Dear Alexei, > now Lanczos coefficients can be evaluated *much* faster. Very nice. I committed the patch. Now that this is much faster and costs fewer memory, would it be an idea to do this at runtime, so that precision can be arbitrary? Best wishes, Chris