betaFeedback

Reading v1View latest version

WorksMathematics

Recent progress in the construction of Hadamard matrices

Abstract

A Hadamard matrix is a square matrix with entries ±1\pm 1 whose rows are mutually orthogonal. Hadamard matrices are conjectured to exist at every order divisible by 44; for approximately the last twenty years, the smallest open case of this conjecture was order 668. In August 2026, Alpöge, Voinov and Reynolds-Haertle announced Hadamard matrices at the twelve orders below 20002000 at which none was known. In this paper we verify and analyse the claimed matrices. We survey results on the existence of Hadamard matrices and list the 9595 orders up to 10 00010\,000 at which no Hadamard matrix is currently known. In pursuing this work we constructed Hadamard matrices at sixteen orders that were open in the published literature. We conclude with a list of open problems.

To Jennifer R. Seberry and Dragomir Ž. Ðoković, pioneers in the existence of Hadamard matrices.

Introduction

Let MM be an n×nn \times n complex matrix with entries of modulus at most 1. Hadamard proved that ∣det⁡M∣≤nn/2\lvert\det M\rvert\le n^{n/2}, with equality if and only if every entry has modulus 1 and MM∗=nIMM^{*}=nI [52]. Over C\mathbb{C} the bound is attained at every order by the character table of Zn\mathbb{Z}_n. Over R\mathbb{R} the extremal matrices are Hadamard matrices, which exist only at orders 1, 2 and 4t4t. The first infinite family is due to Sylvester [96]. Paley described further infinite families and conjectured that a Hadamard matrix exists at every order 4t4t [81].

The conjecture is settled at orders below 92 by the constructions of Sylvester and Paley together with Kronecker products. Order 92 was resolved computationally by Baumert, Golomb and Hall [5], and order 116 by Baumert in 1966 [6]. By 1977 the smallest open order was 268 [54, 89]; it was settled by Sawade in 1985 [83], leaving 428 as the smallest unsettled order. From the construction of order 428 by Kharaghani and Tayfeh-Rezaie [59], the smallest open order was 668. On 12 August 2026 Alpöge, Voinov and Reynolds-Haertle announced Hadamard matrices at the twelve orders

668, 716, 892, 1132, 1244, 1388, 1436, 1676, 1772, 1916, 1948, 1964,

which were the only open orders below 2000 [3]. The announcement gives the matrices but not the method of construction.

Verifying the twelve matrices is a matrix multiplication; their structure is less apparent, and we describe it in detail. Each is a Goethals–Seidel array over an abelian group. Nine are over cyclic groups, four of them with a border of width 4, and three are over non-cyclic groups, with a wider border that is constant on the cosets of a subgroup.

A Hadamard matrix is therefore known at every order below 2000 divisible by 4. We construct Hadamard matrices at sixteen further orders below 10 000, record the orders settled by others while this paper was in preparation, and list the 95 orders n≤10 000n \le10\,000 at which no Hadamard matrix is currently known; the smallest is 2396=4⋅5992396 = 4 \cdot599. As a survey of constructions, the paper is written for readers new to the subject; for a broader introduction we refer to the monographs [56, 27, 92, 93].

Section 2 gives definitions and classical constructions of Hadamard matrices. Section 3 describes the matrices at the twelve orders listed above. Section 4 surveys the known constructions, describes the new matrices, ours and those of others, and tabulates the open orders up to 10 000. Section 5 lists open problems. Every Hadamard matrix constructed in this paper is publicly available at https://www.math74.com/Hadamard/ and was verified by an exact computation of HHTHH^{\mathsf{T}}.

Hadamard matrices: a rapid overview

The identity and all-ones matrices of order nn are InI_n and JnJ_n; the all-ones row vector is jj and RR is the matrix with ones on the back-diagonal, with sizes clear from context; in displayed arrays, ++ and −- stand for 11 and −1-1.

Theorem 1 (Hadamard [52]). Let M=(mij)M=(m_{ij}) be an n×nn \times n real matrix with ∣mij∣≤1\lvert m_{ij}\rvert\le1. Then ∣det⁡M∣≤nn/2\lvert\det M\rvert\le n^{n/2}, with equality if and only if mij∈{±1}m_{ij} \in\{\pm1\} for all i,ji,j and MMT=nInMM^{\mathsf{T}}=nI_n.

The bound follows directly from the inequality ∣det⁡M∣≤∏i∥ri∥\lvert\det M\rvert\le\prod_i \lVert r_i\rVert, where r1,…,rnr_1,\ldots,r_n are the rows of MM, together with ∥ri∥≤n\lVert r_i\rVert\le\sqrt{n}. The inequality was conjectured by Kelvin and proved by Hadamard; see [14] for its history. Hadamard drew attention to real matrices meeting the bound, and constructed examples at orders 12 and 20.

Definition 1. A Hadamard matrix of order nn is an n×nn \times n matrix HH with entries in {±1}\{\pm1\} satisfying HHT=nInHH^{\mathsf{T}}=nI_n. We write H(n)H(n) for a Hadamard matrix of order nn. More generally, a Butson Hadamard matrix BH⁡(n,k)\operatorname{BH}(n,k) is an n×nn \times n matrix HH with entries in the kkth roots of unity and HH∗=nInHH^*=nI_n [16], so that a BH⁡(n,2)\operatorname{BH}(n,2) is a Hadamard matrix.

Permuting and negating rows or columns preserves the defining equation. Two Hadamard matrices related by such operations are equivalent, and every Hadamard matrix is equivalent to one whose first row and column consist of 1s. Comparing the first three rows of such a normalised matrix shows that n∈{1,2}n \in\{1,2\} or 4∣n4 \mid n.

Sylvester observed that H(a)⊗H(b)H(a) \otimes H(b) is a Hadamard matrix of order abab, so that

(+++−)⊗k\left(\begin{array}{cc} + & + \\ + & - \end{array}\right)^{\otimes k}

is an H(2k)H(2^k) [96].

The next substantial contribution to the existence of Hadamard matrices after Hadamard is due to Paley [81]. For an odd prime power qq, let χ\chi be the quadratic character of Fq\mathbb{F}_q with the convention that χ(0)=0\chi(0)=0, and Q=(χ(y−x))x,y∈FqQ=(\chi(y-x))_{x,y\in\mathbb{F}_q} the Jacobsthal matrix. The character sum ∑z∈Fqχ(z)χ(z+s)=−1\sum_{z\in\mathbb{F}_q}\chi(z)\chi(z+s)=-1 for s≠0s\ne0 gives QQT=qI−JQQ^{\mathsf{T}}=qI-J and QJ=JQ=0QJ=JQ=0, and QT=χ(−1)QQ^{\mathsf{T}}=\chi(-1)Q. A Hadamard matrix HH is skew if H+HT=2IH+H^{\mathsf{T}}=2I, and a conference matrix of order nn is a matrix CC with zero diagonal, entries ±1\pm1 off the diagonal, and CCT=(n−1)InCC^{\mathsf{T}}=(n-1)I_n.

Theorem 2 (Paley [81]). Let

S=(0jχ(−1)jTQ).S=\begin{pmatrix} 0 & j\\ \chi(-1)j^{\mathsf{T}} & Q \end{pmatrix}.

If q≡3(mod4)q \equiv3 \pmod4, then SS is skew-symmetric and I+SI+S is a skew Hadamard matrix of order q+1q+1. If q≡1(mod4)q \equiv1 \pmod4, then SS is a symmetric conference matrix and

S⊗(+++−)+Iq+1⊗(+−−−)S\otimes\begin{pmatrix}+&+\\+&-\end{pmatrix} +I_{q+1}\otimes\begin{pmatrix}+&-\\-&-\end{pmatrix}

is a Hadamard matrix of order 2(q+1)2(q+1).

The eigenvalues of QQ are given by Gauss sums, and the consideration of Gauss and Jacobi sums gives rise to a large number of cyclotomic constructions of Hadamard matrices. For the theory of cyclotomic numbers on which these depend see Storer or Hall [95, 53]; for constructions from cyclotomy and character sums see Whiteman [109], Spence [94] and Yamada [116]; and for the related supplementary difference sets the surveys of Seberry and Yamada [92] and of Wallis, Street and Wallis [107]. Cocyclic and group-developed constructions are treated in [27, 56].

Plug-in constructions

Another approach builds H(4m)H(4m) from four m×mm \times m blocks placed in a fixed array. Williamson [112] took AA, BB, CC, DD to be symmetric circulant {±1}\{\pm1\}-matrices of order mm with AAT+BBT+CCT+DDT=4mImAA^{\mathsf T}+BB^{\mathsf T}+CC^{\mathsf T}+DD^{\mathsf T}=4mI_m. Since circulant matrices commute, the array

(ABCD−BA−DC−CDA−B−D−CBA)\begin{pmatrix} A & B & C & D \\ -B & A & -D & C \\ -C & D & A & -B \\ -D & -C & B & A \end{pmatrix}

is then a Hadamard matrix of order 4m4m. The Hadamard matrix of order 92 was obtained in this way, with m=23m=23 [5]. Turyn’s family supplies Williamson matrices of order (q+1)/2(q+1)/2 for every prime power q≡1(mod4)q \equiv1 \pmod{4} [103]. Williamson matrices do not exist at every odd order, however. Đoković showed that there are none of order 35 [32], and Holzmann, Kharaghani and Tayfeh-Rezaie found none of orders 47, 53 or 59 [55].

Goethals and Seidel [49] removed the symmetry requirement of the Williamson array by varying the array into which the circulant matrices are inserted.

Theorem 3 (Goethals–Seidel [49]). Let AA, BB, CC, DD be circulant {±1}\{\pm1\}-matrices of order mm with AAT+BBT+CCT+DDT=4mImAA^{\mathsf T}+BB^{\mathsf T}+CC^{\mathsf T}+DD^{\mathsf T}=4mI_m. Then

(ABRCRDR−BRADTR−CTR−CR−DTRABTR−DRCTR−BTRA)\begin{pmatrix} A & B\mathrm R & C\mathrm R & D\mathrm R \\ -B\mathrm R & A & D^{\mathsf T}\mathrm R & -C^{\mathsf T}\mathrm R \\ -C\mathrm R & -D^{\mathsf T}\mathrm R & A & B^{\mathsf T}\mathrm R \\ -D\mathrm R & C^{\mathsf T}\mathrm R & -B^{\mathsf T}\mathrm R & A \end{pmatrix}

is a Hadamard matrix of order 4m4m.

In 2026, Williamson, Yacobi and Zinn-Justin generated Goethals–Seidel Hadamard matrices of orders up to 252 by local search, guided by a transformer trained on the output of that search [111].

