Simulating particles is a relatively easy task if the particles are spherical. However, in the real world, most particles are not perfectly spherical, but rather have irregular shapes and sizes. Simulating these particles becomes a much more difficult and time-consuming task.
The ability to simulate particles is crucial to understanding their behavior. For example, microplastics have become a new form of pollution due to the proliferation of plastic waste and their uncontrollable breakdown in the environment, either by mechanical means or due to degradation by ultraviolet light. These very small particles are now found almost everywhere around the world. Understanding more about these particles and their behavior is crucial to solving this environmental crisis.
To address this challenge, researchers at the University of Illinois at Urbana-Champaign trained a neural network to predict interactions between irregularly shaped particles and accelerate molecular dynamics simulations. Their method runs simulations up to 23 times faster than traditional simulation methods and can be applied to any irregular shape, given enough training data.
The research, titled “Molecular Dynamics Simulation of Anisotropic Particles Accelerated by Neural Network Predicted Interactions,” Journal of Chemical Physics.
“Microplastics are now everywhere in the environment, and most of them are not spherical, but very heterogeneous, with corners and edges. To address the question of how microplastics behave in the environment, we need to develop new methods and find faster, cheaper and more efficient ways to simulate them,” says Antonia Statt, professor of materials science and engineering.
Spheres are easy to simulate because the only parameter needed to determine how two particles interact is the distance between the centers of the spheres. To move from spheres to more complex shapes such as cubes or cylinders, you need to know not only how far the two particles are from each other, but also the angle and relative position of each particle. For example, the traditional way to simulate a cube is to build the cube out of many smaller spheres.
“Tessellating a cube with little spheres is a very roundabout way of doing things,” Statt explains. “And it's expensive, because you have to calculate the interactions of all the little spheres. To get around that, we used machine learning, or feed-forward neural nets, which in other words say, 'Let me fit some complex function that we don't know.' Neural nets are really good at that. If you give them enough data, they can fit anything they want.”
Using this method, we don't need to calculate all the distances between the little spheres individually – it's much easier and faster, since we only need the distance between the centers of the cubes and their relative orientations. Moreover, this method is just as accurate as the traditional method – we can't improve its accuracy, since it's trained on data generated by the traditional method, but we can improve its efficiency.
In the future, Statt hopes to be able to simulate more complex irregular shapes, or a mix of different shapes, like a cube and a cylinder instead of two cubes. “You'd have to learn all the individual interactions, but this method is general enough that we should be able to do that,” Statt says.
Other contributors to the study include B. Ruşen Argun (Department of Mechanical Engineering, University of Illinois) and Yu Fu (Department of Physics, University of Illinois).
More information:
B. Ruşen Argun et al., Accelerating molecular dynamics simulations of anisotropic particles with neural net-predicted interactions, Journal of Chemical Physics (2024). DOI: 10.1063/5.0206636
Courtesy of the University of Illinois Grainger School of Engineering
Citation: Using Machine Learning to Speed Up Simulations of Irregularly Shaped Particles (August 26, 2024) Retrieved August 26, 2024 from https://phys.org/news/2024-08-machine-simulations-irregularly-particles.html
This document is subject to copyright. It may not be reproduced without written permission, except for fair dealing for the purposes of personal study or research. The content is provided for informational purposes only.