{"id":"326934ab-24bf-46c7-b17d-62c6b2c59c04","arxiv_id":"2506.03705","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Infinitely many smoothly slice knots admit topological slice discs that are non-approximable by smooth slice discs, even up to isotopy.","lead":"The paper constructs infinitely many smoothly slice knots, each with a topological slice disc that cannot be approximated by any smooth slice disc, even allowing topological isotopy. This answers a refined version of a MathOverflow question about the gap between smooth and topological four dimensional knot theory.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The contradiction proof's quantifier order is invalid: the smooth slice disc is only guaranteed near a disc that depends on ε, so it need not lie inside that disc's tubular neighbourhood, and the inclusion chain used to define ι is not established.","rationale":"The reader's weakest assumption concerns the uncited d-invariant results [CK21, Cha21]. Those are important, but the more immediate and more load-bearing problem is that the proof never validly reaches the point where those results apply. The contradiction hypothesis quantifies over ε first and disc second; the proof then selects ε after the disc, which is not justified. In particular, the smooth slice disc D'_{m,ε} obtained for a given ε may lie in ν_ε(D_m(ε)) but outside every tubular neighbourhood of D_m(ε) when ε is chosen before D_m(ε). The paper's own final remark illustrates that isotopic discs can be arranged to contain a fixed smooth disc in arbitrarily small ε-neighbourhoods, but the proof would need the smooth disc to lie inside the tubular neighbourhood of the same isotopic disc. That pairwise inclusion is exactly what is not supplied. If the central claim is true, the proof needs an additional argument or a reformulation of the definition; as written, the contradiction does not follow. Hence the verdict should be conditional on resolving this quantifier gap. The concern is not about the authors' integrity or the plausibility of the theorem, but about a precise step in the written argument.","tokens_in":4583,"tokens_out":30671,"duration_ms":359561,"concrete_test":"Formalize the quantifier structure of the contradiction hypothesis: ∀ε>0 ∃D_m(ε) ∃smooth slice disc F(ε) with F(ε) ⊂ ν_ε(D_m(ε)). Check whether the implication ∀ε∃D_m∃F (F⊂ν_ε(D_m)) ⇒ ∃E∃ε∃F (F⊂ν_ε(E)⊂N(E)) is derivable. It is not, because replacing ε by a smaller ε' can replace E by a different disc. A positive settlement would require a separate lemma producing, from the stated hypothesis, a topological slice disc E isotopic to D_m and a smooth slice disc inside a tubular neighbourhood of E; the current proof contains no such lemma.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The proof assumes for contradiction that for every ε>0 there exists a topological slice disc D_m(ε), isotopic to the constructed D_m, and a smooth slice disc D'_{m,ε} with D'_{m,ε} ⊂ ν_ε(D_m(ε)). It then fixes a tubular neighbourhood N(D_m) and 'Choose[s] ε > 0 small enough such that ν_ε(D_m) ⊆ N(D_m)' (proof paragraph beginning 'For a contradiction', p.2). This chooses ε after D_m, but under the contradiction hypothesis D_m is chosen after ε. The only guaranteed smooth disc lies in ν_ε(D_m(ε)); decreasing ε to force ν_ε(D_m) into N(D_m) generally produces a different disc D_m(ε'), and the hypothesis gives no smooth disc inside ν_ε(D_m(ε')) for the original D_m(ε'). Thus the proof does not establish the existence of a single pair (E,F) with E locally flat, F a smooth slice disc, and F ⊂ ν_ε(E) ⊂ N(E). Without such a pair, the chain M_K → D^4∖N(D_m) → D^4∖νε(D_m) → D^4∖D'_{m,ε} and the resulting map ι are not obtained, so the d-invariant contradiction from [CK21, Cha21] is never reached. This is a logical gap in the central argument, independent of the correctness of the cited algebraic lemmas.