[Paper Review] Response to "Exponential challenges in unbiasing quantum Monte Carlo algorithms with quantum computers"

Joonho Lee, David R. Reichman+5 more|arXiv (Cornell University)|Jul 27, 2022

Quantum Computing Algorithms and Architecture4 citations

TL;DR

This paper responds to concerns about exponential scaling in quantum-classical quantum Monte Carlo (QC-QMC) algorithms, demonstrating that exponential challenges in estimating wavefunction overlaps depend critically on the choice of QMC method, system Hamiltonian, and trial/walker wavefunction forms. Numerical results show QC-QMC can remain efficient for certain parameter regimes—such as the 1D transverse field Ising model at Γ/J = 0.5—where signal-to-noise ratios remain stable with increasing system size, suggesting practical quantum advantage remains possible under favorable conditions.

ABSTRACT

A recent preprint by Mazzola and Carleo numerically investigates exponential challenges that can arise for the QC-QMC algorithm introduced in our work, "Unbiasing fermionic quantum Monte Carlo with a quantum computer." As discussed in our original paper, we agree with this general concern. However, here we provide further details and numerics to emphasize that the prospects for practical quantum advantage in QC-QMC remain open. The exponential challenges in QC-QMC are dependent on (1) the choice of QMC methods, (2) the underlying system, and (3) the form of trial and walker wavefunctions. While one can find difficult examples with a specific method, a specific system, and a specific walker/trial form, for some combinations of these choices, the approach is potentially more scalable than other near-term quantum algorithms. Future research should aim to identify examples for which QC-QMC enables practical quantum advantage.

Motivation & Objective

• To address concerns raised in a preprint about exponential scaling in QC-QMC algorithms.
• To clarify that exponential challenges in wavefunction amplitude estimation are not universally prohibitive across all system and method combinations.
• To demonstrate via numerical simulations that QC-QMC can remain efficient for certain parameter regimes, such as Γ/J = 0.5 in the 1D transverse field Ising model.
• To argue that QC-QMC may offer advantages over other near-term quantum algorithms like VQE and QPE in terms of noise resilience and measurement overhead.
• To encourage future research in identifying specific combinations of systems, methods, and wavefunctions that yield practical quantum advantage in QC-QMC.

Proposed method

• The authors analyze QC-QMC as a hybrid quantum-classical framework that estimates wavefunction overlaps ⟨ΨT|ϕ⟩ on a quantum computer, where |ΨT⟩ is a trial wavefunction and |ϕ⟩ is a walker state.
• They compare QC-AFQMC and QC-GFMC, showing that for non-interacting fermionic systems (e.g., mapped 1D TFIM), QC-AFQMC is exact without sampling, highlighting method dependence.
• Numerical simulations are performed on the 1D transverse field Ising model at Γ/J = 0.5 and Γ/J = 1.0, using both standard and sophisticated walker wavefunctions.
• Overlap distributions and local energy estimates are computed via repeated sampling from |ΨT|², with GFMC energy convergence analyzed as a function of measurement count M.
• The study compares standard computational basis state walkers with more complex walkers generated via spin flips from a reference state |ΨMC⟩, assessing their impact on signal-to-noise and sign problems.
• The authors evaluate QC-QMC against other quantum algorithms (VQE, QPE, QITE), emphasizing lower measurement overhead and noise resilience in specific scaling regimes.

Experimental results

Research questions

• RQ1 Does the exponential scaling of wavefunction overlap estimation in QC-QMC universally limit its practical utility across all systems and methods?
• RQ2 Can QC-QMC remain efficient and scalable for systems where the signal-to-noise ratio does not deteriorate with increasing system size?
• RQ3 How does the choice of QMC method—such as AFQMC versus GFMC—affect the performance and scalability of QC-QMC?
• RQ4 To what extent do the form of the trial and walker wavefunctions influence the exponential challenges in QC-QMC?
• RQ5 Under what specific combinations of system, method, and wavefunction form can QC-QMC demonstrate practical quantum advantage?

Key findings

• For the 1D transverse field Ising model at Γ/J = 0.5, the overlap distribution is sharper than at Γ/J = 1.0, leading to more accurate local energy estimates for high-overlap configurations.
• GFMC energy convergence at Γ/J = 0.5 shows weak system size dependence with increasing measurement count, in contrast to the strong L-dependence observed at Γ/J = 1.0.
• The use of a trial wavefunction |ΨT⟩ = exp[λ∑iσix](|↑↑⋯↑⟩ + |↓↓⋯↓⟩) with λ = 0.127 reproduces the exact ground state energy to within 99.8% for system sizes L = 6 to 12.
• QC-QMC does not exhibit exponential scaling in measurement overhead for the Γ/J = 0.5 regime, suggesting it may be more scalable than other near-term quantum algorithms in this parameter range.
• The introduction of sophisticated walkers—generated via spin flips from |ΨMC⟩—can improve overlap estimation but may reintroduce sign problems in otherwise sign-problem-free models.
• QC-QMC offers lower measurement overhead and greater noise resilience than VQE and QPE when expanding the basis set without increasing particle number, due to efficient electron correlation treatment outside the qubit space.