5

Hi I am trying to come up with a closed form expression for $$ \int_0^{\pi/2} \frac{x^{2p}}{1+\cos^2x}dx,\quad p\geq 0. $$ I am interested in this general case in terms of p. For small p, we can use complex variables trivially, but I am not sure what to do for the general case. This integral for $p=1,p=2$ has very nice results given by $$ \int_0^{\pi/2} \frac{x}{1+\cos^2 x}dx=\frac{ \pi^2+4Li_2(3-2\sqrt{2})-4Li_2(-3+2\sqrt{2}) }{8\sqrt{2}}, $$ $$ \int_0^{\pi/2} \frac{x^2}{1+\cos^2 x}dx=\frac{\pi^3+12\pi Li_2(3-2\sqrt{2})}{24\sqrt{2}}. $$ These integrals are standard in the russian literature.

Jeff Faraci
  • 9,878
  • The point in using $2p$ is that $p$ is integer, is it? – chubakueno Apr 09 '14 at 04:12
  • @chubakueno Yes p is an integer. Thanks. – Jeff Faraci Apr 09 '14 at 04:13
  • 1
    You said it was trivial, but Ron Gordon's answer for $p=1$ says otherwise. If it was, posting your answer there would be nice :) – chubakueno Apr 09 '14 at 04:31
  • Jeff, as one who actually evaluated the case $p=1$ using a contour integral in the complex plane, I would strongly disagree with the derivation being "trivial". Maybe you have something really simple and my derivation was unnecessarily complicated. If so, please point out something you know substantially simpler than the analysis I provided in the link in my answer below. – Ron Gordon Apr 09 '14 at 04:53
  • @chubakueno: I just saw that you typed precisely what I said well before I did, in fewer words. Thanks, and I didn't mean to step on your toes. – Ron Gordon Apr 09 '14 at 04:57
  • @chubakueno Have you ever read math literature before? I am not sure why the word "trivial" is annoying you. You will find this in any math paper you read, and also, for $p=1$, you pick a contour and turn the crank, then you are done. Trivial, yes. – Jeff Faraci Apr 09 '14 at 15:09
  • @RonGordon The answer for $p=1$ is trivial, just takes some calculation force. Your solution is the only way I know of solving the problem, by using residue methods (which is what I have posted in the original post). It is not the simplest integral but once we have the contour, it is straightforward & trivial. – Jeff Faraci Apr 09 '14 at 15:10
  • @Jeff: "...you pick a contour and turn the crank, then you are done. Trivial, yes." Could you point out a published example, preferably in a textbook, of a solution to this problem in a similar manner as I have done? Unfortunately, I do take the word "trivial" a bit personally, because if it is, then I spent a huge amount of time developing and posting the solution to this problem when I could have just put a reference in the comments. – Ron Gordon Apr 09 '14 at 15:58
  • @RonGordon You take the word "trivially" personally? Read a mathematics journal then, the word is everywhere! You spent a lot of time developing a solution, a lot of time doesn't mean the problem is difficult though. I still appreciate your solution, I don't know why you hate the word Trivial, this is common in all math literature – Jeff Faraci Apr 09 '14 at 16:00
  • @Jeff: (last volley on my part - we can agree to disagree, and I'll still like you). Yes. I make my living from words (I am a patent agent) so I take them seriously. "trivial" is a fine word and very appropriate in certain circumstances - it's all a matter of who is your audience, etc. But it's a word that must be used carefully: I'll bet in those journals (some of which I am familiar with) there are instances where the author should regret using that word. (Having spent 3 years as a journal editor myself, I know this all too well.) (cont'd) – Ron Gordon Apr 09 '14 at 16:06
  • 1
    The more trivial approach to that problem is to use the identity $$1 + 2 \sum_{k=1}^{\infty} a^{k} \cos(kx) = \frac{1-a^{2}}{1-2a \cos x +a^{2}} , \ \ |a| <1$$. – Random Variable Apr 09 '14 at 16:07
  • 1
    @RonGordon hahaha, I can tell! You are like a lawyer, I shall avoid all arguments with you from now on LOL. Thanks though for the solution, really. AND I agree, it depends on your audience. I put the word trivial because to me and other people like Random Variable, You, Felix Marin, I expect that contour integration is trivial once we can find a suitable $f(z)$. – Jeff Faraci Apr 09 '14 at 16:08
  • So in this case, it took me a long time to develop the solution from scratch. Now if I am an ignoramus, then so what, and someone would have pointed out long ago that X, Y, Z, A, B, and C did this in the time of Cauchy. But that didn't happen. Rather, I got upvoted and nobody told me of a different way. For months. I have yet to see a similar solution published, although I would not be surprised if it exists. And if so...doesn't make it trivial. – Ron Gordon Apr 09 '14 at 16:11
  • The mathematics is more important than the world "trivial". This is math, not philosophy. I am going to search for literature on this integral (for p=1). I will have to go through some of the old russian books I have. This integral for p=1 was given to me when I was a first year in analytical mechanics at Sofia University, however this was in the 1980's. The only solution I saw to it was using a contour method as you have done. – Jeff Faraci Apr 09 '14 at 16:15
  • @RonGordon what do you think of Random Variables trivial approach? – Jeff Faraci Apr 09 '14 at 16:16
  • @Jeff: actually, I think it is brilliant. He should be able to apply it to this problem in theory, although the higher order integrals may give him fits. Ultimately, though, they should lead to the polylogs we expect. – Ron Gordon Apr 09 '14 at 16:21
  • @RonGordon Great, thanks. – Jeff Faraci Apr 09 '14 at 16:24
  • @RandomVariable If you think you can use that method for this general case, please share. I will award a bounty if Ron cannot calculate $H_k$, or I can award you both +50. Thanks. – Jeff Faraci Apr 09 '14 at 16:26
  • @RonGordon No problem! – chubakueno Apr 10 '14 at 02:01
  • @Jeff It didn't annoy me, it intrigued me. Was that solution really trivial to someone new to this site? Or, is he just bluffing? We will see in the next comment... – chubakueno Apr 10 '14 at 02:10

