{"id":"7ad1f4fc-c81a-4133-a2d7-2417844801f9","arxiv_id":"2412.05038","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Fermionic localization makes the BMS3 torus partition function one-loop exact for irrational or imaginary θ and yields a factorized rational-θ answer.","lead":"Using supersymmetric localization, this paper computes the torus partition function of 3d gravity with zero cosmological constant on the BMS3 coadjoint orbit. For most states the earlier one-loop answer is shown to be exact; for special rational values of the angular potential a closed form is obtained up to an infrared divergent factor.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Exactness result (4.54) rests on an unproved contour deformation: with A_mn pure imaginary under the chosen coefficients, (QF)bosonic is not real on the original field space, so condition (2.6) is not met without a justified complex rotation.","rationale":"The reader's weakest assumption identifies exactly the load-bearing gap: the localization formula (2.9) requires a real positive-definite (QF)bosonic, but the constructed term is complex on the original field space. My independent reading of (4.37)-(4.48) confirms this. The entire exactness proof for irrational and purely imaginary θ reduces to the validity of the contour deformation, and the paper supplies only a heuristic Remark. This is a correctness risk rather than a fatal inconsistency: if the contour deformation can be justified (e.g., by showing the rotated contour is homotopic with vanishing boundary terms and that the superdeterminant is invariant), the result would stand. Since the reader already imposed a CONDITIONAL verdict on this basis, I recommend no change. I do not elevate to REJECT because the one-loop match in (4.22) and the AdS3 consistency check in Section 3 provide independent support, and the missing step is a gap that could be filled rather than a demonstrated contradiction. The rational-θ IR divergence is also acknowledged by the authors and is secondary to the main exactness claim.","tokens_in":20741,"tokens_out":5015,"duration_ms":55066,"concrete_test":"Compute for a single mode with B>0 and A=ic (c real) the Gaussian integral I = ∫_{R^2} dε dα exp(−Bε² − ic εα) on the original real contour. Integrating α first gives 2π/|c|; integrating ε first gives √(4πB)/|c|; the discrepancy shows the original integral is not well-defined without a contour prescription. Then repeat with the rotated contour ε → e^{iπ/4}ε (the diagonal basis of (4.48)) and verify that the result is unique, matches the product formula (4.52), and has no boundary contribution at infinity. If the rotated integral differs or boundary terms appear, Eq. (4.54) is not established.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The localization claim in Section 4.2.2 hinges on Eq. (4.48)-(4.54). The quadratic form (4.37)-(4.39) has A_mn pure imaginary under the positivity choices (4.42) and (4.45), so (QF)bosonic is not real on the original integration domain E_mn in R^4. Hence the bosonic part is not >= 0 as required by (2.6), and the standard localization theorem does not directly apply. The Remark after (4.48) asserts that one deforms the contour to the real basis of the diagonalized matrix, but no proof is given that (i) the deformation passes through a sector where the full interacting action decays, (ii) there are no boundary contributions at infinity in the infinite-dimensional field space, or (iii) the superdeterminant ratio is unchanged by the rotation. The claimed equivalence with a Fourier-transform perspective is also not demonstrated for the original contour: for a pure-imaginary cross term the integral is conditionally convergent and order-dependent. Without this contour argument, Eq. (4.54) is a formal Gaussian evaluation in a complex basis rather than a consequence of localization, so the central one-loop-exactness conclusion is not established as written.