14

Consider the following integral

$$c=\int_0^{\pi/2}\log(1-x\cot x)\, \mathrm{d}x\approx-3.35333726288947201778500718670823032.$$

I suspect it can be analytically computed because by expanding the $\log$ function, $$ c=\sum_{k=1}^{\infty}\frac1k\int_0^{\pi/2}(x \cot x)^k\, \mathrm{d}x $$ it can be integrated term by term, albeit in involved form (combination of logarithms and $\zeta$-functions) $$ \int_0^{\pi/2}x \cot x\, \mathrm{d}x=\frac{1}{2} \pi \log2,\\ \int_0^{\pi/2}x^2 \cot^2 x\, \mathrm{d}x=-\frac{\pi ^3}{24}+\pi \log2,\\ \int_0^{\pi/2}x^3 \cot^3 x\,\mathrm{d}x=-\frac{\pi^3}{16} (1+2\log 2)+\frac{3 \pi}{16} (8\log 2+3 \zeta(3)), $$ and so on.

Just to provide some background to this question, the integral has some significance in theoretical physics. It enters the high-density asymptotics of the quasiparticle renormalization factor of the 3D homogeneous electron gas, see eq. 35 in Phys. Rev. B 70, 035111 (2004) or eqs. 8 and 9 in Phys. Rev. 120, 2041 (1960): $$ Z_{qp}=1+\frac{c}{\pi^2}\alpha r_s$$

However, since there is no parametric dependence, and since it is easy to compute numerically, no one cared to find the analytic form. However, I find it is a lovely little problem.

yarchik
  • 1,181
  • Why do you suspect this? Have you tried to check? – user Jan 19 '20 at 20:19
  • @user The integral has some significance in theoretical physics. However, since there is no parametric dependence, and since it is easy to compute numerically, no one cared to find the analytic form. The term-by-term integration can easily be done with e.g. Mathematica. The result is a combination of logarithms and $\zeta$-functions. – yarchik Jan 19 '20 at 20:34
  • I see no reason to think there is a closed form. – GEdgar Jan 19 '20 at 22:05
  • Since I have not much time left, I set in the log-function $a = 1$ and treat it for any $a$. Now, in general one has to use partial Integration to get rid of the log - function, I would suggest to go this way and use the Limit a->1 at the end. Another way would be differentiation under the integral sign. – stocha Jan 20 '20 at 18:43
  • @stocha I've tried the partial integration, but it did not bring me further. Differentiation is definitely worth trying. – yarchik Jan 20 '20 at 18:49
  • @yarchik: In one of my projects I deal with the laplace transform of a complex valued digamma function, in the calculation I meet some integrals with the form of yours using Feymann’s "trick", but also with an addional denominator. That' why I'm interesting in the solution, whenever one exists. – stocha Jan 20 '20 at 22:44
  • @yarchik: You can write your integral in the form: $$\int_0^{\frac{\pi }{2}} \left(\frac{x \cos (x)}{\sin (x)-x \cos (x)}-\frac{x^2 \csc (x)}{\sin (x)-x \cos (x)}\right) , dx$$, could you please write the expansion? – stocha Jan 21 '20 at 12:30
  • @yarchik: I know, if you integrate it seperately. I would like to look to your solution for the expansion, please provide us. – stocha Jan 21 '20 at 16:18
  • The complexity of the first coefficients leaves little hope, let alone to find the general term of the expansion. –  Jan 21 '20 at 18:36
  • @yarchik: I was able to make some progress, your integral is more general as Arathoon or stocha. One has to deal with it in the same way. – stocha Jan 22 '20 at 10:11
  • @stocha Sounds promising! In the solution of the Arathoon question one makes use of the integral $\int_0^{\pi/2} x \tan(x)^n dx$. There is a complementary formula for the integral $\int_0^{\pi/2} x^n \cot(x) dx$ in the Table Of Integrals, Series And Products of Gradshteyn and Ryzhik (2007) given by Eq. (3.748.2). – yarchik Jan 22 '20 at 10:48
  • @yarchik: Thanks for the note :-) – stocha Jan 22 '20 at 11:23
  • 1
    I don't know how useful this is, but you can reduce your integral to $I=\int_0^{\pi/2} \log(\tan x - x)$ and introducing 2 parameters $\alpha$ and $\beta$ in this fashion: $I=\int_0^{\pi/2} \log(\alpha \tan x - \beta x)$ and then differentiating you find $$\alpha \frac{\partial I}{\partial \alpha}+\beta \frac{\partial I}{\partial \beta}=\frac{\pi}{2}$$But the region of space to solve this PDE in does not look nice... – Gennaro Marco Devincenzis Jan 27 '20 at 18:41

