{"id":"5d17f2d3-c032-476e-a7a3-0dcf8972d0b2","arxiv_id":"1908.08105","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Memristive network dynamics are governed by a projector of the circuit's cycle space; this paper reviews that framework, adds a disorder-averaged equation, proves planar locality bounds do not extend to non-planar graphs, and reports logarithmic relaxation in simulations.","lead":"An exploratory paper on a toy model of memristive circuits links the circuit's graph topology to its nonlinear dynamics, and connects the long-time states of the network to spin-glass physics. It offers a small new result: the planar locality bound cannot be generalized to non-planar circuits using face-bounding cycles, and it presents numerics suggesting glassy, logarithmic relaxation under disorder.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Equation (13) does not reduce to Eq. (1) when disorder is removed, so the disorder-generalized dynamics underlying Fig. 8 is internally inconsistent as stated.","rationale":"The reader's weakest assumption already targeted Eq. (13), and I agree that this is the load-bearing point. The missing derivation is a serious gap, but the concern is stronger: the equation as printed fails a direct consistency check in the homogeneous limit. This does not invalidate the no-go theorem, which is a correct Euler-characteristic argument for the face-counting method, nor the review aspects or the Ising-model connection. It does mean the new numerical observation (Fig. 8) rests on an equation that is not established. Since the paper is a perspective with several independent contributions, the appropriate disposition remains CONDITIONAL: the authors should supply a derivation of Eq. (13), verify the homogeneous limit, and re-run Fig. 8 with the corrected equation and error bars. I therefore keep the reader's verdict unchanged.","tokens_in":9389,"tokens_out":32622,"duration_ms":301234,"concrete_test":"Set N_ii=0 in Eq. (13) and reduce A,B to alpha,beta times the identity; symbolic evaluation yields d w/dt = alpha w - (1/beta)(I + Omega W)^{-1} Omega S. Compare with Eq. (1), which has (I + xi Omega W)^{-1} Omega S. Run this comparison on a two-edge loop (for which Omega has entries 1/2) with xi=2 and a numerical integrator; if the two trajectories do not coincide, Eq. (13) does not generalize Eq. (1) and Fig. 8 is not a valid test of the disordered model.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing assumption is that Eq. (13) is the correct disorder-generalized equation. This is not merely underived; as written it does not reduce to Eq. (1) in the homogeneous limit. Setting N_ii=0 gives T=I and Omega'=Omega, so Eq. (13) becomes d w/dt = alpha w - (1/beta)(I + Omega W)^{-1} Omega S. Equation (1) is d w/dt = alpha w - (1/beta)(I + xi Omega W)^{-1} Omega S. These agree only if xi=1. The paper claims 'if N_ij=0 we recover the previous equation,' but that limit forces xi=0, and even then (I + Omega W)^{-1} Omega S differs from Omega S. Thus the projector construct in Eq. (13) places the disorder-dependent matrix T in a way that cannot reproduce the homogeneous evolution for general xi. Consequently the log-relaxation claim in Fig. 8, obtained by integrating Eq. (13), is unsupported unless Eq. (13) is re-derived or corrected with the homogeneous limit as a consistency check. The no-go theorem itself is independent and, under the standard cellular-embedding interpretation, is sound.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper is a perspective article on a toy model of memristive networks. The authors review the exact differential equation for the internal memory variables in homogeneous circuits (Eq. 1), introduce a generalization to disordered memristors (Eq. 13), prove a new no-go theorem that the face-counting method used in [20] to derive locality bounds cannot be extended from planar to non-planar graphs, and discuss connections to slow relaxation, Ising models, and QUBO optimization. The no-go theorem is the main new theoretical result: using the Euler characteristic and the cycle-space dimension, the authors show that any closed orientable surface embedding of a non-planar graph has too few faces to supply a basis of face-bounding cycles, so the planar locality bound cannot be generalized by this method.","tokens_in":9640,"tokens_out":16903,"duration_ms":147630,"significance":"The no-go theorem is correct and is a genuine, if modest, contribution: it cleanly explains a structural obstruction to extending the planar locality bound, and it is derived parameter-free from the Euler characteristic. The paper is also useful as a concise introduction to the graph-theoretic origin of the projector Omega in memristive circuit equations. However, the disordered generalization (Eq. 13) that underlies the numerical log-relaxation claim is asserted without derivation and does not reduce to the homogeneous equation (1) as claimed, while the Ising-mapping equations (15) are garbled. These issues bear directly on the paper's central claims, so the manuscript needs substantial revision before it can be accepted.","major_comments":[{"comment":"Equation (13) does not reduce to Eq. (1) in the homogeneous limit, contrary to the claim that 'if N_ij = 0 we recover the previous equation.' Setting N_ii = 0 gives T = I and Omega' = Omega, so Eq. (13) becomes d w/dt = A w - B^{-1}(I + Omega W)^{-1} Omega S. For homogeneous parameters A = alpha I and B = beta I, this is alpha w - (1/beta)(I + Omega W)^{-1} Omega S, whereas Eq. (1) contains (I + xi Omega W)^{-1} Omega S. The two expressions agree only for xi = 1. Since Fig. 8 is obtained by integrating Eq. (13), the log-relaxation observation is not supported unless Eq. (13) is re-derived or corrected and the numerics repeated.","section":"..to Statistical Physics, Eq. (13)"},{"comment":"The Ising-mapping equations are garbled: the line 'Sigma = Omega, p/2 = alpha xi. alpha/2 + alpha xi/3 Omega_ii - 1/beta sum_j Omega_ij S_j = r_i - p/2 Sigma_ii' does not parse as a well-formed system of equations, and the surrounding text does not specify how the vector S is obtained from r and Sigma. As written, the asserted mapping between the memristive dynamics and the QUBO functional M(W) cannot be verified or used. Please rewrite Eq. (15) as a proper set of equations and provide the derivation of the mapping.","section":"..to Statistical Physics, Eq. (15)"},{"comment":"The claim that the relaxation is 'compatible with a logarithmic one' is not adequately supported: the average is over only 20 simulations, no error bars are shown, the disorder distribution is not specified (the caption gives sigma = 0.05 but does not define what sigma is, and also says 'homogeneous across the system'), the initialization of the internal variables is not described, and no quantitative goodness-of-fit criterion for the claimed log(t) regime is provided. This is a load-bearing issue because the simulation integrates Eq. (13), whose correctness is in question (see the first major comment).","section":"..to Statistical Physics, Fig. 8"}],"minor_comments":[{"comment":"The manuscript references 'eqn. (13)' in the Introduction before Eq. (13) is introduced; please renumber or reorder the presentation.","section":"Introduction"},{"comment":"The no-go proof should explicitly state that the embedding is cellular; Euler's formula V - E + F = 2 - 2g holds only when all faces are disks, and the argument as written implicitly assumes this.","section":"Locality, Eqs. (10)-(11)"},{"comment":"The dimension formula dim C = |E| - |V| + 1 assumes a connected graph; for a graph with c connected components the correct expression is |E| - |V| + c, which only strengthens the conclusion, but the assumption should be stated.","section":"Locality, Eq. (11)"},{"comment":"The transposition conventions for the cycle matrix A are inconsistent: Eq. (2) writes A^t(ARA^t)^{-1}A while the later definition of the projector uses A(A^T A)^{-1}A^T. Please clarify the dimensions of A so that the projector acts on the edge space consistently.","section":"From Graph theory, Eqs. (2)-(3) and 'Cycle matrix'"},{"comment":"There are numerous typographical errors, including 'highlighlites', 'intendedended', and 'exogeneous'; a careful proofreading is needed.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The paper is largely a perspective based on the authors' prior work, and the self-citation pattern is transparent but heavy. The no-go theorem is correct and independent of the self-citations. The decision hinges on Eq. (13): if the authors cannot provide a proper derivation that recovers Eq. (1) in the homogeneous limit, the log-relaxation claim should be withdrawn or substantially qualified. The garbled Ising mapping also needs to be fixed. The paper is within the scope of the journal as a perspective, but the central new content is currently not reliable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The one solid new result here is the no-go theorem: for a non-planar graph embedded in an orientable surface, the face-bounding cycles cannot be numerous enough to support the planar locality-bound argument. The Euler-characteristic derivation is short, self-contained, and correct. That part should survive peer review and is citable on its own.\n\nThe paper is otherwise an honest, useful perspective on the authors' own prior work. It is transparent that most of the content is not novel, and the graph-theoretic introduction to the projector Ω is clear and readable. The numerical suggestion of logarithmic relaxation in Fig. 8 is interesting, but I have a serious problem with its foundation.\n\nEquation (13) is the load-bearing new ingredient for the disorder case. It is asserted with “it can be shown” and no derivation. Worse, the claimed homogeneous limit does not work as stated. Setting N_ii=0 gives T=I and Ω′=Ω, so (13) becomes d w/dt = αw − (1/β)(I + ΩW)^{-1} ΩS. The homogeneous Eq. (1) has (I + ξΩW)^{-1} ΩS. These agree only if ξ=1. The paper's own parameter ξ is not set to 1 in the homogeneous case, so (13) does not reduce to (1) for general ξ. The stress-test note is right; this is not a cosmetic issue. The log-relaxation claim in Fig. 8 is obtained by integrating (13), so it is unsupported until (13) is re-derived or corrected and the numerics re-run with the homogeneous limit as a consistency check.\n\nOther soft spots are minor in comparison. Fig. 8 has no error bars and little detail on initialization or disorder realizations. The Ising-mapping equations (15) look garbled—the first line seems to have misrendered subscripts and the mapping is not clearly invertible. The citation pattern is heavily self-referential, but for a perspective on the authors' own line of work that is acceptable, and the no-go theorem is independent of those citations.\n\nThe no-go theorem and the review material are worth publishing after major revision. I would not rely on the current Fig. 8 claim, and I would want the equation fixed before accepting. A serious referee should see this, not a desk reject. Send it to peer review, but be prepared for a heavy revision request.","headline":"The no-go theorem is a clean, correct new result, but the disorder-generalized equation (13) does not reduce to the homogeneous Eq. (1) as claimed, which undermines the paper's central numerical observation.","tokens_in":10122,"tokens_out":2341,"would_cite":true,"duration_ms":23156,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["05.45.-a","05.20.-y","07.50.Ek"],"model":"deepseek-v4-flash","headline":"A no-go theorem shows that the planar exponential-decay locality bound for memristive circuits cannot be extended to non-planar graphs by face-bounding cycles, because only the sphere has enough faces.","keywords":["memristive networks","cycle-space projector","Kirchhoff laws","planar locality bound","no-go theorem","graph genus","Ising model","glassy relaxation"],"falsifier":"Simulate a small circuit with two or three memristors having different $\\beta_i$ and $\\alpha_i$ values, integrate the component-level equations together with Kirchhoff's laws, and compare with the paper's equation (13); any disagreement refutes the disorder extension and the logarithmic-relaxation claim. For the no-go theorem, the refuting observation would be a non-planar graph embedded in a closed orientable surface with $|F|\\ge |E|-|V|+1$; Euler's formula makes that impossible, so searching for such an embedding should fail.","tokens_in":9178,"feed_emoji":"⚡","tokens_out":12768,"duration_ms":111494,"temperature":0.7,"pith_summary":"This paper develops a graph-theoretic picture of circuits made only of memristors, where the circuit topology enters the memory dynamics through a projector $\\Omega$ onto the cycle space of the graph. Its new theoretical result is a no-go theorem: the face-bounding-cycle method that yields an exponential locality bound for planar circuits cannot be generalized to non-planar circuits, because requiring $|F|\\ge \\dim C = |E|-|V|+1$ together with the Euler characteristic $\\chi=|V|-|E|+|F|=2-2g$ forces $g\\le 1/2$, so only the sphere qualifies. The paper also states a generalized disorder equation for non-identical memristors and reports numerical evidence that the average internal memory relaxes logarithmically in time, a glassy signature. It then maps the asymptotic memristor states onto an Ising/QUBO functional with exchange coupling proportional to $\\Omega$, connecting the circuit dynamics to mean-field spin-glass physics. A sympathetic reader would care because the no-go theorem sets a precise limit on when memristive circuits can be treated as local, and the Ising mapping gives a concrete electronic playground for disordered-systems questions.","feed_headline":"Only the sphere passes the face-counting test for memristive circuits","feed_subtitle":"If true, non-planar memristive circuits cannot borrow the planar locality argument; new techniques are needed.","key_machinery":"The central object is the cycle-space projector $\\Omega=A(A^T A)^{-1}A^T$, defined from a basis $A$ of the graph's cycle space; it is the only place circuit topology enters the memory dynamics. The proof of the no-go theorem uses the face-counting obstruction: for planar graphs one can choose basis cycles that bound faces and express their inner products through the adjacency matrix of the dual graph, which yields the exponential decay; for non-planar graphs this requires $|F|\\ge \\dim C$, and the Euler characteristic $2-2g=\\chi=|V|-|E|+|F|$ then forbids all genus $g\\ge 1$. On the statistical-physics side, the load-bearing mapping is $\\Sigma=\\Omega$ and $\\frac{p}{2}=\\alpha\\xi$, which identifies the Lyapunov function of the asymptotic dynamics with the Ising/QUBO Hamiltonian, so the circuit's fixed points are the extrema of a binary optimization problem.","core_discovery":"The central claim is that the vectorial memristor dynamics $\\frac{d}{dt}\\vec w = \\alpha \\vec w - \\frac{1}{\\beta}(I+\\xi\\Omega W)^{-1}\\Omega\\vec S$, with $\\Omega = A(A^T A)^{-1}A^T$ the cycle-space projector, encodes Kirchhoff's laws in $\\Omega$ alone, and that for planar circuits the entries of $\\Omega$ decay exponentially with edge distance, $|\\Omega_{ij}|\\le e^{-z\\,d(i,j)+\\tilde\\rho}$. The paper's new no-go theorem asserts that this face-bounding-cycle proof cannot be carried to non-planar graphs: any embedding whose faces could serve as a cycle basis would need $|F|\\ge \\dim C = |E|-|V|+1$, and combining this with $\\chi=|V|-|E|+|F|=2-2g$ gives $g\\le 1/2$, so only the sphere (genus zero) has enough faces. The paper further claims that with disorder the memory dynamics becomes $\\frac{d}{dt}\\vec w = A\\vec w - B^{-1}(I+\\Omega' W)^{-1}\\Omega' T^{-1}\\vec S$, and that numerical simulations on random graphs show the average $\\langle w\\rangle$ settling into a logarithmic relaxation. Finally, asymptotic states are claimed to map onto the QUBO/Ising functional $M(W)=\\sum_i(r_i-\\frac{p}{2}\\Sigma_{ii})w_i - \\frac{p}{2}\\sum_{i\\ne j} w_i \\Sigma_{ij} w_j$ with $\\Sigma=\\Omega$, making the memristive network a heuristic analog optimizer for NP-complete binary problems.","pith_inferences":["Editorial inference: the no-go theorem suggests that for dense non-planar graphs the meaningful locality statement is statistical rather than geometric — $\\Omega_{ij}$ entries are small in a distributional sense (variance $\\sim 1/N$) — and one could test whether correlation functions of the $w_i$ decay like $1/N$ on expander-like circuits.","Editorial inference: logarithmic relaxation in the disordered equation implies that using memristive networks as QUBO heuristics on dense random circuits will be slow; a concrete testable extension is to add a small uniform diffusion term $\\alpha$ and measure whether the log regime crosses over to a power law, which the paper does not report.","Editorial inference: because the disorder matrix $T$ can take negative values in active/passive mixtures, the disorder equation may generate effective frustrated interactions even when bare couplings look ferromagnetic, offering a mechanism for intrinsic spin-glass behavior that could be tested by measuring a spin-glass order parameter in simulations."],"forward_implications":["For non-planar circuits, locality bounds on $|\\Omega_{ij}|$ cannot be obtained by the face-bounding-cycle construction; any such bound must come from a different technique, such as spectral or probabilistic arguments.","The emergent speed-of-light bound $|\\langle w_i(t)w_j(0)\\rangle|\\le K e^{-(d_{ij}-v_{\\rm eff}t)}$ remains established only for planar circuits, so light-cone-like behavior in non-planar memristive networks is an open question.","If the disorder equation (13) is correct, densely connected random memristive networks should show logarithmic, glassy relaxation of the average memory, implying slow approach to computational steady states.","Random-graph $\\Omega$ entries being approximately Gaussian with variance $1/N$ connects the network to mean-field spin-glass physics, so mean-field and replica methods could predict the asymptotic value of $\\langle w\\rangle$."],"supporting_citations":[{"why":"Supplies the planar locality bound $|\\Omega_{ij}|\\le e^{-z d(i,j)+\\tilde\\rho}$ and the face-bounding-cycle method that the no-go theorem rules out for non-planar graphs.","marker":"[20]"},{"why":"Derives the homogeneous vectorial memristor equation that the paper generalizes to disorder in eq. (13).","marker":"[12]"},{"why":"Provides the mean-field Ising/Lyapunov-function connection and the Monte-Carlo comparison used for asymptotic $\\langle w\\rangle$.","marker":"[24]"},{"why":"Introduces the mapping of memristive dynamics to QUBO/Ising optimization that the paper extends to arbitrary $\\Omega$.","marker":"[26]"},{"why":"Reports experimental slow relaxation in atomic switch networks that the numerical logarithmic relaxation is compared with.","marker":"[17]"},{"why":"Reports experimental power-law-like relaxation in atomic switch networks, the glassy analogue for the disordered memristor results.","marker":"[18]"},{"why":"Defines the canonical mean-field spin-glass model, the reference point for the approximately Gaussian distribution of $\\Omega_{ij}$ on random graphs.","marker":"[28]"}],"fun_headline_variants":["Sphere-only proof limits memristive circuit analysis","Non-planar memristive circuits need new techniques","Face-count test: only sphere works for memristive nets","Memristive networks: from Kirchhoff to Ising, but planar only","Topology restricts memristive circuit face-count proofs"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the disorder equation $\\frac{d}{dt}\\vec w = A\\vec w - B^{-1}(I+\\Omega' W)^{-1}\\Omega' T^{-1}\\vec S$, introduced with 'it can be shown' and no derivation, is the correct Kirchhoff-consistent extension of the homogeneous memristor dynamics when memristors are not identical, because the paper's logarithmic-relaxation evidence depends entirely on it.","fun_headline_variants_meta":{"raw":{"variants":["Sphere-only proof limits memristive circuit analysis","Non-planar memristive circuits need new techniques","Face-count test: only sphere works for memristive nets","Memristive networks: from Kirchhoff to Ising, but planar only","Topology restricts memristive circuit face-count proofs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000227,"raw_usage":{"total_tokens":1481,"prompt_tokens":961,"completion_tokens":520,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":577,"completion_tokens_details":{"reasoning_tokens":438}},"tokens_in":577,"tokens_out":520,"duration_ms":6415,"temperature":1.0,"reasoning_tokens":438,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:48:26.420395+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate a small circuit with two or three memristors having different $\\beta_i$ and $\\alpha_i$ values, integrate the component-level equations together with Kirchhoff's laws, and compare with the paper's equation (13); any disagreement refutes the disorder extension and the logarithmic-relaxation claim. For the no-go theorem, the refuting observation would be a non-planar graph embedded in a closed orientable surface with $|F|\\ge |E|-|V|+1$; Euler's formula makes that impossible, so searching for such an embedding should fail.","supporting_citations":[{"cited_title":"Caravelli, Phys","cited_arxiv_id":null,"evidence_quote":"Supplies the planar locality bound $|\\Omega_{ij}|\\le e^{-z d(i,j)+\\tilde\\rho}$ and the face-bounding-cycle method that the no-go theorem rules out for non-planar graphs."},{"cited_title":"Caravelli, F","cited_arxiv_id":null,"evidence_quote":"Derives the homogeneous vectorial memristor equation that the paper generalizes to disorder in eq. (13)."},{"cited_title":"Caravelli, P","cited_arxiv_id":null,"evidence_quote":"Provides the mean-field Ising/Lyapunov-function connection and the Monte-Carlo comparison used for asymptotic $\\langle w\\rangle$."},{"cited_title":"Asymptotic behavior of memristive circuits","cited_arxiv_id":"1712.07046","evidence_quote":"Introduces the mapping of memristive dynamics to QUBO/Ising optimization that the paper extends to arbitrary $\\Omega$."},{"cited_title":"Avizienis et al., PLoS ONE 7(8): e42772","cited_arxiv_id":null,"evidence_quote":"Reports experimental slow relaxation in atomic switch networks that the numerical logarithmic relaxation is compared with."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Reports experimental power-law-like relaxation in atomic switch networks, the glassy analogue for the disordered memristor results."},{"cited_title":"Sherrington, S","cited_arxiv_id":null,"evidence_quote":"Defines the canonical mean-field spin-glass model, the reference point for the approximately Gaussian distribution of $\\Omega_{ij}$ on random graphs."}],"review_version":1}