ResearchCombinatorial designsGSI-R-0002 · Paper · Existence
We construct skew-Hadamard matrices of order 4v for 106 odd v < 2500, and symmetric ones for 85, each listed as undecided in the 2025 census.
A Hadamard matrix of order n is a ±1 matrix H with HHT = nI, skew if H + HT = 2I, symmetric if H = HT. The census of Đoković and Seberry [1] lists 737 odd v < 2500 as undecided for skew and 595 for symmetric matrices of order 4v. We give skew matrices for 106 of the 737 and symmetric ones for 85 of the 595, by four theorems, an extension of Đoković's construction, and computer search.
- Before this work
- The census of Đoković and Seberry [1] (May 2025) lists 737 odd v < 2500 as undecided (‘open’) for skew and 595 for symmetric matrices of order 4v. Later, Đoković [2] settled 3 symmetric and Schumer [3] 1 skew value; no other update is counted.
- Evidence
- Independently checked. The repository verifier rebuilds matrices from certificates and tests them exactly. Our separate program check_h.c (Files) re-tested 77 stored matrices, and skewness or symmetry of 58 skew and 6 symmetric ones, all passing. For the other 64 skew and 97 symmetric orders the verifier alone tests that. No outside review has taken place.
- What you can check
- One half: the matrices. The repository verifier tests each in a few minutes for orders below 12000; the README gives no time for the 3 above (13072, 14352, 14384). Census status and novelty can be checked only against the sources.
- Not established
- That the 45 skew and 36 symmetric values classed as new (see The result) have no earlier construction, published or posted: the literature search did not cover MathSciNet or zbMATH.
- Method
- Public: certificates, search programs and proofs of Theorems A to D in the repository (files below); the manuscript, submitted to arXiv on 7 November 2026, is not yet announced.
What was known
Theorems A and B below are symmetric versions of Yamada's Theorems 3 and 4 [4], the second due to Kiyasu [5], with the same hypotheses. The matrices of Seberry and Yamada [6] in the propus array of Seberry and Balonin [7], introduced by Balonin, give symmetric Hadamard matrices of order 4(q + 2) for prime powers q ≡ 5 (mod 8) with (q + 1)/2 a prime power, and of order 4q for prime powers q ≡ 1 (mod 8) from a symmetric Hadamard matrix of order (q − 1)/2, a case we found stated in neither source. The propus array turns commuting symmetric ±1 matrices A, B, D of order v with A2 + 2B2 + D2 = 4vI into a symmetric Hadamard matrix of order 4v. Miyamoto [8] asserted, for a prime power q ≡ 1 (mod 4), a Hadamard matrix of order 4q with two symmetric blocks, from symmetric Hadamard and conference matrices of orders q − 1 and q + 1; Đoković and Seberry note that his proof does not establish the symmetry [1]. To our knowledge no general skew construction of order 3(q + 1) appears in the census, Cati and Pasechnik [9], Seberry and Yamada [10] or Koukouvinos and Stylianou [11].
We claim no priority for 18 symmetric values: 7 follow from Spence's Corollary 3 [12], 2 from Corollary 1 of [7] corrected by Balonin and Đoković [13], 1 (235) from that corollary with a symmetric conference matrix of order 118 due to J. Y. Z. Ho (personal communication, cited in Brouwer's table [14]), and 8 are values for which Zaitsev [15] had posted quaternary Hadamard matrices of order 2v, not shown equivalent to ours.
