{"id":"365e0087-f0e5-46fd-b157-b3ceaad4e439","arxiv_id":"2502.03440","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The probability that n random d-dimensional lattice vectors are pairwise equidistant is asymptotically an explicit moment-dependent constant divided by d^{(n(n-1)/2 - 1)/2} as d grows.","lead":"Random points with independent integer-valued coordinates almost never form equidistant configurations in high dimensions: the chance decays like a constant divided by a power of the dimension. This paper computes that constant exactly from the first four moments of the coordinate distribution, for any finite support and any number of points.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 3.3 omits the normalization used in its proof: as stated, (3.6) is not scale-invariant and fails for X uniform on {0,2}, whose equidistance probability equals that of X/2 but whose formula constant differs.","rationale":"The central asymptotic and its proof are structurally sound: the lattice analysis in Section 2, the covariance determinant calculation in Propositions A.3 and A.4, and the local CLT step are coherent, and the constants check in the normalized Bernoulli case. The reader's concern about VarX=0 is real but minor: an explicit non-degeneracy assumption is missing. The more load-bearing problem is scale normalization. The equidistance event is invariant under global dilation of X, so any correct asymptotic constant must be scale-invariant; (3.6) is not. The proof implicitly performs a normalization (translate to 0, make integer with gcd 1) but does not record it in the theorem statement. The explicit counterexample X uniform on {0,2} gives the same p_d as X/2 ~ Ber(1/2), yet the printed formula predicts a constant off by exactly the scaling factor. This is not a rejection of the method; the repair is straightforward — either state the normalization and use normalized moments, or carry the lattice volume |LatV| explicitly, as Theorem 3.8 already does through |LatX1| and |LatX2|. With that repair, the central claims should hold. Hence the appropriate verdict is conditional acceptance rather than unconditional acceptance.","tokens_in":22768,"tokens_out":12113,"duration_ms":115898,"concrete_test":"For n=3, take X uniform on {0,2}. Since x → 2x is a bijection, p_d for X equals p_d for Ber(1/2), whose asymptotic constant is 2/(π√3) by (3.1) and also by (3.6). Now evaluate (3.6) directly with VarX=1 and C1=0: the predicted constant becomes 1/(8π√3). If the theorem is meant for arbitrary finite lattice support, this mismatch disproves (3.6) as printed. If it is meant only after translation/scaling normalization, the statement must say so and must use the moments of the normalized variable.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's main lattice result, Theorem 3.3, is not correct as stated for an arbitrary finite lattice distribution. The proof begins a reduction: after translating so that 0 is in the support, the support is made integer with gcd 1, and then |LatV| is computed as 2^{m-n}. But the theorem statement gives the asymptotic directly in terms of the moments of the original X, with no mention of this normalization. This matters because the event that all pairwise Euclidean distances are equal is invariant under X → cX, whereas the right-hand side of (3.6) is not: VarX → c^2 VarX and C1 → c^4 C1 make the square-root factor scale as c^{2(m-1)}, changing the claimed probability by c^{-2(m-1)}. Concretely, for n=3, m=3, take X uniform on {0,2}. The coordinate map x → 2x is a bijection, so p_d for X equals p_d for Ber(1/2), whose asymptotic constant from (3.1)/(3.6) is 2/(π√3). But plugging VarX=1, C1=0 into (3.6) gives the constant 1/(8π√3), off by c^{-2(m-1)} = 1/16. The local CLT machinery is sound if the statement is repaired by evaluating the moments on the normalized variable or by inserting the lattice volume factor explicitly. The degenerate case VarX=0, noted by the reader, is a second, smaller missing hypothesis.