Optimization of the Coupled Cluster Implementation in NWChem on Petascale Parallel Architectures

Area of Science:

• Computational Chemistry
• High-Performance Computing
• Quantum Chemistry

Background:

• The coupled cluster singles and doubles (CCSD) algorithm is crucial for accurate quantum chemical calculations.
• Large-scale CCSD computations face communication bottlenecks, limiting performance and scalability.
• Understanding DNA conformations requires accurate and efficient computational methods.

Purpose of the Study:

• To optimize the CCSD algorithm in NWChem to overcome communication bottlenecks.
• To improve the performance and scalability of CCSD calculations on supercomputers.
• To investigate the conformational energy differences in DNA using various quantum chemical methods.

Main Methods:

• Optimization of the CCSD algorithm within the NWChem software package.
• Execution of large-scale CCSD(T) calculations on the NCSA Blue Waters supercomputer.
• Comparison of MP2, CCSD, and CCSD(T) methods for analyzing DNA conformational energies.

Main Results:

• Achieved a 2-fold to 5-fold speedup in CCSD iteration time and improved scaling to 20,000 nodes.
• Demonstrated a sustained performance of 0.32 petaflop/s for a complete CCSD(T) calculation.
• Observed good agreement between MP2 and CCSD methods for DNA conformational analysis, highlighting MP2's utility.

Conclusions:

• The optimized CCSD algorithm offers significant performance gains and scalability for large quantum chemistry problems.
• MP2 calculations show promise for conformational analysis of DNA, providing results consistent with more computationally intensive methods.
• Discrepancies with classical force fields indicate a need for improved dihedral parameters in molecular simulations of DNA.