2 Answers2

10

$\color{brown}{\textbf{Alternative expressions for the integral.}}$

Firstly, $$I = \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\ln(1-x\cot x)\,\mathrm dx = \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\ln(\sin x - x\cos x)\,\mathrm dx - \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\ln(\sin x)\,\mathrm dx = \dfrac\pi2\ln2 +I_1,$$

wherein $I_1$ allows integration by parts: $$I_1 = \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\ln(\sin x - x\cos x)\,\mathrm dx = x\ln(\sin x-x\cos x)\bigg|_{\ 0}^{\Large^\pi\hspace{-1pt}/_2} - \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^2\sin x}{\sin x - x\cos x}\,\mathrm dx,$$ $$ I_1 =-\int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^2\sin x}{\sin x- x\cos x}\,\mathrm dx = - \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^2}{1-x\cot x}\,\mathrm dx = - J_{21},\tag1$$ where

$$J_{mn} = \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^m}{(1-x\cot x)^n}\,\mathrm dx.\tag2$$

On the other hand, $$J_{21} = \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^2(1-x\cot x + x\cot x)}{1-x\cot x}\,\mathrm dx = \dfrac{\pi^3}{24} + I_2,$$ where $$I_2 = \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^3\cot x}{1 - x\cot x}\,\mathrm dx = \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^3}{\tan x - x}\,\mathrm dx.\tag3$$

Formulas $(3)$ are not suitable for numeric calculations.

But integration by parts is possible, $$I_2 = \dfrac14\int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{1}{\tan x - x}\,\mathrm dx^4 = \dfrac14\dfrac{x^4}{\tan x-x}\bigg|_{\,0}^{\Large^\pi\hspace{-1pt}/_2} + \dfrac14\int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^4(1+\tan^2x -1)}{(\tan x - x)^2}\,\mathrm dx,$$ $$ I_2 = \dfrac14\int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^4}{(1 - x\cot x)^2}\,\mathrm dx = \dfrac14 J_{42},$$

$$I = \dfrac\pi2\ln2 - \dfrac{\pi^3}{24} - \dfrac14 J_{42}.\tag4$$

Formula $(4)$ provides both suitable numeric calculations via Wolfram Alpha by the expression

Expression

with the result

Result

and the further building of the series in the elementary functions via the transformations in the form of $$ J_{42} = \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^4((1 - x\cot x)^2 + 2x\cot x(1 - x\cot x) + x^2\cot^2 x) }{(1 - x\cot x)^2}\,\mathrm dx\\ = \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\left(x^4 + 2\,\dfrac{x^5\cot x}{1-x\cot x} + \dfrac{x^6\cot^2x}{(1 - x\cot x)^2}\right)\,\mathrm dx\\ = \int\limits_0^{\Large^\pi\hspace{-1pt}/_2} x^4\,\mathrm dx + \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\left(\dfrac{2x^5\cot x}{1-x\cot x} + \dfrac{x^6\cot^2x}{(1 - x\cot x)^2}\right)\,\mathrm dx,$$ $$J_{42} = \dfrac{\pi^5}{160} + I_3 + I_4,\tag5$$ where $$I_3 = 2\int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^5\cot x}{1-x\cot x} \,\mathrm dx = 2\int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^5}{\tan x - x}\,\mathrm dx = \dfrac13\int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{1}{\tan x - x}\,\mathrm dx^6\\ = \dfrac13\dfrac{x^6}{\tan x-x}\bigg|_{\,0}^{\Large^\pi\hspace{-1pt}/_2} + \dfrac13\int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^6(1+\tan^2x -1)}{(\tan x - x)^2}\,\mathrm dx = \dfrac13 J_{62},$$ $$I_4 = \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^6\cot^2x}{(1 - x\cot x)^2}\,\mathrm dx = \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^6}{(\tan x - x)^2} \,\mathrm dx = \dfrac17\int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{1}{(\tan x - x)^2} \,\mathrm dx^7\\ = \dfrac27\dfrac{x^7}{(\tan x-x)^3}\bigg|_{\,0}^{\Large^\pi\hspace{-1pt}/_2} + \dfrac27\int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^7(1+\tan^2x -1)}{(\tan x - x)^3}\,\mathrm dx = \dfrac27\int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^7\cot x}{(1 - x\cot x)^3}\,\mathrm dx\\ = \dfrac27\int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^6(1 - (1 - x\cot x))}{(1 - x\cot x)^3}\,\mathrm dx =\dfrac27(J_{63}-J_{62}),$$

