Accurate Molecular Properties via Bootstrap Embedding

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Materials Science

Background:

• Accurate calculation of molecular properties like dipole moments and forces is crucial for understanding chemical systems.
• Traditional methods can be computationally expensive, limiting their application in complex systems and large-scale simulations.
• Machine learning potentials require high-quality training data, often derived from accurate quantum chemical calculations.

Purpose of the Study:

• To demonstrate the effectiveness of back-transformed bootstrap embedding (BE) for one- and two-particle density matrices (1 and 2-PDMs).
• To introduce and validate the Bootstrap Embedding-Direct Matrices (BE-DM) approach for computing molecular properties.
• To assess the potential of BE-DM for enhancing machine learning applications in computational chemistry.

Main Methods:

• Utilized back-transformed bootstrap embedding (BE) to compute one- and two-particle density matrices (1 and 2-PDMs).
• Calculated dipole moments and atomic force gradients for various polar molecular systems.
• Employed the BE2/CCSD level of theory to evaluate the accuracy of the force gradients.

Main Results:

• Achieved a mean absolute error of approximately 0.003 au/bohr in atomic force gradients at the BE2/CCSD level.
• Demonstrated the stability of geometry optimization trajectories enabled by the computed forces, even in challenging environments.
• Showcased the efficiency of BE-DM in evaluating dipole moments and atomic force gradients.

Conclusions:

• Bootstrap Embedding-Direct Matrices (BE-DM) provide an efficient and accurate method for calculating molecular properties.
• The method's ability to generate coupled-cluster-level forces is highly promising for improving machine-learned potentials.
• BE-DM offers a significant advancement for applications requiring accurate molecular forces and properties, particularly in machine learning.