pith. machine review for the scientific record. sign in

arxiv: 2604.24760 · v1 · submitted 2026-04-27 · 🪐 quant-ph · cond-mat.stat-mech· physics.comp-ph

Recognition: unknown

Contracting Tensor Networks with Generalized Belief Propagation

Joseph Tindall , Grace M. Sommers , Hilbert Kappen
Authors on Pith no claims yet

Pith reviewed 2026-05-08 04:08 UTC · model grok-4.3

classification 🪐 quant-ph cond-mat.stat-mechphysics.comp-ph
keywords tensor networksbelief propagationgeneralized belief propagationtensor contractionIsing modelAKLT stateice modelsquantum many-body systems
0
0 comments X

The pith

Generalized belief propagation with overlapping regions approximates contractions of two- and three-dimensional tensor networks.

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

The paper shows how to extend belief propagation to generalized belief propagation for tensor network contraction by passing messages across a hierarchy of overlapping regions. This turns the contraction problem into a set of fixed-point equations that can be solved numerically in general and analytically in special cases. The method is demonstrated on both finite and infinite networks in two and three dimensions. Concrete applications cover the partition function of the fully frustrated Ising model, ground-state degeneracy in three-dimensional ice models, observables on the deformed AKLT state, and norms of random tensor-network states. Standard belief propagation appears as the simplest special case within this framework.

Core claim

We implement the GBP algorithm for a number of different region choices on a range of two- and three-dimensional, infinite and finite tensor networks, solving the corresponding fixed point equations both numerically and, in certain tractable cases, analytically. Our examples include calculating the partition function of the fully frustrated Ising model, computing the ground state degeneracy of three-dimensional ice models, measuring observables on the deformed AKLT quantum state and evaluating the norm of randomly generated tensor network states.

What carries the argument

Generalized belief propagation on a hierarchy of overlapping regions of the tensor network, where messages are passed between regions to obtain consistent fixed-point solutions for the contraction.

Load-bearing premise

The hierarchy of overlapping regions must be chosen so that the resulting fixed-point solutions approximate the true contraction with sufficient accuracy and without systematic bias from the region selection itself.

What would settle it

A direct numerical comparison on any small, exactly solvable tensor network where the GBP result with a given region choice deviates from the known exact value beyond floating-point precision.

Figures

Figures reproduced from arXiv: 2604.24760 by Grace M. Sommers, Hilbert Kappen, Joseph Tindall.

Figure 1
Figure 1. Figure 1: FIG. 1. Examples of some of the GBP parent region choices (and resulting intersections) used in this work for the view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3 view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Results for contracting the norm tensor network for the deformed AKLT state view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. Message passing results for 100 random realizations of an view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. (Generalized) belief propagation on the classical Villain model. view at source ↗
read the original abstract

Recent years have seen a growing interest in the use of belief propagation - an algorithm originally introduced for performing statistical inference on graphical models - for approximate, but highly efficient, tensor network contraction. Here, we detail how to apply generalized belief propagation (GBP) - where messages are passed within a hierarchy of overlapping regions of the tensor network - to approximately contract tensor networks and obtain accurate results. The original belief propagation algorithm is a corner case of this approach, corresponding to a particularly simple choice of regions of the tensor network. We implement the GBP algorithm for a number of different region choices on a range of two- and three-dimensional, infinite and finite tensor networks, solving the corresponding fixed point equations both numerically and, in certain tractable cases, analytically. Our examples include calculating the partition function of the fully frustrated Ising model, computing the ground state degeneracy of three-dimensional ice models, measuring observables on the deformed AKLT quantum state and evaluating the norm of randomly generated tensor network states.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces the application of generalized belief propagation (GBP) to approximate tensor network contraction by passing messages over hierarchies of overlapping regions, generalizing standard belief propagation. It implements the approach for multiple region choices on 2D and 3D finite and infinite tensor networks, solving the resulting fixed-point equations both numerically and analytically in tractable cases. Demonstrations include the partition function of the fully frustrated Ising model, ground-state degeneracy of 3D ice models, observables on the deformed AKLT state, and norms of random tensor network states.

