\ & R^{-} \stackrel{?}{=}\frac{1}{\sqrt[\normalsize{4}]{\ln 2}}. \tag5 \end{align} $$

How can we prove it?

A version of this has been sent to Eric Weisstein, these integrals are on Mathworld as Ramanujan log-trigonometric integrals.

Alwaz
  • 155
Olivier Oloa
  • 122,789
  • My immediate (albeit likely overly optimistic) thought is that their product could be viewed as a double integral, and this "somehow" shown to be equal to unity. – Semiclassical Jul 31 '14 at 20:10
  • @semiclassical The double integral seems frightening :) – Olivier Oloa Jul 31 '14 at 20:17
  • Oh, yes. That somehow is a bit like suggesting that it would "somehow" be possible for me to climb Mount Everest---it may be true, but it's still not bloody likely! On a more topical note, it also vaguely reminds me of certain properties of 'functional determinants' in field theory. (But that doesn't help much either!) – Semiclassical Jul 31 '14 at 20:18
  • 5
    The identity $$\sqrt{(1/2)+(1/2)\cos\theta}=\sin(\theta/2)$$ suggests making the substitution $$\cos\theta=\sqrt{\log^2\cos x/(x^2+\log^2\cos x)}$$ – Gerry Myerson Aug 17 '14 at 01:29
  • Re Gerry Myerson: And then, perhaps, the half angle formulae in the first terms of Vieta's product for $\pi$. – marty cohen Sep 21 '14 at 03:44
  • 6
    The integrals can be rewritten a bit more compactly in the following way: $$R^+:=\frac{2}{\pi}\int_0^{\pi/2}\text{Re}\left [ (ix-\ln(\cos x))^{1/4} \right ]dx\R^-:=\frac{2}{\pi}\int_0^{\pi/2}\text{Re}\left [ \left (-ix-\ln(\cos x)\right )^{-1/4} \right ]dx$$ – SDiv Sep 22 '14 at 23:04
  • @SDiv You are on the right track ;-) – Olivier Oloa Sep 22 '14 at 23:04

3 Answers3

50

Here is an approach.

Theorem. Let $s$ be a real number such that $-1<s<1$. Then

\begin{equation}{\Large\int_{0}^{\!\Large \frac{\pi}{2}}} \frac{\cos \left(\! s \arctan \left(-\frac{x}{\ln \cos x}\right)\right)}{(x^2+\ln^2\! \cos x)^{\Large\frac{s}{2}}}\, \mathrm{d}x = \frac{\pi}{2}\frac{1}{\ln^{\Large s}\!2} \tag1\end{equation}

Proof. First assume that $0<s<1.$ Then we may write $$ \begin{align} \int_{0}^{\!\Large \frac{\pi}{2}} \frac{\cos \left(\! s \arctan \left(-\frac{x}{\ln \cos x}\right)\right)}{(x^2+\ln^2\! \cos x)^{s/2}} \mathrm{d}x & = \frac{1}{\Gamma(s)}\int_{0}^{\!\Large \frac{\pi}{2}}\!\!\int_{0}^{+\infty} u^{s-1} \cos (u x) \:e^{u \ln \cos x}\:\mathrm{d}u \:\mathrm{d}x \tag2\\\\ & = \frac{1}{\Gamma(s)}\int_{0}^{+\infty}u^{s-1}\!\!\int_{0}^{\!\Large \frac{\pi}{2}} \cos^u\! x \cos (u x)\:\mathrm{d}x \:\mathrm{d}u \tag3\\\\ & = \frac{1}{\Gamma(s)}\int_{0}^{+\infty}u^{s-1} \frac{\pi}{2^{u+1}}\:\mathrm{d}u \tag4\\\\ & = \frac{\pi}{2} \frac{1}{\Gamma(s)}\int_{0}^{+\infty}u^{s-1} e^{-u\ln 2}\:\mathrm{d}u \tag5\\\\ & = \frac{\pi}{2}\frac{1}{\Gamma(s)}\frac{\Gamma(s)}{\ln^{s}\!2} \tag6\\\\ & = \frac{\pi}{2}\frac{1}{\ln^{s}\!2} \tag7 \end{align} $$ where we have used Fubini's theorem and the classic results (here and there) $$ \begin{align} & \int_{0}^{+\infty} u^{s-1} \cos (a u) \:e^{-b u}\:\mathrm{d}u = \Gamma (s)\frac{\cos \left(\! s \arctan \left(\frac{a}{b}\right)\right)}{(a^2+b^2)^{s/2}}, \, \left(\Re(s)>0, b>0, a>0 \right) \tag8 \\ & \int_{0}^{\!\Large \frac{\pi}{2}} \cos^u\! x \cos (u x)\:\mathrm{d}x = \frac{\pi}{2^{u+1}}, \quad u>-1. \tag9 \end{align} $$ We may extend identity $(7)$ by analytic continuation to obtain $(1)$.