Therefore,

$$I = \dfrac\pi2\ln2 - \dfrac{\pi^3}{24} - \dfrac{\pi^5}{640} - \dfrac1{84}J_{62} - \dfrac1{14}J_{63}.\tag6$$

Numeric calculations via Mathcad Alpha by the formula $(6)$

Formula (6)

leads to the same result, and this confirms the correctness of the approach.

$\color{brown}{\textbf{Recurrence relations.}}$

For the arbitrary $m,n$ $$ J_{mn} = \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\,(x\cot x + (1-x\cot x))^n \dfrac{x^m}{(1 - x\cot x)^n}\,\mathrm dx = \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\sum\limits_{k=0}^n\binom nk\dfrac{x^{m+k}\cot^k x}{(1 - x\cot x)^k}\,\mathrm dx = \dfrac{\pi^{m+1}}{(m+1)2^{m+1}} + \sum\limits_{k=1}^n\dfrac{\dbinom nk}{m+k+1} \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{\mathrm dx^{m+k+1}}{(\tan x - x)^k} = \dfrac{\pi^{m+1}}{(m+1)2^{m+1}}\\ + \sum\limits_{k=1}^n\dfrac{\dbinom nk}{m+k+1} \left(\dfrac{x^{m+k+1}}{(\tan x - x)^{k}}\bigg|_{\,0}^{\Large^\pi\hspace{-1pt}/_2} + k\int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^{m+k+1}(1+\tan^2x-1)}{(\tan x-x)^{k+1}}\,\mathrm dx\right)\\ = \dfrac{\pi^{m+1}}{(m+1)2^{m+1}} + \sum\limits_{k=1}^n\dfrac{k}{m+k+1} \dbinom nk \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^{m+2}(x\cot x)^{k-1}}{(1 -x\cot x)^{k+1}}\,\mathrm dx\\ = \dfrac{\pi^{m+1}}{(m+1)2^{m+1}} + \sum\limits_{k=1}^n\dfrac{k}{m+k+1} \dbinom nk \int\limits_0^{\Large^\pi\hspace{-1pt}/_2}\dfrac{x^{m+2}(1-(1-x\cot x))^{k-1}}{(1 -x\cot x)^{k+1}}\,\mathrm dx\\ = \dfrac{\pi^{m+1}}{(m+1)2^{m+1}} + \sum\limits_{k=1}^n\dfrac{k}{m+k+1} \dbinom nk \sum\limits_{j=0}^{k-1}(-1)^{k-1-j}\dbinom{k-1}j J_{m+2,\,j+2},$$

$$J_{mn} = \dfrac{\pi^{m+1}}{(m+1)2^{m+1}} + \sum\limits_{j=0}^{n-1} F_{j} J_{m+2,\,j+2},\tag7$$

where

$$F_{j} = \sum\limits_{k=j+1}^n (-1)^{k-1-j} \dfrac{k}{m+k+1}\dbinom nk \dbinom{k-1}j.\tag8 $$

If $(m,n)=(2,1),\ $ then $$F_{0} = \sum\limits_{k=1}^1 (-1)^{k-1} \dfrac{k}{2+k+1}\dbinom1k \dbinom{k-1}0 =\dfrac14,$$ $$J_{21} = \dfrac{\pi^{3}}{3\cdot2^3} + \sum\limits_{j=0}^0 F_{j} J_{4,\,j+2} = \dfrac{\pi^{3}}{24} + J_{42}.$$

If $(m,n)=(4,2),\ $ then $$F_{0} = \sum\limits_{k=1}^2 (-1)^{k-1} \dfrac{k}{4+k+1}\dbinom2k \dbinom{k-1}0 =\dfrac13 - \dfrac27 = \dfrac{1}{21},$$ $$F_{1} = \sum\limits_{k=2}^2 (-1)^{k} \dfrac{k}{4+k+1}\dbinom2k \dbinom{k-1}1 =\dfrac27,$$ $$J_{42} = \dfrac{\pi^{5}}{5\cdot2^5} + \sum\limits_{j=0}^1 F_{j} J_{2,\,j+2} = \dfrac{\pi^5}{160} + \dfrac1{21}J_{62} + \dfrac27J_{63}.$$

Similarly, $$J_{62} = \dfrac{\pi^7}{896}+\dfrac1{36}J_{82}+\dfrac29J_{83}\tag9$$ (see also Wolfram Alpha test).

