{"id":"b83b299b-353c-4fd9-93aa-13bf522af030","arxiv_id":"2501.00932","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A newly constructed 'upper tail field' is the local limit of the KPZ fixed point near a conditioned large value, interpolating between Brownian and KPZ scaling regimes.","lead":"The authors introduce a new random object, the upper tail field of the KPZ fixed point, and prove it emerges when the KPZ interface is conditioned on an extremely large height and then zoomed in near that point. The same field is shown to connect two known behaviors: a Brownian-type limit before the large height and the usual KPZ fixed point after it.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Equal-time bent-contour multipoint formula is the load-bearing unproved premise; if it fails, H_UT's definition and Theorem 1.1 break.","rationale":"Proposition 1.13 is the engine: the m=1 case is the known GUE tail, while m≥2 uses the multipoint formula (3.5). The paper's own Section 1.4 states that equal times were a difficulty in [LW24] and claims the present analysis works for equal times because contours are bent. But the validity of the bent-contour formula is not proved here: Proposition 3.1 is proved by induction from (3.5), and (3.5) itself is imported. The footnote directs to [Liu22a], yet [LW24] had already found equal times an obstruction in exactly this circle of formulas. Thus the equal-time bent-contour variant is the least secure premise on which the existence of H_UT rests. This matches the reader's weakest_assumption. I do not claim the formula is false; the numerical test would settle the m=2 case, and the same reduction can be iterated for m≥3. Secondary issues, such as the sketched proofs of Lemmas 5.3 and 5.4 and uniform bounds that omit factorial factors, affect Proposition 1.8 and some technical estimates but do not threaten the central construction as directly. The reader's CONDITIONAL verdict is therefore appropriate, and my read changes nothing.","tokens_in":48937,"tokens_out":18759,"duration_ms":171163,"concrete_test":"Check the equal-time case m=2, τ1=τ2=0, α1=0, α2=α>0. By Definition 2.4 and independence of H_UT(0,0), the tail function must obey T(0,β;(0,0),(α,0)) = P(B(2α)-2α ≥ β) = 1 - Φ((β+2α)/√(2α)). Evaluate the m=2 version of (2.21) at equal times by truncating the n-sum and using the explicit contours in (3.17)-(3.18), for several parameters, e.g. (α,β)=(1,0),(1,1),(2,-1). An analytic variant is to deform the v-contours in (2.21) to residues at v=1 and show the resulting series telescopes to the same Gaussian tail. Agreement confirms the bent-contour equal-time formula and hence the consistency of \\hat T on time slices; disagreement would invalidate the definition of H_UT and Theorem 1.1.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing step is Proposition 1.13, whose equal-time case is asserted rather than proved. Proposition 1.13 is the sole input converting contour integrals into the convergence (4.1); that convergence is used both in the proof of Theorem 1.1 and, via (2.29), in Proposition 2.5 to verify Kolmogorov consistency of the tail functions \\hat T. The proof starts from the multipoint tail formula (3.5), imported from [Liu22a] and [LW24, (2.4)], with contours 'bent' to allow equal times. But the paper does not prove this bent-contour variant: Proposition 3.1 states it, and the footnote refers to [Liu22a] after Definition 2.25. Equal times were a known obstruction in [LW24], which handled them by a separate probabilistic argument; the present text asserts that bending one side resolves the issue. If the bent-contour formula has a hidden restriction at equal times, T is not the correct limit for same-time points, \\hat T fails consistency on all of R^2, H_UT need not exist as a field, and Proposition 1.5(d) plus Theorem 1.1 at a fixed time lose support. This is a concrete unverified premise, not a demonstrated error.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a new random field, the upper tail field H_UT of the KPZ fixed point, defined by explicit multipoint tail functions T built from contour integrals. The main theorem states that, conditionally on a large value H_KPZ(α̂L^{-1},1+τ̂L^{-3/2})≥L+β̂L^{-1/2}, the rescaled field √L(H_KPZ(αL^{-1},1+τL^{-3/2})−L) converges in finite-dimensional distributions to β̂+H_UT(α−α̂,τ−τ̂); the equality-conditioned version converges to β̂+H_UT^0. The proof proceeds through a multipoint upper tail estimate (Proposition 1.13) obtained by steepest-descent analysis of an explicit multipoint distribution formula for the KPZ fixed point. The paper also proves structural properties of H_UT: H_UT(0,0) is exponential, H_UT^0 is independent of it, the time-zero slice is B_ts(2α)−2|α|, and suitable large-scale limits recover a Brownian-type field in negative times and the KPZ fixed point in positive times.","tokens_in":49175,"tokens_out":9785,"duration_ms":97593,"significance":"If correct, this is a substantial contribution: it identifies a new universal object for the KPZ class in the upper-tail conditioning regime, rigorously connects two previously studied scaling limits, and provides explicit finite-dimensional tail formulas with no free parameters. The construction is not circular: the field is defined from the contour-integral functions T and its existence is verified from the explicit asymptotics of the pre-limit probabilities, as the paper explicitly states after (2.29). The paper also gives exact distributional identities (exponential marginal, Brownian time-zero slice) that are concrete and falsifiable. The main risk is a technical gap in the equal-time case of the multipoint tail formula, which is load-bearing for the consistency of H_UT and for part of Theorem 1.1; this is a fixable gap rather than a demonstrated error.","major_comments":[{"comment":"The equal-time bent-contour version of the multipoint tail formula is asserted rather than proved. The text states that when some times coincide, the contours need to be bent according to the order under ≺, and says 'we bend the contours at the beginning so that the integral is well defined', referring to [Liu22a]. This is precisely the point that [LW24] found to be an obstruction and handled by a separate probabilistic argument for equal times. Proposition 3.1 is the sole input for Proposition 1.13 at equal times, and Proposition 1.13 is used in (4.1) for arbitrary space-time points and, through (2.29), in the consistency proof of Proposition 2.5. The paper should either prove the bent-contour equality from the known formula or give a precise statement in [Liu22a] with its hypotheses verified, including equal times. This is a load-bearing gap, not a demonstrated error.","section":"Section 3.1, Proposition 3.1 and footnote after (3.6)"},{"comment":"The definition of T relies on the assertion, made after (2.17), that the integrals are independent of the specific choices of the Γ-contours as long as they satisfy the stated nesting and angular conditions. This independence is not proved. In the strictly ordered-time case it is presumably a standard contour deformation, but in the equal-time case the bending of contours is exactly the delicate mechanism that Proposition 3.1 is supposed to justify. Since T is the building block of the whole field, the paper should provide a proof of contour independence or an explicit reference covering the bent-contour setting.","section":"Section 2.1, Definition 2.2"},{"comment":"The proofs of Lemmas 5.3 and 5.4 are only sketched: the text says 'we only provide the main steps of the proof and skip the details'. These two lemmas supply both the limit identification and the uniform bounds used to pass to the limit in (5.4), so Proposition 5.1 and hence Proposition 1.8(a) depend on them. In particular, Lemma 5.4 needs a written proof of the exponential decay factor and the summation/integration bounds that justify dominated convergence. The same applies to the unproved uniform bound (5.40) in Section 5.2, which is used to justify the limit in Proposition 5.5. These are routine but nontrivial technical steps, and they are load-bearing for the large-scale limit claims.","section":"Section 5.1, Lemmas 5.3 and 5.4"}],"minor_comments":[{"comment":"The text refers to 'Proposition 2.1' when applying the Cauchy determinant bound; the intended reference is Lemma 2.1.","section":"Section 5.1, proof of Lemma 5.4"},{"comment":"In the displayed expression for the difference of tail probabilities, the second argument contains repeated (α_{k−1},τ_{k−1}); it should be (α_{k+1},τ_{k+1}) after removing the k-th point.","section":"Section 2.2, proof of Proposition 2.5, equation (2.34)"},{"comment":"The statement writes 'Dn(n;z)' where D_n(h;z) is meant, and the bound 'C n1+···+nm' should read 'C^{n_1+···+n_m}'.","section":"Lemma 3.4"},{"comment":"The proposition begins 'For all x, τ, β∈R' but the formula uses the spatial variable α; the notation should be made consistent.","section":"Proposition 1.5(b)"},{"comment":"The paper appeals to the Kolmogorov extension theorem for joint tail probability functions, but Proposition 2.5 only states boundary limits and marginal consistency. It would be clearer to state explicitly that \\hat T inherits monotonicity and right-continuity from the limiting representation (2.29), since those properties are needed for the extension theorem.","section":"Section 2.2, Proposition 2.5"}],"recommendation":"major_revision","confidential_remarks":"The equal-time bent-contour formula is the main correctness risk. I do not think the paper should be rejected: the issue is a missing proof or precise citation of a technical input, not a demonstrated contradiction. If the authors can supply a self-contained proof of Proposition 3.1 at equal times and full details for Lemmas 5.3 and 5.4, the paper would be a strong contribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read Liu\\u2013Zhang's upper tail field paper. The headline: they introduce H_UT, a space-time field describing the KPZ fixed point near a conditioned high point, prove convergence to it under the 1:2:3 scaling, and show it interpolates the Brownian and KPZ regimes. That is a real advance over [LW24] and [NZ22], which had only one-point or fixed-time results. The multipoint upper tail estimate (Prop 1.13) is the technical heart, and the asymptotic analysis in Section 3 is substantial. Proposition 1.5(d), identifying the \\u03c4=0 slice as Brownian motion minus absolute value, is a nice result, and the large-scale limits in Prop 1.8 are exactly what you would hope for.\n\nThe soft spots, in proportion.\n\nThe equal-time case of the multipoint formula is genuinely load-bearing. Proposition 3.1 imports a variation of [Liu22a]'s formula with contours 'bent' so equal times work. The footnote says bending one side resolves the obstruction, but there is no proof. Equal times were a known obstacle in [LW24], which needed a separate probabilistic argument. If the bent-contour formula has hidden restrictions, then T is not established as the limit for same-time points, and the consistency in Prop 2.5 \\u2014 hence the existence of H_UT \\u2014 is unsupported. The paper says the proof of Theorem 1.1 does not depend on the field's existence, but it still depends on the equal-time case of Prop 1.13. This is a real gap. It may well be true, since the authors know this formula deeply, but a referee should ask for a complete proof or a precise statement of which equal-time configurations the formula covers.\n\nSecond, Lemmas 5.3 and 5.4 in Section 5.1, which justify the Brownian limit in the negative-time regime, are sketched ('we only provide the main steps and skip details'). That is probably fine for an early version, but these are nontrivial asymptotics and deserve to be written out.\n\nMinor: the heavy notation in (2.17) could use a small glossary, but that is a matter of taste.\n\nOverall: if the equal-time contour claim holds up, the main theorem is solid and the new field is likely to be useful. The citation to [Liu22a] is appropriate, and the stated consistency verification through tail functions is the right approach. This paper deserves a serious referee. I would make acceptance conditional on a full proof of the equal-time bent-contour formula, or a clear reference with the argument spelled out, plus the Brownian-limit details.","headline":"A genuinely new limiting field with a clean main theorem, but the equal-time contour formula it leans on is asserted, not proved; a referee should demand that proof or a workaround.","tokens_in":49726,"tokens_out":2312,"would_cite":false,"duration_ms":23498,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K35","82B41","60J65"],"pacs":[],"model":"deepseek-v4-flash","headline":"Conditioning the KPZ fixed point on a large height at one point and zooming in at scales L^{-1} in space and L^{-3/2} in time produces a new random field H_UT on all of R^2.","keywords":["KPZ universality class","KPZ fixed point","upper tail field","Tracy-Widom distribution","directed landscape","Brownian motion","random growth"],"falsifier":"Compute the finite-dimensional distributions of H_UT directly from the explicit contour formula for the equality-conditioned case and compare them with a simulation of the KPZ fixed point conditioned on a large height at (0,1) with L=1000; if the sampled joint tails deviate from the T-function predictions at equal times, the bent-contour assumption fails.","tokens_in":1641,"feed_emoji":"","tokens_out":2585,"duration_ms":79811,"temperature":0.7,"pith_summary":"This paper proves that conditioning the KPZ fixed point, the conjectured universal space-time limit of KPZ growth models, on a large height at a point and then zooming into a tiny space-time window yields a new limiting random field H_UT. The window scales space as $L^{{-1}}$, time as $L^{{-3/2}}$, and height fluctuations as $L^{{1/2}}$, which are smaller than the scales of earlier one-point or pre-high-point limits. The new field lives on the whole $R^{2}$ plane, whereas the KPZ fixed point itself only has nonnegative time. Zooming out of H_UT recovers a Brownian-type minimum field for negative times and the KPZ fixed point for positive times, so H_UT interpolates between two previously known conditional regimes.","feed_headline":"New random field appears at a rare high point of the KPZ interface","feed_subtitle":"Zooming into a conditioned tall growth spot yields a universal field that connects Brownian and KPZ scales.","key_machinery":"The load-bearing object is a multipoint upper tail estimate (Proposition 1.13). For ordered space-time points (alpha_1,tau_1) prec ... prec (alpha_m,tau_m) near (0,1), the rescaled joint tail probability 16*pi*$L^{{3/2}}$ $e^{{4/3 L^{3/2}}$} P(cap_{ell=1}^m {H_L(alpha_ell,tau_ell) ≥ beta_ell}) converges to an explicit function T($\\beta$;(alpha_1,tau_1),...,(alpha_m,tau_m)), with a companion statement for the derivative in each beta_k. The function T is defined by contour integrals with Cauchy determinants, and the proof begins from the explicit multipoint distribution formula for the KPZ fixed point with contours bent so that the formula remains valid when several points share the same time. The joint tail functions of H_UT are assembled from T by inserting the conditioned point (0,0) and ordering all points, and the Kolmogorov extension theorem turns these consistent tail functions into a genuine random field once the consistency conditions are checked.","core_discovery":"The central claim is a conditional scaling limit: conditioned on H_KPZ(hat_alpha $L^{{-1}}$,1+hat_tau $L^{{-3/2}}$) ≥ L+hat_beta $L^{{-1/2}}$, the rescaled field $\\sqrt$(L)(H_KPZ($\\alpha$ $L^{{-1}}$,1+tau $L^{{-3/2}}$)-L) converges in finite-dimensional distributions to hat_beta+H_UT($\\alpha$-hat_alpha,tau-hat_tau). Conditioning instead on equality gives hat_beta+$H_UT^{0}$, where $H_UT^{0}$($\\alpha$,tau)=H_UT($\\alpha$,tau)-H_UT(0,0). The new field H_UT satisfies: H_UT(0,0) is an Exponential(2) random variable; $H_UT^{0}$ is independent of H_UT(0,0); at time tau=0 the spatial process $H_UT^{0}$($\\alpha$,0) has the same law as B_ts(2alpha)-2|$\\alpha$| for a two-sided Brownian motion B_ts. Zooming out of H_UT at large scale gives min{B_1(-t)+x, B_2(-t)-x} for negative times and H_KPZ(x,t) for positive times. Thus H_UT is presented as a new universal object attached to an unusually high point of the KPZ fixed point, bridging Brownian and KPZ scaling behaviors.","pith_inferences":["Editorial extension: because the equal-time contour-bending step was the previously known obstruction, the same bent-contour multipoint formula should also yield upper tail estimates for other fields built from the explicit KPZ fixed point distribution, such as fixed-time Airy line ensemble marginals.","Editorial extension: the paper conjectures, with a heuristic based on the flat-initial-condition one-point tail, that H_UT is independent of the initial condition of the KPZ fixed point; a direct test would be to compute the same conditional limit for the flat initial condition and check that its one-point tail matches e^{-2 max{beta,0}}.","Editorial extension: the decomposition H_UT = H_UT^0 + Exponential(2) suggests a general principle for conditioned random surfaces: the conditioned height should factor into a global exponential component and a locally universal shape fluctuation, a splitting that could be tested in other exactly solvable growth models with multipoint formulas."],"forward_implications":["If H_UT is universal within the KPZ universality class, every KPZ-universal growth model conditioned on a rare high point should exhibit the same local field H_UT in the same scaling window.","The field H_UT provides a single interpolation between the previously known Brownian-bridge limit before the high point and the unconditioned KPZ fixed point after it, so the 1:2:3 scaling of the positive-time regime and the 1:2 scaling of the negative-time regime are limits of one object.","The time-zero slice H_UT^0(alpha,0) is explicit, namely B_ts(2alpha)-2|alpha|, giving a concrete Brownian description of the transition layer at the critical time.","The equality-conditioned field H_UT^0 is well defined and independent of the exponential height at the conditioned point, cleanly separating the global rarity of a large height from the local fluctuation around that height."],"supporting_citations":[{"why":"Supplies the explicit multipoint distribution formula of the KPZ fixed point whose tail-probability variation is the starting point for the asymptotic analysis.","marker":"[Liu22a]"},{"why":"Established the Brownian-bridge conditional scaling limit before a large height and identified the equal-time obstruction that the present bent-contour method must overcome.","marker":"[L W24]"},{"why":"Proved the one-point conditional limit of the KPZ fixed point after a large height, the positive-time regime that H_UT must recover when zoomed out.","marker":"[NZ22]"},{"why":"Constructed the KPZ fixed point as the universal space-time object whose conditional scaling is studied here.","marker":"[MQR21]"},{"why":"Provides the GUE and GOE Tracy-Widom upper tail asymptotics used to identify the one-point exponential tail of H_UT and to check boundary limits in the consistency proof.","marker":"[BBD08]"},{"why":"Supplies the directed landscape representation and the metric triangle inequality used to prove that the one-point tail function of H_UT has the correct limits.","marker":"[DOV22]"}],"fun_headline_variants":["Rare KPZ spike spawns a new universal random field","Zooming into a tall KPZ point reveals a Brownian–KPZ bridge","Upper tail field: a new object linking Brownian and KPZ scales","Conditioned on a big KPZ value, a new field emerges","KPZ fixed point's rare high value yields a new scaling limit"],"cache_read_input_tokens":51840,"weakest_assumption_plain":"The result depends on the validity of the bent-contour multipoint distribution formula for the KPZ fixed point when two or more space-time points share the same time, since the equal-time case is known to be a delicate obstruction and the new contour choice is introduced specifically to handle it.","fun_headline_variants_meta":{"raw":{"variants":["Rare KPZ spike spawns a new universal random field","Zooming into a tall KPZ point reveals a Brownian–KPZ bridge","Upper tail field: a new object linking Brownian and KPZ scales","Conditioned on a big KPZ value, a new field emerges","KPZ fixed point's rare high value yields a new scaling limit"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000299,"raw_usage":{"total_tokens":1773,"prompt_tokens":1031,"completion_tokens":742,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":647,"completion_tokens_details":{"reasoning_tokens":648}},"tokens_in":647,"tokens_out":742,"duration_ms":7411,"temperature":1.0,"reasoning_tokens":648,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:40:56.101253+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the finite-dimensional distributions of H_UT directly from the explicit contour formula for the equality-conditioned case and compare them with a simulation of the KPZ fixed point conditioned on a large height at (0,1) with L=1000; if the sampled joint tails deviate from the T-function predictions at equal times, the bent-contour assumption fails.","supporting_citations":[],"review_version":1}