The result
A value is ‘new’ if, to our knowledge, no earlier construction, published or posted, and no combination of published results gives a matrix of order 4v with the property; ‘from published results’ if published ingredients give it with at most a short verification of ours; ‘recorded as known’ if Table 5 of [9], which by its convention records an order as known without a construction, has no entry for it.
| Construction | Values | New | From published results | Recorded as known |
|---|---|---|---|---|
| Symmetric, Theorem A | 58 | 11 | 47 | 0 |
| Symmetric, Theorem B | 18 | 18 | 0 | 0 |
| Symmetric, Theorem C | 3 | 3 | 0 | 0 |
| Symmetric, m(m − 1) and m(m + 3), after Đoković [2] | 2 | 0 | 2 | 0 |
| Symmetric, propus arrays | 4 | 4 | 0 | 0 |
| Symmetric, total | 85 | 36 | 49 | 0 |
| Skew, Theorem D | 59 | 0 | 57 | 2 |
| Skew, difference families | 42 | 41 | 0 | 1 |
| Skew, orthogonal designs (Seberry [16]) | 5 | 4 | 1 | 0 |
| Skew, total | 106 | 45 | 58 | 3 |
Let q ≡ 1 (mod 8) be a prime power. If there is a Hadamard matrix of order (q − 1)/2, then there is a symmetric Hadamard matrix of order 4q.
Of the 11 new values, 10 use an ordinary Hadamard matrix (known at all orders 4k ≤ 2000 [17], [18]), among them the matrix of order 716 announced by Alpöge, Voinov and Reynolds-Haertle on 12 August 2026 [17] for q = 1433; q = 1753 uses the symmetric matrix of order 876 found here.
Let q ≡ 5 (mod 8) be a prime power. If there is a skew-Hadamard matrix of order (q + 3)/2, then there is a symmetric Hadamard matrix of order 4(q + 2).
Yamada's circulants [4] and the skew-Hadamard input assemble a symmetric quaternary Hadamard matrix L of order 2(q + 2) over {±1, ±i}, which Turyn's map [19] turns into a symmetric real one of order 4(q + 2). Theorem B applies to 28 open v = q + 2, for none of which (q + 1)/2 is a prime power; 10 carry no priority claim (8 for Zaitsev [15], 703 and 1183 for Spence [12]), leaving 18 new. It also applies to 183, listed under propus arrays.
Let q ≡ 1 (mod 4) be a prime power. If there is a Hadamard matrix of order q − 1 with block rows (P Q) and (Q P), in particular if there is a periodic Golay pair of length (q − 1)/2 over an abelian group, then there is a symmetric Hadamard matrix of order 4q.
It settles 101, 149 and 389 with periodic Golay pairs (±1 sequences on an abelian group whose periodic autocorrelations sum to zero at every non-zero shift) of lengths 50, 74 and 194. Pairs of length 50 were known (Đoković [20]; Kotsireas and Koukouvinos [21]); Đoković and Kotsireas [22] found the pair of length 74. Ordinary Hadamard matrices of orders 404, 596 and 1556 were known [8]; symmetric ones of order q − 1 also exist, so the values lie within Miyamoto's assertion, which Theorem C does not use.
For every prime power q ≡ 3 (mod 4) there is a skew-Hadamard matrix of order 3(q + 1).
Paley's skew matrix of order q + 1 is, up to simultaneous row and column permutation and a ±1 diagonal similarity, a 4 × 4 array of negacyclic blocks Nkj with N30 = N10T; Yamada [23] gave this for q ≡ 3 (mod 8), and we prove it for q ≡ 7 (mod 8). Seberry, Yamada and Turyn [6] form Aj = ΣTk ⊗ Nkj from T-matrices Tk of order t (Cooper and Wallis [24]). For T = (I, Z, 0, −Z2) of order 3, Z the cyclic shift, A0 + A0T = 2I, so the Goethals–Seidel array of the Aj is skew-Hadamard; that is the new observation. All 59 open v = 3(q + 1)/4, q ≡ 3 (mod 8), are settled; 663 and 879 are recorded as known. Ordinary Hadamard matrices of these orders were known.
The smallest open symmetric value, v = 65, comes from Verma's quaternary Legendre pair (A, B) of length 64 [25], in which A is real and B symmetric, a symmetry Verma does not use. A bordered propus array, the analogue of the bordered construction of Kotsireas and Winterhof [26], gives a symmetric Hadamard matrix of order 260.
