Roadmap for release 0.4.0

[nikoscham]

Following the release of version 0.3.0 (and the 0.3.1 patch), this roadmap outlines the vision for FEAScript 0.4.0. The primary focus of this release is scaling to larger problems: today the linear solver takes almost all of the run time, and medium-sized 2D problems already become slow. The goal for 0.4.0 is to solve noticeably larger 2D problems in a fraction of today's time, on a single thread. Version 0.4.0 also starts moving the API toward a stable 1.0.

---

Numerical Solution & Performance

• Preconditioned conjugate gradient (PCG) solver: A sparse iterative solver that becomes the default for symmetric problems (heat conduction, and later elasticity) (Add a preconditioned conjugate gradient (PCG) solver #115).
• Preconditioners: Jacobi, plus incomplete Cholesky (IC0) or SSOR (Add preconditioners for PCG #116).
• Symmetric Dirichlet conditions: Apply fixed-value conditions by eliminating their columns as well as their rows (moving the known values to the right-hand side), so the system matrix stays symmetric for PCG (Apply Dirichlet conditions symmetrically #117).
• Solver for non-symmetric problems: GMRES or BiCGStab for advection-dominated problems (Add an iterative solver for non-symmetric problems #118).
• Iterative solver robustness: A tighter default tolerance than the current Jacobi default, a clear warning or error when a solver does not converge, and regression tests that check the iterative solvers against the direct solvers (Iterative solver robustness #119).
• Assembly into typed-array CSR: Compute the sparsity pattern from the element connectivity, then assemble the system matrix directly into compressed sparse row (CSR) format with typed arrays instead of a Map per row, to cut memory use (Assemble the system matrix directly into typed-array CSR #120).
• Frontal solver memory: The frontal solver currently runs out of memory on medium-sized problems, because its working front is allocated with the size of the whole system. The front will be sized from the actual maximum front width (computed in a prefront pass), stored in typed arrays, and kept narrow by reordering the elements. This keeps a robust direct solver for problems where iterative solvers struggle, such as creeping flow, and a reference for checking the iterative solvers (Frontal solver runs out of memory on medium-sized problems #121).
• Frontal solver for nonlinear general form PDEs: Use the frontal solver inside the Newton–Raphson iterations, as already done for the linear general form PDE (Support the frontal solver for nonlinear general form PDEs #101).
• Jacobi stopping criterion: The Jacobi solver now stops on the relative residual ‖R − J u‖ / ‖R‖, instead of on the size of the update.
• WebGPU Jacobi stopping criterion: The WebGPU Jacobi solver ("jacobi-gpu") stops on the same relative residual as the CPU Jacobi solver (WebGPU Jacobi solver: stop on the relative residual, as the CPU Jacobi does #113).

API

• Solver options: solver: { method, preconditioner, tolerance } replaces setSolverMethod(). The old method keeps working, with a deprecation warning (Solver options instead of setSolverMethod() #122).
• Named boundary condition objects, e.g. { type: "convection", h: 1, ambientTemperature: 20 } instead of ["convection", 1, 20]. The array form keeps working, with a deprecation warning (Named boundary condition objects #123).
• Model names without the Script suffix, e.g. "heatConduction" instead of "heatConductionScript". The old names keep working, with a deprecation warning (Model names without the Script suffix #124).
• Errors: Invalid input throws a FEAScriptError with an error code, instead of logging the error and continuing (Throw a FEAScriptError on invalid input #125).
• TypeScript declarations (.d.ts) shipped in the npm package (Ship TypeScript declarations in the npm package #126).
• Run report: result.info with whether the solver converged, the iterations, the residual and the solve time (Add a run report to the result #127).
• A version export replacing printVersion (Replace printVersion with a version export #128).
• Remove the "constantTemp" boundary condition (deprecated in 0.3.0); use "constantTemperature" (Remove the "constantTemp" boundary condition #129).

Verification & Benchmarks

• Code_Aster benchmark tests for 2D heat conduction: TPLP302 (strip with a Fourier series solution) and TPLP01 (L-shaped domain with a re-entrant corner) (Add Code_Aster benchmark tests for 2D heat conduction #130).
• Performance benchmarks in CI: Number of unknowns vs solve time and memory for every model, tracked as trends between releases and published next to the verification results (Performance Benchmarking #52). This is part of the wider testing and quality assurance plan proposed in Build Data-Driven Testing & Quality Assurance Infrastructure #50.

Physical Modelling

• Creeping flow fix: No-slip conditions now take precedence at corners shared with a moving wall, so the lid-driven cavity is mirror-symmetric (Creeping flow: no-slip should take precedence at corners shared with a moving wall #131).

Please feel free to

• Propose new features or optimizations
• Express interest in contributing to any of the above tasks
• Help define priorities for the 0.4.0 release
• Share your views on the 1.0 API (model definition, boundary conditions, results)