Email address: 20241032@gdufe.edu.cn (Zhilin Zhang)
Contents
Part I: Introduction 4
1 Introduction 5
Part II: Structural reduction 11
2 The O’Nan–Scott reduction 12
3 Point-imprimitive actions are excluded 35
3.1 Key parameter constraints 35
3.2 Classical groups: point-imprimitive candidates are excluded 36
3.3 General result for point-imprimitive case 36
Part III: The almost simple classification 52
4 Almost simple type (AS) 53
4.1 Key ordering principle for non-symmetric designs 55
4.2 Classification for P S L 2 ( q ) PSL_2(q) P S L 2 ( q ) 56
4.3 Even characteristic 68
4.4 Classical groups (Lie type) 98
4.5 Scope: non-2-homogeneous point actions 98
4.6 The Aschbacher classes 99
4.7 P S L n ( q ) PSL_n(q) P S L n ( q ) (n ≥ 3 n \ge3 n ≥ 3 ) 100
4.8 P S p 2 n ( q ) PSp_{2n}(q) P S p 2 n ( q ) 101
4.9 P S U n ( q ) PSU_n(q) P S U n ( q ) 102
4.10 P Ω n ϵ ( q ) P\Omega_n^{\epsilon}(q) P Ω n ϵ ( q ) 102
4.11 Summary of the classical group classification 103
4.12 Point-imprimitive case for classical groups: exclusion 103
4.13 Classification of AS type for classical groups 105
4.14 Exceptional Lie type groups 106
4.15 Sporadic simple groups 108
4.16 Examples 109
Part I: Introduction
Introduction
A combinatorial design D = ( P , B ) \mathcal{D} = (\mathcal{P}, \mathcal{B}) D = ( P , B ) is a pair consisting of a finite set P \mathcal{P} P of points and a collection B \mathcal{B} B of subsets of P \mathcal{P} P , called blocks. It has strength t t t , for an integer t ≥ 2 t \ge2 t ≥ 2 , if every t t t -subset of P \mathcal{P} P is contained in exactly λ \lambda λ blocks; it is then a t t t -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design, where v = ∣ P ∣ v = |\mathcal{P}| v = ∣ P ∣ and k = ∣ B ∣ k = |B| k = ∣ B ∣ for B ∈ B B \in\mathcal{B} B ∈ B , and a 2-design is the case t = 2 t = 2 t = 2 . The automorphism group Aut ( D ) \operatorname{Aut}(\mathcal{D}) Aut ( D ) consists of all permutations of P \mathcal{P} P that map blocks to blocks, and D \mathcal{D} D is block-primitive when this group acts primitively on B \mathcal{B} B .
The study of designs with automorphism groups that act transitively or primitively on points or blocks has a long history. A fundamental result of Delandtsheer and Doyen [20 ] states that if a block-transitive automorphism group G G G is imprimitive on points, then v ≤ ( k 2 ) v \le\binom{k}{2} v ≤ ( 2 k ) . This result has been used extensively in the classification of block-transitive point-imprimitive designs.
The O’Nan–Scott theorem classifies finite primitive permutation groups into five types: affine (HA), almost simple (AS), diagonal (SD and CD), product action (PA) and twisted wreath (TW), the diagonal type subsuming the simple and compound diagonal cases (Dixon and Mortimer [24 ] ). The quasiprimitive version of Liebeck, Praeger and Saxl [31 ] refines the list to eight types: affine (HA), almost simple (AS), holomorph simple (HS), holomorph compound (HC), simple diagonal (SD), compound diagonal (CD), twisted wreath (TW) and product action (PA). In the diagonal case we separate the two-sided from the one-sided simple diagonal action, a distinction the design-theoretic applications require. Table 2 records the eight types, the structure of the socle, the form of the point and block actions, and the place in this paper where each type is settled.
Table 2. O’Nan–Scott types for quasiprimitive permutation groups. For each type we record the structure of the socle, the form of the point action on P = G / H \mathcal{P}=G/H P = G / H and of the block action on B = G / M \mathcal{B}=G/M B = G / M , and the mechanism and location of the exclusion or classification.
Quasiprimitivity does not imply primitivity. Groups of affine, holomorph simple and holomorph compound type are primitive (Praeger [35 ] ; see also Theorem 3.3 of Chen, Hua, Li and Wu [15 ] ), whereas the other five types admit quasiprimitive actions that are not primitive; for the almost simple type, P S L 2 ( 7 ) PSL_2(7) P S L 2 ( 7 ) acting on the 28 points P S L 2 ( 7 ) / S 3 PSL_2(7)/S_3 P S L 2 ( 7 ) / S 3 is an example, the stabilizer S 3 S_3 S 3 being self-normalizing but not maximal. Point-imprimitive actions therefore have to be excluded separately, which is done in § 2.4 (Proposition 3.4 ) and, family by family, in § 3.1 .
Huber [27 , 28 ] classified flag-transitive Steiner t t t -designs, and Zieschang [43 ] studied flag-transitive automorphism groups of 2-designs with ( r , λ ) = 1 (r,\lambda) = 1 ( r , λ ) = 1 . For symmetric designs, block-primitivity coincides with point-primitivity by duality and the flag-transitive case is well understood: Buekenhout, Delandtsheer and Doyen [9 ] treated linear spaces (λ = 1 \lambda= 1 λ = 1 ), O’Reilly-Regueiro [36 , 37 ] the cases λ = 2 , 3 \lambda= 2, 3 λ = 2 , 3 and λ = 4 \lambda= 4 λ = 4 , Alavi, Bayat, Daneshkhah and Montinaro [2 ] prime λ \lambda λ , and O’Reilly-Regueiro and Rodríguez-Fitta [38 ] the product action. Braić, Golemac, Mandić and Vučićíć [7 , 8 ] classified the primitive symmetric designs with at most 255 and with up to 2500 points, their automorphism groups being affine or almost simple. The symmetric case is not treated here.
For block-transitive designs, Camina and Gagen [13 ] showed that k ∣ v k \mid v k ∣ v forces flag-transitivity, and Camina and Siemons [14 ] studied the 2 2 2 -( v , k , 1 ) (v, k, 1) ( v , k , 1 ) case. Chen, Hua, Li and Wu [15 ] proved that the action of an automorphism group of a non-trivial 2 2 2 -design on its blocks is faithful, and the known block-transitive 2 2 2 -( v , 5 , λ ) (v, 5, \lambda) ( v , 5 , λ ) result gives the HA/AS/PA alternative [30 ] . For point-imprimitive block-transitive designs, Cameron and Praeger [12 ] give v ≤ ( k − 2 ) 2 v \le(k - 2)^2 v ≤ ( k − 2 ) 2 in the non-trivial case, which Mann and Tuan [33 ] sharpened for t = 3 t = 3 t = 3 .
Theorem 1.1 (Block-primitivity implies point-primitivity). Let D D D be a non-trivial 2 2 2 -( v , k , λ ) (v, k, \lambda) ( v , k , λ ) design with 2 < k < v − 1 2 < k < v - 1 2 < k < v − 1 . If Aut ( D ) \operatorname{Aut}(D) Aut ( D ) acts primitively on the blocks of D D D , then it acts primitively on the points.
The proof runs in two parts. For a symmetric design block-primitivity and point-primitivity coincide by duality, so nothing is left to prove. For a non-symmetric 2 2 2 -design the point action is primitive by Theorem 1.3 (2). A design of strength t ≥ 3 t \ge3 t ≥ 3 is either complete or non-complete. A complete design has block set the full set of k k k -subsets, hence Aut ( D ) = Sym ( v ) \operatorname{Aut}(D) = \operatorname{Sym}(v) Aut ( D ) = Sym ( v ) , and Sym ( v ) \operatorname{Sym}(v) Sym ( v ) acts 2 2 2 -transitively on the points; the conclusion therefore holds whether or not the block action is primitive. The non-complete designs of strength t ≥ 3 t \ge3 t ≥ 3 are among those classified in Theorem 1.4 , and the same conclusion applies.
We begin with the non-symmetric 2 2 2 -designs. The product action is excluded for every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 and arbitrary block stabilizers, the point action being primitive (Proposition 3.4 ); a reduction over the admissible primitive degrees, using the degree tables of Dixon–Mortimer [24 ] , leaves the six socles A 5 A_5 A 5 , A 6 A_6 A 6 , P S L 2 ( 7 ) PSL_2(7) P S L 2 ( 7 ) , P S L 2 ( 8 ) PSL_2(8) P S L 2 ( 8 ) , P S L 2 ( 11 ) PSL_2(11) P S L 2 ( 11 ) and P S L 2 ( 13 ) PSL_2(13) P S L 2 ( 13 ) with seventeen primitive degrees, and the sieve of Proposition 7.11 excludes them, the 2-homogeneous pairs being removed by a transitivity argument. The SD, CD and TW types are excluded in Section 2.
Theorem 1.3 (Complete classification). Let D = ( P , B ) \mathcal{D}=(\mathcal{P},\mathcal{B}) D = ( P , B ) be a non-trivial, non-symmetric 2-design with 2 < k < v − 1 2<k<v-1 2 < k < v − 1 , and let G ≤ Aut ( D ) G\leq\operatorname{Aut}(\mathcal{D}) G ≤ Aut ( D ) act primitively on B \mathcal{B} B . Denote by G P G^{\mathcal{P}} G P and G B G^{\mathcal{B}} G B the actions of G G G on the point set P \mathcal{P} P and the block set B \mathcal{B} B , respectively. Then:
(1) G P G^{\mathcal{P}} G P acts quasiprimitively on P \mathcal{P} P (Lemma 2.1).
(2) G P G^{\mathcal{P}} G P cannot act imprimitively on P \mathcal{P} P (Proposition 3.4): by the quasiprimitive O’Nan–Scott classification, an imprimitive quasiprimitive action is of almost simple, twisted wreath, simple diagonal, compound diagonal or product action type (the affine, holomorph simple and holomorph compound types are always primitive), and each of these types is excluded (TW, SD and CD in Section 2; PA by Proposition 7.11 and Lemma 3.11; AS in Section 4). Consequently no non-trivial non-symmetric block-primitive 2-design admits a point-imprimitive automorphism group.
(3) Hence G P G^{\mathcal{P}} G P acts primitively on P \mathcal{P} P , and by the O’Nan–Scott theorem both G P G^{\mathcal{P}} G P and G B G^{\mathcal{B}} G B are of almost simple type (Table 2 , Theorem 5.1).
(4) The point-primitive almost simple case is then classified completely across all families of finite simple groups; the classification is summarized in Table 1 and stated in full in Proposition 4.1.
Table 1. Summary of the classification of point-primitive non-symmetric block-primitive 2-designs (Proposition 4.1 ). The existence results are listed together with the design families and the acting groups; all remaining cases are excluded.
Point-primitive designs with classical, sporadic or exceptional socle are classified in Section 4: the reduction lists the candidate quadruples ( T , G , M , H ) (T,G,M,H) ( T , G , M , H ) and the design conditions they have to satisfy, and the outcome for each family is stated in Proposition 4.1.
Among the designs classified in Theorem 1.3, those of strength t ≥ 3 t\geq3 t ≥ 3 are classified as follows; the full statement, with block counts and automorphism groups, is Theorem 6.3 of Section 6.
Theorem 1.4 (Block-primitive designs of strength t ≥ 3 t\geq3 t ≥ 3 ). Let D \mathcal{D} D be a non-trivial non-complete block-primitive design of strength t ≥ 3 t\geq3 t ≥ 3 . Then D \mathcal{D} D is one of the following:
the inversive planes 3 3 3 -( q 2 + 1 , q + 1 , 1 ) \left(q^2+1,q+1,1\right) ( q 2 + 1 , q + 1 , 1 ) , q ≥ 3 q \ge3 q ≥ 3 a prime power;
the Steiner systems 3 3 3 -( q + 1 , 4 , 1 ) \left(q+1,4,1\right) ( q + 1 , 4 , 1 ) for q ≡ ± 1 ( m o d 8 ) q \equiv\pm1 \pmod{8} q ≡ ± 1 ( mod 8 ) and q ≡ 1 , 3 ( m o d 6 ) q \equiv1,3 \pmod{6} q ≡ 1 , 3 ( mod 6 ) , and 3 3 3 -( q + 1 , 6 , 1 ) \left(q+1,6,1\right) ( q + 1 , 6 , 1 ) for q ≡ 1 ( m o d 20 ) q \equiv1 \pmod{20} q ≡ 1 ( mod 20 ) , attached to the exceptional maximal subgroups S 4 \mathrm{S}_4 S 4 and S 5 \mathrm{S}_5 S 5 of P G L 2 ( q ) \mathrm{PGL}_2(q) PGL 2 ( q ) ;
the five Witt designs 4 4 4 -( 11 , 5 , 1 ) \left(11,5,1\right) ( 11 , 5 , 1 ) , 5 5 5 -( 12 , 6 , 1 ) \left(12,6,1\right) ( 12 , 6 , 1 ) , 3 3 3 -( 22 , 6 , 1 ) \left(22,6,1\right) ( 22 , 6 , 1 ) , 4 4 4 -( 23 , 7 , 1 ) \left(23,7,1\right) ( 23 , 7 , 1 ) , 5 5 5 -( 24 , 8 , 1 ) \left(24,8,1\right) ( 24 , 8 , 1 ) .
In particular no non-complete block-primitive design of strength t ≥ 6 t \ge6 t ≥ 6 exists. The full statement, with block counts and automorphism groups, is Theorem 6.3 , and its proof is in § 4.2 .
§ 2.1 proves that block-primitivity implies point-quasiprimitivity (Lemma 2.1 ) and settles every O’Nan–Scott type other than the almost simple one. § 2.4 excludes point-imprimitive actions by pairing block and point types. § 3.1 classifies the almost simple case: the 2 2 2 -homogeneous actions by the orbit construction, the others family by family, with the ordering principle ∣ H ∣ > ∣ M ∣ |H| > |M| ∣ H ∣ > ∣ M ∣ , the Aschbacher subdivision and the maximal-subgroup data. § 4.1 states the main theorem for 2 2 2 -designs, § 4.2 classifies those of strength t ≥ 3 t \ge3 t ≥ 3 , § 4.3 completes the reduction of the remaining types, and § 6.1 collects the concluding remarks.
Notation and terminology follow [24 ] .
The eight types of Table 2 are mutually exclusive and exhaust all finite quasiprimitive permutation groups, by the O’Nan–Scott theorem [31 ] .
Part II: Structural reduction
The O'Nan–Scott reduction
In this section we prove the structural reduction of non-symmetric block-primitive 2-designs: block-primitivity forces point-quasiprimitivity, and every O’Nan–Scott type other than the almost simple type is excluded or reduced to an explicit algebraic feasibility problem. The affine, holomorph, diagonal, compound diagonal, twisted wreath and product action types are treated in turn; the necessary conditions for the one-sided (point-imprimitive) simple diagonal case, Theorem 2.3 , are derived here and applied to the pairwise analysis of § 2.4 . Point-imprimitive point actions are excluded in § 2.4 , and the surviving almost simple case is classified in § 3.1 .
Lemma 2.1 (Block-primitivity implies point-quasiprimitivity). Let D = ( P , B ) \mathcal{D}=(\mathcal{P},\mathcal{B}) D = ( P , B ) be a finite non-trivial 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design with 2 ≤ k < v 2 \le k < v 2 ≤ k < v and λ > 0 \lambda> 0 λ > 0 , and let G ≤ Aut ( D ) G \le\operatorname{Aut}(\mathcal{D}) G ≤ Aut ( D ) act primitively on B \mathcal{B} B . The block action G B G^{\mathcal{B}} G B is faithful (by [15 Lemma 3.2] ), and G P G^{\mathcal{P}} G P acts quasiprimitively on P \mathcal{P} P .
Proof of Lemma 2.1. First note that a subgroup L ≤ G L \le G L ≤ G transitive on B \mathcal{B} B is transitive on P \mathcal{P} P . Indeed, take a non-empty point-orbit Ω \Omega Ω of L L L on P \mathcal{P} P . Since L L L preserves Ω \Omega Ω and is transitive on blocks, there exists a constant c c c such that ∣ B ∩ Ω ∣ = c |B \cap\Omega| = c ∣ B ∩ Ω∣ = c for all B ∈ B B \in\mathcal{B} B ∈ B . With A A A the point-block incidence matrix and 1 Ω \mathbf{1}_{\Omega} 1 Ω , 1 P \mathbf{1}_{\mathcal{P}} 1 P , 1 B \mathbf{1}_{\mathcal{B}} 1 B the characteristic vectors of Ω \Omega Ω , P \mathcal{P} P , B \mathcal{B} B , respectively, we have
A T 1 Ω = c 1 B , A T 1 P = k 1 B . A^{\mathsf{T}}\mathbf{1}_{\Omega}=c\mathbf{1}_{\mathcal{B}},\qquad A^{\mathsf{T}}\mathbf{1}_{\mathcal{P}}=k\mathbf{1}_{\mathcal{B}}. A T 1 Ω = c 1 B , A T 1 P = k 1 B .
The fundamental design identity A A T = ( r − λ ) I v + λ J v AA^{\mathsf{T}}=(r-\lambda)I_v+\lambda J_v A A T = ( r − λ ) I v + λ J v , together with r > λ r>\lambda r > λ (as r ( k − 1 ) = λ ( v − 1 ) r(k-1)=\lambda(v-1) r ( k − 1 ) = λ ( v − 1 ) and k < v k<v k < v ), shows that A A A has full row rank, so A T A^{\mathsf{T}} A T is injective. Hence 1 Ω = ( c / k ) 1 P \mathbf{1}_{\Omega}=(c/k)\mathbf{1}_{\mathcal{P}} 1 Ω = ( c / k ) 1 P , and since Ω \Omega Ω is non-empty, Ω = P \Omega=\mathcal{P} Ω = P . Thus L L L is point-transitive. To prove quasiprimitivity, take any non-trivial normal subgroup N ⊴ G N\unlhd G N ⊴ G . Since G G G is primitive on B \mathcal{B} B , every non-trivial normal subgroup of G G G is transitive on B \mathcal{B} B (see [24 Proposition 1.5.1] ); hence N N N is block-transitive, and by the first paragraph N N N is transitive on P \mathcal{P} P . Thus every non-trivial normal subgroup of G G G is transitive on P \mathcal{P} P , which means that G P G^{\mathcal{P}} G P acts quasiprimitively on P \mathcal{P} P . Faithfulness of G B G^{\mathcal{B}} G B : if g ∈ G g\in G g ∈ G fixes every block, then g ( B ) = B g(B)=B g ( B ) = B for every B ∈ B B\in\mathcal{B} B ∈ B , so for every point i i i and block B B B the incidence entries satisfy A i , B = A g ( i ) , B A_{i,B}=A_{g(i),B} A i , B = A g ( i ) , B ; the rows i i i and g ( i ) g(i) g ( i ) of the incidence matrix A A A therefore coincide. Since A A A has full row rank v v v (as shown above), its rows are linearly independent, whence g ( i ) = i g(i)=i g ( i ) = i for every point i i i , and g = 1 g=1 g = 1 . Hence G B G^{\mathcal{B}} G B is faithful; this supplies a self-contained proof of [15 Lemma 3.2] . ∎
Note. The argument above uses only the non-triviality condition 2 ≤ k < v 2 \le k < v 2 ≤ k < v and λ > 0 \lambda> 0 λ > 0 . It does not require the design to be non-symmetric, nor does it assume point-transitivity a priori. The conclusion holds for any block-primitive automorphism group of a non-trivial 2-design.
Simple diagonal type (SD). Suppose that both the point action and the block action are of SD type, with socle S = T ℓ S = T^{\ell} S = T ℓ . In SD type, the point stabilizer K = G α K = G_{\alpha} K = G α intersects S S S in a diagonal subgroup D ≅ T D \cong T D ≅ T , so
v = ∣ S : D ∣ = ∣ T ∣ ℓ − 1 . v = |S : D| = |T|^{\ell-1}. v = ∣ S : D ∣ = ∣ T ∣ ℓ − 1 .
Similarly, if the block action is also of SD type, then
b = ∣ T ∣ ℓ − 1 . b = |T|^{\ell-1}. b = ∣ T ∣ ℓ − 1 .
Thus v = b v = b v = b , and the design is symmetric. However, we are only considering non-symmetric designs, so the double-SD case is excluded.
Theorem 2.3 (Necessary conditions for one-sided SD type). Let D \mathcal{D} D be a non-symmetric 2-design with 2 < k < v − 1 2 < k < v - 1 2 < k < v − 1 admitting a block-primitive, point-quasiprimitive automorphism group G G G whose block action is of simple diagonal type. Let N = Soc ( G ) = T ℓ N = \operatorname{Soc}(G) = T^{\ell} N = Soc ( G ) = T ℓ and q = ∣ T ∣ q = |T| q = ∣ T ∣ . Then N N N is transitive on the block set and G B ∩ N ≅ T G_B \cap N \cong T G B ∩ N ≅ T is a diagonal subgroup, so
b = [ N : N ∩ G B ] = q ℓ / q = q ℓ − 1 . b = [N : N \cap G_B] = q^{\ell}/q = q^{\ell-1}. b = [ N : N ∩ G B ] = q ℓ / q = q ℓ − 1 . (i) Since N N N is a non-trivial normal subgroup of G G G and G G G is quasiprimitive on P \mathcal{P} P , N N N is transitive on P \mathcal{P} P , whence
v = [ N : N ∩ G α ] = ∣ N ∣ = q ℓ . v = [N : N \cap G_{\alpha}] = |N| = q^{\ell}. v = [ N : N ∩ G α ] = ∣ N ∣ = q ℓ . In particular, every prime divisor of v v v divides q q q , so gcd ( v − 1 , q ℓ − 1 ) = 1 \gcd(v-1,q^{\ell-1}) = 1 g cd( v − 1 , q ℓ − 1 ) = 1 .
(ii) The design equation v r = b k vr=bk v r = bk (with r r r the replication number) gives r = b k / v = q ℓ − 1 k / v r=bk/v=q^{\ell-1}k/v r = bk / v = q ℓ − 1 k / v , and λ ( v − 1 ) = r ( k − 1 ) \lambda(v-1)=r(k-1) λ ( v − 1 ) = r ( k − 1 ) gives
λ = q ℓ − 1 k ( k − 1 ) v ( v − 1 ) . \lambda=\frac{q^{\ell-1}k(k-1)}{v(v-1)}. λ = v ( v − 1 ) q ℓ − 1 k ( k − 1 ) .
Since v ∣ q ℓ v\mid q^\ell v ∣ q ℓ , we have gcd ( v − 1 , q ℓ − 1 ) = 1 \gcd(v-1,q^{\ell-1})=1 g cd( v − 1 , q ℓ − 1 ) = 1 , and the integrality of λ \lambda λ forces
v − 1 ∣ k ( k − 1 ) , 3 ≤ k ≤ v − 2 , v-1\mid k(k-1),\qquad3\leq k\leq v-2, v − 1 ∣ k ( k − 1 ) , 3 ≤ k ≤ v − 2 ,
where the bounds on k k k follow from the standing assumption 2 < k < v − 1 2<k<v-1 2 < k < v − 1 of the paper.
(iii) If v − 1 v-1 v − 1 is prime, then v − 1 ∣ k ( k − 1 ) v-1\mid k(k-1) v − 1 ∣ k ( k − 1 ) forces k ≡ 0 k\equiv0 k ≡ 0 or 1 ( m o d v − 1 ) 1\pmod{v-1} 1 ( mod v − 1 ) ; but 3 ≤ k ≤ v − 2 < v − 1 3\leq k\leq v-2<v-1 3 ≤ k ≤ v − 2 < v − 1 , and no integer in [ 3 , v − 2 ] [3,v-2] [ 3 , v − 2 ] is congruent to 0 0 0 or 1 1 1 modulo v − 1 v-1 v − 1 . Hence no such k k k exists: no non-trivial block of any kind (single- or multiple-orbit) occurs. More generally, for fixed v v v , the congruence forces k k k to lie in one of at most 2 ω ( v − 1 ) 2^{\omega(v-1)} 2 ω ( v − 1 ) CRT residue classes modulo the prime-power factors of v − 1 v-1 v − 1 ; intersecting with [ 3 , v − 2 ] [3,v-2] [ 3 , v − 2 ] makes the possible k k k explicitly enumerable.
In the single-orbit case k ∣ ∣ G B ∣ = q ∣ K ∣ k\mid|G_B|=q|K| k ∣ ∣ G B ∣ = q ∣ K ∣ as well, which together with (ii) is a finite check. The multiple-orbit case (k = ∑ i k i k=\sum_i k_i k = ∑ i k i , k i ∣ q ∣ K ∣ k_i\mid q|K| k i ∣ q ∣ K ∣ ) reduces the problem to the enumerable candidates of (iii) together with orbit-structure analysis of G B G_B G B on P P P . Note that the argument requires no structural hypothesis on N ∩ G α N\cap G_\alpha N ∩ G α beyond N N N being transitive on P P P .
Remark 2.4. The conditions above are purely arithmetic and depend on q = ∣ T ∣ q=|T| q = ∣ T ∣ , ℓ \ell ℓ and the point stabilizer order only through v = [ N : N ∩ G α ] v=[N:N\cap G_\alpha] v = [ N : N ∩ G α ] . For example, if v − 1 v-1 v − 1 is prime, or if no CRT combination in (iii) lies in [ 3 , v − 2 ] [3,v-2] [ 3 , v − 2 ] , then no non-symmetric block-primitive 2 2 2 -design of one-sided SD type exists for those parameters. This leaves the arithmetic of v − 1 v-1 v − 1 as the decisive ingredient.
Theorem 2.5 (Exclusion of SD/CD block type for λ = 1 \lambda=1 λ = 1 designs). When λ = 1 \lambda=1 λ = 1 , the design is a finite linear space (block = line). Any non-trivial 2 2 2 -( v , k , 1 ) (v,k,1) ( v , k , 1 ) design with a block-primitive automorphism group is of almost simple type: Soc ( G ) = T \operatorname{Soc}(G)=T Soc ( G ) = T for some non-abelian simple group T T T , and G G G acts primitively on blocks with almost simple socle.
Proof. The reduction of block-transitive linear spaces to the almost simple case is the one of Camina and Siemons [14 ] (block-transitive automorphism groups of 2 2 2 -( v , k , 1 ) (v,k,1) ( v , k , 1 ) designs); see also the discussion of linear spaces in § 1.1 . Since SD and CD types require Soc ( G ) = T ℓ \operatorname{Soc}(G)=T^{\ell} Soc ( G ) = T ℓ with ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 , a non-trivial block-primitive 2 2 2 -( v , k , 1 ) (v,k,1) ( v , k , 1 ) design cannot have SD or CD block action type. This also excludes the “one-sided SD” case, as long as “one-sided SD” refers to the block action being of primitive SD type.
Compound diagonal type (CD). Suppose that both the point action and the block action are of CD type, with socle S = T k S=T^k S = T k , where T T T is a non-abelian simple group. In CD type, the k k k simple factors are partitioned into r r r groups of size s s s (so k = r s k=rs k = r s ), and the point stabilizer K = G α K=G_{\alpha} K = G α intersects S S S in a product of r r r diagonal subgroups, each diagonal across one group of s s s factors. Thus
∣ S α ∣ = ∣ T ∣ r , v = ∣ S : S α ∣ = ∣ T ∣ k − r . |S_{\alpha}|=|T|^r,\qquad v=|S:S_{\alpha}|=|T|^{k-r}. ∣ S α ∣ = ∣ T ∣ r , v = ∣ S : S α ∣ = ∣ T ∣ k − r .
Similarly, if the block action is also of CD type, let its CD parameters be k = r ′ s ′ k=r's' k = r ′ s ′ . Then
∣ S B ∣ = ∣ T ∣ r ′ , b = ∣ S : S B ∣ = ∣ T ∣ k − r ′ . |S_B|=|T|^{r'},\qquad b=|S:S_B|=|T|^{k-r'}. ∣ S B ∣ = ∣ T ∣ r ′ , b = ∣ S : S B ∣ = ∣ T ∣ k − r ′ .
Since the design is non-symmetric, b > v b>v b > v , and hence ∣ T ∣ k − r ′ > ∣ T ∣ k − r |T|^{k-r'}>|T|^{k-r} ∣ T ∣ k − r ′ > ∣ T ∣ k − r , which gives r ′ < r r'<r r ′ < r . This is consistent with Fisher’s inequality ∣ S B ∣ < ∣ S α ∣ |S_B|<|S_{\alpha}| ∣ S B ∣ < ∣ S α ∣ , but it is not a contradiction.
The design parameters must also satisfy
b w = v ρ , ρ ( w − 1 ) = λ ( v − 1 ) , bw=v\rho,\qquad\rho(w-1)=\lambda(v-1), b w = v ρ , ρ ( w − 1 ) = λ ( v − 1 ) ,
where w w w is the block size and ρ \rho ρ is the replication number. These divisibility conditions do not by themselves exclude all cases with r ′ < r r'<r r ′ < r .
Theorem 2.7 (Exclusion of CD type in the minimal model). Consider the minimal CD case with n = 6 n=6 n = 6 , where the point action has parameters ( r , s ) = ( 3 , 2 ) (r,s)=(3,2) ( r , s ) = ( 3 , 2 ) (three binary diagonal subgroups H 0 = D 12 × D 34 × D 56 H_0=D_{12}\times D_{34}\times D_{56} H 0 = D 12 × D 34 × D 56 ) and the block action has parameters ( r ′ , s ′ ) = ( 2 , 3 ) (r',s')=(2,3) ( r ′ , s ′ ) = ( 2 , 3 ) (two ternary diagonal subgroups M 0 = D 123 × D 456 M_0=D_{123}\times D_{456} M 0 = D 123 × D 456 ). Then v = ∣ T ∣ 3 v=|T|^3 v = ∣ T ∣ 3 and b = ∣ T ∣ 4 b=|T|^4 b = ∣ T ∣ 4 .
Proof. We show that no non-trivial 2-design can arise from a single double-coset orbit H 0 g M 0 H_0gM_0 H 0 g M 0 in this model. Let q = ∣ T ∣ q=|T| q = ∣ T ∣ , and let k g = [ M 0 : H 0 ∩ M 0 g ] k_g=[M_0:H_0\cap M_0^g] k g = [ M 0 : H 0 ∩ M 0 g ] be the block size. By Lagrange’s theorem, since ∣ M 0 ∣ = q 2 |M_0|=q^2 ∣ M 0 ∣ = q 2 , we have
k g ∣ q 2 , 1 ≤ k g ≤ q 2 . k_g\mid q^2,\qquad1\leq k_g\leq q^2. k g ∣ q 2 , 1 ≤ k g ≤ q 2 .
From the design equation b k = v r bk=vr bk = v r , we get r = q k g r=qk_g r = q k g . The 2-design condition λ ( v − 1 ) = r ( k − 1 ) \lambda(v-1)=r(k-1) λ ( v − 1 ) = r ( k − 1 ) becomes
λ = q k g ( k g − 1 ) q 3 − 1 . \lambda=\frac{qk_g(k_g-1)}{q^3-1}. λ = q 3 − 1 q k g ( k g − 1 ) .
Since gcd ( q , q 3 − 1 ) = 1 \gcd(q,q^3-1)=1 g cd( q , q 3 − 1 ) = 1 , the divisibility condition requires
q 3 − 1 ∣ k g ( k g − 1 ) . (1) q^3-1\mid k_g(k_g-1). \tag*{(1)} q 3 − 1 ∣ k g ( k g − 1 ) . (1)
Now, since k g ∣ q 2 k_g\mid q^2 k g ∣ q 2 and gcd ( q 2 , q 3 − 1 ) = 1 \gcd(q^2,q^3-1)=1 g cd( q 2 , q 3 − 1 ) = 1 , we have gcd ( k g , q 3 − 1 ) = 1 \gcd(k_g,q^3-1)=1 g cd( k g , q 3 − 1 ) = 1 . Therefore, (*) forces
q 3 − 1 ∣ k g − 1. q^3-1\mid k_g-1. q 3 − 1 ∣ k g − 1.
But for any q ≥ 2 q\geq2 q ≥ 2 , we have k g − 1 < q 2 < q 3 − 1 k_g-1<q^2<q^3-1 k g − 1 < q 2 < q 3 − 1 , so the only possibility is k g = 1 k_g=1 k g = 1 .
This corresponds to H 0 ∩ M 0 g = M 0 H_0\cap M_0^g=M_0 H 0 ∩ M 0 g = M 0 , i.e., block size k = 1 k=1 k = 1 , which is a degenerate design and not a non-trivial 2-design.
Thus, for all g ∈ T 6 ⋊ S 6 g\in T^6\rtimes S_6 g ∈ T 6 ⋊ S 6 , the double-coset H 0 g M 0 H_0gM_0 H 0 g M 0 either fails the divisibility condition (when k g > 1 k_g>1 k g > 1 ) or gives a degenerate block (when k g = 1 k_g=1 k g = 1 ). This argument does not depend on the specific structure of the intersection H 0 ∩ M 0 g H_0\cap M_0^g H 0 ∩ M 0 g , nor on whether g g g is a pure socle element or includes a coordinate permutation. □
Theorem 2.8 (Exclusion of single-orbit CD blocks). Let D \mathcal{D} D be a non-symmetric 2-design admitting a block-primitive, point-quasiprimitive automorphism group G G G whose block action is of compound diagonal type with parameters ( r ′ , s ′ ) (r',s') ( r ′ , s ′ ) , n = r ′ s ′ n=r's' n = r ′ s ′ , and top group K M K_M K M ; assume that the point action is of compound diagonal type with parameters ( r , s ) (r,s) ( r , s ) , n = r s n=rs n = r s , and top group K H K_H K H , with r ′ < r r'<r r ′ < r (the non-symmetric case). Write q = ∣ T ∣ q=|T| q = ∣ T ∣ and
v = a q n − r ( a = n ! / ∣ K H ∣ ) , b = c q n − r ′ ( c = n ! / ∣ K M ∣ ) . v=aq^{n-r}\quad(a=n!/|K_H|),\qquad b=cq^{n-r'}\quad(c=n!/|K_M|). v = a q n − r ( a = n ! /∣ K H ∣ ) , b = c q n − r ′ ( c = n ! /∣ K M ∣ ) . Suppose that the block is given by a single double coset H g M HgM H g M , and let k = [ M : M ∩ H g ] k=[M:M\cap H^g] k = [ M : M ∩ H g ] be the block size, i.e. the length of the orbit of M M M on G / H G/H G / H with representative g H gH g H (up to conjugation this equals [ M : H ∩ M g ] [M:H\cap M^g] [ M : H ∩ M g ] ). Then:
k ∣ ∣ M ∣ = q r ′ ( n ! / c ) k \mid|M| = q^{r'}(n!/c) k ∣ ∣ M ∣ = q r ′ ( n ! / c ) ;
the integrality of the design parameters forces
a ′ ( a q n − r − 1 ) ∣ c ′ q r − r ′ k ( k − 1 ) , a'(a q^{n-r}-1)\mid c' q^{r-r'}k(k-1), a ′ ( a q n − r − 1 ) ∣ c ′ q r − r ′ k ( k − 1 ) ,
where a ′ = a / gcd ( a , c ) a'=a/\gcd(a,c) a ′ = a / g cd( a , c ) and c ′ = c / gcd ( a , c ) c'=c/\gcd(a,c) c ′ = c / g cd( a , c ) .
Proof. Indeed, the replication number is ρ = b k / v = ( c / a ) q r − r ′ k \rho=bk/v=(c/a)q^{r-r'}k ρ = bk / v = ( c / a ) q r − r ′ k , and λ ( v − 1 ) = ρ ( k − 1 ) \lambda(v-1)=\rho(k-1) λ ( v − 1 ) = ρ ( k − 1 ) gives λ = ( c / a ) q r − r ′ k ( k − 1 ) / ( a q n − r − 1 ) ∈ Z \lambda=(c/a)q^{r-r'}k(k-1)/(a q^{n-r}-1)\in\mathbb{Z} λ = ( c / a ) q r − r ′ k ( k − 1 ) / ( a q n − r − 1 ) ∈ Z , which is (ii) after clearing denominators. Condition (ii) is a finite, checkable number-theoretic condition once T T T , n n n , r r r , r ′ r' r ′ , a a a , c c c are fixed: for each prime power p e ∥ a ′ ( a q n − r − 1 ) p^e\parallel a'(a q^{n-r}-1) p e ∥ a ′ ( a q n − r − 1 ) with p ∤ c ′ q r − r ′ p\nmid c'q^{r-r'} p ∤ c ′ q r − r ′ , the residue of k k k modulo p e p^e p e must lie in { 0 , 1 } \{0,1\} { 0 , 1 } ; in general one uses the p p p -adic condition
v p ( k ( k − 1 ) ) ≥ e − v p ( c ′ ) − v p ( q r − r ′ ) , v_p(k(k-1))\ge e-v_p(c')-v_p(q^{r-r'}), v p ( k ( k − 1 )) ≥ e − v p ( c ′ ) − v p ( q r − r ′ ) ,
which is vacuous when the right-hand side is non-positive. Hence k k k is one of finitely many CRT combinations once the parameters are fixed, and each candidate can be checked against (i) and the bounds 2 ≤ k ≤ v − 2 2\le k\le v-2 2 ≤ k ≤ v − 2 . The case p ∣ c ′ p\mid c' p ∣ c ′ (equivalently, prime divisors of c ′ c' c ′ inside a ′ ( a q n − r − 1 ) a'(a q^{n-r}-1) a ′ ( a q n − r − 1 ) ) is covered by the p p p -adic form and does not force the { 0 , 1 } \{0,1\} { 0 , 1 } residues.
For general λ > 1 \lambda> 1 λ > 1 , the O’Nan–Scott theorem applies directly to the block action G ↷ B = G / M G \curvearrowright\mathcal{B} = G/M G ↷ B = G / M , since block-primitivity is equivalent to M M M being a maximal subgroup of G G G . It does not directly apply to the point action G ↷ P = G / H G \curvearrowright\mathcal{P} = G/H G ↷ P = G / H , since the point action is imprimitive.
A general lemma on regular normal subgroups.
Lemma 2.10 (Regular normal subgroup). Let D = ( P , B ) \mathcal{D} = (\mathcal{P},\mathcal{B}) D = ( P , B ) be a non-trivial 2 2 2 -design with 2 < k < v 2 < k < v 2 < k < v , and let G ≤ Aut ( D ) G \leq\operatorname{Aut}(\mathcal{D}) G ≤ Aut ( D ) act primitively on B \mathcal{B} B . If G G G has a normal subgroup N N N acting regularly on P \mathcal{P} P , then N N N acts regularly on B \mathcal{B} B and D \mathcal{D} D is symmetric (b = v b = v b = v ).
Proof. The image of N N N in the primitive action on B \mathcal{B} B is a normal subgroup of that action. It cannot be trivial: otherwise N N N fixes every block, and since N N N is transitive on P \mathcal{P} P , an N N N -invariant block B B B would be an N N N -invariant subset of P \mathcal{P} P , hence of size 0 0 0 or v v v , contradicting 1 < k < v 1 < k < v 1 < k < v for a non-trivial block. Hence N N N is transitive on B \mathcal{B} B . With M = G B M = G_B M = G B , the orbit-stabilizer theorem gives b = [ N : N ∩ M ] = v / ∣ N ∩ M ∣ ≤ v b = [N : N \cap M] = v/|N \cap M| \leq v b = [ N : N ∩ M ] = v /∣ N ∩ M ∣ ≤ v . Fisher’s inequality b ≥ v b \geq v b ≥ v forces b = v b = v b = v , so D \mathcal{D} D is symmetric. □
Holomorph affine type (HA).
Lemma 2.11 (HA type forces symmetry). If the point action of G G G is of holomorph affine type, then b = v b = v b = v ; that is, the design is symmetric. Therefore, non-symmetric block-primitive 2 2 2 -designs cannot have HA type.
Proof. In the holomorph affine type the socle N N N of G G G is elementary abelian and acts regularly on P \mathcal{P} P , so N N N is a normal subgroup of G G G regular on the point set; Lemma 2.10 then gives b = v b = v b = v , that is, the design is symmetric. □
Twisted wreath type (TW). If the point action of G G G is of twisted wreath type, then Soc ( G ) = T ℓ \operatorname{Soc}(G) = T^{\ell} Soc ( G ) = T ℓ is regular on P \mathcal{P} P , so v = ∣ T ∣ ℓ v = |T|^{\ell} v = ∣ T ∣ ℓ . By Lemma 2.10 , a non-trivial block-primitive 2 2 2 -design admitting such a group must be symmetric. Hence no non-symmetric block-primitive 2 2 2 -design arises from a point action of TW type.
For completeness we also record the block-action point of view. If G B G^{B} G B is of twisted wreath type, then Soc ( G ) = T ℓ \operatorname{Soc}(G) = T^{\ell} Soc ( G ) = T ℓ is regular on B \mathcal{B} B , so b = ∣ T ∣ ℓ b = |T|^{\ell} b = ∣ T ∣ ℓ and G B ∩ Soc ( G ) = 1 G_B \cap\operatorname{Soc}(G) = 1 G B ∩ Soc ( G ) = 1 . Since G G G is quasiprimitive on P \mathcal{P} P , Soc ( G ) = T ℓ \operatorname{Soc}(G) = T^{\ell} Soc ( G ) = T ℓ is transitive on P \mathcal{P} P , and v = ∣ T ∣ ℓ / ∣ Soc ( G ) α ∣ v = |T|^{\ell}/|\operatorname{Soc}(G)_{\alpha}| v = ∣ T ∣ ℓ /∣ Soc ( G ) α ∣ . If the point action is also of TW type, then Soc ( G ) \operatorname{Soc}(G) Soc ( G ) is regular on P \mathcal{P} P , so v = b v=b v = b and the design is symmetric, as above. In the remaining case ∣ Soc ( G ) α ∣ > 1 \lvert\operatorname{Soc}(G)_{\alpha}\rvert>1 ∣ Soc ( G ) α ∣ > 1 , the design parameters satisfy
k ∣ Soc ( G ) α ∣ = r , r ( k − 1 ) = λ ( v − 1 ) , k\lvert\operatorname{Soc}(G)_{\alpha}\rvert=r,\qquad r(k-1)=\lambda(v-1), k ∣ Soc ( G ) α ∣ = r , r ( k − 1 ) = λ ( v − 1 ) ,
so that k ∣ Soc ( G ) α ∣ ( k − 1 ) = λ ( ∣ T ∣ ℓ / ∣ Soc ( G ) α ∣ − 1 ) k\lvert\operatorname{Soc}(G)_{\alpha}\rvert(k-1)=\lambda(\lvert T\rvert^{\ell}/\lvert\operatorname{Soc}(G)_{\alpha}\rvert-1) k ∣ Soc ( G ) α ∣ ( k − 1 ) = λ (∣ T ∣ ℓ / ∣ Soc ( G ) α ∣ − 1 ) , and the analysis of this arithmetic is subsumed in the general parameter reduction of the structural reduction theorem (§ 4.1 ). (k = ∣ B ∣ k=\lvert B\rvert k = ∣ B ∣ is not in general a divisor of ∣ G B ∣ \lvert G_B\rvert ∣ G B ∣ .) Now we handle the remaining O’Nan–Scott types for G B G^B G B : HS, HC, and PA.
A semiregularity lemma.
Lemma 2.12 (Commuting transitive groups are regular). Let X X X be a transitive permutation group on a set Ω \Omega Ω , and let Y ≤ Sym ( Ω ) Y\leq\operatorname{Sym}(\Omega) Y ≤ Sym ( Ω ) commute elementwise with X X X . If Y Y Y is transitive on Ω \Omega Ω , then Y Y Y is regular on Ω \Omega Ω .
Proof. If y ∈ Y y\in Y y ∈ Y fixes a point α ∈ Ω \alpha\in\Omega α ∈ Ω , then for every β = x α \beta=x\alpha β = xα with x ∈ X x\in X x ∈ X we have y β = y x α = x y α = x α = β y\beta=yx\alpha=xy\alpha=x\alpha=\beta y β = y xα = x y α = xα = β ; since X X X is transitive, y y y fixes Ω \Omega Ω pointwise, whence y = 1 y=1 y = 1 . Thus Y Y Y is semiregular, and being transitive it is regular.
Holomorph simple type (HS). Suppose that G B G^B G B is of holomorph simple type. Then Soc ( G ) = T 1 × T 2 ≅ T × T \operatorname{Soc}(G)=T_1\times T_2\cong T\times T Soc ( G ) = T 1 × T 2 ≅ T × T with two commuting minimal normal subgroups T 1 , T 2 T_1,T_2 T 1 , T 2 , each regular on B ≅ T B\cong T B ≅ T ; in particular b = ∣ B ∣ = ∣ T ∣ b=\lvert B\rvert=\lvert T\rvert b = ∣ B ∣ = ∣ T ∣ . Since G G G is quasiprimitive on P \mathcal{P} P (by Lemma 2.1 ), both T 1 T_1 T 1 and T 2 T_2 T 2 are transitive on P \mathcal{P} P . Lemma 2.12 , applied to ( X , Y ) = ( T 1 , T 2 ) (X,Y)=(T_1,T_2) ( X , Y ) = ( T 1 , T 2 ) and to ( T 2 , T 1 ) (T_2,T_1) ( T 2 , T 1 ) , shows that T 1 T_1 T 1 and T 2 T_2 T 2 are regular on P \mathcal{P} P ; hence v = ∣ P ∣ = ∣ T 1 ∣ = ∣ T ∣ v=\lvert\mathcal{P}\rvert=\lvert T_1\rvert=\lvert T\rvert v = ∣ P ∣ = ∣ T 1 ∣ = ∣ T ∣ . Thus v = b = ∣ T ∣ v=b=\lvert T\rvert v = b = ∣ T ∣ , contradicting the standing assumption that D \mathcal{D} D is non-symmetric. Therefore G B G^B G B cannot be of HS type for non-symmetric 2-designs. □
Holomorph compound type (HC). Suppose that G B G^B G B is of holomorph compound type. Then the socle of G B G^B G B splits as N 1 × N 2 N_1\times N_2 N 1 × N 2 with N 1 ≅ N 2 ≅ T ℓ / 2 N_1\cong N_2\cong T^{\ell/2} N 1 ≅ N 2 ≅ T ℓ /2 (so ℓ \ell ℓ is even) its two minimal normal subgroups; N 1 N_1 N 1 and N 2 N_2 N 2 commute, and each is regular on B ≅ T ℓ / 2 B\cong T^{\ell/2} B ≅ T ℓ /2 , so b = ∣ B ∣ = ∣ T ∣ ℓ / 2 b=\lvert B\rvert=\lvert T\rvert^{\ell/2} b = ∣ B ∣ = ∣ T ∣ ℓ /2 . By quasiprimitivity of G G G on P \mathcal{P} P , both N 1 N_1 N 1 and N 2 N_2 N 2 are transitive on P \mathcal{P} P , and Lemma 2.12 shows that both are regular on P \mathcal{P} P . Hence v = ∣ P ∣ = ∣ N 1 ∣ = ∣ T ∣ ℓ / 2 = b v=\lvert\mathcal{P}\rvert=\lvert N_1\rvert=\lvert T\rvert^{\ell/2}=b v = ∣ P ∣ = ∣ N 1 ∣ = ∣ T ∣ ℓ /2 = b , again contradicting the non-symmetry of D \mathcal{D} D . Therefore G B G^B G B cannot be of HC type. □
Product action type (PA). Throughout this section, T \mathcal{T} T denotes the set of the six socles of Theorem 5.1 , namely
T = { A 5 , A 6 , PSL 2 ( 7 ) , PSL 2 ( 8 ) , PSL 2 ( 11 ) , PSL 2 ( 13 ) } ; \mathcal{T}=\{A_5,A_6,\operatorname{PSL}_2(7),\operatorname{PSL}_2(8),\operatorname{PSL}_2(11),\operatorname{PSL}_2(13)\}; T = { A 5 , A 6 , PSL 2 ( 7 ) , PSL 2 ( 8 ) , PSL 2 ( 11 ) , PSL 2 ( 13 )} ;
we fix T ∈ T T \in\mathcal{T} T ∈ T , a primitive degree m m m of T T T and a point stabilizer H H H of the action. First we combine the O’Nan–Scott theorem for the primitive block action with the product structure of the point action, and record that the block action is necessarily of product action type. Since G G G is primitive on B \mathcal{B} B , it is quasiprimitive on B \mathcal{B} B , so the socle Soc ( G ) = T ℓ \operatorname{Soc}(G) = T^{\ell} Soc ( G ) = T ℓ is transitive on B \mathcal{B} B . The possible O’Nan–Scott types for G B G^{\mathcal{B}} G B are then excluded as follows: the affine type is impossible since Soc ( G ) = T ℓ \operatorname{Soc}(G) = T^{\ell} Soc ( G ) = T ℓ is not elementary abelian; the almost simple type is impossible since Soc ( G ) \operatorname{Soc}(G) Soc ( G ) is not simple for ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 ; the holomorph simple and holomorph compound types are impossible since they require Soc ( G ) \operatorname{Soc}(G) Soc ( G ) to split into two minimal normal subgroups, each regular on B \mathcal{B} B , whereas Soc ( G ) = T ℓ \operatorname{Soc}(G) = T^{\ell} Soc ( G ) = T ℓ has a unique minimal normal subgroup; the simple diagonal type is excluded for ℓ = 2 \ell= 2 ℓ = 2 by Lemma 2.14 below (the diagonal socle-orbit through a point is never a union of point-stabilizer boxes, and the remaining D D D -orbits avoid the point, so no non-trivial block through a point exists; the two non-2-homogeneous cases are verified computationally); the compound diagonal type for ℓ ≥ 3 \ell\ge3 ℓ ≥ 3 is excluded by Proposition 2.15 below; the twisted wreath type is excluded as in § 2.1 : Soc ( G ) \operatorname{Soc}(G) Soc ( G ) would be regular on B \mathcal{B} B , giving b = ∣ T ∣ ℓ b = |T|^{\ell} b = ∣ T ∣ ℓ , and with the point action in product form v = m ℓ v = m^{\ell} v = m ℓ the design equations reduce to the arithmetic of § 4.1 , which leaves no survivor (cf. the remark on the TW type in § 2.1 ). Hence G B G^{\mathcal{B}} G B is of product action type. In the product action on points the block stabilizer meets the socle in Soc ( G ) ∩ G B = K ℓ \operatorname{Soc}(G) \cap G_{\mathcal{B}} = K^{\ell} Soc ( G ) ∩ G B = K ℓ , where K K K is a proper subgroup of T T T with n = [ T : K ] ≥ 2 n = [T : K] \ge2 n = [ T : K ] ≥ 2 . Transitivity of Soc ( G ) \operatorname{Soc}(G) Soc ( G ) on B \mathcal{B} B forces G = Soc ( G ) G B G = \operatorname{Soc}(G)G_{\mathcal{B}} G = Soc ( G ) G B ; in the product action model below this means that the coordinate part of G B G_{\mathcal{B}} G B is the full coordinate group, and consequently the number of blocks is
b = [ Soc ( G ) : Soc ( G ) ∩ G B ] = n ℓ , n = [ T : K ] . b = [\operatorname{Soc}(G) : \operatorname{Soc}(G) \cap G_{\mathcal{B}}] = n^{\ell}, \qquad n = [T : K]. b = [ Soc ( G ) : Soc ( G ) ∩ G B ] = n ℓ , n = [ T : K ] .
Here n n n is the block-coordinate index , to be distinguished from the point-coordinate index m = [ T : H ] m = [T : H] m = [ T : H ] ; the number of points is v = m ℓ v = m^{\ell} v = m ℓ , while the number of blocks is b = n ℓ b = n^{\ell} b = n ℓ ; by Fisher’s inequality n ℓ = b ≥ v = m ℓ n^{\ell} = b \ge v = m^{\ell} n ℓ = b ≥ v = m ℓ , i.e. n ≥ m n \ge m n ≥ m , a restriction satisfied by all listed pairs.
Corollary 2.13 (Block-primitivity forces point-primitivity in the product action). With the notation above, G B G^{\mathcal{B}} G B is primitive of product action type, so by the primitivity criterion for product actions the coordinate permutation group P P P is primitive on { 1 , … , ℓ } \{1,\ldots,\ell\} { 1 , … , ℓ } . Since H H H is maximal in T T T , the point action of G G G on P ≅ Ω ℓ P \cong\Omega^{\ell} P ≅ Ω ℓ is therefore primitive of product action type as well. In the product-action model of this section the point action is assumed of
product action type with primitive coordinate degree m m m (the c = 1 c = 1 c = 1 branch of the quasiprimitive product action, where the point stabilizer meets the socle in H ℓ H^{\ell} H ℓ with H H H maximal, so that the point action is indeed primitive); the point-imprimitive product action, of degree v = c ω ℓ v = c\omega^{\ell} v = c ω ℓ with c ≥ 1 c \ge1 c ≥ 1 the block size of the G G G -invariant partition (Praeger type (viii); cf. Example 1.1 of Zhang, Chen and Zhou [42 ] ), is excluded unconditionally in Lemma 3.11 .
Since G G G is quasiprimitive on P \mathcal{P} P , Soc ( G ) = T ℓ \operatorname{Soc}(G) = T^{\ell} Soc ( G ) = T ℓ is transitive on P \mathcal{P} P , and v = m ℓ ≤ b = n ℓ v = m^{\ell} \le b = n^{\ell} v = m ℓ ≤ b = n ℓ by Fisher’s inequality as recorded above.
Lemma 2.14 (No simple-diagonal block action under a product point action). Let T ∈ T T \in\mathcal{T} T ∈ T , let m m m be a primitive degree of T T T and let G = T 2 ⋊ C 2 G = T^{2} \rtimes C_{2} G = T 2 ⋊ C 2 act in product action on P = Δ 2 \mathcal{P} = \Delta^{2} P = Δ 2 , ∣ Δ ∣ = m |\Delta| = m ∣Δ∣ = m , with point stabilizer G α = H 2 ⋊ C 2 G_{\alpha} = H^{2} \rtimes C_{2} G α = H 2 ⋊ C 2 (H H H maximal in T T T , [ T : H ] = m [T : H] = m [ T : H ] = m , C 2 C_{2} C 2 the coordinate swap). If the block action of G G G on B \mathcal{B} B is of simple diagonal type, i.e. Soc ( G ) ∩ G B = D = { ( t , t ) : t ∈ T } \operatorname{Soc}(G) \cap G_{B} = D = \{(t,t) : t \in T\} Soc ( G ) ∩ G B = D = {( t , t ) : t ∈ T } , then no non-trivial block B ∋ α B \ni\alpha B ∋ α exists.
Proof. Since Soc ( G ) = T 2 \operatorname{Soc}(G) = T^{2} Soc ( G ) = T 2 is normal in G G G , the diagonal subgroup D = Soc ( G ) ∩ G B D = \operatorname{Soc}(G) \cap G_{B} D = Soc ( G ) ∩ G B is normal in G B G_{B} G B ; hence every block B ∋ α B \ni\alpha B ∋ α is a union of D D D -orbits. The orbit of D D D through x = ( x 1 , x 2 ) x = (x_{1},x_{2}) x = ( x 1 , x 2 ) is the diagonal line L x = { ( t x 1 , t x 2 ) : t ∈ T } L_{x} = \{(tx_{1},tx_{2}) : t \in T\} L x = {( t x 1 , t x 2 ) : t ∈ T } , and the orbit sizes of D D D on P = Δ 2 \mathcal{P} = \Delta^{2} P = Δ 2 are recorded in Table 3 ; in particular the orbit of α \alpha α has size m m m . Thus k = ∣ B ∣ k = |B| k = ∣ B ∣ is a sum of orbit sizes containing the entry m m m . Three arithmetic obstructions apply. (i) Design equations: r = b k / v r = bk/v r = bk / v and λ = r ( k − 1 ) / ( v − 1 ) \lambda= r(k - 1)/(v - 1) λ = r ( k − 1 ) / ( v − 1 ) are integers, where b = ∣ B ∣ = ∣ T ∣ b = |B| = |T| b = ∣ B ∣ = ∣ T ∣ and v = m 2 v = m^{2} v = m 2 . (ii) Fisher’s inequality: b ≥ v b \ge v b ≥ v , i.e. ∣ T ∣ ≥ m 2 |T| \ge m^{2} ∣ T ∣ ≥ m 2 . (iii) The Cameron–Praeger criterion Theorem 2.19: with Q d Q_{d} Q d the orbit of G G G on unordered point pairs at Hamming distance d d d (d = 1 , 2 d = 1,2 d = 1 , 2 ), the number of Q d Q_{d} Q d -pairs inside a block is q d = ∣ Q d ∣ ( k 2 ) / ( v 2 ) q_{d} = |Q_{d}| \binom{k}{2}/\binom{v}{2} q d = ∣ Q d ∣ ( 2 k ) / ( 2 v ) ; since ∣ Q d ∣ = ( v 2 ) ( m − 1 ) d / 2 |Q_{d}| = \binom{v}{2}(m - 1)^{d}/2 ∣ Q d ∣ = ( 2 v ) ( m − 1 ) d /2 ,
Table 3. Orbit sizes of the diagonal subgroup D = S o c ( G ) ∩ G B D = Soc(G) \cap G_B D = S oc ( G ) ∩ G B in the product action on Δ 2 \Delta^2 Δ 2 , and the number of block sizes surviving the sieve of Lemma 2.14.
q d = ( d 2 ) ( m − 1 ) d k ( k − 1 ) 2 ( v − 1 ) ∈ Z , d = 1 , 2. (2) q_{d} = \frac{\binom{d}{2}(m - 1)^{d} k(k - 1)}{2(v - 1)} \in\mathbb{Z}, \qquad d = 1,2. \tag*{(2)} q d = 2 ( v − 1 ) ( 2 d ) ( m − 1 ) d k ( k − 1 ) ∈ Z , d = 1 , 2. (2)
For all seventeen primitive degrees of Table 3 , GAP 4.16.1 runs the three conditions over every sum of the orbit sizes containing the entry m m m and finds no admissible block size: the survivor count is zero for every pair. Hence no non-trivial block exists, and the simple diagonal block action is excluded for ℓ = 2 \ell= 2 ℓ = 2 .
Proposition 2.15 (No compound-diagonal block action under a product point action). Let T ∈ T T \in\mathcal{T} T ∈ T , let m m m be a primitive degree of T T T and let G G G act in product action on P = Δ ℓ \mathcal{P} = \Delta^\ell P = Δ ℓ , ∣ Δ ∣ = m |\Delta| = m ∣Δ∣ = m , ℓ ≥ 3 \ell\ge3 ℓ ≥ 3 , with point stabilizer G α = H ℓ ⋊ C ℓ G_\alpha= H^\ell\rtimes C_\ell G α = H ℓ ⋊ C ℓ . Then the block action of G G G cannot be of compound diagonal type : no non-trivial block B ∋ α B \ni\alpha B ∋ α exists for any ℓ ≥ 3 \ell\ge3 ℓ ≥ 3 and any compound diagonal stabilizer S o c ( G ) ∩ G B Soc(G) \cap G_B S oc ( G ) ∩ G B .
Proof. Since S o c ( G ) = T ℓ Soc(G) = T^\ell S oc ( G ) = T ℓ is normal in G G G , the intersection D r ′ = S o c ( G ) ∩ G B D_{r'} = Soc(G) \cap G_B D r ′ = S oc ( G ) ∩ G B is normal in G B G_B G B ; hence every block B ∋ α B \ni\alpha B ∋ α is a union of D r ′ D_{r'} D r ′ -orbits. After relabelling, D r ′ D_{r'} D r ′ acts diagonally on each block of r ′ r' r ′ coordinates, and the sizes of the D r ′ D_{r'} D r ′ -orbits are the orbit sizes of T T T acting diagonally on Δ r ′ \Delta^{r'} Δ r ′ ; for r ′ = 1 r' = 1 r ′ = 1 the unique orbit size is m m m , and for r ′ = 2 r' = 2 r ′ = 2 the orbit sizes are the entries of Table 3 . In particular the orbit of α \alpha α has size m m m , so k = ∣ B ∣ k = |B| k = ∣ B ∣ is a sum of such orbit sizes containing the entry m m m . Also v = m ℓ v = m^\ell v = m ℓ and b = ∣ B ∣ = ∣ T ∣ ℓ − r ′ b = |B| = |T|^{\ell-r'} b = ∣ B ∣ = ∣ T ∣ ℓ − r ′ . The three arithmetic obstructions of Lemma 2.14 apply verbatim: the design equations r = b k / v ∈ Z r = bk/v \in\mathbb{Z} r = bk / v ∈ Z and λ = r ( k − 1 ) / ( v − 1 ) ∈ Z \lambda= r(k - 1)/(v - 1) \in\mathbb{Z} λ = r ( k − 1 ) / ( v − 1 ) ∈ Z ; Fisher’s inequality b ≥ v b \ge v b ≥ v ; and the Cameron–Praeger integrality
q d = ( ℓ d ) ( m − 1 ) d k ( k − 1 ) 2 ( v − 1 ) ∈ Z , d = 1 , … , ℓ , (3) q_d=\frac{\binom{\ell}{d}(m-1)^d k(k-1)}{2(v-1)}\in\mathbb{Z},\qquad d=1,\ldots,\ell, \tag*{(3)} q d = 2 ( v − 1 ) ( d ℓ ) ( m − 1 ) d k ( k − 1 ) ∈ Z , d = 1 , … , ℓ , (3)
with ∣ Q d ∣ = v ( ℓ d ) ( m − 1 ) d / 2 |Q_d|=v^{\binom{\ell}{d}}(m-1)^d/2 ∣ Q d ∣ = v ( d ℓ ) ( m − 1 ) d /2 . The range 3 ≤ ℓ ≤ 9 3\le\ell\le9 3 ≤ ℓ ≤ 9 . For all seventeen primitive degrees of Table 3 , all r ′ = 1 , … , ℓ − 1 r'=1,\ldots,\ell-1 r ′ = 1 , … , ℓ − 1 and all 3 ≤ ℓ ≤ 9 3\le\ell\le9 3 ≤ ℓ ≤ 9 , GAP 4.16.1 evaluates the three conditions over every sum of the D r ′ D_{r'} D r ′ -orbit sizes containing m m m and finds no admissible block size: the survivor count is zero for every pair (scripts gap_cd_sieve.g, gap_cd_ell789.g). The range ℓ ≥ 10 \ell\ge10 ℓ ≥ 10 . Write t t t for the number of orbits of H H H on Δ \Delta Δ (t ≤ 3 t\le3 t ≤ 3 for the six socles); the number s r ′ s_{r'} s r ′ of D r ′ D_{r'} D r ′ -orbits on Δ r ′ \Delta^{r'} Δ r ′ satisfies s r ′ ≤ m r ′ s_{r'}\le m^{r'} s r ′ ≤ m r ′ , and each D r ′ D_{r'} D r ′ -orbit has size at most ∣ T ∣ |T| ∣ T ∣ , so k ≤ m r ′ ∣ T ∣ k\le m^{r'}|T| k ≤ m r ′ ∣ T ∣ . The D r ′ D_{r'} D r ′ -orbit L α L_\alpha L α of α \alpha α has size m m m and contributes m − 1 m-1 m − 1 pairs at Hamming distance r ′ r' r ′ , so (2) with d = r ′ d=r' d = r ′ gives the lower bound
( ℓ r ′ ) ( m − 1 ) r ′ k ( k − 1 ) 2 ( m ℓ − 1 ) ≥ m − 1. (4) \frac{\binom{\ell}{r'}(m-1)^{r'}k(k-1)}{2(m^\ell-1)}\ge m-1. \tag*{(4)} 2 ( m ℓ − 1 ) ( r ′ ℓ ) ( m − 1 ) r ′ k ( k − 1 ) ≥ m − 1. (4)
Two general facts now confine ℓ \ell ℓ to a finite range. First, (4) with k ≤ m r ′ ∣ T ∣ k\le m^{r'}|T| k ≤ m r ′ ∣ T ∣ fails once
( ℓ r ′ ) ( m − 1 ) r ′ m 2 r ′ ∣ T ∣ 2 < 2 ( m − 1 ) ( m ℓ − 1 ) , \binom{\ell}{r'}(m-1)^{r'}m^{2r'}|T|^2<2(m-1)(m^\ell-1), ( r ′ ℓ ) ( m − 1 ) r ′ m 2 r ′ ∣ T ∣ 2 < 2 ( m − 1 ) ( m ℓ − 1 ) ,
which is a polynomial-versus-exponential inequality that fails for all ℓ \ell ℓ beyond an explicit bound. Second, by Zsigmondy’s theorem [44 ] m ℓ − 1 m^\ell-1 m ℓ − 1 has a primitive prime divisor p ℓ ≡ 1 ( m o d ℓ ) p_\ell\equiv1\pmod{\ell} p ℓ ≡ 1 ( mod ℓ ) with p ℓ ∤ m − 1 p_\ell\nmid m-1 p ℓ ∤ m − 1 ; as ( ℓ r ′ ) \binom{\ell}{r'} ( r ′ ℓ ) has no prime divisor exceeding ℓ < p ℓ \ell<p_\ell ℓ < p ℓ , the integrality of q r ′ q_{r'} q r ′ forces p ℓ ∣ k ( k − 1 ) p_\ell\mid k(k-1) p ℓ ∣ k ( k − 1 ) , and since k ( k − 1 ) ≤ m 2 r ′ ∣ T ∣ 2 k(k-1)\le m^{2r'}|T|^2 k ( k − 1 ) ≤ m 2 r ′ ∣ T ∣ 2 is bounded while p ℓ ≥ ℓ + 1 p_\ell\ge\ell+1 p ℓ ≥ ℓ + 1 , this again fails for all ℓ \ell ℓ beyond an explicit bound. Combining the two bounds with Fisher’s inequality leaves, for every pair and every r ′ r' r ′ , only an explicitly bounded range of ℓ \ell ℓ ; GAP 4.16.1 checks these residual ranges for the seventeen primitive degrees (scripts gap_cd_sieve.g, gap_cd_ell789.g, gap_cd_fisher.g) and reports zero survivors in every case. This completes the exclusion of the compound diagonal block action for all ℓ ≥ 3 \ell\ge3 ℓ ≥ 3 .
Step 1: Permutation-module inclusion. The incidence matrix of D \mathcal{D} D induces an embedding of permutation modules R ( P ) ↪ R ( B ) R(\mathcal{P})\hookrightarrow R(\mathcal{B}) R ( P ) ↪ R ( B ) : every irreducible constituent of the permutation character on P \mathcal{P} P must appear in the permutation character on B \mathcal{B} B (a block-transitive 2-design gives such an inclusion). In PA type the permutation character on B \mathcal{B} B is the ℓ \ell ℓ -th tensor power of the permutation character of H H H on Δ \Delta Δ :
π B = π Δ ⊗ ℓ . \pi_{\mathcal{B}}=\pi_{\Delta}^{\otimes\ell}. π B = π Δ ⊗ ℓ .