An alternative approach is to allow a constant negative value off the diagonal in the sum of Gram matrices, which is then absorbed into a border of ±1\pm1 entries. The basic example uses two circulants.

Theorem 4 ([47]). Let AA and BB be circulant {±1}\{\pm1\}-matrices of order ℓ\ell satisfying

AAT+BBT=(2ℓ+2)Iℓ−2Jℓ,(1)AA^{\mathsf T}+BB^{\mathsf T}=(2\ell+2)I_\ell-2J_\ell, \tag*{(1)}

normalised so that every row of AA and of BB sums to 1. Then

\begin{pmatrix} - & - & \vline& j & j \\ - & + & \vline& j & -j \\ \hline j^{\mathsf T} & j^{\mathsf T} & \vline& A & B \\ j^{\mathsf T} & -j^{\mathsf T} & \vline& B^{\mathsf T} & -A^{\mathsf T} \end{pmatrix}

is a Hadamard matrix of order 2ℓ+22\ell+2.

Condition (1) forces the row sums of AA and BB to be ±1\pm1, and so ℓ\ell is odd. For a {±1}\{\pm1\}-sequence a=(a0,…,am−1)a=(a_0,\ldots,a_{m-1}), the periodic autocorrelation Pa(s)=∑i=0m−1aiai+sP_a(s)=\sum_{i=0}^{m-1}a_i a_{i+s}, with indices mod mm, is the inner product of rows ii and i+si+s of the circulant with first row aa. The first rows of AA and BB form a Legendre pair: two {±1}\{\pm1\}-sequences whose periodic autocorrelations sum to −2-2 at every non-zero shift. The name comes from the Legendre sequence of a prime p≡3(mod4)p\equiv3\pmod{4}, whose periodic autocorrelation is −1-1 at every non-zero shift, so that two copies of it form a Legendre pair. Legendre pairs are known at infinitely many lengths and it is conjectured that one exists at every odd length [38], which would give a Hadamard matrix at every order 2ℓ+2≡0(mod4)2\ell+2\equiv0\pmod{4}.

We conclude this section with a construction combining the four circulants of Theorem 3 with the border and constant negative autocorrelation of Theorem 4.

Theorem 5. Let AA, BB, CC, DD be circulant {±1}\{\pm1\}-matrices of even order mm with

AAT+BBT+CCT+DDT=(4m+4)Im−4Jm,(2)AA^{\mathsf{T}}+BB^{\mathsf{T}}+CC^{\mathsf{T}}+DD^{\mathsf{T}}=(4m+4)I_m-4J_m, \tag*{(2)}

normalised so that AA has row sum 2 and BB, CC, DD have row sum 0. Then

(+−−−jjjj−+−−j−jj−j−−+−j−j−jj−−−+jj−j−jjTjTjTjTABRCRDR−jTjTjT−jT−BRADTR−CTR−jT−jTjTjT−CR−DTRABTR−jTjT−jTjT−DRCTR−BTRA)\begin{pmatrix} + & - & - & - & j & j & j & j \\ - & + & - & - & j & -j & j & -j \\ - & - & + & - & j & -j & -j & j \\ - & - & - & + & j & j & -j & -j \\ j^{\mathsf{T}} & j^{\mathsf{T}} & j^{\mathsf{T}} & j^{\mathsf{T}} & A & BR & CR & DR \\ -j^{\mathsf{T}} & j^{\mathsf{T}} & j^{\mathsf{T}} & -j^{\mathsf{T}} & -BR & A & D^{\mathsf{T}}R & -C^{\mathsf{T}}R \\ -j^{\mathsf{T}} & -j^{\mathsf{T}} & j^{\mathsf{T}} & j^{\mathsf{T}} & -CR & -D^{\mathsf{T}}R & A & B^{\mathsf{T}}R \\ -j^{\mathsf{T}} & j^{\mathsf{T}} & -j^{\mathsf{T}} & j^{\mathsf{T}} & -DR & C^{\mathsf{T}}R & -B^{\mathsf{T}}R & A \end{pmatrix}

is a Hadamard matrix of order 4m+44m+4.

Multiplying (2) by JmJ_m shows that the squares of the four row sums add to 4. For even mm the row sums are even, so one of them is ±2\pm2 and the others are 0. Negating or permuting AA, BB, CC, DD preserves (2). By analogy with Legendre pairs, the term Legendre quadruples seems appropriate for four {±1}\{\pm1\}-sequences of the same length whose periodic autocorrelations sum to −4-4 at every non-zero shift. The first rows of AA, BB, CC, DD in Theorem 5 form a Legendre quadruple of even length mm.

The twelve orders settled in 2026

Each of the twelve orders is of the form 4p4p with pp a prime, p≡3(mod4)p\equiv3\pmod{4}: the primes are 167, 179, 223, 283, 311, 347, 359, 419, 443, 479, 487, 491. The announcement encodes the twelve matrices as a single string of signs, together with a short sed program that assembles them, and does not describe how they were found [3]. Each assembled matrix is a Goethals–Seidel array of order 4v4v in four generators X,Y,Z,WX,Y,Z,W over an abelian group GG of order vv, with a border of width bb, so that n=4v+bn=4v+b (Table 1).

nnbbGGvv∑P\sum P∣U∣|U|rr
8920Z223\mathbb{Z}_{223}2230375
11320Z283\mathbb{Z}_{283}2830395
12440Z311\mathbb{Z}_{311}3110563
19480Z487\mathbb{Z}_{487}4870955
19640Z491\mathbb{Z}_{491}4910771
6684Z166\mathbb{Z}_{166}166−4-41166
7164Z178\mathbb{Z}_{178}178−4-41178
16764Z418\mathbb{Z}_{418}418−4-41542
17724Z442\mathbb{Z}_{442}442−4-41250
191612Z2×Z119\mathbb{Z}_2 \times\mathbb{Z}_{119}476−12,+4-12,+412/4a12/4^{a}46/128a46/128^{a}
138820Z2×Z3×Z57\mathbb{Z}_2 \times\mathbb{Z}_3 \times\mathbb{Z}_{57}342−20,+4-20,+4670
143628Z2×Z8×Z11\mathbb{Z}_2 \times\mathbb{Z}_8 \times\mathbb{Z}_{11}352−28,+4-28,+4596

Table 1. The twelve matrices; n=4v+bn=4v+b is the matrix order, bb the border width, GG is a group of order vv, the column ∑P\sum P gives the values of (3) on S∖{0}S\setminus\{0\} and, when S≠GS\ne G, off SS, UU is the multiplier group (Definition 2) and rr the number of its orbits on GG.

Over a non-cyclic group, circulants are replaced by group-developed matrices [f]x,y=f(y−x)[f]_{x,y}=f(y-x), for f:G→{±1}f:G\to\{\pm1\}, and RR by the matrix of the map x↦−xx\mapsto-x; Theorems 3 and 5 then hold as stated. The periodic autocorrelation Pf(d)=∑x∈Gf(x)f(x+d)P_f(d)=\sum_{x\in G}f(x)f(x+d) is the inner product of rows xx and x+dx+d of [f][f]. In all twelve matrices

∑fPf(d)={−bif d∈S∖{0},4if d∉S,(3)\sum_f P_f(d)= \begin{cases} -b & \text{if }d\in S\setminus\{0\},\\ 4 & \text{if }d\notin S, \end{cases} \tag*{(3)}

for a subgroup SS of GG, where the sum is over the generators ff of XX, YY, ZZ and WW. When b∈{0,4}b\in\{0,4\}, GG is cyclic, S=GS=G and (3) is the hypothesis of Theorem 3 or Theorem 5; when b≥12b\geq12, SS is a proper subgroup of index b/4+1b/4+1. Summing (3) over d≠0d\ne0 gives ∑fsf2=4v+∑d≠0∑fPf(d)\sum_f s_f^2=4v+\sum_{d\ne0}\sum_f P_f(d) for the row sums sfs_f of the generators.

Definition 2. Let U≤Aut⁡(G)U\le\operatorname{Aut}(G). A map f:G→{±1}f:G\to\{\pm1\} is UU-invariant, and UU is a multiplier group for ff, if f(σx)=f(x)f(\sigma x)=f(x) for all σ∈U\sigma\in U and x∈Gx\in G, that is, if ff is constant on the UU-orbits of GG.

The five matrices with b=0b=0, at orders 892, 1132, 1244, 1948 and 1964, are given by Theorem 3 with circulants of prime order m=pm=p, and their generators share a multiplier group. In the theory of difference sets a multiplier tt of a set D⊆ZmD\subseteq\mathbb{Z}_m need only satisfy tD=D+gttD=D+g_t for some gt∈Zmg_t\in\mathbb{Z}_m [8]; when gcd⁡(∣U∣,m)=1\gcd(|U|,m)=1 such a DD has a UU-invariant translate, with the same periodic autocorrelation.

Lemma 6. If aa is UU-invariant, then Pa(ts)=Pa(s)P_a(ts)=P_a(s) for all t∈Ut\in U and s∈Zms\in\mathbb{Z}_m.

If the four generators are UU-invariant and UU has rr orbits on Zm\mathbb{Z}_m, each generator is determined by a choice of sign for each orbit. The multiplier group is a restriction chosen for the search. For m=pm=p prime, r=1+(p−1)/∣U∣r=1+(p-1)/|U|. At these five orders ∣U∣=3,3,5,9,7|U|=3,3,5,9,7, and no generator is symmetric.

The four matrices with b=4b=4, at orders 668, 716, 1676 and 1772, are given by Theorem 5, up to the signs of the border, from Legendre quadruples of length m=p−1m=p-1. At orders 668 and 716 the generators have no multiplier, so all 4m4m signs are free; Best, Đoković, Kharaghani and Ramp observed that Turyn-type sequences TT(56)TT(56) and TT(60)TT(60) would give these two orders [7]. At orders 1676 and 1772, m=418=2⋅11⋅19m=418=2\cdot11\cdot19 and m=442=2⋅13⋅17m=442=2\cdot13\cdot17, with multiplier groups of orders 15 and 12. For squarefree m=p1⋯psm=p_1\cdots p_s, rr is the sum of (pi−1)/∣UI∣(p_i-1)/|U_I| over the subsets I⊆{1,…,s}I\subseteq\{1,\ldots,s\}, where UIU_I is the image of UU in ∏i∈IFpi×\prod_{i\in I}\mathbb{F}_{p_i}^{\times}.

