Plasma is a gas in which electrons are stripped from atoms, behaving like a soup of charged particles. To predict how such a soup flows and heats up, you need to know how particles collide and exchange energy. Previously, formulas derived for simple cases were used, but real plasma is more complicated: particles interact with many neighbors at once, and energy transfer is uneven.

In the new work, a computer learned the rules of collisions on its own by watching microscopic simulations where each particle is individually visible. The result is a model that captures subtleties missed by old equations. It was packaged into a compact form for fast computation. Tests showed that the model accurately predicts how plasma conducts heat and responds to changes at various densities and temperatures.

This approach helps bridge the world of individual particles with the behavior of large plasma volumes. It matters for fusion reactors, space plasma, and other problems where old models fail.