{"id":"adf22607-ffcf-41b3-a939-7ab6c329af4f","arxiv_id":"1908.07025","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Trophic coherence, a measure of how aligned a directed network's edges are with a global direction, is shown to tie together non-normality, spectral radius, strong connectivity, and a majority-rule phase transition at critical coherence q_c=1.","lead":"Directed networks can be organized by a hidden global direction, and this trophic coherence turns out to explain why they differ from undirected graphs. This paper unifies several known signatures of directionality, from non-normality to strong connectivity, and shows that even a simple majority-rule dynamics becomes unstable on such networks.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The spectral-radius step behind the non-normality theorem uses trace as a Gelfand norm; trace is not a norm and the expectation/limit interchange is unproved, so Eq. (6), and the bounds built on it, are not established.","rationale":"The reader identified Eq. (3) as the weakest assumption because it is imported without derivation. My concern is adjacent but more specific: even granting Eq. (3), the step from Eq. (3) to Eq. (6) is invalid as stated. Gelfand's formula requires a matrix norm, and trace is not one; the trace of powers counts closed walks and can oscillate. The paper also silently replaces the limit of expected traces with the expected spectral radius, which is not generally justified. Because the non-normality theorem is the main new proof in the paper, this is a load-bearing gap. However, the empirical patterns and simulations may still support the conceptual claim, and the flaw could plausibly be repaired by a more careful spectral-radius argument or by weakening the theorem to an empirical statement. Therefore I do not move the verdict away from conditional acceptance; the paper needs a rigorous treatment of this step before the central theorem can be trusted.","tokens_in":18255,"tokens_out":7821,"duration_ms":91548,"concrete_test":"Recompute the spectral radius by direct eigendecomposition for all networks in Tables S1-S4 and for an ensemble of preferential-preying networks with N=1000, B=20 across q in [0,2], and compare with e^tau from Eq. (4). If the relative error |rho - e^tau|/rho exceeds a few percent in the tau < 0 regime, Eq. (6) fails for the systems where the theorem claims dF is close to 1. As an analytical control, verify that the trace-Gelfand limit exists for a 3-node directed cycle; it does not, although rho = 1.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central quantitative link is Eq. (6), rho = e^tau, obtained by 'taking the norm in Eq. (5) to be the trace'. This is not a valid use of Gelfand's formula: the trace functional is not a matrix norm. Consequently, lim_k (tr A^k)^(1/k) need not equal rho. For a directed cycle, tr A^k vanishes except when the cycle length divides k, so the ordinary limit does not exist even though rho = 1. Moreover, Eq. (3) supplies only an expectation, and no argument justifies exchanging the expectation with the limit before taking the kth root. Since the Theorem's bounds, Eqs. (16)-(19), and the loopful/loopless spectral-radius classification all rely on rho = e^tau, the derivation of the central non-normality result is incomplete. The empirical tables even contain a visible discrepancy: Bridge Brook Lake has tau = -0.53, giving e^tau about 0.59, while Table S1 reports rho = 1. This is the load-bearing step to test.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper argues that a single property of directed networks, trophic coherence, controls a cluster of apparently unrelated features: spectral radius, non-normality, the size of the strongly connected component, cycle abundance, and the stability of majority-rule dynamics. The central theoretical result is a theorem stating that the normalized departure from normality tends to 1 with increasing trophic coherence and exceeds sqrt(1 - 1/<k>) in the loopless regime. The proof proceeds by combining the coherence-ensemble formula for tr(A^k), Eq. (3), with Gelfand's formula to obtain the spectral radius rho = e^tau, and then bounding the sum of squared eigenvalue moduli. The author validates the predictions on 62 empirical networks and with simulations of a preferential preying model, and illustrates the dynamical consequences on the C. elegans neural network.","tokens_in":18400,"tokens_out":10857,"duration_ms":104714,"significance":"If the central derivation were correct, the paper would provide a valuable unification of several directed-network phenomena and strengthen the case that edge-direction alignment is a fundamental organizing principle. The manuscript is clearly written, the data set is diverse, and the stated data/code availability is a strength. However, the key