Pith. sign in

REVIEW 3 major objections 5 minor 51 references

Stochastic binary networks with asymmetric and time-delayed interactions

T0 review · 3 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read With sufficiently long signal delays, the steady-state joint distribution of a stochastically updated Ising-type network becomes uniform — every spin configuration equally likely — no matter the coupling strength, while temporal correlation

desk verdict Solid two-spin delay theory; the N-spin uniformization proof is a self-consistency check, not a uniqueness proof. read the letter →

arxiv 2607.15215 v1 pith:Q5YYUF7M submitted 2026-07-16 physics.app-ph cond-mat.mes-hallcond-mat.stat-mechcs.ETphysics.data-an

Hantao Zhang , Sidra Gibeault , Matthew W. Daniels , Philippe Talatchian , Ursula Ebels , Advait Madhavan , Mark D. Stiles This is my paper · ORCID
classification physics.app-phcond-mat.mes-hallcond-mat.stat-mechcs.ETphysics.data-an
keywords stochasticbinarynetworkstime-delayedinteractionsasymmetriccouplingIsingmodeluniformsteadystatetemporalcorrelationssuperparamagnetictunneljunctionsdoublyMarkovchain
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 studies networks of binary spins whose interactions are both asymmetric (the influence of spin j on i need not equal the influence of i on j) and delayed by finite signal-propagation times. It claims that when delays are sufficiently long relative to the spins' intrinsic fluctuation time, the steady-state probability distribution over spin configurations becomes uniform: every configuration is equally likely, no matter how strong the couplings. This looks like the spins have decoupled, but temporal auto- and cross-correlations remain strongly oscillatory, so the uniform distribution is an out-of-equilibrium feature, not high-temperature randomness. The claim is proved for two spins and argued generally: any stochastic network whose transition rates respect a compact global symmetry, and whose delayed signals are statistically decorrelated from the present, becomes an irreducible memoryless Markov chain with a doubly stochastic transition matrix and hence a uniform steady state. The uniformity is destroyed by bias fields and persists in simulations of five coupled spins.

What carries the argument

The argument rests on the delayed transition rate λ_i(t) = λ0 exp[-(Σ_j J_{i←j} S_j(t - t_{i←j}) + h_i) S_i(t)/kT], where spin i updates using only past states of its neighbors. The proof's engine is the decorrelation step: for min delay ≫ τ_cross, Pr[S_j(t - t_delay) | S_i(t)] ≈ Pr[S_j(t - t_delay)], which converts the delayed update into an effective transition matrix whose entries sum over neighbor states. When transition rates respect a global symmetry, such as spin inversion for Ising spins, the effective matrix becomes doubly stochastic, and the uniqueness theorem for irreducible, memoryless Markov chains forces the uniform steady state. Correlation peaks are explained by round-trip ec

What would settle it

Simulate two coupled spins with J1←2 = J2←1 = kT and vary the common delay from a fraction of the intrinsic spin time τ to many τ; measure the steady-state probabilities P(↑↑) and P(↓↑). If P(↑↑) differs from 1/4 within statistical error at any finite delay, or if the lagged conditional probability Pr[S_2(t - t_{2←1}) | S_1(t)] differs from the marginal Pr[S_2] when the distribution is already uniform, then the decorrelation step in Sec. V B is not what produces uniformity and the central claim, as proved, fails.

Watch

Extended reading notes

Core claim

The central discovery is that adding a time delay to couplings in a stochastic binary network changes the long-time probability landscape in a way that neither asymmetry nor delay alone would predict. For anti-symmetric two-spin couplings, the paper finds — matching measurements on coupled superparamagnetic tunnel junctions — that delay enhances damped oscillations in correlation functions, with autocorrelation peaks appearing at even multiples of the delay and cross-correlation peaks at odd multiples. For any coupling sign, once the smallest delay exceeds the cross-correlation time of the spins, the steady state becomes the uniform distribution, all 2^N configurations equally occupied, rega

Load-bearing premise

The proof rests on the assertion that a spin at time t is almost independent of neighbor states one delay earlier — Pr[S_j(t - t_{i←j}) | S_i(t)] ≈ Pr[S_j(t - t_{i←j})] — asserted for min delay ≫ τ_cross; if delayed signals remain statistically correlated with the present (or, for N spins, jointly independent), the effective transition matrix is not doubly stochastic and uniformity need not hold.

