6

I was trying to compute the value of the integral

$$\ I = \int_1^{\sqrt2} \frac{\operatorname{arctanh}(\sqrt{2-x^2})}{1+x}dx$$

I began by declaring the family of integrals:

$$\ I(a) = \int_1^{\sqrt2} \frac{\operatorname{arctanh}(a\sqrt{2-x^2})}{1+x}dx$$

And then differentiating under the integral sign in accordance to the Leibnitz rule, keeping in mind that:

$$\frac{d}{dx}\operatorname{arctanh}(x) = \frac{1}{1-x^2}$$

I obtained:

$$\ I'(a) = \int_1^{\sqrt2} \frac{\sqrt{2-x^2}dx}{(1-a^2(2-x^2))\cdot (1+x)}$$

Now I substitute $\ x = \sqrt2 sin\theta$. Changing the bounds, I received:

$$\ I'(a) = \int_{\frac{\pi}{4}}^{\frac{\pi}{2}} \frac{2\cos^2\theta d\theta}{(1-2a^2\cos^2\theta)\cdot (1+\sqrt2 \sin\theta)}$$

The problem with this integral is that - When the cosine term in the denominator varies from $\frac{1}{\sqrt2}$ to $0$, it will reach a point within the above range such that

$$\ 1 = 2a^2cos^2\theta$$

For all non zero $a>1$, it will have an asymptote. I do not understand how we will be able to then simplify the integral further. Maybe the radius of convergence of the integral necessitates that $\ a<1$?

If we assume that the integral converges, I think we can evaluate $I'(a)$ by multiplying and dividing by $\frac{-1}{a^2}$.

I am getting a not-so-good expression for $I'(a)$:

$$\ I'(a) = \frac{-1}{a^2}\cdot \big[ \frac{1}{2}\cdot \log(2) - \int_{\frac{\pi}{4}}^{\frac{\pi}{2}} \frac{d\theta}{(1-2a^2\cos^2\theta)(1+\sqrt2\sin\theta)}\big]$$

Note that for the special case $\ a = 1$, it is probably the case that

$$\ I(1) = \frac{\pi^2}{48}$$

This was deduced from WolframAlpha as a possible closed form for the integral. I have not confirmed whether this is the real answer or not, but it does seem to match.

Any help is appreciated. I am also open to other techniques for evaluating the original integral. Thanks for reading!

  • Is your primary interest here in solving the original integral, or this specifically a question about finding a way to make differentiating under the integral sign work? – David H Dec 24 '23 at 09:38
  • @DavidH Apologies, should have framed the question better. I want to find the value of the integral, first and foremost. And also I noticed that issue arises only for $a>1$, that has been corrected. –  Dec 24 '23 at 09:39
  • 1
    You will take limit when $a\to1^-$. So you may let $a<1$. – Bob Dobbs Dec 24 '23 at 10:40
  • @BobDobbs Even then, the integral isn't very nice. And we'll have to integrate it again, which sort of looks like a nightmare to me. –  Dec 24 '23 at 11:47
  • What is the closed form solution for $a = 1$ ? – mick Dec 24 '23 at 11:57
  • @mick It is $\frac{\pi^2}{48}$ if I'm not wrong. –  Dec 24 '23 at 11:59
  • @ScipioAemilianus yes that is correct. Maybe it helps to add such things in the OP ? – mick Dec 24 '23 at 12:07
  • 1
    @mick As you have suggested, I have added it for context. –  Dec 24 '23 at 12:12
  • 2
    The original integral isn't real if $a>1$ because then the integrand is complex at $x=1$. (OK, technically I should prove the integral's imaginary part doesn't cancel...) Anyway, all you want is $\lim_{a\to1^-}\int_0^aI'(b)\mathrm{d}b$. I think $u=-\cot2\theta$ may help. – J.G. Dec 24 '23 at 12:28
  • 1
    @ScipioAemilianus İf it is a nightmare, just run away. Why do you enter it?:) – Bob Dobbs Dec 24 '23 at 13:10
  • 1
    @BobDobbs Okay, I'll try it anyways and let you know what I get. –  Dec 24 '23 at 13:12

1 Answers1

2

Substitute $x=\sqrt2\,\sin y$, exploit symmetry, convert from hyperbolic to logarithmic, and substitute $y=2\arctan z$:

$$\begin{align*} I &= \int_1^{\sqrt2} \frac{\operatorname{artanh} \sqrt{2-x^2}}{1+x} \, dx \\ &= \sqrt2 \int_{\tfrac\pi4}^{\tfrac\pi2} \frac{\cos y \operatorname{artanh}\left(\sqrt2\,\cos y\right)}{1+\sqrt2\,\sin y} \, dy \\ &= \sqrt2 \int_0^{\tfrac\pi4} \frac{\cos \left(y+\frac\pi4\right) \operatorname{artanh}\left(\sqrt2\,\cos \left(y+\frac\pi4\right)\right)}{1+\sqrt2\,\sin \left(y+\frac\pi4\right)} \, d\left(y+\frac\pi4\right) \\ &= \int_0^{\tfrac\pi4} \frac{(\cos y-\sin y) \operatorname{artanh}\left(\cos y-\sin y\right)}{1+\cos y+\sin y} \, dy \tag{$*$} \\ &= \int_{\tfrac\pi2}^{\tfrac\pi4} \frac{\left(\cos\left(\frac\pi2-y\right)-\sin\left(\frac\pi2-y\right)\right) \operatorname{artanh}\left(\cos \left(\frac\pi2-y\right)-\sin \left(\frac\pi2-y\right)\right)}{1+\cos\left(\frac\pi2-y\right)+\sin \left(\frac\pi2-y\right)} \, d\left(\frac\pi2-y\right) \\ &= \int_{\tfrac\pi4}^{\tfrac\pi2} \frac{(\cos y-\sin y) \operatorname{artanh}\left(\cos y-\sin y\right)}{1+\cos y+\sin y} \, dy \tag{$**$} \\ &= \frac14 \int_0^{\tfrac\pi2} \frac{\cos y - \sin y}{1+\cos y+\sin y} \log \frac{1+\cos y-\sin y}{1-\cos y+\sin y} \, dy \tag*{$\frac{(*)+(**)}2$}\\ &= \frac14 \int_0^1 \frac{1-2z-z^2}{(1+z)\left(1+z^2\right)} \log \frac{1-z}{z(1+z)} \, dz \end{align*}$$

Expanding into partial fractions and breaking up the logarithms leads to integrals of the much simpler forms,

$$\int_0^1 \frac{L(z)}{1+z}\,dz \text{ and } \int_0^1 \frac{2z L(z)}{1+z^2} \, dz$$

where $L(z) \in \left\{\log z, \log(1\pm z)\right\}$. Closed forms are listed below. Linked Q&As were found with Approach Zero.

$$\begin{align*} \int_0^1 \frac{\log z}{1+z} \, dz &= -\frac{\pi^2}{12} \tag1 \\ \int_0^1 \frac{\log(1-z)}{1+z} \, dz &= \frac12 \log^22 - \frac{\pi^2}{12} \tag2 \\ \int_0^1 \frac{\log(1+z)}{1+z} \, dz &= \frac12 \log^22 \\[2ex] \int_0^1 \frac{2z \log z}{1+z^2} \, dz &= -\frac{\pi^2}{24} \tag3 \\ \int_0^1 \frac{2z \log(1-z)}{1+z^2} \, dz &= \frac14 \log^22 - \frac{5\pi^2}{48} \tag4 \\ \int_0^1 \frac{2z \log(1+z)}{1+z^2} \, dz &= \frac14 \log^22 + \frac{\pi^2}{48} \tag4 \end{align*}$$

$(1)$, $(2)$, $(3)$, $(4)$


On the other hand, as @Zacky pointed out, by replacing $z\mapsto\dfrac{1-z}{1+z}$ one can collapse $I$ into a multiple of integral $(1)$ :

$$\begin{align*} \int_0^1 \frac{\log \frac{1-z}{1+z}}{1+z} \, dz &= \int_0^1 \frac{\log z}{1+z} \, dz \\[2ex] \int_0^1 \frac{2z \log \frac{1-z}{1+z}}{1+z^2} \, dz &= \int_0^1 \frac{2\log z}{1+z} \, dz - \int_0^1 \frac{2z \log z}{1+z^2} \, dz \\ &= \int_0^1 \frac{2\log z}{1+z} \, dz - \int_0^1 \frac{\frac12 \log z^2}{1+z^2} \, dz^2 \\ &= \frac32 \int_0^1 \frac{\log z}{1+z} \, dz \end{align*}$$

user170231
  • 19,334
  • 2
    Very nice! I think it would be a great improvement if you could elaborate on the transition from the second to the third line, it's not immediately clear. Also, the last part is a bit of an overkill.. just substitute $x\to \frac{1-x}{1+x}$ in the fourth line and average them, then use partial fractions alongside $x^2\to x$ in one of the integrals to arrive at the $(1)$ integral. – Zacky Dec 27 '23 at 09:35
  • @user170231 Thanks for the answer. Can we evaluate an expression for $I(a)$? –  Dec 27 '23 at 10:00
  • 2
    @Zacky Thanks a lot for the feedback! I've added some more details to address your first point. And I really should have spotted your second point, being such a big fan of that substitution lately. – user170231 Dec 27 '23 at 17:38
  • 1
    @ScipioAemilianus Possibly; using these same initial steps on your $I(a)$, one arrives at$$I'(a) = \int_0^1 \frac{\left(z^2+2z-1\right)^2}{(1+z) \left(\left(z^2+1\right)^2 - a^2 \left(z^2+2z-1\right)^2\right)} , dz$$Expand to partial fractions (the denominator factorizes to linear times two quadratics) and integrate. Unfortunately that produces some complicated $\arctan$ expressions that make the $a$-integral a bit tricky to deal with. – user170231 Dec 27 '23 at 18:43