1 Answers1

8

I did the case $p=1$ here. The generalization to higher $p$ may involve higher-order derivatives as follows:

$$\begin{align}K_p &= \int_0^{\pi/2} dx \frac{x^{2 p}}{1+\cos^2{x}} = \frac1{2^{4 p-1}} \int_{-\pi}^{\pi} dy \frac{y^{2 p}}{3+\cos{y}} \end{align}$$

So define, as before,

$$J(a) = \int_{-\pi}^{\pi} dy \frac{e^{i a y}}{3+\cos{y}} $$

Then

$$K_p = \frac{(-1)^p}{2^{4 p-1}} \left [\frac{d^{2 p}}{da^{2 p}} J(a) \right ]_{a=0}$$

In the above reference, I derive an expression for $J(a)$:

$$J(a) = \frac{\pi}{\sqrt{2}} \cos{\pi a} \: (3-2 \sqrt{2})^a - 4 \sin{\pi a}\: PV \int_0^1 dx \frac{x^a}{x^2-6 x+1}$$

where $PV$ denotes the Cauchy principal value. Thus, the above outlines a procedure for expressing the integral $K_p$ in terms of powers of $\log{(3-2 \sqrt{2})}$ and $\operatorname{Li}_{2 k}(3-2 \sqrt{2})$ for $k=1$ to $p$. I hope I can derive a general expression, but for the time being, this outline should suffice for arbitrary positive integer values of $p$.

