M3GNet Accelerates Discovery Of Quaternary Oxides
Researchers systematically assess a universal machine-learning interatomic potential (M3GNet) to accelerate crystal structure prediction in quaternary oxides, exploring Sr-Li-Al-O and Ba-Y-Al-O. They rediscover known phases absent from training and predict seven new thermodynamically and dynamically stable compounds, including a new Sr2LiAlO4 polymorph. The study finds PBE stability predictions need cross-validation with SCAN and RPA, and CSP search algorithms remain bottlenecks.
Key Points
- 1Demonstrates M3GNet rediscovers known phases and predicts seven new thermodynamically stable quaternary oxides
- 2Highlights that PBE-based stability assessments require cross-validation with SCAN and RPA for reliability
- 3Indicates uMLIPs cut CSP computational cost but search-algorithm efficiency becomes the main bottleneck
Scoring Rationale
Finds new, stable quaternary oxides via M3GNet-enabled CSP, but preprint evidence and PBE reliance limit immediate confidence.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
