Loading [MathJax]/jax/element/mml/optable/MathOperators.js

Friday, March 18, 2022

Integral of the day XXXI

Integral of a finite product

Integral involving a partial fraction expansion


Today we show the proof of this integral posted by @integralsbot 0k=1k2x2+k2dx=πn2(2n1) The proof relies on the partial fraction decomposition and some properties of the binomial coefficient.

Proof:

We need the following partial fraction decomposition (inspired by this this exercise ): nk=11x+k2=2nk=1(1)k1k2(x+k2)(nk)!(n+k)!

Here is the proof:

Since deg1<deg(x+1)(x+22)(x+n2) there exists a partial fraction decomposition: 1(x+1)(x+22)(x+n2)=nk=1Akx+k2 Multiplying by (x+1)(x+22)(x+n2) 1=nk=1Akni=1ik(x+i2) Suppose that x=m2 with 1mn Hence 1=Amni=1im(m2+i2)=Am(m2+1)(m2+22)+(m2+(m1)2)(m2+(m+1)2)(m2+n2)=Am((m21))((m222))((m2(m1)2))((m+1)2m2)(n2m2)=Am(1)m1(m1)(m+1)(m2)(m+2)(m(m1))(m+(m1))((m+1)+m)((m+1)m)(nm)(n+m)=Am(1)m1[(m1)(m2)212(nm)][(m+1)(m+2)(m+(m1))(m+(m+1))(n+m)]=Am(1)m1[(m1)!(nm)!][(m+1)(m+2)(m+(m1))(m+(m+1))(n+m)] Multiplying and dividing by (2m)m!(2m)m! 1=Am(1)m1[(m1)!(nm)!][m!(m+1)(m+2)(m+(m1))(2m)(m+(m+1))(n+m)](2m)m!=Am(1)m1[(m1)!(nm)!][(n+m)!](2m)m! Multiplying and dividing by m 1=Am(1)m1m!(nm)!(n+m)!(2m2)m!=Am(1)m1(nm)!(n+m)!2m2 Therefore Am=2(1)m1m2(nm)!(n+m)!1mn Hence nk=11x+k2=2nk=1(1)n1k2(x+k2)(nk)!(n+k)! Back to the integral: I=0k=1k2x2+k2dx=(nk=1k2)0k=11x2+k2dx=2n!20nk=1(1)n1k2(x2+k2)(nk)!(n+k)!dx=2n!2nk=1(1)n1k2(nk)!(n+k)!01x2+k2dx=2n!2nk=1(1)n1(nk)!(n+k)!01(xk)2+1dx=2n!2nk=1(1)n1k(nk)!(n+k)!01w2+1dx(wxk)=2n!2nk=1(1)n1k(nk)!(n+k)![arctan(w)]0=πn!2nk=1(1)n1k(nk)!(n+k)! where we have used the fact that nk=1k2=(nk=1k)(nk=1k)=n!2

Finally, we have to prove that πn!2nk=1(1)n1k(nk)!(n+k)!S=πn2(2n1) Note that S=1(2n)!nk=1(1)k1k(2n)!(nk)!(n+k)!=1(2n)!nk=1(2nn+k)(1)k1k=1(2n)!2nj=n+1(2nj)(1)nj1(jn)(j=n+k)=1(2n)!2nj=n+1(2n2nj)(1)nj1(jn)((vvw)=(vw))=1(2n)!m=0m=n1(2nm)(1)mn1(nm)(m=2nj)=(1)n1(2n)!n1m=0(2nm)(1)m(nm)=(1)n1(2n)![nn1m=0(2nm)(1)mn1m=0(2nm)(1)mm]=(1)n1(2n)![nn1m=0(2nm)(1)m2nn1m=0(2n1m1)(1)m]=(1)n1(2n)![2nn1m=0(2n1m)(1)mnn1m=0(2nm)(1)m]((v+1w)=(vw)+(vw1))=1(2n)![2n(2n2n1)n(2n1n1)](mj=0(1)j(vj)=(1)n(v1m))=1(2n)![12n1(2n)!(n1)!(n1)!12n(2n)!(n1)!(n1)!]=n2(2n1)n!n! Hence I=0k=1k2x2+k2dx=πn!2nk=1(1)k1k(nk)!(n+k)!=πn!2S=πn2(2n1) Therefore, we can conclude 0k=1k2x2+k2dx=πn2(2n1)

Tuesday, March 15, 2022

Integral of the day XXX

Integral evalauted with Fourier series

An integral involving an orthogonal system


Today we show the proof of this integral posted by @integralsbot π0dx(12acos(mx)+a2)(12bcos(nx)+b2)=π(1+anbm)(1a2)(1b2)(1anbm) m,nZ>0 with gcd(m,n)=1 and |a|,|b|1 The proof relies on Fourier series and the theory of orthogonal functions.