The three matrices with b≥12b\ge12 are over non-cyclic groups: at order 1916, G=Z22×Z119G=\mathbb{Z}_2^2\times\mathbb{Z}_{119} and S=Z119S=\mathbb{Z}_{119}; at order 1388, G=Z2×Z3×Z57G=\mathbb{Z}_2\times\mathbb{Z}_3\times\mathbb{Z}_{57} and S=Z57S=\mathbb{Z}_{57}; at order 1436, G=Z22×Z8×Z11G=\mathbb{Z}_2^2\times\mathbb{Z}_8\times\mathbb{Z}_{11} and S=Z28×Z11S=\mathbb{Z}_{28}\times\mathbb{Z}_{11}. Each is given by the following theorem, with k=4k=4, 6 and 8.

Theorem 7. Let GG be an abelian group of order vv, let SS be a proper subgroup of GG of index kk, and let b=4(k−1)b=4(k-1). Let ΠS\Pi_S be the v×kv\times k incidence matrix of elements and cosets of SS, and Π=I4⊗ΠS\mathit{\Pi}=I_4\otimes\Pi_S. Let X,Y,Z,WX,Y,Z,W be group-developed {±1}\{\pm1\}-matrices over GG with

XXT+YYT+ZZT+WWT=(4v+b)Iv−4kΠSΠST+4Iv,XX^{\mathsf{T}}+YY^{\mathsf{T}}+ZZ^{\mathsf{T}}+WW^{\mathsf{T}}=(4v+b)I_v-4k\Pi_S\Pi_S^{\mathsf{T}}+4I_v,

and let MM be the array of Theorem 3 in X,Y,Z,WX,Y,Z,W. Let KK be a Hadamard matrix of order bb, and let EE and FF be {±1}\{\pm1\}-matrices of sizes b×4kb\times4k and 4k×b4k\times b with

EET=4kIb,FFT=4kI4k−4(I4⊗Jk),MΠET+ΠFKT=0.EE^{\mathsf{T}}=4kI_b,\qquad FF^{\mathsf{T}}=4kI_{4k}-4(I_4\otimes J_k),\qquad M\mathit{\Pi}E^{\mathsf{T}}+\mathit{\Pi}FK^{\mathsf{T}}=0.

Then

H=(KEΠTΠFM)H=\begin{pmatrix} K & E\Pi^{\mathsf{T}} \\ \Pi F & M \end{pmatrix}

is a Hadamard matrix of order 4v+b4v+b.

Since ΠSΠST\Pi_S\Pi_S^{\mathsf{T}} is the {0,1}\{0,1\}-matrix of pairs lying in a common coset of SS, the condition on X,Y,Z,WX,Y,Z,W is (3). The theorem follows on computing HHTHH^{\mathsf{T}} blockwise, using ΠTΠ=(v/k)I4k\Pi^{\mathsf{T}}\Pi=(v/k)I_{4k} and ΠSJkΠST=Jv\Pi_SJ_k\Pi_S^{\mathsf{T}}=J_v.

At order 1388 the four generators are fixed by a common group of order 6. At order 1436 they are fixed by multiplication by the quadratic residues modulo 11 on the factor Z11\mathbb{Z}_{11}. The sets of entries −1-1 of the four generators have sizes k1,…,k4k_1,\ldots,k_4. When b≤4b\leq4 they form supplementary difference sets $4$-{v;k1,k2,k3,k4;λ}\{v;k_1,k_2,k_3,k_4;\lambda\}: every non-zero element of GG arises exactly λ\lambda times as a difference of two elements of one of the sets. When b≥12b\geq12 they form a divisible difference family relative to SS [77], in which each element of S∖{0}S\setminus\{0\} arises λ1\lambda_1 times and each element of G∖SG\setminus S arises λ2=λ1+b/4+1\lambda_2=\lambda_1+b/4+1 times. In every case k1+k2+k3+k4−λ1=v+b/4k_1+k_2+k_3+k_4-\lambda_1=v+b/4, with λ1=λ\lambda_1=\lambda when $b\leq 4 ({xref}Table 2`).

For a character χ\chi of GG write f^(χ)=∑x∈Gf(x)χ(x)\widehat{f}(\chi)=\sum_{x\in G}f(x)\chi(x). By (3), ∑f∣f^(χ)∣2=n\sum_f\lvert\widehat{f}(\chi)\rvert^2=n for every character χ\chi not trivial on SS, so a generator has a flat spectrum when ∣f^(χ)∣2=n/4\lvert\widehat{f}(\chi)\rvert^2=n/4 at all such χ\chi. We measure flatness by the mean Φf\Phi_f of (4∣f^(χ)∣2/n−1)2(4\lvert\widehat{f}(\chi)\rvert^2/n-1)^2 over these characters. At b=0b=0, Parseval’s identity gives v(v−1)Φf=∑d≠0Pf(d)2−(sf2−v)2/vv(v-1)\Phi_f=\sum_{d\ne0}P_f(d)^2-(s_f^2-v)^2/v, so Φf\Phi_f is essentially the mean square of Pf(d)/vP_f(d)/\sqrt{v}.

A random sequence has Φf≈1\Phi_f\approx1. To model a typical solution, we treat the values ∣f^(χ)∣2\lvert\widehat{f}(\chi)\rvert^2 at characters in distinct orbits of ⟨U,−1⟩\langle U,-1\rangle as independent exponential variables, as they approximately are for random sequences, conditioned on the four generators summing to nn at each character. A typical solution then has Φf≈0.6\Phi_f\approx0.6, and a typical solution with ∣f^(χ)∣2<n/2\lvert\widehat{f}(\chi)\rvert^2<n/2 at every such character has Φf≈0.3\Phi_f\approx0.3.

The generators of the matrices with b≥4b\geq4 are typical, with $0.27<\Phi_f<0.93 ({xref}Table 2`). Those with b=0b=0 are much flatter: ∣f^(χ)∣2<n/2\lvert\widehat{f}(\chi)\rvert^2<n/2 at every non-trivial character, and 0.08<Φf<0.140.08<\Phi_f<0.14, well below the value 0.3 that this cutoff alone gives. By a Chernoff bound under the model, at most 2−572^{-57} (at 1948) to 2−1412^{-141} (at 1132) of the solutions with the same multiplier group are as flat. This suggests that the search at b=0b=0 was restricted to sequences with nearly flat spectra. At 668 and 716, by contrast, the generators have no multiplier group and are not unusually flat, so there is no evidence that either restriction was used at these orders.

nn(sX,sY,sZ,sW)(s_X,s_Y,s_Z,s_W)44-{v;k1,k2,k3,k4;λ}\{v;k_1,k_2,k_3,k_4;\lambda\}(ΦX,ΦY,ΦZ,ΦW)(\Phi_X,\Phi_Y,\Phi_Z,\Phi_W)
892(11,11,11,23)(11,11,11,23){223;106,106,106,100;195}\{223;106,106,106,100;195\}(0.09,0.09,0.09,0.09)(0.09,0.09,0.09,0.09)
1132(19,19,19,7)(19,19,19,7){283;132,132,132,138;251}\{283;132,132,132,138;251\}(0.08,0.09,0.08,0.08)(0.08,0.09,0.08,0.08)
1244(21,19,21,1)(21,19,21,1){311;145,146,145,155;280}\{311;145,146,145,155;280\}(0.10,0.09,0.10,0.10)(0.10,0.09,0.10,0.10)
1948(17,1,17,37)(17,1,17,37){487;235,243,235,225;451}\{487;235,243,235,225;451\}(0.14,0.14,0.12,0.12)(0.14,0.14,0.12,0.12)
1964(29,27,15,13)(29,27,15,13){491;231,232,238,239;449}\{491;231,232,238,239;449\}(0.13,0.11,0.13,0.10)(0.13,0.11,0.13,0.10)
668(2,0,0,0)(2,0,0,0){166;82,83,83,83;164}\{166;82,83,83,83;164\}(0.36,0.56,0.56,0.44)(0.36,0.56,0.56,0.44)
716(2,0,0,0)(2,0,0,0){178;88,89,89,89;176}\{178;88,89,89,89;176\}(0.50,0.55,0.48,0.59)(0.50,0.55,0.48,0.59)
1676(2,0,0,0)(2,0,0,0){418;208,209,209,209;416}\{418;208,209,209,209;416\}(0.48,0.92,0.28,0.36)(0.48,0.92,0.28,0.36)
1772(2,0,0,0)(2,0,0,0){442;220,221,221,221;440}\{442;220,221,221,221;440\}(0.55,0.62,0.41,0.69)(0.55,0.62,0.41,0.69)
1916(6,6,38,20)(6,6,38,20){476;235,235,219,228;438,442}\{476;235,235,219,228;438,442\}(0.58,0.38,0.34,0.69)(0.58,0.38,0.34,0.69)
1388(6,10,26,24)(6,10,26,24){342;168,166,158,159;304,310}\{342;168,166,158,159;304,310\}(0.49,0.73,0.53,0.45)(0.49,0.73,0.53,0.45)
1436(−14,18,−30,4)(-14,18,-30,4){352;183,167,191,174;356,364}\{352;183,167,191,174;356,364\}(0.67,0.58,0.54,0.78)(0.67,0.58,0.54,0.78)

Table 2. Row sums of the generators, parameters of their sets of entries −1-1, and the flatness Φf\Phi_f of each generator.

Construction of Hadamard matrices of order ≤10 000\le10\,000

Theorems 3, 4 and 5 reduce the construction of a Hadamard matrix to the construction of a few circulant matrices. A circulant is determined by its first row, so the unknowns are sequences, and the hypotheses of the three theorems say that the periodic autocorrelations of the first rows sum to 0, −2-2 and −4-4 respectively at every shift s≠0s \ne0. The non-periodic autocorrelation of a=(a0,…,am−1)a=(a_0,\ldots,a_{m-1}) is Na(s)=∑i=0m−1−saiai+sN_a(s)=\sum_{i=0}^{m-1-s}a_i a_{i+s}.

Sequences whose non-periodic autocorrelations cancel are more rigid, but they can be concatenated and interleaved to give longer sequences of the same kind, and most of the recursive constructions rest on them. The families in use are the following.

  • A Golay pair of length gg is a pair a,ba,b of {±1}\{\pm1\}-sequences with Na(s)+Nb(s)=0N_a(s)+N_b(s)=0 for all s≥1s\ge1 [50]. A complex Golay pair is defined in the same way, with entries in {±1,±i}\{\pm1,\pm i\} and Na(s)=∑iaia‾i+sN_a(s)=\sum_i a_i\overline{a}_{i+s}.

  • Base sequences BS(m,n)BS(m,n) are four {±1}\{\pm1\}-sequences of lengths m,m,n,nm,m,n,n whose non-periodic autocorrelations sum to 0 at every s≥1s\ge1.

