# Bugs with Sum?

In this post, I provided a solution to the integral $$\int_0^{\infty}\left[\int^1_0J_0(v\rho)\cos(\rho^2/2)\rho\,\mathrm{d}\rho \right]^2v\mathrm{d}v$$

However, I found a bug with Sum in my procedure, the following codes show that:

exp=((-1)^(j + k) 2^(-1 - 2 j - 2 k))/((1 + 2 j + 2 k) Gamma[1 + 2 j] Gamma[1 + 2 k]);

Sum[exp, {j, 0, \[Infinity]}, {k, 0, \[Infinity]}]
% // N
(*Sin[1]/4*)
(*0.210368*)

NSum[exp, {j, 0, \[Infinity]}, {k, 0, \[Infinity]}]
(*0.460368 - 1.99638*10^-45 I*)



So is it a bug?

• From bugs tag wiki: "Please do not use this tag for new questions." But that said, I think everyone will think it's a a bug, and the tag would be put on in the end. So leave it, unless there is objection. Commented May 26 at 19:13

Many times in Sum[], n = 0 seems to be a special case connected with bugs. Separate out j == 0 and k == 0:

exp = ((-1)^(j + k)  2^(-1 - 2  j - 2  k)) /
((1 + 2  j + 2  k)  Gamma[1 + 2  j]  Gamma[1 + 2  k]);

Sum[exp, {j, 1, \[Infinity]}, {k, 1, \[Infinity]}] +
2 Sum[exp /. k -> 0, {j, 0, Infinity}] -
exp /. {j -> 0, k -> 0} // Expand

(*  1/4 + Sin[1]/4  *)

• Thanks for the comment on my incorrect answer. Based on your answer, I find Sum[exp, {j, 0, [Infinity]}, {k, 1, [Infinity]}] + Sum[exp /. k -> 0, {j, 0, [Infinity]}] // Expand also works. But seperating j=0 outside does not, although j and k are interchangeable. So it seems the "bug" comes from the first sum(summing k). Commented May 27 at 13:24
• @metroidman Yes, there are several variations that work. IIRC, I tried Sum[exp, {j, 1, Infinity}, {k, 0, Infinity}] + Sum[exp /. k -> 0, {j, 0, Infinity}] first, which is similar to your example, but does not work and supports your claim of dependence on the first sum. Then I tried the above. Then I fiddled with various combinations. The one above and the similar Sum[exp, {j, 1, \[Infinity]}, {k, 1, \[Infinity]}] + Sum[exp /. k -> 0, {j, 1, Infinity}] + exp /. {j -> 0, k -> 0} // Expand seem fastest. Of the two, I'm not sure why I chose the answer; in hindsight, it seems more complicated. Commented May 27 at 14:26

I guess MMA first do the substitution $$j\to n-k$$, and then there is a bug of Sum for $$n=0$$

expr=z^(2 j+2 k)/(Gamma[1+2 j] Gamma[1+2 k])/.j->n-k//Simplify

term=Sum[expr,{k,0,n}]

Table[{term,Sum[expr,{k,0,n}]},{n,0,4}]

Sum[expr,{k,0,n},Assumptions->n==0]

Sum[expr,{k,0,n},Assumptions->n>=0]/.n->0


Sum[1/(Gamma[1+2 k] Gamma[1-2k+n]),{k,0,n}]

Sum[z^(2k)/(Gamma[1+2 k]Gamma[1-2 k+n]),{k,0,n}]

Out[]/.z->1