I would like to perform the integration ($u \in [0,1]$), \begin{equation} \int_{a=-1}^1 \int_{b=-\sqrt{1-a^2}}^{\sqrt{1-a^2}} u^2 \sqrt{-a^2 u^2-b^2+1} \tan ^{-1}\left(\frac{\left| a\right| }{\sqrt{-a^2-b^2+1}}\right) db da . \end{equation}

I also have a companion problem \begin{equation} \int_{a=-1}^1 \int_{b=-\sqrt{1-a^2}}^{\sqrt{1-a^2}} u \sqrt{-a^2-b^2+1} \tan ^{-1}\left(u \left| a\right| \sqrt{-\frac{1}{a^2 u^2+b^2-1}}\right) db da . \end{equation}

Solutions to these two problems would provide a resolution to the question "Compute a certain "separability probability" via a constrained 4D integration over $[-1,1]^4$", posed in https://mathematica.stackexchange.com/questions/193337/compute-a-certain-separability-probability-via-a-constrained-4d-integration-ov

Now, for the component functions of the first integrand, we have \begin{equation} \int_{a=-1}^1 \int_{b=-\sqrt{1-a^2}}^{\sqrt{1-a^2}} u^2 \sqrt{-a^2 u^2-b^2+1} db da = \frac{4}{3} u \left(\sqrt{1-u^2} u+\sin ^{-1}(u)\right), \end{equation} and \begin{equation} \int_{a=-1}^1 \int_{b=-\sqrt{1-a^2}}^{\sqrt{1-a^2}} \tan ^{-1}\left(\frac{\left| a\right| }{\sqrt{-a^2-b^2+1}}\right) db da=\frac{1}{2} (\pi -2) \pi. \end{equation} Can some form of integration-by-parts--or other methodology--be performed? (I realize that more may certainly be needed in addition to the two above results.)

The much-viewed question https://math.stackexchange.com/questions/1167346/integration-by-parts-for-a-double-integral seems relevant here.

  • $\begingroup$ To get rid of at least one integral, first change $a, b$ to polar coordinates $r, \theta$, and change again to $x, y$ with $r = x/\sqrt{1+x^2}, \cos(\theta) = y/x$ with bounds $x \in [0, \infty)$ and $y \in [0, x]$. Switching the integration order to $y \in [0, \infty)$, $x \in [y, \infty)$ then allows you to perform the $x$-integral, and $$\int_0^\infty \! \frac{8 u^2 \sqrt{1 + (1-u^2) \, y^2} \tan^{-1}(y)}{3 (1 + y^2)^2} \, dy$$ remains. Not sure how helpful that actually is though. $\endgroup$ – student Mar 18 '19 at 19:52

Further to my comment, one can set $y = \frac{z}{\sqrt{1-z^2}}$ to obtain $$\frac{8}{3} \, u^2 \int_0^1 \! \sqrt{1 - u^2 z^2} \sin^{-1}(z) \, dz \;,$$ which Mathematica evaluates to $$ \frac{2}{3} \, u \, \bigl(\operatorname{Li}_2(-u) - \operatorname{Li}_2(u) + \pi \sqrt{1 - u^2} \, u - (1 - u^2) \tanh^{-1}(u) - u + \pi \sin^{-1}(u)\bigr) \;, $$ where $\operatorname{Li}_2(x)$ is the dilogarithm.

A very similar derivation for your companion problem yields the final integral $$\frac{8}{3} \, u \int_0^1 \! \sqrt{1 - z^2} \sin^{-1}(u z) \, dz$$ which Mathematica evaluates to $$-\frac{2}{3u} \, \bigl(u^2 \, \operatorname{Li}_2(-u) - u^2 \, \operatorname{Li}_2(u) - (1 - u^2) \tanh^{-1}(u) + u\bigr) \;.$$

For completeness, we can also do the third integral (from the comments) in exactly the same way, yielding $$\frac{8}{\pi} \, u^2 \int_0^1 \! \sqrt{1 - u^2 z^2} \, dz$$ as the final integral, which evaluates to $$\frac{4}{\pi} \, u \, \bigl(\sqrt{1 - u^2} \, u + \sin^{-1}(u)\bigr) \;.$$

Bonus round: If $I_1$, $I_2$, $I_3$ are the three results above, and $$A = \frac{6}{\pi^2} \, (I_2 - I_1) + I_3 \;, \\ B = \frac{12 u^2 \left(u^6+9 u^4-9 u^2-12 \left(u^4+u^2\right) \log (u)-1\right)}{\left(u^2-1\right)^5} \;,$$ then we are also interested in the integral of $AB$ over $u \in [0,1]$. That seems quite impossible, but we are in luck: integrating twice by parts, differentiating each time the factor with dilogarithms, yields a simple integral in $u$ that can be evaluated. The result is $$\frac{256-168 \zeta (3)}{\pi ^2}-38+48 \log (2) \approx 0.747925 \;.$$

| cite | improve this answer | |
  • $\begingroup$ Great! Quick numerical integrations give me agreement with both of your results to eight decimal places. I will incorporate your findings in my related question mathematica.stackexchange.com/questions/193337/… $\endgroup$ – Paul B. Slater Mar 18 '19 at 22:47
  • $\begingroup$ For simplicity, I omitted certain numerical factors that appear in the subject matter context mathematica.stackexchange.com/questions/193337/…. In that context, we have to add together the two results of Student multiplying them by $6/\pi^2$ and add them to a companion third term (continued in next comment) $\endgroup$ – Paul B. Slater Mar 19 '19 at 16:50
  • $\begingroup$ we had already been able to calculate ourselves—that is, the integral of $\frac{3 u^2 \sqrt{-a^2 u^2-b^2+1}}{\pi }$—giving us $\frac{2 \left(\sqrt{(1-u^2) u^2}+\sin ^{-1}\left(u\right)\right)}{\pi }$. This sum of the three terms proves—rather surprisingly—to be precisely equal to the “two-rebit Lovas-Andai separability function” (eq. (2) in arxiv.org/abs/1701.01973). $\endgroup$ – Paul B. Slater Mar 19 '19 at 16:50
  • $\begingroup$ Thanks, student. What would be of major importance to the underlying "separability probability question" would be the integration over $u \in [0,1]$ of the product of each of your two functions with $\frac{12 u^2 \left(u^6+9 u^4-9 u^2-12 \left(u^4+u^2\right) \log (u)-1\right)}{\left(u^2-1\right)^5}$. The result of the first integration times $\frac{6}{\pi^2}$ appears (via numeric means) to equal $-\frac{41 C}{2}-69-\pi (-22+\pi -37 \log (2)+15 \log (3))$, where Catalan's constant is indicated. Similarly, the result of the second integration times $\frac{6}{\pi^2}$ appears $\endgroup$ – Paul B. Slater Mar 20 '19 at 3:12
  • $\begingroup$ (again via numeric means) to equal $\frac{1}{6} (-191 C-395+\pi (139-15 \pi +125 \log (2)+\log (27))) \approx 0.20672$. (I was steered to these two conjectures by wolframalpha.com.) The result of the second integration minus the result of the first integration plus $\frac{919}{5}-264 \log (2)$ should equal the sought "rebit-retrit separability probability" of $-\frac{34 C}{3}+\frac{5609}{30}-264 \log (2)+\frac{1}{6} \pi (7-9 \pi -97 \log (2)+93 \log (3)) \approx 0.747925$. My attempts to use Mathematica to perform these computations has so far been unsuccessful. $\endgroup$ – Paul B. Slater Mar 20 '19 at 3:25

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.