I've been tracking this post today on math.SE where the OP was asking for a proof of convergence for the integral

$$ \int_0^1 \frac{\ln\left(1-\alpha^2x^2 \right)}{\sqrt{1-x^2}}dx. $$

I want to ask a related but distinct question of how to explicitly compute the value of this integral. Something I had learned of today for the first time was the Feynman trick for evaluating integrals with some parameter $\alpha$ included in it (here is a helpful link). Essentially what this comes down to is interpreting the result of this integral as a function of $\alpha$: $$ f(\alpha) \;\; =\;\; \int_0^1 \frac{\ln\left(1-\alpha^2x^2 \right)}{\sqrt{1-x^2}}dx. $$

Feynman's approach to solving this was to essentially uncover a differential equation for $f$ by first differentiating this expression with respect to $\alpha$:

$$ \frac{df}{d\alpha} \;\; =\;\; \int_0^1\frac{-2\alpha x^2}{\sqrt{1-x^2} \left (1-\alpha^2x^2\right )}dx. $$

Next we fix an initial condition for this differential equation by picking a value for $\alpha$ that makes our original integral easy to compute. For instance, $\alpha=0$ yields: $$ f(0) \;\; =\;\; \int_0^1 \frac{\ln\left(1 \right)}{\sqrt{1-x^2}}dx \;\; =\;\; 0, $$

and then to proceed with finding $\frac{df}{d\alpha}$ via the integral expression determined above. Plugging this into Wolfram Alpha shows us that

$$ \int_0^1 \frac{\ln\left(1-\alpha^2x^2 \right)}{\sqrt{1-x^2}}dx \;\; =\;\; -\pi\ln \left (\frac{2}{1+ \sqrt{1-\alpha^2}} \right ) $$

but I would like some guidance on how to actually compute this integral. I had never heard of this method until today and am quite intrigued by how it works.

Note: I did a quick scan of all Feynman integral posts on here and I didn't see an example where the integrand included a square-root expression in the denominator. If I am mistaken I will gladly remove this post.


let $x=\sin t$,

$$ I(\alpha)=\int_0^1 \frac{\ln\left(1-\alpha^2x^2 \right)}{\sqrt{1-x^2}}dx =\int_0^{\pi/2}\ln(1-\alpha^2 \sin^2 t)dt $$

$$I'(\alpha) = \int_0^{\pi/2}\frac{-2\alpha\sin^2 t}{1-\alpha^2 \sin^2 t}dt =\frac2{\alpha}\int_0^{\pi/2}\left(1-\frac{1}{1-\alpha^2 \sin^2 t}\right)dt = \frac{\pi}{\alpha} - J $$


$$J= \frac{2}{\alpha} \int_0^{\pi/2}\frac{dt }{1-\alpha^2 \sin^2 t} = \frac{2}{\alpha} \int_0^{\pi/2}\frac{d(\tan t)}{1+(1-\alpha^2) \tan^2 t} =\frac{\pi}{\alpha\sqrt{1-\alpha^2}}$$


$$I(\alpha) = \int_0^{\alpha} I'(s)ds= \pi \int_0^{\alpha}\left( \frac1{s} -\frac{1}{s\sqrt{1-s^2}}\right)ds =\pi \ln\left(1+\sqrt{1-s^2}\right)_0^{\alpha}$$ $$=-\pi\ln\left(\frac2{1+\sqrt{1-\alpha^2}}\right)$$

  • $\begingroup$ Thanks for your answer. Other than the general steps of finding $\frac{df}{d\alpha}$, are there general principles by which you rearrange functions to compute the derivatives? The computations seem rather ad hoc otherwise. $\endgroup$ – Mnifldz Nov 22 '19 at 7:06
  • $\begingroup$ @Mnifldz - The steps presented in the answer is intended to illustrate the Feynman technique. It may seem ad hoc otherwise, but only because certain non-essential steps are skipped. A smooth presentation would have been quite lengthy. For example, the last integral $\int_0^{\alpha}\frac{1}{s\sqrt{1-s^2}}ds$ could be carried out quiet naturally with a few more steps. $\endgroup$ – Quanto Nov 22 '19 at 15:18
  • $\begingroup$ That's fine. I understand how all of the computations were made, no confusion there. I was just wondering, in addition to the technique itself, if there were underlying methods to carry out the integrals themselves. $\endgroup$ – Mnifldz Nov 22 '19 at 15:46

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.