  • Turyn-type sequences TT(n)TT(n) are four {±1}\{\pm1\}-sequences A,B,C,DA,B,C,D of lengths n,n,n,n−1n,n,n,n-1 with NA(s)+NB(s)+2NC(s)+2ND(s)=0N_A(s)+N_B(s)+2N_C(s)+2N_D(s)=0 at every s≥1s\ge1. The concatenations (C;D)(C;D) and (C;−D)(C;-D), together with AA and BB, form base sequences BS(2n−1,n)BS(2n-1,n) [104].

  • TT-sequences of length tt are four {0,±1}\{0,\pm1\}-sequences, exactly one of which is non-zero in each position, whose non-periodic autocorrelations sum to 0 at every s≥1s\ge1. Base sequences BS(m,n)BS(m,n) give TT-sequences of length m+nm+n [20, 104].

  • TT-matrices of order tt are four circulant {0,±1}\{0,\pm1\}-matrices T1,…,T4T_1,\ldots,T_4 of order tt with disjoint supports and ∑iTiTiT=tIt\sum_i T_iT_i^{\mathsf T}=tI_t. The sums of the periodic autocorrelations of the rows generating the circulants are zero at every s≥1s\ge1. TT-sequences generate TT-matrices but the generating rows of TT-matrices are not necessarily TT-sequences.

  • Williamson-type matrices WT(w)WT(w) are four {±1}\{\pm1\}-matrices A,B,C,DA,B,C,D of order ww with XYT=YXTXY^{\mathsf T}=YX^{\mathsf T} for all X,Y∈{A,B,C,D}X,Y\in\{A,B,C,D\} and AAT+BBT+CCT+DDT=4wIwAA^{\mathsf T}+BB^{\mathsf T}+CC^{\mathsf T}+DD^{\mathsf T}=4wI_w. Plugged into the Williamson array they give an H(4w)H(4w). The Williamson matrices of § 2.1 are the circulant symmetric case.

A theorem of Cooper and Wallis combines TT-matrices of order tt and Williamson-type matrices of order ww into a Hadamard matrix of order 4tw4tw [20]. Taking w=1w=1, TT-sequences of length tt give an H(4t)H(4t), so base sequences BS(m,n)BS(m,n) give an H(4(m+n))H(4(m+n)), and TT(n)TT(n) give an H(4(3n−1))H(4(3n-1)). The Hadamard matrix of order 428 =4⋅107=4\cdot107 was obtained in this way from a TT(36)TT(36) [59]. Golay pairs enter through Turyn’s and Yang’s composition theorems, which combine them with base sequences or TT-sequences to give TT-sequences of greater length [104, 118]. Craigen, Holzmann and Kharaghani showed that complex Golay pairs of lengths xx and yy give a Hadamard matrix of order 8(x+y)8(x+y) [23]. These constructions can be described with orthogonal designs [48].

Not every known order comes from a family. Twenty-nine orders up to 10000 are reached by none of the families in Table 4, only by matrices found by computer. Four of these were obtained by Djokovic and coauthors from supplementary difference sets in the Goethals–Seidel array: 764 [34], 956 [33], and 1004 and 2524 [37]. Twelve are those of § 3; five, 3732, 4492, 5020, 5788 and 5932, were found by the present authors; and eight, described below, were found by Schumer, Cooley and Kumar. Orders first settled by computer but now reached by a family are not counted as sporadic: Sawade’s order 268 [83], and Djokovic’s order 652 =4⋅163=4\cdot163 [30] and our order 8476 =4⋅13⋅163=4\cdot13\cdot163, which the theorem of Cooper and Wallis now gives from TT-matrices of order 163. Cati and Pasechnik have implemented the known constructions up to 1208 in SageMath [17].

ConstructionOrdersCount
Direct constructions
Paley I [81]q+1q+1, q≡3 mod 4q \equiv3 \bmod4.625
Paley II [81]2(q+1)2(q+1), q≡1 mod 4q \equiv1 \bmod4.355
Sylvester [96]2a2^a.12
Spence [94]4(1+q+q2)4(1+q+q^2), with 1+q+q21+q+q^2 prime ≡3,5,7 mod 8\equiv3,5,7 \bmod8 or 2q2+2q+32q^2+2q+3 a prime power.12
Mathon [73]2(q2(q+2)+1)2(q^2(q+2)+1), q≡3 mod 4q \equiv3 \bmod4, q+2q+2 a prime power.3
Seberry–Whiteman [90]2(5⋅92t+1+1)2(5\cdot9^{2t+1}+1).2
Recursive constructions
Cooper–Wallis [20]4tw4tw, given TT-matrices of order t>1t>1 and WT(ww).1950
Paley and Miyamoto blocks [91, 62]4st4st, given TT-matrices of order t>1t>1, with 4s−1≡3 mod 84s-1 \equiv3 \bmod8 a prime power; also 6108, 7532, 9708.1662
Williamson array [112]4w4w, given WT(ww).1595
Miyamoto II [74]WT(qq) for q≡1 mod 4q \equiv1 \bmod4, given a symmetric BH((q−1)/2,4q-1)/2,4); reaches 4tq4tq.1491
Agaian [1]8ab8ab, given H(4a)H(4a), H(4b)H(4b).1230
Kronecker productabab, given H(a)H(a), H(b)H(b).1227
Craigen–Holzmann–Kharaghani [23]8(x+y)8(x+y), given complex Golay pairs of lengths xx and yy.1001
Craigen–Seberry–Zhang [24]16abcd16abcd, given H(4a)H(4a), H(4b)H(4b), H(4c)H(4c), H(4d)H(4d).624
Miyamoto I [74]4q4q, q≡1 mod 4q \equiv1 \bmod4, given H(q−1)H(q-1).200
Seberry–Yamada [92]8q8q, given WT((q−1)/2q-1)/2).165
Yamada [116]4(q+2)4(q+2), q≡5 mod 8q \equiv5 \bmod8, given skew H((q+3)/2)H((q+3)/2).78
Miyamoto III [74 Theorem 7]8q8q, q≡3 mod 4q \equiv3 \bmod4, with qq and 2q−32q-3 prime.46
Yamada [116]4(q+2)4(q+2), q≡1 mod 8q \equiv1 \bmod8, given a symmetric conference matrix of order (q+3)/2(q+3)/2.34
Sporadic (computer search)764 [34], 956 [33], 1004 and 2524 [37]; 668, 716, 892, 1132, 1244, 1388, 1436, 1676, 1772, 1916, 1948, 1964 [3]; 3732, 4492, 5020, 5788, 5932 (this paper); 2092, 2292, 2588 [19]; 3676 [88, 19]; 2148, 2284, 2636, 3964 [62].29

Table 4. Constructions of Hadamard matrices and the number of orders n≤10000n \le10000 each reaches. An order reached by several constructions is counted under each. WT(ww) denotes Williamson-type matrices of order ww.

Table 3 lists the constructions used to cover orders up to 10000, with the number of orders each reaches. Table 3 records where the ingredients of the recursive constructions are known to exist. The remaining recursive constructions take Hadamard matrices or Williamson-type matrices as input, and we refer to [92] for them.

IngredientUsed inKnown to existSmallest open
Golay pairs, length ggTT-sequencesg=2a10b26cg = 2^{a}10^{b}26^{c} [50, 104]; no other g≤105g \le105 [9, 45].106
Turyn-type sequences TT(n)TT(n)TT-sequenceseven n≤46n \le46 [7, 69, 70].48
Base sequences BS(n+1,n)BS(n+1,n)TT-sequencesn≤43n \le43 [35, 108], and n=2a10b26cn = 2^{a}10^{b}26^{c} [35].44
TT-sequences, length ttCooper–Wallisgg, 2g+12g+1 and g+g′g+g' for Golay lengths g,g′g,g'; 2n+12n+1 from BS(n+1,n)BS(n+1,n); 3n−13n-1 from TT(n)TT(n); Yang multiplication [118]. All t≤100t \le100 except possibly t=97t = 97 [35].97
TT-matrices, order ttCooper–Wallisas for TT-sequences; t=97t = 97, 109, 197 [62]; 103 [87]; 151 [88]; 163 and 181 [57]; 151 and 163 also, independently, in this paper.127
Williamson-type matrices WT(w)WT(w)Williamson array, Cooper–Wallis, Seberry–Yamada(q+1)/2(q+1)/2, q≡1(mod4)q \equiv1 \pmod{4} [103]; qq from a symmetric BH((q−1)/2,4)BH((q-1)/2,4), q≡1(mod4)q \equiv1 \pmod{4} [74]; 2uv2uv from WT(u)WT(u), WT(v)WT(v) [92]; the orders below 2000 in [92 Table A.1] except 543 and those resting on its Corollary 8.8(1); every odd w≤69w \le69, and 73, 79, 91, 93, 95, 97, 103, 109, 127, 133, 181 [58]; 111 [62]; 127, from good matrices [31].71
Skew Hadamard matricesYamadaq+1q+1, q≡3(mod4)q \equiv3 \pmod{4} [81]; doubling; Spence family [94]; 4n4n for odd n<250n < 250; 276 and 292 [36]356
Symmetric conference matricesYamadaq+1q+1, q≡1(mod4)q \equiv1 \pmod{4} [81]; q2(q+2)+1q^{2}(q+2)+1, q≡3(mod4)q \equiv3 \pmod{4} and q+2q+2 prime powers [73]; 5⋅92t+1+15 \cdot9^{2t+1}+1 [90]; 66 [51]; 86 [105].118a118^{a}
Complex Golay pairsCraigen–Holzmann–Kharaghani1, 3, 5, 11, 13 and every even n≤26n \le26 except 14 [11]; 2mn2mn for complex Golay lengths m,nm,n [21]; further products [23].29

Table 3. Ingredients of the recursive constructions of Hadamard matrices.

Remark 8. Corollary 2 of [74], and Corollary 8.8(1) of [92], omit the hypothesis that the input be symmetric, which the construction listed in Table 4 as Miyamoto II requires. As a result, order 2076 is wrongly claimed in [74, 87], and order 6108 in [35 Proposition 3.3]. We construct a Hadamard matrix of order 2076 below, and Kumar has constructed one of order 6108 [62].

