Comment on "Beyond-classical computation in quantum simulation"
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.
Kai: We're looking at the paper: "Comment on "Beyond-classical computation in quantum simulation"".
Mira: Here's the gist of the paper. A recent article investigates whether Neural Quantum States can provide competitive results in simulating quantum annealing protocols when accounting for Monte-Carlo noise and large autocorrelation times The authors conclude that NQS can provide competitive results in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times between samples obtained from the final state Comparison of Simulation Methods The paper revisits concerns regarding Neural Quantum States by demonstrating that they can provide competitive results in certain scenarios when considering sampling budget King et al.
1: provided a comprehensive benchmark and comparison of various classical approaches, including tensor networks such as Matrix Product States (MPS), Projected Entangled Pair States (PEPS) as well as Neural Quantum States (NQS) against superconducting quantum processor units (QPU) in performing a quench/annealing dynamics to a spin glass Hamiltonian The quality of the time evolution is characterized at t = 0.6 · ta by the correlation error εc = sP i,j (cij − c˜ij) squared / c˜ij squared (3) where cij = ⟨σ z i σ z j⟩ denotes the two-point correlations obtained from the quantum processor or classical method in question and ˜cij the ground-truth values obtained from well-converged MPS simulations, with the sums running over all lattice points Analysis of Correlation Magnitude and Noise The analysis focuses on C(d) = ⟨ ci,j ⟩ i,j,i−j=d as a measure for the magnitude of the correlations at a certain distance d For some methods, in particular the QPU and NQS, an exact evaluation of the correlations cij is either not possible or becomes prohibitively expensive The samples are drawn according to the Born distribution of the final state and the sample size was NMC = 106 The expected MC-noise affecting the estimation of each long-range correlation is of order σMC ≈ p 1/NMC, (5) where p is defined in Eq.
Mira: (5) Decomposition of Correlation Error The correlation error can be decomposed into a systematic and a statistical contribution as εc(NMC) = r εsys + εstat/NMC (7) where the statistical error contribution is of the form ε(i,j) stat /NMC and a possible systematic, method-specific error ε(i,j) sys The fit approximates the data very well, providing a validation to the presented approach In particular, the correlation error we obtain using NQS lies below the one obtained using QPU as well as NQS as presented in Ref
1: Limitations and Future Directions The presented analysis is restricted to the simplest lattice geometry, as well as to ta ≤ 7ns Ref.
1: on the other hand, shows results obtained using the QPU for more involved geometries and annealing times of up to ta = 20ns They furthermore consider so-called lowprecision inputs, i.e.
Paper summary: Kai: They start by looking at this comparison between Matrix Product States and Neural Quantum States when simulating a spin glass Hamiltonian using a quench or annealing dynamics over different times, specifically at time alpha equals two nanoseconds >
Mira: And they show that for the distance-resolved magnitude of correlations, the data obtained using Neural Quantum States plateaus around zero point three at larger distances, which matches the size of the Monte Carlo error when you use a sampling budget of one hundred thousand samples >
Lev: That means once the noise in estimating those long-range correlations gets bigger than that zero point three value, you can stop waiting for more samples because that's where the estimation error starts to dominate >
Kai: They also look at how the correlation error itself scales with the sampling budget, and they find a fitting function that matches both two nanosecond and seven nanosecond annealing times >
Mira: The paper points out that for larger geometries and longer annealing times up to twenty nanoseconds, they test low precision inputs where the couplings are drawn randomly from a bimodal distribution between negative one and positive one >
Lev: They found that for these rugged energy landscapes, classical methods like tensor networks optimized with belief propagation sometimes show worse correlation errors compared to cases with uniform couplings >
Kai: And they mention that while another paper showed favorable performance of a time dependent variational Monte Carlo approach on larger diamond structured lattices, this study motivates more numerical experiments to really see the limits of Neural Quantum States in a broader context >
Conclusion: Mira: So when you look at the title, "Comment on Beyond-classical computation in quantum simulation," it tells you this paper isn't just reporting new hardware results; it’s a direct commentary on where current classical simulation methods stand versus what actual quantum hardware is capable of >
Kai: The authors are essentially saying that while the raw accuracy might look different at first glance, when you factor in the noise and how slow those samples take to collect, Neural Quantum States can be a viable tool for annealing protocols >
Lev: It’s about being realistic about the simulation budget; if you're not accounting for the Monte Carlo noise or those long autocorrelation times between samples from the final state, any claim about accuracy is misleading >
Mira: So what this means for us is that we can't just look at a single comparison result and assume one method always wins; we have to consider the practical constraints of how much data we can realistically gather and how noisy those measurements are >
