Researchers have built a machine‑learned model that predicts how atoms in CrCoNi alloys interact. This model aims to let scientists run large-scale atom-by-atom simulations while keeping accuracy close to that of expensive quantum calculations. It is designed to work across many different mixtures of chromium, cobalt and nickel, not just the special case where the three elements are equal.

The team created a general‑purpose interatomic potential using the neuroevolution potential (NEP) framework. An interatomic potential is a mathematical model that gives the energy and forces between atoms so that computers can simulate materials. NEP was trained on a wide dataset that includes pure chromium, cobalt and nickel, many binary and ternary alloy compositions, different crystal structures and temperatures. The training data came from spin‑polarized first‑principles (ab initio) calculations, which are quantum mechanical results that include magnetic effects.

At a high level, the model replaces repeated costly quantum calculations with a fast learned function that still matches those calculations closely. According to the paper, the model reproduces many key properties: equations of state (how energy changes with volume), phonons (vibrations in the crystal), elastic constants, how dislocations split, surface and defect energies, melting temperatures and strain‑driven phase changes. It also captures short‑range chemical ordering — tiny local preferences in which atom sits next to which — and how that ordering changes the stacking fault energy. Those stacking fault results agree with both first‑principles calculations and experiments, and they hold for both equimolar and non‑equimolar alloys.

This matters because CrCoNi medium‑entropy alloys are chemically complex and have low stacking fault energies, features that make them hard to model with traditional potentials. Most existing models focus only on equimolar alloys or do poorly for the pure elements. The new potential claims consistent accuracy across the full compositional space while running much faster than quantum methods. That should enable larger and longer atomistic simulations and help researchers explore how changing composition affects mechanical behavior and design non‑equimolar alloys.