{"id":"ae095cc6-bcaf-4959-8758-77f4c1dd741a","arxiv_id":"2412.06772","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Critical clusters of the Gaussian free field on Z^3, Z^4 and Z^5 have tail exponent delta=(d+2)/(d-2) and largest-cluster dimension (d+2)/2, confirming Werner's conjectures.","lead":"For the Gaussian free field on metric graphs, this paper proves that in dimensions 3, 4 and 5 the largest critical cluster in a box of side length r has volume of order r to the power (d+2)/2, and that the cluster of the origin has a power-law volume tail. This confirms conjectures by Werner and fixes two critical exponents for a strongly correlated percolation model below its upper critical dimension.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1.1 for d=4,5 rests on the unverified one-arm bound sup q<∞ imported from [8]; if q grows polylogarithmically on Z^4, the sharp volume exponents in (1.3)-(1.4) would not follow.","rationale":"Both the reader and I identify the same load-bearing assumption: sup_r q(r)<∞, the boundedness of the normalized critical one-arm probability, imported from Cai-Ding [8]. I agree with the reader's ACCEPT-with-moderate-confidence in the sense that the paper's internal proofs appear careful and the dependence on q is explicit and correctly tracked in Proposition 4.1, Lemma 5.1, and (5.11)-(5.12). The argument is not circular and contains no fitted parameters. The reason I would move to CONDITIONAL rather than UNCHANGED is that the sharpness of Theorem 1.1 for d=4,5, the first rigorous volume exponents below d=6, depends on a single external, very recent preprint that is not part of the manuscript and is not even summarized. The paper's own proofs show that the polynomial tail n^{-ν/(2α-ν)} is obtained by absorbing sup q into constants; any unbounded q would introduce logarithmic or power corrections. Because [8] is load-bearing and unverified in this manuscript, acceptance should be conditional on an independent check of that one-arm bound, or on the authors providing a self-contained proof. This is not an objection to the internal logic, which is sound under the stated assumption.","tokens_in":39897,"tokens_out":8899,"duration_ms":86235,"concrete_test":"Independently re-derive the estimate sup_r q(r)<∞ for Z^4 and Z^5 from the proof in [8], paying special attention to Z^5 where α=5<2ν=6. Concretely, check whether the argument in [8] yields q(r)≤C uniformly for these dimensions. If the bound is confirmed, the present proofs of Proposition 4.1 and Theorems 2.2/2.5 go through; if only polylogarithmic growth is available, recompute (2.13) and (2.20) with q(r)=(log r)^C and see whether the announced exponents in Corollary 1.2 for d=4,5 survive.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Section 2, (2.11) states sup_r q(r)<∞ for Z^d, d=3,4,5, citing [8] for d=4,5; for Z^4 the authors' own [15] only gives q(r)≤(log r)^C. This condition is the multiplicative engine of the paper. In Proposition 4.1, the probability of a large hub-cluster is c q(c7 r)^{-4} and its volume is c8 a r^α / q(c7 r)^2. The proof of Theorem 2.2 then fixes r = (2n sup q^2/(c8 a))^{1/α} in (5.11), and (5.12) obtains the tail n^{-ν/(2α-ν)}. Both steps collapse if q is unbounded: the volume lower bound has extra q^{-2} factors, the probability has q^{-4}, and the exponent n^{-ν/(2α-ν)} is only recovered because sup q is absorbed into constants. If on Z^4 only q(r)≤(log r)^C were available, (1.3) would acquire logarithmic corrections and Corollary 1.2 would not follow from the present proof. Since [8] is a June 2024 preprint not reproduced here, the central claim in d=4,5 is exactly as secure as that external result; the paper's own Remark 5.2,2) confirms the condition is necessary, as sup q=∞ on Z^d, d≥7.