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies the probability p_d that n independent d-dimensional vectors with iid entries from a finite-lattice or finite-support distribution X form an equidistant set under Euclidean distance. The authors rewrite the equidistance condition as a sum of iid difference vectors equal to zero, apply the multidimensional local limit theorem of Krafft, and reduce the problem to two lattice/linear-algebra tasks: computing the fundamental volume of the lattice generated by the support of the increments, and computing the determinant of their covariance matrix. For the binary case they obtain the explicit constant in (3.1); for general lattice X they state the closed-form constant (3.6) in terms of VarX and C1=Var((X-EX)^2); and for arbitrary finite support they state a more general formula (3.18) involving lattice volumes of the sets of products of support elements, using a Hamel-basis embedding.","tokens_in":23121,"tokens_out":30291,"duration_ms":244364,"significance":"If the stated formulas hold under the right hypotheses, the result is a clean and useful high-dimensional asymptotic: it identifies the 1/d^{(m-1)/2} rate and gives explicit constants without free parameters. The proof strategy is original and transparent: the lattice basis for the overlapping map H in Theorem 2.1 is explicit and checkable, the derivation of |Lat V|=2^{m-n} in Lemma 3.2 is constructive, and the determinant computation using the spectral decomposition of the line graph of K_n (Propositions A.1, A.3, A.4) is a nice piece of linear algebra. The general finite-support theorem 3.8 is ambitious and, after a careful check, the covariance identity C0-2C1=4(VarY)^∘2 is valid in the stated vector setting. The main weakness is that Theorem 3.3, as stated, is not invariant under scaling of X and therefore is false without an explicit normalization assumption; this is a localized but load-bearing error in the central theorem.","major_comments":[{"comment":"The statement as written is not invariant under scaling of X, although the event is. In the proof in Section 3.2 the authors first translate so that 0 is in the support, multiply by the lcm to make the support integer, and divide by the gcd so that gcd Supp X = 1; only after that normalization is |Lat V| = 2^{m-n} used. Formula (3.6), however, is expressed through the moments of the original X. Under X -> cX, VarX -> c^2 VarX and C1 -> c^4 C1, so the right-hand side of (3.6) changes by c^{-2(m-1)}, while p_d is unchanged. Concretely, for n=3 and X uniform on {0,2}, the equidistance probability is the same as for Ber(1/2), for which (3.1) gives constant 2/(pi sqrt(3)); substituting VarX=1, C1=0 into (3.6) gives 1/(8 pi sqrt(3)), which is smaller by a factor of 16. The theorem must either state the formula for the normalized variable (integer support with gcd 1 after translation), or include the missing lattice-volume/normalization factor explicitly.","section":"Theorem 3.3, Eq. (3.6)"},{"comment":"The theorem does not exclude a degenerate lattice distribution: if X is a constant, then p_d=1 for every d, but VarX=0 and C1=0, so the right-hand side of (3.6) is undefined (division by zero), and the covariance matrix in the local CLT is singular. Add an assumption such as |Supp X| >= 2 or VarX > 0 to Theorem 3.3; Theorem 3.8 similarly needs a non-degeneracy assumption ensuring positive-definite covariance and a positive dimension ell.","section":"Theorem 3.3 and Theorem 3.8, hypotheses"}],"minor_comments":[{"comment":"The first line says 'in particular Theorem 3.3' where the intended reference is Theorem 2.1, since the basis of Lat H is being used.","section":"Proof of Lemma 3.2"},{"comment":"After (3.19), the displayed definitions 'C0 := (Var(Y1-Y2))^2' and 'C1 := Cov(Y1-Y2, Y1-Y3)' do not match the theorem statement; they should read C0 := Var((Y1-Y2)^2) and C1 := Cov((Y1-Y2)^2, (Y1-Y3)^2). As written, the types of these objects are unclear.","section":"Proof of Theorem 3.8"},{"comment":"The word 'denumerator' should be 'denominator' in the sentence after (3.19).","section":"Proof of Theorem 3.8"},{"comment":"The abstract claims the result for all finitely supported X with an explicit constant in terms of the first four moments; for the finite-support case the statement in Theorem 3.8 also depends on lattice volumes |Lat X1| and |Lat X2|, so the abstract should be phrased to avoid suggesting that first moments alone determine the constant in the general finite-support setting.","section":"Abstract and Introduction"}],"recommendation":"major_revision","confidential_remarks":"The scale-invariance problem in Theorem 3.3 is real, but the proof already contains the necessary normalization, so I expect a revision can fix it by restating the theorem and the abstract. I did not find evidence of circularity or parameter fitting."