{"id":"7fee7fd5-47bb-457f-bd0d-dd64b2841369","arxiv_id":"1908.06824","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Rank of the linear-representation incidence matrix over finite fields equals 1 plus (q minus 1) times the number of hyperplanes meeting K, and the corresponding LDPC code is generated by minimum-weight plane words.","lead":"Using character theory, this paper gives a geometric formula for the rank of the incidence matrix of affine points and lines determined by a subset of a finite projective space. It proves a conjecture that the associated LDPC code is generated by its minimum-weight plane words, a structural result for coding theory.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Equation (11) in the proof of Theorem 3.1 is false as printed: the middle equality drops the condition θ(u0)=0 and the final term conflates the original incidence map with the quotient map, so the pivotal induction step of the main conjecture proof rests on an incorrect equation.","rationale":"Good-faith reading: the paper's central claim is that the F-code C is generated by plane words, proved by induction (Theorem 3.1) on |K| and driven by the rank formula of Theorem 1.1. The character-theoretic proof of Theorem 1.1 is sound for q ≠ 0 in F (the harmless extension to a field with a primitive p-th root is legitimate; Lemma 2.1 and Lemma 2.2 are standard and correct). The induction skeleton also checks out: equations (7)–(10) are correct, and the lower bound via the quotient geometry is legitimate. The failure is isolated at (11), a false displayed equation in the pivotal step. This is more serious than the reader's 'presentation issue' label, because the printed chain asserts two equalities that are numerically false; however, the intended correction is evident and I verified it in small cases, so the main theorem is very likely salvageable. Secondary concerns: (a) the q ≠ 0 hypothesis excludes binary/even-q cases exactly as the reader's weakest_assumption states; Remark 1.2 is honest about this, but it does limit the scope of 'Vandendriessche's conjecture' as proven, since LDPC codes are most commonly studied over F_2; (b) boundary bugs at K=∅ in Theorem 1.1's geometric reformulation and in Theorem 3.3's capacitor-word statement are degenerate and do not affect the main applications but should be excluded explicitly. Independent support: the Wenger-graph rank agrees with the eigenvalue multiplicity of [3], the n=3 case was previously known [12], and the LU(3,q) dimensions match [10,13], all of which corroborate the main formulas. Given that the only obstruction is a repairable false equation in the proof of the central theorem, the verdict should remain CONDITIONAL as the reader had it; I recommend no change to the verdict label, but the revision must rewrite (11) with the two η maps distinguished.","tokens_in":11447,"tokens_out":49054,"duration_ms":454802,"concrete_test":"Recompute the induction-step equality of Theorem 3.1 with the two incidence maps distinguished. Take n=2, q=3, K={u1}, u0=u2: rank η_{K′} = 5, rank η_K = 3, so the LHS of (11) is 2; as printed the middle term #{θ : θ(u1)≠0} = 6 and the last term q^{n−1} − rank η_K = 0, so both printed equalities fail. With the correction (keep θ(u0)=0; the last η_K is the quotient T*_0 incidence map) the chain reads 2 = 2 = 3−1. Repeat for n=3, q=2, K={u1,u2} inside a 7-point H: verify dim C = 2 and that the two plane words span it. If the corrected equality fails in any small case, Theorem 3.1 is in doubt.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The key step in the proof of Theorem 3.1 is the displayed equation (11), which computes the rank increment closing the induction. As printed it reads rank_F η_{K′} − rank_F η_K = #{θ ∈ V* | θ(u0)=0, θ(u)≠0 ∀u∈K} = #{θ ∈ V* | θ(u)≠0 ∀u∈K} = q^{n−1} − rank_F η_K. The middle and final equalities are false in general. The second equality drops the condition θ(u0)=0; for n=2, q=3, K={u1}, u0=u2, the first line equals #{c2=0, c1≠0} = 2 while the middle term equals 3·2 = 6. The third equality is dimensionally inconsistent with the original incidence maps: the rank difference of the original geometries is at most q^n, whereas q^{n−1} − rank η_K is only meaningful if the final η_K denotes the incidence map of the quotient geometry T*_{n−2}(K), a different matrix. The intended argument is repairable: rank_F η_{K′} − rank_F η_K = #{θ ∈ V* : θ(u0)=0, θ(u)≠0 ∀u∈K} = #{θ̄ ∈ (V/u0)* : θ̄(ū)≠0 ∀ū∈K} = q^{n−1} − rank_F η_{K,quot}; I verified this numerically for n=2, q=3 (both sides equal 2, versus 6 and 0 as printed). Thus the theorem is very likely correct, but the written proof is invalid at exactly the step that proves Vandendriessche's conjecture; this is a substantive mathematical error, not merely a typo. Two boundary defects compound it: Theorem 1.1's geometric form '1+(q−1)h_K' gives 1 for K=∅ although rank N = 0, and Theorem 3.3's capacitor-word generation fails for K=∅ because every capacitor word has zero coefficient on the trivial character. None affects nonempty K, but the statements should carry a K≠∅ hypothesis.