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper constructs, for each odd integer m ≥ 1, a smoothly slice knot K_m in S^3 and a topological slice disc D_m for K_m. The main theorem asserts that D_m is smoothly non-approximable: there exists ε > 0 such that for every topological slice disc D' that is topologically isotopic to D_m rel. boundary, the ε-neighbourhood of D' contains no smooth slice disc for K_m. The proof builds K_m as a satellite of a knot R_m, distinguishes the knots by Alexander polynomials, and derives a contradiction from the existence of a hypothetical smooth slice disc by combining the Blanchfield form metaboliser structure with d-invariant results imported from Cha and Kim [CK21] and Cha [Cha21].","tokens_in":4861,"tokens_out":16584,"duration_ms":179841,"significance":"If correct, the theorem provides the first examples of smoothly slice knots with smoothly non-approximable topological slice discs, thereby answering a refined version of a MathOverflow question and complementing Venema's approximation theorems. The construction is explicit and the knots are distinguished by elementary Alexander polynomial computations. The algebraic core—the presentation of H_1 of the zero-framed surgery, the metaboliser dimension count, and the reduction to branched covers—is clearly and coherently presented. The decisive d-invariant contradiction is not derived in the paper but is cited from published results of Cha-Kim and Cha; this reliance is standard practice, though it makes the proof less self-contained. The paper is well written overall, but the central argument has a quantifier gap that must be fixed before the proof is valid.","major_comments":[{"comment":"The inclusion chain used to define the map ι is not justified as written. The contradiction hypothesis states that for every ε > 0 there exists a topological slice disc (call it D_m^ε) that is isotopic rel. boundary to the fixed D_m, and a smooth slice disc D'_m,ε contained in νε(D_m^ε). The proof then fixes a tubular neighbourhood N(D_m) of the original D_m, chooses ε with νε(D_m) ⊆ N(D_m), and writes the chain M_{K_m} → D^4∖N(D_m) → D^4∖νε(D_m) → D^4∖D'_m,ε. This chain requires D'_m,ε ⊆ νε(D_m), but the hypothesis only provides D'_m,ε ⊆ νε(D_m^ε), and nothing guarantees that D_m^ε lies in or near νε(D_m) after ε is chosen. Without the inclusion D'_m,ε ⊂ νε(D_m), the final map in the chain (and hence the map ι) is not defined, so the kernel computation and the subsequent d-invariant contradiction do not go through. The gap can be repaired by applying the argument to each D_m^ε instead of to the fixed D_m: because D_m^ε is isotopic to D_m, the curve α_1 still bounds in its exterior, and the chain M_{K_m} → D^4∖N(D_m^ε) → D^4∖νε(D_m^ε) → D^4∖D'_m,ε is legitimate. This yields a contradiction for every ε and proves the theorem. The manuscript should be revised to make this quantifier handling explicit.","section":"Proof, paragraph beginning 'For a contradiction'"}],"minor_comments":[{"comment":"The same symbol D (or D_m) is used for the fixed disc and for the variable isotopic disc in the definition and in the contradiction hypothesis. This is confusing; for example, the sentence 'there exists a topological slice disc D_m ... isotopic rel. boundary to D_m' has two different meanings of D_m. Please use distinct notation, such as D_m^ε, for the variable disc.","section":"Definition of smoothly non-approximable and proof"},{"comment":"The proof relies on [CK21, Lemma 5.2], [Cha21, p.17 Assertion], [CK21, Theorem 5.4], and [Cha21, Lemma 4.1] for the decisive d-invariant computations, but these statements are not restated. Since the contradiction is entirely derived from these results, it would improve the paper to state the precise assertions being used.","section":"d-invariant step"},{"comment":"In the final remark, the assertion that for every ε > 0 the disc D is topologically isotopic to a disc that contains ∆ in its ε-neighbourhood, and hence D fails to be smoothly non-approximable, is stated without proof. A brief justification or reference would be helpful, since this claim is used to explain the role of the isotopy in the definition.","section":"Final remark"},{"comment":"The formula ∆_{R_m}(t) = ((m+1)t - m)(mt - (m+1)) is asserted without derivation. A short explanation from the Seifert matrix of R_m would aid the reader's verification.","section":"Alexander polynomial computation"}],"recommendation":"major_revision","confidential_remarks":"The paper is a short, elegant construction with a single, but central, logical gap in the proof of the theorem. The gap is fixable by restructuring the contradiction argument to work with the isotoped disc D_m^ε rather than the original D_m, as detailed in the major comment. I recommend asking the authors to make this revision and to clean up the notation. The heavy reliance on [CK21] and [Cha21] is acceptable for a research note, though restating the cited lemmas would improve verifiability."