This restricts the possible degree v v v and the possible point stabilizer G α G_\alpha G α ; in particular the rank of G G G on P \mathcal{P} P cannot exceed the rank of G G G on B \mathcal{B} B . We use this condition only as a parameter sieve: the inclusion itself does not determine the O’Nan–Scott type of the point action nor the maximality of H H H , which are handled by the orbit-union sieves of Proposition 7.9 and Proposition 7.11 below.
Step 2: Hamming-distance suborbits. In PA action, fix a block B ∈ B B \in\mathcal{B} B ∈ B . The Hamming distance d d d from B B B defines a G B G_B G B -invariant stratification of B \mathcal{B} B , but each Hamming layer Ω d \Omega_d Ω d is not necessarily a single suborbit. If H H H is not 2-transitive, then the distance-1 layer may split into multiple suborbits. The total size of the distance-1 layer is ℓ ( m − 1 ) \ell(m-1) ℓ ( m − 1 ) , but the actual minimal suborbit length may be smaller.
The intersection numbers x i = ∣ B ∩ B ′ ∣ x_i = |B \cap B'| x i = ∣ B ∩ B ′ ∣ must satisfy the moment equations:
∑ i ∣ Δ i ∣ x i = k ( r − 1 ) , ∑ i ∣ Δ i ∣ x i ( x i − 1 ) = k ( k − 1 ) ( λ − 1 ) . \sum_i |\Delta_i|x_i = k(r-1), \qquad\sum_i |\Delta_i|x_i(x_i-1) = k(k-1)(\lambda-1). i ∑ ∣ Δ i ∣ x i = k ( r − 1 ) , i ∑ ∣ Δ i ∣ x i ( x i − 1 ) = k ( k − 1 ) ( λ − 1 ) .
These equations are valid for any block-transitive design and do not require flag-transitivity.
The cases ℓ = 2 \ell= 2 ℓ = 2 and ℓ ≥ 3 \ell\ge3 ℓ ≥ 3 . The cases ℓ = 2 \ell= 2 ℓ = 2 and ℓ ≥ 3 \ell\ge3 ℓ ≥ 3 are covered in full by the sieve of Proposition 7.9 (six socles, every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 ) and by Proposition 7.11 (six socles and the listed degree pairs, every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 ); the moment equations of Step 2 feed the arithmetic of those sieves.
The exact block-size formula. Fix a point α ∈ B \alpha\in B α ∈ B . The block B B B is G B G_B G B -invariant and contains α \alpha α , so B B B is a union of G B G_B G B -orbits and k = ∣ B ∣ k = |B| k = ∣ B ∣ is the sum of the lengths of those G B G_B G B -orbits; in general it is not true that k = ∣ G B : G B , α ∣ k = |G_B : G_{B,\alpha}| k = ∣ G B : G B , α ∣ , the latter being only the length of the G B G_B G B -orbit of α \alpha α , equal to ∣ B ∣ |B| ∣ B ∣ precisely when G G G is flag-transitive. In the product action on B = Δ ℓ \mathcal{B} = \Delta^\ell B = Δ ℓ , the block stabilizer has the form G B = K ℓ ⋊ C B G_B = K^\ell\rtimes C_B G B = K ℓ ⋊ C B , where K K K is a maximal subgroup of T T T with [ T : K ] = n [T : K] = n [ T : K ] = n and C B C_B C B is the coordinate stabilizer. The point stabilizer is G α = H ℓ ⋊ C α G_\alpha= H^\ell\rtimes C_\alpha G α = H ℓ ⋊ C α , where H H H is a maximal subgroup of T T T with [ T : H ] = m [T : H] = m [ T : H ] = m ; since the point action is primitive, the ℓ \ell ℓ coordinate point stabilizers are conjugate, i.e. the α i \alpha_i α i are of the same type. The block–point stabilizer is F = G B , α = G α ∩ G B F = G_{B,\alpha} = G_\alpha\cap G_B F = G B , α = G α ∩ G B . Both G α G_\alpha G α and G B G_B G B decompose with respect to the same semidirect product T ℓ ⋊ C ℓ T^\ell\rtimes C_\ell T ℓ ⋊ C ℓ (the coordinate groups act by permuting coordinates); hence an element of the intersection has its socle part in H ℓ ∩ K ℓ = ( H ∩ K ) ℓ H^\ell\cap K^\ell= (H \cap K)^\ell H ℓ ∩ K ℓ = ( H ∩ K ) ℓ and its coordinate-permutation part in C α ∩ C B C_\alpha\cap C_B C α ∩ C B , and conversely every pair of these parts gives an element of the intersection. Thus
G α ∩ G B = ( H ∩ K ) ℓ ⋊ ( C α ∩ C B ) , ∣ F ∣ = ∣ H ∩ K ∣ ℓ ⋅ ∣ C α ∩ C B ∣ . G_{\alpha} \cap G_{B} = (H \cap K)^{\ell} \rtimes(C_{\alpha} \cap C_{B}), \qquad|F| = |H \cap K|^{\ell} \cdot|C_{\alpha} \cap C_{B}|. G α ∩ G B = ( H ∩ K ) ℓ ⋊ ( C α ∩ C B ) , ∣ F ∣ = ∣ H ∩ K ∣ ℓ ⋅ ∣ C α ∩ C B ∣.
Under flag-transitivity the block B B B is the single G B G_{B} G B -orbit of α \alpha α , and
k = ∣ G B ∣ ∣ F ∣ = [ K : K ∩ H ] ℓ ⋅ [ C B : C α ∩ C B ] = A ℓ ⋅ j , (5) k = \frac{|G_{B}|}{|F|} = [K : K \cap H]^{\ell} \cdot[C_{B} : C_{\alpha} \cap C_{B}] = A^{\ell} \cdot j, \tag*{(5)} k = ∣ F ∣ ∣ G B ∣ = [ K : K ∩ H ] ℓ ⋅ [ C B : C α ∩ C B ] = A ℓ ⋅ j , (5)
where A = [ K : K ∩ H ] A = [K : K \cap H] A = [ K : K ∩ H ] and 1 ≤ j = ∣ C B ⋅ α ∣ = [ C B : C α ∩ C B ] 1 \le j = |C_{B} \cdot\alpha| = [C_{B} : C_{\alpha} \cap C_{B}] 1 ≤ j = ∣ C B ⋅ α ∣ = [ C B : C α ∩ C B ] is the orbit size of the point α \alpha α under the coordinate stabilizer C B C_{B} C B . Since C B ≤ C ℓ C_{B} \le C_{\ell} C B ≤ C ℓ , the index j j j divides ℓ ! \ell! ℓ ! ; in particular every prime divisor of j j j is at most ℓ \ell ℓ . For a general block-primitive design (not necessarily flag-transitive) the formula (5) need not hold: k k k is only constrained to be the sum of the lengths of the G B G_{B} G B -orbits on B B B , each such orbit being a K K K -box (or a C B C_{B} C B -amalgam of K K K -boxes), cf. Lemma 2.18 . Throughout this section (5) is used in the flag-transitive (symmetric Menon) analysis only; the non-symmetric exclusion uses the orbit-union constraint instead (layers (i)–(ii) of Proposition 7.11).*
The λ \lambda λ -integrality lemma. The replication number r r r is determined by the double count b k = v r bk = vr bk = v r of the 2-design: since b = n ℓ b = n^{\ell} b = n ℓ and v = m ℓ v = m^{\ell} v = m ℓ ,
r = b k v = n ℓ k m ℓ . r = \frac{bk}{v} = \frac{n^{\ell}k}{m^{\ell}}. r = v bk = m ℓ n ℓ k .
which holds for any block-transitive design and does not require flag-transitivity. Together with the standard identity λ ( v − 1 ) = r ( k − 1 ) \lambda(v - 1) = r(k - 1) λ ( v − 1 ) = r ( k − 1 ) , the integrality of λ \lambda λ takes the exact form
λ = n ℓ k ( k − 1 ) m ℓ ( m ℓ − 1 ) ∈ Z . \lambda= \frac{n^{\ell}k(k-1)}{m^{\ell}(m^{\ell}-1)} \in\mathbb{Z}. λ = m ℓ ( m ℓ − 1 ) n ℓ k ( k − 1 ) ∈ Z .
With k = A ℓ j k = A^{\ell}j k = A ℓ j this is the constraint used in the sieve below. (For ℓ ≥ 23 \ell\ge23 ℓ ≥ 23 the prime divisors of m m m , n n n and A A A are at most 13 (the six socles) while every prime divisor of j j j is at most ℓ \ell ℓ .) The integrality of r = n ℓ k / m ℓ r = n^{\ell}k/m^{\ell} r = n ℓ k / m ℓ is an independent necessary condition, checked in the sieve together with the integrality of λ \lambda λ .
Lemma 2.16 (Prime-power congruence). Let p ν ∥ m ℓ − 1 p^{\nu} \mathrel{\|} m^{\ell} - 1 p ν ∥ m ℓ − 1 with p ∤ n ⋅ m ⋅ A p \nmid n \cdot m \cdot A p ∤ n ⋅ m ⋅ A and p > ℓ p > \ell p > ℓ (for ℓ ≥ 23 \ell\ge23 ℓ ≥ 23 the prime divisors of m m m and n n n are at most 13 (six socles), and every prime divisor of j j j is at most ℓ \ell ℓ , so p ∤ n ⋅ m ⋅ A ⋅ j p \nmid n \cdot m \cdot A \cdot j p ∤ n ⋅ m ⋅ A ⋅ j ). Then λ ∈ Z \lambda\in\mathbb{Z} λ ∈ Z implies p ν ∣ k − 1 p^{\nu} \mid k - 1 p ν ∣ k − 1 , i.e. A ℓ j ≡ 1 ( m o d p ν ) A^{\ell}j \equiv1 \pmod{p^{\nu}} A ℓ j ≡ 1 ( mod p ν ) . In particular, if 1 < A ℓ j < p ν 1 < A^{\ell}j < p^{\nu} 1 < A ℓ j < p ν , then no candidate exists.
Proof. The integrality λ ∈ Z \lambda\in\mathbb{Z} λ ∈ Z of the exact form n ℓ k ( k − 1 ) / ( m ℓ ( m ℓ − 1 ) ) n^{\ell}k(k-1)/(m^{\ell}(m^{\ell}-1)) n ℓ k ( k − 1 ) / ( m ℓ ( m ℓ − 1 )) forces m ℓ ( m ℓ − 1 ) ∣ n ℓ k ( k − 1 ) m^{\ell}(m^{\ell}-1) \mid n^{\ell}k(k-1) m ℓ ( m ℓ − 1 ) ∣ n ℓ k ( k − 1 ) , hence v p ( k ( k − 1 ) ) ≥ v p ( m ℓ − 1 ) = ν v_p(k(k-1)) \ge v_p(m^{\ell}-1)=\nu v p ( k ( k − 1 )) ≥ v p ( m ℓ − 1 ) = ν because p ∤ n p \nmid n p ∤ n . Since p ∤ A p \nmid A p ∤ A and p > ℓ p>\ell p > ℓ while every prime divisor of j j j is at most ℓ \ell ℓ , we have p ∤ A ℓ j = k p \nmid A^{\ell}j=k p ∤ A ℓ j = k , so p ∣ k − 1 p \mid k-1 p ∣ k − 1 and p ν ∣ k − 1 p^{\nu} \mid k-1 p ν ∣ k − 1 , that is, A ℓ j ≡ 1 ( m o d p ν ) A^{\ell}j \equiv1 \pmod{p^{\nu}} A ℓ j ≡ 1 ( mod p ν ) . If 1 < A ℓ j < p ν 1<A^{\ell}j<p^{\nu} 1 < A ℓ j < p ν , then k − 1 k-1 k − 1 is a non-zero multiple of p ν p^{\nu} p ν strictly smaller than p ν p^{\nu} p ν , which is impossible, so no candidate exists.
Lemma 2.17 (Large-prime-part exclusion). Let M ′ ′ M^{\prime\prime} M ′′ be the product of the prime-power divisors q v q ( m ℓ − 1 ) q^{v_q(m^{\ell}-1)} q v q ( m ℓ − 1 ) of m ℓ − 1 m^{\ell}-1 m ℓ − 1 with q > ℓ q>\ell q > ℓ , i.e. M ′ ′ = ( m ℓ − 1 ) / ∏ q ≤ ℓ q v q ( m ℓ − 1 ) M^{\prime\prime}=(m^{\ell}-1)/\prod_{q\leq\ell}q^{v_q(m^{\ell}-1)} M ′′ = ( m ℓ − 1 ) / ∏ q ≤ ℓ q v q ( m ℓ − 1 ) . For ℓ ≥ 23 \ell\geq23 ℓ ≥ 23 every prime divisor of m ⋅ n ⋅ A m\cdot n\cdot A m ⋅ n ⋅ A is at most ℓ \ell ℓ (six socles), and every prime divisor of j j j is at most ℓ \ell ℓ , so q ∤ m ⋅ n ⋅ k q\nmid m\cdot n\cdot k q ∤ m ⋅ n ⋅ k for every prime q > ℓ q>\ell q > ℓ ; from the exact integrality form λ = n ℓ k ( k − 1 ) / ( m ℓ ( m ℓ − 1 ) ) \lambda=n^{\ell}k(k-1)/(m^{\ell}(m^{\ell}-1)) λ = n ℓ k ( k − 1 ) / ( m ℓ ( m ℓ − 1 )) this forces M ′ ′ ∣ k − 1 M^{\prime\prime}\mid k-1 M ′′ ∣ k − 1 . In particular, if M ′ ′ > A ℓ ℓ ! − 1 M^{\prime\prime}>A^{\ell}\ell!-1 M ′′ > A ℓ ℓ ! − 1 , then k − 1 = A ℓ j − 1 < M ′ ′ k-1=A^{\ell}j-1<M^{\prime\prime} k − 1 = A ℓ j − 1 < M ′′ for every admissible 1 ≤ j ∣ ℓ ! 1\leq j\mid\ell! 1 ≤ j ∣ ℓ ! , and no candidate exists.
Proof. As in the previous lemma, λ ∈ Z \lambda\in\mathbb{Z} λ ∈ Z forces m ℓ ( m ℓ − 1 ) ∣ n ℓ k ( k − 1 ) m^{\ell}(m^{\ell}-1)\mid n^{\ell}k(k-1) m ℓ ( m ℓ − 1 ) ∣ n ℓ k ( k − 1 ) . For a prime q > ℓ q>\ell q > ℓ one has q ∤ m q\nmid m q ∤ m , since every prime divisor of the primitive degree m m m is at most 13 for the six socles and at most ℓ \ell ℓ for an arbitrary socle; one also has q ∤ A q\nmid A q ∤ A , because A ∣ ∣ K ∣ A\mid|K| A ∣ ∣ K ∣ and every prime divisor of A A A therefore divides m m m ; and q ∤ n q\nmid n q ∤ n , q ∤ j q\nmid j q ∤ j , since every prime divisor of j j j is at most ℓ \ell ℓ . Hence q ∤ n k q\nmid nk q ∤ nk and q v q ( m ℓ − 1 ) ∣ k ( k − 1 ) q^{v_q(m^{\ell}-1)}\mid k(k-1) q v q ( m ℓ − 1 ) ∣ k ( k − 1 ) ; as q ∤ k q\nmid k q ∤ k this gives q v q ( m ℓ − 1 ) ∣ k − 1 q^{v_q(m^{\ell}-1)}\mid k-1 q v q ( m ℓ − 1 ) ∣ k − 1 . Multiplying over all primes q > ℓ q>\ell q > ℓ yields M ′ ′ ∣ k − 1 M^{\prime\prime}\mid k-1 M ′′ ∣ k − 1 . If M ′ ′ > A ℓ ℓ ! − 1 M^{\prime\prime}>A^{\ell}\ell!-1 M ′′ > A ℓ ℓ ! − 1 , then 0 ≤ k − 1 = A ℓ j − 1 < M ′ ′ 0\leq k-1=A^{\ell}j-1<M^{\prime\prime} 0 ≤ k − 1 = A ℓ j − 1 < M ′′ for every admissible j ∣ ℓ ! j\mid\ell! j ∣ ℓ ! , whence k − 1 = 0 k-1=0 k − 1 = 0 , that is k = 1 k=1 k = 1 , which is degenerate.
Lemma 2.18 (Product-action blocks are unions of K K K -boxes). Let B B B be a block in the product-action model of Section 4, with G B ∩ T ℓ = K ℓ G_B\cap T^{\ell}=K^{\ell} G B ∩ T ℓ = K ℓ as in the reduction. Then B B B is a union of K ℓ K^{\ell} K ℓ -orbits. Acting coordinatewise, each K ℓ K^{\ell} K ℓ -orbit has the product shape
K α 1 × ⋯ × K α ℓ ( α = ( α 1 , … , α ℓ ) ∈ B ) , K_{\alpha_1}\times\cdots\times K_{\alpha_\ell}\qquad(\alpha=(\alpha_1,\ldots,\alpha_\ell)\in B), K α 1 × ⋯ × K α ℓ ( α = ( α 1 , … , α ℓ ) ∈ B ) , a “K K K -box” of size ∏ i = 1 ℓ ∣ K α i ∣ \prod_{i=1}^{\ell}|K_{\alpha_i}| ∏ i = 1 ℓ ∣ K α i ∣ , where each ∣ K α i ∣ = ∣ K : K ∩ H i ∣ |K_{\alpha_i}|=|K:K\cap H_i| ∣ K α i ∣ = ∣ K : K ∩ H i ∣ is the size of a K K K -orbit on Δ = T / H \Delta=T/H Δ = T / H (H i H_i H i the stabilizer of α i \alpha_i α i in T T T ). Consequently k = ∣ B ∣ k=|B| k = ∣ B ∣ is a sum of K K K -box sizes; each coordinate projection π i ( B ) \pi_i(B) π i ( B ) is a K K K -invariant subset of Δ \Delta Δ , i.e. a union of K K K -orbits. It does not follow that a non-transitive K K K makes every projection proper: a K K K -invariant subset of Δ \Delta Δ may be a union of several K K K -orbits, possibly all of Δ \Delta Δ ; and the block-size formula (5) (k = A ℓ j k=A^{\ell}j k = A ℓ j ) holds only under flag-transitivity, as recorded there. In general k k k is only constrained to be a sum of K K K -box sizes, possibly amalgamated by the coordinate action of C B C_B C B .
Proof. The subgroup K ℓ = G B ∩ T ℓ K^\ell=G_B\cap T^\ell K ℓ = G B ∩ T ℓ lies in G B G_B G B , so every K ℓ K^\ell K ℓ -orbit meeting B B B is contained in B B B , and B B B is a union of K ℓ K^\ell K ℓ -orbits. Since K ℓ K^\ell K ℓ acts coordinatewise, the orbit through α = ( α 1 , … , α ℓ ) \alpha=(\alpha_1,\ldots,\alpha_\ell) α = ( α 1 , … , α ℓ ) is the product K α 1 × ⋯ × K α ℓ K\alpha_1\times\cdots\times K\alpha_\ell K α 1 × ⋯ × K α ℓ : each coordinate runs through the K K K -orbit of α i \alpha_i α i independently of the others, and the size of the product is ∏ i ∣ K : K ∩ H i ∣ \prod_i\lvert K:K\cap H_i\rvert ∏ i ∣ K : K ∩ H i ∣ . Each coordinate projection π i ( B ) \pi_i(B) π i ( B ) is consequently K K K -invariant, and a K K K -invariant subset of Δ \Delta Δ is a union of K K K -orbits, of size ∑ a ∈ A a \sum_{a\in A}a ∑ a ∈ A a over the K K K -orbit sizes a a a that occur. □
Exclusion of the product action type for the six socles. The exclusion is established in Proposition 7.9 (§ 4.3 ), which splits the seventeen listed primitive degrees of the six socles into three classes and excludes them for every ℓ ≥ 2 \ell\geq2 ℓ ≥ 2 by the fiber double-counting identity (14) and the K K K -box structure of Lemma 2.18; no claim is made for arbitrary socles, embeddings, or unlisted primitive degree pairs.
Theorem 2.19 (Cameron–Praeger criterion). Let D \mathcal{D} D be a block-transitive 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design with automorphism group G G G transitive on blocks, let Q 1 , … , Q r Q_1,\ldots,Q_r Q 1 , … , Q r be the orbits of G G G on unordered pairs of points, and let q i q_i q i be the number of pairs of a block B ∈ B B\in\mathcal{B} B ∈ B lying in Q i Q_i Q i (independent of B B B by block-transitivity). Then
q i ∣ Q i ∣ = ( k 2 ) ( v 2 ) for every i . (6) \frac{q_i}{\lvert Q_i\rvert}=\frac{\binom{k}{2}}{\binom{v}{2}}\qquad\text{for every }i. \tag*{(6)} ∣ Q i ∣ q i = ( 2 v ) ( 2 k ) for every i . (6)
Proof. Double count the incidences { ( B ′ , { x , y } ) } \{(B',\{x,y\})\} {( B ′ , { x , y })} with B ′ ∈ B B'\in\mathcal{B} B ′ ∈ B and { x , y } ∈ Q i \{x,y\}\in Q_i { x , y } ∈ Q i : on the one hand each of the λ \lambda λ blocks through the pair contributes, giving λ ∣ Q i ∣ \lambda\lvert Q_i\rvert λ ∣ Q i ∣ ; on the other hand block-transitivity gives b q i bq_i b q i with b = ∣ B ∣ b=\lvert\mathcal{B}\rvert b = ∣ B ∣ . Hence q i / ∣ Q i ∣ = λ / b = ( k 2 ) / ( v 2 ) q_i/\lvert Q_i\rvert=\lambda/b=\binom{k}{2}/\binom{v}{2} q i / ∣ Q i ∣ = λ / b = ( 2 k ) / ( 2 v ) by the standard identity λ = b k ( k − 1 ) / ( v ( v − 1 ) ) \lambda=bk(k-1)/(v(v-1)) λ = bk ( k − 1 ) / ( v ( v − 1 )) . □
The pure product case L = H ℓ L=H^\ell L = H ℓ : the Z Z Z -criterion and the W 2 W_2 W 2 -criterion. We now isolate the configuration in which the block stabilizer meets the socle in the full point stabilizer, L = N ∩ G B = H ℓ L=N\cap G_B=H^\ell L = N ∩ G B = H ℓ with H H H the point stabilizer in T T T ; this is the branch of the product action in which B B B is a union of N α N_\alpha N α -orbits, each N α N_\alpha N α -orbit being a product H H H -box of size ( m − 1 ) ∣ d ∣ (m-1)^{\lvert d\rvert} ( m − 1 ) ∣ d ∣ indexed by the set d ⊆ { 1 , … , ℓ } d\subseteq\{1,\ldots,\ell\} d ⊆ { 1 , … , ℓ } of coordinates on which the box differs from the fixed point α \alpha α . This is the “case A” of the PA analysis of § 2.1 ; it is the branch in which the block size k k k is realizable as a sum of box sizes ∑ b ∈ S ( m − 1 ) ∣ b ∣ \sum_{b\in S}(m-1)^{\lvert b\rvert} ∑ b ∈ S ( m − 1 ) ∣ b ∣ over a family S S S of coordinate sets. The following criterion is decisive for the six socles.
Lemma 2.20 (The Z Z Z -criterion for the pure product case). Let T T T be a non-abelian simple group acting primitively on Δ \Delta Δ with point stabilizer H H H and degree m = ∣ Δ ∣ m = |\Delta| m = ∣Δ∣ , let ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 , and consider the product action of N = T ℓ N = T^\ell N = T ℓ on P = Δ ℓ \mathcal{P} = \Delta^\ell P = Δ ℓ with v = m ℓ v = m^\ell v = m ℓ fixed point α \alpha α . Suppose B ⊂ P B \subset\mathcal{P} B ⊂ P is a union of N α N_\alpha N α -orbits, k = ∣ B ∣ = ∑ b ∈ S ( m − 1 ) ∣ b ∣ k = |B| = \sum_{b\in S}(m-1)^{|b|} k = ∣ B ∣ = ∑ b ∈ S ( m − 1 ) ∣ b ∣ for some family S ⊆ 2 { 1 , … , ℓ } S \subseteq2^{\{1,\ldots,\ell\}} S ⊆ 2 { 1 , … , ℓ } , and write C k = ( k 2 ) C_k = \binom{k}{2} C k = ( 2 k ) , C v = ( v 2 ) C_v = \binom{v}{2} C v = ( 2 v ) . For each coordinate j j j and each value u ∈ Δ u \in\Delta u ∈ Δ , let B j , u = { γ ∈ B : γ j = u } B_{j,u} = \{\gamma\in B : \gamma_j = u\} B j , u = { γ ∈ B : γ j = u } be the fiber of B B B over u u u in coordinate j j j ; since B B B is N α N_\alpha N α -invariant, each fiber size lies in { 0 , 1 , m − 1 , m } \{0,1,m-1,m\} { 0 , 1 , m − 1 , m } . Then the number q 1 q_1 q 1 of unordered point pairs of B B B at Hamming distance exactly 1 1 1 satisfies
q 1 = ∑ j = 1 ℓ ∑ u ∈ Δ ( ∣ B j , u ∣ 2 ) = ( m − 1 ) ( ℓ k − Z ) 2 , (7) q_1=\sum_{j=1}^{\ell}\sum_{u\in\Delta}\binom{|B_{j,u}|}{2}=\frac{(m-1)(\ell k-Z)}{2}, \tag*{(7)} q 1 = j = 1 ∑ ℓ u ∈ Δ ∑ ( 2 ∣ B j , u ∣ ) = 2 ( m − 1 ) ( ℓ k − Z ) , (7) where
Z = ∑ j = 1 ℓ ∑ u ∈ Δ 1 ∣ B j , u ∣ ∈ { 1 , m − 1 } , (8) Z=\sum_{j=1}^{\ell}\sum_{u\in\Delta}\mathbf{1}_{|B_{j,u}|\in\{1,m-1\}}, \tag*{(8)} Z = j = 1 ∑ ℓ u ∈ Δ ∑ 1 ∣ B j , u ∣ ∈ { 1 , m − 1 } , (8) and Z Z Z admits the orbit-union upper bound
Z ≤ ∑ b ∈ S A b , A b = ∣ b ∣ ( m − 1 ) ∣ b ∣ − 1 + ( ℓ − ∣ b ∣ ) ( m − 1 ) ∣ b ∣ . (9) Z\le\sum_{b\in S}A_b,\qquad A_b=|b|(m-1)^{|b|-1}+(\ell-|b|)(m-1)^{|b|}. \tag*{(9)} Z ≤ b ∈ S ∑ A b , A b = ∣ b ∣ ( m − 1 ) ∣ b ∣ − 1 + ( ℓ − ∣ b ∣ ) ( m − 1 ) ∣ b ∣ . (9) By the Cameron–Praeger criterion (6) (Theorem 2.19 with t = 2 t=2 t = 2 ), if B P B^{\mathcal{P}} B P is a 2 2 2 -design then q 1 = ∣ Q 1 ∣ C k / C v q_1=|Q_1|C_k/C_v q 1 = ∣ Q 1 ∣ C k / C v with ∣ Q 1 ∣ = ℓ v ( m − 1 ) / 2 |Q_1|=\ell v(m-1)/2 ∣ Q 1 ∣ = ℓ v ( m − 1 ) /2 , whence, by (7) ,
Z t a r g e t : = ℓ k − 2 ℓ C k v − 1 (10) Z_{\mathrm{target}}:=\ell k-\frac{2\ell C_k}{v-1} \tag*{(10)} Z target := ℓ k − v − 1 2 ℓ C k (10) must be an integer, must lie in the admissible range [ 0 , ℓ m ℓ − 1 ] [0,\ell m^{\ell-1}] [ 0 , ℓ m ℓ − 1 ] , and must not exceed the fractional-knapsack relaxation
Σ A max : = max { ∑ b ∈ S A b : k = ∑ b ∈ S ( m − 1 ) ∣ b ∣ , # { b ∈ S : ∣ b ∣ = t } ≤ ( ℓ t ) } . (11) \Sigma A_{\max}:=\max\left\{\sum_{b\in S}A_b:\ k=\sum_{b\in S}(m-1)^{|b|},\ \#\{b\in S:|b|=t\}\le\binom{\ell}{t}\right\}. \tag*{(11)} Σ A m a x := max { b ∈ S ∑ A b : k = b ∈ S ∑ ( m − 1 ) ∣ b ∣ , # { b ∈ S : ∣ b ∣ = t } ≤ ( t ℓ ) } . (11) computed by taking, for each layer t t t , as many boxes of layer t t t as the capacity ( ℓ t ) \binom{\ell}{t} ( t ℓ ) allows, in order of decreasing efficiency A t / ( m − 1 ) t A_t/(m-1)^t A t / ( m − 1 ) t , with A t = t ( m − 1 ) t − 1 + ( ℓ − t ) ( m − 1 ) t A_t=t(m-1)^{t-1}+(\ell-t)(m-1)^t A t = t ( m − 1 ) t − 1 + ( ℓ − t ) ( m − 1 ) t . Consequently, if Z t a r g e t ∉ Z Z_{\mathrm{target}}\notin\mathbb{Z} Z target ∈ / Z , or Z t a r g e t ∉ [ 0 , ℓ m ℓ − 1 ] Z_{\mathrm{target}}\notin[0,\ell m^{\ell-1}] Z target ∈ / [ 0 , ℓ m ℓ − 1 ] , or Z t a r g e t > Σ A max Z_{\mathrm{target}}>\Sigma A_{\max} Z target > Σ A m a x , then no N α N_\alpha N α -invariant k k k -subset B B B with B P B^{\mathcal{P}} B P a 2 2 2 -design exists.
Proof. For c ∈ { 0 , 1 , m − 1 , m } c \in\{0,1,m-1,m\} c ∈ { 0 , 1 , m − 1 , m } the identity ( c 2 ) = m − 1 2 c − m − 1 2 1 c ∈ { 1 , m − 1 } \binom{c}{2}=\frac{m-1}{2}c-\frac{m-1}{2}\mathbf{1}_{c\in\{1,m-1\}} ( 2 c ) = 2 m − 1 c − 2 m − 1 1 c ∈ { 1 , m − 1 } holds; summing over all fibers B j , u B_{j,u} B j , u and using ∑ u ∈ Δ ∣ B j , u ∣ = k \sum_{u\in\Delta}|B_{j,u}|=k ∑ u ∈ Δ ∣ B j , u ∣ = k for each j j j gives (6)–(7). For (9) , fix b ∈ S b\in S b ∈ S : its contribution to Z Z Z is at most the number of coordinates j j j with j ∉ b j\notin b j ∈ / b times ( m − 1 ) ∣ b ∣ (m-1)^{|b|} ( m − 1 ) ∣ b ∣ (fiber { 1 } \{1\} { 1 } , one value u u u with ∣ B j , u ∣ = 1 |B_{j,u}|=1 ∣ B j , u ∣ = 1 ) plus the number of coordinates j ∈ b j\in b j ∈ b times ( m − 1 ) ∣ b ∣ − 1 (m-1)^{|b|-1} ( m − 1 ) ∣ b ∣ − 1 (fiber Δ ∖ { α j } \Delta\setminus\{\alpha_j\} Δ ∖ { α j } , each of the m − 1 m-1 m − 1 values u u u giving ∣ B j , u ∣ = m − 1 |B_{j,u}|=m-1 ∣ B j , u ∣ = m − 1 ), i.e. A b A_b A b ; overlapping fibers of distinct boxes only merge values and cannot increase Z Z Z , giving (9) . The Cameron–Praeger criterion yields q 1 = ∣ Q 1 ∣ C k / C v q_1=|Q_1|C_k/C_v q 1 = ∣ Q 1 ∣ C k / C v ; substituting ∣ Q 1 ∣ = ℓ v ( m − 1 ) / 2 |Q_1|=\ell v(m-1)/2 ∣ Q 1 ∣ = ℓ v ( m − 1 ) /2 and C v = v ( v − 1 ) / 2 C_v=v(v-1)/2 C v = v ( v − 1 ) /2 into (7) gives Z = Z t a r g e t Z=Z_{\mathrm{target}} Z = Z target . Since Z ≥ 0 Z\geq0 Z ≥ 0 and each of the ℓ m \ell m ℓ m indicator terms in (8) is at most 1, Z ≤ ℓ m ≤ ℓ m ℓ − 1 Z\leq\ell m\leq\ell m^{\ell-1} Z ≤ ℓ m ≤ ℓ m ℓ − 1 . The bound (11) is the fractional relaxation of (9) , hence an upper bound for every feasible S S S . The three failure conditions therefore contradict the design hypothesis.
Lemma 2.21 (The W 2 W_2 W 2 -criterion for the pure product case). Keep the notation and hypotheses of Lemma 2.20, and let q 2 q_2 q 2 be the number of unordered point pairs of B B B at Hamming distance exactly 2. Write W 2 = q 2 / ( m − 1 ) 2 W_2=q_2/(m-1)^2 W 2 = q 2 / ( m − 1 ) 2 . Then the Cameron–Praeger criterion for the distance-2 orbital Q 2 Q_2 Q 2 , ∣ Q 2 ∣ = ( ℓ 2 ) v ( m − 1 ) 2 / 2 |Q_2|=\binom{\ell}{2}v(m-1)^2/2 ∣ Q 2 ∣ = ( 2 ℓ ) v ( m − 1 ) 2 /2 , gives
W 2 t a r g e t : = ( ℓ 2 ) C k v − 1 , (12) W_2^{\mathrm{target}}:=\frac{\binom{\ell}{2}C_k}{v-1}, \tag*{(12)} W 2 target := v − 1 ( 2 ℓ ) C k , (12) and W 2 W_2 W 2 admits the orbit-union upper bound
W 2 ≤ ∑ b ∈ S B ∣ b ∣ , B t = ( ℓ − t ) ( ℓ − t − 1 ) ( m − 1 ) t + t ( ℓ − t ) ( m − 1 ) t − 1 , (13) W_2\leq\sum_{b\in S}B_{|b|},\qquad B_t=(\ell-t)(\ell-t-1)(m-1)^t+t(\ell-t)(m-1)^{t-1}, \tag*{(13)} W 2 ≤ b ∈ S ∑ B ∣ b ∣ , B t = ( ℓ − t ) ( ℓ − t − 1 ) ( m − 1 ) t + t ( ℓ − t ) ( m − 1 ) t − 1 , (13) whose fractional-knapsack relaxation Σ B max \Sigma B_{\max} Σ B m a x is computed as in Lemma 2.20 with efficiencies B t / ( m − 1 ) t B_t/(m-1)^t B t / ( m − 1 ) t and layer capacities ( ℓ t ) \binom{\ell}{t} ( t ℓ ) . Consequently, if W 2 t a r g e t ∉ Z W_2^{\mathrm{target}}\notin\mathbb{Z} W 2 target ∈ / Z or W 2 t a r g e t > Σ B max W_2^{\mathrm{target}}>\Sigma B_{\max} W 2 target > Σ B m a x , then no N α N_\alpha N α -invariant k k k -subset B B B with B P B^P B P a 2-design exists.
Proof. A distance-2 pair { γ , δ } \{\gamma,\delta\} { γ , δ } with differing coordinates j , j ′ j,j' j , j ′ pairs the fiber pattern of γ \gamma γ on ( j , j ′ ) (j,j') ( j , j ′ ) with the complementary pattern of δ \delta δ ; for a fixed coordinate pair ( j , j ′ ) (j,j') ( j , j ′ ) and a fixed pattern d d d on the remaining ℓ − 2 \ell-2 ℓ − 2 coordinates, the contribution is ( m − 1 ) 2 ( m − 1 ) ∣ d ∣ (m-1)^2(m-1)^{|d|} ( m − 1 ) 2 ( m − 1 ) ∣ d ∣ whenever both complementary boxes lie in S S S (patterns ( 00 , d ) (00,d) ( 00 , d ) and ( 11 , d ) (11,d) ( 11 , d ) , or ( 01 , d ) (01,d) ( 01 , d ) and ( 10 , d ) (10,d) ( 10 , d ) ). Summing and bounding 1 { 00 , d } ∈ S ∧ 1 { 11 , d } ∈ S + 1 { 01 , d } ∈ S ∧ 1 { 10 , d } ∈ S ≤ 1 { 00 , d } ∈ S + 1 { 01 , d } ∈ S \mathbf{1}_{\{00,d\}\in S}\wedge\mathbf{1}_{\{11,d\}\in S}+\mathbf{1}_{\{01,d\}\in S}\wedge\mathbf{1}_{\{10,d\}\in S}\leq\mathbf{1}_{\{00,d\}\in S}+\mathbf{1}_{\{01,d\}\in S} 1 { 00 , d } ∈ S ∧ 1 { 11 , d } ∈ S + 1 { 01 , d } ∈ S ∧ 1 { 10 , d } ∈ S ≤ 1 { 00 , d } ∈ S + 1 { 01 , d } ∈ S and then aggregating over boxes b ∈ S b\in S b ∈ S gives (13) : a box b b b with t = ∣ b ∣ t=|b| t = ∣ b ∣ contributes ( ℓ − t 2 ) ( m − 1 ) t \binom{\ell-t}{2}(m-1)^t ( 2 ℓ − t ) ( m − 1 ) t from pairs j , j ′ ∉ b j,j'\notin b j , j ′ ∈ / b (pattern ( 00 , d ) (00,d) ( 00 , d ) ) and t ( ℓ − t ) ( m − 1 ) t − 1 t(\ell-t)(m-1)^{t-1} t ( ℓ − t ) ( m − 1 ) t − 1 from pairs j ∈ b ∋ j ′ j\in b\ni j' j ∈ b ∋ j ′ (pattern 01 , d 01,d 01 , d ), i.e. B t B_t B t . The target (12) follows from q 2 = ∣ Q 2 ∣ C k / C v q_2=|Q_2|C_k/C_v q 2 = ∣ Q 2 ∣ C k / C v . The conclusion is immediate.
Corollary 2.22 (The pure product case is closed for the six socles and ℓ ≤ 9 \ell\le9 ℓ ≤ 9 ). Let T ∈ { A 5 , A 6 , P S L 2 ( 7 ) , M 11 , M 12 , P S U 3 ( 3 ) } T \in\{A_5,A_6,PSL_2(7),M_{11},M_{12},PSU_3(3)\} T ∈ { A 5 , A 6 , P S L 2 ( 7 ) , M 11 , M 12 , P S U 3 ( 3 )} with primitive degrees m = 5 , 6 , 7 , 11 , 12 , 28 m=5,6,7,11,12,28 m = 5 , 6 , 7 , 11 , 12 , 28 respectively, and let ℓ ∈ { 2 , … , 9 } \ell\in\{2,\ldots,9\} ℓ ∈ { 2 , … , 9 } . In the pure product case L = H ℓ L=H^\ell L = H ℓ of the product action of N = T ℓ N=T^\ell N = T ℓ on Δ ℓ \Delta^\ell Δ ℓ (i.e. B B B a union of N α N_\alpha N α -orbits), there is no non-trivial 2 2 2 -design B P B^P B P : every admissible block size k k k with λ = k ( k − 1 ) / ( v − 1 ) ∈ Z \lambda=k(k-1)/(v-1)\in\mathbb{Z} λ = k ( k − 1 ) / ( v − 1 ) ∈ Z and k = ∑ b ∈ S ( m − 1 ) ∣ b ∣ k=\sum_{b\in S}(m-1)^{|b|} k = ∑ b ∈ S ( m − 1 ) ∣ b ∣ fails the Z Z Z -criterion of Lemma 2.20 or the W 2 W_2 W 2 -criterion of Lemma 2.21 .
Proof. Every admissible block size k k k , 2 ≤ k ≤ v − 2 2\le k\le v-2 2 ≤ k ≤ v − 2 , with k ( k − 1 ) ≡ 0 ( m o d v − 1 ) k(k-1)\equiv0\pmod{v-1} k ( k − 1 ) ≡ 0 ( mod v − 1 ) and k = ∑ b ∈ S ( m − 1 ) ∣ b ∣ k=\sum_{b\in S}(m-1)^{|b|} k = ∑ b ∈ S ( m − 1 ) ∣ b ∣ is tested against Lemma 2.20 and Lemma 2.21 and is excluded by one of the two criteria.
We record that the complement of a union of N α N_\alpha N α -orbits is again a union of N α N_\alpha N α -orbits, and the complement of a 2 2 2 -design is a 2 2 2 -design; the exclusion of a block size k k k therefore excludes v − k v-k v − k as well, which is reflected in the symmetric appearance of the admissible lists (each admissible k k k is paired with v − k v-k v − k ). All admissible candidates are excluded by the two criteria directly; no candidate survives to the complement argument.
Lemma 2.23 (Box orbits on pairs in the product action). Let T T T be a non-abelian simple group acting primitively on Δ \Delta Δ with point stabilizer H H H and degree m = ∣ Δ ∣ m=|\Delta| m = ∣Δ∣ , and let 1 = s 1 < s 2 ≤ ⋯ ≤ s R 1=s_1<s_2\le\cdots\le s_R 1 = s 1 < s 2 ≤ ⋯ ≤ s R be the lengths of the H H H -orbits on Δ \Delta Δ (the suborbits of the primitive action; s 1 = 1 s_1=1 s 1 = 1 corresponds to α j \alpha_j α j itself). Consider the product action of N = T ℓ N=T^\ell N = T ℓ on P = Δ ℓ P=\Delta^\ell P = Δ ℓ , fix α ∈ P \alpha\in P α ∈ P , and let G α G_\alpha G α be the stabilizer of α \alpha α in a group G G G whose socle is N N N (so that the coordinate-permutation part C α ≤ S ℓ C_\alpha\le S_\ell C α ≤ S ℓ permutes the coordinates). For a point β ≠ α \beta\ne\alpha β = α , let w ( β ) = ( i 1 , … , i ℓ ) w(\beta)=(i_1,\ldots,i_\ell) w ( β ) = ( i 1 , … , i ℓ ) be its coordinate word, defined by β j ∈ Δ α j ( i j ) \beta_j\in\Delta_{\alpha_j}^{(i_j)} β j ∈ Δ α j ( i j ) , and for a word w = ( i 1 , … , i ℓ ) w=(i_1,\ldots,i_\ell) w = ( i 1 , … , i ℓ ) set
O ( w ) = { β ∈ P : β j ∈ Δ α j ( i j ) for 1 ≤ j ≤ ℓ } , ∣ O ( w ) ∣ = ∏ j = 1 ℓ s i j . O(w)=\{\beta\in P:\ \beta_j\in\Delta_{\alpha_j}^{(i_j)}\text{ for }1\le j\le\ell\},\qquad|O(w)|=\prod_{j=1}^{\ell}s_{i_j}. O ( w ) = { β ∈ P : β j ∈ Δ α j ( i j ) for 1 ≤ j ≤ ℓ } , ∣ O ( w ) ∣ = j = 1 ∏ ℓ s i j . Then every G α G_\alpha G α -orbit on P ∖ { α } P\setminus\{\alpha\} P ∖ { α } is the union of the boxes O ( w ) O(w) O ( w ) with w w w running over a single C α C_\alpha C α -orbit A A A on the set of words with at least one non-trivial entry, and has size
∣ O ∣ = ∑ w ∈ A ∏ j = 1 ℓ s i j . |O|=\sum_{w\in A}\prod_{j=1}^{\ell}s_{i_j}. ∣ O ∣ = w ∈ A ∑ j = 1 ∏ ℓ s i j .
In the single-type case, where the word has entries i i i on a d d d -element subset J J J and 1 1 1 elsewhere, this reduces to ∣ O ∣ = c s i d |O|=c s_i^d ∣ O ∣ = c s i d with c = ∣ C α ( J ) ∣ c=|C_\alpha(J)| c = ∣ C α ( J ) ∣ , the size of the C α C_\alpha C α -orbit of J J J on the ( ℓ d ) \binom{\ell}{d} ( d ℓ ) -element set of d d d -subsets, so that 1 ≤ c ≤ ( ℓ d ) 1\leq c\leq\binom{\ell}{d} 1 ≤ c ≤ ( d ℓ ) ; in general the mixed word orbits give further sizes ( ℓ ! / ∏ i = 2 R n i ! ) ∏ i = 2 R s i n i (\ell!/\prod_{i=2}^{R}n_i!)\prod_{i=2}^{R}s_i^{n_i} ( ℓ ! / ∏ i = 2 R n i !) ∏ i = 2 R s i n i for the multiplicity patterns ( n 2 , … , n R ) (n_2,\ldots,n_R) ( n 2 , … , n R ) .
Proof. The H H H -suborbits partition Δ \Delta Δ , so the word of β \beta β is well defined and the boxes O ( w ) O(w) O ( w ) partition P \mathcal{P} P ; ∣ O ( w ) ∣ = ∏ s i j |O(w)|=\prod s_{i_j} ∣ O ( w ) ∣ = ∏ s i j follows from the orbit-stabilizer theorem applied coordinatewise. Since C α ≤ S ℓ C_\alpha\leq S_\ell C α ≤ S ℓ permutes the coordinates, it sends O ( w ) O(w) O ( w ) to O ( σ ⋅ w ) O(\sigma\cdot w) O ( σ ⋅ w ) for the coordinatewise action σ ⋅ w = ( i σ − 1 ( 1 ) , … , i σ − 1 ( ℓ ) ) \sigma\cdot w=(i_{\sigma^{-1}(1)},\ldots,i_{\sigma^{-1}(\ell)}) σ ⋅ w = ( i σ − 1 ( 1 ) , … , i σ − 1 ( ℓ ) ) , so the G α G_\alpha G α -orbit of a point of word w w w is the union of the boxes indexed by the C α C_\alpha C α -orbit of w w w , and its size is the displayed sum; the displayed mixed size is the special case in which C α C_\alpha C α contains the full symmetric group on the coordinates (or a subgroup transitive on the words of the given multiplicity pattern).
Proposition 2.24 (The subdirect socle block stabilizer branch: the orbit-wise criterion). Let D \mathcal{D} D be a non-trivial 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design (not assumed symmetric) with 2 < k < v − 1 2<k<v-1 2 < k < v − 1 , and let G ≤ Aut ( D ) G\leq\operatorname{Aut}(\mathcal{D}) G ≤ Aut ( D ) act transitively on points and primitively on blocks, with socle N = T ℓ N=T^\ell N = T ℓ (ℓ ≥ 2 \ell\geq2 ℓ ≥ 2 ) of product action type on points of degree v = m ℓ v=m^\ell v = m ℓ , where m m m is a primitive degree of the non-abelian simple group T T T . Assume that L = N ∩ G β L=N\cap G_\beta L = N ∩ G β is a subdirect subgroup of N N N , hence a product of t t t full diagonal subgroups on pairwise disjoint supports with 1 ≤ t ≤ ℓ − 1 1\leq t\leq\ell-1 1 ≤ t ≤ ℓ − 1 ; then b = [ N : L ] = ∣ T ∣ ℓ − t b=[N:L]=|T|^{\ell-t} b = [ N : L ] = ∣ T ∣ ℓ − t and v = m ℓ v=m^\ell v = m ℓ need not coincide. For the nine socles
T ∈ { A 5 , A 6 , PSL 2 ( 7 ) , PSL 2 ( 8 ) , PSL 2 ( 11 ) , PSL 2 ( 13 ) , M 11 , M 12 , PSU 3 ( 3 ) } T\in\{A_5,A_6,\operatorname{PSL}_2(7),\operatorname{PSL}_2(8),\operatorname{PSL}_2(11),\operatorname{PSL}_2(13),M_{11},M_{12},\operatorname{PSU}_3(3)\} T ∈ { A 5 , A 6 , PSL 2 ( 7 ) , PSL 2 ( 8 ) , PSL 2 ( 11 ) , PSL 2 ( 13 ) , M 11 , M 12 , PSU 3 ( 3 )} and every primitive degree m m m of T T T , with 2 ≤ ℓ ≤ 9 2\leq\ell\leq9 2 ≤ ℓ ≤ 9 , the orbit-wise form of the Cameron–Praeger criterion (6) (Theorem 2.19 ) takes the following explicit form. Fix α ∈ B \alpha\in B α ∈ B , let o o o be the size of a G α G_\alpha G α -orbit (box, Lemma 2.23 ) on P ∖ { α } \mathcal{P}\setminus\{\alpha\} P ∖ { α } , and let Q i = G ⋅ { α , β } Q_i=G\cdot\{\alpha,\beta\} Q i = G ⋅ { α , β } be the G G G -orbit of the unordered pair { α , β } \{\alpha,\beta\} { α , β } with β \beta β in that orbit. Then ∣ Q i ∣ = v o / 2 |Q_i|=vo/2 ∣ Q i ∣ = v o /2 (G α G_\alpha G α has index v v v in G G G , and G { α , β } G_{\{\alpha,\beta\}} G { α , β } has index o o o in G α G_\alpha G α and index 2 2 2 in G α , β G_{\alpha,\beta} G α , β ), so the required in-block count of (6) is
q i = ∣ Q i ∣ k ( k − 1 ) v ( v − 1 ) = o k ( k − 1 ) 2 ( v − 1 ) , q_i=\frac{|Q_i|k(k-1)}{v(v-1)}=\frac{ok(k-1)}{2(v-1)}, q i = v ( v − 1 ) ∣ Q i ∣ k ( k − 1 ) = 2 ( v − 1 ) o k ( k − 1 ) ,
and it must be an integer in [ 0 , min ( ∣ Q i ∣ , ( k 2 ) ) ] = [ 0 , min ( v o / 2 , ( k 2 ) ) ] [0,\min(|\mathcal{Q}_i|,\binom{k}{2})]=[0,\min(v o/2,\binom{k}{2})] [ 0 , min ( ∣ Q i ∣ , ( 2 k ) )] = [ 0 , min ( v o /2 , ( 2 k ) )] . This condition is necessary but not sufficient: it admits numerous admissible candidates, so it does not by itself eliminate the branch. The complete exclusion of the product-action type for non-symmetric designs is instead carried out by the exhaustive K K K -box-union enumeration with the λ \lambda λ -constancy test of layer (ii) and by the SD/CD block-type exclusions of § 2.1 ; the block-transitivity fact N G β = G NG_\beta=G N G β = G used there follows from the maximality of G β G_\beta G β . The surviving candidate ( A 6 , 6 , 2 , 1 , 28 ) (A_6,6,2,1,28) ( A 6 , 6 , 2 , 1 , 28 ) with v = 36 v=36 v = 36 , k = 28 k=28 k = 28 , b = 360 b=360 b = 360 , r = 280 r=280 r = 280 , λ = 216 \lambda=216 λ = 216 is excluded by the complement-design argument in the proof. For each candidate the admissible sizes o o o form a finite set (the words of the multiplicity patterns, R ℓ R^\ell R ℓ many, under the coordinate action of C α C_\alpha C α ).
Proof. By block-primitivity G G G is transitive on blocks, so Theorem 2.19 applies and (5) holds for every orbit Q i \mathcal{Q}_i Q i of G G G on unordered pairs: writing q i q_i q i for the number of pairs of a block B B B lying in Q i \mathcal{Q}_i Q i , the count q i = ∣ Q i ∣ k ( k − 1 ) / ( v ( v − 1 ) ) q_i=|\mathcal{Q}_i|k(k-1)/(v(v-1)) q i = ∣ Q i ∣ k ( k − 1 ) / ( v ( v − 1 )) must be an integer in [ 0 , min ( ∣ Q i ∣ , ( k 2 ) ) ] [0,\min(|\mathcal{Q}_i|,\binom{k}{2})] [ 0 , min ( ∣ Q i ∣ , ( 2 k ) )] (both bounds are automatic for an in-block pair count: q i q_i q i counts pairs of B B B , hence q i ≤ ( k 2 ) q_i\le\binom{k}{2} q i ≤ ( 2 k ) , and q i q_i q i counts members of Q i \mathcal{Q}_i Q i , hence q i ≤ ∣ Q i ∣ q_i\le|\mathcal{Q}_i| q i ≤ ∣ Q i ∣ ). For a fixed point α \alpha α of a block B B B , the G G G -orbits on pairs containing α \alpha α are precisely the G α G_\alpha G α -orbits on P ∖ { α } \mathcal{P}\setminus\{\alpha\} P ∖ { α } ; by Lemma 2.23 each has size o o o as described. If for some admissible box-orbit size o o o the divisibility and range conditions fail, then in particular they fail for the actual size of the orbit, and no such pair orbit can exist; since q i q_i q i is then not an integer, the design cannot exist. These conditions admit numerous admissible candidates, so the orbit-wise criterion alone does not close the branch; the complete exclusion is achieved by the λ \lambda λ -constancy test of layer (ii) and by the SD/CD block-type exclusions of § 2.1 , as stated. The single candidate ( A 6 , 6 , 2 , 1 , 28 ) (A_6,6,2,1,28) ( A 6 , 6 , 2 , 1 , 28 ) that survives the orbit-wise sieve is handled by the complement-design argument that follows.
It remains to dispose of the single survivor ( A 6 , 6 , 2 , 1 , 28 ) (A_6,6,2,1,28) ( A 6 , 6 , 2 , 1 , 28 ) . Here v = 36 v=36 v = 36 , b = ∣ A 6 ∣ = 360 b=|A_6|=360 b = ∣ A 6 ∣ = 360 , k = 28 k=28 k = 28 , r = b k / v = 280 r=bk/v=280 r = bk / v = 280 and λ = r ( k − 1 ) / ( v − 1 ) = 216 \lambda=r(k-1)/(v-1)=216 λ = r ( k − 1 ) / ( v − 1 ) = 216 ; the complement of the design is a 2 2 2 -( 36 , 8 , 16 ) (36,8,16) ( 36 , 8 , 16 ) design with r ′ = 80 r'=80 r ′ = 80 and b ′ = 360 b'=360 b ′ = 360 blocks of size 8 8 8 . Let C C C be a block of the complement containing α = ( 1 , 1 ) \alpha=(1,1) α = ( 1 , 1 ) , and write N C = Stab N ( C ) N_C=\operatorname{Stab}_N(C) N C = Stab N ( C ) with N = A 6 × A 6 N=A_6\times A_6 N = A 6 × A 6 ; since N N N is transitive on blocks and b ′ = 360 b'=360 b ′ = 360 , we have ∣ N C ∣ = ∣ N ∣ / 360 = 360 |N_C|=|N|/360=360 ∣ N C ∣ = ∣ N ∣/360 = 360 . For β ∈ P ∖ { α } \beta\in\mathcal{P}\setminus\{\alpha\} β ∈ P ∖ { α } the number of blocks through { α , β } \{\alpha,\beta\} { α , β } is
λ ′ ( α , β ) = ∣ N ( α , β ) ∣ ∣ N C ∣ ⋅ # { ( c , d ) ∈ C 2 : ( c , d ) ∈ N ⋅ ( α , β ) } , \lambda'(\alpha,\beta)=\frac{|N_{(\alpha,\beta)}|}{|N_C|}\cdot\#\{(c,d)\in C^2:(c,d)\in N\cdot(\alpha,\beta)\}, λ ′ ( α , β ) = ∣ N C ∣ ∣ N ( α , β ) ∣ ⋅ # {( c , d ) ∈ C 2 : ( c , d ) ∈ N ⋅ ( α , β )} ,
where N ( α , β ) N_{(\alpha,\beta)} N ( α , β ) is the ordered-pair stabilizer. The stabilizer N α = A 5 × A 5 N_\alpha=A_5 \times A_5 N α = A 5 × A 5 has three orbits on P ∖ { α } \mathcal{P}\setminus\{\alpha\} P ∖ { α } : O 1 O_1 O 1 (first coordinate moves, 5 points), O 2 O_2 O 2 (second coordinate moves, 5 points) and O 3 O_3 O 3 (both move, 25 points). If β ∈ O 1 \beta\in O_1 β ∈ O 1 , the pairs ( c , d ) ∈ C 2 (c,d)\in C^2 ( c , d ) ∈ C 2 in the orbit of ( α , β ) (\alpha,\beta) ( α , β ) are those with equal second coordinates, and their number is ∑ j ∣ C j ∣ ( ∣ C j ∣ − 1 ) \sum_j |C_j|(|C_j|-1) ∑ j ∣ C j ∣ ( ∣ C j ∣ − 1 ) , where C j C_j C j is the set of points of C C C in column j j j ; the pair stabilizer has order ∣ N α ∣ / ∣ O 1 ∣ = 720 |N_\alpha|/|O_1|=720 ∣ N α ∣/∣ O 1 ∣ = 720 . Hence
16 = λ ′ ( α , β ) = 720 360 ∑ j ∣ C j ∣ ( ∣ C j ∣ − 1 ) = 2 ( ∑ j ∣ C j ∣ 2 − 8 ) . 16=\lambda'(\alpha,\beta)=\frac{720}{360}\sum_j |C_j|(|C_j|-1)=2\left(\sum_j |C_j|^2-8\right). 16 = λ ′ ( α , β ) = 360 720 j ∑ ∣ C j ∣ ( ∣ C j ∣ − 1 ) = 2 ( j ∑ ∣ C j ∣ 2 − 8 ) .
so ∑ j ∣ C j ∣ 2 = 16 \sum_j |C_j|^2=16 ∑ j ∣ C j ∣ 2 = 16 . Symmetrically, with R i R_i R i the points of C C C in row i i i , the case β ∈ O 2 \beta\in O_2 β ∈ O 2 forces ∑ i ∣ R i ∣ 2 = 16 \sum_i |R_i|^2=16 ∑ i ∣ R i ∣ 2 = 16 ; the case β ∈ O 3 \beta\in O_3 β ∈ O 3 imposes the compatible value 72 − ∑ i ∣ R i ∣ 2 − ∑ j ∣ C j ∣ 2 = 40 72-\sum_i |R_i|^2-\sum_j |C_j|^2=40 72 − ∑ i ∣ R i ∣ 2 − ∑ j ∣ C j ∣ 2 = 40 , which is then automatic. The six column sizes sum to 8 and satisfy ∑ j ∣ C j ∣ 2 = 16 \sum_j |C_j|^2=16 ∑ j ∣ C j ∣ 2 = 16 ; the only such distributions are ( 3 , 2 , 1 , 1 , 1 , 0 ) (3,2,1,1,1,0) ( 3 , 2 , 1 , 1 , 1 , 0 ) and ( 2 , 2 , 2 , 2 , 0 , 0 ) (2,2,2,2,0,0) ( 2 , 2 , 2 , 2 , 0 , 0 ) up to permutation, and likewise for the rows.
Now N C N_C N C preserves C C C , hence preserves the set L L L of columns met by C C C and the set R R R of rows met by C C C ; α ∈ C \alpha\in C α ∈ C lies in column 1, so L ∖ { 1 } L\setminus\{1\} L ∖ { 1 } has 3 or 4 elements. The first projection π 1 ( N C ) \pi_1(N_C) π 1 ( N C ) is contained in ( A 6 ) 1 ≅ A 5 (A_6)_1\cong A_5 ( A 6 ) 1 ≅ A 5 acting on { 2 , 3 , 4 , 5 , 6 } \{2,3,4,5,6\} { 2 , 3 , 4 , 5 , 6 } , and it fixes L ∖ { 1 } L\setminus\{1\} L ∖ { 1 } setwise; the stabilizer in A 5 A_5 A 5 of a 3-subset has order 6 (the A 5 A_5 A 5 -orbit of 3-subsets has 10 members) and of a 4-subset has order 12. Thus ∣ π 1 ( N C ) ∣ ≤ 12 |\pi_1(N_C)|\leq12 ∣ π 1 ( N C ) ∣ ≤ 12 , and similarly ∣ π 2 ( N C ) ∣ ≤ 12 |\pi_2(N_C)|\leq12 ∣ π 2 ( N C ) ∣ ≤ 12 . Therefore ∣ N C ∣ ≤ ∣ π 1 ( N C ) ∣ ⋅ ∣ π 2 ( N C ) ∣ ≤ 144 < 360 |N_C|\leq|\pi_1(N_C)|\cdot|\pi_2(N_C)|\leq144<360 ∣ N C ∣ ≤ ∣ π 1 ( N C ) ∣ ⋅ ∣ π 2 ( N C ) ∣ ≤ 144 < 360 , contradicting ∣ N C ∣ = 360 |N_C|=360 ∣ N C ∣ = 360 . This excludes the recorded survivor ( A 6 , 6 , 2 , 1 , 28 ) (A_6,6,2,1,28) ( A 6 , 6 , 2 , 1 , 28 ) ; together with the λ \lambda λ -constancy enumeration of layer (ii) and the SD/CD block-type exclusions this completes the exclusion of the branch as stated.
In the two-orbit product-action configuration (K K K has exactly one non-trivial orbit of size m − 1 m-1 m − 1 on Δ \Delta Δ , so the orbit-square sum Σ 2 = ∑ a ∈ A a 2 \Sigma_2=\sum_{a\in A}a^2 Σ 2 = ∑ a ∈ A a 2 of Proposition 7.9 equals 1 + ( m − 1 ) 2 1+(m-1)^2 1 + ( m − 1 ) 2 ), the fiber double-counting takes the form
m ℓ ⋅ m f i b e r = λ ⋅ Σ 2 ℓ − m ℓ 2 . (14) m^\ell\cdot m_{\mathrm{fiber}}=\lambda\cdot\frac{\Sigma_2^\ell-m^\ell}{2}. \tag*{(14)} m ℓ ⋅ m fiber = λ ⋅ 2 Σ 2 ℓ − m ℓ . (14)
Lemma 2.25 (Product-action box sieve for arbitrary primitive degree). Let D \mathcal{D} D be a non-trivial 2 2 2 -( m ℓ , k , λ ) (m^\ell,k,\lambda) ( m ℓ , k , λ ) design with 2 < k < v − 1 2<k<v-1 2 < k < v − 1 whose automorphism group is of product action type on points with socle T ℓ T^\ell T ℓ , ℓ ≥ 2 \ell\geq2 ℓ ≥ 2 , and assume the two-orbit box framework of Lemma 2.18 : the point stabilizer H H H and the block stabilizer intersection K K K are conjugate in T T T , so that the K K K -orbit lengths on Δ = T / H \Delta=T/H Δ = T / H are { 1 , m − 1 } \{1,m-1\} { 1 , m − 1 } with m = ∣ Δ ∣ m=|\Delta| m = ∣Δ∣ the primitive degree. Then:
(i) for ℓ = 2 \ell= 2 ℓ = 2 , no block size k = 1 + c 1 ( m − 1 ) + c 2 ( m − 1 ) 2 k = 1+c_1(m-1)+c_2(m-1)^2 k = 1 + c 1 ( m − 1 ) + c 2 ( m − 1 ) 2 with c 1 ∈ { 0 , 1 , 2 } c_1 \in\{0,1,2\} c 1 ∈ { 0 , 1 , 2 } , c 2 ∈ { 0 , 1 } c_2 \in\{0,1\} c 2 ∈ { 0 , 1 } , c 1 + c 2 > 0 c_1+c_2 > 0 c 1 + c 2 > 0 , satisfies the integrality λ = k ( k − 1 ) / ( m 2 − 1 ) ∈ Z \lambda= k(k-1)/(m^2-1) \in\mathbb{Z} λ = k ( k − 1 ) / ( m 2 − 1 ) ∈ Z for any m ≥ 3 m \ge3 m ≥ 3 , except the trivial k = m 2 = v k = m^2 = v k = m 2 = v excluded by 2 < k < v − 1 2 < k < v-1 2 < k < v − 1 . In closed form: reducing k ( k − 1 ) ≡ 0 ( m o d m 2 − 1 ) k(k-1) \equiv0 \pmod{m^2-1} k ( k − 1 ) ≡ 0 ( mod m 2 − 1 ) modulo m + 1 m+1 m + 1 (the congruence modulo m − 1 m-1 m − 1 holds automatically since k ≡ 1 ( m o d m − 1 ) k \equiv1 \pmod{m-1} k ≡ 1 ( mod m − 1 ) ) turns it into the constant congruence ( 1 − 2 c 1 + 4 c 2 ) ( 4 c 2 − 2 c 1 ) ≡ 0 ( m o d m + 1 ) (1-2c_1+4c_2)(4c_2-2c_1) \equiv0 \pmod{m+1} ( 1 − 2 c 1 + 4 c 2 ) ( 4 c 2 − 2 c 1 ) ≡ 0 ( mod m + 1 ) ; for m ≥ 20 m \ge20 m ≥ 20 the constant on the left is strictly smaller in absolute value than m + 1 m+1 m + 1 unless it vanishes identically (indeed max c 1 , c 2 ∣ ( 1 − 2 c 1 + 4 c 2 ) ( 4 c 2 − 2 c 1 ) ∣ = 20 \max_{c_1,c_2}|(1-2c_1+4c_2)(4c_2-2c_1)| = 20 max c 1 , c 2 ∣ ( 1 − 2 c 1 + 4 c 2 ) ( 4 c 2 − 2 c 1 ) ∣ = 20 over the allowed pairs, attained at c 1 = 0 , c 2 = 1 c_1 = 0,c_2 = 1 c 1 = 0 , c 2 = 1 , while m + 1 ≥ 21 m+1 \ge21 m + 1 ≥ 21 ), and it vanishes identically exactly for c 1 = c 2 = 0 c_1 = c_2 = 0 c 1 = c 2 = 0 (the trivial k = 1 k = 1 k = 1 ) and c 1 = 2 , c 2 = 1 c_1 = 2,c_2 = 1 c 1 = 2 , c 2 = 1 (giving k = ( 1 + ( m − 1 ) ) 2 = m 2 = v k = (1+(m-1))^2 = m^2 = v k = ( 1 + ( m − 1 ) ) 2 = m 2 = v , trivial); the finitely many remaining values 3 ≤ m ≤ 19 3 \le m \le19 3 ≤ m ≤ 19 are verified directly, together with the fiber identity (14) . Hence the two-orbit PA configuration with ℓ = 2 \ell= 2 ℓ = 2 is excluded for every primitive degree m ≥ 3 m \ge3 m ≥ 3 ;
(ii) for ℓ = 3 , 4 , 5 , 6 \ell= 3,4,5,6 ℓ = 3 , 4 , 5 , 6 , the enumeration of all box multiplicity vectors (bounded by c j ≤ ( ℓ j ) c_j \le\binom{\ell}{j} c j ≤ ( j ℓ ) ) together with the integrality and the fiber identity leaves no survivor; the complete exclusion of the product action configuration is stated in Proposition 7.11.
Consequently, within the two-orbit box framework, the product action configuration is excluded for every primitive degree m ≥ 3 m \ge3 m ≥ 3 when ℓ = 2 \ell= 2 ℓ = 2 , regardless of the underlying non-abelian simple group T T T .
Point-imprimitive actions are excluded
The point-imprimitive case admits no non-trivial non-symmetric block-primitive 2-design.