mathematical step linking Eq. (3) to the spectral radius is invalid as stated, so the theorem and its bounds are not currently proven. The empirical observations remain suggestive, but the main quantitative claim needs a rigorous derivation or a clear rephrasing as a conjecture supported by numerics.","major_comments":[{"comment":"The step 'taking the norm in Eq. (5) to be the trace' is not a legitimate application of Gelfand's formula. The trace functional is not a matrix norm: it is not submultiplicative and does not satisfy the axioms required for the formula. For a directed cycle of length n, tr(A^k) is n when n divides k and 0 otherwise, so the limit in Eq. (5) does not exist even though rho = 1. Moreover, Eq. (3) supplies only the ensemble expectation of tr(A^k), and no argument justifies interchanging the expectation with the limit and the k-th root before taking k to infinity. Since the proof of the theorem and the bounds in Eqs. (16)-(19) rely on rho = e^tau, the central theoretical result is not established.","section":"Results, 'Graph ensembles' and 'Non-normality', Eqs. (5)-(6)"},{"comment":"The theorem's proof hinges entirely on Eq. (3), which is imported from the author's prior work without derivation or a statement of the conditions under which it holds. The main text attributes this result to 'Ref. [1]', but in the reference list [1] is Newman's review article; the intended source appears to be Johnson and Jones (2017), listed as [9]. As it stands, the proof is not self-contained, and the reader cannot verify the key formula on which all subsequent claims rest.","section":"Results, 'Graph ensembles', Eq. (3) and its citation"},{"comment":"The empirical validation in Figures 2 and 3 uses the same 62 networks on which the coherence-ensemble machinery was developed and tested in the prior work. This does not invalidate the theorem, but it weakens the claim that the bounds 'hold for these empirical cases too' as an independent confirmation. The preferential preying model simulations provide some independent support, but the main theoretical claim would benefit from out-of-sample tests on networks not used to calibrate the ensemble.","section":"Results, 'Strong connectivity' and 'Non-normality', Figures 2-3"},{"comment":"The derivation of the critical value q_c = 1 assumes that the ratio k_i^n/k_i^b equals the constant lambda for every non-basal vertex, which is a property of the basal ensemble but only approximately true for the preferential preying model used in the simulations. The text appropriately notes that the mean-field analysis is for the basal ensemble and that the extension to other ensembles is heuristic, but the phrasing 'the same critical value may also apply' makes the status of q_c = 1 for general networks less definite than the abstract suggests.","section":"Methods 0.2, 'Majority rule dynamics on basal ensemble networks'"}],"minor_comments":[{"comment":"The main text repeatedly cites 'Ref. [1]' for the coherence ensemble and the basal ensemble, but reference [1] in the main-text list is Newman (2003). The intended reference is presumably Johnson and Jones (2017), which is listed as [9]. Please correct all such citations.","section":"References throughout"},{"comment":"The abbreviation 'SSC' is introduced for the strongly connected component, but 'SCC' is used everywhere else, including in the tables and figure captions. Please standardize.","section":"Section 'Strong connectivity'"},{"comment":"The caption contains a repeated article: 'the the C. elegans neural network' should be 'the C. elegans neural network'.","section":"Figure 1 caption"},{"comment":"The affiliation line contains spacing errors: 'Edgbasto n' and 'Lo ndon' should be 'Edgbaston' and 'London'.","section":"Author affiliation"},{"comment":"For Bridge Brook Lake, tau = -0.53 gives e^tau approx 0.59, but the table reports rho = 1. The paper does not comment on this discrepancy between the actual spectral radius and the coherence-ensemble expectation, which is relevant because Figures 2 and 3 use tau to place networks on the horizontal axis.","section":"Table S1"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a single-author perspective that builds heavily on the author's prior work (Johnson and Jones 2017). The reference mislabeling in the main text is serious and suggests the manuscript needs careful editorial checking. The primary technical issue is the invalid use of Gelfand's formula with the trace; if the authors can supply a rigorous derivation of rho = e^tau (for example, via the spectral radius of nonnegative matrices and the growth rate of closed walks), the central theorem would be placed on solid ground. I do not recommend rejection, as the empirical observations are suggestive and the gap appears fixable, but the proof as written is not sound."