betaFeedback

WorksMathematics

On tt-designs supported by self-orthogonal codes

Abstract

We study self-orthogonal tt-(v,k,λ)(v,k,λ) designs satisfying the condition that the codes generated by the designs are self-orthogonal codes whose codewords of minimum weight kk correspond to the blocks. A classification of such designs is given under the conditions that t≥⌊k/4⌋+1≥2t\geq \lfloor k/4\rfloor+1\geq2 and t≤11t \le 11. This yields a classification of self-orthogonal codes of minimum weight kk whose codewords of minimum weight form a tt-design under the same conditions.

Introduction

A tt-(v,k,λ)(v,k,\lambda) design D\mathcal{D} is a pair of a set P\mathcal{P} of vv points and a collection B\mathcal{B} of kk-element subsets of P\mathcal{P} (called blocks) such that every tt-element subset of P\mathcal{P} is contained in exactly λ\lambda blocks, where λ\lambda is a positive integer. As usual, we assume that t<k<vt < k < v, throughout this paper. Hereafter, all designs are assumed to have no repeated blocks. Often a tt-(v,k,λ)(v,k,\lambda) design is simply called a tt-design. A tt-design can be represented by its (block-point) incidence matrix A=(aij)A=(a_{ij}), where aij=1a_{ij}=1 if the jjth point is contained in the iith block and aij=0a_{ij}=0 otherwise. Two designs with the same parameters are isomorphic if there is a bijection between their point sets that maps the blocks of the first design into blocks of the second design.

A tt-(v,k,λ)\left(v,k,\lambda\right) design D\mathcal{D} is called self-orthogonal if the block intersection numbers have the same parity as the block size kk (see [19]). The term self-orthogonal is due to a natural connection between such designs and self-orthogonal codes. If D\mathcal{D} is a self-orthogonal tt-(v,k,λ)\left(v,k,\lambda\right) design with kk even, then the code C(D)C(\mathcal{D}) generated by the rows of an incidence matrix of D\mathcal{D} is self-orthogonal. Throughout this paper, we assume that kk is even for a self-orthogonal tt-(v,k,λ)\left(v,k,\lambda\right) design, unless otherwise specified.

We define a saturated design to be a self-orthogonal tt-(v,k,λ)\left(v,k,\lambda\right) design D=(P,B)\mathcal{D}=(\mathcal{P},\mathcal{B}) with kk even, satisfying the following conditions:

C(D) is self-dual, that is, C(D)=C(D)⊥,(1)C(\mathcal{D})\text{ is self-dual, that is, }C(\mathcal{D})=C(\mathcal{D})^{\perp}, \tag*{(1)}
B is the set of supports of the codewords of weight k in C(D),(2)\mathcal{B}\text{ is the set of supports of the codewords of weight }k\text{ in }C(\mathcal{D}), \tag*{(2)}
the minimum weight of C(D) is k,(3)\text{the minimum weight of }C(\mathcal{D})\text{ is }k, \tag*{(3)}

where C(D)⊥C(\mathcal{D})^{\perp} denotes the dual code of C(D)C(\mathcal{D}). A self-orthogonal tt-(v,k,λ)\left(v,k,\lambda\right) design with kk even failing at least one of (1), (2) and (3) is called unsaturated. It is well known that the Witt system D\mathcal{D} on 24 points, that is, the 55-(24,8,1)\left(24,8,1\right) design is saturated, where C(D)C(\mathcal{D}) is the extended Golay code G24G_{24}.

The only known extremal doubly even self-dual codes of length 24ℓ24\ell are the extended Golay code G24G_{24} and the extended quadratic residue code QR48QR_{48} of length 48. In addition, it is known that G24G_{24} (resp. QR48QR_{48}) is the unique extremal doubly even self-dual code of length 24 (resp. 48). The existence of an extremal doubly even self-dual code of length 72 is a long-standing open question. By the Assmus–Mattson theorem [1], the supports of the codewords of minimum weight in an extremal doubly even self-dual code of length 24ℓ24\ell form a self-orthogonal 55-(24ℓ,4ℓ+4,(5ℓ−2ℓ−1))\left(24\ell,4\ell+4,\binom{5\ell-2}{\ell-1}\right) design [15]. As described above, it is well known that the Witt system on 24 points is saturated. For ℓ=2\ell=2, 3 and 4, every self-orthogonal 55-(24ℓ,4ℓ+4,(5ℓ−2ℓ−1))\left(24\ell,4\ell+4,\binom{5\ell-2}{\ell-1}\right) design is saturated [13], [11] and [10], respectively. A powerful tool used in [11] and [13] is the system of fundamental equations, sometimes called the Mendelsohn equations [8] (see also [16] and [19]).

The purpose of this paper is to begin a systematic study of the family of saturated self-orthogonal designs. We show that self-orthogonal designs with certain parameters are saturated, by proving the nonexistence of unsaturated designs with the same parameters. As a typical example, we have the following:

Theorem A. For ℓ=1\ell=1, 2 and 3, every self-orthogonal (ℓ+2)(\ell+2)-(24ℓ,4ℓ+4,λ)\left(24\ell,4\ell+4,\lambda\right) design is a saturated 55-(24ℓ,4ℓ+4,(5ℓ−2ℓ−1))\left(24\ell,4\ell+4,\binom{5\ell-2}{\ell-1}\right) design.

Remark 1.1. (i) Every self-orthogonal 33-(24,8,λ)\left(24,8,\lambda\right) design is isomorphic to the Witt system on 2424 points. Every self-orthogonal 44-(48,12,λ)\left(48,12,\lambda\right) design is isomorphic to the unique self-orthogonal 55-(48,12,8)\left(48,12,8\right) design.

(ii) The divisibility condition shows that none of the 55-(24ℓ,4ℓ+4,5ℓ−2ℓ−1)\left(24\ell,4\ell+4,\frac{5\ell-2}{\ell-1}\right) designs for ℓ=1,2\ell=1,2 and 33 is a 66-design. We refer to the proof of Theorem 5 in [2] for general ℓ\ell.

In Table 1, we give some necessary notation for codes, which is used throughout this paper.

NotationCodes
e7e_7the Hamming [7,4,3][7,4,3] code
e8e_8the extended Hamming [8,4,4][8,4,4] code
R(1,4)R(1,4)the first order Reed–Muller [16,5,8][16,5,8] code
C22C_{22}the unique (singly even) self-dual [22,11,6][22,11,6] code ([18])
G23G_{23}the Golay [23,12,7][23,12,7] code
G24G_{24}the extended Golay [24,12,8][24,12,8] code
C24C_{24}the unique (singly even) self-dual [24,12,6][24,12,6] code ([18])
C32,1,C32,2,…,C32,5C_{32,1}, C_{32,2}, \ldots, C_{32,5}the five doubly even self-dual [32,16,8][32,16,8] codes ([7])
C46C_{46}the unique (singly even) self-dual [46,23,10][46,23,10] code ([13])
QR47QR_{47}the quadratic residue [47,24,11][47,24,11] code
QR48QR_{48}the extended quadratic residue [48,24,12][48,24,12] code
C72C_{72}a putative doubly even self-dual [72,36,16][72,36,16] code

Table 1. Notation for some codes

Lalaude-Labayle [14] gave a classification of self-orthogonal codes of minimum weight kk whose codewords of minimum weight form a tt-design for the case t=3t=3 with k≤10k\leq10 and the case t=5t=5 with k≤18k\leq18.

Theorem 1.2 (Lalaude-Labayle [14]). Let CC be a self-orthogonal code of minimum weight kk whose codewords of minimum weight form a tt-design.

(i) If t=3t=3 and k≤10k\leq10, then CC is equivalent to one of e8e_8, R(1,4)R(1,4), (C22)0(C_{22})_0, C22C_{22}, G23⊥G_{23}^{\perp}, G24G_{24}, C32,1C_{32,1}, C32,2C_{32,2}, …\ldots, C32,5C_{32,5} and C46C_{46}, where (C22)0(C_{22})_0 is the doubly even [22,10,8][22,10,8] subcode of C22C_{22}.

(ii) If t=5t=5 and k≤18k\leq18, then CC is equivalent to one of G24G_{24}, QR48QR_{48} and C72C_{72}.

:::`

The following theorem extends the above theorem, which is one of the main results of this paper.

Theorem B. Suppose that t≥⌊k/4⌋+1t \ge\lfloor k/4 \rfloor+ 1. Let CC be a self-orthogonal code of minimum weight kk whose codewords of minimum weight form a tt-design.

(i) If t=2t = 2, then CC is equivalent to one of e7+e_7^{+}, e8e_8, the unique self-orthogonal [16,6,6][16, 6, 6] code, the unique shortened [21,10,6][21, 10, 6] code of C22C_{22} and C22C_{22}.

(ii) If t=3t = 3, then CC is equivalent to one of e8e_8, R(1,4)R(1,4), (C22)0(C_{22})_0, C22C_{22}, G23⊥G_{23}^{\perp}, G24G_{24}, C32,1C_{32,1}, C32,2C_{32,2}, …\ldots, C32,5C_{32,5} and C46C_{46}, where (C22)0(C_{22})_0 is the doubly even [22,10,8][22, 10, 8] subcode of C22C_{22}.

(iii) If t=4t = 4, then CC is equivalent to one of G23⊥G_{23}^{\perp}, G24G_{24}, QR47⊥QR_{47}^{\perp} and QR48QR_{48}.

(iv) If t=5t = 5, then CC is equivalent to one of G24G_{24}, QR48QR_{48} and C72C_{72}.

(v) If 6≤t≤116 \le t \le11, then there is no such code CC.

This paper is organized as follows. In § 2, we give definitions, notation and basic results, and derive lower bounds on the minimum weight of C(D)⊥C(D)^{\perp} using the Mendelsohn equations for a tt-(v,k,λ)(v,k,\lambda) design DD. In § 3, we examine the existence of self-orthogonal (⌊k/4⌋+1)(\lfloor k/4 \rfloor+1)-(v,k,λ)(v,k,\lambda) designs for even v≤30v \le30, k=8,10,12,14k = 8, 10, 12, 14 and k≤v/2k \le v/2. We give strong restrictions on the parameters of self-orthogonal designs and we establish the nonexistence of self-orthogonal designs for many parameters (Table 2 and Table 3). In § 4, we give possible parameters of unsaturated self-orthogonal tt-(v,k,λ)(v,k,\lambda) designs satisfying t≥⌊k/4⌋+1t \ge\lfloor k/4 \rfloor+1 for v≤10000v \le10000 and k=8,10,…,30k = 8, 10, \ldots, 30 (Table 4). This is used to show that self-orthogonal designs with certain parameters are saturated. As a typical example, we have a proof of Theorem A. In addition, the nonexistence of self-orthogonal designs is also established for many parameters. In § 5, we study self-orthogonal designs satisfying (3). We give strong restrictions on the parameters of self-orthogonal designs satisfying (3) and t≥⌊k/4⌋+1≥2t \ge\lfloor k/4 \rfloor+1 \ge2. This determines possible parameters of self-orthogonal tt-(v,k,λ)(v,k,\lambda) designs with kk even satisfying (3) for k≤40k \le40 (Proposition 5.2). By investigating self-orthogonal tt-(v,k,λ)(v,k,\lambda) designs with parameters listed in Proposition 5.2, we give a classification of self-orthogonal tt-(v,k,λ)(v,k,\lambda) designs with kk even satisfying (2), (3), t≥⌊k/4⌋+1≥2t \ge\lfloor k/4 \rfloor+ 1 \ge2 and t≤11t \le11 in § 6. Theorem B is obtained as a consequence of this classification. As the other consequence, a classification of saturated self-orthogonal designs under the same condition is also obtained (Corollary 6.4).