The strategy is a block-type/point-type pairing argument. Since G G G acts primitively on blocks, the block action falls into one of the five primitive O’Nan–Scott types of Table 2 . Since the point action is quasiprimitive but imprimitive, and since quasiprimitive actions of affine, holomorph simple or holomorph compound type are always primitive (Praeger [35 ] ), only the almost simple, twisted wreath, simple diagonal, compound diagonal and product action point types can occur. Each pair (block type, point type) is then excluded in turn: the parameter constraints and orbit-intersection criteria below dispose of the diagonal and product configurations, and the classical-group analysis rules out all point-imprimitive candidates. For compound socles only the block types PA, SD and TW need to be considered (the diagonal and holomorph block types are handled in § 2.1 ). The outcome is the general exclusion statement of Proposition 3.4 covering every point-imprimitive configuration.
Suppose that G G G acts imprimitively on points. Then there exists a subgroup K K K such that
H = G x < K < G , H = G_x < K < G, H = G x < K < G ,
and the G G G -orbits on the block system of this imprimitivity are the cosets of K K K in G G G . Thus:
s = [ G : K ] = number of blocks in the imprimitivity partition , s = [G : K] = \text{number of blocks in the imprimitivity partition}, s = [ G : K ] = number of blocks in the imprimitivity partition ,
and each block has size:
∣ P i ∣ = [ K : H ] = v / s . |\mathcal{P}_i| = [K : H] = v/s. ∣ P i ∣ = [ K : H ] = v / s .
Key parameter constraints
The design parameters must satisfy additional constraints due to the imprimitivity partition:
(1) Point-pair orbit split. Point pairs split into two types: pairs within the same imprimitivity block, of which there are s ⋅ ( v / s 2 ) s \cdot\binom{v/s}{2} s ⋅ ( 2 v / s ) , and pairs in different imprimitivity blocks, of which there are ( s 2 ) ⋅ ( v / s ) 2 \binom{s}{2} \cdot(v/s)^2 ( 2 s ) ⋅ ( v / s ) 2 . The 2-design condition requires that both types of pairs have the same coverage number λ \lambda λ .
(2) Divisibility conditions. The block size k k k must be compatible with the imprimitivity partition: for a fixed block B B B , with k i = ∣ B ∩ P i ∣ k_i = |B \cap P_i| k i = ∣ B ∩ P i ∣ the intersection sizes with the classes of the partition, one has k = ∑ i = 1 s k i k = \sum_{i=1}^{s} k_i k = ∑ i = 1 s k i ; the multiset { k i } \{k_i\} { k i } of intersection sizes is independent of the chosen block, since G G G permutes the blocks while it permutes the classes.
(3) Parameter equations. v r = b k vr = bk v r = bk and r ( k − 1 ) = λ ( v − 1 ) r(k - 1) = \lambda(v - 1) r ( k − 1 ) = λ ( v − 1 ) .
Classical groups: point-imprimitive candidates are excluded
For classical groups, the point-imprimitive candidates are the chains of parabolic subgroups:
P k < P l < G , k < l , P_k < P_l < G,\qquad k < l, P k < P l < G , k < l ,
with H = P k H = P_k H = P k the stabilizer of a k k k -subspace and K = P l K = P_l K = P l the stabilizer of an l l l -subspace containing it; then v = [ G : P k ] v = [G : P_k] v = [ G : P k ] , s = [ G : P l ] s = [G : P_l] s = [ G : P l ] , and each class of the partition has size [ P l : P k ] [P_l : P_k] [ P l : P k ] .
The block stabilizer M M M is a maximal parabolic subgroup P j P_j P j ; a block is then the set of all k k k -subspaces contained in, or containing, a fixed j j j -subspace. For two points U , V U,V U , V that is, two k k k -subspaces, the number of such blocks through { U , V } \{U,V\} { U , V } equals the number of j j j -subspaces containing U ∨ V U \vee V U ∨ V , which depends on dim ( U ∩ V ) \dim(U \cap V) dim ( U ∩ V ) . Since dim ( U ∩ V ) \dim(U \cap V) dim ( U ∩ V ) takes several values on the point set, the point-pair coverage is not constant, and the 2-design condition fails. Hence no generalized Grassmann design arises in the point-imprimitive setting.
General result for point-imprimitive case
Lemma 3.1 (Block type constrains the point type). Let D \mathcal{D} D be a non-trivial 2-design and let G ≤ Aut ( D ) G \leq\operatorname{Aut}(\mathcal{D}) G ≤ Aut ( D ) act primitively on B \mathcal{B} B and quasiprimitively on P \mathcal{P} P . Since the block action G B G^{\mathcal{B}} G B is faithful (Lemma 2.1 ), the O’Nan–Scott type of the primitive block action restricts the type of the point action as follows (five-type description of Table 2 ):
(1) if G B G^{\mathcal{B}} G B is of affine type, then G P G^{\mathcal{P}} G P is of affine type, hence primitive;
(2) if G B G^{\mathcal{B}} G B is of almost simple type, then G P G^{\mathcal{P}} G P is of almost simple type;
(3) if G B G^{\mathcal{B}} G B is of diagonal, product action or twisted wreath type, then G P G^{\mathcal{P}} G P is neither affine nor almost simple; in particular, an imprimitive quasiprimitive point action is then of simple diagonal, compound diagonal, product action or twisted wreath type.
Proof. The block action is faithful by Lemma 2.1 , so G ≅ G B G \cong G^{\mathcal{B}} G ≅ G B and the minimal normal subgroups of G G G correspond, as abstract groups, to those of G B G^{\mathcal{B}} G B . (1) If G B G^{\mathcal{B}} G B is affine, its socle is an elementary abelian normal subgroup regular on B \mathcal{B} B . The corresponding minimal normal subgroup N N N of G G G is elementary abelian, and quasiprimitivity makes N N N transitive on P \mathcal{P} P , hence regular on P \mathcal{P} P ; G P G^{\mathcal{P}} G P is thus of affine type, which is primitive (Praeger [35 ] ; cf. Theorem 3.3 of Chen, Hua, Li and Wu [15 ] ). (2) If G B G^{\mathcal{B}} G B is almost simple, then G ≅ G B G \cong G^{\mathcal{B}} G ≅ G B is almost simple, so the unique minimal normal subgroup T T T of G G G is non-abelian simple and transitive on P \mathcal{P} P by quasiprimitivity; G P G^{\mathcal{P}} G P is of almost simple type. (3) If G B G^{\mathcal{B}} G B is of diagonal, product action or twisted wreath type, its socle is ≅ T ℓ \cong T^{\ell} ≅ T ℓ with ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 , neither elementary abelian nor simple. The corresponding minimal normal subgroups of G G G are therefore neither elementary abelian nor simple, so G P G^{\mathcal{P}} G P is neither affine nor almost simple. If G P G^{\mathcal{P}} G P is imprimitive, then it is not of holomorph simple or holomorph compound type either, since those types are automatically primitive (Praeger [35 ] ); hence G P G^{\mathcal{P}} G P is of simple diagonal, compound diagonal, product action or twisted wreath type.
The type of a primitive or quasiprimitive action is determined by the pair consisting of the socle and the stabilizer (Praeger [35 ] ); cf. the type-characteristics table of Baddeley, Praeger and Schneider [3 Section 2] . Since the two actions G P G^{\mathcal{P}} G P and G B G^{\mathcal{B}} G B of the same group G G G share their minimal normal subgroups (faithfulness of G B G^{\mathcal{B}} G B , Lemma 2.1 ), the two types can coexist only under the constraints of the following lemma, which we record from Baddeley, Praeger and Schneider [3 ] .
Lemma 3.2 (Blow-up decompositions and type inheritance). Let G G G be a transitive permutation group on Ω \Omega Ω , let E = { Γ 1 , … , Γ ℓ } \mathcal{E} = \{\Gamma_1, \ldots, \Gamma_\ell\} E = { Γ 1 , … , Γ ℓ } be a homogeneous G G G -invariant Cartesian decomposition of Ω \Omega Ω , and let Γ 0 ∈ E \Gamma_0 \in\mathcal{E} Γ 0 ∈ E ; for the notion of a blow-up decomposition see Baddeley, Praeger and Schneider [3 Section 2] .
(1) If E \mathcal{E} E is a blow-up decomposition and G G G is quasiprimitive on Ω \Omega Ω , then the component G Γ 0 G^{\Gamma_0} G Γ 0 is quasiprimitive on Γ 0 \Gamma_0 Γ 0 and Soc ( G ) = ∏ Γ ∈ E Soc ( G Γ ) \operatorname{Soc}(G) = \prod_{\Gamma\in\mathcal{E}} \operatorname{Soc}(G^\Gamma) Soc ( G ) = ∏ Γ ∈ E Soc ( G Γ ) ;
(2) if E \mathcal{E} E is a blow-up decomposition and G Γ 0 G^{\Gamma_0} G Γ 0 is quasiprimitive of non-affine type, then G G G is quasiprimitive on Ω \Omega Ω and the same socle formula holds;
(3) G G G is not quasiprimitive of simple diagonal type (quasiprimitive groups of type S d \mathrm{Sd} Sd preserve no Cartesian decomposition);
(4) if G G G is quasiprimitive of compound diagonal type, then E \mathcal{E} E is a blow-up decomposition and the component G Γ 0 G^{\Gamma_0} G Γ 0 is quasiprimitive of simple or compound diagonal type.
This is Baddeley, Praeger and Schneider [3 Theorem 6.1] .
Remark 3.3. An almost simple group has a unique minimal normal subgroup, which is non-abelian simple, and hence admits no non-trivial blow-up decomposition; consequently every faithful quasiprimitive action of an almost simple group is again of almost simple type. This is the mechanism behind Lemma 3.1 (2). Conversely, a quasiprimitive action of product action or twisted wreath type carries a normal Cartesian decomposition (the coordinate structure of the product action, respectively the natural decomposition of the regular socle; cf. Baddeley, Praeger and Schneider [3 Example 5.5] ); by Lemma 3.2 (3) such an action cannot coexist with a point action of simple diagonal type. Table 4 records the compatible pairs for two faithful quasiprimitive actions of the same group.
Table 4. Compatible O’Nan–Scott types of two faithful quasiprimitive actions of the same group G G G (point action on P \mathcal{P} P , block action on B \mathcal{B} B )
Proposition 3.4 (Point-imprimitive actions are excluded). Let D = ( P , B ) \mathcal{D} = (\mathcal{P}, \mathcal{B}) D = ( P , B ) be a non-trivial 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design with 2 < k < v − 1 2 < k < v-1 2 < k < v − 1 , and let G ≤ Aut ( D ) G \leq\operatorname{Aut}(\mathcal{D}) G ≤ Aut ( D ) act primitively on B \mathcal{B} B . Then G G G acts primitively on P \mathcal{P} P ; in particular, no non-trivial non-symmetric block-primitive 2 2 2 -design is point-imprimitive.
Proof. By Block’s theorem [6 ] , a group transitive on the blocks of a non-trivial 2 2 2 -design is transitive on the points; hence G G G is transitive on P \mathcal{P} P . Assume, for a contradiction, that G G G is imprimitive on P \mathcal{P} P .
Case 1: G G G is not quasiprimitive on P \mathcal{P} P . Then G G G has a non-trivial intransitive normal subgroup on P \mathcal{P} P ; choose 1 < N ◃ G 1 < N \triangleleft G 1 < N ◃ G with N N N intransitive on P \mathcal{P} P . Since G G G is primitive on B \mathcal{B} B , the normal subgroup N B N^{\mathcal{B}} N B of the primitive group G B G^{\mathcal{B}} G B is transitive or trivial on B \mathcal{B} B . If N B N^{\mathcal{B}} N B were transitive, then N N N would be transitive on B \mathcal{B} B , and Block’s theorem [6 ] would force N N N to be transitive on P \mathcal{P} P , a contradiction. Hence N B = 1 N^{\mathcal{B}} = 1 N B = 1 , that is, N N N fixes every block setwise. Consequently every block is a union of N N N -orbits, and the partition E \mathcal{E} E of P \mathcal{P} P into the N N N -orbits is non-trivial and G G G -invariant. Write B 0 = ⋃ E ∈ E ′ E B_0 = \bigcup_{E\in\mathcal{E}'} E B 0 = ⋃ E ∈ E ′ E for a fixed block, with ∅ ≠ E ′ ⊊ E \varnothing\neq\mathcal{E}' \subsetneq\mathcal{E} ∅ = E ′ ⊊ E , since B 0 ∉ { ∅ , P } B_0 \notin\{\varnothing,\mathcal{P}\} B 0 ∈ / { ∅ , P } . Take E ∈ E ′ E \in\mathcal{E}' E ∈ E ′ and E ′ ∈ E ∖ E ′ E' \in\mathcal{E}\setminus\mathcal{E}' E ′ ∈ E ∖ E ′ . For x ≠ y ∈ E x \neq y \in E x = y ∈ E , the number of blocks through { x , y } \{x,y\} { x , y } equals # { g ∈ G : E g − 1 ⊆ B 0 } \#\{g \in G : E^{g^{-1}} \subseteq B_0\} # { g ∈ G : E g − 1 ⊆ B 0 } : a block B = g B 0 B = gB_0 B = g B 0 contains x x x and y y y if and only if it contains the whole orbit E E E , and E ⊆ g B 0 E \subseteq gB_0 E ⊆ g B 0 is equivalent to E g − 1 ⊆ B 0 E^{g^{-1}} \subseteq B_0 E g − 1 ⊆ B 0 . Likewise, for u ∈ E ′ u \in E' u ∈ E ′ , the number of
blocks through { x , u } \{x,u\} { x , u } equals # { g ∈ G : E g − 1 ⊆ B 0 , ( E ′ ) g − 1 ⊆ B 0 } \#\{g\in G:E g^{-1}\subseteq B_0,(E')g^{-1}\subseteq B_0\} # { g ∈ G : E g − 1 ⊆ B 0 , ( E ′ ) g − 1 ⊆ B 0 } . As D \mathcal{D} D is a 2-design these two numbers coincide, while the second set of group elements is contained in the first. The identity belongs to the first set (because E ⊆ B 0 E\subseteq B_0 E ⊆ B 0 ), hence to the second, forcing E ′ ⊆ B 0 E'\subseteq B_0 E ′ ⊆ B 0 , that is, E ′ ∈ E ′ E'\in\mathcal{E}' E ′ ∈ E ′ , a contradiction. Thus G G G is quasiprimitive on P \mathcal{P} P .
Case 2: G G G is quasiprimitive on P \mathcal{P} P . By Lemma 3.1 the type of the primitive block action G B G^{\mathcal{B}} G B (five-type description, Table 2 ) constrains the quasiprimitive point action, and we discuss the block type in turn.
Block action of affine type. By Lemma 3.1 (1) the point action is of affine type, and the affine point type is automatically primitive (Praeger [35 ] ), contradicting the imprimitivity hypothesis. This disposes of all five pairs (HA, ⋅ \cdot ⋅ ) (entry (1) of Table 5 ).
Table 5. The block-type/point-type pairs of Case 2: G G G is primitive on B \mathcal{B} B and quasiprimitive, imprimitive on P \mathcal{P} P . The columns list the quasiprimitive point types, restricted to the five types admitting imprimitive actions (the affine, holomorph simple and holomorph compound types are automatically primitive, Praeger [35 ] ); the entries refer to the exclusion listed below.
Block action of almost simple type. By Lemma 3.1 (2) the point action is of almost simple type; the pairs (AS, SD), (AS, CD), (AS, PA) and (AS, TW) are impossible (entry (2)), and the pair (AS, AS) is disposed of as follows. Imprimitivity means that the point stabilizer H H H is not maximal in G G G ; writing K K K for a maximal subgroup with H < K < G H<K<G H < K < G , the point action preserves the non-trivial partition into the cosets of K K K , and the socle T T T is transitive on P \mathcal{P} P (quasiprimitivity of the AS type is equivalent to G = T H G=TH G = T H ). Such subgroup chains are excluded family by family in § 3.1 : in the classical, exceptional and sporadic families the chains of parabolic subgroups and all other chains fail the constant-coverage test for point pairs (see the point-imprimitive exclusions below).
Block action of simple diagonal type. By Lemma 3.1 (3) the point action is of simple diagonal, compound diagonal, product action or twisted wreath type; the pair (SD, AS) is impossible (entry (3)). The one-sided analysis of § 2.1 excludes all four remaining pairs (SD, ·) (entry (4)): the unconditional necessary conditions of Theorem 2.3 , the one-sided analysis for ℓ = 2 \ell= 2 ℓ = 2 (Proposition 7.6 ) and, for every ℓ ≥ 4 \ell\ge4 ℓ ≥ 4 , the λ \lambda λ -consistency exclusion (Proposition 7.7 ) depend only on the block-diagonal structure b = q ℓ − 1 b = q^{\ell-1} b = q ℓ − 1 and v ∣ q ℓ v \mid q^\ell v ∣ q ℓ , not on the point stabilizer, hence hold verbatim for every quasiprimitive imprimitive point action.
Table 6. The exclusion locations for the pairs of Table 5 in the point-imprimitive case.
Block action of compound diagonal type. Again the point action is of simple diagonal, compound diagonal, product action or twisted wreath type (entry (3) excludes (CD, AS)), and the block-side analysis of Proposition 7.14 excludes all four pairs (CD, ·) (entry (5)): the compound diagonal block stabilizer is shown not to be maximal, independently of the point action.
Block action of holomorph type. The pairs (HS, ·) and (HC, ·) are excluded for every point type (entry (6)): the primitive block action of holomorph type has two commuting transitive minimal normal subgroups M 1 ≅ M 2 M_1 \cong M_2 M 1 ≅ M 2 on B B B ; if M i M_i M i were intransitive on P P P , it would fix every block setwise (each block is a subset of P \mathcal{P} P , and M i M_i M i fixes all points), contradicting its transitivity on B \mathcal{B} B ; hence M 1 , M 2 M_1, M_2 M 1 , M 2 are transitive on P \mathcal{P} P , and two commuting transitive permutation groups are regular (Lemma 2.12), so each M i M_i M i is regular on P \mathcal{P} P and v = ∣ M 1 ∣ = b v = |M_1| = b v = ∣ M 1 ∣ = b , contradicting the non-symmetric hypothesis b ≠ v b \ne v b = v .
Block action of product action or twisted wreath type. By Lemma 3.1 (3) the point action is of simple diagonal, compound diagonal, product action or twisted wreath type. The pairs (PA, SD) and (TW, SD) are excluded by Remark 3.3 (entry (7)): a product action or twisted wreath block action carries a normal Cartesian decomposition (the coordinate structure, respectively the natural decomposition of the regular socle; Baddeley, Praeger and Schneider [3 Example 5.5] ), and groups of type Sd preserve no Cartesian decomposition (Lemma 3.2 (3)). The pairs (PA, TW) and (TW, TW) are excluded by Lemma 2.10 (entry (9)): the point action of twisted wreath type has its socle regular on P \mathcal{P} P , which forces b = v b = v b = v and hence symmetry. The pairs (TW, CD) and (TW, PA) are excluded by Proposition 3.13 (entry (8), first part): a block action of twisted wreath type is impossible altogether for a non-symmetric design, regardless of the point type. The remaining pairs (PA, CD) and (PA, PA) are excluded as follows. The arithmetic reduction of Lemma 3.5 constrains every admissible block size to a union of K ℓ K^\ell K ℓ -orbits on P \mathcal{P} P ; the orbit-intersection criterion Lemma 3.6 together with the non-existence criterion Corollary 3.7 and the primitive-divisor exclusion criterion Lemma 3.8 reduce the exclusion of both pairs to the orbit-integrality condition (*) of Proposition 3.9, and this condition is verified in Lemma 3.10 for the complete A 5 / A 4 A_5/A_4 A 5 / A 4 family over every ℓ ≥ 4 \ell\ge4 ℓ ≥ 4 for both the compound diagonal and the product action point type. For the point-imprimitive product action, of degree v = c ω ℓ v = c\omega^\ell v = c ω ℓ with c ≥ 1 c \ge1 c ≥ 1 the block size of the G G G -invariant partition (Praeger [35 ] , type (viii)), the exclusion is unconditional: by Lemma 3.11 the pair (PA, PA) is excluded for the socles P S L 2 ( q ) PSL_2(q) P S L 2 ( q ) , q ∈ { 5 , 7 , 8 , 9 , 11 } q \in\{5, 7, 8, 9, 11\} q ∈ { 5 , 7 , 8 , 9 , 11 } , for every c ≥ 1 c \ge1 c ≥ 1 , in particular for the c = 1 c = 1 c = 1 and c > 1 c > 1 c > 1 branches alike.
In every case the imprimitive point action contradicts the existence of D \mathcal{D} D . Hence G G G is primitive on P \mathcal{P} P . The geometric candidates arising from chains of parabolic subgroups (generalized Grassmann and building designs) fail the coverage test directly: the coverage number of a point pair { U , V } \{U,V\} { U , V } of k k k -subspaces equals the number of intermediate subspaces containing U ∨ V U \vee V U ∨ V , which depends on dim ( U ∩ V ) \dim(U \cap V) dim ( U ∩ V ) .
Lemma 3.5 (Arithmetic reduction: block type PA or TW with point type
CD or PA). Let D \mathcal{D} D be a non-trivial non-symmetric 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design, and let G ≤ Aut ( D ) G \leq\operatorname{Aut}(\mathcal{D}) G ≤ Aut ( D ) act primitively on B \mathcal{B} B and quasiprimitively, imprimitively on P \mathcal{P} P , with the block action of product action or twisted wreath type and the point action of compound diagonal or product action type. Let N = T ℓ = soc ( G ) N = T^{\ell} = \operatorname{soc}(G) N = T ℓ = soc ( G ) (ℓ ≥ 2 \ell\geq2 ℓ ≥ 2 ), let α ∈ P \alpha\in\mathcal{P} α ∈ P and β ∈ B \beta\in\mathcal{B} β ∈ B with α ∈ β \alpha\in\beta α ∈ β , and write M ω = N ∩ G α M_\omega= N \cap G_\alpha M ω = N ∩ G α and M β = N ∩ G β M_\beta= N \cap G_\beta M β = N ∩ G β .
(1) If the block action is of twisted wreath type, then b = ∣ T ∣ ℓ b = |T|^{\ell} b = ∣ T ∣ ℓ and v = ∣ T ∣ ℓ / ∣ M ω ∣ v = |T|^{\ell}/|M_\omega| v = ∣ T ∣ ℓ /∣ M ω ∣ , so
r = b k v = k ∣ M ω ∣ , λ = k ( k − 1 ) ∣ M ω ∣ 2 ∣ T ∣ ℓ − ∣ M ω ∣ , r = \frac{bk}{v} = k|M_\omega|,\qquad\lambda= \frac{k(k-1)|M_\omega|^2}{|T|^{\ell}-|M_\omega|}, r = v bk = k ∣ M ω ∣ , λ = ∣ T ∣ ℓ − ∣ M ω ∣ k ( k − 1 ) ∣ M ω ∣ 2 ,
and for every 1 ≠ m ∈ M ω 1 \ne m \in M_\omega 1 = m ∈ M ω the order ord ( m ) \operatorname{ord}(m) ord ( m ) divides r = k ∣ M ω ∣ r = k|M_\omega| r = k ∣ M ω ∣ .
(2) If the block action is of product action type, then b = n ℓ b = n^{\ell} b = n ℓ with n = [ T : K ] n = [T:K] n = [ T : K ] the block-coordinate index and K ℓ = M β K^{\ell} = M_\beta K ℓ = M β (K K K a maximal subgroup of T T T ), the block β \beta β is a union of K ℓ K^{\ell} K ℓ -orbits on P \mathcal{P} P (since K ℓ ≤ G β K^{\ell} \leq G_\beta K ℓ ≤ G β ), the orbit of α \alpha α under K ℓ K^{\ell} K ℓ (the “α \alpha α -box”) has size ∣ K ∣ ℓ / ∣ K ℓ ∩ M ω ∣ |K|^{\ell}/|K^{\ell} \cap M_\omega| ∣ K ∣ ℓ /∣ K ℓ ∩ M ω ∣ , and
r = b k v = n ℓ k ∣ M ω ∣ ∣ T ∣ ℓ ∈ Z , λ = r ( k − 1 ) v − 1 ∈ Z . r = \frac{bk}{v} = \frac{n^{\ell}k|M_\omega|}{|T|^{\ell}} \in\mathbb{Z},\qquad\lambda= \frac{r(k-1)}{v-1} \in\mathbb{Z}. r = v bk = ∣ T ∣ ℓ n ℓ k ∣ M ω ∣ ∈ Z , λ = v − 1 r ( k − 1 ) ∈ Z .
(3) If the point action is of compound diagonal type, then M ω M_\omega M ω is a non-simple subdirect subgroup of T ℓ T^{\ell} T ℓ ; by Goursat’s theorem the subdirect subgroups of T 2 T^2 T 2 are the twisted diagonals, which are simple, so the compound diagonal point type forces ℓ ≥ 3 \ell\geq3 ℓ ≥ 3 , and the strip decomposition of M ω M_\omega M ω (the subdirect-product structure of [26 ] ; see also [3 Section 2] ) gives ∣ M ω ∣ = ∣ T ∣ r |M_\omega| = |T|^r ∣ M ω ∣ = ∣ T ∣ r with 1 ≤ r < ℓ 1 \leq r < \ell 1 ≤ r < ℓ . If the point action is of product action type, M ω M_\omega M ω is non-trivial and not subdirect.
Proof. (1) In the twisted wreath block action the socle N N N is regular on B \mathcal{B} B (the type characteristic M β = 1 M_\beta= 1 M β = 1 of Baddley, Praeger and Schneider [3 Section 2] ), so b = ∣ N ∣ = ∣ T ∣ ℓ b = |N| = |T|^{\ell} b = ∣ N ∣ = ∣ T ∣ ℓ ; by quasiprimitivity N N N is transitive on P \mathcal{P} P , so v = ∣ N ∣ / ∣ M ω ∣ v = |N|/|M_\omega| v = ∣ N ∣/∣ M ω ∣ . The displayed values of r r r and λ \lambda λ follow from b k = r v bk = rv bk = r v and λ ( v − 1 ) = r ( k − 1 ) \lambda(v-1) = r(k-1) λ ( v − 1 ) = r ( k − 1 ) . For 1 ≠ m ∈ M ω 1 \ne m \in M_\omega 1 = m ∈ M ω the element m m m fixes α \alpha α and is regular on B \mathcal{B} B (M β = 1 M_\beta= 1 M β = 1 ), hence has no fixed block; it permutes the r r r blocks through α \alpha α without fixed points, so ord ( m ) ∣ r \operatorname{ord}(m) \mid r ord ( m ) ∣ r . (2) In the product action block type the block stabilizer meets the socle in K ℓ K^{\ell} K ℓ ; since K ℓ ≤ G β K^{\ell} \leq G_\beta K ℓ ≤ G β , the block β \beta β is a union of K ℓ K^{\ell} K ℓ -orbits, and the orbit of α \alpha α has the displayed size by orbit–stabilizer. The formulas for r , λ r,\lambda r , λ are the design equations. (3) The classification of the subdirect subgroups of a direct product of non-abelian simple groups (Goursat’s theorem) identifies the subdirect subgroups of T 2 T^2 T 2 with the twisted diagonals { ( t , t φ ) : t ∈ T } \{(t,t^\varphi):t\in T\} {( t , t φ ) : t ∈ T } , all isomorphic to T T T ; a non-simple subdirect subgroup therefore requires ℓ ≥ 3 \ell\ge3 ℓ ≥ 3 . The strip decomposition of a subdirect subgroup of T ℓ T^\ell T ℓ , that is, the subdirect-product structure established in [26 ] (see also [3 Section 2] ), gives ∣ M ω ∣ = ∣ T ∣ r |M_\omega|=|T|^r ∣ M ω ∣ = ∣ T ∣ r with r r r the number of strips; the product action point type has M ω ≠ 1 M_\omega\ne1 M ω = 1 not subdirect by definition.
Lemma 3.6 (Orbit-intersection criterion). Let G G G be transitive on the point set P \mathcal{P} P , let α ∈ P \alpha\in\mathcal{P} α ∈ P , and put M = G α M=G_\alpha M = G α . Let C 1 , … , C r C_1,\ldots,C_r C 1 , … , C r be the M M M -orbits on the complete family of admissible candidate blocks containing α \alpha α , and let m i = ∣ C i ∣ m_i=|C_i| m i = ∣ C i ∣ . Let R 1 , … , R s R_1,\ldots,R_s R 1 , … , R s be the M M M -orbits on P ∖ { α } \mathcal{P}\setminus\{\alpha\} P ∖ { α } . For each i i i choose a representative block B i ∈ C i B_i\in C_i B i ∈ C i ; since M M M is transitive on C i C_i C i , the quantities below are independent of this choice. For 1 ≤ i ≤ r 1\le i\le r 1 ≤ i ≤ r and 1 ≤ j ≤ s 1\le j\le s 1 ≤ j ≤ s define
a i j = ∣ B i ∩ R j ∣ = ∣ ( B i ∖ { α } ) ∩ R j ∣ ( α ∉ R j ) , a_{ij}=|B_i\cap R_j|=|(B_i\setminus\{\alpha\})\cap R_j|\qquad(\alpha\notin R_j), a ij = ∣ B i ∩ R j ∣ = ∣ ( B i ∖ { α }) ∩ R j ∣ ( α ∈ / R j ) , c i j = ∣ { C ∈ C i : β ∈ C } ∣ = m i a i j ∣ R j ∣ ( β ∈ R j ) . c_{ij}=|\{C\in C_i:\beta\in C\}|=\frac{m_i a_{ij}}{|R_j|}\qquad(\beta\in R_j). c ij = ∣ { C ∈ C i : β ∈ C } ∣ = ∣ R j ∣ m i a ij ( β ∈ R j ) . For I ⊆ { 1 , … , r } I\subseteq\{1,\ldots,r\} I ⊆ { 1 , … , r } write C I = ⋃ i ∈ I C i C_I=\bigcup_{i\in I}C_i C I = ⋃ i ∈ I C i and k I = ∑ i ∈ I m i k_I=\sum_{i\in I}m_i k I = ∑ i ∈ I m i . The coverage number of a point β ∈ R j \beta\in R_j β ∈ R j by the blocks of C I C_I C I is
λ j ( I ) = ∑ i ∈ I c i j = 1 ∣ R j ∣ ∑ i ∈ I m i a i j , \lambda_j(I)=\sum_{i\in I}c_{ij}=\frac{1}{|R_j|}\sum_{i\in I}m_i a_{ij}, λ j ( I ) = i ∈ I ∑ c ij = ∣ R j ∣ 1 i ∈ I ∑ m i a ij , and the pairs through α \alpha α are covered uniformly by C I C_I C I if and only if λ 1 ( I ) = ⋯ = λ s ( I ) \lambda_1(I)=\cdots=\lambda_s(I) λ 1 ( I ) = ⋯ = λ s ( I ) . The latter is a necessary condition for C I C_I C I to be a 2 2 2 -( v , k I , λ ) (v,k_I,\lambda) ( v , k I , λ ) design on P \mathcal{P} P ; it is used below as an exclusion test only (if no non-empty I I I yields uniform coverage of the pairs through α \alpha α , then no candidate family is a design), and is not asserted to be sufficient for the full design property, which also requires uniform coverage of the pairs avoiding α \alpha α .
Proof. Count the pairs ( C , β ) (C,\beta) ( C , β ) with C ∈ C i C\in C_i C ∈ C i and β ∈ C ∩ R j \beta\in C\cap R_j β ∈ C ∩ R j . Since M M M is transitive on C i C_i C i , there are m i ⋅ ∣ C ∩ R j ∣ = m i a i j m_i\cdot|C\cap R_j|=m_i a_{ij} m i ⋅ ∣ C ∩ R j ∣ = m i a ij such pairs; since M M M is transitive on R j R_j R j , each point β ∈ R j \beta\in R_j β ∈ R j lies in the same number c i j c_{ij} c ij of blocks of C i C_i C i , giving ∣ R j ∣ c i j |R_j|c_{ij} ∣ R j ∣ c ij . Equating the two counts yields the displayed formula for c i j c_{ij} c ij , and summing over i ∈ I i\in I i ∈ I gives λ j ( I ) \lambda_j(I) λ j ( I ) . A family of blocks is a 2 2 2 -design on P \mathcal{P} P precisely when the coverage number is independent of the point β ≠ α \beta\ne\alpha β = α , i.e. constant in j j j .
Corollary 3.7 (Non-existence criterion). Let C 1 , … , C r \mathcal{C}_1,\ldots,\mathcal{C}_r C 1 , … , C r , m i m_i m i , R j R_j R j , a i j a_{ij} a ij , k I k_I k I and λ j ( I ) \lambda_j(I) λ j ( I ) be as in Lemma 3.6 , and let b I b_I b I be the number of blocks of the candidate family corresponding to I I I . If for every non-empty I ⊆ { 1 , … , r } I \subseteq\{1,\ldots,r\} I ⊆ { 1 , … , r } either the vector ( λ 1 ( I ) , … , λ s ( I ) ) (\lambda_1(I),\ldots,\lambda_s(I)) ( λ 1 ( I ) , … , λ s ( I )) is non-constant or the value
λ I = b I k I ( k I − 1 ) v ( v − 1 ) \lambda_I = \frac{b_I k_I(k_I-1)}{v(v-1)} λ I = v ( v − 1 ) b I k I ( k I − 1 ) is not an integer, then the candidate family yields no non-trivial G G G -invariant 2 2 2 -design.
Lemma 3.8 (Primitive-divisor exclusion criterion). Let q = ∣ T ∣ q=|T| q = ∣ T ∣ , v = q s v=q^s v = q s , and b = n ℓ b=n^\ell b = n ℓ with n = ∣ T : K ∣ n=|T:K| n = ∣ T : K ∣ . For every non-empty union I I I of K ℓ K^\ell K ℓ -orbits on P \mathcal{P} P write k I = ∣ C I ∣ k_I=|\mathcal{C}_I| k I = ∣ C I ∣ . Suppose that for every such I I I there is a prime p I p_I p I with
ν p I ( v − 1 ) > ν p I ( b k I ( k I − 1 ) ) , \nu_{p_I}(v-1)>\nu_{p_I}\bigl(bk_I(k_I-1)\bigr), ν p I ( v − 1 ) > ν p I ( b k I ( k I − 1 ) ) , where ν p \nu_p ν p denotes the p p p -adic valuation. Then λ I = b k I ( k I − 1 ) / ( v ( v − 1 ) ) \lambda_I=bk_I(k_I-1)/(v(v-1)) λ I = b k I ( k I − 1 ) / ( v ( v − 1 )) is non-integral, and no non-empty K ℓ K^\ell K ℓ -orbit union is a 2 2 2 -design.
Proof. Since p I ∤ v p_I\nmid v p I ∤ v , the p I p_I p I -adic valuation of λ I \lambda_I λ I is
ν p I ( λ I ) = ν p I ( b k I ( k I − 1 ) ) − ν p I ( v − 1 ) < 0 , \nu_{p_I}(\lambda_I)=\nu_{p_I}\bigl(bk_I(k_I-1)\bigr)-\nu_{p_I}(v-1)<0, ν p I ( λ I ) = ν p I ( b k I ( k I − 1 ) ) − ν p I ( v − 1 ) < 0 ,
so λ I ∉ Z \lambda_I\notin\mathbb{Z} λ I ∈ / Z , and Corollary 3.7 excludes every non-empty candidate block union. □ \square □
Proposition 3.9 (Orbit-integrality reduction). In the product action block type with point type of compound diagonal or product action, put q = ∣ T ∣ q=|T| q = ∣ T ∣ , v = q s v=q^s v = q s and b = n ℓ b=n^\ell b = n ℓ with n = [ T : K ] n=[T:K] n = [ T : K ] , and let k I k_I k I be the size of an admissible candidate block union, so that 3 ≤ k I ≤ v − 2 3\leq k_I\leq v-2 3 ≤ k I ≤ v − 2 . The two pairs are excluded as soon as the following arithmetic condition holds:
∀ I ≠ ∅ with 3 ≤ k I ≤ v − 2 ∃ p I ∣ ( q s − 1 ) with ν p I ( q s − 1 ) > ν p I ( n ℓ k I ( k I − 1 ) ) . (15) \forall I\ne\varnothing\ \text{with }3\leq k_I\leq v-2\ \exists p_I\mid(q^s-1)\ \text{with }\nu_{p_I}(q^s-1)>\nu_{p_I}\bigl(n^\ell k_I(k_I-1)\bigr). \tag*{(15)} ∀ I = ∅ with 3 ≤ k I ≤ v − 2 ∃ p I ∣ ( q s − 1 ) with ν p I ( q s − 1 ) > ν p I ( n ℓ k I ( k I − 1 ) ) . (15) Since n = [ T : K ] n=[T:K] n = [ T : K ] divides ∣ T ∣ = q |T|=q ∣ T ∣ = q and gcd ( q , q s − 1 ) = 1 \gcd(q,q^s-1)=1 g cd( q , q s − 1 ) = 1 , no prime divisor of q s − 1 q^s-1 q s − 1 divides n ℓ n^\ell n ℓ ; the condition () is therefore equivalent to the divisibility statement that q s − 1 ∤ k I ( k I − 1 ) q^s-1\nmid k_I(k_I-1) q s − 1 ∤ k I ( k I − 1 ) for every admissible I ≠ ∅ I\ne\varnothing I = ∅ . If ( ) holds, then Corollary 3.7 excludes every admissible candidate block union.
Lemma 3.10 (Orbit-integrality condition for the A 5 / A 4 A_5/A_4 A 5 / A 4 family). Let D \mathcal{D} D be a non-trivial non-symmetric 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design, and let G ≤ Aut ( D ) G \leq\operatorname{Aut}(\mathcal{D}) G ≤ Aut ( D ) act primitively on B \mathcal{B} B and quasiprimitively, imprimitively on P \mathcal{P} P , with the block action of product action type and the point action of compound diagonal or product action type, as in Lemma 3.5 . The orbit-integrality condition ( ∗ ) (*) ( ∗ ) of Proposition 3.9 holds for the complete A 5 / A 4 A_5/A_4 A 5 / A 4 family: T = A 5 T=A_5 T = A 5 , K = A 4 K=A_4 K = A 4 (the index-5 5 5 maximal subgroup of A 5 A_5 A 5 ), every ℓ ≥ 4 \ell\geq4 ℓ ≥ 4 , and both point stabilizers M ω = { ( a , b , a , b ) : a , b ∈ T } M_\omega=\{(a,b,a,b):a,b\in T\} M ω = {( a , b , a , b ) : a , b ∈ T } (compound diagonal) and M ω = T × T × { 1 } × { 1 } M_\omega=T \times T \times\{1\} \times\{1\} M ω = T × T × { 1 } × { 1 } (product action), hence Corollary 3.7 excludes every admissible block size for both pairs.
Proof. We apply the orbit-intersection criterion of Lemma 3.6 together with its non-existence criterion Corollary 3.7; for the pairs under consideration the arithmetic of Lemma 3.5 (2)--(3) already excludes every admissible block size at the minimal parameters, and the same integrality obstruction persists throughout the A 5 / A 4 A_5/A_4 A 5 / A 4 family.
Since K = A 4 K=A_4 K = A 4 is maximal in T = A 5 T=A_5 T = A 5 , the double-coset space K \ T / K K\backslash T/K K \ T / K consists of exactly two double cosets, K K K of size 12 12 12 and T ∖ K T\setminus K T ∖ K of size 48 48 48 . The quotient P = T 4 / M ω \mathcal{P}=T^4/M_\omega P = T 4 / M ω is identified with T 2 T^2 T 2 in both cases: for the compound diagonal stabilizer the diagonal strips collapse to the two remaining coordinates, and for the product action stabilizer M ω = T × T × { 1 } × { 1 } M_\omega=T \times T \times\{1\} \times\{1\} M ω = T × T × { 1 } × { 1 } the action is on the last two coordinates. The K 4 K^4 K 4 -orbits on P \mathcal{P} P are the products O i × O j O_i \times O_j O i × O j with O i ∈ { K , T ∖ K } O_i \in\{K,T\setminus K\} O i ∈ { K , T ∖ K } : four orbits of sizes 144 , 576 , 576 , 2304 144,576,576,2304 144 , 576 , 576 , 2304 in the compound diagonal case, and the twenty-five products of the five right K K K -cosets in the product action case (each factor of size 12 12 12 , each orbit of size 144 144 144 ). A block β ∋ α \beta\ni\alpha β ∋ α is a union of K 4 K^4 K 4 -orbits containing the orbit of α \alpha α , so k = 144 , 720 , 1296 , 2448 , 3024 k=144,720,1296,2448,3024 k = 144 , 720 , 1296 , 2448 , 3024 in the compound diagonal case (k = 3600 k=3600 k = 3600 is the whole point set, not an admissible block), and k = 144 t k=144t k = 144 t with 1 ≤ t ≤ 24 1\leq t\leq24 1 ≤ t ≤ 24 in the product action case. Substituting v = 3600 v=3600 v = 3600 and b = 5 4 = 625 b=5^4=625 b = 5 4 = 625 into λ = b k ( k − 1 ) / ( v ( v − 1 ) ) \lambda=bk(k-1)/(v(v-1)) λ = bk ( k − 1 ) / ( v ( v − 1 )) yields a non-integer in every case verified by direct computation with GAP 4.16.1; scripts gap_pa_cd_c6.g, gap_pa_pa_c7.g and the family scan gap_pa_cdpa_c8b.g recorded with the author. Hence no admissible block size survives at the minimal parameters.
The integrality obstruction is not an artifact of the minimal parameters, and in fact the whole family is excluded by Fisher’s inequality alone. In the free-coordinate model the admissible parameters are
b = 5 ℓ , v = 60 ℓ − 2 (compound diagonal) , b = 25 s , v = 60 s (product action) , \begin{aligned}
b &= 5^\ell,\qquad v=60^{\ell-2} \quad\text{(compound diagonal)},\\
b &= 25^s,\qquad v=60^s \quad\text{(product action)},
\end{aligned} b b = 5 ℓ , v = 6 0 ℓ − 2 (compound diagonal) , = 2 5 s , v = 6 0 s (product action) ,
where 60 = ∣ A 5 ∣ 60 = |A_5| 60 = ∣ A 5 ∣ , the compound-diagonal stabilizer M ω M_\omega M ω leaves ℓ − 2 \ell- 2 ℓ − 2 coordinates free, and the product-action stabilizer leaves s s s coordinates free. Since 60 ℓ − 2 > 5 ℓ 60^{\ell-2} > 5^\ell 6 0 ℓ − 2 > 5 ℓ for every ℓ ≥ 4 \ell\ge4 ℓ ≥ 4 (⇔ 12 ℓ > 3600 \Leftrightarrow12^\ell> 3600 ⇔ 1 2 ℓ > 3600 ), and 60 s > 25 s 60^s > 25^s 6 0 s > 2 5 s for every s ≥ 1 s \ge1 s ≥ 1 , in both cases b < v b < v b < v , violating Fisher’s inequality b ≥ v b \ge v b ≥ v for a non-symmetric design. Hence no admissible block size survives for any ℓ ≥ 4 \ell\ge4 ℓ ≥ 4 (compound diagonal) or s ≥ 1 s \ge1 s ≥ 1 (product action). □ \square □
Lemma 3.11 (Fisher exclusion for the point-imprimitive product action). Let D \mathcal{D} D be a non-trivial non-symmetric 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design, and let G ≤ Aut ( D ) G \le\operatorname{Aut}(\mathcal{D}) G ≤ Aut ( D ) act primitively on B \mathcal{B} B and quasiprimitively, imprimitively on P \mathcal{P} P , with the block action of product action type and the point action of product action type. Write b = n ℓ b = n^\ell b = n ℓ with n = [ T : K ] n = [T:K] n = [ T : K ] the block-coordinate index, and v = c ω ℓ v = c\omega^\ell v = c ω ℓ , where ω = ∣ Δ ∣ \omega= |\Delta| ω = ∣Δ∣ is the coordinate degree and c ≥ 1 c \ge1 c ≥ 1 is the block size of the G G G -invariant partition of P \mathcal{P} P (Praeger [35 ] , type (viii); cf. also Example 1.1 of Zhang, Chen and Zhou [42 ] ). Then Fisher’s inequality b ≥ v b \ge v b ≥ v yields
c ω ℓ ≤ n ℓ . c\omega^\ell\le n^\ell. c ω ℓ ≤ n ℓ . In particular, n < ω n < \omega n < ω is impossible for any c ≥ 1 c \ge1 c ≥ 1 . For the socles PSL 2 ( q ) \operatorname{PSL}_2(q) PSL 2 ( q ) with q ∈ { 5 , 7 , 8 , 9 , 11 } q \in\{5,7,8,9,11\} q ∈ { 5 , 7 , 8 , 9 , 11 } , every admissible pair ( n , ω ) (n,\omega) ( n , ω ) consisting of a primitive block degree and a nonprimitive point-coordinate degree satisfies n < ω n < \omega n < ω (verified by direct computation with GAP 4.16.1: the primitive and nonprimitive degree tables are recorded in gap_imp_deg4.log and the exhaustive sieve over the 196 degree pairs in papa_c_all_sieve.py, both kept with the author; the inequality depends only on the degree values, so conjugate representatives of the corresponding subgroup classes are irrelevant); hence c ω ℓ ≥ ω ℓ > n ℓ = b c\omega^\ell\ge\omega^\ell> n^\ell= b c ω ℓ ≥ ω ℓ > n ℓ = b , contradicting Fisher’s inequality for every c ≥ 1 c \ge1 c ≥ 1 . In the boundary case n = ω n = \omega n = ω , Fisher forces c = 1 c = 1 c = 1 and b = v b = v b = v , so that D \mathcal{D} D is symmetric, again contradicting the standing assumption that D \mathcal{D} D is non-symmetric. Consequently no non-trivial non-symmetric block-primitive 2 2 2 -design admits a point-imprimitive product action.
Lemma 3.12 (Pairing count). Let G G G be transitive on the point set P \mathcal{P} P and on the block set B \mathcal{B} B , with N = soc ( G ) N = \operatorname{soc}(G) N = soc ( G ) transitive on P \mathcal{P} P and regular on B \mathcal{B} B . Let α ∈ P \alpha\in\mathcal{P} α ∈ P and let B 0 B_0 B 0 be a block containing α \alpha α . Write M ω = N ∩ G α M_\omega= N \cap G_\alpha M ω = N ∩ G α and identify P \mathcal{P} P with the set of left cosets n M ω nM_\omega n M ω , so that α = M ω \alpha= M_\omega α = M ω . For a point δ ≠ α \delta\ne\alpha δ = α write δ = h δ M ω \delta= h_\delta M_\omega δ = h δ M ω , and define
B 0 h δ − 1 = { y M ω : y h δ M ω ∈ B 0 } . B_0h_\delta^{-1} = \{yM_\omega: yh_\delta M_\omega\in B_0\}. B 0 h δ − 1 = { y M ω : y h δ M ω ∈ B 0 } .
Then
λ ( α , δ ) = ∣ M ω ∣ ⋅ ∣ B 0 ∩ B 0 h δ − 1 ∣ . \lambda(\alpha,\delta)=|M_\omega|\cdot\left|B_0\cap B_0h_\delta^{-1}\right|. λ ( α , δ ) = ∣ M ω ∣ ⋅ B 0 ∩ B 0 h δ − 1 .
Proof. Since N N N is transitive on B \mathcal{B} B , the blocks through α \alpha α are precisely the translates g B 0 gB_0 g B 0 with g M ω ∈ B 0 gM_\omega\in B_0 g M ω ∈ B 0 , and the blocks through both α \alpha α and δ \delta δ are those with g M ω ∈ B 0 gM_\omega\in B_0 g M ω ∈ B 0 and g h δ M ω ∈ B 0 gh_\delta M_\omega\in B_0 g h δ M ω ∈ B 0 . Hence
λ ( α , δ ) = ∣ { g ∈ N : g M ω ∈ B 0 , g h δ M ω ∈ B 0 } ∣ = ∣ { g ∈ N : g M ω ∈ B 0 ∩ B 0 h δ − 1 } ∣ . \begin{aligned}
\lambda(\alpha,\delta)=\left|\{g\in N:gM_\omega\in B_0,\ gh_\delta M_\omega\in B_0\}\right| \\
&=\left|\{g\in N:gM_\omega\in B_0\cap B_0h_\delta^{-1}\}\right|.
\end{aligned} λ ( α , δ ) = ∣ { g ∈ N : g M ω ∈ B 0 , g h δ M ω ∈ B 0 } ∣ = { g ∈ N : g M ω ∈ B 0 ∩ B 0 h δ − 1 } .
Each left coset x M ω ∈ B 0 ∩ B 0 h δ − 1 xM_\omega\in B_0\cap B_0h_\delta^{-1} x M ω ∈ B 0 ∩ B 0 h δ − 1 contributes exactly ∣ M ω ∣ |M_\omega| ∣ M ω ∣ elements g ∈ N g\in N g ∈ N (the coset itself), and distinct left cosets are disjoint, giving the formula. □
Proposition 3.13 (No block action of twisted wreath type). There is no non-trivial non-symmetric 2-design whose automorphism group is primitive on the block set and quasiprimitive on the point set, with the block action of twisted wreath type.
Proof. Assume such a design exists and let N = T ℓ = soc ( G ) N=T^\ell=\operatorname{soc}(G) N = T ℓ = soc ( G ) ; in the twisted wreath block action N N N is regular on B \mathcal{B} B (the type characteristic M β = 1 M_\beta=1 M β = 1 of Baddeley, Praeger and Schneider [3 Section 2] ). Fix α ∈ P \alpha\in\mathcal{P} α ∈ P and a block B 0 ∋ α B_0\ni\alpha B 0 ∋ α , and let M ω = N ∩ G α M_\omega=N\cap G_\alpha M ω = N ∩ G α , so that v = ∣ N ∣ / ∣ M ω ∣ v=|N|/|M_\omega| v = ∣ N ∣/∣ M ω ∣ and b = ∣ N ∣ b=|N| b = ∣ N ∣ .
We first record the structural consequences of the twisted wreath block action.
(a) If M ω = 1 M_\omega=1 M ω = 1 , then v = b v=b v = b and D \mathcal{D} D is symmetric by Fisher’s inequality, contrary to the non-symmetric hypothesis. Hence M ω ≠ 1 M_\omega\ne1 M ω = 1 and the point action of N N N is non-regular. Since N = T ℓ N=T^\ell N = T ℓ is the unique non-abelian minimal normal subgroup of G G G , the point action is not of type HA, HS, HC, AS or TW; it is therefore of type SD, CD or PA.
(b) By Lemma 3.12,
λ ( α , δ ) = ∣ M ω ∣ ⋅ ∣ B 0 ∩ B 0 h δ − 1 ∣ \lambda(\alpha,\delta)=|M_\omega|\cdot\left|B_0\cap B_0h_\delta^{-1}\right| λ ( α , δ ) = ∣ M ω ∣ ⋅ B 0 ∩ B 0 h δ − 1 for every δ ≠ α \delta\ne\alpha δ = α . Summing over δ ≠ α \delta\ne\alpha δ = α and using ∑ δ ≠ α λ ( α , δ ) = r ( k − 1 ) \sum_{\delta\ne\alpha}\lambda(\alpha,\delta)=r(k-1) ∑ δ = α λ ( α , δ ) = r ( k − 1 ) with r = b k / v = ∣ M ω ∣ k r=bk/v=|M_\omega|k r = bk / v = ∣ M ω ∣ k yields the identity
∑ δ ≠ α ∣ B 0 ∩ B 0 h δ − 1 ∣ = k ( k − 1 ) , \sum_{\delta\ne\alpha}\left|B_0\cap B_0h_\delta^{-1}\right|=k(k-1), δ = α ∑ B 0 ∩ B 0 h δ − 1 = k ( k − 1 ) , which is a tautology; in particular the pairing count imposes no further restriction.
(c) Model the point set as the set of left cosets N / M ω N/M_\omega N / M ω and let X = { x ∈ N : x M ω ∈ B 0 } X=\{x\in N:xM_\omega\in B_0\} X = { x ∈ N : x M ω ∈ B 0 } , so that X M ω = X XM_\omega=X X M ω = X and ∣ X ∣ = k ∣ M ω ∣ |X|=k|M_\omega| ∣ X ∣ = k ∣ M ω ∣ . Since N N N is regular on B \mathcal{B} B we have b = ∣ N ∣ b=|N| b = ∣ N ∣ and r = b k / v = k ∣ M ω ∣ r=bk/v=k|M_\omega| r = bk / v = k ∣ M ω ∣ . For two points x M ω xM_\omega x M ω , y M ω yM_\omega y M ω with d = x − 1 y d=x^{-1}y d = x − 1 y , a block g B 0 gB_0 g B 0 contains both points exactly when g − 1 x g^{-1}x g − 1 x and g − 1 y g^{-1}y g − 1 y lie in X X X , i.e. g = x a − 1 g=xa^{-1} g = x a − 1 with a ∈ X a\in X a ∈ X and a d ∈ X ad\in X a d ∈ X ; hence
λ ( x M ω , y M ω ) = ∣ { a ∈ X : a d ∈ X } ∣ . \lambda(xM_\omega,yM_\omega)=|\{a\in X:ad\in X\}|. λ ( x M ω , y M ω ) = ∣ { a ∈ X : a d ∈ X } ∣.
In the integral group ring Z [ N ] \mathbb{Z}[N] Z [ N ] this reads
X ‾ ( − 1 ) X ‾ = ( r − λ ) M ‾ ω + λ N ‾ , r = k ∣ M ω ∣ , (16) \overline{X}^{(-1)}\overline{X}=(r-\lambda)\overline{M}_\omega+\lambda\overline{N},\qquad r=k|M_\omega|, \tag*{(16)} X ( − 1 ) X = ( r − λ ) M ω + λ N , r = k ∣ M ω ∣ , (16)
where Y ‾ = ∑ y ∈ Y y \overline{Y}=\sum_{y\in Y}y Y = ∑ y ∈ Y y and Y ‾ ( − 1 ) = ∑ y ∈ Y y − 1 \overline{Y}^{(-1)}=\sum_{y\in Y}y^{-1} Y ( − 1 ) = ∑ y ∈ Y y − 1 . Taking the sum of the coefficients of (16) gives only the standard parameter identity λ = ∣ M ω ∣ k ( k − 1 ) / ( v − 1 ) \lambda=|M_\omega|k(k-1)/(v-1) λ = ∣ M ω ∣ k ( k − 1 ) / ( v − 1 ) . Equation (16) states that X X X is a relative difference set in N N N with respect to M ω M_\omega M ω : ∣ X ∩ X d − 1 ∣ = λ |X\cap Xd^{-1}|=\lambda ∣ X ∩ X d − 1 ∣ = λ for every d ∈ N ∖ M ω d\in N\setminus M_\omega d ∈ N ∖ M ω and ∣ X ∩ X d − 1 ∣ = r − λ |X\cap Xd^{-1}|=r-\lambda ∣ X ∩ X d − 1 ∣ = r − λ for every 1 ≠ d ∈ M ω 1\ne d\in M_\omega 1 = d ∈ M ω ; this formulation requires no normality of M ω M_\omega M ω in N N N . Were M ω M_\omega M ω normal in N N N , the image of B 0 B_0 B 0 in U = N / M ω U=N/M_\omega U = N / M ω would be a ( ∣ U ∣ , k , λ / ∣ M ω ∣ ) (|U|,k,\lambda/|M_\omega|) ( ∣ U ∣ , k , λ /∣ M ω ∣ ) -difference set. However, in the SD, CD and PA cases of (a) the subgroup M ω = N ∩ G α M_\omega=N\cap G_\alpha M ω = N ∩ G α is respectively a diagonal subgroup T diag T^{\operatorname{diag}} T diag , a strip subgroup of order ∣ T ∣ r |T|^r ∣ T ∣ r with 1 ≤ r < ℓ 1\le r<\ell 1 ≤ r < ℓ , or the coordinate subgroup H ℓ H^\ell H ℓ ; each of these is a proper subgroup of N N N that is not a product of coordinate factors, and the normal subgroups of N = T ℓ N=T^\ell N = T ℓ with T T T non-abelian simple are exactly the coordinate products, so in all three cases M ω M_\omega M ω is not normal in N N N and the difference-set description has to be read at the level of the relative difference set X ⊆ N X\subseteq N X ⊆ N .
(d) The integrality of λ = ∣ M ω ∣ k ( k − 1 ) / ( v − 1 ) \lambda=|M_\omega|k(k-1)/(v-1) λ = ∣ M ω ∣ k ( k − 1 ) / ( v − 1 ) gives the arithmetic obstruction: every prime p p p with p ∣ v − 1 p\mid v-1 p ∣ v − 1 and p ∤ ∣ M ω ∣ p\nmid|M_\omega| p ∤ ∣ M ω ∣ divides k ( k − 1 ) k(k-1) k ( k − 1 ) . In the SD and CD cases gcd ( ∣ T ∣ , v − 1 ) = 1 \gcd(|T|,v-1)=1 g cd( ∣ T ∣ , v − 1 ) = 1 , whence rad ( v − 1 ) ∣ k ( k − 1 ) \operatorname{rad}(v-1)\mid k(k-1) rad ( v − 1 ) ∣ k ( k − 1 ) .
(e) If the point action is of type PA, then by Lemma 3.11 no non-trivial non-symmetric block-primitive 2-design admits a point-imprimitive product action. If the point action is of type SD, then the pair (TW, SD) is excluded by Remark 3.3: as recorded in entry (7) of Table 4 , a group of type SD preserves no normal Cartesian decomposition, whereas the twisted wreath block action carries one, the natural decomposition of the regular socle (Baddeley, Praeger and Schneider [3, Example 5.5]); hence the only surviving point type is CD, for which ∣ M ω ∣ = ∣ T ∣ r |M_\omega|=|T|^r ∣ M ω ∣ = ∣ T ∣ r with 1 ≤ r < ℓ 1\le r<\ell 1 ≤ r < ℓ (Lemma 3.5 (3)).
It remains to rule out the compound diagonal point type CD; by (c)–(e) this amounts to excluding a relative difference set X ⊆ N X\subseteq N X ⊆ N with respect to the strip subgroup M ω M_\omega M ω of order ∣ T ∣ r |T|^r ∣ T ∣ r that satisfies the arithmetic obstruction of (d). The structural, representation-theoretic and arithmetic constraints on such X X X are recorded in Lemma 3.14 , and Lemma 3.14 (v) excludes every admissible configuration arising in the classification.
Lemma 3.14 (Necessary structure for a twisted wreath block action with compound diagonal point type). Let N = T ℓ N=T^\ell N = T ℓ with T T T a non-abelian finite simple group and ℓ ≥ 3 \ell\ge3 ℓ ≥ 3 , and let M ω ≤ N M_\omega\le N M ω ≤ N be a strip subgroup of order ∣ T ∣ r |T|^r ∣ T ∣ r with 1 ≤ r < ℓ 1 \le r < \ell 1 ≤ r < ℓ (the compound diagonal point stabilizer of Lemma 3.5 (3)). Let X ⊆ N X \subseteq N X ⊆ N be a relative difference set with respect to M ω M_\omega M ω as in (16) , with ∣ X ∣ = k ∣ T ∣ r |X|=k|T|^r ∣ X ∣ = k ∣ T ∣ r , λ = k ( k − 1 ) ∣ T ∣ r / ( ∣ T ∣ ℓ − r − 1 ) \lambda=k(k-1)|T|^r/(|T|^{\ell-r}-1) λ = k ( k − 1 ) ∣ T ∣ r / ( ∣ T ∣ ℓ − r − 1 ) and 2 ≤ k ≤ ∣ T ∣ ℓ − r − 2 2 \le k \le|T|^{\ell-r}-2 2 ≤ k ≤ ∣ T ∣ ℓ − r − 2 .
(i) For every 1 ≠ m ∈ M ω 1 \ne m \in M_\omega 1 = m ∈ M ω the order ord ( m ) \operatorname{ord}(m) ord ( m ) divides r = k ∣ T ∣ r r=k|T|^r r = k ∣ T ∣ r (the element m m m fixes α \alpha α and is regular on B \mathcal{B} B , so it permutes the r r r blocks through α \alpha α without fixed points).
(ii) ∣ T ∣ ℓ − r − 1 |T|^{\ell-r}-1 ∣ T ∣ ℓ − r − 1 divides k ( k − 1 ) k(k-1) k ( k − 1 ) ; in particular every primitive prime divisor p p p of ∣ T ∣ ℓ − r − 1 |T|^{\ell-r}-1 ∣ T ∣ ℓ − r − 1 (Zsigmondy’s theorem, with the standard exceptions) satisfies p ≡ 1 ( m o d ℓ − r ) p \equiv1 \pmod{\ell-r} p ≡ 1 ( mod ℓ − r ) , p ∤ ∣ T ∣ p \nmid|T| p ∤ ∣ T ∣ , and p ∣ k p \mid k p ∣ k or p ∣ k − 1 p \mid k-1 p ∣ k − 1 .
(iii) k = ∣ B 0 ∣ k=|B_0| k = ∣ B 0 ∣ is the size of a union of G β G_\beta G β -orbits on the point set P = N / M ω \mathcal{P}=N/M_\omega P = N / M ω , and λ = ∣ X ∩ X d − 1 ∣ \lambda=|X\cap Xd^{-1}| λ = ∣ X ∩ X d − 1 ∣ holds for every d ∈ N ∖ M ω d\in N\setminus M_\omega d ∈ N ∖ M ω as in (16) ; in particular the pair ( k , λ ) (k,\lambda) ( k , λ ) is an integer solution of
k ( k − 1 ) = t ( ∣ T ∣ ℓ − r − 1 ) , λ = t ∣ T ∣ r , 2 ≤ k ≤ ∣ T ∣ ℓ − r − 2 , k(k-1)=t\left(|T|^{\ell-r}-1\right),\qquad\lambda=t|T|^r,\qquad2\le k\le|T|^{\ell-r}-2, k ( k − 1 ) = t ( ∣ T ∣ ℓ − r − 1 ) , λ = t ∣ T ∣ r , 2 ≤ k ≤ ∣ T ∣ ℓ − r − 2 , for some positive integer t t t (here t = λ / ∣ T ∣ r t=\lambda/|T|^r t = λ /∣ T ∣ r ; the integrality of t t t follows from (ii)).
(iv) (Representation-theoretic constraint.) For every irreducible complex representation ρ \rho ρ of N = T ℓ N=T^\ell N = T ℓ whose M ω M_\omega M ω -fixed space is zero, one has ρ ( X ‾ ) = 0 \rho(\overline{X})=0 ρ ( X ) = 0 , where ρ ( X ‾ ) = ∑ x ∈ X ρ ( x ) \rho(\overline{X})=\sum_{x\in X}\rho(x) ρ ( X ) = ∑ x ∈ X ρ ( x ) (this is automatic: X X X is a union of M ω M_\omega M ω -right cosets and ∑ h ∈ M ω ρ ( h ) = ∣ M ω ∣ P H = 0 \sum_{h\in M_\omega}\rho(h)=|M_\omega|P_H=0 ∑ h ∈ M ω ρ ( h ) = ∣ M ω ∣ P H = 0 when the fixed space is zero); more generally, if W W W is the M ω M_\omega M ω -fixed subspace of the representation space of ρ \rho ρ , then the image of (16) under ρ \rho ρ reads
ρ ( X ‾ ) ∗ ρ ( X ‾ ) = ( r − λ ) ∣ M ω ∣ ⋅ id W . \rho(\overline{X})^*\rho(\overline{X})=(r-\lambda)|M_\omega|\cdot\operatorname{id}_W. ρ ( X ) ∗ ρ ( X ) = ( r − λ ) ∣ M ω ∣ ⋅ id W . In the compound diagonal situation this forces, for every r r r -tuple ( τ 1 , … , τ r ) (\tau_1,\ldots,\tau_r) ( τ 1 , … , τ r ) of non-trivial irreducible representations of T T T with ( τ 1 ⊗ ⋯ ⊗ τ r ) T = 0 (\tau_1\otimes\cdots\otimes\tau_r)^T=0 ( τ 1 ⊗ ⋯ ⊗ τ r ) T = 0 , the vanishing
∑ x ∈ X τ 1 ( x 1 ) ⊗ ⋯ ⊗ τ r ( x r ) = 0 \sum_{x\in X}\tau_1(x_1)\otimes\cdots\otimes\tau_r(x_r)=0 x ∈ X ∑ τ 1 ( x 1 ) ⊗ ⋯ ⊗ τ r ( x r ) = 0 (the remaining ℓ − r \ell-r ℓ − r coordinates acting trivially).
(v) The equations of (iii) admit arithmetic solutions for some parameter triples (for example T = A 5 T=A_5 T = A 5 with ℓ − r = 2 \ell-r=2 ℓ − r = 2 has the solution k = 1770 k=1770 k = 1770 , t = 870 t=870 t = 870 ; the complete list for the simple groups arising in the classification is recorded in the accompanying arithmetic sieve, script tw_cd_sieve.py, kept with the author). Each surviving arithmetic candidate must in addition satisfy the orbit-union condition of (iii): the block B 0 B_0 B 0 is a union of G β G_\beta G β -orbits on P = N / M ω \mathcal{P}=N/M_\omega P = N / M ω , where the twisted wreath block stabilizer G β G_\beta G β is described by the structure theory of Baddeley, Praeger and Schneider [3 Section 2] ; the candidates are excluded one by one by that orbit-union condition, as recorded in the output of tw_cd_sieve.py.