The known constructions described above give Hadamard matrices at all but 133 orders n≤10 000n \le10\,000. By computer search we found Hadamard matrices at sixteen of these orders. To construct a matrix at order 2076, we found by computer a symmetric BH(86,4)BH(86,4), built from a doubly bordered array and two matrices of order 42 with entries in {±1,±i}\{\pm1,\pm i\}, one circulant and one back-circulant. It gives Williamson-type matrices of order 173, and so the orders 4⋅173⋅t4 \cdot173 \cdot t for odd t≤13t \le13, closing six orders otherwise open. The matrices at orders 3732, 4492, 5932 and 5020 are given by Theorem 3, with generators invariant under multiplier groups of orders 31, 33, 39 and 50. Those at orders 5788 =4(4⋅192+3)= 4(4 \cdot19^{2}+3) and 8476 =4(4⋅232+3)= 4(4 \cdot23^{2}+3) are given by Theorem 7, with G=Z22×Fq2G = \mathbb{Z}_{2}^{2} \times\mathbb{F}_{q}^{2}, S=Fq2S = \mathbb{F}_{q}^{2} and q=19,23q = 19,23. We also found TT-matrices of orders t=151t = 151 and 163, invariant under multiplication by the subgroup of order 5 (resp. 9) of (Z/tZ)∗(\mathbb{Z}/t\mathbb{Z})^{*}. With Williamson matrices of orders 9, 11 and 13, the theorem of Cooper and Wallis gives the orders 5436, 6644 and 7852 from t=151t = 151, and order 7172 from t=163t = 163. These TT-matrices, and these four orders, had been posted earlier by others; we found ours independently. Schumer gave TT-matrices of order 151 and Hadamard matrices of orders 5436, 6644 and 7852 [88], and Jones gave TT-matrices of order 163 and Hadamard matrices of orders 7172 and 8476 [57].

Twenty-two further orders were settled by others while this paper was in preparation; we checked each of their matrices by computing HHTHH^{T} exactly. Cooley found matrices of orders 2092, 2292 and 2588 by Theorem 5, from Legendre quadruples of lengths 522, 572 and 646, and of order 3676 by Theorem 3 over Z919\mathbb{Z}_{919} [19]; Schumer had earlier given a skew Hadamard matrix of order 3676 by the same theorem [88]. Kumar found a matrix of order 2284 by Theorem 3 over Z571\mathbb{Z}_{571}, and matrices of orders 2636 and 3964 by Theorem 7 with k=4k=4, over Z24×Z41\mathbb{Z}_2^4 \times\mathbb{Z}_{41} and Z22×Z247\mathbb{Z}_2^2 \times\mathbb{Z}_{247} [62]. Kharaghani, Mohammadian, Tayfeh-Rezaie and Zaitsev independently found a BH(1142, 4), which also gives order 2284 [119]. At order 2148 Kumar used a border of width 12 over Z3×Z178\mathbb{Z}_3 \times\mathbb{Z}_{178}, with S=Z178S=\mathbb{Z}_{178} of index 3, where the sum of the periodic autocorrelations is −12-12 on S∖{0}S\setminus\{0\} and 0 off SS.

Kumar constructed thirteen further orders without search, from a product theorem of Seberry and Yamada [91 Theorem 4]: if a Hadamard matrix of order 4s4s is a 4×44\times4 array of s×ss\times s circulants NkjN_{kj}, and T1,…,T4T_1,\ldots,T_4 are TT-matrices of order tt, then the matrices Aj=∑kTk⊗NkjA_j=\sum_k T_k\otimes N_{kj} satisfy ∑jAjAjT=4stI\sum_j A_jA_j^{\mathsf{T}}=4stI, and the Goethals–Seidel array over Zt×Zs\mathbb{Z}_t\times\mathbb{Z}_s gives a Hadamard matrix of order 4st4st. For a prime power q≡3 mod 8q\equiv3\bmod8, Paley’s matrix of order q+1q+1 is equivalent to such an array with s=(q+1)/4s=(q+1)/4 [117, 85]. With TT-matrices of order t=3t=3, 5 and 7 this gives the ten orders 3156, 5260, 5860, 6036, 6356, 6756, 7364, 7660, 8204 and 8620, each of the form 4st4st with 4s−14s-1 prime. Miyamoto’s matrices of order 4q4q [74], built from near-Williamson matrices [58] or from a periodic Golay pair [39], can be arranged in the same way, and give the orders 6108, 7532 and 9708, with q=509q = 509, 269269 and 809809. Finally, order 9172=4⋅22939172 = 4 \cdot2293 follows from Cooley’s matrix of order 22922292 by Miyamoto I. Table 5 lists the remaining 9595 orders n≤10 000n \le10\,000 at which no Hadamard matrix is currently known. It was produced by applying the constructions of Table 4 to all orders currently known, including the twelve of § 3 and those described above. It updates [93 Table A.17], which gives, for each odd qq, the least tt for which a Hadamard matrix of order 2tq2^{t}q was known. The smallest open order is now 2396=4⋅5992396 = 4 \cdot599. Every prime appearing in this table is congruent to 3 mod 43 \mathbin{\bmod} 4.

nn#pp
4p4p69599, 643, 719, 739, 751, 839, 859, 863, 907, 947, 971, 1019, 1031, 1039, 1051, 1087, 1103, 1187, 1223, 1259, 1283, 1291, 1319, 1327, 1367, 1423, 1427, 1439, 1451, 1471, 1487, 1499, 1543, 1559, 1567, 1571, 1579, 1583, 1619, 1663, 1667, 1699, 1747, 1783, 1787, 1831, 1847, 1871, 1879, 1907, 1979, 1987, 2027, 2039, 2063, 2083, 2087, 2099, 2203, 2207, 2251, 2287, 2339, 2347, 2371, 2383, 2399, 2423, 2459
4⋅3p4 \cdot3p7223, 419, 479, 491, 571, 631, 643
4⋅5p4 \cdot5p6191, 223, 359, 419, 443, 463
4⋅7p4 \cdot7p3167, 283, 347
4⋅9p4 \cdot9p1191
4⋅11p4 \cdot11p1179
4⋅13p4 \cdot13p2167, 179
8p8p6599, 719, 739, 839, 1031, 1087
95

Table 5. Orders n≤10 000n \le10\,000 at which no Hadamard matrix is known, grouped by the form of nn. Each entry is the prime pp; the second column gives the number of orders in each row.

Open problems

We collect open problems on Hadamard matrices and on five related topics, with recent references for each. The open cases in Table 3 form a further series of open problems.

Hadamard matrices

Problem 1 (Legendre pairs). Does a Legendre pair exist at every odd length ℓ\ell?

Problem 2 (Kotsireas, 2019). Let pp and qq be odd primes, not necessarily distinct, and let (⋅p)\left(\frac{\cdot}{p}\right) be the Legendre symbol modulo pp. Do the sequences A(p,q)=[1,q⋅(1p),…,q⋅(p−1p)]A_{(p,q)} = \left[1, q \cdot\left(\frac{1}{p}\right), \ldots, q \cdot\left(\frac{p-1}{p}\right)\right], B(p,q)=[1,−q⋅(1p),…,−q⋅(p−1p)]B_{(p,q)} = \left[1, -q \cdot\left(\frac{1}{p}\right), \ldots, -q \cdot\left(\frac{p-1}{p}\right)\right] of length pp always q2q^{2}-uncompress to a Legendre pair of length pq2pq^{2}, that is, are they the q2q^{2}-compressions of some putative Legendre pair of length pq2pq^{2}? Here the q2q^{2}-compression of (a0,…,apq2−1)(a_{0}, \ldots, a_{pq^{2}-1}) is the sequence of length pp with jjth entry ∑i=0q2−1aj+ip\sum_{i=0}^{q^{2}-1} a_{j+ip}. See [60] for computational results.

Problem 3 (Turyn-type sequences). Does TT(n)TT(n) exist for every even nn? Is there an infinite family of Turyn-type sequences?

Paley’s matrices of order q+1q + 1 with q≡3 mod 4q \equiv3 \mathbin{\bmod} 4 are skew, and Seberry conjectured that skew Hadamard matrices exist at every order divisible by 44; see [61].

Problem 4 (Skew Hadamard matrices). Is there a skew Hadamard matrix at every order 4n4n?

The maximal determinant problem

When n>2n > 2 and 4∤n4 \nmid n there is no Hadamard matrix, and one asks for

Dn=max⁡{∣det⁡M∣:M∈{±1}n×n}.D_n = \max\{|\det M| : M \in\{\pm1\}^{n \times n}\}.

For odd nn, Barba’s bound Dn≤Bn=2n−1(n−1)(n−1)/2D_n \le B_n = \sqrt{2n-1}(n-1)^{(n-1)/2} can be attained only if 2n−12n-1 is a square [4]. For n≡2(mod4)n \equiv2 \pmod{4}, the bound Dn≤Wn=(2n−2)(n−2)(n−2)/2D_n \le W_n = (2n-2)(n-2)^{(n-2)/2} of Ehlich and Wojtas can be attained only if 2n−22n-2 is a sum of two squares [40, 114]. For n≡3(mod4)n \equiv3 \pmod{4}, Ehlich gave a sharper bound EnE_n, which for n≥63n \ge63 can be attained only if 7∣n7 \mid n [40]; Cohn showed that En2E_n^2 is an integer only when n=112t2±28t+7n = 112t^2 \pm28t + 7 [18], and Tamura, using the Hasse–Minkowski theorem, that the smallest order at which EnE_n could be attained is at least 511 [100]. The survey [14] gives proofs and the state of the problem in 2021.

At n≡2(mod4)n \equiv2 \pmod{4}, Momihara, Suda and Xiang [76], and Li, Momihara and Xiang [67], constructed infinite families of matrices with ∣det⁡M∣≥n−2n−1Wn|\det M| \ge\frac{n-2}{n-1}W_n; their orders form a set of density zero. Butbaia et al. found the largest known determinants at n=51,107,111,115,119n = 51, 107, 111, 115, 119 [15], and Nuñez Ponasso studied the problem over roots of unity [78]. At n≡3(mod4)n \equiv3 \pmod{4} the best ratio known at a single order is Dn/En≥0.8122D_n/E_n \ge0.8122, at n=119n = 119 [15], and the best known along an infinite family is 0.48 [14].

Problem 5 (Barba bound). Is Dn=BnD_n = B_n at every nn for which 2n−12n-1 is a square? Brouwer [13] and Whiteman [110] showed that it is at n=2q2+2q+1n = 2q^2 + 2q + 1 for every odd prime power qq; the smallest orders left undecided in [14] are 85 and 145.

Problem 6 (Ehlich bound). Is Dn=EnD_n = E_n for any n≡3(mod4)n \equiv3 \pmod{4} with n>3n > 3?

