PDG: A Composite Method Based on the Resolution of the Identity

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Theoretical Chemistry

Background:

• The Gaussian-3 (G3) composite approach is a standard for calculating thermochemical properties.
• Computationally intensive methods like Møller-Plesset perturbation theory (MP2) and coupled-cluster singles-doubles with noniterative triples corrections (CCSD(T)) are key components of G3.
• These methods require significant computational resources, particularly memory for two-electron repulsion integrals (ERIs).

Purpose of the Study:

• To investigate the application of the resolution-of-the-identity (RI) approximation to computationally demanding methods within the G3 framework.
• To develop and evaluate a new variant of the G3 approach, termed RI-G3 or PDG, that utilizes the RI approximation for improved efficiency.
• To assess the impact of the RI approximation on the accuracy and computational performance of G3 calculations.

Main Methods:

• Applied the resolution-of-the-identity (RI) approximation to two-electron repulsion integrals (ERIs) in MP2 and CCSD(T) calculations.
• Implemented RI-based MP2 and CCSD(T) methods using a hybrid distributed/shared memory model (MPI and OpenMP).
• Compared the accuracy and performance of the new RI-G3/PDG scheme against the standard G3 approach using the G3/99 test set for heats of formation.

Main Results:

• The RI approximation introduced negligible changes to the mean absolute errors (<0.1 kcal/mol) in heats of formation compared to standard G3.
• The standard deviations of the errors remained unaltered, indicating preserved accuracy.
• RI-MP2 energy and gradient computations achieved a 7.5× speedup.
• RI-CCSD(T) calculations nearly halved memory demand and provided a 4-5× speedup for CCSD iterations, significantly reducing computation time.

Conclusions:

• The RI-G3/PDG scheme offers a computationally efficient alternative to the standard G3 approach for thermochemical property calculations.
• The RI approximation effectively reduces computational cost and memory requirements without compromising accuracy.
• This optimized approach enables faster and more resource-friendly high-level quantum chemical calculations.