IncfloVelFill LOW K block inverts direction_dependent sign test (zlo inflow/outflow swapped)

[WeiqunZhang]

Severity: high · Category: correctness · Fix order: 6 of 29 — fix this 6th.

_{Filenames are numbered in reverse fix order: 001 = fix last, 029 = fix first. This file is 024.}

Location: src/prob/prob_bc.H:243

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

The defect

src/prob/prob_bc.H:243 — IncfloVelFill LOW K block inverts the direction_dependent sign test: it applies the Dirichlet inflow value when norm_vel <= 0 and copies interior (outflow treatment) when norm_vel > 0, the opposite of the LOW I (line 86) and LOW J (line 165) blocks; copy-pasted from the HIGH-face blocks.

src/prob/prob_bc.H:243 — IncfloVelFill's LOW K block inverts the direction_dependent test: line 243 applies Dirichlet when norm_vel <= 0 and line 251 extrapolates when norm_vel > 0 -- the HIGH-side conditions -- whereas LOW I (lines 85-86,94), LOW J (164-165,173), and the low-z branches of IncfloDenFill/TracFill/TempFill (lines 377,489,599) all use >= 0 / < 0.

Why it matters

F009: 3D run with zlo.type="direction_dependent" and zlo.velocity=(0,0,w>0) (inflow through bottom): velocity ghost cells at zlo are filled by copying interior values while IncfloDenFill/TracFill/TempFill (lines 377/489/599, all '>= 0.') fill inflow Dirichlet values -- inflow velocity BC silently never enforced; with w<0 (outflow) the Dirichlet value is wrongly forced.

F016: 3D run with zlo boundary type 'direction_dependent' ("dd"): inflow at zlo (w>0) is treated as outflow (ghost cells copied from interior) and outflow (w<0) is clamped to the Dirichlet inflow values -- boundary behavior is exactly reversed at the low-z face, unlike every other face.

How to reach it

• 3D build, e.g. test_no_eb_3d/benchmark.inout modified: zlo.type="direction_dependent", zlo.velocity=0. 0. 1., zhi.type="po", x/y walls. No assert/abort gates it; shipped deck already uses direction_dependent on x faces.
• 3D build; e.g. test_no_eb_3d/benchmark.inout with zlo.type="dd", zlo.velocity=0. 0. 1., zhi.type="po", geometry.is_periodic=1 1 0. No abort blocks it; LOW K executes on every velocity fillpatch.

Suggested fix

In the LOW K block of IncfloVelFill, flip the two sign tests to the low-side convention: gate the Dirichlet branch (line 243) on norm_vel >= 0. and the interior-copy branch (line 251) on norm_vel < 0.. On a low domain face the outward normal points in -z, so a positive normal velocity means flow entering the domain and must receive the Dirichlet inflow value; only negative normal velocity (outflow) should extrapolate from the interior. This is exactly the convention already used by the LOW I and LOW J twins in the same function, by the low-z branches of IncfloDenFill/IncfloTracFill/IncfloTempFill (lines 377/489/599), and by the MAC BC code (cc_arr(...,2) >= 0 at low z), and it matches the pre-#125 low-k code before commit 5634287 flipped it. The change restores intended behavior for all direction_dependent users at zlo; no compatibility decision is needed, though the maintainer may want a 3D regression test with zlo.type = "direction_dependent" since no existing test caught the flip.

For src/prob/prob_bc.H:243 (F009):

--- a/src/prob/prob_bc.H
+++ b/src/prob/prob_bc.H
@@ -240,15 +240,15 @@
                 }
 
                 if ( (bc.lo(dir) == amrex::BCType::ext_dir) ||
-                     (bc.lo(dir) == amrex::BCType::direction_dependent && norm_vel <= 0.) )
+                     (bc.lo(dir) == amrex::BCType::direction_dependent && norm_vel >= 0.) )
                 {
                     if ( orig_comp+nc == dir ) {
                         vel(i,j,k,dcomp+nc) = norm_vel;
                     } else {
                         vel(i,j,k,dcomp+nc) =  bcv_vel[amrex::Orientation(amrex::Direction::z,amrex::Orientation::low)][orig_comp+nc];
                     }
                 }
-                else if (bc.lo(dir) == amrex::BCType::direction_dependent && norm_vel > 0.)
+                else if (bc.lo(dir) == amrex::BCType::direction_dependent && norm_vel < 0.)
                 {
                     vel(i,j,k,dcomp+nc) = vel(i,j,domain_box.smallEnd(2),dcomp+nc);
                 }

_{Flips LOW K direction_dependent tests to match LOW I/LOW J and the Den/Trac/Temp low-z fills: Dirichlet when norm_vel >= 0., copy interior when norm_vel < 0. Overlaps F037's LOW-K hunk; apply one first and hand-merge the other (merged line 242-243: 'else if ( (...ext_dir) || (...direction_dependent && norm_vel >= 0.) )').}

_{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

F009 — confirmed (two independent verifier lenses)

Lens 1 (refutation attempt): prob_bc.H:242-243: LOW K Dirichlet branch is '(bc.lo(dir) == amrex::BCType::direction_dependent && norm_vel <= 0.)' and line 251 copies interior when 'norm_vel > 0.' — reversed vs LOW I (line 86 '>= 0.'), LOW J (165), and IncfloDenFill/TracFill/TempFill low-z (377/489/599, all '>= 0.'). MAC code also uses 'cc_arr(...,2) >= 0' at low-z. Regressed in commit 5634287 (#125), unfixed at HEAD.

Lens 2 (reachability/intent): prob_bc.H:242-251 LOW K uses norm_vel<=0 for Dirichlet, >0 for interior copy — opposite of LOW I/J (86,165) and DenFill/TracFill/TempFill LOW K (377,489,599, all >=0). History: 60881d3 had LOW K correct (>=0); 5634287 pasted HIGH-face conditions (bc.hi, <=0) into LOW K; 8643ecc "fix typo" restored bc.lo but missed the signs. Consumed via IncfloVelFill in fillpatch, advection, and nodal-projection inflow enforcement (incflo_apply_nodal_projection.cpp:153).

F016 — confirmed (two independent verifier lenses)

Lens 1 (refutation attempt): Same defect as F009, confirmed in current source: line 243 'norm_vel <= 0.' gates Dirichlet and line 251 'norm_vel > 0.' gates interior copy in the LOW K block — the HIGH-side conditions. All other low-side twins (lines 86/94, 165/173) and scalar low-z fills (377/489/599) use '>= 0.'/'< 0.'. git show 5634287 shows the pre-#125 low-k code correctly used 'norm_vel >= 0.' for Dirichlet; #125 flipped it.

Lens 2 (reachability/intent): prob_bc.H:242-251 LOW K uses norm_vel <= 0.→Dirichlet / > 0.→copy-interior, mirror of LOW I/J (85-86,94) and scalar low-z fills (377,489,599). History: 60881d3 had LOW K correct (>= 0.); 5634287 (#125) pasted HIGH-side logic (bc.hi, <= 0.); 8643ecc (#126 "fix typo") fixed only bc.hi→bc.lo, leaving signs inverted. No assert gates dd; IncfloVelFill fills all velocity fillpatches (incflo_fillpatch.cpp:13-26).

---

_{Based on commit 7307d872, which is also the tree the audit verified against. From an automated audit of src/. Audit finding ids: F009, F016. Reviewer unit(s): Prob, theme:symmetry-twins. 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.}