Fix NavierStokesSLCN DDt projection source shape mismatch (#180)

[lmoresi]

Summary

One-line outdated-API fix. When SNES_MultiComponent_Projection was wired into the SemiLagrangian DDt path, the psi_fn setter and _setup_projections were updated to flatten the source tensor to (1, Nc) via _build_projection_source — but the fallback inside update_pre_solve (taken whenever uw.function.evaluate() raises on expressions containing derivatives) was missed and kept assigning self.psi_fn directly.

Stokes never hits this fallback because its projection source has no derivatives. NavierStokes hits it every solve, producing:

sympy.matrices.exceptions.ShapeError: Matrix size mismatch: (1, 3) + (2, 2).

— (1, 3) is the projector's Nc=3 symmetric-tensor solution variable in 2D; (2, 2) is the raw stress flux tensor that should have been flattened.

Files

• src/underworld3/systems/ddt.py — route the fallback through _build_projection_source.
• tests/test_0610_navier_stokes_slcn_projection.py — regression test: one ns.solve() on a tiny mesh. Reliably raises ShapeError without the fix, passes with it.

Test plan

• pytest tests/test_0610_navier_stokes_slcn_projection.py — passes
• Without the fix the same test fails with the user's exact error (1, 3) + (2, 2)
• Reporter @Chawgai to confirm on their nstt.py reproducer

Reported by @Chawgai. Diagnosed with @ss2098's investigation pointing at the SLCN projection/history-field path as the root cause.

Underworld development team with AI support from Claude Code

[Copilot]

Pull request overview

Fixes a NavierStokes SemiLagrangian (SLCN) DDt projection fallback path where tensor-valued projection sources were not flattened to the (1, Nc) row-matrix shape required by SNES_MultiComponent_Projection, which previously caused SymPy ShapeError during solves.

Changes:

• Route the update_pre_solve() projection fallback through _build_projection_source() so tensor sources are flattened consistently.
• Add a regression test that runs a minimal NavierStokes.solve() and asserts it no longer raises the (1, 3) + (2, 2) shape mismatch error.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
src/underworld3/systems/ddt.py	Updates the SemiLagrangian DDt fallback projection source wiring to apply tensor flattening via _build_projection_source().
tests/test_0610_navier_stokes_slcn_projection.py	Adds a regression test covering the NavierStokes SLCN DDt projection fallback shape mismatch.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[lmoresi]

Review pass — ready to merge

Scope: the DDt-fallback path was assigning self.psi_fn directly to _psi_star_projection_solver.uw_function, which collided with SNES_MultiComponent_Projection's (1, Nc) row-matrix contract — producing the ShapeError("(1, 3) + (2, 2)") reported in #180. The fix routes through _build_projection_source(self.psi_fn) so the same flattening used at setup time applies to the runtime fallback, and adds the matching flat→tensor fan-out so subsequent advection sees the freshly-projected values.

Verification:

• Confirmed _build_projection_source already exists and is what _setup_projections and the psi_fn setter use — the fallback was the only place that had been missed in the multicomponent migration.
• Fan-out mirrors initialise_history() (~line 1540) — same indexing, same symmetric fill.
• The new regression test (tests/test_0610_navier_stokes_slcn_projection.py) reproduces the exact ShapeError("(1, 3) + (2, 2)") from the issue, and is correctly tagged level_2 / tier_a so it runs in standard CI.
• @Chawgai already reported on [BUG] - Matrix Mismatch Regarding NavierStokesSLCN Solver #180 that this branch resolves the crash on his nstt.py reproducer.

Copilot note on stale state: Copilot's earlier comment about the missing fan-out is now stale — the current diff includes exactly the fan-out it asked for (lines 1779–1785). Worth marking resolved.

Suggested follow-up: the fallback exists at all because uw.function.evaluate couldn't handle tensor expressions with derivatives. PR #185 removes that limitation, which means in steady-state operation this fallback path should become rare. Both PRs together close out #180.

LGTM modulo standard maintainer eyeball.

Underworld development team with AI support from Claude Code