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper's construction is genuinely new: the knots Km, built by saturating one band of a two-component link with a Whitehead double, give infinitely many smoothly slice knots, distinguished by Alexander polynomial. The definition of smoothly non-approximable, allowing isotopy of the topological disc, is the right refinement of Winter's question, and the final remark showing why Gompf's construction fails that definition is useful. The authors also clearly explain why Meier's and CHH's earlier examples do not imply their theorem. The homology computation of the Blanchfield form and the metabolic argument before the d-invariant step are explicit and sound.\n\nBut there is a serious flaw in the contradiction proof. The assumption is: for every ε>0 there exists a topological slice disc D_ε, isotopic rel. boundary to the constructed D_m, such that νε(D_ε) contains a smooth slice disc F_ε. The proof then fixes the original D_m, chooses ε small enough that νε(D_m) lies inside a tubular neighbourhood N(D_m), and considers the inclusion chain\nM_K → D^4∖N(D_m) → D^4∖νε(D_m) → D^4∖F_ε.\nThe last inclusion requires F_ε ⊆ νε(D_m). But the assumption only gives F_ε ⊆ νε(D_ε), and D_ε is allowed to depend on ε and need not be close to D_m. The notation in the paper obscures this by reusing D_m for the isotopic disc. The quantifier order is wrong: you cannot pick ε first, get D_ε, and then also demand that νε(D_ε) lies inside a fixed tubular neighbourhood of D_m. If you instead try to use D_ε throughout, you would need ε chosen after D_ε to make νε(D_ε) fit in some N(D_ε), which is circular. As written, the inclusion chain and the map ι are not established, and the d-invariant contradiction is not reached.\n\nThis is a load-bearing gap, not a minor omission. That said, the issue is in the logical setup, not in the algebraic lemmas from [CK21, Cha21], which may well be correct. The paper is short, clearly written, and the underlying idea is likely salvageable, but the current proof does not prove the theorem.\n\nThis paper deserves a serious referee, but not acceptance in this form. It will be of interest to people in knot concordance and 4-manifold topology. I would bring it to a reading group to discuss the gap and possible repairs.","headline":"Nice new construction and a clear question, but the central contradiction proof has a quantifier gap that invalidates the inclusion chain as written.","tokens_in":5383,"tokens_out":11447,"would_cite":false,"duration_ms":125470,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["57K10","57K18","57N70","57R40"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper constructs infinitely many distinct smoothly slice knots, each of which admits a topological slice disc that is smoothly non-approximable.","keywords":["slice discs","Heegaard Floer invariants","non-approximable","topological slice discs","smoothly slice knots","Blanchfield form","branched covers","d-invariant"],"falsifier":"Exhibit, for some m, a smooth slice disc for K_m inside every ε-neighbourhood of a topologically isotopic copy of D_m; alternatively, compute the correction terms d(Σ_r, s_{Σ_r} + k b_{x1}) for a fixed K_m and show that they all vanish, which would remove the nonvanishing step the proof relies on.","tokens_in":4376,"feed_emoji":"🪢","tokens_out":15080,"duration_ms":151673,"temperature":0.7,"pith_summary":"This paper proves that there are infinitely many distinct knots in the 3-sphere that are smoothly slice—each bounds a smooth disc in the 4-ball—yet also admit a topological slice disc that is smoothly non-approximable: no smooth slice disc for the same knot can be found inside a small neighbourhood of any topologically isotopic copy of that disc. The construction starts from a family of ribbon knots and feeds a Whitehead double of the trefoil into one of their bands, producing knots whose Alexander polynomials are all different. The proof that the resulting topological slice discs cannot be smoothed relies on a contradiction between the metaboliser of the Blanchfield form of the knot exterior and Heegaard Floer d-invariants of cyclic branched covers. If the argument is correct, it answers a refined version of a question posed on an online mathematics forum, and it shows that the gap between topological and smooth sliceness persists even when the boundary knot is fixed and smoothable.","feed_headline":"Smoothly slice knots resist smooth approximation of their slice discs","feed_subtitle":"Topological and smooth sliceness diverge even for the same knot and the same boundary.","key_machinery":"The machinery is the metaboliser of the Blanchfield form of the zero-framed surgery M_{K_m}, together with Heegaard Floer d-invariants of branched covers. A slice disc for K_m gives a homomorphism ι from H_1(M_{K_m}; Q[$t^{{±1}}$]) to the homology of the disc complement, and the kernel of ι is a metaboliser: a subspace equal to its own annihilator under the nonsingular Blanchfield pairing. The paper computes H_1(M_{K_m}; Q[$t^{{±1}}$]) ≅ Q ⊕ Q with basis α1, α2, and since α1 is killed by the topological slice disc D_m, the metaboliser condition forces ⟨α1⟩ = ker ι. This equality is transported to the r-fold cyclic branched covers Σ_r, where the lift x1 of α1 generates the kernel of the map H_1(Σ_r; Z) → H_1(V_r; Z) for some prime r. Smoothness of the imagined approximating disc would force the d-invariant d(Σ_r, s_{Σ_r} + k b_{x1}) to vanish for every integer k, while the quoted results of [CK21] and [Cha21] guarantee a nonzero value for some k, giving the contradiction.","core_discovery":"For each odd integer m ≥ 1, let Wh be the positive-clasped zero-twisted Whitehead double of the right-handed trefoil, let U be the unknot, and define K_m = R_m(U, Wh), where R_m is the ribbon knot with 2m+1 band crossings and two distinguished unknotted curves α1, α2 in its complement. Because α1 and α2 link R_m trivially, the satellite operation does not change the Alexander polynomial: Δ_{K_m}(t) = Δ_{R_m}(t) = ((m+1)t − m)(mt − (m+1)), so the K_m are pairwise distinct. Each K_m is smoothly slice, by cutting the right band of its Seifert surface to obtain an unlink and capping off with smooth discs. The authors define a topological slice disc D_m by cutting the left band instead and capping the two strands with two parallel copies of a topological slice disc for Wh, which exists because Δ_Wh = 1. The central claim is that D_m is smoothly non-approximable: there is an ε > 0 such that every topological slice disc topologically isotopic to D_m rel. boundary has no smooth slice disc for K_m inside its ε-neighborhood. The proof assumes such a smooth disc exists, derives that the kernel of the induced map on H_1 with Q[$t^{{±1}}$] coefficients is exactly the subgroup generated by α1, then uses quoted results on Heegaard Floer d-invariants of the r-fold cyclic branched covers Σ_r to obtain a contradiction: vanishing for all spin^c structures would be forced by the smooth slice disc, but a nonzero d-invariant is guaranteed by the cited calculations.","pith_inferences":["Replacing Wh by any knot with Alexander polynomial 1 and suitable nonvanishing d-invariants should produce further families of non-approximable discs, since the proof uses little else about Wh.","Because the argument avoids a homology-ribbon assumption, it may apply to a broader class of topological slice discs than earlier doubly-slice-based constructions; testing whether every smoothly slice knot with a two-generator Alexander module admits such a disc is a natural next step.","The proof gives no explicit value of ε; extracting quantitative bounds from the d-invariant calculations could show how the smoothness gap scales with m.","The non-approximability is a property of the