>> I have one remaining problem. Whereas bflevin_u_sum(1/n^2,n,1) converges
>> very quickly
>> bflevin_u_sum(1/n^(201/100),n,1) does not converge at all. It would be
>> nice if somebody could explain this to me.
>
> I now undersand why that happens. If we sum a rational function then
> the result of levin(s,k,n,beta,omega) is a rational number and hence
> bfloat can compute it very accurately. Otherwise one must be much more
> careful than I have been to estimate rounding errors. So for the moment
> bflevin_u_sum works only reliably for rational functions. It should be
> compared to sumninf which was posted earlier (which I can't run yet).
I should have said: it works reliably for functions taking rational values.
Regards,
Michel