Pith. sign in

REVIEW 2 major objections 4 minor 60 references

Stable valleys in the glassy landscape of a low-density parity-check (LDPC) code

T0 review · 2 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read The paper argues that in a disorder-free classical LDPC code on an expander graph, low-energy valleys of the energy landscape can be thermally (canonically) unstable while remaining protected by an extensive entropy barrier at fixed energy,

desk verdict A careful, honest case study whose canonical-instability result is solid; the marquee microcanonical-canonical gap is suggestive but rests on a finite-size-drifted lower bound with no extrapolation to back it. read the letter →

arxiv 2607.20421 v1 pith:43EJXHUB submitted 2026-07-22 cond-mat.stat-mech

Grace M. Sommers , David A. Huse This is my paper
classification cond-mat.stat-mech
keywords low-densityparity-checkcodespinglasswithoutdisorderenergylandscapevalleysmicrocanonicalversuscanonicalensembleentropybarrierexpandergraphcavitymethod/treerecursionflat-histogramMonteCarlo
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 studies a classical low-density parity-check code, which is a spin-glass model without quenched disorder or frustration, and maps out the stability of its energy valleys. It claims that valleys with low minimum energy leave the valley at a canonical instability energy that is lower than the energy where the microcanonical entropy barrier vanishes. If true, this is a concrete demonstration of nonequivalence between canonical and microcanonical ensembles in a finite-connectivity glassy system with extensive barriers. The authors support this with tree recursion calculations and flat-histogram Monte Carlo on closed random regular graphs.

What carries the argument

Valleys are parameterized by their minimum energy density ε_min. The canonical instability is detected as an inflection point in the within-valley free energy F_valley(E,T), extracted from an adaptive multicanonical Boltzmann factor x(E) whose local maximum gives ε_canonical. The microcanonical instability is probed through the entropy barrier ΔS(E) = S(E,M*(E)) − S(E,M_barrier(E)), where M(σ) is a single reaction coordinate measuring overlap with the valley bottom; ΔS is a lower bound on the true entropic barrier. Tree-based recursion with (α,γ) boundary conditions and a low-temperature expansion in the defect density supplement the closed-graph simulations.

What would settle it

A calculation or simulation that exhibits a valley and an explicit escape path at an energy density below the predicted ε_micro while keeping the reaction coordinate M above its barrier value, or an escape-time measurement showing τ(E) does not grow exponentially with ΔS(E)/T, would disprove the claim that low-energy valleys are microcanonically stable.

Watch

Extended reading notes

Core claim

The central claim is that valleys of the Tanner-Hamming [7,4,3] cLDPC model on locally tree-like expander graphs exhibit a regime where the valley is canonically unstable yet microcanonically stable: heating the valley makes it escape at an energy density ε_canonical, but a state constrained to a fixed energy density below ε_micro cannot leave because of an extensive entropic barrier. The numerical evidence, stated in the main text, is 'consistent with the hypothesis that as ε_min increases, the entropy barrier at a given ε decreases, but ε_micro and ε_canonical remain separated for low ε_min.'

Load-bearing premise

The conclusion that a valley is microcanonically stable relies on estimating the entropy barrier from a single reaction coordinate M(σ); if the true escape path is hidden by this projection, the reported ε_micro is an artifact.

Editorial extensions

If this is right

  • If the central claim is correct, canonical and microcanonical phase boundaries differ for these valleys, so the equilibrium Gibbs measure at a given temperature would leave the valley while energy-constrained dynamics would remain trapped.
  • The canonical instability energy matches the codeword-polarized (memory) transition temperature computed on trees, suggesting that the passive error-correction threshold is set by the canonical instability, while microcanonical decoding might operate at higher energies.
  • The paper's phase diagram implies that high-energy valleys (large ε_min) lose the canonical instability and instead escape only via a vanishing entropy barrier, so there is a bounded region of 'intermediate' valleys.
  • Fine-tuned boundary conditions on trees produce a family of high-energy valleys with a downward cusp in energy at their transition, bounding the region of thermally stable valleys and confirming a rich landscape beyond codeword valleys.

