newtonpf.py: choose solver based on distributed_slack parameter

[rbolgaryn]

+Some small changes

• to make the behavior in lightsim2grid and pandapower consistent in terms of the J matrix, we choose the "standard" vs "single slack" version of the solver bnased on the parameter "distributed_slack".

• We transform the Ybus to csc if it is not yet csc. TODO for pandapower: define Ybus in csc in the first place if lightsim2grid is used (right now implemented via to_csc()).

[rbolgaryn]

I noticed that if I uncomment the division by baseMVA then a test fails in pandapower, due to tolerances. Do you accept tolerance in MVA in newton and translate it in pu internally? Or do you consider the tolerance in a different way?

In pandapower tolerance is considered like this:

def _check_for_convergence(F, tol):
    # calc infinity norm
    return linalg.norm(F, Inf) < tol

Can you please double-check?

[BDonnot]

Can you please double-check?

I can confirm that inside the c++ solver nothing is done with respect to the tolerance.

The c++ code looks like:

bool BaseSolver::_check_for_convergence(const RealVect & F,
                                        real_type tol)
{
    const auto norm_inf = F.lpNorm<Eigen::Infinity>();
    bool res =  norm_inf  < tol;
    return res;
}

So it's exactly the same as the one you mentionned