AdaBoost Ensemble Correction Models for TDDFT Calculated Absorption Energies

Molecular excited states are important for molecular optical properties, which can be feasibly explored by quantum chemical calculations. However, the computation is highly demanding due to their complicated characteristic features. Therefore, high accuracy and unambiguous descriptions are strongly desired for excited state investigations. This paper proposes accurate, robust, and efficient ensemble correction models for absorption calculations with the most used quantum chemical method, time-dependent density functional theory (TDDFT). Models are built by AdaBoost framework with both weak machine learning: support vector machine (SVM), general regression neural network (GRNN), and an ensemble learning: the random forest (RF) regression method. With the models, the low accuracy calculations, TDDFT calculated absorption energies (λ
_max
) for 433 organic molecules with the minimum basis set STO-3G, are significantly improved. The mean absolute error (MAE) and the root mean square error (RMSE) of the calculated λmax are reduced from 0.62 and 0.79 eV to 0.11 and 0.14 eV, respectively. The validation parameters of the proposed correction model can reach up to R
²
(0.97), Q
²
(0.98), and Q
_cv
²
(0.99), which suggests the great goodness-of-fit and predictability. This investigation illustrates that the proposed ensemble correction models by sophisticated algorithms are highly efficient and accurate. Therefore, it may serve as an alternative tool to establish good correction models for TDDFT absorption calculations, which could significantly improve the accuracy of TDDFT calculations and extend machine learning algorithms on other feature calculations of excited states.