(t,v,k)(t,v,k)τ\tauμmax⁡\mu_{\max}d(D)d(\mathcal{D})A(v,k)A_{(v,k)}μBI\mu_{BI}
(3,16,8)(3,16,8)364198 ([17])
(3,22,8)(3,22,8)626330 ([18])1
(3,24,8)(3,24,8)2116, 8759 ([18])
(3,26,8)(3,26,8)716390 ([7])1
(3,28,8)(3,28,8)4206442 ([7])
(3,30,8)(3,30,8)606393 ([7])
(3,20,10)(3,20,10)4134512 ([17])1, 2, 3, 5, 6, 7, 9, 10, 11, 13
(3,22,10)(3,22,10)694, 6700 ([18])1, 3, 5, 7, 9
(3,24,10)(3,24,10)18004, 6, 8960 ([18])
(3,26,10)(3,26,10)3204, 61313 ([7])
(3,28,10)(3,28,10)20261624 ([7])1, 2
(3,30,10)(3,30,10)18361848 ([7])1, 3
(4,24,12)(4,24,12)1586, 82576 ([18])1, 2, ..., 7
(4,26,12)(4,26,12)495062340 ([7])
(4,28,12)(4,28,12)99063717 ([7])
(4,30,12)(4,30,12)55165320 ([7])1
(4,28,14)(4,28,14)66365020 ([7])1, 2, 3
(4,30,14)(4,30,14)143268931 ([7])

Table 2. Self-orthogonal (⌊k/4⌋+1)(\lfloor k/4 \rfloor+1)-(v,k,λ)(v,k,\lambda) designs (v≤30,k=8,10,12,14)(v\leq30,k=8,10,12,14)

(t,v,k)(t,v,k)τ\tauμmax⁡\mu_{\max}μE\mu_EμN\mu_N
(3,16,8)(3,16,8)33661,21,2 (Section 4)
(3,22,8)(3,22,8)66222211
(3,24,8)(3,24,8)21211111
(3,26,8)(3,26,8)771111
(3,20,10)(3,20,10)4413131,2,…,131,2,\ldots,13
(3,22,10)(3,22,10)669988 ([22])1,2,…,7,91,2,\ldots,7,9
(3,26,10)(3,26,10)332020
(3,28,10)(3,28,10)2020221,21,2
(3,30,10)(3,30,10)1818331,2,31,2,3 (Section 4)
(4,24,12)(4,24,12)151588881,2,…,71,2,\ldots,7
(4,30,12)(4,30,12)55551111
(4,28,14)(4,28,14)6666331,2,31,2,3
(4,30,14)(4,30,14)143143221,21,2 (Section 4)

Table 3. Self-orthogonal (⌊k/4⌋+1)(\lfloor k/4\rfloor+1)-(v,k,λ)(v,k,\lambda) designs (v≤30,k=8,10,12,14)(v\le30,k=8,10,12,14)

kk(v,m)(v,m)
8(16, 4), (16, 6), (22, 6), (22, 7), (23, 7)
10(20, 4), (22, 6), (26, 6)
12(23, 7), (23, 8), (24, 8), (36, 8), (47, 11), (68, 16), (156, 36), (311, 71)
14(27, 7), (80, 16), (159, 31)
16(22, 3), (24, 8), (27, 4), (32, 8), (43, 8), (43, 11), (48, 12), (56, 12), (58, 13)
18(36, 8)
20(39, 11)
22(43, 11)
24(48, 12), (72, 16), (96, 20)
26(52, 12)
28(55, 15)
30(59, 15)

Table 4. Solutions (v,m)(v,m) of (11) (k=8,10,…,30)(k=8,10,\ldots,30)

Preliminaries

In this section, we give definitions, notation and basic results used throughout this paper.

Self-orthogonal codes

A (binary) [n,k][n,k] code CC is a kk-dimensional vector subspace of F2n\mathbb{F}_2^n, where F2\mathbb{F}_2 denotes the finite field of order 2. Throughout this paper, all codes are assumed to be binary. The parameters nn and kk are called the length and dimension of CC, respectively. The support supp⁡(x)\operatorname{supp}(x) of a vector x=(x1,x2,…,xn)∈F2nx=(x_1,x_2,\ldots,x_n)\in\mathbb{F}_2^n is {i∣xi≠0}\{i\mid x_i\ne0\} and the weight wt⁡(x)\operatorname{wt}(x) of xx is ∣supp⁡(x)∣|\operatorname{supp}(x)|. A vector of CC is called a codeword. The minimum nonzero weight of all codewords in CC is called the minimum weight of CC. An [n,k,d][n,k,d] code is an [n,k][n,k] code with minimum weight dd. The weight enumerator of CC is given by the polynomial ∑c∈Cywt⁡(c)\sum_{c\in C}y^{\operatorname{wt}(c)}. Two codes are equivalent if one can be obtained from the other by permuting the coordinates.

The dual code C⊥C^\perp of a code CC of length nn is defined as

C⊥={x∈F2n∣x⋅y=0 for all y∈C}.C^\perp=\{x\in\mathbb{F}_2^n\mid x\cdot y=0\text{ for all }y\in C\}.

where x⋅yx\cdot y is the standard inner product. A code CC is called self-orthogonal if C⊆C⊥C\subseteq C^\perp, and CC is called self-dual if C=C⊥C=C^\perp. A code is called doubly even if all codewords have weight divisible by 4. A doubly even code is self-orthogonal. It is known that there is a doubly even self-dual code of length nn if and only if n≡0(mod8)n\equiv0\pmod{8}. A self-dual code is called singly even if there is at least one codeword of weight ≡2(mod4)\equiv2\pmod{4}.

Let D\mathcal{D} be a self-orthogonal tt-(v,k,λ)(v,k,\lambda) design with kk even. Let C(D)C(\mathcal{D}) denote the code generated by the rows of an incidence matrix of D\mathcal{D}. Then C(D)C(\mathcal{D}) is self-orthogonal. In addition, if k≡0(mod4)k\equiv0\pmod{4}, then C(D)C(\mathcal{D}) is doubly even.

A binomial identity

Although the following lemma may be known, we give a proof for completeness.

Lemma 2.1. Let ss and jj be nonnegative integers. Then

∑i=0s(−2)i(2s−is)(2j+1i)=(−4)s(js).\sum_{i=0}^{s}(-2)^i\binom{2s-i}{s}\binom{2j+1}{i}=(-4)^s\binom{j}{s}.

Proof. The assertion is immediate for s=0s=0. We proceed by induction on jj. If j=0j=0, both sides are equal to δs,0\delta_{s,0}. Suppose that j≥1j \ge1 and that the assertion holds for j−1j-1. For s≥1s \ge1, Pascal’s identity gives

(2s−i−1s)−(2s−i−2s)=(2s−i−2s−1)(0≤i≤s−1).\binom{2s-i-1}{s}-\binom{2s-i-2}{s}=\binom{2s-i-2}{s-1}\qquad(0\le i\le s-1).

Using this identity and the induction hypothesis, we obtain

∑i=0s(−2)i(2s−is)(2j+1i)=∑i=0s(−2)i(2s−is)((2j−1i)+2(2j−1i−1)+(2j−1i−2))=(−4)s(j−1s)−4∑i=0s−1(−2)i(2s−i−1s)(2j−1i)+4∑i=0s−2(−2)i(2s−i−2s)(2j−1i)=(−4)s(j−1s)−4∑i=0s−1(−2)i((2s−i−1s)−(2s−i−2s))(2j−1i)=(−4)s(j−1s)−4∑i=0s−1(−2)i(2(s−1)−is−1)(2j−1i)=(−4)s((j−1s)+(j−1s−1))=(−4)s(js).\begin{aligned} \sum_{i=0}^{s}(-2)^{i}\binom{2s-i}{s}\binom{2j+1}{i} &=\sum_{i=0}^{s}(-2)^{i}\binom{2s-i}{s}\left(\binom{2j-1}{i}+2\binom{2j-1}{i-1}+\binom{2j-1}{i-2}\right)\\ &=(-4)^{s}\binom{j-1}{s}-4\sum_{i=0}^{s-1}(-2)^{i}\binom{2s-i-1}{s}\binom{2j-1}{i}\\ &\quad+4\sum_{i=0}^{s-2}(-2)^{i}\binom{2s-i-2}{s}\binom{2j-1}{i}\\ &=(-4)^{s}\binom{j-1}{s}-4\sum_{i=0}^{s-1}(-2)^{i}\left(\binom{2s-i-1}{s}-\binom{2s-i-2}{s}\right)\binom{2j-1}{i}\\ &=(-4)^{s}\binom{j-1}{s}-4\sum_{i=0}^{s-1}(-2)^{i}\binom{2(s-1)-i}{s-1}\binom{2j-1}{i}\\ &=(-4)^{s}\left(\binom{j-1}{s}+\binom{j-1}{s-1}\right)\\ &=(-4)^{s}\binom{j}{s}. \end{aligned}

This completes the proof. □\square

By putting t=s+1t=s+1, the above lemma gives the following identity:

∑i=1ti(−2)i−1(2t−i−1t−1)(2ji)=0(j=1,2,…,t−1),(4)\sum_{i=1}^{t}i(-2)^{i-1}\binom{2t-i-1}{t-1}\binom{2j}{i}=0\quad(j=1,2,\ldots,t-1), \tag*{(4)}

which is used in the proof of Lemma 2.2.

Mendelsohn equations

Let D=(P,B)\mathcal{D}=(\mathcal{P},\mathcal{B}) be a tt-(v,k,λ)(v,k,\lambda) design. It follows that every ii-element subset of points (i≤ti\le t) is contained in exactly λi\lambda_i blocks, where

λi=λ(v−it−i)(k−it−i)=λ(v−i)⋯(v−t+1)(k−i)⋯(k−t+1).(5)\lambda_i=\lambda\frac{\binom{v-i}{t-i}}{\binom{k-i}{t-i}}=\lambda\frac{(v-i)\cdots(v-t+1)}{(k-i)\cdots(k-t+1)}. \tag*{(5)}

The number λ1\lambda_1 is traditionally denoted by rr, and the total number of blocks is b=λ0b=\lambda_0.

Assume that kk is even. We denote the code generated by the rows of an incidence matrix of D\mathcal{D} by C(D)C(\mathcal{D}). Let xx be a vector of C(D)⊥C(\mathcal{D})^\perp and let SS be the support. Suppose that wt⁡(x)=m\operatorname{wt}(x)=m. Counting the number of pairs (I,B)(I,B) with I⊆S∩BI\subseteq S\cap B, ∣I∣=i|I|=i, B∈BB\in\mathcal{B}, we have

∑B∈B(∣B∩S∣i)=λi(mi)(i=0,1,…,t).(6)\sum_{B\in\mathcal{B}}\binom{|B\cap S|}{i}=\lambda_i\binom{m}{i}\quad(i=0,1,\ldots,t). \tag*{(6)}

Let

nj(x)=∣{B∈B∣2j=∣B∩S∣}∣(j=0,1,…).n_j(x)=|\{B\in\mathcal{B}\mid2j=|B\cap S|\}|\quad(j=0,1,\ldots).

