{"id":"02cd9bd5-8074-4f8b-a700-39a5eee8adfc","arxiv_id":"1908.03426","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Interfaces, a species of lattice animals tied to percolation cluster boundaries, satisfy a duality br=(b1/r)^r on triangulated lattices and a strict growth-rate inequality b<a, connecting interface counts to percolation thresholds.","lead":"This paper proves new bounds relating the exponential growth rates of lattice animals and interfaces to percolation thresholds on lattices. It derives a duality formula for interfaces on triangulated lattices and a strict inequality showing interfaces grow strictly slower than general animals.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Inequality (2) depends on a false non-decay claim for Pp(|So|=n) on (1-pc, pc): for p<pc, Aizenman-Barsky gives exponential decay, and |So|<=|Co|.","rationale":"The paper contains substantial correct-looking material, especially Theorem 1.2 and the duality theorems, but the headline inequality (2) is not proved. The figure caption's reliance on [23] appears to be a misreading: [23] treats supercritical large finite clusters by volume, not interface length in the subcritical window. Because the interface is contained in the occupied cluster, exponential decay of the volume distribution transfers to exponential decay of the interface-size distribution. Hence the asserted failure of exponential decay on (1−pc, pc) is impossible for pc>1/2. Since (2) is the basis for the abstract's claimed translation formula and for the lower bounds on a(Z^d) in the companion paper, this is a load-bearing gap. The reader's weakest assumption about imported interface uniqueness is plausible but less directly falsifiable; our concern is checkable against [23] and elementary percolation bounds.","tokens_in":30356,"tokens_out":43815,"duration_ms":422087,"concrete_test":"Verify the actual statement of Kesten–Zhang [23]; I expect its main theorem bounds Pp(|Co|≥n) for p>pc by exp(-c n^{(d-1)/d}), which does not imply failure of exponential decay of Pp(|So|=n) on [1−pc, pc]. Also test the square lattice at p=0.5 (pc≈0.593): Aizenman–Barsky gives Pp(|Co|=n)≤e^{-cn}, so Pp(|So|=n)≤e^{-cn}, contradicting the caption. This settles whether the cited support for inequality (2) is valid.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's only explicit justification for inequality (2) is the Figure 1 caption, which asserts that Kesten–Zhang [23] proved failure of exponential decay of Pp(|So|=n) for p in [1−pc, pc]. This is inconsistent with Aizenman–Barsky: for any p strictly between 1−pc and pc with pc>1/2, p<pc, so Pp(|Co|=n) decays exponentially; since the interface is a subgraph of the cluster (Theorem 3.5 gives P⊂E(C)), Pp(|So|=n)≤Pp(|Co|≥n)≤e^{-cn}. Thus the interval claim is false. Inequality (2), b(G)≥f(r(pc)), requires b_{r(pc)}=f(r(pc)), which the paper derives solely from that false interval claim via Theorem 1.2. No alternative proof of failure at pc is supplied. A correct proof may exist via χ(pc)=∞, but it is not given. If (2) is the advertised dictionary formula, the central claim is unsupported as written.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a percolation-based method for bounding the exponential growth rates of lattice animals and of a subclass called interfaces. It introduces growth rates b_r(G) of interfaces with a fixed surface-to-volume ratio r, proves the universal upper bound b_r(G) ≤ f(r) with f(r)=(1+r)^{1+r}/r^r, and proves an equivalence (Theorem 1.2) between exponential decay of the interface size distribution at parameter p and the strictness of the inequality b_{r(p)}<f(r(p)). It then derives duality relations b_r=(b_{1/r})^r for triangulated lattices, continuity and log-concavity of b_r, an interface analogue of the Cheeger constant giving new upper bounds on p_c, a strict inequality between interface and animal growth rates, and an appendix result on continuity of the cluster-size decay exponent. The abstract and introduction also claim improved asymptotic bounds on a(Z^d) as d→∞, but Section 1.2 states that those bounds are obtained in the follow-up paper [15].","tokens_in":30579,"tokens_out":7968,"duration_ms":92009,"significance":"If the main dictionary formula (2) were established, the paper would provide a genuinely new two-way