{"id":"e1816b26-065b-4684-a495-39e71eac7115","arxiv_id":"2411.17809","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Defects of continuously adjustable dimension p=2+δ are defined and analyzed in the O(N) model, yielding new interfaces and non-local 3d CFTs.","lead":"This paper introduces conformal defects whose dimension can be any continuous number, not just lines or surfaces, by using a new parameter δ alongside the standard ε expansion. It computes properties of such 'transdimensional defects' in the O(N) model and shows they interpolate between known defects and give access to new conformal theories.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The load-bearing extrapolation to δ=1 and δ=1−ε is not proven: f(δ) is a divergent-series analytic continuation with a pole at δ=-2/3, and Eq. (2.28)'s new-interface spectrum relies on the conjectured value f(1)=1/2 from Padé, so the headline claims should remain conditional.","rationale":"The reader's weakest assumption is exactly the one I would flag: the paper's perturbative series in δ are used at δ=O(1) where convergence is absent. I agree with CONDITIONAL. The strongest claim -- a new interface at δ=1−ε and a non-local 3d CFT at δ=1 -- depends on (i) f(1)=1/2, inferred from Padé (2.25) and explicitly flagged as 'tempting', and (ii) the rational resummation 1/(1+δ) for the geometric series, which at δ=1 is an analytic continuation of a divergent series. The paper gives clean free-field results and nontrivial cross-checks (Dirichlet at δ=1−ε in free theory, large-N matching at d=4 and d=6), which is real evidence for the framework, but those checks do not cover the interacting interface spectrum. A direct computation at finite δ would settle whether the resummations are the correct analytic functions. Because the concern is an unproven extrapolation rather than an internal inconsistency, the appropriate verdict is the reader's CONDITIONAL, unchanged.","tokens_in":22304,"tokens_out":6663,"duration_ms":60385,"concrete_test":"Compute the O(ε) anomalous dimension γ_φ(δ) directly from the exact integrals (A.16)-(A.21) without first expanding in δ, e.g. by numerical evaluation of the two-loop integrals in d=4−ε at fixed p=3−ε and p=3 for small ε (ε=0.01, 0.05, 0.1), and compare with (2.26) and (2.29). If the numerical results deviate from 1/(1+δ) at δ=1 beyond truncation error, the resummation fails. Separately, extend the β coefficients (A.26) to k=15 and form diagonal Padé/Borel approximants for f(1); if f(1) is not stable at 0.500±0.001, Eq. (2.28)'s claimed spectrum is not numerically fixed. The first test is the decisive one.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The manuscript's most consequential outputs are the 'new interface' spectrum (2.28) and the non-local 3d CFT dimensions (2.29), both obtained by evaluating first-order-in-ε results at δ=1−ε or δ=1. Those evaluations are not protected by the all-orders-in-δ computation. The function f(δ) in (2.24) is only known through ten terms; its series has radius 2/3 because of an apparent pole at δ=-2/3, and (2.25) assigns f(1)=0.4999(3) via Padé and then states 'It is tempting to conclude f(1)=1/2, and we use this value in what follows.' This is an inference, not a derivation. Likewise, Eq. (2.26) replaces Σ(-δ)^k by 1/(1+δ); at δ=1 that geometric series is divergent and the replacement is a choice of analytic continuation (Abel/Cesàro sum), not a proven property of the field theory. If the true functions have additional non-perturbative corrections in δ, the coefficient of ε in (2.28) and the lightest-singlet comparison ΔS<ΔD are not established. The paper itself flags the tentative nature of f(1), so the conditional verdict is appropriate; but as it stands the central claim of a genuinely new interface depends on an unproven resummation.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces the notion of a 'transdimensional defect', i.e. a conformal defect whose dimension p is a continuously variable parameter, and studies it in two main settings. In the free O(N) scalar theory in d=4−ε with a defect of dimension p=2+δ, the defect RG flow is solved exactly to all orders in δ and first orders in ε, giving the defect operator dimensions, one-point functions, and OPE coefficients. In the interacting O(N) vector model, the authors compute the defect beta function and anomalous dimensions to first order in ε and to high or all orders in δ, and then extrapolate to δ=1−ε (an interface in d=3−ε) and δ=1 (a three-dimensional defect in d=4−ε). In