Section: New Results

CREST: Chemical Reactivity Exploration with Stochastic Trees

Participants : Leonard Jaillet, Stephane Redon.

We have proposed the CREST method (Chemical Reactivity Exploration with Stochastic Trees), a new simulation tool to assess the chemical reaction paths of molecular systems. First, it builds stochastic trees based on motion planning principles to search for relevant pathways inside a system's state space. This generates low energy paths transforming a reactant to a given product. Then, a nudged elastic band optimization step locally improves the quality of the initial solutions. The consistency of our approach has been evaluated through tests in various scenarios. It shows that CREST allows to appropriately describe conformational changes as well as covalent bond breaking and formations present in chemical reactions (see figure 14).

This contribution appears in continuity of our previous work regarding the development of a geenric Motion planning architecture for nanosystems. Important features have been added to specifically treat the case of chemical reaction, such as structure alignment, exploration based on multiple trees, automatic resizing of the sampling volume, etc.

Figure 14. Fictive chemical reaction transforming a pentane into a cyclopentane with a H2 molecule. Hydrogen atoms leading to the H2 molecule are colored. The path obtained with CREST (top) is able to capture the CH3 internal rotations that approaches the two H2 Hydrogens and thus, lead to a low energy barrier. By comparison, a method based on linear interpolation (Lin) gives intermediate broken structures after local path optimization (down). The plot on the right shows the respective energies along the paths. This represents Scenario 3 described in our Results section.