Proof:

We will use the following Fourier series (Proof in the appendix): 1+2k=1pkcos(kx)=1p212pcosx+p2|p|| Let m,nZ>0 with gcd(m,n)=1 and |a|,|b|1 Then π0dx(12acos(mx)+a2)(12bcos(nx)+b2)=1(1a2)(1b2)π01a2(12acos(mx)+a2)1b2(12bcos(mx)+b2)dx=1(1a2)(1b2)π0(1+2k=1akcos(kmx))(1+2j=1bjcos(jnx))dx=1(1a2)(1b2)[π0dx+2k=1akπ0cos(kmx)dx+2j=1bjπ0cos(jnx)dx+4k=1j=1akbjπ0cos(kmx)cos(jnx)dx] Recall that a system of real functions ϕ0(x),ϕ1(x),ϕ2(x),...,ϕn(x),... is said to be orthogonal on the interval [a,b] if baϕn(x)ϕm(x)dx=0nm,m=0,1,2,... We shall also assume that baϕ2n(x)dx0n=0,1,2,... Consider the system 1,cos(x),cos(2x),...,cos(nx),... This sytem is orthogonal on the interval [0,π]. In fact π0cos(nx)dx=[sin(nx)n]π0=0n=1,2,... which means that the functions cos(nx) and 1 are orthogonal. Moreover, we have π0cos(nx)cos(mx)dx=12π0[cos(n+m)x+cos(nm)x]dx=12π0cos(n+m)xdx+12π0cos(nm)xdx=0nm=0 Finally, note that π0cos2(nx)dx=12π01+cos(2nx)dx=π2 Hence, π0dx(12acos(mx)+a2)(12bcos(nx)+b2)=1(1a2)(1b2)[π0dx+2k=1akπ0cos(kmx)dx=0+2j=1bjπ0cos(jnx)dx=0+4k=1j=1akbjπ0cos(kmx)cos(jnx)dx]from (1)=1(1a2)(1b2)[π+4k=1j=1akbjπ0cos(kmx)cos(jnx)dxJ] From (2) note that J=π0cos(kmx)cos(jnx)dx={0 if kmjnπ2 if km=jn However, since gcd(m,n)=1 then n Therefore km = jn \Longleftrightarrow k = in \wedge j=im \quad i\in \mathbb{Z}\setminus\left\{0\right\} Hence if \displaystyle k = in \wedge j=im with i\in \mathbb{Z}_{>0} \textrm{ and } |a|,|b|\lt 1 \begin{align*} \sum_{k=1}^{\infty}\sum_{j=1}^{\infty}a^kb^j\int_{0}^{\pi} \cos(kmx)\cos(jnx) dx =& \sum_{i=1}^{\infty}a^{in}b^{im}\int_{0}^{\pi} \cos(inmx)\cos(imnx) dx\\ =& \sum_{i=1}^{\infty}a^{in}b^{im}\int_{0}^{\pi} \cos^2(inmx) dx\\ =& \frac{\pi}{2} \sum_{i=1}^{\infty}a^{in}b^{im} \quad \textrm{ from (3) } \\ =& \frac{\pi}{2} \sum_{i=1}^{\infty}\left(a^{n}b^{m}\right)^{i}\\ =& \frac{\pi}{2} \frac{a^nb^{m}}{(1-a^nb^{m})} \tag{4} \end{align*} Therefore \begin{align*} I =& \frac{1}{(1-a^2)(1-b^2)}\left[\pi + 4\sum_{k=1}^{\infty}\sum_{j=1}^{\infty}a^kb^j\int_{0}^{\pi} \cos(kmx)\cos(jnx) dx \right]\\ =& \frac{1}{(1-a^2)(1-b^2)}\left[\pi + \frac{2\pi a^nb^{m}}{(1-a^nb^{m})} \right] \quad \textrm{ from (4)} \\ =& \frac{\pi }{(1-a^2)(1-b^2)}\left[\frac{(1-a^nb^{m})}{(1-a^nb^{m})} + \frac{2 a^nb^{m}}{(1-a^nb^{m})} \right]\\ =& \frac{\pi (1+a^nb^{m})}{(1-a^2)(1-b^2)(1-a^nb^{m})}\\ \end{align*} Hence \boxed{ \displaystyle \int_{0}^{\pi} \frac{dx}{(1-2a\cos(mx)+a^2)(1-2b\cos(nx)+b^2)} =\frac{\pi (1+a^nb^{m})}{(1-a^2)(1-b^2)(1-a^nb^{m})} }

