mlsanityData to 14 Oct 2026

Derived Fields Preserve Fine-Scale Detail in Budgeted Neural Simulators

arXiv 2603.29224 ↗First posted 31 Mar 2026cs.LG cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

Fine-scale-faithful neural simulation under fixed storage budgets remains challenging. Many existing methods reduce high-frequency error by improving architectures, training objectives, or rollout strategies. However, under budgeted coarsen-quantize-decode pipelines, fine detail can already be lost when the carried state is constructed. In the canonical periodic incompressible Navier-Stokes setting, we show that primitive and derived fields undergo systematically different retained-band distortions under the same operator. Motivated by this observation, we formulate Derived-Field Optimization (DerivOpt), a general state-design framework that chooses which physical fields are carried and how storage budget is allocated across them under a calibrated channel model. Across the full time-dependent forward subset of PDEBench, DerivOpt not only improves pooled mean rollout nRMSE, but also delivers a decisive advantage in fine-scale fidelity over a broad set of strong baselines. More importantly, the gains are already visible at input time, before rollout learning begins. This indicates that the carried state is often the dominant bottleneck under tight storage budgets. These results suggest a broader conclusion: in budgeted neural simulation, carried-state design should be treated as a first-class design axis alongside architecture, loss, and rollout strategy.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 26 of its 50 references that are in the map. First posted on arXiv 31 Mar 2026.

Builds on

26 of 50 references.

References · 26

  1. 2020Fourier Neural Operator for Parametric Partial Differential Equationscited by 2,908
  2. 2020Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domainscited by 2,200
  3. 2018On the Spectral Bias of Neural Networkscited by 1,341
  4. 2018Variational image compression with a scale hyperpriorcited by 935
  5. 2020Learning Mesh-Based Simulation with Graph Networkscited by 868
  6. 2021Physics-Informed Neural Operator for Learning Partial Differential Equationscited by 527
  7. 2021Machine learning accelerated computational fluid dynamicscited by 471
  8. 2022PDEBENCH: An Extensive Benchmark for Scientific Machine Learningcited by 413
  9. 2022Message Passing Neural PDE Solverscited by 386
  10. 2019Machine Learning for Fluid Mechanicscited by 385
  11. 2021Factorized Fourier Neural Operatorscited by 249
  12. 2021Multiwavelet-based Operator Learning for Differential Equationscited by 234
14 more
  1. 2022U-NO: U-shaped Neural Operatorscited by 230
  2. 2023PDE-Refiner: Achieving Accurate Long Rollouts with Neural PDE Solverscited by 216
  3. 2021Enhancing Computational Fluid Dynamics with Machine Learningcited by 138
  4. 2022MultiScale MeshGraphNetscited by 98
  5. 2018Latent-space Physics: Towards Learning the Temporal Evolution of Fluid Flowcited by 97
  6. 2023β-Variational autoencoders and transformers for reduced-order modelling of fluid flowscited by 77
  7. 2020Deep Neural Networks for Nonlinear Model Order Reduction of Unsteady Flowscited by 36
  8. 2024Toward a Better Understanding of Fourier Neural Operators from a Spectral Perspectivecited by 25
  9. 2023A Neural PDE Solver with Temporal Stencil Modelingcited by 22
  10. 2024Pi-fusion: Physics-informed diffusion model for learning fluid dynamicscited by 9
  11. 2023HelmFluid: Learning Helmholtz Dynamics for Interpretable Fluid Predictioncited by 7
  12. 2025LOGLO-FNO: Efficient Learning of Local and Global Features in Fourier Neural Operatorscited by 6
  13. 2025FD-Bench: A Modular and Fair Benchmark for Data-driven Fluid Simulationcited by 5
  14. 2019Learning Generalized Quasi-Geostrophic Models Using Deep Neural Numerical Modelscited by 3

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Towards Infinitely Long Neural Simulations: Self-Refining Neural Surrogate Models for Dynamical Systems
  2. 2026Sharp Stability Threshold and Certification for Designing Stable Residual Architectures
  3. 2025Tensor-Compressed and Fully-Quantized Training of Neural PDE Solvers
  4. 2026Hybrid Neural World Models
  5. 2026Bias-Constrained Diffusion Schedules for PDE Emulations: Reconstruction Error Minimization and Efficient Unrolled Training
  6. 2026Neither Precision Nor Architecture Alone: Controlled Tests of Failure Remedies for Physics-Informed Neural Networks
  7. 2026Learning to Discretize: Diffusion-Based Adaptive Mesh with Spectral Guidance
  8. 2025Expressive Power of Implicit Models: Rich Equilibria and Test-Time Scaling