Challenges for Kinetics Predictions via Neural Network Potentials: A Wilkinson's Catalyst Case.

Area of Science:

• Computational Chemistry
• Chemical Kinetics
• Machine Learning in Chemistry

Background:

• Ab initio kinetic studies are crucial for designing new chemical reactions.
• The Artificial Force Induced Reaction (AFIR) method is efficient but computationally expensive for exploring reaction networks.
• Accelerating these studies is essential for advancing chemical reaction design.

Purpose of the Study:

• To investigate the use of Neural Network Potentials (NNP) for accelerating ab initio kinetic studies.
• To develop and apply a novel NNP-powered AFIR method for reaction path network exploration.
• To identify limitations of general-purpose NNP models in this context.

Main Methods:

• Theoretical study of ethylene hydrogenation using the AFIR method.
• Generative Topographic Mapping for reaction path network analysis.
• Training a state-of-the-art NNP model using calculated geometries.
• Implementing NNP to replace expensive ab initio calculations during AFIR searches.

Main Results:

• Successfully performed the first NNP-powered reaction path network exploration using AFIR.
• Identified challenges and limitations of general-purpose NNP models for accelerating kinetic studies.
• Demonstrated that NNP can significantly speed up reaction path searches.

Conclusions:

• NNP shows potential for accelerating ab initio kinetic studies.
• Complementing NNP with semiempirical methods can overcome current limitations.
• The proposed framework lays the groundwork for exploring larger chemical systems.