{"id":"5a01155d-aa87-45d6-b763-b440aae48a41","arxiv_id":"2608.10076","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For delay-coupled oscillator networks, synchronization stability can be reduced to a low-dimensional master stability condition on the adjacency eigenvalues, and optimized networks become heterogeneous, directed master-slave structures.","lead":"The authors extend the master stability function formalism to delay-coupled networks with heterogeneous, weighted, and directed connections. They find that such networks can synchronize more robustly than homogeneous ones and provide an optimization method for designing them.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Central claim overreaches: the reduction to a single d* MSF is proven only for τ→0 and identity coupling, and is explicitly false for the LK model; the general conclusion needs the multi-indegree intersection condition.","rationale":"The paper's core contribution is real: it extends master stability analysis to delay-coupled systems with directed, weighted, heterogeneous adjacency matrices, and the subsystem-level condition in Eq. (8) is a meaningful generalization. The exact proofs for identical-indegree and master-slave networks, together with the independent LK time-series validation in SM Sec. SIII, give genuine support for the restricted claims. The reader's conditional verdict is therefore appropriate. I focused on a different load-bearing step than the reader's weakest_assumption: rather than the approximate simultaneous triangularization (which the authors quantify and explicitly label as an approximation), the decisive gap is the reduction from ∩_j Ω_j to a single Ω(d*). This reduction is what turns the framework into the paper's one-parameter spectral-placement claim. It is proven only in a special small-delay, identity-coupling limit; the proof's algebra is not valid for general complex ν; and the paper itself demonstrates monotonicity failure for the LK model. Since the central conclusion is stated without these qualifications, it is broader than what the derivation supports. The correct fix is modest: either restrict the central claim to cases where the monotone ordering of Ω_j is verified, or state the general result as requiring all subsystems. Because the underlying framework and the optimization results can survive that correction, rejection is not warranted; conditional acceptance with mandatory revision remains the right outcome.","tokens_in":19979,"tokens_out":12079,"duration_ms":128698,"concrete_test":"Take the optimal SL network at τ=10 from Fig. 2(b,c), the case with residual ||eΔ_lower||_F = 2.7×10^-2. For each eigenvalue ν of A and each diagonal entry eΔ_jj of eΔ = P^{-1}ΔP, compute the subsystem MTLE λmax(D(0)f + eΔ_jj D(0)h, νD(τ)h) using DDE-BIFTOOL. If max over j and ν of these values differs from the d*-based prediction λmax(D(0)f + d*D(0)h, νD(τ)h) at the same ν—in particular, if some ν lies in Ω(d*) but outside Ω_j for some j—then the d*-reduction in the central conclusion fails for that network. Repeating the same check at τ=0.5 and τ=5 would delimit the regime in which the central claim actually holds.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central conclusion asserts that synchronization stability is entirely determined by λmax(J'_1,J'_2(ν)) with J'_1 = D(0)f + d*D(0)h. This requires the stability region Ω = ∩_j Ω_j to collapse to Ω(d*), where d* = min_j d_j. That collapse is established only in SM Sec. SI under the restrictive assumptions τ→0 and D(τ)h = −D(0)h = I_n. Even there the proof is incomplete: Eq. (S7) shifts μ by −eΔ_jj and concludes the spectrum moves left, but the actual exponents are λ = μ/(1+ντ) with complex ν, so the real-part shift is −eΔ_jj Re(1/(1+ντ)), which is not always stabilizing. For large τ or non-identity coupling, no proof is given; the text says monotonicity is 'observed numerically' but no large-τ monotonicity plot is presented. The paper itself shows the failure of monotonicity for the LK model (Fig. 3b), where Ω shrinks with increasing d*, and the authors concede that all subsystems must be analyzed. Therefore the general, unqualified central claim is not supported. The valid result is conditional: if Ω_j is monotonically ordered by eΔ_jj, then d* suffices; otherwise one must check λmax(D(0)f + eΔ_jj D(0)h, νD(τ)h) < 0 for all j and all eigenvalues ν of A.