{"id":"1d07c269-c257-4a9f-b4cb-89d2810cfcb6","arxiv_id":"1908.03974","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The singular part of a one-loop four-point Witten diagram with a double-particle cut equals a product of tree-level subdiagram coefficients divided by a mean-field-theory coefficient.","lead":"This paper studies loop corrections in anti-de Sitter space and shows that their singular parts factorize into products of simpler tree-level diagrams, as in flat space. It provides a bulk derivation of large-N conformal field theory relations and introduces AdS Cutkosky rules for computing loop corrections.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Locality assumption in Sec. 4.1 restricts the factorization (4.12)-(4.14) to local bulk theories; non-local higher-spin application remains conditional.","rationale":"The paper is a careful, technical derivation of the bulk-side origin of the large-N factorization relations for double-particle cuts. The reader's verdict CONDITIONAL rests on the locality assumption for the b-factor singularities, and the stress-test confirms that this is the weakest point in the argument. The derivation leading to the factorization (4.12)-(4.14) relies on the analytic structure of the spectral integral (3.10), which is determined by the singularities of IL and IR in ν1 and ν2. The paper argues that these are exactly the b-factor poles for local theories, but explicitly acknowledges that non-local theories can generate additional singularities from infinite sums of derivative terms. Section 6.5 then applies the formalism to higher-spin theories, which are known to be non-local, and the paper itself flags this as a potentially tricky step. Since the paper aims to provide a general bulk derivation and lists higher-spin theories as an application, the locality assumption is genuinely load-bearing. If it fails, the pinching analysis and the factorization formulas would need modification. The paper does not resolve this, so a CONDITIONAL verdict is appropriate. No change to the reader's verdict is needed; the concern is real but already captured.","tokens_in":27522,"tokens_out":10290,"duration_ms":104518,"concrete_test":"Construct a concrete non-local bulk theory, e.g. a quartic vertex with a pseudo-local operator V(□) acting at the vertex (a finite-order polynomial in □ with order k, then study k→∞, or a higher-spin vertex from the construction in ref. [33]). Compute the tree-level four-point coefficient function IL(ν1,ν2;ν) in the split representation. Check whether IL has poles or branch cuts in ν1 or ν2 beyond those of b(h-iν, h+iν1, h+iν2). If additional singularities appear, repeat the contour-pinching analysis of Section 4.1 for the integral (3.10) and verify whether the double-trace residues still satisfy (4.12). If the residues differ, the factorization requires modification for non-local theories.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the singular part of a one-loop four-point amplitude factorizes into on-shell tree-level subamplitudes (4.7)-(4.14) rests on the analytic structure of the integral (3.10) in the spectral parameters ν1 and ν2. Section 4.1 asserts that all singularities of the tree-level coefficient functions IL and IR in these parameters are produced by the three-point b-factors (4.1), because derivative interactions contribute only polynomial factors, e.g. (4.2), which cannot create new poles. The paper explicitly notes the caveat: 'unless we are dealing with a non-local theory, in which such polynomial terms may sum up to a singularity.' This caveat is not academic. Section 6.5 applies the formalism to higher-spin theories, which are holographically reconstructed as non-local in a conventional sense (refs. [32,63]), and states that the extension to off-shell external lines is assumed to have 'the same analytic structure in ν1 and ν2 — see (4.1)'. If this assumption fails, the pinching analysis of Section 4.1 would have to include additional singularities, changing the residue computation (4.7) and hence the factorization (4.12)-(4.14). The paper's stated scope (abstract, Section 6.5) includes such non-local theories, so the central claim is conditional on an unproven locality assumption for an important class of applications. This is the most load-bearing unresolved point.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper analyzes singularities of one-loop four-point Witten diagrams in Euclidean AdS using the conformal partial wave expansion. After expressing the loop diagram as a double spectral integral over tree-level subdiagrams via the split representation and the bubble integral (Eqs. (3.1)-(3.10)), the author studies pinch singularities of the integral and extracts the leading double-trace residues. The main result (Eqs. (4.7)-(4.9), translated to (4.12)-(4.14)) states that the singular part of the loop coefficient function factorizes into on-shell tree-level coefficient functions divided by the mean-field conformal block coefficient. The paper then compares this with the large-N OPE analysis on the boundary and discusses extensions to higher loops, spinning fields, AdS