J62 Test

Besides, $$J_{63} = \dfrac{\pi^7}{896}+\dfrac1{120}J_{82} + \dfrac1{15}J_{83} + \dfrac3{20}J_{84}.\tag{10}$$

$\color{brown}{\textbf{Simple series.}}$

Obtained results are not the best way to get required series of the arbitrary length.

$$\boxed{ \begin{matrix} I & = & -3.35333726288947201778500718670823032009876022464933939598 \\ \frac\pi2\ln2 & = &1.088793045151801065250344449118806973669291850184643147162 \\ J_{21} & = & 4.442130308041273083035351635930890531086461245854584994170 \\ \frac{\pi^3}{24} & = & 1.291928195012492507311513127795891466759387023578546153922 \\ J_{42} & = & 12.60080845211512230289535403253999625730829688910415536099 \\ \frac{\pi^5}{160} & = & 1.912623029908009082892133187771472540501879416425468690959 \\ J_{62} & = & 9.357325953756236734147158157553707227832359838953032605558 \\ J_{63} & = & 35.84909465209885681432007993043088180418373451454989791084 \\ \frac{\pi^7}{896} & = & 3.370862977429455432493534032446475258836420173320761453966 \\ J_{82} & = & 13.21743446830609099759197972403428192140938899336281280188 \\ J_{83} & = & 25.28690408493225448274231109747825862030555487117486858192 \\ J_{84} & = & 102.2743092725712233044348622015074565154951081384648503713 \\ \end{matrix}}$$

On the other hand, using of the simple Laurent series for the function $$g(y) = \dfrac{35}{1-y\sqrt{15}\cot y\sqrt{15}} = \dfrac7{y^2}-\sum\limits_{i=0}^\infty c_iy^{2i}\tag{11}$$

g(y), Laurent series

gives evidently convergent series $$J_{21} = \dfrac1{35}\int\limits_0^{\Large^\pi\hspace{-1pt}/_2} \left(7 - \sum\limits_{i=0}^\infty c_i\left(\dfrac{x^2}{15}\right)^{i+1}\right)\,\mathrm dx,$$

$$J_{21} = \dfrac32\pi - \dfrac3{14}\pi\sum\limits_{i=0}^\infty \dfrac{c_i}{2i+3}\left(\dfrac{\pi^2}{60}\right)^{i+1}\,\mathrm dx,\tag{12}$$

wherein the first $8$ terms provide the accuracy of $8$ decimal digits.

3

Not the answer but too long for a comment : One "ansatz" to find a reformulation of the integral is:

1.) Consider: $$\mathcal{I}\left( k \right) = \int (x \cot (x))^k \, dx$$ 2.) Set $s'(x)=\cot ^k(x)$ and $v(x)=x^k$

3.) Perform partial integration: $$\int \frac{\partial (s(x) v(x))}{\partial x} \, dx=\int v(x) s'(x) \ \, dx+\int s(x) v'(x) \, dx$$ to reduce the power of $v(x)=x^k$

4.) For $$\mathcal{J}\left( k \right) =\int \cot ^k(x) \, dx$$ use Tunk-Fey to reduce the k-power under the integral.

5.) Collect the resulting terms … $-\frac{\cot ^{n-1}(x)}{n-1}$ to an infinity sum.

6.) Use for the expression:

$$ \int (x^{p} (\sum_{k} ...+...\int(\cot (y))^p\, dy)) \, dx$$ the formula in the Table Of Integrals, Series And Products of Gradshteyn and Ryzhik (2007) given by Eq. (3.748.2).

Edit : I validated it numerically for the sum over k=1...2 and it worked!

stocha
  • 703
  • 1
  • 7
  • 18
  • 1
    Thanks for outlining the program to attack problem. So its seems the integral can be integrated term-by-term which makes it feasible to represent in terms of series. – yarchik Jan 24 '20 at 16:19