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This Letter proposes a master stability function (MSF) generalization for delay-coupled networks of identical oscillators with weighted, directed, and heterogeneous degree structure. The authors linearize the coupled system around an identical synchronous solution, introduce a similarity transformation P that simultaneously Jordanizes the adjacency matrix A and lower-triangularizes the indegree matrix Δ, and obtain a staircase system of n-dimensional variational modes (Eqs. (6)-(9)). They define per-subsystem stability regions Ω_j and, under an assumed monotonicity in the diagonal entries eΔ_jj, reduce the stability condition to a single MSF λmax(J'_1,J'_2(ν)) with J'_1 = D(0)f + d*D(0)h and J'_2 = νD(τ)h, where d* is the minimal indegree. They apply the framework to Stuart-Landau oscillators and Lang-Kobayashi laser arrays, compute MSF landscapes, and develop a constrained optimization algorithm that minimizes the maximum transverse Lyapunov exponent. The numerical results identify optimal networks with directed master-slave-like, heterogeneous structures and show that, for small delays, delayed coupling can produce a deeper stability region than the non-delayed case. Independent time-series simulations of the LK model are used to test stability predictions for all-to-all networks.","tokens_in":20273,"tokens_out":7541,"duration_ms":66569,"significance":"If the low-dimensional reduction were valid in the stated generality, the paper would be practically important: synchronization stability of a high-dimensional delay-coupled network would reduce to the spectral placement of the adjacency matrix eigenvalues in a low-dimensional stability region, enabling systematic network design. The special-case derivations for identical indegree and master-slave networks are correct, and the independent LK time-series validation that distinguishes predicted stable and unstable regimes is a genuine strength; the availability of code and data is also commendable. However, the central claim as stated is too broad: the paper itself shows that the monotonicity in indegree fails for the LK model, and the general heterogeneous case relies on an approximate simultaneous triangularization without an error bound. The conditional version of the result and a proper validation on optimized heterogeneous networks would make this a solid contribution; in its current form the central conclusion is not established.","major_comments":[{"comment":"The central conclusion that synchronization stability is 'entirely determined' by λmax(J'_1,J'_2(ν)) with J'_1 = D(0)f + d*D(0)h is not established for general heterogeneous networks. The reduction requires two conditions: an exact simultaneous similarity transformation putting A in Jordan form and Δ in lower-triangular form, and monotone inclusion of the per-subsystem stability regions Ω_j in the diagonal entries eΔ_jj. The paper proves the simultaneous transformation only for identical-indegree and master-slave networks; for the optimized networks in Figs. 2 and 4 the non-lower-triangular part of eΔ is nonzero as reported in SM Eqs. (S4)-(S5). The monotonicity is proven only in the limit τ→0 with D(τ)h = -D(0)h = I_n (SM Sec. SI), and the LK model in Fig. 3(b) displays the opposite monotonicity, with the stability region shrinking as d* increases. The text itself concedes that all subsystems must be analyzed in that case. The unconditional statement should therefore be replaced by a conditional statement: if the regions Ω_j are monotonically ordered in eΔ_jj, then d* suffices; otherwise the intersection Ω = ∩_j Ω_j must be checked for each subsystem Ξ_j.","section":"Main text, 'MSF generalization for delay-coupled systems', Eqs. (6)-(9) and central conclusion"},{"comment":"The proof that Ω_j expands monotonically with eΔ_jj is incomplete. In Eq. (S7) the authors show that increasing eΔ_jj shifts μ leftward, but with their definition μ = λ_ℓ(1+ντ) the actual characteristic exponents are λ_ℓ = μ/(1+ντ). Therefore the real part of the shift is -δ Re(1/(1+ντ)) rather than -δ; for complex ν with Re(1/(1+ντ)) negative, the shift can be destabilizing. The conclusion that the entire spectrum is 'uniformly shifted leftward' does not follow from the displayed equation. In addition, the main text states that for large delays the monotonicity is 'observed numerically', but no large-delay monotonicity plot is presented; Fig. 3 and Fig. S2 only show small-delay comparisons. The authors should either provide a valid proof of the monotonicity condition, state precisely the conditions under which it holds, or base the general criterion on the intersection Ω = ∩_j Ω_j.","section":"Supplemental Material, Sec. SI, Eq. (S7)"},{"comment":"The numerical evidence for simultaneous triangularization of the optimized heterogeneous networks