Optimization of the Linear-Scaling Local Natural Orbital CCSD(T) Method: Improved Algorithm and Benchmark...

Area of Science:

• Computational chemistry
• Quantum chemistry
• Method development

Background:

• Coupled-cluster (CC) methods are accurate but computationally expensive for large systems.
• Local approximations reduce computational cost but require efficient implementations.
• Previous local second-order Møller-Plesset (LMP2) schemes demonstrated efficiency gains.

Purpose of the Study:

• To develop and optimize an efficient local natural orbital coupled-cluster [LNO-CCSD(T)] method.
• To extend the efficient domain construction technique from LMP2 to the CC level.
• To enable accurate and feasible quantum chemical calculations for large molecular systems.

Main Methods:

• Integral-direct, in-core domain construction adapted from LMP2.
• Integration of redundancy-free LMP2 and Laplace-transformed (T) algorithms.
• Significant improvements in memory demand, domain/LNO construction, auxiliary basis compression, and integral transformation.
• Careful examination of approximation accuracy on medium-to-large systems.

Main Results:

• The LNO-CCSD(T) method achieves asymptotically linear-scaling operation count and constant data storage.
• Benchmark calculations on systems up to 63 atoms show average errors <0.07% (correlation) and <0.34 kcal/mol (reaction) vs. canonical CCSD(T).
• Demonstrated feasibility of LNO-CCSD(T) calculations for a 2380-atom protein on a single processor.

Conclusions:

• The optimized LNO-CCSD(T) implementation provides a highly efficient and accurate approach for large molecular systems.
• The method inherits beneficial properties of local approximations, enabling unprecedented computational scales.
• This development opens possibilities for accurate electronic structure calculations on complex biological molecules.