Institute of Quantum Control, Forschungszentrum J¨ulich · Institute of Theoretical Physics, University of Regensburg · Institute for Theoretical Physics, University of Cologne
quant-ph
Submitted: 2026-07-09
Updated: 2026-07-09
Comments: 3 pages, 1 figure; Comment on arXiv:2403.00910v2
Journal ref: SciPost Phys. Core 9, 074 (2026)
DOI: 10.21468/SciPostPhysCore.9.4.074
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 79/100
The gist: A recent article investigates whether Neural Quantum States can provide competitive results in simulating quantum annealing protocols when accounting for Monte-Carlo noise and large autocorrelation…
Key concepts
- Neural Quantum States (NQS)
- NQS is a versatile method using artificial neural networks to approximate the many-body wave function in quantum systems. It acts as a variational ansatz, meaning it tries to find a good approximation of the true complex quantum state using trainable parameters within the neural network structure.
- Monte-Carlo Noise ($\sigma_{MC}$)
- This refers to random errors introduced during the simulation process, specifically when drawing samples from the final state distribution. The noise magnitude is estimated to be on the order of $p/N_{MC}$, where $p$ relates to the correlation function and $N_{MC}$ is the number of Monte-Carlo samples used, affecting how accurately long-range correlations are estimated.
- Correlation Error ($\epsilon_c$)
- This metric quantifies the difference between the two-point correlations calculated from a simulation method (like NQS or QPU) and the true ground-truth values. The analysis decomposes this error into systematic and statistical parts, showing that NQS's correlation error can be lower than that of the quantum processor under certain conditions.
Terminology
Summary
A recent article investigates whether Neural Quantum States can provide competitive results in simulating quantum annealing protocols when accounting for Monte-Carlo noise and large autocorrelation times The authors conclude that NQS can provide competitive results in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times between samples obtained from the final state
Comparison of Simulation Methods
The paper revisits concerns regarding Neural Quantum States by demonstrating that they can provide competitive results in certain scenarios when considering sampling budget King et al. [1] provided a comprehensive benchmark and comparison of various classical approaches, including tensor networks such as Matrix Product States (MPS), Projected Entangled Pair States (PEPS) as well as Neural Quantum States (NQS) against superconducting quantum processor units (QPU) in performing a quench/annealing dynamics to a spin glass Hamiltonian The quality of the time evolution is characterized at t = 0. p. 0 [1] provided a comprehensive benchmark and comparison of various classical
approaches, including tensor networks such as Matrix
Product States (MPS), Projected Entangled Pair States
(PEPS) as well as Neural Quantum States (NQS) against
superconducting quantum processor units (QPU) in performing a quench/annealing dynamics to a spin glass
Hamiltonian on several different lattice geometries and
for different annealing times. The quench Hamiltonian
takes on the form
H(t) = Γ(t/ta)HD + J (t/ta)HP (1)
HD = −σ
x
, HP =
X
i<j
Jijσ
z
i σ
z
j
(2)
with the annealing time ta and the driving/classical Ising
problem Hamiltonian HD/HP , respectively.Where the paper says it
Analysis of Correlation Magnitude and Noise
The analysis focuses on C(d) = ⟨ ci,j ⟩ i,j,i−j=d as a measure for the magnitude of the correlations at a certain distance d For some methods, in particular the QPU and NQS, an exact evaluation of the correlations cij is either not possible or becomes prohibitively expensive The samples are drawn according to the Born distribution of the final state and the sample size was NMC = 106 The expected MC-noise affecting the estimation of each long-range correlation is of order σMC ≈ p 1/NMC, (5) where p is defined in Eq. (5)
Decomposition of Correlation Error
The correlation error can be decomposed into a systematic and a statistical contribution as εc(NMC) = r εsys + εstat/NMC (7) where the statistical error contribution is of the form ε(i,j) stat /NMC and a possible systematic, method-specific error ε(i,j) sys The fit approximates the data very well, providing a validation to the presented approach In particular, the correlation error we obtain using NQS lies below the one obtained using QPU as well as NQS as presented in Ref [1]
Limitations and Future Directions
The presented analysis is restricted to the simplest lattice geometry, as well as to ta ≤ 7ns Ref. [1] on the other hand, shows results obtained using the QPU for more involved geometries and annealing times of up to ta = 20ns They furthermore consider so-called lowprecision inputs, i. p. 2 3
for more involved geometries and annealing times of up
to ta = 20ns. They furthermore consider so-called lowprecision inputs, i.Where the paper says it
Conclusion on Competitive Performance
NQS can provide competitive results for performing annealing protocols, even surpassing the QPU in terms of the correlation error for the considered instance of the disorder realization The authors conclude that NQS can provide competitive results in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times between samples obtained from the final state The correlation error we obtain using NQS lies below the one obtained using QPU as well as NQS as presented in Ref [1] The authors acknowledge insightful comments on the results by A. King and R. p. 0 In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al. In
this comment we revisit these concerns, demonstrating that NQS can provide competitive results
in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times