is only approximate: the normalized Frobenius norm of the non-lower-triangular part of eΔ is up to 2.7×10^{-2} for the SL model at τ=10 and 2.5×10^{-2} for the LK model (SM Eqs. (S4)-(S5)). Because the mode-decoupling in Eq. (8) is exact only when this part vanishes, the manuscript needs a quantitative perturbation bound connecting this norm to the error in λmax or to a corrected stability condition. Without such a bound, the eigenvalue-placement plots in Fig. 2(c) do not prove that the optimized heterogeneous networks synchronize. The independent LK time-series tests in SM Sec. SIII use all-to-all networks, not the optimized structures of Fig. 4, and therefore do not close this gap.","section":"Supplemental Material, Sec. SI, 'Analysis for the optimal network structures'"},{"comment":"The verification that the eigenvalues of optimized networks lie inside the stability region is partly self-referential. The stability region Ω is computed from the MSF using the minimal indegree d* of the same optimized network, and the optimization minimizes λmax evaluated through that same MSF. The agreement therefore confirms consistency of the optimization with its own objective function rather than providing independent confirmation that the full delay-coupled system synchronizes. An independent test for at least one optimized heterogeneous network, such as direct integration of Eq. (1), would substantially strengthen the claim; the current independent time-series evidence is limited to the all-to-all LK configurations in SM Sec. SIII.","section":"Main text, Fig. 2(c) and 'Optimizing networks for synchronization'"}],"minor_comments":[{"comment":"The phrase 'approximately transforms D into a lower triangular matrix' appears to refer to the indegree matrix Δ, not to the matrix D; the notation should be corrected.","section":"Supplemental Material, Sec. SI"},{"comment":"The diagonal entries of the transformed indegree matrix are denoted eΔ_jj in the main text and eD_jj in the SM; please use one consistent notation throughout.","section":"Supplemental Material, Sec. SI and main text"},{"comment":"The Taylor expansion e^{λτ} ≈ 1 + λτ in the derivation of Eq. (S7) appears to produce a sign inconsistency in the definition of μ; please check whether μ = λ(1+ντ) or μ = λ(1-ντ) is intended.","section":"Supplemental Material, Sec. SI, Eqs. (S6)-(S7)"},{"comment":"The phrase 'stationary synchronous states of the form E_j(t) = r* e^{iΩ0 t}' is not literally stationary in the original time variable; please clarify that this is a state that is stationary in a rotating frame.","section":"Main text, 'MSF analysis for non-diffusive coupling'"},{"comment":"Please state the number of nodes M used for the optimized networks in Fig. 2(b), since the eigenvalue placement and the stability conclusions depend on M.","section":"Fig. 2 caption"}],"recommendation":"major_revision","confidential_remarks":"The paper has a promising idea and useful numerical machinery, but the central claim is currently stated more strongly than the supporting proof allows. The LK counterexample and the approximate triangularization are not minor caveats; they affect the main theorem. I would encourage the editor to require a revision that states the conditional theorem precisely, adds the missing validation for optimized heterogeneous networks, and either proves or explicitly checks the monotonicity condition for each model and delay. If the authors rework the claim in this way, the paper could be suitable for this journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper does real work. It extends the MSF to delay-coupled networks that are degree-heterogeneous, weighted, and directed, which is a genuine gap in the literature. The simultaneous Jordan/triangular reduction is a clean idea, and the special cases (identical indegree, master-slave with distinct or identical diagonal entries) are handled correctly. The LK time-series validation is a strong point: predicted stable and unstable transitions are confirmed by independent simulation, and the code and data are public. The optimization results, pointing to master-slave directed structures as optimal for delayed coupling, are interesting and well-illustrated.