Research map / Large language models / Agents, code and math
LLEMA: Accelerating Materials Design via LLM-Guided Evolutionary Search
Nikhil Abhyankar, Sanchit Kabra, Saaketh Desai, Chandan K. Reddy
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add iterative feedback memory to a plain LLM (ablation) | LLM · 4-task ablation | hit rate: 4.4 -> 15.1 | Improved |
| add iterative feedback memory to a plain LLM (ablation) | LLM · 4-task ablation | memorization rate: 95.3 -> 58.3 | Improved |
| add multi-island mutation and crossover on top of memory (ablation) | LLM w/ Memory · 4-task ablation | hit rate: 15.1 -> 29.8 | Improved |
| add chemistry-informed generation rules on top of mutation and crossover (ablation) | LLEMA · 4-task ablation | stability: 21.5 -> 27.6 | Improved |
| add chemistry-informed generation rules on top of mutation and crossover (ablation) | LLEMA · 4-task ablation | memorization rate: 25.3 -> 16.6 | Improved |
| add chemistry-informed generation rules on top of mutation and crossover (ablation) | LLEMA · 4-task ablation | hit rate: 29.8 -> 30.2 | No clear effect |
| add CGCNN/ALIGNN surrogate models to the oracle (ablation) | LLEMA · Wide-bandgap semiconductors | hit rate: 4.8 -> 30.2 | Improved |
| add CGCNN/ALIGNN surrogate models to the oracle (ablation) | LLEMA · Wide-bandgap semiconductors | stability: 3.4 -> 27.6 | Improved |
| use 5 islands instead of 1 (ablation) | LLEMA · Wide-Bandgap Semiconductors | hit rate: 18.62 -> 33.62 | Improved |
| use 5 islands instead of 1 (ablation) | LLEMA · Hard, Stiff Ceramics | hit rate: 5.65 -> 30.99 | Improved |
| use 5 islands instead of 10 (ablation) | LLEMA · SAW/BAW Acoustic Substrates | hit rate: 26.78 -> 59.88 | Improved |
| use 5 islands instead of 10 (ablation) | LLEMA · High-k Dielectrics | hit rate: 20.34 -> 19.96 | No clear effect |
Closest papers by meaning
- MADE: Benchmark Environments for Closed-Loop Materials Discovery (ICML 2026 · similarity 0.88)
- A Sober Look at LLMs for Material Discovery: Are They Actually Good for Bayesian Optimization Over Molecules? (ICML 2024 · similarity 0.88)
- FlowLLM: Flow Matching for Material Generation with Large Language Models as Base Distributions (NeurIPS 2024 · similarity 0.88)
- MatExpert: Decomposing Materials Discovery By Mimicking Human Experts (ICLR 2025 · similarity 0.88)
- LLM Meets Diffusion: A Hybrid Framework for Crystal Material Generation (NeurIPS 2025 · similarity 0.88)
- Fine-Tuned Language Models Generate Stable Inorganic Materials as Text (ICLR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
