Nodal projection in/out solvability rescales m_leveldata velocity instead of the vel argument

[WeiqunZhang]

Severity: medium · Category: correctness · Fix order: 12 of 29 — fix this 12th.

_{Filenames are numbered in reverse fix order: 001 = fix last, 029 = fix first. This file is 018.}

Location: src/projection/incflo_apply_nodal_projection.cpp:171

Based on commit 7307d872 (line numbers refer to that tree).

The defect

What is wrong — The in/out solvability block aliases m_leveldata[lev]->velocity instead of the vel[lev] parameter, so when the low-level ApplyNodalProjection is called with a different velocity vector, solvability is enforced on the wrong field and the state velocity is modified.

Why it matters

Nodal projection + a direction_dependent BC (has_inout_bndry=true) + incflo.do_initial_pressure_proj=1: InitialPressureProjection (init.cpp:610) passes a local gravity-based vel; enforceInOutSolvability rescales ld.velocity's boundary cells (corrupting the state) while the projected field keeps its influx/outflux imbalance, giving a non-solvable nodal solve and wrong hydrostatic p/gp.

Suggested fix

In the low-level ApplyNodalProjection overload, build the enforceInOutSolvability aliases from the vel parameter rather than m_leveldata[lev]->velocity. HydroUtils::enforceInOutSolvability mutates the passed field in place (it rescales outflow boundary values so total outflux matches influx), and the nodal projector at lines 185/189 consumes vel — so the field being projected must be the one rebalanced, exactly as the MAC twin does by passing mac_vec (the same object later given to macproj->setUMAC) in incflo_compute_MAC_projected_velocities.cpp. Aliasing vel is consistent with the rest of the function, which already operates on vel[lev] for setBndry and the inflow physbc fill. In the normal timestep path vel aliases the state velocity, so behavior there is unchanged; the maintainer should just confirm no caller passing a separate vector (e.g. InitialPressureProjection) relied on the old side effect of the state velocity being rescaled.

--- a/src/projection/incflo_apply_nodal_projection.cpp
+++ b/src/projection/incflo_apply_nodal_projection.cpp
@@ -167,10 +167,9 @@
         Vector<Array<MultiFab, AMREX_SPACEDIM>> vel_vec(finest_level+1);
 
         for (int lev = 0; lev <= finest_level; lev++) {
-            auto& ld = *m_leveldata[lev];
-            AMREX_D_TERM(vel_vec[lev][0] = MultiFab(ld.velocity, amrex::make_alias, 0, 1);,
-                         vel_vec[lev][1] = MultiFab(ld.velocity, amrex::make_alias, 1, 1);,
-                         vel_vec[lev][2] = MultiFab(ld.velocity, amrex::make_alias, 2, 1););
+            AMREX_D_TERM(vel_vec[lev][0] = MultiFab(*vel[lev], amrex::make_alias, 0, 1);,
+                         vel_vec[lev][1] = MultiFab(*vel[lev], amrex::make_alias, 1, 1);,
+                         vel_vec[lev][2] = MultiFab(*vel[lev], amrex::make_alias, 2, 1););
         }
 
         HydroUtils::enforceInOutSolvability(GetVecOfArrOfPtrs(vel_vec), get_velocity_bcrec().data(), geom, true);

_{Aliases the vel parameter (the field actually projected at lines 185/189) instead of m_leveldata state, matching the MAC twin which passes mac_vec. HydroUtils::enforceInOutSolvability (AMReX-Hydro Utils/hydro_utils.H) rescales the passed MultiFabs in place. The now-unused 'auto& ld' local is removed to avoid an unused-variable warning. Behavior unchanged for the main overload where vel[lev] == &ld.velocity.}

_{Diff(s) are against 7307d872, written from the current source and verified only with git apply --check — never compiled, never run, never applied to the tree. Treat them as precise intent, not tested patches.}

Verification evidence

F038 — confirmed (one verifier lens)

Lens 1 (refutation attempt): Lines 171-173: vel_vec[lev][0] = MultiFab(ld.velocity, amrex::make_alias, 0, 1) inside the overload whose parameter is vel; line 176 passes these aliases to enforceInOutSolvability, while line 185/189 projects vel. init.cpp:610 calls this overload with a local gravity-based vel and set_inflow_bc=true; has_inout_bndry is true for direction_dependent BCs. AMReX-Hydro's correct_outflow rescales the passed field in place, so the state velocity is mutated and the projected field keeps its imbalance. The MAC twin (line 269 of incflo_compute_MAC_projected_velocities.cpp) correctly uses mac_vec. No fix in git history.

---

_{Based on commit 7307d872, which is also the tree the audit verified against. From an automated audit of src/. Audit finding id: F038. Reviewer unit(s): Projection. Nothing here was compiled or run — the failure scenarios are code reasoning, so the reaching configuration above is the cheapest way to confirm or refute it.}