[Paper Review] Enhancing automated reaction discovery with boxed molecular dynamics in energy space

Rafael A. Jara‐Toro, Gustavo A. Pino+3 more|arXiv (Cornell University)|Oct 3, 2019

Machine Learning in Materials Science103 references27 citations

TL;DR

This paper proposes integrating the Boxed eXtended Dynamics (BXDE) method with the AutoMeKin automated reaction discovery framework to enhance the detection of intermediates and transition states in complex reaction networks. By applying BXDE in energy space, the method locates reaction pathways that are ~50 kcal/mol more stable and more densely connected than those found with standard AutoMeKin, significantly improving efficiency and completeness in reaction discovery for the ozonolysis of α-pinene.

ABSTRACT

The rare event acceleration method BXDE is interfaced in the present work with the automated reaction discovery method AutoMeKin. To test the efficiency of the combined AutoMeKin-BXDE procedure, the ozonolysis of a-pinene is studied in comparison with standard AutoMeKin. AutoMeKin-BXDE locates intermediates and transition states that are more densely connected with each other and approximately 50 kcal/mol more stable than those found with standard AutoMeKin. Other than the different density of edges between the nodes, both networks are scale-free and display small-world properties, mimicking the network of organic chemistry. Finally, while AutoMeKin-BXDE finds more transition states than those previously reported for O3 + a-pinene, the standard procedure fails to locate some of the previously published reaction pathways using the same simulation time of 2.5 ns. In summary, the mixed procedure is very promising and clearly outperforms the standard simulation algorithms implemented in AutoMeKin. BXDE will be available in the next release of AutoMekin.

Motivation & Objective

• To improve the efficiency and completeness of automated reaction discovery in complex organic reaction networks.
• To address the limitation of standard AutoMeKin in locating key intermediates and transition states within limited simulation times.
• To test the effectiveness of combining BXDE with AutoMeKin in discovering reaction pathways for the O3 + α-pinene system.
• To evaluate whether the hybrid AutoMeKin-BXDE approach produces networks with structural properties resembling those of organic chemistry networks.
• To demonstrate that BXDE enhances the discovery of previously reported and new reaction pathways compared to standard AutoMeKin.

Proposed method

• The BXDE method is interfaced with AutoMeKin to enable enhanced sampling in energy space, focusing on rare reaction events.
• BXDE applies a box potential in energy space to confine trajectories near the potential energy surface, improving sampling of transition states and intermediates.
• The method uses a 2.5 ns simulation time to explore the reaction coordinate space, with energy constraints to accelerate rare events.
• Reaction networks are constructed from identified intermediates and transition states, with topological analysis performed to assess network properties.
• Networks are evaluated for scale-free and small-world characteristics, and compared to standard AutoMeKin results.
• The approach is validated using the ozonolysis of α-pinene as a benchmark reaction.

Experimental results

Research questions

• RQ1 Can the integration of BXDE with AutoMeKin significantly improve the discovery of transition states and intermediates in complex reaction networks?
• RQ2 How do the topological properties (e.g., scale-free, small-world) of the reaction networks generated by AutoMeKin-BXDE compare to those from standard AutoMeKin?
• RQ3 To what extent does AutoMeKin-BXDE recover previously reported reaction pathways for O3 + α-pinene compared to standard AutoMeKin?
• RQ4 Are the intermediates and transition states identified by AutoMeKin-BXDE more thermodynamically stable than those found with standard AutoMeKin?
• RQ5 Does the enhanced sampling in energy space via BXDE lead to a denser connectivity of reaction nodes in the network?

Key findings

• AutoMeKin-BXDE locates intermediates and transition states that are approximately 50 kcal/mol more stable than those found with standard AutoMeKin.
• The reaction network generated by AutoMeKin-BXDE exhibits a higher density of connections between nodes (edges) compared to standard AutoMeKin.
• Both the AutoMeKin-BXDE and standard AutoMeKin networks display scale-free and small-world topological properties, resembling the structure of organic chemistry networks.
• AutoMeKin-BXDE discovers more transition states than previously reported for the O3 + α-pinene system.
• Standard AutoMeKin fails to locate some of the previously published reaction pathways within the same 2.5 ns simulation time, highlighting the advantage of BXDE integration.
• The BXDE method is set to be included in the next release of AutoMeKin, indicating its potential for broader application in automated reaction discovery.