{"id":"50e3e3a3-f685-4b2d-a655-dfb4c4e0207a","arxiv_id":"2608.11931","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Stochastic completeness at infinity on weighted graphs is equivalent to uniqueness of bounded pointwise solutions of the generalized porous medium equation for every bounded initial datum.","lead":"On weighted graphs, a random walk that never escapes to infinity is shown to be exactly equivalent to an unusual uniqueness property: the nonlinear filtration equation has at most one bounded solution for every bounded starting configuration. The paper also gives a mass-balance version in which the killing term's dissipation is accounted for, and shows the conditions are sharp.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Central dichotomy is conditional on companion preprint [4]: uniqueness half depends on trapping property (3.1) and comparison Lemma 3.1, neither proven here.","rationale":"I read the full manuscript in good faith and found the internal argument coherent. The uniqueness proof via the nonlinear modulus Lemma 4.2 and the Omori-Yau principle is sound given its stated assumptions; the nonuniqueness construction in Theorem 5.2 is carefully executed, including the barrier separation with correctly handled killing terms; the mass-balance theorems and the sharpness examples are consistent with the main dichotomy. The single load-bearing concern is exactly the one identified by the reader: Section 3, which carries the finite-subgraph comparison principle, existence of extremal bounded solutions, and the trapping property, is imported wholesale from the companion preprint [4] with no proofs. Both directions of the central characterization, as well as the mass-balance theorem's use of uniqueness, rest on these imported results. If any of them fails for non-locally finite graphs or arbitrary killing, the main theorem collapses. Since the paper does not reproduce or independently establish those results, a conditional verdict is appropriate. My review does not change the reader's assessment, so the verdict remains CONDITIONAL.","tokens_in":28409,"tokens_out":26802,"duration_ms":267855,"concrete_test":"Obtain the companion preprint arXiv:2607.23091 and independently verify that its Theorem 3.4, quoted here as Theorem 3.3, establishes for every phi in I, kappa>=0, and graphs not assumed locally finite: (a) existence of global bounded pointwise solutions u_A,u_B; (b) the trapping inequality (3.1) for every bounded pointwise solution v with A<=v<=B; and (c) that Lemma 3.1, the finite-subgraph parabolic comparison principle, holds without local finiteness. If any of (a)-(c) fails or requires extra hypotheses, Theorem 6.2 is not proven; if all pass, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's main theorem (Theorem 6.2) is not self-contained. In the uniqueness direction (Theorem 4.3), the proof reduces arbitrary bounded solutions v1,v2 to the extremal pair u_A,u_B via the trapping property (3.1) imported from [4, Theorem 3.4]; the subsequent modulus argument (Lemmas 4.1-4.2) then shows u_A=u_B. If the trapping property holds only for a restricted class of solutions, or if the finite-subgraph comparison Lemma 3.1 fails for graphs that are not locally finite and have kappa>0, the reduction collapses. In the nonuniqueness direction (Theorem 5.2), the finite-subgraph approximate solutions and the comparison principle again come from [4]. The paper states these results without proof and without indicating how [4] handles non-locally finite graphs or arbitrary killing. Thus the correctness of the central 'if and only if' is contingent on the companion preprint; no internal contradiction was found, but the proof is incomplete as presented.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the generalized porous medium (filtration) equation (∂t + ΔΦ)u = 0 on weighted graphs with a formal Laplacian that includes a killing term κ, where Φu = φ∘u for a continuous, increasing, nonconstant φ with φ(0)=0. The main claim (Theorem A / Theorem 6.2) is that stochastic completeness at infinity (SC∞) is equivalent to uniqueness of bounded pointwise solutions for every bounded initial datum and every admissible nonlinearity; if SC∞ fails, every bounded initial datum admits infinitely many global bounded pointwise solutions, pairwise distinct on every finite time interval (Theorem 5.2). The paper also proves a mass-balance characterization (Theorem B / Theorem 7.4): under finite total measure or the linear-growth condition (LG), SC∞ is equivalent to the generalized mass balance (MB) holding for every bounded positive solution with finite-mass datum, with sharpness examples showing that both the growth condition and boundedness are needed. A