Editorial extensions

If this is right

  • Delay time becomes a usable control parameter: uniform, long delays make steady-state distributions insensitive to couplings, while non-uniform delays restore coupling-dependent distributions.
  • In stochastic or neuromorphic hardware, communication delays are not necessarily errors: the model reproduces measured correlation oscillations in coupled magnetic junctions, and delay can be tuned to shape correlations.
  • Bias fields break the Z2 symmetry that produces uniformity; the resulting delayed steady states depend on couplings in a way qualitatively different from instantaneous Ising equilibrium, giving concrete predictions for biased networks.
  • The uniformization theorem extends beyond Ising spins to any compact-symmetry stochastic network, including Potts, Kuramoto, Heisenberg, and nonlinear-sigma-model-type systems; the paper notes this explains the vanishing collective frequency in delayed Kuramoto oscillators.
  • Uniform occupation alone cannot distinguish the delay-driven steady state from infinite temperature; oscillatory correlation functions are the diagnostic that separates them.

Reading between the lines

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

  • A direct test of the decorrelation condition — computing the mutual information between S_i(t) and S_j(t - t_delay) in simulation — should show uniformization switching on exactly as that mutual information vanishes; the paper does not report this quantity.
  • For finite networks, irreducibility of the effective chain is assumed; if topology, asymmetric delays, or strong coupling create nearly decoupled components, uniformization could still hold asymptotically but over timescales far beyond the single-spin dwell time — a regime not explored here.
  • The peak-location rule, autocorrelation peaks at even multiples of the delay and cross-correlation peaks at odd multiples, turns delay lines into readable memory in correlation space; engineering non-uniform delays could synthesize desired correlation spectra in oscillator-based or p-bit hardware.
  • For biological neural networks with directional delays, the result warns that flat steady-state firing probabilities do not imply uncoupled dynamics; oscillatory correlations can carry the coupling information.
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

3 major / 5 minor

Summary. The paper studies stochastic binary (Ising) networks with asymmetric and time-delayed interactions, motivated by coupled superparamagnetic tunnel junctions. It first analyzes the no-delay two-spin case, giving the steady-state distribution (Eq. 4) and an eigenvalue criterion (Eq. 6c) for oscillatory versus monotonic correlation functions. It then adds finite delays and argues, both numerically and analytically, that sufficiently long delays make the steady-state joint distribution uniform regardless of coupling strength, while temporal correlations remain strongly oscillatory. A five-spin simulation section and a general symmetry-based appendix (Appendix C) are used to claim the phenomenon extends to arbitrary Z2-, Zn-, O(n)-, and U(n)-symmetric networks. The paper also compares the delayed model with unpublished experimental correlation data and discusses implications for neuromorphic hardware.

Significance. If the central claim is correct, the paper identifies a genuinely novel nonequilibrium regime: interaction-induced temporal structure coexisting with a featureless uniform steady state, with implications for neuromorphic computing and for the modeling of asymmetric delayed networks. The paper has real strengths: the two-spin no-delay analysis is exact and the eigenvalue criterion is clearly correct; the experiment-theory comparison is concrete; and the numerical parameters are stated with enough detail to reproduce the simulations. However, the advertised generality and the strong-coupling claim rest on a proof whose load-bearing step is invalid. The 'uniform distribution' is shown to be a self-consistent fixed point, but uniqueness is not established; under the paper's own decorrelation approximation a polarized fixed point exists for a fully connected ferromagnetic network with N≥3 and strong coupling. Thus the broad claim is not merely unproved but appears false as stated.