Reading between the lines

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

  • If this ensemble inequivalence is generic, it should appear in other sparse glassy models such as Gallager codes and diluted p-spin models, which share the same structure of extensive barriers and non-random interactions.
  • The single-reaction-coordinate estimate of the entropy barrier is a lower bound; a stronger test would be to compute escape times and compare their scaling with ΔS, or to search for escape paths orthogonal to M that would invalidate the reported ε_micro.
  • The finite-size drift in the barrier-vanishing energy (seen in the Supplemental Material) leaves open the possibility that the separation between ε_canonical and ε_micro shrinks or vanishes at larger system sizes, which is directly testable.
  • An analogous thermally-unstable-but-microcanonically-stable regime might exist in quantum LDPC codes, which are built from products of classical LDPC codes and exhibit topological quantum spin glass order.
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 / 4 minor

Summary. The paper studies a classical LDPC model—the Tanner-Hamming [7,4,3] code on locally tree-like random regular graphs—as a statistical-mechanics model with glassy behavior but no quenched disorder or frustration. Using cavity-method recursion with a two-parameter (α,γ) family of boundary conditions, the authors derive a low-temperature expansion for the bulk free energy, energy, and configurational entropy, and construct high-energy valleys via fine-tuned node compositions. On closed graphs, adaptive multicanonical Monte Carlo is used to detect (i) the canonical instability (vanishing of the within-valley free-energy minimum) and (ii) a lower bound on the microcanonical entropic barrier using a one-dimensional overlap coordinate M(σ). The central claim is that for valleys with low minimum energy density ε_min, the canonical instability occurs at an energy below the microcanonical barrier-vanishing energy, giving an intermediate regime that is canonically unstable but microcanonically stable—an inequivalence of ensembles. The paper presents the phase diagram as a heuristic and labels the microcanonical boundary a 'rough estimate' and 'conjecture'.

Significance. If the central claim were established in the thermodynamic limit, the model would provide a concrete, non-random, unfrustrated example of microcanonical/canonical ensemble inequivalence in a finite-connectivity system with extensive barriers, complementing the rigorous spin-glass results of Ref. [35] and connecting to quantum LDPC constructions. The paper's strengths include a transparent analytic low-temperature expansion cross-checked against population dynamics and the exact α=γ=2 second moment; a clear and honest acknowledgment that the intermediate phase is a hypothesis; and careful numerical methodology with multiple system sizes (N_v = 218, 1298, 7778) and explicit algorithmic details. The tree-level calculations and the agreement between the codeword-polarized tree prediction for the canonical instability and the closed-graph data at N_v = 7778 are convincing.