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the volume of critical and near-critical clusters of the metric graph Gaussian free field on graphs satisfying polynomial volume growth and Green's function decay. The main result is Theorem 1.1: for Z^d with d in {3,4,5}, the cluster size tail P(|K_0| >= n) is of order n^{-(d-2)/(d+2)}, and the largest cluster in a box of radius r has volume of order r^{(d+2)/2}. The paper also proves an off-critical lower bound with exponential correction, moment bounds, and a general version (Theorems 2.2, 2.5, Corollary 2.3) valid under the standing condition sup_r q(r) < infinity, where q is the normalized one-arm probability. The proofs use the isomorphism with random interlacements, a hub-and-spokes construction (Proposition 4.1), entropy bounds, and second-moment arguments. The Zd applications for d=4,5 rely on the external one-arm bound of [8].","tokens_in":40187,"tokens_out":10438,"duration_ms":99914,"significance":"If correct, these are the first rigorous volume exponents for level-set percolation of the Gaussian free field below the upper-critical dimension, settling the value of the fractal dimension conjectured by Werner and the tail exponent conjectured in [16]. The paper is well structured: the conditional theorems are proved in full, the assumptions (V_alpha), (G_nu), (p0) are explicit, and the general graph formulation is a genuine strength. The stress-test concern about the external one-arm bound does land, but it does not undermine the paper's own contribution: Theorem 2.2 is conditional on sup q < infinity, and the dependence on [8] for Z^4,Z^5 is openly displayed in (2.11) and Remark 2.1. The unconditional status of Theorem 1.1 in those dimensions is exactly the status of [8]; I view this as a caveat about provenance rather than an internal flaw.","major_comments":[],"minor_comments":[{"comment":"The sentence 'sup q < infinity, when alpha > 2nu and on Z^alpha, alpha = 3,4,5' is ambiguous: for Z^4 one has alpha = 2nu, so the reader cannot tell whether the alpha > 2nu clause or the Z^alpha clause is the one being invoked. Please rephrase to state explicitly that for Z^3 the bound comes from alpha > 2nu, while for Z^4 and Z^5 it is imported from [8].","section":"Remark 2.1, (2.11)"},{"comment":"The condition 'r >= C((n/|a|)^{1/alpha} wedge n^{2/(2alpha-nu)})' should use a maximum (vee), not a minimum (wedge); the cluster K^a_r must fit inside the ball of radius r, so r must be at least the larger of the two scales. As written with wedge the remark is false in the parameter range where the two scales differ.","section":"Remark 5.2, item 4)"},{"comment":"The reference 'Theorem 2.2,(i)' is incorrect: Theorem 2.2 has a single assertion, not two items. It should read 'Theorem 2.2'.","section":"Proof of Corollary 2.3, page 26"},{"comment":"The phrase 'and call denote by ~B(x,r)' contains a typo; it should read 'and denote by ~B(x,r)'.","section":"Section 2, after (2.3)"},{"comment":"The statement repeats 'x in G' and reads awkwardly: it first says 'For all x in G and for all K as in (3.1)...' and then later 'for all s,r>=1, a>=0 and x in G such that...'. Please reword to avoid the duplicated quantifier.","section":"Lemma 3.1 statement"},{"comment":"Consider adding a sentence near Theorem 1.1 noting that the verification of sup q < infinity for d=4,5 is due to [8], a recent preprint, so that the reader is immediately aware that the unconditional statement in those dimensions depends on an external input not proved in this paper.","section":"Introduction, after Theorem 1.1"}],"recommendation":"minor_revision","confidential_remarks":"The paper is a strong contribution and the conditional results are carefully proved. The only reason I do not recommend outright acceptance is the set of small but real presentation issues listed, especially the wedge/vee mistake in Remark 5.2(4), which could mislead a reader applying the bound. The reliance on [8] is a normal citation practice and should not, by itself, block publication; an explicit caveat in the introduction would be sufficient."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe headline: this paper gets the critical cluster volume exponent δ=(d+2)/(d-2) and the largest-cluster dimension (d+2)/2 for the metric graph GFF on Z^d, d=3,4,5, matching Werner's conjecture. The new content is the lower bounds, and they're real. The upper bounds were already in the air from [7,8,16]; the matching lower bounds required a new mechanism, and Proposition 4.1 delivers it: build one big cluster by taking a high-capacity 'hub' cluster in a box and wiring it to the rest via interlacement trajectories. The proof also covers a general α,ν framework, which I thought was well executed.\n\nI read the stress-test note and disagree with any reading that this is essentially conditional. Yes, the lower bounds rest on sup q<∞, imported from [8] for d=4,5 (and from [15,17] for d=3). Yes, for Z^4 the best previously available bound was q(r)≤(log r)^C, so if [8] is wrong the sharp power law (1.3) would not follow from this proof. But the paper states this assumption explicitly in (2.11), and Remark 5.2,2) explains why the condition is necessary. That's honest dependence, not circularity. I did not find an internal flaw. The constant bookkeeping in the r-scaling does hold: the sup q absorbed into the constants is why (5.12) gets the clean exponent. If you drop sup q<∞ you get extra q factors, but the theorem is stated under that assumption.\n\nThe one thing I'd flag as a mild weakness: Theorem 1.3 and Corollary 1.4 give lower bounds only; matching upper bounds at small a are open, and the paper says so. That's fine, but it's worth knowing before citing the off-critical exponent.\n\nFor a referee: give this to someone who knows the one-arm bounds in this area, and have them check [8]. The result is important and the proof is serious. I'd bring it to reading group and cite it if my work touched volume tails in GFF.","headline":"Genuinely new matching lower bounds for critical GFF cluster volumes in d=3,4,5; solid proof, honestly conditioned on a recent one-arm estimate.","tokens_in":40758,"tokens_out":2448,"would_cite":true,"duration_ms":26999,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K35","60G15","82B43"],"pacs":[],"model":"deepseek-v4-flash","headline":"In d = 3, 4, 5, the critical cluster of the Gaussian free field has volume of order r^{(d+2)/2}, proving the conjectured exponents.","keywords":["Gaussian free field","metric graphs","percolation","critical exponents","cluster volumes","random interlacements","one-arm probability"],"falsifier":"On $Z^{4}$, simulate the normalized one-arm probability q(r) = r P(0 connects to the boundary of ~B(0,r)) for r up to $10^{4}$; if q(r) grows without bound, the standing assumption sup q<∞ fails and the hub construction collapses. Independently, on $Z^{3}$ estimate the slope of log P(|$K^{0}$|≥n) against log n for large n; a slope apart from -1/5 would contradict Theorem 1.1 directly.","tokens_in":39683,"feed_emoji":"🕸️","tokens_out":9049,"duration_ms":82556,"temperature":0.7,"pith_summary":"This paper determines the volume of critical clusters in the percolation of level sets of the Gaussian free field on metric graphs, in the three genuinely intermediate dimensions 3, 4 and 5. It proves that the probability that the cluster of the origin contains at least n vertices is, up to constants, $n^{{-(d-2)/(d+2)}}$, and that the largest cluster inside a box of side length r has volume of order $r^{{(d+2)/2}}$ with high probability. These are the exponents conjectured for this model, and they are not the mean-field values that hold in dimensions 7 and above; before this paper, no volume exponent below d=6 was known rigorously. The same method applies to any transient graph with polynomial volume growth and polynomial Green-function decay once a normalized critical one-arm probability stays bounded.","feed_headline":"Critical clusters in 3,4,5 dims scale as r^{(d+2)/2}","feed_subtitle":"First rigorous proof of the non-mean-field volume exponents below the upper critical dimension d=6.","key_machinery":"The load-bearing object is the normalized critical one-arm probability q(r) = $r^{{ν/2}}$ sup_{x∈G} P(x connects to ∂~B(x,r)), where ν is the exponent of the Green-function decay (ν=d-2 on Z^d). The proof's core construction, Proposition 4.1, selects the critical cluster with largest capacity inside a box, threads independent random-interlacement trajectories through it, and keeps every critical cluster touched by those trajectories; the isomorphism between the metric-graph Gaussian free field and random interlacements turns this decorated set into a genuine level-set cluster with both large volume and large capacity. A change-of-measure ('entropic repulsion') bound then converts the presence of such a cluster into the volume tails of Theorems 1.1 and 1.3. The whole mechanism runs only when q(r) stays bounded, which is exactly what the one-arm results provide on $Z^{3}$, $Z^{4}$ and $Z^{5}$.","core_discovery":"The central result, Theorem 1.1, is a two-sided estimate on Z^d for d∈{3,4,5}: there are constants c,C such that c $n^{{-(d-2)/(d+2)}}$ ≤ P(|$K^{0}$| ≥ n) ≤ C $n^{{-(d-2)/(d+2)}}$ for every n≥1, and for r,t with $r^{{(d+2)/2}}$ ≥ C t, with probability at least 1-C $t^{{-c}}$ the largest critical cluster in a box of radius r has cardinality between (1/t)$r^{{(d+2)/2}}$ and t $r^{{(d+2)/2}}$. Hence the critical exponents δ = lim_n log n / log P(|$K^{0}$|≥n) and d_f = lim_r log $M^{0}$_r / log r exist and equal (d+2)/(d-2) and (d+2)/2, resolving the conjectured values below the upper critical dimension d=6. The paper also proves Theorem 1.3, a lower bound on the near-critical tail: for a∈[-1,1] and n≥1, P(n≤|K^a|<∞) ≥ c P(n≤|$K^{0}$|<∞) exp{-C|a|^{(d+2)/d} $n^{{(d-2)/d}}$}. In the general graph setting these statements take the form (2.13) and (2.19), with the exponents ν/(2α-ν) and α-ν/2 replacing the Z^d-specific ones.","pith_inferences":["A natural test of the method is to push the same hub construction to $d=6$, where $q(r)$ grows only subpolynomially; the paper's Remark 5.2,3) indicates that the resulting tails should carry the same exponential cost with subpolynomial corrections, which would complete the exponent table at the upper critical dimension.","Because the argument only needs capacity-volume comparability and a bounded one-arm probability, it should transfer to any long-range-correlated percolation model in the same universality class, such as the discrete excursion sets of the Gaussian free field, for which the lower bound (1.11) already yields a sub-exponential tail.","One concrete observable to check numerically is the typical count of critical clusters with volume of order $r^{(d+2)/2}$ and diameter at least $sr$; Remark 5.2,5) predicts the expected number is of constant order, which a simulation on $\\mathbb{Z}^3$ could verify directly.","If matching upper bounds for (1.12) are proved, the moment exponents of Corollary 1.4 would establish the conjectured value $\\Delta=2d/(d-2)-1$, fixing the full set of gap exponents for this model."],"forward_implications":["On $\\mathbb{Z}^3$, $\\mathbb{Z}^4$ and $\\mathbb{Z}^5$, the critical cluster-volume exponents are $\\delta=(d+2)/(d-2)$ and $d_f=(d+2)/2$, matching the fractal-dimension conjecture for this model; this is the first rigorous determination below $d=6$.","The same bounds hold on every graph satisfying the volume-growth and Green-function assumptions $(V_\\alpha),(G_\\nu)$ with $\\sup_r q(r)<\\infty$; for instance, a $\\mathbb{Z}^2$-times-Sierpinski-gasket graph has explicit but non-algebraic exponents $\\delta\\approx 3.9299$ and $d_f\\approx 2.5339$.","In the near-critical regime the paper gives $P(n\\le |K^a|<\\infty) \\ge c\\, n^{-\\nu/(2\\alpha-\\nu)} \\exp\\{-C|a|^{2-\\nu/\\alpha} n^{\\nu/\\alpha}\\}$, and the resulting moment lower bound $\\mathbb{E}[|K^a|^k; |K^a|<\\infty] \\ge c|a|^{1-k(2\\alpha-\\nu)/\\nu}$ matches the conjectured gap exponent $\\Delta=2d/(d-2)-1$ on $\\mathbb{Z}^d$ once upper bounds are added.","The statement for the largest cluster fails for $d\\ge 7$, where the volume of the largest cluster is of order $r^4$; the assumption $\\sup_r q(r)<\\infty$ is therefore not a technical convenience but a necessary hypothesis, as the paper notes in Remark 5.2,2)."],"supporting_citations":[{"why":"establishes sup_r q(r)<∞ on Z^3, Z^4 and Z^5, the standing one-arm input on which the hub construction rests","marker":"[8]"},{"why":"proves the intermediate-dimension one-arm bounds for graphs with α>2ν and α=2ν that are needed in the general setup","marker":"[15, 17]"},{"why":"supplies the critical capacity tail (1.2), the entropy/change-of-measure bound (3.7), and the radius-to-capacity strategy adapted here to volume","marker":"[16]"},{"why":"provides the isomorphism between the metric-graph Gaussian free field and loop soups/random interlacements, including [32, Prop. 5.2] used for the upper bound on E[M^0_r]","marker":"[32]"},{"why":"formulates the fractal-dimension (d+2)/2 conjecture for low dimensions and the mean-field r^4 comparison that the theorem proves for d≤5","marker":"[44]"},{"why":"shows the mean-field one-arm behavior in d≥7 and thereby pinpoints where the largest-cluster conclusion is false (Remark 5.2,2)","marker":"[7]"},{"why":"supplies the general transient-graph framework, the isomorphism lemma, and the local-uniqueness estimates for interlacements used in Proposition 4.1","marker":"[12]"}],"fun_headline_variants":["Cluster volume exponents proven for dimensions 3,4,5","Gaussian free field critical clusters: non-mean-field exponents","Werner's conjecture on cluster volumes resolved below d=6","Largest critical cluster volume order r^{(d+2)/2} in low dims","Exact critical cluster volume exponents for d=3,4,5"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof assumes that the normalized critical one-arm probability q(r) = $r^{{ν/2}}$ sup_x P(x connects to the boundary of a box of radius r) stays bounded as r grows; should q(r) diverge, the hub cluster with large capacity would no longer be constructed, and the volume tails would not follow.","fun_headline_variants_meta":{"raw":{"variants":["Cluster volume exponents proven for dimensions 3,4,5","Gaussian free field critical clusters: non-mean-field exponents","Werner's conjecture on cluster volumes resolved below d=6","Largest critical cluster volume order r^{(d+2)/2} in low dims","Exact critical cluster volume exponents for d=3,4,5"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000971,"raw_usage":{"total_tokens":4176,"prompt_tokens":1043,"completion_tokens":3133,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":659,"completion_tokens_details":{"reasoning_tokens":3039}},"tokens_in":659,"tokens_out":3133,"duration_ms":22990,"temperature":1.0,"reasoning_tokens":3039,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T19:20:48.855792+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"On $Z^{4}$, simulate the normalized one-arm probability q(r) = r P(0 connects to the boundary of ~B(0,r)) for r up to $10^{4}$; if q(r) grows without bound, the standing assumption sup q<∞ fails and the hub construction collapses. Independently, on $Z^{3}$ estimate the slope of log P(|$K^{0}$|≥n) against log n for large n; a slope apart from -1/5 would contradict Theorem 1.1 directly.","supporting_citations":[{"cited_title":"Drewitz, A","cited_arxiv_id":null,"evidence_quote":"supplies the critical capacity tail (1.2), the entropy/change-of-measure bound (3.7), and the radius-to-capacity strategy adapted here to volume"},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"provides the isomorphism between the metric-graph Gaussian free field and loop soups/random interlacements, including [32, Prop. 5.2] used for the upper bound on E[M^0_r]"},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"formulates the fractal-dimension (d+2)/2 conjecture for low dimensions and the mean-field r^4 comparison that the theorem proves for d≤5"},{"cited_title":"Drewitz, A","cited_arxiv_id":null,"evidence_quote":"supplies the general transient-graph framework, the isomorphism lemma, and the local-uniqueness estimates for interlacements used in Proposition 4.1"}],"review_version":1}