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper computes the torus partition function for the geometric action of 3d flat gravity on a constant coadjoint orbit of \\hat{BMS}_3 using fermionic localization. After constructing a Q-exact localization term with a general ansatz and imposing Q-closure and positivity on the quadratic fluctuation matrix, the authors evaluate the localized path integral and obtain Z = e^{-S_0} \\prod_{m,n}(m-n\\theta)^{-1} for irrational or purely imaginary \\theta, matching the perturbative one-loop result (4.22). For rational \\theta they find Z = N Z_{1\\text{-loop}} with N = \\prod_{n\\theta\\in\\mathbb{Z}}(M_0+n^2)^{-1/2}, up to an IR-divergent zero-mode integral. The same technique is first applied to the AdS_3/Virasoro case, reproducing the known one-loop exactness.","tokens_in":20998,"tokens_out":3169,"duration_ms":236960,"significance":"If established, the BMS_3 result would constitute a nontrivial exactness statement for the flat-space gravitational path integral and would extend fermionic localization to a phase space for which no Kähler structure is known. A notable strength is that the localization-term coefficients are not tuned to match the one-loop answer; the superdeterminant computation reproduces (4.22) independently. The rational-\\theta analysis also goes beyond the irrational case by exhibiting a partially resummed result. However, the central derivation is formal: the positivity condition required by the localization theorem is not verified on the actual integration domain, and the contour deformation invoked to repair this is not proved.","major_comments":[{"comment":"The load-bearing condition (2.6) is not satisfied on the original real integration domain. Under the choices made in (4.42) and (4.45), the coefficient A_{mn} in (4.38) is purely imaginary, so the expression (4.39) written in terms of the real modes E_{mn} is not real, let alone nonnegative. Thus the matrix M_{mn} in (4.40) cannot be regarded as a positive quadratic form on the original \\mathbb{R}^4 field space. The diagonalization leading to (4.48) implicitly performs a complex linear change of variables, but this is precisely the point that must be justified, not assumed.","section":"Section 4.2.1, Eqs. (4.37)-(4.48)"},{"comment":"The paper states that the contour is deformed from E_{mn}\\in\\mathbb{R}^4 to \\tilde{E}_{mn}\\in\\mathbb{R}^4 and that this is equivalent to a Fourier-transform perspective, but no proof is supplied that the deformation has no boundary contributions at infinity, that it passes through a region where the full interacting action is decaying, or that the superdeterminant ratio is invariant under the rotation. For a purely imaginary cross term, the corresponding Gaussian integral is conditionally convergent and order-dependent, so the equivalence with a Fourier transform is not automatic. Without this argument, Eq. (4.54) is a formal Gaussian evaluation in a complexified basis rather than a consequence of the localization theorem.","section":"Remark after Eq. (4.48)"},{"comment":"For rational \\theta the localization manifold is infinite-dimensional and the final expression contains the IR-divergent factor \\int \\prod_{n\\theta\\in\\mathbb{Z}} d\\alpha_n. The comparison Z = N Z_{1\\text{-loop}} in (4.64) therefore requires a prescription for regularizing this divergence, but no such prescription is given. The statement in Section 5 that after stripping off the IR divergence the remaining finite factor is meaningful is not made precise. This does not invalidate the irrational-\\theta claim by itself, but it leaves the rational-\\theta result at the level of a formal statement.","section":"Section 4.2.2, Eqs. (4.60)-(4.65)"}],"minor_comments":[{"comment":"The infinite products over m,n in (4.22) and (4.54) are not absolutely convergent and are implicitly zeta-regularized, but the precise regularization scheme is not specified. A short remark on how the products are defined would improve rigor without changing the known one-loop results.","section":"Section 4.1, Eq. (4.22) and Section 4.2.2, Eq. (4.54)"},{"comment":"The path-integral measure is fixed only up to dimensionless numerical factors, and the choice L = k^{-1} is made without discussing scheme dependence of the final rational-\\theta factor N. It would be helpful to state explicitly which factors in (4.65) are scheme-independent.","section":"Section 4.1, Eq. (4.21)"},{"comment":"The sentence 'the two perspectives are equivalent and lead to same result' should cite or outline a condition under which complex contour rotation and Fourier transformation yield identical regularized integrals; as written this is an assertion rather than a demonstrated equivalence.","section":"Remark after Eq. (4.48)"},{"comment":"The summary says a Kähler metric was used in [10,28] for the