Cutkosky rules, and higher-spin theories.","tokens_in":27770,"tokens_out":12735,"duration_ms":123328,"significance":"If correct, this provides a purely bulk derivation of factorization relations that have been used extensively in the large-N bootstrap and holographic loop computations (e.g., [7,17,22]). The derivation is not circular: it uses analytic structure of spectral integrals rather than the large-N CFT relations as input, and a_M is a kinematical quantity, not a fitted parameter. The paper is careful to state what is captured (\"less singular terms\") and what is not (single-trace contributions, regular terms). The main limitation is that the proof relies on a locality assumption that is not verified for the non-local higher-spin theories discussed in Sec. 6.5; this conditionality is acknowledged in the text but not resolved.","major_comments":[{"comment":"The factorization in (4.7)-(4.14) is derived under the assumption that all ν1,ν2-singularities of I_L and I_R are those of the three-point b-factors. The text correctly observes that derivative interactions are harmless in local theories, but explicitly allows that in non-local theories polynomial factors may sum to a singularity. Section 6.5 applies the formalism to higher-spin theories, which are non-local in the conventional sense (refs. [32,63]), and merely assumes that the off-shell extension has \"the same analytic structure in ν1 and ν2\". If this assumption fails, the pinching analysis in Section 4.1 and the residue formula (4.7) must be modified. The manuscript should either prove the required analytic structure for the non-local cases it claims to cover or restrict the statement of the main result to local bulk theories and present the higher-spin application as conditional. As written, the abstract's general claim of factorization goes beyond what is demonstrated.","section":"Section 4.1, Eqs. (4.1)-(4.7); Section 6.5"},{"comment":"The status of shadow singularities in the proposed AdS Cutkosky rule is not fully specified. The text says that the explicit term also has shadow double-trace singularities that \"should be ignored\", and Appendix A shows for one toy integral that the deformed contour reproduces the physical residues despite extra shadow poles. However, no general prescription is given for identifying and removing the shadow contributions when (4.7)/(4.9) is used as an input to a loop computation, and Section 6.2 explicitly lists this removal as an open problem. Because the paper advertises (6.4) as the AdS version of Cutkosky rules, the rule is not yet a complete algorithm for computing the singular part of a general amplitude; this should be stated as a limitation rather than as a finished result.","section":"Section 4.2, Eq. (4.7); Section 6.2, Eq. (6.4)"}],"minor_comments":[{"comment":"Equation (4.13) appears to contain a typo: the right-hand side shows a^[0]_L times a^[0]_L, but consistency with (4.12) requires a^[0]_L times a^[0]_R.","section":"Eq. (4.13)"},{"comment":"The step from (3.7) to (3.8) is not shown in the text (\"This is a straightforward computation\"). Since this is a central formula, providing the intermediate Symanzik-star evaluation would improve reproducibility.","section":"Section 3.1, Eqs. (3.7)-(3.8)"},{"comment":"The phrase \"exact formula\" for (4.13) is potentially overstatement; the derivation includes \"less singular terms\", so the text should say explicitly that the formula is exact for the coefficient of the leading double-trace singularity in the non-degenerate case.","section":"Section 4.4"},{"comment":"The higher-loop and higher-point extension is presented schematically without a complete pinching analysis for the multi-integral case. The text should state explicitly that this is an extrapolation of the one-loop argument rather than a fully worked proof.","section":"Section 6.1"},{"comment":"The \"reasonable assumptions\" under which (4.13) applies to higher-spin theories should be enumerated; the footnote about non-locality and the phrase \"same analytic structure\" are too brief for a reader wanting to apply the result.","section":"Section 6.5"}],"recommendation":"major_revision","confidential_remarks":"The paper is a methods/justification paper rather than a new computational breakthrough, but it is well within JHEP scope and likely to be useful. The author is unusually transparent about the caveats (non-locality, shadow singularities). My main concern is that this transparency is not matched by a resolution: the higher-spin application in Sec. 6.5 is conditional on an unproven assumption about analytic structure, and the shadow-singularity issue leaves the advertised Cutkosky rule incomplete. I do not see any reason to doubt the local bulk derivation, so rejection is not warranted. A major revision that either proves or explicitly scopes the non-local and shadow-related claims would be appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this is the paper that puts the double-trace factorization of one-loop AdS diagrams on a general footing. The bubble case was already in Fitzpatrick–Kaplan, but Ponomarev derives it for an arbitrary one-loop four-point diagram with a double-particle cut, purely from the bulk side, and shows how to iterate it to higher loops and higher points. The final relations (4.13) and (4.14) are the ones people have been using in 1/N^4 bootstrap computations; here they come out of the analytic structure of spectral integrals rather than being assumed from CFT large-N logic.