Researchers at Microsoft Research have shown that polymer properties can be predicted using only artificial intelligence trained on quantum calculations — with no experimental data in the training. The study, called SimPoly, was published on November 9, 2026, in Nature Communications.
The team developed Vivace, a fast and scalable machine learning force field trained on the quantum-chemical dataset PolyData. With it, the model predicts the measured densities of a wide range of polymers as accurately as, or better than, classical force fields — and it also captures the glass transition signature in the density-temperature curve, which makes it possible to estimate the materials' glass transition temperatures.
A benchmark with 130 polymers
To test the model, the authors created PolyArena, a benchmark with experimental properties for 130 polymers — densities and glass transition temperatures under standard conditions, compiled from the Bicerano handbook. The set covers families such as polyolefins, polyesters, polyacrylates, polycarbonates, polyimides and polystyrenes, with elements including H, C, N, O, F, Si, S and Cl.
The work was led by Gregor N. C. Simm, Jean Hélie and Lixin Sun of Microsoft Research AI for Science in Cambridge, together with colleagues from Microsoft Quantum in Redmond. The research was fully funded by Microsoft Research.
Why it matters
Polymers are in clothes, food, electronics, aerospace components and medical devices, but simulating them is hard: classical methods lose accuracy outside the conditions they were fitted for, and quantum calculations are too expensive for large systems and long timescales. A model that gets measured properties right without relying on experimental data opens the way to designing new materials entirely on the computer — from tougher plastics to lighter components.
Image credit: illustrative image of a polymer-chain simulation (University of Vienna). It is not a figure from the SimPoly paper.
Sources: Nature Communications — SimPoly (DOI: 10.1038/s41467-026-78736-x); arXiv — preprint of the study.
Transparency: this article was produced with the help of artificial intelligence, based on cross-checking public posts on X and sources on the internet, and reviewed against the original sources cited. Please consult the original sources to confirm the details.
By Muse