the former case they claim a new interface CFT that is not equivalent to two copies of the O(N) boundary CFT, and in the latter they claim a new non-local 3d CFT. A third section studies symmetry-breaking defects in the large-N O(N) model for 4<d<6, using δ as a regulator around p=(d−2)/2, and checks the results against known line and surface defect limits. The paper is self-contained and includes extensive Feynman integral computations in the appendices.","tokens_in":22639,"tokens_out":3653,"duration_ms":34588,"significance":"If the extrapolations in δ are valid, the paper opens a genuinely new direction in defect CFT: interpolating between defects of different integer dimensions in the same way that the ε expansion interpolates between spacetime dimensions. The free-field results are exact and clean, and the interacting O(N) calculation is a substantial perturbative effort with detailed appendices; the large-N analysis also provides nontrivial analytic continuation checks. The strength of the paper lies in the internal consistency of the perturbative framework and in the verification against known integer-dimension limits. However, the two headline claims—the new interface spectrum (2.28) and the non-local 3d CFT dimensions (2.29)—depend on evaluating resummed series at δ=O(1), outside the radius of convergence of the δ expansion, and on the conjectured value f(1)=1/2. This makes the paper's most interesting claims conditional rather than established. The technical work is a solid contribution even if the extrapolations are later revised, but the current presentation does not fully separate the exact perturbative results from the conjectural analytic continuations.","major_comments":[{"comment":"The value f(1)=1/2 is inferred, not derived. The function f(δ) is known only through ten terms in (2.24); its series has radius 2/3 because of an apparent pole at δ=−2/3, and the estimate f(1)=0.4999(3) comes from Padé/conformal approximations. The text states 'It is tempting to conclude that f(1)=1/2, and we use this value in what follows.' This is a numerical conjecture, and it is load-bearing: the interface spectrum (2.28) and the non-local CFT dimensions (2.29) use f(1) directly in the O(ε) terms. Unless f(1)=1/2 can be proven or supported by an independent argument, the headline claims should be explicitly presented as conditional on this conjecture.","section":"§2.2, Eqs. (2.24)–(2.25)"},{"comment":"The all-orders-in-δ resummation of the geometric series Σ(−δ)^k as 1/(1+δ) is an analytic continuation that is not derived from the field theory. At δ=1 the original series diverges, and replacing it by 1/2 is a choice of summation (Abel/Cesàro). If the true beta-function coefficients contain non-perturbative corrections in δ, the O(ε) coefficient in (2.26) and hence also the lightest-singlet comparison in (2.28) would change. The paper does not provide a mechanism (e.g. a Borel summability argument or a known non-perturbative consistency condition) that would justify this analytic continuation uniquely. This is not a minor technicality: it directly affects the claimed difference between the transdimensional interface and two decoupled copies of the Dirichlet boundary CFT.","section":"§2.2, Eqs. (2.26) and (A.31)"},{"comment":"The central claim that the δ=1−ε interface is a 'completely new interface' rests on the inequality ΔS<ΔD at first order in ε. This inequality follows from (2.28), which in turn relies on f(1)=1/2 and on the geometric-series resummation discussed above. Moreover, the paper itself notes that higher orders in ε could restore equality and that a two-loop extrapolation 'somewhat supports' the identity at ε=1. Given this tension, the conclusion in §4 that this is 'therefore a completely new interface' is stronger than what the presented evidence establishes. The authors should either provide a robustness check (e.g. show that the inequality survives within a range of f(1) values consistent with the Padé error estimate) or rephrase the conclusion as a conjecture.","section":"§2.2, Eq. (2.28)"},{"comment":"The decomposition of two-regulator integrals into the form (A.3) with a unique set of poles and a regular remainder is asserted but not proved. Since footnote 1 acknowledges that in the interacting case the minimal subtraction scheme requires a convention for how to minimize over two regulators, the uniqueness of (A.3) is an additional assumption, not a consequence of dimensional analysis. If integrals with two regulators can also produce terms