major comments (3)
  1. [Sec. V B, Eqs. (11)-(12)] The N-spin uniqueness argument is a self-consistency check, not a proof. Equation (11) defines effective transition probabilities by marginalizing over the unknown stationary joint distribution Pr(s_1,...,s_N). Substituting the uniform ansatz to obtain Eq. (12) and verifying detailed balance only shows that uniform is a fixed point of a nonlinear self-consistency map. The cited theorem [48] applies to a fixed, memoryless, irreducible Markov chain; here the transition matrix is determined by the very distribution that is to be found, so the standard linear uniqueness theorem does not apply. The statement 'we have completed our proof these distributions become uniform' therefore overreaches. The two-spin case is a legitimate fixed-point calculation, but the N-spin generalization is not proved.
  2. [Appendix C, Eq. (C3)] The derivation of double stochasticity assumes that the marginalized delayed-neighbor distribution is G-invariant ('this incoming noise bath is perfectly symmetric'). The long-delay condition stated between Eqs. (8) and (9) only gives Pr[S_j(t-t_{i←j}) | S_i(t)] ≈ Pr[S_j(t-t_{i←j})]; it does not make that marginal symmetric under G. In a symmetry-broken steady state, such as a ferromagnetic phase, the delayed-neighbor distribution is polarized, and the index-shift in Eq. (C7) is not legitimate. The proof is circular: the G-invariance of the stationary marginal is exactly what must be shown, not a consequence of the delay being long.
  3. [Abstract and Sec. V B (strong-coupling claim)] The advertised result that uniformization holds 'even in strongly coupled systems' is contradicted by a simple fixed point under the paper's own decorrelation approximation. For a fully connected ferromagnetic N-spin network with N≥3 and βJ=5, the self-consistency equation m = (1 - r^{N-1})/(1 + r^{N-1}) with r = [p e^{-βJ} + (1-p) e^{βJ}]/[p e^{βJ} + (1-p) e^{-βJ}], p=(1+m)/2, admits a polarized solution m≈0.98. Thus uniform is not the unique fixed point. In an ordered phase the correlation time is long, so the premise min(t_{i←j}) ≫ τ_cross is itself problematic; either way, the strong-coupling statement in the abstract is not supported. The five-spin simulations in Fig. 6 use J=0.2kT, which is the weak-coupling regime, and do not test the strong-coupling assertion.
minor comments (5)
  1. [Sec. IV A] Typo: 'prinicpal' should be 'principal'.
  2. [Sec. V C] Duplicate wording: 'the the peaks' should be 'the peaks'.
  3. [Eq. (11)] Notation inconsistency: Pr(s_1,· · · ,S_N) mixes lower-case and upper-case variables; use Pr(s_1,...,s_N).
  4. [Appendix C] Typo: 'or a spherical manifold' should likely read 'on a spherical manifold'.
  5. [Sec. V B] The analogy to NP verifiability is misleading: verifying the uniform ansatz is not an instance of verifying a solution to a hard problem in the NP sense, and it should be removed or rephrased.

Circularity Check

2 steps flagged · score 6.0 of 10

The N-spin and Appendix C uniformization proofs insert the uniform/G-invariant stationary distribution as an ansatz, verify it as a fixed point, and then invoke a uniqueness theorem that does not apply to the resulting self-consistent transition matrix.

  1. self definitional [Section V B, N-spin generalization, via Eqs. (11) and (12) and following paragraph]
    "Suppose the joint states are uniformly distributed; then Eq. (11) becomes ... An important property of Eq. (12) is that the equation is invariant under flipping s_i to −s_i ... Consequently, Pr[S_i(t+ 1) =−s_i|S_i(t) =s_i] = Pr[S_i(t+ 1) =s_i|S_i(t) =−s_i]. This fact immediately tells us Pr(s_i) = 1−Pr(s_i) = 1/2, thus the distribution is uniform. Having verified the uniform distribution satisfies the detailed balance equations, we proceed to prove that this uniform distribution is the unique solution."

    Equation (11) defines the effective single-spin transition probabilities as sums over the unknown stationary joint distribution Pr(s_1,...,s_N). The proof substitutes the uniform distribution into Eq. (11) to form Eq. (12), checks detailed balance for that substituted chain, and then invokes the uniqueness theorem for irreducible memoryless Markov chains [48]. That theorem applies only to a fixed, linear transition matrix; here the matrix is defined in terms of the very stationary distribution being sought, making the fixed-point problem nonlinear. Uniformity is shown to be one self-consistent fixed point, not a derived unique solution; the argument does not exclude other fixed points. Under the same decorrelation assumption, a fully connected ferromagnetic N-spin network with N≥3 admits a

  2. self definitional [Appendix C, paragraph beginning 'To understand why time delay induces uniformity', through the row-sum proof]
    "Because the delay is significantly longer than the system’s correlation time, the signals arriving from neighbors are statistically independent of the receiving spin’s current state. If the interactions between spins respect a global symmetry, this incoming noise bath is perfectly symmetric. Consequently, the effective transition probability of the receiving spin depends only on the relative difference between its initial and final states, rather than its absolute state."

    The proof of double stochasticity requires the shift-invariance T(gσ'|gσ)=T(σ'|σ), which is obtained by marginalizing over delayed neighbors using a 'perfectly symmetric' noise bath. The symmetry of the noise bath is exactly the G-invariance (and hence, for Z2, O(n), or U(n), the uniformity) of the stationary distribution that the appendix aims to prove. The G-invariance of the interaction functions alone does not force the state distribution to be G-invariant; a symmetry-broken stationary distribution would produce a polarized bath and would not yield the constant-row-sum argument. Thus the Appendix C derivation assumes the G-invariant steady state as a premise.

full rationale

The experimental correlation analysis and the two-spin fixed-point calculation are not circular: the two-spin equations solve a genuine self-consistency problem for the one-spin marginals under the stated decorrelation assumption, and the simulations and correlation calculations use the original delayed dynamics. No load-bearing self-citation was found; the uniqueness reference [48] is an external textbook, not a self-citation. However, the advertised analytic extension to N spins and to general symmetric networks is circular at the load-bearing step. Equation (11) builds the effective transition probabilities from the unknown stationary joint distribution; the proof then inserts the uniform distribution (Eq. (12)), verifies detailed balance for that substituted chain, and invokes a linear-Markov uniqueness theorem that is inapplicable to a transition matrix defined self-consistently in terms of the target stationary distribution. Appendix C repeats the same structure by assuming a symmetric 'noise bath,' which is equivalent to assuming the G-invariant/uniform steady state. Because these proofs underlie the abstract's claims about strongly coupled systems and the broad class of Potts, Kuramoto, Heisenberg, and related models, the central uniformization derivation is substantially circular, and the score is 6.0.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The paper introduces no parameters fitted to data; all simulation values are hand-chosen representative values, and the experiment-model comparison in Fig. 1 is qualitative rather than a fit. The central delay-uniformity claim rests on postulates that are the paper's own: the decorrelation of delayed signals from the present state (Sec. V B), and, for the general version, G-invariance of the neighbor marginals used in the marginalization (Appendix C) — the latter is a circular postulate for Z2. The uniqueness conclusion imports the standard memoryless-Markov-chain theorem [48] into a system that is only approximately memoryless. No new physical entities (particles, fields, conserved quantities) are invented.

free parameters (6)
  • discretization ratio τ/Δt = 4.98 (λ0Δt ≈ 0.2) = 4.98
    Hand-chosen in all Monte Carlo runs (Figs. 2-7); sets the flip-attempt probability per step. At J=kT it produces single-step flip probabilities up to ≈0.55, violating the stated λΔt ≪ 1 constraint (Sec. III).
  • energy barrier ΔE_i = 4kT = 4kT
    Hand-chosen for all simulations to represent the SMTJ operating regime; sets the base flip rate scale λ0.
  • two-spin coupling J1←2 = −J2←1 = kT (and J = ±kT symmetric cases) = kT
    Hand-chosen for the two-spin distribution/correlation calculations (Figs. 2-5) and the experiment comparison (Fig. 1f,g); not fitted to the data.
  • five-spin coupling J = 0.2kT = 0.2kT
    Hand-chosen for the N=5 calculations (Figs. 6-7); a weak coupling that keeps the no-delay entropy high and makes uniformization visible.
  • delay times t_i←j ∈ {100, 200, 500, 5000} Δt = 100-5000 Δt (≈20 to 1000 τ)
    Hand-chosen to explore short- and long-delay regimes. The experimental delay (microseconds) is never measured or fitted; the Fig. 1 comparison uses 100 Δt ≈ 20τ.
  • bias fields h_i = kT (two-spin), 0.2kT (five-spin) = kT / 0.2kT
    Hand-chosen to break Z2 symmetry in the bias studies (Fig. 4, Figs. 6c,d, Appendix A).
assumptions (5)
  • domain assumption The transition rate has the exponential (Néel-Brown) form of Eq. (3), valid when the effective field changes slowly relative to 1/λ0_i(T).
    Modeling basis for everything; the validity condition is stated in Sec. III and assumed for the theory.
  • ad hoc to paper Decoupling: for min(t_i←j) ≫ τ_cross, Pr[S_j(t−t_i←j) | S_i(t)] ≈ Pr[S_j(t−t_i←j)].
    The load-bearing premise of the uniformization proof (Sec. V B, between Eqs. 8 and 9); asserted, not derived, and τ_cross is coupling-dependent.
  • ad hoc to paper G-invariance of the delayed-neighbor marginal used in the Appendix C marginalization.
    Double-stochasticity of T requires a G-invariant measure in the marginalization; for Z2 this equals the uniformity conclusion, so the general proof is circular.
  • standard math The delayed system can be treated as a memoryless irreducible Markov chain with a unique steady state (Ref. [48]).
    Used to conclude uniqueness of the uniform distribution; the true delayed dynamics are not memoryless, and the self-consistent fixed-point equations can have additional polarized solutions for strong coupling.
  • domain assumption λ_i Δt ≪ 1 (at most one flip per time step).
    Stated in Sec. III but violated in the J=kT simulations (flip probability per step up to λ0Δt·e ≈ 0.55).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Stochastic binary networks with asymmetric and time-delayed interactions." pith.science (2026). https://pith.science/paper/Q5YYUF7M

@misc{pith2026260715215,
  author       = {Pith},
  title        = {Pith review of: Stochastic binary networks with asymmetric and time-delayed interactions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Q5YYUF7M}},
  note         = {Machine review of arXiv:2607.15215}
}
read the original abstract

Stochastic binary networks are widely used to describe collective dynamics in complex systems and to perform neuromorphic computation, yet realistic networks often contain both asymmetric interactions and finite signal propagation times that fall outside conventional theories. Here we study stochastic binary networks with asymmetric and time-delayed interactions motivated by experimental observations in coupled superparamagnetic tunnel junctions. We find that time delay fundamentally reshapes the dynamics induced by anti-symmetric couplings, producing strong oscillatory temporal correlations consistent with experiment. At the same time, sufficiently long delays drive the steady-state probabilities toward equal state occupations even in strongly coupled systems. These apparently featureless probability distributions coexist with pronounced temporal correlations, distinguishing them from equilibrium high-temperature behavior. We further show analytically that delay-induced uniform distributions emerge in a broad class of stochastic networks, while symmetry-breaking bias fields restore interaction-dependent steady states with qualitatively modified behavior. Simulations of networks with five coupled spins demonstrate that these effects persist beyond minimal systems with only two spins. Our results establish a unified framework for stochastic binary networks in the intermediate regime between symmetric instantaneous interactions and asymmetric or time-delayed interactions, and suggest that asymmetry and delay can be exploited as functional resources in neuromorphic hardware and complex network dynamics.

Figures

Figures reproduced from arXiv: 2607.15215 by the authors.

Figure 1
Figure 1. FIG. 1. (a) Two coupled SMTJs, modeled as two Ising [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. (a) Steady-state probability distribution as char [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. FIG. 4. Probability distributions in the presence of bias and [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figures from the paper (5 more)
Figure 5
Figure 5. Figure 5: FIG. 5. Correlation functions with short time delay, long [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6. Entropy of five coupled spins. (a) evolution of entropy [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 10
Figure 10. Figure 10: FIG. 10. Steady-state probability distribution dependence on [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]
Figure 9
Figure 9. Figure 9: FIG. 9. Steady-state probability distribution dependence on [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 11
Figure 11. Figure 11: FIG. 11. Experimental auto- ( [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 4 canonical work pages

  1. [48]

    Ching and M

    W.-K. Ching and M. K. Ng,Markov chains: models, al- gorithms and applications(Springer, 2006)

    2006

  2. [1]

