Static Subspace Approximation for Random Phase Approximation Correlation Energies: Implementation and Performance.

Area of Science:

• Computational Chemistry
• Materials Science
• Quantum Mechanics

Background:

• Accurate theoretical modeling of complex interfacial reactions requires methods beyond semilocal density functional theory.
• Many-body perturbation theory methods like Random Phase Approximation (RPA) offer higher accuracy but face computational challenges.

Purpose of the Study:

• To implement and evaluate Random Phase Approximation (RPA) calculations within the BerkeleyGW framework.
• To demonstrate the computational feasibility and performance of RPA for large, complex systems relevant to catalysis and electrochemistry.

Main Methods:

• Implementation of RPA calculations leveraging the static subspace approximation for efficient polarizability representation.
• Utilizing GPU-based supercomputing resources to overcome computational complexity.
• Analysis of computational cost scaling with system size and parallel performance (strong scaling).

Main Results:

• The implemented RPA method shows favorable computational performance on large, complex systems.
• Computational cost for RPA correlation energy scales linearly with system size up to 50,000 bands.
• Excellent strong scaling results across multiple supercomputers indicate high performance and portability.

Conclusions:

• Advances in computational power, particularly GPUs, make advanced methods like RPA accessible for complex chemical systems.
• The developed implementation provides a computationally efficient and accurate tool for studying catalysis and electrochemistry.
• The linear scaling of computational cost signifies a breakthrough in applying RPA to industrially relevant problems.