the function simplify_sum fails to compute the serie :
sum((k*sin((k*%pi)/3))/(4*k^2-1),k,1,inf)
I know the serie converge to
(sqrt(3)*%pi)/16
=0.3400873807939158...
but simplify_sum find a closed form which is wrong (from its floating point approximation)
'(5*3^(3/2)*sqrt(%pi)*%i*erf(%i)-10*sqrt(3)*%e)/48
=2.564242426927912...
The bug could comes from the Zeilberger/Gosper algorithm or a lack of knowledge on some special functions. Note that the website wolframalpha find a complex closed form but with the correct floating point value :
sum_(k=1)^∞ (k sin((π k)/3))/(4 k^2 - 1) = 1/6 ((-1)^(1/6) 2F1(1/2, 2, 5/2, -(-1)^(2/3)) - (-1)^(5/6) 2F1(1/2, 2, 5/2, (-1)^(1/3)))≈0.340087
see maxima mailing list for further discussions.
here is another example where maxima (and sagemath) fails to simplify a summation . Wolframalpha.com also fails to find the pi/16 value but gives different closed forms and a numerical value compatible with pi/16. Here is a minimal example showing the problem :
you can check that the value given by simplify_sum is far from the numerical value, and this one is compatible with pi/16 (which has been obtained by Dirichlet Theorem for the Fourier series of some piecewise constant function):
Last edit: Robert Dodier 2023-05-14