# Machine learning photodynamics reveal intersystem-crossing-driven ladderdiene ring opening

## TL;DR

This paper uses machine learning to design a light-induced ring-opening reaction that efficiently produces cyclooctatetraene under mild conditions.

## Contribution

The study introduces a machine learning-accelerated photodynamics approach to predict and optimize a visible light-driven ring-opening reaction.

## Key findings

- The carbonyl group lowers the S1-energy to 3.65 eV, making the reaction accessible to near-visible light.
- ML simulations predict a quantum yield of 89% for the formation of carbonyl-functionalized COT.
- The reaction mechanism involves fast intersystem crossings within 20 ps, leading to efficient ring opening.

## Abstract

Photochemical ring-opening reactions have become an essential tool for chemical syntheses under mild conditions with high atom economy. We propose a near-visible light-induced electrocyclic ring-opening reaction to afford cyclooctatetraene (COT) using carbonyl-functionalized tricyclooctadiene (1) based on our machine learning (ML) accelerated photodynamics simulations. Our CAM-B3LYP/cc-pVDZ calculations show that carbonyl group reduce the S1-energy of 1 to 3.65 eV (340 nm) from 6.25 eV, approaching the visible light range. The multiconfigurational CASSCF(12,11)/ANO-RCC-VDZP calculations show small S1 and T1 energy gaps near an S1-minimum region. Our ML-photodynamics simulations with 1000 FSSH trajectories revealed a stepwise ring-opening mechanism of 1 from the S1, dominated by relatively fast S1/T1 intersystem crossings in 20 ps. The trajectories predict that the quantum yield to carbonyl-functionalized COT is 89%, suggesting the light-induced ring-opening reaction of 1 is highly efficient. This work demonstrates a predictive ML-photodynamics application for photochemical reaction design.

Photochemical ring-opening reactions have become an essential tool for chemical syntheses under mild conditions with high atom economy.