{"id":"90b0d583-e860-413e-ae4f-afef23588fbc","arxiv_id":"2608.13246","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Forbidden singularities in heavy-light Virasoro blocks are shown to be the critical point of an instanton-gas phase transition via AGT, with critical fugacity equal to the known singularity location.","lead":"The paper maps the “forbidden singularities” in thermal autocorrelators of 2D conformal field theories to a phase transition in a gas of instantons in a 4D supersymmetric gauge theory, using the AGT correspondence. It derives the critical fugacity of the transition and shows it exactly matches the singularity location, giving a statistical-mechanics picture of eigenstate thermalization.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Symmetric saddle-point exclusion is asserted, not proved; if it fails, r < z* and the central equality is lost.","rationale":"The reader's weakest assumption is the same as the main threat I find: the symmetric saddle-point solutions are excluded by an unproved Lefschetz-thimble assertion. The paper's own Section IV.D contains the conflict with the central claim, so this is not an external or contrived objection. The finite-c zero plots and the free-energy scaling in Figures 1-2 are real evidence for a transition at z* at the level of the exact Virasoro block, but they cannot select which saddle of the effective instanton-gas action controls the large-nu growth of the coefficients; that is exactly what the saddle-point analysis must establish. The chosen solution reproduces z*, while the symmetric solutions, if contributing, would give a smaller radius. The derivation is therefore incomplete at its decisive step. The authors are transparent about this gap, so the critique is about the argument, not about any misrepresentation. A conditional verdict is appropriate because the physical claim may well be correct, but the proof as written does not yet rule out the competing saddles.","tokens_in":34246,"tokens_out":4388,"duration_ms":53995,"concrete_test":"Truncate the integral (64)/(71) at X_max = 120 for alpha_H = 10, h_L = 1, locate all saddle points of (72)-(73) including the symmetric roots of (90), and compute their Lefschetz thimbles by integrating the downward-flow equation dz/ds = -partial f/partial z from each saddle. The symmetric saddles are irrelevant only if no thimble intersects the original real cone. Repeat at X_max = 60, 120, 240 to check convergence of the flow and thimble intersections. If any symmetric thimble intersects the contour, recompute r as e^{-Re(lambda_sym)}; a value below z* falsifies the central claim as stated. A complementary check is to compute exact finite-c Nekrasov coefficients Z_nu for c = 100..400 and nu up to about 25 and extrapolate (Z_nu)^{-1/nu}; convergence toward e^{-Re(lambda_sym)} rather than z* would confirm the concern.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central equality r = z* is obtained by solving the saddle-point equations (72)-(73) with the maximally symmetry-breaking ansatz y2 = 0, leading to the on-shell values lambda_n in (82) and the claimed radius (84). But Section IV.D immediately exhibits symmetric solutions y_X(omega) in (90) whose numerical roots have Re(lambda_sym) > ln(1/z*), as shown in Figure 6. Such solutions would give a contribution e^{nu Re(lambda_sym)} to Zbar_nu, hence a radius e^{-Re(lambda_sym)} < z*, directly contradicting (84). The paper dismisses these solutions with the assertion that their Lefschetz thimbles do not intersect the integration contour of (64), and with a speculation that the associated z-singularity lies on another branch of the free energy F(z). No flow equations, no intersection check, and no branch-cut computation are supplied; the text itself calls understanding this obstruction a future problem. Since the path integral is defined over a real cone, every saddle with Re(lambda) larger than the chosen one must be explicitly shown not to contribute before the large-nu growth of Zbar_nu can be claimed. The branch-of-F(z) argument cannot be used as a proof: whether a CFT free-energy branch contains a singularity at z = e^{-lambda_sym} is precisely the statement that the saddle-point computation is meant to determine. A secondary gap is the regulator Delta -> 0 limit of the effective action (footnote 5), but the unproved exclusion of the symmetric saddle is the load-bearing concern.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper argues that the forbidden singularity in the heavy-light vacuum Virasoro block at large central charge is exactly the critical fugacity of an instanton-gas phase transition. Using the AGT correspondence, the authors rewrite the vacuum block as an SU(2) Nekrasov partition function and derive a c-independent effective action I0 for pairs of Young tableaux in the heavy-light limit. At large instanton number ν they propose a string-like scaling ansatz, convert the tableau sum into a saddle-point integral, and solve the saddle-point equations with the maximally symmetry-breaking ansatz y2=0. This yields on-shell values λ_n and a radius of convergence r = 1 - e^{-2π/α_H} = z*, matching the known forbidden singularity. The paper also discusses the high-fugacity phase, reproduces the z→1 OPE singularity through a complex saddle in the total number of rows, and interprets the discontinuity at c=∞ as an essential singularity. Numerical checks include finite-c zeros against anti-Stokes curves, free-energy scaling changes across z*, and convergence of the ratio Z_{ν+1}/Zν toward z*. The central claim is explicit and falsifiable, but it depends on several unproven analytic-continuation and saddle-exclusion steps.","tokens_in":1842,"tokens_out":5956,"duration_ms":88699,"significance":"If the central claim holds, the paper provides a concrete statistical-mechanical mechanism for forbidden singularities in eigenstate thermalization: the singularity is not a kinematic artifact but an instability toward instanton proliferation, analogous to a Lee-Yang transition. This is an original and potentially influential connection between ETH signatures, Virasoro blocks, and 4d instanton gases. The manuscript contains several genuine non-trivial checks: the reproduced radius z*, the numerical convergence in Figure 5, and the finite-c zero distributions in Figure 1. The derivation is not parameter-fitted; the critical fugacity emerges from saddle-point equations. However, the main equality r = z* is load-bearing and currently rests on the unproven exclusion of competing symmetric saddle-points. The paper would be a strong contribution if that gap is closed; in its present form the central claim is defensible but not established.","major_comments":[{"comment":"The symmetric saddle-point solutions displayed in Eq. (90) have numerically determined on-shell values with Re λ_sym > ln(1/z*), as shown in Figure 6. If they contributed to the path integral (64), their growth e^{ν Re λ_sym} would dominate Zν and force a radius of convergence r < z*, directly contradicting the central equality (84). The paper excludes them by asserting that their Lefschetz thimbles do not intersect the defining integration contour and by speculating that the associated z-singularity lies on another branch of F(z). No flow equations, no intersection check, and no branch-cut computation are supplied; the text itself calls this a future problem. This exclusion is load-bearing: the path integral is defined over a real cone, so every saddle with larger Re λ than the chosen one must be shown not to contribute before the growth of Zν can be claimed. The branch-of-F(z) argument cannot serve as a proof, because whether the singularity lies on the principal branch of F(z) is precisely what the saddle-point computation is meant to determine.","section":"§IV.D, Eq. (90) and Figure 6"},{"comment":"The reduction of the Young-tableau sum to the saddle-point integral (64) relies on two unproven assumptions. First, the string-like scaling ansatz (57) is proposed rather than derived; Section IV.E gives a scaling argument that two-dimensional tableaux have I0 ∼ ν^{1/2}, but it does not exclude other growth patterns, and the integration measure is explicitly dropped. Second, the analytic continuation of the real-cone integral (70) to complex saddle-points is assumed to be unobstructed, with the text stating 'we assume that there is no subtle obstruction'. Since the final answer is a radius of convergence, the dominance of the chosen saddle over all other sectors and the validity of the deformation are essential, not technical, points. A Picard-Lefschetz analysis or a rigorous large-deviation estimate for the sum over all tableaux is needed.","section":"§IV.B–C, Eqs. (57), (64), (70)–(71)"},{"comment":"The physical vacuum block corresponds to internal dimension Δ = 0, but the effective action (47) contains terms such as 2 ln[Γ(Δ)/Γ(Y_2^1 + Δ)] and ln[Γ(2h_L + Δ − 1)/Γ(Y_2^2 + 2h_L + Δ − 1)] that are singular or delicate as Δ → 0. The paper avoids this by assuming a generic non-zero Δ of order one in the c → ∞ limit and states that the modification is negligible. No argument is given that the Δ → 0 limit commutes with the large-ν saddle-point analysis, nor that the saddle-point solution y2 = 0 makes the potential divergences harmless. Since the claim concerns the vacuum block at Δ = 0, this regulator issue must be resolved or explicitly shown to be irrelevant.","section":"Footnote 5 and §IV.A, Eqs. (48)–(49)"},{"comment":"The derivation of the normalization constant k in Eq. (85) interchanges sums and integrals and uses the identity (1 - e^{-λ})^{iα_H - 1} = 0 to discard a formally divergent factor involving ∑_{i=1}^∞ 1. This step is not justified, and while k does not enter the radius of convergence, the same resummation technique is used to obtain the boundary-condition equation (81) from which λ_n is determined. Please provide a careful justification of the order of summation and the treatment of the divergent intermediate expressions, or derive the boundary condition by an alternative method.","section":"§IV.C, Eqs. (80)–(86)"}],"minor_comments":[{"comment":"The subsection heading contains a typo: 'Method of monodomy' should read 'Method of monodromy'.","section":"§II.A"},{"comment":"The phrase 'back hole information paradox' should be 'black hole information paradox', and the abstract's 'bare resemblance' should be 'bear resemblance'.","section":"Introduction"},{"comment":"The caption describes the figure as the distribution of rescaled complex effective actions, but the horizontal axis is labeled Log[Z_{ν+1}/Zν]; please clarify what exactly is plotted and how the distribution relates to the ratio.","section":"Figure 5"},{"comment":"Reference [4] is missing the journal name and volume information, and reference [28] lists only an arXiv identifier with no journal reference; please complete these entries.","section":"References"},{"comment":"The high-fugacity analysis assumes that the total number of rows n can be analytically continued to a complex variable w with unspecified analytic properties of B_w(α_H,c); since this section is secondary to the main claim, a brief justification or a comment on the required analytic structure would suffice.","section":"§V, Eqs. (104)–(106)"}],"recommendation":"major_revision","confidential_remarks":"The paper is interesting and the central identification r = z* is compelling, but the missing Lefschetz-thimble analysis for the symmetric saddle-points is essential. I would support publication if the authors can either prove the non-contribution of the symmetric saddles or identify a well-defined prescription under which those saddles are excluded. The Δ → 0 regulator issue should also be addressed. The current version is not ready for acceptance in its present form."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Main thing to know: this paper is worth a referee's time, but the central claim is not yet airtight. The genuinely new step is the large-ν treatment of the AGT instanton gas: string-like Young tableau ansatz, a homogeneous effective action, and a saddle-point evaluation that extracts r = 1 − e^{−2π/α_H} as the critical fugacity. The location z* was already known from monodromy methods, but deriving it from the proliferation of instantons is a real conceptual step, and the numerical checks in Figures 1, 2, and 5 give honest support. The authors also flag several of their own gaps, which is to their credit.\n\nThe soft spot is load-bearing. In Section IV.D the authors exhibit symmetric saddle-point solutions whose on-shell Re λ is larger than the chosen solution's, which would give r < z* and contradict the advertised equality. They dismiss these saddles with two arguments: that the Lefschetz thimbles do not intersect the defining integration contour, and that the associated singularity lies on another branch of the free energy F(z). Neither is demonstrated. The branch-cut speculation is close to circular, since locating singularities of F(z) is exactly what the saddle-point computation is supposed to determine. This is a genuine gap, not a technical nitpick. The authors seem aware of it, and even call understanding the obstruction a future problem, but the central result of the paper depends on it.\n\nTwo secondary issues: the Δ → 0 vacuum-block limit is assumed rather than shown (footnote 5), and the analytic continuation of the Young-tableaux integral from the real cone to complex saddles is asserted with a soft 'we assume no subtle obstruction.' These are probably fixable, but they belong on the list for revision. The citation pattern is fine; the heavy reliance on [19] is appropriate since that is where the Lee-Yang analogy originates, and one of the authors is a coauthor there.\n\nWho is this for: anyone working on ETH in 2D CFTs, large-c conformal blocks, or AGT instanton counting. It deserves a serious referee, but I would send it back demanding that the symmetric saddles be dealt with—either by an actual thimble computation, by a sharpened statement of what is conjectured, or by softening the claim. As written, the central equality is conditional.","headline":"A serious, inventive paper that recasts the forbidden singularity as a critical fugacity in an AGT instanton gas, but the central equality r = z* depends on an unproved exclusion of competing saddle-points in Section IV.D.","tokens_in":35071,"tokens_out":1796,"would_cite":true,"duration_ms":22221,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The forbidden singularity of CFT thermalization is an instanton-gas phase transition.","keywords":["eigenstate thermalization","forbidden singularities","Virasoro conformal blocks","heavy-light limit","AGT correspondence","Nekrasov instanton partition function","instanton gas","Lee-Yang phase transition"],"falsifier":"One could settle the claim by computing the Lefschetz thimbles of the symmetric saddle-point solutions (90) and checking whether any intersects the original contour (70); a single intersection would put a singularity with radius $r < z_*$ on the principal branch. Numerically, one can also extract the radius of convergence of the finite- but large-$c$ vacuum block from Zamolodchikov recursion and look for a first singularity below $z_* = 1 - e^{-2\\pi/\\alpha_H}$.","tokens_in":34021,"feed_emoji":"⚛️","tokens_out":10151,"duration_ms":91822,"temperature":0.7,"pith_summary":"In two-dimensional CFTs, eigenstate thermalization predicts apparent singularities in heavy-light correlators — the 'forbidden singularities' — that are not true singularities of the underlying conformal block. This paper identifies the first such singularity, $z_* = 1 - e^{-2\\pi/\\alpha_H}$, with the critical fugacity of an emergent instanton gas. Via the AGT correspondence, the heavy-light vacuum Virasoro block is recast as the grand canonical partition function of instantons in an $\\mathcal{N}=2$ supersymmetric $SU(2)$ gauge theory. At criticality the gas undergoes instanton proliferation, controlled by a complex saddle point of string-like Young tableaux, analogous to a Lee-Yang transition. If the identification is right, a signature of eigenstate thermalization becomes a true phase transition in eigenstates.","feed_headline":"The forbidden singularity in CFTs is an instanton critical point","feed_subtitle":"The heavy-light Virasoro block breaks down exactly where the AGT instanton gas proliferates.","key_machinery":"The load-bearing object is the AGT correspondence, which equates a Virasoro conformal block with the Nekrasov instanton partition function of an $\\mathcal{N}=2$ supersymmetric $SU(2)$ gauge theory on the $\\Omega$-background, mapping the cross-ratio $z$ to the instanton fugacity. In the heavy-light limit the central charge $c$ becomes the volume of the instanton gas, and the large-$c$ reduction of the instanton counting formula yields the effective action $I_0(Y_1,Y_2)$ for two Young tableaux. The analysis then passes to the large-$\\nu$ statistical theory with the anisotropic ansatz $Y = \\nu\\, y$ (string-like tableaux); the saddle-point equation $\\Delta^{1,1}_X / \\Delta^{1,1}_{X-1} = e^{-\\lambda} (X - i\\alpha_H)/(X+1)$, together with normalization and boundary conditions, produces the discrete family $e^{-\\lambda_n} = 1 - e^{-2\\pi n/\\alpha_H}$. The $n=1$ value sets the critical fugacity $z_*$.","core_discovery":"The paper's central claim is that the forbidden singularity at $z_* = 1 - e^{-2\\pi/\\alpha_H}$ in the heavy-light vacuum Virasoro block is exactly the critical fugacity $r$ of a phase transition in the AGT-dual instanton gas. The transition is tied to the non-commutativity of the limits $c \\to \\infty$ and $\\nu \\to \\infty$: after taking the large-central-charge limit first, the instanton sum is governed by a $c$-independent effective action $I_0(Y_1,Y_2)$ for a pair of Young tableaux. At large instanton number $\\nu$, the dominant configuration is a string-like pair growing as $Y = \\nu y$, and the saddle-point equations admit a complex solution with $y^2_X = 0$ and $y^1_X$ built from binomial series in $\\alpha_H$. The boundary condition $y^1_\\infty = 0$ forces $e^{-\\lambda_n} = 1 - e^{-2\\pi n/\\alpha_H}$, and the $n=1$ saddle gives the radius of convergence of the fugacity expansion, reproducing $z_*$. Below $z_*$ the gas is dilute with $c$-independent free energy; above it the dominant string-like configurations have actions $\\propto c\\,\\ln\\nu$, and the same machinery reproduces the true OPE singularity at $z=1$.","pith_inferences":["Not in the paper: perturbing the heavy-light ratio while keeping $c$ large should shift $z_*$ and the critical fugacity together; measuring that shift from the finite-$c$ block would test the robustness of the identification.","Not in the paper: if the Lefschetz-thimble exclusion is right, the symmetric saddle-points should still contribute as trans-series terms somewhere in the complex-$c$ plane; a resurgence analysis in $1/c$ could reveal them.","Not in the paper: the string-like Young-tableau saddle resembles the thermodynamic limit of a one-dimensional chain, suggesting a possible integrable/TBA description of the phase transition.","Not in the paper: in the gauge-theory language the transition is a divergence of the average instanton number, so a localization-based computation of $\\langle \\nu \\rangle$ as a function of $z$ could probe the transition without mentioning CFT."],"forward_implications":["The first forbidden singularity $z_* = 1 - e^{-2\\pi/\\alpha_H}$ is a genuine critical point of the instanton gas rather than an artifact of the large-$c$ limit, with the central charge acting as the system volume.","The infinite tower of forbidden singularities $z_n = 1 - e^{-2\\pi n/\\alpha_H}$ reappears as an infinite family of saddle-point fugacities, connecting the vacuum block to the un-physical blocks.","At finite but large $c$, the sharp transition is smoothed into a condensation of Virasoro-block zeros along an anti-Stokes curve, in direct analogy with Lee-Yang zeros.","Above $z_*$, the instanton gas is dominated by string-like Young tableaux with action $\\propto c\\,\\ln\\nu$, and the same effective theory reproduces the heavy-light behavior at the true singularity $z=1$.","The two limits $c \\to \\infty$ and $\\nu \\to \\infty$ do not commute, so the forbidden singularity encodes the non-commutativity of the thermodynamic and large-central-charge limits."],"supporting_citations":[{"why":"Supplies the monodromy method and the leading-order accessory parameter whose poles are the forbidden singularities $z_n=1-e^{-2\\pi n/\\alpha_H}$.","marker":"[14]"},{"why":"Provides the finite-c zero condensation, branch-cut resolution, and numerical evidence that motivate the Lee-Yang phase-transition picture.","marker":"[19]"},{"why":"Introduces the AGT correspondence identifying Virasoro blocks with four-dimensional instanton partition functions.","marker":"[21]"},{"why":"Defines the Nekrasov instanton partition function that forms the gauge-theory side of the correspondence.","marker":"[22]"},{"why":"Gives the Young-tableaux expansion of Nekrasov functions used as the micro-state representation of the instanton gas.","marker":"[23]"},{"why":"Supplies Lee-Yang theory of condensation of zeros that structures the phase-transition interpretation.","marker":"[32]"},{"why":"Completes the Lee-Yang framework for zeros and phase transitions that the paper's analogy relies on.","marker":"[33]"},{"why":"Motivates the large-$\\nu$ string-like scaling ansatz $Y=\\nu y$ used in the saddle-point analysis.","marker":"[38]"},{"why":"Gives the heavy-light $z\\to1$ behavior of the Virasoro block that the high-fugacity phase must reproduce.","marker":"[20]"}],"fun_headline_variants":["Forbidden singularity is an instanton critical point","Instanton proliferation drives CFT phase transition","Eigenstate thermalization meets instanton criticality","Heavy-light CFT limit exposes instanton gas transition","CFT forbidden singularity from instanton gas deconfinement"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The conclusion that the radius of convergence is exactly $z_*$ rests on excluding the symmetric saddle-point solutions of Section IV.D, which have larger $\\mathrm{Re}\\,\\lambda$ and would imply $r < z_*$, on the grounds that their Lefschetz thimbles do not meet the original integration contour; the paper asserts this exclusion but does not prove it.","fun_headline_variants_meta":{"raw":{"variants":["Forbidden singularity is an instanton critical point","Instanton proliferation drives CFT phase transition","Eigenstate thermalization meets instanton criticality","Heavy-light CFT limit exposes instanton gas transition","CFT forbidden singularity from instanton gas deconfinement"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001165,"raw_usage":{"total_tokens":4871,"prompt_tokens":1041,"completion_tokens":3830,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":657,"completion_tokens_details":{"reasoning_tokens":3754}},"tokens_in":657,"tokens_out":3830,"duration_ms":24393,"temperature":1.0,"reasoning_tokens":3754,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:26:38.830006+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"One could settle the claim by computing the Lefschetz thimbles of the symmetric saddle-point solutions (90) and checking whether any intersects the original contour (70); a single intersection would put a singularity with radius $r < z_*$ on the principal branch. Numerically, one can also extract the radius of convergence of the finite- but large-$c$ vacuum block from Zamolodchikov recursion and look for a first singularity below $z_* = 1 - e^{-2\\pi/\\alpha_H}$.","supporting_citations":[{"cited_title":"Cardy, Quantum quenches to a critical point in one di- mension: some further results, Journal of Statistical Me- chanics: Theory and Experiment2016, 023103 (2016)","cited_arxiv_id":null,"evidence_quote":"Supplies the monodromy method and the leading-order accessory parameter whose poles are the forbidden singularities $z_n=1-e^{-2\\pi n/\\alpha_H}$."}],"review_version":1}