Artificial Intelligence Driving Materials Discovery? Perspective on the Article: Scaling Deep Learning for Materials Discovery

Anthony K. Cheetham,Ram Seshadri
DOI: https://doi.org/10.1021/acs.chemmater.4c00643
IF: 10.508
2024-04-29
Chemistry of Materials
Abstract:The discovery of new crystalline inorganic compounds-novel compositions of matter within known structure types, or even compounds with completely new crystal structures-constitutes an important goal of solid-state and materials chemistry. Some fractions of new compounds can eventually lead to new structural and functional materials that enhance the efficiency of existing technologies or even enable completely new technologies. Materials researchers eagerly welcome new approaches to the discovery...
materials science, multidisciplinary,chemistry, physical
What problem does this paper attempt to address?
The paper attempts to address the effectiveness and practicality of using Artificial Intelligence (AI) and Machine Learning (ML) technologies for the discovery of new materials in the field of materials science. Specifically, the paper evaluates the research results published by Merchant et al. in the journal "Nature," which claim to have predicted a large number of new inorganic material structures through deep learning methods, and assert that the number of these predicted structures is ten times that of known stable materials. However, the authors of this paper point out some key issues through a detailed review of some new compounds in the GNoME Explorer database and the Stable Structure database: 1. **Novelty and Practicality**: Many of the predicted new materials lack novelty and do not show practical application value. Additionally, some compounds, although theoretically predicted to be stable, may not be achievable in actual synthesis processes. 2. **Structural Symmetry Issues**: Many of the predicted material structures exhibit unusually low symmetry, which is rarely seen in actual experiments. This low symmetry may be due to theoretical calculations neglecting the entropy effects at finite temperatures. 3. **Use of Radioactive Elements**: The database contains a large number of compounds with radioactive elements (such as americium, promethium, etc.), which are rarely found in nature and are difficult to use in practical applications. 4. **Atomic Ordering Issues**: Many of the predicted compounds exhibit impossible atomic ordering, especially among rare earth elements. In summary, although AI/ML methods have great potential in material discovery, current methods still need further improvement to enhance their reliability and practicality.