Proof. (i) is exactly the argument of Lemma 3.5 (1), which is recalled in the statement. (ii) The integrality of λ = k ( k − 1 ) ∣ T ∣ r / ( ∣ T ∣ ℓ − r − 1 ) \lambda=k(k-1)|T|^r/(|T|^{\ell-r}-1) λ = k ( k − 1 ) ∣ T ∣ r / ( ∣ T ∣ ℓ − r − 1 ) together with gcd ( ∣ T ∣ ℓ − r − 1 , ∣ T ∣ r ) = 1 \gcd(|T|^{\ell-r}-1,|T|^r)=1 g cd( ∣ T ∣ ℓ − r − 1 , ∣ T ∣ r ) = 1 gives ∣ T ∣ ℓ − r − 1 ∣ k ( k − 1 ) |T|^{\ell-r}-1\mid k(k-1) ∣ T ∣ ℓ − r − 1 ∣ k ( k − 1 ) ; the second sentence is Zsigmondy’s theorem. (iii) The relative difference set equations (16) read ∣ X ∩ X d − 1 ∣ = λ |X\cap Xd^{-1}|=\lambda ∣ X ∩ X d − 1 ∣ = λ for d ∈ N ∖ M ω d\in N\setminus M_\omega d ∈ N ∖ M ω ; writing k ( k − 1 ) = λ ( ∣ T ∣ ℓ − r − 1 ) / ∣ T ∣ r k(k-1)=\lambda(|T|^{\ell-r}-1)/|T|^r k ( k − 1 ) = λ ( ∣ T ∣ ℓ − r − 1 ) /∣ T ∣ r and taking t = λ / ∣ T ∣ r t=\lambda/|T|^r t = λ /∣ T ∣ r gives the displayed equations, and t t t is a positive integer by (ii). (iv) The vanishing statement follows from the coset decomposition X = ⋃ i x i M ω X=\bigcup_i x_iM_\omega X = ⋃ i x i M ω and ∑ h ∈ M ω ρ ( h ) = ∣ M ω ∣ P H = 0 \sum_{h\in M_\omega}\rho(h)=|M_\omega|P_H=0 ∑ h ∈ M ω ρ ( h ) = ∣ M ω ∣ P H = 0 ; the operator identity is the image of (16) under ρ \rho ρ . (v) The orbit-union computation of the G β G_\beta G β -orbits on P = N / M ω \mathcal{P}=N/M_\omega P = N / M ω and the sieve over the arithmetic candidates of (iii) are described in the statement.
Part III: The almost simple classification
Almost simple type (AS)
The almost simple case is treated in detail. By Lemma 2.1 the block action of G G G on B \mathcal{B} B is faithful; let T = soc ( G ) T = \operatorname{soc}(G) T = soc ( G ) be the unique non-abelian minimal normal subgroup. Then T ≤ G ≤ Aut ( T ) T \leq G \leq\operatorname{Aut}(T) T ≤ G ≤ Aut ( T ) .
The classification proceeds in three steps. First, the 2-homogeneous point actions are disposed of by an orbit construction (Lemma 4.2 ): each such action produces its designs from subgroup orbits and requires no further analysis. Second, the ordering principle (Lemma 4.3 ) reduces the non-symmetric case to pairs of maximal subgroups ( H , M ) (H, M) ( H , M ) with ∣ H ∣ > ∣ M ∣ |H| > |M| ∣ H ∣ > ∣ M ∣ , and block-primitivity forces M M M to be maximal in G G G . Third, every family of finite simple groups is treated in turn: the groups PSL 2 ( q ) \operatorname{PSL}_2(q) PSL 2 ( q ) directly, the higher-rank classical groups through the Aschbacher subdivision of their maximal subgroups, the exceptional Lie type groups through their rank-3 building actions, and the sporadic groups by exhaustive enumeration of maximal-subgroup orbits. Each candidate quadruple ( T , G , H , M ) (T, G, H, M) ( T , G , H , M ) is checked against the design equations by orbit-union analysis and by the integrality of r r r and λ \lambda λ . The outcome is the classification recorded in Proposition 4.1 .
By Theorem 1.3 the point action is primitive, and the classification below is restricted to point actions that are primitive but not 2-homogeneous; the point-primitive case is classified completely as follows.
Proposition 4.1 (Point-primitive non-2-homogeneous almost simple classification). We classify non-symmetric block-primitive 2-designs with almost simple automorphism group G G G acting primitively but not 2-homogeneously on points, across all families of finite simple groups, following the ordering principle ∣ H ∣ > ∣ M ∣ |H| > |M| ∣ H ∣ > ∣ M ∣ for non-symmetric designs and, for the classical families, the Aschbacher subdivision of the maximal subgroups:
(1) Classical Lie type: for PSL n ( q ) \operatorname{PSL}_n(q) PSL n ( q ) (n ≥ 3 n \geq3 n ≥ 3 ), PSp 2 n ( q ) \operatorname{PSp}_{2n}(q) PSp 2 n ( q ) , PSU n ( q ) \operatorname{PSU}_n(q) PSU n ( q ) and PΩ n ε ( q ) \operatorname{P\Omega}^{\varepsilon}_n(q) PΩ n ε ( q ) , the designs are the unital 2 2 2 -( q 3 + 1 , q + 1 , 1 ) (q^3 + 1, q + 1, 1) ( q 3 + 1 , q + 1 , 1 ) carried by the rank-3 action on a polar space (its parameter set is recorded in the existence column of Table 1 ; the unital parameters are classical, see [5 ] ) and the C 8 C_8 C 8 cross-type orbit designs of Proposition 4.11, with their parameter sets recorded there; for PSL 2 ( q ) \operatorname{PSL}_2(q) PSL 2 ( q ) the same theorem gives no design when q q q is odd, the eight designs of Table 7 when q = 8 q = 8 q = 8 , the two designs 2 2 2 -( 496 , 186 , 74 ) (496, 186, 74) ( 496 , 186 , 74 ) and 2 2 2 -( 496 , 310 , 206 ) (496, 310, 206) ( 496 , 310 , 206 ) when q = 32 q = 32 q = 32 , and no design when q = 16 q = 16 q = 16 , the even-characteristic subfield block stabilizers being delimited in Theorem 4.6 ; the generalized Grassmann candidates arising from
Table 7. q = 8 q = 8 q = 8 : the unions of orbits with constant covering numbers. The rows marked yes give block-primitive designs; in the rows marked no the block action is imprimitive, since S 3 < D 18 ≤ G S_3 < D_{18} \leq G S 3 < D 18 ≤ G .
chains of parabolic subgroups fail the constant-coverage test and yield no design.
(2) Sporadic groups: on the primitive non-2 2 2 -homogeneous point actions the classification is exhaustive: the only designs are J 2 J_2 J 2 on 100 points, 2 2 2 -( 100 , 56 , 560 ) (100,56,560) ( 100 , 56 , 560 ) , and H S HS H S on 100 points, 2 2 2 -( 100 , 64 , 1680 ) (100,64,1680) ( 100 , 64 , 1680 ) ; no such design exists on M 12 ( 66 ) M_{12}(66) M 12 ( 66 ) , M 22 ( 176 ) M_{22}(176) M 22 ( 176 ) , M 23 ( 253 ) M_{23}(253) M 23 ( 253 ) , M 24 ( 276 ) M_{24}(276) M 24 ( 276 ) , M c L ( 275 ) McL(275) M c L ( 275 ) , H e ( 2058 ) He(2058) H e ( 2058 ) , R u ( 4060 ) Ru(4060) R u ( 4060 ) , S u z ( 1782 ) Suz(1782) S u z ( 1782 ) or C o 2 ( 2300 ) Co_2(2300) C o 2 ( 2300 ) . The 2 2 2 -homogeneous actions of C o 3 Co_3 C o 3 on 276 points and of H S HS H S on 176 points give designs by the orbit construction of Lemma 4.2 .
(3) Exceptional Lie type: the designs are the Suzuki–Tits ovoid designs and the Ree unital designs 2 2 2 -( q 3 + 1 , q + 1 , 1 ) (q^3+1,q+1,1) ( q 3 + 1 , q + 1 , 1 ) , with the group and the action recorded for each row of Table 10 ; the further primitive actions on building points are recorded there together with the reason for their exclusion.
Table 10. The exceptional Lie type groups, the designs they carry and the reason for exclusion.
We first dispose of the 2 2 2 -homogeneous point actions, which produce their designs by an orbit construction and therefore need no further analysis. Then we prove a key structural result for the AS case.
Lemma 4.2 (Two-homogeneous point actions produce designs from subgroup orbits). Let G G G act 2 2 2 -homogeneously on a finite point set P \mathcal{P} P with v = ∣ P ∣ v=|\mathcal{P}| v = ∣ P ∣ , and let M M M be a maximal subgroup of G G G with an orbit union Ω \Omega Ω on P \mathcal{P} P (a union of M M M -orbits) of size k k k , 2 ≤ k ≤ v − 2 2\leq k\leq v-2 2 ≤ k ≤ v − 2 . Then the stabilizer of Ω \Omega Ω in G G G is exactly M M M , and the translates Ω g \Omega^g Ω g , g ∈ G g\in G g ∈ G , form the block set of a non-trivial 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design whose automorphism group contains G G G acting primitively on blocks.
Proof. Since M ≤ G Ω M\leq G_{\Omega} M ≤ G Ω and M M M is maximal, the stabilizer G Ω G_{\Omega} G Ω is either M M M or G G G ; the latter is impossible because Ω \Omega Ω is a proper non-empty subset of P \mathcal{P} P (2 ≤ k ≤ v − 2 2\leq k\leq v-2 2 ≤ k ≤ v − 2 ). Hence G Ω = M G_{\Omega}=M G Ω = M and G G G acts primitively on the block set { Ω g : g ∈ G } \{\Omega^g:g\in G\} { Ω g : g ∈ G } . It remains to check the design property. As G G G is 2 2 2 -homogeneous, it is transitive on the unordered pairs of points; hence the number λ \lambda λ of translates of Ω \Omega Ω containing a given unordered pair { x , y } \{x,y\} { x , y } is independent of the pair:
λ = ∣ { g ∈ G : { x , y } ⊆ Ω g } ∣ ∣ G Ω ∣ , \lambda=\frac{|\{g\in G:\{x,y\}\subseteq\Omega^g\}|}{|G_{\Omega}|}, λ = ∣ G Ω ∣ ∣ { g ∈ G : { x , y } ⊆ Ω g } ∣ ,
and transitivity on pairs shows this is a constant. The design is non-trivial by 2 ≤ k ≤ v − 2 2\leq k\leq v-2 2 ≤ k ≤ v − 2 . □
Consequently, a 2 2 2 -homogeneous point action yields its non-trivial 2 2 2 -designs by this orbit construction alone, and the classification below is restricted to point actions that are primitive but not 2 2 2 -homogeneous.
By Lemma 2.1 and Proposition 3.4 , G G G acts transitively on the point set P \mathcal{P} P . The AS classification therefore reduces to the analysis of pairs of maximal subgroups ( H , M ) (H,M) ( H , M ) of G G G , where H = G x H=G_x H = G x is the point stabilizer and M = G B M=G_B M = G B is the block stabilizer. The block B B B must be an M M M -invariant subset of P = G / H \mathcal{P}=G/H P = G / H .
Key ordering principle for non-symmetric designs
A fundamental observation for the classification of non-symmetric designs is the following:
Lemma 4.3 (Ordering of stabilizer orders). For a non-symmetric block-primitive 2-design (b > v b>v b > v ), the point stabilizer H H H has strictly larger order than the block stabilizer M M M :
Proof. By definition, v = [ G : H ] v=[G:H] v = [ G : H ] and b = [ G : M ] b=[G:M] b = [ G : M ] . The non-symmetric condition b > v b>v b > v is equivalent to:
∣ G ∣ ∣ M ∣ > ∣ G ∣ ∣ H ∣ ⟺ ∣ H ∣ > ∣ M ∣ . \frac{|G|}{|M|}>\frac{|G|}{|H|}\Longleftrightarrow|H|>|M|. ∣ M ∣ ∣ G ∣ > ∣ H ∣ ∣ G ∣ ⟺ ∣ H ∣ > ∣ M ∣.
This completes the proof. □ \square □
This ordering principle is the cornerstone of our classification: for each family of almost simple groups, we enumerate all maximal subgroups, order them by increasing order, and consider only pairs ( H , M ) (H,M) ( H , M ) where H H H comes after M M M in this ordering (i.e., ∣ H ∣ > ∣ M ∣ |H|>|M| ∣ H ∣ > ∣ M ∣ ). This dramatically reduces the number of candidate pairs, as we only need to consider O ( n 2 / 2 ) O(n^2/2) O ( n 2 /2 ) pairs instead of O ( n 2 ) O(n^2) O ( n 2 ) .
We distinguish two cases for the point action:
Point-primitive: H H H is a maximal subgroup of G G G . This is the case analyzed in detail in the following sections.
Point-imprimitive: H H H is not a maximal subgroup of G G G . This means there exists an intermediate subgroup K K K with H < K < G H<K<G H < K < G , giving a non-trivial G G G -invariant partition of P \mathcal{P} P .
By Lemma 3.1 (2) the only point-imprimitive candidates left after the reduction are the almost simple subgroup chains; we record their exclusion family by family, in agreement with Proposition 3.4 . In every family the point-imprimitive candidates are excluded, and the classification below covers the point-primitive case alone.
Classification for PSL2(q)
The smallest classical group T = P S L 2 ( q ) T = PSL_2(q) T = P S L 2 ( q ) , q = p f q = p^{f} q = p f , is treated in full. In the statements of Cases 1 and 2 the symbols D q − 1 D_{q-1} D q − 1 and D q + 1 D_{q+1} D q + 1 denote the dihedral groups of order q − 1 q-1 q − 1 and q + 1 q+1 q + 1 . When q q q is even the integers q ∓ 1 q \mp1 q ∓ 1 are odd, so neither group occurs in P S L 2 ( q ) PSL_2(q) P S L 2 ( q ) ; the dihedral maximal subgroups then have orders 2 ( q − 1 ) 2(q-1) 2 ( q − 1 ) and 2 ( q + 1 ) 2(q+1) 2 ( q + 1 ) , and the two cases are rebuilt from those subgroups in Subsection 4.3.
The action of P S L 2 ( q ) PSL_2(q) P S L 2 ( q ) on the projective line P G ( 1 , q ) PG(1,q) P G ( 1 , q ) with v = q + 1 v=q+1 v = q + 1 , corresponding to the Borel subgroup H = q : ( q − 1 ) / 2 H = q : (q-1)/2 H = q : ( q − 1 ) /2 , is 2-transitive for every prime power q q q . Since the classification in this paper concerns non-2-homogeneous point actions, this whole case is excluded from the classification below; the 2-homogeneous case is classical (a 2-homogeneous group acting on P \mathcal{P} P produces a 2-design from every block orbit, with constant λ \lambda λ , and such designs of P S L 2 ( q ) PSL_2(q) P S L 2 ( q ) on the projective line are covered by the literature on orbit designs). We focus on non-2-homogeneous point actions, where the point stabilizer H H H is a maximal subgroup of a different type.
Case 1: point stabilizer H = D q − 1 H = D_{q-1} H = D q − 1 (dihedral of order q − 1 q-1 q − 1 )
Here v = [ P S L 2 ( q ) : D q − 1 ] = q ( q + 1 ) / 2 v = [PSL_2(q) : D_{q-1}] = q(q+1)/2 v = [ P S L 2 ( q ) : D q − 1 ] = q ( q + 1 ) /2 . By Dickson’s classification of the maximal subgroups of P S L 2 ( q ) PSL_2(q) P S L 2 ( q ) [22 ] , every maximal subgroup of P S L 2 ( q ) PSL_2(q) P S L 2 ( q ) , q = p f q = p^{f} q = p f , is, up to conjugacy, the Borel subgroup q : ( q − 1 ) / 2 q : (q-1)/2 q : ( q − 1 ) /2 , one of the two dihedral subgroups D q ± 1 D_{q \pm1} D q ± 1 (each maximal unless it is contained in a Borel or subfield subgroup), one of the exceptional subgroups A 4 A_4 A 4 , S 4 S_4 S 4 , A 5 A_5 A 5 (subject to the congruences q ≡ ± 1 ( m o d 8 ) q \equiv\pm1 \pmod{8} q ≡ ± 1 ( mod 8 ) , q 2 ≡ 1 ( m o d 16 ) q^2 \equiv1 \pmod{16} q 2 ≡ 1 ( mod 16 ) , q ≡ ± 1 ( m o d 5 ) q \equiv\pm1 \pmod{5} q ≡ ± 1 ( mod 5 ) , respectively), or a subfield subgroup P S L 2 ( q 0 ) PSL_2(q_0) P S L 2 ( q 0 ) or P G L 2 ( q 0 ) PGL_2(q_0) P G L 2 ( q 0 ) with q = q 0 f q = q_0^{f} q = q 0 f , f f f prime. In the block search below we range over the maximal subgroups M M M with ∣ M ∣ < ∣ H ∣ |M| < |H| ∣ M ∣ < ∣ H ∣ : by Fisher’s inequality the remaining maximal subgroups (with [ G : M ] ≤ [ G : H ] = v [G : M] \leq[G : H] = v [ G : M ] ≤ [ G : H ] = v ) cannot carry a non-symmetric design, and the equality case ∣ M ∣ = ∣ H ∣ |M| = |H| ∣ M ∣ = ∣ H ∣ gives a symmetric design (outside the present classification by duality). Since ∣ D q − 1 ∣ = q − 1 |D_{q-1}| = q-1 ∣ D q − 1 ∣ = q − 1 , the admissible candidates in Case 1 are as follows.
(i) M = D q + 1 M = D_{q+1} M = D q + 1 : here b = [ P S L 2 ( q ) : D q + 1 ] = q ( q − 1 ) / 2 < v b = [PSL_2(q) : D_{q+1}] = q(q-1)/2 < v b = [ P S L 2 ( q ) : D q + 1 ] = q ( q − 1 ) /2 < v , contradicting Fisher’s inequality; excluded for every q q q .
(ii) M ∈ { A 5 , S 4 , A 4 } M \in\{A_5, S_4, A_4\} M ∈ { A 5 , S 4 , A 4 } : the block B B B is the M M M -orbit of the base point, of size k = ∣ M ∣ / ∣ M ∩ H g ∣ k = |M|/|M \cap H^g| k = ∣ M ∣/∣ M ∩ H g ∣ for the appropriate conjugate. Since the elements of A 5 ∩ D q − 1 A_5 \cap D_{q-1} A 5 ∩ D q − 1 have order in { 1 , 2 , 3 , 5 } \{1, 2, 3, 5\} { 1 , 2 , 3 , 5 } and those of S 4 ∩ D q − 1 S_4 \cap D_{q-1} S 4 ∩ D q − 1 , respectively A 4 ∩ D q − 1 A_4 \cap D_{q-1} A 4 ∩ D q − 1 , have order in { 1 , 2 , 3 , 4 } \{1,2,3,4\} { 1 , 2 , 3 , 4 } , respectively { 1 , 2 , 3 } \{1,2,3\} { 1 , 2 , 3 } , the realizable block sizes are k ∈ { 12 , 20 , 30 , 60 } k \in\{12,20,30,60\} k ∈ { 12 , 20 , 30 , 60 } for M = A 5 M=A_5 M = A 5 , k ∈ { 4 , 6 , 8 , 12 , 24 } k \in\{4,6,8,12,24\} k ∈ { 4 , 6 , 8 , 12 , 24 } for M = S 4 M=S_4 M = S 4 , and k ∈ { 4 , 6 , 12 } k \in\{4,6,12\} k ∈ { 4 , 6 , 12 } for M = A 4 M=A_4 M = A 4 . With b = q ( q 2 − 1 ) / ( 2 ∣ M ∣ ) b=q(q^2-1)/(2|M|) b = q ( q 2 − 1 ) / ( 2∣ M ∣ ) and v = q ( q + 1 ) / 2 v=q(q+1)/2 v = q ( q + 1 ) /2 , the integrality of r = b k / v = ( q − 1 ) k / ∣ M ∣ r=bk/v=(q-1)k/|M| r = bk / v = ( q − 1 ) k /∣ M ∣ and of λ = r ( k − 1 ) / ( v − 1 ) = 2 k ( k − 1 ) / ( ∣ M ∣ ( q + 2 ) ) \lambda=r(k-1)/(v-1)=2k(k-1)/(|M|(q+2)) λ = r ( k − 1 ) / ( v − 1 ) = 2 k ( k − 1 ) / ( ∣ M ∣ ( q + 2 )) must hold. The divisibility condition ∣ M ∣ ( q + 2 ) ∣ 2 k ( k − 1 ) |M|(q+2) \mid2k(k-1) ∣ M ∣ ( q + 2 ) ∣ 2 k ( k − 1 ) is decided for the finitely many listed k k k : for M = A 5 M=A_5 M = A 5 , q + 2 ∣ 118 q+2 \mid118 q + 2 ∣ 118 (k = 60 k=60 k = 60 ), q + 2 ∣ 29 q+2 \mid29 q + 2 ∣ 29 (k = 30 k=30 k = 30 ), k = 20 , 12 k=20,12 k = 20 , 12 give non-integral quotients; the only admissible prime powers are q = 27 q=27 q = 27 (k = 30 k=30 k = 30 ), which violates q ≡ ± 1 ( m o d 5 ) q \equiv\pm1 \pmod{5} q ≡ ± 1 ( mod 5 ) , so A 5 ≰ P S L 2 ( 27 ) A_5 \nleq PSL_2(27) A 5 ≰ P S L 2 ( 27 ) . For M = S 4 M=S_4 M = S 4 , q + 2 ∣ 46 q+2 \mid46 q + 2 ∣ 46 (k = 24 k=24 k = 24 ) gives no prime power, q + 2 ∣ 11 q+2 \mid11 q + 2 ∣ 11 (k = 12 k=12 k = 12 ) gives q = 9 q=9 q = 9 , but then H = D 8 H=D_8 H = D 8 is not maximal in P S L 2 ( 9 ) ≅ A 6 PSL_2(9) \cong A_6 P S L 2 ( 9 ) ≅ A 6 ; k = 8 , 6 , 4 k=8,6,4 k = 8 , 6 , 4 give non-integral quotients. For M = A 4 M=A_4 M = A 4 , q + 2 ∣ 22 q+2 \mid22 q + 2 ∣ 22 (k = 12 k=12 k = 12 ) gives q ∈ { 9 , 20 } q \in\{9,20\} q ∈ { 9 , 20 } , where A 4 A_4 A 4 is not maximal in P S L 2 ( 9 ) PSL_2(9) P S L 2 ( 9 ) and q = 20 q=20 q = 20 is not a prime power; q + 2 ∣ 5 q+2 \mid5 q + 2 ∣ 5 (k = 6 k=6 k = 6 ) gives q = 3 q=3 q = 3 with k = v k=v k = v ; k = 4 k=4 k = 4 has no solution. Hence the exceptional candidates admit no design in Case 1.
(iii) the Borel subgroup has order q ( q − 1 ) / 2 > q − 1 q(q-1)/2 > q-1 q ( q − 1 ) /2 > q − 1 and the subfield subgroups have order at least q 0 ( q 0 2 − 1 ) > q − 1 q_0(q_0^2-1)>q-1 q 0 ( q 0 2 − 1 ) > q − 1 (q = q 0 f q=q_0^f q = q 0 f , f ≥ 2 f \ge2 f ≥ 2 ); both violate ∣ M ∣ < ∣ H ∣ |M|<|H| ∣ M ∣ < ∣ H ∣ and are excluded by Fisher’s inequality.
Thus no non-symmetric block-primitive 2-design arises with point stabilizer H = D q − 1 H=D_{q-1} H = D q − 1 .
Case 2: point stabilizer H = D q + 1 H=D_{q+1} H = D q + 1 (dihedral of order q + 1 q+1 q + 1 )
Here v = [ P S L 2 ( q ) : D q + 1 ] = q ( q − 1 ) / 2 v=[PSL_2(q):D_{q+1}]=q(q-1)/2 v = [ P S L 2 ( q ) : D q + 1 ] = q ( q − 1 ) /2 . As in Case 1 the Borel and subfield subgroups have order exceeding q + 1 q+1 q + 1 and are excluded by Fisher’s inequality, and M = H M=H M = H gives the symmetric case. The admissible candidates with ∣ M ∣ < ∣ H ∣ = q + 1 |M|<|H|=q+1 ∣ M ∣ < ∣ H ∣ = q + 1 are M = D q − 1 M=D_{q-1} M = D q − 1 and the exceptional subgroups.
Throughout this case a candidate base block is a union B 0 B_0 B 0 of M M M -orbits containing p 0 p_0 p 0 , and the condition that λ \lambda λ be constant forces a linear system which we record once for reference. Let ℓ o = ∣ o ∣ \ell_o=|o| ℓ o = ∣ o ∣ be the length of the M M M -orbit o o o on P \mathcal{P} P , let j j j run over the G G G -orbits of point pairs, let μ j \mu_j μ j be the number of pairs in j j j , and let c o o ′ ( j ) c_{oo'}^{(j)} c o o ′ ( j ) be the number of pairs { p , r } \{p,r\} { p , r } with p ∈ o p\in o p ∈ o , r ∈ o ′ r\in o' r ∈ o ′ that lie in j j j (the case o = o ′ o=o' o = o ′ , that is, both points in one orbit, is allowed). Then a union B 0 B_0 B 0 of M M M -orbits with k = ∣ B 0 ∣ k=|B_0| k = ∣ B 0 ∣ has constant λ \lambda λ only if
∑ o ℓ o x o = k , ∑ o , o ′ c o o ′ ( j ) x o x o ′ = μ j λ b ( 1 ≤ j ≤ s ) , x o ∈ { 0 , 1 } , (17) \sum_o \ell_o x_o = k,\qquad
\sum_{o,o'} c_{oo'}^{(j)}x_ox_{o'}=\frac{\mu_j\lambda}{b}\quad(1\le j\le s),\qquad x_o\in\{0,1\}, \tag*{(17)} o ∑ ℓ o x o = k , o , o ′ ∑ c o o ′ ( j ) x o x o ′ = b μ j λ ( 1 ≤ j ≤ s ) , x o ∈ { 0 , 1 } , (17)
where x o = 1 x_o = 1 x o = 1 exactly when o ⊆ B 0 o \subseteq B_0 o ⊆ B 0 : the second group of equations is the double count of the pairs of the orbit j j j against the b b b blocks, each pair lying in λ \lambda λ of them. Note that (17) is a necessary condition only, so an infeasible system excludes the corresponding unions, while a feasible one still requires the direct constancy test.
(i) M = D q − 1 M = D_{q-1} M = D q − 1 : a dihedral subgroup D q − 1 D_{q-1} D q − 1 and a conjugate of D q + 1 D_{q+1} D q + 1 have trivial intersection in their rotation subgroups (gcd ( ( q − 1 ) / 2 , ( q + 1 ) / 2 ) = 1 \gcd((q - 1)/2,(q + 1)/2) = 1 g cd(( q − 1 ) /2 , ( q + 1 ) /2 ) = 1 for odd q q q ), so ∣ M ∩ H g ∣ ∈ { 1 , 2 } |M \cap H^g| \in\{1,2\} ∣ M ∩ H g ∣ ∈ { 1 , 2 } , and M M M has on P = G / H \mathcal{P} = G/H P = G / H exactly the two orbit lengths ( q − 1 ) / 2 (q - 1)/2 ( q − 1 ) /2 and q − 1 q - 1 q − 1 . Since G G G is block-transitive but not necessarily flag-transitive, the base block B 0 B_0 B 0 is a union of M M M -orbits containing p 0 p_0 p 0 , not in general a single M M M -orbit; write ∣ B 0 ∣ = k |B_0| = k ∣ B 0 ∣ = k .
k = ( q − 1 ) / 2 k = (q - 1)/2 k = ( q − 1 ) /2 : if B 0 B_0 B 0 were the single M M M -orbit of length ( q − 1 ) / 2 (q - 1)/2 ( q − 1 ) /2 , then r = b k / v = ( q + 1 ) / 2 r = bk/v = (q + 1)/2 r = bk / v = ( q + 1 ) /2 and λ = ( q − 3 ) / ( 2 ( q − 2 ) ) \lambda= (q - 3)/(2(q - 2)) λ = ( q − 3 ) / ( 2 ( q − 2 )) , satisfying 0 < λ < 1 / 2 0 < \lambda< 1/2 0 < λ < 1/2 for q ≥ 5 q \ge5 q ≥ 5 ; not a positive integer, excluded. More generally, every union of M M M -orbits has size ( q − 1 ) ( a / 2 + b ) (q - 1)(a/2 + b) ( q − 1 ) ( a /2 + b ) with a a a the number of orbits of length ( q − 1 ) / 2 (q - 1)/2 ( q − 1 ) /2 and b b b the number of orbits of length q − 1 q - 1 q − 1 , so ( q − 1 ) ( a / 2 + b ) = ( q − 1 ) / 2 (q - 1)(a/2 + b) = (q - 1)/2 ( q − 1 ) ( a /2 + b ) = ( q − 1 ) /2 forces a = 1 a = 1 a = 1 and b = 0 b = 0 b = 0 : no union other than this single orbit has that size.
The flag-transitive subcase (B 0 B_0 B 0 a single M M M -orbit, so k = ∣ M ∣ / ∣ M ∩ H g ∣ ∈ { ( q − 1 ) / 2 , q − 1 } k = |M|/|M \cap H^g| \in\{(q - 1)/2, q - 1\} k = ∣ M ∣/∣ M ∩ H g ∣ ∈ {( q − 1 ) /2 , q − 1 } ) is covered by the classification of point-primitive designs with dihedral point stabilizers of Chen, Wu and Xia [16 ] , Theorem 1.2: a locally transitive point-primitive design with dihedral point stabilizer is one of (a)–(f), and none has v = q ( q − 1 ) / 2 v = q(q - 1)/2 v = q ( q − 1 ) /2 or a block stabilizer of type D q − 1 D_{q-1} D q − 1 . Hence the single-orbit base block B 0 = O ( p 0 ) B_0 = O(p_0) B 0 = O ( p 0 ) of size q − 1 q - 1 q − 1 is excluded for every q q q .
In the general case k k k ranges over the sizes of the M M M -orbit unions containing p 0 p_0 p 0 . The integrality of r = b k / v = ( q + 1 ) k / ( q − 1 ) r = bk/v = (q + 1)k/(q - 1) r = bk / v = ( q + 1 ) k / ( q − 1 ) and of λ = r ( k − 1 ) / ( v − 1 ) = 2 k ( k − 1 ) / ( ( q − 1 ) ( q − 2 ) ) \lambda= r(k - 1)/(v - 1) = 2k(k - 1)/((q - 1)(q - 2)) λ = r ( k − 1 ) / ( v − 1 ) = 2 k ( k − 1 ) / (( q − 1 ) ( q − 2 )) is a purely arithmetic sieve over the two orbit lengths; for q ≤ 29 q \le29 q ≤ 29 it leaves, besides k = q − 1 k = q - 1 k = q − 1 and the complementary size v − ( q − 1 ) v - (q - 1) v − ( q − 1 ) , only the intermediate values k ∈ { 40 , 96 } k \in\{40, 96\} k ∈ { 40 , 96 } (q = 17 q = 17 q = 17 ) and k ∈ { 99 , 154 } k \in\{99, 154\} k ∈ { 99 , 154 } (q = 23 q = 23 q = 23 ). For every admissible k k k the design condition is that λ ( p 0 , x ) = # { g M : { p 0 , x } ⊆ g B 0 } \lambda(p_0, x) = \#\{gM : \{p_0, x\} \subseteq gB_0\} λ ( p 0 , x ) = # { g M : { p 0 , x } ⊆ g B 0 } equals λ \lambda λ for every point x x x , where g g g ranges over a right transversal of M M M in G G G and the blocks are the g B 0 gB_0 g B 0 with p 0 g − 1 ∈ B 0 p_0^{g^{-1}} \in B_0 p 0 g − 1 ∈ B 0 . An exhaustive check over all
such unions B 0 B_0 B 0 (script_gap_case2_lambda_correct.g, appended as electronic supplementary material) shows that for every prime power q ≤ 31 q \le31 q ≤ 31 no union B 0 B_0 B 0 has constant λ \lambda λ ; for example q = 11 q = 11 q = 11 , k = q − 1 = 10 k = q - 1 = 10 k = q − 1 = 10 gives λ ( p 0 , x ) ∈ { 1 , 2 , 4 } \lambda(p_0,x) \in\{1,2,4\} λ ( p 0 , x ) ∈ { 1 , 2 , 4 } while λ = 2 \lambda= 2 λ = 2 , and the intermediate sizes above are likewise excluded. In particular the flag-transitive block-size formula k = ∣ M ∣ / ∣ M ∩ H g ∣ k = |M|/|M \cap H^g| k = ∣ M ∣/∣ M ∩ H g ∣ is not used here, since the block need not be a single M M M -orbit.
For q ≥ 37 q \ge37 q ≥ 37 the arithmetic sieve reduces to a closed condition. Writing k = ( q − 1 ) t / 2 k = (q - 1)t/2 k = ( q − 1 ) t /2 with 1 ≤ t ≤ q − 1 1 \le t \le q - 1 1 ≤ t ≤ q − 1 , we have r = ( q + 1 ) t / 2 r = (q + 1)t/2 r = ( q + 1 ) t /2 and λ = t ( ( q − 1 ) t − 2 ) / ( 2 ( q − 2 ) ) \lambda= t((q - 1)t - 2)/(2(q - 2)) λ = t (( q − 1 ) t − 2 ) / ( 2 ( q − 2 )) ; since q − 1 ≡ 1 ( m o d q − 2 ) q - 1 \equiv1 \pmod{q - 2} q − 1 ≡ 1 ( mod q − 2 ) and q − 2 q - 2 q − 2 is odd, the integrality of λ \lambda λ is equivalent to q − 2 ∣ t ( t − 2 ) q - 2 \mid t(t - 2) q − 2 ∣ t ( t − 2 ) , so that t ≡ 0 t \equiv0 t ≡ 0 or 2 ( m o d ℓ e ) 2 \pmod{\ell^e} 2 ( mod ℓ e ) for every prime power ℓ e ∥ q − 2 \ell^e \parallel q - 2 ℓ e ∥ q − 2 , and t t t is paired with q − t q - t q − t by complementation. For q ≡ 3 ( m o d 4 ) q \equiv3 \pmod4 q ≡ 3 ( mod 4 ) the candidate design, if it existed, would be quasisymmetric with block intersection numbers ( t 2 ) \binom{t}{2} ( 2 t ) and ( t 2 ) + t ( q − t ) / ( 2 ( q − 2 ) ) \binom{t}{2} + t(q - t)/(2(q - 2)) ( 2 t ) + t ( q − t ) / ( 2 ( q − 2 )) and block graph T ‾ ( q + 1 ) \overline{T}(q + 1) T ( q + 1 ) , the complement of the triangular graph: indeed, with W W W the natural q q q -dimensional module of G = P S L 2 ( q ) G = PSL_2(q) G = P S L 2 ( q ) on L = P G ( 1 , q ) L = PG(1,q) L = P G ( 1 , q ) , one has 1 M G − 1 H G = χ W 1_M^G - 1_H^G = \chi_W 1 M G − 1 H G = χ W (verified class by class by the module computation of Appendix A), and H H H acts regularly on L L L for q ≡ 3 ( m o d 4 ) q \equiv3 \pmod4 q ≡ 3 ( mod 4 ) , so dim W H = 0 \dim W^H = 0 dim W H = 0 and ker N ≅ W \ker N \cong W ker N ≅ W is the unique copy in the block space; the intersection formula then follows from N T N = ( r − λ ) ( I − P W ) + c J N^T N = (r - \lambda)(I - P_W) + cJ N T N = ( r − λ ) ( I − P W ) + c J . By the classification of the quasisymmetric designs with block graph T ‾ ( m ) \overline{T}(m) T ( m ) of Coster and Haemers [18 ] (their Proposition 1.2 identifies the parameters as v = ( m − 1 2 ) v = \binom{m-1}{2} v = ( 2 m − 1 ) , k = a ( m − 2 ) / 2 k = a(m - 2)/2 k = a ( m − 2 ) /2 , μ = a ( m − a − 1 ) / ( 2 ( m − 3 ) ) \mu= a(m - a - 1)/(2(m - 3)) μ = a ( m − a − 1 ) / ( 2 ( m − 3 )) , which is exactly our m = q + 1 m = q + 1 m = q + 1 , a = t a = t a = t ), their Theorem 3.2 and Corollary 3.3 are stated for all m m m , but their conclusions are necessary conditions only. Corollary 3.3(v) forces m ≡ 4 ( m o d 16 ) m \equiv4 \pmod{16} m ≡ 4 ( mod 16 ) as soon as m ≡ 4 ( m o d 8 ) m \equiv4 \pmod8 m ≡ 4 ( mod 8 ) , so every q ≡ 11 ( m o d 16 ) q \equiv11 \pmod{16} q ≡ 11 ( mod 16 ) is excluded outright, with no table and no bound on q q q . Three of the four residue classes of q q q modulo 16 with q ≡ 3 ( m o d 4 ) q \equiv3 \pmod4 q ≡ 3 ( mod 4 ) are now settled without computation. Recall that τ ∗ \tau^* τ ∗ denotes the square-free part of τ \tau τ , so that τ = τ ∗ s 2 \tau= \tau^*s^2 τ = τ ∗ s 2 with τ ∗ \tau^* τ ∗ square free; thus τ ∗ \tau^* τ ∗ is even exactly when v 2 ( τ ) v_2(\tau) v 2 ( τ ) is odd.
Lemma 4.4. Let q = p f q = p^f q = p f be a prime power with q ≡ 3 ( m o d 16 ) q \equiv3 \pmod{16} q ≡ 3 ( mod 16 ) , put m = q + 1 m = q + 1 m = q + 1 , and let t t t satisfy 3 ≤ t ≤ q − 3 3 \le t \le q - 3 3 ≤ t ≤ q − 3 and q − 2 ∣ t ( t − 2 ) q - 2 \mid t(t - 2) q − 2 ∣ t ( t − 2 ) , so that τ = t ( q − t ) / ( 2 ( q − 2 ) ) \tau= t(q - t)/(2(q - 2)) τ = t ( q − t ) / ( 2 ( q − 2 )) is a positive integer. Assume p ∤ t p \nmid t p ∤ t .
Then the equation of [18 Corollary 3.3(v)] ,
2 τ X 2 + ( m − 1 ) Y 2 + m − 2 2 Z 2 = τ ( m − 1 ) ( m − 2 ) W 2 2\tau X^2+(m-1)Y^2+\frac{m-2}{2}Z^2=\tau(m-1)(m-2)W^2 2 τ X 2 + ( m − 1 ) Y 2 + 2 m − 2 Z 2 = τ ( m − 1 ) ( m − 2 ) W 2
has no solution in integers with X Y Z ≠ 0 XYZ\ne0 X Y Z = 0 .
Proof. Since q ≡ 3 ( m o d 4 ) q\equiv3\pmod{4} q ≡ 3 ( mod 4 ) is a prime power but not a square, f f f is odd; hence p ≡ 3 ( m o d 4 ) p\equiv3\pmod{4} p ≡ 3 ( mod 4 ) , so ( − 1 p ) = − 1 \left(\frac{-1}{p}\right)=-1 ( p − 1 ) = − 1 and p ≠ 2 p\ne2 p = 2 . Put a = 2 τ a=2\tau a = 2 τ , b = m − 1 = q b=m-1=q b = m − 1 = q and c = ( m − 2 ) / 2 c=(m-2)/2 c = ( m − 2 ) /2 . Then a b c = τ ( m − 1 ) ( m − 2 ) abc=\tau(m-1)(m-2) ab c = τ ( m − 1 ) ( m − 2 ) , so the equation reads a X 2 + b Y 2 + c Z 2 = a b c W 2 aX^2+bY^2+cZ^2=abcW^2 a X 2 + b Y 2 + c Z 2 = ab c W 2 .
We have p ∣ b p\mid b p ∣ b with v p ( b ) = f v_p(b)=f v p ( b ) = f odd, while p ∤ a p\nmid a p ∤ a and p ∤ c p\nmid c p ∤ c : indeed p ∤ 2 p\nmid2 p ∤ 2 and
τ ≡ t ( q − t ) 2 ( q − 2 ) ≡ t ⋅ ( − t ) − 4 = t 2 4 ( m o d p ) \tau\equiv\frac{t(q-t)}{2(q-2)}\equiv\frac{t\cdot(-t)}{-4}=\frac{t^2}{4}\pmod p τ ≡ 2 ( q − 2 ) t ( q − t ) ≡ − 4 t ⋅ ( − t ) = 4 t 2 ( mod p )
is meaningful because 2 ( q − 2 ) ≡ − 4 ≠ 0 ( m o d p ) 2(q-2)\equiv-4\ne0\pmod p 2 ( q − 2 ) ≡ − 4 = 0 ( mod p ) , and p ∤ τ p\nmid\tau p ∤ τ because p ∤ t p\nmid t p ∤ t ; and m − 2 = q − 1 ≡ − 1 ( m o d p ) m-2=q-1\equiv-1\pmod p m − 2 = q − 1 ≡ − 1 ( mod p ) and m − 2 ≡ 2 ( m o d 16 ) m-2\equiv2\pmod{16} m − 2 ≡ 2 ( mod 16 ) , so p ∤ c p\nmid c p ∤ c .
Reducing each coefficient by its square part, which amounts to rescaling the corresponding variable, and then cancelling a prime common to two coefficients by the equivalence a X 2 + b Y 2 = c Z 2 ⟺ ( a / p ) X 2 + ( b / p ) Y 2 = ( c p ) Z 2 aX^2+bY^2=cZ^2\Longleftrightarrow(a/p)X^2+(b/p)Y^2=(cp)Z^2 a X 2 + b Y 2 = c Z 2 ⟺ ( a / p ) X 2 + ( b / p ) Y 2 = ( c p ) Z 2 , we obtain a square-free pairwise coprime triple ( a ′ , b ′ , c ′ ) (a',b',c') ( a ′ , b ′ , c ′ ) . Since p p p divides only b b b , and with odd exponent f f f , the exponent of p p p in b b b remains odd at every step, so p p p is never common to two of the three coefficients; hence p ∣ b ′ p\mid b' p ∣ b ′ and a ′ c ′ ≡ a c = τ ( m − 2 ) ( m o d p ) a'c'\equiv ac=\tau(m-2)\pmod p a ′ c ′ ≡ a c = τ ( m − 2 ) ( mod p ) .
By [18 Lemma 2.3] a necessary condition for a solution is ( a ′ c ′ p ) = 1 \left(\frac{a'c'}{p}\right)=1 ( p a ′ c ′ ) = 1 whenever p ∣ b ′ p\mid b' p ∣ b ′ . But
( a ′ c ′ p ) = ( τ p ) ( m − 2 p ) = 1 ⋅ ( − 1 ) = − 1 , \left(\frac{a'c'}{p}\right)=\left(\frac{\tau}{p}\right)\left(\frac{m-2}{p}\right)=1\cdot(-1)=-1, ( p a ′ c ′ ) = ( p τ ) ( p m − 2 ) = 1 ⋅ ( − 1 ) = − 1 ,
because τ ≡ t 2 / 4 \tau\equiv t^2/4 τ ≡ t 2 /4 is a quotient of two squares modulo p p p , and m − 2 ≡ − 1 m-2\equiv-1 m − 2 ≡ − 1 with p ≡ 3 ( m o d 4 ) p\equiv3\pmod4 p ≡ 3 ( mod 4 ) . This contradiction proves the assertion.
Thus every pair with q ≡ 3 ( m o d 16 ) q\equiv3\pmod{16} q ≡ 3 ( mod 16 ) and p ∤ t p\nmid t p ∤ t is excluded. The complementary case p ∣ t p\mid t p ∣ t forces t ≥ p t\ge p t ≥ p , which together with $3\le t ≤ q − 3 t \leq q - 3 t ≤ q − 3 can occur only for f ≥ 2 f \geq2 f ≥ 2 ; in the range q ≤ 3000 q \leq3000 q ≤ 3000 it occurs only for q = 1331 = 11 3 q = 1331 = 11^3 q = 1331 = 1 1 3 , and there the remaining Legendre conditions already fail.
If q ≡ 7 ( m o d 16 ) q \equiv7 \pmod{16} q ≡ 7 ( mod 16 ) then m ≡ 8 ( m o d 16 ) m \equiv8 \pmod{16} m ≡ 8 ( mod 16 ) and branch (i) of [18 Corollary 3.3] requires τ ∗ \tau^* τ ∗ to be even; hence every t t t with τ ∗ \tau^* τ ∗ odd is excluded.
It remains to consider q ≡ 15 ( m o d 16 ) q \equiv15 \pmod{16} q ≡ 15 ( mod 16 ) , so m ≡ 0 ( m o d 16 ) m \equiv0 \pmod{16} m ≡ 0 ( mod 16 ) and p ≡ 15 ( m o d 16 ) p \equiv15 \pmod{16} p ≡ 15 ( mod 16 ) . Here [18 Corollary 3.3(i)] reads τ X 2 + ( m − 2 ) Y 2 = 2 ( m − 1 ) Z 2 \tau X^2 + (m - 2)Y^2 = 2(m - 1)Z^2 τ X 2 + ( m − 2 ) Y 2 = 2 ( m − 1 ) Z 2 , that is a X 2 + b Y 2 = c Z 2 aX^2 + bY^2 = cZ^2 a X 2 + b Y 2 = c Z 2 with ( a , b , c ) = ( τ , m − 2 , 2 ( m − 1 ) ) (a,b,c) = (\tau,m-2,2(m-1)) ( a , b , c ) = ( τ , m − 2 , 2 ( m − 1 )) and a b c = 2 τ ( m − 1 ) ( m − 2 ) abc = 2\tau(m-1)(m-2) ab c = 2 τ ( m − 1 ) ( m − 2 ) . Since p ∣ c = 2 q p \mid c = 2q p ∣ c = 2 q while p ∤ a p \nmid a p ∤ a and p ∤ b p \nmid b p ∤ b , the criterion attached to p p p is
( − a b p ) = ( − 1 p ) ( τ p ) ( m − 2 p ) = ( − 1 ) ⋅ 1 ⋅ ( − 1 ) = + 1 , \left(\frac{-ab}{p}\right) = \left(\frac{-1}{p}\right)\left(\frac{\tau}{p}\right)\left(\frac{m-2}{p}\right) = (-1)\cdot1\cdot(-1) = +1, ( p − ab ) = ( p − 1 ) ( p τ ) ( p m − 2 ) = ( − 1 ) ⋅ 1 ⋅ ( − 1 ) = + 1 ,
so this condition is satisfied identically: the arithmetic route through [18 ] terminates at this residue class, and no prime divisor of q q q excludes it. The residue class q ≡ 15 ( m o d 16 ) q \equiv15 \pmod{16} q ≡ 15 ( mod 16 ) is nevertheless empty. The reason is geometric, it applies to every admissible t t t at once and requires no bound on q q q , and it is recorded in Lemma 4.5 below, after the enumeration of the remaining block stabilizers.
For q = 47 q = 47 q = 47 all M M M -orbit unions of size 460 were enumerated by computer: the 42,504 unions of twenty short orbits (script gap_case2_q47_b0.g), the 1,480,556 unions of eighteen short orbits and one long orbit (script q47_b1_exact.py), the 40,450,905 unions of sixteen short orbits and two long orbits (script q47_b2_exact.py), and the 323,607,240 unions of fourteen short orbits and three long orbits (script q47_b3_exact.py, 165 triples of long orbits); none has constant λ \lambda λ . The remaining mixed unions with at least four long orbits are subject to the system (17) ; its 0–1 linearization is infeasible (script q47_milp_check.py, solved with HiGHS), so these unions do not give constant λ \lambda λ either, and the exclusion of q = 47 q = 47 q = 47 is complete.
For q = 71 q = 71 q = 71 the arithmetic sieve leaves t ∈ { 2 , 23 , 48 , 69 } t \in\{2, 23, 48, 69\} t ∈ { 2 , 23 , 48 , 69 } , hence the only intermediate parameter is t = 23 t = 23 t = 23 , i.e. k = 805 k = 805 k = 805 , r = 828 r = 828 r = 828 , λ = 268 \lambda= 268 λ = 268 (with t = 48 t = 48 t = 48 its complement). Since q ≡ 3 ( m o d 4 ) q \equiv3 \pmod{4} q ≡ 3 ( mod 4 ) , such a design would be quasisymmetric with intersection numbers x = ( 23 2 ) = 253 x = \binom{23}{2} = 253 x = ( 2 23 ) = 253 and y = 261 y = 261 y = 261 , and block graph K G ( 72 , 2 ) KG(72,2) K G ( 72 , 2 ) or its
complement. On the 71-dimensional eigenspace of K G ( 72 , 2 ) KG(72,2) K G ( 72 , 2 ) with eigenvalue 68 68 68 the Gram matrix N T N = k I + x ( J − I ) + ( y − x ) A N^{T}N = kI + x(J-I) + (y-x)A N T N = k I + x ( J − I ) + ( y − x ) A would have eigenvalue k − x + ( y − x ) ⋅ 68 = 1096 k-x+(y-x)\cdot68 = 1096 k − x + ( y − x ) ⋅ 68 = 1096 , whereas the design equations force N T N N^{T}N N T N to have eigenvalues r − λ = 560 r-\lambda= 560 r − λ = 560 (multiplicity 71 71 71 ) and 0 0 0 (multiplicity 2484 2484 2484 ) on the non-trivial spaces; on the eigenvalue-( − 2 ) (-2) ( − 2 ) space it gives 536 ≠ 0 536 \ne0 536 = 0 . For the complement the corresponding eigenvalues are 0 0 0 and 560 560 560 , again violating these constraints. Hence q = 71 q=71 q = 71 is excluded by the block-graph spectral condition (script q71_spectral_check.py).
In the range $37 \le q \le 125$ there are fourteen prime powers q ≡ 1 ( m o d 4 ) q \equiv1 \pmod{4} q ≡ 1 ( mod 4 ) , namely 37 , 41 , 49 , 53 , 61 , 73 , 81 , 89 , 97 , 101 , 109 , 113 , 121 , 125 37,41,49,53,61,73,81,89,97,101,109,113,121,125 37 , 41 , 49 , 53 , 61 , 73 , 81 , 89 , 97 , 101 , 109 , 113 , 121 , 125 (with 49 = 7 2 49=7^2 49 = 7 2 , 81 = 3 4 81=3^4 81 = 3 4 , 121 = 11 2 121=11^2 121 = 1 1 2 ). For q = 49 , 61 , 73 , 81 , 109 q=49,61,73,81,109 q = 49 , 61 , 73 , 81 , 109 the condition q − 2 ∣ t ( t − 2 ) q-2 \mid t(t-2) q − 2 ∣ t ( t − 2 ) has no solution with 3 ≤ t ≤ q − 3 3 \le t \le q-3 3 ≤ t ≤ q − 3 , so these five values are already excluded at the integrality stage; for the remaining nine values q = 37 , 41 , 53 , 89 , 97 , 101 , 113 , 121 , 125 q=37,41,53,89,97,101,113,121,125 q = 37 , 41 , 53 , 89 , 97 , 101 , 113 , 121 , 125 one has dim W H = 1 \dim W^{H}=1 dim W H = 1 , so the copy of W W W in the block space is not unique and the intersection formula used for q ≡ 3 ( m o d 4 ) q \equiv3 \pmod{4} q ≡ 3 ( mod 4 ) does not automatically hold. For q = 37 q=37 q = 37 the candidate parameter is t = 7 t=7 t = 7 , i.e. a 2 2 2 -( 666 , 126 , 25 ) (666,126,25) ( 666 , 126 , 25 ) design, and the full enumeration of the 191,520 unions of M M M -orbits of size 126 was carried out against the system (17) (script q37_enum.py); none satisfies it. For q = 41 q=41 q = 41 the candidate parameter is t = 26 t=26 t = 26 , i.e. a 2 2 2 -( 820 , 520 , 346 ) (820,520,346) ( 820 , 520 , 346 ) design, and the same enumeration of the 83,061,552 unions of M M M -orbits of size 520 was carried out (script q41_enum.py); none satisfies it. For q = 53 q=53 q = 53 the candidate parameter is t = 17 t=17 t = 17 , i.e. a 2 2 2 -( 1378 , 442 , 147 ) (1378,442,147) ( 1378 , 442 , 147 ) design, and the enumeration of the 10,165,373,049 unions of M M M -orbits of size 442 was carried out (script q53_enum.py); none satisfies it. The remaining six cases q = 89 , 97 , 101 , 113 , 121 , 125 q=89,97,101,113,121,125 q = 89 , 97 , 101 , 113 , 121 , 125 are decided by the same system (17) . The integrality prerequisite μ j λ ≡ 0 ( m o d b ) \mu_j\lambda\equiv0 \pmod{b} μ j λ ≡ 0 ( mod b ) holds for every j j j in each of the six cases, so no candidate is removed at that stage, and the 0–1 linearization of (17) (with y o o ′ = x o x o ′ y_{oo'}=x_ox_{o'} y o o ′ = x o x o ′ , 0 ≤ y o o ′ ≤ ℓ o ℓ o ′ 0 \le y_{oo'} \le\ell_o\ell_{o'} 0 ≤ y o o ′ ≤ ℓ o ℓ o ′ and the three constraints y o o ′ ≤ ℓ o ℓ o ′ x o y_{oo'} \le\ell_o\ell_{o'}x_o y o o ′ ≤ ℓ o ℓ o ′ x o , y o o ′ ≤ ℓ o ℓ o ′ x o ′ y_{oo'} \le\ell_o\ell_{o'}x_{o'} y o o ′ ≤ ℓ o ℓ o ′ x o ′ , y o o ′ ≥ ℓ o ℓ o ′ ( x o + x o ′ − 1 ) y_{oo'} \ge\ell_o\ell_{o'}(x_o+x_{o'}-1) y o o ′ ≥ ℓ o ℓ o ′ ( x o + x o ′ − 1 ) ) is infeasible in each case. The resulting programs carry only 67–94 binary variables and 2211–4465 continuous pairing variables, and HiGHS certifies infeasibility in at most 17 seconds in every case ($q=89:
3.9 s 3.9\ \mathrm{s} 3.9 s ; q = 97 q = 97 q = 97 : 5.6 s 5.6\ \mathrm{s} 5.6 s ; q = 101 q = 101 q = 101 : 6.6 s 6.6\ \mathrm{s} 6.6 s ; q = 113 q = 113 q = 113 : 10.6 s 10.6\ \mathrm{s} 10.6 s ; q = 121 q = 121 q = 121 : 14.7 s 14.7\ \mathrm{s} 14.7 s ; q = 125 q = 125 q = 125 : 16.9 s 16.9\ \mathrm{s} 16.9 s ; script case2_milp_compact.py). Hence none of the unions of M M M -orbits of size k k k has constant λ \lambda λ , and the exclusion of these six values is complete.
(ii) M ∈ { A 5 , S 4 , A 4 } M \in\{A_5, S_4, A_4\} M ∈ { A 5 , S 4 , A 4 } : the realizable block sizes are as in Case 1 (the intersection orders are constrained by the element orders of D q + 1 D_{q+1} D q + 1 and of the exceptional group), and λ = 2 k ( k − 1 ) / ( ∣ M ∣ ( q − 2 ) ) \lambda= 2k(k - 1)/(|M|(q - 2)) λ = 2 k ( k − 1 ) / ( ∣ M ∣ ( q − 2 )) must be an integer.
M = A 5 M = A_5 M = A 5 , k ∈ { 12 , 20 , 30 , 60 } k \in\{12, 20, 30, 60\} k ∈ { 12 , 20 , 30 , 60 } : the divisibility q − 2 ∣ k ( k − 1 ) / 30 q - 2 \mid k(k - 1)/30 q − 2 ∣ k ( k − 1 ) /30 gives q − 2 ∣ 29 q - 2 \mid29 q − 2 ∣ 29 (k = 30 k = 30 k = 30 ) and q − 2 ∣ 118 q - 2 \mid118 q − 2 ∣ 118 (k = 60 k = 60 k = 60 ), i.e. q = 31 q = 31 q = 31 with λ = 1 \lambda= 1 λ = 1 and q = 61 q = 61 q = 61 with λ = 2 \lambda= 2 λ = 2 ; k = 20 , 12 k = 20, 12 k = 20 , 12 give non-integral quotients. Both survivors are excluded by the constancy test (computed in gap_finalcheck4.g: λ ( p 0 , x ) \lambda(p_0, x) λ ( p 0 , x ) is not constant on the H H H -suborbits).
M = S 4 M = S_4 M = S 4 , k ∈ { 4 , 6 , 8 , 12 , 24 } k \in\{4, 6, 8, 12, 24\} k ∈ { 4 , 6 , 8 , 12 , 24 } : the divisibility q − 2 ∣ k ( k − 1 ) / 12 q - 2 \mid k(k - 1)/12 q − 2 ∣ k ( k − 1 ) /12 gives q − 2 ∣ 46 q - 2 \mid46 q − 2 ∣ 46 (k = 24 k = 24 k = 24 ), i.e. q = 25 q = 25 q = 25 with λ = 2 \lambda= 2 λ = 2 (the value q − 2 ∣ 11 q - 2 \mid11 q − 2 ∣ 11 from k = 12 k = 12 k = 12 gives q = 13 q = 13 q = 13 , violating q 2 ≡ 1 ( m o d 16 ) q^2 \equiv1\pmod{16} q 2 ≡ 1 ( mod 16 ) , so S 4 ≰ P S L 2 ( 13 ) S_4 \not\le PSL_2(13) S 4 ≤ P S L 2 ( 13 ) ). The survivor q = 25 q = 25 q = 25 is excluded by the constancy test (computed in gap_finalcheck4.g).
M = A 4 M = A_4 M = A 4 , k ∈ { 4 , 6 , 12 } k \in\{4, 6, 12\} k ∈ { 4 , 6 , 12 } : since λ = 2 k ( k − 1 ) / ( 12 ( q − 2 ) ) \lambda= 2k(k - 1)/(12(q - 2)) λ = 2 k ( k − 1 ) / ( 12 ( q − 2 )) , the divisibility 12 ( q − 2 ) ∣ 2 k ( k − 1 ) 12(q - 2) \mid2k(k - 1) 12 ( q − 2 ) ∣ 2 k ( k − 1 ) is required. For k = 12 k = 12 k = 12 it gives q − 2 ∣ 22 q - 2 \mid22 q − 2 ∣ 22 , i.e. q ∈ { 3 , 4 , 13 , 24 } q \in\{3, 4, 13, 24\} q ∈ { 3 , 4 , 13 , 24 } ; q = 13 q = 13 q = 13 yields λ = 2 \lambda= 2 λ = 2 and r = 14 r = 14 r = 14 (both integral, with b = 91 > v = 78 b = 91 > v = 78 b = 91 > v = 78 ), and is excluded by the constancy test (computed in gap_finalcheck4.g), while q = 24 q = 24 q = 24 is not a prime power and q = 3 , 4 q = 3, 4 q = 3 , 4 are degenerate. For k = 6 k = 6 k = 6 it gives q − 2 ∣ 5 q - 2 \mid5 q − 2 ∣ 5 , i.e. q ∈ { 3 , 7 } q \in\{3, 7\} q ∈ { 3 , 7 } , and A 4 A_4 A 4 is not maximal in P S L 2 ( 7 ) PSL_2(7) P S L 2 ( 7 ) . For k = 4 k = 4 k = 4 it gives q − 2 ∣ 2 q - 2 \mid2 q − 2 ∣ 2 , i.e. q ∈ { 3 , 4 } q \in\{3, 4\} q ∈ { 3 , 4 } , both degenerate. Hence no candidate survives.
Lemma 4.5. Let q = p f q = p^f q = p f be a prime power with q ≡ 15 ( m o d 16 ) q \equiv15 \pmod{16} q ≡ 15 ( mod 16 ) , and let t t t satisfy 3 ≤ t ≤ q − 3 3 \le t \le q - 3 3 ≤ t ≤ q − 3 and q − 2 ∣ t ( t − 2 ) q - 2 \mid t(t - 2) q − 2 ∣ t ( t − 2 ) , so that λ = t ( ( q − 1 ) t − 2 ) / ( 2 ( q − 2 ) ) \lambda= t((q - 1)t - 2)/(2(q - 2)) λ = t (( q − 1 ) t − 2 ) / ( 2 ( q − 2 )) is a positive integer. Then in Case 2(i) no union of M M M -orbits has constant λ \lambda λ ; equivalently, no design arises with point stabilizer H = D q + 1 H = D_{q+1} H = D q + 1 and block stabilizer M = D q − 1 M = D_{q-1} M = D q − 1 .
Proof. Let I I I be the set of internal points of the conic C : X 0 X 2 − X 1 2 = 0 C : X_0X_2 - X_1^2 = 0 C : X 0 X 2 − X 1 2 = 0 of P G ( 2 , q ) PG(2,q) P G ( 2 , q ) . By [34 Section 2.2.1] , T = P S L 2 ( q ) T = PSL_2(q) T = P S L 2 ( q ) acts transitively on I I I with ∣ I ∣ = q ( q − 1 ) / 2 = v |I| = q(q-1)/2 = v ∣ I ∣ = q ( q − 1 ) /2 = v and point stabilizer D q + 1 D_{q+1} D q + 1 ; since the subgroups D q + 1 D_{q+1} D q + 1 form a single conjugacy class of T T T , the T T T -sets P = T / H P = T/H P = T / H and I I I are isomorphic, and orbit lengths are preserved under this isomorphism.
Put M = G B M = G_B M = G B . The block B B B is M M M -invariant, hence a union of M M M -orbits on I I I . Let Ω \Omega Ω be an M M M -orbit of maximal length ℓ \ell ℓ . Since T B = T ∩ G B ≤ M T_B = T \cap G_B \leq M T B = T ∩ G B ≤ M , every M M M -orbit is a union of T B T_B T B -orbits, so ℓ ≥ q − 1 \ell\geq q - 1 ℓ ≥ q − 1 : for q ≡ − 1 ( m o d 4 ) q \equiv-1 \pmod{4} q ≡ − 1 ( mod 4 ) the T B T_B T B -orbits of maximal length are the ( q − 1 ) (q-1) ( q − 1 ) -arcs C h ∗ C_h^* C h ∗ , and by [34 Proposition 2.7(i) and Corollary 2.8(i)] there are exactly ( q − 3 ) / 4 (q - 3)/4 ( q − 3 ) /4 of them. Let Ω 1 , … , Ω N \Omega_1,\ldots,\Omega_N Ω 1 , … , Ω N be the M M M -orbits of length ℓ \ell ℓ . Then
1 ≤ N ≤ ∣ I ∣ ℓ ≤ q ( q − 1 ) / 2 q − 1 = q 2 < q . 1 \leq N \leq\frac{|I|}{\ell} \leq\frac{q(q-1)/2}{q-1} = \frac{q}{2} < q. 1 ≤ N ≤ ℓ ∣ I ∣ ≤ q − 1 q ( q − 1 ) /2 = 2 q < q .
As G G G is transitive on blocks, every block contains the same number s s s of orbits among Ω 1 , … , Ω N \Omega_1,\ldots,\Omega_N Ω 1 , … , Ω N , so b s b\,s b s counts the pairs ( B , Ω i ) (B,\Omega_i) ( B , Ω i ) with Ω i ⊆ B \Omega_i \subseteq B Ω i ⊆ B . For fixed i i i the orbit Ω i \Omega_i Ω i is M M M -invariant, so a block meets Ω i \Omega_i Ω i if and only if it contains Ω i \Omega_i Ω i ; hence the blocks containing Ω i \Omega_i Ω i are exactly the r r r blocks through a fixed x ∈ Ω i x \in\Omega_i x ∈ Ω i . The parameter λ \lambda λ counts blocks through a prescribed pair, not through a single point, so the count gives
b s = N r . (18) b\,s = N\,r. \tag*{(18)} b s = N r . (18)
Now b = q ( q + 1 ) / 2 b = q(q + 1)/2 b = q ( q + 1 ) /2 . Indeed, the parameter t t t indexes the block size k = t ( q − 1 ) / 2 k = t(q - 1)/2 k = t ( q − 1 ) /2 as in item (i) above, and with v = q ( q − 1 ) / 2 v = q(q - 1)/2 v = q ( q − 1 ) /2 the design equation b k ( k − 1 ) = λ v ( v − 1 ) b\,k(k - 1) = \lambda\,v(v - 1) b k ( k − 1 ) = λ v ( v − 1 ) together with the displayed value of λ \lambda λ gives b = λ v ( v − 1 ) / ( k ( k − 1 ) ) = q ( q + 1 ) / 2 b = \lambda v(v - 1)/(k(k - 1)) = q(q + 1)/2 b = λ v ( v − 1 ) / ( k ( k − 1 )) = q ( q + 1 ) /2 ; the same equation gives r = k b / v = t ( q + 1 ) / 2 r = kb/v = t(q + 1)/2 r = k b / v = t ( q + 1 ) /2 . Since q ∣ b q \mid b q ∣ b and 0 < N < q 0 < N < q 0 < N < q , the identity b s = N r b\,s = N\,r b s = N r forces q ∣ r q \mid r q ∣ r , and gcd ( q , ( q + 1 ) / 2 ) = 1 \gcd(q,(q + 1)/2) = 1 g cd( q , ( q + 1 ) /2 ) = 1 then forces q ∣ t q \mid t q ∣ t . This contradicts 3 ≤ t ≤ q − 3 3 \leq t \leq q - 3 3 ≤ t ≤ q − 3 , so no union of M M M -orbits has constant λ \lambda λ .
The four values t = 1 , 2 , q − 2 , q − 1 t = 1, 2, q - 2, q - 1 t = 1 , 2 , q − 2 , q − 1 left out of the statement are covered by items 1 and 2 above and by complementation, so every t t t with 1 ≤ t ≤ q − 1 1 \leq t \leq q - 1 1 ≤ t ≤ q − 1 is excluded. The same argument applies to any G G G with socle P S L 2 ( q ) PSL_2(q) P S L 2 ( q ) : beyond the existence of an orbit of length q − 1 q - 1 q − 1 , only the inequality ℓ ≥ q − 1 \ell\geq q - 1 ℓ ≥ q − 1 and the bound N < q N < q N < q are used.
Thus no non-symmetric block-primitive 2-design arises with point stabilizer H = D q + 1 H = D_{q+1} H = D q + 1 .
Case 3: point stabilizer of exceptional type
Let H ∈ { A 4 , S 4 , A 5 } H \in\{A_4,S_4,A_5\} H ∈ { A 4 , S 4 , A 5 } , so v = q ( q 2 − 1 ) / ( 2 ∣ H ∣ ) v=q(q^2-1)/(2|H|) v = q ( q 2 − 1 ) / ( 2∣ H ∣ ) . For any admissible block stabilizer M M M with ∣ M ∣ < ∣ H ∣ |M|<|H| ∣ M ∣ < ∣ H ∣ , the block size satisfies k ≤ ∣ M ∣ k\leq|M| k ≤ ∣ M ∣ , and λ ≥ 1 \lambda\geq1 λ ≥ 1 (the design is non-trivial) forces k ( k − 1 ) ≥ ( v − 1 ) ∣ M ∣ / ∣ H ∣ k(k-1)\geq(v-1)|M|/|H| k ( k − 1 ) ≥ ( v − 1 ) ∣ M ∣/∣ H ∣ , hence
( v − 1 ) ≤ ∣ H ∣ ( ∣ M ∣ − 1 ) < ∣ H ∣ 2 . (v-1)\leq|H|(|M|-1)<|H|^2. ( v − 1 ) ≤ ∣ H ∣ ( ∣ M ∣ − 1 ) < ∣ H ∣ 2 .
