caelum0x 771f89b099 feat(v2-A/v2-B): weak-form, implicit dynamics, FROLS/SSR/trapping, SRBench harness
Four expansion-v2 items (finished after a session-limit interruption):

Weak/integral-form discovery (crates/lawsynth-weakform, specs/weak-form) — the top
noise-robustness technique from PySINDy: integrate candidates against compactly-
supported test functions, NO differentiation of noisy data. ~23x lower coefficient
error than finite-difference strong-form under noise. 20 tests, deterministic.

Implicit/rational dynamics (crates/lawsynth-implicit, specs/implicit-dynamics) —
SINDy-PI alternating-LHS nullspace solve; recovers rational laws (Michaelis-Menten)
explicit regression can't express. Honest degree-2 identifiability limit documented
(multi-dimensional nullspace); residual-floor + derivative-preferring tie-break for
canonical selection. 10 tests.

Optimizer breadth (lawsynth-sparse) — FROLS (orthogonal-least-squares, ERR
ordering), SSR (stepwise), trapping (stability-biased); --solver frols|ssr|trapping.
24 tests; default path byte-identical.

SRBench harness (benchmarks/) — 18 Strogatz ODE systems + scoring + a determinism
column (byte-identical on replay — a metric no stochastic competitor can report).
3/3 Strogatz recovered at R²=1.0. srbench_report.py + 5 tests.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-08-18 14:49:27 +03:00
..

Weak / integral-form discovery boundary (v2-A)

This directory specifies weak/integral-form system identification — the noise-robust formulation implemented in crates/lawsynth-weakform. It is a boundary specification in the house style: it states what a conforming implementation MUST do.

Motivation

Strong-form SINDy fits Ẋ = Θ(X)Ξ using derivatives Ẋ estimated from the data. Numerical differentiation amplifies observation noise, so the strong form degrades sharply as noise grows. The weak form multiplies the ODE by a smooth, compactly-supported test function φ and integrates by parts, moving the derivative onto the analytic φ:

∫ φ̇_k · x dt  =  ∫ φ_k · Θ(x) dt · Ξ

No derivative of the noisy data is ever taken; only integrals against φ and φ̇.

Requirements

  1. No differentiation of observed data. A conforming weak-form discovery MUST form its targets and library rows from integrals of the data against test functions (and their analytic derivatives), never from a finite-difference / smoothing derivative of the observations.
  2. Compactly-supported test functions. Each φ_k MUST vanish (with its derivative) at the boundary of its subdomain, so integration by parts carries no boundary term. The standard bump φ(t) = (1 − ((t−c)/r)²)^p, p ≥ 2, is sufficient; the implementation MUST document its family, order, and radius.
  3. Determinism. Test-function placement (centers/radii), quadrature, and the sparse solve MUST be deterministic — centers derived from the usable time window by a fixed rule, quadrature a fixed scheme, any sampling seeded from content. Identical inputs MUST yield bit-identical output.
  4. Honest scope. The weak form estimates the SAME law family as the strong form (the sparse coefficient matrix Ξ over the same candidate library); it is a more noise-robust estimator of the same object, not a different model class. The implementation MUST report the number of test functions used and a conditioning/health signal so a caller can judge whether the placed test functions excite the candidate columns independently.
  5. Noise-robustness claim. Any claim that the weak form beats the strong form under noise MUST be backed by a reproducible comparison on the same noisy data (the reference implementation's noise_robustness test does this).

Non-goals

Full PDE / spatiotemporal weak forms and adaptive test-function selection are out of scope for this boundary; they may be added as extensions with their own contracts.