Appendix

1+2\sum_{k=1}^{\infty} p^k\cos(kx) = \frac{1-p^2}{1-2p\cos x + p^2} \quad |p|\lt 1 Proof \begin{align*} 1+2\sum_{k=1}^{\infty} p^k\cos(kx) =& 1+2\Re\left(\sum_{k=1}^{\infty} p^ke^{ixk}\right)\\ =& 1+2\Re\left(\frac{pe^{ix}}{1-pe^{ix}}\right)\\ =& 1+2\Re\left(\frac{p\cos x +ip\sin x }{1-p\cos x -ip\sin x}\right)\\ =& 1+2\Re\left(\frac{p\cos x +ip\sin x}{1-p\cos x -ip\sin x}\frac{1-p\cos x +ip\sin x}{1-p\cos x+ip\sin x}\right)\\ =& 1+2\Re\left(\frac{p\cos x -p^2 +i(p^2\cos x \sin x +p\sin x (1-p\cos x))}{(1-p\cos)^2+ p^2\sin^2 x}\right)\\ =& 1+\frac{2p\cos x-2p^2}{1-2p\cos x + p^2}\\ =& \frac{1-2p\cos + p^2}{1-2p\cos + p^2}+\frac{2p\cos x-2p^2}{1-2p\cos + p^2}\\ =& \frac{1-p^2}{1-2p\cos + p^2}\\ \end{align*}

Thursday, March 10, 2022

Integral of the day XXIX

Integral involving the arctanh function

Integral representation of the \arcsin(a) function


Today we show the proof of this integral posted by @integralsbot \int_{0}^{1} \ln\left(\frac{1+ax}{1-ax} \right)\frac{dx}{x\sqrt{1-x^2}} = \pi \arcsin(a) \quad |a|\leq 1 In the proof we use the residue theorem to evaluate it.

Proof:

Recall the logarithmic definition of \operatorname{arcanh}(x) \operatorname{arcanh}(x) = \frac{1}{2}\ln\left(\frac{1+x}{1-x} \right) \quad |x|\lt 1 Hence I(a) = \int_{0}^{1} \ln\left(\frac{1+ax}{1-ax} \right)\frac{dx}{x\sqrt{1-x^2}} = 2\int_{0}^{1} \frac{\operatorname{arcanh}(ax)}{x\sqrt{1-x^2}} dx Hence \begin{align*} I'(a) = 2\int_{0}^{1} \frac{1}{\sqrt{1-x^2}(1-a^2x^2)} dx = & 2\int_{0}^{\frac{\pi}{2}} \frac{1}{1-a^2\sin^2(w)} dw \quad (x \mapsto \sin(w)) \\ =& 4\int_{0}^{\frac{\pi}{2}} \frac{1}{a^2\cos(2w)-a^2+2} dw \\ =& 2\int_{0}^{\pi } \frac{1}{a^2\cos \theta -a^2+2} d\theta \quad (\theta \mapsto 2w)\\ =& \int_{-\pi}^{\pi } \frac{1}{a^2\cos \theta -a^2+2} d\theta \quad \textrm{(The integrand is even)} \end{align*} If we make the following change of variable \cos\theta = \frac{z+z^{-1}}{2} d\theta = \frac{dz}{zi} I'(a) = \int_{-\pi}^{\pi } \frac{1}{a^2\cos \theta -a^2+2} d\theta = \frac{2}{i}\oint_{|z|=1} \frac{dz}{a^2z^2+(4-2a^2)z+a^2} The integrand has two poles: z_{1} = \frac{a^2-2+2\sqrt{1-a^2}}{a^2} z_{2} = \frac{a^2-2-2\sqrt{1-a^2}}{a^2} Only z_{1} is inside the unit circle. Suppose |a|\leq 1: \begin{align*} \left| \frac{a^2-2+2\sqrt{1-a^2}}{a^2}\right|\lt 1 \Longrightarrow& -a^2\lt a^2-2+2\sqrt{1-a^2}\lt a^2 \\ \Longrightarrow& -2a^2\lt-2+2\sqrt{1-a^2}\lt0\\ \Longrightarrow& 1-a^2\lt \sqrt{1-a^2}\lt 1\\ \Longrightarrow& a\neq 0 \; \wedge \; a\neq 1 \end{align*} Then, if we add the assumption |a| \lt 1 \;\wedge\; a\neq 0 \Longrightarrow |z_{1}|\lt1

By the residue theorem: \begin{align*} I'(a) = \int_{-\pi}^{\pi } \frac{1}{a^2\cos \theta -a^2+2} d\theta =& \frac{2}{i}\oint_{|z|=1} \frac{dz}{a^2z^2+(4-2a^2)z+a^2}\\ =& \frac{4\pi}{a^2} \lim_{z \to \frac{a^2-2+2\sqrt{1-a^2}}{a^2}} \frac{1}{\left(z-\frac{a^2-2-2\sqrt{1-a^2}}{a^2}\right)}\\ =& \frac{4\pi}{\left(a^2-2+2\sqrt{1-a^2}-\left(a^2-2-2\sqrt{1-a^2}\right)\right)}\\ =& \frac{\pi}{\sqrt{1-a^2}} \end{align*} Therefore I(a) = \int_{0}^{1} \ln\left(\frac{1+ax}{1-ax} \right)\frac{dx}{x\sqrt{1-x^2}} = \pi \arcsin(a) + C If we let a \to 0 then C \to 0.