Example 1. We have

$$ \int_{0}^{\pi/2}\displaystyle \sqrt{\sqrt{x^2 + \ln^2\!\cos x}-\ln\! \cos x}\,\,\mathrm{d}x = \frac{\pi}{2}\sqrt{2\ln 2} \tag{10} $$

and

$$ \int_{0}^{\pi/2}\displaystyle \frac{1}{\sqrt{x^2 + \ln^2\!\cos x}} \sqrt{\sqrt{x^2 + \ln^2\!\cos x}-\ln\! \cos x}\,\,\mathrm{d}x = \frac{\pi}{\sqrt{2\ln 2}} \tag{11} $$

Proof. Let $0<x<\frac{\pi}{2}$ and set $t:=\arctan \left(-\frac{x}{\ln \cos x}\right)$. Observe that $0 < t < \frac{\pi}{2}$ and $$ \cos t=\cos \left(\!\arctan \left(-\frac{x}{\ln \cos x}\right)\right)= -\frac{\ln \cos x}{\sqrt{x^2 + \ln^2\!\cos x}}, $$ $$ \cos \left(\frac{t}{2}\right) = \sqrt{ \frac{1}{2}+ \frac{1}{2} \cos t} $$ then put successively $\displaystyle s=-\frac{1}{2}$, $\displaystyle s=\frac{1}{2}$ in $(1)$ to obtain $(10)$ and $(11)$.

Example 2.

$$ \frac{2}{\pi}\int_{0}^{\pi/2}\sqrt[\normalsize{8}]{x^2 + \ln^2\!\cos x} \sqrt{ \frac{1}{2}+\frac{1}{2}\sqrt{ \frac{1}{2}+ \frac{1}{2} \sqrt{ \frac{\ln^{2}\!\cos x}{ x^2 + \ln^2\! \cos x}}}}\,\mathrm{d}x = \sqrt[\normalsize{4}]{\ln 2} \tag{12} $$

and

$$ \frac{2}{\pi}\int_{0}^{\pi/2}\frac{1}{\sqrt[\normalsize{8}]{x^2 + \ln^2\!\cos x}} \sqrt{ \frac{1}{2}+\frac{1}{2}\sqrt{ \frac{1}{2}+ \frac{1}{2} \sqrt{ \frac{\ln^{2}\!\cos x}{ x^2 + \ln^2\! \cos x}}}}\,\mathrm{d}x=\frac{1}{\sqrt[\normalsize{4}]{\ln 2}} \tag{13} $$

Proof. Let $0<x<\frac{\pi}{2}$ and set $t:=\arctan \left(-\frac{x}{\ln \cos x}\right)$. Observe that $0 < t < \frac{\pi}{2}$ and $$ \cos t=\cos \left(\!\arctan \left(-\frac{x}{\ln \cos x}\right)\right)=\sqrt{ \frac{\ln^{2}\!\cos x}{x^2 + \ln^2\! \cos x}}, $$ $$ \cos \left(\frac{t}{4}\right) = \sqrt{\frac{1}{2}+\frac{1}{2}\sqrt{ \frac{1}{2}+ \frac{1}{2} \cos t}}, $$ then put successively $\displaystyle s=-\frac{1}{4}$, $\displaystyle s=\frac{1}{4}$ in $(1)$ to obtain $(12)$ and $(13)$.

Example n.

Set

$$ R_n^{+}:=\frac{2}{\pi}\int_{0}^{\pi/2}\sqrt[\Large {2^n}]{x^2 + \ln^2\!\cos x} \sqrt{\frac{1}{2}+\frac{1}{2}\sqrt{\frac{1}{2}+\cdots+\frac{1}{2}\sqrt{ \frac{1}{2}+ \frac{1}{2}\sqrt{ \frac{\ln^{2}\!\cos x}{ x^2 + \ln^2\! \cos x}}}}}\,\mathrm{d}x $$

and

$$ R_n^{-}:=\frac{2}{\pi}\int_{0}^{\pi/2}\frac{1}{\sqrt[\Large {2^n}]{x^2 + \ln^2\!\cos x}} \sqrt{\frac{1}{2}+\frac{1}{2}\sqrt{\frac{1}{2}+\cdots+\frac{1}{2}\sqrt{ \frac{1}{2}+ \frac{1}{2}\sqrt{ \frac{\ln^{2}\!\cos x}{ x^2 + \ln^2\! \cos x}}}}}\,\mathrm{d}x. $$

