One model simulates sand, fluids and proteins, particle by particle

Researchers at MIT and the University of British Columbia say their WorldParticle system uses one method to simulate materials from sand to proteins. The accuracy and speed results come from the team's own demonstrations.

Share
One model simulates sand, fluids and proteins, particle by particle
Illustration: AI-generated

Researchers say one AI system can simulate a pile of sand collapsing and a protein changing shape, using the same method for both. The team, at MIT in the United States, the University of British Columbia (UBC) and elsewhere, calls it WorldParticle and describes it in an MIT News article dated 9 November 2026.

Scientists usually have to pick a separate tool for each job, such as a solver built for flowing water or one for protein folding. Each is hand-coded around its own equations. The team says it has replaced that patchwork with one AI system.

WorldParticle breaks an object into thousands or millions of tiny 3D dots, called particles. A neural network, a program built from many simple connected calculations, takes in their positions and physical properties and outputs them for the next frame, one step at a time.

Peter Yichen Chen, a former MIT postdoctoral researcher who is now an assistant professor at UBC, compares the design to ChatGPT. Instead of predicting the next word, he says, it predicts the next movement of the particles.

In the researchers' own demonstrations, WorldParticle modelled six different physical phenomena, including deformable objects (things that change shape), granular materials (such as sand) and proteins. They say it was more accurate than numerical simulation methods, the standard programs that solve the physics equations.

In one of the team's demonstrations, it showed a granular object collapsing into a pile, while some comparison methods left sand particles floating in the air.

In the team's own tests, the method modelled how protein structures change over time about 100 times faster than numerical simulators, says Minghao Guo, an MIT postdoctoral researcher and co-lead author.

Large objects can mean millions of particles, which takes a great deal of computing. The team used a merging technique that combines and compresses particles while keeping fine detail.

The MIT article names possible uses: simulating sloshing oil hitting an elastic valve in a storage tank, or studying how sand or soil moves when a soft robot touches it.

The work is in an open-access paper that will be presented at the ACM SIGGRAPH Conference and Exhibition on Computer Graphics and Interactive Techniques in Asia.

"We want to have a GPT moment for simulation," Chen says. His vision is one general model that works for any scenario without coding a new simulator. "This is a first step in that direction."

The team says it next wants to adapt WorldParticle to other systems, including tasks where soft robots handle things, RNA evolution and merging galaxies. It also wants to build towards a full, general-purpose set of simulation tools for scientific discovery.

Sources