For 12 odd v < 1000 (191, 225, 277, 299, 543, 579, 589, 611, 817, 897, 899, 909) we construct skew-Hadamard matrices of order 2tv, t = 3 or 4; Table 5 of [9] records the least t known as 226 for 191 and 4 to 6 for the other 11. They remain open.
How to check it
git clone https://github.com/vatskishan03/Hadamard
cd Hadamard
git checkout d885022144ba6bf4efa8315e14871a3736947527
# 260; Theorems A, B, C, D (3236, 4924, 404, 996); a search (1004)
python3 verify/verify.py symmetric/260 symmetric/3236 symmetric/4924 symmetric/404 skew/996 skew/1004
cc -O2 verify/check_matrix.c -o check_matrix
gzip -dc symmetric/260/H260.csv.gz | ./check_matrixsymmetric/260 ok H H^T = 260 I sha256 948586d6b68bc9a3c52ce2259903ad13f46ec54a800ee5cb8511b5a69d83eabe (symmetric: H = H^T)A pass shows a Hadamard matrix of the stated order and symmetry. An earlier construction at a value called new refutes only the label.
What it changes
Of the 737 open skew values, 107 are settled (919 by Schumer [3], the 106 in the table), leaving 630, and of the 595 open symmetric values, 106 are settled (189, 473, 1139 by Đoković [2], the 85 in the table, the 18 without a priority claim), leaving 489.
Limits
- Not established: that the 45 skew and 36 symmetric values classed as new have no earlier construction. The searches, 12 October to 6 November 2026 (575, 1657 and 1801 only 3 to 6 November, without broader web searches), omitted MathSciNet and zbMATH and read Miyamoto's paper [8] only through its abstract and a zbMATH review.
- For the symmetric values 133, 183 and 857, classed as new, Zaitsev [15] lists quaternary Hadamard matrices of orders 266, 366 and 1714, with no non-equivalence claim in the manuscript; a nauty check (programs not in the repository) found none of the three monomially equivalent to its transpose, hence none to a symmetric matrix.
- Theorem B needs a skew-Hadamard matrix of order (v + 1)/2, and we know of none for the 6 open v = q + 2 (1711, 1879, 2215, 2359, 2383, 2391).
- For the 57 open v = q with q ≡ 5 (mod 8), Theorem C needs (q − 1)/4 to be a sum of two squares, which fails for 44; for 10 of the other 13 (q = 773, 821, 1109, 1493, 1621, 1733, 1973, 2197, 2309, 2341) we know of no input.
How it was found
K. Kumar takes responsibility for the mathematical arguments, computational results, attribution and references of the manuscript.
Files
- Repository at the cited commitmatrices, certificates, verifier and search programs, version of 6 November 2026
- check_h.cour second checker, not in the repository; its output on the 77 stored matrices is check_h.out beside it
- README.mdthe repository's own counts, by order and with its own status labels, and the verification commands
- skew/README.mdthe skew orders, with constructions and status labels
- symmetric/README.mdthe symmetric orders, with constructions and status labels
- skew/PALEY_SKEW.mdTheorem D: construction and proof
- symmetric/PROPUS_WEIGHING.mdTheorem A: proof, and the ten values with no symmetric input
- symmetric/YAMADA_KIYASU.mdTheorem B: proof
- symmetric/PERIODIC_GOLAY.mdTheorem C: proof, and the periodic Golay pairs
- verify/verify.pyrebuilds every matrix from its certificate and tests it
- NOVELTY.mdstatus labels, sources searched, sources not read
References
- [1]D. Ž. Đoković, J. Seberry, Census of symmetric and skew Hadamard matrices of order 4v for odd v < 2500, manuscript, 8th Workshop on Design Theory, Hadamard Matrices and Applications (HADAMARD 2025), Seville, May 2025.
- [2]D. Ž. Đoković, Orthogonal designs and symmetric Hadamard matrices, arXiv:2508.17141 (23 August 2025). Link
- [3]A. Schumer, Periodic T-matrices of order 151 and a skew spin Goethals–Seidel family over Z/919Z: certificates and Hadamard constructions of orders 3676, 5436, 6644, and 7852, Zenodo, 17 September 2026. Link
- [4]M. Yamada, Some new series of Hadamard matrices, J. Austral. Math. Soc. Ser. A 46 (1989) 371–383.