The case when a = 1 can be obtained as a limitng case easily.

Therefore \boxed{ \int_{0}^{1} \ln\left(\frac{1+ax}{1-ax} \right)\frac{dx}{x\sqrt{1-x^2}} = \pi \arcsin(a) \quad |a|\leq 1}

Tuesday, March 8, 2022

The trilogarithm function

The trilogarithm function

The strong connection between the \operatorname{Li}_{s} function and \phi vol. III


Today we show the proof of this integral posted by @integralsbot \int_{0}^{\ln \phi} \frac{x^2(e^{2x}+1)}{(e^{2x}-1)} dx = \frac{\zeta(3)}{10} The proof relies on some identities of the trilogarithm and dilogarithm function. We evaluated similar integrals and series here, here.

Proof:

\begin{align*} \int_{0}^{\ln \phi} \frac{x^2(e^{2x}+1)}{(e^{2x}-1)} dx =& \frac{1}{8}\int_{1}^{\phi^2} \frac{\ln^2(w)(w+1)}{w(w-1)}dw \quad (w\mapsto e^{2x})\\ =& \frac{1}{8} \int_{1}^{\phi^2} \frac{\ln^2(w)}{w-1+}dw + \frac{1}{8}\int_{1}^{\phi^2} \frac{\ln^2(w)}{w(w-1)}dw \\ =& \frac{1}{8} \int_{1}^{\phi^2} \frac{\ln^2(w)}{w-1}dw + \frac{1}{8} \int_{1}^{\phi^2} \frac{\ln^2(w)}{w-1}dw - \frac{1}{8}\int_{1}^{\phi^2} \frac{\ln^2(w)}{w}dw \\ =& \frac{1}{4} \underbrace{\int_{1}^{\phi^2} \frac{\ln^2(w)}{w-1}dw}_{J} - \frac{1}{8}\underbrace{\int_{1}^{\phi^2} \frac{\ln^2(w)}{w}dw}_{K} \end{align*} \begin{align*} K = \int_{1}^{\phi^2} \frac{\ln^2(w)}{w}dw =& \int_{0}^{\ln \phi^2 } s^2 ds \quad (s \mapsto \ln(w))\\ =& \left[\frac{s^3}{3}\right]_{0}^{\ln \phi^2}\\ =& \frac{\ln^3(\phi^2)}{3} \end{align*} \begin{align*} J = \int_{1}^{\phi^2} \frac{\ln^2(w)}{w-1}dw = & \int_{0}^{\phi} \frac{\ln^2(1+s)}{s} ds \quad (s \mapsto w-1, \phi^2-1 = \phi) \\ \stackrel{IBP}{=}& \ln(s)\ln^2(1+s)\Big|_{0}^{\phi} - 2\int_{0}^{\phi} \frac{\ln(s)\ln(1+s)}{1+s} ds\\ =& \ln(\phi)\ln^2(1+\phi) - 2\int_{1}^{\phi^2} \frac{\ln(t-1)\ln(t)}{t}dt \quad (t \mapsto s+1 )\\ =& \ln(\phi)\ln^2(1+\phi) - 2\int_{1}^{\phi^2} \frac{\ln\left(t\left(1-\frac{1}{t}\right)\right)\ln(t)}{t}dt\\ =& \ln(\phi)\ln^2(1+\phi) - 2\int_{1}^{\phi^2} \frac{\ln^2(t)}{t}dt - 2\int_{1}^{\phi^2} \frac{\ln\left(1-\frac{1}{t}\right)\ln(t)}{t}dt\\ =& \ln(\phi)\ln^2(1+\phi) - \frac{2}{3}\ln^3(\phi^2) -2\int_{1}^{\phi^{-2}} \frac{\ln\left(1-y\right)\ln(y)}{y}dy\\ \stackrel{IBP}{=}& \ln(\phi)\ln^2(1+\phi^2) - \frac{2}{3}\ln^3(\phi^2) +2\operatorname{Li}_{2}(y)\ln(y)\Big|_{1}^{\phi^{-2}}-2\int_{1}^{\phi^{-2}} \frac{\operatorname{Li}_{2}(y)}{y}dy \\ =& \ln(\phi)\ln^2(1+\phi) - \frac{2}{3}\ln^3(\phi^2) +2\operatorname{Li}_{2}(\phi^{-2})\ln(\phi^{-2})-2\int_{1}^{0} \frac{\operatorname{Li}_{2}(y)}{y}dy -2\int_{0}^{\phi^{-2}} \frac{\operatorname{Li}_{2}(y)}{y}dy \\ =& \ln(\phi)\ln^2(1+\phi) - \frac{2}{3}\ln^3(\phi^2) +2\operatorname{Li}_{2}(\phi^{-2})\ln(\phi^{-2})+2\int_{0}^{1} \frac{\operatorname{Li}_{2}(y)}{y}dy -2\int_{0}^{\phi^{-2}} \frac{\operatorname{Li}_{2}(y)}{y}dy \\ =& \ln(\phi)\ln^2(1+\phi) - \frac{2}{3}\ln^3(\phi^2) +2\operatorname{Li}_{2}(\phi^{-2})\ln(\phi^{-2})+2\operatorname{Li}_{3}(1) -2\operatorname{Li}_{3}(\phi^{-2}) \\ \end{align*} Note that the polylogarithm function is defined by a power series in z: \operatorname{Li}_{s} (z) = \sum_{k=1}^{\infty} \frac{z^k}{k^s} If we put s=3 and z=1 : \operatorname{Li}_{3} (1) = \sum_{k=1}^{\infty} \frac{1}{k^3} = \zeta(3) Hence J = \ln(\phi)\ln^2(1+\phi) - \frac{2}{3}\ln^3(\phi^2) +2\operatorname{Li}_{2}(\phi^{-2})\ln(\phi^{-2})+2\zeta(3) -2\operatorname{Li}_{3}(\phi^{-2}) Landen proved the following identity: \operatorname{Li}_{3}(\phi^{-2}) = \frac{4}{5}\zeta(3) + \frac{\pi^2}{15} \ln(\phi^{-2})- \frac{1}{12} \ln^3(\phi^{-2}) while Euler proved \operatorname{Li}_{2} (\phi^{-2}) = \frac{\pi^2}{15} -\ln^2(\phi) Therefore J = \ln(\phi)\ln^2(1+\phi) - \frac{2}{3}\ln^3(\phi^2) - 2\ln^2(\phi)\ln(\phi^{-2}) + \frac{2}{5} \zeta(3) + \frac{\ln^3(\phi^{-2})}{6} Additionally, from the identities: \ln(\phi)\ln^2(\phi+1) = 4\ln^3(\phi) \ln^2(\phi)\ln(\phi^{-2}) = -2\ln^3(\phi) \ln^3(\phi^{-2}) = -8\ln^3(\phi) we have J = \frac{4}{3}\ln^3(\phi)+ \frac{2}{5} \zeta(3) Hence I = \frac{1}{4}J -\frac{1}{8}K = \frac{1}{3}\ln^3(\phi)+ \frac{\zeta(3)}{10} - \frac{1}{3}\ln^3(\phi) Therefore, we can conclude \boxed{ \int_{0}^{\ln \phi} \frac{x^2(e^{2x}+1)}{(e^{2x}-1)} dx = \frac{\zeta(3)}{10}}

