Mechanism discovery with thermodynamically consistent and atom conserving chemical reaction neural networks
Felix Döppel · Mauro Bracconi · Matteo Maestri
Abstract
Chemical reaction neural networks are a promising tool for the automated discovery of chemical mechanisms from reactor data. For the first time, thermodynamic consistency of the discovered mechanisms is enforced through DeDonder hard-constraints, leading to physical plausibility, improved convergence, and guaranteed evolution towards the physically correct equilibrium composition.
Chat is not available.
Successful Page Load