Machine Learning K-Means Clustering Algorithm for Interpolative Separable Density Fitting to Accelerate Hybrid...

Area of Science:

• Computational Chemistry
• Materials Science
• Machine Learning

Background:

• Hartree-Fock exchange (HFX) calculations using numerical atomic orbitals (NAOs) are computationally expensive and memory-intensive.
• The interpolative separable density fitting (ISDF) method offers a low-rank decomposition approach to mitigate these costs.
• Current ISDF methods rely on computationally demanding procedures like QR factorization with column pivoting (QRCP) for selecting interpolation points.

Purpose of the Study:

• To introduce a novel machine learning K-means clustering algorithm for selecting interpolation points in ISDF.
• To accelerate hybrid functional calculations employing NAOs within the HONPAS package.
• To evaluate the accuracy and computational efficiency of the K-means-based ISDF method.

Main Methods:

• Implementation of a K-means clustering algorithm for ISDF interpolation point selection.
• Integration of the K-means-ISDF approach into the HONPAS computational chemistry package.
• Performance benchmarking against traditional QRCP methods for HFX calculations.

Main Results:

• The K-means-based ISDF method achieves comparable accuracy to existing methods for both molecular and solid-state systems.
• A significant reduction in computational cost for selecting interpolation points, nearly two orders of magnitude lower than QRCP.
• An overall speedup of approximately 10 times for ISDF-based HFX calculations.

Conclusions:

• The K-means clustering algorithm provides a computationally efficient and accurate alternative for ISDF point selection.
• This machine learning approach effectively accelerates hybrid functional calculations with NAOs.
• The developed method offers a practical solution for reducing the computational burden of large-scale electronic structure calculations.