Thursday, March 3, 2022

Special case of the Gaussian function

Integral evaluated with the Guassian function


Today we show the proof of this integral posted by \int_{0}^{1} \frac{dx}{(x^2-x+1)\sqrt[3]{x(1-x)}} = \frac{4\pi}{3\sqrt{3}} The proof relies on a special case of the Gaussian function involving one of the roots of the unity, e^{\frac{i\pi}{3}}

Proof:

First note that \Im \left(\frac{4}{\sqrt{3}(1-i\sqrt{3}-2x)}\right) = \frac{1}{x^2-x+1 } Hence \begin{align*} I = \int_{0}^{1} \frac{dx}{(x^2-x+1)\sqrt[3]{x(1-x)}} = & \Im \left( \frac{4}{\sqrt{3}}\int_{0}^{1} \frac{dx}{(1-i\sqrt{3}-2x)\sqrt[3]{x(1-x)}}\right)\\ =& \Im \left( \frac{4}{\sqrt{3}(1-i\sqrt{3})}\int_{0}^{1} \frac{dx}{\left(1-\frac{2}{1-i\sqrt{3}}x\right)\sqrt[3]{x(1-x)}}\right)\\ \end{align*} Note that \frac{2}{1-i\sqrt{3}} = \frac{2}{1-i\sqrt{3}}\left(\frac{1+i\sqrt{3}}{1+i\sqrt{3}}\right) = \frac{1+i\sqrt{3}}{2} = e^{\frac{i\pi}{3}} Hence I = \Im \left( \frac{2}{\sqrt{3}}e^{\frac{i\pi}{3}} \underbrace{\int_{0}^{1} \frac{dx}{\left(1-e^{\frac{i\pi}{3}}x\right)\sqrt[3]{x(1-x)}}}_{J}\right) Recall the integral representation of the Gaussian function: F(a,b;c;z) = \frac{\Gamma(c)}{\Gamma(b)\Gamma(c-b)} \int_{0}^{1} x^{b-1}(1-x)^{c-b-1}(1-xz)^{-a} dx This is called the Pochhammer integral and converges whenever \quad \Re(c)>\Re(b)>0 \quad \arg\left(|1-z|\right)\lt \pi

