SIESTA-SIPs: Massively parallel spectrum-slicing eigensolver for an ab initio molecular dynamics package.

Area of Science:

• Computational materials science
• Electronic structure theory
• High-performance computing

Background:

• Integration of the Shift-and-Invert Parallel Spectral Transformation (SIPs) eigensolver from the SLEPc library into the SIESTA ab initio molecular dynamics package.
• Demonstration of the code's effectiveness on polyethylene chains, boron nitride sheets, and bulk water clusters.

Discussion:

• Performance of the SLEPc eigensolver is matrix-dependent, favoring sparser matrices in lower-dimensional systems (polyethylene, boron nitride) over denser matrices in bulk systems (water).
• SIESTA-SIPs demonstrates superior performance compared to SIESTA's default ScaLAPACK solver for large numbers of orbitals and cores.
• A novel iterative load-balancing method exploiting eigenvalue spectrum knowledge improves computational efficiency during self-consistency cycles.

Key Insights:

• The SIESTA-SIPs integration provides a significant performance boost for large-scale electronic structure calculations.
• Matrix sparsity is a critical factor influencing eigensolver performance in molecular dynamics simulations.
• Adaptive load balancing enhances computational efficiency in iterative electronic structure methods.

Outlook:

• Potential for accelerating large-scale quantum mechanical simulations in materials science and chemistry.
• Further optimization of eigensolvers for complex, bulk material systems.
• Exploration of advanced load-balancing techniques for even greater computational efficiency.