ML-DiCE | Machine Learned Diffusion Coefficient Estimator

Installation

pip install mldice

Usage

Create a conda environment as

conda create --name myEnv

Activate the environment created

conda activate myEnv

In the activated environment, run

mldice -options

mldice will featurize your alloy or impure metal and predict diffusion coefficient in m^2/s. The options can be set through following arguments

Use -de to specify diffusing element -dm to specify diffusion medium. Examples:

• -de Fe would take iron as diffusing element.
• -dm Ni75.5Cu24Co0.5 select Ni75.5Cu24Co0.5 is the diffusion medium where constituent elements expressed as percentage
• -t 500 would select the temperature (say, 500K in this case) of diffusion process in Kelvin.
• -m self select self diffusion mechanism. Mechanisms include self, impurity and chemical modes.
• -e RF would select Random Forest regression as prediction algorithm. DNN selects neural network based prediction.

Essentially, the run command shall be as follows:

mldice -de [diffusing element] -dm [diffusion medium] -t [temperature] -m [diffusion mechanism] -e [algorithm chosen for prediction]

Output

All outputs can be found in the Prediction.md file. It contains the following information:

Predicted parameters

Property	Value
Predicted D	-- m^2/s
RMSE	-- m^2/s
MAE	-- m^2/s
Uncertainty	-- m^2/s

Online Ressources

• https://arjun.skv.net/SI (Supporting Information)
• https://arjunskv.net/main (Main article)

Citation

@article{mldice,
    author = {Kulathuvayal, Arjun S. and Rao, Yi and Su, Yanqing},
    title = "{Elemental diffusion coefficient prediction in conventional alloys using machine learning}",
    journal = {Chemical Physics Reviews},
    volume = {5},
    number = {4},
    pages = {041402},
    year = {2024},
    month = {10},
    issn = {2688-4070},
    doi = {10.1063/5.0222001},
    url = {https://doi.org/10.1063/5.0222001},
    eprint = {https://pubs.aip.org/aip/cpr/article-pdf/doi/10.1063/5.0222001/20208332/041402\_1\_5.0222001.pdf},
}

Under development

Advanced featurization for alloys: New featurization schemes are under developing