final measure-change result (Proposition 7.8) reduces SC∞ to ordinary stochastic completeness after absorbing the killing term into the vertex measure.","tokens_in":28560,"tokens_out":59123,"duration_ms":484535,"significance":"If correct, this resolves the nonlinear parabolic characterization problem for graphs raised by Schmidt–Zimmermann and extends the manifold results of Grillo–Ishige–Muratori–Punzo to weighted graphs with killing and no local-finiteness assumption. The internal proofs are detailed and careful: the modulus-of-continuity argument in Lemma 4.2, the barrier construction in Theorem 5.2, and the no-flux identity in Lemma 7.1 are all substantive and appear correct. The paper is also honest about its hypotheses and provides explicit birth–death chain examples showing that the growth and boundedness assumptions in the mass-balance theorem are sharp. There are no fitted parameters and no circular deductions; the equivalences are genuinely new statements. The principal weakness is that the paper is not self-contained: the finite-subgraph comparison principle and the existence/trapping results on which the central dichotomy rests are imported from the companion preprint [4].","major_comments":[{"comment":"The central 'if and only if' is conditional on the companion preprint [4]. Lemma 3.1 (finite-subgraph parabolic comparison) and Theorem 3.3 (existence of global bounded pointwise solutions together with the trapping property (3.1)) are stated without proof. These are load-bearing in both directions of Theorem 6.2: Theorem 4.3 uses the trapping property (3.1) to reduce arbitrary bounded solutions v1,v2 to the extremal pair uA,uB, and Theorem 5.2 uses Lemma 3.1 together with the finite-subgraph construction from [4, proof of Theorem 3.4]. Theorem 7.4 also uses Lemma 3.1 repeatedly. Since [4] is an unreviewed companion preprint, the manuscript as submitted does not provide a complete proof of its main characterization. The authors should either include full proofs of Lemma 3.1 and Theorem 3.3 in an appendix, or state precisely which conditions on the graph (non-local finiteness, arbitrary κ≥0) are verified in [4] and make the paper's acceptance conditional on the companion being published in final form.","section":"Section 3, Lemma 3.1 and Theorem 3.3"}],"minor_comments":[{"comment":"The notation 'u_0' for the minimal positive solution in Theorem 3.3 clashes with the notation for the initial datum u0; this becomes confusing in the proof of Theorem 7.4, where 'u0 = lim_n U_n' is used. Please use a different symbol, e.g. u_min or u_*.","section":"Section 3 / Section 7"},{"comment":"The inequalities in (5.7) are automatically satisfied because β_α ≤ β_α and β_α ≥ β_α; as written they are vacuous. Either remove them or state the intended nontrivial inequalities explicitly.","section":"Section 5, Step 2"},{"comment":"The proof says 'a common diagonal extraction' produces ordered limit solutions u1 ≤ u2 for the two exterior values 0 and α2. This requires a simultaneous Arzelà–Ascoli extraction for the pair of sequences; the details should be spelled out, since the two finite-subgraph solutions are ordered for each n by Lemma 3.1 but the extraction must preserve the ordering in the limit.","section":"Theorem 7.4, proof of Step 2"},{"comment":"In the finite ODE argument, Peano's theorem gives only local existence; the comparison principle is then used to show that solutions remain in a fixed bounded box and hence extend globally. This implication should be stated explicitly, because it is the comparison principle that supplies the global extension.","section":"Section 5, Step 1"}],"recommendation":"major_revision","confidential_remarks":"This is a strong paper and I have found no internal contradiction in the parts proved here. The sole serious risk is the reliance on the unreviewed companion preprint [4] for the finite-subgraph comparison principle, the existence theorem, and the trapping property. If the authors can make Section 3 self-contained—or if [4] is already accepted and available in final form—I would support acceptance. The mass-balance part and the sharpness examples are especially convincing."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a real result, the graph analogue of Grillo–Ishige–Muratori–Punzo, with killing terms and no local finiteness. The proof structure is sound: the modulus-of-continuity lemma and the weighted-time-integral argument in Lemma 4.2 are clever, the nonuniqueness construction with localized barriers looks correct, and the sharpness examples in Section 7 are a real asset. I read the main theorems carefully and found no internal contradiction and no fitted parameter.\n\nWhat is new: the parabolic characterization itself—SC_infinity iff uniqueness of bounded pointwise solutions to the GPME, for all nonlinearities and all bounded data—plus the mass-balance equivalence. The paper clearly resolves an open problem of Schmidt and Zimmermann, and the sanity check that the linear case recovers the classical characterization is in place.\n\nThe soft spot is exactly what the stress-test flags: the paper is not self-contained. Section 3 imports the finite-subgraph comparison principle (Lemma 3.1) and the existence/trapping property of extremal solutions (Theorem 3.3, especially (3.1)) from the authors' companion preprint [4], and those statements are load-bearing for both directions. Theorem 4.3 needs the trapping property to reduce arbitrary bounded solutions to the extremal pair; Theorem 5.2 needs the finite-subgraph comparison to run the exhaustion. If [4]'s results fail for non-locally-finite graphs or with arbitrary killing, the main dichotomy collapses. This is a serious conditional, but it is not a flaw in the reasoning here—it is a missing proof in the submitted package. A referee can handle it by verifying [4] or asking the authors to fold the needed statements into an appendix.\n\nMinor point: the finite ODE systems in Theorem 5.2 have continuous, not Lipschitz, right-hand sides, so Peano gives a local solution but possibly non-unique; the comparison principle then keeps everything in a box, which is fine, but worth a remark. The mass-balance theorem requires phi^{-1}(R+) nonempty; that is stated clearly.\n\nBottom line: this deserves a serious referee. I would send it to review, with the explicit instruction that the companion preprint [4] be checked (or the authors asked to include the proofs). For anyone working on graphs and nonlinear parabolic equations, this is worth reading and citing.","headline":"Genuinely new graph analogue of the nonlinear parabolic characterization of stochastic completeness; solid internal proofs, but the main theorem is conditional on the companion preprint [4] — worth refereeing.","tokens_in":29126,"tokens_out":4835,"would_cite":true,"duration_ms":50186,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35K55","35R02","05C63","60J27"],"pacs":[],"model":"deepseek-v4-flash","headline":"A weighted graph is stochastically complete at infinity exactly when every bounded initial datum for the filtration equation has a unique bounded pointwise solution, and failure of the property produces infinitely many such solutions.","keywords":["stochastic completeness at infinity","weighted graphs","generalized porous medium equation","filtration equation","uniqueness of bounded pointwise solutions","mass balance","graph Laplacian","Omori–Yau maximum principle"],"falsifier":"Take a stochastically complete at infinity birth–death chain (for instance the chain in Example 7.7(a)) and run the finite-graph exhaustion construction with two different constant exterior data for $\\varphi(s)=s|s|^{-1/2}$ and a compactly supported bounded initial datum. The theorem predicts the two subsequential limits coincide as bounded pointwise solutions on every $[0,T]$; finding two different bounded limits on some positive time interval would falsify the uniqueness half, while uniqueness of the limit corroborates it.","tokens_in":28179,"feed_emoji":"♾️","tokens_out":12295,"duration_ms":117806,"temperature":0.7,"pith_summary":"This paper proves a nonlinear version of a classical probabilistic notion for weighted graphs: heat is lost only through the killing term and never at infinity exactly when a nonlinear filtration equation has at most one bounded solution for every bounded starting profile. The equation is $(\\partial_t+\\Delta\\Phi)u=0$, where $\\Delta$ is the graph Laplacian and $\\Phi u=\\varphi\\circ u$ for a continuous, increasing, nonconstant function $\\varphi$ with $\\varphi(0)=0$; it includes the heat equation, the porous medium powers, and the fast diffusion powers. The main theorem says stochastic completeness at infinity holds if and only if every bounded initial datum admits a unique bounded pointwise solution on every finite time horizon, and when the property fails, every bounded datum admits infinitely many distinct bounded solutions. A second theorem detects the same property through a generalized mass balance: the mass remaining at time $t$ plus the mass removed by killing up to $t$ equals the initial mass. These results extend manifold parabolic characterizations to weighted graphs without assuming local finiteness.","feed_headline":"Graphs leak heat at infinity only when nonlinear solutions multiply","feed_subtitle":"Uniqueness of bounded porous-medium solutions is equivalent to no heat leaking to infinity on weighted graphs.","key_machinery":"The load-bearing mechanism is the trapping of all bounded solutions between two extremal solutions. For constants $A$ and $B$ bracketing the initial datum, the companion construction [4] supplies global bounded pointwise solutions $u_A\\le u_B$ such that every bounded solution lying between $A$ and $B$ satisfies $u_A\\le v\\le u_B$; uniqueness under stochastic completeness at infinity therefore reduces to proving $u_A=u_B$. Subtracting the equations gives $q=u_B-u_A$ and $\\rho=\\varphi(u_B)-\\varphi(u_A)$ with $\\partial_t q+\\Delta \\rho=0$, and a concave modulus of continuity $\\omega$ for $\\varphi$ on $[A,B]$ (Lemma 4.1) gives the pointwise bound $0\\le \\rho\\le \\omega(q)$. Lemma 4.2 turns any such pair into a bounded function $W$ with $-\\Delta W\\ge \\omega^{-1}(W)$; positivity of $W$ would contradict the weak Omori–Yau maximum principle, which is equivalent to stochastic completeness at infinity. In the incomplete case a normalized defect function $V$, with $\\sup_X V=1$ and $\\Delta V\\le -(V+1)-K$ for $K=\\kappa/\\mu$, builds barriers that separate, on the super-level sets of $V$, the limits obtained from finite-graph exhaustions with different constant exterior data, forcing infinitely many distinct solutions.","core_discovery":"For any weighted graph $G=(X,w,\\kappa,\\mu)$, let $\\Delta$ be the formal nonnegative Laplacian and consider the filtration equation $(\\partial_t+\\Delta\\Phi)u=0$ with $\\Phi u=\\varphi\\circ u$ for $\\varphi$ in the class $\\mathcal I$ of continuous increasing nonconstant functions vanishing at zero. The central result (Theorem 6.2) proves that $G$ satisfies stochastic completeness at infinity if and only if, for every $T>0$, every $u_0\\in\\ell^\\infty(X)$, and every $\\varphi\\in\\mathcal I$, the equation has a unique bounded pointwise solution on $[0,T]\\times X$; detection is possible with a single datum, with the zero datum alone, and by a global comparison principle. If $G$ fails the property, Theorem 5.2 constructs, for every $\\varphi\\in\\mathcal I$ and every bounded datum, infinitely many global bounded pointwise solutions that are pairwise distinct on every finite time interval. Theorem 7.4 adds an equivalent mass formulation: under finite total measure or under the sharp linear-growth condition $\\limsup_{r\\downarrow 0}\\varphi(r)/r<\\infty$, stochastic completeness at infinity is equivalent to the generalized mass balance (MB) holding for every bounded positive solution with finite-mass data; when $\\kappa=0$ the balance becomes conservation of mass.","pith_inferences":["Beyond the paper: the single-datum and zero-datum detection in Theorem 6.2 suggests a computational test on any candidate graph: integrate the filtration equation from zero initial datum and look for two bounded profiles, which would certify stochastic incompleteness at infinity numerically.","Beyond the paper: Example 7.7 shows that on infinite-measure graphs the fast-diffusion range can break the mass balance even under stochastic completeness, so the linear-growth condition marks a real phase boundary where nonlinear flux outgrows summability before the property itself is lost.","Beyond the paper: the measure-change Proposition 7.8 converts any known criterion for ordinary stochastic completeness of killing-free graphs into a criterion for stochastic completeness at infinity by replacing the vertex measure with the vertex measure plus the killing weight, which may yield new geometric sufficient conditions.","Beyond the paper: the dichotomy suggests that non-uniqueness at infinity is a generic nonlinear signature of boundary leakage, so similar characterizations may hold for other Markov generators and for nonlinearities satisfying only mild continuity and monotonicity."],"forward_implications":["For the identity nonlinearity, the characterization reduces to the classical linear statement: stochastic completeness at infinity is equivalent to uniqueness of bounded heat-equation solutions, and when the killing term vanishes this is ordinary stochastic completeness.","Uniqueness for the linear heat equation automatically implies uniqueness for every nonlinearity in the allowed class, so nonlinear effects cannot create new ambiguity once heat is known not to escape.","On graphs that are stochastically incomplete at infinity, every bounded initial datum, including zero, has infinitely many bounded solutions on every finite time interval; for zero initial data the extra solutions can be chosen of one sign.","The mass-balance theorem ties the property to an observable conservation law: on finite-measure graphs or under linear growth of the nonlinearity at zero, the total mass of a bounded positive solution plus the mass removed by killing equals the initial mass for every finite-mass datum."],"supporting_citations":[{"why":"Companion preprint supplying the finite-subgraph parabolic comparison principle, global bounded pointwise solutions, and the trapping property of extremal solutions used in both halves of the characterization.","marker":"[4]"},{"why":"General nonlinear characterization on Riemannian manifolds whose exterior-Dirichlet construction and concave modulus of continuity are adapted to graphs; the result being generalized.","marker":"[20]"},{"why":"Earlier nonlinear characterization of stochastic completeness via concave fast-diffusion filtration equations, which motivates the uniqueness half.","marker":"[19]"},{"why":"Defines stochastic completeness at infinity and proves the equivalent resolvent and bounded injectivity criteria used in Proposition 2.4.","marker":"[33]"},{"why":"Provides the weak Omori–Yau characterization, resolvent conservation identities, and birth–death chain criterion used in Lemmas 4.2, 7.1, and Example 7.7.","marker":"[34]"},{"why":"Original weak Omori–Yau maximum principle for graphs without killing, cited for the characterization used in the uniqueness argument.","marker":"[28]"},{"why":"Generalized conservation property for Schrodinger heat semigroups on manifolds, whose measure-change formulation is translated to graphs in Proposition 7.8.","marker":"[41]"}],"fun_headline_variants":["Nonlinear diffusion uniqueness marks graph infinity completeness","Heat leak at infinity is nonlinear solution nonuniqueness","Graph completeness at infinity equals unique nonlinear heat flow","Mass balance signals uniqueness of nonlinear heat solutions","Unique nonlinear solutions prove no heat escape at infinity"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof imports from the companion study [4] the guarantee that extremal bounded solutions exist, trap all other bounded solutions between them, and obey a finite-subgraph comparison principle for every nonlinearity, killing term, and graph in the stated generality; if any of those companion results fails, the characterization collapses.","fun_headline_variants_meta":{"raw":{"variants":["Nonlinear diffusion uniqueness marks graph infinity completeness","Heat leak at infinity is nonlinear solution nonuniqueness","Graph completeness at infinity equals unique nonlinear heat flow","Mass balance signals uniqueness of nonlinear heat solutions","Unique nonlinear solutions prove no heat escape at infinity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000767,"raw_usage":{"total_tokens":3503,"prompt_tokens":1152,"completion_tokens":2351,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":768,"completion_tokens_details":{"reasoning_tokens":2280}},"tokens_in":768,"tokens_out":2351,"duration_ms":17071,"temperature":1.0,"reasoning_tokens":2280,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T00:23:14.975359+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a stochastically complete at infinity birth–death chain (for instance the chain in Example 7.7(a)) and run the finite-graph exhaustion construction with two different constant exterior data for $\\varphi(s)=s|s|^{-1/2}$ and a compactly supported bounded initial datum. The theorem predicts the two subsequential limits coincide as bounded pointwise solutions on every $[0,T]$; finding two different bounded limits on some positive time interval would falsify the uniqueness half, while uniqueness of the limit corroborates it.","supporting_citations":[{"cited_title":"The generalized porous medium equation on graphs: well-posedness, extinction, and mass conservation","cited_arxiv_id":"2607.23091","evidence_quote":"Companion preprint supplying the finite-subgraph parabolic comparison principle, global bounded pointwise solutions, and the trapping property of extremal solutions used in both halves of the characterization."},{"cited_title":"Nonlinear characterizations of stochastic completeness","cited_arxiv_id":null,"evidence_quote":"Earlier nonlinear characterization of stochastic completeness via concave fast-diffusion filtration equations, which motivates the uniqueness half."},{"cited_title":"Stochastic incompleteness for graphs and weak Omori–Yau maximum principle","cited_arxiv_id":null,"evidence_quote":"Original weak Omori–Yau maximum principle for graphs without killing, cited for the characterization used in the uniqueness argument."},{"cited_title":"A generalized conservation property for the heat semigroup on weighted manifolds","cited_arxiv_id":null,"evidence_quote":"Generalized conservation property for Schrodinger heat semigroups on manifolds, whose measure-change formulation is translated to graphs in Proposition 7.8."}],"review_version":1}