← All AI/ML open problems

Symbolic regression & equation discovery · Time-series learning / Open question

Discovering ODEs Without a Supplied Library

Sparse regression is powerful once the right nonlinear coordinates and terms are supplied. In many applications, discovering that library is the main difficulty. Efficiency payoff: Learn a small nonlinear dynamical law without constructing an exponentially large library or fitting a large black-box simulator.

OpenEditorially scopedFramework needs definitionCompiled 2026-11-20

Seed catalog · Showing the bundled record.

Problem statement

For x˙=f∗(x)\dot x=f_*(x) in a forward-invariant cube, let each coordinate of f∗f_* be a sum of at most ss terms c ψ(a⊤x)c\,\psi(a^\top x), with ψ∈{z,z2,z3,sin⁡z}\psi\in\{z,z^2,z^3,\sin z\} and unknown bounded rational c,ac,a of at most bb bits. From MM length-TT trajectories observed every Δt\Delta t with independent Gaussian measurement noise, learn the active projections/operators and their coefficients without enumerating all candidate aa. Determine finite-data recovery and computation bounds under a quantitative trajectory-separation condition.

Assumptions and scope

  • Initial states are known and sampled independently from a supplied distribution; solutions stay in the cube and vector fields have supplied Lipschitz/derivative bounds. The observed coordinates are the physical state in this starting model; no exact derivatives are available.
  • Identify complete vector fields up to functional equivalence, not an arbitrary decomposition into identical/split terms. Require inequivalent candidates to have a stated positive L2 separation in their sampled trajectory laws; finite sampling may alias distinct continuous-time fields.
  • Specify the neural/KAN representation, weak or integral loss, optimizer and symbolic extraction rule. Charge denoising, quadrature, projection search and integration tolerances. A sparse linear regression on a known correct library is the established baseline.

What would settle it?

A finite-noise, finite-step identification or prediction guarantee with explicit M,T,Delta t,s,b and separation dependence, or a computational/observational lower bound locating a tractable subfamily.

Selected for a theoretical or computational target. A fitted scaling curve or improved benchmark score alone does not establish the stated learning-theoretic guarantee.

Status and evidence

SINDy, derivative-free weak SINDy and weak-surrogate convergence are established. Higher-Order-KANDy already learns flexible terms using delay/jet representations. The added target is a quantitative learned-library and symbolic-recovery guarantee; the finite grammar is an explicit inductive bias. This is an editorial, source-motivated research target. The preliminary review through 2026-11-20 does not certify novelty or exclude an unlocated resolution.

Preliminary source check; not an expert-verified open status or an independent audit of proposed proofs. The compilation date is not a literature-review cutoff.

Sources and further reading

  1. Brunton, Proctor and Kutz — Discovering Governing Equations from Data: Sparse Identification of Nonlinear Dynamical Systems ↗related algorithm · source checked 2026-11-20

    Primary abstract checked: sparse equation discovery, parameterized/time-varying/forced examples and interpretable models are established. A fixed candidate library is an inductive assumption, not automatic discovery of every physical law.

  2. Messenger and Bortz — Weak SINDy: Galerkin-Based Data-Driven Model Selection ↗partial theory · source checked 2026-11-20

    Primary search abstract checked: weak formulations avoid pointwise differentiation and improve noisy coefficient/model recovery. Derivative-free equation learning is already available.

  3. Slote and Fish — Higher-Order Kolmogorov-Arnold Networks for Dynamics ↗related algorithm · source checked 2026-11-20

    Primary search abstract checked: delay/jet embeddings and KAN layers for higher-order noisy dynamics without a fixed candidate library. Coarse/noisy recovery is empirical, not a general finite-data identifiability theorem.

  4. Russo and Laiu — Convergence of Weak-SINDy Surrogate Models ↗partial theory · source checked 2026-11-20

    Primary search abstract checked: projection-based surrogate-dynamics and solution-error analysis under bounded composition-operator assumptions. Basic weak-SINDy convergence is not an open existence problem.