major comments (2)
  1. [Main text, 'Microcanonical ensemble'; SM Sec. VI.D, Fig. S7] The existence of the intermediate regime (canonically unstable, microcanonically stable) depends on ε_micro > ε_canonical in the thermodynamic limit. The quantity ε_micro is read off from the vanishing of the lower-bound barrier ΔS(E) in SM Eq. (S64). For ε_min=0, SM Fig. S7a shows a significant downward drift of the barrier-vanishing energy with increasing N_v (218 → 1298 → 7778); for ε_min>0, the data exist only at N_v=1298. No finite-size extrapolation is provided, and the paper itself labels ε_micro a 'rough estimate'. If the downward drift persists, extrapolated ε_micro could fall below ε_canonical, eliminating the proposed intermediate phase. The thermodynamic-limit existence of the ensemble inequivalence is therefore not established.
  2. [Main text, 'Microcanonical ensemble' and Fig. 3; SM Eq. (S64), Fig. S6] Even apart from finite-size effects, ε_micro as defined is the energy where a lower bound on the true entropic barrier vanishes, not necessarily the energy where the true barrier vanishes. The paper's own Fig. S6 shows that a one-dimensional projection can overestimate the entropy at the barrier coordinate, making the bound less tight; the paper notes this in SM Sec. VI.D. Consequently, the reported ε_micro may underestimate the true microcanonical instability. A positive separation at finite N_v therefore does not directly imply a positive separation in the thermodynamic limit; the manuscript should either supply a scaling analysis of ε_micro or clearly restrict the claim to finite systems.
minor comments (4)
  1. [Abstract] The abstract states that the numerical study 'reveals the inequivalence of the microcanonical and canonical ensemble', whereas the main text (Discussion and 'Dynamics on closed graphs') describes 'suggestive evidence' and a 'hypothesis'. Given the finite-size caveats, the abstract overstates the current support; recommend softening to 'provides evidence for' or 'suggests'.
  2. [Fig. 3a] The thick light-blue curve is called a 'rough estimate' of the barrier-vanishing energy. Since this curve is load-bearing for the central claim, the main figure should display the finite-size data points (or at least the range of drift seen in SM Fig. S7) directly on the phase diagram, so the reader can assess the uncertainty.
  3. [SM Sec. VI.D, Fig. S7 caption] The caption states that error bars are smaller than markers for N_v=218 and 1298, but does not comment on N_v=7778. Clarify whether error bars are shown for the largest size and how the drift is quantified.
  4. [Notation] The subscript notation for ε_min is inconsistent between the main text (ε_min) and the SM (εmin, e.g., Sec. VI.A). Unify the notation throughout.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: tree-level predictions are compared with, not fitted to, closed-graph numerics; acknowledged limitations do not close a definitional loop.

full rationale

The paper's derivation chain is self-contained at the level of its claims. The tree recursion (Eqs. 2-5 and SM Secs. I-II) produces bulk free energy, energy, complexity s_conf = s_para − s(β,1,1) (Eq. 6), and transition temperatures from stated (α,γ) boundary-condition ensembles; these are analytic/recursive outputs, not fits. The canonical instability on closed graphs is read off as a local maximum of the adaptively learned x(E) (Eqs. 11-12), and the quoted agreement with the tree value ε(γ=∞, x_mem) is a comparison of independent calculations, not a parameter fitted to the closed-graph data. The microcanonical barrier ΔS = S(E,M*) − S(E,M_barrier) is explicitly described as a lower bound on the true barrier (SM Sec. VI.C), and the paper acknowledges finite-size drift and the single-reaction-coordinate caveat (SM Figs. S6-S7); these are strength-of-evidence limitations, not circular reductions. The main self-citation, Ref. [35], supplies a proof of spin-glass order for a broader class of cLDPC codes and motivates the model, but the paper explicitly states that the Tanner-Hamming code lacks a formal proof of code expansion/redundancy, so the new numerical claims do not reduce to Ref. [35]. No fitted parameter is renamed as a prediction, and no boundary-condition ensemble is defined by the quantity it is claimed to predict. The only editorial defect noticed is an unfilled '(cite source)' marker in the SM discussion of the orthogonal reaction coordinate, which is a completeness issue, not circularity.

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

The ledger has three classes of hand-set parameters: the boundary-condition weights (α,γ), the node-composition matrix N for high-energy valleys, and the numerical cutoffs. No parameters are fit to data in the tree-level theory; the low-T expansion has no fudge constants. The axioms are standard cavity-method assumptions plus two model-specific unproven claims (no global redundancies; tree-to-graph transfer), and the microcanonical barrier projection is the most fragile assumption, acknowledged in the SM.