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the point-line incidence system T^*_{n-1}(K) obtained from AG(n,q) by taking as lines all affine lines whose point at infinity lies in a subset K of the hyperplane at infinity. Its main result is a character-theoretic formula (Theorem 1.1) for the rank, over any field F with q nonzero, of the incidence matrix N, expressed as 1+(q−1)h_K, where h_K counts hyperplanes meeting K. From this it derives the dimension of the LDPC code C = ker N and, in Theorem 3.1, proves a conjecture of Vandendriessche that C is generated by plane words of weight 2q. The paper also studies the transpose code D = ker N^T, introduces capacitor words as generators, gives minimum-distance bounds and explicit examples, and applies the results to Wenger graphs and hyperovals.","tokens_in":11912,"tokens_out":14363,"duration_ms":130474,"significance":"If Theorem 3.1 is correct after the repairs noted below, the paper settles Vandendriessche's conjecture for all n and arbitrary K, a genuine advance in the theory of LDPC codes from finite geometries. The character-theoretic proof of Theorem 1.1 is elegant and self-contained, and it provides a uniform explanation of rank computations that were previously done case-by-case. The applications to Wenger graphs and hyperovals are concrete and useful. A clear strength is that the rank formula and the plane-word generation are derived from first principles rather than fitted to examples. The main limitation is the standing assumption q != 0 in F, which is explicitly acknowledged in Remark 1.2; the paper offers no result when q = 0 in F, so its scope is narrower than the most general coding-theoretic setting.","major_comments":[{"comment":"The displayed chain of equalities in (11) is incorrect as printed. The second equality removes the condition θ(u0)=0, which is essential; for instance, for n=2, q=3, K={u1} and u0=u2, the first term equals 2 while the middle term equals 6. The third equality is also wrong: q^{n-1} − rank_F η_K uses the original incidence map η_K, whereas the intended count is q^{n-1} − rank_F η_{\\bar K} for the quotient geometry T^*_{n-2}(\\bar K). In the same example, q^{n-1} − rank_F η_K equals 0, while the left-hand side equals 2. Because this equation is precisely the induction step that proves C'=C, the proof of Theorem 3.1 is not valid as written. The argument appears repairable by tracking the quotient and invoking the induction hypothesis there, but the correction is substantive and must be made explicit.","section":"§3.1, Eq. (11)"},{"comment":"The geometric formula 1+(q−1)h_K is false for K=∅, since h_∅=0 but rank_F N=0 rather than 1. This boundary case is used in the induction of Theorem 3.1, where K=∅ is the base case and (11) with the intended correction relies on rank_F η_∅=0. The authors should either state Theorem 1.1 for nonempty K and treat the empty case separately, or add a convention that makes the formula correct for K=∅.","section":"Theorem 1.1, statement"}],"minor_comments":[{"comment":"The proof should be expanded. The sentence 'Each such point t is also met by |K| lines, each of which contains a point u ... Therefore the sets ... both have cardinality at least |K|' does not by itself imply the lower bound on the positive support; a short double-counting argument between the positive and negative supports is needed to conclude that the positive support also has size at least |K|.","section":"Theorem 3.6"},{"comment":"The proof applies Lemma 2.2 to conclude that a character appearing with nonzero coefficient in a capacitor word lies in W. This requires W to be an FV-submodule of F^P; this is true because the set of capacitor words is translation-invariant, but the fact is not stated and should be mentioned.","section":"Theorem 3.3"},{"comment":"The name 'Gallagher' should be 'Gallager' to match reference [5].","section":"Introduction, first paragraph"},{"comment":"Once (11) is repaired, the image of K in the quotient should be denoted \\bar K throughout to avoid confusion with the original set K.","section":"§3.1, after Eq. (11)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript has a strong central idea, and the errors identified are localized and appear repairable. In my view the paper can be made acceptable with a careful revision, provided the authors correct equation (11), fix the empty-set boundary case in Theorem 1.1, and complete the proof of Theorem 3.6."