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First, the bad news: Theorem 3.3 is not correct as stated. The event of equidistance is invariant under scaling X by a positive constant, but the right-hand side of (3.6) is not. Take X uniform on {0,2}: the equidistance probability equals that of Ber(1/2), so the asymptotic constant should be 2/(π√3), but plugging VarX=1 and C1=0 into (3.6) gives 1/(8π√3), off by a factor of 1/16. The proof itself normalizes: it translates the support to contain 0, rescales to integer support with gcd 1, and then uses the lattice volume 2^{m-n} for that normalized variable. The theorem statement never imports that normalization back into the moments, so the stated formula is missing a scale-dependent lattice factor. That is a load-bearing flaw, not a cosmetic one.\n\nThat said, the paper has real substance. The binary case, Theorem 3.1, is complete and correct. Theorem 2.1, the explicit lattice basis for the overlapping map H, is a clean and genuinely useful result. The idea of applying Krafft's multidimensional local CLT and reducing the covariance determinant to the line-graph eigenvalue structure is elegant, and the determinant computations in the appendix are checkable. For a properly normalized lattice variable, the proof strategy works; the fix is to state Theorem 3.3 for a normalized X or to include the explicit lattice volume factor as a function of the scaling.\n\nThe degenerate case VarX=0 is a second, smaller missing hypothesis: the formula blows up while the true probability is 1. That is easy to patch by excluding degenerate X.\n\nI have not gone line-by-line through Theorem 3.8, but the lattice-volume notation there seems to carry the scaling more honestly, so the general finite-support theorem may survive. Still, the authors should verify the scaling behavior explicitly. The general-case proof is sketchier in places, but the components are present.\n\nBottom line: the paper deserves a serious referee—the method is original and the main flaw is repairable—but it should not be accepted in its current form. A referee should ask the authors to make the normalization explicit and exclude the degenerate case.","headline":"Theorem 3.3 as stated is not scale-invariant and fails for X uniform on {0,2}, but the proof gives a sound method for a normalized version; the paper deserves a serious referee.","tokens_in":23600,"tokens_out":3331,"would_cite":false,"duration_ms":29505,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60F05","60G50","05C50","11H06"],"pacs":[],"model":"deepseek-v4-flash","headline":"In d dimensions, the chance that n iid lattice vectors are pairwise equidistant decays as an explicit constant times a power of 1/d, with the exponent fixed by n.","keywords":["Multidimensional local limit theorem","Discrete lattice","Line graphs","Spectral graph theory","Equidistant points","Equidistant binary codes","Lattice distributions","Johnson graphs"],"falsifier":"Take $n=3$ and let $X$ be the constant random variable $X\\equiv 0$. Then for every $d$ all three vectors coincide, all distances are $0$, and $p_d=1$; substituting $\\operatorname{Var} X=0$ and $C_1=0$ into formula (3.6) makes the denominator zero, so the asymptotic statement as written fails for this degenerate input.","tokens_in":22602,"feed_emoji":"🎲","tokens_out":17560,"duration_ms":134971,"temperature":0.7,"pith_summary":"Let $X_1,\\dots,X_n$ be independent $d$-dimensional vectors whose entries are iid samples from a finite lattice distribution $X$, and