Then

$$ R_n^{+}= \sqrt[\Large {2^n}]{\ln 2} \tag{14} $$

and

$$ R_n^{-}= \frac{1}{\sqrt[\Large {2^n}]{\ln 2}} \tag{15}. $$

Proof. Let $0<x<\frac{\pi}{2}$ and set $t:=\arctan \left(-\frac{x}{\ln \cos x}\right)$. Observe that $0 < t < \frac{\pi}{2}$ and $$ \cos t=\sqrt{ \frac{\ln^{2}\!\cos x}{x^2 + \ln^2\! \cos x}}, $$ $$ \cos \left(\frac{t}{2^n}\right) = \sqrt{\frac{1}{2}+\frac{1}{2}\sqrt{\frac{1}{2}+\frac{1}{2}\sqrt{\frac{1}{2}+\cdots+\frac{1}{2}\sqrt{ \frac{1}{2}+ \frac{1}{2} \cos t}}}}, $$ then put successively $\displaystyle s=-\frac{1}{2^n}$, $\displaystyle s=\frac{1}{2^n}$, $n\geq 1,$ in $(1)$ to obtain $(14)$ and $(15)$.

Venus
  • 10,041
Olivier Oloa
  • 122,789
  • Very brilliant approach & great answer, but how to prove equation (8) & (9)? The cited links don't contain the proofs. +1 anyway – Anastasiya-Romanova 秀 Sep 24 '14 at 09:19
  • 8
    Much more professional than my answer! I knew you were holding out on us. – SDiv Sep 24 '14 at 17:55
  • @SDiv I really congratulate you for having found a route! Maybe it would be interesting to have your real name to cite your answer in an eventual paper on the subject. Thanks. – Olivier Oloa Sep 24 '14 at 20:16
  • 2
    OK fine, I am able to prove equation (8) & (9) by myself – Anastasiya-Romanova 秀 Sep 25 '14 at 09:53
  • Mr. Oloa, I wanna ask something for the Theorem (1). When I plug in $s=1$, the numeric value of the integral is $0.695384$ but $\dfrac{\pi}{2\ln2}\approx2.26618$. Why $RHS\neq LHS$? – Anastasiya-Romanova 秀 Sep 25 '14 at 13:19
  • For $s=1$, $(1)$ evaluates to $\frac{\pi}{2 \ln 2} - \frac{\pi}{2}$ http://math.stackexchange.com/questions/325651/definite-integral-int-0-pi-2-frac-log-cos-xx2-log2-cos-xdx-f My guess is that the problem arises when you switch the order of integration. – Random Variable Sep 25 '14 at 15:05
  • @RandomVariable The OP that you cite, is that wrong since it's evaluated as $\dfrac{\pi}{2}\left(1-\dfrac{1}{\log 2}\right)$? – Anastasiya-Romanova 秀 Sep 25 '14 at 16:16
  • @Anastasiya-Romanova $(1)$ evaluated at $s=1$ is the negative of that integral. – Random Variable Sep 25 '14 at 16:21
  • @RandomVariable How come it's the negative of the other integral? Plug in $s=1$ to (1), the integral is $$\int_0^{\pi/2} \frac{\log(\cos x)}{x^2+\log^2(\cos x)}dx$$Anyway, plug in $s=2$ to (1), according to W|A the integral doesn't converge – Anastasiya-Romanova 秀 Sep 25 '14 at 16:26
  • 1
    @Anastasiya-Romanova Because the positive square root of $\log^{2} (\cos x)$ for $0 \le x \le \frac{\pi}{2}$ is $-\log \cos x$. Perhaps $s$ needs to be further restricted. – Random Variable Sep 25 '14 at 16:47
  • @RandomVariable Hhmmm, That makes sense – Anastasiya-Romanova 秀 Sep 25 '14 at 17:09
  • @Anastasiya-Romanova Identity $(1)$ is true for $-1<s<1,$ further extension need some care. I've fixed it. Thanks! – Olivier Oloa Sep 25 '14 at 17:43
  • @OlivierOloa Hhmmm, I deserve a credit for my comment because it's helpful, lol. (kidding) Anyway, could you please extend your theorem for $s\ge1$? Especially for $s=1$, I'm curious. Thanks... ≥Ö‿Ö≤ – Anastasiya-Romanova 秀 Sep 25 '14 at 18:23
  • 1
    Wow, this is very impressive, but the only thing I haven't understood yet is how did you get the equality $(2)$, may I have miss something in your reasonning or it is obvious? :) – ParaH2 Oct 29 '15 at 20:07
  • this is sick (+1)! – tired Jan 31 '16 at 00:53
  • I’m confused as to how you knew to start from the integral$${\Large\int_{0}^{!\Large \frac{\pi}{2}}} \frac{\cos \left(! s \arctan \left(-\frac{x}{\ln \cos x}\right)\right)}{(x^2+\ln^2! \cos x)^{\Large\frac{s}{2}}}, \mathrm{d}x = \frac{\pi}{2}\frac{1}{\ln^{\Large s}!2}$$What lend you to start off with it and how did you know it would have a nice closed form? – Frank W Jul 11 '18 at 23:33
18

