True SDC O4 normalizes species before enforcing positive cell-centered density

[zingale]

Summary

In the 4th-order true SDC path, Castro converts cell averages to cell centers and immediately normalizes species on the centered state before enforcing a positive density. Near sharp gradients, the average-to-center conversion can produce negative or very small centered densities, so the species normalization step can generate invalid partial densities or NaNs before the later burn guards run.

Affected code

• Source/sdc/Castro_sdc.cpp
• Source/sdc/Castro_sdc.cpp
• Source/sdc/Castro_sdc.cpp
• Source/sdc/Castro_sdc.cpp
• Source/sdc/Castro_sdc_util.H
• Source/driver/Castro.cpp

Problem

normalize_species_sdc() assumes u(i,j,k,URHO) is valid and positive. It clamps each rho X_k against small_x * rho and then rescales by rho / sum(rho X_k).

That is unsafe immediately after make_cell_center(...), because Castro already recognizes elsewhere that the average-to-center conversion can make rho < 0 near steep gradients. The main computeTemp() path explicitly enforces minimum density after the same conversion before continuing thermodynamics work, but the SDC path does not.

Why this matters

• Negative centered density can make the clamp bounds nonsensical.
• Very small or negative rho_xn_sum can lead to large rescaling or NaNs.
• This happens before okay_to_burn() gets a chance to reject the zone, so the state can already be corrupted.

Likely impact

• Corrupted species in the O4 true SDC reaction update.
• Spurious failures or NaNs in difficult zones near discontinuities.
• Potentially non-reproducible behavior because this depends on local reconstructed centered states.

Suggested fix

Before calling normalize_species_sdc() on a centered state, enforce a positive density first, or fold a density floor into normalize_species_sdc() itself.

At minimum, the following two paths should be protected:

• The O4 update path after make_cell_center(...) on Sborder
• construct_old_react_source() when sdc_order == 4 && input_is_average

Possible acceptance criteria

• Add a regression test or problem setup with a steep interface that exercises the O4 true SDC path.
• Verify no NaNs appear in the centered state after normalization.
• Verify species remain bounded and sum to density after the fix.