Note that we often simply denote it by njn_j. Then we have the following system of equations:

∑j=0⌊m/2⌋(2ji)nj=λi(mi)(i=0,1,…,t).(7)\sum_{j=0}^{\lfloor m/2\rfloor}\binom{2j}{i}n_j=\lambda_i\binom{m}{i}\quad(i=0,1,\ldots,t). \tag*{(7)}

These fundamental equations, which are sometimes called Mendelsohn equations [8] (see also [16] and [19]), are a powerful tool used throughout this paper.

The following lemma gives a necessary condition for the solvability of systems of the same form as the Mendelsohn equations (7).

Lemma 2.2. Let b1,b2,…,btb_1,b_2,\ldots,b_t be nonnegative integers. If the following system of equations

∑j=1t−1(2ji)xj=bi(i=1,2,…,t)\sum_{j=1}^{t-1}\binom{2j}{i}x_j=b_i\quad(i=1,2,\ldots,t)

has a solution (x1,x2,…,xt−1)(x_1,x_2,\ldots,x_{t-1}), then

∑i=1tkibi=0,\sum_{i=1}^{t}k_i b_i=0,

where

ki=i(−2)i−1(2t−i−1t−1).(8)k_i=i(-2)^{i-1}\binom{2t-i-1}{t-1}. \tag*{(8)}

Proof. Using the given equations and (4), we have

∑i=1tkibi=∑i=1tki∑j=1t−1(2ji)xj=∑j=1t−1xj∑i=1tki(2ji)=0.\sum_{i=1}^{t} k_i b_i=\sum_{i=1}^{t}k_i\sum_{j=1}^{t-1}\binom{2j}{i}x_j=\sum_{j=1}^{t-1}x_j\sum_{i=1}^{t}k_i\binom{2j}{i}=0.

The result follows.

Bounds on minimum weights of C(D)⊥C(\mathcal{D})^\perp

Throughout this subsection, let D\mathcal{D} be a tt-(v,k,λ)(v,k,\lambda) design with kk even. We give bounds on the minimum weight d(C(D)⊥)d(C(\mathcal{D})^\perp) of the dual code of C(D)C(\mathcal{D}). The assertions (i) and (ii) in the following lemma were first proved in [19] for t=2t=2 and in [13] for t=3t=3, respectively. Our proof is similar to those in [13] and [19]$.

Lemma 2.3. (i) If tt is even with t≥2t\geq2, then d(C(D)⊥)≥(t−1)+λt−1/λd(C(\mathcal{D})^\perp)\geq(t-1)+\lambda_{t-1}/\lambda.

(ii) If tt is odd with t≥3t\geq3, then d(C(D)⊥)>(t−2)+λt−2/λt−1d(C(\mathcal{D})^\perp)> (t-2)+\lambda_{t-2}/\lambda_{t-1}.

Proof. The system of equations (7) has a solution if we take m=d(C(D)⊥)m=d(C(\mathcal{D})^\perp). It follows from [13 (3.2)] that

∑j=⌈t+12⌉⌊m2⌋nj>0.(9)\sum_{j=\lceil\frac{t+1}{2}\rceil}^{\lfloor\frac{m}{2}\rfloor} n_j>0. \tag*{(9)}

Suppose that t=2et=2e, where ee is a positive integer. From (9), we have m≥2em\geq2e. Hence, the Mendelsohn equations (7) yield the following:

∑j=e⌊m2⌋2j(2j−1)⋯(2j−(2e−2))nj=m(m−1)⋯(m−(2e−2))λ2e−1,\sum_{j=e}^{\lfloor\frac{m}{2}\rfloor}2j(2j-1)\cdots(2j-(2e-2))n_j=m(m-1)\cdots(m-(2e-2))\lambda_{2e-1},
∑j=e⌊m2⌋2j(2j−1)⋯(2j−(2e−1))nj=m(m−1)⋯(m−(2e−1))λ.\sum_{j=e}^{\lfloor\frac{m}{2}\rfloor}2j(2j-1)\cdots(2j-(2e-1))n_j=m(m-1)\cdots(m-(2e-1))\lambda.

Hence, we have

m(m−1)⋯(m−(2e−2))((m−(2e−1))λ−λ2e−1)=∑j=e+1⌊m/2⌋2j(2j−1)⋯(2j−(2e−2))(2j−2e)nj≥0,\begin{aligned} &m(m-1)\cdots(m-(2e-2))((m-(2e-1))\lambda-\lambda_{2e-1})\\ &\quad=\sum_{j=e+1}^{\lfloor m/2\rfloor}2j(2j-1)\cdots(2j-(2e-2))(2j-2e)n_j\geq0, \end{aligned}

where equality holds if and only if ∑j=e+1⌊m/2⌋nj=0\sum_{j=e+1}^{\lfloor m/2\rfloor}n_j=0. Thus, m≥(2e−1)+λ2e−1/λm\geq(2e-1)+\lambda_{2e-1}/\lambda. This completes the proof of the assertion (i).

Now suppose that t=2e+1t=2e+1, where ee is a positive integer. Since a tt-(v,k,λ)(v,k,\lambda) design is a (t−1)(t-1)-(v,k,λt−1)(v,k,\lambda_{t-1}) design, the above discussion given in the proof of the assertion (i) can be applied. Thus, it follows that d(C(D)⊥)≥(t−2)+λt−2/λt−1d(C(D)^\perp)\geq(t-2)+\lambda_{t-2}/\lambda_{t-1}, where equality holds if and only if ∑j=e+1⌊m/2⌋nj=0\sum_{j=e+1}^{\lfloor m/2\rfloor}n_j=0. It follows from (9) that ∑j=e+1⌊m/2⌋nj>0\sum_{j=e+1}^{\lfloor m/2\rfloor}n_j>0. The assertion (ii) follows.

Self-orthogonal tt-(v,k,λ)(v,k,\lambda) designs with vv even

Throughout this section, we assume that vv is even with v≤30v\leq30. We examine the existence of self-orthogonal (⌊k/4⌋+1)(\lfloor k/4\rfloor+1)-(v,k,λ)(v,k,\lambda) designs for k=8,10,12,14k=8,10,12,14 and k≤v/2k\leq v/2.

Lemma 3.1. Let DD be a self-orthogonal 3-(v,k,λ)(v,k,\lambda) design such that kk is even and k≥4k\geq4. If there is a vector of weight 4 in C(D)⊥C(D)^\perp, then v=2kv=2k.

Proof. Let xx be a vector of weight 4 in C(D)⊥C(D)^\perp. From the Mendelsohn equations (7), we have

2n1+4n2=4λ(v−1)(v−2)(k−1)(k−2),2n_1+4n_2=4\lambda\frac{(v-1)(v-2)}{(k-1)(k-2)},
n1+6n2=6λv−2k−2n_1+6n_2=6\lambda\frac{v-2}{k-2}

and

4n2=4λ.4n_2=4\lambda.

This implies that (v−2k)(v−k)=0(v-2k)(v-k)=0. The result follows.

Let dmax⁡(v)d_{\max}(v) denote the largest minimum weight among all self-dual codes of length vv (see [6 Table I] for dmax⁡(v)d_{\max}(v) (v≤30v\leq30)).

Proposition 3.2. There is no self-orthogonal 3-(v,k,λ)(v,k,\lambda) design for (v,k)=(18,8)(v,k)=(18,8) and (20,8)(20,8).

Proof. Suppose that D\mathcal{D} is a self-orthogonal 3-(v,k,λ)(v,k,\lambda) design with (v,k)=(18,8)(v,k)=(18,8) or (20,8)(20,8). By Lemma 2.3 (ii) and dmax⁡(18)=dmax⁡(20)=4d_{\max}(18)=d_{\max}(20)=4, C(D)C(\mathcal{D}) is contained in some self-dual code with parameters [18,9,4][18,9,4] and [20,10,4][20,10,4], respectively. Thus, there is a vector of weight 4 in C(D)⊥C(\mathcal{D})^{\perp}. The result follows from Lemma 3.1.

Let D=(X,B)\mathcal{D}=(X,\mathcal{B}) be a self-orthogonal tt-(v,k,λ)(v,k,\lambda) design with kk even. Then C(D)C(\mathcal{D}) is a self-orthogonal code containing at least ∣B∣|\mathcal{B}| codewords of weight kk. If vv is even, then C(D)C(\mathcal{D}) is contained in some self-dual [v,v/2,d][v,v/2,d] code containing at least ∣B∣|\mathcal{B}| codewords of weight kk. The possible minimum weights dd are determined by Lemma 2.3 and dmax⁡(v)d_{\max}(v). We denote by d(D)d(\mathcal{D}) the set of the possible minimum weights dd. Let A(v,k,d)A_{(v,k,d)} denote the largest number of codewords of weight kk among self-dual [v,v/2,d][v,v/2,d] codes. Let

A(v,k)=max⁡{A(v,k,d)∣d∈d(D)}.A_{(v,k)}=\max\{A_{(v,k,d)}\mid d\in d(\mathcal{D})\}.

Since ∣B∣≤A(v,k)|\mathcal{B}|\le A_{(v,k)}, we have λ≤⌊A(v,k)(vk)(tk)⌋\lambda\le\left\lfloor A_{(v,k)}\frac{\binom{v}{k}}{\binom{t}{k}}\right\rfloor. Let τ\tau be the least positive integer such that

τ(v−it−i)(k−it−i)\tau\frac{\binom{v-i}{t-i}}{\binom{k-i}{t-i}}

is an integer for every i=0,1,…,ti=0,1,\ldots,t. Then (5) implies that τ\tau divides λ\lambda. Let

μmax⁡=⌊A(v,k)(vt)(kt)τ⌋.\mu_{\max}=\left\lfloor\frac{A_{(v,k)}\binom{v}{t}}{\binom{k}{t}\tau}\right\rfloor.

Of course, if there is a self-orthogonal tt-(v,k,μτ)(v,k,\mu\tau) design, then μ≤μmax⁡\mu\le\mu_{\max}. In Table 2, we list τ\tau, μmax⁡\mu_{\max} and d(D)d(\mathcal{D}) for v≤30v\le30, k=8,10,12,14k=8,10,12,14 and k≤v/2k\le v/2, except (v,k)=(18,8)(v,k)=(18,8) and (20,8)(20,8). We also list A(v,k)A_{(v,k)} along with references. The column “μBI\mu_{BI}” is explained below.

Let D=(X,B)\mathcal{D}=(X,\mathcal{B}) be a self-orthogonal tt-(v,k,λ)(v,k,\lambda) design with kk even. Let BB be a block of D\mathcal{D}. Let

m2j=∣{B′∈B∖{B}∣2j=∣B∩B′∣}∣(j=0,1,…,⌊k2⌋−1).m_{2j}=\left|\{B'\in\mathcal{B}\setminus\{B\}\mid2j=|B\cap B'|\}\right|\quad\left(j=0,1,\ldots,\left\lfloor\frac{k}{2}\right\rfloor-1\right).