free parameters (3)
  • (α,γ) boundary-condition weights = scanned over ranges; e.g., γ=α with γ from 0.02 to ∞, α=1 with γ→1+
    These two parameters define the weighted boundary-condition ensemble used in the tree recursion; no fit to data, but they select which valleys are probed and therefore parameterize the phase diagram.
  • Fine-tuned node-composition matrix N (entries N_{ij}) for high-energy valleys = e.g., N_max with N_{00}=2, N_{01}=3, N_{10}=0, N_{12}=6, N_{20}=4; families with N_{01}=1 or 3 and varying N_{12}
    Chosen by hand to construct stable high-energy valleys; constrained by stability conditions, but different choices generate different valley families that are not derived from first principles.
  • Numerical cutoffs (M_min, E_max, learning-interval increments) = tuned per simulation; not quoted
    Flat-histogram sampling of within-valley free energy requires rejecting configurations outside M_min/E_max; the paper notes the choice of M_min requires finesse, so these are simulation hyperparameters that influence measured barriers.
assumptions (5)
  • domain assumption The high-girth random regular graph is locally tree-like, so the infinite rooted-tree recursion captures the thermodynamic-limit behavior on closed graphs.
    Girth is Θ(log N_v), but finite cycles exist; equivalence between tree fixed-point distributions and closed-graph pure states is a standard but unproven step (main text 'Solution via tree recursion'; SM Sec. II).
  • domain assumption For the Tanner-Hamming [7,4,3] code there are no redundancies among checks on different vertices, so the global partition function factorizes and the ground-state degeneracy is 2^{N_v/2}.
    Paper: 'this inequality is empirically saturated: there are no redundancies between the checks on different vertices [35]' — stated as empirical, not proven.
  • standard math The cavity-method distributional recursion (Eq. 4) converges to a fixed point Q*_{β,α,γ}, and α=γ=1 yields a stationary point of the averaged free energy.
    Standard in cavity method; the paper shows the stationary condition (Eq. S20) but assumes existence/uniqueness of the fixed point (SM Sec. II.B).
  • domain assumption The lower bound on the entropy barrier computed from the single-coordinate M(σ) projection is representative of the true high-dimensional barrier.
    If true escape paths are hidden by the projection (SM Fig. S6), the microcanonical stability claim fails; authors acknowledge this.
  • domain assumption The low-temperature expansion remains accurate at the temperatures used to locate transitions (e.g., x_mem=1/25, T_G=0.176).
    The series is truncated at O(x^3, x^{γ+1}, x^{γ/α+1}); at x=0.04 truncation errors are non-negligible, so comparisons at T_G rely on the truncation being adequate.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Stable valleys in the glassy landscape of a low-density parity-check (LDPC) code." pith.science (2026). https://pith.science/paper/43EJXHUB

@misc{pith2026260720421,
  author       = {Pith},
  title        = {Pith review of: Stable valleys in the glassy landscape of a low-density parity-check (LDPC) code},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/43EJXHUB}},
  note         = {Machine review of arXiv:2607.20421}
}
read the original abstract

Classical low-density parity-check (cLDPC) codes defined on expander graphs are a fundamental ingredient in the construction of good quantum LDPC codes, a recent milestone in quantum error correction. They also define interesting statistical mechanics models in their own right, as they include examples of spin glass order without quenched randomness or frustration. We investigate this via a case study of a cLDPC code on a locally tree-like expander graph. Recursive techniques on trees, made possible by the locally tree-like property, probe a menagerie of stable, incongruent valleys induced by imposing different boundary conditions at low temperature. A complementary numerical study of the valleys on closed finite graphs reveals the inequivalence of the microcanonical and canonical ensemble for certain valleys.

Figures

Figures reproduced from arXiv: 2607.20421 by the authors.