ADDENDUM

I have found a formalism in which I can derive a more explicit expression for the $K_p$. Consider the following function:

$$f(a) = \frac{\pi}{\sqrt{2}} \cos{\pi a} \: (3-2 \sqrt{2})^a - 4 \sin{\pi a} \, x^a$$

We seek to take $p$ successive second derivatives of $f$ with respect to $a$. Note how both terms of $f$ are a trig function times a number $q$ raised to the $a$th power. It turns out that

$$\frac{d^2}{da^2} \left ( \begin{array} & \sin{\pi a} \, q^a \\ \cos{\pi a} \, q^a \end{array}\right ) = \left ( \begin{array} & \log^2{q}-\pi^2 & 2 \pi \log{q} \\ -2 \pi \log{q} & \log^2{q}-\pi^2 \end{array} \right ) \left ( \begin{array} & \sin{\pi a} \, q^a \\ \cos{\pi a} \, q^a \end{array}\right )$$

We can express this in terms of a rotation matrix:

$$\frac{d^2}{da^2} \left ( \begin{array} & \sin{\pi a} \, q^a \\ \cos{\pi a} \, q^a \end{array}\right ) = \left ( \log^2{q}+\pi^2\right )\left ( \begin{array} & \cos{\theta} & \sin{\theta} \\ -\sin{\theta} & \cos{\theta} \end{array} \right ) \left ( \begin{array} & \sin{\pi a} \, q^a \\ \cos{\pi a} \, q^a \end{array}\right )$$

where $\cos{\theta} = (\log^2{q}-\pi^2)/(\log^2{q}+\pi^2)$ and $\sin{\theta} = 2 \pi \log{q}/(\log^2{q}+\pi^2)$. We then have an explicit expression for the $2 p$th derivative of the terms of $f$:

$$\begin{align}\left [\frac{d^{2 p}}{da^{2 p}} \left ( \begin{array} & \sin{\pi a} \, q^a \\ \cos{\pi a} \, q^a \end{array}\right )\right ]_{a=0} &= \left ( \log^2{q}+\pi^2\right )^p\left ( \begin{array} & \cos{p \theta} & \sin{p \theta} \\ -\sin{p \theta} & \cos{p \theta} \end{array} \right ) \left ( \begin{array} & 0 \\ 1 \end{array}\right )\\ &= \left ( \log^2{q}+\pi^2\right )^p \left ( \begin{array} & \sin{p \theta} \\ \cos{p \theta} \end{array}\right ) \end{align}$$

The above elements are simply Chebyshev polynomials of the second and first kind, respectively, in $\cos{\theta}$. As an example, let's do the first few values of $p$:

$p=1$: $$\left [\frac{d^{2}}{da^{2}} f(a) \right ]_{a=0} = \frac{\pi}{\sqrt{2}} \left (\log^2{(3-2 \sqrt{2})}-\pi^2 \right ) - 8 \pi \log{x}$$

which agrees with the result in the linked solution.

$p=2$: $$\left [\frac{d^{4}}{da^{4}} f(a) \right ]_{a=0} = \frac{\pi}{\sqrt{2}} \left [ \left (\log^2{(3-2 \sqrt{2})}-\pi^2 \right )^2-4 \pi^2 \log^2{(3-2 \sqrt{2})}\right ] - 16 \pi \log{x} \left (\log^2{x}-\pi^2 \right )$$

$p=3$: $$\left [\frac{d^{6}}{da^{6}} f(a) \right ]_{a=0} = \frac{\pi}{\sqrt{2}} \left [\left (\log^2{(3-2 \sqrt{2})}-\pi^2 \right )^3-12 \pi^2 \log^2{(3-2 \sqrt{2})} \left (\log^2{(3-2 \sqrt{2})}-\pi^2 \right ) \right ] \\ - 4 \left [6 \pi (\log^2{x}-\pi^2)^2 \log{x} - 8 \pi^3 \log^3{x} \right ]$$

and so on. Once this is done, the $\log{x}$ terms need to thrown into the integral above, i.e., we need to compute

$$H_k = PV \int_0^1 dx \frac{\log^{2 k-1}{x}}{x^2-6 x+1} $$

for $k = 1,\ldots,p$. Each $H_k$ will be expressible in terms of a $\operatorname{Li}_{2 k}(3-2 \sqrt{2})$ plus some other terms. If I have more time, I will post an explicit expression for $H_k$ and finally get an explicit expression for the original integral. But for now, this represents a lot of progress.

