Integrated Software/Hardware Execution Models for High-Accuracy Methods in Chemistry

TL;DR

The paper addresses the need for high-accuracy quantum chemistry on heterogeneous hardware by proposing a hybrid DMRG-DUCC workflow that combines Hermitian CC downfolding for dynamical correlation with DMRG for static correlation. It demonstrates a practical execution model by pairing Micron's CXL memory for memory-intensive downfolding tasks with Azure Quantum Elements cloud resources for compute-intensive DMRG, using ExaChem TAMM as the integration layer. The authors show that the DUCC-based downfolded Hamiltonians converge systematically and improve upon CCSD for challenging systems like Fe–nitrosyl and retinal cis–trans isomerization, while also highlighting limitations at large diradical character. The work provides a scalable, hardware-aware strategy for embedding quantum chemistry workflows across future exascale platforms and outlines clear directions for extending the approach with higher-rank many-body terms and classical/quantum hybrids.

Abstract

The effective deployment and application of advanced methodologies for quantum chemistry is inherently linked to the optimal usage of emerging and highly diversified computational resources. This paper examines the synergistic utilization of Micron memory technologies and Azure Quantum Element cloud computing in Density Matrix Renormalization Group (DMRG) simulations leveraging coupled-cluster (CC) downfolded/effective Hamiltonians based on the double unitary coupled cluster (DUCC) Ansatz. We analyze the performance of the DMRG-DUCC workflow, emphasizing the proper choice of hardware that reflects the numerical overheads associated with specific components of the workflow. We report a hybrid approach that takes advantage of Micron CXL hardware for the memory capacity intensive CC downfolding phase while employing AQE cloud computing for the less resource-intensive DMRG simulations. Furthermore, we analyze the performance of the scalable ExaChem suite of electronic simulations conducted on Micron prototype systems.

Figures (8)

• Figure 1: CreteA and CreteB clusters (each using 1 rack of equipment) built for PNNL under the AMAIS project. There are four power suplies at the top, two hosts in the middle and the bottom, 1 large chassis for the CXL switch, and 4 large chassis with multiple FPGA boards.
• Figure 2: Updating the ExaChem software stack to support CXL shared memory
• Figure 3: Comparison of CCSD per-iteration runtimes using [Fe$(\rm H_2O)_6$](3+) with the cc-pVTZ basis set (no frozen core): CXL FAM based implementation vs Global Arrays (GA) distributed memory implementation
• Figure 4: Isomers of the iron-nitrosyl complex. The orange color is for iron, gray for carbon, blue for nitrogen and red for oxygen.
• Figure 5: The energy differences (in eV) between Flat - Standard (F-S) and Reverse - Standard (R-S) isomers, computed using DMRG and DMRG-DUCC methods. The used active space is CAS(8,8), CAS(16,16) and CAS(32,32).