Hierarchical Quantum Embedding by Machine Learning for Large Molecular Assemblies.

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Machine Learning

Background:

• Quantum-classical hybrid models are used for large molecules, but accuracy can be limited in large quantum regions.
• Approximate electronic structure models are often necessary for computationally expensive quantum regions.
• Accurate description of molecular interactions is crucial in drug discovery and materials science.

Purpose of the Study:

• To develop a quantum-in-quantum embedding strategy to improve the accuracy of quantum-classical hybrid models.
• To introduce the concept of "quantum cores" for accurate electronic structure calculations within larger quantum regions.
• To enhance machine learning potentials using high-accuracy data from quantum cores.

Main Methods:

• A quantum-in-quantum embedding strategy is coupled with machine learning potentials.
• Huzinaga-type projection-based embedding is used to obtain accurate electronic energies for quantum cores.
• Transfer learning is applied to improve machine learning potentials using high-accuracy data.
• Alchemical free energy and nonequilibrium switching simulations are used for binding free energy calculations.

Main Results:

• The proposed method enhances the accuracy of quantum-classical hybrid models for large molecules.
• Accurate electronic energies from quantum cores are effectively utilized to improve machine learning potentials.
• The strategy shows potential for accurate binding free energy calculations in protein-ligand complexes.

Conclusions:

• The quantum-in-quantum embedding strategy offers a promising approach to improve the accuracy of computational models for large molecular systems.
• This method effectively bridges the gap between high-accuracy quantum calculations and computationally efficient classical environments.
• The developed technique has significant implications for molecular simulations in fields like drug discovery.