Figure 1
Figure 1. FIG. 1. The LDPC code studied in this work is defined on [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Bulk energy density for valleys induced by various [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. (a) Heuristic phases of stability for valleys parameterized by [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

60 extracted references · 1 canonical work pages

  1. [1]

    2 codewords with| P i σi|=n L: the all up codeword and itsZ 2 partner, the all down state

  2. [2]

    7 codewords with P i σi = 1: 1 1 1−1 1−1−1 and cyclic permutations

  3. [3]

    broadcasting

    7 codewords with P i σi =−1, obtained by flipping all spins in the codewords of set (2). C. Ensembles of boundary conditions Recall that we definedQ (r) β,α,γ as the probability distribution of root magnetizations in a depthrrooted tree, where the distribution is taken over an ensemble of boundary conditions weighted byZ σ∂ (γβ/α) α. In the Tanner-Hamming...

  4. [49]

    , b, independently from the weighted populationQ(r) β,α,γ , with probability proportional to the weightw i

    Draw (m i, µi),i= 1, . . . , b, independently from the weighted populationQ(r) β,α,γ , with probability proportional to the weightw i. Letm= (m 1, . . . , mb) andµ= (µ 1, . . . , µb)

  5. [50]

    First, ifγ=α, thenm=µand we only need to store tuples (m, µ)

    Set thejth element ofQ (r+1) β,α,γ equal to (m, µ, w) where m=g(m, β), µ=g(µ, γβ/α), w=z(µ, γβ/α) α bY i=1 wi.(S13) There are two simplifying cases. First, ifγ=α, thenm=µand we only need to store tuples (m, µ). Second, ifα= 1, then we can follow the recursion using twounweightedpopulations. The trick is to consider ˜Qinstead of Q, so that factors ofz(µ, β...

  6. [51]

    , σb with Boltzmann weighte −γβE v(σ0,σ)

    Sample the spin configurationσ=σ 1, . . . , σb with Boltzmann weighte −γβE v(σ0,σ)

  7. [52]

    b, independently samplemi from population ˜Q(r) σi,β,1,γ

    For eachσ i,i= 1, . . . b, independently samplemi from population ˜Q(r) σi,β,1,γ

  8. [53]

    The process is iterated up to a predeterminedr max, chosen large enough so thatQor ˜Qhas reached an approximate steady state, as indicated by a plateau in⟨|m|⟩

    Set thejth element of the population ˜Q(r+1) σ0,β,1,γ equal to (m, µ) wherem=g(m, β), µ=g(µ, γβ). The process is iterated up to a predeterminedr max, chosen large enough so thatQor ˜Qhas reached an approximate steady state, as indicated by a plateau in⟨|m|⟩. 4 FIG. S1. Methods of recursion on trees. (a) Joining togetherbbranches with conditional magnetiza...

Show all 60 references
  1. [54]

    M´ ezard, G

    M. M´ ezard, G. Parisi, and M. A. Virasoro, SK Model: The Replica Solution without Replicas, Europhysics Let- ters (EPL)1, 77 (1986)

    1986

  2. [55]

    M´ ezard, G

    M. M´ ezard, G. Parisi, and M. A. Virasoro,Spin glass the- ory and beyond: An Introduction to the Replica Method and Its Applications, Vol. 9 (World Scientific Publishing Company, 1987)

    1987

  3. [56]

    M´ ezard and G

    M. M´ ezard and G. Parisi, The bethe lattice spin glass revisited, The European Physical Journal B-Condensed Matter and Complex Systems20, 217 (2001)

    2001

  4. [57]

    M. M´ ezard, Cavity method: message-passing from a physics perspective, inStatistical Physics, Optimiza- tion, Inference, and Message-Passing Algorithms: Lec- ture Notes of the Les Houches School of Physics: Special Issue, October 2013, edited by F. Krzakala, F. Ricci- Terseng...