From the Mendelsohn equations (7), we have the following system of equations:

∑i=0⌊k/2⌋−1(2ij)m2i=(λj−1)(kj)(j=0,1,…,t).(10)\sum_{i=0}^{\left\lfloor k/2\right\rfloor-1}\binom{2i}{j}m_{2i}=(\lambda_j-1)\binom{k}{j}\quad(j=0,1,\ldots,t). \tag*{(10)}

Since each m2im_{2i} counts blocks, (10) must have a solution in nonnegative integers. For each (t,v,k)(t,v,k) in Table 2, we consider λ=τμ\lambda=\tau\mu with 1≤μ≤μmax⁡1\le\mu\le\mu_{\max}. The column “μBI\mu_{BI}” lists the values of μ\mu for which (10) has no solution in nonnegative integers. Thus, no self-orthogonal tt-(v,k,τμ)(v,k,\tau\mu) design exists for any value of μ\mu listed in this column. For example, when (t,v,k)=(3,22,8)(t,v,k)=(3,22,8), we have τ=6\tau=6 and μmax⁡=2\mu_{\max}=2. From (10) we have m6=2μ−4m_6=2\mu-4, so μ=1\mu=1 is excluded by the condition m6≥0m_6\ge0. Hence, the only possible λ\lambda is 12.

Proposition 3.3. There is no self-orthogonal tt-(v,k,λ)(v,k,\lambda) design for

(t,v,k)=(3,28,8),(3,30,8),(3,24,10),(4,26,12) and (4,28,12).(t,v,k)=(3,28,8),(3,30,8),(3,24,10),(4,26,12)\text{ and }(4,28,12).

Proof. For each of the listed parameter sets, Table 2 gives μmax⁡=0\mu_{\max}=0. However, the existence of such a design requires λ=τμ\lambda=\tau\mu for a positive integer μ\mu satisfying μ≤μmax⁡\mu\le\mu_{\max}, a contradiction.

Proposition 3.4. There is no self-orthogonal tt-(v,k,λ)(v,k,\lambda) design for (t,v,k)=(3,26,8),(3,28,10),(4,30,12)(t,v,k)=(3,26,8),(3,28,10),(4,30,12) and (4,28,14)(4,28,14).

Proof. For each of the listed parameter sets, Table 2 shows that (10) has no nonnegative integer solution for any λ=τμ\lambda=\tau\mu with 1≤μ≤μmax⁡1\le\mu\le\mu_{\max}. Since these are all possible values of λ\lambda, the result follows.

Proposition 3.5. (i) There is a self-orthogonal 33-(22,8,λ)(22,8,\lambda) design if and only if λ=12\lambda=12.

(ii) There is a self-orthogonal 33-(24,8,λ)(24,8,\lambda) design if and only if λ=21\lambda=21.

(iii) There is a self-orthogonal 44-(24,12,λ)(24,12,\lambda) design if and only if λ=120\lambda=120.

In each case, the design is unique, up to isomorphism.

Proof. From Table 2, the only possible values of λ\lambda are 1212, 2121 and 120120 in (i), (ii) and (iii), respectively. The existence and uniqueness up to isomorphism of the designs with these parameters follow from [19], Remark 1.1 and [22], respectively.

Proposition 3.6. There is a self-orthogonal 33-(22,10,λ)(22,10,\lambda) design if and only if λ=48\lambda=48. The design is unique, up to isomorphism.

Proof. Up to isomorphism, there is a unique self-orthogonal 33-(22,10,48)(22,10,48) design [22]. From Table 2, it is sufficient to show that there is no self-orthogonal 33-(22,10,6μ)(22,10,6\mu) design for μ∈{2,4,6}\mu\in\{2,4,6\}.

Suppose that D\mathcal{D} is a self-orthogonal 33-(22,10,6μ)(22,10,6\mu) design with μ∈{2,4,6}\mu\in\{2,4,6\}. By Lemma 2.3 (ii) and Lemma 3.1, C(D)⊥C(\mathcal{D})^\perp contains no vectors of weight 2 or 4. Since 1∈C(D)⊥\mathbf{1}\in C(\mathcal{D})^\perp, two disjoint blocks yield a vector of weight 2 in C(D)⊥C(\mathcal{D})^\perp, where 1\mathbf{1} denotes the all-one vector. Similarly, two blocks intersecting in 8 points yield a vector of weight 4. Hence, we have m0=m8=0m_0=m_8=0 in (10). Multiplying these equations in (10) for j=0,1,2,3j=0,1,2,3 by −16-16, 1111, −6-6, 22, respectively, and adding, we obtain

16(m8−m0)=8μ−64.16(m_8-m_0)=8\mu-64.

Thus, μ=8\mu=8, a contradiction.

Proposition 3.7. There is no self-orthogonal 33-(20,10,λ)(20, 10, \lambda) design for any positive integer λ\lambda.

Proof. Let D=(X,B)\mathcal{D}=(X,\mathcal{B}) be a self-orthogonal 33-(20,10,λ)(20,10,\lambda) design. By Lemma 2.3 (ii) and dmax⁡(20)=4d_{\max}(20)=4, C(D)C(\mathcal{D}) is contained in a self-dual [20,10,4][20,10,4] code CC. Up to equivalence, there are 77 such codes [17].

We verified by Magma [4] that only one of these codes has the property that every 33-subset of the coordinate set is contained in the support of a codeword of weight 1010. This is the unique self-dual [20,10,4][20,10,4] code with weight enumerator

1+5y4+80y6+250y8+352y10+250y12+80y14+5y16+y20.1+5y^{4}+80y^{6}+250y^{8}+352y^{10}+250y^{12}+80y^{14}+5y^{16}+y^{20}.

Since λ>0\lambda>0, CC must be this code, up to equivalence. We also verified that some 33-subset is contained in exactly 3232 supports of codewords of weight 1010. Hence, λ≤32\lambda\leq32 and ∣B∣=19λ/2≤304|\mathcal{B}|=19\lambda/2\leq304.

Choose a codeword y∈Cy\in C of weight 1010 with supp⁡(y)∉B\operatorname{supp}(y)\notin\mathcal{B}. Since C⊆C(D)⊥C\subseteq C(\mathcal{D})^{\perp}, the Mendelsohn equations (7) apply to yy and give

n4(y)=n0(y)−4n5(y)+3λ16=n0(y)+3λ16.n_{4}(y)=n_{0}(y)-4n_{5}(y)+\frac{3\lambda}{16}=n_{0}(y)+\frac{3\lambda}{16}.

Adding yy to the incidence vectors of the blocks counted by n4(y)n_{4}(y) gives distinct codewords of weight 44 in CC. Thus, we have n4(y)≤5n_{4}(y)\leq5. Since n0(y)n_{0}(y) and n4(y)n_{4}(y) are nonnegative integers, we have

16∣λand0<3λ16≤5.16\mid\lambda\quad\text{and}\quad0<\frac{3\lambda}{16}\leq5.

Therefore, we have λ=16\lambda=16, that is, D\mathcal{D} is a 33-(20,10,16)(20,10,16) design.

We now show that no such design exists. For a fixed block B∈BB\in\mathcal{B}, (10) gives

m2=4−4m0andm8=m0−1.m_{2}=4-4m_{0}\quad\text{and}\quad m_{8}=m_{0}-1.

Since m0∈{0,1}m_{0}\in\{0,1\} and m8≥0m_{8}\geq0, we have m0=1m_{0}=1 and m2=m8=0m_{2}=m_{8}=0. Thus, B\mathcal{B} is closed under complementation and ∣B∩B′∣∈{0,4,6}|B\cap B'|\in\{0,4,6\} for any distinct B,B′∈BB,B'\in\mathcal{B}. Let Γ\Gamma be the graph with vertex set

V(Γ)={{x,1+x}∣x∈C, wt⁡(x)=10},V(\Gamma)=\bigl\{\{x,1+x\}\mid x\in C,\ \operatorname{wt}(x)=10\bigr\},

where two distinct vertices {x,1+x}\{x,1+x\} and {y,1+y}\{y,1+y\} are adjacent if wt⁡(x+y)∈{4,16}\operatorname{wt}(x+y)\in\{4,16\}. This definition is independent of the choice of representatives, since replacing either representative by its complement interchanges weights 4 and 16. The following set

{{x,1+x}∣x∈C, supp⁡(x)∈B}\{\{x,1+x\}\mid x\in C,\ \operatorname{supp}(x)\in B\}

is an independent set of Γ\Gamma of size 152/2=76152/2=76. However, we verified by Magma [4] that Γ\Gamma has independence number 70, a contradiction.

For (t,v,k)(t,v,k) given in Table 2 (except the cases with μmax⁡=0\mu_{\max}=0), we summarize in Table 3 the existence and nonexistence results discussed in this paper for self-orthogonal tt-(v,k,μτ)(v,k,\mu\tau) designs. Some of these results are established in § 4, as indicated in the table. In the column “μE\mu_E”, μ\mu indicates that the existence of a self-orthogonal tt-(v,k,μτ)(v,k,\mu\tau) design is known, and in the column “μN\mu_N”, μ\mu indicates that the nonexistence of a self-orthogonal tt-(v,k,μτ)(v,k,\mu\tau) design is known.

Unsaturated self-orthogonal designs

In this section, in order to show that self-orthogonal designs with certain parameters are saturated, we establish the nonexistence of unsaturated designs with the parameters.

Lemma 4.1. Let D\mathcal{D} be a self-orthogonal tt-(v,k,λ)(v,k,\lambda) design with kk even. Suppose that D\mathcal{D} is unsaturated. Then there is a nonzero vector u∈C(D)⊥u \in C(\mathcal{D})^\perp such that ∣B∩S∣≤k/2|B \cap S| \le k/2 for every block BB, where S=supp⁡(u)S = \operatorname{supp}(u).

Proof. If C(D)≠C(D)⊥C(\mathcal{D}) \ne C(\mathcal{D})^\perp, then choose a vector uu of minimum weight in a nontrivial coset of C(D)C(\mathcal{D}) in C(D)⊥C(\mathcal{D})^\perp. Otherwise, since D\mathcal{D} is unsaturated, (2) or (3) fails. Hence, there is a nonzero codeword uu of minimum weight in C(D)C(\mathcal{D}) whose support is not a block.

In either case, uu is a nonzero vector and supp⁡(u)\operatorname{supp}(u) is not a block. For any block BB, let χB\chi_B denote its incidence vector. Then u+χBu+\chi_B is nonzero and lies in the same coset of C(D)C(\mathcal{D}) as uu. By the choice of uu, we have

wt⁡(u)≤wt⁡(u+χB)=wt⁡(u)+k−2∣B∩supp⁡(u)∣.\operatorname{wt}(u) \le\operatorname{wt}(u+\chi_B) = \operatorname{wt}(u)+k-2|B \cap\operatorname{supp}(u)|.

Thus, we conclude that ∣B∩supp⁡(u)∣≤k/2|B \cap\operatorname{supp}(u)| \le k/2.

