A generative model reveals the secrets of material disorder

Scientists from the National University of Singapore (NUS) used generative machine learning models to explore the different methods by which atoms enter adjacent crystals in a piezoelectric material, which are materials that generate a low electrical voltage when the application of a mechanical constraint, can undergo disparities. This revelation uncovers the ways in which disorder emerges in such materials.

In materials science, a long-standing question is understanding whether different structural disorders in complex materials serve useful functions, with the major challenge being identifying the types of disorders within a particular sample.

An NUS research team has addressed this challenge by condensing a wide range of structural disorders within the domain boundaries of a piezoelectric material into a small set of simple, multi-scale probabilistic rules. With these rules, they created a generative machine learning model spanning three orders of magnitude in length scales, enabling the study of statistical properties of the material beyond practical measurement limits.

Led by Assistant Professor Ne-te Duane Loh of the Department of Physics and Department of Biological Sciences at NUS, the research team discovered that experimentally observed structural disorder along the domain boundaries of potassium niobate piezoelectric films and sodium could be distilled in surprising ways. compact set of simple probabilistic rules. These rules could be decomposed into two sets that dominate at distinct length scales: the Markov chain and random kernels. Using these two sets of rules creates the set of domain boundaries for a specific material sample.

The team translated these probabilistic rules into the “vocabulary” and “grammar” of an interpretable machine learning model to generate and study a broad spectrum of realistic disordered domain boundaries that are indistinguishable from experimental measurements. This generative model provided access to orders of magnitude more observations than would be possible through hands-on experimentation or costly first-principles calculations.

Using this model, the authors discovered previously undetected domain boundary patterns in the material, which are chain-like structures, shedding light on factors that could affect its piezoelectric response. They also found evidence that these domain boundaries maximize entropy. This advance suggests that interpretable machine learning models can understand the complex nature of disorder in materials, paving the way for understanding their function and design.

The research results were published in the journal Scientists progress.

This research continues the team’s continued integration of statistical learning with atomic resolution electron microscopy to image complex materials. Dr. Jiadong Dan, the first author and Eric and Wendy Schmidt AI in Science Fellow, said: “Our work can be generally extended and applied to other important systems where disorder plays a critical role in controlling the physical properties of materials. »

The team also plans to further investigate the functional importance of newly discovered structural patterns, highlighting the potential for understanding and designing complex materials.

Professor Loh added: “This work complements our previous learning of atomic pattern hierarchies. Together, they push us towards creating complementary artificial intelligence (AI) alongside microscopes to provide rapid and unprecedented feedback.”