- [5]Z. Kiyasu, Hadamard Matrices and their Applications (in Japanese), Denshi Tsushin Gakkai, Tokyo, 1980.
- [6]J. Seberry, M. Yamada, On the products of Hadamard matrices, Williamson matrices and other orthogonal matrices using M-structures, J. Combin. Math. Combin. Comput. 7 (1990) 97–137.
- [7]J. Seberry, N. A. Balonin, Two infinite families of symmetric Hadamard matrices, Australas. J. Combin. 69 (2017) 349–357.
- [8]M. Miyamoto, A construction of Hadamard matrices, J. Combin. Theory Ser. A 57 (1991) 86–108.
- [9]M. Cati, D. V. Pasechnik, A database of constructions of Hadamard matrices, arXiv:2411.18897 (v2, 30 August 2025). Link
- [10]J. Seberry, M. Yamada, Hadamard matrices, sequences, and block designs, in: J. H. Dinitz, D. R. Stinson (eds.), Contemporary Design Theory: A Collection of Surveys, Wiley, New York, 1992, 431–560.
- [11]C. Koukouvinos, S. Stylianou, On skew-Hadamard matrices, Discrete Math. 308 (2008) 2723–2731.
- [12]E. Spence, An infinite family of Williamson matrices, J. Austral. Math. Soc. Ser. A 24 (1977) 252–256.
- [13]N. A. Balonin, D. Ž. Đoković, Symmetric Hadamard matrices of orders 268, 412, 436 and 604, Inform. Upravl. Sist. 2018, no. 4, 2–8; arXiv:1803.08787. Link
- [14]A. E. Brouwer, Parameters of strongly regular graphs, online tables, accessed 6 November 2026. Link
- [15]V. Zaitsev, Quaternary Hadamard matrices, web page, University of Lethbridge, accessed 6 November 2026. Link
- [16]J. Seberry, On skew Hadamard matrices, Ars Combin. 6 (1978) 255–275.
- [17]L. Alpöge, P. Voinov, S. Reynolds-Haertle, Hadamard matrices of orders 668, 716, 892, 1132, 1244, 1388, 1436, 1676, 1772, 1916, 1948 and 1964, announcement, 12 August 2026; verified and analysed in [18]. Link
- [18]R. Egan, I. S. Kotsireas, S. London, P. Ó Catháin, Recent progress in the construction of Hadamard matrices, arXiv:2610.27175 (v2, 3 November 2026). Link
- [19]R. J. Turyn, Complex Hadamard matrices, in: Combinatorial Structures and their Applications, Gordon and Breach, New York, 1970, 435–437.
- [20]D. Ž. Đoković, Cyclic (v; r, s; λ) difference families with two base blocks and v ≤ 50, Ann. Comb. 15 (2011) 233–254.
- [21]I. S. Kotsireas, C. Koukouvinos, Periodic complementary binary sequences of length 50, Int. J. Appl. Math. 21 (2008) 509–514.
- [22]D. Ž. Đoković, I. S. Kotsireas, Some new periodic Golay pairs, Numer. Algorithms 69 (2015) 523–530. Link
- [23]M. Yamada, Hadamard matrices of generalized quaternion type, Discrete Math. 87 (1991) 187–196.
- [24]J. Cooper, J. Wallis, A construction for Hadamard arrays, Bull. Austral. Math. Soc. 7 (1972) 269–277.
- [25]H. Verma, A quaternary Legendre pair of length 64, arXiv:2609.25573 (v2, 30 October 2026). Link
- [26]I. S. Kotsireas, A. Winterhof, Quaternary Legendre pairs, in: C. J. Colbourn, J. H. Dinitz (eds.), New Advances in Designs, Codes and Cryptography, Fields Inst. Commun. 86, Springer, Cham, 2024, 289–304. Link