Problem 7 (Near-optimal families). Is there an infinite set of orders n≡3(mod4)n \equiv3 \pmod{4} along which Dn/En→1D_n/E_n \to1? Does Dn/Wn→1D_n/W_n \to1 as n→∞n \to\infty through all n≡2(mod4)n \equiv2 \pmod{4}? For the second, the best general bound is due to Brent, Osborn and Smith: Dn/Wn>1/πD_n/W_n > 1/\pi for large nn whenever a Hadamard matrix of order n−2n-2 exists [10].

Butson Hadamard matrices

A complex Hadamard matrix of order nn is an n×nn \times n matrix HH with entries of modulus 1 and HH∗=nIHH^* = nI; the character table of Zn\mathbb{Z}_n is one. The Butson Hadamard matrices BH⁡(n,k)\operatorname{BH}(n,k) of Section 2 are the complex Hadamard matrices with entries in the kkth roots of unity. The inner product of two rows is a vanishing sum of roots of unity, and the theorem of Lam and Leung [64] shows that a BH⁡(n,pr)\operatorname{BH}(n,p^r) with pp prime and n>1n > 1 exists only if p∣np \mid n. A BH⁡(n,4)\operatorname{BH}(n,4) gives a Hadamard matrix of order 2n2n [102]. No such condition arises for k=6k = 6, and Craigen and Szőllősi constructed a BH⁡(p2,6)\operatorname{BH}(p^2,6) for every prime pp [97 Theorem 1.4.41]. Tadej and Życzkowski survey complex Hadamard matrices [99], and Szőllősi’s thesis treats the Butson case [97]. A non-existence criterion using algebraic number theory is due to Winterhof [113].

The first BH⁡(94,4)\operatorname{BH}(94,4) was found by Szőllősi, from circulant {±1}\{\pm1\}-matrices of order 47 placed in a modification of a block construction of Kharaghani and Seberry [98], and the first BH⁡(118,4)\operatorname{BH}(118,4) by Kharaghani, Mohammad­ian, Tayfeh-Rezaie and Zaitsev, from four circulant matrices of order 59 of which three are symmetric [58]. Czifra, Matolcsi and Szőllősi derived a necessary condition for the existence of a BH⁡(n,k)\operatorname{BH}(n,k) from positive definite functions on finite groups, and constructed the first BH⁡(62,6)\operatorname{BH}(62,6), BH⁡(82,6)\operatorname{BH}(82,6) and BH⁡(146,6)\operatorname{BH}(146,6) [25].

Problem 8 (Fourth roots). Does a BH⁡(n,4)\operatorname{BH}(n,4) exist at every even order nn? This is a conjecture of Seberry [97 Conjecture 1.4.11], and it implies the Hadamard conjecture.

Problem 9 (Sixth roots). Does a BH⁡(2p,6)\operatorname{BH}(2p,6) exist for every prime pp? This is a conjecture of Szőllősi [97 Conjecture 1.4.47]. According to [25], the undecided even orders below 100 are 46, 74, 86 and 94, all of this form.

Modular Hadamard matrices

Marrero and Butson defined an mm-modular Hadamard matrix of order nn to be an n×nn \times n {±1}\{\pm1\}-matrix HH with HHT≡nI(modm)HH^{\mathsf{T}} \equiv nI \pmod{m} [71, 72]. The entries of HHTHH^{\mathsf{T}} lie in [−n,n][-n,n], so an mm-modular Hadamard matrix of order n<mn<m is a Hadamard matrix. The mm-modular Hadamard conjecture asserts that an mm-modular Hadamard matrix exists at every order divisible by 44; the Hadamard conjecture is equivalent to this assertion for infinitely many mm. It is known for m=12m=12 [71, 72], m=5m=5 [65] and m=32m=32, and so for their divisors. Eliahou and Kervaire settled m=32m=32 using modular Golay quadruples, which give modular Hadamard matrices through the Goethals–Seidel array [43]. Kuperberg settled m=7m=7 and m=11m=11 at all but finitely many orders [63]. Eliahou and Kervaire survey the subject [44]. Eliahou constructed a 6464-modular Hadamard matrix of order 668668, and announced 6464-modular Hadamard matrices at all orders 4ℓ4\ell [42, 41].

Problem 10 (128-modular Hadamard matrices). Does a 128128-modular Hadamard matrix exist at every order divisible by 44?

Problem 11 (7-modular Hadamard matrices). Does a 77-modular Hadamard matrix exist at every order divisible by 44?

Asymptotic existence

The asymptotic existence theorems of Seberry and her successors have the following form: there are constants a,b≥0a,b\geq0 such that for every odd qq there is a Hadamard matrix of order 2tq2^tq for every t≥a+blog⁡2qt\geq a+b\log_2q. The Hadamard conjecture is the case a=2a=2, b=0b=0. Seberry showed that one may take b=2b=2 [106]. Craigen improved the bound using signed groups [22]; bb was reduced in stages to 3/83/8 [26], and Livinskyi reduced it further [68]. The introduction of [26] summarises these results. De Launey showed that for every ε>0\varepsilon>0 the odd qq for which there is a Hadamard matrix of order 22+⌊εlog⁡2q⌋q2^{2+\lfloor\varepsilon\log_2q\rfloor}q form a set of positive density [26]. Paley’s constructions give Ω(x/log⁡x)\Omega(x/\log x) orders up to xx at which a Hadamard matrix exists; de Launey and Gordon showed that Kronecker products of Paley matrices give at least xlog⁡xexp⁡((C+o(1))(log⁡log⁡log⁡x)2)\frac{x}{\log x}\exp\bigl((C+o(1))(\log\log\log x)^2\bigr), with C=0.8178…C=0.8178\ldots, and that the other known constructions do not improve this bound [28]. Alexeev, Jasper and Mixon showed that for every α<17/92\alpha<17/92 and every sufficiently large nn there is an n×nn\times n {±1}\{\pm1\}-matrix with condition number at most 1+n−α1+n^{-\alpha}, and that for 4∤n4\nmid n the condition number is at least 1+clog⁡n/n1+c\log n/n for an absolute constant c>0c>0 [2].

Problem 12 (Arithmetic progressions). Is there an infinite arithmetic progression {a+bk:k≥0}\{a+bk:k\geq0\} of positive integers such that a Hadamard matrix of order 4(a+bk)4(a+bk) exists for every kk?

Problem 13 (Bounded exponent). Is there an absolute constant cc such that a Hadamard matrix of order 2tq2^tq exists for every odd qq and every t≥ct\geq c?

Group-developed matrices and Gauss sums

Let GG be an abelian group of order nn and exponent ee, and let f:G→{±1}f:G\to\{\pm1\}. The matrix [f(y−x)]x,y∈G[f(y-x)]_{x,y\in G} is a Hadamard matrix if and only if ∣f^(χ)∣2=n|\widehat{f}(\chi)|^2=n for every character χ\chi of GG, where f^(χ)=∑g∈Gf(g)χ(g)\widehat{f}(\chi)=\sum_{g\in G}f(g)\chi(g) lies in Z[ζe]\mathbb{Z}[\zeta_e]. The existence of group-developed Hadamard matrices is thus a question about algebraic integers of absolute value n\sqrt{n} in a cyclotomic field, and the same holds for Butson Hadamard matrices. Gauss sums are the basic source of such integers. The quadratic Gauss sum underlies Paley’s construction, and evaluations of Gauss and Jacobi sums of higher order give difference sets and strongly regular graphs [115], skew Hadamard difference sets [46], the matrices of large determinant [76], and the group-invariant BH⁡(q2,6)\operatorname{BH}(q^2,6) of Ong and Schmidt [79]. In the other direction, the prime ideal factorisation of an element of absolute value n\sqrt{n} is constrained, and this gives the main nonexistence results, by Turn’s self-conjugacy argument [101] and Schmidt’s field descent [84], developed further by Leung and Schmidt [66]. Ryser conjectured that there is no circulant Hadamard matrix of order greater than 44 [82]. Schmidt’s monograph [86] develops these methods in detail. In October 2026

OpenAI released a manuscript, produced by one of its models, that claims a proof of Ryser’s conjecture; its main theorem is formalised in Lean [80].

A difference set DD in GG is skew Hadamard if GG is the disjoint union of DD, −D-D and {0}\{0\}. The quadratic residues of Fq\mathbb{F}_q, q≡3(mod4)q \equiv3 \pmod{4}, form one, and every known example in an abelian group lies in an elementary abelian group.

Problem 14 (Skew Hadamard difference sets). If an abelian group contains a skew Hadamard difference set, must it be elementary abelian? Momihara showed that one counterexample would give infinitely many [75].

Problem 15 (Momihara, Suda and Xiang). Momihara, Suda and Xiang constructed three families of matrices MnM_n of order n=q+1n=q+1, with q≡1(mod4)q \equiv1 \pmod{4} a prime power, whose determinants they expressed through Gauss and Jacobi sums without evaluating them [76 Theorem 1.6, Classes (4)–(6)]. Is ∣det⁡Mn∣≥0.8Wn|\det M_n| \ge0.8 W_n for every such n≥30n \ge30 in each family? They checked that it is for n≤1000n \le1000.

A Hadamard matrix is cocyclic over a group GG of order nn if it is equivalent to [ψ(g,h)]g,h∈G[\psi(g,h)]_{g,h\in G} for a 2-cocycle ψ ⁣:G×G→{±1}\psi\colon G \times G \to\{\pm1\}; when ψ\psi is a coboundary the matrix is equivalent to a group-developed one. The Sylvester and Paley matrices are cocyclic [56, 29].

Problem 16 (Cocyclic Hadamard matrices). Is there a cocyclic Hadamard matrix at every order divisible by 44? This is a conjecture of de Launey and Horadam [56].

Statement of AI Use

The authors used Anthropic’s Claude (Claude Opus 5.5, via Claude Code, September–October 2026) extensively in this work. Claude wrote the Python and C++ programs used to analyse the twelve matrices of [3], to check constructions claimed in the literature and the matrices found by others described in Section 4, and to construct the symmetric BH(86, 4), the TT-matrices of orders 151 and 163, and the Hadamard matrices of orders 5788 and 8476 of Section 4. Theorem 7 was formulated from this analysis, then checked and rewritten by hand. Sections of the paper were drafted using Claude, then checked and rewritten by hand. Each section was proof-read by all four authors. Every Hadamard matrix constructed in this paper is publicly available at https://www.math74.com/Hadamard/ and was verified by an exact computation of HHTHH^{\mathsf{T}}. The authors take full responsibility for all claims made in this paper.

