> Source URL: /sciences.guide
---
title: "Prompt Engineering for Chemistry & Physics"
description: "How LLMs and foundation models are accelerating the physical sciences — and what you could build with them in this course."
---

[@styles]: ./styles.css

# Prompt Engineering for Chemistry & Physics

Chemistry and physics labs are increasingly AI-native: generative models propose candidate materials, LLM agents plan syntheses, and foundation models trained on millions of DFT calculations are replacing expensive simulation steps. Grad schools and national labs expect students who can work alongside these systems.

## Where this is showing up in Chemistry & Physics

- Microsoft's **[MatterGen](https://www.microsoft.com/en-us/research/blog/mattergen-property-guided-materials-design/)** is a diffusion model for inorganic materials design that generates 2.9x more stable structures than prior state-of-the-art, with property-conditioned adapters for chemistry, symmetry, magnetic, and mechanical targets.
- Meta FAIR's **[Open Materials 2024 (OMat24)](https://arxiv.org/abs/2410.12771)** released 110M+ DFT calculations and the **EquiformerV2** pre-trained models, achieving F1 > 0.9 on the Matbench Discovery leaderboard for stability prediction.
- **[ChemCrow](https://www.nature.com/articles/s42256-024-00832-8)** (Bran, Cox, Schilter, Baldassari, White, Schwaller — *Nature Machine Intelligence*, 2024) pairs GPT-4 with 18 expert-designed chemistry tools and has autonomously planned syntheses of an insect repellent, three organocatalysts, and guided discovery of a novel chromophore.
- Boiko et al., ["Autonomous chemical research with large language models"](https://www.nature.com/articles/s41586-023-06792-0) (*Nature*, 2023), demonstrated LLM agents driving real lab instruments — the template for the autonomous-lab work now underway at ORNL, Argonne, and industrial R&D.

## Projects you could build in this course

- A RAG assistant over an open chemistry or physics literature subset (e.g., arXiv cond-mat, PubChem)
- A structured-extraction tool that pulls reaction conditions and yields from lab notebooks or papers
- An agent that plans a simple synthesis or simulation workflow by calling specialized tools

[← Back to Thinking With Machines](./index.path.md)


---

## Backlinks

The following sources link to this document:

- [Chemistry & Physics](/index.path.llm.md)