Question about subproblem handling in AnnulusBasis

[Wang Zhen]

Dear Dedalus Development Team,

I'm working with eigenvalue problems on different domains and noticed an inconsistency in how subproblems are handled between rectangular and annular domains that I hope you can clarify.

For the square domain, solver.solve_dense(solver.subproblems[j], rebuild_matrices=True) and solver.solve_dense(solver.subproblems_by_group[(None, j)], rebuild_matrices=True) yield identical eigenvalue spectra when (j) varies from (-N_y/2) to (N_y/2), corresponding to the Fourier wavenumber (k_y) in the (y)-direction.

For the annulus domain, solver.solve_dense(solver.subproblems[j], rebuild_matrices=True) and solver.solve_dense(solver.subproblems_by_group[(j, None)], rebuild_matrices=True) produce different eigenvalue spectra, even when I expect (j) to correspond to the Fourier wavenumber in the (\theta)-direction. 

Could you explain why these two access methods behave differently for annular domains?

Best regards,

Zhen