\n\nThe soft spot is the central conclusion. The claim that stability is “entirely determined” by the MSF with the minimum indegree d* is not supported in the generality stated. The proof in the SM is limited to the small-delay limit with identity Jacobian coupling, and even there the spectral shift argument is questionable: the shift depends on Re(1/(1+ντ)), not simply on −eΔ_jj, so it does not always move the spectrum left. The paper itself shows the LK model violates the monotonicity assumption (Fig. 3b), and the authors concede that all subsystems must be analyzed for that model. So the valid statement is conditional: if the stability regions Ω_j are monotonically ordered by the diagonal entries of Δ̃, then d* suffices; otherwise one must check every subsystem. That conditional version is useful, but the unqualified conclusion overreaches.\n\nA second concern is the approximate simultaneous triangularization for the optimized heterogeneous networks. The SM reports Frobenius norms of the non-lower-triangular part up to 2.7×10−2, which is small but not a proof that mode decoupling holds. For the specific master-slave outputs (LK optimization), the exact equivalence is proven, so those examples are fine. But the SL optimized networks are not in a provably triangularizable class, and the paper relies on numerical smallness without a sensitivity analysis. This is mitigable, but it should be addressed.\n\nI also note the monotonicity claim for large delays is said to be “observed numerically” but no large-τ monotonicity plot is provided; the SL small-τ results match theory, which is fine, but the statement outruns the evidence.\n\nWho is this for? Researchers working on synchronization in delay-coupled lasers, multi-agent systems, and anyone using MSF for network design. They will find the framework and the optimization algorithm useful, provided they read the conditions carefully.\n\nVerdict: send to peer review, but require the authors to qualify the central theorem, fix or clarify the SM proof, and address the approximate triangularization. The core is solid enough to merit referee time, but the broadest claims need revision.","headline":"Genuinely useful MSF extension for delayed heterogeneous directed networks, but the headline claim as stated is too strong; the exact reduction to a single d* MSF is proven only under restrictive conditions and is false for the LK model.","tokens_in":20789,"tokens_out":1586,"would_cite":true,"duration_ms":18222,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05C82","34D06","34K20"],"pacs":["05.45.Xt","42.55.Px","89.75.-k"],"model":"deepseek-v4-flash","headline":"Synchronization stability of delay-coupled networks reduces to one low-dimensional master stability function over adjacency eigenvalues.","keywords":["master stability function","delay-coupled networks","synchronization","heterogeneous networks","directed networks","network optimization","Stuart-Landau oscillators","Lang-Kobayashi lasers"],"falsifier":"Take one of the optimized Stuart-Landau networks at $\\tau=10$ (or a Lang-Kobayashi network at large coupling), compute the residual $P^{-1}\\Delta P$ after Jordanizing $A$, then compare the MSF-predicted maximum transverse Lyapunov exponent against a direct numerical solution of the full $Mn$-dimensional variational delay equation; a systematic discrepancy growing with the residual norm would falsify the reduction. A cheaper test: construct a directed network with identical indegrees plus a small strictly upper-triangular perturbation of $\\Delta$ and check whether the mode-decoupling prediction holds to first order.","tokens_in":19782,"feed_emoji":"🕸️","tokens_out":5887,"duration_ms":52568,"temperature":0.7,"pith_summary":"The paper extends the master stability function (MSF) framework, previously limited to degree-homogeneous delay-coupled networks, to weighted, directed, and degree-heterogeneous networks. Its central claim is that synchronization stability of the full M n-dimensional delay-coupled system is decided by a low-dimensional MSF, evaluated at the complex eigenvalues of the adjacency matrix, provided the network satisfies a simultaneous triangularization condition. On this basis the paper shows that time delays shrink the MSF stability region, so degree-homogeneous structures such as all-to-all networks can lose stability and become suboptimal. An optimization algorithm then produces master-slave, nonreciprocal, heterogeneous networks whose eigenvalues sit inside the stability region and can synchronize more stably than homogeneous ones, including in a small-delay regime where delayed coupling outperforms its non-delayed counterpart. The framework is demonstrated on Stuart-Landau oscillators, representing systems near a Hopf bifurcation, and on coupled semiconductor lasers.","feed_headline":"Delays make heterogeneous directed networks the best synchronizers","feed_subtitle":"A generalized master stability function shows that hub-like master-slave wiring beats all-to-all in delayed