New SpeND method uses neural networks for accurate mesh-free physics simulations

Researchers have developed a novel method called Spectral-like Neural Discretisation (SpeND) to create mesh-free numerical operators for physics simulations. This technique uses a neural network to learn stencil weights, conditioned on local node geometry, to better approximate spectral operators. A key feature is a projection layer that ensures polynomial consistency by construction, eliminating the need for reference solutions and making the training physics-agnostic. Modal analysis indicates that SpeND operators exhibit superior accuracy over a wider wavenumber band compared to traditional methods like LABFM or finite differences on structured grids, while maintaining fourth-order convergence. AI

IMPACT SpeND offers a more accurate and efficient approach to mesh-free simulations, potentially advancing fields requiring high-fidelity physics modeling.

RANK_REASON The cluster contains an academic paper detailing a new computational method. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• LABFM
• Lucas Gerken Starepravo
• Spectral-like Neural Discretisation
• Sphaeria

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →