{"id":"c45c764b-ca65-455a-801d-5b654e78fb90","arxiv_id":"2607.28781","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"In an athermal random-field Ising model with non-reciprocal couplings, a spontaneous time-oscillatory (time-crystal) phase exists for intermediate disorder strength on complete graphs and in 3D, but not in 2D.","lead":"A two-species spin model with push-pull (non-reciprocal) interactions and random local fields is shown to enter a persistently oscillating 'time crystal' phase at intermediate disorder, without any periodic driving. The same model in two dimensions does not order in time, while on a complete graph and in 3D the oscillation lifetime grows with system size.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Mean-field closure in Eq. 9 is not the exact complete-graph GGD dynamics; the Hopf bifurcation and analytic phase diagram may be artifacts of an uncontrolled approximation.","rationale":"The reader identified the same load-bearing assumption: the K s_iB → K m_B mean-field closure used in the dynamics, validated only at fixed points, not in the oscillatory regime. I agree and sharpen it: the exact complete-graph N→∞ dynamics can be written down without that closure, and it differs from Eq. 9 in the nonlinear terms that determine the limit cycle and phase boundary. This is the most load-bearing concern because the analytic claim of 'solving the dynamics exactly' and the Hopf-bifurcation phase diagram both rest on Eq. 9. The 3D numerical evidence is also weak, but the paper itself qualifies it, whereas the closure is presented as exact. The proposed drift-measurement test settles the issue directly. If the corrected ODE still produces a stable limit cycle, the central claim is supported; if not, the analytic foundation collapses, though the simulations might still indicate a time-crystal-like phase. Since the paper is already CONDITIONAL and this concern is addressable, the verdict need not change.","tokens_in":15906,"tokens_out":11887,"duration_ms":125006,"concrete_test":"Simulate the actual complete-graph GGD for N = 500, 1000, 2000 at representative COP points (e.g., K=0.5, σ=0.4; K=0.3, σ=0.5; K=0.8, σ=0.1). Measure the conditional drift E[Δm_A|m_A,m_B] per MC step from the simulation and compare it with the RHS of Eq. 9 and with the conditional mean-field ODE dm_A/dt = -m_A + (1+m_B)/2 erf((Jm_A+K)/√2σ) + (1-m_B)/2 erf((Jm_A-K)/√2σ) (similarly for m_B). If the simulation drift matches the conditional ODE but not Eq. 9, solve the conditional ODE numerically for the limit-cycle amplitude and phase boundary; if it still has a stable limit cycle in the same K–σ region, the central claim survives with corrected analytics, otherwise the analytic time-crystal proof is refuted.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central analytic result is Eq. 9 (and Appendix B, Eqs. B1–B2): dm_A/dt = -m_A + erf((J m_A + K m_B)/√2σ), obtained by replacing K s_iB with K m_B inside the local field before taking the sign. This is not the N→∞ limit of the actual greedy-Glauber dynamics on a complete graph. The exact steady-state calculation in Appendix A correctly conditions on the discrete value s_iB = ±1 (Eqs. A3–A8); the corresponding dynamical drift is dm_A/dt = -m_A + (1+m_B)/2 erf((J m_A + K)/√2σ) + (1-m_B)/2 erf((J m_A - K)/√2σ) (and analogously for B with K → -K s_iA), not Eq. 9. Replacing a discrete ±1 random variable by its mean inside a nonlinear sign/erf is uncontrolled and can change the limit-cycle amplitude, the oscillation frequency, and the location of the ordered–oscillatory boundary. The paper validates the K m_B closure only for fixed points (Fig. 6), not in the oscillatory regime. Eq. 9 is also asserted via dm/dt = -∂f/∂m (Ref. [51]), but the non-reciprocal coupled dynamics has no joint free-energy potential; composing two single-species gradient flows with different arguments is not itself a gradient flow. If this closure fails, the Hopf bifurcation and the complete-graph phase diagram are not established for the actual model, and the 3D prediction (already based on only small L with selection-biased spectra) loses its analytic underpinning.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a two-species random-field Ising model with non-reciprocal same-site couplings, evolved with athermal greedy Glauber dynamics. The authors claim an exact solution on the complete graph, leading to a K–σ phase diagram with ordered, disordered, and 'chaotic ordered' phases. They identify the oscillatory phase as a spontaneous time crystal, supported by Monte Carlo simulations on complete graphs and cubic lattices, and report its absence in two dimensions. The central analytic result is Eq. (9), a pair of mean-field ODEs for m_A and m_B, from which a Hopf bifurcation and the phase boundaries are derived.","tokens_in":16361,"tokens_out":12683,"duration_ms":130345,"significance":"If correct, the paper would add a clean classical mechanism—quenched disorder plus non-reciprocity—for spontaneous time-translation symmetry breaking without periodic driving, and the complete-graph exactness would make it analytically tractable. The numerical part provides honest finite-size data, includes order parameters R and L, Fourier transforms, and autocorrelation scaling, and does not appear to fit parameters to force oscillations. However, the analytical derivation of Eq. (9) is not valid as an exact complete-graph result, and the 3D evidence is based on small system sizes; the significance of the paper therefore hinges on a revision that puts the mean-field dynamics on a solid footing.","major_comments":[{"comment":"Eq. (9) is not the exact complete-graph GGD dynamics. Since s_iB=±1, averaging sgn(Jm_A+Ks_iB+h) gives dm_A/dt = -m_A + (1+m_B)/2 erf((Jm_A+K)/(√2σ)) + (1-m_B)/2 erf((Jm_A-K)/(√2σ)), not erf((Jm_A+Km_B)/(√2σ)). Replacing Ks_iB by Km_B inside erf is uncontrolled; Fig. 6 validates it only at fixed points. At (0,0) the correct Jacobian has off-diagonal erf(K/(√2σ)) and diagonal -1+(J/σ)√(2/π)exp(-K²/(2σ²)), so the Hopf condition is σ=J√(2/π)exp(-K²/(2σ²)), not Eq. (B3). The analytic phase diagram in Figs. 4/9 is therefore not established.","section":"Appendix B, Eq. (9); Appendix A"},{"comment":"The dynamics is asserted as dm_A/dt=-∂f_A/∂m_A with f_A the RFIM free energy, citing Ref. [51]. The coupled non-reciprocal system has no joint free energy; f_A(m_A;m_B) is a single-species functional with m_B as a parameter, and two such gradient flows do not form a gradient system. Ref. [51]'s Model A dynamics therefore does not justify Eq. (9). The equation must be derived from the microscopic update or explicitly treated as a closure and tested in the oscillatory regime.","section":"Eqs. (8)-(9), Appendix B"},{"comment":"The 3D claim rests on small sizes (L=8,10,20 in Fig. 12; L=20,30 in Fig. 17) and τ~L^{1.55} from Fig. 18; the text admits lack of self-averaging, and Fig. 17 selects the 'most prominent peak'. With no valid complete-graph analytic control (see above), the numerical evidence is not strong enough to support the abstract's statement that the autocorrelation time diverges in three dimensions.","section":"Cubic lattice / Figs. 5, 12, 17, 18"}],"minor_comments":[{"comment":"The abstract calls the phase 'chaotic time-oscillatory' while also reporting 'stable time oscillations'; a limit cycle and a chaotic attractor are different objects. Clarify, also in relation to 'time quasi-crystal' in the Discussion.","section":"Abstract and Discussion"},{"comment":"Time unit inconsistency: Eq. (9) sets δ to one Monte Carlo step, Appendix C defines one Monte Carlo run as two sweeps, and Fig. 3 captions use 'Monte Carlo runs'. Define t unambiguously.","section":"Eq. (9); Appendix C"},{"comment":"The figure selects the realization with the most prominent peak; state the selection criterion and how many realizations were used, to avoid peak-selection bias.","section":"Fig. 17"},{"comment":"The sentence 'for K>1 ... system exists in a chaotic state with small amplitude oscillations for all σ' seems to extend the oscillatory phase to all large K, in tension with 'intermediate K' in the abstract. Clarify whether this is the same collective time-crystal phase or single-site oscillations.","section":"Text near Fig. 7"},{"comment":"The derivation of the OP/COP boundary K=(J-√(π/2)σ)/(2√2) is compressed; present the steps leading to the bound on sin(4θ)/(3+cos(4θ)) more explicitly.","section":"Appendix B(b)"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a timely question and contains valuable simulations, but the central analytic claim is overstated. I would ask the authors to re-derive the mean-field equations from the microscopic GGD (the correct conditional-average equation is straightforward to write down) and recompute the phase diagram; if the revised equations still produce a Hopf bifurcation, the paper's qualitative conclusions may survive, but the present version should not be accepted as an exact solution. The self-cited Ref. [50] is used for the free-energy potential; please check that the citation is appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Punchline: the model is worth looking at, but the central analytic result is a mean-field closure that the paper never derives or validates in the oscillatory regime. Until that is fixed, the time-crystal claim is a conjecture supported by simulation, not by the exact solution advertised in the abstract.\n\nWhat's actually new: a two-species random-field Ising model with non-reciprocal couplings run under greedy Glauber dynamics, where quenched disorder provides the stochasticity. The complete-graph phase diagram — ordered, chaotic-oscillatory, disordered — and the finite-dimension contrast (3D shows growing autocorrelation time, 2D does not) are new and plausible. The simulations are extensive for the complete graph, and the R and L order parameters do give a clean qualitative check of the phase boundaries. The paper is also candid about the limits of its 3D runs.\n\nThe soft spots are real. Eq. 9 is asserted, not derived from the microscopic GGD. Replacing K s_iB by K m_B before taking the sign is a mean-field approximation; it is not the N→∞ dynamics. The exact steady-state equations in Appendix A condition on s_iB = ±1, and the corresponding drift would be a weighted mix of two erf terms, not simply erf((J m_A + K m_B)/√2σ). The authors only check this closure for fixed points (Fig. 6), not for the limit cycle. Since the Hopf bifurcation and the entire analytic phase diagram come from Eq. 9, the analytic support is not established unless the closure can be justified. That is the main problem.\n\nSmaller concerns: the 3D evidence relies on L=20 and 30, with Fourier peaks selected from the most favorable realization and no error bars on those spectra. The autocorrelation-time scaling (τ ~ L^1.55) is based on very few points. And the paper does not clearly separate its contribution from Hanai's order-by-disorder time crystal or the limit-cycle phase already known in the Avni et al. non-reciprocal Ising model.\n\nBottom line: this deserves a serious referee, but the authors need to either derive the correct complete-graph dynamics or carefully qualify the approximation, and they need stronger 3D data. I'd send it out with the referee explicitly asked to scrutinize Eq. 9.","headline":"Interesting model and honest numerics, but the analytic phase diagram rests on an uncontrolled mean-field closure, so the time-crystal claim is not yet established for the actual dynamics.","tokens_in":16764,"tokens_out":3775,"would_cite":false,"duration_ms":43484,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["82C20","82C26","82D30"],"pacs":[],"model":"deepseek-v4-flash","headline":"Disorder and non-reciprocal interactions alone can produce a time crystal in a classical athermal spin model, without any external periodic driving.","keywords":["time crystal","random field Ising model","non-reciprocal interactions","greedy Glauber dynamics","Hopf bifurcation","limit cycle","athermal dynamics","quenched disorder"],"falsifier":"Simulate greedy Glauber dynamics on a complete graph at, say, K = 0.5, σ = 0.4 for N up to a few thousand and compare the measured oscillation amplitude and period with the numerical solution of dm_α/dt = −m_α + erf((J m_α ± K m_β)/√(2σ²)). If the autocorrelation time fails to grow with N while the closure predicts a limit cycle, or if the limit cycle disappears when the closure is relaxed (e.g., by solving the full set of spin equations), the central claim fails.","tokens_in":15781,"feed_emoji":"⏰","tokens_out":2380,"duration_ms":25972,"temperature":0.7,"pith_summary":"This paper studies a two-species random field Ising model with non-reciprocal interactions under athermal greedy Glauber dynamics. It claims that the combination of quenched random-field disorder and non-reciprocal coupling is sufficient to generate a time crystal — sustained, synchronized oscillations whose autocorrelation time grows with system size — even though no external drive is applied. On a complete graph the dynamics are solved exactly, yielding a phase diagram with a chaotic oscillatory phase; the same phase appears in three dimensions, but not in two. If correct, this provides a new route to time-crystalline order in a simple classical disordered system.","feed_headline":"Disorder plus non-reciprocal spins make a time crystal without a drive","feed_subtitle":"A classical spin model shows sustained oscillations that grow with system size, in 3D but not 2D.","key_machinery":"The central machinery is the mean-field closure for the interspecies coupling, replacing the local term K s_iB by K m_B (and similarly for A), which turns the greedy Glauber dynamics on a complete graph into a pair of coupled ordinary differential equations with error-function nonlinearities. These equations admit a Hopf bifurcation at the disordered-phase boundary and a transition to an ordered steady state; the phase boundaries in the K–σ plane are derived by analyzing the polar-coordinate flow and the existence of nontrivial fixed points.","core_discovery":"The paper establishes that a two-species random field Ising model with non-reciprocal interactions, evolving under greedy Glauber dynamics, exhibits a time-crystal phase at intermediate values of the non-reciprocal coupling K and disorder strength σ. On a complete graph, the exact mean-field dynamics reduce to dm_α/dt = −m_α + erf((J m_α ± K m_β)/√(2σ²)), which undergo a Hopf bifurcation at σ = √(2/π)J, producing a stable limit cycle with collective oscillations. The autocorrelation time τ diverges with system size as τ ∼ N^0.35 on the complete graph and as τ ∼ L^1.55 on a cubic lattice, while in two dimensions τ saturates and no time crystal forms.","pith_inferences":["The mean-field closure replaces the local interspecies field by its spatial average, but the closure is only validated at fixed points; the oscillatory regime may be sensitive to correlations that the closure ignores, so the exact Hopf boundary is not yet fully established for the original dynamics.","The same mechanism — disorder replacing thermal noise while non-reciprocity provides the frustration that prevents relaxation — may generalize to other non-reciprocal spin models, perhaps even continuous spins, offering a route to design time crystals with only static randomness.","The τ ∼ L^1.55 scaling in 3D, if confirmed with larger system sizes, would be a new dynamic exponent for this disorder-induced limit cycle, and could be compared with the known scaling of the non-reciprocal Ising model without disorder.","A testable extension would be to check whether the oscillation amplitude and period are robust to changing the disorder distribution (e.g., bimodal instead of Gaussian), which would distinguish a mechanism based on the shape of the error-function nonlinearity from one relying on rare large fields."],"forward_implications":["This predicts a time crystal in a classical, athermal, disordered spin system without periodic driving, so time-translation symmetry is broken spontaneously rather than by an external clock.","The phase diagram on a complete graph supplies an exact reference: ordered phase at low σ, chaotic ordered (time-crystal) phase at intermediate σ, and disordered phase at large σ, with the Hopf line at σ = √(2/π)J.","The autocorrelation-time scaling τ ∼ N^0.35 on the complete graph and τ ∼ L^1.55 on a cubic lattice imply a genuine many-body time crystal in three dimensions, whereas the absence of scaling in two dimensions indicates a lower critical dimension between 2 and 3.","For K > J, all nonzero fixed points vanish, so the model reduces to small-amplitude single-site oscillations for any disorder strength, marking the upper edge of the time-crystal regime."],"fun_headline_variants":["Disorder and non-reciprocity create time crystal without drive","Random fields plus non-reciprocity yield time crystal in 3D","Time crystal arises from non-reciprocal random-field spins in 3D but not 2D","Classical spins form time crystal from disorder and non-reciprocity"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The analytics assume the local interspecies coupling can be replaced by its mean-field average (K s_iB ≈ K m_B) throughout, but this approximation is checked only for fixed points, not for the oscillatory regime where the time crystal lives.","fun_headline_variants_meta":{"raw":{"variants":["Disorder and non-reciprocity create time crystal without drive","Random fields plus non-reciprocity yield time crystal in 3D","Time crystal arises from non-reciprocal random-field spins in 3D but not 2D","Classical spins form time crystal from disorder and non-reciprocity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000828,"raw_usage":{"total_tokens":3440,"prompt_tokens":714,"completion_tokens":2726,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":458,"completion_tokens_details":{"reasoning_tokens":2641}},"tokens_in":458,"tokens_out":2726,"duration_ms":16035,"temperature":1.0,"reasoning_tokens":2641,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T00:24:03.029348+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate greedy Glauber dynamics on a complete graph at, say, K = 0.5, σ = 0.4 for N up to a few thousand and compare the measured oscillation amplitude and period with the numerical solution of dm_α/dt = −m_α + erf((J m_α ± K m_β)/√(2σ²)). If the autocorrelation time fails to grow with N while the closure predicts a limit cycle, or if the limit cycle disappears when the closure is relaxed (e.g., by solving the full set of spin equations), the central claim fails.","supporting_citations":[],"review_version":1}