such as log(ε/δ) or other non-polynomial dependence, the extracted beta-function coefficients and the all-orders-in-δ expressions would be scheme-dependent. The authors should justify (A.3) more explicitly or state it as an axiom of the two-regulator scheme they adopt.","section":"Appendix A, Eq. (A.3); footnote 1"}],"minor_comments":[{"comment":"The one-point function coefficient a_S is said to be 'guessed' after expanding in ε and δ; it would be helpful to state explicitly the order to which the guess has been verified and whether the closed form is proven or conjectural.","section":"§2.1, Eq. (2.10)"},{"comment":"The plot of f(δ) has no error bars or indication of the Padé error estimate; adding a shaded band reflecting the uncertainty (2.25) would help the reader judge how reliable the extrapolation to δ=1 is.","section":"§2.2, Fig. 1"},{"comment":"The notation ˆ∆|3−ε is used without definition; it is presumably the dimension in d=3−ε, but this should be stated explicitly to avoid confusion with the defect dimension.","section":"§2.2, Eq. (2.28)"},{"comment":"There is a typo: 'would like the acknowledge' should read 'would like to acknowledge'.","section":"Acknowledgments"}],"recommendation":"major_revision","confidential_remarks":"The paper is technically impressive and the free-field and large-N sections are solid, but the two headline claims are contingent on analytic continuations that are not established. The authors are honest about the tentative nature of f(1) and about the two-loop counter-evidence for the new-interface claim, which makes this a fixable issue: either prove or sharpen the resummation, or explicitly demote the claims to conjectures with a clear statement of what would falsify them. I would not recommend rejection, because the core perturbative framework is sound and will be useful even if the δ=O(1) extrapolations need revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this paper deserves a serious referee. The concept of transdimensional defects — taking a defect dimension p=2+δ and treating δ as a perturbative parameter alongside ε — is genuinely new as far as I can tell, and the authors put in the work to make it concrete. The free-field section is exact and clean; the interacting O(N) computation to all orders in δ at leading order in ε is a substantial piece of algebra, and the large-N section for 1≤p≤2 in 4≤d≤6 reproduces known line and surface limits, which is a good sanity check.\n\nWhat I want you to know before diving in: the more exciting claims — the new interface at δ=1−ε and the non-local 3d CFT at δ=1 — are not protected by the all-orders-in-δ computation. They require evaluating resummed divergent series at δ=O(1). The geometric series Σ(−δ)^k becomes 1/(1+δ), and f(δ), known through ten terms with a pole at δ=−2/3, is assigned f(1)=0.4999(3) via Padé, and the authors then say \"it is tempting to conclude f(1)=1/2, and we use this value.\" That is an inference, not a derivation. The paper is honest about this, which I respect, but it means the interface spectrum in (2.28) and the singlet-vs-displacement ordering are conditional. If there are non-perturbative corrections in δ, those numbers are not established.\n\nWhere the paper is solid: the free-field exact results, the two-loop surface-defect extension in Appendix B, and the large-N interpolation that matches [7] and [13] in the overlapping limits. The self-citations are not load-bearing; the new results stand on their own. The writing is clear and the limitations are mostly stated up front.\n\nMy recommendation: accept for peer review. The idea is strong enough and the calculations careful enough to justify referee time. In the report, I would ask the authors to move the δ=1 and δ=1−ε results into an explicitly conjectural subsection, or at minimum to state clearly that f(1)=1/2 is a Padé-based assumption rather than a proven value, and to add any available check at higher order in ε that might support or falsify the extrapolation. As it stands, the reader should treat the framework as a real advance and the headline interface as a well-motivated conjecture.","headline":"A genuinely new framework for continuously variable defect dimension with careful, honest calculations; the headline interface result hinges on an unproven resummation, so treat the extrapolations as conjectures.","tokens_in":23170,"tokens_out":3675,"would_cite":true,"duration_ms":28970,"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":"Conformal defects can be defined at a continuously