Proposition 4.2. Let D\mathcal{D} be a self-orthogonal tt-(v,k,λ)(v,k,\lambda) design with kk even. Suppose that D\mathcal{D} is unsaturated. Let t0=⌊k4⌋+1t_0 = \left\lfloor\frac{k}{4}\right\rfloor+1 and let m=∣S∣m = |S|, where SS is given as in Lemma 4.1. Assume t≥t0t \ge t_0. Then

∑i=1t0i(−2)i−1(2t0−i−1t0−1)(mi)∏j=it0−1v−jk−j=0.(11)\sum_{i=1}^{t_0} i(-2)^{i-1}\binom{2t_0-i-1}{t_0-1}\binom{m}{i}\prod_{j=i}^{t_0-1}\frac{v-j}{k-j}=0. \tag*{(11)}

Proof. Since SS is the support of a vector in C(D)⊥C(\mathcal{D})^\perp, each block meets SS in an even number of points. By Lemma 4.1, this number is at most k/2k/2. Hence, the Mendelsohn equations (7) give

∑j=1t0−1(2ji)nj=λi(mi)(i=1,2,…,t0).(12)\sum_{j=1}^{t_0-1}\binom{2j}{i}n_j=\lambda_i\binom{m}{i}\qquad(i=1,2,\ldots,t_0). \tag*{(12)}

Applying Lemma 2.2 to (12) with t=t0t=t_0 and bi=λi(mi)b_i=\lambda_i\binom{m}{i}, we obtain

∑i=1t0i(−2)i−1(2t0−i−1t0−1)λi(mi)=0.\sum_{i=1}^{t_0}i(-2)^{i-1}\binom{2t_0-i-1}{t_0-1}\lambda_i\binom{m}{i}=0.

From (5), we have

λi=λt0∏j=it0−1v−jk−j(i=1,2,…,t0).\lambda_i=\lambda_{t_0}\prod_{j=i}^{t_0-1}\frac{v-j}{k-j}\qquad(i=1,2,\ldots,t_0).

Substituting these expressions into the sum above and dividing by λt0>0\lambda_{t_0}>0 gives (11).

For k=4k=4 or 66, (11) reduces to

2m(v−(k−1)m+(k−2))k−1=0.\frac{2m\left(v-(k-1)m+(k-2)\right)}{k-1}=0.

Under the conditions k<vk<v and 0<m<v0<m<v, the integer solutions are

(v,m)=((k−1)m−(k−2),m)(m=3,4,…).(13)(v,m)=((k-1)m-(k-2),m)\qquad(m=3,4,\ldots). \tag*{(13)}

For k=8,10,…,30k=8,10,\ldots,30, all integer solutions of (11) satisfying k<v≤10000k<v\le10000 and 0<m<v0<m<v are listed in Table 4. This was found by Magma [4].

We are now in a position to give a proof of Theorem A.

Proof of Theorem A. Suppose that (v,k)=(24,8)(v,k)=(24,8), (48,12)(48,12) or (72,16)(72,16). Since these (v,k)(v,k) are not in Table 4, a self-orthogonal (⌊k/4⌋+1)(\lfloor k/4\rfloor+1)-(v,k,λ)(v,k,\lambda) design DD is saturated by Proposition 4.2. This means that C(D)C(D) is an extremal doubly even self-dual code of length vv. Hence, by (2) and the Assmus–Mattson theorem [1], DD is a 55-(v,k,5(v/24)−2(v/24)−1)(v,k,\frac{5^{(v/24)-2}}{(v/24)-1}) design. This completes a proof of Theorem A. □

We also apply Proposition 4.2 to have the following:

Proposition 4.3. There is no self-orthogonal tt-(v,k,λ)(v,k,\lambda) design for (t,v,k)=(3,30,10)(t,v,k)=(3,30,10) and (4,30,14)(4,30,14).

Proof. Suppose that D\mathcal{D} is a self-orthogonal tt-(v,k,λ)(v,k,\lambda) design with (t,v,k)=(3,30,10)(t,v,k)=(3,30,10) or (4,30,14)(4,30,14). If D\mathcal{D} is unsaturated, then Proposition 4.2 gives an integer solution of (11) with 0<m<v0<m<v. However, Table 4 contains no such solution for (v,k)=(30,10)(v,k)=(30,10) or (30,14)(30,14). Thus, D\mathcal{D} is saturated. Hence, C(D)C(\mathcal{D}) is a self-dual code of length 30 and minimum weight k∈{10,14}k\in\{10,14\}. This contradicts dmax⁡(30)=6d_{\max}(30)=6. □\square

Next we consider the parameter pairs (v,k)(v,k) given in Table 4. For the cases with even v≤36v\le36 not included in Table 2, the corresponding bounds are given in Table 5. Note that the numbers A(v,k)A_{(v,k)} are known for v≤36v\le36. The column “μBI\mu_{BI}” lists the values of μ\mu for which the system (10) with λ=τμ\lambda=\tau\mu has no solution in nonnegative integers. The column “μU\mu_U” lists the values of μ\mu for which the existence of a self-orthogonal tt-(v,k,τμ)(v,k,\tau\mu) design is unknown. From Table 5, we have the following:

(⌊k/4⌋+1)(\lfloor k/4\rfloor+1)-(v,k,λ)(v,k,\lambda)τ\tauμmax⁡\mu_{\max}d(D)d(\mathcal{D})A(v,k)A_{(v,k)}μBI\mu_{BI}μU\mu_U
(4,36,12)(4,36,12)155810387 ([12])
(5,22,16)(5,22,16)7280677 ([18])
(5,24,16)(5,24,16)7816, 8759 ([18])
(5,32,16)(5,32,16)39206, 836518 ([3], [7])1, 2, …\ldots, 20
(5,36,18)(5,36,18)35466, 871232 ([12])1, 2, …\ldots, 31; 33, …\ldots, 4632

Table 5. (⌊k/4⌋+1)(\lfloor k/4\rfloor+1)-(v,k,λ)(v,k,\lambda) designs with parameters in Table 4 (vv even)

Proposition 4.4. (i) There is no self-orthogonal 5-(22,16,λ)(22,16,\lambda) design.

(ii) There is a self-orthogonal 5-(24,16,λ)(24,16,\lambda) design if and only if λ=78\lambda=78.

(iii) If there is a self-orthogonal 5-(36,18,λ)(36,18,\lambda) design, then λ=1120\lambda=1120.

The existence of a self-orthogonal 5-(24,16,78)(24,16,78) design in (ii) follows by taking the complementary design of the Witt system on 24 points.

Proposition 4.5. There is no self-orthogonal 44-(36,12,λ)(36,12,\lambda) design for any positive integer λ\lambda.

Proof. Let D\mathcal{D} be a self-orthogonal 44-(36,12,15μ)(36,12,15\mu) design. From Table 5, we have μ∈{1,2,3,4,5}\mu\in\{1,2,3,4,5\}. Since C(D)C(\mathcal{D}) is self-orthogonal, there is a self-dual code CC containing C(D)C(\mathcal{D}). Since 36≠0(mod8)36 \ne0 \pmod{8}, CC is not doubly even. Hence, there is a vector x∈C⊆C(D)⊥x \in C \subseteq C(\mathcal{D})^\perp of weight 2s2s, where ss is odd.

We apply the Mendelsohn equations (7) to the vector xx with m=2sm=2s. Multiplying these equations for i=1,2,3,4i=1,2,3,4 by −5,10,−12,8-5,10,-12,8, respectively, adding them, and dividing by 1616, we obtain

8(n4+5n5+15n6)=5μs(s3−14s2+66s−104).8(n_4+5n_5+15n_6)=5\mu s(s^3-14s^2+66s-104).

Since ss is odd, 5s(s3−14s2+66s−104)5s(s^3-14s^2+66s-104) is odd. Thus, we have 8∣μ8\mid\mu, a contradiction.

Now we examine the existence of self-orthogonal (⌊k/4⌋+1)(\lfloor k/4\rfloor+1)-(v,k,λ)(v,k,\lambda) designs for (v,k)(v,k) given in Table 4, where vv is odd or vv is even with v≥38v\ge38. Let D=(X,B)\mathcal{D}=(X,\mathcal{B}) be a self-orthogonal tt-(v,k,λ)(v,k,\lambda) design. Then C(D)C(\mathcal{D}) is a self-orthogonal code containing at least ∣B∣+1|\mathcal{B}|+1 codewords. Since ∣B∣+1≤2⌊v/2⌋|\mathcal{B}|+1\le2^{\lfloor v/2\rfloor}, we have λ≤⌊2⌊v/2⌋(kt)/(vt)⌋\lambda\le\left\lfloor2^{\lfloor v/2\rfloor}\binom{k}{t}/\binom{v}{t}\right\rfloor. Let τ\tau be as defined in § 3. Let

μmax⁡=⌊2⌊v/2⌋(kt)(vt)τ⌋.\mu_{\max}=\left\lfloor\frac{2^{\lfloor v/2\rfloor}\binom{k}{t}}{\binom{v}{t}\tau}\right\rfloor.

Of course, if there is a self-orthogonal tt-(v,k,μτ)(v,k,\mu\tau) design, then μ≤μmax⁡\mu\le\mu_{\max}. For these parameter pairs, Table 6 lists τ\tau and μmax⁡\mu_{\max}, except for (t,v,k)=(4,156,12),(4,311,12)(t,v,k)=(4,156,12),(4,311,12) and (4,159,14)(4,159,14). For these three parameters, we have

(t,v,k)(t,v,k)τ\tauμmax⁡\mu_{\max}(t,v,k)(t,v,k)τ\tauμmax⁡\mu_{\max}
(3,23,8)(3,23,8)88(5,58,16)(5,58,16)529841
(4,23,12)(4,23,12)912(6,39,20)(6,39,20)12051
(4,47,12)(4,47,12)92586(6,43,22)(6,43,22)56145
(4,68,12)(4,68,12)91160253(7,48,24)(7,48,24)285276
(4,27,14)(4,27,14)776(7,72,24)(7,72,24)19958092
(4,80,14)(4,80,14)2347946717(7,96,24)(7,96,24)20939106891
(5,27,16)(5,27,16)21840(7,52,26)(7,52,26)2310142
(5,43,16)(5,43,16)56169(8,55,28)(8,55,28)1771193
(5,48,16)(5,48,16)273156(8,59,30)(8,59,30)3450410
(5,56,16)(5,56,16)427308

Table 6. (⌊k4⌋+1)(\lfloor\frac{k}{4} \rfloor+ 1)-(v,k,λ)(v,k,\lambda) designs with parameters in Table 4

τ(μmax⁡+1)>(v−tk−t).\tau(\mu_{\max}+1)>\binom{v-t}{k-t}.

Thus, the bound μ≤μmax⁡\mu\le\mu_{\max} gives no additional restriction beyond λ≤(v−tk−t)\lambda\le\binom{v-t}{k-t}.

Proposition 4.6. There is no self-orthogonal 55-(27,16,λ)(27,16,\lambda) design for any positive integer λ\lambda.

Proof. For (t,v,k)=(5,27,16)(t,v,k)=(5,27,16), Table 6 gives τ=2184\tau=2184 and μmax⁡=0\mu_{\max}=0. However, such a design has λ=τμ\lambda=\tau\mu for a positive integer μ≤μmax⁡\mu\le\mu_{\max}, a contradiction.

Now let D=(X,B)\mathcal{D}=(X,\mathcal{B}) be a self-orthogonal tt-(v,k,λ)(v,k,\lambda) design with vv odd. Let C′(D)\mathcal{C}'(\mathcal{D}) be the self-orthogonal code obtained from C(D)\mathcal{C}(\mathcal{D}) by appending a zero coordinate. Then C′(D)\mathcal{C}'(\mathcal{D}) is contained in a self-dual code of length v+1v+1 containing at least ∣B∣|\mathcal{B}| codewords of weight kk. Let A(v+1,k)′A'_{(v+1,k)} denote the largest number of codewords of weight kk among self-dual codes of length v+1v+1. Let

μmax⁡′=⌊Av+1,k′(kt)(vt)τ⌋.\mu'_{\max}=\left\lfloor\frac{A'_{v+1,k}\binom{k}{t}}{\binom{v}{t}\tau}\right\rfloor.