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is worth a serious referee, but the central proof step does not hold. Eq. (6), rho = e^tau, is derived by \"taking the norm in Eq. (5) to be the trace.\" Trace is not a matrix norm. Gelfand's formula applies to norms, and the trace functional does not satisfy the required properties. For a directed cycle, tr A^k vanishes except when the cycle length divides k, so the ordinary limit in Gelfand's formula does not even exist, although the spectral radius is well defined. You also need to justify exchanging the ensemble expectation in Eq. (3) with the limit in k before taking the kth root. Neither issue is addressed. This matters because the non-normality bounds (Eqs. 16-19) and the loopful/loopless spectral-radius classification all rest on rho = e^tau. The data tables make the problem visible: Bridge Brook Lake has tau = -0.53, so e^tau is about 0.59, yet Table S1 reports rho = 1. So the empirical support for the theorem is weaker than the text implies.\n\nNow the credit. The paper does something genuinely useful: it shows that several apparently unrelated directed-network phenomena—spectral radius, strong connectivity, cycle growth, non-normality—all track a single scalar, the loop exponent tau. The figures are persuasive as empirical regularities. The new results are the non-normality bounds and the majority-rule phase transition at q_c = 1. The mean-field calculation for the basal ensemble is clearly laid out, and the simulation support on a different generative model is a nice touch. The author is also honest about the provenance of the coherence ensemble from Ref. [1] and about the fact that the same data set is used for validation, which is a mild circularity but not a fatal one given the prior work is published.\n\nThe soft spots, beyond the Gelfand issue: the link between cycle decay and SCC size is heuristic, and the majority-rule mean-field assumes a constant ratio k_i^n/k_i^b, which is only approximate for the preferential preying model. These are secondary.\n\nWho should read this: network scientists and complex-systems people who want a compact synthesis of directionality effects. What it needs: a proper derivation of Eq. (6) or a recasting of the theorem as a numerically supported conjecture. I would send it to peer review with a request for major revision; the empirical sections can stand, but the theoretical claim needs to be made rigorous or honestly downgraded.","headline":"The unifying story is appealing but the central theorem is not proven: Eq. (6) treats the trace as a Gelfand norm, and the one visible counterexample in the tables shows rho != e^tau for at least one network.","tokens_in":18995,"tokens_out":3180,"would_cite":false,"duration_ms":35629,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05C20","05C82"],"pacs":["89.75.Hc","89.75.-k"],"model":"deepseek-v4-flash","headline":"This paper argues that the most distinctive features of directed networks—trophic coherence, non-normality, spectral radius, strong connectivity, and cycle counts—share one origin: how strongly edge directions align with a global direction.","keywords":["directed networks","trophic coherence","non-normality","spectral radius","strongly connected component","majority-rule dynamics","network ensembles","complex systems"],"falsifier":"Measure the spectral radius and SCC fraction on empirical directed networks with measured trophic coherence; any network with $\\tau<0$ that still has a giant strongly connected component, or whose spectral radius stays far above $e^{\\tau}$, would refute the regime split. In generated networks, pushing $q\\to 0$ and finding $\\rho$ not collapsing to $e^{\\tau}$ would directly falsify Eq. (3).","tokens_in":17972,"feed_emoji":"🧭","tokens_out":13971,"duration_ms":133420,"temperature":0.7,"pith_summary":"Directed networks are often treated as undirected graphs with arrows added at random. This paper tries to establish that the arrows are the story: in many real systems they align with a global direction, and that single ordering explains a cluster of otherwise unrelated anomalies. Using the coherence ensemble, the paper shows that trophic coherence and non-normality are two sides of the same coin, and that the sign of one loop exponent separates networks with large strongly connected components and abundant cycles from networks where both vanish. It also shows—on the worm neural network and on generated graphs—that the same ordering flips the stability of the simplest majority-rule dynamics, with a critical coherence at $q_c=1$. If this is right, a single measurable property of edge directions organizes much of directed-network structure and dynamics.","feed_headline":"Directed networks split into two regimes by one number","feed_subtitle":"A single measure of arrow alignment ties