ADDENDUM II

OK, it is in fact practical and relatively straightforward, albeit a huge mess, to evaluate the $H_k$. The basis for the evaluation lies in the expression of $H_k$ as

$$H_k = \frac1{4 \sqrt{2}} \left [\int_0^1 dx \frac{\log^{2 k-1}{x}}{x-(3+2 \sqrt{2})} - PV \int_0^1 dx \frac{\log^{2 k-1}{x}}{x-(3-2 \sqrt{2})} \right ]$$

The first integral is straightforward:

$$\int_0^1 dx \frac{\log^{2 k-1}{x}}{x-(3+2 \sqrt{2})} = (2k-1)! \operatorname{Li}_{2 k}(3-2 \sqrt{2})$$

The 2nd one, however, is somewhat challenging. Note that the principal value carries into this integral because there is a bona fide singularity, but recall that the principal value was not just are up to avoid this singularity. Rather, it was a consequence of the way we treated the pole on the branch cut.

Anyway, the second integral is given by

$$PV \int_0^1 dx \frac{\log^{2 k-1}{x}}{x-(3-2 \sqrt{2})} = - i \pi \log^{2 k-1}(3-2 \sqrt{2}) + (2 k-1)! \operatorname{Li}_{2 k}(3 + 2 \sqrt{2})$$

Now, it turns out that the we may actually determine a value for the polylog of a positive real number greater than $1$ by a general inversion relation:

$$\operatorname{Li}_{2 k}(3 - 2 \sqrt{2}) + \operatorname{Li}_{2 k}(3 + 2 \sqrt{2}) = (-1)^{k+1} \frac{(2 \pi)^{2 k}}{(2 k)!} B_{2 k} \left (1-i\frac1{2 \pi} \log{(3+2 \sqrt{2})} \right ) $$

where $B_{2 k}$ is the $(2 k)$th Bernoulli polynomial. The leads to a formulation for evaluating each $H_k$. Note that the imaginary pieces should cancel out. For example,

$$4 \sqrt{2} H_1 = 2 \operatorname{Li}_2(3-2 \sqrt{2}) - \frac{\pi^2}{3} + \frac12 \log^2{(3-2\sqrt{2})}$$

$$4 \sqrt{2} H_2 = 2 \operatorname{Li}_4(3-2 \sqrt{2}) - \frac{2 \pi^4}{15} - \pi^2 \log^2{(3-2\sqrt{2})} +\frac14 \log^4{(3-2 \sqrt{2})}$$

$$4 \sqrt{2} H_3 = 2 \operatorname{Li}_6(3-2 \sqrt{2}) - \frac{16 \pi^6}{63} - \frac{4 \pi^4}{3} \log^2{(3-2\sqrt{2})} - \frac{5 \pi^2}{3} \log^4{(3-2 \sqrt{2})}+\frac{1}{6} \log^6{(3-2 \sqrt{2})}$$

Ron Gordon
  • 138,521
  • this is a big step in the right direction Ron. Thanks a lot for this. I will wait to check it as the answer (only because I want to see if somebody will provide a complete solution). I hope this does not bother you. If you can provide an explicit expression for $H_k$, let me know, (I will attach a bounty). I am aware this problem is not as trivial, for the general case. – Jeff Faraci Apr 09 '14 at 15:12
  • @Jeff: you're welcome. Actually, I feel these steps are less difficult than those I took solving the initial ($p=1$) problem. Just a matter of bookkeeping. Nevertheless, we will need $H_k$, yes. – Ron Gordon Apr 09 '14 at 16:01
  • Yes these steps are less difficult, however I still did not see them! If you do spend lots of time on $H_k$, just let me know before you post a solution, and I will bounty the problem. Thanks again – Jeff Faraci Apr 09 '14 at 16:03
  • Thanks for the update to the post in Addendum II. I did not get notified of it, and just realized now having going through this integral. Thanks again. As I said in a previous comment if you work out $H_k$, I will put this question up as a bounty. I have put the question on Bounty, and will award you +50 after 23 hours passes and they allow me to. – Jeff Faraci Apr 27 '14 at 20:51
  • +50 Thanks again – Jeff Faraci Apr 28 '14 at 22:12