{"id":"4487594d-fdd9-4e25-a2e1-492cefd63e00","arxiv_id":"2501.03532","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The authors pack every d-sphere of radii 1 to 2 into a set whose δ-neighborhood has measure ≲ |log δ|^{-2/d}, matching the lower bound in two dimensions.","lead":"This mathematics paper constructs compact sets that contain a sphere of every radius between 1 and 2 in (d+1)-dimensional space, with a δ-neighborhood whose measure is at most a power of |log δ|. The result improves a 1999 construction and proves the optimal bound in two dimensions, a step forward in the curved Kakeya problem.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"As typeset, the proof of Proposition 2.1 uses 2^{Md} intervals but the binary grouping argument requires 2^{M^d}; for d≥2 the grid has more points than Md, so the key estimate (2.19)–(2.22) is not proved for most x′.","rationale":"The reader’s verdict ACCEPT rested on the assumption that the only gaps are minor sketches in Lemma 2.3. In fact, Lemma 2.3 is trivially true via |n_j - n_i| ≤ j - i, so that is not load-bearing. The load-bearing issue is the scaling in Proposition 2.1: the written 2^{-M d} (2^{Md} intervals) is inconsistent with the binary decomposition and with the theorem’s exponent |log δ|^{-(1+α)/d}. The proof as typeset therefore fails for d ≥ 2, exactly where the paper claims the optimal improvement. This is not a matter of style but of correctness of the central estimate. I recommend CONDITIONAL: the paper should be accepted only if the exponent is corrected to 2^{-M^d} (with the binary-grouping argument rechecked) and the typo is explicitly resolved. If the authors confirm the 2^{M^d} scaling, the argument is coherent; if not, the central claim for d ≥ 2 is unproved and the contradiction with the lower bound indicates a substantive error.","tokens_in":13933,"tokens_out":48167,"duration_ms":414871,"concrete_test":"Recount the binary-grouping step (2.16) under both scalings. Set d=2, M=10, so the text gives 2^{20} intervals and m≈25 grid points. Pick x′ whose closest grid point is x_25; show that (2.16) has no valid p and that (2.19) would need a group thickness ≥ δ0 2^{5} M^{-1-α}, violating (2.15). Then repeat with 2^{M^2}=2^{100} intervals, for which m < M^2 and j ≤ M^2 always holds; verify the algebra of (2.20)–(2.22) goes through uniformly. This check distinguishes a harmless typo from a genuine gap.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The construction in Proposition 2.1 divides the parameter interval into 2^{Md} pieces, so each n ∈ {0,...,2^{Md}-1} has a binary expansion of length Md. However, the grid of tangency points has m = (M/2-1)^d + λ points. For d ≥ 2 and large M, m is much larger than Md (e.g., d=2: m ≈ M^2/4 while Md = 2M). The measure estimate fixes a point x′, chooses the closest grid point x_j, and writes n = p + Σ_{i=1}^j ε_i 2^{i-1} with p a multiple of 2^j. This decomposition is only valid for j ≤ Md; for j > Md the list of p is empty and the argument yields a single group whose thickness bound contains a factor 2^{j-Md} ≥ 1, which destroys the claimed δ0 M^{-1-α} fiber bound. Since most grid points have index j > Md, the bound (2.15) is not established for typical x′. Moreover, if the written 2^{Md} scaling were correct, Proposition 3.1 would give |N_δ| ≲ |log δ|^{-1-α} for the sphere family when d=2, contradicting the Kolasa–Wolff lower bound |N_δ| ≳ |log δ|^{-1}. The theorem’s stated exponent |log δ|^{-(1+α)/d} requires the number of intervals to be 2^{M^d}, so the proof appears to rely on an unstated correction from 2^{Md} to 2^{M^d}.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper constructs, for each δ∈(0,1), a compact set Kδ in R^{d+1} that contains a translated copy of every d-sphere of radius between 1 and 2, with |N_δ(Kδ)| ≲_d |log δ|^{-2/d}. For d=2 this is the sharp order predicted by the Kolasa–Wolff lower bound. The proof proceeds by first establishing Proposition 2.1, a packing result for graphs of regular curved functions, via a discrete translation algorithm that compresses the graph family at a grid of tangency points. Proposition 3.1 then uses a partition of unity and the graph result to handle general curved hypersurfaces, and the sphere case follows as a corollary. The paper also states a generalisation to Hölder-continuous families of C^{2,α} hypersurfaces of nonzero Gaussian curvature.","tokens_in":14295,"tokens_out":16284,"duration_ms":121883,"significance":"If the proof is correct, the main result is a genuine improvement over the Kolasa–Wolff construction, removing a (log|log δ|)^{2/d} factor and establishing the sharp order of the δ-neighbourhood for d=2. The method is self-contained, uses only elementary