Pith. sign in

REVIEW 4 major objections 5 minor 58 references

Autoregressive pairwise Graphical Models efficiently find ground state representations of stoquastic Hamiltonians

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Simple pairwise model beats neural nets at stoquastic ground states

desk verdict A useful, simple exactly-samplable ansatz with a resource-limited performance claim that is supported only qualitatively; the pairwise-sufficiency motivation is thin, and the benchmark is uneven. read the letter →

arxiv 2505.06798 v1 pith:RONQM4TU submitted 2025-05-11 quant-ph

Yuchen Pang , Abhijith Jayakumar , Evan McKinney , Carleton Coffrin , Marc Vuffray , Andrey Y. Lokhov This is my paper · ORCID
classification quant-ph
keywords autoregressivegraphicalmodelsstoquasticHamiltoniansvariationalMonteCarlopairwiseenergyAnsatzexactsamplingfrustratedspinsystemsground-staterepresentationenergy-based
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper introduces Autoregressive Graphical Models (AGMs): variational states built by factoring a stoquastic ground-state wavefunction into conditional probabilities, each modeled by a logistic function of an energy containing at most pairwise spin interactions. Working on small systems where the exact autoregressive representation can be computed, the authors find that pairwise terms dominate higher-order terms across the parameter range studied, motivating the same pairwise Ansatz for larger systems. They then show that this linear pairwise AGM, trained with the first-order ADAM optimizer, matches or beats a neural-network quantum state of equal parameter count trained with stochastic reconfiguration within the same wall-clock time on $10\times10$ transverse-field Ising and XXZ models, and converges faster on frustrated disordered and ANNNI models when compute is limited. The practical upshot is that a transparent pairwise model with exact sampling can be a better use of limited compute than a larger nonlinear black-box Ansatz for stoquastic spin problems.

What carries the argument

The machinery is the pairwise autoregressive conditional $P(\sigma_i \mid \sigma_{>i};\theta^{(i)}) = 1/(1+\exp(-2\sigma_i(\theta^{(i)}_i + \sum_{j>i} \theta^{(i)}_j \sigma_j)))$ for a fixed ordering of the $n$ spins. Multiplying these conditionals gives a normalized distribution from which independent samples are drawn in one forward pass, so variational Monte Carlo energy and gradient estimates need no Markov chain; this is the Autoregressive Graphical Model. The pairwise truncation is the load-bearing simplification: it keeps the model linear in the couplings, gives $O(n^2)$ evaluation cost, and matches the dominant order found by exact learning on small systems. Higher-order polynomial terms can be added to the same conditional form, so the pairwise choice is a deliberate modeling decision rather than an architectural ceiling.

What would settle it

Diagonalize a small frustrated 2D cluster exactly (for example an 8-spin ANNNI or $\pm J$ transverse Ising plaquette), compute its full autoregressive representation by interaction screening in the infinite-sample limit, and compare the $\ell^1$ norms of pairwise versus fourth-order parameters; if any higher-order norm exceeds the pairwise norm, the modeling premise fails for that case. A complementary check is to test whether the pairwise AGM's variational-energy error, measured against the exact ground state, decreases with $N_s$ in the frustrated regime or plateaus.

Watch

Extended reading notes

Core claim

The central claim is that a simple pairwise-energy AGM is sufficient to capture the ground-state properties of a variety of stoquastic Hamiltonians, including frustrated ones, and that this transparent model often outperforms more complex nonlinear models when both are compared under realistic wall-clock and sample budgets. The claim rests on exact learning of the full autoregressive representation for 7-spin 1D chains: in both the transverse-field Ising and XXZ ground states, the norm of the order-2 parameters dominates order-4 and other terms, indicating that pairwise interactions carry the essential structure. The paper argues this dominance justifies truncating each conditional in Eq. (7) at pairwise order, which yields an $O(n^2)$ Ansatz that is exactly samplable, and that the resulting model is especially effective in the quantum-dominated regime of non-frustrated models and in the short-time regime of frustrated models, where Markov-chain sampling is slow to mix.

Load-bearing premise

The load-bearing premise is that pairwise dominance measured by exact learning on 7-spin 1D chains transfers to larger 2D systems and to frustrated Hamiltonians under a fixed, unoptimized spin ordering, so the pairwise truncation in the autoregressive conditionals remains accurate there.

Editorial extensions

If this is right

  • On $10\times10$ ferromagnetic transverse-field Ising and antiferromagnetic XXZ models, the pairwise AGM reaches variational energies comparable to a neural-network state with the same parameter count, and outperforms it in the quantum-dominated regime, within the same compute-time budget.
  • On frustrated systems (disordered transverse Ising model and the 2D ANNNI model at small transverse field), the AGM converges to lower variational energy faster; the neural-network baseline catches up only after orders of magnitude more time.
  • Raising the number of samples $N_s$ systematically lowers the AGM's variational energy, while imposing known spin-flip and lattice-reflection symmetries helps mainly at low sample counts, suggesting the model learns the symmetries from data as samples increase.
  • Because the AGM is exactly samplable, it avoids slow-mixing Markov chains in frustrated regimes, making it a candidate warm-start or cheap first pass before more expensive variational simulations are run.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If pairwise dominance holds beyond the tested models, the AGM makes variational ground-state search for stoquastic systems nearly parameter-light: its structure is fixed by the Hamiltonian graph and a chosen variable order, leaving only a small set of coupling weights to learn.
  • The paper does not optimize the autoregressive variable ordering; comparing exact-learning norm profiles across random spin orderings is a direct way to test whether the pairwise advantage survives in stronger-frustration or larger systems.
  • A wall-clock comparison conflates per-sample cost with statistical efficiency; measuring gradient variance at equal sample counts could separate the benefit of exact i.i.d. sampling from the benefit of the linear parameterization.
  • The pairwise dominance found in the conditionals echoes the Jastrow wavefunction structure the paper mentions, so AGMs may offer a generative-model route to explaining when Jastrow-type Ansätze work and when higher-order corrections matter.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper introduces Autoregressive Graphical Models (AGMs) whose conditionals have pairwise energy functions, as a variational Ansatz for stoquastic spin Hamiltonians. For small 1D systems (7 spins), exact learning via Interaction Screening shows that pairwise terms dominate the autoregressive conditionals; this motivates using the pairwise AGM for larger systems. The authors compare AGMs trained with first-order stochastic gradient descent against NetKet neural-network quantum states (NNQS) trained with stochastic reconfiguration, and against tensor-network benchmarks, on 10×10 2D transverse-field Ising (TIM) and XXZ models, and on frustrated disordered TIM and ANNNI models. They find that AGMs achieve comparable or better variational energies in resource-limited settings, especially for frustrated systems where exact sampling avoids MCMC mixing problems. The paper claims that simple pairwise AGMs can efficiently represent ground states of a variety of stoquastic models.

Significance. If the results hold, the paper offers a simple, interpretable, and exactly samplable variational Ansatz that is competitive with more complex neural-network states in resource-limited settings. The use of exact learning to guide Ansatz selection is a valuable methodological idea, and the observation that exactly samplable models help in frustrated systems is practically relevant. However, the evidence has important gaps: the representational justification rests on small 1D systems, and the performance comparisons lack statistical robustness. With additional validation, the paper could make a useful contribution to the variational quantum Monte Carlo toolbox.

major comments (4)
  1. [III.A, III.B, III.C] The pairwise truncation in Eq. (7) is justified only by exact learning on 7-spin 1D chains (Fig. 1a,b). No exact-learning check is performed for 2D or frustrated systems, and the variable ordering in the autoregressive factorization is never specified. Because the conditional P(σ_i|σ_{>i}) is order-dependent, the pairwise Ansatz may not be able to represent the ground state for the 10×10 TIM/XXZ, D-TIM, or ANNNI models even approximately; the observed energy advantage then reflects only the optimization dynamics of an underparameterized model. The paper itself lists a principled ordering choice as future work (§IV), acknowledging that this representational question is unresolved. I recommend adding exact-learning diagnostics on small 2D and frustrated instances (e.g., 4×4 TIM and small frustrated systems) or, if infeasible, restricting the representational claim to the tested models and framing the results as an empirical resource-limited comparison only.
  2. [III.B, III.C (Figs. 2-4)] The central performance claim that AGM 'outperforms' NNQS is supported only by single-run energy-vs-time curves (Figs. 2-4) and a disorder-averaged difference (Fig. 3b) that is shown without error bars or the number of disorder realizations. For a statement about variational optimization, run-to-run fluctuations from weight initialization and stochastic sampling are substantial; without multiple seeds and standard errors, the claimed advantage could be within noise. The disorder-averaged experiment in Fig. 3b should report the mean and standard error over at least tens of disorder realizations.
  3. [III.B, III.C (and Appendix A.2)] The NNQS baseline is trained with a fixed set of hyperparameters (§A.2), while the AGM receives a low-cost hyperparameter optimization (§A.3). The authors state that the NN performance 'can presumably be increased further by using more intense hyperoptimization' (§III.B). This asymmetry biases the comparison in favor of AGM. To make the 'outperforms' claim credible, the NN baseline should either receive comparable hyperparameter tuning or the paper should report the sensitivity of the NN results to hyperparameter choices.
  4. [III.B, III.C] For the 2D and frustrated systems, the paper reports only variational energies relative to NNQS and a non-variational PEPS benchmark. It never reports the exact ground state energy (e.g., from exact diagonalization on small 2D systems or from extrapolated DMRG on quasi-1D systems) or the overlap/fidelity of the AGM state with the true ground state. Consequently, the title's claim that the method 'efficiently find[s] ground state representations' is not quantitatively supported: lower energy than a competing variational method does not imply that the state is a good representation of the ground state. Adding an accuracy benchmark (e.g., exact ground state energies for 4×4 and 6×6 systems) would substantially strengthen the paper.
minor comments (5)
  1. [Section II] 'Given an-qubit HamiltonianH' should read 'Given an n-qubit Hamiltonian H'; there are also broken sentences such as 'and this condition is satisfied for all local Hamiltonains'.
  2. [Section III.A] The text says 'we look at the sum of absolute values in the solution of (9) for every order (||θ*_k||_1 for order k)', but the caption of Fig. 1 says 'The magnitude of the largest term at each order'. These are different quantities; please clarify.
  3. [Section III.C] The description of the ANNNI model says 'anti-ferro magnetic interactions are only present in the horizontal directions', but the n.n.n. interactions are along the y-axis; this appears to be a typo for 'vertical directions'.
  4. [Sections II and III] The variable ordering used in the autoregressive factorization (e.g., raster-scan order) is never specified in the main text or the appendix; please state it explicitly.
  5. [Figure 2 caption] In Fig. 2, the TN result is shown as a line on the energy-vs-time plot, but the TN algorithm does not have a time-dependent energy curve; please clarify that this is a reference value.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation: the pairwise AGM claim is benchmarked against external DMRG/PEPS/NetKet results; small-system interaction screening is a diagnostic, not a fitted input.

full rationale

The derivation chain is not circular. The Ansatz P(σ_i|σ>i) = 1/(1+exp(-2σ_i(θ_i + Σ_{j>i} θ_j σ_j))) in Eq. (7) is a definition, not an output of the Hamiltonian, and the paper then tests it empirically. The 'pairwise dominance' finding in Sec. III.A is obtained by exact learning: 'we use this method as it gives a convex optimization method for finding the AGM... artificially taking the infinite sample limit' (Sec. III.A), and the resulting coefficient norms in Fig. 1 are a diagnostic, not parameters that are reused in the variational energies. The energies in Figs. 2-4 come from the Hamiltonian expectation in Eqs. (3)-(6) and are compared with external benchmarks: 'We compare the AGM Ansatz with a Neural network quantum state (NN) [45] with the same number of trainable parameters, and a Tensor network (TN) based approach' (Sec. III.B). Therefore the central claim is externally falsifiable. The paper does cite the authors' own interaction-screening consistency results ('From the consistency of the Interaction screening estimator (see [17, 47]) the solution of the optimization problem will uniquely represent the ground state in the autoregressive form'), but this is a published statistical theorem whose stated assumptions do not include the target result, so under the review rules it counts as independent support rather than circularity. The main limitation is an extrapolation, not circularity: pairwise dominance is measured on 7-spin 1D chains ('AGMs learned from these models are dominated by pairwise terms which implies the applicability of the pairwise autoregressive model as a variational Ansatz for finding these ground states'), and the paper itself postpones 'a principled way of choosing the order of variables producing the best autoregressive decompositions' (Conclusions). An untested extrapolation to 2D/frustrated models is a correctness risk, not a circularity. Score 1 reflects only the prevalence of self-citations in the methodological scaffolding, not any reduction of a prediction to a fitted input.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on standard variational Monte Carlo machinery plus two extra assumptions: the transferability of pairwise dominance from small 1D systems to larger 2D and frustrated systems, and the adequacy of a fixed variable ordering. Hyperparameters for AGM and fixed baselines for NNQS are tuned or chosen, and the PEPS reference uses post-hoc selection.

free parameters (3)
  • AGM hyperparameters alpha0 and gamma = alpha0 in 1e-2 to 1e-4, gamma in 0.8 to 1.0 (per model)
    Initial learning rate and 1000-step decay rate chosen by Hyperopt for each Hamiltonian; they affect the optimization trajectory and are fitted to make training work.
  • NNQS baseline hyperparameters = learning rate 0.01, diagonal shift 0.01
    Fixed for all NN runs, not tuned per model; this affects the fairness of the comparison and is an ad hoc choice.
  • PEPS benchmark settings = best of five settings (e.g., D=2 FU for TIM, D=6 SU for XXZ)
    Reference TN energies are the lowest among five update/bond-dimension settings, chosen post hoc; this is a benchmark selection rather than a model parameter.
assumptions (5)
  • standard math Ground states of stoquastic Hamiltonians can be chosen non-negative in the computational basis (Perron-Frobenius).
    Used in Sec II to reduce the variational problem to a classical probability distribution over spin configurations; cited [4].
  • domain assumption The Interaction Screening estimator is consistent and identifies the true conditional distribution in the infinite-sample limit.
    Used in Sec III.A to claim the convex objective in Eq (9) recovers the exact autoregressive representation; prior results cited [17,47].
  • ad hoc to paper The fixed ordering of variables in the autoregressive factorization yields conditionals that are well approximated by pairwise terms in the systems studied.
    The ordering is never specified or optimized, and the pairwise truncation is assumed to hold for larger 2D and frustrated systems based on 7-spin 1D evidence; this is load-bearing for the Ansatz.
  • ad hoc to paper The pairwise AGM Ansatz contains a sufficiently accurate approximation to the true ground state for the target Hamiltonians.
    The method's success depends on this representational assumption; supported only by the numerical benchmarks, not by an a priori argument.
  • standard math Monte Carlo estimates of the energy and gradient are unbiased with i.i.d. samples from the variational model.
    Standard VMC; used in Eqs (3)-(6).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Autoregressive pairwise Graphical Models efficiently find ground state representations of stoquastic Hamiltonians." pith.science (2026). https://pith.science/paper/RONQM4TU

@misc{pith2026250506798,
  author       = {Pith},
  title        = {Pith review of: Autoregressive pairwise Graphical Models efficiently find ground state representations of stoquastic Hamiltonians},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RONQM4TU}},
  note         = {Machine review of arXiv:2505.06798}
}
read the original abstract

We introduce Autoregressive Graphical Models (AGMs) as an Ansatz for modeling the ground states of stoquastic Hamiltonians. Exact learning of these models for smaller systems show the dominance of the pairwise terms in the autoregressive decomposition, which informs our modeling choices when the Ansatz is used to find representations for ground states of larger systems. We find that simple AGMs with pairwise energy functions trained using first-order stochastic gradient methods often outperform more complex non-linear models trained using the more expensive stochastic reconfiguration method. We also test our models on Hamiltonians with frustration and observe that the simpler linear model used here shows faster convergence to the variational minimum in a resource-limited setting.

Figures

Figures reproduced from arXiv: 2505.06798 by the authors.

Figure 1
Figure 1. FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5: The lowest energies per site found by the PEPS algorithm during ITE for different settings. The x axes are [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6: Energy vs Time for TIM model on a 10 [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7: Energy vs Time for the anti-ferromagnetic XXZ model on a 10 [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

58 extracted references · 48 canonical work pages

  1. [1]

    This implies that for lower values of g, HANNNI should be a hard model for MCMC-based VQMC methods compared to the exact sampling method

    has shown that this model has a highly degenerate ground state α = 1/3, and for an L×L lattice the num- ber of groundstates of the classical model is known to scale asO(2cL) [36, 40, 41] for a constant c. This implies that for lower values of g, HANNNI should be a hard model for MCMC-based VQMC methods compared to the exact sampling method. While for larg...

    work page

  2. [2]

    In our benchmarks, we adopt the Quimb li- brary [16] and the ITensor library [15] as reference im- plementations of tensor network algorithms

    T ensor networks Tensor networks are one of the state-of-arts techniques to solve the ground states of many-body quantum sys- tems [33]. In our benchmarks, we adopt the Quimb li- brary [16] and the ITensor library [15] as reference im- plementations of tensor network algorithms. The algorithm used for 1D TIM is the well-known density matrix renormalizatio...

    work page

  3. [3]

    Instead of defining the quantum state from a parameterized family of prob- ability distributions, now we define a ”Neural Quantum State” [5] (NQS), as the variational Ansatz

    Neural network quantum states We are interested in the same problem as before, to find the ground state by minimizing the expected value of the energy of our variational state. Instead of defining the quantum state from a parameterized family of prob- ability distributions, now we define a ”Neural Quantum State” [5] (NQS), as the variational Ansatz. In th...

    work page

  4. [4]

    Then the gradient could be estimated efficiently as in (6) with the help of automatic differentiation

    Autoregressive Graphical Models Each conditional in the AGM Ansatz in (7) can be seen as neural net with a single linear layer on σn−1, followed by multiplication with σi and a final sigmoid layer. Then the gradient could be estimated efficiently as in (6) with the help of automatic differentiation. So this Ansatz can be implemented as a simple neural net...

    work page

  5. [5]

    Solving the quan- tum many-body problem with artificial neural networks

    Giuseppe Carleo and Matthias Troyer. Solving the quan- tum many-body problem with artificial neural networks. Science, 355(6325):602–606, 2017

    2017

  6. [6]

    Deep learning of representations for un- supervised and transfer learning

    Yoshua Bengio. Deep learning of representations for un- supervised and transfer learning. In Proceedings of ICML workshop on unsupervised and transfer learning , pages 17–36. JMLR Workshop and Conference Proceedings, 2012

    work page 2012

  7. [7]

    Making a science of model search: Hyperparameter optimization in hundreds of dimensions for vision architectures

    James Bergstra, Daniel Yamins, and David Cox. Making a science of model search: Hyperparameter optimization in hundreds of dimensions for vision architectures. In International conference on machine learning, pages 115–

    work page

  8. [8]

    Gauge-invariant method for the±j spin-glass model

    JA Blackman and J Poulter. Gauge-invariant method for the±j spin-glass model. Physical Review B , 44(9):4374, 1991

    work page 1991

Show all 58 references
  1. [9]

    Monte carlo simulation of stoquastic hamiltonians

    Sergey Bravyi. Monte carlo simulation of stoquastic hamiltonians. arXiv preprint arXiv:1402.2295 , 2014

    2014 arXiv

  2. [10]

    Two- dimensional frustrated j 1- j 2 model studied with neural network quantum states

    Kenny Choo, Titus Neupert, and Giuseppe Carleo. Two- dimensional frustrated j 1- j 2 model studied with neural network quantum states. Physical Review B , 100(12):125124, 2019

    2019

  3. [11]

    Monte carlo simulation of a many-fermion study

    David Ceperley, Geoffrey V Chester, and Malvin H Ka- los. Monte carlo simulation of a many-fermion study. Physical Review B , 16(7):3081, 1977

    1977

  4. [12]

    Critical behavior of the ising spin- glass models in a transverse field

    Bikas K Chakrabarti. Critical behavior of the ising spin- glass models in a transverse field. Physical Review B , 24(7):4062, 1981

    1981

  5. [13]

    Symmetries and many-body excitations with neural-network quantum states

    Kenny Choo, Giuseppe Carleo, Nicolas Regnault, and Titus Neupert. Symmetries and many-body excitations with neural-network quantum states. Physical review let- ters, 121(16):167204, 2018

    2018

  6. [14]

