Prescribed (non-zero) displacements for shell models

[mkofler96]

Problem

The pure-shell solve path rejects prescribed non-zero displacements: shellDirichlet in web/src/workers/solver.worker.ts throws "prescribed (non-zero) displacements are not supported for shell models yet". This is a deliberate loud error rather than a silent pin-to-zero (the issue #216 failure mode), but it is a real capability gap versus the solid path, which supports inhomogeneous essential BCs via prescribed_dofs (solve_mfem.cpp::add_prescribed_dofs).

The gap is in the engine: solve_shell (engine/cpp/solve_shell.cpp) accepts only fixed_vertices / fixed_dofs / point_loads — there is no inhomogeneous essential BC input, and shell_core's CG solve eliminates fixed DOFs by zeroing rows/columns, which only expresses u = 0.

Fix

1. engine/cpp/shell_core.cpp: support prescribed DOF values — the standard lifting approach: move K·u_prescribed to the right-hand side, then constrain those DOFs like fixed ones. Unit-test natively in engine/tests/shell_validation.cpp (e.g. a plate with a prescribed edge translation reproduces the equivalent force-loaded solution).
2. engine/cpp/solve_shell.cpp: accept a prescribed_dofs array {vertex, dof, value} mirroring the solid solver's contract, with the same out-of-range validation (bc_validation.h, solve_shell.cpp silently drops out-of-range vertex/DOF indices in fixed DOFs, fixed vertices, and point loads (shell-solve equivalent of #362/#367) #379/Reject out-of-range BC/load indices loudly (issue #362) #380).
3. web/src/workers/solver.worker.ts: replace the throw in shellDirichlet with grouping into prescribed_dofs, mirroring groupDirichlet's split for the solid path.

Existing coverage to mirror: examples/validation/prescribed-displacement.test.mjs exercises the solid contract end-to-end.

Related

• dirichlet boundary conditions do not work #216 (prescribed displacements must not be silently pinned to zero)
• solve_shell.cpp silently drops out-of-range vertex/DOF indices in fixed DOFs, fixed vertices, and point loads (shell-solve equivalent of #362/#367) #379 / Reject out-of-range BC/load indices loudly (issue #362) #380 (BC ingestion validation conventions in solve_shell.cpp)

[mkofler96]

Closing — commit 9b055f7 (#431, "Support prescribed (non-zero) displacements on shell models", Fixes KOF-210) implements all three fix items requested here: shell_core.cpp moves the K·u_prescribed contribution to the RHS instead of only supporting u=0 elimination, solve_shell.cpp/solve_coupled.cpp accept a prescribed_dofs array mirroring the solid solver's contract, and solver.worker.ts's shellDirichlet now groups into prescribed vs. homogeneous constraints instead of throwing. Covered by 6 native checks in engine/tests/shell_validation.cpp plus examples/validation/shell-prescribed-displacement.test.mjs.

---

Generated by Claude Code