With v = q ( q 2 − 1 ) / ( 2 ∣ H ∣ ) v=q(q^2-1)/(2|H|) v = q ( q 2 − 1 ) / ( 2∣ H ∣ ) this yields q ( q 2 − 1 ) ≤ 2 ∣ H ∣ 2 ( ∣ M ∣ − 1 ) q(q^2-1)\leq2|H|^2(|M|-1) q ( q 2 − 1 ) ≤ 2∣ H ∣ 2 ( ∣ M ∣ − 1 ) , and since ∣ M ∣ < ∣ H ∣ |M|<|H| ∣ M ∣ < ∣ H ∣ we get q 3 < 2 ∣ H ∣ 3 q^3<2|H|^3 q 3 < 2∣ H ∣ 3 . Hence q ≤ 75 q\leq75 q ≤ 75 for H = A 5 H=A_5 H = A 5 , q ≤ 30 q\leq30 q ≤ 30 for H = S 4 H=S_4 H = S 4 , and q ≤ 15 q\leq15 q ≤ 15 for H = A 4 H=A_4 H = A 4 . Each prime power q q q below these bounds lies in the range q ≤ 127 q\leq127 q ≤ 127 covered by the finite check of Appendix § 6.2 , which finds no surviving configuration: every pair ( H , M ) (H,M) ( H , M ) of maximal subgroup classes with ∣ M ∣ < ∣ H ∣ |M|<|H| ∣ M ∣ < ∣ H ∣ fails Fisher’s inequality, the integrality of r r r or λ \lambda λ , or the constancy of λ \lambda λ . Hence no design arises with exceptional point stabilizer.
Case 4: subfield point stabilizer
Let H = P S L 2 ( q 0 ) H=PSL_2(q_0) H = P S L 2 ( q 0 ) or P G L 2 ( q 0 ) PGL_2(q_0) P G L 2 ( q 0 ) with q = q 0 f q=q_0^f q = q 0 f , f ≥ 2 f\geq2 f ≥ 2 prime, so that v = q ( q 2 − 1 ) / ( 2 ∣ H ∣ ) v=q(q^2-1)/(2|H|) v = q ( q 2 − 1 ) / ( 2∣ H ∣ ) .
(i) f ≥ 3 f\geq3 f ≥ 3 : a maximal subgroup M M M of G G G is either a subfield subgroup, an exceptional subgroup, or one of the subgroups of order exceeding ∣ H ∣ |H| ∣ H ∣ . The subfield subgroups satisfy M ≤ H M\leq H M ≤ H . Since M M M is maximal in G G G and H < G H<G H < G , the inclusion M ≤ H M\leq H M ≤ H forces M = H M=H M = H , contradicting ∣ M ∣ < ∣ H ∣ |M|<|H| ∣ M ∣ < ∣ H ∣ ; hence this case cannot occur. The subgroups of type D q ± 1 D_{q\pm1} D q ± 1 and the Borel subgroup have order q ± 1 q\pm1 q ± 1 , respectively q ( q − 1 ) / 2 q(q-1)/2 q ( q − 1 ) /2 , exceeding ∣ H ∣ = q 0 ( q 0 2 − 1 ) |H|=q_0(q_0^2-1) ∣ H ∣ = q 0 ( q 0 2 − 1 ) for f ≥ 3 f\geq3 f ≥ 3 ; excluded by Fisher’s inequality. For M M M exceptional (∣ M ∣ ≤ 60 |M|\leq60 ∣ M ∣ ≤ 60 ), λ ≥ 1 \lambda\geq1 λ ≥ 1 forces ( v − 1 ) ≤ ∣ H ∣ ( ∣ M ∣ − 1 ) ≤ 59 q 0 ( q 0 2 − 1 ) (v-1)\leq|H|(|M|-1)\leq59q_0(q_0^2-1) ( v − 1 ) ≤ ∣ H ∣ ( ∣ M ∣ − 1 ) ≤ 59 q 0 ( q 0 2 − 1 ) , and since v = q 0 3 f − 6 ( q 0 2 − 1 ) / ( 2 ( q 0 2 − 1 ) ) v=q_0^{3f-6}(q_0^2-1)/(2(q_0^2-1)) v = q 0 3 f − 6 ( q 0 2 − 1 ) / ( 2 ( q 0 2 − 1 )) this gives q 0 3 f − 6 ≤ 118 q_0^{3f-6}\leq118 q 0 3 f − 6 ≤ 118 ; at f = 3 f=3 f = 3 it reads q 0 ( q 0 2 + q 0 + 1 ) ≤ 118 ( q 0 2 − 1 ) q_0(q_0^2+q_0+1)\leq118(q_0^2-1) q 0 ( q 0 2 + q 0 + 1 ) ≤ 118 ( q 0 2 − 1 ) , which leaves q 0 ∈ { 2 , 3 , 4 } q_0\in\{2,3,4\} q 0 ∈ { 2 , 3 , 4 } , while f ≥ 5 f\geq5 f ≥ 5 forces q 0 = 1 q_0=1 q 0 = 1 . Thus q ∈ { 8 , 27 , 64 } q\in\{8,27,64\} q ∈ { 8 , 27 , 64 } , all within the range of the exhaustive enumeration, which finds no survivor. Hence f ≥ 3 f\geq3 f ≥ 3 is excluded.
(ii) f = 2 f=2 f = 2 , q = q 0 2 q=q_0^2 q = q 0 2 : write H = P G L 2 ( q 0 ) H=PGL_2(q_0) H = P G L 2 ( q 0 ) , v = q 0 ( q 0 2 + 1 ) / 2 v=q_0(q_0^2+1)/2 v = q 0 ( q 0 2 + 1 ) /2 (the P S L 2 ( q 0 ) PSL_2(q_0) P S L 2 ( q 0 ) case is smaller and handled identically).
M = D q 0 ± 1 M=D_{q_0^{\pm1}} M = D q 0 ± 1 : here ∣ M ∣ = q 0 2 ± 1 < ∣ H ∣ |M|=q_0^2\pm1<|H| ∣ M ∣ = q 0 2 ± 1 < ∣ H ∣ and b = q 0 2 ( q 0 2 ± 1 ) / 2 > v b=q_0^2(q_0^2\pm1)/2>v b = q 0 2 ( q 0 2 ± 1 ) /2 > v , while k = ∣ M ∣ / ∣ M ∩ H g ∣ k=|M|/|M\cap H^g| k = ∣ M ∣/∣ M ∩ H g ∣ and r = b k / v r=bk/v r = bk / v . For M = D q 0 2 + 1 M=D_{q_0^2+1} M = D q 0 2 + 1 one has gcd ( q 0 2 + 1 , q 0 ( q 0 2 − 1 ) ) = 2 \gcd(q_0^2+1,q_0(q_0^2-1))=2 g cd( q 0 2 + 1 , q 0 ( q 0 2 − 1 )) = 2 , so ∣ M ∩ H g ∣ ∈ { 1 , 2 } |M\cap H^g|\in\{1,2\} ∣ M ∩ H g ∣ ∈ { 1 , 2 } and $k\in
{ q 0 2 + 1 , ( q 0 2 + 1 ) / 2 } \{q_0^2+1,(q_0^2+1)/2\} { q 0 2 + 1 , ( q 0 2 + 1 ) /2 } . If k = q 0 2 + 1 k=q_0^2+1 k = q 0 2 + 1 then r = q 0 k r=q_0k r = q 0 k and λ = 2 q 0 k ( k − 1 ) / ( q 0 3 + q 0 − 2 ) = 2 q 0 2 + 4 q 0 2 / ( q 0 3 + q 0 − 2 ) \lambda=2q_0k(k-1)/(q_0^3+q_0-2)=2q_0^2+4q_0^2/(q_0^3+q_0-2) λ = 2 q 0 k ( k − 1 ) / ( q 0 3 + q 0 − 2 ) = 2 q 0 2 + 4 q 0 2 / ( q 0 3 + q 0 − 2 ) , where 0 < 4 q 0 2 < q 0 3 + q 0 − 2 0<4q_0^2<q_0^3+q_0-2 0 < 4 q 0 2 < q 0 3 + q 0 − 2 for q 0 ≥ 4 q_0\geq4 q 0 ≥ 4 ; for q 0 = 3 q_0=3 q 0 = 3 the remainder is 8 and for q 0 = 2 q_0=2 q 0 = 2 one gets k = v k=v k = v , which is trivial. If k = ( q 0 2 + 1 ) / 2 k=(q_0^2+1)/2 k = ( q 0 2 + 1 ) /2 then r = q 0 k r=q_0k r = q 0 k and λ = q 0 ( q 0 4 − 1 ) / ( 2 ( q 0 3 + q 0 − 2 ) ) = q 0 2 − q 0 ( q 0 − 1 ) 2 / ( q 0 3 + q 0 − 2 ) \lambda=q_0(q_0^4-1)/(2(q_0^3+q_0-2))=q_0^2-q_0(q_0-1)^2/(q_0^3+q_0-2) λ = q 0 ( q 0 4 − 1 ) / ( 2 ( q 0 3 + q 0 − 2 )) = q 0 2 − q 0 ( q 0 − 1 ) 2 / ( q 0 3 + q 0 − 2 ) , where 0 < q 0 ( q 0 − 1 ) 2 < q 0 3 + q 0 − 2 0<q_0(q_0-1)^2<q_0^3+q_0-2 0 < q 0 ( q 0 − 1 ) 2 < q 0 3 + q 0 − 2 . Hence no pair ( q 0 , k ) (q_0,k) ( q 0 , k ) survives in this branch. For M = D q 0 2 − 1 M=D_{q_0^2-1} M = D q 0 2 − 1 the integrality of r = q 0 ( q 0 2 − 1 ) k / ( q 0 2 + 1 ) r=q_0(q_0^2-1)k/(q_0^2+1) r = q 0 ( q 0 2 − 1 ) k / ( q 0 2 + 1 ) alone suffices: if q 0 q_0 q 0 is odd then gcd ( q 0 2 + 1 , q 0 ( q 0 2 − 1 ) ) = 2 \gcd(q_0^2+1,q_0(q_0^2-1))=2 g cd( q 0 2 + 1 , q 0 ( q 0 2 − 1 )) = 2 , so ( q 0 2 + 1 ) / 2 (q_0^2+1)/2 ( q 0 2 + 1 ) /2 divides k k k , but ( q 0 2 + 1 ) / 2 (q_0^2+1)/2 ( q 0 2 + 1 ) /2 is coprime to q 0 2 − 1 q_0^2-1 q 0 2 − 1 while k k k divides q 0 2 − 1 q_0^2-1 q 0 2 − 1 , which forces k = 0 k=0 k = 0 ; for q 0 = 2 q_0=2 q 0 = 2 the requirement 5 ∣ k 5\mid k 5 ∣ k contradicts k ≤ 3 k\leq3 k ≤ 3 . Hence both branches are excluded, and the degree q = q 0 2 q=q_0^2 q = q 0 2 never arises for q 0 ≥ 3 q_0\geq3 q 0 ≥ 3 .
M M M a subfield subgroup: then M ≤ H M\leq H M ≤ H by the subfield chain, the block is { p 0 } \{p_0\} { p 0 } , k = 1 k=1 k = 1 and λ = 0 \lambda=0 λ = 0 ; excluded.
M ∈ { A 5 , S 4 , A 4 } M\in\{A_5,S_4,A_4\} M ∈ { A 5 , S 4 , A 4 } with M ≰ H M\nleq H M ≰ H : here r = q 0 ( q 0 2 − 1 ) k / ∣ M ∣ r=q_0(q_0^2-1)k/|M| r = q 0 ( q 0 2 − 1 ) k /∣ M ∣ and λ = 2 q 0 ( q 0 2 − 1 ) k ( k − 1 ) / ( ∣ M ∣ ( q 0 3 + q 0 − 2 ) ) \lambda=2q_0(q_0^2-1)k(k-1)/(|M|(q_0^3+q_0-2)) λ = 2 q 0 ( q 0 2 − 1 ) k ( k − 1 ) / ( ∣ M ∣ ( q 0 3 + q 0 − 2 )) , where k = ∣ M ∣ / ∣ M ∩ H g ∣ k=|M|/|M\cap H^g| k = ∣ M ∣/∣ M ∩ H g ∣ is a co-divisor of ∣ M ∣ |M| ∣ M ∣ . Write 2 k ( k − 1 ) / ∣ M ∣ = c n / c d 2k(k-1)/|M|=c_n/c_d 2 k ( k − 1 ) /∣ M ∣ = c n / c d in lowest terms. Then λ \lambda λ is an integer only if c d ( q 0 3 + q 0 − 2 ) c_d(q_0^3+q_0-2) c d ( q 0 3 + q 0 − 2 ) divides c n q 0 ( q 0 2 − 1 ) c_nq_0(q_0^2-1) c n q 0 ( q 0 2 − 1 ) , and the division of c n q 0 ( q 0 2 − 1 ) c_nq_0(q_0^2-1) c n q 0 ( q 0 2 − 1 ) by c d ( q 0 3 + q 0 − 2 ) c_d(q_0^3+q_0-2) c d ( q 0 3 + q 0 − 2 ) has the constant quotient c n / c d c_n/c_d c n / c d and the remainder 2 c n ( 1 − q 0 ) 2c_n(1-q_0) 2 c n ( 1 − q 0 ) . Since c n / c d ≤ 118 c_n/c_d\leq118 c n / c d ≤ 118 (attained at M = A 5 M=A_5 M = A 5 , k = 60 k=60 k = 60 ), for q 0 ≥ 16 q_0\geq16 q 0 ≥ 16 one has 0 < 2 c n ( q 0 − 1 ) < c d ( q 0 3 + q 0 − 2 ) 0<2c_n(q_0-1)<c_d(q_0^3+q_0-2) 0 < 2 c n ( q 0 − 1 ) < c d ( q 0 3 + q 0 − 2 ) for every admissible k k k , so λ \lambda λ is not an integer. Substituting the prime powers q 0 < 16 q_0<16 q 0 < 16 leaves exactly six survivors of the two integrality conditions, namely ( M , k ) = ( S 4 , 8 ) (M,k)=(S_4,8) ( M , k ) = ( S 4 , 8 ) at q 0 = 3 q_0=3 q 0 = 3 , ( S 4 , 12 ) (S_4,12) ( S 4 , 12 ) , ( A 4 , 12 ) (A_4,12) ( A 4 , 12 ) , ( A 5 , 12 ) (A_5,12) ( A 5 , 12 ) at q 0 = 4 q_0=4 q 0 = 4 , ( A 5 , 30 ) (A_5,30) ( A 5 , 30 ) at q 0 = 7 q_0=7 q 0 = 7 , and ( S 4 , 24 ) (S_4,24) ( S 4 , 24 ) at q 0 = 9 q_0=9 q 0 = 9 , with ( λ , r ) = ( 4 , 8 ) , ( 10 , 30 ) , ( 20 , 60 ) , ( 4 , 12 ) , ( 28 , 168 ) , ( 45 , 720 ) (\lambda,r)=(4,8),(10,30),(20,60),(4,12),(28,168),(45,720) ( λ , r ) = ( 4 , 8 ) , ( 10 , 30 ) , ( 20 , 60 ) , ( 4 , 12 ) , ( 28 , 168 ) , ( 45 , 720 ) respectively. All six have q = q 0 2 ∈ { 9 , 16 , 49 , 81 } ⊂ [ 9 , 81 ] q=q_0^2\in\{9,16,49,81\}\subset[9,81] q = q 0 2 ∈ { 9 , 16 , 49 , 81 } ⊂ [ 9 , 81 ] , inside the range q ≤ 127 q\leq127 q ≤ 127 of the exhaustive enumeration of gap_finalcheck4.g, which finds no surviving configuration. If M ≤ H M\leq H M ≤ H , maximality of M M M in G G G forces M = H M=H M = H , contradicting ∣ M ∣ < ∣ H ∣ |M|<|H| ∣ M ∣ < ∣ H ∣ ; so this case cannot occur. Hence f = 2 f=2 f = 2 is excluded.
We have therefore proved the following.
Theorem 4.6 (P S L 2 ( q ) PSL_2(q) P S L 2 ( q ) non-2-homogeneous classification). Let D D D be a non-trivial non-symmetric block-primitive 2-design whose automorphism group G G G has socle T = P S L 2 ( q ) T=PSL_2(q) T = P S L 2 ( q ) and acts point-primitively but not 2-homogeneously
on P \mathcal{P} P . If q q q is odd, no such D \mathcal{D} D exists. If q q q is even, then D \mathcal{D} D is one of the two designs of Proposition 4.28 when q = 8 q = 8 q = 8 , or one of the two designs of Proposition 4.29 when q = 32 q = 32 q = 32 , and no such D \mathcal{D} D exists when q = 16 q = 16 q = 16 or q = 64 q = 64 q = 64 . For a dihedral block stabilizer Lemma 4.8 reduces the block to a union of a a a orbits with ( q − 2 ) / 2 ∣ a ( a − 1 ) (q - 2)/2 \mid a(a - 1) ( q − 2 ) /2 ∣ a ( a − 1 ) and 1 ≤ a ≤ q / 2 1 \leq a \leq q/2 1 ≤ a ≤ q /2 , and Theorem 4.26 together with Corollary 4.27 removes the two extreme values a = 1 a = 1 a = 1 and a = q / 2 − 1 a = q/2 - 1 a = q /2 − 1 at every even degree, which settles q = 16 q = 16 q = 16 , q = 64 q = 64 q = 64 and every even degree with ( q − 2 ) / 2 (q - 2)/2 ( q − 2 ) /2 prime without enumerating any union of orbits; the enumeration of the unions of orbits is needed at q = 8 q = 8 q = 8 and q = 32 q = 32 q = 32 , where it produces nothing beyond the four designs listed, while at the remaining even degrees with 2 ≤ a ≤ q / 2 − 2 2 \leq a \leq q/2 - 2 2 ≤ a ≤ q /2 − 2 the admissible unions are exactly the 0-1 solutions of the class equations of Proposition 4.34 ; a block stabilizer of type S 3 S_3 S 3 is never admissible, at any even degree, because the subfield subgroup S 3 S_3 S 3 fixes a point of P \mathcal{P} P and is therefore not maximal (Lemma 4.7 ), so the subfield case at q = 32 q = 32 q = 32 is void as well; the subfield block stabilizers at q = 64 q = 64 q = 64 (of type A 5 A_5 A 5 ; the companion subfield subgroup P S L 2 ( 8 ) PSL_2(8) P S L 2 ( 8 ) is excluded by Fisher’s inequality) are excluded by Proposition 4.42 ; for the subfield stabilizer P S L 2 ( r ) PSL_2(r) P S L 2 ( r ) with r = 2 d ≥ 4 r = 2^d \geq4 r = 2 d ≥ 4 and q = r p q = r^p q = r p , p p p an odd prime (for p = 2 p = 2 p = 2 Fisher’s inequality already excludes it, since ∣ P S L 2 ( r ) ∣ = r ( r 2 − 1 ) |PSL_2(r)| = r(r^2 - 1) ∣ P S L 2 ( r ) ∣ = r ( r 2 − 1 ) exceeds 2 ( q ± 1 ) 2(q \pm1) 2 ( q ± 1 ) , and r = 4 r = 4 r = 4 , p = 3 p = 3 p = 3 is the degree q = 64 q = 64 q = 64 of Proposition 4.42 ), every block is a union of orbits of the three types of Lemma 4.32 , unions of at most two orbits are removed for p = 3 p = 3 p = 3 by Corollary 4.38 , while for p ≥ 5 p \geq5 p ≥ 5 Corollary 4.39 removes every union with T 2 u ( u − 1 ) ≤ ( u p − 2 ) ( u + 1 ) T^2u(u - 1) \leq(u^p - 2)(u + 1) T 2 u ( u − 1 ) ≤ ( u p − 2 ) ( u + 1 ) and every union of at most n 0 ( u , p ) n_0(u,p) n 0 ( u , p ) orbits, with n 0 ( u , p ) n_0(u,p) n 0 ( u , p ) as in Corollary 4.39, while a union beyond that range is likewise reduced to the class equations of Proposition 4.34 , a feasibility problem that lies outside the scope of this paper.
Proof. The Borel case is 2-transitive and excluded at the start of this subsection. For odd q q q the point stabilizer H H H is dihedral, exceptional, or subfield, and the four cases above exclude every admissible block stabilizer M M M , using Fisher’s inequality, the integrality of r r r and λ \lambda λ , the constancy of λ \lambda λ (verified for every prime power q ≤ 127 q \leq127 q ≤ 127 by the finite check of Appendix § 6.2 ), and the closed divisibility arguments recorded for each case, and, in Case 2(i), Lemma 4.4 together with the arithmetic conditions of [18 ] for the residue classes q ≡ 3 , 7 , 11 ( m o d 16 ) q \equiv3, 7, 11 \pmod{16} q ≡ 3 , 7 , 11 ( mod 16 ) and Lemma 4.5 for q = 15 q = 15 q = 15 (mod 16). For even q q q the dihedral maximal subgroups have orders 2 ( q − 1 ) 2(q - 1) 2 ( q − 1 ) and 2 ( q + 1 ) 2(q + 1) 2 ( q + 1 ) rather than q ∓ 1 q \mp1 q ∓ 1 , so the four cases above cover no subgroup; Lemma 4.8 instead fixes the orbit lengths and the arithmetic condition, the two extreme values of a a a are then removed for every even degree by Theorem 4.26 and Corollary 4.27, and the unions of orbits are enumerated exactly only at q = 8 q=8 q = 8 and q = 32 q=32 q = 32 , which gives Proposition 4.28 and Proposition 4.29 . At q = 16 q=16 q = 16 no design arises, the only maximal subgroups being D 30 D_{30} D 30 , D 34 D_{34} D 34 and the subfield group A 5 A_{5} A 5 of order 60 > 34 60>34 60 > 34 and hence not below ∣ H ∣ \lvert H\rvert ∣ H ∣ .
Even characteristic
Let q = 2 f q=2^{f} q = 2 f with f ≥ 2 f\ge2 f ≥ 2 , so that G = PSL 2 ( q ) = SL 2 ( q ) G=\operatorname{PSL}_{2}(q)=\operatorname{SL}_{2}(q) G = PSL 2 ( q ) = SL 2 ( q ) and ∣ G ∣ = q ( q 2 − 1 ) \lvert G\rvert=q(q^{2}-1) ∣ G ∣ = q ( q 2 − 1 ) . Both q − 1 q-1 q − 1 and q + 1 q+1 q + 1 are odd, so G G G has no dihedral subgroup of order q − 1 q-1 q − 1 or order q + 1 q+1 q + 1 ; the dihedral maximal subgroups of G G G have orders 2 ( q − 1 ) 2(q-1) 2 ( q − 1 ) and 2 ( q + 1 ) 2(q+1) 2 ( q + 1 ) . Cases 1 and 2, whose statements name subgroups of order q ∓ 1 q\mp1 q ∓ 1 , therefore cover no subgroup when q q q is even. By Dickson’s list the maximal subgroups of PSL 2 ( 2 f ) \operatorname{PSL}_{2}(2^{f}) PSL 2 ( 2 f ) are these two dihedral groups together with the subfield groups PSL 2 ( 2 d ) \operatorname{PSL}_{2}(2^{d}) PSL 2 ( 2 d ) with d ∣ f d\mid f d ∣ f , d < f d<f d < f , and d ≥ 2 d\ge2 d ≥ 2 whenever f f f is odd: the case d = 1 d=1 d = 1 gives S 3 S_{3} S 3 , which is never maximal, since it fixes the point { z , z 2 } \{z,z^{2}\} { z , z 2 } with z ∈ F 4 ∖ F 2 z\in\mathbb{F}_{4}\setminus\mathbb{F}_{2} z ∈ F 4 ∖ F 2 (Lemma 4.7 ). At q = 16 q=16 q = 16 the only subfield group is A 5 A_{5} A 5 , of order 60 above the orders 30 and 34 of the dihedral groups and hence not admissible. The two dihedral stabilizers to be examined are therefore D 2 ( q − 1 ) D_{2(q-1)} D 2 ( q − 1 ) and D 2 ( q + 1 ) D_{2(q+1)} D 2 ( q + 1 ) . The even degrees below this subsection, q = 2 q=2 q = 2 and q = 4 q=4 q = 4 , need no enumeration: PSL 2 ( 2 ) = S 3 \operatorname{PSL}_{2}(2)=S_{3} PSL 2 ( 2 ) = S 3 acts 2-transitively on 3 points, and for PSL 2 ( 4 ) = A 5 \operatorname{PSL}_{2}(4)=A_{5} PSL 2 ( 4 ) = A 5 the stabilizers A 4 A_{4} A 4 and D 10 D_{10} D 10 give 2-transitive actions on 5 and 6 points, while the stabilizer S 3 S_{3} S 3 gives v = 10 v=10 v = 10 , where the condition ∣ M ∣ < 6 \lvert M\rvert<6 ∣ M ∣ < 6 admits no maximal subgroup. We treat H = D 2 ( q + 1 ) H=D_{2(q+1)} H = D 2 ( q + 1 ) , for which v = [ G : H ] = q ( q − 1 ) / 2 v=[G:H]=q(q-1)/2 v = [ G : H ] = q ( q − 1 ) /2 , against the block stabilizer M = D 2 ( q − 1 ) M=D_{2(q-1)} M = D 2 ( q − 1 ) . The exchanged pair, H = D 2 ( q − 1 ) H=D_{2(q-1)} H = D 2 ( q − 1 ) against M = D 2 ( q + 1 ) M=D_{2(q+1)} M = D 2 ( q + 1 ) , has v = q ( q + 1 ) / 2 v=q(q+1)/2 v = q ( q + 1 ) /2 while [ G : M ] = q ( q − 1 ) / 2 [G:M]=q(q-1)/2 [ G : M ] = q ( q − 1 ) /2 , so b < v b<v b < v and Fisher’s inequality excludes it as soon as M M M is dihedral; with the subfield block stabilizer instead no block-primitive design arises, since S 3 S_{3} S 3 is not maximal in G G G , the two unions of S 3 S_{3} S 3 -orbits at the 36 points of q = 8 q=8 q = 8 which are 2-designs having an imprimitive block action (Proposition 4.28 ).
Lemma 4.7 (subfield S 3 S_{3} S 3 is never maximal). Let q = 2 f q=2^{f} q = 2 f with f ≥ 3 f\ge3 f ≥ 3 odd, let M = PSL 2 ( 2 ) = S 3 ≤ PSL 2 ( q ) = G M=\operatorname{PSL}_{2}(2)=S_{3}\le\operatorname{PSL}_{2}(q)=G M = PSL 2 ( 2 ) = S 3 ≤ PSL 2 ( q ) = G be the subfield subgroup of order 6, and let P P P be the set of points { x , x q } \{x,x^{q}\} { x , x q } of F q 2 ∖ F q \mathbb{F}_{q^{2}}\setminus\mathbb{F}_{q} F q 2 ∖ F q . Let z ∈ F 4 ∖ F 2 z\in\mathbb{F}_{4}\setminus\mathbb{F}_{2} z ∈ F 4 ∖ F 2 , so that z 2 = z + 1 z^{2}=z+1 z 2 = z + 1 , z 3 = 1 z^{3}=1 z 3 = 1 and z q = z 2 z^{q}=z^{2} z q = z 2 . Then p 0 = { z , z 2 } p_{0}=\{z,z^{2}\} p 0 = { z , z 2 } is a point of P P P and M ≤ G p 0 = D 2 ( q + 1 ) M\le G_{p_{0}}=D_{2(q+1)} M ≤ G p 0 = D 2 ( q + 1 ) ; in particular M < G p 0 M<G_{p_{0}} M < G p 0 , so M M M is not a maximal subgroup of G G G . Consequently no block-primitive design with socle PSL 2 ( q ) \operatorname{PSL}_{2}(q) PSL 2 ( q ) has a block stabilizer of type S 3 S_{3} S 3 when f f f is odd.
Proof. Since f f f is odd we have 2 f ≡ 2 ( m o d 3 ) 2^{f} \equiv2 \pmod{3} 2 f ≡ 2 ( mod 3 ) , and as z 3 = 1 z^{3}=1 z 3 = 1 this gives z q = z 2 f = z 2 z^{q}=z^{2^{f}}=z^{2} z q = z 2 f = z 2 . A subfield F 2 m \mathbb{F}_{2^{m}} F 2 m is contained in F 2 f \mathbb{F}_{2^{f}} F 2 f exactly when m ∣ f m \mid f m ∣ f , and 2 ∤ f 2 \nmid f 2 ∤ f , so F 4 ⊈ F q \mathbb{F}_{4} \nsubseteq\mathbb{F}_{q} F 4 ⊈ F q and z ∉ F q z \notin\mathbb{F}_{q} z ∈ / F q ; hence the point of P \mathcal{P} P containing z z z is { z , z q } = { z , z 2 } \{z,z^{q}\}=\{z,z^{2}\} { z , z q } = { z , z 2 } , and this pair is p 0 = P 1 ( F 4 ) ∖ P 1 ( F 2 ) p_{0}=\mathbb{P}^{1}(\mathbb{F}_{4})\setminus\mathbb{P}^{1}(\mathbb{F}_{2}) p 0 = P 1 ( F 4 ) ∖ P 1 ( F 2 ) . Every element of M M M is induced by a matrix with entries in F 2 \mathbb{F}_{2} F 2 , and a Möbius map with such entries carries F 4 ∪ { ∞ } \mathbb{F}_{4}\cup\{\infty\} F 4 ∪ { ∞ } onto itself; so every element of M M M permutes P 1 ( F 4 ) \mathbb{P}^{1}(\mathbb{F}_{4}) P 1 ( F 4 ) and fixes the subset p 0 p_{0} p 0 . Therefore M ≤ G p 0 M\leq G_{p_{0}} M ≤ G p 0 , and G p 0 G_{p_{0}} G p 0 is dihedral of order 2 ( q + 1 ) ≥ 18 > 6 = ∣ M ∣ 2(q+1)\geq18>6=|M| 2 ( q + 1 ) ≥ 18 > 6 = ∣ M ∣ . Since block-primitivity of the design is equivalent to maximality of the block stabilizer in G G G , the subgroup M M M can serve as a block stabilizer only of a design which is not block-primitive. □ \square □
Lemma 4.8. Assume q = 2 f q=2^{f} q = 2 f with f ≥ 2 f\geq2 f ≥ 2 , H = D 2 ( q + 1 ) H=D_{2(q+1)} H = D 2 ( q + 1 ) and M = D 2 ( q − 1 ) M=D_{2(q-1)} M = D 2 ( q − 1 ) . Then ∣ M ∩ H g ∣ = 2 |M\cap H^{g}|=2 ∣ M ∩ H g ∣ = 2 for every g ∈ G g\in G g ∈ G . Consequently every orbit of M M M on P \mathcal{P} P has length q − 1 q-1 q − 1 , and M M M has exactly q / 2 q/2 q /2 orbits. If a base block is a union of a a a orbits of M M M with p 0 p_{0} p 0 in one of them, then
k = a ( q − 1 ) , b = [ G : M ] = q ( q + 1 ) 2 , r = a ( q + 1 ) , λ = 2 a ( a q − a − 1 ) q − 2 , \begin{aligned}
k&=a(q-1), && b=[G:M]=\frac{q(q+1)}{2},\\
r&=a(q+1), && \lambda=\frac{2a(aq-a-1)}{q-2},
\end{aligned} k r = a ( q − 1 ) , = a ( q + 1 ) , b = [ G : M ] = 2 q ( q + 1 ) , λ = q − 2 2 a ( a q − a − 1 ) , and λ \lambda λ is an integer exactly when ( q − 2 ) / 2 (q-2)/2 ( q − 2 ) /2 divides a ( a − 1 ) a(a-1) a ( a − 1 ) ; also a ≤ q / 2 a\leq q/2 a ≤ q /2 .
Proof. Each conjugate H g H^{g} H g is dihedral of order 2 ( q + 1 ) 2(q+1) 2 ( q + 1 ) , so its elements of odd order form a cyclic group of order q + 1 q+1 q + 1 , which is coprime to q − 1 q-1 q − 1 . Hence M ∩ H g M\cap H^{g} M ∩ H g contains no element of odd order other than 1 1 1 , and as a subgroup all of whose non-identity elements are involutions it has order at most 2. This gives ∣ M ∩ H g ∣ ≤ 2 |M\cap H^{g}|\leq2 ∣ M ∩ H g ∣ ≤ 2 .
For the equality we count involutions. Every involution of SL 2 ( q ) \operatorname{SL}_{2}(q) SL 2 ( q ) is unipotent, hence a transvection [22 ] , and the involutions of G G G form a single conjugacy class of size q 2 − 1 q^{2}-1 q 2 − 1 . Each H g H^{g} H g contains q + 1 q+1 q + 1 involutions, so the number of pairs ( τ , H g ) (\tau,H^{g}) ( τ , H g ) with τ \tau τ an involution of H g H^{g} H g equals v ( q + 1 ) v(q+1) v ( q + 1 ) , and each involution τ \tau τ lies in v ( q + 1 ) / ( q 2 − 1 ) = q / 2 v(q+1)/(q^{2}-1)=q/2 v ( q + 1 ) / ( q 2 − 1 ) = q /2 conjugates of H H H . A point x x x is fixed by τ \tau τ exactly when τ ∈ H x \tau\in H^{x} τ ∈ H x , so τ \tau τ has q / 2 q/2 q /2 fixed points. If two involutions of M M M fixed a common point x x x , then M ∩ H x M\cap H^{x} M ∩ H x would contain both and would have order at least 4, against the bound just proved. The q − 1 q-1 q − 1 involutions of M M M therefore have pairwise disjoint fixed-point sets, each of size q / 2 q/2 q /2 , and ( q − 1 ) q / 2 = v (q-1)q/2=v ( q − 1 ) q /2 = v covers every point of P \mathcal{P} P . Hence ∣ M ∩ H x ∣ = 2 |M\cap H^{x}|=2 ∣ M ∩ H x ∣ = 2 for every point x x x .
Each M M M -orbit then has length ∣ M ∣ / ∣ M ∩ H x ∣ = 2 ( q − 1 ) / 2 = q − 1 |M|/|M\cap H^{x}|=2(q-1)/2=q-1 ∣ M ∣/∣ M ∩ H x ∣ = 2 ( q − 1 ) /2 = q − 1 , and there are v / ( q − 1 ) = q / 2 v/(q-1)=q/2 v / ( q − 1 ) = q /2 of them. A base block which is the union of a a a such orbits has k = a ( q − 1 ) k=a(q-1) k = a ( q − 1 ) . Since M M M is a subgroup of index b = [ G : M ] = ∣ G ∣ / ∣ M ∣ = q ( q 2 − 1 ) / ( 2 ( q − 1 ) ) = q ( q + 1 ) / 2 b=[G:M]=|G|/|M|=q(q^2-1)/(2(q-1))=q(q+1)/2 b = [ G : M ] = ∣ G ∣/∣ M ∣ = q ( q 2 − 1 ) / ( 2 ( q − 1 )) = q ( q + 1 ) /2 , the replication number is r = b k / v = a ( q + 1 ) r=bk/v=a(q+1) r = bk / v = a ( q + 1 ) . Fisher’s inequality forces b ≥ v b\geq v b ≥ v , that is q ( q + 1 ) / 2 ≥ q ( q − 1 ) / 2 q(q+1)/2\geq q(q-1)/2 q ( q + 1 ) /2 ≥ q ( q − 1 ) /2 , and gives no further restriction; the bound a ≤ q / 2 a\leq q/2 a ≤ q /2 is the number of orbits. Finally v − 1 = ( q − 2 ) ( q + 1 ) / 2 v-1=(q-2)(q+1)/2 v − 1 = ( q − 2 ) ( q + 1 ) /2 , so
λ = r ( k − 1 ) v − 1 = a ( q + 1 ) ( a ( q − 1 ) − 1 ) ( q − 2 ) ( q + 1 ) / 2 = 2 a ( a q − a − 1 ) q − 2 . \lambda=\frac{r(k-1)}{v-1}=\frac{a(q+1)(a(q-1)-1)}{(q-2)(q+1)/2}=\frac{2a(aq-a-1)}{q-2}. λ = v − 1 r ( k − 1 ) = ( q − 2 ) ( q + 1 ) /2 a ( q + 1 ) ( a ( q − 1 ) − 1 ) = q − 2 2 a ( a q − a − 1 ) .
Write q − 2 = 2 m q-2=2m q − 2 = 2 m with m m m odd. Then λ = a ( a q − a − 1 ) / m \lambda=a(aq-a-1)/m λ = a ( a q − a − 1 ) / m is an integer exactly when m m m divides a ( a q − a − 1 ) a(aq-a-1) a ( a q − a − 1 ) , and a q − a − 1 ≡ a − 1 ( m o d m ) aq-a-1\equiv a-1\pmod m a q − a − 1 ≡ a − 1 ( mod m ) because q ≡ 1 ( m o d m ) q\equiv1\pmod m q ≡ 1 ( mod m ) . Hence λ ∈ Z \lambda\in\mathbb{Z} λ ∈ Z exactly when m = ( q − 2 ) / 2 m=(q-2)/2 m = ( q − 2 ) /2 divides a ( a − 1 ) a(a-1) a ( a − 1 ) .
Lemma 4.9. Let q = 2 f q=2^f q = 2 f with f ≥ 2 f\geq2 f ≥ 2 , let P P P be the v = q ( q − 1 ) / 2 v=q(q-1)/2 v = q ( q − 1 ) /2 points { x , x q } \{x,x^q\} { x , x q } , and let P ~ \widetilde{P} P be the partition of P P P into the q / 2 q/2 q /2 orbits of M = D 2 ( q − 1 ) M=D_{2(q-1)} M = D 2 ( q − 1 ) . Then the setwise stabilizer of P ~ \widetilde{P} P in PGL 2 ( q ) \operatorname{PGL}_2(q) PGL 2 ( q ) is the monomial group
N = { x ↦ λ x , x ↦ λ / x : λ ∈ F q × } , N=\{x\mapsto\lambda x,\ x\mapsto\lambda/x:\lambda\in F_q^{\times}\}, N = { x ↦ λ x , x ↦ λ / x : λ ∈ F q × } , of order 2 ( q − 1 ) = ∣ M ∣ 2(q-1)=|M| 2 ( q − 1 ) = ∣ M ∣ , and N ≤ PSL 2 ( q ) = G N\leq\operatorname{PSL}_2(q)=G N ≤ PSL 2 ( q ) = G ; in particular N = M N=M N = M . Since ∣ G ∣ = q ( q 2 − 1 ) > 2 ( q − 1 ) |G|=q(q^2-1)>2(q-1) ∣ G ∣ = q ( q 2 − 1 ) > 2 ( q − 1 ) , some element of G G G lies outside N N N , and G G G does not stabilize P ~ \widetilde{P} P .
Proof. By Lemma 4.8 the orbit of M M M through X = { z , z q } X=\{z,z^q\} X = { z , z q } is O X = { X a : a ∈ F q × } O_X=\{X_a:a\in F_q^\times\} O X = { X a : a ∈ F q × } with X a = { a z , a z q } X_a=\{az,az^q\} X a = { a z , a z q } , that is, the set of scalar multiples of X X X . Let T T T be a Möbius map with entries in F q F_q F q which carries every M M M -orbit into a single M M M -orbit, and write T ( x ) = ( α x + β ) / ( γ x + δ ) T(x)=(\alpha x+\beta)/(\gamma x+\delta) T ( x ) = ( α x + β ) / ( γ x + δ ) with α δ − β γ ≠ 0 \alpha\delta-\beta\gamma\ne0 α δ − β γ = 0 . For a ∈ F q × a\in F_q^\times a ∈ F q × the points T ( a x ) T(ax) T ( a x ) and T ( x ) T(x) T ( x ) then lie in the same M M M -orbit, so T ( a x ) = b ( a ) T ( x ) T(ax)=b(a)T(x) T ( a x ) = b ( a ) T ( x ) for some b ( a ) ∈ F q × b(a)\in F_q^\times b ( a ) ∈ F q × , and comparing the coefficients of the identity
( α a x + β ) ( γ x + δ ) = b ( a ) ( α x + β ) ( γ a x + δ ) (\alpha ax+\beta)(\gamma x+\delta)=b(a)(\alpha x+\beta)(\gamma ax+\delta) ( α a x + β ) ( γ x + δ ) = b ( a ) ( α x + β ) ( γ a x + δ )
gives α γ a = b ( a ) α γ a \alpha\gamma a=b(a)\alpha\gamma a α γ a = b ( a ) α γ a and α a δ + β γ = b ( a ) ( α δ + β γ a ) \alpha a\delta+\beta\gamma=b(a)(\alpha\delta+\beta\gamma a) α a δ + β γ = b ( a ) ( α δ + β γ a ) . If α γ ≠ 0 \alpha\gamma\ne0 α γ = 0 , the first relation gives b ( a ) = 1 b(a)=1 b ( a ) = 1 for every a a a , and the second then gives ( a − 1 ) ( α δ − β γ ) = 0 (a-1)(\alpha\delta-\beta\gamma)=0 ( a − 1 ) ( α δ − β γ ) = 0 , against α δ − β γ ≠ 0 \alpha\delta-\beta\gamma\ne0 α δ − β γ = 0 . Hence α γ = 0 \alpha\gamma=0 α γ = 0 . If α = 0 \alpha=0 α = 0 and δ ≠ 0 \delta\ne0 δ = 0 , the constant terms give b ( a ) = 1 b(a)=1 b ( a ) = 1 and the coefficients of x x x give γ = γ a \gamma=\gamma a γ = γ a , whence γ = 0 \gamma=0 γ = 0 and α δ − β γ = 0 \alpha\delta-\beta\gamma=0 α δ − β γ = 0 , again a contradiction; so α = δ = 0 \alpha=\delta=0 α = δ = 0 and T ( x ) = β / ( γ x ) T(x)=\beta/(\gamma x) T ( x ) = β / ( γ x ) . If γ = 0 \gamma=0 γ = 0 and β ≠ 0 \beta\ne0 β = 0 , the constant terms give b ( a ) = 1 b(a)=1 b ( a ) = 1 and the coefficients of x x x give α a = α \alpha a=\alpha α a = α , again against a ≠ 1 a\ne1 a = 1 ; so γ = β = 0 \gamma=\beta=0 γ = β = 0 and T ( x ) = α x / δ T(x)=\alpha x/\delta T ( x ) = α x / δ . Thus every such T T T is of the form x ↦ λ x x\mapsto\lambda x x ↦ λ x or x ↦ λ / x x\mapsto\lambda/x x ↦ λ / x .
Conversely x ↦ λ x x\mapsto\lambda x x ↦ λ x carries O X O_X O X onto O λ X O_{\lambda X} O λ X and x ↦ λ / x x\mapsto\lambda/x x ↦ λ / x carries O X O_X O X onto O λ / X O_{\lambda/X} O λ / X , so both stabilize P ~ \widetilde{\mathcal{P}} P , and N N N is the full setwise stabilizer, of order 2 ( q − 1 ) 2(q-1) 2 ( q − 1 ) . In characteristic 2 the group F q × F_q^\times F q × has odd order q − 1 q-1 q − 1 , so squaring is a bijection of F q × F_q^\times F q × and every λ \lambda λ is a square; the matrices λ I \lambda I λ I and ( 0 λ 1 0 ) \begin{pmatrix}0&\lambda\\1&0\end{pmatrix} ( 0 1 λ 0 ) therefore belong to S L 2 ( q ) SL_2(q) S L 2 ( q ) , whence N ≤ G N\leq G N ≤ G . By Lemma 4.8 the group M M M has order 2 ( q − 1 ) 2(q-1) 2 ( q − 1 ) and fixes each orbit of P ~ \widetilde{\mathcal{P}} P , so M ≤ N M\leq N M ≤ N and hence M = N M=N M = N . Finally ∣ G ∣ = q ( q 2 − 1 ) > 2 ( q − 1 ) = ∣ N ∣ |G|=q(q^2-1)>2(q-1)=|N| ∣ G ∣ = q ( q 2 − 1 ) > 2 ( q − 1 ) = ∣ N ∣ , so G ⊈ N G\not\subseteq N G ⊆ N and G G G does not stabilize P ~ \widetilde{\mathcal{P}} P .
The partition P ~ \widetilde{\mathcal{P}} P is thus not an imprimitivity system of G G G : a design whose point set is P \mathcal{P} P and whose block stabilizer is M M M has no G G G -invariant partition of its points arising from the orbits of M M M , and the imprimitivity-based bounds for block-transitive designs, which presuppose such a partition, cannot be applied to it. The same conclusion holds at every even degree, independently of a a a and of q q q .
The value a = 1 a=1 a = 1 of Lemma 4.8 , in which the base block is a single orbit of M M M and λ = 2 \lambda=2 λ = 2 , can be settled for every even degree without enumerating unions of orbits. Two ingredients are needed: an invariant that separates the orbits of M M M , and an explicit transport of pairs that the invariant leaves together.
For X = { z , z q } ∈ P X=\{z,z^q\}\in\mathcal{P} X = { z , z q } ∈ P put Z = z q Z=z^q Z = z q , N = z Z = z q + 1 N=zZ=z^{q+1} N = z Z = z q + 1 and u = z 2 / N u=z^2/N u = z 2 / N , so that u q + 1 = 1 u^{q+1}=1 u q + 1 = 1 ; the orbit of M = D 2 ( q − 1 ) M=D_{2(q-1)} M = D 2 ( q − 1 ) through X X X is O X = { X a : a ∈ F q × } O_X=\{X_a:a\in F_q^\times\} O X = { X a : a ∈ F q × } with X a = { a z , a z q } X_a=\{az,az^q\} X a = { a z , a z q } . For distinct X = { z , z q } X=\{z,z^q\} X = { z , z q } and Y = { w , w q } Y=\{w,w^q\} Y = { w , w q } of P \mathcal{P} P put
cr ( X , Y ) = ( z + w ) ( z q + w q ) ( z + w q ) ( z q + w ) . \operatorname{cr}(X,Y)=\frac{(z+w)(z^q+w^q)}{(z+w^q)(z^q+w)}. cr ( X , Y ) = ( z + w q ) ( z q + w ) ( z + w ) ( z q + w q ) .
the cross-ratio of the four points z , w , z q , w q z,w,z^q,w^q z , w , z q , w q . It lies in F q × F_q^\times F q × , it is invariant under G G G , and it is defined only up to inversion: replacing the representative w w w of Y Y Y by w q w^q w q , or the representative z z z of X X X by z q z^q z q , inverts the value (Lemma 4.12 ). A pair { X , Y } \{X,Y\} { X , Y } thus carries the two values ρ \rho ρ and ρ − 1 \rho^{-1} ρ − 1 ; and ρ = 1 \rho=1 ρ = 1 would force ( z + z q ) ( w + w q ) = 0 (z+z^q)(w+w^q)=0 ( z + z q ) ( w + w q ) = 0 , that is Y ∈ P 1 ( F q ) Y\in\mathbb{P}^1(F_q) Y ∈ P 1 ( F q ) , so ρ = 1 \rho=1 ρ = 1 does not occur for two distinct points of P \mathcal{P} P .
Lemma 4.10. Let q = 2 f q=2^f q = 2 f with f ≥ 2 f\geq2 f ≥ 2 , let X = { z , z q } ∈ P X=\{z,z^q\}\in\mathcal{P} X = { z , z q } ∈ P , and put Z = z q Z=z^q Z = z q , α = z + Z \alpha=z+Z α = z + Z , N = z Z N=zZ N = z Z and u = z 2 / N u=z^2/N u = z 2 / N . Then κ = u + 1 / u \kappa=u+1/u κ = u + 1/ u satisfies
κ = Z z + z Z = Z 2 + z 2 z Z = ( Z + z ) 2 z Z = α 2 N . \kappa=\frac{Z}{z}+\frac{z}{Z}=\frac{Z^2+z^2}{zZ}=\frac{(Z+z)^2}{zZ}=\frac{\alpha^2}{N}. κ = z Z + Z z = z Z Z 2 + z 2 = z Z ( Z + z ) 2 = N α 2 .
and κ ≠ 0 \kappa\ne0 κ = 0 depends on the orbit O X O_X O X of M M M through X X X only, not on the point X X X chosen in it.
Proof. Since u = z 2 / ( z Z ) = z / Z u=z^2/(zZ)=z/Z u = z 2 / ( z Z ) = z / Z we have 1 / u = Z / z 1/u=Z/z 1/ u = Z / z , which gives the first equality, and Z 2 + z 2 = ( Z + z ) 2 Z^2+z^2=(Z+z)^2 Z 2 + z 2 = ( Z + z ) 2 holds in characteristic 2. Replacing z z z by a z az a z with a ∈ F q × a\in F_q^\times a ∈ F q × multiplies z z z , Z Z Z , α \alpha α and N N N by a a a , a a a , a 2 a^2 a 2 and a 2 a^2 a 2 , so α 2 / N \alpha^2/N α 2 / N is unchanged and κ \kappa κ is an invariant of O X O_X O X . Finally u ≠ 1 u\ne1 u = 1 because z ∉ F q z\notin F_q z ∈ / F q , so κ ≠ 0 \kappa\ne0 κ = 0 .
Lemma 4.11. Let t , s ∈ F q × ∖ { 1 } t,s\in F_q^\times\setminus\{1\} t , s ∈ F q × ∖ { 1 } , let Y t = { z t , z t q } = { z t , Z t } Y_t=\{zt,zt^q\}=\{zt,Zt\} Y t = { z t , z t q } = { z t , Z t } and Y s = { z s , Z s } Y_s=\{zs,Zs\} Y s = { z s , Z s } be points of O X O_X O X .
If s = 1 / t s=1/t s = 1/ t , then g ( x ) = N / x g(x)=N/x g ( x ) = N / x belongs to G G G and satisfies g ( X ) = X g(X)=X g ( X ) = X and g ( Y t ) = Y s g(Y_t)=Y_s g ( Y t ) = Y s .
If s + 1 / s = t + 1 / t + κ s+1/s=t+1/t+\kappa s + 1/ s = t + 1/ t + κ , put r = s t r=st r = s t , b = r α / ( 1 + r ) b=r\alpha/(1+r) b = r α / ( 1 + r ) , c = ( s + t ) / ( r α ) c=(s+t)/(r\alpha) c = ( s + t ) / ( r α ) and g ( x ) = ( x + b ) / ( c x + 1 ) g(x)=(x+b)/(cx+1) g ( x ) = ( x + b ) / ( c x + 1 ) . Then g g g belongs to G G G , g ( z ) = Z g(z)=Z g ( z ) = Z and g ( z t ) = z s g(zt)=zs g ( z t ) = z s , so g ( X ) = X g(X)=X g ( X ) = X and g ( Y t ) = Y s g(Y_t)=Y_s g ( Y t ) = Y s ; moreover det g = ( 1 + s ) ( 1 + t ) / ( 1 + s t ) ≠ 0 \det g=(1+s)(1+t)/(1+st)\ne0 det g = ( 1 + s ) ( 1 + t ) / ( 1 + s t ) = 0 .
In both cases the unordered pairs { X , Y t } \{X,Y_t\} { X , Y t } and { X , Y s } \{X,Y_s\} { X , Y s } lie in one orbit of G G G on the unordered pairs of P \mathcal{P} P .
Proof. In (i), g ( z ) = N / z = Z g(z)=N/z=Z g ( z ) = N / z = Z and g ( Z ) = N / Z = z g(Z)=N/Z=z g ( Z ) = N / Z = z , so g g g interchanges z z z and Z Z Z and fixes X X X ; while g ( z t ) = N / ( z t ) = Z / t = z ( 1 / t ) g(zt)=N/(zt)=Z/t=z(1/t) g ( z t ) = N / ( z t ) = Z / t = z ( 1/ t ) and g ( Z t ) = z / t g(Zt)=z/t g ( Z t ) = z / t , so g g g sends Y t Y_t Y t to Y 1 / t = Y s Y_{1/t}=Y_s Y 1/ t = Y s . The coefficients 0 , N , 1 , 0 0,N,1,0 0 , N , 1 , 0 lie in F q F_q F q .
In (ii), the condition g ( z ) = Z g(z)=Z g ( z ) = Z reads z + b = Z ( c z + 1 ) = c N + Z z+b=Z(cz+1)=cN+Z z + b = Z ( cz + 1 ) = c N + Z , that is b + c N = α b+cN=\alpha b + c N = α , whence c = ( b + α ) / N c=(b+\alpha)/N c = ( b + α ) / N . The condition g ( z t ) = z s g(zt)=zs g ( z t ) = z s reads z t + b + z s ( c z t + 1 ) = 0 zt+b+zs(czt+1)=0 z t + b + z s ( cz t + 1 ) = 0 , and with z 2 = α z + N z^2=\alpha z+N z 2 = α z + N this becomes
z ( c s t α + t + s ) + ( c s t N + b ) = 0 z(cst\alpha+t+s)+(cstN+b)=0 z ( cs t α + t + s ) + ( cs tN + b ) = 0
so c s t α = t + s cst\alpha=t+s cs t α = t + s and c s t N = b cstN=b cs tN = b . The second of these combined with c = ( b + α ) / N c=(b+\alpha)/N c = ( b + α ) / N gives b ( 1 + s t ) = s t α b(1+st)=st\alpha b ( 1 + s t ) = s t α , that is b = r α / ( 1 + r ) b=r\alpha/(1+r) b = r α / ( 1 + r ) with r = s t r=st r = s t ; the first then gives c = ( s + t ) / ( r α ) c=(s+t)/(r\alpha) c = ( s + t ) / ( r α ) . The two expressions agree precisely when
( s + t ) ( 1 + s t ) = s t α 2 N , (s+t)(1+st)=st\frac{\alpha^2}{N}, ( s + t ) ( 1 + s t ) = s t N α 2 ,
which is t + 1 / t + s + 1 / s = α 2 / N = κ t+1/t+s+1/s=\alpha^2/N=\kappa t + 1/ t + s + 1/ s = α 2 / N = κ , the hypothesis of (ii) by Lemma 4.10 . All coefficients lie in F q F_q F q , and 1 + s t ≠ 0 1+st\ne0 1 + s t = 0 because s ≠ 1 / t s\ne1/t s = 1/ t ; finally det g = 1 + b c = ( 1 + s ) ( 1 + t ) / ( 1 + s t ) ≠ 0 \det g=1+bc=(1+s)(1+t)/(1+st)\ne0 det g = 1 + b c = ( 1 + s ) ( 1 + t ) / ( 1 + s t ) = 0 since s , t ≠ 1 s,t\ne1 s , t = 1 .
An element of G G G has coefficients in F q \mathbb{F}_q F q and therefore commutes with x ↦ x q x \mapsto x^q x ↦ x q , so it maps the point { w , w q } \{w,w^q\} { w , w q } to a point. Hence in both cases g g g carries the pair { X , Y t } \{X,Y_t\} { X , Y t } to { X , Y s } \{X,Y_s\} { X , Y s } .
The transport of Lemma 4.11 rests on the map x ↦ N / x x \mapsto N/x x ↦ N / x , and that map inverts the cross-ratio based at X X X . This yields a parity law for the class counts of an arbitrary orbit, and singles out the one class which may occur an odd number of times.
Lemma 4.12 (cross-ratio inversion). Let q = 2 f q=2^f q = 2 f with f ≥ 2 f \ge2 f ≥ 2 , let X = { z , z q } ∈ P X=\{z,z^q\}\in\mathcal{P} X = { z , z q } ∈ P , and put Z = z q Z=z^q Z = z q and N = z Z N=zZ N = z Z . Let τ \tau τ be the map x ↦ N / x x\mapsto N/x x ↦ N / x on P 1 ( F q 2 ) \mathbb{P}^1(\mathbb{F}_{q^2}) P 1 ( F q 2 ) . Then τ ∈ G \tau\in G τ ∈ G , τ ( X ) = X \tau(X)=X τ ( X ) = X , and τ \tau τ is an involution; and for every Y = { w , w q } ∈ P Y=\{w,w^q\}\in\mathcal{P} Y = { w , w q } ∈ P with Y ≠ X Y\ne X Y = X ,
cr ( X , τ ( Y ) ) = cr ( X , Y ) − 1 , with τ ( Y ) = { N / w , N / w q } . \operatorname{cr}(X,\tau(Y))=\operatorname{cr}(X,Y)^{-1}, \qquad\text{with }\tau(Y)=\{N/w,N/w^q\}. cr ( X , τ ( Y )) = cr ( X , Y ) − 1 , with τ ( Y ) = { N / w , N / w q } . the two values being computed with the representatives displayed for X X X and for Y Y Y , respectively for τ ( Y ) \tau(Y) τ ( Y ) . Replacing the representative of either point by its conjugate inverts the value, so cr \operatorname{cr} cr is an invariant of the unordered pair { X , Y } \{X,Y\} { X , Y } only up to inversion, and τ \tau τ realizes that inversion. Since τ ∈ G \tau\in G τ ∈ G fixes X X X , the unordered pairs { X , Y } \{X,Y\} { X , Y } and { X , τ ( Y ) } \{X,\tau(Y)\} { X , τ ( Y )} lie in one orbit of G G G on the unordered pairs of P \mathcal{P} P .
Proof. The map τ \tau τ is induced by the matrix ( 0 N 1 0 ) \begin{pmatrix}0 & N \\ 1 & 0\end{pmatrix} ( 0 1 N 0 ) , whose entries lie in F q \mathbb{F}_q F q and whose determinant N ≠ 0 N\ne0 N = 0 is a square, because q − 1 q-1 q − 1 is odd and hence squaring is a bijection of F q × \mathbb{F}_q^\times F q × ; after rescaling the determinant is 1 1 1 , so τ ∈ S L 2 ( q ) = G \tau\in SL_2(q)=G τ ∈ S L 2 ( q ) = G . The map τ \tau τ is an involution with τ ( z ) = N / z = Z \tau(z)=N/z=Z τ ( z ) = N / z = Z and τ ( Z ) = z \tau(Z)=z τ ( Z ) = z , so it fixes the point X X X ; and τ \tau τ carries points to points because an element of G G G has coefficients in F q \mathbb{F}_q F q and therefore commutes with x ↦ x q x\mapsto x^q x ↦ x q .
For the identity we compare the four factors of the two cross-ratios. Since N ∈ F q N\in\mathbb{F}_q N ∈ F q we have N q = N N^q=N N q = N , and hence
z − τ ( w ) = z w − N w = z ( w − Z ) w , z q − τ ( w ) q = Z w q − N w q = Z ( w q − z ) w q , z-\tau(w)=\frac{zw-N}{w}=\frac{z(w-Z)}{w}, \qquad z^q-\tau(w)^q=\frac{Zw^q-N}{w^q}=\frac{Z(w^q-z)}{w^q}, z − τ ( w ) = w z w − N = w z ( w − Z ) , z q − τ ( w ) q = w q Z w q − N = w q Z ( w q − z ) ,
z − τ ( w ) q = z w q − N w q = z ( w q − Z ) w q , z q − τ ( w ) = Z w − N w = Z ( w − z ) w . z-\tau(w)^q=\frac{zw^q-N}{w^q}=\frac{z(w^q-Z)}{w^q}, \qquad z^q-\tau(w)=\frac{Zw-N}{w}=\frac{Z(w-z)}{w}. z − τ ( w ) q = w q z w q − N = w q z ( w q − Z ) , z q − τ ( w ) = w Z w − N = w Z ( w − z ) .
Multiplying the first two and dividing by the product of the last two, the common factor z Z / ( w w q ) zZ/(ww^q) z Z / ( w w q ) cancels and
cr ( X , τ ( Y ) ) = ( w − Z ) ( w q − z ) ( w q − Z ) ( w − z ) = ( w + Z ) ( w q + z ) ( w q + Z ) ( w + z ) . \operatorname{cr}(X,\tau(Y))=\frac{(w-Z)(w^q-z)}{(w^q-Z)(w-z)}=\frac{(w+Z)(w^q+z)}{(w^q+Z)(w+z)}. cr ( X , τ ( Y )) = ( w q − Z ) ( w − z ) ( w − Z ) ( w q − z ) = ( w q + Z ) ( w + z ) ( w + Z ) ( w q + z ) .
whereas
cr ( X , Y ) = ( z + w ) ( Z + w q ) ( z + w q ) ( Z + w ) = ( w + z ) ( w q + Z ) ( w + Z ) ( w q + z ) , \operatorname{cr}(X,Y)=\frac{(z+w)(Z+w^{q})}{(z+w^{q})(Z+w)}=\frac{(w+z)(w^{q}+Z)}{(w+Z)(w^{q}+z)}, cr ( X , Y ) = ( z + w q ) ( Z + w ) ( z + w ) ( Z + w q ) = ( w + Z ) ( w q + z ) ( w + z ) ( w q + Z ) ,
the two right-hand sides being reciprocal, since characteristic 2 makes ( w + z ) ( w q + Z ) (w+z)(w^{q}+Z) ( w + z ) ( w q + Z ) and ( w + Z ) ( w q + z ) (w+Z)(w^{q}+z) ( w + Z ) ( w q + z ) the very same product in reverse order. Likewise swapping w w w with w q w^{q} w q , or z z z with Z Z Z , in the four factors turns the value into its reciprocal, so the unordered pair { X , Y } \{X,Y\} { X , Y } carries the two values ρ \rho ρ and ρ − 1 \rho^{-1} ρ − 1 and nothing else. Finally τ ∈ G \tau\in G τ ∈ G and τ ( X ) = X \tau(X)=X τ ( X ) = X give τ ( { X , Y } ) = { X , τ ( Y ) } \tau(\{X,Y\})=\{X,\tau(Y)\} τ ({ X , Y }) = { X , τ ( Y )} , so the two pairs are in one orbit of G G G .
Corollary 4.13 (even class counts). Let q = 2 f q=2^{f} q = 2 f with f ≥ 2 f\geq2 f ≥ 2 , let X ∈ P X\in\mathcal{P} X ∈ P , and let O O O be an orbit of M = D 2 ( q − 1 ) M=D_{2(q-1)} M = D 2 ( q − 1 ) on P \mathcal{P} P with X ∉ O X\notin O X ∈ / O . Write O = { Y a : a ∈ F q × } O=\{Y_a:a\in\mathbb{F}_q^\times\} O = { Y a : a ∈ F q × } with Y a = { a w , a w q } Y_a=\{aw,aw^{q}\} Y a = { a w , a w q } for a representative Y 1 = { w , w q } Y_1=\{w,w^{q}\} Y 1 = { w , w q } of O O O , put k = N ( X ) / N ( Y 1 ) k=N(X)/N(Y_1) k = N ( X ) / N ( Y 1 ) , and let a 0 ∈ F q × a_0\in\mathbb{F}_q^\times a 0 ∈ F q × be the unique solution of a 0 2 = k a_0^2=k a 0 2 = k . Then τ X \tau_X τ X acts on O O O as a ↦ k / a a\mapsto k/a a ↦ k / a , its only fixed point is Y a 0 Y_{a_0} Y a 0 , and the number of points Y ∈ O Y\in O Y ∈ O with { X , Y } \{X,Y\} { X , Y } in a prescribed class C C C is even for every C C C except the class of { X , Y a 0 } \{X,Y_{a_0}\} { X , Y a 0 } , for which it is odd. Moreover N ( Y a 0 ) = N ( X ) N(Y_{a_0})=N(X) N ( Y a 0 ) = N ( X ) .
Proof. By Lemma 4.12 the map τ X \tau_X τ X lies in G G G , fixes X X X , and sends Y a = { a w , a w q } Y_a=\{aw,aw^{q}\} Y a = { a w , a w q } to { N ( X ) / ( a w ) , N ( X ) / ( a w q ) } = Y k / a \{N(X)/(aw),N(X)/(aw^{q})\}=Y_{k/a} { N ( X ) / ( a w ) , N ( X ) / ( a w q )} = Y k / a with k = N ( X ) / N ( Y 1 ) k=N(X)/N(Y_1) k = N ( X ) / N ( Y 1 ) ; so τ X \tau_X τ X preserves O O O and acts on it as a ↦ k / a a\mapsto k/a a ↦ k / a . The fixed points of that involution are the solutions of a 2 = k a^2=k a 2 = k , and there is exactly one, since q − 1 q-1 q − 1 is odd and squaring is injective on F q × \mathbb{F}_q^\times F q × . Every other point of O O O is paired with a distinct point of O O O in the same class, because { X , τ X ( Y ) } = τ X ( { X , Y } ) \{X,\tau_X(Y)\}=\tau_X(\{X,Y\}) { X , τ X ( Y )} = τ X ({ X , Y }) and τ X ∈ G \tau_X\in G τ X ∈ G ; hence the class counts are even away from the fixed point and odd at it. Finally N ( Y a 0 ) = a 0 2 N ( Y 1 ) = k N ( Y 1 ) = N ( X ) N(Y_{a_0})=a_0^2N(Y_1)=kN(Y_1)=N(X) N ( Y a 0 ) = a 0 2 N ( Y 1 ) = k N ( Y 1 ) = N ( X ) .
Lemma 4.14 (norm fibres are transversals). Let q = 2 f q=2^{f} q = 2 f with f ≥ 2 f\geq2 f ≥ 2 , and for ν ∈ F q × \nu\in\mathbb{F}_q^\times ν ∈ F q × put P ν = { Y ∈ P : N ( Y ) = ν } \mathcal{P}_{\nu}=\{Y\in\mathcal{P}:N(Y)=\nu\} P ν = { Y ∈ P : N ( Y ) = ν } , where N ( Y ) = w w q N(Y)=ww^{q} N ( Y ) = w w q for Y = { w , w q } Y=\{w,w^{q}\} Y = { w , w q } . Then ∣ P ν ∣ = q / 2 |\mathcal{P}_{\nu}|=q/2 ∣ P ν ∣ = q /2 , and every orbit of M = D 2 ( q − 1 ) M=D_{2(q-1)} M = D 2 ( q − 1 ) meets P ν \mathcal{P}_{\nu} P ν in exactly one point. Consequently the point Y a 0 Y_{a_0} Y a 0 of Corollary 4.13 is the unique point of O O O with N ( Y ) = N ( X ) N(Y)=N(X) N ( Y ) = N ( X ) .
Proof. For Y a = { a w , a w q } Y_a=\{aw,aw^{q}\} Y a = { a w , a w q } one has N ( Y a ) = a 2 N ( Y 1 ) N(Y_a)=a^2N(Y_1) N ( Y a ) = a 2 N ( Y 1 ) , and a ↦ a 2 a\mapsto a^2 a ↦ a 2 is a bijection of F q × \mathbb{F}_q^\times F q × , since q − 1 q-1 q − 1 is odd; as a a a runs over F q × \mathbb{F}_q^\times F q × the value N ( Y a ) N(Y_a) N ( Y a ) therefore runs over F q × \mathbb{F}_q^\times F q × exactly once. So each orbit O O O of length q − 1 q-1 q − 1 meets P ν \mathcal{P}_{\nu} P ν in exactly one point, and summing over the q / 2 q/2 q /2 orbits gives ∣ P ν ∣ = q / 2 |\mathcal{P}_{\nu}|=q/2 ∣ P ν ∣ = q /2 . The last assertion is then the identification N ( Y a 0 ) = N ( X ) N(Y_{a_0})=N(X) N ( Y a 0 ) = N ( X ) of Corollary 4.13 .