tunable dimension, and the new parameter δ produces a conformal interface unlike any previously known one.","keywords":["conformal defects","defect CFT","epsilon expansion","transdimensional defects","O(N) vector model","non-local CFT","renormalization group","conformal interfaces"],"falsifier":"Continue the series for f(δ) to one more order in δ (δ¹¹) or evaluate it by a non-perturbative method: if f(1) differs from exactly 1/2 by more than the stated error, the interface spectrum (2.28) shifts and the claimed novelty of the interface is falsified. Alternatively, compute the δ=1−ε interface spectrum to second order in ε: if  ̂S and  ̂D become degenerate (or  ̂S becomes heavier) at ε², the claim that this is a new interface rather than two Dirichlet copies is falsified.","tokens_in":22082,"feed_emoji":"🌀","tokens_out":8949,"duration_ms":72879,"temperature":0.7,"pith_summary":"This paper introduces \"transdimensional defects,\" conformal defects whose spatial dimension p is a continuously adjustable parameter rather than a fixed integer. Working in d=4−ε with a defect of dimension p=2+δ, the authors compute defect observables to all orders in δ at fixed order in ε, and show that the defect flows to an infrared fixed point for every δ. The payoff is two new objects: at δ=1−ε the defect becomes an interface whose low-lying spectrum differs from the standard O(N) Dirichlet interface, and at δ=1 the same construction produces a non-local three-dimensional CFT whose operator dimensions are unlike anything previously catalogued. If the resummation is sound, defect CFTs form continuous families parameterized by δ, not just isolated integer-dimensional points.","feed_headline":"Tune the defect dimension and a new conformal interface appears","feed_subtitle":"Resumming to all orders in δ morphs a surface defect into an interface and a non-local 3d CFT.","key_machinery":"The load-bearing object is the new parameter δ entering every defect integral through the defect dimension p=2+δ, on top of the bulk dimension shift d=4−ε. The technical engine is the all-orders-in-δ resummation of the infinite class of \"hopping\" diagrams on the defect: the geometric series Σ(−δ)^k is replaced by its analytic continuation 1/(1+δ), turning previously integer-only defect calculations into closed rational functions of δ (Eqs. (A.31), (A.35), (2.26)–(2.27)). A second ingredient is the function f(δ) of Eq. (2.24), which controls the dimension of the singlet defect operator  ̂S; its first ten series coefficients are packaged in a compact exponential form, and its value at δ=1 is estimated as f(1)=0.4999(3) via Padé-conformal approximants and then taken to be 1/2. That identification is what makes the interface spectrum (2.28) differ from two Dirichlet copies of the boundary theory.","core_discovery":"The central claim is that a conformal defect of non-integer dimension p=2+δ is a consistent, renormalizable object, and that the δ-expansion can be resummed to all orders so that one can genuinely interpolate between integer-dimensional defects. For the free and interacting O(N) vector models in d=4−ε, the defect beta function has an IR fixed point h∗=δ+ε+O(ε²), and the dimensions of the defect operators  ̂φ and ∂⊥ ̂φ become rational functions of δ after resummation (Eqs. (2.26) and (2.27)). Setting δ=1−ε gives an interface whose singlet operator  ̂S is lighter than the displacement operator at first order in ε; this is the paper's evidence that the interface is not two copies of the ordinary O(N) boundary CFT. Setting δ=1 gives a non-local three-dimensional CFT with spectrum (2.29)–(2.30), which the authors state does not match any theory known to them. The same δ-machinery is applied at large N for 4≤d≤6, where it interpolates between line and surface defects and matches known results at both ends.","pith_inferences":["If the resummation is valid at δ of order one, the same method should produce continuous defect families in other bulk theories (fermionic, multi-scalar, or gauged) wherever a weakly relevant defect deformation exists in non-integer dimension.","The apparent exactness of f(1)=1/2 suggests that f(δ) might have a closed analytic form; discovering it would replace the Padé estimate with a proof and might reveal a hidden integrable structure.","The δ=1−ε interface is a natural target for the conformal bootstrap: a singlet-sector gap computation would independently test whether  ̂S stays lighter than the displacement operator at ε=1.","Transdimensional defects may offer a perturbative