{"id":"6fca594c-595e-471b-ae4f-1ab82d39e77a","arxiv_id":"2411.13419","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"In a one-dimensional forest fire model with random delays, almost surely some fire never stops (an 'infinite fire'), with a phase transition at delays decaying like c/x with c=1.","lead":"This paper studies a forest fire model on a line where fires spread with random delays, and proves that if delays are sometimes positive, almost surely some fire will burn infinitely many trees. It also finds a sharp boundary between delays that allow infinite fires and those that do not.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the induction gap in Theorem 2.1 is closed by a simple renewal argument, and the conditional Borel-Cantelli step admits a standard repair.","rationale":"I examined the central claim, Theorem 2.1, and the proof in Section 2. The induction step that the reader flagged is repairable in a few lines: once a tree appears at x+1, it remains occupied until a fire from x succeeds; since the attempt times f_{x,i}+\\Delta_{x,i} go to infinity, the first attempt after the appearance time necessarily finds an occupied, non-burning tree and ignites it. The same argument applies after every burn, so x+1 is burnt infinitely often. This supplies the missing renewal argument without changing the proof's structure. The lower bound in (2.1) is also sound because sites beyond m_k have never been burnt, so their tree processes are independent and the product lower bound follows from Lemma 2.2. The only genuine technical flaw is the invocation of conditional Borel-Cantelli for the tail events A^c_{k+1}, which are not measurable with respect to the stopped sigma-algebras F_{k+1}. However, a standard product-of-conditional-probabilities argument over the events E_k={no success among the first k record fires} yields the same conclusion, since the conditional probability of success at each step is at least \\epsilon_0\\delta_2. Thus the paper's central claim is very likely correct; the written proof needs minor amendments but no substantive reworking. The reader's CONDITIONAL verdict is therefore unchanged by this stress-test.","tokens_in":16056,"tokens_out":41886,"duration_ms":459772,"concrete_test":"Re-derive Section 2 with the renewal argument made explicit: in the induction, define A_j and T_i=f_{x,i}+\\Delta_{x,i}, prove that the first T_i exceeding A_j occurs while x+1 is occupied and non-burning, and verify that the product estimate P(E_{k+1}|F_k)\\le(1-\\epsilon_0\\delta_2)1_{E_k} replaces the cited conditional Borel-Cantelli step. If both checks go through, Theorem 2.1 is fully proved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The reader's stated concern about Theorem 2.1 does not land as a fatal gap. For the induction step, let A_j be the j-th time a tree appears at site x+1 and let T_i=f_{x,i}+\\Delta_{x,i}. Since the induction hypothesis gives f_{x,i}\\to\\infty and \\Delta_{x,i} are finite a.s., T_i\\to\\infty; the first T_i after A_j finds x+1 occupied and not burning, so it ignites x+1. Repeating after each burn and subsequent exponential appearance time yields infinitely many fires at x+1. The conditional Borel-Cantelli citation is formally imperfect because A^c_{k+1} is a tail event, not F_{k+1}-measurable; however, defining E_k as the event that the first k record fires are all finite gives P(E_{k+1}|F_k)\\le(1-\\epsilon_0\\delta_2)1_{E_k}, hence P(E_k)\\le(1-\\epsilon_0\\delta_2)^{k-1}\\to 0. Thus the central claim is supported modulo minor expository additions.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a generalization of the one-dimensional forest fire model with ignition at the origin, introduced in [15], by allowing non-zero burning times θ and random spread delays Δ. The main results are: (i) if P(Δ>0)>0, then an infinite fire occurs almost surely (Theorem 2.1); (ii) for θ≡0 with i.i.d. delays of positive mean and positive variance, there is exactly one infinite fire (Theorem 3.1); (iii) for location-dependent delays with asymptotic order c/x, there is a threshold at c=1 separating a.s. absence and a.s. occurrence of infinite fire (Theorem 3.6); (iv) if Δ≡0, no infinite fire occurs (Theorem 4.1), and the record maxima satisfy a doubly exponential lower bound (Theorem 4.7). A conditional O(log n) bound for the first burning time is proved under Conjecture 4.4 (Theorem 4.5).","tokens_in":16268,"tokens_out":18472,"duration_ms":176184,"significance":"The paper establishes a robust new phenomenon for a natural one-dimensional forest-fire model: random spread delays can cause a fire to propagate to infinity, a behavior that is impossible in the zero-delay models of [9] and [15]. The threshold result in Section 3.2 is particularly interesting, as it connects the tail behavior of the delay distribution to a percolation-type exponent. The arguments are self-contained and use standard tools (renewal, conditional Borel-Cantelli, random walk recurrence, Hoeffding's inequality). The main unconditional theorems appear correct in outline and are potentially useful for further work on self-organized critical systems. The paper is honest about the dependence of Theorem 4.5 on Conjecture 4.4, though this dependence should be made more prominent in the presentation.","major_comments":[{"comment":"The induction step asserting that a new tree at site x+1 will be burnt by a fire from x is stated without proof; the text only notes that a tree appears eventually and that f_{x,i}→∞. Because a tree may be burnt before the next fire from x attempts to spread, one needs a renewal argument: after each burn at x+1, the appearance time of the next tree is independent exponential, and the first subsequent spread attempt from x occurs after this appearance with probability one. Please supply this argument explicitly. In addition, the invocation of the conditional Borel-Cantelli lemma should be replaced by a short proof (e.g., bounding P(E_k) by (1−ε0δ2)^{k−1}) or a precise statement of the lemma.","section":"Section 2, proof of Theorem 2.1"},{"comment":"These results are conditional on Conjecture 4.4, which is not proved. The manuscript labels 4.5 as 'Theorem' without making the conditional nature prominent in the section or abstract. If the conjecture remains open, the authors should relabel the statement as a conditional proposition and clearly state in the introduction that the only conditional result of the paper is Theorem 4.5 and Lemma 4.6(b).","section":"Section 4, Theorem 4.5 and Lemma 4.6(b)"},{"comment":"The equality P(X|Y,D_{n,i}) = P(X|Y) is asserted without justification. It is true because X is determined by the first i−1 fires, which are independent of the i-th spread times and of the tree appearances after the i-th fire starts, but this independence should be spelled out. Please add a sentence explaining why X is jointly independent of (Y,D_{n,i}).","section":"Section 3.2, proof of (3.4)"}],"minor_comments":[{"comment":"Conjecture 4.4 contains a typo: 'such f_{m_i,2}' should read 'such that f_{m_i,2}'.","section":"Section 4, Conjecture 4.4"},{"comment":"The same symbol F is used for the sigma-algebra F_k in Section 2 and for the event F_n in Lemma 4.3; please use different notation (e.g., \\mathcal{F}_k and E_n) to avoid ambiguity.","section":"General notation"},{"comment":"The chain of inequalities leading to the definition of δ2 is hard to follow; consider presenting the conditioning argument in a separate paragraph to improve readability.","section":"Lemma 2.2"},{"comment":"The condition on Δ_x is rendered as 'P(lim |Δ_x|?x=0)=1' in the text; please make the mathematical expression unambiguous (presumably P(lim |Δ_x|√x=0)=1).","section":"Theorem 3.6(d)"},{"comment":"The abstract refers to arXiv:0907.1821; for consistency with the body, cite it as [15] instead.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid contribution to the probability literature on forest-fire models. The main unconditional theorems appear correct, and the conditional result is clearly marked, though it should not be presented as a proved theorem. The induction and conditional Borel-Cantelli gaps in Theorem 2.1 are readily repairable, and the independence issue in (3.4) is also fixable with a short clarification. I recommend major revision rather than rejection; I do not see a need for additional data or experiments."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take on arXiv:2411.13419. The paper adds delays in fire spread and nonzero burn times to the Z+ forest fire model with ignition at 0, and proves a new phenomenon: if delays are sometimes positive, an infinite fire occurs a.s., which is impossible in the zero-delay models of Comets–Menshikov–Volkov and Volkov. The cleanest result is the phase transition in Section 3.2: for spread times with mean about c/x, the infinite fire exists iff c>1. That is new and worth knowing.\n\nThe paper does a lot of things right. The graphical representation is helpful. The constant-delay coupling with Volkov's model is elegant, and the proof that there is at most one infinite fire in the i.i.d. positive-mean delay case uses a zero-mean random walk recurrence argument that is correct and short. Section 4 has a nice double-exponential lower bound on successive maxima that does not depend on the conjecture.\n\nThe soft spots are all in the proof details, not the main ideas. Theorem 2.1's induction claims that fires reach every site infinitely often; the written justification is too quick and needs a renewal argument after each burn at x+1. The repair is standard and I don't doubt the statement. In the proof of Theorem 3.6(d), the step labeled (3.4) is not justified by the argument given: conditioning on the first fire to reach n does not make the earlier-fires event independent of the current delays in the way the text asserts. The inequality itself looks true (the event that the i-th fire reaches n is increasing in the delays, while the 'small sum' event is decreasing), but it needs a clean monotonicity argument. Finally, Theorem 4.5 is explicitly conditional on Conjecture 4.4; that is clearly stated, so it is a limitation, not an error.\n\nNone of these look fatal. The central claims are supported by the sketches, and the gaps are of the kind a referee can ask to be filled without changing the results. The citation pattern is clean; the only self-citation is to the base model in [15], which is appropriate. This is a genuine advance for the forest fire / SOC subfield, not a breakthrough for all of probability.\n\nI'd send it out for review. For a reading group, it is a good example of how a small model change creates a real phase transition, and it has several instructive Borel–Cantelli arguments.","headline":"A genuinely new infinite-fire phase transition in a natural generalization of the one-sided forest fire model; proofs need tightening but the claims are credible.","tokens_in":16787,"tokens_out":9387,"would_cite":true,"duration_ms":92359,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60G55","60K35","60F20"],"pacs":[],"model":"deepseek-v4-flash","headline":"Any positive delay in fire spread makes an infinite forest fire inevitable almost surely.","keywords":["forest fire model","infinite fire","delayed fire spread","burning time","self-organized criticality","phase transition","Borel-Cantelli","random walk"],"falsifier":"For a fixed delay law with $P(\\Delta>0)>0$, simulate the process and measure the conditional probability, given the history up to $T_k$, that the next fire reaching beyond $m_k$ continues forever; Theorem 2.1 predicts this probability is never below $\\epsilon_0 \\delta_2 > 0$. If simulations with large $k$ show it decaying to zero, the uniform lower bound is false.","tokens_in":15812,"feed_emoji":"🔥","tokens_out":6211,"duration_ms":66408,"temperature":0.7,"pith_summary":"The paper studies a one-dimensional forest fire model in which fires start only at the origin, burn trees for a positive random time, and take a positive random time to jump from one tree to the next. Its central claim is that if the spread delay has any chance of being positive, then almost surely some fire will burn infinitely many trees, an 'infinite fire' that cannot occur in the classic zero-delay model. The authors see this as a genuinely new phenomenon introduced by delay, and they go on to map when it happens: with delayed spread an infinite fire is the rule rather than the exception, while with instantaneous spread it can never happen.","feed_headline":"Any positive fire-spread delay makes an infinite fire inevitable","feed_subtitle":"Proof shows any non-zero spread time makes some blaze burn forever, a phenomenon impossible with instant spread.","key_machinery":"The central machinery is the sequence of successive maxima $m_1 < m_2 < \\cdots$ of the rightmost burnt site, together with the $\\sigma$-algebras $F_k$ generated by the process up to the time $T_k$ when site $m_k$ stops burning. Lemma 2.2 gives a uniform lower bound on the probability that cumulative delays grow at least linearly; combined with the product in (2.1) this yields a strictly positive conditional probability $\\epsilon_0 \\delta_2$ that the next fire is infinite, which the conditional Borel-Cantelli lemma turns into an almost-sure event.","core_discovery":"Theorem 2.1 asserts that whenever $P(\\Delta > 0) > 0$, with probability one there is a fire that never stops: $P(\\inf\\{k : n_k = \\infty\\} < \\infty) = 1$. The proof tracks the successive rightmost points $m_k$ reached by fires and shows, via a conditional Borel-Cantelli argument, that from any such maximum there is a uniformly positive chance that the next fire burns forever. The same framework yields a uniqueness result when burning is instantaneous and delays have positive finite variance (exactly one infinite fire), a phase transition at average delay $c/x$ with threshold $c=1$, and a no-infinite-fire theorem when spread is instantaneous.","pith_inferences":["The paper leaves open the exact law of the time $T$ when the infinite fire begins; a natural next step is to estimate its tail as a function of the delay distribution.","Because Theorem 2.1 needs only $P(\\Delta>0)>0$, heavy-tailed delays should also produce infinite fires; quantifying how the waiting time scales with the delay tail would test the robustness of the mechanism.","The constant-delay coupling suggests that after the infinite fire, the distribution of fire sizes is governed by the known zero-delay model, which could give explicit asymptotics for the number of trees burnt by later fires.","In the instant-spread setting, Conjecture 4.4 would imply $f_{n,1}=O(\\log n)$; checking numerically whether $f_{n,1}/\\log n$ stays bounded for large $n$ would provide evidence for the conjecture and for Theorem 4.5."],"forward_implications":["If the main theorem holds, the model exhibits an almost-sure infinite fire under any non-degenerate delay distribution, a phenomenon absent from the zero-delay models of [9] and [15].","With instantaneous burning and non-degenerate finite-variance delays, the infinite fire is unique: no later fire can reach infinity again.","With a fixed positive delay $a>0$, the process after the infinite fire starts is exactly the 2009 zero-delay forest fire model, so the infinite fire acts as a reset point after which known finite-fire statistics apply.","For location-dependent delays with mean about $c/x$, there is a sharp threshold: $c \\le 1$ suppresses infinite fires, while $c > 1$ makes one almost sure.","When spread is instantaneous, no infinite fire occurs even with positive burn times, and first-passage times grow logarithmically in the distance."],"supporting_citations":[{"why":"Defines the original ignition-at-zero forest fire model that the delayed process generalizes and with which the constant-delay process couples.","marker":"[15]"},{"why":"Supplies the previous generalization of the forest-fire model in which no infinite fire occurs, setting the contrast for Theorem 2.1.","marker":"[9]"},{"why":"Introduces the signal-recovery/forest-fire framework from which the model descends.","marker":"[3]"},{"why":"Provides the recurrence of zero-mean one-dimensional random walks used to prove uniqueness of the infinite fire in Theorem 3.1.","marker":"[14]"}],"fun_headline_variants":["Any positive fire-spread delay guarantees an infinite fire","Nonzero spread time forces some blaze to burn forever","Delays make infinite fires inevitable: new proof","With any fire delay, an endless blaze occurs almost surely","Infinite fire becomes certain once spread takes any time"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The induction step of Theorem 2.1 assumes that at each site, the infinitely many delayed attempts to ignite the next site will eventually occur while that next site has a tree, even though earlier fires may have burnt it; this renewal property is asserted rather than proved.","fun_headline_variants_meta":{"raw":{"variants":["Any positive fire-spread delay guarantees an infinite fire","Nonzero spread time forces some blaze to burn forever","Delays make infinite fires inevitable: new proof","With any fire delay, an endless blaze occurs almost surely","Infinite fire becomes certain once spread takes any time"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00049,"raw_usage":{"total_tokens":2314,"prompt_tokens":751,"completion_tokens":1563,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":367,"completion_tokens_details":{"reasoning_tokens":1487}},"tokens_in":367,"tokens_out":1563,"duration_ms":13407,"temperature":1.0,"reasoning_tokens":1487,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T16:29:05.673840+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a fixed delay law with $P(\\Delta>0)>0$, simulate the process and measure the conditional probability, given the history up to $T_k$, that the next fire reaching beyond $m_k$ continues forever; Theorem 2.1 predicts this probability is never below $\\epsilon_0 \\delta_2 > 0$. If simulations with large $k$ show it decaying to zero, the uniform lower bound is false.","supporting_citations":[{"cited_title":"Forest fires onZ` with ignition only at0","cited_arxiv_id":null,"evidence_quote":"Defines the original ignition-at-zero forest fire model that the delayed process generalizes and with which the constant-delay process couples."},{"cited_title":"Generalizations of forest fires with ignition at the origin","cited_arxiv_id":null,"evidence_quote":"Supplies the previous generalization of the forest-fire model in which no infinite fire occurs, setting the contrast for Theorem 2.1."},{"cited_title":"A signal-recovery system: asymptotic properties, and construction of an infinite-volume process","cited_arxiv_id":null,"evidence_quote":"Introduces the signal-recovery/forest-fire framework from which the model descends."},{"cited_title":"Principles of random walk","cited_arxiv_id":null,"evidence_quote":"Provides the recurrence of zero-mean one-dimensional random walks used to prove uniqueness of the infinite fire in Theorem 3.1."}],"review_version":1}