New algorithm for tensor contractions on multi-core CPUs, GPUs, and accelerators enables CCSD and EOM-CCSD...

Area of Science:

• Computational Chemistry
• High-Performance Computing
• Quantum Mechanics

Background:

• Tensor contraction is a fundamental operation in quantum chemistry.
• Existing methods often lack hardware agnosticism and efficient data management.
• Large-scale electronic structure calculations require optimized algorithms.

Discussion:

• The presented algorithm offers hardware-agnostic tensor contraction for arbitrary symmetry and sparsity.
• It efficiently repackages tensor blocks for GPU acceleration and utilizes asynchronous data transfer.
• The implementation is available as open-source libxm, integrated with libtensor and Q-Chem.

Key Insights:

• Enables canonical coupled-cluster (CCSD) and equation-of-motion coupled-cluster (EOM-CCSD) calculations with over 1000 basis functions on a single quad-GPU.
• Demonstrates theoretical O(N^6) scaling for CCSD calculations, independent of disk data size.
• Achieves efficient GPU-enabled computations without altering higher-level quantum chemistry codes.

Outlook:

• Potential for accelerating a wide range of quantum chemistry and materials science simulations.
• Facilitates larger and more complex electronic structure investigations.
• Further integration with other computational chemistry packages is anticipated.