Fig. 4. Monte Carlo tree search (MCTS) to identify the lowest energy pathway in a complex reaction network. CO + H2O reaction on Cu(111) is utilized to illustrate the MCTS, and how MCTS enquires the R-Pat and K-Info units for reaction information is illustrated by the HCOO + H reaction as indicated in the dotted purple box. (A) MCTS procedure that includes four phases: selection, rollout, reward and expansion, which are iteratively performed to explore the reaction network. Red arrow: the interactive section traverses the tree by selecting the node with maximum priority score V. Green arrow: the fast rollout evaluation starts from the selected node. Blue arrow: update the reward according to the rollout result for nodes traversed in the branch from the selected node to the root node. (B) The flowchart of the expansion phase showing the working mechanism, where the R-Pat unit and K-Info unit are utilized to obtain the kinetics information so that the child nodes are ranked. (C) R-Pat unit to predict reaction patterns for a given reactant. (D) K-Info unit to predict quantitatively the reaction barrier and energy.