Taylor expansions and a discrete combinatorial inequality, and is clearly presented. The generalisation to nonconvex curved hypersurfaces with nonzero Gaussian curvature is a useful extension of previous work. The paper also draws a helpful conceptual distinction between cinematic curvature and the curvature condition used here. The main risk is the scaling consistency of the construction, which is discussed in the major comments.","major_comments":[{"comment":"The manuscript consistently types the exponent as 'M d', e.g. '2^{-M d}', 'M d/3', 'M d − m ≥ M d/3'. If this is read literally as M·d, the construction divides the parameter interval into 2^{M d} pieces, but the grid of tangency points has m = (M/2−1)^d + λ points. For d ≥ 2 and large M, m ≫ M d (e.g., d=2 gives m ≈ M^2/4 while M d = 2M). The binary decomposition (2.16) is then valid only for j ≤ M d; for j > M d the only possible p is 0, so the grouping is trivial and the thickness bound (2.19) cannot be proved for typical x′, whose closest grid point has index j > M d. Moreover, the line after (2.14), 'M d − m ≥ M d/3', is false under the literal reading for d ≥ 2. The intended notation is evidently M^d: this is consistent with the theorem exponent |log δ|^{-2/d}, with Proposition 3.1's final scale 2^{-N M_0^d} = 2^{-M^d}, and with the inequality M^d − m ≥ M^d/3. The proof therefore requires a systematic correction of every occurrence of 'M d' to 'M^d' in the exponents, and a re-verification of all estimates under that scaling. Without this correction, the measure bound (2.15) is not established and the claimed result would contradict the Kolasa–Wolff lower bound when d=2.","section":"§2.1–2.4, in particular (2.2), (2.3), (2.16), (2.19)–(2.22)"},{"comment":"The key grid-path estimate (2.22) is reduced to Lemma 2.3, but the lemma is only proved for the exponent 2, and the remark asserts that the same bound holds for any exponent β by induction without giving the argument. Since the application requires β = 1+α with α ∈ (0,1], a complete proof is needed. This is a load-bearing estimate: if it fails, the bound (2.21) and hence the fiber measure estimate (2.15) do not follow. The claim is plausible, but as written it is a gap in the proof.","section":"§2.5, Lemma 2.3 and its remark"}],"minor_comments":[{"comment":"The repeated appearance of 'M d' instead of 'M^d' is more than a typographical nuisance: it makes the proof impossible to follow as typeset and obscures the central scaling. The authors should carefully correct all exponents and ensure that the notation is unambiguous.","section":"Throughout, especially (2.7), (2.20), (2.21)"},{"comment":"The text says 'we can replace ∂_a f(a_n, x_i) by ∂_a f(a_{⌊n⌋_i}, x_i)', but the expression in (2.20) is evaluated at x, not at x_i. The intended statement is that ∂_a f(a_n, x) is replaced by ∂_a f(a_{⌊n⌋_i}, x) using (1.3), and similarly for ∇_x f. This should be corrected to avoid confusing the reader.","section":"§2.4, after (2.20)"},{"comment":"The abstract claims a single compact set K containing spheres of every radius in [1,2], while Theorem 1.1 produces a set K_δ for each δ. The passage from the theorem to the abstract is said to follow by a standard iterative argument, but no details are given. A brief explanation of that diagonal argument would improve readability.","section":"§1, abstract and Theorem 1.1"},{"comment":"The construction of the continuous path x̃_j for general d is described only by induction and an example. Since the path order is used in the key estimate (2.22), a precise recursive definition for all d would be helpful.","section":"§2.1.1"}],"recommendation":"major_revision","confidential_remarks":"The scaling problem identified in the major comments appears to be a LaTeX/transcription issue in which superscripts were flattened to 'M d'. If the authors intended M^d, the main proof is largely sound and the result is significant. I recommend inviting a revision with the explicit instruction to correct the exponent notation throughout and to supply the missing proof of Lemma 2.3 for the needed range of exponents. The paper's novelty relative to [YZ24] and its consistency with the known Kolasa–Wolff lower bound should be re-examined after the correction."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know before you read this paper. First, the main result is likely right and is a genuine improvement: it removes the log log factor from Kolasa-Wolff and gets the sharp order in d=2, with a clean generalization to C^{2,α} hypersurfaces. Second, as typeset, the proof has a systematic exponent error: everywhere it writes 2^{Md}, it needs 2^{M^d}. That error is load-bearing.