    J. J. Hopfield, Neural networks and physical systems with emergent collective computational abilities., Proceedings of the National Academy of Sciences79, 2554 (1982), https://www.pnas.org/doi/pdf/10.1073/pnas.79.8.2554

    doi:10.1073/pnas.79.8.2554 1982

  3. [2]

    D. H. Ackley, G. E. Hinton, and T. J. Sejnowski, A learn- ing algorithm for boltzmann machines, Cognitive Science 9, 147 (1985)

    1985

  4. [3]

    Kirkpatrick, C

    S. Kirkpatrick, C. D. Gelatt, and M. P. Vecchi, Optimiza- tion by simulated annealing, Science220, 671 (1983), https://www.science.org/doi/pdf/10.1126/science.220.4598.671

    doi:10.1126/science.220.4598.671 1983

  5. [4]

    Lucas, Ising formulations of many np problems, Fron- tiers in Physics2, 74887 (2014)

    A. Lucas, Ising formulations of many np problems, Fron- tiers in Physics2, 74887 (2014)

    2014

  6. [5]

    J. A. Hertz,Introduction to the theory of neural compu- tation(Crc Press, 2018)

    2018

  7. [6]

    Ising, Beitrag zur theorie des ferromagnetismus, Zeitschrift f¨ ur Physik31, 253 (1925)

    E. Ising, Beitrag zur theorie des ferromagnetismus, Zeitschrift f¨ ur Physik31, 253 (1925)

    1925

  8. [7]

    S. F. Edwards and P. W. Anderson, Theory of spin glasses, Journal of Physics F: Metal Physics5, 965 (1975)

    1975

Show all 51 references
  1. [8]

    Sherrington and S

    D. Sherrington and S. Kirkpatrick, Solvable model of a spin-glass, Phys. Rev. Lett.35, 1792 (1975)

    1975

  2. [9]

    Dayan and L

    P. Dayan and L. F. Abbott,Theoretical neuroscience: computational and mathematical modeling of neural sys- tems(MIT press, 2005)

    2005

  3. [10]

    Schneidman, M

    E. Schneidman, M. J. Berry, R. Segev, and W. Bialek, Weak pairwise correlations imply strongly correlated net- work states in a neural population, Nature440, 1007 (2006)

    2006

  4. [11]

    Meshulam and W

    L. Meshulam and W. Bialek, Statistical mechanics for networks of real neurons, Rev. Mod. Phys.97, 045002 (2025)

    2025

  5. [12]

    GALAM, Sociophysics: A review of galam models, In- ternational Journal of Modern Physics C19, 409 (2008), https://doi.org/10.1142/S0129183108012297

    S. GALAM, Sociophysics: A review of galam models, In- ternational Journal of Modern Physics C19, 409 (2008), https://doi.org/10.1142/S0129183108012297

    2008 doi

  6. [13]

    Castellano, S

    C. Castellano, S. Fortunato, and V. Loreto, Statistical physics of social dynamics, Rev. Mod. Phys.81, 591 (2009)

    2009

  7. [14]

    Grollier, D

    J. Grollier, D. Querlioz, and M. D. Stiles, Spintronic nan- odevices for bioinspired computing, Proceedings of the IEEE104, 2024 (2016)

    2024

  8. [15]

    Grollier, D

    J. Grollier, D. Querlioz, K. Y. Camsari, K. Everschor- Sitte, S. Fukami, and M. D. Stiles, Neuromorphic spin- tronics, Nature electronics3, 360 (2020)

    2020

  9. [16]

    Mohseni, P

    N. Mohseni, P. L. McMahon, and T. Byrnes, Ising ma- chines as hardware solvers of combinatorial optimization problems, Nature Reviews Physics4, 363 (2022)

    2022

  10. [17]

    Zhang, Q

    T. Zhang, Q. Tao, B. Liu, A. Grimaldi, E. Raimondo, M. Jim´ enez, M. J. Avedillo, J. Nunez, B. Linares- Barranco, T. Serrano-Gotarredona,et al., A review of ising machines implemented in conventional and emerg- 15 ing technologies, IEEE Transactions on Nanotechnology 23, 704 (2024)

    2024

  11. [18]

    H. Lee, J. P. Kim, and S. Kim, Fundamental challenges, physical implementations, and integration strategies for ising machines in large-scale optimization tasks, Ad- vanced Electronic Materialsn/a, e00682 (2026)