let $p_d$ be the probability that all their pairwise Euclidean distances are equal. This paper proves that as $d \\to \\infty$, $p_d \\sim C_n / d^{(m-1)/2}$ with $m = \\binom{n}{2}$, where the constant $C_n$ is given explicitly in terms of the first four moments of $X$ (Theorem 3.3). The same method also handles the symmetric Bernoulli case, all finitely supported distributions (Theorem 3.8), and other distances such as $L^p$ distances. The result gives the asymptotic density of equidistant families in high-dimensional discrete spaces, connecting a probabilistic question to equidistant binary codes and to the geometry of lattices.","feed_headline":"Equal-distance chance of n lattice points decays as a power law","feed_subtitle":"The paper pins down the exponent and constant from the first four moments of the entry distribution.","key_machinery":"The mechanism has three parts. First, the multidimensional local limit theorem for lattice distributions converts the probability into $| \\operatorname{Lat} V | (2\\pi d)^{-(m-1)/2} |\\operatorname{Var}(V)|^{-1/2}$ for the increment vector $V = ((X_i-X_j)^2-(X_1-X_2)^2)_{(i,j)\\ne(1,2)}$. Second, the paper analyzes the image of the 'overlapping map' $H:\\{0,1\\}^n \\to \\{0,1\\}^{\\binom{n}{2}}$, $(v_1,\\dots,v_n)\\mapsto(\\mathbf{1}_{v_i\\ne v_j})_{i<j}$; Theorem 2.1 gives a basis of this lattice consisting of $H([1:i])$ for $i<n$ together with all $2e_{i,j}$ with $j<n$, and after removing the $(1,2)$ coordinate this yields a universal basis with fundamental volume $2^{m-n}$. Third, the covariance determinant is computed by recognizing the $m\\times m$ matrix of overlap counts as the adjacency matrix of the line graph $L(K_n)$ (the triangular graph), whose eigenvalues are $2n-4$, $n-4$, and $-2$ with multiplicities $1$, $n-1$, and $m-n$; the determinant then follows from a block-arrowhead calculation in the basis $\\{1, H_{1,2}, H_2,\\dots,H_n\\}$.","core_discovery":"The central claim (Theorem 3.3) is that for iid $d$-dimensional vectors with entries from a lattice distribution $X$, the probability $p_d$ that all $\\binom{n}{2}$ pairwise Euclidean distances are equal satisfies, as $d \\to \\infty$,\n$$p_d \\sim \\frac{1}{$d^{{(m-1)/2}}$} \\left( m(2\\pi)^{m-1}(\\operatorname{Var} X)^{2(m-n)}(4(\\operatorname{Var} X)^2 + (n-2)C_1)^{n-1} \\right)^{-1/2},$$\nwith $m = \\binom{n}{2}$ and $C_1 = \\operatorname{Var}((X - \\mathbb{E}[X])^2)$. The proof reduces the event to a partial-sum probability for iid increment vectors, applies a multidimensional local limit theorem, and computes the two ingredients explicitly: the lattice spanned by the increments has fundamental volume $2^{m-n}$ for every non-degenerate integer-valued $X$ (after rescaling and taking gcd $1$), and the determinant of the covariance matrix of the increments is evaluated through the spectrum of the adjacency matrix of the line graph of the complete graph $K_n$. The general finite-support case (Theorem 3.8) keeps the same structure after embedding the support into $\\mathbb{Q}^{\\ell}$.","pith_inferences":["A natural extension would be to replace 'all distances equal' with a prescribed multiset of distances, e.g. exactly two distinct distance values; the same lattice and covariance machinery would apply to principal submatrices of the same line-graph matrix, and the exponent should again be $d^{-\\ell(m-1)/2}$ with the same $\\ell$.","The explicit eigenvectors of $L(K_n)$ used here also suggest a route to second-order corrections via Edgeworth expansions of multidimensional lattice distributions, refining the '$\\sim$' relation to an asymptotic expansion in powers of $d^{-1/2}$.","The degenerate-case breakdown indicates that a separate analysis of near-constant distributions (with variance tending to $0$ slowly in $d$) could reveal a crossover regime where the equidistance probability transitions from $1$ to the power law; this regime is not treated in the paper."],"forward_implications":["In the binary case $X\\sim\\operatorname{Ber}(1/2)$, equidistant vectors are exactly equidistant binary codes; the theorem yields their asymptotic density as $\\sqrt{2^{3m-2n-1}/(m\\pi^{m-1})}\\,d^{-(m-1)/2}$.","Because the lattice volume factor $2^{m-n}$ is independent of $X$, the main-order constant for any integer-valued $X$ with $\\gcd(\\operatorname{Supp}X)=1$ depends on $X$ only through $\\operatorname{Var}X$ and $C_1$, i.e. its first four moments.","For fixed $n$, the probability decays as $d^{-(\\binom{n}{2}-1)/2}$: triples become linearly rare in $d$, while larger equidistant families vanish faster.","The method extends to $L^p$ distances for integer-valued $X$ and integer $p$, and Theorem 3.8 covers arbitrary finite supports by embedding the support into $\\mathbb{Q}^{\\ell}$; so the same power law governs a wide family of distance notions."],"supporting_citations":[{"why":"Supplies the multidimensional local limit theorem (eq. 3.3) that converts the partial-sum probability into the asymptotic ratio involving the lattice fundamental volume and the covariance determinant.","marker":"Krafft [1967]"},{"why":"Provides the standard line-graph eigenvalue fact (Proposition 1.4.1) underlying the spectrum of the triangular graph used in Proposition A.1.","marker":"[Brouwer and Haemers, 2012]"},{"why":"Characterises the (-2)-eigenspace of line graphs via the cycle/matroid structure, which the paper uses to describe the eigenvectors for L(K_n).","marker":"Doob [1973]"},{"why":"Origin of the multidimensional local limit theorem for lattice distributions on which the method builds.","marker":"Meizler et al. [1949]"}],"fun_headline_variants":["Equidistance probability in lattice vectors decays as d^{-(m-1)/2}","Random lattice points: all-equal distances shrink as d^{-(m-1)/2}","Exact power law for equidistant events in high-dimensional lattices","Four moments give explicit constant for n-point equidistance probability","Lattice vectors in high-D: equal-distance chance decays as d^{-(m-1)/2}"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The asymptotic formula is derived from a multidimensional local limit theorem that needs the increment vectors to have a non-singular covariance matrix and a support spanning a full-rank lattice; the paper verifies this for non-degenerate $X$, but it never explicitly excludes a constant random variable, for which the formula is undefined although the equidistance probability is exactly $1$.","fun_headline_variants_meta":{"raw":{"variants":["Equidistance probability in lattice vectors decays as d^{-(m-1)/2}","Random lattice points: all-equal distances shrink as d^{-(m-1)/2}","Exact power law for equidistant events in high-dimensional lattices","Four moments give explicit constant for n-point equidistance probability","Lattice vectors in high-D: equal-distance chance decays as d^{-(m-1)/2}"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001178,"raw_usage":{"total_tokens":4911,"prompt_tokens":1031,"completion_tokens":3880,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":647,"completion_tokens_details":{"reasoning_tokens":3773}},"tokens_in":647,"tokens_out":3880,"duration_ms":28242,"temperature":1.0,"reasoning_tokens":3773,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T04:46:03.029991+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take $n=3$ and let $X$ be the constant random variable $X\\equiv 0$. Then for every $d$ all three vectors coincide, all distances are $0$, and $p_d=1$; substituting $\\operatorname{Var} X=0$ and $C_1=0$ into formula (3.6) makes the denominator zero, so the asymptotic statement as written fails for this degenerate input.","supporting_citations":[{"cited_title":"A multidimensional local limit theorem for lattice distributions, 1967","cited_arxiv_id":null,"evidence_quote":"Supplies the multidimensional local limit theorem (eq. 3.3) that converts the partial-sum probability into the asymptotic ratio involving the lattice fundamental volume and the covariance determinant."},{"cited_title":"An interrelation between line graphs, eigenvalues, and matroids","cited_arxiv_id":null,"evidence_quote":"Characterises the (-2)-eigenspace of line graphs via the cycle/matroid structure, which the paper uses to describe the eigenvectors for L(K_n)."}],"review_version":1}