Selected work

Representative Papers

Physics-Informed Coarsening for Multigrid Graph Neural Surrogates

May 29, 2026

This work proposes a multigrid graph neural network surrogate model for complex solid mechanics problems involving nonlinear elasticity, plasticity, and transient dynamics. The method employs an encoder-processor-decoder architecture and introduces a novel node-scoring mechanism based on physical residuals to drive adaptive coarsening on unstructured graphs, preferentially preserving regions of high strain or stress concentration in multiscale modeling. In contrast to conventional geometry-based downsampling heuristics, this strategy effectively maintains long-range interactions, significantly enhancing the stability and accuracy of long-time rollouts. Experimental results demonstrate that, across a range of linear and nonlinear transient scenarios, the proposed approach consistently outperforms standard sampling baselines in both predictive accuracy and rollout stability.

0 citationsRead paper

Mesh Based Simulations with Spatial and Temporal awareness

May 02, 2026

This work addresses key limitations of existing machine learning approaches in computational fluid dynamics, which are constrained by node-level supervision and explicit Euler time stepping, rendering them ill-suited for capturing stiff dynamics and flux continuity inherent in partial differential equations. To overcome these challenges, the authors propose a unified framework that integrates geometric deep learning with numerical analysis principles: stencil-level supervision enforces spatial derivative consistency, temporal cross-attention replaces unstable explicit integration schemes, and a novel 3D rotational positional encoding (RoPE) captures rotational symmetries on unstructured meshes. The resulting method achieves substantial improvements in accuracy and stability for long-horizon rollouts across diverse physical datasets and backbone architectures, while also generalizing effectively to unseen subtasks such as wall shear stress and pressure prediction.

0 citationsRead paper

Recent publications

Latest Papers

Physics-Informed Coarsening for Multigrid Graph Neural Surrogates

May 29, 2026

This work proposes a multigrid graph neural network surrogate model for complex solid mechanics problems involving nonlinear elasticity, plasticity, and transient dynamics. The method employs an encoder-processor-decoder architecture and introduces a novel node-scoring mechanism based on physical residuals to drive adaptive coarsening on unstructured graphs, preferentially preserving regions of high strain or stress concentration in multiscale modeling. In contrast to conventional geometry-based downsampling heuristics, this strategy effectively maintains long-range interactions, significantly enhancing the stability and accuracy of long-time rollouts. Experimental results demonstrate that, across a range of linear and nonlinear transient scenarios, the proposed approach consistently outperforms standard sampling baselines in both predictive accuracy and rollout stability.

0 citationsRead paper

Mesh Based Simulations with Spatial and Temporal awareness

May 02, 2026

This work addresses key limitations of existing machine learning approaches in computational fluid dynamics, which are constrained by node-level supervision and explicit Euler time stepping, rendering them ill-suited for capturing stiff dynamics and flux continuity inherent in partial differential equations. To overcome these challenges, the authors propose a unified framework that integrates geometric deep learning with numerical analysis principles: stencil-level supervision enforces spatial derivative consistency, temporal cross-attention replaces unstable explicit integration schemes, and a novel 3D rotational positional encoding (RoPE) captures rotational symmetries on unstructured meshes. The resulting method achieves substantial improvements in accuracy and stability for long-horizon rollouts across diverse physical datasets and backbone architectures, while also generalizing effectively to unseen subtasks such as wall shear stress and pressure prediction.

0 citationsRead paper