Therefore if we put \displaystyle a=1, b= \frac{2}{3} and \displaystyle c=\frac{4}{3} we have \begin{align*} J = \int_{0}^{1}\left(1-e^{\frac{i\pi}{3}} x\right)^{-1} x^{-\frac{1}{3}}(1-x)^{-\frac{1}{3}} dx =& \frac{\Gamma^2\left(\frac{2}{3}\right)}{\Gamma\left(\frac{4}{3}\right)} F\left(1, \frac{2}{3};\frac{4}{3};e^{\frac{i\pi}{3}}\right) \end{align*} The Gaussian hypergeometric function satisfies the following property: F\left(3a,\tfrac{1}{3}+a;\tfrac{2}{3}+2a;{\mathrm{e}}^{\frac{\mathrm{i}\pi}{3% }}\right)=\sqrt{\pi}{\mathrm{e}}^{\frac{\mathrm{i}\pi a}{2}}\left(\frac{16}{2% 7}\right)^{(3a+1)/6}\frac{\Gamma\left(\frac{5}{6}+a\right)}{\Gamma\left(\frac{% 2}{3}+a\right)\Gamma\left(\frac{2}{3}\right)}, If we put a = \frac{1}{3} F\left(1,\tfrac{2}{3};\tfrac{4}{3};{\mathrm{e}}^{\frac{\mathrm{i}\pi}{3% }}\right)=\sqrt{\pi}{\mathrm{e}}^{\frac{\mathrm{i}\pi }{6}}\left(\frac{16}{2 7}\right)^{\frac{1}{3}}\frac{\Gamma\left(\frac{7}{6}\right)}{\Gamma\left(\frac{2}{3}\right)} = \sqrt{\pi}{\mathrm{e}}^{\frac{\mathrm{i}\pi }{6}}\frac{2\sqrt[3]{2}}{3}\frac{\Gamma\left(\frac{7}{6}\right)}{\Gamma\left(\frac{2}{3}\right)} Therefore \begin{align*} J = \int_{0}^{1}\left(1-e^{\frac{i\pi}{3}} x\right)^{-1} x^{-\frac{1}{3}}(1-x)^{-\frac{1}{3}} dx =& \sqrt{\pi}{\mathrm{e}}^{\frac{\mathrm{i}\pi }{6}}\frac{2\sqrt[3]{2}}{3}\frac{\Gamma\left(\frac{7}{6}\right)\Gamma\left(\frac{2}{3}\right)}{\Gamma\left(\frac{4}{3}\right)} \\ =& \pi{\mathrm{e}}^{\frac{\mathrm{i}\pi }{6}}\frac{2}{3} \end{align*} Hence \begin{align*} I = \int_{0}^{1} \frac{dx}{(x^2-x+1)\sqrt[3]{x(1-x)}} = & \Im \left( \frac{2}{\sqrt{3}}e^{\frac{i\pi}{3}} \int_{0}^{1} \frac{dx}{\left(1-e^{\frac{i\pi}{3}}x\right)\sqrt[3]{x(1-x)}}\right)\\ =& \Im \left(\pi{\mathrm{e}}^{\frac{\mathrm{i}\pi }{2}}\frac{4}{3\sqrt{3}}\right)\\ =& \Im \left(i\pi\frac{4}{3\sqrt{3}}\right)\\ =& \frac{4\pi}{3\sqrt{3}} \end{align*} Therefore, we can conclude \boxed{\int_{0}^{1} \frac{dx}{(x^2-x+1)\sqrt[3]{x(1-x)}} = \frac{4\pi}{3\sqrt{3}}}

Wednesday, March 2, 2022

Generalized hypergeometric functions IX

Another hypergeometric series

Series invoving a quotient of binomial coefficients


We show the proof of this nice result: \sum_{k=0}^{n} \frac{{\binom{n}{k}}^3}{\binom{3n}{k}} = \frac{(2n)!^3}{n!^3(3n)!} For the proof we use the Saalschütz theorem for hypergeometric functions.

Proof:

We will transform this series to an hypergeometric series

Note that \binom{v}{m} = \frac{(v-m+1)_{m}}{m!} where (x)_{m} = x(x+1)(x+2)\cdots (x+m-1) is the Pochhammer polynomial (rising factorial)

