Artificial Neural Network-Based Density Functional Approach for Adiabatic Energy Differences in Transition Metal...

Area of Science:

• Computational chemistry and physics
• Quantum chemistry
• Materials science

Background:

• Approximate Kohn-Sham density functional theory (DFT) is widely used but struggles with spin state energetics.
• Accurate prediction of spin states is crucial for understanding transition metal complex behavior.

Purpose of the Study:

• To develop an enhanced exchange-correlation functional for improved prediction of adiabatic energy differences in transition metal complexes.
• To leverage machine learning to correct deficiencies in existing DFT functionals.

Main Methods:

• Developed a machine learning approach using an artificial neural network correction.
• Trained the correction on a small dataset of electronic densities, atomization energies, and spin state energetics.
• Employed a bioinspired particle swarm optimization for training the functional.

Main Results:

• The machine-learned meta-generalized gradient approximation (mGGA) functional significantly outperforms existing DFT functionals.
• Demonstrated superior accuracy in predicting adiabatic energy differences for diverse transition metal complexes.
• Showcased the effectiveness of the mGGA functional across various coordination environments and metal types.

Conclusions:

• The developed machine learning-enhanced DFT functional offers a promising advancement for accurate spin state energetics prediction.
• This approach provides a more reliable tool for computational studies involving transition metal complexes.
• Highlights the potential of integrating machine learning with DFT for complex chemical problems.