New Particle Simulation

 

Making particles in a lab is relatively simple when the particles are spherical. Real world particles are irregular and have varying size. Making these particles is laborious and slow. Simulating particles is important for understanding their behavior.

The newest form of pollution, microplastics, are found everywhere in the world. We must understand how the particles look and behave to understand how to eliminate them.

Scientists at the University of Illinois Urbana-Champaign are working to combat this challenge. They have used networks to predict interactions between irregular particles. This is to accelerate molecular dynamics simulations. The simulations are 23 times faster than normal testing and can be applied to any nonstandard shape. 

The research was published in the journal Chemical Physics. It is titled, “Molecular dynamics simulations of anisotropic particles accelerated by neural-net predicted interactions.”

Antonia Statt is a professor of materials science and engineering. She states, “Microplastics are now present everywhere in the environment and most of them are not spheres, they are very heterogeneous and they have corners and edges. Tackling the problem of how they behave in the environment requires us to develop new methods, finding ways to simulate them faster, cheaper and more efficiently.”

Spheres are relatively easy to simulate. This is because the only thing needed to determine how two particles interact is the distance between the sphere’s centers. When cubes or cylinders are used, all of the angles and relative positions must be known of each particle. The orthodox technique of creating cubes requires building the cube out of many tiny spheres.

Start reports, “it’s a very roundabout way of describing a cube, to tessellate it with small spheres. It’s also expensive because you have to calculate the interactions of all the little spheres with each other. To bypass that, we used machine learning, a feed-forward neural net- which is a fancy way of saying ‘let’s fit a complicated function that we don’t know.’ And neural nets are really good at that. If you provide them enough data, they can fit anything you like.”

Using machine learning, all the distances of the little spheres don’t have to be calculated. Only the center to center distance and relative orientation is required. This makes calculations faster and easier.

Other researchers on this work include B. Rusen Argun and Yu Fu, both from the University of Illinois.