5
$\begingroup$

Bug introduced in 9.0 or earlier and fixed in 11.0.0


Calculating this sum on Mathematica 10.3

Sum[(-1)^(r - 1)/((a^2 + r^2)r), {r, 1, Infinity}]

gives the answer

$$-\frac{1}{2a^4}+\frac{\pi^2}{12a^2}+\frac{\pi\;\text{Csch}(a\pi)}{2a^3}$$

but this is not right, and there shouldn't be any neat answer like this in terms of elementary functions. As $a\to0$, the answer should tend to $(3/4)\zeta(3)$, which doesn't have a simple form in terms of a rational multiple of $\pi^4$, as suggested by the above formula. (Typically the answer would be expressed in terms of the digamma function.)

My question is, am I doing something wrong? How should one avoid this kind of problem and be confident that the sums returned are correct?

$\endgroup$
  • $\begingroup$ Bug still exists in Mathematica 10.4 $\endgroup$ – user58955 Mar 17 '16 at 15:52
2
$\begingroup$

Seems like a bug to me. The following is a workaround, at least on this example.

sum0 = Sum[(-1)^(r - 1)/((1. a^2 + r^2) r), {r, 1, Infinity}]

Solve::ratnz: Solve was unable to solve the system with inexact coefficients. The answer was obtained by solving a corresponding exact system and numericizing the result. >>

(*
  1/(1. + 1. a^2) 1. *
    HypergeometricPFQ[
     {1., 1.,
      (0. + 1. I) ((0. - 1. I) + (1. + 0. I) a),
      (0. - 1. I) ((0. + 1. I) + (1. + 0. I) a)},
     {2., (0. + 1. I) ((0. - 2. I) + (1. + 0. I) a),
      (0. - 1. I) ((0. + 2. I) + (1. + 0. I) a)},
     -1.]
*)

sum = sum0 // Rationalize // Simplify
(*
  (1/(4 a^2))(Log[16] + PolyGamma[0, 1/2 - (I a)/2] - 
    PolyGamma[0, 1 - (I a)/2] + PolyGamma[0, 1/2 + (I a)/2] - 
    PolyGamma[0, 1 + (I a)/2])
*)

Limit[sum, a -> 0]
% - 3/4 Zeta[3] // FullSimplify
(*
  1/16 (-PolyGamma[2, 1/2] + PolyGamma[2, 1])
  0
*)
$\endgroup$
  • $\begingroup$ Note, FWIW: Changing exact expressions to inexact ones often changes the methods and decisions made by Mathematica. Normally, exact expressions give better results (that often take longer to compute). When faced with a bug, it's worth a shot trying an inexact coefficient. Numbers like 1. and 0. seem like good ones to start with, since there is no round-off error in them to start with. $\endgroup$ – Michael E2 Feb 25 '16 at 23:44
  • $\begingroup$ Thank you for your answer. Is this kind of bug common? I am new to Mathematica and this is pretty much the first thing I tried, so I am wondering how reliable it is in general. I suppose what might be helpful would be some guidance of the form "avoid infinite sums with alternating signs, because they are known to be problematic" or some such. $\endgroup$ – Alex Selby Feb 25 '16 at 23:45
  • $\begingroup$ @AlexSelby Not very common. Currently there are 9 questions tagged bugs and summation. It might be better to say that Mathematica is sometimes overambitious, in that it has sophisticated algorithms that will attempt to tackle tricky cases and sometimes get tricked. OTOH, this might be a programming error. Hard to say which it is in this case. You should report it to Wolfram support (menu Help > Give Feedback...). $\endgroup$ – Michael E2 Feb 25 '16 at 23:51
  • $\begingroup$ OK, thanks. Looks like I was just unlucky then. (I wonder what caused it to come up with this particular formula.) $\endgroup$ – Alex Selby Feb 25 '16 at 23:57
4
$\begingroup$

Any of the methods for the Regularization option to Sum provide the same result with the expected limit.

Union[FullSimplify[
    Sum[(-1)^(r - 1)/((a^2 + r^2) r), {r, 1, Infinity}, 
     Regularization -> #]] & /@
  {"Abel", "Borel", "Cesaro", "Dirichlet", 
   "Euler"}]

(*  {(1/(4*a^2))*(HarmonicNumber[
          -(1/2) - (I*a)/2] - 
        HarmonicNumber[-((I*a)/2)] - 
        HarmonicNumber[(I*a)/2] + 
        HarmonicNumber[(1/2)*I*
            (I + a)] + Log[16])}  *)

Limit[%, a -> 0] // FunctionExpand // Simplify

(*  {(3*Zeta[3])/4}  *)

sum[a_] = (1/(4*a^2))*(
    HarmonicNumber[-(1/2) - (I*a)/2] -
     HarmonicNumber[-((I*a)/2)] -
     HarmonicNumber[(I*a)/2] +
     HarmonicNumber[(1/2)*I*(I + a)] + Log[16]);

Plot[sum[a], {a, -5, 5},
 Epilog -> {Red, AbsolutePointSize[5],
   Point[{0, 3 Zeta[3]/4}]}]

enter image description here

sum[a] == sum[-a] // Simplify

(*  True  *)
$\endgroup$

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.