Hi,
I changed the demo file at http://thegate.bertold.org:8081/levin
to include an example where bflevin_u_sum does not converge
due to rounding errors (the Euler-Mascheroni constant:
sum(1/n-log(1+1/n),n,1,inf)). In that case one can still
use levin_u_sum with a specified number of terms.
I think it is quite easy to make an estimate for the internal precision
necessary to make bflevin_u_sum converge more robustly (this is in the
literature of course). If I find time I will look into this.
Regards,
Michel