[ChatBattery] Expert-Guided LLM Reasoning for Battery Discovery: From AI-Driven Hypothesis to Synthesis and Characterization

Authors: Shengchao Liu^*, Hannan Xu^*, Yan Ai^*, Huanxin Li⁺, Yoshua Bengio⁺, Harry Guo⁺

1 Environment Setup

conda create -n ChatBattery python=3.9
conda activate ChatBattery

pip install pandas
pip install openai==0.28
pip install Levenshtein
pip install pymatgen==2024.4.13
pip install ase
pip install scikit-learn
pip install xgboost
pip install mp-api==0.41.2
pip install Flask
pip install -e .

2 Data Preprocess

In our case, we download all the chemical formula including Lithium from ICSDN. Then after preprocessing, the file is data/Li_battery/preprocessed.csv.

3 Run Scripts

You need to set up two keys:

• First set the OpenAI API key, type this in terminal export OPENAI_API_KEY=xxxxx.
• Then is the MP API key, you can change it in the file ChatBattery/search_agent and reset MP_api_key = 'xxx'.

3.1 Exploration Phase

For stage 1 to 4, run this CMD in terminal

python main.py

The default address is 127.0.0.1:5000. Simply open this address in your browser, and you will see the interface displayed like this:

Then please follow the instructions from stage 1 to stage 4. Enter Ctrl + C if you want to stop the program. The results are collected into a csv file, which will be fed into the next stage.

3.2 Exploitation Phase

• Stage 5, first please get prepared the input file

cd stage_05_deduplication
python stage_05_deduplication.py --input_file=xxx

• Stage 6, then run the following CMD

cd stage_06_ranking
bash stage_06_ranking.sh