Lemma 4.15 (uniform partner counts). Let q = 2 f q = 2^f q = 2 f with f ≥ 2 f \ge2 f ≥ 2 , let C \mathcal{C} C be one of the ( q − 2 ) / 2 (q - 2)/2 ( q − 2 ) /2 classes of unordered pairs of points of P \mathcal{P} P (Proposition 4.34 ), and let X ∈ P X \in\mathcal{P} X ∈ P . Then exactly q + 1 q + 1 q + 1 points Y ≠ X Y \ne X Y = X of P \mathcal{P} P satisfy { X , Y } ∈ C \{X,Y\} \in\mathcal{C} { X , Y } ∈ C .
Proof. For a class C \mathcal{C} C put n C ( X ) = # { Y ≠ X : { X , Y } ∈ C } n_{\mathcal{C}}(X) = \#\{Y \ne X : \{X,Y\} \in\mathcal{C}\} n C ( X ) = # { Y = X : { X , Y } ∈ C } . Every g ∈ G g \in G g ∈ G maps X X X to g X gX g X and maps C \mathcal{C} C , being an orbit of G G G on unordered pairs, onto itself; hence n C ( g X ) = n C ( X ) n_{\mathcal{C}}(gX) = n_{\mathcal{C}}(X) n C ( g X ) = n C ( X ) , and since G G G is transitive on P \mathcal{P} P the number n C n_{\mathcal{C}} n C does not depend on X X X . Counting the pairs of C \mathcal{C} C from both of their ends gives ∣ P ∣ n C = 2 ∣ C ∣ |\mathcal{P}|n_{\mathcal{C}} = 2|\mathcal{C}| ∣ P ∣ n C = 2∣ C ∣ , and Proposition 4.34 gives ∣ P ∣ = q ( q − 1 ) / 2 |\mathcal{P}| = q(q - 1)/2 ∣ P ∣ = q ( q − 1 ) /2 and ∣ C ∣ = q ( q 2 − 1 ) / 4 |\mathcal{C}| = q(q^2 - 1)/4 ∣ C ∣ = q ( q 2 − 1 ) /4 , whence
n C = 2 ∣ C ∣ ∣ P ∣ = q ( q 2 − 1 ) q ( q − 1 ) = q + 1. n_{\mathcal{C}} = \frac{2|\mathcal{C}|}{|\mathcal{P}|} = \frac{q(q^2 - 1)}{q(q - 1)} = q + 1. n C = ∣ P ∣ 2∣ C ∣ = q ( q − 1 ) q ( q 2 − 1 ) = q + 1. Thus the counts n C n_{\mathcal{C}} n C are the same for all classes, so that every linear functional of the class counts of a base block is determined by k k k alone and can neither confirm nor exclude a design. □
Proposition 4.16 (norm fibres meet the classes bijectively). Let q = 2 f q = 2^f q = 2 f with f ≥ 2 f \ge2 f ≥ 2 , let ν ∈ F q \nu\in\mathbb{F}_q ν ∈ F q , and let P ν = { Y ∈ P : N ( Y ) = ν } \mathcal{P}_\nu= \{Y \in\mathcal{P} : N(Y) = \nu\} P ν = { Y ∈ P : N ( Y ) = ν } as in Lemma 4.14. Since q − 1 q - 1 q − 1 is odd, ν \nu ν has a unique square root η ∈ F q \eta\in\mathbb{F}_q η ∈ F q , and every point of P ν \mathcal{P}_\nu P ν is { η e , η / e } \{\eta e,\eta/e\} { η e , η / e } for a unique class { e , 1 / e } \{e,1/e\} { e , 1/ e } of elements e ∈ F q 2 e \in\mathbb{F}_{q^2} e ∈ F q 2 with e q + 1 = 1 e^{q+1} = 1 e q + 1 = 1 and e ≠ 1 e \ne1 e = 1 . For X = { η e , η / e } X = \{\eta e,\eta/e\} X = { η e , η / e } and Y = { η f , η / f } Y = \{\eta f,\eta/f\} Y = { η f , η / f } in P ν \mathcal{P}_\nu P ν ,
cr ( X , Y ) = ( e + f 1 + e f ) 2 . \operatorname{cr}(X,Y) = \left(\frac{e+f}{1+ef}\right)^2. cr ( X , Y ) = ( 1 + e f e + f ) 2 . Consequently the q / 2 − 1 q/2 - 1 q /2 − 1 points of P ν ∖ { X } \mathcal{P}_\nu\setminus\{X\} P ν ∖ { X } represent the ( q − 2 ) / 2 (q - 2)/2 ( q − 2 ) /2 classes exactly once each. Moreover, for every class C \mathcal{C} C the map X ↦ F C ( X ) X \mapsto F_{\mathcal{C}}(X) X ↦ F C ( X ) , where F C ( X ) F_{\mathcal{C}}(X) F C ( X ) denotes the point of P ν \mathcal{P}_\nu P ν with { X , F C ( X ) } ∈ C \{X,F_{\mathcal{C}}(X)\} \in\mathcal{C} { X , F C ( X )} ∈ C , is a fixed-point-free involution of P ν \mathcal{P}_\nu P ν ; hence C \mathcal{C} C contains exactly q / 4 q/4 q /4 pairs inside P ν \mathcal{P}_\nu P ν , exactly q 2 ( q − 1 ) / 4 q^2(q - 1)/4 q 2 ( q − 1 ) /4 pairs whose two points have equal norm, and exactly q 2 ( q − 1 ) / 4 q^2(q - 1)/4 q 2 ( q − 1 ) /4 pairs whose two points have different norms.
Proof. Write Y = { y , y q } Y = \{y,y^q\} Y = { y , y q } with y y q = ν yy^q = \nu y y q = ν and put e = y / η e = y/\eta e = y / η ; then e q + 1 = e e q = y y q / η 2 = 1 e^{q+1} = ee^q = yy^q/\eta^2 = 1 e q + 1 = e e q = y y q / η 2 = 1 , and the two representatives y y y and y q y^q y q of Y Y Y correspond to e e e and 1 / e 1/e 1/ e . The element e = 1 e = 1 e = 1 gives y = η ∈ F q y = \eta\in\mathbb{F}_q y = η ∈ F q , which is not a point of P \mathcal{P} P , so the q q q remaining elements e e e give the q / 2 q/2 q /2 points of P ν \mathcal{P}_\nu P ν , in agreement with Lemma 4.14.
For the displayed value, the four factors of the cross-ratio are
x + y = η ( e + f ) , x q + y q = η ( e ˉ + f ˉ ) , x + y q = η ( e + f ˉ ) , x q + y = η ( e ˉ + f ) , x+y=\eta(e+f),\quad x^{q}+y^{q}=\eta(\bar{e}+\bar{f}),\quad x+y^{q}=\eta(e+\bar{f}),\quad x^{q}+y=\eta(\bar{e}+f), x + y = η ( e + f ) , x q + y q = η ( e ˉ + f ˉ ) , x + y q = η ( e + f ˉ ) , x q + y = η ( e ˉ + f ) ,
with e ˉ = e q = 1 / e \bar{e}=e^{q}=1/e e ˉ = e q = 1/ e ; the common factor η 2 = ν \eta^{2}=\nu η 2 = ν cancels, and since ( e + f ) ( 1 / e + 1 / f ) = ( e + f ) 2 / ( e f ) (e+f)(1/e+1/f)=(e+f)^{2}/(ef) ( e + f ) ( 1/ e + 1/ f ) = ( e + f ) 2 / ( e f ) while ( e + 1 / f ) ( 1 / e + f ) = ( 1 + e f ) 2 / ( e f ) (e+1/f)(1/e+f)=(1+ef)^{2}/(ef) ( e + 1/ f ) ( 1/ e + f ) = ( 1 + e f ) 2 / ( e f ) , the common factor 1 / ( e f ) 1/(ef) 1/ ( e f ) cancels as well and
cr ( X , Y ) = ( e + f ) 2 ( 1 + e f ) 2 = ( e + f 1 + e f ) 2 . \operatorname{cr}(X,Y)=\frac{(e+f)^{2}}{(1+ef)^{2}}=\left(\frac{e+f}{1+ef}\right)^{2}. cr ( X , Y ) = ( 1 + e f ) 2 ( e + f ) 2 = ( 1 + e f e + f ) 2 .
Now f ↦ ( e + f ) / ( 1 + e f ) f\mapsto(e+f)/(1+ef) f ↦ ( e + f ) / ( 1 + e f ) is induced by the matrix ( 1 e e 1 ) \begin{pmatrix}1&e\\ e&1\end{pmatrix} ( 1 e e 1 ) , whose determinant 1 + e 2 = ( 1 + e ) 2 1+e^{2}=(1+e)^{2} 1 + e 2 = ( 1 + e ) 2 is non-zero because e ≠ 1 e\ne1 e = 1 ; the map is therefore injective on F q ∗ ∪ { ∞ } \mathbb{F}_{q}^{\ast}\cup\{\infty\} F q ∗ ∪ { ∞ } , and 1 + e f ≠ 0 1+ef\ne0 1 + e f = 0 whenever Y ≠ X Y\ne X Y = X . Replacing f f f by 1 / f 1/f 1/ f inverts ( e + f ) / ( 1 + e f ) (e+f)/(1+ef) ( e + f ) / ( 1 + e f ) , and a class carries the two values ρ \rho ρ and ρ − 1 \rho^{-1} ρ − 1 only, so two points Y , Y ′ Y,Y' Y , Y ′ of P ν ∖ { X } \mathcal{P}_{\nu}\setminus\{X\} P ν ∖ { X } lie in one class exactly when f ′ = f f'=f f ′ = f or f ′ = 1 / f f'=1/f f ′ = 1/ f , that is exactly when Y ′ = Y Y'=Y Y ′ = Y . As there are q / 2 − 1 = ( q − 2 ) / 2 q/2-1=(q-2)/2 q /2 − 1 = ( q − 2 ) /2 such points and the same number of classes, the correspondence between points and classes is a bijection.
Finally fix ν \nu ν and C C C . By what precedes, each X ∈ P ν X\in\mathcal{P}_{\nu} X ∈ P ν is paired by C C C with exactly one other point of P ν \mathcal{P}_{\nu} P ν , so F C F_{C} F C is a fixed-point-free involution pairing the q / 2 q/2 q /2 points of P ν \mathcal{P}_{\nu} P ν into q / 4 q/4 q /4 pairs. Summing over the q − 1 q-1 q − 1 values of ν \nu ν gives q ( q − 1 ) / 4 q(q-1)/4 q ( q − 1 ) /4 pairs of C C C with equal norms; and C C C has q ( q 2 − 1 ) / 4 q(q^{2}-1)/4 q ( q 2 − 1 ) /4 pairs in all, so q ( q 2 − 1 ) / 4 − q ( q − 1 ) / 4 = q 2 ( q − 1 ) / 4 q(q^{2}-1)/4-q(q-1)/4=q^{2}(q-1)/4 q ( q 2 − 1 ) /4 − q ( q − 1 ) /4 = q 2 ( q − 1 ) /4 of them join points of different norms.
Corollary 4.17 (orbit matchings). Let q = 2 f q=2^{f} q = 2 f with f ≥ 2 f\ge2 f ≥ 2 , let O \mathcal{O} O be the set of q / 2 q/2 q /2 orbits of M = D 2 ( q − 1 ) M=D_{2(q-1)} M = D 2 ( q − 1 ) on P \mathcal{P} P , and for distinct O , O ′ ∈ O \mathcal{O},\mathcal{O}'\in\mathcal{O} O , O ′ ∈ O let { X , X ′ } = π ( O , O ′ ) \{X,X'\}=\pi(\mathcal{O},\mathcal{O}') { X , X ′ } = π ( O , O ′ ) , where X ∈ O X\in\mathcal{O} X ∈ O and X ′ X' X ′ is the unique point of O ′ \mathcal{O}' O ′ with N ( X ′ ) = N ( X ) N(X')=N(X) N ( X ′ ) = N ( X ) . Then π ( O , O ′ ) \pi(\mathcal{O},\mathcal{O}') π ( O , O ′ ) lies in a class that does not depend on the choice of X ∈ O X\in\mathcal{O} X ∈ O ; for every class C C C the pairs π ( O , O ′ ) \pi(\mathcal{O},\mathcal{O}') π ( O , O ′ ) of class C C C form a perfect matching of O \mathcal{O} O ; each class contains exactly q / 4 q/4 q /4 pairs of orbits; and the ( q − 2 ) / 2 (q-2)/2 ( q − 2 ) /2 matchings decompose the complete graph on O \mathcal{O} O .
Proof. By Lemma 4.14 the orbit O ′ \mathcal{O}' O ′ meets the fibre P N ( X ) \mathcal{P}_{N(X)} P N ( X ) in exactly one point, so X ′ X' X ′ exists and is unique. If X X X is replaced by λ X \lambda X λ X with λ ∈ F q ∗ \lambda\in\mathbb{F}_{q}^{\ast} λ ∈ F q ∗ , then λ X ′ ∈ O ′ \lambda X'\in\mathcal{O}' λ X ′ ∈ O ′ satisfies N ( λ X ′ ) = λ 2 N ( X ′ ) = λ 2 N ( X ) = N ( λ X ) N(\lambda X')=\lambda^{2}N(X')=\lambda^{2}N(X)=N(\lambda X) N ( λ X ′ ) = λ 2 N ( X ′ ) = λ 2 N ( X ) = N ( λ X ) , so λ X ′ \lambda X' λ X ′ is the norm-matching point of λ X \lambda X λ X and π ( O , O ′ ) = { X , X ′ } \pi(\mathcal{O},\mathcal{O}')=\{X,X'\} π ( O , O ′ ) = { X , X ′ } is merely scaled; the class is therefore independent of X ∈ O X\in\mathcal{O} X ∈ O , because the map y ↦ λ y y\mapsto\lambda y y ↦ λ y lies in M ≤ G M\le G M ≤ G and G G G preserves classes.
Fix O O O and C C C . The point X ∈ O X \in O X ∈ O has, by Proposition 4.16 , exactly one point of class C C C in the whole fibre P N ( X ) \mathcal{P}_{N(X)} P N ( X ) , and by Lemma 4.14 that point lies in exactly one of the q / 2 − 1 q/2-1 q /2 − 1 orbits other than O O O . Hence O O O is incident with exactly one class-C C C pair of orbits, the graph of class C C C on O O O is 1-regular and therefore a perfect matching with q / 4 q/4 q /4 edges. Summing over the ( q − 2 ) / 2 (q-2)/2 ( q − 2 ) /2 classes gives q ( q − 2 ) / 8 = ( q / 2 2 ) q(q-2)/8 = \binom{q/2}{2} q ( q − 2 ) /8 = ( 2 q /2 ) pairs of orbits, that is all of them, so the matchings decompose the complete graph on O O O .
Lemma 4.18 (fibre rigidity). Let q = 2 f q=2^f q = 2 f with f ≥ 2 f \ge2 f ≥ 2 , and let B B B be a union of a a a orbits of M = D 2 ( q − 1 ) M=D_{2(q-1)} M = D 2 ( q − 1 ) , so that ∣ B ∣ = a ( q − 1 ) |B|=a(q-1) ∣ B ∣ = a ( q − 1 ) . For ν ∈ F q × \nu\in\mathbb{F}_q^\times ν ∈ F q × put S ν = B ∩ P ν S_\nu=B\cap\mathcal{P}_\nu S ν = B ∩ P ν , and for a class C C C let m C ( S ν ) m_C(S_\nu) m C ( S ν ) be the number of pairs of S ν S_\nu S ν that lie in C C C . Then ∣ S ν ∣ = a |S_\nu|=a ∣ S ν ∣ = a , and m C ( S ν ) m_C(S_\nu) m C ( S ν ) does not depend on ν \nu ν .
Proof. The group M M M contains every scalar map σ λ \sigma_\lambda σ λ with σ λ ( Y ) = λ Y \sigma_\lambda(Y)=\lambda Y σ λ ( Y ) = λY and λ ∈ F q × \lambda\in\mathbb{F}_q^\times λ ∈ F q × , and B B B is a union of orbits of M M M ; hence σ λ ( B ) = B \sigma_\lambda(B)=B σ λ ( B ) = B for all λ \lambda λ . Since N ( λ Y ) = λ 2 N ( Y ) N(\lambda Y)=\lambda^2N(Y) N ( λY ) = λ 2 N ( Y ) , the map σ λ \sigma_\lambda σ λ sends P μ \mathcal{P}_\mu P μ onto P λ 2 μ \mathcal{P}_{\lambda^2\mu} P λ 2 μ . As q − 1 q-1 q − 1 is odd, λ ↦ λ 2 \lambda\mapsto\lambda^2 λ ↦ λ 2 is a bijection of F q × \mathbb{F}_q^\times F q × (Lemma 4.14), so each ν \nu ν has exactly one square root λ \lambda λ and S ν = σ λ ( S 1 ) S_\nu=\sigma_\lambda(S_1) S ν = σ λ ( S 1 ) . Finally all four terms of the cross-ratio are multiplied by λ \lambda λ , so cr ( λ X , λ Y ) = cr ( X , Y ) \operatorname{cr}(\lambda X,\lambda Y)=\operatorname{cr}(X,Y) cr ( λ X , λY ) = cr ( X , Y ) and σ λ \sigma_\lambda σ λ preserves the classes; hence m C ( S ν ) = m C ( S 1 ) m_C(S_\nu)=m_C(S_1) m C ( S ν ) = m C ( S 1 ) .
Corollary 4.19 (fibre and far decomposition). In the situation of Lemma 4.18 , let A C ( B ) A_C(B) A C ( B ) be the number of pairs of B B B in the class C C C , and let A C f a r ( B ) A_C^{\mathrm{far}}(B) A C far ( B ) be the number of those whose two points lie in distinct fibres. Put m C = m C ( S 1 ) m_C=m_C(S_1) m C = m C ( S 1 ) . Then
A C ( B ) = ( q − 1 ) m C + A C f a r ( B ) , ∑ C m C = ( a 2 ) , A_C(B)=(q-1)m_C+A_C^{\mathrm{far}}(B), \qquad\sum_C m_C=\binom{a}{2}, A C ( B ) = ( q − 1 ) m C + A C far ( B ) , C ∑ m C = ( 2 a ) , and ∑ C A C f a r ( B ) = a 2 ( q − 1 ) ( q − 2 ) / 2 \sum_C A_C^{\mathrm{far}}(B)=a^2(q-1)(q-2)/2 ∑ C A C far ( B ) = a 2 ( q − 1 ) ( q − 2 ) /2 . Thus every class count of B B B is the sum of a term carried by the single fibre P 1 \mathcal{P}_1 P 1 and of a far term; the fibre term is the same for all fibres and for all blocks inducing the same a a a -subset of P 1 \mathcal{P}_1 P 1 .
Proof. By Lemma 4.18 each of the q − 1 q-1 q − 1 fibres P ν \mathcal{P}_\nu P ν contains exactly m C m_C m C pairs of B B B of class C C C , which gives the first summand of the first identity, while the remaining pairs of B B B have their two ends in distinct fibres. Summing that identity over the ( q − 2 ) / 2 (q-2)/2 ( q − 2 ) /2 classes and using ∑ C m C = ( a 2 ) \sum_C m_C=\binom{a}{2} ∑ C m C = ( 2 a ) together with ∑ C A C ( B ) = ( k 2 ) \sum_C A_C(B)=\binom{k}{2} ∑ C A C ( B ) = ( 2 k ) , k = a ( q − 1 ) k=a(q-1) k = a ( q − 1 ) , gives
∑ C A C f a r ( B ) = a ( q − 1 ) ( a ( q − 1 ) − 1 ) 2 − ( q − 1 ) a ( a − 1 ) 2 = a 2 ( q − 1 ) ( q − 2 ) 2 . \sum_C A_C^{\mathrm{far}}(B)=\frac{a(q-1)(a(q-1)-1)}{2}-\frac{(q-1)a(a-1)}{2}=\frac{a^2(q-1)(q-2)}{2}. C ∑ A C far ( B ) = 2 a ( q − 1 ) ( a ( q − 1 ) − 1 ) − 2 ( q − 1 ) a ( a − 1 ) = 2 a 2 ( q − 1 ) ( q − 2 ) .
Corollary 4.20 (the far term as a scalar sum). In the situation of Corollary 4.19, put S = B ∩ P 1 S=B\cap P_1 S = B ∩ P 1 and, for δ ∈ F q × ∖ { 1 } \delta\in F_q^\times\setminus\{1\} δ ∈ F q × ∖ { 1 } and a class C C C ,
n C ( S , δ S ) = # { { X , δ Y } : X , Y ∈ S , { X , δ Y } ∈ C } , n_C(S,\delta S)=\#\{\{X,\delta Y\}:X,Y\in S,\ \{X,\delta Y\}\in C\}, n C ( S , δ S ) = # {{ X , δ Y } : X , Y ∈ S , { X , δ Y } ∈ C } , the number of pairs of class C C C with one end in S ⊂ P 1 S\subset P_1 S ⊂ P 1 and the other in δ S ⊂ P δ 2 \delta S\subset P_{\delta2} δ S ⊂ P δ 2 . Then
A C f a r ( B ) = q − 1 2 ∑ δ ≠ 1 n C ( S , δ S ) = q − 1 2 ( λ − 2 m C ) , A_C^{\mathrm{far}}(B)=\frac{q-1}{2}\sum_{\delta\ne1}n_C(S,\delta S)=\frac{q-1}{2}(\lambda-2m_C), A C far ( B ) = 2 q − 1 δ = 1 ∑ n C ( S , δ S ) = 2 q − 1 ( λ − 2 m C ) , where λ = 2 a ( a ( q − 1 ) − 1 ) / ( q − 2 ) \lambda=2a\bigl(a(q-1)-1\bigr)/(q-2) λ = 2 a ( a ( q − 1 ) − 1 ) / ( q − 2 ) is the parameter of the putative design, whose number of blocks and replication number are q ( q + 1 ) / 2 q(q+1)/2 q ( q + 1 ) /2 and a ( q + 1 ) a(q+1) a ( q + 1 ) . Consequently B G B^G B G is the block set of a 2-design if and only if
∑ δ ≠ 1 n C ( S , δ S ) = λ − 2 m C for every class C , \sum_{\delta\ne1}n_C(S,\delta S)=\lambda-2m_C\qquad\text{for every class }C, δ = 1 ∑ n C ( S , δ S ) = λ − 2 m C for every class C , that is, if and only if a single fibre S S S equidistributes the classes of the q − 1 q-1 q − 1 scalar lines through its points.
Proof. Fix δ ≠ 1 \delta\ne1 δ = 1 and let ν \nu ν run over F q × F_q^\times F q × . Since δ 2 ≠ 1 \delta^2\ne1 δ 2 = 1 in characteristic 2, the map ν ↦ δ 2 ν \nu\mapsto\delta^2\nu ν ↦ δ 2 ν has no fixed point, and its cycles have odd length, so the q − 1 q-1 q − 1 sets { ν , δ 2 ν } \{\nu,\delta^2\nu\} { ν , δ 2 ν } are the q − 1 q-1 q − 1 edges of these cycles; a pair { μ , μ ′ } \{\mu,\mu'\} { μ , μ ′ } of distinct elements of F q × F_q^\times F q × occurs among them exactly when μ ′ μ − 1 ∈ { δ , δ − 1 } \mu'\mu^{-1}\in\{\delta,\delta^{-1}\} μ ′ μ − 1 ∈ { δ , δ − 1 } . By Lemma 4.18 each of these q − 1 q-1 q − 1 unordered pairs of fibres contributes the same number n C ( S , δ S ) n_C(S,\delta S) n C ( S , δ S ) of class-C C C pairs, so the value of δ \delta δ alone contributes ( q − 1 ) n C ( S , δ S ) (q-1)n_C(S,\delta S) ( q − 1 ) n C ( S , δ S ) . Summing over the q − 2 q-2 q − 2 values of δ ≠ 1 \delta\ne1 δ = 1 counts each unordered pair of distinct fibres exactly twice, because δ \delta δ and δ − 1 \delta^{-1} δ − 1 give the same pair while δ ≠ δ − 1 \delta\ne\delta^{-1} δ = δ − 1 ; hence the first identity. For the second, add ( q − 1 ) m C (q-1)m_C ( q − 1 ) m C and use Corollary 4.19:
A C ( B ) = q − 1 2 ( 2 m C + ∑ δ ≠ 1 n C ( S , δ S ) ) = ( q − 1 ) λ 2 , A_C(B)=\frac{q-1}{2}\left(2m_C+\sum_{\delta\ne1}n_C(S,\delta S)\right)=\frac{(q-1)\lambda}{2}, A C ( B ) = 2 q − 1 2 m C + δ = 1 ∑ n C ( S , δ S ) = 2 ( q − 1 ) λ ,
the required value ( k 2 ) / s = k ( k − 1 ) / ( q − 2 ) \binom{k}{2}/s=k(k-1)/(q-2) ( 2 k ) / s = k ( k − 1 ) / ( q − 2 ) of a design with k = a ( q − 1 ) k=a(q-1) k = a ( q − 1 ) and s = ( q − 2 ) / 2 s=(q-2)/2 s = ( q − 2 ) /2 , since ( k 2 ) / s = a ( q − 1 ) ( a ( q − 1 ) − 1 ) / ( q − 2 ) = ( q − 1 ) λ / 2 \binom{k}{2}/s=a(q-1)\bigl(a(q-1)-1\bigr)/(q-2)=(q-1)\lambda/2 ( 2 k ) / s = a ( q − 1 ) ( a ( q − 1 ) − 1 ) / ( q − 2 ) = ( q − 1 ) λ /2 . Thus all class counts are constant and equal to ( k 2 ) / s \binom{k}{2}/s ( 2 k ) / s exactly when the displayed identity holds for every C C C , which is the criterion for B G B^G B G to be a 2-design.
Lemma 4.21 (label duality). Let q = 2 f q=2^{f} q = 2 f with f ≥ 2 f\ge2 f ≥ 2 , let X ≠ Y X\ne Y X = Y be points of P 1 \mathcal{P}_{1} P 1 , and let δ , δ ′ ∈ F q × ∖ { 1 } \delta,\delta'\in F_{q}^{\times}\setminus\{1\} δ , δ ′ ∈ F q × ∖ { 1 } . Then
cr ( X , δ Y ) = cr ( X , δ ′ Y ) ⟺ δ ′ = δ or δ δ ′ = 1. \operatorname{cr}(X,\delta Y)=\operatorname{cr}(X,\delta'Y)\Longleftrightarrow\delta'=\delta\ \text{ or }\ \delta\delta'=1. cr ( X , δ Y ) = cr ( X , δ ′ Y ) ⟺ δ ′ = δ or δ δ ′ = 1. Consequently the q − 2 q-2 q − 2 values cr ( X , δ Y ) \operatorname{cr}(X,\delta Y) cr ( X , δ Y ) with δ ≠ 1 \delta\ne1 δ = 1 consist of exactly ( q − 2 ) / 2 (q-2)/2 ( q − 2 ) /2 distinct values, each taken at the two elements δ \delta δ and δ − 1 \delta^{-1} δ − 1 .
Proof. Regard X , Y X,Y X , Y as elements of F q 2 F_{q^{2}} F q 2 and write N ( Z ) = Z Z q N(Z)=ZZ^{q} N ( Z ) = Z Z q ; since X , Y ∈ P 1 X,Y\in\mathcal{P}_{1} X , Y ∈ P 1 we have N ( X ) = N ( Y ) = 1 N(X)=N(Y)=1 N ( X ) = N ( Y ) = 1 . For δ ∈ F q × \delta\in F_{q}^{\times} δ ∈ F q × the four points x , δ y , x q , δ y q x,\delta y,x^{q},\delta y^{q} x , δ y , x q , δ y q give, with T = x y q + x q y T=xy^{q}+x^{q}y T = x y q + x q y and S = x y + x q y q S=xy+x^{q}y^{q} S = x y + x q y q ,
cr ( X , δ Y ) = ( x + δ y ) ( x q + δ y q ) ( x + δ y q ) ( x q + δ y ) = 1 + T δ + δ 2 1 + S δ + δ 2 . \operatorname{cr}(X,\delta Y)=\frac{(x+\delta y)(x^{q}+\delta y^{q})}{(x+\delta y^{q})(x^{q}+\delta y)}=\frac{1+T\delta+\delta^{2}}{1+S\delta+\delta^{2}}. cr ( X , δ Y ) = ( x + δ y q ) ( x q + δ y ) ( x + δ y ) ( x q + δ y q ) = 1 + S δ + δ 2 1 + T δ + δ 2 . Neither denominator vanishes: if x q = δ y x^{q}=\delta y x q = δ y , then 1 = N ( X ) = x x q = δ 2 y y q = δ 2 1=N(X)=xx^{q}=\delta^{2}yy^{q}=\delta^{2} 1 = N ( X ) = x x q = δ 2 y y q = δ 2 , hence δ = 1 \delta=1 δ = 1 , and the factor x + δ y q x+\delta y^{q} x + δ y q is excluded by the same argument. An expansion gives
( 1 + T δ + δ 2 ) ( 1 + S δ ′ + δ ′ 2 ) − ( 1 + T δ ′ + δ ′ 2 ) ( 1 + S δ + δ 2 ) = ( δ − δ ′ ) ( T − S ) ( 1 − δ δ ′ ) . (1+T\delta+\delta^{2})(1+S\delta'+\delta'^{2})-(1+T\delta'+\delta'^{2})(1+S\delta+\delta^{2})=(\delta-\delta')(T-S)(1-\delta\delta'). ( 1 + T δ + δ 2 ) ( 1 + S δ ′ + δ ′ 2 ) − ( 1 + T δ ′ + δ ′ 2 ) ( 1 + S δ + δ 2 ) = ( δ − δ ′ ) ( T − S ) ( 1 − δ δ ′ ) . In characteristic 2 one has T − S = T + S = ( x + x q ) ( y + y q ) T-S=T+S=(x+x^{q})(y+y^{q}) T − S = T + S = ( x + x q ) ( y + y q ) , and both factors are nonzero because the second coordinate of x + x q x+x^{q} x + x q is the nonzero second coordinate of X X X , and likewise for Y Y Y . Hence the difference vanishes exactly when δ = δ ′ \delta=\delta' δ = δ ′ or δ δ ′ = 1 \delta\delta'=1 δ δ ′ = 1 , and the first assertion follows. The second follows because the only element of F q × F_{q}^{\times} F q × equal to its inverse is 1: the set F q × ∖ { 1 } F_{q}^{\times}\setminus\{1\} F q × ∖ { 1 } splits into the ( q − 2 ) / 2 (q-2)/2 ( q − 2 ) /2 pairs { δ , δ − 1 } \{\delta,\delta^{-1}\} { δ , δ − 1 } , and by the first assertion distinct pairs give distinct values.
Proposition 4.22 (balanced class multiplicities). In the situation of Lemma 4.21 put s = ( q − 2 ) / 2 s=(q-2)/2 s = ( q − 2 ) /2 , and for a class C C C with representative ρ \rho ρ let w C ( X , Y ) w_C(X,Y) w C ( X , Y ) be half the number of δ ∈ F q × ∖ { 1 } \delta\in F_{q}^{\times}\setminus\{1\} δ ∈ F q × ∖ { 1 } with cr ( X , δ Y ) ∈ { ρ , ρ − 1 } \operatorname{cr}(X,\delta Y)\in\{\rho,\rho^{-1}\} cr ( X , δ Y ) ∈ { ρ , ρ − 1 } . Then
w C ( X , Y ) ∈ { 0 , 1 , 2 } , ∑ C w C ( X , Y ) = s , w_C(X,Y)\in\{0,1,2\},\qquad\sum_C w_C(X,Y)=s, w C ( X , Y ) ∈ { 0 , 1 , 2 } , C ∑ w C ( X , Y ) = s , and the classes that are doubled are as numerous as those that are missed:
# { C : w C ( X , Y ) = 0 } = # { C : w C ( X , Y ) = 2 } . \#\{C:w_C(X,Y)=0\}=\#\{C:w_C(X,Y)=2\}. # { C : w C ( X , Y ) = 0 } = # { C : w C ( X , Y ) = 2 } .
Proof. By Lemma 4.21 the q − 2 q-2 q − 2 values of δ ≠ 1 \delta\ne1 δ = 1 are the values of the ( q − 2 ) / 2 (q-2)/2 ( q − 2 ) /2 pairs { δ , δ − 1 } \{\delta,\delta^{-1}\} { δ , δ − 1 } , so w C ( X , Y ) w_C(X,Y) w C ( X , Y ) is the number of these pairs whose common value lies in { ρ , ρ − 1 } \{\rho,\rho^{-1}\} { ρ , ρ − 1 } . A class contains the two values ρ \rho ρ and ρ − 1 \rho^{-1} ρ − 1 and, by the same lemma, all ( q − 2 ) / 2 (q-2)/2 ( q − 2 ) /2 values are distinct; hence at most two of them lie in { ρ , ρ − 1 } \{\rho,\rho^{-1}\} { ρ , ρ − 1 } , that is w C ( X , Y ) ≤ 2 w_C(X,Y) \le2 w C ( X , Y ) ≤ 2 . Every value lies in exactly one class, which gives the sum s s s . As w C ( X , Y ) w_C(X,Y) w C ( X , Y ) takes only the values 0 , 1 , 2 0,1,2 0 , 1 , 2 , the sum
∑ C ( w C ( X , Y ) − 1 ) = 0 \sum_C (w_C(X,Y)-1)=0 C ∑ ( w C ( X , Y ) − 1 ) = 0
is a difference of the number of classes counted once and the number of classes counted twice, and the last assertion follows. □
Corollary 4.23. Let S ⊂ P 1 S \subset\mathcal{P}_1 S ⊂ P 1 with ∣ S ∣ = a |S|=a ∣ S ∣ = a , let X ≠ Y X \ne Y X = Y run over the ordered pairs of S S S , and put L C = ∑ X ≠ Y w C ( X , Y ) L_C=\sum_{X\ne Y}w_C(X,Y) L C = ∑ X = Y w C ( X , Y ) and s = ( q − 2 ) / 2 s=(q-2)/2 s = ( q − 2 ) /2 . Then
∑ C L C = a ( a − 1 ) s \sum_C L_C=a(a-1)s C ∑ L C = a ( a − 1 ) s for every S S S . Thus the total of the class counts is fixed in advance and independent of S S S : a failure of the class equations of Corollary 4.20 is necessarily a redistribution among the classes with total zero, and no argument that uses only the sum of the class counts can exclude an a a a -subset.
Proof. Sum the identity ∑ C w C ( X , Y ) = s \sum_C w_C(X,Y)=s ∑ C w C ( X , Y ) = s of Proposition 4.22 over the a ( a − 1 ) a(a-1) a ( a − 1 ) ordered pairs of distinct points of S S S . □
Lemma 4.25. Let q = 2 f q=2^f q = 2 f with f ≥ 2 f\ge2 f ≥ 2 , let κ = κ ( O ) ≠ 0 \kappa=\kappa(O)\ne0 κ = κ ( O ) = 0 be as in Lemma 4.10 , and let
T q = { t + 1 / t : t ∈ F q × } = { 0 } ∪ { τ ≠ 0 : Tr ( 1 / τ ) = 0 } , T_q=\{t+1/t:t\in F_q^{\times}\}=\{0\}\cup\{\tau\ne0:\operatorname{Tr}(1/\tau)=0\}, T q = { t + 1/ t : t ∈ F q × } = { 0 } ∪ { τ = 0 : Tr ( 1/ τ ) = 0 } ,
where Tr \operatorname{Tr} Tr is the trace of F q F_q F q over F 2 F_2 F 2 . Then
# { τ ∈ T q ∖ { 0 } : τ + κ ∈ T q ∖ { 0 } } ≥ 1 8 ( q − 2 q − 7 ) . \#\{\tau\in T_q \setminus\{0\} : \tau+ \kappa\in T_q \setminus\{0\}\} \ge\frac{1}{8}(q - 2\sqrt{q} - 7). # { τ ∈ T q ∖ { 0 } : τ + κ ∈ T q ∖ { 0 }} ≥ 8 1 ( q − 2 q − 7 ) .
which is at least 1 for every even q ≥ 16 q \ge16 q ≥ 16 .
Proof. Put H = { x ∈ F q : Tr ( x ) = 0 } H = \{x \in F_q : \operatorname{Tr}(x) = 0\} H = { x ∈ F q : Tr ( x ) = 0 } , of order q / 2 q/2 q /2 , and let x = 1 / τ x = 1/\tau x = 1/ τ , y = 1 / ( τ + κ ) y = 1/(\tau+ \kappa) y = 1/ ( τ + κ ) ; then x , y ∈ H ∖ { 0 } x,y \in H \setminus\{0\} x , y ∈ H ∖ { 0 } and κ = 1 / x + 1 / y \kappa= 1/x + 1/y κ = 1/ x + 1/ y , that is y = x / ( 1 + κ x ) y = x/(1+\kappa x) y = x / ( 1 + κ x ) . Conversely such a pair ( x , y ) (x,y) ( x , y ) of elements of H ∖ { 0 } H \setminus\{0\} H ∖ { 0 } with κ = 1 / x + 1 / y \kappa= 1/x + 1/y κ = 1/ x + 1/ y gives τ = 1 / x \tau= 1/x τ = 1/ x with τ + κ = 1 / y \tau+ \kappa= 1/y τ + κ = 1/ y in T q ∖ { 0 } T_q \setminus\{0\} T q ∖ { 0 } . The map x ↦ y x \mapsto y x ↦ y is an involution of the set of admissible x x x , and it has no fixed point because x = y x = y x = y would give κ = 1 / x + 1 / x = 0 \kappa= 1/x + 1/x = 0 κ = 1/ x + 1/ x = 0 . Hence the number in the statement equals Sol ( κ ) / 2 \operatorname{Sol}(\kappa)/2 Sol ( κ ) /2 , where
Sol ( κ ) = # { x ∈ F q : Tr ( x ) = 0 , Tr ( x / ( 1 + κ x ) ) = 0 , x ≠ 0 , x ≠ 1 / κ } . \operatorname{Sol}(\kappa) = \#\{x \in F_q : \operatorname{Tr}(x) = 0,\ \operatorname{Tr}(x/(1+\kappa x)) = 0,\ x \ne0,\ x \ne1/\kappa\}. Sol ( κ ) = # { x ∈ F q : Tr ( x ) = 0 , Tr ( x / ( 1 + κ x )) = 0 , x = 0 , x = 1/ κ } .
Extend f ( x ) = x / ( 1 + κ x ) f(x) = x/(1+\kappa x) f ( x ) = x / ( 1 + κ x ) to all of F q F_q F q by f ( 1 / κ ) = 0 f(1/\kappa) = 0 f ( 1/ κ ) = 0 . Since 1 Tr ( u ) = 0 = 1 2 ( 1 + ( − 1 ) Tr ( u ) ) \mathbf{1}_{\operatorname{Tr}(u)=0} = \frac{1}{2}(1+(-1)^{\operatorname{Tr}(u)}) 1 Tr ( u ) = 0 = 2 1 ( 1 + ( − 1 ) Tr ( u ) ) ,
∑ x ∈ F q 1 Tr ( x ) = 0 1 Tr ( f ( x ) ) = 0 = 1 4 ( q + ∑ x ( − 1 ) Tr ( x ) + ∑ x ( − 1 ) Tr ( f ( x ) ) ) + 1 4 ∑ x ( − 1 ) Tr ( x + f ( x ) ) . \sum_{x\in F_q}\mathbf{1}_{\operatorname{Tr}(x)=0}\mathbf{1}_{\operatorname{Tr}(f(x))=0}
= \frac{1}{4}\left(q+\sum_x(-1)^{\operatorname{Tr}(x)}+\sum_x(-1)^{\operatorname{Tr}(f(x))}\right)
+\frac{1}{4}\sum_x(-1)^{\operatorname{Tr}(x+f(x))}. x ∈ F q ∑ 1 Tr ( x ) = 0 1 Tr ( f ( x )) = 0 = 4 1 ( q + x ∑ ( − 1 ) Tr ( x ) + x ∑ ( − 1 ) Tr ( f ( x )) ) + 4 1 x ∑ ( − 1 ) Tr ( x + f ( x )) .
The first sum vanishes because Tr \operatorname{Tr} Tr is not the zero map. The map f f f is a bijection of F q F_q F q taking 1 / κ 1/\kappa 1/ κ to 0 0 0 and 0 0 0 to 1 / κ 1/\kappa 1/ κ , so the second sum equals 1 − ( − 1 ) Tr ( 1 / κ ) 1-(-1)^{\operatorname{Tr}(1/\kappa)} 1 − ( − 1 ) Tr ( 1/ κ ) . With w = 1 + κ x w=1+\kappa x w = 1 + κ x one has x + f ( x ) = κ x 2 / ( 1 + κ x ) = ( w + 1 / w ) / κ x+f(x)=\kappa x^2/(1+\kappa x)=(w+1/w)/\kappa x + f ( x ) = κ x 2 / ( 1 + κ x ) = ( w + 1/ w ) / κ , so the third sum equals
K ( κ ) + ( − 1 ) Tr ( 1 / κ ) , K ( κ ) = ∑ w ∈ F q ∗ ( − 1 ) Tr ( ( w + 1 / w ) / κ ) . K(\kappa)+(-1)^{\operatorname{Tr}(1/\kappa)},\qquad K(\kappa)=\sum_{w\in F_q^\ast}(-1)^{\operatorname{Tr}((w+1/w)/\kappa)}. K ( κ ) + ( − 1 ) Tr ( 1/ κ ) , K ( κ ) = w ∈ F q ∗ ∑ ( − 1 ) Tr (( w + 1/ w ) / κ ) .
Thus the total is 1 4 ( q + 1 + K ( κ ) ) \frac{1}{4}(q+1+K(\kappa)) 4 1 ( q + 1 + K ( κ )) . Removing x = 0 x=0 x = 0 , which contributes 1, and x = 1 / κ x=1/\kappa x = 1/ κ , which contributes 1 Tr ( 1 / κ ) = 0 \mathbf{1}_{\operatorname{Tr}(1/\kappa)=0} 1 Tr ( 1/ κ ) = 0 , gives
Sol ( κ ) = 1 4 ( q + 1 + K ( κ ) ) − 1 − 1 Tr ( 1 / κ ) = 0 ≥ 1 4 ( q − 2 q − 7 ) . \operatorname{Sol}(\kappa)=\frac{1}{4}(q+1+K(\kappa))-1-\mathbf{1}_{\operatorname{Tr}(1/\kappa)=0}\ge\frac{1}{4}(q-2\sqrt{q}-7). Sol ( κ ) = 4 1 ( q + 1 + K ( κ )) − 1 − 1 Tr ( 1/ κ ) = 0 ≥ 4 1 ( q − 2 q − 7 ) .
by the Weil bound ∣ K ( κ ) ∣ ≤ 2 q |K(\kappa)|\le2\sqrt{q} ∣ K ( κ ) ∣ ≤ 2 q for the Kloosterman sum [40 ] . For q ≥ 16 q\ge16 q ≥ 16 the right-hand side is positive, and Sol ( κ ) \operatorname{Sol}(\kappa) Sol ( κ ) is even, so it is at least 2 and the number in the statement is at least 1.
Theorem 4.26. Let q = 2 f q = 2^{f} q = 2 f with f ≥ 4 f \ge4 f ≥ 4 , let H = D 2 ( q + 1 ) H = D_{2(q+1)} H = D 2 ( q + 1 ) and M = D 2 ( q − 1 ) M = D_{2(q-1)} M = D 2 ( q − 1 ) , and let B B B be a single orbit of M M M on P \mathcal{P} P . Then B G B^{G} B G is not a 2-design.
Proof. By Lemma 4.8 an orbit of M M M has length q − 1 q - 1 q − 1 and a block consisting of one orbit has k = q − 1 k = q - 1 k = q − 1 , b = q ( q + 1 ) / 2 b = q(q + 1)/2 b = q ( q + 1 ) /2 , r = q + 1 r = q + 1 r = q + 1 and λ = 2 \lambda= 2 λ = 2 . Suppose that B G B^{G} B G is a 2-design, and let O = O X O = O_{X} O = O X be the orbit of M M M which is B B B .
By Proposition 4.30 together with the count s = ( q − 2 ) / 2 s = (q - 2)/2 s = ( q − 2 ) /2 of orbits of G G G on pairs in Proposition 4.34 , every orbit C C C of G G G on unordered pairs satisfies
P C ( B ) = ( k 2 ) / s = k ( k − 1 ) q − 2 = q − 1. P_{C}(B) = \binom{k}{2}/s = \frac{k(k - 1)}{q - 2} = q - 1. P C ( B ) = ( 2 k ) / s = q − 2 k ( k − 1 ) = q − 1.
On the other hand, by Lemma 4.25 and q ≥ 16 q \ge16 q ≥ 16 there are t , s ∈ F q × ∖ { 1 } t,s \in\mathbb{F}_{q}^{\times} \setminus\{1\} t , s ∈ F q × ∖ { 1 } with s + 1 / s = t + 1 / t + κ s + 1/s = t + 1/t + \kappa s + 1/ s = t + 1/ t + κ , and by Lemma 4.11 (ii) the pairs { X , Y t } \{X,Y_{t}\} { X , Y t } and { X , Y s } \{X,Y_{s}\} { X , Y s } are in one orbit of G G G . For a ∈ F q × a \in\mathbb{F}_{q}^{\times} a ∈ F q × the element x ↦ ( a ′ / a ) x x \mapsto(a'/a)x x ↦ ( a ′ / a ) x of G G G carries { X a , X a t } \{X_{a},X_{at}\} { X a , X a t } to { X a ′ , X a ′ t } \{X_{a'},X_{a't}\} { X a ′ , X a ′ t } , so the q − 1 q - 1 q − 1 pairs { X a , X a t } \{X_{a},X_{at}\} { X a , X a t } form one orbit C 1 C_{1} C 1 , and likewise the q − 1 q - 1 q − 1 pairs { X a , X a s } \{X_{a},X_{as}\} { X a , X a s } form one orbit C 2 C_{2} C 2 . These two orbits contain the equivalent pairs { X , Y t } \{X,Y_{t}\} { X , Y t } and { X , Y s } \{X,Y_{s}\} { X , Y s } , hence C 1 = C 2 = C C_{1} = C_{2} = C C 1 = C 2 = C , and this single orbit contains the $2(q - 1)$ pairs of B B B labelled t t t and $s. This contradicts P C ( B ) = q − 1 P_{C}(B) = q - 1 P C ( B ) = q − 1 .
At q = 8 q = 8 q = 8 the statement fails: the orbit of D 14 D_{14} D 14 on the 28 points is the block of the design 2 2 2 -( 28 , 7 , 2 ) (28, 7, 2) ( 28 , 7 , 2 ) of Proposition 4.28 , and q = 8 q = 8 q = 8 is the only even degree in which a single orbit carries a design.
The complementary block of a design is again a design, and at the opposite extreme of the range 1 ≤ a ≤ q / 2 1 \le a \le q/2 1 ≤ a ≤ q /2 of Lemma 4.8 the complement is a single orbit. This removes the second extreme value of a a a .
Corollary 4.27. Let q = 2 f q = 2^{f} q = 2 f with f ≥ 4 f \ge4 f ≥ 4 , and let B B B be a union of a a a orbits of M = D 2 ( q − 1 ) M = D_{2(q-1)} M = D 2 ( q − 1 ) on the v = q ( q − 1 ) / 2 v = q(q - 1)/2 v = q ( q − 1 ) /2 points of P \mathcal{P} P such that B G B^{G} B G is a 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design with λ > 0 \lambda> 0 λ > 0 and k < v k < v k < v . Then 2 ≤ a ≤ q / 2 − 2 2 \le a \le q/2 - 2 2 ≤ a ≤ q /2 − 2 ; in particular neither a = 1 a = 1 a = 1 nor a = q / 2 − 1 a = q/2 - 1 a = q /2 − 1 occurs. If moreover ( q − 2 ) / 2 (q - 2)/2 ( q − 2 ) /2 is prime, then no such B B B exists at all.
Proof. The value a = 1 a = 1 a = 1 is excluded by Theorem 4.26 . Let a = q / 2 − 1 a = q/2 - 1 a = q /2 − 1 , put m = ( q − 2 ) / 2 m = (q - 2)/2 m = ( q − 2 ) /2 so that q = 2 m + 2 q = 2m + 2 q = 2 m + 2 , and note that Lemma 4.8 gives k = a ( q − 1 ) = v − ( q − 1 ) k = a(q - 1) = v - (q - 1) k = a ( q − 1 ) = v − ( q − 1 ) : the complement B c B^{c} B c consists of a single orbit of M M M . Complementation carries the b b b blocks B g B^{g} B g onto the blocks ( B c ) g (B^{c})^{g} ( B c ) g and preserves the block stabilizer, so ( B c ) G (B^{c})^{G} ( B c ) G is a 2 2 2 -( v , q − 1 , λ ′ ) (v,q - 1,\lambda') ( v , q − 1 , λ ′ ) design with b ′ = b = q ( q + 1 ) / 2 b' = b = q(q + 1)/2 b ′ = b = q ( q + 1 ) /2 , r ′ = b ′ k ′ / v = q + 1 r' = b'k'/v = q + 1 r ′ = b ′ k ′ / v = q + 1 and λ ′ = b − 2 r + λ \lambda' = b - 2r + \lambda λ ′ = b − 2 r + λ . In this case Lemma 4.8 gives λ = 2 a ( a q − a − 1 ) / ( q − 2 ) = 2 m 2 + m − 1 \lambda= 2a(aq-a-1)/(q-2) = 2m^2 + m - 1 λ = 2 a ( a q − a − 1 ) / ( q − 2 ) = 2 m 2 + m − 1 , whence
λ ′ = q ( q + 1 ) 2 − 2 a ( q + 1 ) + λ = ( m + 1 ) ( 2 m + 3 ) − 2 m ( 2 m + 3 ) + 2 m 2 + m − 1 = 2 , \lambda' = \frac{q(q+1)}{2} - 2a(q+1) + \lambda= (m+1)(2m+3) - 2m(2m+3) + 2m^2 + m - 1 = 2, λ ′ = 2 q ( q + 1 ) − 2 a ( q + 1 ) + λ = ( m + 1 ) ( 2 m + 3 ) − 2 m ( 2 m + 3 ) + 2 m 2 + m − 1 = 2 ,
so B c B^c B c would be a single orbit of M M M whose G G G -images form a design, which Theorem 4.26 excludes. Since a = q / 2 a=q/2 a = q /2 gives k = v k=v k = v , this leaves 2 ≤ a ≤ q / 2 − 2 2 \leq a \leq q/2-2 2 ≤ a ≤ q /2 − 2 . Finally, by Lemma 4.8 the integrality of λ \lambda λ is equivalent to m ∣ a ( a − 1 ) m \mid a(a-1) m ∣ a ( a − 1 ) . If m m m is prime, then a ≡ 0 a \equiv0 a ≡ 0 or a ≡ 1 ( m o d m ) a \equiv1 \pmod{m} a ≡ 1 ( mod m ) , and as 1 ≤ a ≤ q / 2 = m + 1 1 \leq a \leq q/2=m+1 1 ≤ a ≤ q /2 = m + 1 the only such values are a = 1 a=1 a = 1 , a = m a=m a = m and a = m + 1 a=m+1 a = m + 1 , all of them excluded above.
Proposition 4.28 (q = 8 q=8 q = 8 ). Let q = 8 q=8 q = 8 and G = P S L 2 ( 8 ) G=PSL_2(8) G = P S L 2 ( 8 ) . The group G G G acts point-primitively and not 2-homogeneously on a set of 28 points, with point stabilizer H = D 18 H=D_{18} H = D 18 , and on a set of 36 points, with point stabilizer H = D 14 H=D_{14} H = D 14 . On the 28 points the block stabilizer M = D 14 M=D_{14} M = D 14 gives the designs 2-(28, 7, 2) and 2-(28, 21, 20), and these two are block-primitive. The subfield subgroup M = S 3 M=S_3 M = S 3 gives further unions with constant covering numbers, namely 2-(28, 9, 8), 2-(28, 19, 38), 2-(28, 10, 10), 2-(28, 18, 34) on the 28 points and 2-(36, 6, 2), 2-(36, 30, 58) on the 36 points; but S 3 S_3 S 3 is not maximal in G G G (Lemma 4.7 ), and since ∣ G B ∣ = 6 |G_B|=6 ∣ G B ∣ = 6 while b = 84 b=84 b = 84 , the subgroup D 18 ≤ G D_{18} \leq G D 18 ≤ G of order 18 cannot be transitive on the 84 blocks, so its orbits form a nontrivial G G G -invariant partition of the block set and the block action of these six unions is imprimitive. None of the six is therefore block-primitive, and exactly two designs with socle P S L 2 ( 8 ) PSL_2(8) P S L 2 ( 8 ) are. Table 7 records all eight unions with constant covering numbers and marks which of them are block-primitive.
Proof. By Dickson’s list the maximal subgroups of P S L 2 ( 8 ) PSL_2(8) P S L 2 ( 8 ) are exactly D 18 D_{18} D 18 and D 14 D_{14} D 14 , the two dihedral groups of orders 2 ( q ± 1 ) 2(q \pm1) 2 ( q ± 1 ) ; the subfield group S 3 = P S L 2 ( 2 ) S_3=PSL_2(2) S 3 = P S L 2 ( 2 ) is contained in D 18 D_{18} D 18 and hence is not maximal (Lemma 4.7 ). A block stabilizer of a block-primitive design is maximal and satisfies ∣ G B ∣ < ∣ H ∣ |G_B|<|H| ∣ G B ∣ < ∣ H ∣ , so the only admissible pair is M = D 14 M=D_{14} M = D 14 against H = D 18 H=D_{18} H = D 18 , and the 36 points admit none: every other maximal subgroup has order above ∣ H ∣ = 14 |H|=14 ∣ H ∣ = 14 , while M = H M=H M = H is excluded. On the 28 points D 14 D_{14} D 14 has 4 orbits of 7 points, and exactly one of them is the block of a design; its 36 images form 2-(28, 7, 2) and the complementary blocks form 2-(28, 21, 20), with r = b k / v = 36 ⋅ 7 / 28 = 9 r=bk/v=36\cdot7/28=9 r = bk / v = 36 ⋅ 7/28 = 9 and λ ′ = b − 2 r + λ = 36 − 18 + 2 = 20 \lambda'=b-2r+\lambda=36-18+2=20 λ ′ = b − 2 r + λ = 36 − 18 + 2 = 20 . The subfield S 3 S_3 S 3 has 7 orbits of lengths 3, 6, 6, 3, 1, 3, 6 on the 28 points and 8 orbits of lengths 3, 3, 3, 3, 6, 6, 6, 6 on the 36 points; every block with the prescribed block stabilizer is a union of these orbits, so it suffices to run through the 2 4 2^4 2 4 , 2 7 2^7 2 7 and 2 8 2^8 2 8 unions, identify a union with its complement, and compute the covering numbers of the resulting block set on all pairs of points. The covering numbers are constant exactly for the eight unions of Table 7 , with the values of λ \lambda λ recorded there, the complementary blocks being accounted for by λ ′ = b − 2 r + λ \lambda' = b - 2r + \lambda λ ′ = b − 2 r + λ ; this reproduces the pairs ( 28 , 7 , 2 ) / ( 28 , 21 , 20 ) (28, 7, 2)/(28, 21, 20) ( 28 , 7 , 2 ) / ( 28 , 21 , 20 ) , ( 28 , 9 , 8 ) / ( 28 , 19 , 38 ) (28, 9, 8)/(28, 19, 38) ( 28 , 9 , 8 ) / ( 28 , 19 , 38 ) , ( 28 , 10 , 10 ) / ( 28 , 18 , 34 ) (28, 10, 10)/(28, 18, 34) ( 28 , 10 , 10 ) / ( 28 , 18 , 34 ) and ( 36 , 6 , 2 ) / ( 36 , 30 , 58 ) (36, 6, 2)/(36, 30, 58) ( 36 , 6 , 2 ) / ( 36 , 30 , 58 ) . In the six unions with G B = S 3 G_B = S_3 G B = S 3 one has ∣ G B ∣ = 6 |G_B| = 6 ∣ G B ∣ = 6 and b = 84 b = 84 b = 84 , so the subgroup D 18 ≤ G D_{18} \leq G D 18 ≤ G of order 18 is not transitive on the 84 blocks and its orbits form a nontrivial G G G -invariant partition of the block set; these unions are thus 2-designs whose block action is imprimitive, which is why Table 7 marks them as not block-primitive. The permutation rank of G G G is 4 on the 28 points and 5 on the 36 points, so G G G is 2-homogeneous on neither set.
Proposition 4.29 (q = 32 q = 32 q = 32 ). Let q = 32 q = 32 q = 32 , G = P S L 2 ( 32 ) G = PSL_2(32) G = P S L 2 ( 32 ) , H = D 66 H = D_{66} H = D 66 and M = D 62 M = D_{62} M = D 62 , so that v = 496 v = 496 v = 496 and b = 528 b = 528 b = 528 . Each orbit of M M M on the 496 points has 31 points. Among the unions of a a a such orbits exactly two carry designs: a union B B B of 6 orbits, whose 528 images under G G G form a 2-(496, 186, 74) design, and its complement, a 2-(496, 310, 206) design. Both are point-primitive, block-primitive and non-symmetric, and the permutation rank of G G G on the 496 points is 16.
Proof. By Lemma 4.8 the admissible block sizes are k = a ( q − 1 ) = 31 a k = a(q - 1) = 31a k = a ( q − 1 ) = 31 a with 15 ∣ a ( a − 1 ) 15 \mid a(a - 1) 15 ∣ a ( a − 1 ) , that is a ∈ { 1 , 6 , 10 , 15 , 16 } a \in\{1, 6, 10, 15, 16\} a ∈ { 1 , 6 , 10 , 15 , 16 } , and the corresponding pairs ( k , λ ) (k, \lambda) ( k , λ ) are ( 31 , 2 ) (31, 2) ( 31 , 2 ) , ( 186 , 74 ) (186, 74) ( 186 , 74 ) , ( 310 , 206 ) (310, 206) ( 310 , 206 ) , ( 465 , 464 ) (465, 464) ( 465 , 464 ) and ( 496 , 528 ) (496, 528) ( 496 , 528 ) . Since G G G is transitive on blocks, the covering number depends only on the orbit of G G G on unordered pairs; there are 15 such orbits, each of size μ t = 8184 \mu_t = 8184 μ t = 8184 , so a block must contain exactly n t = μ t λ / b = 1147 n_t = \mu_t\lambda/b = 1147 n t = μ t λ / b = 1147 pairs from each orbit. Enumerating the 16, 8008, 8008, 16 and 1 unions for the five values of a a a , and comparing the exact pair counts with n t n_t n t , one finds that a = 1 a = 1 a = 1 and a = 15 a = 15 a = 15 admit no union, that a = 16 a = 16 a = 16 gives the complete design, and that a = 6 a = 6 a = 6 and a = 10 a = 10 a = 10 admit exactly one union each, namely B B B and its complement. For these two the covering numbers of all 528 blocks on all ( 496 2 ) = 122760 \binom{496}{2} = 122760 ( 2 496 ) = 122760 pairs are 74 and 206 respectively; the values r = b k / v = 198 r = bk/v = 198 r = bk / v = 198 and r ( k − 1 ) / ( v − 1 ) = 198 ⋅ 185 / 495 = 74 r(k - 1)/(v - 1) = 198 \cdot185/495 = 74 r ( k − 1 ) / ( v − 1 ) = 198 ⋅ 185/495 = 74 confirm the parameters. Maximality of D 66 D_{66} D 66 and of D 62 D_{62} D 62 is Dickson’s list [22 ] . □ \square □
The parameters of Lemma 4.8 are computed for the first even degrees in Table 8 , where m = ( q − 2 ) / 2 m = (q - 2)/2 m = ( q − 2 ) /2 .
Table 8. Even q q q : admissible block sizes and pair counts.
Table 8 is the even-characteristic counterpart of the arithmetic sieve of Case 2. The fourth column lists the replication numbers k = a ( q − 1 ) k = a(q - 1) k = a ( q − 1 ) and the fifth the values of λ \lambda λ for the admissible a a a in increasing order; the last column records whether some union of orbits of M M M gives a constant pair count on the 28, 120, 496 and 2016 points respectively. For every q q q the largest admissible value a = q / 2 a = q/2 a = q /2 gives k = v k = v k = v , the complete design, which the hypothesis excludes; the complement of a union of a a a orbits is a union of q / 2 − a q/2 - a q /2 − a orbits, so a a a and q / 2 − a q/2 - a q /2 − a carry complementary designs whenever either of them does, and at q = 32 q = 32 q = 32 the two surviving values a = 6 a = 6 a = 6 and a = 10 a = 10 a = 10 are complementary. The entry for q = 64 q = 64 q = 64 refers to the two dihedral block stabilizers; of the subfield block stabilizers of that degree, P S L 2 ( 8 ) PSL_2(8) P S L 2 ( 8 ) is excluded by Fisher’s inequality (its index 520 is smaller than v = 2016 v = 2016 v = 2016 ), and the case M = A 5 M = A_5 M = A 5 is excluded by Proposition 4.42. By Theorem 4.26 and Corollary 4.27 the two extreme values a = 1 a = 1 a = 1 and a = q / 2 − 1 a = q/2 - 1 a = q /2 − 1 yield no design for any even q ≥ 16 q \ge16 q ≥ 16 , so among the entries of the last column only 2 ≤ a ≤ q / 2 − 2 2 \le a \le q/2 - 2 2 ≤ a ≤ q /2 − 2 can produce one, and none does when ( q − 2 ) / 2 (q - 2)/2 ( q − 2 ) /2 is prime; the entry at q = 8 q = 8 q = 8 records the genuine design 2 2 2 -( 28 , 7 , 2 ) (28, 7, 2) ( 28 , 7 , 2 ) rather than an exception to that exclusion.
A criterion for orbit unions
The enumerations of unions of orbits carried out above decide the same question in each degree, namely whether a prescribed union of M M M -orbits is the block of a 2 2 2 -design. The criterion below separates the part of that question which is arithmetic from the part which depends on how the pairs of points are distributed among the orbits of G G G on pairs, and it requires no hypothesis on G G G beyond transitivity.
Proposition 4.30 (classwise criterion). Let G G G be a transitive permutation group on a finite set P \mathcal{P} P with ∣ P ∣ = v \lvert\mathcal{P}\rvert= v ∣ P ∣ = v . Let B ⊆ P B \subseteq\mathcal{P} B ⊆ P , put M = G B M = G_B M = G B , and suppose that G G G is transitive on B G B^G B G . Let C 1 , … , C s \mathcal{C}_1,\ldots,\mathcal{C}_s C 1 , … , C s be the orbits of G G G on the unordered pairs of distinct points of P \mathcal{P} P , and for an unordered pair π \pi π let G π G_\pi G π be its stabilizer in G G G . Put
P C ( B ) = ∣ { π ⊆ B : π ∈ C } ∣ . P_{\mathcal{C}}(B) = \lvert\{\pi\subseteq B : \pi\in\mathcal{C}\}\rvert. P C ( B ) = ∣{ π ⊆ B : π ∈ C }∣ . Then B G B^G B G is a 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design with λ > 0 \lambda> 0 λ > 0 if and only if the product ∣ G π ∣ P C ( B ) \lvert G_\pi\rvert P_{\mathcal{C}}(B) ∣ G π ∣ P C ( B ) is the same for all classes C \mathcal{C} C , and then
λ = ∣ G π ∣ P C ( B ) ∣ M ∣ \lambda= \frac{\lvert G_\pi\rvert P_{\mathcal{C}}(B)}{\lvert M\rvert} λ = ∣ M ∣ ∣ G π ∣ P C ( B ) for every class C \mathcal{C} C .
Proof. For a pair π \pi π let N ( π ) N(\pi) N ( π ) be the number of blocks of B G B^G B G containing π \pi π . A block equals B g B^g B g for exactly ∣ M ∣ \lvert M\rvert ∣ M ∣ elements g g g , so the number of elements g g g with π ⊆ B g \pi\subseteq B^g π ⊆ B g is ∣ M ∣ N ( π ) \lvert M\rvert N(\pi) ∣ M ∣ N ( π ) . As π ⊆ B g \pi\subseteq B^g π ⊆ B g is equivalent to π g − 1 ⊆ B \pi^{g^{-1}} \subseteq B π g − 1 ⊆ B ,
∣ M ∣ N ( π ) = ∣ { g ∈ G : π g ⊆ B } ∣ = ∑ π ′ ⊆ B ∣ { g ∈ G : π ′ g = π } ∣ . \lvert M\rvert N(\pi) = \lvert\{g \in G : \pi^g \subseteq B\}\rvert= \sum_{\pi' \subseteq B}\lvert\{g \in G : \pi'^g = \pi\}\rvert. ∣ M ∣ N ( π ) = ∣{ g ∈ G : π g ⊆ B }∣ = π ′ ⊆ B ∑ ∣{ g ∈ G : π ′ g = π }∣ .
If π ′ \pi' π ′ lies in the class C \mathcal{C} C of π \pi π then π ′ g = π \pi'^g = \pi π ′ g = π for some g g g , and the elements g g g with π ′ g = π \pi'^g = \pi π ′ g = π form the coset g 0 G π g_0G_\pi g 0 G π , of order ∣ G π ∣ \lvert G_\pi\rvert ∣ G π ∣ ; if π ′ \pi' π ′ lies in another class this set is empty. Hence
∣ M ∣ N ( π ) = ∣ G π ∣ P C ( B ) , C the class of π . \lvert M\rvert N(\pi) = \lvert G_\pi\rvert P_{\mathcal{C}}(B), \qquad\mathcal{C}\text{ the class of }\pi. ∣ M ∣ N ( π ) = ∣ G π ∣ P C ( B ) , C the class of π .
The pair π \pi π is covered λ \lambda λ times by B G B^G B G exactly when N ( π ) N(\pi) N ( π ) takes the same value λ \lambda λ for every π \pi π , which is the stated condition; dividing by ∣ M ∣ \lvert M\rvert ∣ M ∣ gives the formula for λ \lambda λ . ∎
Corollary 4.31. In the situation of Proposition 4.30 , ∑ C P C ( B ) = ( k 2 ) \sum_C P_C(B)=\binom{k}{2} ∑ C P C ( B ) = ( 2 k ) . If in addition ∣ G π ∣ = μ \lvert G_\pi\rvert=\mu ∣ G π ∣ = μ for every pair π \pi π , then
λ = μ k ( k − 1 ) 2 s ∣ M ∣ . \lambda=\frac{\mu k(k-1)}{2s\lvert M\rvert}. λ = 2 s ∣ M ∣ μ k ( k − 1 ) . where s s s is the number of orbits of G G G on unordered pairs.