Oloa was right, I was on the right track last night (it makes me wonder if he has already solved it and just didn't tell us?). The final trick is to realize $(ix-\ln(\cos x))^{1/4}=\ln(1+i\tan(x))^{1/4}$ and then make the substitution $u=1+i\tan x$. This gives $R^+=\frac{2}{\pi}\text{Re}\lbrace\frac{i}{2}\Gamma(5/4) \text{PolyLog}(5/4, 2)\rbrace=(\ln 2)^{1/4}$. Details to follow in an edit in the next few minutes.

Edit: details follow So, this may not be the best way but at least it's a way: To get to the compact form I posted in the above comment, let $\theta(x)$ be such that $$\cos(\theta(x))=\frac{-\ln(\cos x)}{\sqrt{x^2+\ln^2(\cos x)}}.$$ Considering the right triangle with legs $x$ and $-\ln(\cos x)$ we notice that the hypotenuse is given by $h(x)=\sqrt{x^2+\ln^2(\cos x)}$ and further that $\sin(\theta(x))=\frac{x}{h(x)}$. This gives, using the correct identity $\sqrt{1/2+1/2\cos x}=\cos(x/2)$: $$R^+:=\frac{2}{\pi}\int_0^{\pi/2}\left(\frac{x}{\sin(\theta[x])}\right )^{1/4}\cos\left (\frac{\theta[x]}{4}\right )dx\\R^-:=\frac{2}{\pi}\int_0^{\pi/2}\left(\frac{x}{\sin(\theta[x])}\right )^{-1/4}\cos\left (\frac{\theta[x]}{4}\right )dx. $$

The trick is now to turn that $\theta/4$ into some kind of 4th root. We use the exponential definition of cosine: $\cos(\theta/4)=\frac{1}{2}\left ( e^{i\theta/4}+e^{-i\theta/4} \right )=\frac{1}{2}\left ( \left [\cos\theta+i\sin\theta \right ]^{1/4}+\left [\cos\theta-i\sin\theta \right ]^{1/4} \right )$. Then we have $$\left(\frac{x}{\sin(\theta[x])}\right )^{1/4}\cos\left (\frac{\theta[x]}{4}\right )=\frac{1}{2}\left ( [x\cot\theta+ix]^{1/4}+[x\cot\theta-ix]^{1/4} \right ).$$ Considering the triangle again this gives $$\left(\frac{x}{\sin(\theta[x])}\right )^{1/4}\cos\left (\frac{\theta[x]}{4}\right )=\frac{1}{2}\left ( [-\ln(\cos x)+ix]^{1/4}+[-\ln(\cos x)-ix]^{1/4} \right )\\=\text{Re}\left \lbrace [-\ln(\cos x)+ix]^{1/4} \right \rbrace. $$ Similar methods show $$\left(\frac{x}{\sin(\theta[x])}\right )^{-1/4}\cos\left (\frac{\theta[x]}{4}\right )=\text{Re}\left \lbrace [-\ln(\cos x)-ix]^{-1/4} \right \rbrace.$$ Now we use the addition property of logarithms and the exponental form of cosine again:$$-\ln(\cos x)+ix=-\ln(e^{ix}/2+e^{-ix}/2)+\ln(e^{ix})=\ln\left ( \frac{2e^{ix}}{e^{ix}+e^{-ix}} \right ) \\=\ln(1+i\tan x).$$ Similarly, $-\ln(\cos x)-ix=\ln(1-i\tan x).$ Okay, putting this together gives, using the fact that the integral over a real part is the real part of an integral: $$R^+=\frac{2}{\pi}\text{Re}\left \lbrace \int_0^{\pi/2}\ln(1+i\tan x)^{1/4}dx \right \rbrace \\ R^-=\frac{2}{\pi}\text{Re}\left \lbrace \int_0^{\pi/2}\ln(1-i\tan x)^{-1/4}dx \right \rbrace.$$ It seems that Mathematica still couldn't solve this, but with a simple substitution it works. For $R^+$ let $u=1+i\tan x$. Then $$ R^+=\frac{2}{\pi}\text{Re}\left \lbrace i\int_1^{1+i\infty}\frac{\ln(u)^{1/4}}{u^2-2u}du \right \rbrace\\=\frac{2}{\pi}\text{Re}\lbrace\frac{i}{2}\Gamma(5/4) \text{PolyLog}(5/4, 2)\rbrace\\=(\ln 2)^{1/4}.$$ I assume someone skilled in contour integration can do whatever Mathematica did to get this answer. For $R^-$ let $u=1-i\tan x$. Then $$ R^-=\frac{2}{\pi}\text{Re}\left \lbrace i\int_1^{1-i\infty}\frac{\ln(u)^{-1/4}}{u^2-2u}du \right \rbrace\\=\frac{2}{\pi}\text{Re}\lbrace\frac{i}{2}\Gamma(3/4) \text{PolyLog}(3/4, 2)\rbrace\\=(\ln 2)^{-1/4}.$$ I had to fudge a minus sign on this one, probably something with direction of contour integration or something. Anyway, conjecture confirmed. I assume similar methods can be used to confirm the nth order Ramanujan Log-Trigonometric Integral as mentioned by Oloa. Specifically, it seems we have $$R_n^+=\frac{2}{\pi}\text{Re}\left \lbrace \int_0^{\pi/2}\ln(1+i\tan x)^{1/(2^n)}dx \right \rbrace \\ R_n^-=\frac{2}{\pi}\text{Re}\left \lbrace \int_0^{\pi/2}\ln(1-i\tan x)^{-1/(2^n)}dx \right \rbrace.$$

