Learning Autoregressive Neural Operators on Solver-Defined Physical Constraint Manifolds
Organizations: School of Computer Science, Sichuan University, Chengdu, China · National Key Laboratory of Fundamental Algorithms and Models for Engineering Numerical Simulation, Chengdu, China
Abstract
Autoregressive neural operators approximate time-dependent PDE solvers by repeatedly composing a learned transition, although the resulting states need not remain feasible under the discrete constraints of the numerical problem. We study this issue for incompressible flow, where each velocity state must satisfy solver-level incompressibility together with prescribed boundary values. We formulate autoregressive operator learning as a map between solver-defined affine feasible sets and introduce PCM-NO, a projected-residual realization of this feasible-transition principle. The neural backbone predicts a candidate state increment, a fixed tangent projection extracts its admissible component, and a metric retraction maps the advanced state to the prescribed feasible set. With exact affine constraint solves, the resulting transition is closed on the feasible set, while an affine identity separates admissible-update parameterization from state-level closure. With approximate solves, tangent-only residuals admit a cumulative upper bound, whereas per-step retraction resets feasibility to the residual achieved by the current solve. Experiments on periodic Navier--Stokes flow and lid-driven cavity flow demonstrate the superiority of PCM-NO in preserving solver-level constraints over long autoregressive rollouts and under parameter, boundary-value, and resolution shifts, while achieving lower rollout errors and better flow-specific diagnostics across both in-distribution and stress settings.
Figures & tables
| Dataset | Method | |||||
| P-NS | FNO | N.F. | N.F. | — | ||
| DivReg | — | |||||
| PINO | — | |||||
| ClawNO | — | |||||
| FinalProj | — | |||||
| PCM-NO | — |
| Settings | Method | Energy/Profile Err. | |||
| P-NS-Re | FNO | N.F. | N.F. | — | N.F. |
| DivReg | — | ||||
| PINO | — | ||||
| FinalProj | — | ||||
| PCM-NO | — | ||||
| LC-Re | FNO |
| Dataset | Raw RelErr. | Projected RelErr. | Normal Frac. (%) | Pythag. Gap |
| P-NS | ||||
| LC |
| Variant | ||||
| TangentOnly | ||||
| RetractOnly | ||||
| DivOnly | ||||
| PCM-NO |
Appendix figures & tables5 assets
Supplementary material from the paper’s appendix.
Appendix
| Setting | P-NS | LC |
| Domain | ||
| Generator | Dealiased pseudo-spectral RK4 solver | MAC-grid projection solver |
| State representation | Velocity field | Staggered velocity field |
| Spatial resolution | ||
| Stored states per trajectory | ||
| One-step transitions per trajectory |
| Setting | P-NS | LC |
| Spatial resolution | ||
| Train/validation/test trajectories | ||
| FNO layers / width / Fourier modes | ||
| Optimizer | AdamW | AdamW |
| Learning rate / weight decay | ||
| Maximum epochs |
| Dataset | Batch | Method | Step ms/traj. | 50-step ms/traj. | Speedup vs. solver | Cost vs. FNO |
| P-NS | 1 | Reference solver | – | |||
| 1 | FNO | |||||
| 1 | FinalProj | |||||
| 1 | PCM-NO | |||||
| 8 | Reference solver | – | ||||
| 8 | FNO |
| Method | Time/step (ms) | |||
| ClawNO | ||||
| FinalProj | ||||
| PCM-NO |
PDE Operator Learning
Partial differential equation (PDE) operator learning builds models that map inputs such as coefficients and initial conditions to solution fields.
See all 250 papers