← All AI/ML open problems

Symbolic regression & equation discovery · Statistical inference / Open question

Which PDE Can Coarse Measurements Reveal?

A compact equation fitted to coarse observations may describe only the discretization or unresolved fluctuations, rather than the underlying constitutive law. Efficiency payoff: Identify when a small reusable PDE model can replace expensive field-level prediction without requiring unnecessarily fine observations.

OpenEditorially scopedFramework needs definitionCompiled 2026-11-20

Seed catalog · Showing the bundled record.

Problem statement

On the periodic unit interval, consider ut=∂x(D(u)ux)+R(u)u_t=\partial_x(D(u)u_x)+R(u), where D,RD,R are degree-qq polynomials with at most ss total nonzero coefficients, bounded by BB, and D(z)≥λ>0D(z)\geq\lambda>0 on the observed amplitude range. Observe noisy cell averages on spacing hh at time interval Δt\Delta t for MM initial conditions. Determine the minimax sample/resolution requirements for identifying coefficient support and estimating D,RD,R to error ε\varepsilon, including the bias between a coarse effective equation and the underlying continuum PDE.

Assumptions and scope

  • Specify smooth periodic initial conditions, observation duration, Gaussian measurement-noise variance and a uniform solution-regularity bound on that duration. Use coefficient l2 error and a minimum nonzero magnitude beta_min for support recovery.
  • A lower restricted singular-value bound kappa for the noiseless weak-form feature matrix across experiments is an explicit identifiability condition, not something inferred merely from having many grid points. A traveling wave or one constant state need not distinguish different PDEs.
  • Account for nonlinear averaging: D(average u) is not generally average D(u). Charge weak-test selection, denoising, quadrature, temporal aliasing and any neural meta-training. State the attainable error floor at fixed h and Delta t rather than replacing cell averages by exact point values.

What would settle it?

A finite-resolution confidence/support-recovery theorem with matching ambiguity or lower bounds, together with a costed estimator. Show which additional experiments or resolutions distinguish continuum coefficients from coarse closures.

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

PDE-FIND already uses multiple initial conditions to disambiguate laws; WSINDy supplies noise-robust weak PDE discovery and efficient convolutions. UNIDE already amortizes term selection across PDE families. The residual target is the joint finite-noise, coarse-observation and continuum-identification frontier, not the invention of PDE learning. 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. Rudy et al. — Data-Driven Discovery of Partial Differential Equations ↗related algorithm · source checked 2026-11-20

    Primary abstract checked: sparse derivative/nonlinearity discovery and multiple initial conditions to disambiguate equations. Fitting a single trajectory need not uniquely determine a PDE.

  2. Messenger and Bortz — Weak SINDy for Partial Differential Equations ↗partial theory · source checked 2026-11-20

    Primary search abstract checked: convolutional weak forms, noise robustness and FFT complexity already exist. Coarse-cell nonlinear closure and finite-sample identification must be analyzed rather than assumed away.

  3. Xie — UNIDE: Cross-PDE Learning for Noise-Robust Discovery of Governing Equations ↗related algorithm · source checked 2026-11-20

    Primary search abstract checked: amortized denoising, term selection and weak-form regression across six PDE families. The high-noise recovery figures are synthetic-benchmark results; they are not uniform support/continuum guarantees.