{"id":"2d060a6d-5a7c-4b20-a7f8-f6b1cd364604","arxiv_id":"2505.22838","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":2.0,"correctness_risk":"low","formal_verification":"none","parameter_count":1,"one_line_summary":"Several classical bounds in design and coding theory are shown to follow from the single inequality S0S2 - S1^2 ≥ 0.","lead":"This expository note shows that the variance method, the simple fact that a sum of squared deviations is nonnegative, gives a single proof template for several classic bounds in design theory and coding theory. A general reader might read it to see how Fisher's inequality, the Johnson bound, and other classical results all fall out of one inequality.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 2 delegates the crucial Fisher simplification to [4] without showing it; the omission is the main load-bearing gap, though the algebra itself is correct and easily checkable.","rationale":"The reader's weakest_assumption is exactly the right target. All other applications—Plackett-Burman, Johnson, Stanton-Kalbfleisch, Mullin-Vanstone—are self-contained: from the displayed S0, S1, S*, S2 values, the bound follows by a few lines of algebra. Fisher is the exception: the crucial algebra is outsourced to a 1940 paper. Because the abstract and introduction promise that all five bounds are derived by applying Theorems 1.1 and 1.3, a gap at the first derivation would undermine the central claim. My independent check shows the algebra is correct, so the gap is repairable and does not invalidate the paper. The secondary issue in Theorem 8.1—'any integer ell' despite ell^2+ell in the denominator and a proof that requires ell >= 1 for the sign of 2*ell*S1—is real but peripheral; it does not bear on the central unification claim. Thus I do not change the reader's conditional verdict, but I do emphasize the Fisher step as the single step that must be fixed.","tokens_in":9202,"tokens_out":11747,"duration_ms":125705,"concrete_test":"Use a CAS to substitute b=vr/k and lambda=r(k-1)/(v-1) into (b-1)(k(k-1)(lambda-1)+k(r-1))-k^2(r-1)^2, factor, and confirm the result is exactly (r-k)(v-k)(r-lambda). If the factor matches, Section 2's omitted algebra is correct and the unification claim survives; if it differs, the flagship derivation fails. As a secondary check, test Theorem 8.1 with ell=0 and ell=-1 to confirm the statement needs the restriction ell >= 1 (or ell^2+ell > 0).","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that Fisher's inequality is a direct specialization of Theorem 1.1. The counting identities (9)-(11) and the substitution into (2) are correct, but the transition from (12) to (13) is asserted with 'a considerable amount of non-obvious simplification' and the reader is sent to Fisher [4]. That step is load-bearing: if it failed, the unification program would break at its first and flagship example. I checked the algebra using (7) and (8): writing b=vr/k and lambda=r(k-1)/(v-1), the left side of (12) factors exactly as (r-k)(v-k)(r-lambda), so the inequality is true. Thus the concern is not correctness of the mathematics but incompleteness of the proof as presented: no derivation is exhibited, and the paper cannot be verified independently from its own text at this step. This supports the conditional verdict rather than a clean accept.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a unified treatment of several classical inequalities in design theory and coding theory by viewing them as consequences of the variance nonnegativity inequality S0S2 - S1^2 >= 0 (Theorem 1.1) and a variant S0C - B^2 >= 0 for situations where one has a lower bound on S1 (Theorem 1.3). The author derives Fisher's inequality, the Plackett-Burman bound, the second Johnson bound, the Stanton-Kalbfleisch bound, the Mullin-Vanstone bound, and the two-point sampling error bound as specializations of these inequalities, and then discusses an extension based on sum_i (a_i - l)(a_i - l - 1) >= 0. Each application sets up the relevant S0, S1, and S* from design or code identities and reduces the resulting polynomial inequality.","tokens_in":9321,"tokens_out":10116,"duration_ms":110726,"significance":"If the claims hold, the note succeeds in showing that these classic bounds are not merely analogous but are direct instances of a single quadratic nonnegativity principle. The main strengths are the clean template, the absence of circularity (none of the target inequalities is used as an input), and the fact that the author's own previously published results are re-derived in the note rather than assumed. The paper is also honest about an earlier proof error in [14]. The contribution is expository rather than original, but it could serve as a useful reference for the variance method in combinatorial designs and coding theory.","major_comments":[{"comment":"The proof of Fisher's inequality stops at (12) and asserts that 'a considerable amount of non-obvious simplification' yields (r - k)(v - k)(r - lambda) >= 0, referring the reader to [4]. This step is load-bearing because Fisher's inequality is the flagship application of the variance template, and (13) is not self-evident from (12), (7), and (8). I checked the algebra: substituting b = vr/k and lambda = r(k - 1)/(v - 1) into the left side of (12) factors exactly as (r - k)(v - k)(r - lambda), so the statement is correct. However, the manuscript should supply this factorization or a few lines of algebra rather than delegating it to a 1940 paper; otherwise the first derivation is not verifiable from the paper's own text.","section":"Section 2, Eq. (12)-(13)"},{"comment":"Theorem 8.1 states the bound for 'any integer l', but the displayed formula contains l^2 + l in the denominator and the proof divides by this quantity. For l = 0 and l = -1 the bound is undefined. Moreover, in the chain '0 = ... <= (v - k)(v - k - 1) - 2lk(v - k) + (l^2 + l)(b - 1)', replacing S1 by its lower bound k(v - k) in the term -2lS1 is only valid in that direction when l >= 0; for negative l the inequality sign reverses. The intended application l = floor((v - 1)/k) is always at least 1, so the main example is unaffected, but the theorem as stated needs a qualification such as 'for every integer l >= 1', and the case l <= -2, if kept, requires a separate argument.","section":"Theorem 8.1, Eq. (26)"}],"minor_comments":[{"comment":"The sentence 'It is interesting to to note' contains a duplicated 'to'.","section":"Remark 1.2"},{"comment":"The theorem referred to as the 'Erdos-de Bruijn Theorem' is more standardly called the de Bruijn-Erdos theorem, matching the order of the authors in reference [1].","section":"Section 5"},{"comment":"The symbol A is used both for the orthogonal array and for the randomized algorithm, which makes sentences such as 'we run the algorithm A with randomness specified by the points in row r of A' slightly confusing. A different letter for one of these objects would improve readability.","section":"Section 7"},{"comment":"There are small grammatical repetitions such as 'If I is a a no-instance' and 'a a yes-instance' in the discussion of the yes-biased algorithm.","section":"Section 7"}],"recommendation":"major_revision","confidential_remarks":"The paper is explicitly expository and contains a large share of references to the author's own work; this is not a reason for rejection, but the editor should consider whether the venue expects original research contributions. The major issues identified are local and fixable: the missing algebra in Section 2 and the unqualified statement of Theorem 8.1. Once these are corrected, the note would be a serviceable unified account of the variance method."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThis is a well-written expository note, not a research paper. The one genuinely new thing is the framing: two variance inequalities, Theorems 1.1 and 1.3, as a common template for Fisher, Plackett-Burman, second Johnson, Stanton-Kalbfleisch, Mullin-Vanstone, plus the two-point sampling analysis and a couple of the author's own bounds. The derivations in Sections 3–7 are clean and check out; the counting identities are right and the algebra is simple. The author is also upfront about the earlier error in [14], which is more than most would do.\n\nThe main soft spot is Section 2. The proof of Fisher's inequality stops at inequality (12) and says a 'considerable amount of non-obvious simplification' using (7)–(8) turns it into (13), sending the reader to Fisher's 1940 paper. That step is load-bearing: it is the flagship demonstration of the unified method, and the note cannot be verified from its own text at that point. I checked the algebra myself and it does factor exactly as (r-k)(v-k)(r-λ), so the mathematics is fine; the exposition is incomplete. A referee should ask for the two or three lines of algebra to be shown.\n\nSecond minor issue: Theorem 8.1 states the bound 'for any integer ℓ', but the proof only makes sense for ℓ ≥ 0. The division by ℓ²+ℓ is undefined at ℓ=0 and ℓ=-1, and the direction of the inequality involving S1 would switch for negative ℓ. Since the recommended choice is positive anyway, this is a statement error, easily fixed.\n\nThird small thing: the novelty is organizational. The paper explicitly re-derives known bounds, and the underlying inequality is Cauchy-Schwarz in disguise. That is not a flaw for what it is, but it means the value is pedagogical and archival, not new mathematics.\n\nWho is this for? Anyone teaching design theory or coding theory who wants a single template to present several classical bounds. It would make a useful survey or teaching note if the Fisher algebra is filled in and the ℓ issue is fixed. I would not cite it as a research source, but I would point students to it.\n\nRecommendation: this deserves a serious referee, but with the expectation of a light revision, mostly the omitted algebra and the quantification in Theorem 8.1. The central unification claim holds up; it just needs to be fully self-contained.","headline":"Useful expository unification of classical variance-method bounds; the Fisher step is omitted but correct, so the note is worth reviewing with light revision.","tokens_in":9903,"tokens_out":3339,"would_cite":false,"duration_ms":36494,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05B05","05B15","94B65"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper shows that five classical bounds in design and coding theory all reduce to one variance inequality, $S_0S_2 - S_1^2 \\geq 0$.","keywords":["variance method","Fisher's inequality","Plackett-Burman bound","second Johnson bound","Stanton-Kalbfleisch bound","Mullin-Vanstone bound","orthogonal arrays","pairwise balanced designs"],"falsifier":"One can settle the Fisher step directly: expand the left side of (12) symbolically with the identities $vr=bk$ and $\\lambda(v-1)=r(k-1)$ substituted, and check whether the difference between the two sides reduces identically to zero under the BIBD identities; a single parameter tuple $(v,b,r,k,\\lambda)$ satisfying those identities for which it does not would refute the paper's derivation of Fisher's inequality.","tokens_in":8941,"feed_emoji":"🧮","tokens_out":9469,"duration_ms":97646,"temperature":0.7,"pith_summary":"The paper argues that a single elementary inequality, the nonnegativity of the variance of a list of real numbers written as $S_0S_2 - S_1^2 \\geq 0$, is a common engine behind several celebrated bounds in design theory and coding theory. A companion inequality, $S_0C - B^2 \\geq 0$, handles cases where only a lower bound on $S_1$ is known. The note shows that Fisher's inequality, the Plackett-Burman bound, the second Johnson bound, the Stanton-Kalbfleisch bound, and the Mullin-Vanstone bound all fall out by plugging counts of incidences, intersections, or weights into this template. It also applies the template to the error analysis of two-point sampling derandomization and describes an integer variant that strengthens two of the bounds. If the unification is correct, these results are not separate tricks but specializations of one proof scheme.","feed_headline":"One variance inequality unifies six classic bounds","feed_subtitle":"Counting intersections and applying one nonnegativity identity recovers Fisher, Johnson, Plackett–Burman, and more.","key_machinery":"The central object is the nonnegativity of variance, written as $S_0S_2 - S_1^2 \\geq 0$ for a list $a_1,\\dots,a_n$ with $S_0=n$, $S_1=\\sum a_i$, and $S_2=\\sum a_i^2$. In combinatorial applications one computes $S^*=\\sum \\binom{a_i}{2}$, converts it via $S_2=2S^*+S_1$, and substitutes the resulting expressions into the identity. When only a lower bound $S_1\\geq B$ is known and all $a_i\\geq 1$, Theorem 1.3 uses monotonicity of the function $f(\\epsilon)=(B+\\epsilon)^2/(C+\\epsilon)$ to obtain $S_0C-B^2\\geq 0$, with $C=2S^*+B$. The integer extension replaces the squared deviation by $\\sum (a_i-\\ell)(a_i-\\ell-1)\\geq 0$, whose equality condition is $a_i\\in\\{\\ell,\\ell+1\\}$; choosing $\\ell$ optimally strengthens the Stanton-Kalbfleisch and Johnson estimates.","core_discovery":"On the paper's own terms, the central claim is that Theorems 1.1 and 1.3 provide a proof calculus: from the variance inequality $S_0S_2 - S_1^2 \\geq 0$ (and the lower-bound variant $S_0C - B^2 \\geq 0$) each classical bound follows by substituting counts of incidences, intersections, or weights. For a BIBD one counts intersections of blocks with a fixed block; for an orthogonal array one counts occurrences of a fixed symbol in rows; for a constant-weight code one counts column weights of a 0-1 matrix; for a pairwise balanced design one deletes a fixed block and counts the sizes of the remaining blocks; for an $(r,\\lambda)$-design one counts point incidences directly. The same machinery gives the error analysis of two-point sampling derandomization and, through the integer identity $\\sum (a_i-\\ell)(a_i-\\ell-1)\\geq 0$, strengthens the Stanton-Kalbfleisch and Johnson bounds. The paper presents the bounds as immediate specializations, with the equality conditions of the identities translating into structural descriptions of extremal cases.","pith_inferences":["Beyond the paper's list, the same substitution recipe could be applied to other incidence structures, such as $t$-designs or covering designs, where a suitable $S^*$ can be counted; any bound produced this way would inherit the equality characterization for free.","Theorem 1.3's monotonicity trick is not limited to $\\epsilon=0$; if a counting problem has $S_1=B+\\epsilon$ with $\\epsilon$ bounded below, the same monotonicity could convert that slack into a sharper bound than $S_0C-B^2\\geq 0$.","The integer identity (25) is an optimization over $\\ell$; for fixed parameter sets one could compute the largest lower bound over integer $\\ell$ and compare with the classical bounds, giving a testable strengthening on known design and code tables.","The equality condition $a_i\\in\\{\\ell,\\ell+1\\}$ of the extension suggests that extremal designs satisfying the strengthened bounds must have almost-uniform block sizes; identifying the objects that meet the improved bounds is a natural classification problem."],"forward_implications":["Fisher's inequality $b\\geq v$ for every $(v,b,r,k,\\lambda)$-BIBD follows directly from the variance template, since the simplified expression forces $r-k\\geq 0$ when blocks are incomplete.","An orthogonal array $\\mathrm{OA}_\\lambda(k,n)$ can exist only when $\\lambda \\geq (k(n-1)+1)/n^2$, and a repeated-row version gives $\\lambda \\geq m(k(n-1)+1)/n^2$.","Any 0-1 matrix with row weight $r$ and pairwise inner product at most $\\lambda$ has at most $n(r-\\lambda)/(r^2-n\\lambda)$ rows (when the denominator is positive), yielding the second Johnson bound for constant-weight binary codes.","A pairwise balanced design on $v$ points containing a block of size $k$ has at least $1+k^2(v-k)/(v-1)$ blocks; equality forces a projective plane or a near-pencil, and the general bound implies the de Bruijn-Erdős theorem.","Every $(r,\\lambda)$-design satisfies $b\\geq r^2v/(r+\\lambda(v-1))$, which bounds nonincident point-line sets in a projective plane of order $q$ by $s\\leq 1+(q+1)(\\sqrt{q}-1)$."],"supporting_citations":[{"why":"Supplies Fisher's original variance proof and the non-obvious simplification to $(r-k)(v-k)(r-\\lambda)\\geq 0$ that the paper delegates rather than re-derives.","marker":"[4]"},{"why":"States the Plackett-Burman bound for orthogonal arrays that Section 3 derives from Theorem 1.1.","marker":"[9]"},{"why":"Gives the second Johnson bound and the integer-strengthened inequality reused in Section 8.","marker":"[6]"},{"why":"States the Stanton-Kalbfleisch bound for pairwise balanced designs containing a block of size $k$.","marker":"[13]"},{"why":"States the Mullin-Vanstone bound for $(r,\\lambda)$-designs used in Section 6.","marker":"[8]"},{"why":"Provides the unified orthogonal-array analysis of two-point sampling whose error bound Section 7 reproves via the variance identity.","marker":"[5]"},{"why":"Supplies the projective-plane application in which the Mullin-Vanstone bound bounds nonincident point-line sets.","marker":"[16]"},{"why":"Introduces the author's variance-method extensions, including the strengthened Stanton-Kalbfleisch bound stated in Theorem 8.1.","marker":"[14]"}],"fun_headline_variants":["Variance method unifies six classic bounds","One variance inequality yields six classical bounds","From variance identity to Fisher, Johnson, and more","Single nonnegativity step proves many classic inequalities","Variance trick: recover six bounds with one identity"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing step is the assertion that inequality (12), after using $vr=bk$ and $\\lambda(v-1)=r(k-1)$, simplifies to $(r-k)(v-k)(r-\\lambda)\\geq 0$; the simplification is not shown in the paper and is delegated to Fisher's 1940 paper, so the flagship Fisher-inequality application stands or falls on that algebra.","fun_headline_variants_meta":{"raw":{"variants":["Variance method unifies six classic bounds","One variance inequality yields six classical bounds","From variance identity to Fisher, Johnson, and more","Single nonnegativity step proves many classic inequalities","Variance trick: recover six bounds with one identity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000271,"raw_usage":{"total_tokens":1564,"prompt_tokens":815,"completion_tokens":749,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":431,"completion_tokens_details":{"reasoning_tokens":679}},"tokens_in":431,"tokens_out":749,"duration_ms":9060,"temperature":1.0,"reasoning_tokens":679,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T13:00:14.429770+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"One can settle the Fisher step directly: expand the left side of (12) symbolically with the identities $vr=bk$ and $\\lambda(v-1)=r(k-1)$ substituted, and check whether the difference between the two sides reduces identically to zero under the BIBD identities; a single parameter tuple $(v,b,r,k,\\lambda)$ satisfying those identities for which it does not would refute the paper's derivation of Fisher's inequality.","supporting_citations":[{"cited_title":"Mullin and S.A","cited_arxiv_id":null,"evidence_quote":"States the Mullin-Vanstone bound for $(r,\\lambda)$-designs used in Section 6."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies Fisher's original variance proof and the non-obvious simplification to $(r-k)(v-k)(r-\\lambda)\\geq 0$ that the paper delegates rather than re-derives."},{"cited_title":"Plackett and J.P","cited_arxiv_id":null,"evidence_quote":"States the Plackett-Burman bound for orthogonal arrays that Section 3 derives from Theorem 1.1."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the second Johnson bound and the integer-strengthened inequality reused in Section 8."},{"cited_title":"Proc. Second Chapel Hill Conf. on Combinatorics","cited_arxiv_id":null,"evidence_quote":"States the Stanton-Kalbfleisch bound for pairwise balanced designs containing a block of size $k$."},{"cited_title":"Gopalakrishnan and D.R","cited_arxiv_id":null,"evidence_quote":"Provides the unified orthogonal-array analysis of two-point sampling whose error bound Section 7 reproves via the variance identity."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the projective-plane application in which the Mullin-Vanstone bound bounds nonincident point-line sets."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the author's variance-method extensions, including the strengthened Stanton-Kalbfleisch bound stated in Theorem 8.1."}],"review_version":1}