\n\nThe derivation is in good shape. The split representation plus bubble integral reduces the loop to a double spectral integral over tree-level coefficient functions; the pinching analysis is done carefully, with explicit statements of which singularities are captured and which are ignored. The appendix toy example is helpful. Section 5 shows the result is consistent with the large-N expansion on the boundary, but that consistency check is not an input to the bulk computation, so there is no circularity. The notation is heavy but the logic is clean.\n\nSoft spots: the locality assumption in Section 4.1 is the one to watch. The argument that derivative interactions only add polynomial factors in nu_1 and nu_2, and therefore no new poles, is fine for local theories. The paper itself notes it can fail for non-local theories. Then Section 6.5 applies the formalism to higher-spin gravity, which is known to be non-local in the standard sense, with the assumption that the off-shell extension has the same analytic structure. That makes the higher-spin application genuinely conditional. The author says ‘this may be a tricky step’ and leaves it to future work, so it is not a hidden flaw, but anyone using (4.13) for higher-spin loops needs to know this has not been established. Also, one step in the derivation of (3.8) is dismissed as ‘straightforward’ and left to the reader; that is minor, but a referee might ask for the one-line version.\n\nOverall: solid methodology paper for local AdS theories. It does not change the paradigm, but it justifies and generalizes a widely used result. I would send it to peer review and expect it to pass with minor comments. The higher-spin part should be read as an open problem, not a proven application.","headline":"A solid bulk-side derivation of the double-trace cut factorization for one-loop Witten diagrams, with the higher-spin application still resting on a flagged locality assumption.","tokens_in":28307,"tokens_out":2388,"would_cite":true,"duration_ms":26395,"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 singular part of a one-loop Witten diagram with a two-particle cut is fixed entirely by on-shell tree-level subdiagrams, through a factorization identity that matches large-N CFT expectations.","keywords":["AdS/CFT","Witten diagrams","conformal partial waves","conformal blocks","loop amplitudes","large-N expansion","Cutkosky rules","double-trace operators"],"falsifier":"Evaluate a one-loop four-point amplitude in a bulk theory with a non-local vertex, e.g. an exponential of a d'Alembertian acting on the cut lines, and compare the residue at the double-trace location with the right side of (4.12). If the derivative factors produce additional contour pinches in $\\nu_1,\\nu_2$, the residue will differ and the factorization fails.","tokens_in":27302,"feed_emoji":"🌌","tokens_out":5824,"duration_ms":55977,"temperature":0.7,"pith_summary":"This paper argues that loop-level scattering amplitudes in anti-de Sitter space are not new data: cutting a one-loop four-point diagram along two internal lines leaves a product of tree-level amplitudes evaluated on shell, integrated over the cut particles' phase space. If true, AdS loop amplitudes inherit their singular structure from simpler tree-level building blocks, exactly as flat-space unitarity dictates. The paper derives the factorization identity in the conformal partial wave representation and checks that it reproduces the relations the boundary large-N expansion predicts.","feed_headline":"Loop singularities in AdS factor into on-shell tree amplitudes","feed_subtitle":"Cut a Witten diagram in two and its singular part is built from tree-level data, matching the large-N boundary expansion.","key_machinery":"The machinery is the conformal partial wave expansion of bulk amplitudes combined with the split representation of bulk-to-bulk propagators and the bubble integral formula for integrated products of conformal partial waves. The split representation turns a cut propagator into an integral over boundary points and spectral parameters, so cutting two lines expresses the loop as an integrated product of tree-level subamplitude expansions. The bubble integral then collapses the pair of integrated boundary points back to a single partial wave, leaving a double spectral integral whose contour-pinching singularities are evaluated by residues; those residues are the factorization identity.","core_discovery":"The central claim is that the singular part of a one-loop four-point scalar amplitude in Euclidean AdS, associated with a double-particle cut, factorizes: it equals the product of the on-shell tree-level subamplitudes obtained by cutting the two propagators, with the cut particles integrated over their AdS phase space. In conformal