Of course, if there is a self-orthogonal tt-(v,k,μτ)(v,k,\mu\tau) design, then μ≤μmax⁡′\mu\leq\mu'_{\max}. In the case (t,v,k)=(3,23,8)(t,v,k)=(3,23,8), we have μmax⁡′=3\mu'_{\max}=3, which improves the result in Table 6, where A(24,8)′=759A'_{(24,8)}=759.

For the cases in Table 6 with vv odd and 0<μmax⁡≤1000 < \mu_{\max} \le100, we give further restrictions in Table 7. For (t,v,k)=(3,23,8)(t,v,k)=(3,23,8), we use the improved bound μmax⁡′=3\mu'_{\max}=3 in the column “μmax⁡\mu_{\max}”. The column “μBI\mu_{BI}” has the same meaning as in Table 2. From Table 7, we have the following:

(t,v,k)(t,v,k)τ\tauμmax⁡\mu_{\max}μBI\mu_{\mathrm{BI}}
(3,23,8)(3,23,8)831
(4,23,12)(4,23,12)9121,2,…,7,9,…,121,2,\ldots,7,9,\ldots,12
(4,27,14)(4,27,14)7761,2,…,61,2,\ldots,6
(6,39,20)(6,39,20)120511,2,…,31,33,…,511,2,\ldots,31,33,\ldots,51
(6,43,22)(6,43,22)561451,2,…,451,2,\ldots,45

Table 7. (⌊k4⌋+1)(\lfloor\frac{k}{4} \rfloor+ 1)-(v,k,λ)(v,k,\lambda) designs with parameters in Table 4 (vv odd)

Proposition 4.7. (i) There is a self-orthogonal 44-(23,12,λ)(23,12,\lambda) design if and only if λ=72\lambda=72.

(ii) There is no self-orthogonal tt-(v,k,λ)(v,k,\lambda) design for (t,v,k)=(4,27,14)(t,v,k)=(4,27,14) and (6,43,22)(6,43,22).

(iii) If there is a self-orthogonal 66-(39,20,λ)(39,20,\lambda) design, then λ=3840\lambda=3840.

The existence of a self-orthogonal 44-(23,12,72)(23,12,72) design in (i) is established in [22].

Proposition 4.8. There is a self-orthogonal 33-(23,8,λ)(23,8,\lambda) design if and only if λ=16\lambda=16. The design is unique, up to isomorphism.

Proof. Up to isomorphism, there is a unique self-orthogonal 33-(23,8,16)(23,8,16) design [19]. From Table 7, it is sufficient to show that there is no self-orthogonal 33-(23,8,24)(23,8,24) design.

Let D=(X,B)\mathcal{D}=(X,\mathcal{B}) be a self-orthogonal 33-(23,8,24)(23,8,24) design. Fix a block B∈BB\in\mathcal{B}. From (10), we have m6=2m_6=2. Thus, there is a block B′B' with ∣B∩B′∣=6|B\cap B'|=6. The sum of the incidence vectors of BB and B′B' is a vector of weight 44 in C(D)⊆C(D)⊥C(\mathcal{D})\subseteq C(\mathcal{D})^\perp. This contradicts Lemma 2.3 (ii), which gives d(C(D)⊥)≥5d(C(\mathcal{D})^\perp)\ge5. □

Proposition 4.9. There is no self-orthogonal 66-(39,20,λ)(39,20,\lambda) design for any positive integer λ\lambda.

Proof. By Proposition 4.7 (iii), it is sufficient to show that there is no self-orthogonal 66-(39,20,3840)(39,20,3840) design. Let D=(X,B)\mathcal{D}=(X,\mathcal{B}) be such a design. Since ∣B∣=323232>218|\mathcal{B}|=323232>2^{18} and C(D)C(\mathcal{D}) is self-orthogonal, the dimension of C(D)C(\mathcal{D}) is 1919. In addition, the following code

C=⟨{(x,0)∣x∈C(D)},1⟩C=\langle\{(x,0)\mid x\in C(\mathcal{D})\},1\rangle

is a doubly even self-dual code of length 4040.

Puncturing CC in the last coordinate gives C(D)⊥C(\mathcal{D})^\perp. By Lemma 2.3 (i), we have

d(C(D)⊥)≥5+λ5λ=5+3415>7.d(C(\mathcal{D})^\perp)\ge5+\frac{\lambda_5}{\lambda}=5+\frac{34}{15}>7.

Thus, CC is an extremal doubly even self-dual code of length 40. In addition, CC has exactly 525504 codewords of weight 20 [15]. Since 1∈C1 \in C, these codewords occur in pairs {x,1+x}\{x,1+x\}, with exactly one member of each pair having last coordinate 0. Hence, exactly 262752 codewords of weight 20 have last coordinate 0. The incidence vectors of the blocks, with a zero coordinate appended, are among these codewords. This contradicts ∣B∣=323232|\mathcal{B}| = 323232.

For some cases (k,v,m)(k,v,m) given in Table 4, we mention the existence of an unsaturated self-orthogonal (⌊k/4⌋+1)(\lfloor k/4\rfloor+1)-(v,k,λ)(v,k,\lambda) design DD such that C(D)⊥C(D)^\perp contains a vector uu of weight mm given in Lemma 4.1.

  • (k,v,m)=(8,16,4)(k,v,m)=(8,16,4) and (8,16,6)(8,16,6): There are exactly five nonisomorphic 3-(16,8,3)(16,8,3) designs. Since these designs are Hadamard 3-designs, these are self-orthogonal. We denote the design corresponding to the Sylvester Hadamard matrix by D16,8,1D_{16,8,1}. It is known that the code C(D16,8,1)C(D_{16,8,1}) is R(1,4)R(1,4), which is a self-orthogonal [16,5,8][16,5,8] code. In addition, we verified by Magma [4] that 63 nontrivial cosets of C(D16,8,1)C(D_{16,8,1}) in C(D16,8,1)⊥C(D_{16,8,1})^\perp have minimum weights 4 and 6.

We constructed by Magma [4] a self-orthogonal 3-(16,8,6)(16,8,6) design D16,8,2=(X,B)D_{16,8,2}=(X,\mathcal{B}) as follows. Let GG be the permutation group of order 96 on X={1,2,…,16}X=\{1,2,\ldots,16\} generated by

(1,16,3,10)(2,15,4,9)(5,12,7,14)(6,11,8,13),(1,16,3,10)(2,15,4,9)(5,12,7,14)(6,11,8,13),
(3,6,7)(4,5,8)(9,15,13,10,16,14)(11,12),(3,6,7)(4,5,8)(9,15,13,10,16,14)(11,12),
(9,10)(11,12)(13,14)(15,16).(9,10)(11,12)(13,14)(15,16).

The block set B\mathcal{B} is the union of the GG-orbits of the following base blocks:

{1,2,5,6,9,10,13,14}, {1,4,5,8,9,12,13,16},\{1,2,5,6,9,10,13,14\},\ \{1,4,5,8,9,12,13,16\},
{1,3,6,8,9,11,14,16}, {1,2,3,4,5,6,7,8},\{1,3,6,8,9,11,14,16\},\ \{1,2,3,4,5,6,7,8\},
{3,4,9,10,11,12,15,16}, {2,4,5,8,10,12,14,15},\{3,4,9,10,11,12,15,16\},\ \{2,4,5,8,10,12,14,15\},
{3,4,5,6,9,10,13,14}.\{3,4,5,6,9,10,13,14\}.

There are two inequivalent doubly even self-dual codes of length 16 and the two codes are denoted by e82e_8^2 and d16+d_{16}^+ (see [6 Table I]). We verified by Magma [4] that C(D16,8,2)C(D_{16,8,2}) is equivalent to d16+d_{16}^+.

These designs are unsaturated, failing (1) for λ=3\lambda=3 and (3) for λ=6\lambda=6.

  • (k,v,m)=(8,22,6)(k, v, m) = (8, 22, 6) and (8,22,7)(8, 22, 7): We denote by D22,8D_{22,8} the unique self-orthogonal 33-(22,8,12)(22, 8, 12) design given in [19]. We verified by MAGMA [4] that the code C(D22,8)C(D_{22,8}) is a self-orthogonal [22,10,8][22, 10, 8] code. In addition, we verified by MAGMA [4] that three nontrivial cosets of C(D22,8)C(D_{22,8}) in C(D22,8)⊥C(D_{22,8})^{\perp} have minimum weights 6 and 7. Hence, D22,8D_{22,8} is unsaturated, failing (1).

  • (k,v,m)=(8,23,7)(k, v, m) = (8, 23, 7): We denote by D23,8D_{23,8} the unique self-orthogonal 33-(23,8,16)(23, 8, 16) design given in [19]. The code C(D23,8)C(D_{23,8}) is a self-orthogonal [23,11,8][23, 11, 8] code, which is equivalent to G23⊥G_{23}^{\perp}. In addition, we verified by MAGMA [4] that the unique nontrivial coset of C(D23,8)C(D_{23,8}) in C(D23,8)⊥C(D_{23,8})^{\perp} has minimum weight 7. Hence, D23,8D_{23,8} is unsaturated, failing (1).

  • (k,v,m)=(10,22,6)(k, v, m) = (10, 22, 6): Up to isomorphism, there is a unique self-orthogonal 33-(22,10,48)(22, 10, 48) design and the code is equivalent to C22C_{22} [22]. Since C22C_{22} has minimum weight 6, this design is unsaturated, failing (3).

  • (k,v,m)=(12,23,7)(k, v, m) = (12, 23, 7) and (12,23,8)(12, 23, 8): We denote by D23,12D_{23,12} the unique self-orthogonal 44-(23,12,72)(23, 12, 72) design [22]. We verified by MAGMA [4] that the code C(D23,12)C(D_{23,12}) is equivalent to G23⊥G_{23}^{\perp}. As described above, the unique nontrivial coset of C(D23,12)C(D_{23,12}) in C(D23,12)⊥C(D_{23,12})^{\perp} has minimum weight 7. Since C(D23,12)C(D_{23,12}) has parameters [23,11,8][23, 11, 8], D23,12D_{23,12} is unsaturated, failing both (1) and (3).

  • (k,v,m)=(12,24,8)(k, v, m) = (12, 24, 8): Up to isomorphism, there is a unique self-orthogonal 44-(24,12,120)(24, 12, 120) design and the code is equivalent to G24G_{24} [22]. Since G24G_{24} has minimum weight 8, this design is unsaturated, failing (3).

  • (k,v,m)=(12,47,11)(k, v, m) = (12, 47, 11): We denote by D47,12D_{47,12} the unique self-orthogonal 44-(47,12,36)(47, 12, 36) design [13]. We verified by MAGMA [4] that the code C(D47,12)C(D_{47,12}) is equivalent to QR47⊥QR_{47}^{\perp}, which is a self-orthogonal [47,23,12][47, 23, 12] code. In addition, we verified by MAGMA [4] that the unique nontrivial coset of C(D47,12)C(D_{47,12}) in C(D47,12)⊥C(D_{47,12})^{\perp} has minimum weight 11. Hence, D47,12D_{47,12} is unsaturated, failing (1).

  • (k,v,m)=(16,24,8)(k, v, m) = (16, 24, 8): A self-orthogonal 55-(24,16,78)(24, 16, 78) design D24,16D_{24,16} is obtained as the complementary design of the Witt system on 24 points. Thus, the code C(D24,16)C(D_{24,16}) is equivalent to G24G_{24}. Since G24G_{24} has minimum weight 8, D24,16D_{24,16} is unsaturated, failing (3).