References

References

  1. [1]S. S. Agaian. Hadamard matrices and their applications, volume 1168 of Lecture Notes in Mathematics. Springer-Verlag, Berlin, 1985.
  2. [2]B. Alexeev, J. Jasper, and D. G. Mixon. Asymptotically optimal approximate Hadamard matrices, 2025. arXiv:2511.14653.arxiv.org/abs/2511.14653
  3. [3]L. Alpöge, P. Voinov, and S. Reynolds-Haertle. Hadamard matrices of orders 668, 716, 892, 1132, 1244, 1388, 1436, 1676, 1772, 1916, 1948 and 1964, 2026. Announcement, 12 August 2026; see https://epoch.ai/frontiermath/open-problems/hadamard.
  4. [4]G. Barba. Intorno al teorema di Hadamard sui determinanti a valore massimo. Giorn. Mat. Battaglini, 71:70–86, 1933.
  5. [5]L. Baumert, S. W. Golomb, and M. Hall. Discovery of an Hadamard matrix of order 92. Bull. Amer. Math. Soc., 68(3):237–238, 1962.DOI
  6. [6]L. D. Baumert. Hadamard matrices of orders 116 and 232. Bull. Amer. Math. Soc., 72(2):237, 1966.DOI
  7. [7]D. Best, D. Ž. Đoković, H. Kharaghani, and H. Ramp. Turyn-type sequences: classification, enumeration, and construction. J. Combin. Des., 21(1):24–35, 2013.DOI
  8. [8]T. Beth, D. Jungnickel, and H. Lenz. Design theory. Vol. I, volume 69 of Encyclopedia of Mathematics and its Applications. Cambridge University Press, Cambridge, second edition, 1999.
  9. [9]P. Borwein and R. A. Ferguson. A complete description of Golay pairs for lengths up to 100. Math. Comp., 73(246):967–985, 2004.DOI
  10. [10]R. P. Brent, J. H. Osborn, and W. D. Smith. Probabilistic lower bounds on maximal determinants of binary matrices. Australas. J. Combin., 66:350–364, 2016.arxiv.org/abs/1501.06235
  11. [11]C. Bright, I. Kotsireas, A. Heinle, and V. Ganesh. Complex Golay pairs up to length 28: a search via computer algebra and programmatic SAT. J. Symbolic Comput., 102:153–172, 2021.
  12. [12]A. E. Brouwer. Parameters of strongly regular graphs. Online tables, https://aeb.win.tue.nl/graphs/srg/srgtab.html, accessed 16 October 2026.
  13. [13]A. E. Brouwer. An infinite series of symmetric designs, volume 202 of Afdeling Zuivere Wiskunde. Mathematisch Centrum, Amsterdam, 1983.
  14. [14]P. Browne, R. Egan, F. Hegarty, and P. Ó Catháin. A survey of the Hadamard maximal determinant problem. The Electronic Journal of Combinatorics, 28(4), 2021.DOI
  15. [15]G. Butbaia, P. Vipulanan­dan, J. Tan, X. Huang, T. Saunders-A’Court, L. Fagan, D. Passaro, M. Tarquini, and S. Gukov. New records for the Hadamard maximal determinant problem in dimensions 51, 107, 111, 115, and 119, 2026. arXiv:2608.22518.
  16. [16]A. T. Butson. Generalized Hadamard matrices. Proc. Amer. Math. Soc., 13:894–898, 1962.DOI
  17. [17]M. Cati and D. V. Pasechnik. A database of constructions of Hadamard matrices, 2024. arXiv:2411.18897.arxiv.org/abs/2411.18897
  18. [18]J. H. E. Cohn. Almost D-optimal designs. Util. Math., 57:121–128, 2000.
  19. [19]T. Cooley. Middletownbooks/Hadamard. https://github.com/Middletownbooks/Hadamard, 2026. Branch claude/focused-planck-j1645f, October 2026.
  20. [20]J. Cooper and J. Wallis. A construction for Hadamard arrays. Bull. Austral. Math. Soc., 7:269–277, 1972.DOI
  21. [21]R. Craigen. Complex Golay sequences. J. Combin. Math. Combin. Comput., 15:161–169, 1994.
  22. [22]R. Craigen. Signed groups, sequences, and the asymptotic existence of Hadamard matrices. J. Combin. Theory Ser. A, 71(2):241–254, 1995.DOI
  23. [23]R. Craigen, W. Holzmann, and H. Kharaghani. Complex Golay sequences: structure and applications. Discrete Math., 252:73–89, 2002.DOI
  24. [24]R. Craigen, J. Seberry, and X. M. Zhang. Product of four Hadamard matrices. J. Combin. Theory Ser. A, 59(2):318–320, 1992.DOI
  25. [25]D. Czifra, M. Matolcsi, and F. Szöllősi. A nonexistence criterion and new constructions for Butson Hadamard matrices. J. Algebraic Combin., 64(2):Paper No. 29, 15, 2026.arxiv.org/abs/2511.11230
  26. [26]W. de Launey. On the asymptotic existence of Hadamard matrices. J. Combin. Theory Ser. A, 116(4):1002–1008, 2009.DOI
  27. [27]W. de Launey and D. Flannery. Algebraic design theory. Mathematical Surveys and Monographs, vol. 175. American Mathematical Society, Providence, RI, 2011.
  28. [28]W. de Launey and D. M. Gordon. On the density of the set of known Hadamard orders. Cryptogr. Commun., 2(2):233–246, 2010.
  29. [29]W. de Launey and R. M. Stafford. On cocyclic weighing matrices and the regular group actions of certain Paley matrices. Discrete Appl. Math., 102(1-2):63–101, 2000. Coding, cryptography and computer security (Lethbridge, AB, 1998).
  30. [30]D. Ž. Đoković. Construction of some new Hadamard matrices. Bull. Austral. Math. Soc., 45(2):327–332, 1992.
  31. [31]D. Ž. Đoković. Good matrices of orders 33, 35 and 127 exist. J. Combin. Math. Combin. Comput., 14:145–152, 1993.
  32. [32]D. Ž. Đoković. Williamson matrices of order 4n for n = 33, 35, 39. Discrete Math., 115:267–271, 1993.
  33. [33]D. Ž. Đoković. Two Hadamard matrices of order 956 of Goethals–Seidel type. Combinatorica, 14(3):375–377, 1994.
  34. [34]D. Ž. Đoković. Hadamard matrices of order 764 exist. Combinatorica, 28(4):487–489, 2008.
  35. [35]D. Ž. Đoković. Small orders of Hadamard matrices and base sequences, 2010. arXiv:1008.2043.
  36. [36]D. Ž. Đoković. Hadamard matrices: skew of orders 276, 292 and symmetric 372, 2023. arXiv:2301.02751.
  37. [37]D. Ž. Đoković, O. Golubitsky, and I. S. Kotsireas. Some new orders of Hadamard and skew-Hadamard matrices. J. Combin. Des., 22(6):270–277, 2014.DOI
  38. [38]D. Ž. Đoković and I. S. Kotsireas. Compression of periodic complementary sequences and applications. Des. Codes Cryptogr., 74(2):365–377, 2015.
  39. [39]D. Ž. Đoković and I. S. Kotsireas. Some new periodic Golay pairs. Numer. Algorithms, 69(3):523–530, 2015.
  40. [40]H. Ehlich. Determinantenabschätzungen für binäre Matrizen. Math. Z., 83:123–132, 1964.
  41. [41]S. Eliahou. Personal communication.
  42. [42]S. Eliahou. A 64-modular Hadamard matrix of order 668. Australas. J. Combin., 93(2):422–427, 2025.
  43. [43]S. Eliahou and M. Kervaire. Modular sequences and modular Hadamard matrices. J. Combin. Des., 9(3):187–214, 2001.DOI
  44. [44]S. Eliahou and M. Kervaire. A survey on modular Hadamard matrices. Discrete Math., 302(1–3):85–106, 2005.DOI
  45. [45]S. Eliahou, M. Kervaire, and B. Saffari. A new restriction on the lengths of Golay complementary sequences. J. Combin. Theory Ser. A, 55(1):49–59, 1990.DOI
  46. [46]T. Feng and Q. Xiang. Cyclotomic constructions of skew Hadamard difference sets. J. Combin. Theory Ser. A, 119(1):245–256, 2012.arxiv.org/abs/1101.2994
  47. [47]R. J. Fletcher, M. Gysin, and J. Seberry. Application of the discrete Fourier transform to the search for generalised Legendre pairs and Hadamard matrices. Australas. J. Combin., 23:75–86, 2001.
  48. [48]A. V. Geramita and J. Seberry. Orthogonal Designs: Quadratic Forms and Hadamard Matrices. Marcel Dekker, New York, 1979.
  49. [49]J.-M. Goethals and J. J. Seidel. A skew Hadamard matrix of order 36. J. Austral. Math. Soc., 11:343–344, 1970.
  50. [50]M. J. E. Golay. Complementary series. IRE Trans. Inform. Theory, IT-7:82–87, 1961.DOI
  51. [51]O. Gritsenko. On strongly regular graph with parameters (65, 32, 15, 16), 2021. arXiv:2102.05432.arxiv.org/abs/2102.05432
  52. [52]J. Hadamard. Résolution d’une question relative aux déterminants. Bull. Sci. Math., 17:240–246, 1893.DOI
  53. [53]M. Hall, Jr. Combinatorial theory. Wiley-Interscience Series in Discrete Mathematics. John Wiley & Sons Inc., New York, second edition, 1986.
  54. [54]A. Hedayat and W. D. Wallis. Hadamard matrices and their applications. Ann. Statist., 6(6):1184–1238, 1978.DOI
  55. [55]W. H. Holzmann, H. Kharaghani, and B. Tayfeh-Rezaie. Williamson matrices up to order 59. Des. Codes Cryptogr., 46(3):343–352, 2008.DOI
  56. [56]K. J. Horadam. Hadamard matrices and their applications. Princeton University Press, Princeton, NJ, 2007.
  57. [57]J. D. Jones. Hadamard-T: T-matrix witnesses and the Hadamard orders they close. https://github.com/JD-Jones-ASES/Hadamard-T, 2026. Version 0.1.0, 29 August 2026.
  58. [58]H. Kharaghani, A. Mohammadian, and B. Tayfeh-Rezaie. A search for Hadamard matrices of Williamson type, 2026. arXiv:2605.08661.arxiv.org/abs/2605.08661
  59. [59]H. Kharaghani and B. Tayfeh-Rezaie. A Hadamard matrix of order 428. J. Combin. Des., 13(6):435–440, 2005.DOI
  60. [60]I. Kotsireas, R. Gallardo-Cava, A. I. Gómez, and D. Gómez-Pérez. On the search of binary Legendre pairs of length pq². J. Symbolic Comput., 138:Paper No. 102606, 20, 2027.
  61. [61]C. Koukouvanos and S. Stylianou. On skew-Hadamard matrices. Discrete Math., 308(13):2723–2731, 2008.DOI
