Improving the accuracy of density-functional theory calculation: the genetic algorithm and neural network approach

Area of Science:

• Computational chemistry
• Quantum mechanics
• Machine learning

Background:

• Density functional theory (DFT) calculations are crucial for predicting molecular properties.
• Improving the accuracy of DFT, particularly for optical absorption energies, remains an active research area.
• Existing DFT methods can exhibit significant root-mean-square (rms) deviations in predicting optical properties.

Purpose of the Study:

• To develop and validate a hybrid approach combining genetic algorithms and neural networks (GANN) to enhance DFT calculation accuracy.
• To apply the GANN approach to predict the optical absorption energies of a diverse set of 150 organic molecules.
• To quantify the improvement in accuracy compared to standard DFT methods.

Main Methods:

• Implementation of a combined quantum mechanical calculation and GANN correction framework.
• Utilizing the Time-Dependent DFT (TD-DFT) method with the B3LYP/6-31G(d) basis set for initial calculations.
• Training and applying a neural network model, optimized by a genetic algorithm, to correct DFT-derived optical absorption energies.

Main Results:

• The standard TD-DFT calculation yielded an initial rms deviation of 0.47 eV for optical absorption energies.
• The neural network correction alone reduced the rms deviation to 0.22 eV.
• The integrated GANN correction approach further improved accuracy, achieving a final rms deviation of 0.16 eV.

Conclusions:

• The GANN approach offers a significant improvement in the accuracy of DFT calculations for optical absorption energies.
• This hybrid method demonstrates a powerful strategy for refining quantum mechanical predictions of molecular properties.
• The GANN correction is effective for a range of organic molecules, highlighting its potential applicability in computational chemistry.