Significance. If the reported accuracies hold, the work provides an efficient message-passing framework for tensor network contraction that scales to higher dimensions and infinite systems, complementing existing methods like TRG or MPS-based approaches. The inclusion of analytic solutions in selected cases offers concrete validation points and could facilitate further theoretical analysis of GBP fixed points in tensor networks.

major comments (2)
  1. [§5] §5 (numerical results on Ising and ice models): the manuscript reports agreement with known analytic or exact values but provides no a-priori error bound or convergence analysis quantifying how the chosen region hierarchy controls truncation error; different region selections can yield distinct fixed points, so the sufficiency of accuracy for arbitrary target observables remains an assumption rather than a derived property.
  2. [§6] §6 (AKLT and random TN examples): while numerical and analytic checks are presented, the text does not derive or test a general relation between the GBP fixed-point distance to the exact contraction and the error in the reported observables (e.g., norm or expectation values), leaving open the possibility of systematic bias from region truncation.
minor comments (2)
  1. [§3] The definition of the region hierarchy and message-update rules in §3 would benefit from an explicit diagram or pseudocode to clarify the overlap structure for readers unfamiliar with GBP.
  2. [§3 and §4] Notation for the fixed-point equations (e.g., the normalization of messages) is occasionally inconsistent between the general GBP section and the specific model implementations; a unified table of symbols would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript and for the constructive comments. We address each major point below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [§5] §5 (numerical results on Ising and ice models): the manuscript reports agreement with known analytic or exact values but provides no a-priori error bound or convergence analysis quantifying how the chosen region hierarchy controls truncation error; different region selections can yield distinct fixed points, so the sufficiency of accuracy for arbitrary target observables remains an assumption rather than a derived property.

    Authors: We agree that an a priori error bound or general convergence analysis relating region hierarchy to truncation error would be desirable. Deriving such a bound is challenging because GBP fixed-point equations are nonlinear and the accuracy depends on both the tensor network structure and the specific region choice. In the manuscript we address this empirically by (i) solving the fixed-point equations analytically in tractable cases to verify the messages, (ii) comparing numerical results against known exact or analytic values for the fully frustrated Ising model and 3D ice models, and (iii) explicitly testing several region hierarchies to show how accuracy improves with larger overlapping regions. We will revise §5 to add a dedicated paragraph discussing the empirical nature of the validation, the sensitivity to region selection, and the fact that accuracy for arbitrary observables is supported by these comparisons rather than proven by a general theorem. revision: yes

  2. Referee: [§6] §6 (AKLT and random TN examples): while numerical and analytic checks are presented, the text does not derive or test a general relation between the GBP fixed-point distance to the exact contraction and the error in the reported observables (e.g., norm or expectation values), leaving open the possibility of systematic bias from region truncation.

    Authors: We acknowledge that a general theoretical relation between the distance of a GBP fixed point to the exact contraction and the resulting error in observables is not derived. Establishing such a relation without further assumptions on the tensor network is non-trivial. In the examples we compute observables directly from the approximate messages and validate them: for the deformed AKLT state we recover the expected analytic behavior, and for random tensor networks we obtain norms that match the exact value within the reported precision. These checks indicate that systematic bias remains small for the chosen regions. We will revise §6 to include an explicit discussion of this limitation together with the empirical evidence from the analytic and random cases. revision: yes

Circularity Check

0 steps flagged

No significant circularity; algorithmic procedure with direct numerical/analytical validation.

full rationale

The paper presents GBP as a standard algorithmic extension of belief propagation for tensor network contraction. It specifies region hierarchies, solves the resulting fixed-point equations numerically or analytically for concrete 2D/3D networks (Ising partition functions, ice models, AKLT norms), and reports observables obtained from those solutions. No derivation step reduces by construction to a fitted parameter, self-definition, or load-bearing self-citation; the results are explicit outputs of the iteration applied to the input tensors. The method is self-contained against external benchmarks via the reported examples, with no renaming of known results or smuggling of ansatzes.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the standard assumptions of message-passing convergence in GBP and the representability of tensor networks; no free parameters, invented entities, or ad-hoc axioms are stated in the abstract.

axioms (1)
  • domain assumption Fixed-point equations of generalized belief propagation on chosen regions admit stable solutions that approximate the tensor network contraction.
    Invoked when the paper states it solves the equations numerically and analytically for the examples.

pith-pipeline@v0.9.0 · 5469 in / 1236 out tokens · 53898 ms · 2026-05-08T04:08:12.867028+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

75 extracted references · 15 canonical work pages · 2 internal anchors

  1. [1]

    empty” region or a region that we have already “discovered

    Equation refeq:F is the difference of the expected energy −P x p(x) logT (x) and the en- tropy −P x p(x) logp (x). The minimum of F is minp F(p) =−logZ. For the norm network, Z resembles the partition sum of a probability distribution, with the important difference that the tv are positive semi-definite (psd) Hermitian matrices instead of positive scalars...

