Machine Learning-Corrected Simplified Time-Dependent DFT for Systems With Inverted Single-t-o-Triplet Gaps of...

Area of Science:

• Materials Science
• Photocatalysis
• Computational Chemistry

Background:

• Solar-driven photocatalytic water splitting (PWS) offers a sustainable route to hydrogen gas (H₂) production.
• Inverted singlet-triplet (INVEST) organic materials are promising for PWS due to reduced destructive triplet states and maximized excited-state lifetimes.
• Aza-triangulenes are a class of organic molecules that often exhibit the desirable INVEST property.

Purpose of the Study:

• To develop a systematic and efficient method for predicting the INVEST property in aza-triangulenes.
• To identify novel aza-triangulene molecules suitable for efficient, solar-driven photocatalytic water splitting.
• To explore the vast chemical space of aza-triangulenes for PWS applications.

Main Methods:

• Utilized machine learning (ML) to generate post hoc corrections for simplified Tamm-Dancoff approximation density functional theory (sTDA-DFT) calculations.
• Calculated singlet and triplet excitation energies and singlet-to-triplet (ΔE_S1T1) gaps for PWS systems.
• Validated ML model predictions against second-order algebraic diagrammatic construction (ADC(2)) results.

Main Results:

• The ML model achieved accuracy within 28-50 meV of ADC(2) for excitation energies and ΔE_S1T1 gaps.
• The developed ML model successfully recalled 85% of PWS candidate systems identified by ADC(2).
• Screening a database of 4025 aza-triangulenes identified 78 promising molecules for solar-driven PWS.

Conclusions:

• Machine learning provides an efficient and accurate approach to predict INVEST properties for photocatalytic applications.
• The study successfully identified a significant number of novel aza-triangulene candidates for efficient solar-driven hydrogen production.
• This work facilitates the exploration of large chemical spaces for designing advanced photocatalytic materials.