\n\nIn Proposition 2.1 the parameter interval is cut into 2^{Md} pieces, so each n has Md bits, but the grid of tangency points has about (M/2)^d points, which for d≥2 is far larger than Md. The binary grouping in Section 2.4 only makes sense for j ≤ Md; for the typical grid point with j > Md, the decomposition n = p + Σ ε_i 2^{i-1} collapses (p=0) and the thickness bound contains a factor 2^{j-Md} ≥ 1, destroying the M^{-1-α} fiber estimate. The same slip makes Proposition 3.1 claim a |log δ|^{-1-α} bound for spheres in d=2, contradicting the Kolasa-Wolff lower bound |N_δ| ≳ |log δ|^{-1}. The theorem's stated exponent (1+α)/d is what you get from 2^{M^d} scaling.\n\nIf you mentally replace M d with M^d everywhere, the proof is coherent and the arguments are largely standard: the implicit function theorem step, the Taylor expansions, and the grid-path estimate (2.22) are plausible. Lemma 2.3 is sketched but the summation-by-parts claim is believable. The deduction of the single compact set in the abstract from the per-δ theorem is asserted rather than written out, but that is a routine iteration.\n\nThe paper is worth engaging with after the typo is fixed. It builds on Yang-Zhong and extends it genuinely; the novelty is real, not a repackaging. The citation practice is fine. My recommendation: send it to a serious referee, but the referee should insist the M^d/M d confusion be resolved explicitly. It may be a LaTeX error, but as submitted the central estimate is not proved for d≥2.","headline":"Right idea, wrong exponent: the proof as typeset uses 2^{Md} where it needs 2^{M^d}, and that error breaks the central estimate for d≥2.","tokens_in":14764,"tokens_out":3966,"would_cite":false,"duration_ms":33986,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["28A75","42B25"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper constructs a compact set in $\\mathbb{R}^{d+1}$ containing a $d$-sphere of every radius in $[1,2]$ whose $\\delta$-neighbourhood has measure $\\lesssim_d |\\log\\delta|^{-2/d}$; for $d=2$ this order is optimal.","keywords":["Kakeya-type sets","sphere packing","hypersurface packing","δ-neighbourhood measure","Gaussian curvature","grid-path estimate","tangent compression"],"falsifier":"For a fixed dimension $d\\ge3$ and a small exponent $\\alpha$ (say $\\alpha=0.1$), compute the left side of (2.22) numerically for the back-and-forth path produced by the paper's induction, at $x=x_j$ and over a range of $M$ and $j$. If the ratio to $2^j M^{-1-\\alpha}$ grows without bound, the key estimate (2.22) is false. A simpler check is to search exhaustively for a unit-step path in $\\{1,\\dots,J\\}^d$ that violates Lemma 2.3.","tokens_in":13760,"feed_emoji":"📐","tokens_out":10786,"duration_ms":87283,"temperature":0.7,"pith_summary":"The paper proves that for every dimension $d \\ge 1$, one can place a $d$-sphere of every radius between 1 and 2 into a compact subset of $\\mathbb{R}^{d+1}$ whose $\\delta$-neighbourhood has Lebesgue measure at most a constant (depending only on $d$) times $|\\log \\delta|^{-2/d}$. For $d=2$ this order is the best possible, since a known lower bound forces $|N_\\delta(K)| \\gtrsim |\\log \\delta|^{-1}$. The construction works by a dyadic compression: curves are translated so that representative graphs become tangent at successive grid points, and the whole measure estimate is reduced to an elementary inequality for paths on a grid. The same mechanism applies to any Hölder-continuous one-parameter family of $C^{2,\\alpha}$ hypersurfaces with Gaussian curvature bounded away from zero, yielding $|\\log \\delta|^{-(1+\\alpha)/d}$.","feed_headline":"Every sphere radius 1 to 2 fits in volume |log δ|^(-2/d)","feed_subtitle":"In two dimensions the bound is optimal, and the previous log-log factor is gone.","key_machinery":"The engine is an iterative tangent-compression scheme. The parameter interval is split into $2^{Md}$ pieces, and the corresponding graph pieces are translated so that their representative functions match both value and gradient at successive points $x_j$ along a back-and-forth Hamiltonian path (visiting every interior grid point exactly once) through the interior points of a $J^d$ grid. At step $j$ the translation vectors have size about $\\delta_0 2^{j-Md}$, and the total accumulated translation stays tiny because each piece is only moved when its binary index says so. The measure bound is reduced to the grid-path estimate (2.22), proved through Lemma 2.3: along any unit-step path $\\{n_i\\}$ in $\\{1,\\dots,J\\}^d$ one has $\\sum_{i=1}^j 2^i |n_j-n_i|^{1+\\alpha} \\lesssim 2^j M^{-1-\\alpha}$, and this inequality is what removes the logarithmic loss.","core_discovery":"The central claim, stated on the paper's own terms, is that the previously known