MLTB: Enhancing Transferability and Extensibility of Density Functional Tight-Binding Theory with Many-body...

Area of Science:

• Computational materials science
• Quantum chemistry
• Machine learning in physics

Background:

• Standard self-consistent charge density functional tight-binding (SCC-DFTB) offers a computationally efficient method for electronic structure calculations.
• Existing DFTB models often lack accuracy for complex systems due to limitations in describing interatomic interactions.
• Machine learning potentials, like HIP-NN, show promise for improving interaction descriptions but can be computationally expensive or lack transferability.

Purpose of the Study:

• To develop a hybrid computational model that combines the efficiency of DFTB with the accuracy of machine learning potentials.
• To create a more transferable and extensible model for accurate electronic structure calculations.
• To provide a practical framework for enhancing DFTB with many-body corrections.

Main Methods:

• A hybrid machine learning tight-binding (MLTB) model was developed by integrating a machine learning neural network potential (HIP-NN) into the standard self-consistent charge density functional tight-binding (SCC-DFTB) formalism.
• The HIP-NN potential was employed as a many-body correction to the repulsive term within the SCC-DFTB framework.
• The developed MLTB model was applied to study the nanocluster structures of the thorium-oxygen system (ThO2).

Main Results:

• The MLTB model demonstrated significantly improved transferability and extensibility compared to standalone SCC-DFTB and HIP-NN models.
• The hybrid approach successfully enhanced the accuracy of DFTB calculations, bringing them closer to the accuracy of more computationally intensive DFT methods.
• An accurate MLTB model for the thorium-oxygen system was developed and utilized for nanocluster structure investigations.

Conclusions:

• The MLTB model offers a practical and accurate computational framework for materials science research.
• This hybrid approach effectively incorporates many-body corrections into DFTB, enhancing its predictive power.
• The developed method paves the way for more reliable and accurate simulations of complex material systems.