`PoissonSolve` descales `rho` after every space-charge solve

[nikitakuklev]

<claude-ish-text>

Summary

src/particles/spacecharge/PoissonSolve.cpp:153-157 unconditionally multiplies rho by -ε₀ on every level after ablastr::fields::computePhi returns. In the current ablastr, computePhi already restores rho to its original value on both the MLMG and IGF paths, so this post-loop overwrite is wrong.

Code

src/particles/spacecharge/PoissonSolve.cpp:153-157:

// fix side effect on rho from previous call
for (int lev=0; lev<=finest_level; lev++) {
    using namespace ablastr::constant::SI;
    rho[lev].mult(-1._rt * epsilon_0);
}

Details

In the current ablastr computePhi (warpx/Source/ablastr/fields/PoissonSolver.H), rho is symmetrically scaled and un-scaled on the MLMG path:

• Line 288 (entering MLMG): rho[lev]->mult(-1._rt / ablastr::constant::SI::epsilon_0);
• Line 446 (after MLMG solve, still inside the per-level loop): rho[lev]->mult(-ablastr::constant::SI::epsilon_0); // Multiply rho by epsilon again

The IGF path (PoissonSolver.H:277-284) continues before reaching either of those lines, so it never touches rho at all.

Net effect of computePhi on rho:

• IGF path: unchanged
• MLMG path: unchanged (scale and un-scale cancel exactly)

Then ImpactX runs the post-loop at PoissonSolve.cpp:153-157, which multiplies rho by -ε₀ on every level for both paths.

</claude-ish-text>

Reproducer

(boilerplate removed)

sim.track_particles()                # one SC slice
R_post  = gather(sim.rho(0))         # post-PoissonSolve state
sim.deposit_charge()                 # re-deposit on same geometry
R_redep = gather(sim.rho(0))         # clean reference

# elementwise ratio on cells with appreciable magnitude:
nz = abs(R_redep) > R_redep_max * 1e-6
median_ratio = numpy.median(R_post[nz] / R_redep[nz])
# observed: -8.854188e-12 (== -EPS0) on both fft and multigrid

Deleting PoissonSolve.cpp:153-157 seems ok

[ax3l]

That's a good point. We fixed this eventually in ABLASTR but forgot to take out the work around here. Will fix, thank you for the analysis!

[ax3l]

@nikitakuklev since you found it, so you like to have the honors & credits of fixing it here via PR? Happy to review it and check that we have tests in place.