{"id":"d797220f-8c8e-4ff8-9f36-b9a7b7eea1eb","arxiv_id":"2411.13395","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The homogeneous and original arithmetic Kakeya inequalities are equivalent, leading to new upper bounds for the d-dimensional arithmetic Kakeya constant and new Minkowski dimension lower bounds for (n,d)-Besicovitch sets.","lead":"The authors show that two versions of a central conjecture in additive combinatorics, the arithmetic Kakeya conjecture, are actually equivalent. This lets them prove new bounds for higher-dimensional variants and for the dimension of Besicovitch sets containing disks in every direction.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Equation (10) applies Lemma 5 to a mis-scaled sumset: with I0=d{0,...,m-1} and d=y/j, the printed term |D_y-(y/j)I0| has dilation (y/j)^2, not j^{-1}; the lower bound on Pr[y in B_I] used for the rest of Theorem 1 does not follow as written.","rationale":"The reader's weakest assumption was Lemma 5 itself; my concern is more specific: the way Lemma 5 is applied in equation (10) is mismatched with the printed definitions. With the text as written, I0 already includes the random factor d, so the summand in (10) carries an extra factor y/j and falls outside the scope of Lemma 5. The natural correction is to take I0 as the unscaled interval, in which case the argument works, and the typo is very likely. The same section also contains the probabilistic tail step flagged by the reader, which can be repaired by standard reverse Markov with a slightly smaller threshold. Neither issue appears fatal to the central claim, but the proof as printed is not complete; a revision should fix the notation in (10) and justify the tail bound. I therefore agree with the CONDITIONAL verdict: the theorem is probably correct, but the exposition needs correction before the proof is rigorous.","tokens_in":10946,"tokens_out":37647,"duration_ms":406285,"concrete_test":"Re-derive equation (10) with the intended definition I0={0,...,m-1} and I=t+d I0. Then the summand becomes |D_y-(y/j)I0|, and Lemma 5 (with A=-y^{-1}D_y and I={0,...,m-1}) applies exactly, yielding Pr[y in B_I] >= m p^{-1-2eps}. Alternatively, keep the printed definition I0=d{0,...,m-1}; in that case verify whether Lemma 5's conclusion can be applied to |D_y-(y/j)I0| = |D_y-(y/j)^2{0,...,m-1}| — it cannot, because the dilation factor is j^{-2}, not j^{-1}.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Section 4, I0 is defined as I0={dj : j=0,...,m-1}, and later the condition y in J is rewritten as d=y/j for some j in [m/2,m]. Conditioned on d=y/j, the progression is I=t+(y/j){0,...,m-1}, so the correct probability summand is |D_y-(y/j){0,...,m-1}|. Equation (10), however, sums |D_y-(y/j)I0|. Under the printed definition of I0, (y/j)I0 = (y/j)^2{0,...,m-1}, so Lemma 5, which controls A+j^{-1}I, does not apply to the printed summand. The subsequent claim that Lemma 5 gives |D_y-(y/j)I0| >= p^{1-eps} for many j is therefore unsupported as written. Since this lower bound is the only mechanism forcing the random arithmetic progression to meet many fibers D_y, the constructed pair (X',Y') is not shown to inherit the entropy bounds needed for the beta(R) inequality, and the proof of Theorem 1 collapses at this step unless the notation is repaired. A secondary but related gap is the assertion Pr[|B_I| >= p^{-2eps} m] >= p^{-2eps}, which does not follow from E|B_I| >= m p^{-2eps} without a second-moment or concentration argument; it is fixable by reverse Markov with a smaller threshold, but as printed it is not justified.