Re: A Bessel integral

*To*: mathgroup at smc.vnet.net*Subject*: [mg36848] Re: [mg36779] A Bessel integral*From*: Vladimir Bondarenko <vvb at mail.strace.net>*Date*: Sun, 29 Sep 2002 02:55:09 -0400 (EDT)*Reply-to*: Vladimir Bondarenko <vvb at mail.strace.net>*Sender*: owner-wri-mathgroup at wolfram.com

Roberto Brambilla <rlbrambilla at cesi.it> wrote on Wed, 25 Sep 2002 01:50:58 : RB> I am considering the following integral RB> W[m_,n_]:=Integrate[BesselJ[m, x]*BesselJ[n, x], {x, 0, Infinity}] RB> where m,n are reals >=0. With Mathematica 4.1 I obtain: RB> If[Re[m+n]>-1, -Cos[(m-n)Pi/2]/(2 Pi)* RB> (2 EulerGamma + Log[4] + RB> PolyGamma[0, 1/2(1 + m - n)] + RB> PolyGamma[0, 1/2(1 - m + n)] + RB> 2PolyGamma[0, 1/2(1 + m + n)]) RB> Any explanation about the analytical expression will be RB> gratefully accepteed. The expression for W[m_,n_] returned by Mathematica is wrong. To prove, just substitute m = n = 0 which is exactly what you had done and observe that the output you had had W[0,0]=-(2 EulerGamma + Log[4] + 4 PolyGamma[0, 1/2])/(2 Pi) = 0.84564 was incorrect. The correct answer is 1/2. Mathematica can handle the numeric integration successfully In[1] := NIntegrate[BesselJ[1, x]*BesselJ[0, x], {x, 0, Infinity}, Method -> Oscillatory] (* The warnings are skipped *) Out[1] = 0.5 Using NIntegrate[BesselJ[0, x]*BesselJ[0, x], {x, 0, Infinity}] without Method -> Oscillatory is not the optimal choice as the integrand oscillates fairly rapidly over the integration region. RB> I suspect that these integrals are divergent (*). In fact, not exactly. Integrate[BesselJ[1, x]*BesselJ[0, x], {x, 0, Infinity}] is equal to 1/2, and Mathematica 4.1 for Microsoft Windows (November 2, 2000) does it correctly, while Mathematica 4.2 for Microsoft Windows (February 28, 2002) concocts a strange mixture of a wrong divergence message and the warning that it cannot check the convergence [should I trust to the second warning? or the first?] As a matter of fact, Integrate[BesselJ[1, x]*BesselJ[0, x], {x, 0, Infinity}] converges because the integrand is regular at x=0, bounded over the whole right semi-axis, and decays as 2*Cos[Pi/4 - x]*Cos[(3*Pi)/4 - x]/(Pi*x) + o(1/x) at x -> Infinity . Say, calculate Normal[Series[BesselJ[1, x], {x, Infinity, 1}]] Normal[ Series[BesselJ[0, x], {x, Infinity, 1}]] // InputForm -> (2*(Cos[Pi/4 - x] - Sin[Pi/4 - x]/(8*x))*(Cos[(3*Pi)/4 - x] + (3*Sin[(3*Pi)/4 - x])/(8*x)))/(Pi*x) then Plot[%,{x,1,10}] and Plot[BesselJ[1,x]*BesselJ[0,x],{x,1,10}] and you could hardly see the difference. Generally, to get to the convergence domain for W in terms of m and n is easy via the asymtotics of the Bessel functions (use something like Expand[Normal[Series[BesselJ[m, x], {x, Infinity, 1}]]Normal[ Series[BesselJ[n, x], {x, Infinity, 1}]]] then analyze the main term). Best wishes, Vladimir Bondarenko Mathematical Director Symbolic Testing Group Email: vvb at mail.strace.net Web : http://www.CAS-testing.org/ http://maple.bug-list.org/VER2/ (under tuning) http://maple.bug-list.org/VER3/ (under tuning) http://maple.bug-list.org/VER1/ (under tuning) http://www.beautyriot.com/ (teamwork) http://www.ohaha.com/ (teamwork) Voice: (380)-652-447325 Mon-Fri 9 a.m. - 6 p.m. Mail : 76 Zalesskaya Str, Simferopol, Crimea, Ukraine