    2013

  5. [58]

    M´ ezard and A

    M. M´ ezard and A. Montanari, Reconstruction on trees and spin glass transition, Journal of statistical physics 124, 1317 (2006)

    2006

  6. [59]

    Krzakala, A

    F. Krzakala, A. Montanari, F. Ricci-Tersenghi, G. Semer- jian, and L. Zdeborov´ a, Gibbs states and the set of so- lutions of random constraint satisfaction problems, Pro- ceedings of the National Academy of Sciences104, 10318 (2007)

    2007

  7. [60]

    Krzakala and L

    F. Krzakala and L. Zdeborov´ a, Following gibbs states adiabatically —the energy landscape of mean-field glassy systems, Europhysics Letters90, 66002 (2010)

    2010

  8. [61]

    Zdeborov´ a and F

    L. Zdeborov´ a and F. Krzakala, Generalization of the cav- ity method for adiabatic evolution of gibbs states, Phys. 8 Rev. B81, 224205 (2010)

    2010

  9. [62]

    [60, 61] as a way to adiabatically evolve within a valley sampled from the global Gibbs measure, where it was applied to dilutedp-spin and antiferromagnetic Potts models

    The subfamily withα= 1 appeared in Refs. [60, 61] as a way to adiabatically evolve within a valley sampled from the global Gibbs measure, where it was applied to dilutedp-spin and antiferromagnetic Potts models. To our knowledge, the two-parameter family (α, γ) has not been an...

  10. [63]

    When the superscript is omitted,Q β,α,γ should be inter- preted as the fixed point distribution

  11. [64]

    The SM also con- tains additional numerics and a review of prior work

    See Supplemental Material for details on recursion rela- tions for the Tanner-Hamming code, calculation of bulk observables, the low-temperature expansion, high-energy valleys,and flat histogram techniques. The SM also con- tains additional numerics and a review of prior work

  12. [65]

    interval- ley

    This decomposition of the total entropy into an “interval- ley” and “intravalley” contribution is reminiscent of the discussion of the clustered solution space of XORSAT problems in Ref. [25], where the total ground state en- tropy splits into “core” (configurational) and “non...

  13. [66]

    [23, 29, 60, 61] for discussions of these two transitions in dilutedp-spin models

    Those steeped in the replica symmetry breaking liter- ature might prefer to callT c(1,1) the dynamical glass (1RSB) transition andT c(∞) the spinodal temperature; see, for example, Refs. [23, 29, 60, 61] for discussions of these two transitions in dilutedp-spin models. We emph...

  14. [67]

    B. A. Berg and T. Neuhaus, Multicanonical algorithms for first order phase transitions, Physics Letters B267, 249 (1991)

    1991

  15. [68]

    B. A. Berg and T. Neuhaus, Multicanonical ensemble: A new approach to simulate first-order phase transitions, Phys. Rev. Lett.68, 9 (1992)

    1992

  16. [69]

    Janke, Multicanonical Monte Carlo simulations, Physica A: Statistical Mechanics and its Applications 254, 164 (1998)

    W. Janke, Multicanonical Monte Carlo simulations, Physica A: Statistical Mechanics and its Applications 254, 164 (1998)

    1998

  17. [70]

    Mau and D

    S.-C. Mau and D. A. Huse, Stacking entropy of hard- sphere crystals, Phys. Rev. E59, 4396 (1999)

    1999

  18. [71]

    Nishino and K

    T. Nishino and K. Okunishi, Corner transfer matrix renormalization group method, Journal of the Physical Society of Japan65, 891 (1996)

    1996

  19. [72]

    X. Wang, Z. Nussinov, and G. Ortiz, Emergence of a Boundary-Sensitive Phase in Hyperbolic Ising Models (2025), arXiv:2507.21044

    2025

  20. [73]

    Tillich and G

    J.-P. Tillich and G. Zemor, Quantum ldpc codes with positive rate and minimum distance proportional to n½, in2009 IEEE International Symposium on Information Theory(2009) pp. 799–803

    2009

  21. [74]

    Leverrier, J.-P

    A. Leverrier, J.-P. Tillich, and G. Z´ emor, Quantum ex- pander codes, in2015 IEEE 56th Annual Symposium on Foundations of Computer Science(2015) pp. 810–824

