Abstract:Iterative solvers are widely used to accurately simulate physical systems. These solvers require initial guesses to generate a sequence of improving approximate solutions. In this contribution, we introduce a novel method to accelerate iterative solvers for physical systems with graph networks (GNs) by predicting the initial guesses to reduce the number of iterations. Unlike existing methods that aim to learn physical systems in an end-to-end manner, our approach guarantees long-term stability and therefore leads to more accurate solutions. Furthermore, our method improves the run time performance of traditional iterative solvers. To explore our method we make use of position-based dynamics (PBD) as a common solver for physical systems and evaluate it by simulating the dynamics of elastic rods. Our approach is able to generalize across different initial conditions, discretizations, and realistic material properties. Finally, we demonstrate that our method also performs well when taking discontinuous effects into account such as collisions between individual rods. Finally, to illustrate the scalability of our approach, we simulate complex 3D tree models composed of over a thousand individual branch segments swaying in wind fields. A video showing dynamic results of our graph learning assisted simulations of elastic rods can be found on the project website available at <a class="link-external link-http" href="http://computationalsciences.org/publications/shao-2021-physical-systems-graph-learning.html" rel="external noopener nofollow">this http URL</a> .

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to improve the performance of iterative solvers for physical systems, especially by predicting initial guesses through Graph Networks (GNs) to reduce the number of iterations. Traditional iterative solvers require an initial guess when simulating physical systems to generate a series of progressively refined approximate solutions. The performance of these solvers largely depends on the calculation of the initial guess. This paper proposes a novel method that uses graph networks to predict initial guesses, thereby accelerating the iterative process, reducing the required number of iterations, and improving the running efficiency of the solver. In addition, this method ensures long - term stability, can more accurately simulate physical systems, and is applicable to different initial conditions, discretizations, and real - world material properties. In particular, the paper uses Position - Based Dynamics (PBD) as a common physical system solver and verifies the effectiveness of the method by simulating the dynamics of elastic rods. ### Main contributions: 1. **Accelerating iterative solvers**: Demonstrates how to use graph networks to accelerate iterative solvers and reduce the number of iterations by predicting initial guesses. 2. **Long - term stability**: Ensures the long - term stability of the classical PBD solver while improving runtime performance. 3. **Real - world simulation of elastic rods**: Demonstrates the effectiveness of the method by simulating the dynamics of elastic rods. 4. **Wide applicability**: Proves the accuracy and generalization ability of the method in different scenarios and various mechanical properties, including collisions and complex topologies (such as dynamic tree simulations). ### Method overview: - **Graph networks**: Use graph networks to predict the results of constraint projection steps as initial guesses. - **Position - Based Dynamics**: Adopt a PBD solver and satisfy pre - defined constraint conditions through constraint projection steps. - **Constraint projection**: Predict the initial guesses of constraint projection through graph networks, reduce the number of iterations, and improve computational efficiency. ### Experimental results: - **Speed improvement**: As the number of nodes increases, the inference time of COPINGNet decreases significantly, with a speed improvement of up to 50%. - **Generalization ability**: The method shows good generalization ability under different initial conditions, geometries, discretizations, and material parameters. - **Complex systems**: Successfully simulates the swaying of a 3D tree model consisting of more than 1,000 branch segments in a wind field, demonstrating the effectiveness of the method in large - scale systems. ### Conclusion: The method proposed in this paper not only improves the performance of iterative solvers but also ensures long - term stability and is applicable to the simulation of multiple physical systems. By predicting initial guesses through graph networks, the number of iterations is significantly reduced, computational efficiency is improved, and the broad application potential in the simulation of complex physical systems is demonstrated.

Accurately Solving Physical Systems with Graph Learning

Learning Solid Dynamics with Graph Neural Network

Learning Mesh-Based Simulation with Graph Networks

Towards complex dynamic physics system simulation with graph neural ordinary equations

Learning to Simulate Complex Physical Systems: A Case Study

The Novel Graph Transformer-Based Surrogate Model for Learning Physical Systems

Learning Physical Dynamics with Subequivariant Graph Neural Networks

On the Relationships between Graph Neural Networks for the Simulation of Physical Systems and Classical Numerical Methods

Towards Complex Dynamic Physics System Simulation with Graph Neural ODEs

Grounding Graph Network Simulators using Physical Sensor Observations

PhyMPGN: Physics-encoded Message Passing Graph Network for spatiotemporal PDE systems

Accelerating Particle and Fluid Simulations with Differentiable Graph Networks for Solving Forward and Inverse Problems

Constraint-based graph network simulator

GNN-based physics solver for time-independent PDEs

A Scalable Graph Neural Network Method For Developing An Accurate Force Field Of Large Flexible Organic Molecules

Graph Neural PDE Solvers with Conservation and Similarity-Equivariance

Learning Controllable Adaptive Simulation for Multi-resolution Physics

GNS: A generalizable Graph Neural Network-based simulator for particulate and fluid modeling

A Physics-driven GraphSAGE Method for Physical Process Simulations Described by Partial Differential Equations

Physics-informed graph neural network emulation of soft-tissue mechanics

A Physics‐Driven GraphSAGE Method for Physical Field Simulations Described by Partial Differential Equations