  2. [2]

    such that a = √ 3 directly corresponds to the AKLT state. The state |ψ(a)⟩ is the unique ground state of the parent Hamiltonian H= X ⟨i,j⟩ D−1 i (a)D−1 j (a)hi,jD−1 i (a)D−1 j (a) (26) with hi,j = (Si ·S j) +116 243(Si ·S j)2 + 16 243(Si ·S j)3.(27) The deformation a̸ = √ 3 breaks the SU(2) symmetry of the AKLT model to a Z2 symmetry around the z directio...

  3. [3]

    ground truth

    for large a. We compare results from BP, GBP with R1 regions and a “ground truth” from our im- plementation of CTMRG for infinite hexagonal C3v tensor networks [ 56, 57]. We use a large CTMRG 11 FIG. 4. Results for contracting the norm tensor network for the deformed AKLT state |ψ(a)⟩ (see Eq. (25)) on the honeycomb lattice in the thermodynamic limit.A.Er...

  4. [4]

    S. R. White, Density matrix formulation for quantum renormalization groups, Physical Review Letters69, 2863 (1992)

    1992

  5. [5]

    Verstraete, M

    F. Verstraete, M. M. Wolf, D. Perez-Garcia, and J. I. Cirac, Criticality, the Area Law, and the Compu- tational Power of Projected Entangled Pair States, Physical Review Letters96, 220601 (2006)

    2006

  6. [6]

    I. V. Oseledets, Tensor-Train Decomposition, SIAM Journal on Scientific Computing33, 2295 (2011)

    2011

  7. [7]

    Kourtis, C

    S. Kourtis, C. Chamon, E. Mucciolo, and A. Ruck- enstein, Fast counting with tensor networks, SciPost Physics7, 060 (2019)

    2019

  8. [8]

    Lubasch, P

    M. Lubasch, P. Moinier, and D. Jaksch, Multigrid renormalization, Journal of Computational Physics 372, 587 (2018)

    2018

  9. [9]

    Gourianov, M

    N. Gourianov, M. Lubasch, S. Dolgov, Q. Y. van den Berg, H. Babaee, P. Givi, M. Kiffner, and D. Jaksch, A quantum-inspired approach to exploit turbulence structures, Nature Computational Science2, 30 (2022)

    2022

  10. [10]

    V. Murg, F. Verstraete, and J. I. Cirac, Variational study of hard-core bosons in a two-dimensional op- tical lattice using projected entangled pair states, Physical Review A75, 033605 (2007)

    2007

  11. [11]

    Lubasch, J

    M. Lubasch, J. I. Cirac, and M.-C. Ba˜ nuls, Unifying projected entangled pair state contractions, New Journal of Physics16, 033014 (2014)

    2014

  12. [12]

    Nishino and K

    T. Nishino and K. Okunishi, Corner Transfer Ma- trix Renormalization Group Method, Journal of the Physical Society of Japan65, 891 (1996)

    1996

  13. [13]

    Or´ us and G

    R. Or´ us and G. Vidal, Simulation of two-dimensional quantum systems on an infinite lattice revisited: Cor- ner transfer matrix for tensor contraction, Physical Review B80, 094403 (2009)

    2009

  14. [14]

    Gray and S

    J. Gray and S. Kourtis, Hyper-optimized tensor net- work contraction, Quantum5, 410 (2021)

    2021

  15. [15]

    Pan and P

    F. Pan and P. Zhang, Simulation of Quantum Cir- cuits Using the Big-Batch Tensor Network Method, Physical Review Letters128, 030501 (2022)

    2022

  16. [16]

    Robeva and A

    E. Robeva and A. Seigal, Duality of graphical mod- els and tensor networks (2017), arXiv:1710.01437 [math.ST]

    work page arXiv 2017

  17. [17]

    Alkabetz and I

    R. Alkabetz and I. Arad, Tensor networks contrac- tion and the belief propagation algorithm, Physical 14 Review Research3, 023073 (2021)

    2021

  18. [18]

    Tindall and M

    J. Tindall and M. Fishman, Gauging tensor net- works with belief propagation, SciPost Phys.15, 222 (2023)

    2023

  19. [19]

    Evenbly, N

    G. Evenbly, N. Pancotti, A. Milsted, J. Gray, and G. K.-L. Chan, Loop series expansions for tensor networks (2025), arXiv:2409.03108 [quant-ph]

    work page arXiv 2025

  20. [20]

    J. Gray, G. Park, G. Evenbly, N. Pancotti, E. F. Kjønstad, and G. K.-L. Chan, Tensor network loop cluster expansions for quantum many-body problems (2025), arXiv:2510.05647 [quant-ph]

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  21. [21]

    Midha and Y

    S. Midha and Y. F. Zhang, Beyond belief prop- agation: Cluster-corrected tensor network con- traction with exponential convergence (2025), arXiv:2510.02290 [quant-ph]

    work page arXiv 2025

  22. [23]

    Belief Propagation and Tensor Network Expansions for Many-Body Quantum Systems: Rigorous Results and Fundamental Limits

    S. Midha, G. M. Sommers, J. Tindall, and D. A. Abanin, Belief propagation and tensor net- work expansions for many-body quantum systems: Rigorous results and fundamental limits (2026), arXiv:2604.03228 [quant-ph]

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  23. [24]

    B. J. Frey and D. MacKay, A revolution: Belief prop- agation in graphs with cycles, inAdvances in Neural Information Processing Systems, Vol. 10, edited by M. Jordan, M. Kearns, and S. Solla (MIT Press, 1997)

    1997

  24. [25]

    Krzakala, M

    F. Krzakala, M. M´ ezard, F. Sausset, Y. F. Sun, and L. Zdeborov´ a, Statistical-physics-based reconstruc- tion in compressed sensing, Phys. Rev. X2, 021005 (2012)

    2012

  25. [26]

    H. A. Bethe, Statistical theory of superlattices, Pro- ceedings of the Royal Society of London. Series A- Mathematical and Physical Sciences150, 552 (1935)

    1935

  26. [27]

    M´ ezard and A

    M. M´ ezard and A. Montanari, Reconstruction on trees and spin glass transition, Journal of Statistical Physics124, 1317 (2006)

    2006

  27. [28]

    Tindall, A

    J. Tindall, A. Mello, M. Fishman, M. Stoudenmire, and D. Sels, Dynamics of disordered quantum sys- tems with two- and three-dimensional tensor net- works (2025), arXiv:2503.05693 [quant-ph]

    work page arXiv 2025

  28. [29]

    Tindall, M

    J. Tindall, M. Fishman, M. Stoudenmire, and D. Sels, Efficient tensor network simulation of IBM’s kicked Ising experiment (2023), arXiv:2306.14887 [quant- ph]

    work page arXiv 2023

  29. [30]

    Begušić, J

    T. Beguˇ si´ c, J. Gray, and G. K.-L. Chan, Fast and converged classical simulations of evidence for the utility of quantum computing before fault tolerance, Science Advances10, eadk4321 (2024), https://www.science.org/doi/pdf/10.1126/sciadv.adk4321

    work page doi:10.1126/sciadv.adk4321 2024

  30. [31]

    G. Park, J. Gray, and G. K.-L. Chan, Simulating quantum dynamics in two-dimensional lattices with tensor network influence functional belief propaga- tion, Physical Review B112, 10.1103/7jzt-xhn6 (2025)

    work page doi:10.1103/7jzt-xhn6 2025

  31. [32]

    Vovrosh, S

    J. Vovrosh, S. Juli` a-Farr´ e, W. Krinitsin, M. Kaicher, F. Hayes, E. Gottlob, A. Kshetrimayum, K. Bidzhiev, S. B. J¨ ager, M. Schmitt, J. Tindall, C. Dalyac, T. Mendes-Santos, and A. Dauphin, Simulating dy- namics of the two-dimensional transverse-field ising model: a comparative study of large-scale classical numerics (2025), arXiv:2511.19340 [quant-ph]

    work page arXiv 2025

  32. [33]