    2015

  22. [75]

    N. P. Breuckmann and J. N. Eberhardt, Quantum low- density parity-check codes, PRX Quantum2, 040101 (2021)

    2021

  23. [76]

    Rakovszky and V

    T. Rakovszky and V. Khemani, The Physics of (good) LDPC Codes II. Product constructions (2024), arXiv:2402.16831

    2024 arXiv

  24. [77]

    Placke, T

    B. Placke, T. Rakovszky, N. P. Breuckmann, and V. Khe- mani, Topological quantum spin glass order and its real- ization in qldpc codes (2024), arXiv:2412.13248 [quant- ph]

    2024 arXiv

  25. [78]

    Linial and M

    N. Linial and M. Simkin, A randomized construction of high girth regular graphs (2019), arXiv:1911.09640

    2019 arXiv

  26. [79]

    T. P. Eggarter, Cayley trees, the Ising problem, and the thermodynamic limit, Physical Review B9, 2989 (1974)

    1974

  27. [80]

    M¨ uller-Hartmann and J

    E. M¨ uller-Hartmann and J. Zittartz, New type of phase transition, Physical Review Letters33, 893 (1974)

    1974

  28. [81]

    Matsuda, Infinite susceptibility without spontaneous magnetization: Exact properties of the ising model on the cayley tree, Progress of Theoretical Physics51, 1053 (1974)

    H. Matsuda, Infinite susceptibility without spontaneous magnetization: Exact properties of the ising model on the cayley tree, Progress of Theoretical Physics51, 1053 (1974)

    1974

  29. [82]

    J. T. Chayes, L. Chayes, J. P. Sethna, and D. J. Thou- less, A mean field spin glass with short-range interac- tions, Communications in Mathematical Physics106, 41 (1986)

    1986

  30. [83]

    P. M. Bleher, J. Ruiz, and V. A. Zagrebnov, On the pu- rity of the limiting gibbs state for the Ising model on the Bethe lattice, Journal of Statistical Physics79, 473 (1995)

    1995

  31. [84]

    Ioffe, On the extremality of the disordered state for the Ising model on the Bethe lattice, Letters in Mathematical Physics37, 137 (1996)

    D. Ioffe, On the extremality of the disordered state for the Ising model on the Bethe lattice, Letters in Mathematical Physics37, 137 (1996)

    1996

  32. [85]

    J. M. Magan and A. Sharma, Using memory to identify phase transitions on a cayley tree (2013), arXiv:1304.7008

    2013 arXiv

  33. [86]

    Gerschenfeld and A

    A. Gerschenfeld and A. Montanari, Reconstruction for models on random graphs, in48th Annual IEEE Sym- posium on Foundations of Computer Science (FOCS’07) (2007) pp. 194–204

    2007

  34. [87]

    D. A. Huse and D. S. Fisher, Pure states in spin glasses, Journal of Physics A: Mathematical and General20, L997 (1987)

    1987

  35. [88]

    C. M. Newman and D. L. Stein, Free energy difference fluctuations in short-range spin glasses, Journal of Sta- tistical Physics191, 10.1007/s10955-024-03334-4 (2024)

    2024 doi

  36. [89]

    Abou-Chacra, D

    R. Abou-Chacra, D. J. Thouless, and P. W. Anderson, A selfconsistent theory of localization, Journal of Physics C: Solid State Physics6, 1734 (1973)

    1973

  37. [90]

    M´ ezard and G

    M. M´ ezard and G. Parisi, The Cavity Method at Zero Temperature, Journal of Statistical Physics111, 1 (2003)

    2003

  38. [91]

    Yedidia, W

    J. Yedidia, W. Freeman, and Y. Weiss, Construct- ing Free-Energy Approximations and Generalized Belief Propagation Algorithms, IEEE Transactions on Informa- tion Theory51, 2282 (2005)

    2005

  39. [92]