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies entropic arithmetic Kakeya constants. It defines β(R) by H(Y) ≤ β(R) max_r H(X+rY) and the homogeneous variant β_h(R) by adding (β_h(R)-1)H(X|Y) to the left-hand side. Theorem 1 claims that β_h(R)=β(R) for every finite R⊂Q with |R|≥2. The proof passes through a discretized graph G'⊂F_p^2 and, using a new sumset lemma (Lemma 5), constructs from a random arithmetic progression a smaller graph G_I whose entropy data inherit the original inequality. Theorem 2 and Proposition 6 iterate the homogeneous inequality to bound β(R^d), yielding Corollary 1 on the Minkowski dimension of (n,d)-Besicovitch sets.","tokens_in":11305,"tokens_out":11663,"duration_ms":113498,"significance":"The equivalence in Theorem 1, if established, is an elegant and useful reduction: it makes existing bounds on β(R) automatically valid for the homogeneous form needed in Oberlin-style iterations, without rewriting the proof of Katz and Tao. The new sumset lemma is a substantive ingredient, and the iterative argument in Section 5 is clean and appears to work. The paper is fully self-contained in its main dependencies and does not fit parameters to force the conclusion. Because the proof of Theorem 1 currently has a local but load-bearing gap, the significance is conditional on the repair.","major_comments":[{"comment":"With I0 defined as I0={dj : 0≤j<m}, the summand |D_y-(y/j)I0| is inconsistent with the conditioning d=y/j. Conditioned on d=y/j, the progression is t+(y/j){0,...,m-1}, so the correct summand is |D_y-(y/j){0,...,m-1}|. With the printed definition, (y/j)I0=(y/j)^2{0,...,m-1}, and Lemma 5, which controls sets of the form A+j^{-1}I for a fixed interval I, does not apply to the printed expression. Thus the stated lower bound on Pr[y∈B_I] is not justified, and this is a load-bearing step for Theorem 1. The gap can be repaired by replacing I0 with the fixed interval {0,...,m-1} in the conditioning and in the subsequent application of Lemma 5, but as written the proof is unsupported at this point.","section":"§4, Eq. (10)"},{"comment":"The inference from E|B_I| ≥ m p^{-2ε} and |B_I| ≤ m to Pr[|B_I| ≥ p^{-2ε}m] ≥ p^{-2ε} is not valid: reverse Markov bounds require the threshold to be strictly below the expectation lower bound, and no such conclusion follows when the threshold equals the mean. This matters because the construction of G_I requires a positive-probability event. The gap is fixable by choosing a smaller threshold such as p^{-3ε}m and then adjusting ε in the entropy estimates, and a union bound with (9) should be stated explicitly; but as printed the step is unjustified.","section":"§4, after Eq. (10)"}],"minor_comments":[{"comment":"The symbol I0 is overloaded: it denotes the random progression {dj}, but in the intended application of Lemma 5 it must be the fixed interval {0,...,m-1}. This overloading is the source of the error in Eq. (10) and should be corrected.","section":"§4, notation"},{"comment":"The proof concludes a bound for at least half of the indices j in the set J of primes in [m/2,m]; the passage from this to the stated ⌊m p^{-ε}⌋ indices in the full interval [m/2,m] should be written out explicitly.","section":"Lemma 5"},{"comment":"The displayed formula for the Minkowski dimension lower bound appears garbled in the typeset text; please check whether it should be d/β(R)^n or d/β(R)^d n and correct the formatting.","section":"Corollary 1"},{"comment":"The coefficient computation after the weighted sum is compressed; since positivity of the coefficients is used before averaging over permutations, an explicit expansion of the grouped coefficients would improve readability.","section":"§5, after Eq. (12)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is potentially publishable in math.CO if the Section 4 gap is repaired. The error in Eq. (10) appears to be a notational/conditioning slip rather than a conceptual flaw, but because Theorem 1 is the main engine of the paper, it must be fixed before acceptance. I found no evidence of circularity or fitted constants."