    J. S. Yedidia, W. Freeman, and Y. Weiss, Generalized belief propagation, inAdvances in Neural Informa- tion Processing Systems, Vol. 13, edited by T. Leen, T. Dietterich, and V. Tresp (MIT Press, 2000)

    2000

  33. [34]

    Kikuchi, A theory of cooperative phenomena, Phys

    R. Kikuchi, A theory of cooperative phenomena, Phys. Rev.81, 988 (1951)

    1951

  34. [35]

    Pearl,Probabilistic reasoning in intelligent sys- tems: Networks of Plausible Inference(Morgan Kauf- mann, San Francisco, California, 1988)

    J. Pearl,Probabilistic reasoning in intelligent sys- tems: Networks of Plausible Inference(Morgan Kauf- mann, San Francisco, California, 1988)

    1988

  35. [36]

    K. P. Murphy, Y. Weiss, and M. I. Jordan, Loopy belief propagation for approximate inference: an em- pirical study, inProceedings of the Fifteenth Confer- ence on Uncertainty in Artificial Intelligence, UAI’99 (Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 1999) p. 467–475

    1999

  36. [37]

    Kolafa, Residual entropy of ices and clathrates from monte carlo simulation, The Journal of Chemi- cal Physics140, 204507 (2014), pMID: 24880301

    J. Kolafa, Residual entropy of ices and clathrates from monte carlo simulation, The Journal of Chemi- cal Physics140, 204507 (2014), pMID: 24880301

    2014

  37. [38]

    Hayashi, C

    T. Hayashi, C. Muguruma, and Y. Okamoto, Cal- culation of the residual entropy of ice ih by monte carlo simulation with the combination of the replica- exchange wang–landau algorithm and multicanonical replica-exchange method, The Journal of Chemical Physics154, 10.1063/5.0038157 (2021)

    work page doi:10.1063/5.0038157 2021

  38. [39]

    Vanderstraeten, B

    L. Vanderstraeten, B. Vanhecke, and F. Verstraete, Residual entropies for three-dimensional frustrated spin systems with tensor networks, Phys. Rev. E98, 042145 (2018)

    2018

  39. [40]

    Pancotti and J

    N. Pancotti and J. Gray, One-step replica symme- try breaking in the language of tensor networks (2023), arXiv:2306.15004 [cond-mat, physics:physics, physics:quant-ph]

    work page arXiv 2023

  40. [41]

    M. X. Cao and P. O. Vontobel, Double-edge factor graphs: Definition, properties, and examples, in2017 IEEE Information Theory Workshop (ITW)(IEEE,

  41. [42]

    Kappen and W

    H. Kappen and W. Wiegerinck, Novel iteration schemes for the cluster variation method, inAdvances in Neural Information Processing Systems, Vol. 14, edited by T. Dietterich, S. Becker, and Z. Ghahra- mani (MIT Press, 2001)

    2001

  42. [43]

    J. S. Yedidia, W. T. Freeman, and Y. Weiss, Under- standing belief propagation and its generalizations, inExploring Artificial Intelligence in the New Mil- lennium(Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 2003) p. 239–269

    2003

  43. [44]

    paramagnetic

    A simple example where uniform initialization of all messages isnota good choice is in the ferromagnetic phase of the Ising model, where this uniform initial- ization is an unstable “paramagnetic” fixed point of the update equations

  44. [45]

    C. Guo, D. Poletti, and I. Arad, Block belief propaga- tion algorithm for two-dimensional tensor networks, Phys. Rev. B108, 125111 (2023)

    2023

  45. [46]

    Villain, Spin glass with non-random interactions, Journal of Physics C: Solid State Physics10, 1717 (1977)

    J. Villain, Spin glass with non-random interactions, Journal of Physics C: Solid State Physics10, 1717 (1977)

    1977

  46. [47]

    M. E. Fisher and J. Stephenson, Statistical Mechan- ics of Dimers on a Plane Lattice. II. Dimer Corre- 15 lations and Monomers, Physical Review132, 1411 (1963)