chosen disc, not of the knot, so detecting it will require invariants of slice discs rather than knots alone."],"forward_implications":["The knots K_m are pairwise distinct and form an infinite family of smoothly slice knots whose topological slice discs cannot be smoothed while the boundary knot is fixed.","Each K_m is topologically doubly slice and smoothly slice but not smoothly doubly slice, giving a new infinite family of that kind.","The theorem answers the refined online question affirmatively, and the allowance of topological isotopy rel. boundary is essential, since the earlier capped-tower construction only yields a weaker statement without it.","The result contrasts with the fact that every topological slice disc can be approximated by a smooth disc when the boundary curve is allowed to move: the obstruction lives in fixing the boundary knot."],"supporting_citations":[{"why":"Shows that the satellite operations on α1 and α2 do not change the Alexander polynomial, so Δ_{K_m}(t) = Δ_{R_m}(t) and the knots K_m are distinct.","marker":"[Sei50]"},{"why":"Supplies the topological slice disc for the Whitehead double Wh, which is used to cap off the left band and build D_m.","marker":"[Fre84]"},{"why":"Gives the tubular neighbourhood theorem for locally flat discs, which provides the exterior of D_m and its boundary identification with the zero-framed surgery M_{K_m}.","marker":"[FQ90]"},{"why":"Provides the theorem that the kernel of the inclusion-induced map is a metaboliser for the Blanchfield form of a slice knot exterior.","marker":"[COT03]"},{"why":"Provides the same metaboliser statement as an alternative reference, used together with [COT03].","marker":"[Hil12]"},{"why":"Supplies the lemmas and theorem converting the metaboliser condition into a statement about kernels of branched-cover maps and guaranteeing a nonvanishing d-invariant.","marker":"[CK21]"},{"why":"Supplies companion assertions and a lemma giving the nonvanishing d-invariant needed for the contradiction.","marker":"[Cha21]"},{"why":"States that a smooth slice disc forces all relevant d-invariants of branched covers to vanish, the zero side of the contradiction.","marker":"[GRS08]"},{"why":"Defines the Heegaard Floer d-invariant used in the argument.","marker":"[OS03]"}],"fun_headline_variants":["Smooth slice knots defy smooth slice disc approximation","Topological slice discs never smooth: new knot family","Knots slice smoothly, but their slice discs resist smoothing","Non-approximable slice discs found for smoothly slice knots","Smoothly slice knots with non-smoothable topological discs"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument depends on two technical results quoted from [CK21] and [Cha21] without proof: the metaboliser condition forces a branched-cover kernel to be generated by the lifted curve for some prime r, and the Heegaard Floer correction term of that cover is nonzero in some spin^c structure; if either of these cited results fails, the contradiction collapses.","fun_headline_variants_meta":{"raw":{"variants":["Smooth slice knots defy smooth slice disc approximation","Topological slice discs never smooth: new knot family","Knots slice smoothly, but their slice discs resist smoothing","Non-approximable slice discs found for smoothly slice knots","Smoothly slice knots with non-smoothable topological discs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000661,"raw_usage":{"total_tokens":2996,"prompt_tokens":894,"completion_tokens":2102,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":510,"completion_tokens_details":{"reasoning_tokens":2021}},"tokens_in":510,"tokens_out":2102,"duration_ms":16078,"temperature":1.0,"reasoning_tokens":2021,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T10:57:32.844170+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Exhibit, for some m, a smooth slice disc for K_m inside every ε-neighbourhood of a topologically isotopic copy of D_m; alternatively, compute the correction terms d(Σ_r, s_{Σ_r} + k b_{x1}) for a fixed K_m and show that they all vanish, which would remove the nonvanishing step the proof relies on.","supporting_citations":[],"review_version":1}