[Paper Review] Large-Scale Simulation of Quantum Computational Chemistry on a New Sunway Supercomputer

Honghui Shang, Li Shen+12 more|arXiv (Cornell University)|Jul 8, 2022

Cloud Computing and Resource Management4 citations

TL;DR

This paper presents Q2 Chemistry, a high-performance VQE simulator leveraging Matrix Product States (MPS) and Density Matrix Embedding Theory (DMET) to enable large-scale quantum computational chemistry simulations on the Sunway supercomputer. It achieves nearly linear weak and strong scaling up to 20 million cores, simulating systems with up to 100 atoms and 1000 qubits, demonstrating feasibility for benchmarking near-term quantum chemistry experiments.

ABSTRACT

Quantum computational chemistry (QCC) is the use of quantum computers to solve problems in computational quantum chemistry. We develop a high performance variational quantum eigensolver (VQE) simulator for simulating quantum computational chemistry problems on a new Sunway supercomputer. The major innovations include: (1) a Matrix Product State (MPS) based VQE simulator to reduce the amount of memory needed and increase the simulation efficiency; (2) a combination of the Density Matrix Embedding Theory with the MPS-based VQE simulator to further extend the simulation range; (3) A three-level parallelization scheme to scale up to 20 million cores; (4) Usage of the Julia script language as the main programming language, which both makes the programming easier and enables cutting edge performance as native C or Fortran; (5) Study of real chemistry systems based on the VQE simulator, achieving nearly linearly strong and weak scaling. Our simulation demonstrates the power of VQE for large quantum chemistry systems, thus paves the way for large-scale VQE experiments on near-term quantum computers.

Motivation & Objective

• To develop a scalable classical simulator for Variational Quantum Eigensolver (VQE) simulations to study quantum computational chemistry on near-term quantum computers.
• To overcome the exponential memory scaling of full configuration interaction (FCI) methods for large molecular systems.
• To enable large-scale, high-accuracy simulations of real-world quantum chemistry problems, such as protein-ligand interactions, prior to execution on noisy intermediate-scale quantum (NISQ) devices.
• To achieve near-linear strong and weak scaling on a supercomputer with up to 20 million cores, enabling benchmarking of near-term quantum algorithms.
• To validate the method on real chemical systems, including SARS-CoV-2 main protease with ligands, to predict binding affinities with quantum chemistry accuracy.

Proposed method

• The VQE simulator uses Matrix Product State (MPS) representation to reduce memory requirements and improve simulation efficiency for large quantum systems.
• It integrates Density Matrix Embedding Theory (DMET) with the MPS-based VQE to extend the simulation range beyond the limits of pure MPS.
• A three-level parallelization scheme is implemented—MPI-based domain decomposition, OpenMP threading, and GPU acceleration—enabling scaling to 20 million cores.
• Julia is used as the primary programming language, combining high-level expressiveness with performance close to native C/Fortran via Julia’s JIT compilation and optimized libraries.
• The simulator performs iterative VQE optimization with parametric quantum circuits, using Hartree-Fock geometry optimization and energy evaluation for ligand binding affinity prediction.
• Communication overhead is minimized via efficient MPI_Allreduce and MPI_Bcast operations, with less than 0.001 seconds per iteration for data transfer at scale.

Experimental results

Research questions

• RQ1 Can a classical VQE simulator scale to simulate large quantum chemistry systems with up to 1000 qubits on a modern supercomputer?
• RQ2 How does the combination of MPS and DMET improve the accuracy and scalability of VQE simulations for large molecules?
• RQ3 To what extent can the VQE simulator achieve strong and weak scaling on a supercomputer with up to 20 million cores?
• RQ4 Can the simulator accurately predict binding energies for real protein-ligand complexes, such as SARS-CoV-2 main protease with known inhibitors?
• RQ5 How does the performance of the Julia-based implementation compare to traditional HPC languages in terms of productivity and computational efficiency?

Key findings

• The Q2 Chemistry simulator achieved nearly linear weak and strong scaling on up to 20 million cores, with communication costs below 0.001 seconds per VQE iteration.
• The method successfully simulated a 103-atom system (Atazanavir) with 378 electrons and 1000 qubits using DMET-MPS-VQE, marking the largest reported quantum chemistry simulation with this approach.
• The binding energy of Candesartan cilexetil was predicted as -6.8 eV, in good agreement with experimental observations.
• Nirmatrelvir (in Paxlovid) showed a lower binding energy of -7.3 eV, indicating stronger binding than Candesartan cilexetil, consistent with its known antiviral potency.
• The use of Julia enabled high productivity and performance comparable to hand-optimized C/Fortran code, with efficient integration of MPI and optimized BLAS/LAPACK libraries.
• The simulator demonstrated that VQE can be reliably benchmarked on classical supercomputers before deployment on near-term quantum hardware with ~100 qubits.