  • (k,v,m)=(16,48,12)(k,v,m)=(16,48,12): We denote by D48,16\mathcal{D}_{48,16} the self-orthogonal 55-(48,16,1365)(48,16,1365) design, whose blocks are the supports of the codewords of weight 16 in QR48QR_{48}. We verified by Magma [4] that the code C(D48,16)C(\mathcal{D}_{48,16}) is the same as QR48QR_{48}. Since QR48QR_{48} has minimum weight 12, D48,16\mathcal{D}_{48,16} is unsaturated, failing (3).

Self-orthogonal designs satisfying (3)

In this section, we work with a slightly weaker hypothesis than the assumption that D\mathcal{D} is saturated. We assume that C(D)C(\mathcal{D}) has minimum weight kk.

Proposition 5.1. Let D\mathcal{D} be a self-orthogonal tt-(v,k,λ)(v,k,\lambda) design with kk even, and let t0=⌊k4⌋+1t_0=\left\lfloor\frac{k}{4}\right\rfloor+1. If C(D)C(\mathcal{D}) has minimum weight kk and

t≥t0≥2,t\ge t_0\ge2,

then

λs=(−1)t0−12t0−1(k/2k/2−t0)∑i=1t0i(−2)i−1(2t0−i−1t0−1)(ki)∏j=0t0−1v−jk−j∏j=st0−1v−jk−j(14)\lambda_s= \frac{(-1)^{t_0-1}2^{t_0-1}\binom{k/2}{k/2-t_0}} {\sum_{i=1}^{t_0}i(-2)^{i-1}\binom{2t_0-i-1}{t_0-1}\binom{k}{i}\prod_{j=0}^{t_0-1}\frac{v-j}{k-j}} \prod_{j=s}^{t_0-1}\frac{v-j}{k-j} \tag*{(14)}

is an integer for s=0,1,…,t0s=0,1,\ldots,t_0.

Proof. Fix a block B0B_0 of D\mathcal{D}. Note that D\mathcal{D} is a t0t_0-design. Since C(D)C(\mathcal{D}) has minimum weight kk, for any block BB we have ∣B∩B0∣≤k/2|B\cap B_0|\le k/2 unless B=B0B=B_0. Thus, we have

nj=0for j>⌊k4⌋ and j≠k2,n_j=0\quad\text{for }j>\left\lfloor\frac{k}{4}\right\rfloor\text{ and }j\ne\frac{k}{2},

and obviously nk/2=1n_{k/2}=1. From (6), we have

∑j=1t0−1(2ji)nj=(λi−1)(ki)(i=1,2,…,t0).\sum_{j=1}^{t_0-1}\binom{2j}{i}n_j=(\lambda_i-1)\binom{k}{i}\quad(i=1,2,\ldots,t_0).

By Lemma 2.2, we have

∑i=1t0ki(λi−1)(ki)=0,\sum_{i=1}^{t_0}k_i(\lambda_i-1)\binom{k}{i}=0,

where kik_i is given in (8) with t=t0t=t_0. From (5), we have

λt0∑i=1t0ki(ki)∏j=it0−1v−jk−j=∑i=1t0ki(ki).\lambda_{t_0}\sum_{i=1}^{t_0} k_i\binom{k}{i}\prod_{j=i}^{t_0-1}\frac{v-j}{k-j}=\sum_{i=1}^{t_0} k_i\binom{k}{i}.

By Lemma 2.1 with s=t0−1s=t_0-1 and j=k/2−1j=k/2-1, we have

∑i=1t0ki(ki)=k(−4)t0−1(k/2−1t0−1)=(−1)t0−122t0−1t0(k/2k/2−t0)≠0.\begin{aligned} \sum_{i=1}^{t_0} k_i\binom{k}{i}=k(-4)^{t_0-1}\binom{k/2-1}{t_0-1} \\ &=(-1)^{t_0-1}2^{2t_0-1}t_0\binom{k/2}{k/2-t_0}\ne0. \end{aligned}

Thus, the coefficient of λt0\lambda_{t_0} is nonzero. Solving for λt0\lambda_{t_0} and applying (5) gives (14).

As a special case of (14), we have

λt0=(−1)t0−122t0−1t0(k/2k/2−t0)∑i=1t0i(−2)i−1(2t0−i−1t0−1)(ki)∏j=it0−1v−jk−j(15)\lambda_{t_0}=\frac{(-1)^{t_0-1}2^{2t_0-1}t_0\binom{k/2}{k/2-t_0}}{\sum_{i=1}^{t_0}i(-2)^{i-1}\binom{2t_0-i-1}{t_0-1}\binom{k}{i}\prod_{j=i}^{t_0-1}\frac{v-j}{k-j}} \tag*{(15)}

Proposition 5.2. Let D\mathcal{D} be a self-orthogonal tt-(v,k,λ)(v,k,\lambda) design with kk even and let t0=⌊k4⌋+1t_0=\left\lfloor\frac{k}{4}\right\rfloor+1. If C(D)C(\mathcal{D}) has minimum weight kk and

t≥t0≥2andk≤40,t\ge t_0\ge2\quad\text{and}\quad k\le40,

then

(t0,v,k,λt0)=(2,7,4,2),(2,8,4,3),(2,9,4,6),(t_0,v,k,\lambda_{t_0})=(2,7,4,2),(2,8,4,3),(2,9,4,6),
(2,16,6,2),(2,21,6,4),(2,22,6,5),(2,24,6,10),(2,25,6,20),(2,16,6,2),(2,21,6,4),(2,22,6,5),(2,24,6,10),(2,25,6,20),
(3,16,8,3),(3,22,8,12),(3,23,8,16),(3,24,8,21),(3,16,8,3),(3,22,8,12),(3,23,8,16),(3,24,8,21),
(3,26,8,28),(3,29,8,16),(3,30,8,12),(3,32,8,7),(3,26,8,28),(3,29,8,16),(3,30,8,12),(3,32,8,7),
(3,26,10,3),(3,42,10,9),(3,46,10,8),(3,50,10,6),(3,26,10,3),(3,42,10,9),(3,46,10,8),(3,50,10,6),
(4,36,12,15),(4,47,12,36),(4,48,12,44),(4,51,12,2640),(4,36,12,15),(4,47,12,36),(4,48,12,44),(4,51,12,2640),
(5,56,16,42),(5,64,16,91),(5,72,16,78),(5,56,16,42),(5,64,16,91),(5,72,16,78),
(7,120,24,231).(16)(7,120,24,231). \tag*{(16)}

Proof. For each even kk with 4≤k≤404 \le k \le40, we verified by MAGMA [4] that v<γ(k)v < \gamma(k) whenever v>kv > k and λt0\lambda_{t_0} in (15) is a positive integer, where γ(k)\gamma(k) is given in Table 8. Then we determined by MAGMA [4] the set V(k)V(k) of integers vv with k<v<γ(k)k < v < \gamma(k) for which λ0,λ1,…,λt0\lambda_0, \lambda_1, \ldots, \lambda_{t_0} are positive integers. The values of γ(k)\gamma(k) and V(k)V(k) are listed in Table 8. To illustrate our approach, we give details for k=10k = 10. If k=10k = 10, then

kkγ(k)\gamma(k)V(k)V(k)kkγ(k)\gamma(k)V(k)V(k)
410{7,8,9}\{7, 8, 9\}24151{120}\{120\}
626{16,21,22,24,25}\{16, 21, 22, 24, 25\}26195∅\varnothing
845{16,22,23,24,26,29,30,32}\{16, 22, 23, 24, 26, 29, 30, 32\}28139∅\varnothing
1075{26,42,46,50}\{26, 42, 46, 50\}30158∅\varnothing
1252{36,47,48,51}\{36, 47, 48, 51\}32224∅\varnothing
1470∅\varnothing34256∅\varnothing
16104{56,64,72}\{56, 64, 72\}36183∅\varnothing
181∅\varnothing38202∅\varnothing
2095∅\varnothing40284∅\varnothing
22114∅\varnothing

Table 8. γ(k)\gamma(k) and V(k)V(k) in the proof of Proposition 5.2

λt0=1152v2−84v+1892.\lambda_{t_0} = \frac{1152}{v^2 - 84v + 1892}.

Since λt0≥1\lambda_{t_0} \ge1, v2−84v+740=(v−10)(v−74)≤0v^2 - 84v + 740 = (v - 10)(v - 74) \le0. Thus, we may take γ(10)=75\gamma(10) = 75. Checking all integers vv with 10<v<7510 < v < 75, we verified by MAGMA [4] that V(10)={26,42,46,50}V(10) = \{26, 42, 46, 50\}.

Remark 5.3. The following cases

(t0,v,k,λt0)=(3,26,8,28),(3,30,8,12) and (4,36,12,15)(t_0, v, k, \lambda_{t_0}) = (3, 26, 8, 28), (3, 30, 8, 12) \text{ and } (4, 36, 12, 15)

in (16) are excluded by Propositions 3.4, 3.3 and 4.5, respectively.

Self-orthogonal codes and designs satisfying (2) and (3)

The aim of this section is to give a proof of Theorem B. Since kk is even and t<kt<k, the assumptions 2≤t≤112\leq t\leq11 and t≥⌊k/4⌋+1t\geq\lfloor k/4\rfloor+1 imply 4≤k≤424\leq k\leq42. For k=42k=42, we have t0=11t_0=11. We verified by Magma [4] that λ11\lambda_{11} in (15) is not a positive integer for any integer vv with 42<v<31642<v<316. Write λ11=N/D(v)\lambda_{11}=N/D(v), where N>0N>0 is the numerator and D(v)D(v) is the denominator in (15). We also verified by Magma [4] that all coefficients of D(316+x)−ND(316+x)-N are positive. Hence, D(v)>N>0D(v)>N>0 for all v≥316v\geq316, and therefore 0<λ11<10<\lambda_{11}<1. Thus, Proposition 5.1 excludes the case k=42k=42. The cases k=8,10,16,18k=8,10,16,18 are covered by Theorem 1.2. Therefore, we classify self-orthogonal (⌊k/4⌋+1)(\lfloor k/4\rfloor+1)-(v,k,λ)(v,k,\lambda) designs satisfying (2) and (3) for the remaining even values of kk with 4≤k≤404\leq k\leq40.

Lemma 6.1. Let D\mathcal{D} be a self-orthogonal tt-(v,k,λ)(v,k,\lambda) design with kk even. Then D\mathcal{D} satisfies (2) and (3) if and only if C(D)C(\mathcal{D}) has minimum weight kk and the supports of its codewords of weight kk are precisely the blocks of D\mathcal{D}.

Proof. This follows directly from (2) and (3).

Remark 6.2. Let CC be a self-orthogonal code of minimum weight kk whose codewords of minimum weight form a tt-design D\mathcal{D}. Then D\mathcal{D} satisfies (2) and (3), and C(D)⊆CC(\mathcal{D})\subseteq C, but the inclusion may be strict. Thus, to classify such codes CC, we must also determine whether there are self-orthogonal codes properly containing C(D)C(\mathcal{D}) with minimum weight kk and exactly the same codewords of weight kk.