  62. [62]K. Kumar. New Hadamard matrices. https://github.com/vatskishan03/Hadamard, 2026. Commit 2d45dbb, 18 October 2026.
  63. [63]V. Kuperberg. Hadamard matrices modulo p and small modular Hadamard matrices. J. Combin. Des., 24(9):393–405, 2016.
  64. [64]T. Y. Lam and K. H. Leung. On vanishing sums of roots of unity. J. Algebra, 224(1):91–109, 2000.DOI
  65. [65]M. H. Lee and F. Szőllősi. Hadamard matrices modulo 5. J. Combin. Des., 22(4):171–178, 2014.
  66. [66]K. H. Leung and B. Schmidt. New restrictions on possible orders of circulant Hadamard matrices. Des. Codes Cryptogr., 64(1-2):143–151, 2012.DOI
  67. [67]S. Li, K. Momihara, and Q. Xiang. New constructions of square (±1)-matrices with large determinants. Linear Algebra Appl., 752:66–106, 2027.DOI
  68. [68]I. Livinskiy. Asymptotic existence of Hadamard matrices, M.Sc. Thesis. University of Manitoba, 2012.
  69. [69]S. London and I. Kotsireas. New Turny-type sequences. Cryptogr. Commun., 17(5):1601–1610, 2025.
  70. [70]S. London and I. Kotsireas. Further results on Turyn-type sequences. Submitted, 2026.
  71. [71]O. Marrero and A. T. Butson. Modular Hadamard matrices and related designs. II. Canad. J. Math., 24(6):1100–1109, 1972.DOI
  72. [72]O. Marrero and A. T. Butson. Modular Hadamard matrices and related designs. J. Combin. Theory Ser. A, 15(3):257–269, 1973.
  73. [73]R. Mathon. Symmetric conference matrices of order pq² + 1. Canad. J. Math., 30(2):321–331, 1978.DOI
  74. [74]M. Miyamoto. A construction of Hadamard matrices. J. Combin. Theory Ser. A, 57(1):86–108, 1991.DOI
  75. [75]K. Momihara. A recursive construction for skew Hadamard difference sets. Electron. J. Combin., 27(3):Paper No. 3.36, 2020.DOI
  76. [76]K. Momihara, S. Suda, and Q. Xiang. Square (1, −1)-matrices with large determinants and near-Hadamard matrices. Linear Algebra Appl., 654:28–55, 2022.
  77. [77]K. Momihara and M. Yamada. Divisible difference families from Galois rings GR(4,n) and Hadamard matrices. Des. Codes Cryptogr., 73(3):897–909, 2014.
  78. [78]G. Nuñez Ponasso. Maximal determinants of matrices over the roots of unity. Linear Algebra Appl., 723:201–243, 2025.arxiv.org/abs/2503.11114
  79. [79]C. Y. Ong and B. Schmidt. Butson Hadamard matrices from pairs of Gauss sums, 2026. arXiv:2609.22754.arxiv.org/abs/2609.22754
  80. [80]OpenAI. The circulant Hadamard conjecture. Model-generated manuscript, family 179, https://github.com/openai/math, 2026. Released 6 October 2026; main theorem formalised in Lean.hub.valency.io/works/W_xcu3hkq8
  81. [81]R. Paley. On orthogonal matrices. J. Math. Phys., 12:311–320, 1933.DOI
  82. [82]H. J. Ryser. Combinatorial mathematics. The Carus Mathematical Monographs, No. 14. Published by The Mathematical Association of America, 1963.
  83. [83]K. Sawade. A Hadamard matrix of order 268. Graphs Combin., 1:185–187, 1985.DOI
  84. [84]B. Schmidt. Cyclotomic integers and finite geometry. J. Amer. Math. Soc., 12(4):929–952, 1999.DOI
  85. [85]B. Schmidt. Williamson matrices and a conjecture of Ito’s. Des. Codes Cryptogr., 17(1-3):61–68, 1999.DOI
  86. [86]B. Schmidt. Characters and Cyclotomic Fields in Finite Geometry, volume 1797 of Lecture Notes in Mathematics. Springer, Berlin, 2002.DOI
  87. [87]L. Schneider. hadamard-2060. https://github.com/schneiderlo/hadamard-2060, 2026. T-matrices of order 103; included in SageMath, pull request #42876, October 2026.
  88. [88]A. Schumer. Periodic T-matrices of order 151 and a skew spin Goethals–Seidel family over Z/19Z: certificates and Hadamard constructions of orders 3676, 5436, 6644, and 7852. Zenodo, https://doi.org/10.5281/zenodo.2282997, 2026. Version 1.0.0, 17 September 2026.
  89. [89]J. Seberry. A computer listing of Hadamard matrices. In Combinatorial Mathematics (Proc. Internat. Conf., Canberra, 1977), volume 686 of Lecture Notes in Math., pages 275–281. Springer, Berlin, 1978.DOI
  90. [90]J. Seberry and A. L. Whiteman. New Hadamard matrices and conference matrices obtained via Mathon’s construction. Graphs Combin., 4:355–377, 1988.DOI
  91. [91]J. Seberry and M. Yamada. On the products of Hadamard matrices, Williamson matrices and other orthogonal matrices using M-structures. J. Combin. Math. Combin. Comput., 7:97–137, 1990.
  92. [92]J. Seberry and M. Yamada. Hadamard matrices, sequences, and block designs. In Contemporary design theory, Wiley-Intersci. Ser. Discrete Math. Optim., pages 431–560. Wiley, New York, 1992.
  93. [93]J. Seberry and M. Yamada. Hadamard matrices–constructions using number theory and algebra. John Wiley & Sons, Inc., Hoboken, NJ, 2020.
  94. [94]E. Spence. Skew-Hadamard matrices of the Goethals–Seidel type. Canadian J. Math., 27(3):555–560, 1975.DOI
  95. [95]T. Storer. Cyclotomy and difference sets. Lectures in Advanced Mathematics, No. 2. Markham Publishing Co., Chicago, Ill., 1967.
  96. [96]J. Sylvester. Thoughts on inverse orthogonal matrices, simultaneous sign successions, and tessellated pavements in two or more colours, with applications to Newton’s rule, ornamental tile-work, and the theory of numbers. Phil. Mag., 34(1):461–475, 1867.
  97. [97]F. Szöllősi. Construction, classification and parametrization of complex Hadamard matrices. PhD thesis, Central European University, 2011. arXiv:1110.5590.arxiv.org/abs/1110.5590
  98. [98]F. Szöllősi. A complex Hadamard matrix of order 94. J. Combin. Theory Ser. A, 225:Paper No. 106254, 2027.arxiv.org/abs/2603.09572
  99. [99]W. Tadej and K. Życzkowski. A concise guide to complex Hadamard matrices. Open Syst. Inf. Dyn., 13(2):133–177, 2006.DOI
  100. [100]H. Tamura. D-optimal designs and group divisible designs. J. Combin. Des., 14(6):451–462, 2006.
  101. [101]R. J. Turyn. Character sums and difference sets. Pacific J. Math., 15:319–346, 1965.DOI
  102. [102]R. J. Turyn. Complex Hadamard matrices. In Combinatorial Structures and their Applications (Proc. Calgary Internat. Conf., Calgary, Alta., 1969), pages 435–437. Gordon and Breach, New York, 1970.
  103. [103]R. J. Turyn. An infinite class of Williamson matrices. J. Combin. Theory Ser. A, 12:319–321, 1972.DOI
  104. [104]R. J. Turyn. Hadamard matrices, Baumert–Hall units, four-symbol sequences, pulse compression, and surface wave encodings. J. Combin. Theory Ser. A, 16:313–333, 1974.DOI
  105. [105]H. Verma. A symmetric conference matrix of order 86, 2026. arXiv:2610.08893.arxiv.org/abs/2610.08893
  106. [106]J. S. Wallis. On the existence of Hadamard matrices. J. Combinatorial Theory Ser. A, 21(2):188–195, 1976.DOI
  107. [107]W. D. Wallis, A. P. Street, and J. S. Wallis. Combinatorics: Room squares, sum-free sets, Hadamard matrices. Lecture Notes in Mathematics, Vol. 292. Springer-Verlag, Berlin-New York, 1972.DOI
  108. [108]X. Wang and J. Zhu. On base, normal and near-normal sequences, 2025. arXiv:2506.20296.arxiv.org/abs/2506.20296
  109. [109]A. L. Whiteman. An infinite family of Hadamard matrices of Williamson type. J. Combin. Theory Ser. A, 14(3):334–340, 1973.DOI
  110. [110]A. L. Whiteman. A family of D-optimal designs. Ars Combin., 30:23–26, 1990.
  111. [111]G. Williamson, O. Yacobi, and P. Zinn-Justin. Generating Hadamard matrices with transformers, 2026. arXiv:2604.11101.arxiv.org/abs/2604.11101
  112. [112]J. Williamson. Hadamard’s determinant theorem and the sum of four squares. Duke Math. J., 11:65–81, 1944.DOI
  113. [113]A. Winterhof. On the non-existence of generalized Hadamard matrices. J. Statist. Plann. Inference, 84(1-2):337–342, 2000.DOI
  114. [114]M. Wojtas. On Hadamard’s inequality for the determinants of order non-divisible by 4. Colloq. Math., 12(1):73–83, 1964.DOI
  115. [115]Q. Xiang. Cyclotomy, Gauss sums, difference sets and strongly regular Cayley graphs. In Sequences and their applications—SETA 2012, volume 7280 of Lecture Notes in Comput. Sci., pages 245–256. Springer, Heidelberg, 2012.DOI
  116. [116]M. Yamada. Some new series of Hadamard matrices. J. Austral. Math. Soc. Ser. A, 46(3):371–383, 1989.
  117. [117]M. Yamada. Hadamard matrices of generalized quaternion type. Discrete Math., 87(2):187–196, 1991.DOI
  118. [118]C. H. Yang. On composition of four-symbol δ-codes and Hadamard matrices. Proc. Amer. Math. Soc., 107(3):763–776, 1989.
  119. [119]V. Zaitsev, H. Kharaghani, A. Mohammadian, and B. Tayfeh-Rezaie. Quaternary Hadamard matrices. https://www.cs.uleth.ca/~zaitsevv/quaternary.html, 2026. Last updated 14 October 2026.DOI

Paper details

vlnc:W_xadvt4we
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 3

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents