Machine-Learning Based Selection and Synthesis of Candidate Metal-Insulator Transition Metal Oxides

Aug 3, 2026

Upload by Liam

The discovery of materials that exhibit a metal-insulator transition (MIT) is crucial for developing efficient microelectronic and optoelectronic devices, yet identification is slow due to expensive electronic-structure calculations, complex mechanisms, and labor-intensive experimental validation. A machine learning classification model is used to rapidly screen a high-throughput crystal structure database for thermally driven MIT candidates. Three oxide candidates—Ca2Fe3O8, CaCo2O4, and CaMn2O4—are then analyzed for MIT mechanisms using high-fidelity density functional theory, followed by probabilistic guidance on synthesis routes. The workflow integrates physics-informed ML, DFT, and ML-suggested synthesis to accelerate discovery.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages20

Reading Time50 min

AccessFree to download

Abstract

Introduction

Challenges in discovering metal-insulator transition materials

Thermally driven MITs and competing mechanisms

Role of machine learning in accelerating materials discovery

arXiv :2404 .08653v1 [ cond-mat .mtrl-sci ] 21 Mar 2024 Machine-Learning Based Selection and Synthesis of Candidate Metal-Insulator Transition Metal Oxides Alexandru B. Georgescu, ∗ ,† Peiwen Ren,‡ Christopher Karpovich,¶ Elsa Olivetti,¶ and James M. Rondinelli ∗ ,‡ †Department of Chemistry, 800 East Kirkwood Avenue, Indiana University, Bloomington, Indiana 47405, United States ‡Department of Materials Science and Engineering, Northwestern University, Evanston, IL 60208, USA ¶Department of Materials Science and Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139 USA E-mail: [georgesc@iu.edu](georgesc@iu.edu) ; [jrondinelli@northwestern.edu](jrondinelli@northwestern.edu) Abstract The discovery of materials that exhibit a metal-insulator transition (MIT) is key to the development of multiple types of novel efficient microelectronic and optoelectronic devices. However, identifying MIT materials is challenging due to a combination of high computational cost of electronic structure calculations needed to understand their mechanism, the mechanisms’ complexity, and the labor-intensive experimental validation process. To that end, we use a machine learning classification model to rapidly screen a high-throughput crystal structure database to identify candidate compounds exhibiting thermally-driven MITs. We focus on three candidate oxides, Ca2 Fe3 O8 , CaCo2 O4 , and CaMn2 O4 , and identify their MIT mechanism using high-fidelity density functional theory calculations. Then, we provide a probabilistic estimate of which synthesis reactions may lead to their realization. Our approach couples physics-informed machine learning, density functional theory calculations, and machine learning-suggested synthesis to reduce the time to discovery and synthesis of new technologically relevant materials. Introduction Electronically active correlated electron materials and their properties can be challenging to predict, requiring high-fidelity, resource-intensive calculations to understand their electronic structure, followed by significant time and resource investment for their synthesis, as well as the accurate experimental measurement of their electronic properties. Further, the dependence of the predicted property on the exact electronic structure methodology-which often includes adjustable parameters- , combined with the high time and resource cost of synthesizing a novel material, make the cycle of discovery relatively slow. This is particularly true for metal-insulator transition (MIT) materials, which undergo a phase transition from a metallic state to an insulating state, or vice versa, through changes in external conditions, such as temperature, pressure, or doping, 1 ,2 leading to a wide variety of applications. 3–8 Here, we focus on thermally-driven MITs, as the state of these materials usually can be switched not just by temperature, but also with an applied electric field, mechanical stress, or light, a group with strong potential technological interest-however, there are only under 70 stoichiometrically distinct materials that display this property, which we have previously provided a database for 9 . Despite the same macroscopic outcome-a change in electrical resistivity with temperature-the mechanisms, the stoichiometry, and the crystal structure of these materials are often dissimilar, and their interplay is often the subject of debate-and often the exact mechanism is the subject of debate. The transition from the metallic phase to the insulating phase is usually triggered by a coupled change in symmetry in both the electronic phase and in crystal structure, which is often difficult to decouple experimentally and theoretically; however recent experiments have made progress using for example isotope effects 10 and superlattice structures 11–14 . Theoretical methods to simulate the ground state show dependence on the computational methodology, making the prediction of new compoun

FAQs

Why is finding metal-insulator transition (MIT) materials difficult?

MIT materials require high-fidelity electronic-structure calculations, whose cost is high, and their mechanisms are complex. They also need labor-intensive experimental validation after prediction.

How does the proposed workflow identify candidate MIT oxides?

It uses a machine learning classification model to rapidly screen a high-throughput crystal structure database for compounds likely to show thermally driven MITs.

What methods are used after ML screening for the selected oxides?

High-fidelity density functional theory calculations are used to identify the MIT mechanism, and probabilistic estimates are provided for synthesis reactions that may realize the candidates.

Aug 3, 2026

Upload by Liam