Let D\mathcal{D} be a self-orthogonal (⌊k/4⌋+1)(\lfloor k/4\rfloor+1)-(v,k,λ)(v,k,\lambda) design satisfying (2) and (3), where kk is even with 4≤k≤404\leq k\leq40. Set t=⌊k/4⌋+1t=\lfloor k/4\rfloor+1. By Proposition 5.2, (t,v,k,λ)(t,v,k,\lambda) is listed in (16). For (v,k)=(9,4),(25,6),(29,8)(v,k)=(9,4),(25,6),(29,8) and (51,12)(51,12), it follows from (13) and Table 4 that (11) has no integer solution with 0<m<v0<m<v. By Proposition 4.2, D\mathcal{D} must be saturated, which is impossible since vv is odd. The following cases

(t,v,k,λ)=(3,26,8,28),(3,30,8,12) and (4,36,12,15)(t,v,k,\lambda)=(3,26,8,28),(3,30,8,12)\text{ and }(4,36,12,15)

are excluded by Propositions 3.4, 3.3 and 4.5, respectively. We consider the remaining cases in (16) below.

  • (t,k)=(2,4)(t,k)=(2,4): The remaining parameters are (v,λ)=(7,2)(v,\lambda)=(7,2) and (8,3)(8,3), for which ∣B∣=7|\mathcal{B}|=7 and 1414, respectively. Since

∣B∣+1≤∣C(D)∣≤2⌊v/2⌋,|\mathcal{B}|+1\leq|C(\mathcal{D})|\leq2^{\lfloor v/2\rfloor},

C(D)C(\mathcal{D}) has parameters [7,3,4][7,3,4] and [8,4,4][8,4,4], respectively. These codes are unique up to equivalence and are e7e_7 and e8e_8, respectively. From (2), D\mathcal{D} is therefore isomorphic to the unique 22-(7,4,2)(7,4,2) design or the unique 33-(8,4,1)(8,4,1) design, respectively. Both codes are maximal self-orthogonal.

  • (t,k)=(2,6)(t,k)=(2,6): Distinct blocks meet in 0 or 2 points, since an intersection of size 4 yields a codeword of weight 4, contrary to (3). Thus, D\mathcal{D} is symmetric when v=16v=16 and quasi-symmetric otherwise.

For v=16v=16, let B1B_1, B2B_2 and B3B_3 denote the three nonisomorphic symmetric 22-(16,6,2)(16,6,2) designs listed in [9 Section 15.8].2 We verified by MAGMA [4] that only B1B_1 satisfies (2) and (3), and that C(B1)C(B_1) has parameters [16,6,6][16,6,6]. This self-orthogonal code is unique up to equivalence, and no self-orthogonal [16,7,6][16,7,6] code exists [5]. Hence, no self-orthogonal code of minimum weight 6 properly contains C(B1)C(B_1).

For v=21v=21, there is a unique quasi-symmetric 22-(21,6,4)(21,6,4) design, up to isomorphism. It satisfies (2) and (3), and its code is equivalent to a [21,10,6][21,10,6] code obtained by shortening C22C_{22} in one coordinate [20]. All such shortened codes are equivalent.

For v=22v=22, every quasi-symmetric 22-(22,6,5)(22,6,5) design is isomorphic to the unique 33-(22,6,1)(22,6,1) design, which satisfies (2) and (3) and generates a code equivalent to C22C_{22} [20]. Both the [21,10,6][21,10,6] code and C22C_{22} are maximal self-orthogonal.

For v=24v=24, Proposition 4.2 and (13) imply that D\mathcal{D} is saturated. Hence, C(D)C(\mathcal{D}) is equivalent to C24C_{24}, which has only 64 codewords of weight 6 [18], contradicting ∣B∣=184|\mathcal{B}|=184.

  • (t,k)=(4,12)(t,k)=(4,12): For (v,λ)=(47,36)(v,\lambda)=(47,36) and (48,44)(48,44), there is a unique self-orthogonal 44-(v,12,λ)(v,12,\lambda) design, up to isomorphism. These designs satisfy (2) and (3), and their codes are equivalent to QR47⊥QR^{\perp}_{47} and QR48QR_{48}, respectively [13]. Both codes are maximal self-orthogonal.

  • (t,k)=(3,8),(3,10)(t,k)=(3,8),(3,10) and (5,16)(5,16): These cases are covered by Theorem 1.2.

Proposition 6.3. There is no self-orthogonal 77-(120,24,231)(120,24,231) design.

Proof. Let D\mathcal{D} be a self-orthogonal 77-(120,24,231)(120,24,231) design. By Proposition 4.2 and Table 4, D\mathcal{D} is saturated. Thus, C(D)C(\mathcal{D}) is an extremal doubly even self-dual [120,60,24][120,60,24] code whose codewords of minimum weight have supports forming D\mathcal{D}. This contradicts [2 Theorem 5], since D\mathcal{D} is also a 66-design. □\square

We are now in a position to give a proof of Theorem B.

Proof of Theorem B. Let CC be as in Theorem B, and suppose that 2≤t≤112 \le t \le11. Then 4≤k≤424 \le k \le42. By Lemma 6.1 and the above arguments, we obtain a classification of self-orthogonal codes of minimum weight kk whose codewords of minimum weight form a (⌊k/4⌋+1)(\lfloor k/4\rfloor+ 1)-design, for 4≤k≤424 \le k \le42. For t=2,3,4t=2,3,4 and 55, the assertions of Theorem B follow from this classification, the known design properties of these codes and the integrality conditions in (5).

Suppose that 6≤t≤116 \le t \le11. The above classification gives k≤16k \le16. Since a tt-design is also a 55-design, Theorem 1.2 (ii) implies that CC is equivalent to one of G24G_{24}, QR48QR_{48} and C72C_{72}. However, the supports of their minimum-weight codewords do not form 66-designs by Remark 1.1 (ii). This completes the proof of Theorem B. □\square

In addition, as a consequence of Theorem B, we have the following classification of certain saturated designs.

Corollary 6.4. Let D\mathcal{D} be a self-orthogonal tt-(v,k,λ)(v,k,\lambda) design, where kk is even, 4≤k≤404 \le k \le40 and t≥⌊k/4⌋+1t \ge\lfloor k/4\rfloor+ 1. Then D\mathcal{D} is saturated if and only if D\mathcal{D} is isomorphic to one of the following:

  1. the unique 33-(8,4,1)(8,4,1) design,

  2. the unique 33-(22,6,1)(22,6,1) design,

  3. the unique 55-(24,8,1)(24,8,1) design,

  4. the five nonisomorphic self-orthogonal 33-(32,8,7)(32,8,7) designs,

  5. the unique self-orthogonal 33-(46,10,8)(46,10,8) design,

  6. the unique self-orthogonal 55-(48,12,8)(48,12,8) design and

  7. a putative self-orthogonal 55-(72,16,78)(72,16,78) design.

Proof. If DD is saturated, then C(D)\mathcal{C}(D) is self-dual and has minimum weight kk. Applying Theorem B with t=⌊k/4⌋+1t=\lfloor k/4\rfloor+1 and using (2) gives the designs listed above. Conversely, every listed design satisfies (1), (2) and (3) (see [11], [13], [20] and [21]).

Acknowledgments. This work was supported by JSPS KAKENHI Grant Number 23K25784. The results presented in this paper were mainly obtained during discussions with Akihiro Munemasa around 2014 and 2015. The author is deeply grateful to him for his useful suggestions, continuous discussions and for allowing the author to publish this paper as a single author.

Use of generative AI. In October 2026, while preparing this paper for submission, the author used OpenAI’s ChatGPT (GPT-6 Astra) to assist in checking the proofs, simplifying arguments and improving the exposition.

References

  1. [1]E.F. Assmus, Jr. and H.F. Mattson, Jr., New 5-designs, J. Combin. Theory 6 (1969), 122–151.DOI
  2. [2]E. Bannai, M. Koike, M. Shinohara and M. Tagami, Spherical designs attached to extremal lattices and the modulo p property of Fourier coefficients of extremal modular forms, Mosc. Math. J. 6 (2006), 225–264.DOI
  3. [4]W. Bosma, J. Cannon and C. Playoust, The Magma algebra system I: The user language, J. Symbolic Comput. 24 (1997), 235–265.DOI
  4. [5]I. Bouyukliev, S. Bouyuklieva, T.A. Gulliver and P.R.J. Östergård, Classification of optimal binary self-orthogonal codes, J. Combin. Math. Combin. Comput. 59 (2006), 33–87.
  5. [6]J.H. Conway and N.J.A. Sloane, A new upper bound on the minimal distance of self-dual codes, IEEE Trans. Inform. Theory 36 (1990), 1319–1333.DOI
  6. [7]J.H. Conway, V. Pless and N.J.A. Sloane, The binary self-dual codes of length up to 32: a revised enumeration, J. Combin. Theory Ser. A 60 (1992), 183–195.DOI
  7. [8]J. de la Cruz and W. Willems, 5-designs related to binary extremal self-dual codes of length 24m, Theory and applications of finite fields, 75–80, Contemp. Math., 579, Amer. Math. Soc., Providence, RI, 2012.DOI
  8. [9]M. Hall, Jr., Combinatorial Theory (second edition), John Wiley & Sons, New York, 1986.
  9. [10]M. Harada, Remark on a 5-design related to a putative extremal doubly-even self-dual [96, 48, 20] code, Des. Codes Cryptogr. 37 (2005), 355–358.DOI
  10. [11]M. Harada, M. Kitazume and A. Munemasa, On a 5-design related to an extremal doubly even self-dual code of length 72, J. Combin. Theory Ser. A 107 (2004), 143–146.DOI
  11. [13]M. Harada, A. Munemasa and V.D. Tonchev, A characterization of designs related to an extremal doubly-even self-dual code of length 48, Ann. Comb. 9 (2005), 189–198.DOI
  12. [14]M. Lalaude-Labayle, On binary linear codes supporting t-designs, IEEE Trans. Inform. Theory 47 (2001), 2249–2255.
  13. [15]C.L. Mallows and N.J.A. Sloane, An upper bound for self-dual codes, Inform. Control 22 (1973), 188–200.DOI
  14. [16]N.S. Mendelsohn, Intersection numbers of t-designs, In: Studies in Pure Mathematics (presented to Richard Rado), Academic Press, London 1971, 145–150.
  15. [17]V. Pless, A classification of self-orthogonal codes over GF (2), Discrete Math. 3 (1972), 209–246.DOI
  16. [18]V. Pless and N.J.A. Sloane, On the classification and enumeration of self-dual codes, J. Combin. Theory Ser. A 18 (1975), 313–335.DOI
  17. [19]V.D. Tonchev, A characterization of designs related to the Witt system S(5, 8, 24), Math. Z. 191 (1986), 225–230.
  18. [20]V.D. Tonchev, Quasi-symmetric designs and self-dual codes, Europ. J. Combin. 7 (1986), 67–73.DOI
  19. [21]V.D. Tonchev, Quasi-symmetric 2-(31, 7, 7) designs and a revision of Hamada’s conjecture, J. Combin. Theory Ser. A 42 (1986), 104–110.DOI
  20. [22]V.D. Tonchev, A characterization of designs related to dodecads in the Witt system S(5, 8, 24), J. Combin. Theory Ser. A 43 (1986), 219–227.DOI

Paper details

vlnc:W_ug6ubvnx
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 9

downloads · 0

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

  • Mathematics

GPT-6 Astra (bare)

View statement
Looking for related papers…
Contents