{"id":"cbaf5527-6b1b-4a09-8c08-94d391994b53","arxiv_id":"2608.13302","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Semiglobal input-to-state stability is certified for weakly reversible networks with inputs in the toric locus and for certain non-weakly-reversible networks via linear conjugacy or reconstruction.","lead":"This paper proves new conditions under which chemical reaction networks tolerate time-varying reaction rates without losing stability, using input-to-state stability theory. It extends prior results to networks with nonzero deficiency, multiple linkage classes, and some non-weakly-reversible networks, and applies them to parallel molecular computations.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 3.7's forward-completeness proof assumes a uniform derivative bound on the input, which the stated piecewise locally Lipschitz class does not supply; without it, the Gronwall argument in (3.10)-(3.11) does not rule out finite-time blow-up.","rationale":"The reader's weakest_assumption matches my reading. The proof of Step 2 of Theorem 3.7 is the linchpin: all main theorems inherit its forward-completeness conclusion, either directly or through Proposition 3.5. The stated input class does not imply a uniform derivative bound on [0,t_max), so the Gronwall argument is not valid as written. I do not see a more serious internal inconsistency; the transformation framework, examples, and statements are coherent, and the gap appears repairable by adding a uniform Lipschitz assumption or a sharper forward-completeness argument. Hence this is not a rejection, but the central theorem is not established for the declared input class. The verdict remains conditional, so no change from the reader's assessment.","tokens_in":26311,"tokens_out":13795,"duration_ms":162425,"concrete_test":"Verify the gap by taking T=1, u(t)=u^*+\\epsilon_0 \\sin(1/(1-t)) on [0,1), with u^* in the toric locus of a weakly reversible network and \\epsilon_0 small enough that u(t) stays in a compact U\\subset T (e.g., the p53 network of Example 3.9). Compute L(t)=sup_{s\\in[0,t]}|u'(s)|; for this input L(t)\\to\\infty as t\\to 1. Then retrace Step 2 with L(t) in place of L and check whether the Gronwall inequality can be integrated with a finite constant; it cannot, so the proof of forward completeness fails for this admissible input. If the theorem is claimed to survive, supply an independent proof of boundedness for this input or restrict the input class.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim rests on semiglobal ISS for all weakly reversible networks with compact input-value set U inside the toric locus, and this rests on Step 2 of Theorem 3.7 proving forward completeness. The proof differentiates W(t)=V(x(t),x*(u(t))) and needs (3.7): |\\dot x_i^*(u(t))| \\leq C1 L r, with a single finite L bounding |du_j/dt| on [0,t_max). The stated input class (Section 2.3) is only 'piecewise locally Lipschitz' on R_{\\ge0}. On a finite maximal interval this gives, for each \\epsilon>0, a Lipschitz constant L_\\epsilon on [0,t_max-\\epsilon], but L_\\epsilon may grow without bound as \\epsilon\\to 0. For example, u(t)=u^*+\\epsilon_0 \\sin(1/(T-t)) lies in a compact U\\subset T for small \\epsilon_0, is smooth (hence piecewise locally Lipschitz) on [0,T), but sup_{[0,T)}|du/dt|=\\infty. Then C3=C1C2Lr in (3.10) is not a finite constant, and the Gronwall bound (3.11) cannot contradict lim_{t\\to T} W(t)=\\infty. Therefore forward completeness is unproven for the declared input class, and since Proposition 3.5 is used in Theorems 4.2 and 4.14 without an independent forward-completeness argument, the gap propagates to the non-weakly-reversible extensions. This is a genuine proof gap; it may be repairable by strengthening the input hypothesis (uniform Lipschitz on [0,t_max), or u with essentially bounded derivative) or by an alternative forward-completeness proof.