    1963

  47. [48]

    M. E. Fisher, Statistical Mechanics of Dimers on a Plane Lattice, Physical Review132, 1411 (1963)

    1963

  48. [49]

    Forgacs, Ground-state correlations and univer- sality in two-dimensional fully frustrated systems, Physical Review B22, 4473 (1980)

    G. Forgacs, Ground-state correlations and univer- sality in two-dimensional fully frustrated systems, Physical Review B22, 4473 (1980)

    1980

  49. [50]

    Caravelli, Some exactly solvable and tunable frus- trated spin models, Physica A: Statistical Mechanics and its Applications594, 127007 (2022)

    F. Caravelli, Some exactly solvable and tunable frus- trated spin models, Physica A: Statistical Mechanics and its Applications594, 127007 (2022)

    2022

  50. [51]

    J. F. Nagle, Lattice statistics of hydrogen bonded crystals. i. the residual entropy of ice, Journal of Mathematical Physics7, 1484 (1966)

    1966

  51. [52]

    Pauling, The structure and entropy of ice and of other crystals with some randomness of atomic arrangement, Journal of the American Chemical So- ciety57, 2680 (1935)

    L. Pauling, The structure and entropy of ice and of other crystals with some randomness of atomic arrangement, Journal of the American Chemical So- ciety57, 2680 (1935)

    1935

  52. [53]

    J. C. Slater, Theory of the transition in kh2po4, The Journal of Chemical Physics9, 16 (1941)

    1941

  53. [54]

    R. J. Baxter,Exactly solved models in statistical mechanics(Courier Corporation, 2007)

    2007

  54. [55]

    crossover

    More precisely, the series expansion up to order n quoted in Table II sums over loop corrections to the BP fixed point from Eulerian cycles containing up to n degree-2 vertices and any number of degree-4 (“crossover”) vertices. Loops containing odd-degree vertices have weight zero [48]

  55. [56]

    [18] cannot be immediately applied since dense loops containing many degree-4 vertices do not decay sufficiently fast

    We further note that Nagle’s series involves discon- nected loops, which are problematic for convergence; while a cluster or cluster cumulant expansion might be preferred, the loose convergence guarantee from Ref. [18] cannot be immediately applied since dense loops containing many degree-4 vertices do not decay sufficiently fast

  56. [57]

    Huang, M

    C.-Y. Huang, M. A. Wagner, and T.-C. Wei, Emer- gence of xy-like phase in deformed spin- 3 2 aklt sys- tems, Phys. Rev. B94, 165130 (2016)

    2016

  57. [58]

    Pomata, C.-Y

    N. Pomata, C.-Y. Huang, and T.-C. Wei, Phase transitions of a two-dimensional deformed affleck- kennedy-lieb-tasaki model, Phys. Rev. B98, 014432 (2018)

    2018

  58. [59]

    Yang and P

    Q. Yang and P. Corboz, Efficient ipeps simulation on the honeycomb lattice via qr-based corner transfer matrix renormalization group, Phys. Rev. B113, 085109 (2026)

    2026

  59. [60]

    I. V. Lukin and A. G. Sotnikov, Variational optimiza- tion of tensor-network states with the honeycomb- lattice corner transfer matrix, Phys. Rev. B107, 054424 (2023)

    2023

  60. [61]

    Gonz´ alez-Garc´ ıa, S

    S. Gonz´ alez-Garc´ ıa, S. Sang, T. H. Hsieh, S. Boixo, G. Vidal, A. C. Potter, and R. Vasseur, Random insights into the complexity of two-dimensional ten- sor network calculations, Phys. Rev. B109, 235102 (2024)

    2024

  61. [62]

    Verstraete and J

    F. Verstraete and J. I. Cirac,Renormalization algo- rithms for Quantum-Many Body Systems in two and higher dimensions, Tech. Rep. (2004) arXiv:cond- mat/0407066 type: article

    work page arXiv 2004

  62. [63]

    Gray and G

    J. Gray and G. K.-L. Chan, Hyperoptimized approx- imate contraction of tensor networks with arbitrary geometry, Phys. Rev. X14, 011009 (2024)

    2024

  63. [64]