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Sin, Sorci and Xiang prove something genuinely new: a character-theoretic rank formula for the incidence matrix of any linear representation T*_{n−1}(K) over fields of characteristic not dividing q, and then use it to prove Vandendriessche's conjecture that the code C is generated by plane words. Theorem 1.1 is a clean unification of scattered cases (Wenger graphs, hyperovals, affine geometries), and the proof via Fourier analysis on the additive group is self-contained and convincing. Theorem 3.1 settles the conjecture for all n. The capacitor-word generating set for D (Theorem 3.3) is a nice additional result, and the applications to Wenger graphs and hyperovals give correct dimensions.\n\nSo the main ideas are sound and the paper is worth a careful referee. But the written proof of Theorem 3.1 contains a genuine error at the pivotal step. Equation (11) is false as printed. The middle equality drops the condition θ(u0)=0, and the last term uses the original incidence map η_K when the argument requires the incidence map of the quotient geometry T*_{n−2}(\\bar K). This is not a missing symbol: the third equality is dimensionally wrong for the original η_K. A concrete check (n=2, q=3, K={u1}, u0=u2) gives 2 on the left, 6 for the printed middle term, and 0 for the printed final term. The repair is straightforward: replace the last η_K by the quotient map η_{\\bar K}, which is exactly what the projection argument earlier in the proof produces. With that fix the induction closes.\n\nTwo smaller boundary defects: Theorem 1.1's geometric form 1+(q−1)h_K gives 1 when K=∅ but rank is 0, and Theorem 3.3's capacitor words do not generate D_K for K=∅ because the trivial character never appears in a capacitor word. Neither affects any nonempty K, but the statements should carry a K≠∅ hypothesis.\n\nThe citation pattern is honest; the self-citations are to cases the paper generalizes. The character theory is standard but applied cleanly. In short: the theorem is very likely correct, the present text is not. The authors should be asked to fix equation (11) and the K=∅ caveats before publication. This is exactly the kind of paper a serious editor should send to a knowledgeable referee rather than desk-reject.","headline":"Genuinely new rank formula and proof of Vandendriessche's conjecture, but the printed proof of Theorem 3.1 has a repairable gap in equation (11); this deserves a serious referee and conditional acceptance.","tokens_in":12447,"tokens_out":7557,"would_cite":true,"duration_ms":69895,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05B25","51E20","94B05"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that over any field in which $q$ is nonzero, the rank of a finite-geometry incidence matrix is $1+(q-1)h_K$, where $h_K$ is the number of hyperplanes meeting $K$, and that the resulting LDPC code is generated by its…","keywords":["finite geometries","linear representations","LDPC codes","incidence matrix rank","plane words","capacitor words","minimum distance","character theory"],"falsifier":"Take $n=2$, $q=3$, and $K$ any two points of the four-point line at infinity; Theorem 1.1 predicts rank $5$ over any field of characteristic not $3$ ($1 + (3-1)\\cdot 2$). Computing the rank of the $9\\times 6$ point-line incidence matrix over, say, $\\mathbb{F}_2$ and getting any value other than $5$ would refute the theorem. To probe the excluded case, compute the same rank over $\\mathbb{F}_3$; a value strictly below $5$ would confirm that the $q=0$ restriction in Remark 1.2 is genuinely necessary.","tokens_in":11262,"feed_emoji":"📐","tokens_out":12097,"duration_ms":110785,"temperature":0.7,"pith_summary":"The paper proves a geometric formula for the rank of the point-line incidence matrix of a finite-geometry code: for a set $K$ of directions at infinity, over any coefficient field in which $q$ is nonzero, the rank equals $1 + (q-1)h_K$, where $h_K$ is the number of hyperplanes of the space at infinity that meet $K$. It uses this rank theorem to show that the low-density parity-check (LDPC) code whose parity-check matrix is the incidence matrix is generated by plane words, the words of minimum weight $2q$, settling a conjecture that had been proved only in the smallest dimension. The same character-theoretic machinery produces a generating set of capacitor words for the transposed code and lower bounds on its minimum distance, together with explicit rank formulas for special geometries such as the Wenger graph family and hyperovals. The interest is that purely geometric counts determine coding-theoretic parameters: dimension becomes a hyperplane intersection count, and minimum-weight codewords are visible as configurations of affine planes.","feed_headline":"Incidence-matrix rank becomes a hyperplane count","feed_subtitle":"The rank formula proves plane words generate the LDPC code, settling a long-standing conjecture.","key_machinery":"The load-bearing object is the incidence map $\\eta: F^L \\to F^P$ sending each line to the sum of the characteristic functions of its $q$ points, viewed as a homomorphism of permutation modules for the additive group $V$ acting regularly on the affine points. The key identity is that the image of $\\eta$ is spanned by the group characters $\\lambda_\\theta$ for which the corresponding $\\mathbb{F}_q$-linear functional $\\theta$ vanishes at some point of $K$; Lemma 2.2 shows that the nonroot characters of any translated line form a basis of the submodule it generates. The geometric translation is that nonzero $\\theta$ define hyperplanes of the space at infinity $H$, and $q-1$ scalar multiples define the same hyperplane, yielding the count $1 + (q-1)h_K$. Plane words—differences of the sums of the $q$ affine lines through two distinct points at infinity inside an affine plane—are then shown, by induction on $|K|$ using a projection argument, to generate all of $C$.","core_discovery":"Let $H = \\mathrm{PG}(V)$ be a hyperplane of $\\mathrm{PG}(E)$, let $K$ be a subset of $H$, and let $N$ be the point-line incidence matrix of the affine space $P = \\mathrm{PG}(E)\\setminus H$ with lines whose direction lies in $K$. The paper's central result, Theorem 1.1, states that over any field $F$ with $q \\neq 0$ the rank of $N$ is the number of linear functionals on $V$ that vanish on at least one point of $K$, equivalently $1 + (q-1)h_K$ with $h_K$ the number of hyperplanes of $H$ meeting $K$. The argument treats $F^P$ and $F^L$ as permutation modules for the additive group $V$ acting regularly on $P$; after adjoining a primitive $p$-th root of unity, the image of the incidence map is spanned by characters $\\lambda_\\theta$ whose associated functional $\\theta$ has a zero in $K$, so rank reduces to counting such $\\theta$. From this the paper derives Theorem 3.1: the code $C = \\ker N$ is spanned by plane words, each of weight $2q$, so $C$ is generated by its minimum-weight words. For the transposed code $D = \\ker N^T$ the paper shows capacitor words generate $D$ and that over totally ordered fields the minimum distance of $D$ is at least $2|K|$.","pith_inferences":["The same character-theoretic reduction may apply to incidence systems built from higher-dimensional affine subspaces rather than lines, with the 'vanishing at a point of $K$' condition replaced by the condition that a functional vanish on an entire subspace meeting $K$; if so, the rank formulas would become counts of hyperplanes meeting $K$ in prescribed dimensions.","The paper leaves the $q = 0$ in $F$ case open; a natural program is to compute modular ranks for small $n$ and $K$ in characteristic dividing $q$ and look for a formula involving p-adic or modular representation data rather than plain hyperplane counts.","Because $C$ is generated by plane words, one could design decoding or syndrome-based algorithms that operate on the local affine-plane structure of minimum-weight configurations, not just on generic sparse parity checks.","The capacitor-word construction suggests that the true minimum distance of $D$ is governed by how the set $K$ sits relative to hyperplane arrangements in $H$; testing whether the $2|K|$ bound is tight for arbitrary $K$ would refine the paper's small examples."],"forward_implications":["If $K$ contains a line of the hyperplane at infinity, $N$ has full rank $q^n$, so the corresponding LDPC code has the smallest possible dimension for its length.","For the Wenger graph family, the rank over any field with $q \\neq 0$ equals the number of polynomials over $\\mathbb{F}_q$ of degree at most $n-1$ that have a root in $\\mathbb{F}_q$, giving an explicit dimension formula for the associated codes.","For $n=3$ and $K$ a hyperoval, the rank over fields of characteristic not 2 is $1 + (q-1)\\binom{q+2}{2}$, and the formula is unchanged if one point is removed from $K$.","The minimum distance of the transposed code $D$ is at least $2|K|$ over any totally ordered field, with explicit codewords meeting this bound in small cases.","Whenever $C$ is nonzero, plane words have minimum weight $2q$ and generate $C$, so $C$ is generated by its minimum-weight codewords."],"supporting_citations":[{"why":"Conjectured the plane-word generation result and proved it for $n=3$; also supplies the theorem that plane words have minimum weight $2q$.","marker":"[12]"},{"why":"Computes dimensions of the LU(3,q) codes over fields with $q \\neq 0$, the special case this paper recovers and extends.","marker":"[13]"},{"why":"Introduced the type-I and type-II geometry-based LDPC codes for $K=H$ over $\\mathbb{F}_2$ and gave the basic minimum-distance lower bounds.","marker":"[14]"},{"why":"Studied these codes for arbitrary $K$ over $\\mathbb{F}_2$ and supplied the $|K|+1$ and $q+1$ lower bounds on minimum distance.","marker":"[11]"},{"why":"Determined the spectrum of the Wenger graphs, giving the zero-eigenvalue multiplicity used to compare with the rank formula.","marker":"[3]"},{"why":"Proved the dimension formula for the binary LU(3,q) codes, a precursor of the general rank theorem.","marker":"[10]"},{"why":"Analyzed small-weight codewords in these LDPC codes, groundwork for identifying plane words as minimum-weight words.","marker":"[9]"}],"fun_headline_variants":["Hyperplane count gives rank of affine incidence","Plane words generate LDPC code, settling conjecture","Incidence rank tied to hyperplane intersections","Geometry yields LDPC code generation proof","Rank formula proves plane words span the code"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire argument depends on $q$ being nonzero in the coefficient field $F$; when $q = 0$ in $F$ the rank formula is not claimed to hold and the paper gives no conjecture for that case.","fun_headline_variants_meta":{"raw":{"variants":["Hyperplane count gives rank of affine incidence","Plane words generate LDPC code, settling conjecture","Incidence rank tied to hyperplane intersections","Geometry yields LDPC code generation proof","Rank formula proves plane words span the code"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000218,"raw_usage":{"total_tokens":1447,"prompt_tokens":960,"completion_tokens":487,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":576,"completion_tokens_details":{"reasoning_tokens":418}},"tokens_in":576,"tokens_out":487,"duration_ms":5113,"temperature":1.0,"reasoning_tokens":418,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:35:36.547259+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take $n=2$, $q=3$, and $K$ any two points of the four-point line at infinity; Theorem 1.1 predicts rank $5$ over any field of characteristic not $3$ ($1 + (3-1)\\cdot 2$). Computing the rank of the $9\\times 6$ point-line incidence matrix over, say, $\\mathbb{F}_2$ and getting any value other than $5$ would refute the theorem. To probe the excluded case, compute the same rank over $\\mathbb{F}_3$; a value strictly below $5$ would confirm that the $q=0$ restriction in Remark 1.2 is genuinely necessary.","supporting_citations":[{"cited_title":"Vandendriessche, LDPC codes associated with linear representations of geome tries, Advances in Mathematics of Communications 4 (3) (2010), 405–417","cited_arxiv_id":null,"evidence_quote":"Conjectured the plane-word generation result and proved it for $n=3$; also supplies the theorem that plane words have minimum weight $2q$."},{"cited_title":"Vandendriessche, Some low-density parity-check codes derived from finite geo metries, Designs, Codes and Cryptography 54 (3) (2010), 287–297","cited_arxiv_id":null,"evidence_quote":"Computes dimensions of the LU(3,q) codes over fields with $q \\neq 0$, the special case this paper recovers and extends."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduced the type-I and type-II geometry-based LDPC codes for $K=H$ over $\\mathbb{F}_2$ and gave the basic minimum-distance lower bounds."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Studied these codes for arbitrary $K$ over $\\mathbb{F}_2$ and supplied the $|K|+1$ and $q+1$ lower bounds on minimum distance."},{"cited_title":"Cioaba, F","cited_arxiv_id":null,"evidence_quote":"Determined the spectrum of the Wenger graphs, giving the zero-eigenvalue multiplicity used to compare with the rank formula."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Proved the dimension formula for the binary LU(3,q) codes, a precursor of the general rank theorem."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Analyzed small-weight codewords in these LDPC codes, groundwork for identifying plane words as minimum-weight words."}],"review_version":1}