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is worth taking seriously. It proves β_h(R)=β(R) and uses that to get the first nontrivial bounds for higher-dimensional arithmetic Kakeya constants, with a Minkowski-dimension corollary for (n,d)-Besicovitch sets. The equivalence was missing from the literature; the product argument and the sumset lemma are new. Most of the structure is coherent and the authors are honest about what is open.\n\nThe main soft spot is in Section 4. After defining I0 = {d j}, equation (10) sums |D_y - (y/j)I0|, but the arithmetic progression that meets D_y is t + (y/j){0,...,m-1}, so the summand should be |D_y - (y/j){0,...,m-1}|. As printed, (y/j)I0 contains an extra factor (y/j), so Lemma 5 does not apply. This looks like a notation slip rather than a fatal flaw, but as written the proof of Theorem 1 does not go through at that step. A clean rewrite of the progression notation and the application of Lemma 5 is needed. The later step Pr[|B_I| ≥ p^{-2ε}m] ≥ p^{-2ε} also doesn't follow from the expectation alone; it needs a second-moment or a smaller threshold. That's minor by comparison and fixable.\n\nIf these are repaired, the central claims are plausible and well-motivated. The reduction to the finite-field setting, the Fourier analysis in Lemma 5, and the Shearer-based iteration in Section 5 all check out at the level I read them. Citation practice looks appropriate: Oberlin defined β_h, Katz-Tao gave the bound, and the authors are honest about not having exact values.\n\nRecommend sending to peer review. The referee should focus on Section 4's scaling. This is a paper for specialists in additive combinatorics and Kakeya-type problems; I'd bring it to reading group once the fix is in.","headline":"New equivalence and higher-dimensional bounds, but Section 4 has a mis-scaled sumset that needs a fix before the proof works as written.","tokens_in":11827,"tokens_out":2295,"would_cite":true,"duration_ms":21496,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11B30"],"pacs":[],"model":"deepseek-v4-flash","headline":"For every finite set of rational directions, the homogeneous arithmetic Kakeya inequality matches the original one.","keywords":["arithmetic Kakeya","homogeneous inequality","Shannon entropy","additive combinatorics","Besicovitch sets","finite-field sumsets","Minkowski dimension","Kakeya conjecture"],"falsifier":"One could count good indices in Lemma 5 numerically: for a moderately large prime $p$, take $m\\approx p^{1/2}$, pick $A$ of size about $p/m$, and test how many $j\\in[m/2,m]$ satisfy $|A+j^{-1}I|\\ge p^{1-\\varepsilon}$; fewer than $mp^{-\\varepsilon}$ good indices would refute the lemma. Alternatively, an exhaustive search over small sets $R\\subset\\mathbb{Q}$ and small-support random variables $(X,Y)$ could look for a violation of $\\beta_h(R)=\\beta(R)$; the theorem asserts none exists.","tokens_in":10759,"feed_emoji":"📐","tokens_out":12248,"duration_ms":119299,"temperature":0.7,"pith_summary":"The paper proves that the 'homogeneous' version of the arithmetic Kakeya inequality, which includes an extra conditional-entropy term, is equivalent to the original inequality for every finite set of rational directions $R\\subset\\mathbb{Q}$ (Theorem 1). The homogeneous form is the one that can be iterated, so the equivalence means any bound on the ordinary arithmetic Kakeya constant $\\beta(R)$ automatically upgrades to the stronger inequality that higher-dimensional Kakeya problems need. Using this, the paper proves a quantitative bound on the $d$-dimensional arithmetic Kakeya constant $\\beta(R^d)$ in terms of $\\beta(R)$, and as a corollary obtains a new lower bound on the Minkowski dimension of $(n,d)$-Besicovitch sets, the sets containing a translate of every $d$-dimensional unit disk. A reader should care because these Besicovitch sets are the disk analogue of Kakeya sets, where the central question is how small such sets can be; the new bound improves the power of the relevant iteration.","feed_headline":"Arithmetic Kakeya inequality upgrades for free","feed_subtitle":"The upgraded inequality iterates to new lower bounds