    Fermionic neural-network states for ab-initio electronic structure

    Kenny Choo, Antonio Mezzacapo, and Giuseppe Carleo. Fermionic neural-network states for ab-initio electronic structure. Nature communications, 11(1):2368, 2020

    2020

  7. [15]

    White, and E

    Matthew Fishman, Steven R. White, and E. Miles Stoudenmire. The ITensor software library for tensor network calculations, 2020

    2020

  8. [16]

    Renormaliza- tion and tensor product states in spin chains and lat- tices

    J Ignacio Cirac and Frank Verstraete. Renormaliza- tion and tensor product states in spin chains and lat- tices. Journal of physics a: mathematical and theoretical , 42(50):504004, 2009

    2009

  9. [17]

    Markov random fields in statistics

    Peter Clifford. Markov random fields in statistics. Dis- order in physical systems: A volume in honour of John M. Hammersley , pages 19–32, 1990

    1990

  10. [18]

    Quan- tum entanglement in neural network states

    Dong-Ling Deng, Xiaopeng Li, and S Das Sarma. Quan- tum entanglement in neural network states. Physical Re- view X , 7(2):021021, 2017

    2017

  11. [19]

    Transverse ising spin-glass model

    RR Dos Santos, Rita Zorzenon dos Santos, and M Kischinhevsky. Transverse ising spin-glass model. Physical Review B , 31(7):4694, 1985

    1985

  12. [20]

    Learning energy-based representations of quan- tum many-body states

    Abhijith Jayakumar, Marc Vuffray, and Andrey Y Lokhov. Learning energy-based representations of quan- tum many-body states. Physical Review Research , 6(3):033201, 2024

    2024

  13. [21]

    quimb: a python library for quantum in- formation and many-body calculations

    Johnnie Gray. quimb: a python library for quantum in- formation and many-body calculations. Journal of Open Source Software, 3(29):819, 2018

    2018

  14. [22]

    Learning of discrete graphical models with neu- ral networks

    Abhijith J, Andrey Lokhov, Sidhant Misra, and Marc Vuffray. Learning of discrete graphical models with neu- ral networks. Advances in Neural Information Processing Systems, 33:5610–5620, 2020

    2020

  15. [23]

    Many-body problem with strong forces

    Robert Jastrow. Many-body problem with strong forces. Physical Review, 98(5):1479, 1955

    1955

  16. [24]

    Discrete distributions are learnable from metastable samples

    Abhijith Jayakumar, Andrey Y Lokhov, Sidhant Misra, and Marc Vuffray. Discrete distributions are learnable from metastable samples. arXiv preprint arXiv:2410.13800, 2024

    2024 arXiv

  17. [25]

    From architectures to applica- tions: A review of neural quantum states

    Hannah Lange, Anka Van de Walle, Atiye Abedinnia, and Annabelle Bohrdt. From architectures to applica- tions: A review of neural quantum states. Quantum Sci- ence and Technology, 2024

    2024

  18. [26]

    Ac- curate determination of tensor network state of quantum lattice models in two dimensions

    Hong-Chen Jiang, Zheng-Yu Weng, and Tao Xiang. Ac- curate determination of tensor network state of quantum lattice models in two dimensions. Physical review letters , 101(9):090603, 2008

    2008

  19. [27]

    Classical simulation of infinite-size quantum lattice systems in two spatial di- mensions

    Jacob Jordan, Roman Or´ us, Guifre Vidal, Frank Ver- straete, and J Ignacio Cirac. Classical simulation of infinite-size quantum lattice systems in two spatial di- mensions. Physical review letters , 101(25):250602, 2008

    2008

  20. [28]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In Yoshua Bengio and Yann LeCun, editors, 3rd International Conference on Learn- ing Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings , 2015

    2015

  21. [29]

    Statistical mechanics: algorithms and computations, volume 13

    Werner Krauth. Statistical mechanics: algorithms and computations, volume 13. OUP Oxford, 2006

    2006

  22. [30]

    Symmetry- projected jastrow mean-field wave function in variational monte carlo

    Ankit Mahajan and Sandeep Sharma. Symmetry- projected jastrow mean-field wave function in variational monte carlo. The Journal of Physical Chemistry A , 123(17):3911–3921, 2019