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies input-to-state stability (ISS) for mass-action chemical reaction networks with time-varying reaction-rate inputs. It claims two main extensions of earlier ISS results of Chaves and Sontag: first, semiglobal ISS for all weakly reversible networks with arbitrary deficiency and linkage classes, provided the input-value set is a compact subset of the toric locus; second, semiglobal ISS for certain non-weakly-reversible networks via linear conjugacy and reconstruction. The paper then proves a convergence result when the input tends to a constant and applies it to parallel molecular computation, with numerical examples illustrating the claims.","tokens_in":26661,"tokens_out":6310,"duration_ms":68675,"significance":"If the proofs are correct, the results form a useful generalization of the existing ISS theory for rate-controlled biochemical networks. The use of the toric locus to characterize admissible input sets is natural and avoids restrictive deficiency and linkage-class assumptions, and the extensions via linear conjugacy and reconstruction enlarge the class of networks for which robustness can be certified. The paper also gives concrete examples and simulations, and it connects ISS to a molecular-computation setting. However, the central forward-completeness proof has a gap that affects the main ISS theorems, and the same gap propagates to the non-weakly-reversible extensions, so the paper cannot be accepted in its current form.","major_comments":[{"comment":"The forward-completeness proof is not valid for the stated input class. The proof requires a single finite constant L in Eq. (3.7) bounding |du_j/dt| on the whole interval [0,t_max), but the input class defined in Section 2.3 is only piecewise locally Lipschitz. Such a function on a finite interval need not have a uniform Lipschitz constant; for example, u(t)=u^*+epsilon sin(1/(T-t)) lies in a compact U subset of T, is smooth on [0,T), yet has sup_{[0,T)} |du/dt|=infinity. In that case C3 in Eq. (3.10) need not be finite, and the Gronwall bound (3.11) cannot contradict lim_{t->t_max} W(t)=infinity. Since forward completeness is part of Definition 3.2(i), the semiglobal ISS claim of Theorem 3.7 is not established for all declared inputs. This is repairable by strengthening the input hypothesis (e.g., uniform Lipschitz or essentially bounded derivative on the maximal interval) or by an independent forward-completeness argument, but as written the proof is incomplete.","section":"Section 3.2, Theorem 3.7, Step 2, Eqs. (3.7)-(3.11)"},{"comment":"The extensions to non-weakly-reversible networks inherit the forward-completeness gap and add a second one. Their proofs invoke Proposition 3.5, which explicitly assumes that the system is R^n_{>0}-forward complete, but the proofs do not verify this property for the original system or for the reconstructed system. Lemma 4.3 establishes existence, uniqueness, and smoothness of equilibria, not forward completeness; Lemmas 4.17 and 4.18 have the same scope. Therefore, even after repairing Theorem 3.7, Theorems 4.2 and 4.14 require an explicit forward-completeness argument for the transformed dynamics before Proposition 3.5 can be applied.","section":"Section 4, Theorems 4.2 and 4.14"},{"comment":"The proof of Lemma 3.4 asserts that the function tilde_alpha_F(s) = inf_{|x-x*|>=s, x in F cap (x*+S) cap R^n_{>0}} [-nabla_x^T V0(x,u*) f(x;u*)] is continuous and nondecreasing on [0,s_max]. While nondecreasing is clear, continuity is not justified: an infimum over a moving compact set can jump upward when the argmin changes. The subsequent construction of a class K function alpha_F with alpha_F <= tilde_alpha_F/2 depends on this regularity, so the proof as written has a gap. A direct construction of a continuous lower bound would repair it, but that construction must be supplied.","section":"Section 3.2, Lemma 3.4"}],"minor_comments":[{"comment":"The phrasing 'and moreover, for all t>=0 we have x(s) in F for any s in [0,t]' is ambiguous: it reads as if the existence of beta and gamma depends on the trajectory remaining in F. The conditional form of the semiglobal ISS estimate should be stated explicitly.","section":"Definition 3.2(ii)"},{"comment":"The constants C1 and C2 should be stated as uniform over u in U. This likely follows from continuity of x*(u) and compactness of U subset of R^r_{>0}, but the uniformity should be made explicit.","section":"Theorem 3.7, Step 2, Eq. (3.8)"},{"comment":"The proof chooses a compact set F with x(t) in F for all t>=0, but permanence gives eventual bounds only. The initial transient should be absorbed into F, and this step should be stated explicitly.","section":"Section 5, Theorem 5.1"},{"comment":"The title and some headers contain typographical spacing artifacts, e.g., 'INPUT-TO-ST A TE ST ABILITY' in the header; these should be corrected in revision.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The forward-completeness gap is the main technical obstacle; it is repairable but requires nontrivial additional work. The paper's novelty is moderate: it applies toric-locus theory to ISS and extends earlier results via existing network-transformation notions. The dependence on the authors' prior work [25,26] and on [30] is significant, and the reader should ask the authors to clarify which statements are new versus imported. The paper is likely appropriate for the journal if the proofs are repaired, but I would not accept it in the present form."