Edit: It was silly for me to choose different representations for the real part. Better to chose the same one in which case we have: $$R_n^\pm=\frac{2}{\pi}\text{Re}\left \lbrace \int_0^{\pi/2}\ln(1+i\tan x)^{\pm 1/(2^n)}dx \right \rbrace.$$

SDiv
  • 2,610
2

The following is another way to prove Theorem $(1)$ in Olivier Oloa's answer; that is, to show $$I(s) = \int_{0}^{\pi/2}\frac{\cos \left( s \arctan \left(-\frac{x}{\ln (\cos x)}\right)\right)}{\left(x^2+\ln^2( \cos x)\right)^{s/2}} \, \mathrm d x = \frac{\pi}{2}\frac{1}{\ln^{ s}(2)}, \quad s <1.$$

Assume that $\ln$ is the principal branch of the logarithm.

First make the substitution $u = \tan (x)$ to get $$\begin{align} I(s)&=\int_{0}^{\infty} \frac{\cos \left(s \arctan \left(\frac{2 \arctan(u)}{\ln(1+u^{2})} \right) \right)}{\left(\arctan^{2}(u) + \frac{1}{4} \ln^{2}(1+u^{2}) \right)^{s/2}} \frac{du}{1+u^{2}} \\ &= \frac{1}{2} \, \Re \int_{-\infty}^{\infty} \frac{1}{\ln^{s}(1-iu)} \frac{1}{1+u^{2}} \, \mathrm du. \end{align}$$

For $-1<s<1$,$ s \ne 0$, the function $$f(z) = \frac{1}{\ln^{s}(1-iz)} \frac{1}{1+z^{2}}$$ has a branch cut on $[0, -i \infty)$ and a simple pole at $z=i$.

Integrate the function $f(z)$ around a contour consisting of the real axis (with a small clockwise-oriented semicircle about the origin) and a large semicircle in the upper half-plane.

The integral along the large semicircle vanishes as the semicircle's radius goes to infinity, and the contribution from the small semicircle about the origin vanishes in the limit since $s<1$.

So we have $$\int_{-\infty}^{\infty} \frac{1}{\ln^{s}(1-ix)} \frac{1}{1+x^{2}} \, \mathrm dx = 2 \pi i \operatorname*{Res}_{z=i} f(z) = \frac{\pi}{\ln^{s}(2)}, $$ from which it follows that $$I(s) = \frac{\pi}{2} \frac{1}{\ln^{s}(2)}. $$

The result obviously holds for $s=0$, and it also holds for $s \le -1$.

For $s=1$, we have $$I(1) = \frac{1}{2} \, \Re \operatorname{PV} \int_{-\infty}^{\infty} \frac{1}{\ln(1-ix)} \frac{1}{1+x^{2}} \, \mathrm dx = \frac{\pi}{2} \left(\frac{1}{\ln (2)}-1 \right). $$

And for $s>1$, $I(s)$ doesn't converge.