/kinn

Kinetics-Informed Neural Networks

Primary LanguageJupyter NotebookGNU General Public License v3.0GPL-3.0

KINNs: Kinetics-Informed Neural Networks

Pareto + Regularization Approach

arXiv: https://doi.org/10.48550/arXiv.2011.14473

KINNs training example for the dcs reaction type

animated

is the automatic differentatiation of a the neural-network over time points . The physical mode (microkinetic model) is denoted by with parameters .

The kinetic model represents the following fully-reversible chemical reactions.
The latent Kinetics type d involves ad/desorption steps, intermediate species (radicals, ) that do not have a corresponding gas phase species is part of the reaction, type c.
Reaction between radicals formula, formula, and formula, adds further complexity to surface reaction, type s.

Raw preliminary JAX-based source code can be found under kinn.

Reference Jupyter Notebooks

  1. Combined data generation can be found here.
  2. Data digestion and plot generation here.