block language this reads $a^{(0)}_O = a^{(0)}_L a^{(0)}_R / a_M$ for generic external dimensions, and in the degenerate (identical-operator) case the second-derivative coefficient satisfies $a^{(2)}_O = a^{(1)}_L a^{(1)}_R / a_M$, with $a_M$ the mean-field-theory coefficient. The paper proves this by representing the loop diagram through its conformal partial wave expansion, analyzing the spectral integrals' pinching singularities, and computing residues; the same analysis yields AdS Cutkosky rules in which a propagator is replaced by $\\Pi_\\Delta - \\Pi_{d-\\Delta}$.","pith_inferences":["A testable extension: apply the same pinching analysis to a theory with an infinite-derivative (non-local) vertex; the paper's caveat predicts the factorization (4.12) can break exactly where derivative factors cease to be polynomial in the spectral parameters.","If the reconstruction-from-singularities program succeeds, bulk loop amplitudes could be assembled from their cut data plus Regge bounds, making direct two-loop integration unnecessary in practice.","For spinning fields, the mean-field sewing shortcut suggests $a^{(0)}_O = a^{(0)}_L a^{(0)}_R / a_M$ holds with spinning conformal block coefficients and the same inverse mean-field coefficient, but only for tensor structures present in mean field theory; explicit bubble integrals would settle the general case.","The shadow poles produced by the AdS Cutkosky replacement might be removed by continuing to Lorentzian signature or by monodromy projection; if they cannot be cleanly projected, the cutting rule needs refinement."],"forward_implications":["The double-cut singularity of any one-loop four-point scalar amplitude can be computed from tree-level data alone, without evaluating the full loop integral.","Generic-dimension double-trace coefficients obey $a^{(0)}_O = a^{(0)}_L a^{(0)}_R / a_M$; for identical external operators the second-derivative coefficient obeys $a^{(2)}_O = a^{(1)}_L a^{(1)}_R / a_M$.","The AdS cutting rule $\\Pi_\\Delta \\to \\Pi_\\Delta - \\Pi_{d-\\Delta}$ gives the singular part of a diagram, with shadow artifacts that still need projection out.","By iterating the bubble reduction, higher-loop and higher-point cuts reduce to multiple-trace singularities whose coefficients are built from tree-level data in the same factorized way.","On the CFT side, order-$1/N^2$ double-trace data determines the leading singular (double-trace) part of the four-point function at order $1/N^4$, consistent with large-N bootstrap relations."],"supporting_citations":[{"why":"Supplies the split representation for propagators and the conformal partial wave expansion of tree-level amplitudes, the starting point of the loop analysis.","marker":"[20]"},{"why":"Provides the exact bubble-diagram computation and the factorization of conformal block coefficients that the paper generalizes.","marker":"[21]"},{"why":"Establishes the one-loop/mean-field relation and the unitarity-based factorization for the bubble diagram that (4.13) extends.","marker":"[22]"},{"why":"Contains the bubble integral formula used to integrate products of conformal partial waves over two boundary points.","marker":"[34, 35]"},{"why":"Supplies the standard contour-pinching methods for analyzing singularities of spectral integrals.","marker":"[26]"},{"why":"The higher-spin one-loop computation whose assumptions the paper's factorization justifies.","marker":"[17]"}],"fun_headline_variants":["AdS loop cuts reduce to tree amplitudes","Cutting AdS loops yields tree-level factors","Loop singularities in AdS: factorized cuts","From AdS loops to boundary large-N via cuts","AdS Cutkosky rules: cuts give tree factors"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The tree-level coefficient functions $I_L$ and $I_R$ have all their $\\nu_1,\\nu_2$ singularities coming from three-point $b$-factors, so derivative corrections merely multiply by polynomials and create no new spectral poles; the paper notes this can fail for non-local theories.","fun_headline_variants_meta":{"raw":{"variants":["AdS loop cuts reduce to tree amplitudes","Cutting AdS loops yields tree-level factors","Loop singularities in AdS: factorized cuts","From AdS loops to boundary large-N via cuts","AdS Cutkosky rules: cuts give tree factors"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000182,"raw_usage":{"total_tokens":1259,"prompt_tokens":839,"completion_tokens":420,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":455,"completion_tokens_details":{"reasoning_tokens":346}},"tokens_in":455,"tokens_out":420,"duration_ms":4731,"temperature":1.0,"reasoning_tokens":346,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:55:28.870417+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Evaluate a one-loop four-point amplitude in a bulk theory with a non-local vertex, e.g. an exponential of a d'Alembertian acting on the cut lines, and compare the residue at the double-trace location with the right side of (4.12). If the derivative factors produce additional contour pinches in $\\nu_1,\\nu_2$, the residue will differ and the factorization fails.","supporting_citations":[],"review_version":1}