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nShort version: the paper does what it says. The main new content is Theorem 3.7: any weakly reversible mass-action network with input values constrained to a compact subset of its toric locus is semiglobal ISS with respect to the associated equilibria, with no restriction on deficiency or number of linkage classes. That is a genuine generalization of Chaves–Sontag, who needed zero deficiency and one linkage class. The extensions in Section 4—via linear conjugacy and via reconstruction—are also new statements, and the molecular-computation convergence result (Theorem 5.1) is a clean application that does not just restate ISS. The paper is written in standard CRN language, the examples are helpful, and the toric-locus idea is used naturally.\n\nThe main proofs are mostly coherent. The one real soft spot is Step 2 of Theorem 3.7. The Gronwall argument needs a uniform bound on |du/dt| over [0,t_max). The declared input class is \"piecewise locally Lipschitz on R_{\\ge0}\". If that means locally Lipschitz at every point of the half-line, then by compactness such a bound exists, including at t_max, and the stress-test concern about sin(1/(T-t)) does not land because that function is not locally Lipschitz at T. If the authors intend the weaker reading (locally Lipschitz on each side of each switch, with possible accumulation), then the bound is not guaranteed and the theorem is not proven for that class. Either way, the manuscript should state the intended regularity precisely; adding \"u has an essential derivative bound on finite intervals\" fixes it. This is patchable.\n\nAlso on Lemma 3.4: the reader worried that the infimum function could jump upward. I do not see that; on a compact set, the infimum of a continuous positive function as a function of the threshold is continuous by uniform continuity. That concern is not a real issue.\n\nExample 5.2 invokes Theorem 6.4 of [15] for permanence of the conjugate time-varying system. I have not checked that theorem, but if it covers only constant rates, the time-varying permanence needs explicit justification. That is a minor gap, not a structural one.\n\nBottom line: the central idea is sound and the extension is worth publishing. The proof details need cleaning up, especially the input regularity in Theorem 3.7. I would send it to peer review; a good referee will catch the same issue and the authors can patch it.","headline":"A useful extension of Chaves–Sontag ISS to arbitrary-deficiency weakly reversible networks and to some non-weakly-reversible ones; the main proof has an input-regularity loose end that is likely patchable.","tokens_in":27206,"tokens_out":5072,"would_cite":true,"duration_ms":55281,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["92C45","93D09","93D25"],"pacs":[],"model":"deepseek-v4-flash","headline":"The authors prove semiglobal input-to-state stability for weakly reversible chemical reaction networks with arbitrary deficiency and multiple linkage classes, and extend the guarantee to some non-weakly reversible networks via linear…","keywords":["chemical reaction networks","input-to-state stability","mass-action kinetics","toric locus","weak reversibility","linear conjugacy","reconstruction","parallel biomolecular computation"],"falsifier":"For a weakly reversible mass-action system with a compact input set $U$ inside its toric locus, construct a piecewise locally Lipschitz input $u(t)\\in U$ whose derivative $\\dot u(t)$ is unbounded near $t_{\\max}$ and whose state escapes to infinity at $t_{\\max}$; the theorem's hypotheses would be met while its forward-completeness conclusion would fail.","tokens_in":26096,"feed_emoji":"⚗️","tokens_out":18743,"duration_ms":162896,"temperature":0.7,"pith_summary":"Input-to-state stability (ISS) certifies that a system driven by bounded, time-varying inputs stays near its nominal trajectory, with the state deviation controlled by the input magnitude. The paper proves this property for a substantially broader class of mass-action chemical reaction networks than previously known: every weakly reversible network—one in which every reaction lies on a directed cycle—is semiglobally ISS regardless of its deficiency or number of linkage classes, provided the time-varying rate constants stay in a compact subset of the network's toric locus—the set of rate