Time-dependent density-functional theory in massively parallel computer architectures: the OCTOPUS project.

Area of Science:

• Computational Physics
• Quantum Chemistry
• Materials Science

Background:

• Density-Functional Theory (DFT) is a powerful quantum mechanical modeling method.
• Time-Dependent DFT (TDDFT) extends DFT to study excited states and dynamic properties.
• Efficient computation for large systems is crucial for advancing scientific discovery.

Purpose of the Study:

• To present parallelization efforts for the Octopus code.
• To enhance the performance of real-time TDDFT calculations.
• To enable the study of large molecular systems on modern parallel architectures.

Main Methods:

• Focus on real-time TDDFT, directly propagating time-dependent Kohn-Sham equations.
• Implement a multi-level parallelization scheme combining real-time TDDFT scalability with real-space domain partitioning.
• Develop a scalable Poisson solver for efficiency.
• Utilize blocks of Kohn-Sham states for data parallelism on GPUs and standard processors.

Main Results:

• Demonstrated the scalability of real-time TDDFT in Octopus on massively parallel systems.
• Achieved efficient execution on supercomputers with thousands of processors and GPUs.
• Showed applicability of the GPU strategy for standard processor optimization.

Conclusions:

• Real-time TDDFT in Octopus is a viable method for excited-state calculations of large systems.
• The parallelization strategy effectively leverages modern supercomputing architectures.
• This work paves the way for more complex and larger-scale simulations in computational chemistry and physics.