    2019

  23. [31]

    Hybrid con- volutional neural network and projected entangled pair states wave functions for quantum many-particle states

    Xiao Liang, Shao-Jun Dong, and Lixin He. Hybrid con- volutional neural network and projected entangled pair states wave functions for quantum many-particle states. Physical Review B , 103(3):035138, 2021

    2021

  24. [32]

    Optimal structure and param- eter learning of Ising models

    Andrey Y Lokhov, Marc Vuffray, Sidhant Misra, and Michael Chertkov. Optimal structure and param- eter learning of Ising models. Science advances , 4(3):e1700791, 2018

    2018

  25. [33]

    Algorithms for finite projected entangled pair states

    Michael Lubasch, J Ignacio Cirac, and Mari-Carmen Banuls. Algorithms for finite projected entangled pair states. Physical Review B , 90(6):064425, 2014

    2014

  26. [34]

    Lack of ergodic- ity in the infinite-range ising spin-glass

    ND Mackenzie and Allan Peter Young. Lack of ergodic- ity in the infinite-range ising spin-glass. Physical Review Letters, 49(5):301, 1982

    1982

  27. [35]

    Efficient 2d tensor network simu- lation of quantum systems

    Yuchen Pang, Tianyi Hao, Annika Dugad, Yiqing Zhou, and Edgar Solomonik. Efficient 2d tensor network simu- lation of quantum systems. In SC20: International Con- ference for High Performance Computing, Networking, Storage and Analysis , pages 1–14. IEEE, 2020

    2020

  28. [36]

    Ground state of liquid he 4

    William Lauchlin McMillan. Ground state of liquid he 4. Physical Review, 138(2A):A442, 1965

    1965

  29. [37]

    Information, physics, and computation

    Marc Mezard and Andrea Montanari. Information, physics, and computation . Oxford University Press, 2009

    2009

  30. [38]

    A practical introduction to tensor net- works: Matrix product states and projected entangled pair states

    Rom´ an Or´ us. A practical introduction to tensor net- works: Matrix product states and projected entangled pair states. Annals of physics , 349:117–158, 2014. 10

    2014

  31. [39]

    Thermodynamic limit of density matrix renormalization

    Stellan ¨Ostlund and Stefan Rommer. Thermodynamic limit of density matrix renormalization. Physical review letters, 75(19):3537, 1995

    1995

  32. [40]

    Two-dimensional ising mod- els with competing interaction—a monte carlo study

    W Selke and ME Fisher. Two-dimensional ising mod- els with competing interaction—a monte carlo study. Zeitschrift f¨ ur Physik B Condensed Matter , 40:71–77, 1980

    1980

  33. [41]

    One-dimensional ising chain with competing interactions: Exact results and connection with other sta- tistical models

    S Redner. One-dimensional ising chain with competing interactions: Exact results and connection with other sta- tistical models. Journal of Statistical Physics , 25:15–23, 1981

    1981

  34. [42]

    Group con- volutional neural networks improve quantum state accu- racy

    Christopher Roth and Allan H MacDonald. Group con- volutional neural networks improve quantum state accu- racy. arXiv preprint arXiv:2104.05085 , 2021

    2021 arXiv

  35. [43]

    The 2d ±j ising spin glass: exact partition functions in polynomial time

    Lawrence Saul and Mehran Kardar. The 2d ±j ising spin glass: exact partition functions in polynomial time. Nu- clear Physics B , 432(3):641–667, 1994

    1994

  36. [44]

    The density-matrix renormalization group in the age of matrix product states

    Ulrich Schollw¨ ock. The density-matrix renormalization group in the age of matrix product states. Annals of physics, 326(1):96–192, 2011

    2011

  37. [45]

    Netket 3: Machine learning tool- box for many-body quantum systems

    Filippo Vicentini, Damian Hofmann, Attila Szab´ o, Dian Wu, Christopher Roth, Clemens Giuliani, Gabriel Pes- cia, Jannes Nys, Vladimir Vargas-Calderon, Nikita As- trakhantsev, et al. Netket 3: Machine learning tool- box for many-body quantum systems. arXiv preprint arXiv:2112....

