conservation constraint enforcement

Existing neural operators often fail to strictly enforce physical conservation laws—such as mass and energy conservation—leading to degraded accuracy, limited generalization, and reliance on problem-specific architectures in time-dependent PDE solving. To address this, we propose a universal plug-and-play neural operator framework that explicitly models and rigorously enforces conservation constraints via an extensible encoder–decoder module for conserved quantities, operating atop arbitrary backbone neural operators without modifying their internal architecture. This is the first approach to decouple conservation enforcement from data-driven learning, providing theoretical guarantees on conservation preservation while enhancing predictive performance. Experiments across diverse PDEs—including adiabatic systems, the shallow water equations, and the Allen–Cahn equation—demonstrate significant improvements in prediction accuracy, strict adherence to multiple conservation laws, and strong cross-domain generalization capability.

Numerical solvers for nonlinear hyperbolic PDEs often suffer from spurious shocks, instability, violation of conservation laws, or convergence to nonphysical solutions. Method: We propose the first end-to-end formal verification framework for such solvers, built on Racket metaprogramming and a custom theorem prover. It integrates symbolic automatic differentiation with floating-point algebra-aware modeling, enabling user-defined physical models and automatic generation of mathematically verified C code. Contribution/Results: The framework formally verifies L² stability, flux conservation, and physical validity, and has been integrated into the Gkeyll multiphysics platform. Experiments demonstrate that the generated solvers achieve both high performance and physical fidelity in strongly nonlinear regimes—including shock waves and turbulence—while providing, for the first time, full formal correctness guarantees from high-order algorithm specification to executable code.

Neural networks solving partial differential equations (PDEs) often violate fundamental physical principles—such as mass conservation, entropy production, positivity, and shock dynamics. To address this, we propose the Constraint Projection Learning (CPL) framework, which encodes conservation laws, the Rankine–Hugoniot condition, entropy conditions, and positivity constraints into differentiable projection operators embedded throughout training. Integrated with total variation diminishing (TVD) regularization and rollout curriculum learning, CPL enforces strict physical compliance at every optimization step. Experiments on the Burgers and Euler equations demonstrate that solutions exhibit exact conservation, bounded total variation, no error accumulation, long-term stability, and machine-precision adherence to physical laws. This work presents the first end-to-end, differentiable, and compact integration of multiple physics-based constraints in neural PDE solvers, significantly enhancing solution reliability and generalizability.

This work addresses the lack of physical consistency in conventional data-driven convolutional neural networks when predicting compressible flow fields governed by conservation laws, particularly their difficulty in accurately capturing nonlinear features such as shock waves. The authors propose an architecture-agnostic physics-regularization framework that, for the first time, embeds the conservation mechanism of the finite volume method directly into the deep learning training process. Specifically, the CNN outputs are interpreted as structured-grid variables, and numerical fluxes are explicitly enforced to satisfy conservation laws. This approach significantly enhances physical fidelity while preserving computational efficiency. In transonic airfoil flow predictions, the method reduces mean drag error by approximately 15%, with improvements reaching up to 34% at extreme angles of attack, demonstrating especially pronounced advantages in low-data regimes.

This work systematically identifies and rectifies seven critical errors in Liu, Madhavan, and Tegmark’s machine learning–based discovery of conservation laws from a one-dimensional damped harmonic oscillator—including mis-specified physical priors, conflation of mathematical definitions of conserved quantities, inappropriate error metrics, and omission of differential equation validation. To address these, we introduce a physics-constrained error analysis framework, rigorous analytical verification, and formal scrutiny of conservation law definitions, thereby falsifying the original method’s applicability to dissipative systems. Our principal contributions are threefold: (i) establishing the first empirically testable theoretical validation standard for ML-driven conservation law discovery; (ii) rigorously distinguishing *invariance* (under symmetry transformations) from *conservation* (time-independence along trajectories); and (iii) proposing a robust modeling paradigm integrating differential geometry and dynamical systems theory—substantially enhancing methodological rigor and reproducibility in physics-guided machine learning.

This work addresses the frequent mismatch between user-specified physical intent and the actual behavior of multiphysics simulation code generated by large language models, often due to erroneous implementations of partial differential equations (PDEs). To bridge this gap, we propose a PDE-structure-based intent verification method that deterministically reconstructs the governing equations implicitly encoded in the generated code and compares them against the user’s intended PDEs, enabling semantic correctness validation and iterative refinement. We introduce, for the first time, a formal metric termed the Intent Fidelity Score (IFS) to quantify alignment with physical intent, establish a PDE-driven feedback loop, and demonstrate compatibility with major PDE frameworks including MOOSE, FEniCS, and FreeFEM. Evaluated on 220 cases in MooseBench, our approach substantially improves IFS—by 0.22–0.41 on challenging instances with initial IFS < 0.7—while audits reveal that execution-only repair strategies still yield physically incorrect results in 39–40% of cases.

研究对比了自动微分与基于离散化的约束在解决偏微分方程中的优劣，通过系统实验揭示了非线性增强时基于离散化方法的优势。

This work addresses the absence of theoretically grounded, real-time solvers for partial differential equations in current scientific machine learning, which hinders reliable simulation validation. The authors propose a physics-informed, data-driven reduced-order model that preserves conservation structures by leveraging exterior calculus to encode topological properties and Gaussian processes to model state-to-flux relationships, yielding a Dirichlet-to-Neumann map with closed-form posterior uncertainty. A novel interface between mixed finite element spaces and Gaussian process regression is established, recasting training as an optimal recovery problem subject to conservation constraints, augmented by an efficient Schur complement strategy. Within a reproducing kernel Hilbert space framework, rigorous posterior error bounds are derived. Numerical experiments demonstrate that the method enables real-time, high-fidelity estimation of boundary fluxes, with its posterior distribution effectively replacing conventional error estimators while offering both interpretability and principled uncertainty quantification.

This work addresses a key limitation of conventional LLM-based PDE solvers, which implicitly embed numerical strategies within generated code, making pre-execution validation and post-failure correction challenging. To overcome this, the authors propose AutoPDE, the first framework to explicitly model solution strategies as revisable, decoupled objects separate from implementation code. AutoPDE employs a three-stage pipeline—PDE type identification, numerical method selection, and adaptive parameter tuning—augmented by low-overhead trial solves and a reusable skill library to construct and refine strategies prior to code generation. Evaluated on the PDE Agent Bench, AutoPDE achieves a 54.5% pass rate, outperforming the strongest baseline by 14.2 percentage points, thereby substantially improving both the reliability and interpretability of AI-driven PDE solving.

This work addresses the high computational cost of traditional numerical methods for solving parametric hyperbolic conservation laws and the tendency of existing neural surrogates to produce non-physical solutions, suffer from unstable rollouts, and exhibit poor generalization due to neglecting underlying physical structures. The authors propose a structure-preserving graph neural network (GNN) solver that explicitly formulates the GNN as a learnable conservative reconstruction and upwind flux operator, inherently satisfying local conservation and upwinding properties. By integrating this formulation within an ADER framework, the method enables high-order spatiotemporal predictions while maintaining physical consistency and supporting stable long-time integration with large time steps. On challenging supersonic flow benchmarks, the approach significantly outperforms both low-order numerical schemes and current neural surrogates in long-term prediction accuracy and achieves speedups of several orders of magnitude over conventional high-resolution simulations.