OnlyDense: Reduced-Order Modeling for Lagrangian simulation

Researchers introduce OnlyDense, a machine learning framework that reduces computational costs for Lagrangian particle simulation methods like SPH and MPM by representing massive particle systems as functions in Hilbert space rather than discrete particle sets. The method achieves 0.99+ R² accuracy using just 32 basis functions on million-particle simulations, combining classical reduced-order modeling with deep learning.

OnlyDense addresses a critical computational bottleneck in scientific simulation by proposing a novel approach to modeling large-scale particle systems. Traditional Lagrangian methods like Smooth Particle Hydrodynamics and Material Point Methods provide physical accuracy but become prohibitively expensive when simulating complex phenomena—void growth in materials, hypervelocity impacts on spacecraft, or extreme deformation events. The breakthrough lies in shifting from discrete particle representation to continuous function representation in Hilbert space, enabling dramatic dimensionality reduction without sacrificing accuracy.

This work builds on decades of reduced-order modeling research but innovates by combining it with learned neural basis functions. Unlike graph-based approaches that track individual particles or nonlinear latent-space methods requiring iterative optimization, OnlyDense enables direct projection and explicit basis function access. The framework essentially learns spatial modes analogous to Proper Orthogonal Decomposition, creating an interpretable latent representation where coefficients have physical meaning.

The implications extend across computational physics, materials science, and engineering simulation. Reduced computational requirements enable faster design iteration, more accessible simulations for researchers without massive computing budgets, and potential deployment of complex physics predictions in resource-constrained environments. The demonstrated performance—0.99+ R² with 32 basis functions on million-particle systems—suggests practical applicability to real industrial problems.

Future development will likely focus on extending the method to multi-physics problems, exploring generalization across different simulation parameters, and investigating whether learned basis functions transfer between related systems. The work represents meaningful progress toward making expensive high-fidelity simulations practical for broader applications.

• →OnlyDense reduces computational cost for million-particle simulations using learned neural basis functions with 0.99+ accuracy
• →The method treats particle systems as continuous functions in Hilbert space rather than discrete particles, enabling direct dimensionality reduction
• →Framework unifies classical reduced-order modeling with deep learning while maintaining invariance to discretization point count
• →Achieves extreme compression: 32 basis functions replace millions of particles while preserving dynamic event fidelity including fragmentation
• →Addresses major computational barriers in materials science and spacecraft engineering simulations of extreme deformation events