constructions, whereas the main text constructs QF without relying on such a metric; this wording is slightly misleading and should be adjusted.","section":"Section 5"}],"recommendation":"major_revision","confidential_remarks":"The referee's main concern is the unproved contour deformation in Section 4.2.1. This is a technical gap rather than an obvious fatal error, because the paper is explicit about the deformation being needed and the rest of the computation is internally consistent. I would therefore recommend major revision rather than rejection: the authors should either prove the deformation (including absence of boundary terms and invariance of the superdeterminant) or replace the localization claim by a weaker statement that the Gaussian evaluation reproduces the one-loop result, and then justify the word 'exact' separately. The rational-\\theta IR divergence also needs a cleaner treatment before the comparison in Eq. (4.64) can be called a resummation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear Colleague,\n\nYou asked about Simón–Yu, BMS3 fermionic localization. My take: this is a serious computation that likely gets the right answer, but the central exactness proof is not complete as written. The paper's main new contribution is a fermionic localization derivation of the torus partition function for constant coadjoint orbits of BMS3, extending the vacuum one-loop exactness of Cotler et al. [41] to conical defects and flat space cosmologies. The localization term is constructed by ansatz, the coefficients are chosen to satisfy positivity conditions, and the resulting superdeterminant reproduces the known one-loop formula (4.22). That cross-check is genuinely good, and the rational-θ analysis leading to the factor N in (4.65) is new and interesting, even if it carries an IR divergence.\n\nWhere I have trouble is the localization argument for the irrational and purely imaginary θ cases. The bosonic part of the Q-exact term (QF)_bosonic has pure imaginary cross terms (A_mn imaginary) under the chosen positivity conditions, so it is not real, let alone non-negative, on the original real integration domain. Condition (2.6) of their own review is therefore not met. The Remark after (4.48) asserts a contour deformation to the eigenbasis of M_mn, but there is no proof that the deformation has no boundary contributions, preserves the measure, and leaves the superdeterminant unchanged. The stress-test is right on this point. It is not fatal, because the Fourier-transform alternative they sketch—integrating out α_mn first, which gives δ(ε_mn)—is a plausible way to justify the result, and the final answer matches the independent one-loop computation. But as written, the claim that (4.54) \"proves\" one-loop exactness is overstrong. The authors should either make the contour argument rigorous (or at least careful) for infinite-dimensional field space, or soften the claim and state the Fourier-transform delta-function argument as the justification.\n\nThe rational-θ result (4.63)-(4.65) is also not fully under control: it has an IR-divergent factor ∫dα_n, and the prefactor N depends on M0 but is independent of the torus moduli. The paper acknowledges this. The 1-loop comparison (4.26) itself is only at a single saddle, not the full sum, so the claim of \"resumming\" all saddles is suggestive but not demonstrated.\n\nOverall: the paper is technically serious, well organized, and the core result is likely correct. It deserves a serious referee and publication after the localization gap is addressed. The abstract should be toned down from \"exact\" to \"argued exact\" or \"conditionally exact\" until the contour issue is settled.","headline":"Plausible and likely correct, but the central localization proof rests on an unproved contour deformation; deserves a serious referee and a request for revision.","tokens_in":21538,"tokens_out":8124,"would_cite":true,"duration_ms":74040,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Fermionic localization on BMS3 orbits computes the exact flat 3d gravity torus partition function, proving one-loop exactness for irrational or purely imaginary angular potential and reducing rational angles to a modular-independent factor.","keywords":["fermionic localization","BMS3","flat space gravity","coadjoint orbit","torus partition function","one-loop exactness","geometric action","superdeterminant"],"falsifier":"Compute the first perturbative correction beyond one loop in the BMS3 geometric action at a saddle with irrational $\\theta$; if any finite two-loop term survives, the exact partition function cannot equal the one-loop result. A more direct check is to evaluate the finite-mode Gaussian integral (4.39) on the original real contour for one Fourier mode and compare it with the result obtained after diagonalizing $M_{mn}$; any discrepancy would show that the contour deformation used in (4.48) changes the value of the path integral.","tokens_in":20519,"feed_emoji":"🧮","tokens_out":17765,"duration_ms":171131,"temperature":0.7,"pith_summary":"The paper asks whether the one-loop approximation to the torus partition function of three-dimensional pure gravity with zero cosmological constant is actually exact. Working with the geometric action on a coadjoint orbit of the centrally extended BMS3 group labelled by constant charges, the authors build a fermionic localization term and evaluate the path integral exactly. For irrational or purely imaginary angular potential $\\theta$, the exact result coincides with the one-loop determinant, $Z = e^{-S_0}\\prod_{m,n}(m-n\\theta)^{-1}$, proving one-loop exactness. For rational $\\theta$, the exact partition function equals the one-loop result around the constant saddle times a factor $N=\\prod_{n\\theta\\in\\mathbb{Z}}(M_0+n^2)^{-1/2}$ that is independent of the continuous torus modular parameter. This matters because it turns a perturbative gravitational computation into a closed-form statement about the full quantum theory.","feed_headline":"For generic angular potential, flat 3d gravity is one-loop exact","feed_subtitle":"Localization on BMS3 orbits gives the full torus partition function; rational values differ by a modular-independent factor.","key_machinery":"The carrying mechanism is the $Q$-exact localization term $Q_F=\\int Q(\\psi_f(D_1\\epsilon+D_2\\tilde\\alpha)+\\psi_\\alpha(D_3\\epsilon+D_4\\tilde\\alpha))$ built from the fermionic symmetry $Q\\epsilon=\\psi_f$, $Q\\tilde\\alpha=\\psi_\\alpha$, $Q\\psi_f=-i\\partial_y\\epsilon$, $Q\\psi_\\alpha=\\epsilon'-i\\partial_y\\tilde\\alpha$. The conditions $Q^2F=0$ and positive-definite bosonic part are solved by $D_2=D_3$, $D_4=0$ and by tuning the constant coefficients in the first-order operators $D_1,D_2$; the resulting quadratic form on Fourier modes has pure imaginary cross terms, so positivity is achieved after a complex rotation of the real integration variables. In the localized integral, the bosonic determinant and the ghost determinant cancel pairwise, leaving the superdeterminant ratio $\\prod_{m,n}(m-n\\theta)^{-1}$, the one-loop result. For rational $\\theta$ the same term with $a_4\\neq 0$ selects a smaller localization locus on which the Euclidean action is constant, reducing the remaining path integral to a Gaussian integral.","core_discovery":"On its own terms, the discovery is that fermionic localization applies to the constant-representative coadjoint orbit of $\\widehat{\\mathrm{BMS}}_3$ even though no Kähler structure on that phase space is known. The localization locus is unique when $\\theta$ is irrational or purely imaginary, and the superdeterminant of the deformation operator factors so that all dependence on the arbitrary localization parameters cancels, leaving $Z=e^{-S_0}\\prod_{m,n}(m-n\\theta)^{-1}$. This is exactly the perturbative one-loop partition function computed earlier, so the one-loop calculation is exact for those values of $\\theta$. When $\\theta$ is rational, a choice $a_4\\neq 0$ in the localization term collapses the localization locus to zero superrotation fluctuation and arbitrary supertranslation modes; the integral over the remaining locus is trivial because the Euclidean action is constant there, and the final result is $Z = N Z_{\\text{1-loop}}$ with $N=\\prod_{n\\theta\\in\\mathbb{Z}}(M_0+n^2)^{-1/2}$, independent of the continuous modular parameter. The rational case thus resums the infinite family of saddles into a single closed factor.","pith_inferences":["The authors do not pursue it, but $N$ may encode the rational saddle sum in (4.27) as a reduced determinant on the zero-mode sublattice; if that can be made precise, the rational partition function would be a fully closed product rather than an IR-divergent integral.","One testable extension is to run the same $Q_F$ ansatz for higher-spin or supersymmetric flat-space gravity, where the algebraic constraints $D_2=D_3$, $D_4=0$ and positivity of the bosonic quadratic form should still be solvable without a Kähler structure.","If the contour-deformation assumption is eventually proved, the exact partition function becomes a sharp non-perturbative test for proposed dual descriptions of asymptotically flat 3d gravity: any candidate must reproduce the same $\\theta$-dependence."],"forward_implications":["For irrational or purely imaginary $\\theta$, all higher-loop corrections to the torus partition function on the constant BMS3 orbit cancel; the one-loop determinant is the complete answer.","For rational $\\theta$, the exact partition function is the one-loop result around the constant saddle multiplied by $N=\\prod_{n\\theta\\in\\mathbb{Z}}(M_0+n^2)^{-1/2}$, so the full saddle family contributes only through a factor independent of the continuous torus modular parameter.","The localization term is constructed algebraically rather than from a Kähler metric, so the same ansatz can be used for other coadjoint orbits or symmetry groups without a known positivity structure.","One-loop exactness now covers conical-defect and flat-cosmology saddles with real irrational or purely imaginary $\\theta$, not only the Minkowski vacuum."],"supporting_citations":[{"why":"Supplies the geometric-action and coadjoint-orbit formulation for Virasoro/AdS3 and the localization computation that the BMS3 argument is modeled on.","marker":"[10]"},{"why":"Supplies the BMS3 geometric action, one-form, symplectic form and constant-orbit data reproduced in (4.6)-(4.9).","marker":"[14]"},{"why":"Supplies the perturbative one-loop torus partition function and BMS3 character that (4.54) must match.","marker":"[22]"},{"why":"Supplies the induced-representation BMS3 character and one-loop flat-gravity partition function used as the comparison target.","marker":"[30]"},{"why":"Duistermaat-Heckman theorem, the basis for the claim that Q-exact Q-closed deformations leave the partition function unchanged.","marker":"[33]"},{"why":"Supplies the localization formula (2.9) reducing the path integral to a superdeterminant on the fixed-point locus.","marker":"[42, 43]"}],"fun_headline_variants":["BMS3 fermionic localization gives exact flat 3d gravity partition","Flat 3d gravity one-loop exact via BMS3 fermionic localization","Exact torus partition from BMS3 fermionic localization","Fermionic localization exact for flat 3d gravity without Kahler","No Kahler? BMS3 localization still yields exact flat 3d gravity"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The exactness result assumes that rotating the integration variables to make the localization term positive leaves the value of the path integral unchanged, even though the bosonic integrand is not positive on the original real fields and the paper does not prove that the rotated contour has no boundary contribution or preserves the measure.","fun_headline_variants_meta":{"raw":{"variants":["BMS3 fermionic localization gives exact flat 3d gravity partition","Flat 3d gravity one-loop exact via BMS3 fermionic localization","Exact torus partition from BMS3 fermionic localization","Fermionic localization exact for flat 3d gravity without Kahler","No Kahler? BMS3 localization still yields exact flat 3d gravity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001309,"raw_usage":{"total_tokens":5272,"prompt_tokens":820,"completion_tokens":4452,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":436,"completion_tokens_details":{"reasoning_tokens":4355}},"tokens_in":436,"tokens_out":4452,"duration_ms":35052,"temperature":1.0,"reasoning_tokens":4355,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T20:58:01.146872+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the first perturbative correction beyond one loop in the BMS3 geometric action at a saddle with irrational $\\theta$; if any finite two-loop term survives, the exact partition function cannot equal the one-loop result. A more direct check is to evaluate the finite-mode Gaussian integral (4.39) on the original real contour for one Fourier mode and compare it with the result obtained after diagonalizing $M_{mn}$; any discrepancy would show that the contour deformation used in (4.48) changes the value of the path integral.","supporting_citations":[],"review_version":1}