How would this be with the generalized eigenvalue problem?

[PetrKryslUCSD]

Any chance I could get it working with a GEP?
I have published previously a paper on approximate solution of the GEP
using ad hoc MOR https://onlinelibrary.wiley.com/doi/full/10.1002/nme.6358.
So I have a fairly good idea of the approximate eigenspace. I wonder if the
algorithm would work in that case.