Pith. sign in

REVIEW 2 major objections 3 minor 2 cited by

Algorithms for variational Monte Carlo calculations of fermion projected entangled pair states in the swap gates formulation and the detailed balance of tensor network sequential sampling

T0 review · 2 major / 3 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper derives a complete variational Monte Carlo pipeline for swap-gates fermion PEPS and claims a proof of detailed balance for the sequential sampling sweep.

desk verdict Useful swap-gates fPEPS VMC derivation, but the detailed balance theorem is false and must be corrected before publication. read the letter →

arxiv 2506.20106 v3 pith:FEL2BVPY submitted 2025-06-25 cond-mat.str-el

Yantao Wu , Zhehao Dai This is my paper
classification cond-mat.str-el
keywords fermionicprojectedentangledpairstatesswapgatesformulationvariationalMonteCarlosequentialsamplingdetailedbalanceboundarymatrixproductstateJordan-Wignersignstochasticreconfiguration
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

This paper supplies a complete variational Monte Carlo (VMC) recipe for fermionic projected entangled pair states (fPEPS) written in the swap-gates formulation, itemizing the amplitude contraction, the log-derivative, the local energy with Jordan-Wigner signs, and the Monte Carlo sweep. It argues that this formulation is no harder to implement than the bosonic one, and that the resulting VMC reaches energy-density errors around $10^{-5}$ on $10\times 10$ and $12\times 12$ free-fermion benchmarks while outperforming a plain boson PEPS in the same sampling framework. The paper also claims a proof of detailed balance for sequential sampling, the fixed-order one-coordinate-at-a-time update used in PEPS VMC, which it says was previously confused and unproved. If that proof is right, the sweep used throughout the algorithm has the Born distribution as its stationary law, so the Monte Carlo averages are justified. The practical payoff would be a transferable variational tool for interacting fermion models where auxiliary-field Monte Carlo suffers from sign problems.

What carries the argument

The load-bearing object is the swap-gates fPEPS tensor network, where every physical leg is extended to the bottom left and each crossing with a virtual bond inserts a $(-1)^{P_s P_k}$ sign, controlled by parity constraints on the tensors. On the sampling side, the central mechanism is Algorithm 1: a deterministic sweep over coordinates, each resampled from a conditional distribution that itself satisfies detailed balance with respect to $\pi$. The authors' proof of Theorem 1 is the multiplication of the conditional detailed-balance equations for the sweep steps, which is meant to yield the global detailed-balance equation $p(s\to s')\pi(s)=p(s'\to s)\pi(s')$. The boundary-MPS environments $\mathrm{LE}[x]$ and $\mathrm{RE}[x]$ do the computational work, making amplitude, derivative, and local-energy contractions polynomial and letting one sweep cost $O(L_x L_y D'^2 D^4)$.

What would settle it

Enumerate all transition probabilities of Algorithm 1 on a two-variable target with correlations, e.g. $\pi(s_1,s_2)\propto\exp(J s_1 s_2)$ with each $p_i$ chosen to satisfy conditional detailed balance, and compute the 4-by-4 transition matrix $T$ of one sweep. If any pair satisfies $T(s\to s')\pi(s)\neq T(s'\to s)\pi(s')$, the claim as stated is false; a correlated two-variable example is already enough to exhibit such a violation.

Watch

Extended reading notes

Core claim

The central claim is that fPEPS variational Monte Carlo can be carried out entirely in the swap-gates formulation with four clear subroutines: boundary-MPS environments give the amplitude $\langle s|\Psi\rangle$; log-derivatives are single-site hole contractions; local hopping energies are amplitudes of a locally changed sample multiplied by the Jordan-Wigner phase $(-1)^{\#_{ij}}$; and a Metropolis sweep over nearest-neighbor bonds, reusing left and right environments, generates samples. As a separate theoretical claim, the paper states Theorem 1: Algorithm 1, the fixed-order sequential sweep in which spin $i$ is resampled from a conditional kernel obeying conditional detailed balance, satisfies global detailed balance with stationary distribution $\pi(s)$; the proof multiplies the conditional detailed-balance equations for the sweep's steps. Accepted, this establishes reversibility of the sampling method introduced in Ref. [20] and guarantees, with irreducibility and aperiodicity, convergence of energy and gradient estimators to their Born-rule expectations.

Load-bearing premise

The reversibility claim hinges on multiplying the per-coordinate detailed-balance equations in sweep order to obtain the global detailed-balance equation; this assumes the reverse of a sweep is literally the reverse sequence of the same conditional moves with the same conditioning coordinates. If that assumption fails, the proof's conclusion is not established, and the sweep's stationary distribution is not guaranteed to be the target $\pi(s)$.

Editorial extensions

If this is right

  • The swap-gates formulation now has a documented VMC pipeline: any sign-carrying local Hamiltonian term can be estimated from amplitudes of locally flipped samples, so implementing a new fermion model amounts to writing its hopping graph and Jordan-Wigner phase.
  • Because the stationary distribution is claimed to be the Born distribution $p(s)=|\langle s|\Psi\rangle|^2/\langle\Psi|\Psi\rangle$, optimizing gradients via stochastic reconfiguration is consistent with minimizing the variational energy rather than a biased surrogate.
  • The benchmarks expect energy-density errors around $10^{-5}$ for free-fermion insulator, Dirac semimetal, and Chern-band systems, indicating the algorithm is ready for interacting targets such as chiral spin liquids and Hubbard-model superconductivity.
  • The comparison with boson PEPS indicates that, in this variational sampling approach, the Jordan-Wigner string costs nothing in gradient computation, so fPEPS should dominate boson PEPS for local fermion Hamiltonians as system size grows.

Reading between the lines

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

  • Editorial inference: if the proof's multiplication step can be repaired by making the reverse sweep sample coordinates in reverse order with reverse conditional kernels, then the correct stationary distribution is still $\pi$; the sweep as written may need that modification or a bias-correction term.
  • Editorial inference: the theorem, if valid, applies to any tensor-network probability distribution, not only fermionic Born distributions, so a correct proof would justify sequential Gibbs-type sweeps for general graphical models.
  • Editorial inference: a direct numerical test of Theorem 1 on a small correlated distribution, such as two or three spins with full enumeration of the transition matrix, would settle the reversibility question independently of the proof and should be the first check before production calculations.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 3 minor

Summary. The manuscript presents a derivation of variational Monte Carlo (VMC) algorithms for fermionic projected entangled pair states (fPEPS) in the swap-gates formulation. It explains how to compute amplitudes, log-derivatives, and local energies, describes a sequential Monte Carlo sweep with reuse of boundary MPS environments, and benchmarks the approach on free-fermion models. As a separate advertised contribution, it claims in Theorem 1 that the sequential sampling procedure satisfies detailed balance with respect to the target distribution.

Significance. If correct, the VMC derivation would be a useful reference for fPEPS calculations, giving explicit fermion-sign bookkeeping and complexity estimates. The numerical benchmarks on free-fermion systems support the practical utility of the algorithm. However, the paper's separate key theoretical result, the detailed-balance theorem for sequential sampling, is false as stated. This invalidates the main advertised novelty, although the algorithm derivation in Sections III.A-III.E is internally consistent and could underpin a corrected or weakened theoretical claim.

major comments (2)
  1. [III.F, Theorem 1 and Eq. (14)] The theorem is false. For N=2, the forward sweep transition is K((s1,s2)->(s1',s2')) = p1(s1'|s2) p2(s2'|s1'), while the reverse sweep transition is K((s1',s2')->(s1,s2)) = p1(s1|s2') p2(s2|s1). The product of the conditional detailed-balance equations used in the proof gives, on the reverse side, p1(s1|s2) p2(s2|s1') times pi(s'), which is not K(s'->s)pi(s') whenever the conditioning spins have changed. A concrete counterexample is pi(00)=pi(11)=0.4, pi(01)=pi(10)=0.1 with exact Gibbs conditionals: then pi(00)K(00->10)=0.4*0.2*0.2=0.016, whereas pi(10)K(10->00)=0.1*0.8*0.8=0.064, violating Eq. (14). The individual local updates are pi-invariant, so the sweep remains pi-invariant, but the advertised detailed-balance result is invalid.
  2. [III.E and III.F] The convergence statement that sequential sampling 'is guaranteed to converge to a unique stationary distribution p(s)' rests on Eq. (14). Since Eq. (14) is false, the manuscript lacks a valid proof of the stated guarantee. A correct argument would show that the composition of pi-invariant conditional updates is pi-invariant and then invoke irreducibility and aperiodicity on the finite state space; the authors should supply this argument or weaken the claim accordingly. This is load-bearing because the sequential sweep is the Monte Carlo update used in the VMC algorithm.
minor comments (3)
  1. [III.F, Algorithm 1] The phrase 'Denote the spin configuration as s' is ambiguous because after updating earlier coordinates the configuration is (s'_1,...,s'_{i-1}, s_i,...,s_N); please state the current configuration explicitly inside the loop.
  2. [II.A, Eq. (2)] The rendering '̸=' should be the standard not-equal symbol; as typeset it may confuse readers.
  3. [III.D, Eq. (13)] The local-energy formula would benefit from an explicit statement of how s' is obtained from s by moving a fermion between sites i and j, including the phase convention for the Jordan-Wigner string.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the detailed-balance proof is self-contained, and the one self-citation is to published prior work and is not load-bearing.

full rationale

The paper's advertised separate result, Theorem 1 in Sec. III.F, is proved entirely within the paper: Algorithm 1 is defined by local conditional updates required to satisfy per-coordinate detailed balance, and the theorem then attempts to derive the global detailed-balance equation (14) by multiplying the conditional equations. Whether or not that multiplication is mathematically valid is a correctness question, not a circularity one: the target equation is not assumed as an input, and the proof does not fit any parameter to the target distribution. The VMC benchmarks in Sec. IV are self-contained against exactly solvable free-fermion Hamiltonians with known ground-state energies, so there is no fitted input renamed as a prediction. The only prior-work support involving overlapping authors is the citation of Ref. [21] for the well-definedness and locality of the swap-gate fPEPS construction; that is an independently published mathematical result about the formalism and is not used to establish the detailed-balance theorem or any numerical result. No ansatz is smuggled in via citation for the central proof, and no known result is merely renamed. Accordingly, there is no significant circularity. If a reader believes the proof of Theorem 1 is invalid, that is a soundness objection, not a circularity objection; the appropriate challenge would be a counterexample to Eq. (14), not a claim that the conclusion is identical to its assumptions.

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

The central derivation is self-contained apart from the fPEPS well-definedness result imported from Ref. 21 and the erroneous commutativity-of-reversibility assumption inside the detailed balance proof. The benchmark section introduces hand-chosen hyperparameters that do not feed into the theorem.

free parameters (5)
  • fPEPS bond dimension D = 4
    Hand-chosen hyperparameter in benchmark calculations; numerical demonstrations depend on it but the algorithm derivation does not.
  • boundary MPS bond dimension D' = 12 (20 for Fermi)
    Hand-chosen compression bond dimension; affects accuracy of amplitude evaluation.
  • SR regulator epsilon = 0.001
    Regularizer added to the quantum metric in stochastic reconfiguration; chosen by hand.
  • step size delta = 0.1
    Learning rate in the gradient descent; chosen by hand.
  • number of MC samples per step = 4096
    Sample count per optimization step; chosen by hand.
assumptions (3)
  • ad hoc to paper The composition of individual reversible coordinate updates is itself reversible with respect to the same target distribution.
    Implicitly assumed in the proof of Theorem 1; it is false for deterministic systematic scans, so the theorem's derivation depends on an invalid premise.
  • domain assumption fPEPS well-definedness and locality of the swap gates formulation, as established in Ref. 21.
    Adopted from a prior paper by the same research group (Dai, Wu, Wang, Zaletel); the algorithm assumes this property.
  • domain assumption Z2 parity gauge constraint on fPEPS tensors (Eq. 2), making the amplitude independent of the physical leg crossing order.
    Standard defining property of fPEPS in the swap gates formulation; used throughout the amplitude derivation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Algorithms for variational Monte Carlo calculations of fermion projected entangled pair states in the swap gates formulation and the detailed balance of tensor network sequential sampling." pith.science (2026). https://pith.science/paper/FEL2BVPY

@misc{pith2026250620106,
  author       = {Pith},
  title        = {Pith review of: Algorithms for variational Monte Carlo calculations of fermion projected entangled pair states in the swap gates formulation and the detailed balance of tensor network sequential sampling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FEL2BVPY}},
  note         = {Machine review of arXiv:2506.20106}
}
read the original abstract

In recent years, the variational Monte Carlo (VMC) calculations of projected entangled pair states (PEPS) has emerged as a competitive method for computing the ground states of many-body quantum systems. This method is particularly important for fermion systems where sign problems are abundant. We derive and explain the algorithms for the VMC calculations of fermion PEPS in the swap gates formulation. As a separate key result, we prove the detailed balance of sequential sampling of tensor networks.

Figures

Figures reproduced from arXiv: 2506.20106 by the authors.

Figure 1
Figure 1. FIG. 1: A 4 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2: Energy density error [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3: Comparison between fermion PEPS versus [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Solving the Hubbard model with Neural Quantum States

    cond-mat.str-el 2025-07 conditional novelty 6.0 of 10

    A transformer-based neural quantum state with a new optimizer achieves state-of-the-art variational energies for the 2D Hubbard model and predicts a horizontal half-filled stripe ground state for t'=-0.2 at 1/8 doping.

  2. Functional Neural Wavefunction Optimization

    cond-mat.str-el 2025-07 conditional novelty 5.0 of 10

    The authors show that stochastic reconfiguration is L2 gradient descent on the unit sphere, and they derive projected inverse iteration, a shifted inverse iteration method for neural-network wavefunctions, with faster...

Reference graph

Works this paper leans on

35 extracted references · 31 canonical work pages · cited by 2 Pith papers

  1. [1]

    The different ways are equivalent due to the gauge constraint Eq

    Well-definedness: as one extends a physical leg, there are different ways by which virtual bonds are crossed by the physical leg. The different ways are equivalent due to the gauge constraint Eq. 2

    work page

  2. [2]

    1, only the tensors in that local region change

    Locality: when acted on by a fermion operator in a local region, even if separated by a long JW string, e.g.c † 5c9 in Fig. 1, only the tensors in that local region change. See the appendix of Ref. [21] for a detailed expla- nation of the above. In our algorithm, we adopt the configuration drawn in Fig. 1, where the physical legs are bent flat enough so t...

    work page

  3. [3]

    computing the amplitude⟨s|Ψ⟩

    work page

  4. [4]

    computing the log-derivative ∂⟨s|Ψ⟩/∂θ α ⟨s|Ψ⟩

    work page

  5. [5]

    computing the local energy ⟨s|H|Ψ⟩ ⟨s|Ψ⟩

    work page

  6. [6]

    We consider an fPEPS of sizeL x ×L y and bond dimensionD

    a Monte Carlo update to samplesfromp(s) We now detail the above in the context of fPEPS. We consider an fPEPS of sizeL x ×L y and bond dimensionD. A. fPEPS tensors after sampling For any given samples, the index of each physical leg at siteiis fixed tos i. The 5-leg tensorA[i] s lrdu then becomes a 4-leg tensor ˜A[i]lrdu =A[i] si lrdu with a totalZ 2 char...

    work page

  7. [7]

    before and after one step of sequential sampling. By the conditional detailed balance: p1(s1 →s′ 1|s2, s3)π(s) =p 1(s′ 1 →s1|s2, s3)π(s′ 1,s 2:3) p2(s2 →s′ 2|s′ 1, s3)π(s′ 1,s 2:3) =p 2(s′ 2 →s2|s′ 1, s3)π(s′ 1:2, s3) p3(s3 →s′ 3|s′ 1, s′ 2)π(s′ 1:2, s3) =p 3(s′ 3 →s3|s′ 1, s′ 2)π(s′) 5 Multiplying the three equations gives the detailed balance of sequent...

    work page 2000

  8. [8]

    E. Y. Loh, J. E. Gubernatis, R. T. Scalettar, S. R. White, D. J. Scalapino, and R. L. Sugar, Sign prob- lem in the numerical simulation of many-electron systems, Phys. Rev. B41, 9301 (1990)

    work page 1990

Show all 35 references
  1. [9]

    Jordan and E

    P. Jordan and E. P. Wigner, About the Pauli exclu- sion principle, Z. Phys.47, 631 (1928)

    1928

  2. [10]

    S. R. White, Density matrix formulation for quan- tum renormalization groups, Phys. Rev. Lett.69, 2863 (1992)

    1992

  3. [11]

    C. V. Kraus, N. Schuch, F. Verstraete, and J. I. Cirac, Fermionic projected entangled pair states, Phys. Rev. A81, 052338 (2010)

    2010

  4. [12]

    Barthel, C

    T. Barthel, C. Pineda, and J. Eisert, Contraction of fermionic operator circuits and the simulation of strongly correlated fermions, Phys. Rev. A80, 042333 (2009)

    2009

  5. [13]

    Piˇ zorn and F

    I. Piˇ zorn and F. Verstraete, Fermionic implemen- tation of projected entangled pair states algorithm, Phys. Rev. B81, 245110 (2010)

    2010

  6. [14]

    Shi, S.-H

    Q.-Q. Shi, S.-H. Li, J.-H. Zhao, and H.-Q. Zhou, Graded projected entangled-pair state represen- tations and an algorithm for translationally in- variant strongly correlated electronic systems on infinite-size lattices in two spatial dimensions, arXiv preprint arXiv:0907.5520 (2009)

    2009 arXiv

  7. [16]

    Corboz and G

    P. Corboz and G. Vidal, Fermionic multiscale entan- 7 glement renormalization ansatz, Phys. Rev. B80, 165129 (2009)

    2009

  8. [17]

    Z.-C. Gu, F. Verstraete, and X.-G. Wen, Grass- mann tensor network states and its renormalization for strongly correlated fermionic and bosonic states (2010), arXiv:1004.2563 [cond-mat.str-el]

    2010 arXiv

  9. [18]

    W.-Y. Liu, H. Zhai, R. Peng, Z.-C. Gu, and G. K.- L. Chan, Accurate simulation of the hubbard model with finite fermionic projected entangled pair states (2025), arXiv:2502.13454 [cond-mat.str-el]

    2025 arXiv

  10. [19]

    Verstraete and J

    F. Verstraete and J. I. Cirac, Renormalization algo- rithms for quantum-many body systems in two and higher dimensions (2004)

    2004

  11. [20]

    Lubasch, J

    M. Lubasch, J. I. Cirac, and M.-C. Ba˜ nuls, Algo- rithms for finite projected entangled pair states, Phys. Rev. B90, 064425 (2014)

    2014

  12. [21]

    W.-Y. Liu, S. Dong, C. Wang, Y. Han, H. An, G.- C. Guo, and L. He, Gapless spin liquid ground state of the spin- 1 2 J1 −J 2 heisenberg model on square lattices, Phys. Rev. B98, 241109 (2018)

    2018

  13. [22]

    Liu, S.-S

    W.-Y. Liu, S.-S. Gong, Y.-B. Li, D. Poilblanc, W.- Q. Chen, and Z.-C. Gu, Gapless quantum spin liquid and global phase diagram of the spin-1/2J 1 −J 2 square antiferromagnetic Heisenberg model, Science Bulletin67, 1034 (2022)

    2022

  14. [23]

    W.-Y. Liu, J. Hasik, S.-S. Gong, D. Poilblanc, W.- Q. Chen, and Z.-C. Gu, Emergence of gapless quan- tum spin liquid from deconfined quantum critical point, Phys. Rev. X12, 031039 (2022)

    2022

  15. [24]

    Liu, S.-S

    W.-Y. Liu, S.-S. Gong, W.-Q. Chen, and Z.-C. Gu, Emergent symmetry in quantum phase transition: From deconfined quantum critical point to gap- less quantum spin liquid, Science Bulletin69, 190 (2024)

    2024

  16. [25]

    W.-Y. Liu, D. Poilblanc, S.-S. Gong, W.-Q. Chen, and Z.-C. Gu, Tensor network study of the spin- 1 2 square-latticeJ 1−J2−J3 model: Incommensurate spiral order, mixed valence-bond solids, and multi- critical points, Phys. Rev. B109, 235116 (2024)

    2024

  17. [26]

    Liu, X.-T

    W.-Y. Liu, X.-T. Zhang, Z. Wang, S.-S. Gong, W.-Q. Chen, and Z.-C. Gu, Quantum criticality with emergent symmetry in the extended shastry- sutherland model, Phys. Rev. Lett.133, 026502 (2024)

    2024

  18. [27]

    Liu, Y.-Z

    W.-Y. Liu, Y.-Z. Huang, S.-S. Gong, and Z.-C. Gu, Accurate simulation for finite projected entangled pair states in two dimensions, Phys. Rev. B103, 235155 (2021)

    2021

  19. [28]

    Z. Dai, Y. Wu, T. Wang, and M. P. Zaletel, Fermionic isometric tensor network states in two di- mensions, Phys. Rev. Lett.134, 026502 (2025)

    2025

  20. [29]

    An fPEPS is a holomorphic ansatz

  21. [30]

    Sorella, Green function monte carlo with stochas- tic reconfiguration, Phys

    S. Sorella, Green function monte carlo with stochas- tic reconfiguration, Phys. Rev. Lett.80, 4558 (1998)

    1998

  22. [31]

    J. P. Provost and G. Vallee, Riemannian structure on manifolds of quantum states, Communications in Mathematical Physics76, 289 (1980)

    1980

  23. [32]

    H. C. Jiang, Z. Y. Weng, and T. Xiang, Accurate determination of tensor network state of quantum lattice models in two dimensions, Phys. Rev. Lett. 101, 090603 (2008)

    2008

  24. [33]

    Jordan, R

    J. Jordan, R. Or´ us, G. Vidal, F. Verstraete, and J. I. Cirac, Classical simulation of infinite-size quantum lattice systems in two spatial dimensions, Phys. Rev. Lett.101, 250602 (2008)

    2008

  25. [34]

    Corboz, R

    P. Corboz, R. Or´ us, B. Bauer, and G. Vidal, Simu- lation of strongly correlated fermions in two spatial dimensions with fermionic projected entangled-pair states, Phys. Rev. B81, 165104 (2010)

    2010

  26. [35]

    Hauschild and F

    J. Hauschild and F. Pollmann, Efficient numerical simulations with tensor networks: Tensor network python (tenpy), SciPost Physics Lecture Notes , 005 (2018)

    2018

  27. [36]

    Hauschild, J

    J. Hauschild, J. Unfried, S. Anand, B. Andrews, M. Bintz, U. Borla, S. Divic, M. Drescher, J. Geiger, M. Hefel, K. H´ emery, W. Kadow, J. Kemp, N. Kirchner, V. S. Liu, G. M¨ oller, D. Parker, M. Rader, A. Romen, S. Scalet, L. Schoonderwoerd, M. Schulz, T. Soejima, P. Thoma, Y....

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.