parameters for which the network is complex balanced, meaning each complex's production and consumption fluxes balance individually. It then extends the guarantee, through linear conjugacy and reconstruction, to certain networks that are not weakly reversible at all. Finally, it shows that if the input rate settles to a constant and the trajectory is permanent—uniformly bounded away from zero and infinity—then the state converges to the equilibrium belonging to that limiting input. This supplies a structural robustness certificate for parallel biomolecular computation, where coupled modules perturb one another through exactly such time-varying effective rates.","feed_headline":"Chemical reaction networks stay bounded under fluctuating rates","feed_subtitle":"Covers weakly reversible networks with any deficiency and any number of linkage classes, plus transformed non-reversible ones.","key_machinery":"The load-bearing object is the toric locus $T$ of the network, the set of rate constants at which the mass-action system is complex balanced; it supplies both the admissible input set $U\\subseteq T$ and, through the smooth map $u\\mapsto x^*(u)$, a moving target equilibrium. The proof combines this with the logarithmic free-energy Lyapunov function $V(x)=\\sum_i (x_i(\\ln x_i-\\ln x_i^*-1)+x_i^*)$: over a compact $U$, the time derivative of $V$ splits into a nonpositive dissipation term coming from complex balancing and a term proportional to $|u-u^*|$. For non-weakly reversible networks, linear conjugacy and reconstruction act as coordinate changes that carry the weakly reversible ISS property back to the original network, with reconstruction eliminating conserved variables through a conserved matrix before lifting the bound back.","core_discovery":"The central discovery, stated as Theorem 3.7, is that weak reversibility alone—without any zero-deficiency or single-linkage-class assumption—is enough for semiglobal input-to-state stability of the time-varying mass-action system (2.7), as long as the input set $U$ is a compact subset of the toric locus $T$. For any $u^*\\in U$, the system is semiglobal ISS with respect to $(x^*,u^*)$, where $x^*$ is the unique equilibrium of the frozen system at $u^*$ in the invariant stoichiometric class containing the initial state. Theorems 4.2 and 4.14 transfer this guarantee to networks that are linearly conjugate or reconstructable to weakly reversible networks, with the admissible input set described by equations (4.2) and (4.9). Theorem 5.1 then converts ISS plus permanence into convergence to the equilibrium associated with the limiting input, which is exactly what makes layer-by-layer molecular computation by parallel chemical reactions work.","pith_inferences":["The toric-locus restriction on the input set suggests a testable relaxation: allow inputs to leave the complex-balanced set $T$ for short intervals and see whether an ISS-type bound can be expressed in terms of the total time spent outside $T$; the smooth-dependence machinery would still apply between excursions.","Because linear conjugacy and reconstruction are coordinate-level transformations, the same argument could certify ISS for high-dimensional conservative networks by first reducing to the free variables and then lifting the ISS bound back through the conserved matrix.","The algebraic equations defining the admissible input set $U$ in Propositions 4.6 and 4.7 could be repurposed as a computational workflow: search for a weakly reversible realization, compute $U$, and certify ISS without simulation.","In the molecular-computation application, Theorem 5.1 reduces composability to permanence; combining it with structural permanence criteria for weakly reversible systems would give fully structural, simulation-free composability conditions."],"forward_implications":["Weak reversibility plus toric-locus-restricted inputs implies semiglobal ISS for arbitrary deficiency and linkage classes, generalizing the earlier zero-deficiency, single-linkage result.","Non-weakly reversible networks that admit a linearly conjugate or reconstructable weakly reversible partner inherit the ISS guarantee, and Proposition 4.8 gives explicit conditions under which every positive input profile is admissible.","When the input converges to a constant and the system is permanent, the state converges to the equilibrium of the limit system, so transient fluctuations do not leave a permanent offset.","Coupled molecular-computation modules can be composed in parallel: an upstream module's converging output acts as a time-varying rate input for a downstream