This repository contains the code associated with the manuscript:

"Nonlinearity of the post-spinel transition and its expression in slabs and plumes worldwide"

If you use the code and/or database here, please cite this article: Dong, J. et al. Nonlinearity of the post-spinel transition and its expression in slabs and plumes worldwide. Nat Commun 16, 1039 (2025). https://doi.org/10.1038/s41467-025-56231-z

Manuscript Authors

• Junjie Dong
• Rebecca A. Fischer
• Lars P. Stixrude
• Matthew C. Brennan
• Kierstin Daviau
• Terry-Ann Suer
• Katlyn M. Turner
• Yue Meng
• Vitali B. Prakapenka

Code Author

• Junjie Dong

Summary

This repository consists of two main parts:

1. Model Selection with Supervised Learning
2. Phase Diagram Construction

The code analyzes phase equilibria observations and computes a globally optimized phase diagram through multi-class logit regression and supervised learning. The code can be used generally, but the construction of phase relations in Mg₂SiO₄ at mantle transition zone conditions is used as a benchmark case to ensure reproducibility and ease of use, and the data on phase equilibria observations of Mg₂SiO₄ are provided.

Usage

Follow the instructions below or in the Python scripts to reproduce the analyses discussed in the manuscript.

Prerequisites

The primary packages used in this code are:

• matplotlib=3.5.0
• numpy=1.19.5
• pandas=1.3.5
• python=3.9.20
• scikit-learn=0.23.2
• scipy=1.10.0

For full package compatibility, you can use the provided conda environment file or requirements file.

Getting Started

1. Set up the conda environment:

   conda env create -f environment.yml
   conda activate mlpd
   

2. Alternatively, you can install dependencies using pip:

   pip install -r requirements.txt
   

3. To tune the hyperparameters for the model, run 'logit-reg-model-selection.py':

   python logit-reg-model-selection.py
   

4. You can find the optimized degree and other hyperparameters in the 'log.txt' file. After updating the optimized degree and other hyperparameters, run 'logit-reg-fit.py':

   python logit-reg-fit.py