for Besicovitch sets of disks.","key_machinery":"The main new mechanism is Lemma 5, a finite-field sumset lemma. It states that if $A\\subset\\mathbb{F}_p$ has size at least $p/m$ and $I=\\{1,\\dots,m\\}$, then for at least $mp^{-\\varepsilon}$ of the indices $j\\in[m/2,m]$ the dilated sumset $A+j^{-1}I$ has size at least $p^{1-\\varepsilon}$. The proof of Theorem 1 uses this lemma on the fibers $D_y$ of the auxiliary graph: it guarantees that enough random arithmetic progressions meet enough fibers, so that the restricted graph $(X',Y')$ inherits entropy bounds close to those of $(X,Y)$. Applying the ordinary inequality (2) to $(X',Y')$ and rearranging yields the homogeneous inequality (4). Theorem 2 then rests on an iterative telescoping identity that applies the homogeneous inequality one coordinate at a time, with a permutation-averaging step producing the factor $(1-1/(\\delta_1\\cdots\\delta_d))^{-1}$ where $\\delta_i=\\beta(R_i)/(\\beta(R_i)-1)$.","core_discovery":"On its own terms, the paper's central claim is that for any finite $R\\subset\\mathbb{Q}$ with $|R|\\geq 2$, $\\beta_h(R)=\\beta(R)$. Here $\\beta(R)$ is the least $\\beta$ such that $H(Y)\\leq \\beta\\max_{r\\in R}H(X+rY)$ for all finitely supported integer random variables $X,Y$, and $\\beta_h(R)$ is the least $\\beta$ such that $H(Y)+(\\beta-1)H(X|Y)\\leq \\beta\\max_{r\\in R}H(X+rY)$. Theorem 1 gives the nontrivial direction $\\beta_h(R)\\leq\\beta(R)$: the homogeneous inequality comes for free once the ordinary one is known. The proof converts the entropy data into a graph in $\\mathbb{F}_p^2$, restricts it to a random arithmetic progression, applies the original inequality to the restricted random variables, and then rearranges the resulting entropy identities. With homogeneity in hand, Proposition 6 proves Theorem 2, the bound $\\beta(R^d)\\leq d\\,\\frac{(\\beta/(\\beta-1))^d}{(\\beta/(\\beta-1))^d-1}$, and Corollary 1 gives $\\dim_M K\\geq \\frac{d}{\\beta(R)}n$ for every $(n,d)$-Besicovitch set $K$. Numerically, using the best current bound $\\beta(R)\\leq\\alpha=1.675\\ldots$ for suitable $R$, this reads $\\dim_M K\\geq \\frac{(\\alpha/(\\alpha-1))^d-1}{(\\alpha/(\\alpha-1))^d}n$.","pith_inferences":["The equality $\\beta_h(R)=\\beta(R)$ suggests a broader principle: for entropy inequalities that sit naturally in a tensor product, the homogeneous strengthening may often be automatic, and the proof of Theorem 1 gives a concrete model for discovering such upgrades.","The new upper bound for $\\beta(R^d)$ is probably not tight: since any affine-spanning $R\\subset\\mathbb{Q}^d$ has $d\\le\\beta(R)\\le d+1$, the formula interpolates between the trivial endpoints, and sharper $d$-dimensional bounds would immediately improve the Besicovitch dimension constant.","Lemma 5 is asymptotic in the prime $p$; extracting an effective quantitative version would make Corollary 1's constant explicit for fixed $n,d$, and then random constructions of Besicovitch sets could be tested against the predicted dimension."],"forward_implications":["Any bound on the one-dimensional arithmetic Kakeya constant $\\beta(R)$ immediately gives the same bound on the homogeneous constant $\\beta_h(R)$; in particular the known $\\alpha=1.675\\ldots$ bound now applies to the homogeneous inequality without rewriting its proof.","The bound $\\beta(R^d)\\le d\\,(\\beta/(\\beta-1))^d/((\\beta/(\\beta-1))^d-1)$ gives explicit numerical constants for product sets; for example $\\beta(\\{0,1\\}^d)\\le d\\,2^d/(2^d-1)$.","For $(n,d)$-Besicovitch sets, the Minkowski dimension lower bound becomes $\\dim_M K\\ge \\bigl((\\alpha/(\\alpha-1))^d-1\\bigr)/(\\alpha/(\\alpha-1))^d\\,n$ with $\\alpha=1.675\\ldots$.","Because the homogeneous inequality iterates cleanly, the proof also supplies a template for turning similar one-dimensional entropy inequalities into higher-dimensional geometric estimates, and the same iterative structure applies to finite-field