module, and the downstream state follows the composite function of the limiting upstream signal."],"supporting_citations":[{"why":"Supplies the original semiglobal ISS framework for zero-deficiency, single-linkage weakly reversible networks that this paper generalizes.","marker":"[7]"},{"why":"Provides the Lyapunov-function-to-ISS criterion (Lemma 3.7) used to turn Lyapunov-type conditions into semiglobal ISS.","marker":"[8]"},{"why":"Introduces the toric locus and proves smooth dependence of the complex-balanced equilibrium on rate parameters, used to define the input set and bound the Lyapunov derivative.","marker":"[12]"},{"why":"Supplies complex-balanced equilibrium theory and the logarithmic free-energy Lyapunov function behind the dissipation inequality.","marker":"[24]"},{"why":"Introduces linear conjugacy, the transformation used in Theorem 4.2 to transfer ISS to non-weakly reversible networks.","marker":"[27]"},{"why":"Introduces reconstruction and reverse reconstruction, including the lemma on nonfree conserved variables, used in Theorem 4.14.","marker":"[30]"},{"why":"Sets up the dynamical-composability problem for layer-by-layer molecular computation that Theorem 5.1 addresses.","marker":"[26]"}],"fun_headline_variants":["Weak reversibility alone ensures ISS for chemical reaction networks","Stability for weakly reversible networks with any deficiency","Molecular computing stays stable under fluctuating rates","No zero deficiency needed: CRN stability under time-varying inputs","Robustness guaranteed for broader class of reaction networks"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof of forward completeness assumes that every input has a uniformly bounded derivative on any finite time interval, whereas the theorem only states that inputs are piecewise locally Lipschitz; if that derivative bound is not available, the Gronwall argument may fail to rule out finite-time blow-up.","fun_headline_variants_meta":{"raw":{"variants":["Weak reversibility alone ensures ISS for chemical reaction networks","Stability for weakly reversible networks with any deficiency","Molecular computing stays stable under fluctuating rates","No zero deficiency needed: CRN stability under time-varying inputs","Robustness guaranteed for broader class of reaction networks"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000618,"raw_usage":{"total_tokens":2860,"prompt_tokens":932,"completion_tokens":1928,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":548,"completion_tokens_details":{"reasoning_tokens":1854}},"tokens_in":548,"tokens_out":1928,"duration_ms":13722,"temperature":1.0,"reasoning_tokens":1854,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:13:51.242312+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a weakly reversible mass-action system with a compact input set $U$ inside its toric locus, construct a piecewise locally Lipschitz input $u(t)\\in U$ whose derivative $\\dot u(t)$ is unbounded near $t_{\\max}$ and whose state escapes to infinity at $t_{\\max}$; the theorem's hypotheses would be met while its forward-completeness conclusion would fail.","supporting_citations":[{"cited_title":"State-estimators for Chemical Reaction Networks of","cited_arxiv_id":null,"evidence_quote":"Supplies the original semiglobal ISS framework for zero-deficiency, single-linkage weakly reversible networks that this paper generalizes."},{"cited_title":"SIAM Journal on Control and Optimization , volume=","cited_arxiv_id":null,"evidence_quote":"Provides the Lyapunov-function-to-ISS criterion (Lemma 3.7) used to turn Lyapunov-type conditions into semiglobal ISS."},{"cited_title":"Communications Biology , volume=","cited_arxiv_id":null,"evidence_quote":"Supplies complex-balanced equilibrium theory and the logarithmic free-energy Lyapunov function behind the dissipation inequality."},{"cited_title":"Input-to-state stability-based chemical reaction networks composition for molecular computations","cited_arxiv_id":"2506.12056","evidence_quote":"Introduces linear conjugacy, the transformation used in Theorem 4.2 to transfer ISS to non-weakly reversible networks."},{"cited_title":"Journal of Mathematical Chemistry , volume=","cited_arxiv_id":null,"evidence_quote":"Introduces reconstruction and reverse reconstruction, including the lemma on nonfree conserved variables, used in Theorem 4.14."},{"cited_title":"Archive for Rational Mechanics and Analysis , volume=","cited_arxiv_id":null,"evidence_quote":"Sets up the dynamical-composability problem for layer-by-layer molecular computation that Theorem 5.1 addresses."}],"review_version":1}