Paper

Paper URL: https://doi.org/10.1038/s41524-025-01522-8

Cite the following article to refer to this work.

@article{kowt2025,
  title = {Physics-informed {Bayesian} optimization suitable for extrapolation of materials growth},
  author = {W. Kobayashi and Takuma Otsuka and Yuki K. Wakabayashi and G. Tei},
  journal = {npj Computational Materials},
  volume = {11},
  pages = {36},
  doi = {https://doi.org/10.1038/s41524-025-01522-8},
  year = {2025}
}

How to run

Use run_PIBO_1st.py to reproduce Figure 4 in our paper. Similarly, run_PIBO_2nd.py produces Figure 7.

You can specify the target composition (x, y) by --target option. For example, run_PIBO_1st.py --target 0.19,0.42, which means the target values are x=0.19 and y=0.42.

Software version

Codes are confirmed to run with the following libraries. Likely to be compatible with newer versions.

• python: 3.11.5
• numpy: 1.24.3
• scipy: 1.11.1
• sklearn: 1.3.0
• matplotlib: 3.7.2
• seaborn: 0.12.2

Files

• README.md: This file.
• LICENSE.md: Document of agreement for using this sample code. Read this carefully before using the code.
• code: Contains codes
  • run_PIBO_1st.py: Script to execute PIBO for the first experiment (Fig. 4 in our paper).
  • run_PIBO_2nd.py: Script to execute PIBO for the second experiment (Fig. 7 in our paper).
  • BO_target.py: Implements BO class.
  • utils.py: Contains internal functions.
  • lhsmdu.py: Latin hypercube sampling package for acquisition function. Repository: https://dx.doi.org/10.5281/zenodo.2578780
• data: Contains data
  • data_1st.csv: Experimental data of 6 trials for run_PIBO_1st.py.
  • data_2nd.csv: Experimental data for run_PIBO_2nd.py. Former 7 points were collected through the first experiment. Latter 5 points were measured in the second phase of experiment.