David-Berghaus
commented
1 year ago One way to find a suitable normalization for these cases would be to test different normalizations in cheap double precision computations to figure out the ones that work (i.e., for which stable solutions exist). This seems to be however quite ugly and tedious to implement.
It can happen that no Victor-Miller basis exists for modular/cusp form spaces of subgroups with g>0.
These cases seem to be impossible to determine in advance and are currently not caught. While it is still possible to compute these examples, this requires a lot of trial and error by hand and ugly code modifications (see: https://github.com/David-Berghaus/q-expansion/tree/Compute-H_5-Eisenstein-series).