handle on fractional-dimensional defects in condensed-matter settings, where the defect dimension could be tuned by geometry or by ensemble averaging rather than by analytic continuation."],"forward_implications":["Defect CFTs become continuous families: for every real δ in the range studied there is a conformal defect fixed point, not only at integer p.","The δ=1−ε interface provides a concrete new conformal interface in d=4−ε whose singlet spectrum is calculable and distinct from the Dirichlet-interface spectrum.","The δ=1 defect realizes a non-local three-dimensional CFT with operator dimensions (2.29)–(2.30) that are not reproduced by any known CFT construction.","In the large-N limit the same δ-continuum connects symmetry-breaking line defects near d=4 with surface defects near d=6, recovering known results at both endpoints.","Nearly marginal defect operators can be made exactly marginal by choosing δ as a function of the coupling, generating conformal manifolds of transdimensional defects."],"supporting_citations":[{"why":"Supplies the ε-expansion paradigm of dimensional continuation that the paper generalizes from spacetime dimension to defect dimension.","marker":"[1]"},{"why":"Provides the surface-defect O(N) computation in d=4−ε that the transdimensional framework extends to p=2+δ, and the d=6−ε symmetry-breaking defect results used in Section 3.","marker":"[13]"},{"why":"Parallel surface-defect analysis supplying the recursive all-orders defect propagator and one-point function used in Section 2.","marker":"[15]"},{"why":"Padé-conformal method used to extrapolate f(δ) to δ=1 and conclude f(1)=0.4999(3).","marker":"[27]"},{"why":"Conjecture that the p=2 defect in d=3 flows to two copies of the ordinary fixed point; the benchmark that the new interface is shown to deviate from.","marker":"[31]"},{"why":"Ordinary boundary/interface fixed point data (boundary operator dimensions) used to establish that the δ=1−ε interface differs from two Dirichlet copies.","marker":"[28-30]"},{"why":"Recent O(N) surface-defect computation suggesting factorization of  ̂S into boundary operators; the paper argues its interface avoids that factorization.","marker":"[47]"},{"why":"Long-range Ising model, the prior example of a defect description of a non-local CFT, used to contextualize the δ=1 generalised free field family.","marker":"[24]"}],"fun_headline_variants":["Defects with continuous dimension: a new conformal interface","Resumming δ-expansion reveals new conformal defect phases","From line to surface: continuous defect dimension via δ","Delta-expansion bridges integer defects, creates new interfaces","All-order resummation morphs defects into novel CFTs"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole construction stands on the assumption that the perturbative series in δ, although divergent, can be resummed and then continued to δ of order one — in particular to δ=1 and δ=1−ε — without changing the fixed point.","fun_headline_variants_meta":{"raw":{"variants":["Defects with continuous dimension: a new conformal interface","Resumming δ-expansion reveals new conformal defect phases","From line to surface: continuous defect dimension via δ","Delta-expansion bridges integer defects, creates new interfaces","All-order resummation morphs defects into novel CFTs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000291,"raw_usage":{"total_tokens":1704,"prompt_tokens":950,"completion_tokens":754,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":566,"completion_tokens_details":{"reasoning_tokens":672}},"tokens_in":566,"tokens_out":754,"duration_ms":6638,"temperature":1.0,"reasoning_tokens":672,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:49:03.456782+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Continue the series for f(δ) to one more order in δ (δ¹¹) or evaluate it by a non-perturbative method: if f(1) differs from exactly 1/2 by more than the stated error, the interface spectrum (2.28) shifts and the claimed novelty of the interface is falsified. Alternatively, compute the δ=1−ε interface spectrum to second order in ε: if  ̂S and  ̂D become degenerate (or  ̂S becomes heavier) at ε², the claim that this is a new interface rather than two Dirichlet copies is falsified.","supporting_citations":[{"cited_title":"Critical exponents in 3.99 dimensions","cited_arxiv_id":null,"evidence_quote":"Supplies the ε-expansion paradigm of dimensional continuation that the paper generalizes from spacetime dimension to defect dimension."}],"review_version":1}