    2026

  12. [19]

    Carleo, I

    G. Carleo, I. Cirac, K. Cranmer, L. Daudet, M. Schuld, N. Tishby, L. Vogt-Maranto, and L. Zdeborov´ a, Machine learning and the physical sciences, Rev. Mod. Phys.91, 045002 (2019)

    2019

  13. [20]

    Y. Avni, M. Fruchart, D. Martin, D. Seara, and V. Vitelli, Nonreciprocal ising model, Phys. Rev. Lett. 134, 117103 (2025)

    2025

  14. [21]

    Y. Avni, M. Fruchart, D. Martin, D. Seara, and V. Vitelli, Dynamical phase transitions in the nonrecip- rocal ising model, Phys. Rev. E111, 034124 (2025)

    2025

  15. [22]

    K. Blom, U. Thiele, and A. c. v. Godec, Local order con- trols the onset of oscillations in the nonreciprocal ising model, Phys. Rev. E111, 024207 (2025)

    2025

  16. [23]

    G. A. Weiderpass, M. Sharma, and S. Sethi, Solving the kinetic ising model with nonreciprocity, Phys. Rev. E 111, 024107 (2025)

    2025

  17. [24]

    Di Carlo, Off-equilibrium kinetic ising model: The metric case, Phys

    L. Di Carlo, Off-equilibrium kinetic ising model: The metric case, Phys. Rev. Res.7, 013250 (2025)

    2025

  18. [25]

    L. S. Tsimring and A. Pikovsky, Noise-induced dynam- ics in bistable systems with delay, Phys. Rev. Lett.87, 250602 (2001)

    2001

  19. [26]

    Masoller, Distribution of residence times of time- delayed bistable systems driven by noise, Phys

    C. Masoller, Distribution of residence times of time- delayed bistable systems driven by noise, Phys. Rev. Lett. 90, 020601 (2003)

    2003

  20. [27]

    Huber and L

    D. Huber and L. S. Tsimring, Dynamics of an ensem- ble of noisy bistable elements with global time delayed coupling, Phys. Rev. Lett.91, 260601 (2003)

    2003

  21. [28]

    Huber and L

    D. Huber and L. S. Tsimring, Cooperative dynamics in a network of stochastic elements with delayed feedback, Phys. Rev. E71, 036150 (2005)

    2005

  22. [29]

    Franosch, M

    T. Franosch, M. Grimm, M. Belushkin, F. M. Mor, G. Foffi, L. Forr´ o, and S. Jeney, Resonances arising from hydrodynamic memory in brownian motion, Nature478, 85 (2011)

    2011

  23. [30]

    Sompolinsky and I

    H. Sompolinsky and I. Kanter, Temporal association in asymmetric neural networks, Phys. Rev. Lett.57, 2861 (1986)

    1986

  24. [31]

    Bratsun, D

    D. Bratsun, D. Volfson, L. S. Tsimring, and J. Hasty, Delay-induced stochastic oscillations in gene regulation, Proceedings of the Na- tional Academy of Sciences102, 14593 (2005), https://www.pnas.org/doi/pdf/10.1073/pnas.0503858102

    2005 doi

  25. [32]

    G. A. Bocharov and F. A. Rihan, Numerical modelling in biosciences using delay differential equations, Journal of Computational and Applied Mathematics125, 183 (2000)

    2000

  26. [33]

    Yamamoto, K

    Y. Yamamoto, K. Aihara, T. Leleu, K.-i. Kawarabayashi, S. Kako, M. Fejer, K. Inoue, and H. Takesue, Coherent ising machines—optical neural networks operating at the quantum limit, npj Quantum Information3, 49 (2017)

    2017

  27. [34]

    Yamamoto, T

    Y. Yamamoto, T. Leleu, S. Ganguli, and H. Mabuchi, Coherent ising machines—quantum optics and neu- ral network perspectives, Applied Physics Letters117, 10.1063/5.0016140 (2020)

    2020 doi

  28. [35]

    N. A. Aadit, A. Grimaldi, M. Carpentieri, L. Theogara- jan, J. M. Martinis, G. Finocchio, and K. Y. Camsari, Massively parallel probabilistic computing with sparse ising machines, Nature Electronics5, 460 (2022)

    2022

  29. [36]

    Gibeault, T

    S. Gibeault, T. N. Adeyeye, L. A. Pocher, D. P. Lath- rop, M. W. Daniels, M. D. Stiles, J. J. McClelland, W. A. Borders, J. T. Ryan, P. Talatchian, U. Ebels, and A. Madhavan, Programmable electrical coupling between stochastic magnetic tunnel junctions, Phys. Rev. Appl. 21, 03...