    2021 arXiv

  38. [46]

    Finite-size behaviour of the two- dimensional annni model

    Walter Selke. Finite-size behaviour of the two- dimensional annni model. Zeitschrift f¨ ur Physik B Con- densed Matter, 43(4):335–344, 1981

    1981

  39. [47]

    Deep autoregressive models for the efficient variational simulation of many-body quantum systems

    Or Sharir, Yoav Levine, Noam Wies, Giuseppe Carleo, and Amnon Shashua. Deep autoregressive models for the efficient variational simulation of many-body quantum systems. Physical review letters , 124(2):020503, 2020

    2020

  40. [48]

    Renormalization algorithms for quantum-many body systems in two and higher dimensions

    Frank Verstraete and J Ignacio Cirac. Renormalization algorithms for quantum-many body systems in two and higher dimensions. arXiv preprint cond-mat/0407066 , 2004

    2004 arXiv

  41. [49]

    Matrix product states, projected entangled pair states, and variational renormalization group methods for quan- tum spin systems

    Frank Verstraete, Valentin Murg, and J Ignacio Cirac. Matrix product states, projected entangled pair states, and variational renormalization group methods for quan- tum spin systems. Advances in physics , 57(2):143–224, 2008

    2008

  42. [50]

    Density-matrix algorithms for quantum renormalization groups

    Steven R White. Density-matrix algorithms for quantum renormalization groups. Physical review b , 48(14):10345, 1993

    1993

  43. [51]

    On the other hand, for NNQS models the SR preconditioning gives an advantage over first-order methods [45]

    also observes that ADAM to be more suitable for models using an auto-regressive factorization when com- pared to SR. On the other hand, for NNQS models the SR preconditioning gives an advantage over first-order methods [45]. For the initialization & hyperparameters setting, we...

  44. [52]

    Restricted boltzmann machines for quantum states with non-abelian or anyonic symmetries

    Tom Vieijra, Corneel Casert, Jannes Nys, Wesley De Neve, Jutho Haegeman, Jan Ryckebusch, and Frank Verstraete. Restricted boltzmann machines for quantum states with non-abelian or anyonic symmetries. Physical review letters, 124(9):097201, 2020

    2020

  45. [53]

    Effi- cient learning of discrete graphical models

    Marc Vuffray, Sidhant Misra, and Andrey Lokhov. Effi- cient learning of discrete graphical models. Advances in Neural Information Processing Systems, 33:13575–13585, 2020

    2020

  46. [54]

    Interaction screening: Efficient and sample-optimal learning of Ising models

    Marc Vuffray, Sidhant Misra, Andrey Lokhov, and Michael Chertkov. Interaction screening: Efficient and sample-optimal learning of Ising models. In Advances in Neural Information Processing Systems , pages 2595– 2603, 2016

    2016

  47. [55]

    Generalization properties of neural network approxima- tions to frustrated magnet ground states

    Tom Westerhout, Nikita Astrakhantsev, Konstantin S Tikhonov, Mikhail I Katsnelson, and Andrey A Bagrov. Generalization properties of neural network approxima- tions to frustrated magnet ground states. Nature com- munications, 11(1):1–8, 2020

    2020

  48. [56]

    From tensor network quantum states to tensorial recurrent neural networks

    Dian Wu, Riccardo Rossi, Filippo Vicentini, and Giuseppe Carleo. From tensor network quantum states to tensorial recurrent neural networks. arXiv preprint arXiv:2206.12363, 2022

    2022 arXiv

  49. [57]

    Solving statisti- cal mechanics using variational autoregressive networks

    Dian Wu, Lei Wang, and Pan Zhang. Solving statisti- cal mechanics using variational autoregressive networks. Physical review letters , 122(8):080602, 2019

    2019

  50. [58]

    How transferable are features in deep neural net- works? Advances in neural information processing sys- tems, 27, 2014

    Jason Yosinski, Jeff Clune, Yoshua Bengio, and Hod Lip- son. How transferable are features in deep neural net- works? Advances in neural information processing sys- tems, 27, 2014. 11 200 400 600 −1.80 −1.75 −1.70 −1.65 HTIM, g = 0.5 200 400 600 −2.54 −2.52 −2.50 −2.48 −2.46 −...

    2014

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.