PNAS Nexus · 9 October 2026 · From issue 49
TIP-INN: A thermodynamics-informed machine learning framework for the mechanical behavior of rocks
A new rock-behavior computer model trains much faster and makes far fewer errors than older tools.
Kangan Li, Tushar Mittal et al.
A new computer model learns the hidden rules that control how rocks bend, crack, and flow under stress. It trains using real stress and strain data, and it never breaks the basic energy rules that real rocks must follow. Compared to older computer tools, it trains 10 to 30 times faster. It also makes errors up to 100 times smaller when tested on new conditions it has not seen before.
How much faster and more accurate the new model is compared to older physics-based tools.
Lab data and computer models only, not tested underground yet.
Integrity screen: passed (3 checks) Checked 24 October 2026. Retraction record: none. DOI resolves at doi.org. Metadata record found (PNAS Nexus). Read the source
This screen checks process signals. It does not tell you the findings are true.