StochasticGW-GPU: Rapid Quasi-Particle Energies for Molecules beyond 10,000 Atoms

Area of Science:

• Computational Physics
• Quantum Chemistry
• Materials Science

Background:

• Accurate computation of quasi-particle (QP) energies is crucial for understanding molecular and material properties.
• The GW approximation provides a robust theoretical framework for these calculations.
• Scaling computational cost with system size has limited the application of GW methods to smaller systems.

Purpose of the Study:

• To introduce StochasticGW-GPU, a new implementation of the Stochastic GW code optimized for GPUs.
• To demonstrate the enhanced performance and scalability of StochasticGW-GPU for large-scale electronic structure calculations.
• To compute QP energies and band gaps for large hydrogenated silicon clusters.

Main Methods:

• Utilizing the stochastic Resolution of the Identity (sROI) technique for semilinear scaling.
• Implementing main bottleneck steps on Graphics Processing Units (GPUs) for performance acceleration.
• Applying the StochasticGW-GPU code to hydrogenated silicon clusters (Si_xH_y) up to 10,001 atoms.

Main Results:

• Achieved substantial performance improvements over previous Stochastic GW versions.
• Successfully computed QP energies for systems with up to 35,144 electrons.
• Obtained individual QP energies with a statistical precision better than ±0.03 eV in under 1 hour.

Conclusions:

• StochasticGW-GPU enables efficient and accurate computation of QP energies for large molecular and material systems.
• The GPU acceleration significantly reduces computation time, making previously intractable system sizes accessible.
• This advancement facilitates high-throughput screening and discovery of novel materials with desired electronic properties.