Parallel Calculation of CCSD and CCSD(T) Analytic First and Second Derivatives

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• High-Performance Computing

Background:

• Coupled-cluster theory is a cornerstone of accurate electronic structure calculations.
• Calculating molecular properties with high accuracy often requires computationally intensive methods like CCSD(T).
• Efficient parallelization is crucial for tackling large-scale quantum chemistry problems on modern hardware.

Purpose of the Study:

• To develop and implement a parallel adaptation of efficient coupled-cluster algorithms (CCSD and CCSD(T)).
• To enable the calculation of energies, gradients, and analytic second derivatives for large molecular systems.
• To demonstrate the feasibility of large-scale calculations on affordable cluster architectures.

Main Methods:

• A minimal-effort strategy focusing on parallelizing time-determining steps of CCSD and CCSD(T) calculations.
• Amplitude-replicated and communication-minimized implementation for parallel processing.
• Utilizing standard communication networks (e.g., Gigabit Ethernet) on compute nodes with sufficient memory and disk space.

Main Results:

• Successful parallel implementation of CCSD and CCSD(T) energies, gradients, and analytic second derivatives.
• Demonstrated efficiency for large-scale coupled-cluster calculations, reducing overall computational time.
• Feasibility of CCSD(T) calculations for systems exceeding 1000 basis functions, including vibrational frequencies and NMR chemical shifts.

Conclusions:

• The parallel adaptation provides an efficient scheme for large-scale coupled-cluster computations on affordable hardware.
• The implementation supports various reference types (UHF, ROHF) and methods (CCSD, CCSD(T), EOM-CCSD).
• Benchmark calculations and applications confirm the efficiency and capability for complex molecular property predictions.