    J. Chen, J. Jiang, D. Hangleiter, and N. Schuch, Sign problem in tensor-network contraction, PRX Quantum6, 010312 (2025)

    2025

  64. [65]

    Z. Lin, Riemannian geometry of symmetric positive definite matrices via cholesky decomposition, SIAM Journal on Matrix Analysis and Applications40, 1353 (2019), https://doi.org/10.1137/18M1221084

    work page doi:10.1137/18m1221084 2019

  65. [66]

    Bergmann, Manopt.jl: Optimization on manifolds in Julia, Journal of Open Source Software7, 3866 (2022)

    R. Bergmann, Manopt.jl: Optimization on manifolds in Julia, Journal of Open Source Software7, 3866 (2022)

    2022

  66. [67]

    A. L. Yuille and A. Rangarajan, The convex-concave principle, inAdvances in Neural Information Process- ing Systems (NIPS 14), edited by T. G. Dietterich, S. Becker, and Z. Ghahramani (MIT Press, 2002) pp. 1031–1038

    2002

  67. [68]

    Heskes, K

    T. Heskes, K. Albers, and B. J. Kappen, Approxi- mate inference and constrained optimization, inPro- ceedings of the 19th Conference on Uncertainty in Ar- tificial Intelligence (UAI)(Morgan Kaufmann, 2003) pp. 313–320

    2003

  68. [69]

    M. S. Rudolph and J. Tindall, Simulating and sam- pling from quantum circuits with 2d tensor networks (2025), arXiv:2507.11424 [quant-ph]

    work page arXiv 2025

  69. [70]

    Fishman, S

    M. Fishman, S. R. White, and E. M. Stoudenmire, Codebase release 0.3 for ITensor, SciPost Phys. Code- bases , 4 (2022)

    2022

  70. [71]

    factor” and its adjacent vertex “variables,

    E. H. Lieb, Residual entropy of square ice, Phys. Rev. 162, 162 (1967). 16 Appendix A: Analytical solution to (G)BP Equations for Villain model In this Appendix we solve for fixed points of Eq. (9) for the Villain model. In Fig. 3, we represented the partition function as a tensor network where each four-index tensor includes interactions between four pai...

    1967

  71. [72]

    paramagnetic

    Simple BP In simple BP, the messages are passed from edges to vertices. There are two types of vertices: those involved in four ferromagnetic interactions, which we will denote + vertices, and those involved in two fer- romagnetic and two antiferromagnetic interactions, denoted−. As a result, there are four types of mes- sages, as indicated in Fig. 6B: (1...

  72. [73]

    correlation parameter

    GBP with plaquette regions Simple BP struggles with the Villain model because it does not account for the model’s most salient fea- ture: the frustration around each plaquette. In the main text, GBP R1 accounts for this frustration by including two types of parent regions, around vertices and plaquettes. In the factor graph language, the choice is even si...

  73. [74]

    physical

    =    a2 all indices 0 or all indices 1 1 P i zi =P i z′ i = 1 or P i zi =P i z′ i = 2 0 otherwise. (C2) Since the same double factor appears on each site, we seek solutions with identical messages mi(zi, z′ i) = m(z, z′) on each edge i = 1 , 2, 3. The messages m(z, z′) are psd Hermitian matrices, which we will take to be real symmetric and P z,z ′ m(...

  74. [75]

    On the endpoints of this interval, it is marginally stable, and coincides with fixed points (1) and (3) respectively

    (µ = 0, c = 1) is a physical fixed point for all a, but is only stable in the interval (1 , √ 5). On the endpoints of this interval, it is marginally stable, and coincides with fixed points (1) and (3) respectively. 3. µ=± q a2−5 a2−1 , c= 4 a2−1 is unphysical for a < √ 5 and stable fora > √ 5. As a result, BP has three stable fixed points, one for each p...

  75. [76]

    By sandwiching the operator Sx, Sy, or Sz in be- tween the bra and ket and contracting BP messages along the virtual legs, we can obtain local expec- tation values

    At the boundaries between these regimes, two fixed points merge and become marginally stable (|λ0| = 1) which means that the distance from the fixed point decays only algebraically with the number of BP iterations rather than exponentially. By sandwiching the operator Sx, Sy, or Sz in be- tween the bra and ket and contracting BP messages along the virtual...