    2024

  30. [37]

    Y. Gao, G. Chen, L. Qi, W. Fu, Z. Yuan, and A. J. Danner, Photonic ising machines for combinato- rial optimization problems, Applied Physics Reviews11, 10.1063/5.0216656 (2024)

    2024 doi

  31. [38]

    Onizawa and T

    N. Onizawa and T. Hanyu, A unified performance–cost landscape of parallel p-bit ising machines based on up- date dynamics, Scientific Reports 10.1038/s41598-026- 47285-0 (2026)

    2026 doi

  32. [39]

    K. Y. Camsari, R. Faria, B. M. Sutton, and S. Datta, Stochasticp-bits for invertible logic, Phys. Rev. X7, 031014 (2017)

    2017

  33. [40]

    K. Y. Camsari, B. M. Sutton, and S. Datta, P-bits for probabilistic spin logic, Applied Physics Reviews6, 10.1063/1.5055860 (2019)

    2019 doi

  34. [41]

    Chowdhury, A

    S. Chowdhury, A. Grimaldi, N. A. Aadit, S. Niazi, M. Mohseni, S. Kanai, H. Ohno, S. Fukami, L. Theogara- jan, G. Finocchio,et al., A full-stack view of probabilistic computing with p-bits: Devices, architectures, and algo- rithms, IEEE Journal on Exploratory Solid-State Com- p...

    2023

  35. [42]

    Rieke, D

    F. Rieke, D. Warland, R. d. R. Van Steveninck, and W. Bialek,Spikes: exploring the neural code(MIT press, 1996)

    1996

  36. [43]

    Schneidman, W

    E. Schneidman, W. Bialek, and M. J. Berry, Synergy, re- dundancy, and independence in population codes, Jour- nal of Neuroscience23, 11539 (2003)

    2003

  37. [44]

    Selcuk, N

    K. Selcuk, N. A. Aadit, C. Delacour, J. Q. Silva, N. San- jay Singh, H. Kaneko, S. Kanai, Y.-J. Wu, Y.-H. Chen, Y.-S. Chen, Y. C. Ong, K.-C. Huang, H. Chuang, H. Ohno, S. Fukami, and K. Y. Camsari, Dac-free p- bits: Asynchronous self-coloring and on-chip annealing, in2025 IEEE...

    2025

  38. [45]

    Niebur, H

    E. Niebur, H. G. Schuster, and D. M. Kammen, Collec- tive frequencies and metastability in networks of limit- cycle oscillators with time delay, Phys. Rev. Lett.67, 2753 (1991)

    1991

  39. [46]

    Giacomelli, F

    G. Giacomelli, F. Marino, M. A. Zaks, and S. Yanchuk, Coarsening in a bistable system with long-delayed feed- back, Europhysics Letters99, 58005 (2012)

    2012

  40. [47]

    Giacomelli, F

    G. Giacomelli, F. Marino, M. A. Zaks, and S. Yanchuk, Nucleation in bistable dynamical systems with long delay, Phys. Rev. E88, 062920 (2013)

    2013

  41. [49]

    Buzs´ aki,Rhythms of the Brain(Oxford University Press, 2006)

    G. Buzs´ aki,Rhythms of the Brain(Oxford University Press, 2006)

    2006

  42. [50]

    R. D. Fields, A new mechanism of nervous system plas- ticity: activity-dependent myelination, Nature Reviews Neuroscience16, 756 (2015)

    2015

  43. [51]

    Gerstner and W

    W. Gerstner and W. M. Kistler,Spiking neuron models: Single neurons, populations, plasticity(Cambridge uni- versity press, 2002)

    2002

Pith tools

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