PDE Surrogate Modeling
PDE: Partial Differential Equation
Overview
PDE surrogate modeling studies how to approximate the mapping from physical conditions, geometry, and parameters to a PDE’s solution, reducing the cost of repeated simulations. Methods range from reduced-order models such as POD and latent dynamics to neural operators such as Fourier Neural Operators and DeepONets, as well as mesh-based graph networks, transformers, and generative models. Many approaches combine these learned models with physical structure, including conservation laws, symmetries, boundary conditions, and numerical solvers, to improve reliability and support inverse problems or design optimization. A central challenge is balancing speed with accuracy under new geometries or regimes, stable long rollouts, fine-scale fidelity, and calibrated uncertainty; whether a surrogate saves time overall depends on training and deployment costs as well as inference speed.
Latest papers
366 papers38 papers in the last four weeks, down 31% on the four weeks before. 0.3% of all new papers.
Surrogate Modeling of 3D Rayleigh-Benard Convection with Equivariant Autoencoders
Latent Evolution Operator (LEO): State-to-State Forecasting of Parametric PDEs
Fast localized urban microclimate emulator: wind and temperatures in unseen urban morphologies
Residuals Are Not Enough: Limits of Physics-Informed Pre-Training for Scientific Foundation Models
nice,'' e.g., structured, well-aligned settings: it enhances generalization and reduces data dependence, compared to data-only pre-training. However, these advantages diminish significantly as the downstream tasks become harder,'' e.g., as they involve discontinuities or deviate from the pre-training distribution. In complex or structurally different problems, such as those involving new boundary conditions or PDE operators, physics-informed models may perform only on par with---or even worse---than data-driven baselines. While residual-based pre-training helps in idealized regimes, realizing broadly transferable SciFMs will likely require subtler spatiotemporal inductive biases and more principled integration of physical knowledge into model architectures.