Surface Hopping Nested Instances Training Set for Excited-state Learning

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Photochemistry

Background:

• Theoretical studies of molecular photochemistry and photophysics are crucial for understanding natural processes.
• Computationally intensive quantum chemical calculations limit direct simulations and machine learning (ML) model development.
• Existing data limitations hinder progress in modeling excited-state properties.

Purpose of the Study:

• To introduce SHNITSEL, a comprehensive data repository for ab-initio calculations of organic molecules.
• To provide a robust benchmark dataset for training and validating machine learning models.
• To accelerate the development of ML-based approaches for predicting excited-state properties.

Main Methods:

• Generated a dataset of 418,870 ab-initio data points for nine organic molecules.
• Included high-accuracy quantum chemical properties for ground and excited states (singlet/triplet).
• Incorporated properties related to electronic state coupling, such as nonadiabatic couplings, transition dipoles, and spin-orbit couplings.

Main Results:

• Established SHNITSEL, a large-scale, high-quality dataset for molecular excited-state studies.
• The dataset contains diverse quantum chemical properties essential for photophysical and photochemical research.
• SHNITSEL enables rigorous benchmarking of existing and novel ML models.

Conclusions:

• SHNITSEL significantly lowers the barrier for developing ML models in excited-state chemistry.
• The repository facilitates advancements in computational modeling of photochemistry and photophysics.
• This resource is expected to drive innovation in predicting and understanding molecular excited states.