LLM Framework DyRIS Predicts Space Groups of Double Perovskites
A recent preprint on arXiv (2608.10483) presents DyRIS (Dynamic and Diversity-enhanced Few-shot Retrieval and Rule-Guided Inference for Space-Group Prediction), a framework utilizing an LLM-agent to forecast the space groups (SGs) of double perovskites (DPs). While double perovskites allow for extensive compositional flexibility, determining stable structures’ space groups is difficult due to datasets that disproportionately represent major SGs over minor ones. DyRIS tackles this issue through diversity-enhanced dynamic few-shot prompting, which retrieves pertinent in-context examples while mitigating the prevalence of common SGs. Additionally, it employs rule-guided inference based on B/B' cation ordering and major-SG bias control to refine the top three SG candidates. The framework was assessed on 3,528 thermodynamically filtered DP entries. Researchers authored the paper, which is crucial for materials science and computational chemistry, as it may expedite the discovery of new double perovskite materials. The preprint can be accessed at https://arxiv.org/abs/2608.10483.
Key facts
- DyRIS is an LLM-agent-based framework for predicting space groups of double perovskites.
- It uses diversity-enhanced dynamic few-shot prompting to handle imbalanced datasets.
- Rule-guided inference incorporates B/B' cation ordering and quantitative indicators.
- The framework ranks Top-3 SG candidates for a given DP composition.
- Evaluation was performed on 3,528 thermodynamically filtered DP entries.
- The paper is available on arXiv with ID 2608.10483.
- Double perovskites have broad compositional tunability but SG prediction is difficult.
- The framework aims to control major-SG bias in predictions.
Entities
Institutions
- arXiv