First-Principles Calculations and Machine Learning for Nonradiative Relaxation of Charge Carriers at Nanostructured...

Area of Science:

• Computational materials science
• Quantum chemistry
• Machine learning applications

Background:

• Nonradiative processes dominate hot carrier relaxation in silicon materials.
• Understanding relaxation dynamics of silver clusters on Si(111) is crucial for materials science.
• Previous methods relied on complex quantum mechanical calculations.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting nonradiative relaxation rate constants (k_e/h) of charge carriers.
• To explore the effectiveness of Redfield tensor elements as features for ML models.
• To simplify the prediction of complex excited-state relaxation processes.

Main Methods:

• Computed nonadiabatic couplings (NACs) and Redfield formalism for dissipative rates.
• Generated a dataset of k_e/h for Ag_n (n=0-7) clusters on Si(111) slabs.
• Trained and evaluated four ML regressors: Ridge, Random Forest, XGBoost, and MLP, using stratified 5-fold cross-validation.

Main Results:

• All evaluated ML models demonstrated strong predictive performance, with test set R^2 values exceeding 0.96.
• Ridge regression and XGBoost achieved the highest test set R^2 values of 0.99.
• The performance similarity between linear and nonlinear models suggests an approximately linear relationship in the feature space.

Conclusions:

• Machine learning models can effectively learn complex excited-state relaxation processes from Redfield tensor elements.
• Relatively simple ML models provide accurate predictions for nonradiative relaxation rates.
• This approach offers a computationally efficient alternative for studying carrier dynamics in materials.