between samples obtained from the final state. In a recent paper, King et al.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
--- Page 4 ---
The numerical simulations were performed using the jVMC codebase [5] The authors acknowledge insightful comments on the results by A. King and R. Wiersema
--- Page 5 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5] p. 2 King and R. The numerical
simulations were performed using the jVMC codebase [5].Where the paper says it
--- Page 6 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 7 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 8 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 9 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 10 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 11 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 12 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 13 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 14 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 15 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 16 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 17 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 18 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 19 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 20 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 21 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 22 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 23 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 24 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 25 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 26 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 27 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 28 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 29 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 30 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 31 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 32 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 33 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 34 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 35 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 36 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 37 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 38 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 39 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 40 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 41 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 42 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 43 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 44 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 45 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 46 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 47 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 48 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 49 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 50 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 51 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 52 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 53 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
--- Page 54 ---
The authors acknowledge insightful comments on the results by A. King and R. Wiersema The numerical simulations were performed using the jVMC codebase [5]
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improved NQS systems can provide competitive results in quantum annealing simulations by accounting for Monte Carlo noise and large autocorrelation times between samples obtained from the final state. This allows them to achieve accuracy levels comparable to quantum processor units for specific lattice geometries and annealing times, as shown when considering
the mean over the absolute values of the twopoint correlations
. -
The analytical form of the correlation error can be decomposed into a systematic part and a statistical part: epsilon c(NMC) = r epsilon sys + epsilon stat / NMC. This decomposition allows researchers to gauge which error contribution, systematic or statistical, is more dominant for a given simulation setup.
-
NQS-based methods can be used to perform annealing protocols with accuracy that surpasses the QPU in terms of the correlation error for specific instances of disorder realization. This suggests that NQS can be a viable alternative for simulating certain quantum dynamics on classical hardware when properly accounting for sampling budget limitations.
Abstract
A recent article [Science 388, 199-204 (2025)] investigates the applicability of numerical methods and a quantum processor unit in simulating a quantum annealing protocol. One of the findings indicates that Neural Quantum States - a versatile variational ansatz for the many-body wave function based on artificial neural networks - fail to reach the same accuracy as the quantum processor. In this comment we revisit these concerns, demonstrating that NQS can provide competitive results in some of the cases when accounting for the Monte-Carlo noise and large autocorrelation times between samples obtained from the final state.
Sources
- Evaluating classical simulations with a quantum processor
- Challenging the Quantum Advantage Frontier with Large-Scale Classical Simulations of Annealing Dynamics
Related papers
- Computational regimes in matrix-product-state-based quantum trajectory simulations
- Toward chemically accurate quantum simulations on quantum hardware: A quantum-HPC hybrid approach
- Transversal surface-code game with reconfigurable qubits
- Fulqrum: a generalized framework for quantum subspace diagonalization
- Flux-Activated Resonant Control of a Bosonic Quantum Memory
- Machine learning the effects of many quantum measurements