systems.","key_machinery":"The central object is the generalized master stability function for delay-coupled systems, built on a simultaneous Jordanization-and-triangularization matrix $P$: a similarity transformation that sends the adjacency matrix $A$ to its Jordan form and the diagonal indegree matrix $\\Delta$ to a lower-triangular matrix. Acting on the variational equation, it yields a staircase system of $M$ $n$-dimensional mode equations in which each mode depends only on earlier modes; if every mode subsystem $\\Xi_j$ has negative maximum transverse Lyapunov exponent, the synchronous state is stable. The paper also uses the characteristic equation $\\det(J_1 + J_2 e^{-\\lambda_\\ell \\tau} - \\lambda_\\ell I_{Mn}) = 0$ for linear delay systems and, under small-delay and identity-coupling conditions, proves that the stability region $\\Omega_j$ expands with the diagonal entry $\\tilde{\\Delta}_{jj}$, reducing the analysis to the smallest indegree $d_*$. The MSF maps each complex eigenvalue $\\nu$ of $A$ to the maximum transverse Lyapunov exponent, turning network design into a spectral placement problem with an indegree constraint.","core_discovery":"The paper's central conclusion is that the synchronization stability of the delay-coupled system is entirely determined by the MSF function $\\lambda_{\\max}(J'_1, J'_2(\\nu))$, which maps the complex eigenvalue $\\nu$ of the adjacency matrix $A$ to the maximum transverse Lyapunov exponent computed from the $n\\times n$ matrices $J'_1 = D^{(0)}f + d_* D^{(0)}h$ and $J'_2 = \\nu D^{(\\tau)}h$. When a similarity matrix $P$ simultaneously puts $A$ in Jordan form and the indegree matrix $\\Delta$ in lower-triangular form, the variational equation splits into staircase-coupled modes; stability of the full system follows by induction from the stability of the $n$-dimensional mode subsystems. The paper proves this reduction for identical-indegree and master-slave networks and verifies it approximately for optimized heterogeneous networks. Using this reduction, it shows that increasing delay shrinks the stability region $\\Omega$, so homogeneous networks with a single longitudinal eigenvalue $\\nu_0 = d$ eventually lose synchronization; optimized networks instead place their whole eigenvalue spectrum inside $\\Omega$ and are directed, weighted, and heterogeneous, typically master-slave. For small delays, the delayed MSF can be deeper than the non-delayed one in the Lang-Kobayashi laser model, so delaying coupling can strengthen synchronization rather than only weaken it.","pith_inferences":["If the spectral-placement view is taken seriously, one can invert the logic and design networks by prescribing eigenvalue locations inside the stability region and then finding adjacency matrices with those spectra and indegree bounds; the paper shows existence only indirectly through optimization, so a constructive spectral-synthesis algorithm is a natural next step.","The master-slave optima suggest a testable engineering rule: in delayed multi-agent control and photonic arrays, adding a few dominant, nonreciprocal hub connections may improve synchronization tolerance to delay more than increasing overall coupling strength.","The Lang-Kobayashi result that small delays deepen the stability region implies delays can be exploited as a resource rather than only a nuisance; a concrete extension would be a delay-tuning experiment on semiconductor arrays sweeping $\\tau$ through the predicted 0.05–0.27 ns window and measuring phase coherence.","Because the carrier variable in Lang-Kobayashi is uncoupled from other nodes, higher indegree shrinks the stability region; this predicts that any master-slave network with an uncoupled internal variable in its coupling function will invert the usual degree–stability monotonicity."],"forward_implications":["Stability analysis of a delay-coupled network can be performed by computing the MSF over the complex plane and checking that all eigenvalues of the adjacency matrix lie inside the stability region, reducing an $Mn$-dimensional problem to $n$ dimensions.","Degree-homogeneous networks, including all-to-all coupling, become increasingly suboptimal as delay grows; their longitudinal eigenvalue can leave the stability region even when transverse modes remain stable.","Deliberately directed, weighted, heterogeneous (master-slave-like) topologies can synchronize more stably than homogeneous ones, and in small-delay laser arrays can outperform the optimal non-delayed configuration.","For systems near a Hopf bifurcation, represented by Stuart-Landau oscillators, larger minimal indegree expands the stability region, while delay shrinks it.","Network optimization guided by the MSF converges to hierarchical structures with high edge-weight variance, high maximum outdegree and eigenvector centrality, and low clustering."],"supporting_citations":[{"why":"Introduces the original master stability function formalism whose assumptions this paper relaxes for delays.","marker":"[24]"},{"why":"Shows how non-diagonalizable network matrices can be handled in the master stability framework, providing the basis for the Jordan-block treatment of directed networks.","marker":"[26]"},{"why":"Previous delay-system MSF extension restricted to simultaneously diagonalizable (identical-indegree) cases, the limitation this paper removes.","marker":"[37]"},{"why":"Supplies the characteristic equation for linear delay systems used here to define the maximum transverse Lyapunov exponent.","marker":"[41]"},{"why":"Provides the numerical method used to compute the characteristic roots and validate the MSF landscapes.","marker":"[44]"},{"why":"Defines the Lang-Kobayashi model used as the non-diffusive coupled-laser example.","marker":"[48]"}],"fun_headline_variants":["Heterogeneous directed networks outperform all-to-all in delayed sync","Delays make nonreciprocal hub wiring optimal for synchronization","Master-slave topology wins in delay-coupled networks","Heterogeneity and directionality boost sync under delay","Generalized MSF: delayed nets favor asymmetric wiring"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole reduction rests on the existence of a similarity matrix that simultaneously puts the adjacency matrix in Jordan form and the indegree matrix in lower-triangular form; this is proven for identical-indegree and master-slave networks but only verified approximately for the optimized networks, where the leftover upper-triangular piece has Frobenius norm up to $2.7\\times 10^{-2}$.","fun_headline_variants_meta":{"raw":{"variants":["Heterogeneous directed networks outperform all-to-all in delayed sync","Delays make nonreciprocal hub wiring optimal for synchronization","Master-slave topology wins in delay-coupled networks","Heterogeneity and directionality boost sync under delay","Generalized MSF: delayed nets favor asymmetric wiring"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000296,"raw_usage":{"total_tokens":1719,"prompt_tokens":947,"completion_tokens":772,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":563,"completion_tokens_details":{"reasoning_tokens":694}},"tokens_in":563,"tokens_out":772,"duration_ms":7518,"temperature":1.0,"reasoning_tokens":694,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T04:14:44.946111+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take one of the optimized Stuart-Landau networks at $\\tau=10$ (or a Lang-Kobayashi network at large coupling), compute the residual $P^{-1}\\Delta P$ after Jordanizing $A$, then compare the MSF-predicted maximum transverse Lyapunov exponent against a direct numerical solution of the full $Mn$-dimensional variational delay equation; a systematic discrepancy growing with the residual norm would falsify the reduction. A cheaper test: construct a directed network with identical indegrees plus a small strictly upper-triangular perturbation of $\\Delta$ and check whether the mode-decoupling prediction holds to first order.","supporting_citations":[{"cited_title":"Gjurchinovski, E","cited_arxiv_id":null,"evidence_quote":"Introduces the original master stability function formalism whose assumptions this paper relaxes for delays."},{"cited_title":"Acharyya, P","cited_arxiv_id":null,"evidence_quote":"Shows how non-diagonalizable network matrices can be handled in the master stability framework, providing the basis for the Jordan-block treatment of directed networks."},{"cited_title":"B¨ orner, P","cited_arxiv_id":null,"evidence_quote":"Previous delay-system MSF extension restricted to simultaneously diagonalizable (identical-indegree) cases, the limitation this paper removes."},{"cited_title":"Mahdavi, M","cited_arxiv_id":null,"evidence_quote":"Supplies the characteristic equation for linear delay systems used here to define the maximum transverse Lyapunov exponent."},{"cited_title":"[44–47], for details on the theoretical analysis, cou- pling classes, and network optimization method","cited_arxiv_id":null,"evidence_quote":"Provides the numerical method used to compute the characteristic roots and validate the MSF landscapes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the Lang-Kobayashi model used as the non-diffusive coupled-laser example."}],"review_version":1}