upper bound for packing spheres, which carried an extra $\\log |\\log \\delta|$ factor, can be sharpened to $|N_\\delta(K)| \\lesssim_d |\\log \\delta|^{-2/d}$. The proof establishes more: for any regular family of curved hypersurfaces $S_a$, defined by $\\Phi(a,x)=0$ with Gaussian curvature uniformly bounded away from zero and $C^{2,\\alpha}$ regularity, the same compression gives $|N_\\delta(K)| \\lesssim |\\log \\delta|^{-(1+\\alpha)/d}$. The sphere case follows because the family $\\Phi(a,x)=|x|-a$, $1 \\le a \\le 2$, satisfies these hypotheses with $\\alpha=1$.","pith_inferences":["The grid-path inequality is the only step whose proof is sketched rather than fully written; a natural test is whether the same inequality holds for the specific paths used when $d \\ge 3$, since the paper constructs those paths by induction without giving the explicit ordering.","The paper's remark that cylinders pack better than spheres suggests that the packing rate is governed by the dimension of the set of normal directions of the family; one could test this by computing sharp exponents for intermediate families that interpolate between cylinders and spheres.","For $d=1$, where optimality of $|\\log \\delta|^{-2}$ remains open, the same tangent-compression scheme might be inverted to build a lower-bound example, because the grid-path mechanism does not obviously favour upper bounds over lower bounds."],"forward_implications":["For $d=2$, the new construction meets the lower bound $|N_\\delta(K)| \\gtrsim |\\log \\delta|^{-1}$, so the exact order of the $\\delta$-neighbourhood measure for packing all circles of radii in $[1,2]$ is now known.","The $\\delta$-level construction passes through a standard iterative argument to give a compact set of Lebesgue measure zero that still contains a $d$-sphere of every radius in $[1,2]$.","The same algorithm proves a general packing bound $|\\log \\delta|^{-(1+\\alpha)/d}$ for Hölder-continuous one-parameter families of $C^{2,\\alpha}$ hypersurfaces with Gaussian curvature bounded away from zero, including hyperbolic paraboloids.","The result holds in every dimension $d \\ge 1$ and removes the $(\\log|\\log\\delta|)^{2/d}$ factor from the previous sphere-packing construction."],"supporting_citations":[{"why":"Supplies the prior sphere-packing construction with the extra log-log factor and the lower bound for $d\\ge2$ that the paper improves.","marker":"[KW99]"},{"why":"Provides the earlier graph-packing scheme whose Section 2 iteration this paper extends, simplifies, and generalizes to higher dimensions.","marker":"[YZ24]"},{"why":"Gives the lower bound $|\\log\\delta|^{-1}$ for plane needle packing that makes the $d=2$ case sharp.","marker":"[Cór93]"},{"why":"Gives the weaker lower bound for packing circles in the plane that defines the open $d=1$ case.","marker":"[Wol97]"}],"fun_headline_variants":["Log-log factor gone: spheres 1–2 pack in |log δ|^{-2/d}","All sphere radii 1–2 fit in volume |log δ|^{-2/d}","2D optimal: all radii 1–2 packed in |log δ|^{-1}","Tighter sphere packing: radii 1–2 in |log δ|^{-2/d}"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The measure bound rests on the grid-path inequality (2.22), whose proof in Lemma 2.3 is only sketched: the summation-by-parts step is asserted, and the construction of the back-and-forth Hamiltonian path for general dimension is stated by induction. If that inequality failed for some $d$ or some $\\alpha$, Proposition 2.1 and hence the sphere-packing theorem would not follow.","fun_headline_variants_meta":{"raw":{"variants":["Log-log factor gone: spheres 1–2 pack in |log δ|^{-2/d}","All sphere radii 1–2 fit in volume |log δ|^{-2/d}","2D optimal: all radii 1–2 packed in |log δ|^{-1}","Tighter sphere packing: radii 1–2 in |log δ|^{-2/d}"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001967,"raw_usage":{"total_tokens":7630,"prompt_tokens":835,"completion_tokens":6795,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":451,"completion_tokens_details":{"reasoning_tokens":6697}},"tokens_in":451,"tokens_out":6795,"duration_ms":53261,"temperature":1.0,"reasoning_tokens":6697,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T21:52:26.100710+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a fixed dimension $d\\ge3$ and a small exponent $\\alpha$ (say $\\alpha=0.1$), compute the left side of (2.22) numerically for the back-and-forth path produced by the paper's induction, at $x=x_j$ and over a range of $M$ and $j$. If the ratio to $2^j M^{-1-\\alpha}$ grows without bound, the key estimate (2.22) is false. A simpler check is to search exhaustively for a unit-step path in $\\{1,\\dots,J\\}^d$ that violates Lemma 2.3.","supporting_citations":[],"review_version":1}