bridge between percolation thresholds and enumerative growth constants, with the universal function f(r) playing a parameter-free role. The duality formula b_r=(b_{1/r})^r, the continuity and log-concavity results for b_r, and the strict comparison b(G)<a(G) are interesting and potentially influential. The paper is also commendably explicit about the dependence on imported results, especially the unique-interface theorem from [14]. However, the central inequality (2) is currently supported by a claimed interval of non-decay that is inconsistent with Aizenman–Barsky, so the main advertised conclusion is not established as written.","major_comments":[{"comment":"The derivation of inequality (2), b(G) ≥ f(r(p_c(G))), rests on the claim, made only in the caption of Figure 1, that exponential decay of P_p(|S_o|=n) fails for all p in [1-p_c, p_c] and that this follows from Kesten–Zhang [23]. This interval claim is false when p_c>1/2. Indeed, by Theorem 3.5 the interface is a subgraph of the cluster, P⊂E(C), so |S_o|≤|C_o|. For p∈(1-p_c,p_c) with p<p_c, Aizenman–Barsky [1] gives exponential decay of P_p(|C_o|=n), hence P_p(|S_o|=n)≤P_p(|C_o|≥n) decays exponentially. By the 'only if' direction of Theorem 1.2 this forces b_{r(p)}(G)<f(r(p)) on that interval, contradicting the plateau claimed in Figure 1. Since the equality b_{r(p_c)}=f(r(p_c)) is needed for (2), and no alternative proof of non-decay at p=p_c is supplied in the paper, the central formula (2) is unsupported as written. The authors should either prove non-decay at p_c directly or explicitly restrict and re-derive all claims that depend on it.","section":"§4, Figure 1 caption; Theorem 1.2; Eq. (2)"},{"comment":"The abstract states that the paper improves the best known asymptotic bounds on a(Z^d) as d→∞, but Section 1.2 says these bounds are obtained in the follow-up paper [15] and the present paper only sets up the machinery. This is a mismatch between the advertised contribution and the actual content. The abstract and introduction should be rewritten so that the claims made for this paper are limited to the results proved here, with the asymptotic bounds clearly attributed to [15].","section":"Abstract and §1.2"},{"comment":"Theorem 1.1 is stated as b(G)<a(G) for every G∈S, but the proof in Section 8 is carried out only for site-interfaces and site-animals, giving ˙b(G)<˙a(G) under additional assumptions (5) and (14). The manuscript does not explain how the site result implies the bond-interface statement b(G)<a(G), nor does it define clearly whether Theorem 1.1 is meant in the bond or site sense. This leaves a gap between the theorem as stated and the proof as written; the statement and proof should be reconciled, or the theorem should be reformulated for the site objects actually treated.","section":"§8, Theorem 8.1 vs §1.3, Theorem 1.1"}],"minor_comments":[{"comment":"There are numerous typographical and grammatical slips, including 'in this vain' for 'in this vein', missing spaces such as 'LetSo', and inconsistent use of superscripts in the definitions of b°_r and b⊙_r. A careful proofreading pass is needed.","section":"Throughout"},{"comment":"The notation b°_r and b⊙_r is introduced with a degree symbol and a circled dot, but the superscripts are dropped immediately afterward; the reader must infer which variant is meant. Please keep the notation explicit for at least the statements of Lemmas 4.2 and Proposition 4.6.","section":"§4, Definition 4.1"},{"comment":"The proof invokes the fact that θ is not analytic at p_c and refers to [14, Corollary 4.14] for analyticity of the inclusion-exclusion expansion, but it does not give a reference or argument for non-analyticity of θ at p_c. A citation or a short explanation would help.","section":"§7, Theorem 7.1"},{"comment":"The paper assumes without comment that quasi-transitive planar lattices are 2-connected after a vertex-deletion operation, and it states that this operation preserves p_c. This is plausible but should be justified or referenced, since the class S is used throughout.","section":"§2.4 and §3"}],"recommendation":"major_revision","confidential_remarks":"The central issue is the Kesten–Zhang attribution in the caption of Figure 1. Before resubmission, the authors should verify the original theorem and provide a self-contained argument for non-decay at p=p_c if that is what the dictionary formula requires. I would also ask the editor to confirm whether the paper under review is intended as the first part of a two-paper series; if so, the abstract's claims about asymptotic bounds should be adjusted accordingly."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth your time if you work on percolation thresholds or enumeration of lattice animals, but treat inequality (2) with suspicion. The paper has three genuinely new pieces: the duality b_r = (b_{1/r})^r for triangulated lattices, the strict inequality b(G) < a(G) between interface and animal growth rates, and an interface Cheeger constant I(G) giving p_c <= 1/(I(G)+1), sometimes beating the Benjamini–Schramm bound. Those arguments look coherent and do not rely on the questionable part.\n\nThe soft spot is exactly the central formula (2), the promised translation between upper bounds on p_c and lower bounds on a(G). The only justification in the text is the caption of Figure 1, which attributes to Kesten–Zhang [23] the claim that P_p(|S_o|=n) fails to decay exponentially for p in [1-p_c, p_c]. That is false when p_c > 1/2: for p in (1-p_c, p_c) with p < p_c, Aizenman–Barsky gives exponential decay for the cluster size, and because the interface is a subgraph of the cluster (|S_o| <= |C_o|), the interface size distribution decays exponentially too. Kesten–Zhang's paper is about supercritical percolation and does not prove failure at p_c. No alternative proof of non-exponential decay at p_c is supplied. That failure is exactly what would turn Proposition 4.4/4.5 into equality at r(p_c) and yield (2). So as written the dictionary formula is unsupported. It may well be repairable—non-exponential decay at criticality is likely provable for the graphs in S via divergence of susceptibility—but that work is not here.\n\nAlso, the abstract overreaches: it says the paper improves the best known asymptotic bounds on a(Z^d), while Section 1.2 correctly defers those improvements to the follow-up [15]. That is a real mismatch between abstract and content.\n\nWhat does hold up? The continuity of the decay exponent c(p), the log-concavity and continuity of b_r, the duality theorems, and the strict inequalities b < a and p_c < 1/(h+1). The proof of Lemma 5.5 is intricate; I did not find a clear hole, but it deserves close referee scrutiny. The dependence on the authors' own interface machinery from [14] is heavy but legitimate—Theorem 3.5 is restated from that work and is not obviously wrong.\n\nRecommendation: send to a serious referee, but flag the status of (2) and the Kesten–Zhang citation explicitly. If the critical-point non-decay can be supplied, this is a strong paper; as it stands, the advertised dictionary formula is not proven.","headline":"Genuinely new results on interface growth rates, but the advertised 'dictionary' inequality (2) is not proven as written—it depends on a false claim about exponential decay at pc.","tokens_in":31096,"tokens_out":8764,"would_cite":true,"duration_ms":90660,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K35","05A16","82B43"],"pacs":[],"model":"deepseek-v4-flash","headline":"A universal inequality turns percolation thresholds into lower bounds on cluster growth rates.","keywords":["percolation threshold","lattice animals","interfaces","exponential growth rates","Bernoulli percolation","cluster size distribution","duality","Cheeger constant"],"falsifier":"Enumerate occurring interfaces in one percolation instance on the square lattice and look for two distinct outermost boundary layers that meet at a vertex; the paper's Lemma 3.2 forbids this, and finding one would falsify the uniqueness theorem on which the counting estimate rests. Alternatively, search numerically for an instance with more than a constant times $n$ occurring interfaces of size $n$ — the bound $N_n\\le l n+1$ would fail, and the chain from counting to Theorem 1.2 would need to be re-examined.","tokens_in":30166,"feed_emoji":"📈","tokens_out":13699,"duration_ms":138142,"temperature":0.7,"pith_summary":"Lattice animals are the connected finite subgraphs that statistical physics counts; interfaces are a thin sub-species, the outermost layer separating a cluster from infinity. This paper proves that these two counting problems are quantitatively tied to Bernoulli percolation through one inequality: $b(G)\\ge f(r(p_c(G)))$ with $f(r)=(1+r)^{1+r}/r^r$ and $r(p)=(1-p)/p$, where $b(G)$ is the exponential growth rate of interfaces. The matching Theorem 1.2 says that the interface-size distribution $P_p(|S_o|=n)$ fails to decay exponentially exactly when $b_{r(p)}(G)=f(r(p))$, giving a precise boundary between exponential and non-exponential decay. Since lattice animals are the coarsest case of the same interface construction, the same inequality recovers and refines the known animal-counting results, and for triangle-generated lattices it yields the duality $b_r=(b_{1/r})^r$. If correct, any improved upper bound on a percolation threshold automatically improves a lower bound on animal growth rates, and vice versa.","feed_headline":"Formula turns percolation thresholds into growth-rate bounds","feed_subtitle":"The same inequality works both ways, so better percolation bounds improve animal counts and vice versa.","key_machinery":"The load-bearing object is $b_r(G)$, the exponential growth rate of interfaces of size $n$ with boundary size close to $rn$, together with the universal comparison function $f(r)=\\frac{(1+r)^{1+r}}{r^r}$. Interfaces are defined from a chosen basis of the cycle space; at the coarse extreme they are all lattice animals, and at the fine extreme they are thin layers around clusters. The engine of the proof is the first-moment estimate $p(1-p)^r\\le 1/b_r(G)$, obtained by counting occurring interfaces through the unique-interface theorem and bounding this count by a quasi-geodesic argument, together with a large-deviation lemma that most occurring interfaces have surface-to-volume ratio near $r(p)=(1-p)/p$. These two ingredients produce the threshold theorem, and the triangle-basis duality comes from a box-gluing construction that shows interfaces with fractal shapes do not dominate the count $b_r$.","core_discovery":"The central discovery is a two-way dictionary between percolation and enumeration. For every graph in the paper's class $\\mathcal{S}$, the exponential growth rate $b_r(G)$ of interfaces with surface-to-volume ratio $r$ satisfies $b_{r(p)}(G)\\le f(r(p))$ for every $p\\in(0,1)$, and equality holds precisely when $P_p(|S_o|=n)$ does not decay exponentially. This equivalence is what converts an upper bound on the percolation threshold $p_c(G)$ into a lower bound on $b(G)$, hence on the lattice-animal growth rate $a(G)$, and conversely. The paper further proves that for any basis of the cycle space made of bounded cycles, interfaces are exponentially rarer than lattice animals, so the resulting inequality is strict: $a(G)>f(r(p_c(G)))$. It also establishes, for lattices whose cycle space has a triangle basis, the self-duality $b_r=(b_{1/r})^r$ of the growth-rate function, and, as a by-product, the continuity of the exponential decay rate of the cluster size distribution on $(0,1)$.","pith_inferences":["A testable extension is to check the dictionary on all 1-ended vertex-transitive graphs; the paper's own proof suggests the large-deviation lemma is the part that would need to be rebuilt in that generality.","The interface constant $I(G)$ opens a route to threshold bounds on amenable graphs, where the classical Cheeger constant vanishes but $I(G)$ can remain positive; the paper proves the one-sided bound $p_c\\le 1/(I(G)+1)$ and does not explore whether that bound can be sharp.","The continuity of the decay exponent $c(p)$ suggests that other percolation observables near criticality, such as finite-size scaling exponents, might inherit Lipschitz regularity; the paper proves continuity only for $c(p)$ and its interface analogue."],"forward_implications":["Any upper bound on the percolation threshold of a lattice in $\\mathcal{S}$ automatically becomes a lower bound on the exponential growth rate of its lattice animals, and any such lower bound becomes a threshold upper bound.","For every interface basis made of bounded cycles, $b(G)<a(G)$, so the translated inequality is strict: $a(G)>f(r(p_c(G)))$.","The interface-size distribution $P_p(|S_o|=n)$ decays exponentially exactly when $b_{r(p)}(G)<f(r(p))$; in particular it also fails to decay at $p=1-p_c$ for the triangle-generated lattices.","On triangle-generated lattices, $b_r=(b_{1/r})^r$ for every $r>0$, so computing the branch $r<1$ determines the branch $r>1$ and vice versa.","The exponential decay rate of the cluster size distribution, $c(p)=\\lim_n P_p(|C_o|=n)^{1/n}$, is a continuous function of $p$ on $(0,1)$."],"supporting_citations":[{"why":"Supplies the unique-interface theorem (Theorem 3.5) and the definition of interfaces that the counting arguments rely on.","marker":"[14]"},{"why":"Proved the animal analogue of the threshold theorem that the interface version extends and compares against.","marker":"[17]"},{"why":"Derivation of the animal exponential-decay threshold that the animal version builds on.","marker":"[10]"},{"why":"Gives the Cheeger-constant bound $p_c\\le 1/(h+1)$ that the paper strengthens to a strict inequality.","marker":"[8]"},{"why":"Locates the interval where interface probabilities fail to decay exponentially, used with Theorem 1.2 to identify where $b_r=f(r)$.","marker":"[23]"},{"why":"Sharpness of the phase transition, giving exponential decay of cluster sizes below $p_c$.","marker":"[1]"},{"why":"Pattern-theorem method adapted to show that interfaces are exponentially rarer than lattice animals.","marker":"[22]"}],"fun_headline_variants":["Percolation thresholds improve lattice animal growth bounds","Two-way formula links percolation and animal growth","Convert percolation thresholds into lattice animal bounds","From percolation thresholds to growth-rate bounds"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing assumption is geometric — every finite cluster separating the origin from infinity has exactly one outermost boundary layer, and at most a linear number of such layers of size $n$ can touch a fixed path from the origin — and if either half fails, the first-moment estimate $p(1-p)^r\\le 1/b_r$ collapses with the whole dictionary.","fun_headline_variants_meta":{"raw":{"variants":["Percolation thresholds improve lattice animal growth bounds","Two-way formula links percolation and animal growth","Convert percolation thresholds into lattice animal bounds","From percolation thresholds to growth-rate bounds"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000537,"raw_usage":{"total_tokens":2560,"prompt_tokens":905,"completion_tokens":1655,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":521,"completion_tokens_details":{"reasoning_tokens":1595}},"tokens_in":521,"tokens_out":1655,"duration_ms":14156,"temperature":1.0,"reasoning_tokens":1595,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:13:20.679667+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Enumerate occurring interfaces in one percolation instance on the square lattice and look for two distinct outermost boundary layers that meet at a vertex; the paper's Lemma 3.2 forbids this, and finding one would falsify the uniqueness theorem on which the counting estimate rests. Alternatively, search numerically for an instance with more than a constant times $n$ occurring interfaces of size $n$ — the bound $N_n\\le l n+1$ would fail, and the chain from counting to Theorem 1.2 would need to be re-examined.","supporting_citations":[{"cited_title":"Georgakopoulos and C","cited_arxiv_id":null,"evidence_quote":"Supplies the unique-interface theorem (Theorem 3.5) and the definition of interfaces that the counting arguments rely on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Proved the animal analogue of the threshold theorem that the interface version extends and compares against."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Derivation of the animal exponential-decay threshold that the animal version builds on."},{"cited_title":"Benjamini and O","cited_arxiv_id":null,"evidence_quote":"Gives the Cheeger-constant bound $p_c\\le 1/(h+1)$ that the paper strengthens to a strict inequality."},{"cited_title":"Kesten and Y","cited_arxiv_id":null,"evidence_quote":"Locates the interval where interface probabilities fail to decay exponentially, used with Theorem 1.2 to identify where $b_r=f(r)$."},{"cited_title":"Aizenman and D","cited_arxiv_id":null,"evidence_quote":"Sharpness of the phase transition, giving exponential decay of cluster sizes below $p_c$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Pattern-theorem method adapted to show that interfaces are exponentially rarer than lattice animals."}],"review_version":1}