/SAC-Scaling-Laws

Training Scaling Laws for Single-atom Catalysts on the Support Using Machine Learning

Primary LanguageJupyter NotebookMIT LicenseMIT

Training Scaling Laws for Single-atom Catalysts

This repository contains the workflow of training scaling laws based on physical descriptors (features obtained from density functional theory calculations) using various machine learning methods.

The scaling laws are useful surrogate models for fast prediction of desired properties and catalyst material screening, saving computing time by doing fewer quantumn calculations.

toc

Developers

Scaling Relationships are developed for

  • Ebind, the binding energy of single-metal atom on a support
  • Ea, the activation barrier for metal atom diffusion

Ebind and Ea represents the thermodynamic and kinetic stability of single metal atom catalysts, respectively.

Dataset

The dataset includes properties of the single-atoms on the support calculated from density functional theory (DFT) in Ea_data.csv

  • 9 types of supports
  • 11 types of metals: Ag, Au, Co, Cu, Fe, Ir, Ni, Pd, Pt, Rh, Ru
  • 99 sample points

metal_support

Machine Learning Methods Used:

  • LASSO regression
  • Ridge regression
  • Elastic net
  • Ordinary Least Square (OLS) regression
  • Genetic Programming (GP) based on sybomlic regression

Getting Started

Dependencies

  • Numpy: Used for vector and matrix operations
  • Matplotlib: Used for plotting
  • Scipy: Used for linear algebra calculations
  • Pandas: Used to import data from Excel files
  • Sklearn: Used for training machine learning models
  • Seaborn: Used for plotting
  • Gplearn: Used for training genetic programming models
  • Graphviz: Used for symbolic tree visualization

Related Publication

Su, Y.; Zhang, L.; Wang, Y.; Liu, J.; Muravev, V.; Alexopoulos, K.; Filot, A. W.; Vlachos, D. G.; Hensen, E. J. M. Stability of Heterogeneous Single-Atom Catalysts : A Scaling Law Mapping Thermodynamics to Kinetics (2019). (Submitted)

Special Thanks

Dr. Ya-qiong Su (DFT data)