Abstract:Computational simulations have become essential for understanding reaction mechanisms, identifying optimal catalysts, studying materials, and discovering chemical pathways. Even with the current advanced computational resources, determining competing reaction pathways and the associated transition states is extremely challenging. The problem is further exasperated as the system size increases due to the increasing number of comparable energy isomers and intermediates. In this work, we present a novel computational protocol to efficiently explore the potential energy surface and identify reaction pathways, by combining an adaptive learning global optimization procedure and neural networks for predicting transition states. For a given reaction, the global optimizer is first used to obtain putative low-lying minima in the basins of reactants and products. In the second step, generative adversarial networks (GAN) are used to estimate the transition states that connect any two initial and final states of a given reaction. This computational interface (GlobOptRx) can quickly identify the transition state structures along a reaction pathway and aid in the discovery of new reactions, determine generalized systems descriptors and facilitate the construction of kinetic models.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to solve the challenge of determining competitive reaction paths and their associated transition states in chemical reactions. Specifically, as the system size increases, finding low - energy isomers and intermediates becomes increasingly complex, and these are the keys to understanding reaction mechanisms, identifying optimal catalysts, and designing new materials. Traditional methods usually rely on known equilibrium structures in experimental and theoretical studies and often require a good guess of the transition - state geometry, which makes the calculation very time - consuming and difficult. To solve these problems, the authors propose a new computational protocol (GlobOptRx), which combines an adaptive - learning global - optimization procedure and a neural network to predict transition states, thereby efficiently exploring the potential - energy surface and identifying reaction paths. This method can quickly determine the transition - state structures on the reaction path, help discover new reactions, determine general system descriptors, and promote the construction of kinetic models. #### Summary of main problems: 1. **Reaction - path exploration in complex systems**: Existing methods have difficulty dealing with numerous energy - similar structures and transition states in large atomic and molecular assemblies. 2. **Computational challenges in transition - state location**: Traditional gradient - based optimization algorithms are prone to getting stuck in local minima rather than global minima. 3. **Kinetic studies of multi - step reactions**: Understanding multiple possible reaction paths is crucial for studying actual chemical problems, but existing methods have difficulty efficiently conducting such exploration. By introducing global - optimization algorithms and generative adversarial networks (GAN), this study provides a faster and more reliable method to identify reaction paths and transition states, especially in cases involving multiple reactants and products. This not only improves computational efficiency but also provides new tools for understanding and designing complex chemical reactions.

Adaptive Learning Reaction Pathways Search

Generative Model for Constructing Reaction Path from Initial to Final States

Heuristics-Guided Exploration of Reaction Mechanisms

React-OT: Optimal Transport for Generating Transition State in Chemical Reactions

Minima Hopping Guided Path Search: An Efficient Method for Finding Complex Chemical Reaction Pathways

Transition State Searching Accelerated by Deep Learning Potential

Chemical Networks from Scratch with Reaction Prediction and Kinetics-Guided Exploration

Robust Gaussian Process Regression method for efficient reaction pathway optimization: application to surface processes

Constrained Broyden Dimer Method with Bias Potential for Exploring Potential Energy Surface of Multistep Reaction Process.

Computer‐assisted synthetic planning considering reaction kinetics based on transition state automated generation method

ChemDyME: Kinetically Steered, Automated Mechanism Generation through Combined Molecular Dynamics and Master Equation Calculations

Exploration of Reaction Pathways and Chemical Transformation Networks

Combined Molecular Dynamics and Coordinate Driving Method for Automatically Searching Complicated Reaction Pathways

Concentration-Flux-Steered Mechanism Exploration with an Organocatalysis Application

Machine-learning driven global optimization of surface adsorbate geometries

Learning to Plan Chemical Syntheses

Systematic and unbiased pathway exploration by artificial force application to a generic neural network potential

Integrated Reaction Path Processing from Sampled Structure Sequences

CatTSunami: Accelerating Transition State Energy Calculations with Pre-trained Graph Neural Networks

A human-machine interface for automatic exploration of chemical reaction networks

Deep reaction network exploration at a heterogeneous catalytic interface