In many solvers, a user-provided initial guess is optional, and I think it should be for panocpy as well. As a first solution, we could use all zeros as a default, which is what most users would provide anyway. We can think of a better initialization later.

When a pa.Problem instance is called, no checks are currently performed to ensure that the dimension of the primal and dual initial guess are n and m respectively, which may silently lead to wrong results.

Instantiating a panocpy.PANOCSolver object without any arguments results in the error

TypeError: __init__(): incompatible constructor arguments. The following argument types are supported:
    1. panocpy._panocpy.PANOCSolver(arg0: panocpy._panocpy.PANOCParams, arg1: panocpy._panocpy.LBFGSDirection)
    2. panocpy._panocpy.PANOCSolver(arg0: panocpy._panocpy.PANOCParams, arg1: panocpy._panocpy.LBFGSParams)
    3. panocpy._panocpy.PANOCSolver(arg0: panocpy._panocpy.PANOCParams, arg1: panocpy._panocpy.PANOCDirection)

However, it would be convenient if default behavior was implemented, where arg0 simply is replaced with a new PANOCParams object and arg1 with a default instance of LBFGSParams.