arXiv Science⌕ Search
Home/Archive/arXiv

arXiv · 2610.13885

Necessary conditions for benchmarking next-generation AI in materials discovery

Daniel E Widdowson·Tatiana V Kurlina·Olga D Anosova·Vitaliy A Kurlin
View original source ↗Read source document ↗Export metadata ↓

Abstract

Current AI predicts materials by speeding up energy minimisation and training on large databases. However, scaling data generation prevents manual analysis by domain experts. If novelty is evaluated by chemical compositions or traditional representations in Crystallographic Information Files, or standard structure matchers based on manually chosen thresholds, one can easily generate millions of ``new'' compositions or files without discovering truly novel materials. This paper uses a hierarchy of crystal invariants to continuously trace and quantify structural novelty based on the new mathematical definition of a crystal structure and asymptotic guarantees for invariant descriptors. The invariant-based analysis through cascade computations on a modest computer exposes the unexpected scale of duplication in the world's largest materials databases, shown here in the high-profile case of GNoME datasets. To prevent current ``embarrassment of over-prediction'', we propose necessary conditions for next-generation AI that should be automatically enforced before claiming any new materials.

Explore related subjects

cond-mat.mtrl-scimath.MG

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Daniel E Widdowson, Tatiana V Kurlina, Olga D Anosova, Vitaliy A Kurlin. 2026-10-09. Necessary conditions for benchmarking next-generation AI in materials discovery. https://arxiv.org/abs/2610.13885

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

KEEP EXPLORING

Related papers

arXiv2026

Electric-Field Driven Nuclear Dynamics of Liquids and Solids from a Multi-Valued Machine-Learned Dipolar Model

The driving of vibrational motion by external electric fields is a topic of continued interest, due to the possibility of assessing new or metastable material phases with desirable properties. Here, we combine ab initio molecular dynamics within the electric-dipole approximation with machine-learning neural networks (NNs) to develop a general, efficient and accurate method to perform electric-field-driven nuclear dynamics for molecules, solids, and liquids. We train equivariant and autodifferentiable NNs for the interatomic potential and the dipole, modifying the model infrastructure to account for the multi-valued nature of the latter in periodic systems. We showcase the method by addressing property modifications induced by electric field interactions in a polar liquid and a polar solid from nanosecond-long molecular dynamics simulations with quantum-mechanical accuracy. For liquid water, we present a calculation of the dielectric function in the GHz to THz range and the electrofreezing transition, showing that nuclear quantum effects enhance this phenomenon. For the ferroelectric perovskite LiNbO$_3$, we simulate the ferroelectric to paraelectric phase transition and the non-equilibrium dynamics of driven phonon modes related to the polarization switching mechanisms, showing that a full polarization switch is not achieved in the simulations.

cond-mat.mtrl-sci↗
arXiv2026

Deterministic domain selection of antiferromagnets via magnetic fields

Antiferromagnets (AFMs) hold promise for applications in digital logic. However, switching AFM domains is challenging, as magnetic fields do not couple to the bulk antiferromagnetic order parameter. Here we show that magnetic-field-driven switching of AFM domains can in many cases be enabled by a generic reduction of magnetic exchange at surfaces. We use statistical mechanics and Monte Carlo simulations to demonstrate that an inequivalence in magnetic exchange between top and bottom surface moments, combined with the enhanced magnetic susceptibility of surface spins, can enable deterministic selection of antiferromagnetic domains depending on the magnetic-field ramping direction. We further show that this mechanism provides a natural interpretation for experimental observations of hysteresis in magneto-optical response of the van der Waals AFM $\mathrm{MnBi_2Te_4}$. Our findings highlight the critical role of surface spins in responses of antiferromagnets to magnetic fields. Furthermore, our results suggest that antiferromagnetic domain selection via purely magnetic means may be a more common and experimentally accessible phenomenon than previously assumed.

cond-mat.mtrl-sci↗
arXiv2026

A Generalizable Machine Learning Framework for Half-Heusler Thermoelectrics

Machine learning (ML) has emerged as a powerful approach for accelerating the discovery of thermoelectric (TE) materials, yet many models suffer from limited generalizability. Here we present a robust ML framework for predicting the thermoelectric figure of merit (zT) in half-Heusler (hH) compounds, emphasizing prediction of previously unseen compositions. A principal component analysis (PCA)-based train/test splitting strategy is introduced to generate representative partitions of chemical space. Bayesian hyperparameter optimization is combined with correlation-based k-best feature selection for Random Forest, XGBoost, and Neural Network models, together with the interpretable SISSO symbolic-regression method. Relative to the full set of 118 input features, the optimized feature subsets reduce the test RMSE by 16-27%. SHAP and SISSO analyses identify temperature, A-site dopant concentration, and A-site heat of vaporization as the most important descriptors of zT. The framework is validated through prediction of previously unexplored compositions, recovery of literature-reported optimal zT values for established hH systems, and first-principles calculations for the new compounds CeCoBi and CeIrBi, which exhibit favorable electronic transport properties. The framework is further applied to high-throughput screening of the doped hH chemical space, covering about 2.46 million elemental substitution combinations across 631 thermodynamically screened parent matrices under stability constraints. This screening yields a concise set of promising candidates for future study. Overall, the work establishes a generalizable workflow for reliable data-driven thermoelectric materials discovery.

cond-mat.mtrl-sci↗