coherence, non-normality, and a sharp stability switch.","key_machinery":"The workhorse is the coherence ensemble: a random directed graph ensemble with fixed in- and out-degree sequences and a prescribed trophic coherence $q$, together with the basal ensemble in which each non-basal vertex receives the same fraction of basal in-edges. From that ensemble comes the identity $\\mathrm{tr}(A^k)=\\frac{\\tilde\\alpha \\tilde q}{\\alpha q}e^{\\tau k}$, which turns the loop exponent $\\tau$ into the spectral radius via Gelfand's formula. The sign of $\\tau$ then acts as an order parameter: positive $\\tau$ means exponentially growing counts of directed circuits and a giant strongly connected component; negative $\\tau$ means exponentially suppressed circuits and a vanishing SCC. The non-normality theorem needs only this spectral-radius result plus the bound $|\\lambda_i|\\le\\rho$, and the majority-rule stability calculation uses the basal-ensemble ratio $k_i^{\\mathrm n}/k_i^{\\mathrm b}=L/L_B-1\\equiv\\lambda$ to reduce flips to a binomial probability, yielding $q_c=1$.","core_discovery":"The central claim is that a single number, the loop exponent $\\tau = \\ln\\alpha + \\frac{1}{2}\\tilde q^2 - \\frac{1}{2}q^2$, with branching factor $\\alpha=\\langle k^{\\mathrm{in}}k^{\\mathrm{out}}\\rangle/\\langle k\\rangle$ and trophic incoherence $q$, governs whether a directed network lives in a loopful or loopless regime. In the coherence ensemble the expected trace of $A^k$ is a single exponential, so Gelfand's formula gives the spectral radius $\\rho=e^{\\tau}$. The paper proves that the normalized departure from normality tends to $1$ as $q\\to 0$, and exceeds $\\sqrt{1-1/\\langle k\\rangle}$ whenever $\\tau<0$; coherent networks are therefore highly non-normal, with spectral radius and strongly connected component both collapsing. It further derives, from a mean-field basal-ensemble calculation, a critical value $q_c=1$ for majority-rule dynamics: below that coherence the mean activity flips sign persistently, above it the dynamics is stable. The empirical networks included in the paper line up with the predicted regime split, and the C. elegans connectome shows the predicted coherence-dependent bistability.","pith_inferences":["A natural extension the author does not pursue: because the same exponent controls cycle counts and SCC size, one could use measured directed-cycle distributions as a proxy for trophic coherence, avoiding the need to solve for trophic levels in networks without basal vertices.","A natural extension the author does not pursue: because non-normality is known to cause transient amplification, coherent directed networks should show strong transient responses before reaching their stable state; this could be tested with linearized dynamics on the same empirical networks.","The basal-ensemble derivation assumes each non-basal vertex receives the same fraction of basal inputs; breaking that assumption in the preferential-preying model should shift or smear the predicted $q_c=1$, giving a controlled test of the mean-field logic.","The $q_c=1$ transition suggests a design rule for engineered information-flow networks: tuning coherence through rewiring should let one switch between stable consensus and sensitive novelty detection, a trade-off directly relevant to sensory and regulatory systems."],"forward_implications":["If the coherence-ensemble description is correct, the sign of $\\tau$ is a regime label: trophically coherent networks with $\\tau<0$ should have near-zero spectral radii and no giant strongly connected component, while $\\tau>0$ networks should be loopful and strongly connected.","Since $d_F\\to 1$ as $q\\to 0$, trophic coherence and non-normality are not independent properties; a mechanism that aligns edge directions automatically makes the adjacency matrix far from normal, so explanations of stability invoking one implicitly involve the other.","The majority-rule result implies that coherence can reverse the direction of stability: the same degree sequence is dynamically stable when incoherent and unstable when coherent, so degree statistics alone are insufficient to predict dynamical outcomes on directed networks.","The mean-field transition at $q_c=1$ gives a quantitative target: real networks with $q$ below 1 should exhibit persistent sign flips under majority-rule updates, while those above 1 should settle into a stable consensus."],"supporting_citations":[{"why":"Supplies the coherence ensemble, the exponential trace identity of Eq. (3), and the empirical network set the paper builds on.","marker":"[1]"},{"why":"Supplies the non-normality measure and the observation that many real directed networks are highly non-normal.","marker":"[7]"},{"why":"Defines trophic coherence and links it to ecosystem stability, framing the ordering phenomenon.","marker":"[8]"},{"why":"Provides the preferential-preying model used to generate networks with tunable coherence in the simulations.","marker":"[12]"},{"why":"Defines the directed configuration ensemble on which the coherence and basal ensembles are based.","marker":"[17]"},{"why":"Gives Gelfand's spectral-radius formula used to convert the trace result into $\\rho=e^{\\tau}$.","marker":"[18]"},{"why":"Reports the suppression of directed cycles in real networks that the loopless regime explains.","marker":"[19]"},{"why":"Specifies the majority-rule update dynamics whose stability transition is analyzed.","marker":"[21]"},{"why":"Supplies the C. elegans neural architecture used to demonstrate coherence-dependent bistability.","marker":"[36]"}],"fun_headline_variants":["One number splits directed networks into loopful and loopless","Loop exponent: the scalar that explains directionality's effects","Non-normality and coherence share a single cause","Directionality's key: one exponent controls dynamical regime","How a single number sets the stability of directed networks"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument depends on Eq. (3)—that in the coherence ensemble closed walks of length $k$ are exponentially numerous at rate $e^{\\tau k}$—an assumption imported from earlier work without derivation here, so if real directed networks deviate from that ensemble expectation the spectral-radius, non-normality, and strong-connectivity conclusions lose their quantitative grounding.","fun_headline_variants_meta":{"raw":{"variants":["One number splits directed networks into loopful and loopless","Loop exponent: the scalar that explains directionality's effects","Non-normality and coherence share a single cause","Directionality's key: one exponent controls dynamical regime","How a single number sets the stability of directed networks"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000765,"raw_usage":{"total_tokens":3353,"prompt_tokens":864,"completion_tokens":2489,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":480,"completion_tokens_details":{"reasoning_tokens":2411}},"tokens_in":480,"tokens_out":2489,"duration_ms":20765,"temperature":1.0,"reasoning_tokens":2411,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:29:44.606223+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the spectral radius and SCC fraction on empirical directed networks with measured trophic coherence; any network with $\\tau<0$ that still has a giant strongly connected component, or whose spectral radius stays far above $e^{\\tau}$, would refute the regime split. In generated networks, pushing $q\\to 0$ and finding $\\rho$ not collapsing to $e^{\\tau}$ would directly falsify Eq. (3).","supporting_citations":[{"cited_title":"Looplessness in networks is linked t o trophic coherence,","cited_arxiv_id":null,"evidence_quote":"Supplies the coherence ensemble, the exponential trace identity of Eq. (3), and the empirical network set the paper builds on."},{"cited_title":"Disturbance, re- source supply, and food-web architecture in streams,","cited_arxiv_id":null,"evidence_quote":"Supplies the non-normality measure and the observation that many real directed networks are highly non-normal."},{"cited_title":"Trophic interactions in Caribbean coral reefs,","cited_arxiv_id":null,"evidence_quote":"Defines trophic coherence and links it to ecosystem stability, framing the ordering phenomenon."},{"cited_title":"Complex trophic interactions in deserts: an empirical c ritique of food-web theory,","cited_arxiv_id":null,"evidence_quote":"Provides the preferential-preying model used to generate networks with tunable coherence in the simulations."},{"cited_title":"Does food web theory work for marine ecosystems?,","cited_arxiv_id":null,"evidence_quote":"Defines the directed configuration ensemble on which the coherence and basal ensembles are based."},{"cited_title":"Predators, pa rasitoids and pathogens: species richness, trophic generality and body sizes in a natural foo d web,","cited_arxiv_id":null,"evidence_quote":"Gives Gelfand's spectral-radius formula used to convert the trace result into $\\rho=e^{\\tau}$."},{"cited_title":"Spatial and temporal variation in the structur e of a freshwater food web,","cited_arxiv_id":null,"evidence_quote":"Reports the suppression of directed cycles in real networks that the loopless regime explains."},{"cited_title":"Construction of a larg e Caribbean food web,","cited_arxiv_id":null,"evidence_quote":"Specifies the majority-rule update dynamics whose stability transition is analyzed."},{"cited_title":"The structure of the nervous system of the nematode caenorhabditis elegans,","cited_arxiv_id":null,"evidence_quote":"Supplies the C. elegans neural architecture used to demonstrate coherence-dependent bistability."}],"review_version":1}