Interpretable AI reveals hidden structure in nuclear masses

Abstract Predicting nuclear masses with high precision remains a central challenge in nuclear physics, with implications ranging from stellar nucleosynthesis to determining the boundaries of the nuclear landscape. Approaches using artificial intelligence (AI) have recently achieved remarkable accuracy on this task, but their usefulness is limited by their black-box nature and by the difficulty of assessing their reliability when extrapolating beyond measured nuclei. Here, we present an interpretable AI model tha…

By the Truvace deskOctober 22, 20262 min read1 sourcePeer-reviewedTRV-2026-2180· cite & verify

Interpretable AI reveals hidden structure in nuclear masses — Binding energy. Public domain

In brief

Researchers built an AI model for predicting nuclear masses that is designed to be interpretable rather than a black box. Its accuracy equals the best existing methods, and its internal representations take the form of a double helix linking proton and neutron embeddings to the most stable nuclei in each isotopic chain.

The work matters because earlier AI approaches were accurate but hard to trust when extrapolating beyond measured nuclei. Being able to read the learned structure offers a route to checking reliability and extending nuclear models. The abstract reports no tests on unmeasured nuclei, so how well the findings generalize remains open.

Main points

The model matches the best existing precision for predicting nuclear masses.
Its learned representations form a double helix linking proton and neutron embeddings.
Links in the helix correspond to the most stable nuclei within each isotopic chain.
The AI autonomously recovered the liquid-drop model's macroscopic components.
It rediscovered localized Jaffe factorization, which accounts for much of its performance gains.

The gain

An interpretable AI model matched the best existing nuclear mass prediction precision while revealing physical structure in binding energies, including a rediscovered principle called localized Jaffe factorization.

The rundown

Predicting nuclear masses precisely is a central challenge in nuclear physics, with implications from stellar nucleosynthesis to the boundaries of the nuclear landscape. Prior AI approaches reached remarkable accuracy but were limited by their black-box nature.

The authors analyzed the model's learned features and found that it reproduced established macroscopic components such as the liquid-drop model without being told to do so. It also surfaced a long-forgotten principle, localized Jaffe factorization, that explains much of its improved performance.

The authors conclude that interpretable AI can reproduce and extend nuclear models by uncovering latent physical symmetries directly from data.