Scalable and Physics-Informed Multireference Implementation with Spin-Orbit Couplings via Modern HPC Clusters | JoVE Visualize

Area of Science:

Computational Chemistry
Quantum Mechanics
High-Performance Computing

Background:

Multireference (MR) calculations are crucial for accurately describing systems with strong electron correlation.
Existing computational methods face challenges in scalability and efficiency for large, complex molecular systems.
Physics-informed (PI) approaches offer potential for optimizing computational chemistry algorithms.

Purpose of the Study:

To develop a scalable and efficient physics-informed computational implementation for determinant-based MR calculations.
To enhance the performance and parallel scalability of MR calculations on modern hardware architectures.
To incorporate relativistic effects, including spin-orbit coupling (SOC), into the MR framework.

Main Methods:

A physics-informed (PI) philosophy is integrated into both the methodology and its computational implementation.
Orbital entanglement guides the reconstruction of selected configuration interaction (CI) wave functions.
PI kernel optimization (PIKO) and PI parallel optimization (PIPO) strategies are employed for efficient computation and hardware acceleration.
An entropy-based performance model and hierarchical load balancing are used for optimizing the MR module on HPC clusters.

Main Results:

Single-GPU accelerations achieve approximately 460 times the performance of full CPU cores.
Strong scaling efficiency exceeds 92.5% on up to 4000 GPUs in High-Performance Computing (HPC) clusters.
The method consistently incorporates scalar relativistic effects and allows for the inclusion of spin-orbit coupling (SOC).

Conclusions:

The developed PI-driven approach offers remarkable heterogeneous computing efficiency and parallel scalability for determinant-based MR calculations.
This methodology significantly advances the capability to perform accurate quantum chemical calculations on large and complex systems.
The implementation provides a robust framework for including relativistic effects in advanced electronic structure calculations.