Hence S= \sum_{k=0}^{n} \frac{{\binom{n}{k}}^3}{\binom{3n}{k}} = \sum_{k=0}^{n} \frac{(n-k+1)_{k}(n-k+1)_{k}(n-k+1)_{k}}{(3n-k+1)_{k}k!k!} Recall that the rising factorial obey the reflection formula: (-x)_{k} = (x-k+1)_{k}(-1)^k \tag{*} Therefore \begin{align*} S =\sum_{k=0}^{n} \frac{{\binom{n}{k}}^3}{\binom{3n}{k}} =& \sum_{k=0}^{n} \frac{(n-k+1)_{k}(n-k+1)_{k}(n-k+1)_{k}}{(3n-k+1)_{k}k!k!}\\ = & \sum_{k=0}^{n} \frac{(-1)^k(n-k+1)_{k}(-1)^k(n-k+1)_{k}(-1)^k(n-k+1)_{k}}{(-1)^k(3n-k+1)_{k}k!k!}\\ = & \sum_{k=0}^{n} \frac{(-n)_{k}(-n)_{k}(-n)_{k}}{(-3n)_{k} k!k!}\\ = & \sum_{k=0}^{n} \frac{(-n)_{k}(-n)_{k}(-n)_{k}}{(-3n)_{k} (1)_{k}}\frac{1}{k!} \quad \left(k! = (1)_{k} \right) \end{align*} Recall the generalized hypergeometric function {}_{p}F_{q} \left[ {a_{1},\cdots a_{p}\atop b_{1}\cdots b_{q}}; z\right] = \sum_{k=0}^{\infty} \frac{(a_{1})_{k}(a_{2})_{k}\cdots (a_{p})_{k}}{(b_{1})_{k}(b_{2})_{k}\cdots (b_{q})_{k}} \frac{z^k}{k!} if one of the a_{i} is a negative integer, for example a_{i} = -n with n\in \mathbb{N} we say that the series "terminates" and {}_{p}F_{q} \left[ {a_{1},\cdots,-n,\cdots, a_{p}\atop b_{1}\cdots b_{q}}; z\right] = \sum_{k=0}^{n} \frac{(a_{1})_{k}(a_{2})_{k}\cdots(-n)_{k} \cdots (a_{p})_{k}}{(b_{1})_{k}(b_{2})_{k}\cdots (b_{q})_{k}} \frac{z^k}{k!} To show this note that (-n)_{k} = (-n)(-n+1)\cdots(-n+k-1) Hence (-n)_{k} = 0 \quad \textrm{ if } \quad k\geq n+1 From this is easy to see that S = \sum_{k=0}^{n} \frac{{\binom{n}{k}}^3}{\binom{3n}{k}} = \sum_{k=0}^{n} \frac{(-n)_{k}(-n)_{k}(-n)_{k}}{(-3n)_{k} (1)_{k}}\frac{1}{k!} = {}_{3}F_{2} \left[ {-n, -n, -n \atop -3n, 1}; 1 \right] This series satisfies the Saalschütz theorem: {{}_{3}F_{2}}\left[{-n,a,b\atop c,d};1\right]=\frac{{\left(c-a\right)_{n}}{% \left(c-b\right)_{n}}}{{\left(c\right)_{n}}{\left(c-a-b\right)_{n}}}, whenever \displaystyle c+d = a+b+1-n, \quad n=0,1,...

Therefore \begin{align*} S = {}_{3}F_{2} \left[ {-n, -n, -n \atop -3n, 1}; 1 \right] =& \frac{(-2n)_{n}(-2n)_{n}}{(-3n)_{n}(-n)_{n}}\\ =& \frac{(-1)^n(n+1)_{n}(-1)^n(n+1)_{n}}{(-1)^n(2n+1)_{n}(-1)^n(1)_{n}} \quad \textrm{ from (*)} \\ =& \frac{(n+1)_{n}(n+1)_{n}}{(2n+1)_{n}(1)_{n}}\\ =& \frac{\Gamma(2n+1)\Gamma(2n+1)\Gamma(2n+1)}{\Gamma(n+1)\Gamma(n+1)\Gamma(3n+1)n!}\\ =& \frac{(2n)!^3}{n!^3(3n)!} \end{align*} Therefore we can conclude \boxed{ \sum_{k=0}^{n} \frac{{\binom{n}{k}}^3}{\binom{3n}{k}} = \frac{(2n)!^3}{n!^3(3n)!}}

Integral of the day XXVIII

Integral involving the Gaussian function

Integral involving the Gaussian fucntion {}_{2}F_{1}


Today we show the proof of this integral posted by @integralsbot \int_{-\infty}^{\infty} \frac{dx}{(1+2a+\cosh x)^s } = \frac{2^{s}\Gamma^2(s)}{\Gamma(2s)}{}_{2}F_{1}\left[{s,s \atop s+\frac{1}{2}}; -a\right] \quad |a|\lt 1 For the proof we will use the generalized binomial theorem, some properties of the Pochhammer symbol (risng facorial) and the duplication formula for the Gamma function

Proof:

Suppose |a|\lt 1 \begin{align*} I= \int_{-\infty}^{\infty} \frac{dx}{(1+2a+\cosh x)^s } =& 2\int_{0}^{\infty} \frac{dx}{(1+2a+\cosh x)^s } \quad \textrm{(The integrand is even)}\\ =& 2\int_{0}^{\infty} \frac{dx}{\left(2a+2\cosh^2 \left(\frac{x}{2}\right)\right)^s } \quad \left( 1+\cosh(x) = 2\cosh^2\left(\frac{x}{2}\right)\right)\\ =& \frac{1}{2^{s-1}}\int_{0}^{\infty} \frac{dx}{\left(a+\cosh^2 \left(\frac{x}{2}\right)\right)^s } \\ =& \frac{1}{2^{s-1}}\int_{0}^{\infty} \sum_{j=0}^{\infty} \binom{-s}{j} \cosh^{2(-s-j)}\left(\frac{x}{2}\right)a^j dx \quad \textrm{(Generalized bionomial theorem)} \\ =& \frac{1}{2^{s-2}}\sum_{j=0}^{\infty} \binom{-s}{j}a^j \int_{0}^{\infty} \cosh^{2(-s-j)}\left(w\right)dw \\ =& \frac{1}{2^{s-2}}\sum_{j=0}^{\infty} \binom{-s}{j}a^j \int_{0}^{\infty} \operatorname{sech}^{2(s+j)}\left(w\right)dw \\ =& \frac{1}{2^{s-2}}\sum_{j=0}^{\infty} \binom{-s}{j}a^j \int_{0}^{1} t^{2(s+j)-1}(1-t^2)^{-\frac{1}{2}} dt \quad (t \mapsto \operatorname{sech}(x) )\\ =& \frac{1}{2^{s-1}}\sum_{j=0}^{\infty} \binom{-s}{j}a^j \int_{0}^{1} s^{(s+j)-1}(1-s)^{-\frac{1}{2}} ds \quad (s \mapsto t^2 )\\ =& \frac{1}{2^{s-1}}\sum_{j=0}^{\infty} \binom{-s}{j}a^j B\left(s+j,\frac{1}{2}\right)\\ \end{align*} Using the fact that \binom{v}{m} = \frac{(v-m+1)_{m}}{m!} and the recursion formula (x-n+1)_{n} = (-1)^n(-x)_{n} where (x)_{n} = x(x+1)(x+2)\cdots (x+n-1) we have \begin{align*} I =& \frac{1}{2^{s-1}}\sum_{j=0}^{\infty} \binom{-s}{j}a^j B\left(s+j,\frac{1}{2}\right)\\ =& \frac{1}{2^{s-1}}\sum_{j=0}^{\infty} \frac{(-s-j+1)_{j}}{j!}a^j \frac{\Gamma(s+j)\sqrt{\pi}}{\Gamma\left(s+j+\frac{1}{2}\right)}\\ =& \frac{1}{2^{s-1}}\sum_{j=0}^{\infty} \frac{(s)_{j}}{j!}(-a)^j \frac{\Gamma(s+j)\sqrt{\pi}}{\Gamma\left(s+j+\frac{1}{2}\right)}\\ \end{align*} From the definition of the Pochhammer symbol: (x)_{n} = \frac{\Gamma(n+x)}{\Gamma(x)} we have I =\frac{1}{2^{s-1}}\sum_{j=0}^{\infty} \frac{(s)_{j}}{j!}(-a)^j \frac{\Gamma(s+j)\sqrt{\pi}}{\Gamma\left(s+j+\frac{1}{2}\right)} = \frac{\Gamma(s)}{2^{s-1}\sqrt{\pi}\Gamma\left(s+\frac{1}{2}\right)}\sum_{j=0}^{\infty} \frac{(s)_{j}(s)_{j}}{\left(s+\frac{1}{2}\right)_{j}}\frac{(-a)^j}{j!} From the duplication formula \Gamma(2x) = \frac{4^x}{2\sqrt{\pi}} \Gamma(x)\Gamma\left(\frac{1}{2}+x\right) we have \begin{align*} I = \frac{\Gamma(s)}{2^{s-1}\sqrt{\pi}\Gamma\left(s+\frac{1}{2}\right)}\sum_{j=0}^{\infty} \frac{(s)_{j}(s)_{j}}{\left(s+\frac{1}{2}\right)_{j}}\frac{(-a)^j}{j!} & = \frac{2^{s}\Gamma^2(s)}{\Gamma(2s)}\sum_{j=0}^{\infty} \frac{(s)_{j}(s)_{j}}{\left(s+\frac{1}{2}\right)_{j}}\frac{(-a)^j}{j!}\\ =& \frac{2^{s}\Gamma^2(s)}{\Gamma(2s)}{}_{2}F_{1}\left[{s,s \atop s+\frac{1}{2}}; -a\right] \end{align*} Finally, we can conclude \boxed{ \int_{-\infty}^{\infty} \frac{dx}{(1+2a+\cosh x)^s } = \frac{2^{s}\Gamma^2(s)}{\Gamma(2s)}{}_{2}F_{1}\left[{s,s \atop s+\frac{1}{2}}; -a\right] \quad |a|\lt 1 }

Series of the day

Series involving the digamma and the zeta functions The sum \displaystyle \sum\frac{1}{(n+1)^pn^q} ...