arXiv Science⌕ Search
Home/Archive/arXiv

arXiv · 2603.12183

Proof-Carrying Materials: Falsifiable Safety Certificates for Machine-Learned Interatomic Potentials

Abhinaba Basu·Pavan Chakraborty
View original source ↗Read source document ↗Export metadata ↓

Abstract

Machine-learned interatomic potentials (MLIPs) are deployed for high-throughput materials screening without formal reliability guarantees. We show that a single MLIP used as a stability filter misses 93% of density functional theory (DFT)-stable materials (recall 0.07) on a 25,000-material benchmark. Proof-Carrying Materials (PCM) closes this gap through three stages: adversarial falsification across compositional space, bootstrap envelope refinement with 95% confidence intervals, and Lean 4 formal certification. Auditing CHGNet, TensorNet and MACE reveals architecture-specific blind spots with near-zero pairwise error correlations (r <= 0.13; n = 5,000), confirmed by independent Quantum ESPRESSO validation (20/20 converged; median DFT/CHGNet force ratio 12x). A risk model trained on PCM-discovered features predicts failures on unseen materials (AUC-ROC = 0.938 +/- 0.004) and transfers across architectures (cross-MLIP AUC-ROC ~ 0.70; feature importance r = 0.877). In a thermoelectric screening case study, PCM-audited protocols discover 62 additional stable materials missed by single-MLIP screening - a 25% improvement in discovery yield.

Explore related subjects

cond-mat.mtrl-scics.AIcs.LGphysics.comp-ph

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Abhinaba Basu, Pavan Chakraborty. 2026-03-13. Proof-Carrying Materials: Falsifiable Safety Certificates for Machine-Learned Interatomic Potentials. https://arxiv.org/abs/2603.12183

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

A hierarchy of thermodynamics learning frameworks for inelastic constitutive modeling

Recent advances in physics-augmented neural networks have enabled thermodynamically consistent data-driven constitutive modeling of complex inelastic materials. Most existing approaches, however, implicitly adopt a specific thermodynamic framework and embed structural assumptions such as normality, dual dissipation potentials, or other structure from manually constructed models directly into the learning architecture. Consequently, differences in predictive performance may arise not only from data or network design, but also from the underlying theoretical assumptions. In this work, we present a unified comparison of several thermodynamically consistent inelastic modeling frameworks from a machine learning perspective. We consider internal-variable formulations with dissipation potential, generalized standard materials, and metriplectic structures, and we analyze their structural assumptions, admissible dependencies, convexity requirements, and implications for dissipation and evolution. Each framework is implemented within a common neural potential architecture based on invariant representations and neural ordinary differential equations. This unified setting ensures that performance differences can be attributed to thermodynamic structure rather than architectural variation. The models are trained and evaluated on three representative inelastic datasets generated from high-fidelity representative volume element simulations: an elastoplastic alloy, a viscoelastic composite, and a rate-dependent crystal plasticity polycrystal. By isolating the role of thermodynamic structure, we assess how restrictions such as duality, normality, operator-based evolution, and convexity influence learnability, expressiveness, stability, and generalization.

cond-mat.mtrl-sci↗
arXiv2026

Berry Curvature Driven Transport in Silicon-Compatible Altermagnetic $α$-MnTe Thin Films

Integrating spin-dependent functionality with mainstream semiconductor technology is a central goal of modern spintronics, yet most candidate materials remain incompatible with silicon-based platforms. Here, we report the direct epitaxial integration of $α$-MnTe thin films on Si(111) via molecular beam epitaxy and demonstrate a robust anomalous Hall effect (AHE) in this silicon-compatible altermagnetic system. Despite the absence of net bulk magnetization, the films exhibit a pronounced hysteretic Hall response, providing transport evidence consistent with finite Berry curvature generated by symmetry breaking in the thin-film geometry. High-resolution structural and spectroscopic characterization confirms phase-pure, epitaxial growth with hexagonal NiAs-type symmetry, while magnetotransport measurements reveal correlated hysteresis in both transverse and longitudinal channels with systematic temperature evolution. First-principles calculations reveal substantial uncompensated Berry curvature arising from the spin-split band structure, consistent with altermagnetic symmetry and the origin of the observed Hall response. These results establish MnTe/Si(111) as a silicon-compatible altermagnetic platform and chart a concrete pathway for embedding Berry-phase-driven functionalities into scalable semiconductor device architectures.

cond-mat.mtrl-sci↗
arXiv2026

All-Optical Control of Interfacial Polarization in MoS$_2$/WSe$_2$ Heterobilayers

All-optical tuning of van der Waals heterostructures with coherent radiation offers a promising path toward ultrafast memory and optoelectronic devices. In the first-principles framework of real-time time-dependent density functional theory, we predict the induction of a persistent, long-lived out-of-plane polarization in MoS$_2$/WSe$_2$ heterobilayers, resonantly driven by intense ultrafast pulses. While weak fields preserve the intrinsic type-II band alignment, intermediate intensities trigger a four-fold enhancement of interlayer charge transfer. By analyzing the high-harmonic generation spectrum, we identify a transition from the perturbative to the strong-field regime inducing photoinduced interfacial polarity. We additionally show that lattice strain, ubiquitously present in heterobilayers, can be used as additional knob to adjust the resonant condition without compromising the permanent dipole induction. Our findings provide a theoretical blueprint for the all-optical manipulation of polar phases in low-dimensional heterostructures at the femtosecond scale.

cond-mat.mtrl-sci↗