variants."],"supporting_citations":[{"why":"Supplies the entropy-to-sumset equivalence used as Lemma 3 to convert the entropy inequality into a finite graph.","marker":"[21]"},{"why":"Provides the reduction of such graphs to a finite field $\\mathbb{F}_p^2$, used after the entropy-to-set conversion.","marker":"[12]"},{"why":"Gives the previous bound $\\alpha=1.675\\ldots$ for the one-dimensional arithmetic Kakeya constant and the iteration scheme used to state the corollary.","marker":"[15]"},{"why":"Introduced the homogeneous iteration for disk Kakeya; Theorem 1 provides the homogeneous input this iteration requires, yielding the new dimension corollary.","marker":"[20]"},{"why":"Bridges arithmetic Kakeya bounds to Minkowski-dimension lower bounds for Euclidean Kakeya sets, which grounds Corollary 1.","marker":"[4]"}],"fun_headline_variants":["Homogeneous arithmetic Kakeya inequality comes free","Iterative argument upgrades arithmetic Kakeya bounds","Free homogeneity boosts Besicovitch set dimensions","Generalized arithmetic Kakeya: sum-difference inequality upgrade","Arithmetic Kakeya: homogeneous inequality for free"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument stands on Lemma 5, the new claim that for every large subset $A$ of a finite field and every short interval $I$, many dilates $j^{-1}I$ expand $A$ to size at least $p^{1-\\varepsilon}$; if that claim fails, the random-progression restriction does not preserve the entropy bounds and Theorem 1 collapses.","fun_headline_variants_meta":{"raw":{"variants":["Homogeneous arithmetic Kakeya inequality comes free","Iterative argument upgrades arithmetic Kakeya bounds","Free homogeneity boosts Besicovitch set dimensions","Generalized arithmetic Kakeya: sum-difference inequality upgrade","Arithmetic Kakeya: homogeneous inequality for free"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001032,"raw_usage":{"total_tokens":4367,"prompt_tokens":983,"completion_tokens":3384,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":599,"completion_tokens_details":{"reasoning_tokens":3323}},"tokens_in":599,"tokens_out":3384,"duration_ms":25371,"temperature":1.0,"reasoning_tokens":3323,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T16:30:53.324579+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"One could count good indices in Lemma 5 numerically: for a moderately large prime $p$, take $m\\approx p^{1/2}$, pick $A$ of size about $p/m$, and test how many $j\\in[m/2,m]$ satisfy $|A+j^{-1}I|\\ge p^{1-\\varepsilon}$; fewer than $mp^{-\\varepsilon}$ good indices would refute the lemma. Alternatively, an exhaustive search over small sets $R\\subset\\mathbb{Q}$ and small-support random variables $(X,Y)$ could look for a violation of $\\beta_h(R)=\\beta(R)$; the theorem asserts none exists.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the entropy-to-sumset equivalence used as Lemma 3 to convert the entropy inequality into a finite graph."},{"cited_title":"Green, I","cited_arxiv_id":null,"evidence_quote":"Provides the reduction of such graphs to a finite field $\\mathbb{F}_p^2$, used after the entropy-to-set conversion."},{"cited_title":"Katz and T","cited_arxiv_id":null,"evidence_quote":"Gives the previous bound $\\alpha=1.675\\ldots$ for the one-dimensional arithmetic Kakeya constant and the iteration scheme used to state the corollary."},{"cited_title":"Two bounds for the X-ray transform","cited_arxiv_id":null,"evidence_quote":"Introduced the homogeneous iteration for disk Kakeya; Theorem 1 provides the homogeneous input this iteration requires, yielding the new dimension corollary."},{"cited_title":"Bourgain, λ p-sets in analysis: results, problems and related aspects, i n: Handbook of the geometry of Banach spaces, Vol","cited_arxiv_id":null,"evidence_quote":"Bridges arithmetic Kakeya bounds to Minkowski-dimension lower bounds for Euclidean Kakeya sets, which grounds Corollary 1."}],"review_version":1}