Learning the Dynamics of Physical Systems from Sparse Observations with Finite Element Networks
Marten Lienen, Stephan Günnemann
Abstract
We propose a new method for spatio-temporal forecasting on arbitrarily distributed points. Assuming that the observed system follows an unknown partial differential equation, we derive a continuous-time model for the dynamics of the data via the finite element method. The resulting graph neural network estimates the instantaneous effects of the unknown dynamics on each cell in a meshing of the spatial domain. Our model can incorporate prior knowledge via assumptions on the form of the unknown PDE, which induce a structural bias towards learning specific processes. Through this mechanism, we derive a transport variant of our model from the convection equation and show that it improves the transfer performance to higher-resolution meshes on sea surface temperature and gas flow forecasting against baseline models representing a selection of spatio-temporal forecasting methods. A qualitative analysis shows that our model disentangles the data dynamics into their constituent parts, which makes it uniquely interpretable. Our implementation is available at https://www.daml.in.tum.de/finite-element-networks/
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
15 top-tier papers have cited it since. Know what they found before your reviewers do.
Lune can draw on
the full paper · the top-tier papers that cite it · related work
Free to start. No credit card required.
Lineage
- Inducing Point Operator Transformer: A Flexible and Scalable Architecture for Solving PDEs cites Combining Differentiable PDE Solvers and Graph Neural Networks for Fluid Flow Prediction
- From Zero to Turbulence: Generative Modeling for 3D Flow Simulation cites Einops: Clear and Reliable Tensor Manipulations with Einstein-like Notation
- Learning continuous-time PDEs from sparse data with graph neural networks cites Physics-aware Difference Graph Networks for Sparsely-Observed Dynamics
- SineNet: Learning Temporal Dynamics in Time-Dependent Partial Differential Equations cites Transform Once: Efficient Operator Learning in Frequency Domain
- BENO: Boundary-embedded Neural Operators for Elliptic PDEs cites Combining Differentiable PDE Solvers and Graph Neural Networks for Fluid Flow Prediction
Cited by top-tier papers15
- From Zero to Turbulence: Generative Modeling for 3D Flow SimulationMarten Lienen, David Lüdke, Jan Hansen-Palmus, Stephan GünnemannICLR 2024 · 56 citations
- Transform Once: Efficient Operator Learning in Frequency DomainMichael Poli, Stefano Massaroli, Federico Berto, Jinkyoo Park et al.NeurIPS 2022 · 29 citations
- Inducing Point Operator Transformer: A Flexible and Scalable Architecture for Solving PDEsSeungjun Lee, Taeil OhAAAI 2024 · 22 citations
- BENO: Boundary-embedded Neural Operators for Elliptic PDEsHaixin Wang, Jiaxin Li, Anubhav Dwivedi, Kentaro Hara et al.ICLR 2024 · 17 citations
- SineNet: Learning Temporal Dynamics in Time-Dependent Partial Differential EquationsXuan Zhang, Jacob Helwig, Yuchao Lin, Yaochen Xie et al.ICLR 2024 · 14 citations
Builds on5
- Combining Differentiable PDE Solvers and Graph Neural Networks for Fluid Flow PredictionFilipe de Avila Belbute-Peres, Thomas D. Economon, J. Zico KolterICML 2020 · 271 citations
- Einops: Clear and Reliable Tensor Manipulations with Einstein-like NotationAlex RogozhnikovICLR 2022 · 124 citations
- Learning continuous-time PDEs from sparse data with graph neural networksValerii Iakovlev, Markus Heinonen, Harri LähdesmäkiICLR 2021 · 81 citations
- Physics-aware Difference Graph Networks for Sparsely-Observed DynamicsSungyong Seo, Chuizheng Meng, Yan LiuICLR 2020 · 65 citations
- Learning Similarity Metrics for Numerical SimulationsGeorg Kohl, Kiwon Um, Nils ThuereyICML 2020 · 17 citations
Related papers
- Continuous PDE Dynamics Forecasting with Implicit Neural RepresentationsYuan Yin, Matthieu Kirchmeyer, Jean-Yves Franceschi, Alain Rakotomamonjy et al.ICLR 2023 · 14 citations
- PhyMPGN: Physics-encoded Message Passing Graph Network for spatiotemporal PDE systemsBocheng Zeng, Qi Wang, Mengtao Yan, Yang Liu et al.ICLR 2025
- Space and time continuous physics simulation from partial observationsSteeven Janny, Madiha Nadri, Julie Digne, Christian WolfICLR 2024 · 10 citations
- Learning Spatiotemporal Dynamical Systems from Point Process ObservationsValerii Iakovlev, Harri LähdesmäkiICLR 2025
- Neural Dynamics on Complex NetworksChengxi Zang, Fei WangKDD 2020 · 4 citations