    Parisi, Order parameter for spin-glasses, Phys

    G. Parisi, Order parameter for spin-glasses, Phys. Rev. Lett.50, 1946 (1983)

    1946

  40. [93]

    Barrat and R

    A. Barrat and R. Zecchina, Time scale separation and heterogeneous off-equilibrium dynamics in spin models over random graphs, Phys. Rev. E59, R1299(R) (1999)

    1999

  41. [94]

    Franz, M

    S. Franz, M. Leone, F. Ricci-Tersenghi, and R. Zecchina, Exact solutions for diluted spin glasses and optimization problems, Phys. Rev. Lett.87, 127209 (2001)

    2001

  42. [95]

    Ricci-Tersenghi, Being glassy without being hard to solve, Science330, 1639 (2010)

    F. Ricci-Tersenghi, Being glassy without being hard to solve, Science330, 1639 (2010)

    2010

  43. [96]

    Richardson and R

    T. Richardson and R. Urbanke,Modern coding theory (Cambridge university press, 2008)

    2008

  44. [97]

    Gopalan, C

    P. Gopalan, C. Huang, H. Simitci, and S. Yekhanin, On the locality of codeword symbols, IEEE Transactions on Information Theory58, 6925 (2012)

    2012

  45. [98]

    Stable valleys in the glassy landscape of a low-density parity-check (LDPC) code

    V. Guruswami, C. Xing, and C. Yuan, How long can optimal locally repairable codes be?, IEEE Transactions on Information Theory65, 3662 (2019). Supplemental Information to: “Stable valleys in the glassy landscape of a low-density parity-check (LDPC) code” The Supplemental Mater...

    2019

  46. [99]

    Pairing trees with magnetizationsm 0, m1 at the roots incurs a shiftY delta(m0, m1) in the quantity of interest (Fig

    Pair up the trees and join them at the root spins, makingb+ 1 Bethe lattices withb+ 1 fewer spins than we started with. Pairing trees with magnetizationsm 0, m1 at the roots incurs a shiftY delta(m0, m1) in the quantity of interest (Fig. S1b)

  47. [100]

    messages

    Make two groups ofb+ 1 branches and join the branches in each group at a single node, making 2 Bethe lattices with 2 more nodes than we started with. Joining branches with magnetizations (m 0,m) at their roots incurs a shift ofY int(m0,m, β) (Fig. S1c). Therefore, Ybulk =Y int...

  48. [101]

    microcanonical

    Multicanonical method When trying to quantify the height of the barrier and the temperature at which it vanishes, standard single- temperature Metropolis encounters a problem: the states at the top of the free energy barrier are exponentially rare in ∆F. The multicanonical met...

  49. [102]

    escape energy

    Space of allowed configurations Let us elaborate on the space of allowed configurations Σ, which is an input parameter to Alg. 1 and its subroutine, MCSweep()(line 21). To properly sample only the within-valley free energy and entropy, we need a way to prevent the state from e...

  50. [103]

    magnetization

    Generalizations Note that while the update criterion Eq. (S61) aims for a flat histogram on the interval [E 1, E2], we can modify it to target any functional form (i.e. a target histogramn target(E)) by taking: xnew(E) = n(x(E), E) n(x(E), E+ 1) ntarget(E+ 1) ntarget(E) x(E).(...

  51. [104]

    [78] (see App

    Simulation details To perform simulations on closed graphs, we first use a version of the method described in Ref. [78] (see App. C of Ref. [35] for details) to generate a high-girth random regular graph (HGRRG). The greedy algorithm succeeds with high probability if, for a ta...

  52. [105]

    Entropy barrier numerics Fig. 3c of the main text shows the entropy landscapeS(m)−S(m ∗) around valleys containing ground states (Emin = 0), on closed graphs withN v = 1298 and a range of microcanonical energy windows above the canonical instability energy. To substantiate our...

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.