Proof. The ( k 2 ) \binom{k}{2} ( 2 k ) pairs of B B B lie in exactly one class each, so the classes partition them into the s s s parts P C ( B ) P_C(B) P C ( B ) , and each part equals ( k 2 ) / s \binom{k}{2}/s ( 2 k ) / s as soon as ∣ G π ∣ P C ( B ) \lvert G_\pi\lvert P_C(B) ∣ G π ∣ P C ( B ) is constant. Substituting into the formula of Proposition 4.30 gives λ = μ ( k 2 ) / ( s ∣ M ∣ ) = μ k ( k − 1 ) / ( 2 s ∣ M ∣ ) \lambda=\mu\binom{k}{2}/(s\lvert M\rvert)=\mu k(k-1)/(2s\lvert M\rvert) λ = μ ( 2 k ) / ( s ∣ M ∣) = μ k ( k − 1 ) / ( 2 s ∣ M ∣) .
The criterion reproduces the pair count used in the proof of Proposition 4.29 rather than adding to it. There G = P S L 2 ( 32 ) G=PSL_2(32) G = P S L 2 ( 32 ) has order 32736 = 4 ⋅ 8184 32736=4\cdot8184 32736 = 4 ⋅ 8184 , the 15 classes of pairs have 8184 members each, so μ = 4 \mu=4 μ = 4 and s = 15 s=15 s = 15 , and Corollary 4.31 with ∣ M ∣ = 62 \lvert M\rvert=62 ∣ M ∣ = 62 gives λ = k ( k − 1 ) / 465 \lambda=k(k-1)/465 λ = k ( k − 1 ) /465 , that is 2 , 74 , 206 , 464 , 528 2,74,206,464,528 2 , 74 , 206 , 464 , 528 for k = 31 , 186 , 310 , 465 , 496 k=31,186,310,465,496 k = 31 , 186 , 310 , 465 , 496 , the five values recorded in Table 8 ; the number of pairs of a block that must lie in a fixed class is then λ ∣ M ∣ / μ = 1147 \lambda\lvert M\rvert/\mu=1147 λ ∣ M ∣ / μ = 1147 , which is the count n t n_t n t of the proof. At q = 8 q=8 q = 8 with M = S 3 M=S_3 M = S 3 , s = 3 s=3 s = 3 and μ = 4 \mu=4 μ = 4 give λ = k ( k − 1 ) / 9 \lambda=k(k-1)/9 λ = k ( k − 1 ) /9 , that is 8 , 38 , 10 , 34 8,38,10,34 8 , 38 , 10 , 34 for k = 9 , 19 , 10 , 18 k=9,19,10,18 k = 9 , 19 , 10 , 18 , the four S 3 S_3 S 3 entries of Table 7 , each of them with an imprimitive block action.
Finally we record the parameters of the subfield configuration to which this criterion has to be applied. Let q = u p q=u^p q = u p with u = 2 m u=2^m u = 2 m , m ≥ 2 m\ge2 m ≥ 2 and p p p an odd prime, let H = D 2 ( q + 1 ) H=D_{2(q+1)} H = D 2 ( q + 1 ) and M = P G L 2 ( u ) ≅ P S L 2 ( u ) ≤ G M=PGL_2(u)\cong PSL_2(u)\le G M = P G L 2 ( u ) ≅ P S L 2 ( u ) ≤ G . The degree [ q : u ] = p [q:u]=p [ q : u ] = p is prime, so M M M is maximal in G G G by Dickson’s list, and ∣ M ∣ = u ( u 2 − 1 ) \lvert M\rvert=u(u^2-1) ∣ M ∣ = u ( u 2 − 1 ) . With v = q ( q − 1 ) / 2 v=q(q-1)/2 v = q ( q − 1 ) /2 this gives
b = ∣ G ∣ ∣ M ∣ = q ( q 2 − 1 ) u ( u 2 − 1 ) = u p − 1 ( u 2 p − 2 + u 2 p − 4 + ⋯ + u 2 + 1 ) . b=\frac{\lvert G\rvert}{\lvert M\rvert}=\frac{q(q^2-1)}{u(u^2-1)}=u^{p-1}(u^{2p-2}+u^{2p-4}+\cdots+u^2+1). b = ∣ M ∣ ∣ G ∣ = u ( u 2 − 1 ) q ( q 2 − 1 ) = u p − 1 ( u 2 p − 2 + u 2 p − 4 + ⋯ + u 2 + 1 ) .
Since ∣ M ∩ H g ∣ \lvert M\cap H^g\rvert ∣ M ∩ H g ∣ divides gcd ( ∣ M ∣ , ∣ H ∣ ) = gcd ( u ( u 2 − 1 ) , 2 ( u p + 1 ) ) = 2 ( u + 1 ) \gcd(\lvert M\rvert,\lvert H\rvert)=\gcd(u(u^2-1),2(u^p+1))=2(u+1) g cd(∣ M ∣ , ∣ H ∣) = g cd( u ( u 2 − 1 ) , 2 ( u p + 1 )) = 2 ( u + 1 ) , the orbit of a point x x x has length ∣ M ∣ / ∣ M ∩ H g ∣ \lvert M\rvert/\lvert M\cap H^g\rvert ∣ M ∣ / ∣ M ∩ H g ∣ with ∣ M ∩ H g ∣ \lvert M\cap H^g\rvert ∣ M ∩ H g ∣ dividing 2 ( u + 1 ) 2(u+1) 2 ( u + 1 ) ; the block size is the sum of the lengths of the orbits that the block contains, and by Lemma 4.32 the only values of ∣ M ∩ H g ∣ \lvert M\cap H^g\rvert ∣ M ∩ H g ∣ which occur are 1 1 1 , 2 2 2 and 2 ( u + 1 ) 2(u+1) 2 ( u + 1 ) , so that the three orbit lengths are those of Corollary 4.33 . The replication number is r = b k / v r=bk/v r = bk / v . Moreover b > v b>v b > v for every u ≥ 2 u\ge2 u ≥ 2 and every prime p ≥ 3 p\ge3 p ≥ 3 , because ( u 2 p − 1 ) / ( u 2 − 1 ) ≥ u 2 p − 2 (u^{2p}-1)/(u^2-1)\ge u^{2p-2} ( u 2 p − 1 ) / ( u 2 − 1 ) ≥ u 2 p − 2 gives b ≥ u 3 p − 3 > u 2 p / 2 > v b\ge u^{3p-3}>u^{2p}/2>v b ≥ u 3 p − 3 > u 2 p /2 > v , so Fisher’s inequality imposes no restriction on this configuration and the block condition is the one supplied by Proposition 4.30 .
The orbit structure of the subfield configuration
Proposition 4.30 leaves open which unions of orbits can carry a block, and the answer turns on the orbit lengths. For a dihedral block stabilizer these are the single length q − 1 q-1 q − 1 of Lemma 4.8 ; for a subfield stabilizer there are three lengths, one for each type of point, and they are pinned down by the census below. Write
P = { { x , x q } : x ∈ F q 2 ∖ F q } , v = ∣ P ∣ = q ( q − 1 ) / 2 , \mathcal{P}=\{\{x,x^q\}:x\in F_{q^2}\setminus F_q\},\qquad v=|\mathcal{P}|=q(q-1)/2, P = {{ x , x q } : x ∈ F q 2 ∖ F q } , v = ∣ P ∣ = q ( q − 1 ) /2 ,
let σ : t ↦ t q \sigma:t\mapsto t^q σ : t ↦ t q be the Frobenius of F q 2 F_{q^2} F q 2 over F q F_q F q and σ u : t ↦ t u \sigma_u:t\mapsto t^u σ u : t ↦ t u the Frobenius of F u 2 p F_{u^{2p}} F u 2 p over F u F_u F u , and recall q = u p q=u^p q = u p and F q ∩ F u 2 = F u F_q\cap F_{u^2}=F_u F q ∩ F u 2 = F u . Every element of M = P G L 2 ( u ) M=PGL_2(u) M = P G L 2 ( u ) has coefficients in F u F_u F u and so commutes with both σ \sigma σ and σ u \sigma_u σ u ; conversely a Möbius transformation commuting with σ u \sigma_u σ u has coefficients in F u F_u F u and hence lies in M M M . Both directions are used in the proof.
Lemma 4.32. Let u = 2 m u=2^m u = 2 m with m ≥ 2 m\geq2 m ≥ 2 , q = u p q=u^p q = u p with p p p an odd prime, H = D 2 ( q + 1 ) H=D_{2(q+1)} H = D 2 ( q + 1 ) and M = P G L 2 ( u ) ≅ P S L 2 ( u ) ≤ G = P S L 2 ( q ) M=PGL_2(u)\cong PSL_2(u)\leq G=PSL_2(q) M = P G L 2 ( u ) ≅ P S L 2 ( u ) ≤ G = P S L 2 ( q ) . Then M M M has exactly
1 + ( u p − 1 − 1 ) + u ( u 2 p − 2 − u p + u − 1 ) 2 ( u 2 − 1 ) 1+(u^{p-1}-1)+\frac{u(u^{2p-2}-u^p+u-1)}{2(u^2-1)} 1 + ( u p − 1 − 1 ) + 2 ( u 2 − 1 ) u ( u 2 p − 2 − u p + u − 1 ) orbits on P \mathcal{P} P , namely 1 orbit of length u ( u − 1 ) / 2 u(u-1)/2 u ( u − 1 ) /2 whose points have stabilizer D 2 ( u + 1 ) D_{2(u+1)} D 2 ( u + 1 ) , then u p − 1 − 1 u^{p-1}-1 u p − 1 − 1 orbits of length u ( u 2 − 1 ) / 2 u(u^2-1)/2 u ( u 2 − 1 ) /2 whose points have stabilizer of order 2, and finally u ( u 2 p − 2 − u p + u − 1 ) / ( 2 ( u 2 − 1 ) ) u(u^{2p-2}-u^p+u-1)/(2(u^2-1)) u ( u 2 p − 2 − u p + u − 1 ) / ( 2 ( u 2 − 1 )) orbits of length u ( u 2 − 1 ) u(u^2-1) u ( u 2 − 1 ) whose points have trivial stabilizer.
Proof. (i) Points over F u 2 F_{u^2} F u 2 . Since F q ∩ F u 2 = F u F_q\cap F_{u^2}=F_u F q ∩ F u 2 = F u , a point { x , x q } \{x,x^q\} { x , x q } has x ∈ F u 2 x\in F_{u^2} x ∈ F u 2 exactly when x ∈ F u 2 ∖ F u x\in F_{u^2}\setminus F_u x ∈ F u 2 ∖ F u ; for such x x x one has x q = x u x^q=x^u x q = x u , because u 2 − 1 u^2-1 u 2 − 1 divides u p − u u^p-u u p − u when p p p is odd. These points are the pairs of roots of the irreducible quadratic polynomials over F u F_u F u and therefore number ( u 2 − u ) / 2 = u ( u − 1 ) / 2 (u^2-u)/2=u(u-1)/2 ( u 2 − u ) /2 = u ( u − 1 ) /2 .
They form a single orbit. Given two of them, { x , x u } \{x,x^u\} { x , x u } and { y , y u } \{y,y^u\} { y , y u } , let h h h be the Möbius transformation with h ( x ) = y h(x)=y h ( x ) = y and h ( x u ) = y u h(x^u)=y^u h ( x u ) = y u ; the two transformations h σ u h\sigma_u h σ u and σ u h \sigma_u h σ u h agree at the two distinct points x x x and x u x^u x u , hence h σ u = σ u h h\sigma_u=\sigma_u h h σ u = σ u h , so h h h has coefficients in F u F_u F u and h ∈ M h\in M h ∈ M .
For the stabilizer, an element of M M M fixing the point either sends x x x to x x x or sends x x x to x u x^u x u . Those sending x x x to x x x are the F u F_u F u -linear maps of the two-dimensional F u F_u F u -space F u ( x ) F_u(x) F u ( x ) commuting with multiplication by x x x , that is multiplication by the elements of F u ( x ) × F_u(x)^\times F u ( x ) × of norm 1; this is a cyclic group of order u + 1 u+1 u + 1 , a nonsplit torus of M M M . Those sending x x x to x u x^u x u are non-empty: the transformation interchanging x x x and x u x^u x u commutes with σ u \sigma_u σ u by the argument just given, hence lies in M M M . So the stabilizer has order 2 ( u + 1 ) 2(u+1) 2 ( u + 1 ) and the orbit has length ∣ M ∣ / ( 2 ( u + 1 ) ) = u ( u − 1 ) / 2 |M|/(2(u+1)) = u(u-1)/2 ∣ M ∣/ ( 2 ( u + 1 )) = u ( u − 1 ) /2 , which is the whole set of points over F u 2 F_{u^2} F u 2 .
(ii) *Points outside F u 2 F_{u^2} F u 2 . Let x ∉ F u 2 x \notin F_{u^2} x ∈ / F u 2 . The σ u \sigma_u σ u -orbit of x x x lies in F u 2 p F_{u^{2p}} F u 2 p and its length divides 2 p 2p 2 p ; it is not 1 1 1 , since x ∉ F u x \notin F_u x ∈ / F u , not 2 2 2 , since x ∉ F u 2 x \notin F_{u^2} x ∈ / F u 2 , and not p p p , since x ∉ F q = F u p x \notin F_q = F_{u^p} x ∈ / F q = F u p . So its length is 2 p 2p 2 p . An element of M M M fixing the point p p p pointwise fixes these 2 p 2p 2 p points and hence is 1 1 1 ; an element interchanging the two members of p p p is unique, since two of them differ by a pointwise fixer, and is an involution, since its square fixes x x x . Hence M p M_p M p is either trivial or of the form ⟨ τ ⟩ \langle\tau\rangle ⟨ τ ⟩ with τ \tau τ an involution of M M M .
(iii) Counting the points with stabilizer of order 2. As in the proof of Lemma 4.8 , every involution of M = SL 2 ( u ) M = \operatorname{SL}_2(u) M = SL 2 ( u ) is unipotent, hence a transvection [22 ] , and the involutions of M M M form a single conjugacy class, of size u 2 − 1 u^2-1 u 2 − 1 . A point p = { x , x q } p = \{x,x^q\} p = { x , x q } has stabilizer of order 2 exactly when some involution τ \tau τ satisfies
τ ( x ) = x q (19) \tau(x) = x^q \tag*{(19)} τ ( x ) = x q (19)
and then τ \tau τ is the unique involution fixing p p p .
We count the solutions of (), first for a single τ \tau τ . Since the involutions form one class and σ \sigma σ commutes with M M M , conjugation by an element of M M M carries the solutions of ( ) for τ \tau τ onto the solutions for τ g \tau^g τ g , so the count does not depend on τ \tau τ , and it suffices to treat the transvection τ 0 : t ↦ t + 1 \tau_0:t \mapsto t+1 τ 0 : t ↦ t + 1 . On P 1 ( F q 2 ) P^1(F_{q^2}) P 1 ( F q 2 ) the equation τ 0 ( x ) = σ ( x ) \tau_0(x) = \sigma(x) τ 0 ( x ) = σ ( x ) has the solution ∞ \infty ∞ , and for finite x x x it reads X q + X = 1 X^q+X=1 X q + X = 1 . On F q 2 F_{q^2} F q 2 the map L : X ↦ X q + X L:X \mapsto X^q+X L : X ↦ X q + X is F q F_q F q -linear with kernel F q F_q F q ; it is the trace of the extension F q 2 / F q F_{q^2}/F_q F q 2 / F q , so its image lies in F q F_q F q , and having dimension 2 − 1 = 1 2-1=1 2 − 1 = 1 over F q F_q F q that image is F q F_q F q itself. Hence L ( X ) = 1 L(X)=1 L ( X ) = 1 is soluble and its solutions form a coset of the kernel, of size q q q . Thus (*) has exactly q + 1 q+1 q + 1 solutions in P 1 ( F q 2 ) P^1(F_{q^2}) P 1 ( F q 2 ) , of which the only one not lying in P P P is the fixed point of τ \tau τ : for τ 0 \tau_0 τ 0 this is ∞ \infty ∞ , and for an arbitrary involution it is its fixed point, which lies in P 1 ( F u ) P^1(F_u) P 1 ( F u ) because a transvection is unipotent and its fixed point is the eigenvector of a Jordan block with entries in F u F_u F u .
Among the q q q solutions in F q 2 F_{q^2} F q 2 one distinguishes two kinds. Those in F u 2 F_{u^2} F u 2 satisfy τ ( x ) = x u \tau(x)=x^u τ ( x ) = x u and are exactly the points over F u 2 F_{u^2} F u 2 fixed by τ \tau τ ; since all involutions are conjugate and the points over F u 2 F_{u^2} F u 2 form one orbit of length u ( u − 1 ) / 2 u(u-1)/2 u ( u − 1 ) /2 whose stabilizers are of order 2 ( u + 1 ) 2(u+1) 2 ( u + 1 ) , each involution fixes exactly
u ( u − 1 ) 2 ⋅ u + 1 u 2 − 1 = u 2 \frac{u(u-1)}{2}\cdot\frac{u+1}{u^2-1}=\frac{u}{2} 2 u ( u − 1 ) ⋅ u 2 − 1 u + 1 = 2 u
of these points, that is u u u of the q q q solutions. The remaining q − u = u p − u q-u=u^p-u q − u = u p − u solutions are points of type (ii) with stabilizer ⟨ τ ⟩ \langle\tau\rangle ⟨ τ ⟩ , and they form ( u p − u ) / 2 (u^p-u)/2 ( u p − u ) /2 distinct points.
It remains to count these points over all τ \tau τ . The number of incidences ( τ , p ) (\tau,p) ( τ , p ) with p p p of type (ii) and M p = ⟨ τ ⟩ M_p=\langle\tau\rangle M p = ⟨ τ ⟩ equals
( u 2 − 1 ) ⋅ u p − u 2 , (u^2-1)\cdot\frac{u^p-u}{2}, ( u 2 − 1 ) ⋅ 2 u p − u ,
and each such p p p is incident with exactly one τ \tau τ , its stabilizer being of order 2. Hence there are ( u 2 − 1 ) ( u p − u ) / 2 (u^2-1)(u^p-u)/2 ( u 2 − 1 ) ( u p − u ) /2 points of type (ii) with nontrivial stabilizer, and, the orbit of such a point having length ∣ M ∣ / 2 = u ( u 2 − 1 ) / 2 |M|/2=u(u^2-1)/2 ∣ M ∣/2 = u ( u 2 − 1 ) /2 , they form exactly u p − 1 − 1 u^{p-1}-1 u p − 1 − 1 orbits.
(iv) The remaining points. The points not covered above number
v − u ( u − 1 ) 2 − ( u p − 1 − 1 ) u ( u 2 − 1 ) 2 = u 2 ( u 2 p − 2 − u p + u − 1 ) 2 , v-\frac{u(u-1)}{2}-(u^{p-1}-1)\frac{u(u^2-1)}{2}
=\frac{u^2(u^{2p-2}-u^p+u-1)}{2}, v − 2 u ( u − 1 ) − ( u p − 1 − 1 ) 2 u ( u 2 − 1 ) = 2 u 2 ( u 2 p − 2 − u p + u − 1 ) ,
as one checks from v = u p ( u p − 1 ) / 2 v=u^p(u^p-1)/2 v = u p ( u p − 1 ) /2 ; this is the product of u ( u 2 p − 2 − u p + u − 1 ) / ( 2 ( u 2 − 1 ) ) u(u^{2p-2}-u^p+u-1)/(2(u^2-1)) u ( u 2 p − 2 − u p + u − 1 ) / ( 2 ( u 2 − 1 )) , the number of orbits, with the orbit length ∣ M ∣ = u ( u 2 − 1 ) |M|=u(u^2-1) ∣ M ∣ = u ( u 2 − 1 ) . The points counted here have trivial stabilizer by (ii), so each of them spans an orbit of the latter length, and the three counts add up to v v v . □ \square □
Corollary 4.33. In the situation of Lemma 4.32, a union of M M M -orbits has size
k = a u ( u − 1 ) 2 + b u ( u 2 − 1 ) 2 + c u ( u 2 − 1 ) , k=a\frac{u(u-1)}{2}+b\frac{u(u^2-1)}{2}+cu(u^2-1), k = a 2 u ( u − 1 ) + b 2 u ( u 2 − 1 ) + c u ( u 2 − 1 ) , where
a ∈ { 0 , 1 } , 0 ≤ b ≤ u p − 1 − 1 , 0 ≤ c ≤ u ( u 2 p − 2 − u p + u − 1 ) 2 ( u 2 − 1 ) , a\in\{0,1\},\qquad0\le b\le u^{p-1}-1,\qquad0\le c\le\frac{u(u^{2p-2}-u^p+u-1)}{2(u^2-1)}, a ∈ { 0 , 1 } , 0 ≤ b ≤ u p − 1 − 1 , 0 ≤ c ≤ 2 ( u 2 − 1 ) u ( u 2 p − 2 − u p + u − 1 ) , and every such value is attained by a union of orbits.
Proof. Immediate from Lemma 4.32, which gives the three orbit lengths u ( u − 1 ) / 2 u(u-1)/2 u ( u − 1 ) /2 (one orbit, degenerate at u = 2 u=2 u = 2 ), u ( u 2 − 1 ) / 2 u(u^2-1)/2 u ( u 2 − 1 ) /2 and u ( u 2 − 1 ) u(u^2-1) u ( u 2 − 1 ) , together with the numbers 1 1 1 , u p − 1 − 1 u^{p-1}-1 u p − 1 − 1 and u ( u 2 p − 2 − u p + u − 1 ) / ( 2 ( u 2 − 1 ) ) u(u^{2p-2}-u^p+u-1)/(2(u^2-1)) u ( u 2 p − 2 − u p + u − 1 ) / ( 2 ( u 2 − 1 )) of orbits of each type; a union of a a a , b b b , c c c orbits of the three types therefore has the displayed size, and every triple within the stated ranges is realised by choosing any a a a , b b b , c c c such orbits. □ \square □
For u = 2 u = 2 u = 2 the first type degenerates: L A = 1 L_A = 1 L A = 1 and the single point of that orbit is fixed by all of M = S 3 M = S_3 M = S 3 , since 2 ( u + 1 ) = 6 = ∣ M ∣ 2(u + 1) = 6 = |M| 2 ( u + 1 ) = 6 = ∣ M ∣ . That fixed point is the structural origin of the failure of maximality at u = 2 u = 2 u = 2 : it is a point of P \mathcal{P} P whose stabilizer is dihedral of order 2 ( q + 1 ) 2(q + 1) 2 ( q + 1 ) , so S 3 S_3 S 3 is contained in a proper subgroup of G G G and is not maximal (Lemma 4.7 ). The subfield unions which are 2 2 2 -designs at q = 8 q = 8 q = 8 are therefore not block-primitive, their block action being imprimitive (Proposition 4.28 ); they are those of Table 7 . For u ≥ 4 u \ge4 u ≥ 4 the three lengths are distinct, u ( u − 1 ) / 2 < u ( u 2 − 1 ) / 2 < u ( u 2 − 1 ) u(u - 1)/2 < u(u^2 - 1)/2 < u(u^2 - 1) u ( u − 1 ) /2 < u ( u 2 − 1 ) /2 < u ( u 2 − 1 ) , and the orbit count of Lemma 4.32 equals u ( u 2 + u + 1 ) / 2 u(u^2 + u + 1)/2 u ( u 2 + u + 1 ) /2 when p = 3 p = 3 p = 3 , taking there the values 7 , 42 , 292 , 2184 7, 42, 292, 2184 7 , 42 , 292 , 2184 at u = 2 , 4 , 8 , 16 u = 2, 4, 8, 16 u = 2 , 4 , 8 , 16 , and it is generally 1 + ( u p − 1 − 1 ) + u ( u 2 p − 2 − u p + u − 1 ) / ( 2 ( u 2 − 1 ) ) 1 + (u^{p-1} - 1) + u(u^{2p-2} - u^p + u - 1)/(2(u^2 - 1)) 1 + ( u p − 1 − 1 ) + u ( u 2 p − 2 − u p + u − 1 ) / ( 2 ( u 2 − 1 )) .
The pairs of the point set
Proposition 4.30 measures how the pairs of a base block are distributed among the orbits of G G G on pairs, and Corollary 4.31 converts that distribution into a value of λ \lambda λ as soon as two constants of the action are known, namely the number s s s of orbits of G G G on the unordered pairs of distinct points of P \mathcal{P} P and the common order μ = ∣ G π ∣ \mu= |G_\pi| μ = ∣ G π ∣ of a pair stabilizer. Both constants are determined below for every even q q q , uniformly in q q q and without enumeration.
Proposition 4.34 (pair orbits). Let q = 2 f q = 2^f q = 2 f with f ≥ 2 f \ge2 f ≥ 2 , let G = PSL 2 ( q ) G = \operatorname{PSL}_2(q) G = PSL 2 ( q ) act on P \mathcal{P} P , where ∣ P ∣ = v = q ( q − 1 ) / 2 |\mathcal{P}| = v = q(q - 1)/2 ∣ P ∣ = v = q ( q − 1 ) /2 , and let H = D 2 ( q + 1 ) H = D_{2(q+1)} H = D 2 ( q + 1 ) be a point stabilizer. Then G G G has exactly ( q − 2 ) / 2 (q - 2)/2 ( q − 2 ) /2 orbits on the unordered pairs of distinct points of P \mathcal{P} P , each of length q ( q 2 − 1 ) / 4 q(q^2 - 1)/4 q ( q 2 − 1 ) /4 . Equivalently, the setwise stabilizer G π G_\pi G π of every unordered pair π \pi π has order 4 4 4 .
Proof. (i) Involutions. By the proof of Lemma 4.8 , every involution of G G G is unipotent, the involutions of G G G form a single conjugacy class of size q 2 − 1 q^2 - 1 q 2 − 1 , and each of them fixes exactly q / 2 q/2 q /2 points of P \mathcal{P} P .
(ii) The pair stabilizer splits. Let p 1 ≠ p 2 p_1 \ne p_2 p 1 = p 2 be points and π = { p 1 , p 2 } \pi= \{p_1, p_2\} π = { p 1 , p 2 } . Put A = G p 1 ∩ G p 2 A = G_{p_1} \cap G_{p_2} A = G p 1 ∩ G p 2 and B = { g ∈ G : g p 1 = p 2 } B = \{g \in G : gp_1 = p_2\} B = { g ∈ G : g p 1 = p 2 } . The group G G G is transitive on P \mathcal{P} P ([34 Section 2.2.1] ), so B B B is not empty; and for g 0 ∈ B g_0 \in B g 0 ∈ B every g ∈ B g \in B g ∈ B satisfies g 0 − 1 g ∈ A g_0^{-1}g \in A g 0 − 1 g ∈ A , whence B = g 0 A B = g_0A B = g 0 A . Therefore
∣ G π ∣ = ∣ A ∣ + ∣ B ∣ = 2 ∣ A ∣ . |G_\pi| = |A| + |B| = 2|A|. ∣ G π ∣ = ∣ A ∣ + ∣ B ∣ = 2∣ A ∣.
(iii) At most one element of A A A is an involution. Write p i = { x i , x i q } p_i = \{x_i, x_i^q\} p i = { x i , x i q } . An element of A A A permutes the four points x 1 , x 1 q , x 2 , x 2 q x_1, x_1^q, x_2, x_2^q x 1 , x 1 q , x 2 , x 2 q and preserves each of the two pairs. If it interchanged the members of exactly one pair, then it would fix the two members of the other pair, and its square, fixing four points, would be the identity; so it would be an involution fixing two points of P 1 ( F q 2 ) P^1(F_{q^2}) P 1 ( F q 2 ) . But an involution g g g of G G G is represented by M ∈ SL 2 ( q ) M \in\operatorname{SL}_2(q) M ∈ SL 2 ( q ) with M 2 = I M^2=I M 2 = I , and in characteristic 2 one has ( M − I ) 2 = M 2 + I = 0 (M-I)^2=M^2+I=0 ( M − I ) 2 = M 2 + I = 0 , so M − I M-I M − I has rank 1 and g g g fixes exactly one point of P 1 ( F q 2 ) P^1(F_{q^2}) P 1 ( F q 2 ) . Hence no element of A A A interchanges the members of exactly one pair, and an element of A A A either fixes the four points, and is the identity, or interchanges the members of both pairs. The latter is determined by its images on three of the four points, so there is at most one such element. Thus A ⊆ { 1 , τ } A \subseteq\{1,\tau\} A ⊆ { 1 , τ } for a single involution τ \tau τ , and ∣ A ∣ ≤ 2 |A| \leq2 ∣ A ∣ ≤ 2 .
(iv) The two counts agree. By (iii) an unordered pair of points is fixed by at most one involution. The number of incidences ( τ , π ) (\tau,\pi) ( τ , π ) , where τ \tau τ is an involution fixing both points of the pair π \pi π , is therefore at most ( v 2 ) \binom{v}{2} ( 2 v ) ; computed per involution it equals
( q 2 − 1 ) ( q / 2 2 ) = ( q 2 − 1 ) q ( q − 2 ) 8 = ( v 2 ) . (q^2-1)\binom{q/2}{2}=\frac{(q^2-1)q(q-2)}{8}=\binom{v}{2}. ( q 2 − 1 ) ( 2 q /2 ) = 8 ( q 2 − 1 ) q ( q − 2 ) = ( 2 v ) .
since v − 1 = ( q − 2 ) ( q + 1 ) / 2 v-1=(q-2)(q+1)/2 v − 1 = ( q − 2 ) ( q + 1 ) /2 . Hence every unordered pair is fixed by exactly one involution, so ∣ A ∣ = 2 |A|=2 ∣ A ∣ = 2 and ∣ G π ∣ = 4 |G_\pi|=4 ∣ G π ∣ = 4 by (ii). Each orbit of G G G on pairs has length ∣ G ∣ / 4 = q ( q 2 − 1 ) / 4 |G|/4=q(q^2-1)/4 ∣ G ∣/4 = q ( q 2 − 1 ) /4 , and the number of orbits is
( v 2 ) / ∣ G ∣ 4 = q ( q 2 − 1 ) ( q − 2 ) / 8 q ( q 2 − 1 ) / 4 = q − 2 2 . \binom{v}{2}\bigg/\frac{|G|}{4}=\frac{q(q^2-1)(q-2)/8}{q(q^2-1)/4}=\frac{q-2}{2}. ( 2 v ) / 4 ∣ G ∣ = q ( q 2 − 1 ) /4 q ( q 2 − 1 ) ( q − 2 ) /8 = 2 q − 2 .
Corollary 4.35. Let q = u p q=u^p q = u p with u = 2 m u=2^m u = 2 m , m ≥ 2 m\geq2 m ≥ 2 and p p p an odd prime, let M = PGL 2 ( u ) ≅ PSL 2 ( u ) M=\operatorname{PGL}_2(u)\cong\operatorname{PSL}_2(u) M = PGL 2 ( u ) ≅ PSL 2 ( u ) be the subfield subgroup of G = PSL 2 ( q ) G=\operatorname{PSL}_2(q) G = PSL 2 ( q ) , and let B B B be a union of M M M -orbits with ∣ B ∣ = k |B|=k ∣ B ∣ = k . If B G B^G B G is a 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design with λ > 0 \lambda>0 λ > 0 , then every one of the pair classes contains the same number of pairs of B B B ,
P C ( B ) = ( k 2 ) s = k ( k − 1 ) q − 2 , P_C(B)=\frac{\binom{k}{2}}{s}=\frac{k(k-1)}{q-2}, P C ( B ) = s ( 2 k ) = q − 2 k ( k − 1 ) , and
λ = ∣ G π ∣ P C ( B ) ∣ M ∣ = 4 k ( k − 1 ) ( q − 2 ) u ( u 2 − 1 ) . \lambda=\frac{|G_\pi|P_C(B)}{|M|}=\frac{4k(k-1)}{(q-2)u(u^2-1)}. λ = ∣ M ∣ ∣ G π ∣ P C ( B ) = ( q − 2 ) u ( u 2 − 1 ) 4 k ( k − 1 ) . In particular q − 2 q-2 q − 2 divides k ( k − 1 ) k(k-1) k ( k − 1 ) and u ( u 2 − 1 ) u(u^2-1) u ( u 2 − 1 ) divides 4 P C ( B ) 4P_C(B) 4 P C ( B ) , and λ \lambda λ is determined by k k k alone.
Proof. By Proposition 4.34 the action has s = ( q − 2 ) / 2 s=(q-2)/2 s = ( q − 2 ) /2 pair classes, each of length ∣ G ∣ / 4 |G|/4 ∣ G ∣/4 , and ∣ G π ∣ = 4 |G_{\pi}|=4 ∣ G π ∣ = 4 . By Proposition 4.30 the product ∣ G π ∣ P C ( B ) |G_{\pi}|P_C(B) ∣ G π ∣ P C ( B ) does not depend on C C C , so P C ( B ) P_C(B) P C ( B ) does not either, and ∑ C P C ( B ) = ( s 2 ) \sum_C P_C(B)=\binom{s}{2} ∑ C P C ( B ) = ( 2 s ) over the s s s classes gives the first formula. Substituting into the formula of Corollary 4.31 and using ∣ M ∣ = ∣ PGL 2 ( u ) ∣ = u ( u 2 − 1 ) |M|=|\operatorname{PGL}_2(u)|=u(u^2-1) ∣ M ∣ = ∣ PGL 2 ( u ) ∣ = u ( u 2 − 1 ) gives the second.
The proposition also accounts for the pair counts recorded earlier in this subsection, which were obtained there by direct enumeration in each degree. At q = 8 q=8 q = 8 it gives s = 3 s=3 s = 3 classes of 126 pairs, which is ( 2 8 2 ) = 378 \binom{2^8}{2}=378 ( 2 2 8 ) = 378 ; at q = 32 q=32 q = 32 it gives s = 15 s=15 s = 15 classes of 8184 pairs and μ = 4 \mu=4 μ = 4 , the figures used in the proof of Proposition 4.29 and in the self-check following Corollary 4.31 ; at q = 64 q=64 q = 64 it gives s = 31 s=31 s = 31 classes of 65520 pairs, which is ( 2016 2 ) = 2031120 \binom{2016}{2}=2031120 ( 2 2016 ) = 2031120 . In each case μ = ∣ G π ∣ = 4 \mu=|G_{\pi}|=4 μ = ∣ G π ∣ = 4 , so the values of μ \mu μ and s s s used in the arithmetic sieve of this subsection are instances of the general formulas rather than separate computations.
The value of λ \lambda λ now depends on k k k alone, and k k k in turn is a linear function of the orbit counts of the block. We record this parametrization and its first consequence, which removes every block consisting of a single orbit of M M M .
Lemma 4.36. In the situation of Corollary 4.35 let B B B consist of a a a orbits of the first type, b b b orbits of the second type and c c c orbits of the third type of Lemma 4.32 , where a ∈ { 0 , 1 } a\in\{0,1\} a ∈ { 0 , 1 } , 0 ≤ b ≤ u p − 1 − 1 0\le b\le u^{p-1}-1 0 ≤ b ≤ u p − 1 − 1 and 0 ≤ c ≤ u ( u 2 p − 2 − u p + u − 1 ) 2 ( u 2 − 1 ) 0\le c\le\frac{u(u^{2p-2}-u^p+u-1)}{2(u^2-1)} 0 ≤ c ≤ 2 ( u 2 − 1 ) u ( u 2 p − 2 − u p + u − 1 ) . Put
T = a + ( b + 2 c ) ( u + 1 ) . T=a+(b+2c)(u+1). T = a + ( b + 2 c ) ( u + 1 ) . Then k = u ( u − 1 ) 2 T k=\frac{u(u-1)}{2}T k = 2 u ( u − 1 ) T , and if B G B^G B G is a 2 2 2 -( v , k , λ ) ({v},{k},{\lambda}) ( v , k , λ ) design with λ > 0 \lambda>0 λ > 0 , then
λ = T ( u ( u − 1 ) T − 2 ) ( u p − 2 ) ( u + 1 ) . \lambda=\frac{T\bigl(u(u-1)T-2\bigr)}{(u^p-2)(u+1)}. λ = ( u p − 2 ) ( u + 1 ) T ( u ( u − 1 ) T − 2 ) .
Proof. By Corollary 4.33 ,
k = a u ( u − 1 ) 2 + b u ( u 2 − 1 ) 2 + c u ( u 2 − 1 ) = u ( u − 1 ) 2 ( a + ( b + 2 c ) ( u + 1 ) ) , k=a\frac{u(u-1)}{2}+b\frac{u(u^2-1)}{2}+cu(u^2-1)=\frac{u(u-1)}{2}\bigl(a+(b+2c)(u+1)\bigr), k = a 2 u ( u − 1 ) + b 2 u ( u 2 − 1 ) + c u ( u 2 − 1 ) = 2 u ( u − 1 ) ( a + ( b + 2 c ) ( u + 1 ) ) ,
since u 2 − 1 = ( u − 1 ) ( u + 1 ) u^2-1=(u-1)(u+1) u 2 − 1 = ( u − 1 ) ( u + 1 ) . Substituting this value of k k k into the formula of Corollary 4.35 and using u p = q u^p=q u p = q and u 2 − 1 = ( u − 1 ) ( u + 1 ) u^2-1=(u-1)(u+1) u 2 − 1 = ( u − 1 ) ( u + 1 ) ,
λ = 4 k ( k − 1 ) ( q − 2 ) u ( u 2 − 1 ) = 4 ⋅ u ( u − 1 ) T 2 ( u ( u − 1 ) T 2 − 1 ) ( q − 2 ) u ( u − 1 ) ( u + 1 ) = T ( u ( u − 1 ) T − 2 ) ( q − 2 ) ( u + 1 ) . \lambda=\frac{4k(k-1)}{(q-2)u(u^2-1)}=\frac{4\cdot\frac{u(u-1)T}{2}\left(\frac{u(u-1)T}{2}-1\right)}{(q-2)u(u-1)(u+1)}=\frac{T\bigl(u(u-1)T-2\bigr)}{(q-2)(u+1)}. λ = ( q − 2 ) u ( u 2 − 1 ) 4 k ( k − 1 ) = ( q − 2 ) u ( u − 1 ) ( u + 1 ) 4 ⋅ 2 u ( u − 1 ) T ( 2 u ( u − 1 ) T − 1 ) = ( q − 2 ) ( u + 1 ) T ( u ( u − 1 ) T − 2 ) .
Corollary 4.37. Let q = u p q = u^p q = u p with u = 2 m u = 2^m u = 2 m , m ≥ 2 m \ge2 m ≥ 2 and p p p an odd prime, and let B B B be a single orbit of M M M on P \mathcal{P} P , of any of the three types. Then B G B^G B G is not a 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design with λ > 0 \lambda> 0 λ > 0 .
Proof. Apply Lemma 4.36 to the three admissible triples ( a , b , c ) = ( 1 , 0 , 0 ) (a,b,c) = (1,0,0) ( a , b , c ) = ( 1 , 0 , 0 ) , ( 0 , 1 , 0 ) (0,1,0) ( 0 , 1 , 0 ) and ( 0 , 0 , 1 ) (0,0,1) ( 0 , 0 , 1 ) , for which T = 1 T = 1 T = 1 , u + 1 u+1 u + 1 and 2 ( u + 1 ) 2(u+1) 2 ( u + 1 ) . This gives
λ = u − 2 u p − 2 , λ = u 3 − u − 2 u p − 2 , λ = 4 ( u 3 − u − 1 ) u p − 2 . \lambda= \frac{u-2}{u^p-2}, \qquad\lambda= \frac{u^3-u-2}{u^p-2}, \qquad\lambda= \frac{4(u^3-u-1)}{u^p-2}. λ = u p − 2 u − 2 , λ = u p − 2 u 3 − u − 2 , λ = u p − 2 4 ( u 3 − u − 1 ) . In the first case 0 ≤ u − 2 < u p − 2 0 \le u-2 < u^p-2 0 ≤ u − 2 < u p − 2 because u < u p u < u^p u < u p , so λ \lambda λ lies in [ 0 , 1 ) [0,1) [ 0 , 1 ) and, being an integer, equals 0 only at u = 2 u=2 u = 2 , when the orbit is a single point. In the second case 0 < u 3 − u − 2 < u p − 2 0 < u^3-u-2 < u^p-2 0 < u 3 − u − 2 < u p − 2 , so 0 < λ < 1 0 < \lambda< 1 0 < λ < 1 . In the third case u 3 − u − 1 = ( u 3 − 2 ) − ( u − 1 ) u^3-u-1 = (u^3-2)-(u-1) u 3 − u − 1 = ( u 3 − 2 ) − ( u − 1 ) and 0 < 4 ( u − 1 ) < u 3 − 2 0 < 4(u-1) < u^3-2 0 < 4 ( u − 1 ) < u 3 − 2 , so λ = 4 − 4 ( u − 1 ) / ( u 3 − 2 ) \lambda= 4 - 4(u-1)/(u^3-2) λ = 4 − 4 ( u − 1 ) / ( u 3 − 2 ) lies strictly between 3 and 4 when p = 3 p=3 p = 3 , while 0 < 4 ( u 3 − u − 1 ) < u 5 − 2 ≤ u p − 2 0 < 4(u^3-u-1) < u^5-2 \le u^p-2 0 < 4 ( u 3 − u − 1 ) < u 5 − 2 ≤ u p − 2 when p ≥ 5 p \ge5 p ≥ 5 , so then 0 < λ < 1 0 < \lambda< 1 0 < λ < 1 . A design has λ \lambda λ a positive integer, so none of the three single-orbit blocks qualifies.
Corollary 4.38. Let q = u 3 q = u^3 q = u 3 with u = 2 m u = 2^m u = 2 m , m ≥ 2 m \ge2 m ≥ 2 , and let B B B be the union of exactly two orbits of M M M on P \mathcal{P} P . Then B G B^G B G is not a 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design with λ > 0 \lambda> 0 λ > 0 .
Proof. Since there is exactly one orbit of the first type, the two orbits are of types A A A and s p sp s p , A A A and g e n gen g e n , s p sp s p and s p sp s p , s p sp s p and g e n gen g e n , or g e n gen g e n and g e n gen g e n , so by Lemma 4.36 the relevant values of T T T are
u + 2 , 2 u + 3 , 2 u + 2 , 3 u + 3 , 4 u + 4. u+2,\qquad2u+3,\qquad2u+2,\qquad3u+3,\qquad4u+4. u + 2 , 2 u + 3 , 2 u + 2 , 3 u + 3 , 4 u + 4. Put D = ( u 3 − 2 ) ( u + 1 ) D=(u^3-2)(u+1) D = ( u 3 − 2 ) ( u + 1 ) , so that D λ = T ( u ( u − 1 ) T − 2 ) D\lambda=T\bigl(u(u-1)T-2\bigr) D λ = T ( u ( u − 1 ) T − 2 ) , and subtract from the numerator the multiple of D D D , or of u 3 − 2 u^3-2 u 3 − 2 , indicated in each line:
Table 15.
For T = 2 u + 2 T=2u+2 T = 2 u + 2 , T = 3 u + 3 T=3u+3 T = 3 u + 3 and T = 4 u + 4 T=4u+4 T = 4 u + 4 the factor u + 1 u+1 u + 1 cancels between the numerator and D = ( u 3 − 2 ) ( u + 1 ) D=(u^3-2)(u+1) D = ( u 3 − 2 ) ( u + 1 ) , which is how the last three numerators and the last three subtracted terms above are written.
In every line the remainder is nonzero and its absolute value is smaller than the modulus used in that line, which is D D D in the first two lines and u 3 − 2 u^3-2 u 3 − 2 in the last three. For the first line, 0 < 2 u 3 − 4 u − 2 0 < 2u^3-4u-2 0 < 2 u 3 − 4 u − 2 and D − ( 2 u 3 − 4 u − 2 ) = u 4 − u 3 + 2 u > 0 D-(2u^3-4u-2)=u^4-u^3+2u>0 D − ( 2 u 3 − 4 u − 2 ) = u 4 − u 3 + 2 u > 0 . For the second, 4 u 3 − 3 u 2 − 5 u + 2 4u^3-3u^2-5u+2 4 u 3 − 3 u 2 − 5 u + 2 equals 12 at u = 2 u=2 u = 2 and increases with u u u , while D − ( 4 u 3 − 3 u 2 − 5 u + 2 ) = u 4 − 3 u 3 + 3 u 2 + 3 u − 4 D-(4u^3-3u^2-5u+2)=u^4-3u^3+3u^2+3u-4 D − ( 4 u 3 − 3 u 2 − 5 u + 2 ) = u 4 − 3 u 3 + 3 u 2 + 3 u − 4 equals 6 at u = 2 u=2 u = 2 and for u ≥ 3 u\geq3 u ≥ 3 exceeds u 3 ( u − 3 ) + 3 u 2 + 3 u − 4 > 0 u^3(u-3)+3u^2+3u-4>0 u 3 ( u − 3 ) + 3 u 2 + 3 u − 4 > 0 . For the last three, 4 u − 4 < u 3 − 2 4u-4<u^3-2 4 u − 4 < u 3 − 2 , and 9 u − 12 < u 3 − 2 9u-12<u^3-2 9 u − 12 < u 3 − 2 , 16 u − 24 < u 3 − 2 16u-24<u^3-2 16 u − 24 < u 3 − 2 for u ≥ 4 u\geq4 u ≥ 4 . Since u = 2 m u=2^m u = 2 m with m ≥ 2 m\geq2 m ≥ 2 we have u ≥ 4 u\geq4 u ≥ 4 throughout, so in each line a nonzero remainder of absolute value below the modulus is not divisible by it, and λ \lambda λ is not an integer.
Corollary 4.39. Let q = u p q=u^p q = u p with u = 2 m ≥ 4 u=2^m\geq4 u = 2 m ≥ 4 and p ≥ 5 p\geq5 p ≥ 5 prime, and let B B B be a union of orbits of M M M on P \mathcal{P} P with
T 2 u ( u − 1 ) ≤ ( u p − 2 ) ( u + 1 ) . T^2u(u-1)\leq(u^p-2)(u+1). T 2 u ( u − 1 ) ≤ ( u p − 2 ) ( u + 1 ) . Then B G B^G B G is not a 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design with λ > 0 \lambda>0 λ > 0 . In particular no union of at most
n 0 ( u , p ) = ⌊ ( u p − 2 ) ( u + 1 ) / ( u ( u − 1 ) ) − 1 2 ( u + 1 ) ⌋ n_0(u,p)=\left\lfloor\frac{\sqrt{(u^p-2)(u+1)/(u(u-1))}-1}{2(u+1)}\right\rfloor n 0 ( u , p ) = ⌊ 2 ( u + 1 ) ( u p − 2 ) ( u + 1 ) / ( u ( u − 1 )) − 1 ⌋ orbit(s) qualifies.
Proof. By Lemma 4.36 one has k = u ( u − 1 ) T / 2 k=u(u-1)T/2 k = u ( u − 1 ) T /2 with T = a + ( b + 2 c ) ( u + 1 ) ≥ 1 T=a+(b+2c)(u+1)\geq1 T = a + ( b + 2 c ) ( u + 1 ) ≥ 1 , and 0 < u ( u − 1 ) T − 2 < T u ( u − 1 ) 0<u(u-1)T-2<Tu(u-1) 0 < u ( u − 1 ) T − 2 < T u ( u − 1 ) as soon as u ≥ 4 u\geq4 u ≥ 4 and T ≥ 1 T\geq1 T ≥ 1 . Under the hypothesis therefore
0 < λ = T ( u ( u − 1 ) T − 2 ) ( u p − 2 ) ( u + 1 ) < T 2 u ( u − 1 ) ( u p − 2 ) ( u + 1 ) ≤ 1 , 0<\lambda=\frac{T(u(u-1)T-2)}{(u^p-2)(u+1)}<\frac{T^2u(u-1)}{(u^p-2)(u+1)}\leq1, 0 < λ = ( u p − 2 ) ( u + 1 ) T ( u ( u − 1 ) T − 2 ) < ( u p − 2 ) ( u + 1 ) T 2 u ( u − 1 ) ≤ 1 ,
so λ \lambda λ is not a positive integer. A union of at most n n n orbits has T = a + ( b + 2 c ) ( u + 1 ) T=a+(b+2c)(u+1) T = a + ( b + 2 c ) ( u + 1 ) with a ∈ { 0 , 1 } a\in\{0,1\} a ∈ { 0 , 1 } and b + c ≤ n b+c\leq n b + c ≤ n , hence T ≤ 1 + 2 n ( u + 1 ) T\leq1+2n(u+1) T ≤ 1 + 2 n ( u + 1 ) , the value at b = 0 b=0 b = 0 , c = n c=n c = n . Consequently T 2 u ( u − 1 ) ≤ ( 1 + 2 n ( u + 1 ) ) 2 u ( u − 1 ) T^2u(u-1)\leq(1+2n(u+1))^2u(u-1) T 2 u ( u − 1 ) ≤ ( 1 + 2 n ( u + 1 ) ) 2 u ( u − 1 ) , which does not exceed ( u p − 2 ) ( u + 1 ) (u^p-2)(u+1) ( u p − 2 ) ( u + 1 ) as soon as 1 + 2 n ( u + 1 ) ≤ ( u p − 2 ) ( u + 1 ) / ( u ( u − 1 ) ) 1+2n(u+1)\leq\sqrt{(u^p-2)(u+1)/(u(u-1))} 1 + 2 n ( u + 1 ) ≤ ( u p − 2 ) ( u + 1 ) / ( u ( u − 1 )) , that is, as soon as n ≤ n 0 ( u , p ) n\leq n_0(u,p) n ≤ n 0 ( u , p ) . The factor 2 is the coefficient of the third-type orbits in T T T , which enter with weight 2 ( u + 1 ) 2(u+1) 2 ( u + 1 ) rather than u + 1 u+1 u + 1 ; a bound T ≤ 1 + n ( u + 1 ) T\leq1+n(u+1) T ≤ 1 + n ( u + 1 ) , which counts them as second-type orbits, is too small by a factor up to 2 and does not yield the stated range.
Lemma 4.40 (rigidity of an admissible orbit union). Let q = u p q = u^p q = u p with u = 2 m u = 2^m u = 2 m , m ≥ 2 m \ge2 m ≥ 2 and p p p an odd prime, and let B B B be a union of orbits of M M M on P \mathcal{P} P which is a block of a 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design D = B G \mathcal{D} = B^G D = B G with λ > 0 \lambda> 0 λ > 0 . Then
r = T ( u p + 1 ) u + 1 , b = q ( q 2 − 1 ) u ( u 2 − 1 ) = ∣ G ∣ ∣ M ∣ , λ = T ( u ( u − 1 ) T − 2 ) ( u p − 2 ) ( u + 1 ) . r = \frac{T(u^p+1)}{u+1}, \qquad b = \frac{q(q^2-1)}{u(u^2-1)} = \frac{|G|}{|M|}, \qquad\lambda= \frac{T(u(u-1)T-2)}{(u^p-2)(u+1)}. r = u + 1 T ( u p + 1 ) , b = u ( u 2 − 1 ) q ( q 2 − 1 ) = ∣ M ∣ ∣ G ∣ , λ = ( u p − 2 ) ( u + 1 ) T ( u ( u − 1 ) T − 2 ) . and the setwise stabilizer of B B B in G G G is exactly M M M . In particular the number of blocks does not depend on the block size: every admissible block of the subfield family produces a design with the same q ( q 2 − 1 ) u ( u 2 − 1 ) \frac{q(q^2-1)}{u(u^2-1)} u ( u 2 − 1 ) q ( q 2 − 1 ) blocks.
Proof. The number of blocks of a 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design is b = λ v ( v − 1 ) / ( k ( k − 1 ) ) b = \lambda v(v-1)/(k(k-1)) b = λ v ( v − 1 ) / ( k ( k − 1 )) ; with v = q ( q − 1 ) / 2 v = q(q-1)/2 v = q ( q − 1 ) /2 , k = u ( u − 1 ) T / 2 k = u(u-1)T/2 k = u ( u − 1 ) T /2 and the value of λ \lambda λ in Lemma 4.36 ,
b = T ( u ( u − 1 ) T − 2 ) ( u p − 2 ) ( u + 1 ) ⋅ q ( q − 1 ) ( q 2 − q − 2 ) 4 ⋅ 4 u ( u − 1 ) T ( u ( u − 1 ) T − 2 ) = q ( q − 1 ) ( q 2 − q − 2 ) ( q − 2 ) ( u + 1 ) u ( u − 1 ) = q ( q 2 − 1 ) u ( u 2 − 1 ) , \begin{aligned}
b &= \frac{T(u(u-1)T-2)}{(u^p-2)(u+1)} \cdot\frac{q(q-1)(q^2-q-2)}{4} \cdot\frac{4}{u(u-1)T(u(u-1)T-2)} \\
&= \frac{q(q-1)(q^2-q-2)}{(q-2)(u+1)u(u-1)} = \frac{q(q^2-1)}{u(u^2-1)},
\end{aligned} b = ( u p − 2 ) ( u + 1 ) T ( u ( u − 1 ) T − 2 ) ⋅ 4 q ( q − 1 ) ( q 2 − q − 2 ) ⋅ u ( u − 1 ) T ( u ( u − 1 ) T − 2 ) 4 = ( q − 2 ) ( u + 1 ) u ( u − 1 ) q ( q − 1 ) ( q 2 − q − 2 ) = u ( u 2 − 1 ) q ( q 2 − 1 ) ,
since q 2 − q − 2 = ( q − 2 ) ( q + 1 ) q^2-q-2 = (q-2)(q+1) q 2 − q − 2 = ( q − 2 ) ( q + 1 ) and q = u p q = u^p q = u p . On the other hand D = B G \mathcal{D} = B^G D = B G has b = ∣ G ∣ / ∣ Stab G ( B ) ∣ b = |G|/|\operatorname{Stab}_G(B)| b = ∣ G ∣/∣ Stab G ( B ) ∣ blocks, distinct blocks being the cosets of the setwise stabilizer, while ∣ G ∣ = q ( q 2 − 1 ) |G| = q(q^2-1) ∣ G ∣ = q ( q 2 − 1 ) and ∣ M ∣ = u ( u 2 − 1 ) |M| = u(u^2-1) ∣ M ∣ = u ( u 2 − 1 ) . Hence ∣ Stab G ( B ) ∣ = u ( u 2 − 1 ) = ∣ M ∣ |\operatorname{Stab}_G(B)| = u(u^2-1) = |M| ∣ Stab G ( B ) ∣ = u ( u 2 − 1 ) = ∣ M ∣ , and M ⊆ Stab G ( B ) M \subseteq\operatorname{Stab}_G(B) M ⊆ Stab G ( B ) forces Stab G ( B ) = M \operatorname{Stab}_G(B) = M Stab G ( B ) = M . The displayed value of r r r is λ ( v − 1 ) / ( k − 1 ) \lambda(v-1)/(k-1) λ ( v − 1 ) / ( k − 1 ) . □
Remark 4.41. At u = 2 u = 2 u = 2 , that is q = 8 q = 8 q = 8 , the last of these configurations is genuine: one orbit of length 3 3 3 together with one of the three orbits of length 6 6 6 of M = S 3 M = S_3 M = S 3 give k = 9 k = 9 k = 9 and λ = 8 \lambda= 8 λ = 8 , a union recorded in Table 7 which is however not block-primitive, its block set being partitioned into seven parts by the subgroup D 18 ≤ G D_{18} \leq G D 18 ≤ G (Proposition 4.28 ). This is the same degeneracy of the first type noted after Corollary 4.33 , where L A = 1 L_A = 1 L A = 1 and 2 ( u + 1 ) = ∣ M ∣ 2(u+1) = |M| 2 ( u + 1 ) = ∣ M ∣ .
The enumeration of Table 8 does not reach the block stabilizer A 5 ≅ P G L 2 ( 4 ) A_5 \cong PGL_2(4) A 5 ≅ P G L 2 ( 4 ) at q = 64 q = 64 q = 64 , and this case is settled here: the class criterion restricts the block size to eight values, and at the two values which survive it the orbits of the third type are few enough to be searched exhaustively. The stabilizer S 3 S_3 S 3 is void at every even degree (Lemma 4.7 ), in particular at q = 32 q = 32 q = 32 ; for the subfield stabilizer P S L 2 ( r ) PSL_2(r) P S L 2 ( r ) , r = 2 d ≥ 4 r = 2^d \geq4 r = 2 d ≥ 4 , q = r p q = r^p q = r p with p p p an odd prime, every block is a union of orbits of the three types of Lemma 4.32 , and unions of at most two orbits are removed for p = 3 p = 3 p = 3 by Corollary 4.38 , while for p ≥ 5 p \ge5 p ≥ 5 Corollary 4.39 removes every union with T 2 u ( u − 1 ) ≤ ( u p − 2 ) ( u + 1 ) T^{2}u(u-1) \le(u^{p}-2)(u+1) T 2 u ( u − 1 ) ≤ ( u p − 2 ) ( u + 1 ) , hence every union of at most n 0 ( u , p ) n_{0}(u,p) n 0 ( u , p ) orbits; what lies beyond the reach of this argument is a union of more than n 0 ( u , p ) n_{0}(u,p) n 0 ( u , p ) orbits, reduced to the class equations of Proposition 4.34 ; for every admissible union Lemma 4.40 fixes the number of blocks and the replication number in closed form, so that the remaining case is governed by the class equations of Proposition 4.34 alone. The sieve that settles q = 64 q = 64 q = 64 below does not extend to p ≥ 5 p \ge5 p ≥ 5 : at u = 4 u = 4 u = 4 , p = 3 p = 3 p = 3 the integrality of λ \lambda λ leaves eight sizes besides the empty and the complete block, splitting into 46 orbit profiles over the 42 orbits of M M M and 31 class equations, while at u = 4 u = 4 u = 4 , p = 5 p = 5 p = 5 the same sieve leaves 275 values of T T T , hence 34 231 orbit profiles over 8 857 orbits governed by 511 class equations, and at u = 8 u = 8 u = 8 , p = 5 p = 5 p = 5 it leaves 1 041 values and 2 122 935 profiles over 1 063 236 orbits.
Proposition 4.42 (q = 64 q = 64 q = 64 , subfield block stabilizer). Let q = 64 q = 64 q = 64 , v = 2016 v = 2016 v = 2016 , G = P S L 2 ( 64 ) G = PSL_{2}(64) G = P S L 2 ( 64 ) , and let M ≅ A 5 = P S L 2 ( 4 ) M \cong A_{5} = PSL_{2}(4) M ≅ A 5 = P S L 2 ( 4 ) be a subfield subgroup of G G G . Let B B B be a union of orbits of M M M on P \mathcal{P} P with k = ∣ B ∣ ≤ v − 2 k = |B| \le v - 2 k = ∣ B ∣ ≤ v − 2 . Then B G B^{G} B G is not a 2 2 2 -( 2016 , k , λ ) (2016,k,\lambda) ( 2016 , k , λ ) design with λ > 0 \lambda> 0 λ > 0 . The only union of orbits of M M M which satisfies the class equalities of Proposition 4.30 is P \mathcal{P} P itself, and it is fixed by G G G .
Proof. By Lemma 4.32 with u = 4 u = 4 u = 4 the group M M M has 42 orbits on P \mathcal{P} P , namely one of length 6, fifteen of length 30 and twenty-six of length 60; so k = 6 a + 30 b + 60 c k = 6a + 30b + 60c k = 6 a + 30 b + 60 c with a ∈ { 0 , 1 } a \in\{0,1\} a ∈ { 0 , 1 } , 0 ≤ b ≤ 15 0 \le b \le15 0 ≤ b ≤ 15 and 0 ≤ c ≤ 26 0 \le c \le26 0 ≤ c ≤ 26 . By Proposition 4.34 the group G G G has 31 classes of pairs, each of 65520 = ( 2016 2 ) / 31 65520 = \binom{2016}{2}/31 65520 = ( 2 2016 ) /31 pairs, and by Corollary 4.35 a union B B B is a block exactly when P C ( B ) P_{\mathcal{C}}(B) P C ( B ) takes the same value on all 31 classes, that value being k ( k − 1 ) / ( q − 2 ) = k ( k − 1 ) / 62 k(k-1)/(q-2) = k(k-1)/62 k ( k − 1 ) / ( q − 2 ) = k ( k − 1 ) /62 ; in particular 31 divides k ( k − 1 ) k(k-1) k ( k − 1 ) .
The values k < 2016 k < 2016 k < 2016 of the form 6 a + 30 b + 60 c 6a + 30b + 60c 6 a + 30 b + 60 c with 31 dividing k ( k − 1 ) k(k-1) k ( k − 1 ) are
156, 186, 900, 930, 1086, 1116, 1830, 1860.
For each of them the orbit indicators x i x_i x i must satisfy
∑ i w i x i = k , ∑ i ≤ j N i j ( C ) x i x j = k ( k − 1 ) 62 for each of the 31 classes, \sum_i w_i x_i = k,\qquad\sum_{i \le j} N_{ij}(\mathcal{C})x_i x_j = \frac{k(k-1)}{62}\quad\text{for each of the 31 classes,} i ∑ w i x i = k , i ≤ j ∑ N ij ( C ) x i x j = 62 k ( k − 1 ) for each of the 31 classes,
where w i w_i w i is the length of the i i i -th orbit and N i j ( C ) N_{ij}(\mathcal{C}) N ij ( C ) the number of pairs of C \mathcal{C} C with one point in the i i i -th and one point in the j j j -th orbit. For k = 156 , 186 , 930 , 1086 , 1830 , 1860 k = 156, 186, 930, 1086, 1830, 1860 k = 156 , 186 , 930 , 1086 , 1830 , 1860 this 0-1 system has no solution. For k = 900 k = 900 k = 900 the size equation forces a = 0 a = 0 a = 0 and b + 2 c = 30 b + 2c = 30 b + 2 c = 30 , and for k = 1116 k = 1116 k = 1116 it forces a = 1 a = 1 a = 1 and b + 2 c = 37 b + 2c = 37 b + 2 c = 37 ; each of these two block sizes therefore splits into 8 profiles ( a , b , c ) (a, b, c) ( a , b , c ) , and in each profile the 31 equations, once the orbits of the second type are fixed, involve only the c c c indicators of the orbits of the third type. All 16 profiles are infeasible. Twelve of them are decided by a search over the orbits of both types at once; the remaining four, ( 0 , 6 , 12 ) (0, 6, 12) ( 0 , 6 , 12 ) , ( 0 , 10 , 10 ) (0, 10, 10) ( 0 , 10 , 10 ) , ( 1 , 5 , 16 ) (1, 5, 16) ( 1 , 5 , 16 ) and ( 1 , 9 , 14 ) (1, 9, 14) ( 1 , 9 , 14 ) , are decided by fixing the orbits of the second type one choice at a time and solving the resulting system in the orbits of the third type, which is 16 016 systems in all.
The search is exhaustive, and each verdict is a certificate of branch and bound over 0-1 variables rather than a sample: the six systems at k = 156 , 186 , 930 , 1086 , 1830 , 1860 k = 156, 186, 930, 1086, 1830, 1860 k = 156 , 186 , 930 , 1086 , 1830 , 1860 are returned infeasible by two independent solvers, and the 16 016 systems of the four profiles by a solver whose encoding is calibrated on the control below. The control is the union of all 42 orbits, that is k = v = 2016 k = v = 2016 k = v = 2016 : for it the same encoding returns the single solution x i = 1 x_i = 1 x i = 1 for every i i i , so infeasibility is distinguished from a mis-encoded system.
At k = v k = v k = v the union is P \mathcal{P} P itself, whose setwise stabilizer is G G G and not M M M . The block set then consists of a single block and the design is the complete one, with automorphism group Sym ( v ) \operatorname{Sym}(v) Sym ( v ) ; a complete design is excluded by the hypothesis that the design be non-trivial with block stabilizer M M M a proper subgroup of G G G , and it is precisely the case in which the class equalities hold, one block being all of P \mathcal{P} P .
Classical groups (Lie type)
The classical socle comes first, and only for the non-2-homogeneous point action.
For classical simple groups, we use the Aschbacher classification of maximal subgroups as the framework. Let T T T be a classical simple group, and T ≤ G ≤ Aut ( T ) T \leq G \leq\operatorname{Aut}(T) T ≤ G ≤ Aut ( T ) . If T B = G B ∩ T T_B = G_B \cap T T B = G B ∩ T is a maximal subgroup of T T T , then it belongs to one of the geometric families C i C_i C i (1 ≤ i ≤ 8 1 \leq i \leq8 1 ≤ i ≤ 8 ) or the non-geometric family S S S .
Scope: non-2-homogeneous point actions
We first note the standard 2-transitive actions of classical groups:
PSL 2 ( q ) \operatorname{PSL}_2(q) PSL 2 ( q ) on the projective line P G ( 1 , q ) PG(1,q) P G ( 1 , q ) , 2-transitive;
P S L n ( q ) PSL_n(q) P S L n ( q ) on projective points P G ( n − 1 , q ) PG(n-1,q) P G ( n − 1 , q ) , 2-transitive for n ≥ 3 n \ge3 n ≥ 3 ;
S p 2 n ( 2 ) Sp_{2n}(2) S p 2 n ( 2 ) on the non-zero vectors of the natural module: this action has rank 3 and is 2-transitive only for n = 1 n=1 n = 1 ; for n = 2 n=2 n = 2 the isomorphism S p 4 ( 2 ) ≅ S 6 Sp_4(2) \cong S_6 S p 4 ( 2 ) ≅ S 6 yields a separate 2-transitive action on 6 points.
These 2-transitive cases give automatic 2-designs and are not discussed in detail below. We focus on non-2-homogeneous point actions, where the design condition requires verification.
The Aschbacher classes
Every maximal subgroup of a classical group lies in one of the classes C 1 C_1 C 1 –C 8 C_8 C 8 or in the almost simple class S S S of Aschbacher’s theorem, in the numbering of that theorem. The geometric classes C 1 C_1 C 1 (subspace stabilizers) and C 8 C_8 C 8 (form stabilizers) come first, since their orbits carry the geometric designs; the remaining classes have index too large to satisfy the ordering principle and are removed in the family-by-family analysis.
Screening the Aschbacher classes
Let G G G be one of the four classical groups treated below, H H H the stabilizer of a point and M M M the stabilizer of a block. By Aschbacher’s theorem every maximal subgroup of G G G lies in one of the classes C 1 C_1 C 1 –C 8 C_8 C 8 or in the class S S S ; we screen these classes in the numbering of that theorem, applying the ordering principle (Lemma 4.3 ) first, so that only pairs with ∣ H ∣ > ∣ K ∣ |H| > |K| ∣ H ∣ > ∣ K ∣ need to be examined.
Two families survive this filter. In class C 1 C_1 C 1 the parabolic chains P k < P l P_k < P_l P k < P l fix a subspace and act on the subspaces of one dimension contained in it; these are the generalized Grassmann candidates, and the constant-coverage test removes them in every family (the count of blocks through a pair depends on the dimension of the intersection, as computed for P S L n ( q ) PSL_n(q) P S L n ( q ) below). The rank-3 actions of the classes C 1 C_1 C 1 and C 8 C_8 C 8 carry the suborbit designs; those are symmetric, so they do not enter the present classification.
For the remaining classes the block stabilizer fixes a direct sum decomposition, a subfield, a tensor structure or a non-geometric simple subgroup. Its orbits on the point set then split into at least two types, and a fixed point pair meets blocks of one type while missing blocks of the other, so the coverage number varies and the 2-design condition fails. This removes C 2 C_2 C 2 , C 3 C_3 C 3 , C 4 C_4 C 4 , C 5 C_5 C 5 , C 6 C_6 C 6 , C 7 C_7 C 7 and S S S in all four families, and the outcome is collected in Subsection 4.11.
PSL(n,q), n at least 3
2-transitive case (excluded)
The natural action of P S L n ( q ) PSL_n(q) P S L n ( q ) on projective points P G ( n − 1 , q ) PG(n-1,q) P G ( n − 1 , q ) is 2-transitive for n ≥ 3 n \ge3 n ≥ 3 , with:
v = q n − 1 q − 1 . v=\frac{q^n-1}{q-1}. v = q − 1 q n − 1 .
This is the C 1 C_1 C 1 parabolic subgroup P 1 P_1 P 1 as point stabilizer. We exclude this 2-transitive case and focus on non-2-homogeneous point actions.
Non-2-transitive case: k k k -subspace action (1 < k < n − 1 1<k<n-1 1 < k < n − 1 )
The most important non-2-homogeneous action is the action on k k k -dimensional subspaces of F q n \mathbb{F}_q^n F q n , with:
v = [ n k ] q = ( q n − 1 ) ( q n − 1 − 1 ) ⋯ ( q n − k + 1 − 1 ) ( q k − 1 ) ( q k − 1 − 1 ) ⋯ ( q − 1 ) . v=\begin{bmatrix} n \\ k \end{bmatrix}_q
=\frac{(q^n-1)(q^{n-1}-1)\cdots(q^{n-k+1}-1)}
{(q^k-1)(q^{k-1}-1)\cdots(q-1)}. v = [ n k ] q = ( q k − 1 ) ( q k − 1 − 1 ) ⋯ ( q − 1 ) ( q n − 1 ) ( q n − 1 − 1 ) ⋯ ( q n − k + 1 − 1 ) .
The point stabilizer is the parabolic subgroup P k P_k P k .
Candidate block stabilizer: P k + 1 P_{k+1} P k + 1 (stabilizer of ( k + 1 ) (k+1) ( k + 1 ) -subspaces). Then:
b = [ n k + 1 ] q . b=\begin{bmatrix} n \\ k+1 \end{bmatrix}_q. b = [ n k + 1 ] q .
Fisher’s inequality b > v b>v b > v is equivalent to:
[ n k + 1 ] q > [ n k ] q ⟹ q n − k − 1 q k + 1 − 1 > 1 ⟹ n > 2 k + 1. \begin{bmatrix} n \\ k+1 \end{bmatrix}_q>
\begin{bmatrix} n \\ k \end{bmatrix}_q
\Longrightarrow
\frac{q^{n-k}-1}{q^{k+1}-1}>1
\Longrightarrow n>2k+1. [ n k + 1 ] q > [ n k ] q ⟹ q k + 1 − 1 q n − k − 1 > 1 ⟹ n > 2 k + 1.
Orbit structure: Fix a k k k -subspace U U U stabilized by H H H . The block stabilizer M = P k + 1 M=P_{k+1} M = P k + 1 fixes a ( k + 1 ) (k+1) ( k + 1 ) -subspace W W W . The orbits of M M M on the set of all k k k -subspaces P \mathcal{P} P are classified by the intersection dimension with W W W :
Δ 1 \Delta_1 Δ 1 : k k k -subspaces contained in W W W , of size [ k + 1 k ] q = q k + q k − 1 + ⋯ + 1 \begin{bmatrix} k+1 \\ k \end{bmatrix}_q=q^k+q^{k-1}+\cdots+1 [ k + 1 k ] q = q k + q k − 1 + ⋯ + 1 ;
Δ 2 \Delta_2 Δ 2 : k k k -subspaces meeting W W W in dimension k − 1 k-1 k − 1 ;
further orbits, indexed by the remaining values of the intersection dimension.
Design candidate: B = Δ 1 B = \Delta_1 B = Δ 1 The block B B B consists of all k k k -subspaces contained in the fixed ( k + 1 ) (k+1) ( k + 1 ) -subspace W W W . This candidate fails the constant-coverage test: the number of blocks through a point pair { U , V } \{U,V\} { U , V } of k k k -subspaces equals the number of ( k + 1 ) (k+1) ( k + 1 ) -subspaces containing U ∨ V U \lor V U ∨ V , namely the Gaussian binomial coefficient ( n − ( 2 k − d ) ( k + 1 ) − ( 2 k − d ) ) q \binom{n-(2k-d)}{(k+1)-(2k-d)}_q ( ( k + 1 ) − ( 2 k − d ) n − ( 2 k − d ) ) q when 2 k − d ≤ k + 1 2k-d \le k+1 2 k − d ≤ k + 1 , and 0 0 0 otherwise, where d = dim ( U ∩ V ) d = \dim(U \cap V) d = dim ( U ∩ V ) ; since d d d assumes several values on the point set, the coverage number is not constant and the 2-design condition fails. Hence no generalized Grassmann design arises from this candidate.
PSp( 2 n , q ) (2n,q) ( 2 n , q )
The projective symplectic action
For n = 2 n = 2 n = 2 and q = 2 q = 2 q = 2 one has P S p 4 ( 2 ) ≅ S 6 PSp_4(2) \cong S_6 P S p 4 ( 2 ) ≅ S 6 ; the natural symplectic projective actions below are rank-3 for n ≥ 2 n \ge2 n ≥ 2 (they are 2-transitive only for n = 1 n = 1 n = 1 ).
Rank-3 suborbit structure and suborbit designs
The natural action of P S p 2 n ( q ) PSp_{2n}(q) P S p 2 n ( q ) is on the set of all 1-dimensional subspaces of the symplectic space F q 2 n \mathbb{F}_q^{2n} F q 2 n . This is a rank-3 action:
The point stabilizer is a parabolic subgroup P 1 P_1 P 1 (stabilizer of a totally singular 1-space). The degree is:
v = q 2 n − 1 q − 1 = 1 + q + ⋯ + q 2 n − 1 . v = \frac{q^{2n}-1}{q-1} = 1 + q + \cdots+ q^{2n-1}. v = q − 1 q 2 n − 1 = 1 + q + ⋯ + q 2 n − 1 .
Rank-3 structure: Fix a point (a 1-space) ⟨ v ⟩ \langle v\rangle ⟨ v ⟩ . The other points split into two orbits:
Δ 1 \Delta_1 Δ 1 : points orthogonal to v v v , that is, ⟨ w ⟩ \langle w\rangle ⟨ w ⟩ with ⟨ v , w ⟩ = 0 \langle v,w\rangle= 0 ⟨ v , w ⟩ = 0 , of size q ( q 2 n − 2 − 1 ) / ( q − 1 ) q(q^{2n-2}-1)/(q-1) q ( q 2 n − 2 − 1 ) / ( q − 1 ) ;
Δ 2 \Delta_2 Δ 2 : points non-orthogonal to v v v , of size q 2 n − 1 q^{2n-1} q 2 n − 1 .
This is a rank-3 permutation group, so both suborbits automatically give 2-designs (the point graph and its complement). These suborbit designs are symmetric (b = v b = v b = v , since the block stabilizer of a suborbit coincides with the point stabilizer), hence they are outside the non-symmetric classification of this paper; they are recorded here for completeness.
Design from suborbit Δ 1 \Delta_1 Δ 1 : Taking Δ 1 \Delta_1 Δ 1 as the block set gives a 2-design with parameters:
2 - ( v , k , λ ) , 2\text{-}(v,k,\lambda), 2 - ( v , k , λ ) ,
where k = ∣ Δ 1 ∣ k = |\Delta_1| k = ∣ Δ 1 ∣ and λ \lambda λ is determined by the rank-3 intersection numbers.
PSU(n,q)
Non-2-transitive case: Hermitian polar space action
The natural action of P S U n ( q ) PSU_n(q) P S U n ( q ) is on the set of totally singular 1-dimensional subspaces of the Hermitian space F q 2 n \mathbb{F}_{q^2}^{n} F q 2 n . This is a rank-3 action for most values of n n n .
The point stabilizer is a parabolic subgroup P 1 P_1 P 1 , and the degree is:
v = ( q n − ( − 1 ) n ) ( q n − 1 − ( − 1 ) n − 1 ) q 2 − 1 v=\frac{(q^n-(-1)^n)(q^{n-1}-(-1)^{n-1})}{q^2-1} v = q 2 − 1 ( q n − ( − 1 ) n ) ( q n − 1 − ( − 1 ) n − 1 )
For example, when n = 3 n=3 n = 3 :
v = ( q 3 + 1 ) ( q 2 − 1 ) / ( q 2 − 1 ) = q 3 + 1. v=(q^3+1)(q^2-1)/(q^2-1)=q^3+1. v = ( q 3 + 1 ) ( q 2 − 1 ) / ( q 2 − 1 ) = q 3 + 1.
Rank-3 structure: Fix a totally singular point ⟨ v ⟩ \langle v\rangle ⟨ v ⟩ . The other points split into two orbits based on whether the Hermitian form value is zero or non-zero. This rank-3 action gives automatic 2-designs from both suborbits.
Hermitian designs: Totally singular subspaces of Hermitian geometry give Witt-type designs and Hermitian designs, which are classical examples of 2-designs with unitary socle.
POmega(n,q)
Non-2-transitive case: Orthogonal polar space action
The set of totally singular 1-dimensional subspaces of an orthogonal space forms a projective polar space. This is a rank-3 or rank-4 action depending on the type of orthogonal space.
For P Ω 2 m + ( q ) P\Omega^+_{2m}(q) P Ω 2 m + ( q ) (plus type):
v = ( q m − 1 ) ( q m − 1 + 1 ) q − 1 v=\frac{(q^m-1)(q^{m-1}+1)}{q-1} v = q − 1 ( q m − 1 ) ( q m − 1 + 1 )
For P Ω 2 m − ( q ) P\Omega^-_{2m}(q) P Ω 2 m − ( q ) (minus type):
v = ( q m + 1 ) ( q m − 1 − 1 ) q − 1 v=\frac{(q^m+1)(q^{m-1}-1)}{q-1} v = q − 1 ( q m + 1 ) ( q m − 1 − 1 )
For P Ω 2 m + 1 ( q ) P\Omega_{2m+1}(q) P Ω 2 m + 1 ( q ) (odd dimension, q q q odd):
v = ( q m − 1 ) ( q m + 1 ) q − 1 = q 2 m − 1 q − 1 v=\frac{(q^m-1)(q^m+1)}{q-1}=\frac{q^{2m}-1}{q-1} v = q − 1 ( q m − 1 ) ( q m + 1 ) = q − 1 q 2 m − 1
Rank structure: The rank between two points is divided into 2 to 3 non-diagonal orbits according to their orthogonality relation: the points orthogonal to the fixed point, and the points non-orthogonal to it (further subdivided by norm value in some cases). This rank-3 or rank-4 structure gives automatic 2-designs from the suborbits, known as Witt designs and polar space designs.
Summary of the classical group classification
The verification for the classical groups follows the Aschbacher classes. For each Aschbacher class, we check candidate pairs by the ordering principle ∣ H ∣ > ∣ M ∣ |H| > |M| ∣ H ∣ > ∣ M ∣ , verify parameter divisibility, and confirm constant λ \lambda λ via orbit analysis. Table 9 records the outcome for each class.
Table 9. The Aschbacher classes of a classical group and their outcome in the four classical families of this section.
Note. The entries in the rows C 1 C_1 C 1 and C 8 C_8 C 8 refer to the designs examined in this section. The unital designs 2 2 2 -( q 3 + 1 , q + 1 , 1 ) ({q^3 + 1}, q + 1, 1) ( q 3 + 1 , q + 1 , 1 ) , arising in C 1 C_1 C 1 from rank-3 actions and in C 8 C_8 C 8 from rank-3 polar space actions, are non-symmetric and are listed in the existence column of Table 1 .
Point-imprimitive case for classical groups: exclusion
The point-imprimitive candidates all come from chains of subgroups, and each chain is excluded in turn.
Chains of parabolic subgroups
The point-imprimitive candidates for classical groups are the chains of parabolic subgroups:
P k < P l < G , k < l , P_k < P_l < G,\quad k < l, P k < P l < G , k < l ,
where P i P_i P i is the stabilizer of an i i i -dimensional subspace, H = P k H = P_k H = P k is the point stabilizer and M = P l M = P_l M = P l the block stabilizer, an intermediate subgroup with k < l k < l k < l ; then v = [ G : P k ] v = [G : P_k] v = [ G : P k ] , s = [ G : P l ] s = [G : P_l] s = [ G : P l ] and each class of the partition has size [ P l : P k ] [P_l : P_k] [ P l : P k ] .
Exclusion of the generalized Grassmann candidates
If the block stabilizer M M M is a parabolic subgroup P j P_j P j with j > l j > l j > l , a block is the set of all k k k -subspaces contained in a fixed j j j -subspace. For a point pair { U , V } \{U,V\} { U , V } of k k k -subspaces, the number of blocks through it equals the number of j j j -subspaces containing U ∨ V U \vee V U ∨ V . If dim ( U ∩ V ) = d \dim(U \cap V) = d dim ( U ∩ V ) = d , this number is the Gaussian binomial coefficient ( n − ( 2 k − d ) j − ( 2 k − d ) ) q \binom{n-(2k-d)}{j-(2k-d)}_q ( j − ( 2 k − d ) n − ( 2 k − d ) ) q when 2 k − d ≤ j 2k-d \leq j 2 k − d ≤ j , and 0 0 0 otherwise; it depends on d = dim ( U ∩ V ) d = \dim(U \cap V) d = dim ( U ∩ V ) . Since dim ( U ∩ V ) \dim(U \cap V) dim ( U ∩ V ) assumes several values on the point set, the coverage number is not constant and the 2-design condition fails. The same argument excludes the dual candidates (blocks consisting of all k k k -subspaces through a fixed j j j -subspace). Hence no generalized Grassmann design of this form arises in the point-imprimitive setting: the argument excludes the block families consisting of all k k k -subspaces contained in a fixed j j j -subspace, and dually all k k k -subspaces containing a fixed j j j -subspace. It does not exclude every 2-design with the parameter triple 2 − ( ( n k ) q , ( l k ) q , λ ) 2-\left(\binom{n}{k}_q,\binom{l}{k}_q,\lambda\right) 2 − ( ( k n ) q , ( k l ) q , λ ) ; what it excludes is their realization by these parabolic orbits.
Exclusion of non-parabolic subgroup chains
No other point-imprimitive design arises from non-parabolic subgroup chains.
The remaining chains fall into three cases.
(1) Case 1: H H H parabolic (C 1 C_1 C 1 ). If M M M is also parabolic, the candidate is the generalized Grassmann construction, which fails the constant-coverage test: the coverage number of { U , V } \{U,V\} { U , V } depends on dim ( U ∩ V ) \dim(U \cap V) dim ( U ∩ V ) . If M M M is C 2 C_2 C 2 (reducible), M M M stabilizes a direct sum decomposition V = V 1 ⊕ V 2 V = V_1 \oplus V_2 V = V 1 ⊕ V 2 , and pairs of k k k -subspaces lying in V 1 V_1 V 1 versus V 2 V_2 V 2 have different intersection behaviors, so the point-pair coverage is not constant. If M M M is C 3 C_3 C 3 (field extension) or C 5 C_5 C 5 (subfield), the orbit structure is not uniform and the design condition fails. If M M M is S S S (almost simple), the embedding is non-geometric and the point-pair coverage is not constant.
(2) Case 2: H H H reducible (C 2 C_2 C 2 ). If M M M is also C 2 C_2 C 2 with a finer decomposition, the candidate is a degenerate partition design and is not non-symmetric. If H < M H < M H < M with M = P j M = P_j M = P j parabolic, then H H H stabilizes a j j j -subspace; since H = G L ( V 1 ) × G L ( V 2 ) H = GL(V_1) \times GL(V_2) H = G L ( V 1 ) × G L ( V 2 ) acts irreducibly on V 1 V_1 V 1 and on V 2 V_2 V 2 , the only H H H -invariant subspaces are 0 0 0 , V 1 V_1 V 1 , V 2 V_2 V 2 , V V V , so j ∈ { d 1 , d 2 } j \in\{d_1,d_2\} j ∈ { d 1 , d 2 } . For such j j j one has ∣ P j ∣ = q d 1 d 2 ∣ G L ( d 1 , q ) ∣ ∣ G L ( d 2 , q ) ∣ = q d 1 d 2 ∣ H ∣ > ∣ H ∣ |P_j| = q^{d_1d_2}|GL(d_1,q)||GL(d_2,q)| = q^{d_1d_2}|H| > |H| ∣ P j ∣ = q d 1 d 2 ∣ G L ( d 1 , q ) ∣∣ G L ( d 2 , q ) ∣ = q d 1 d 2 ∣ H ∣ > ∣ H ∣ , contradicting ∣ M ∣ < ∣ H ∣ |M| < |H| ∣ M ∣ < ∣ H ∣ ; hence no such chain occurs.
(3) Case 3: other Aschbacher classes. For C 3 , C 4 , C 5 , C 6 , C 7 , C 8 , S C_3,C_4,C_5,C_6,C_7,C_8,S C 3 , C 4 , C 5 , C 6 , C 7 , C 8 , S the block stabilizer fixes an extension field, a tensor decomposition, a subfield, a symplectic structure or a non-geometric simple subgroup; its orbits on the points then split into at least two types, and the coverage number of a fixed point pair varies.
Classification result for the classical groups
Proposition 4.43 (Classical point-imprimitive exclusion). No non-trivial non-symmetric block-primitive 2-design with classical socle admits a point-imprimitive action. In particular, the generalized Grassmann candidates arising from chains of parabolic subgroups P k < P l < P j P_k < P_l < P_j P k < P l < P j fail the constant-coverage test, and no other chain from the Aschbacher classes C 2 , … , C 8 , S C_2,\ldots,C_8,S C 2 , … , C 8 , S yields a design. This completes the classification of the classical group case, which is therefore point-primitive only.
Proof. The three cases (1)–(3) above exhaust the subgroup chains with ∣ M ∣ < ∣ H ∣ |M| < |H| ∣ M ∣ < ∣ H ∣ : a parabolic point stabilizer gives the generalized Grassmann candidates, whose coverage number depends on dim ( U ∩ V ) \dim(U \cap V) dim ( U ∩ V ) and is therefore not constant, and whose other block types are excluded as recorded; a reducible point stabilizer admits no parabolic block stabilizer, since the only H H H -invariant subspaces are 0 0 0 , V 1 V_1 V 1 , V 2 V_2 V 2 , V V V and the corresponding parabolic subgroup is larger than H H H ; and the remaining Aschbacher classes C 3 , … , C 8 , S C_3,\ldots,C_8,S C 3 , … , C 8 , S give orbits of at least two types, so the coverage number of a fixed pair varies. No case yields a design.
Classification of AS type for classical groups
For a classical simple group T ≤ G ≤ Aut ( T ) T \leq G \leq\operatorname{Aut}(T) T ≤ G ≤ Aut ( T ) acting primitively on points, only four classes of subgroups need to be examined, the remaining classes C 2 C_2 C 2 , C 4 C_4 C 4 , C 5 C_5 C 5 , C 6 C_6 C 6 , C 7 C_7 C 7 having been excluded by Cases 2 and 3 of the point-imprimitive analysis in § 2.4 :
C 1 C_1 C 1 (parabolic subgroups): natural projective space actions
C 3 C_3 C 3 (extension field subgroups)
C 8 C_8 C 8 (classical subgroups: symplectic, unitary, orthogonal)
S S S (non-geometric almost simple subgroups)
The four subsections above examine these classes family by family, and Table 9 records the outcome: C 1 C_1 C 1 , C 3 C_3 C 3 and S S S fail the constant-coverage test, and the C 8 C_8 C 8 candidates are the rank-3 polar space designs, whose unital members 2 2 2 -( q 3 + 1 , q + 1 , 1 ) (q^3+1,q+1,1) ( q 3 + 1 , q + 1 , 1 ) are non-symmetric and appear in the existence column of Table 1 ; any symmetric rank-3 suborbit design lies outside the classification.
Exceptional Lie type groups
Exceptional Lie type simple groups include:
Rank 1: 2 B 2 ( q ) {}^{2}B_2(q) 2 B 2 ( q ) (Suzuki), 2 G 2 ( q ) {}^{2}G_2(q) 2 G 2 ( q ) (Ree)
Rank 2: G 2 ( q ) G_2(q) G 2 ( q ) , 3 D 4 ( q ) {}^{3}D_4(q) 3 D 4 ( q ) , 2 F 4 ( q ) {}^{2}F_4(q) 2 F 4 ( q ) (Ree)
Rank ≥ 3 \ge3 ≥ 3 : F 4 ( q ) F_4(q) F 4 ( q ) , E 6 ( q ) E_6(q) E 6 ( q ) , 2 E 6 ( q ) {}^{2}E_6(q) 2 E 6 ( q ) , E 7 ( q ) E_7(q) E 7 ( q ) , E 8 ( q ) E_8(q) E 8 ( q )
Scope: non-2-homogeneous point actions
Among the exceptional groups of Lie type only 2 B 2 ( q ) {}^{2}B_2(q) 2 B 2 ( q ) and 2 G 2 ( q ) {}^{2}G_2(q) 2 G 2 ( q ) have a 2-transitive permutation representation, on q 2 + 1 q^2+1 q 2 + 1 and q 3 + 1 q^3+1 q 3 + 1 points (Dixon and Mortimer [24 Section 7.7] ); their designs are the Suzuki–Tits and Ree designs of Table 10 . Every other primitive action examined in the classification below is non-2-homogeneous.
Classification procedure
The classification enumerates the maximal subgroups of each exceptional group and checks:
Fisher’s inequality: b = [ G : M ] > v = [ G : H ] b=[G:M]>v=[G:H] b = [ G : M ] > v = [ G : H ] .
Parameter divisibility: existence of integers k k k , r r r , λ \lambda λ satisfying v r = b k vr=bk v r = bk and r ( k − 1 ) = λ ( v − 1 ) r(k-1)=\lambda(v-1) r ( k − 1 ) = λ ( v − 1 ) .
Orbit consistency: M M M -orbits on P = G / H \mathcal{P}=G/H P = G / H yield constant point-pair coverage numbers.
Maximal subgroups of the exceptional Lie type groups
Following the ordering principle (Lemma 4.3 ), we enumerate maximal subgroups of exceptional groups by increasing order, and consider only pairs ( H , M ) (H,M) ( H , M ) with ∣ H ∣ > ∣ M ∣ |H|>|M| ∣ H ∣ > ∣ M ∣ .
The maximal subgroups of exceptional groups fall into five families:
Parabolic subgroups (C 1 C_1 C 1 analogue): stabilizers of building flags, of order increasing with the flag dimension.
Semisimple subgroups: lower-rank groups embedded naturally, as A 2 ( q ) A_2(q) A 2 ( q ) in G 2 ( q ) G_2(q) G 2 ( q ) .
Local subgroups: normalizers of unipotent elements or of tori.
Subfield subgroups: groups defined over subfields.
Diagonal subgroups: diagonally embedded groups.
With ∣ H ∣ > ∣ M ∣ |H|>|M| ∣ H ∣ > ∣ M ∣ the candidate pairs fall into two families:
Rank-3 building designs: Point stabilizer H H H is a maximal parabolic subgroup, block stabilizer M M M is a smaller subgroup. The rank-3 action on building points gives automatic 2-designs.
Chains of parabolic subgroups: H = P k H=P_k H = P k , M = P j M=P_j M = P j with k > j k>j k > j , giving generalized building designs.
All non-parabolic subgroups are excluded by orbit-structure analysis: their orbits on the building never give a constant point-pair coverage number.
Summary for the exceptional Lie type groups
For exceptional Lie type socles the classification is the following. Table 10 records the families, the designs found and the reason for exclusion.
Proposition 4.44 (Exceptional point-primitive classification). The point-primitive non-symmetric block-primitive 2-designs with exceptional Lie type socle and non-2-homogeneous point action arise from:
Rank-3 building designs from natural projective actions on building points.
Generalized building designs from chains of parabolic subgroups.
Special low-rank cases: Suzuki-Tits ovoid designs and Ree unital designs.
Non-parabolic maximal subgroups (reducible or irreducible) are excluded by orbit-structure analysis, as their orbits on the building do not yield constant point-pair coverage.
Proof. The list is read off from Table 10 , which records, family by family, the rank of the building, the designs found and the reason for the exclusion of the remaining subgroups: the parabolic chains produce the generalized building designs of items (1) and (2), and the two rank-1 families 2 B 2 ( q ) ^{2}B_{2}(q) 2 B 2 ( q ) and 2 G 2 ( q ) ^{2}G_{2}(q) 2 G 2 ( q ) give the Suzuki–Tits ovoid designs and the Ree unital designs of item (3), while the families of rank at least 2 carry no admissible pair. The non-parabolic maximal subgroups are reducible or irreducible and fail the constancy of the point-pair coverage as recorded in the paragraphs above.
Sporadic simple groups
Let T T T be a sporadic simple group with T ≤ G ≤ Aut ( T ) T \leq G \leq\operatorname{Aut}(T) T ≤ G ≤ Aut ( T ) . Every point stabilizer is a maximal subgroup of G G G , and by the ordering principle (Lemma 4.3) the block stabilizer M M M is maximal with ∣ H ∣ > ∣ M ∣ |H| > |M| ∣ H ∣ > ∣ M ∣ . A candidate block is a union of M M M -orbits on P = G / H P=G/H P = G / H , and the design conditions are
b = [ G : M ] > v = [ G : H ] ; (20) b=[G:M]>v=[G:H]; \tag*{(20)} b = [ G : M ] > v = [ G : H ] ; (20)
r = b k / v and λ = r ( k − 1 ) / ( v − 1 ) are positive integers ; (21) r=bk/v\quad\text{and}\quad\lambda=r(k-1)/(v-1)\text{ are positive integers}; \tag*{(21)} r = bk / v and λ = r ( k − 1 ) / ( v − 1 ) are positive integers ; (21)
(3) every point pair of P \mathcal{P} P lies in the same number λ \lambda λ of blocks.
The maximal subgroups of the sporadic groups and their indices are recorded in the ATLAS [17 ] . For each primitive action the three conditions are decided by the finite check of Appendix § 6.2 , which enumerates the unions of M M M -orbits and tests condition (3) exactly; the outcome is collected in Table 11 .
Table 11. The sporadic groups and the point-primitive non-symmetric block-primitive 2 2 2 -designs carried by their non-2 2 2 -homogeneous actions.
Two classes of actions need no separate treatment. The Mathieu groups M 11 , M 12 , M 22 , M 23 , M 24 M_{11}, M_{12}, M_{22}, M_{23}, M_{24} M 11 , M 12 , M 22 , M 23 , M 24 act 2-homogeneously on their natural domains, and H S HS H S and C o 3 Co_{3} C o 3 act 2-homogeneously on 176 and 276 points; the designs arising from these actions are given by the orbit construction of Lemma 4.2 , are classified in [39 ] , and lie outside the non-2-homogeneous classification. The symmetric designs 2 2 2 -( 176 , 50 , 14 ) (176, 50, 14) ( 176 , 50 , 14 ) of H S HS H S and 2 2 2 -( 276 , 100 , 36 ) (276, 100, 36) ( 276 , 100 , 36 ) of C o 3 Co_{3} C o 3 belong to the symmetric case.
Among the non-2-homogeneous actions exactly two carry a design: 2 2 2 -( 100 , 56 , 560 ) (100, 56, 560) ( 100 , 56 , 560 ) for J 2 J_{2} J 2 , from the rank-3 action on the cosets of U 3 ( 3 ) U_{3}(3) U 3 ( 3 ) , and 2 2 2 -( 100 , 64 , 1680 ) (100, 64, 1680) ( 100 , 64 , 1680 ) for H S HS H S , from the rank-3 action on the cosets of M 22 M_{22} M 22 . In both cases the block stabilizer is maximal, of order 336 and 10,752 respectively, so the block action is primitive. Proposition 4.45 records this; the remaining groups are listed in Table 11 together with the reason for exclusion.
Examples
The exhaustive enumeration shows that the non-2-homogeneous primitive actions of the sporadic groups admit exactly the following non-symmetric block-primitive 2-designs:
J 2 J_{2} J 2 : 2 2 2 -( 100 , 56 , 560 ) (100, 56, 560) ( 100 , 56 , 560 ) , b = 1800 b = 1800 b = 1800 , r = 1008 r = 1008 r = 1008 , from the rank-3 action on 100 points (the Hall–Janko graph).
H S HS H S (Higman-Sims): 2 2 2 -( 100 , 64 , 1680 ) (100, 64, 1680) ( 100 , 64 , 1680 ) , b = 4125 b = 4125 b = 4125 , r = 2640 r = 2640 r = 2640 , from the rank-3 action on 100 points (the Higman–Sims graph).
The 2-homogeneous actions of C o 3 Co_{3} C o 3 on 276 points and of H S HS H S on 176 points give block-primitive 2-designs by the orbit construction of Lemma 4.2 ; their parameters are classified in [39 ] . No other sporadic group admits a point-primitive non-symmetric block-primitive 2-design with non-2-homogeneous point action.
Table for the sporadic groups
The complete classification of point-primitive non-symmetric block-primitive 2 2 2 -designs with sporadic socle and non-2 2 2 -homogeneous point action, across the sporadic simple groups, is collected in Table 11 .
For each group the check covers every maximal subgroup M M M with ∣ M ∣ < ∣ H ∣ |M|<|H| ∣ M ∣ < ∣ H ∣ , and for each pair it enumerates the unions of M M M -orbits as candidate blocks and tests the three conditions above exactly. For M 11 M_{11} M 11 the maximal subgroups have orders 720, 660, 144, 120 and 48, and in degree 165 the point stabilizer has order 48, so no maximal subgroup of smaller order exists and no block arises. For J 3 J_{3} J 3 the nine classes correspond to the primitive degrees 6156, 14688 (twice), 17442, 20520, 23256, 25840, 26163 and 43605. The parameter quadruples ( 66 , 20 , 6 ) (66,20,6) ( 66 , 20 , 6 ) , ( 275 , 11 , 1 ) (275,11,1) ( 275 , 11 , 1 ) , ( 2058 , 594 , 162 ) (2058,594,162) ( 2058 , 594 , 162 ) , ( 4060 , 1265 , 392 ) (4060,1265,392) ( 4060 , 1265 , 392 ) , ( 1782 , 416 , 100 ) (1782,416,100) ( 1782 , 416 , 100 ) and ( 2300 , 891 , 324 ) (2300,891,324) ( 2300 , 891 , 324 ) already fail the divisibility condition r ( k − 1 ) = λ ( v − 1 ) r(k-1)=\lambda(v-1) r ( k − 1 ) = λ ( v − 1 ) . For the Monster M M M the maximal subgroups are completely classified by Dietrich, Lee and Popiel [23 ] , whose list has 46 classes. Besides the classes recorded in the ATLAS, it establishes the two classes with socles L 2 ( 13 ) L_{2}(13) L 2 ( 13 ) and U 3 ( 4 ) U_{3}(4) U 3 ( 4 ) , of orders 2184 and 249600, and excludes the socles L 2 ( 27 ) L_{2}(27) L 2 ( 27 ) , S z ( 8 ) Sz(8) S z ( 8 ) and U 3 ( 8 ) U_{3}(8) U 3 ( 8 ) . The lists used above are complete.
Proposition 4.45 (Sporadic point-primitive classification). The point-primitive non-symmetric block-primitive 2 2 2 -designs with sporadic socle and non-2 2 2 -homogeneous point action are exactly
J 2 : 2 ( 100 , 56 , 560 ) , H S : 2 ( 100 , 64 , 1680 ) , J_{2}: \ 2(100,56,560), \qquad HS:\ 2(100,64,1680), J 2 : 2 ( 100 , 56 , 560 ) , H S : 2 ( 100 , 64 , 1680 ) ,
with b = 1800 b = 1800 b = 1800 , r = 1008 r = 1008 r = 1008 and b = 4125 b = 4125 b = 4125 , r = 2640 r = 2640 r = 2640 , respectively. No such design exists with socle M 12 M_{12} M 12 , M 22 M_{22} M 22 , M 23 M_{23} M 23 , M 24 M_{24} M 24 , M c L McL M c L , H e He H e , R u Ru R u , S u z Suz S u z or C o 2 Co_{2} C o 2 . The 2-homogeneous actions of C o 3 Co_{3} C o 3 on 276 points and of H S HS H S on 176 points yield designs by the orbit construction of Lemma 4.2 .
Proof. Every primitive action of a sporadic group and every maximal subgroup M M M with ∣ M ∣ < ∣ H ∣ |M| < |H| ∣ M ∣ < ∣ H ∣ is covered by the exact enumeration of Appendix § 6.2 , which tests the three design conditions on every union of M M M -orbits; Table 11 records the outcome for each group. The only surviving pairs are the two displayed, with block stabilizers of orders 336 and 10,752, which are maximal, so the block action is primitive; in the remaining groups the parameter quadruples fail the divisibility condition as recorded above, or no maximal subgroup of smaller order exists. The 2-homogeneous actions are the orbit-construction cases of Lemma 4.2 .
Point-imprimitive case for sporadic groups: exclusion
For sporadic simple groups, the point-imprimitive case corresponds to chains of maximal subgroups H < K < G H < K < G H < K < G .
Since sporadic groups have finitely many maximal subgroups, the classification is carried out by enumeration.
Subgroup chains
For each sporadic group T T T , we list all pairs of maximal subgroups ( K , M ) (K, M) ( K , M ) such that K < M K < M K < M in the subgroup lattice, and check:
Fisher’s inequality: b = [ T : M ] > v = [ T : H ] b = [T : M] > v = [T : H] b = [ T : M ] > v = [ T : H ] ;
parameter divisibility: existence of integers k k k , r r r , λ \lambda λ ;
orbit consistency: constant intersection sizes with imprimitivity blocks.
Table for the point-imprimitive sporadic case
No non-trivial non-symmetric block-primitive 2-design with sporadic socle admits a point-imprimitive action: the candidates listed in Table 12 all fail at least one of Fisher’s inequality, the parameter divisibility, or the orbit-consistency condition; note that the designs 2-(11, 5, 2), 2-(22, 7, 2), 2-(23, 7, 2), 2-(24, 8, 5) of the natural 2-homogeneous actions of M 11 M_{11} M 11 , M 22 M_{22} M 22 , M 23 M_{23} M 23 , M 24 M_{24} M 24 , and the designs from the 2-homogeneous actions of H S HS H S and C o 3 Co_{3} C o 3 , are point-primitive and do not belong to the point-imprimitive setting.
Table 12. The sporadic groups and the point-imprimitive subgroup chains, with the reason for exclusion.
Result for the point-imprimitive sporadic case
Proposition 4.46 (Sporadic point-imprimitive exclusion). No point-imprimitive non-symmetric block-primitive 2-design with sporadic socle exists. Every candidate subgroup chain fails at least one of Fisher’s inequality, the parameter divisibility, or the orbit-consistency condition. Every design with sporadic socle is point-primitive, and the point-primitive instances (arising from 2-transitive or rank-3 primitive actions) fall into the classification of Proposition Proposition 4.45 .
Proof. The candidates of Table 12 are the chains H < K < G H<K<G H < K < G of maximal subgroups of the sporadic simple groups, with ∣ K ∣ < ∣ H ∣ |K|<|H| ∣ K ∣ < ∣ H ∣ ; for each chain the three conditions of Fisher’s inequality, the parameter divisibility and the orbit consistency are tested exactly, and the table records which condition fails. Since every chain falls in one of these rows, no point-imprimitive design with sporadic socle exists, and the designs with sporadic socle are the point-primitive ones of Proposition 4.45 . □
Point-imprimitive case for exceptional Lie type groups: exclusion
For exceptional Lie type groups, the point-imprimitive case corresponds to chains of parabolic subgroups or chains of other maximal subgroups.
Chains of parabolic subgroups
As in the classical case, chains of parabolic subgroups P k < P l < G P_k < P_l < G P k < P l < G give generalized building candidates. For a point pair { U , V } \{U,V\} { U , V } of subspaces of the building, the coverage number equals the number of intermediate subspaces containing U ∨ V U \vee V U ∨ V , which depends on dim ( U ∩ V ) \dim(U \cap V) dim ( U ∩ V ) ; hence the coverage is not constant and the 2-design condition fails.
Exclusion of non-parabolic subgroup chains
The non-parabolic maximal subgroups fall into two categories:
Reducible subgroups (stabilizers of geometric structures)
Irreducible almost simple subgroups (embeddings of smaller simple groups)
For reducible non-parabolic subgroups (e.g., stabilizers of polar space structures in G 2 ( q ) G_2(q) G 2 ( q ) ), the orbit structure on the natural building does not yield constant point-pair coverage, and the design condition fails.
For irreducible almost simple subgroups the embedding is non-geometric, and the point-pair coverage is not constant.
Exceptional case: Ree groups 2 G 2 ( q ) {}^{2}G_2(q) 2 G 2 ( q )
The Ree groups 2 G 2 ( q ) {}^{2}G_2(q) 2 G 2 ( q ) are a special case. On the q 3 + 1 q^3+1 q 3 + 1 points their natural action is primitive of permutation rank 3, while the building has rank 1; the two ranks are distinct notions and are recorded separately, and it yields the Ree unital designs
2 − ( q 3 + 1 , q + 1 , 1 ) , 2-(q^3+1,q+1,1), 2 − ( q 3 + 1 , q + 1 , 1 ) ,
which are point-primitive designs recorded in Table 10 . No point-imprimitive candidate arises for Ree groups.
Result for the point-imprimitive exceptional case
Proposition 4.47 (Exceptional point-imprimitive exclusion). No point-imprimitive non-symmetric block-primitive 2-design with exceptional Lie type socle exists. The generalized building candidates fail the constant-coverage test, and the Ree unital designs 2 − ( q 3 + 1 , q + 1 , 1 ) 2-(q^3+1,q+1,1) 2 − ( q 3 + 1 , q + 1 , 1 ) are point-primitive designs arising from the primitive rank-3 action of 2 G 2 ( q ) {}^{2}G_2(q) 2 G 2 ( q ) ; they are recorded in Table 10 , and their classification is the one of [24 Section 7.7] for the natural 2-transitive action.
Proof. The subgroup chains are of two kinds. A chain of parabolic subgroups gives a generalized building candidate, for which the coverage number of a point pair { U , V } \{U, V\} { U , V } equals the number of intermediate subspaces containing U ∨ V U \vee V U ∨ V and depends on dim ( U ∩ V ) \dim(U \cap V) dim ( U ∩ V ) , so it is not constant. A non-parabolic block stabilizer is reducible or irreducible; in both cases the point-pair coverage is not constant, as recorded above. For 2 G 2 ( q ) {}^{2}G_{2}(q) 2 G 2 ( q ) the natural action on q 3 + 1 q^{3} + 1 q 3 + 1 points is primitive of rank 3, so the Ree unital designs 2 2 2 -( q 3 + 1 , q + 1 , 1 ) \left(q^{3} + 1, q + 1, 1\right) ( q 3 + 1 , q + 1 , 1 ) it yields are point-primitive and appear in Table 10 . □ \square □
Part IV: The main theorem and concluding remarks
Main theorem
The main result is the following.
Theorem 5.1 (Structural Reduction Theorem). Let D = ( P , B ) \mathcal{D} = (\mathcal{P}, \mathcal{B}) D = ( P , B ) be a non-trivial 2 2 2 -design with 2 < k < v − 1 2 < k < v - 1 2 < k < v − 1 , and let G ≤ Aut ( D ) G \leq\operatorname{Aut}(\mathcal{D}) G ≤ Aut ( D ) act primitively on B \mathcal{B} B . By Lemma 2.1 the block action G B G^{\mathcal{B}} G B is faithful and G P G^{\mathcal{P}} G P is quasiprimitive on P \mathcal{P} P . Then:
(i) Symmetric case. If b = v b = v b = v , that is, if D \mathcal{D} D is symmetric, then points and blocks play interchangeable roles, and the study of block-primitive symmetric designs coincides, by duality, with that of point-primitive symmetric designs, a classical subject with an extensive literature; this case is isolated as a separate class and is not part of the structural reduction below.
(ii) Non-symmetric case. Assume from now on that b > v b > v b > v (Fisher’s inequality). Then G G G acts quasiprimitively on P \mathcal{P} P .
(iii) Excluded types. The following O’Nan–Scott types for G G G are strictly excluded:
(1) Holomorph affine type (HA): the design must be symmetric (Lemma 2.10 ), hence no non-symmetric example arises.
(2) Holomorph compound type (HC): the socle splits into two commuting minimal normal subgroups, each regular on B \mathcal{B} B and forced to be regular on P \mathcal{P} P , so v = b v = b v = b and the design is symmetric (§ 2.1 ).
(3) Holomorph simple type (HS): the socle is T 1 × T 2 T_{1} \times T_{2} T 1 × T 2 with two commuting minimal normal subgroups, each regular on B \mathcal{B} B and forced to be regular on P \mathcal{P} P , so v = b v = b v = b and the design is symmetric (§ 2.1 ).
(4) Two-sided simple diagonal type (S D 2 \mathrm{SD}_{2} SD 2 ): the design is forced to be symmetric.
(5) Twisted wreath product type (TW): the socle T ℓ T^{\ell} T ℓ is regular on points, so the design is symmetric by Lemma 2.10 .
(iv) Reduced types. After the exclusions of (iii) and the further exclusions of the one-sided simple diagonal type and of the compound diagonal type (Propositions 7.6 and 7.14 ), no non-AS type survives: the two remaining cases are settled as follows.
(1) Product action type (PA): excluded completely (cf. Corollary 2.13). By the three-class structural analysis of Proposition 7.9 and the five-layer sieve of Proposition 7.11 the product action configuration is excluded for the six socles with the seventeen listed primitive degrees and every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 , with arbitrary block stabilizers; since non-symmetric block-primitive point actions are primitive and the admissible primitive degrees of non-2-homogeneous point actions are enumerated by the degree tables of Dixon–Mortimer [24 ] , this closes the product action type for non-symmetric designs. In particular, the necessary condition that Ind L T ( 1 ) \operatorname{Ind}_{L}^{T}(1) Ind L T ( 1 ) contains every irreducible character of T T T is not sufficient, since it holds for T = A 5 T = A_{5} T = A 5 and L = C 2 L = C_{2} L = C 2 without yielding a design; this is precisely why the five-layer sieve of Proposition 7.11 has to impose the layer-by-layer consistency conditions, and it does so there.
(2) Almost simple type (AS): G G G is almost simple, meaning there exists a non-abelian simple group T T T such that T ≤ G ≤ Aut ( T ) T \le G \le\operatorname{Aut}(T) T ≤ G ≤ Aut ( T ) . The problem reduces to the analysis of quadruples ( T , G , H , M ) (T,G,H,M) ( T , G , H , M ) where H = G x H = G_{x} H = G x is the point stabilizer and M = G B M = G_{B} M = G B is the block stabilizer, and is classified completely.
(v) Point-imprimitive actions are excluded. No non-trivial non-symmetric block-primitive 2-design admits a point-imprimitive automorphism group (Proposition 3.4 ): a non-quasiprimitive point action forces every block to be a union of the orbits of an intransitive normal subgroup, contradicting the 2-design condition; a quasiprimitive imprimitive point action has type AS, TW, SD, CD or PA by the quasiprimitive O’Nan–Scott classification [35 ] (the affine, holomorph simple and holomorph compound types are always primitive), and each of these types is excluded: TW by Lemma 2.10 , SD and CD in § 2.1 , PA by Proposition 7.9 and Proposition 7.11 for the point-primitive product action and by Lemma 3.11 for the point-imprimitive product action, and AS family by family in § 3.1 . The geometric candidates arising from chains of parabolic subgroups fail the constant-coverage test directly: the coverage number of a point pair { U , V } \{U,V\} { U , V } of k k k -subspaces equals the number of intermediate subspaces containing U ∨ V U \vee V U ∨ V , which depends on dim ( U ∩ V ) \dim(U \cap V) dim ( U ∩ V ) . In particular, the classical, exceptional and sporadic candidates (generalized Grassmann, building and subgroup-chain constructions) yield no 2-design in the point-imprimitive setting.
(vi) Point-primitive AS classification. For the AS type, we have also completed the classification of all point-primitive, non-2-homogeneous cases, across all families of finite simple groups, the even-characteristic P S L 2 ( q ) PSL_{2}(q) P S L 2 ( q ) subfield block stabilizers delimited in Theorem 4.6 being reduced there to the class equations of Proposition 4.34 :
Classical Lie type socle: All designs arise from rank-3 polar space designs (unitals, C 8 C_{8} C 8 cross-type embeddings); the generalized Grassmann candidates from C 1 C_{1} C 1 parabolic chains fail the constant-coverage test and yield no design. All other Aschbacher classes are excluded.
Sporadic socle: exactly two non-2-homogeneous actions yield point-primitive non-symmetric block-primitive 2-designs, namely J 2 J_{2} J 2 on 100 points 2-(100, 56, 560) and H S HS H S on 100 points 2-(100, 64, 1680); the 2-homogeneous actions of C o 3 Co_{3} C o 3 on 276 points and of H S HS H S on 176 points yield designs by the orbit construction of Lemma 4.2 ; all remaining sporadic actions are excluded.
Exceptional Lie type socle: All designs arise from parabolic subgroup chains (generalized building designs), plus special rank-1 cases: Suzuki ovoid designs and Ree unital designs. All non-parabolic subgroups are excluded.
(vii) Every O’Nan–Scott type other than almost simple is excluded (HA, HC, HS, two-sided S D 2 SD_{2} S D 2 , TW; the one-sided SD type for every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 and every non-abelian simple T T T (Lemmas 7.1 , 7.3 and 7.5 ); the compound diagonal CD type for n = 6 n=6 n = 6 and the six small simple groups; and the product action type, cf. Proposition 7.9 and Proposition 7.11). The AS case, which by (v) is necessarily point-primitive, is classified across all families of finite simple groups, the even-characteristic P S L 2 ( q ) PSL_{2}(q) P S L 2 ( q ) subfield block stabilizers delimited in Theorem 4.6 being reduced there to the class equations of Proposition 4.34 .
Proof. Part (i) isolates the symmetric case: for a symmetric design, points and blocks play interchangeable roles, and block-primitive symmetric designs coincide with point-primitive symmetric designs by duality, a classical subject covered by an extensive literature; we refer to it rather than repeat the classification here. Part (ii) follows from Lemma 2.1 , which proves that block-primitivity implies point-quasiprimitivity via the incidence-matrix rank argument. Part (iii) follows from the structural exclusion arguments of § 2.1 : HA and TW are excluded by Lemma 2.10 (a normal subgroup regular on points forces symmetry), HC by the centralizer argument, HS by the permutation-module obstruction under non-conjugate stabilizers, and two-sided S D 2 SD_2 S D 2 by symmetry forcing. Part (iv) states the exclusions: PA is excluded completely by Proposition 7.9 and Proposition 7.11, and AS reduces to the quadruple ( T , G , H , M ) (T, G, H, M) ( T , G , H , M ) analysis of § 3.1 . The one-sided SD and CD types are excluded in § 2.1 . Part (v) excludes point-imprimitive actions via Proposition 3.4 , and part (vi) records the complete point-primitive classification of the AS case.
Table 13. The block-type/point-type pairs in the point-primitive non-symmetric case. In the non-symmetric setting b > v b > v b > v by Fisher’s inequality, and the point action is primitive by Proposition 3.4 ; the entries record the outcome for each pair (block type, point type): sym means that the pair forces a symmetric design and hence is excluded for non-symmetric designs, excl means that the pair is excluded by the argument cited in the proof of Theorem 5.1 , and classified means that the pair is completely classified in § 3.1 .
Block-primitive designs of strength t ≥ 3 t \ge3 t ≥ 3
The block-primitive designs of strength t ≥ 3 t \ge3 t ≥ 3 are classified in Theorem 6.3 . The proof builds on the classification of Proposition 4.1 : every non-trivial t t t -design with t ≥ 3 t \ge3 t ≥ 3 is in particular a non-trivial 2 2 2 -design, so the reduction of § 2.1 and the classification of Proposition 4.1 apply verbatim (the dihedral case of Theorem 4.6 is excluded by Lemma 4.5 , so it does not enter this section), and it remains to decide which of the classified designs have strength at least 3. No symmetric design occurs among them: a non-trivial symmetric design has strength at most 2, indeed, if a symmetric 2 2 2 -( v , k , λ ) (v, k, \lambda) ( v , k , λ ) were also a 3 3 3 -design, the derived design at a point would be a 2 2 2 -( v − 1 , k − 1 , λ ) (v-1, k-1, \lambda) ( v − 1 , k − 1 , λ ) design with b ′ = r = k b' = r = k b ′ = r = k blocks, so Fisher’s inequality b ′ ≥ v ′ b' \ge v' b ′ ≥ v ′ would give k ≥ v − 1 k \ge v-1 k ≥ v − 1 , that is k = v − 1 k = v-1 k = v − 1 (complete) or k ≤ 1 k \le1 k ≤ 1 (trivial).
Every non-trivial t t t -design with t ≥ 3 t \ge3 t ≥ 3 is in particular a non-trivial 2-design, and the parameters λ s \lambda_s λ s of the induced s s s -design satisfy the elementary recursion
λ s = λ t ( v − s t − s ) ( k − s t − s ) , ( 1 ≤ s ≤ t ) , (22) \lambda_s = \lambda_t \frac{\binom{v-s}{t-s}}{\binom{k-s}{t-s}}, \qquad(1 \le s \le t), \tag*{(22)} λ s = λ t ( t − s k − s ) ( t − s v − s ) , ( 1 ≤ s ≤ t ) , (22)
with λ t \lambda_t λ t the number of blocks through a t t t -subset, so that λ 2 = λ t ( v − 2 t − 2 ) ( k − 2 t − 2 ) \lambda_2 = \lambda_t \frac{\binom{v-2}{t-2}}{\binom{k-2}{t-2}} λ 2 = λ t ( t − 2 k − 2 ) ( t − 2 v − 2 ) and, for t = 3 t = 3 t = 3 , λ 3 = λ 2 ( k − 2 ) / ( v − 2 ) \lambda_3 = \lambda_2(k-2)/(v-2) λ 3 = λ 2 ( k − 2 ) / ( v − 2 ) . In particular λ t \lambda_t λ t is an integer whenever the design has strength at least t t t ; this is a necessary condition only, not a sufficient one (the integrality of λ t \lambda_t λ t does not by itself force every t t t -subset to be met by a constant number of blocks). The classification below therefore does not rest on it: the candidates are the designs listed in Proposition 4.1 together with the 2-homogeneous orbit designs of Lemma 4.2 , and their strengths are computed directly from (22) and checked design by design, the claimed upper bound on the strength being obtained from the failure of the integrality of λ s \lambda_s λ s for s s s one above the claimed strength. The outcome is simple: beyond the inversive planes and the five Witt designs, two Steiner families arising from the exceptional maximal subgroups of P G L 2 ( q ) PGL_2(q) P G L 2 ( q ) occur, and, among the non-complete designs, none has strength t ≥ 6 t \ge6 t ≥ 6 . (The complete designs whose block set is the full set of k k k -subsets are trivial: they have strength min ( k , v − k ) \min(k,v-k) min ( k , v − k ) and automorphism group containing S v S_v S v , which is primitive on the blocks; they are excluded from the classification, as is customary.)
The first observation governs the point action of any block-transitive design of strength at least 3; its dichotomy is classical, and the block-primitive case is then settled by a normal-subgroup argument.
Lemma 6.1 (Block-transitive designs of strength t ≥ 3 t \ge3 t ≥ 3 ). Let D = ( P , B ) \mathcal{D} = (\mathcal{P}, \mathcal{B}) D = ( P , B ) be a non-trivial t t t -design with t ≥ 3 t \ge3 t ≥ 3 , and let G ≤ Aut ( D ) G \le\operatorname{Aut}(\mathcal{D}) G ≤ Aut ( D ) act transitively on B \mathcal{B} B . Then G G G is 2-homogeneous on P \mathcal{P} P , or G G G is of holomorph affine type on P \mathcal{P} P and D \mathcal{D} D is an affine hyperplane design 3 − ( 2 d , 2 d − 1 , 2 d − 2 − 1 ) 3-(2^d, 2^{d-1}, 2^{d-2}-1) 3 − ( 2 d , 2 d − 1 , 2 d − 2 − 1 ) , d ≥ 3 d \ge3 d ≥ 3 . In the affine case G G G is not primitive on B \mathcal{B} B . Consequently, if G G G is primitive on B \mathcal{B} B , then G G G is 2-homogeneous on P \mathcal{P} P .
Proof. The dichotomy is the classical rigidity theorem of Dembowski for block-transitive designs of strength at least 3 [21 ] : a fixed-triangle counting forces the point action of a block-transitive group to be 2-homogeneous, with the single exceptional family of affine hyperplane designs, on which the acting group is of holomorph affine type. Now suppose G G G is of holomorph affine type with regular normal subgroup N N N on P \mathcal{P} P . The affine hyperplane design on v = 2 d v = 2^d v = 2 d points has b = 2 ( 2 d − 1 ) b = 2(2^d - 1) b = 2 ( 2 d − 1 ) blocks, and the translations in N N N permute the blocks in the 2 d − 1 2^d - 1 2 d − 1 parallel classes of pairs of parallel hyperplanes; in particular N N N is not transitive on B \mathcal{B} B . But a group that is primitive on B \mathcal{B} B admits no non-trivial intransitive normal subgroup on B \mathcal{B} B , since the orbits of a normal subgroup always form a G G G -invariant partition. Hence the affine case is incompatible with block-primitivity, and the last assertion follows. □
By Lemma 6.1 , every block-primitive design of strength at least 3 has a 2-homogeneous point action; by Burnside’s theorem it is therefore almost simple, and we may inspect the families classified in Proposition 4.1 and the 2-homogeneous orbit designs of Lemma 4.2 .
Proposition 6.2 (The strength of the classified designs). Among the non-trivial block-primitive 2-designs of Proposition 4.1 and the 2-homogeneous orbit designs of Lemma 4.2 (the P S L 2 ( q ) PSL_2(q) P S L 2 ( q ) family being classified via the 3-design classification of [4 ] ), the following have strength at least 3:
(1) the inversive planes 3 − ( q 2 + 1 , q + 1 , 1 ) 3-(q^2 + 1, q + 1, 1) 3 − ( q 2 + 1 , q + 1 , 1 ) , q ≥ 3 q \ge3 q ≥ 3 a prime power;
(2) the Steiner systems 3 − ( q + 1 , 4 , 1 ) 3-(q + 1, 4, 1) 3 − ( q + 1 , 4 , 1 ) for q ≡ ± 1 ( m o d 8 ) q \equiv\pm1 \pmod{8} q ≡ ± 1 ( mod 8 ) and q ≡ 1 , 3 ( m o d 6 ) q \equiv1, 3 \pmod{6} q ≡ 1 , 3 ( mod 6 ) , and 3 − ( q + 1 , 6 , 1 ) 3-(q + 1, 6, 1) 3 − ( q + 1 , 6 , 1 ) for q ≡ 1 ( m o d 20 ) q \equiv1 \pmod{20} q ≡ 1 ( mod 20 ) , attached to the exceptional maximal subgroups S 4 S_4 S 4 and S 5 S_5 S 5 of P G L 2 ( q ) PGL_2(q) P G L 2 ( q ) ;
(3) the five Witt designs.
Every other family has strength exactly 2.
Proof. We inspect the families of Proposition 4.1 and the 2-homogeneous orbit designs of Lemma 4.2 . For the P S L 2 ( q ) PSL_2(q) P S L 2 ( q ) family, the point action on P G ( 1 , q ) PG(1,q) P G ( 1 , q ) is 2-transitive; whether an orbit of the k k k -set stabilizers gives a design of strength 3 is decided by the integrality conditions of (22) ; the design 2-( 28 , 7 , 2 ) (28, 7, 2) ( 28 , 7 , 2 ) of Proposition 4.28 has λ 3 = 2 ⋅ 5 / 26 \lambda_3 = 2 \cdot5/26 λ 3 = 2 ⋅ 5/26 , so it is of strength exactly 2. By Dickson’s classification of the maximal subgroups [22 ] , the subfield groups P G L 2 ( q 0 ) PGL_2(q_0) P G L 2 ( q 0 ) with q = q 0 2 q = q_0^2 q = q 0 2 give the inversive planes [5 ] , and the exceptional subgroups S 4 S_4 S 4 (for q ≡ ± 1 ( m o d 8 ) q \equiv\pm1 \pmod{8} q ≡ ± 1 ( mod 8 ) ) and S 5 S_5 S 5 (for q ≡ ± 1 ( m o d 5 ) q \equiv\pm1 \pmod{5} q ≡ ± 1 ( mod 5 ) ) give orbits of size 4 and 6, whose strength-3 designs are classified by Cameron, Maimani, Omidi and Tayfeh-Rezaie [10 ] and by Cameron, Omidi and Tayfeh-Rezaie [11 ] . The integrality conditions of (22) reduce these to the stated ranges; for the 4-sets this is exactly Hanani’s condition q + 1 ≡ 2 , 4 ( m o d 6 ) q + 1 \equiv2, 4 \pmod{6} q + 1 ≡ 2 , 4 ( mod 6 ) for quadruple systems [25 ] , and for the 6-sets one obtains q ≡ 1 ( m o d 20 ) q \equiv1 \pmod{20} q ≡ 1 ( mod 20 ) . The special
case q = 9 q = 9 q = 9 of the 4-set family is the inversive plane of order 3. The other classical and exceptional Lie-type point actions examined in Proposition 4.1 are either 2-transitive, namely the natural actions of P S L d ( q ) PSL_d(q) P S L d ( q ) , P S U 3 ( q ) PSU_3(q) P S U 3 ( q ) , S z ( q ) Sz(q) S z ( q ) and 2 G 2 ( q ) {}^{2}G_2(q) 2 G 2 ( q ) , whose orbit designs enter through Lemma 4.2 , while the natural action of S p 2 m ( 2 ) Sp_{2m}(2) S p 2 m ( 2 ) on non-zero vectors has rank 3 and enters through Lemma 6.1 for m ≥ 2 m \ge2 m ≥ 2 , or of rank at least 3, and the latter are excluded by Lemma 6.1 . For the unital designs of P S U 3 ( q ) PSU_3(q) P S U 3 ( q ) and 2 G 2 ( q ) {}^{2}G_2(q) 2 G 2 ( q ) the strength-3 parameter λ 3 = ( q − 1 ) / ( q 3 − 1 ) \lambda_3 = (q - 1)/(q^3 - 1) λ 3 = ( q − 1 ) / ( q 3 − 1 ) is not integral, and the ovoid designs of S z ( q ) Sz(q) S z ( q ) fail the same check; both families therefore have strength 2. Among the sporadic socles, the 2-homogeneous orbit designs of C o 3 Co_3 C o 3 on 276 points and of H S HS H S on 176 points have parameters as in [39 ] , and the integrality check of (22) fails for λ 3 \lambda_3 λ 3 ; all remaining sporadic actions are not 2-homogeneous and are excluded likewise.
Combining Lemma 6.1 with Proposition 6.2 , we obtain the complete classification promised above, the even-characteristic P S L 2 ( q ) PSL_2(q) P S L 2 ( q ) subfield block stabilizers being reduced in Theorem 4.6 to the class equations of Proposition 4.34 .
Theorem 6.3 (Block-primitive designs of strength t ≥ 3 t \ge3 t ≥ 3 ). The non-trivial non-complete block-primitive designs of strength t ≥ 3 t \ge3 t ≥ 3 are precisely the following designs:
the inversive planes 3 3 3 -( q 2 + 1 , q + 1 , 1 ) (q^2 + 1, q + 1, 1) ( q 2 + 1 , q + 1 , 1 ) for q ≥ 3 q \ge3 q ≥ 3 a prime power, with q ( q 2 + 1 ) q(q^2 + 1) q ( q 2 + 1 ) blocks and automorphism group G ≤ P Γ L ( 2 , q 2 ) G \le P\Gamma L(2,q^2) G ≤ P Γ L ( 2 , q 2 ) ;
the Steiner systems 3 3 3 -( q + 1 , 4 , 1 ) (q + 1, 4, 1) ( q + 1 , 4 , 1 ) for q q q a prime power with q ≡ ± 1 ( m o d 8 ) q \equiv\pm1 \pmod{8} q ≡ ± 1 ( mod 8 ) and q ≡ 1 , 3 ( m o d 6 ) q \equiv1,3 \pmod{6} q ≡ 1 , 3 ( mod 6 ) , with q ( q 2 − 1 ) / 24 q(q^2 - 1)/24 q ( q 2 − 1 ) /24 blocks, and 3 3 3 -( q + 1 , 6 , 1 ) (q + 1, 6, 1) ( q + 1 , 6 , 1 ) for q q q a prime power with q ≡ 1 ( m o d 20 ) q \equiv1 \pmod{20} q ≡ 1 ( mod 20 ) , with q ( q 2 − 1 ) / 120 q(q^2 - 1)/120 q ( q 2 − 1 ) /120 blocks, arising respectively from the orbits of the exceptional subgroups S 4 S_4 S 4 [10 ] and S 5 S_5 S 5 [11 ] of P G L 2 ( q ) PGL_2(q) P G L 2 ( q ) on P G ( 1 , q ) PG(1,q) P G ( 1 , q ) ; for q = 9 q = 9 q = 9 the first of these is the inversive plane of order 3;
the five Witt designs 4 4 4 -( 11 , 5 , 1 ) (11,5,1) ( 11 , 5 , 1 ) , 5 5 5 -( 12 , 6 , 1 ) (12,6,1) ( 12 , 6 , 1 ) , 3 3 3 -( 22 , 6 , 1 ) (22,6,1) ( 22 , 6 , 1 ) , 4 4 4 -( 23 , 7 , 1 ) (23,7,1) ( 23 , 7 , 1 ) , 5 5 5 -( 24 , 8 , 1 ) (24,8,1) ( 24 , 8 , 1 ) , with 66, 132, 77, 253, 759 blocks and Mathieu groups M 11 M_{11} M 11 , M 12 M_{12} M 12 , M 22 M_{22} M 22 , M 23 M_{23} M 23 , M 24 M_{24} M 24 .
Their strengths are 3, 3, and 4, 5, 3, 4, 5, respectively. In particular, apart from the trivial complete designs, there is no block-primitive design of strength t ≥ 6 t \ge6 t ≥ 6 .
Proof. Let D \mathcal{D} D be a non-trivial non-complete block-primitive design of strength t ≥ 3 t \ge3 t ≥ 3 with automorphism group G G G . Complete designs are excluded by hypothesis: a complete design with block set the full set of k k k -subsets has strength min ( k , v − k ) \min(k,v-k) min ( k , v − k ) , has b = ( v k ) > v b=\binom{v}{k} > v b = ( k v ) > v for 2 ≤ k ≤ v − 2 2 \le k \le v-2 2 ≤ k ≤ v − 2 , and has automorphism group Sym ( v ) \operatorname{Sym}(v) Sym ( v ) , which acts 2-transitively on the points; on the blocks, Sym ( v ) \operatorname{Sym}(v) Sym ( v ) is primitive when k ≠ v − k k \ne v-k k = v − k and imprimitive by complementation when k = v − k k=v-k k = v − k . These trivial designs are outside the classification, and when they are block-primitive the point action is primitive in any case. By Lemma 6.1 the point action is 2-homogeneous, hence almost simple, and by Proposition 6.2 the design is one of those listed in (1)–(3). Conversely, each listed design has b > v b>v b > v in every case: q ( q 2 + 1 ) > q 2 + 1 q(q^{2}+1)>q^{2}+1 q ( q 2 + 1 ) > q 2 + 1 , q ( q 2 − 1 ) / 24 > q + 1 q(q^{2}-1)/24>q+1 q ( q 2 − 1 ) /24 > q + 1 for q ≥ 7 q \ge7 q ≥ 7 , q ( q 2 − 1 ) / 120 > q + 1 q(q^{2}-1)/120>q+1 q ( q 2 − 1 ) /120 > q + 1 for q ≥ 41 q \ge41 q ≥ 41 , and 66 , 132 , 77 , 253 , 759 > v 66,132,77,253,759>v 66 , 132 , 77 , 253 , 759 > v for the Witt designs. Block-primitivity holds because the stabilizer of a block is a maximal subgroup of the acting group: for the inversive planes the stabilizer of a circle is PΓL ( 2 , q ) \operatorname{P\Gamma L}(2,q) PΓL ( 2 , q ) , maximal in PΓL ( 2 , q 2 ) \operatorname{P\Gamma L}(2,q^{2}) PΓL ( 2 , q 2 ) ; for the two Steiner families the stabilizer of a block is the corresponding exceptional subgroup S 4 S_{4} S 4 or S 5 S_{5} S 5 , maximal in PGL 2 ( q ) \operatorname{PGL}_{2}(q) PGL 2 ( q ) ; and for the Witt designs the stabilizer of a block is a maximal subgroup of the relevant Mathieu group [41 ] . The strengths are as stated, by (22) : the three 3-designs have non-integral λ 4 = λ 3 ( k − 3 ) / ( v − 3 ) \lambda_{4}=\lambda_{3}(k-3)/(v-3) λ 4 = λ 3 ( k − 3 ) / ( v − 3 ) , and the Witt designs have strengths 4 , 5 , 3 , 4 , 5 4,5,3,4,5 4 , 5 , 3 , 4 , 5 as listed. Hence the classification of the non-complete designs is complete, and no non-complete design of strength t ≥ 6 t \ge6 t ≥ 6 exists.
Reduction of the remaining O'Nan–Scott types
For the special case λ = 1 \lambda= 1 λ = 1 (finite linear spaces), the block-primitive condition forces almost simple type, which excludes both SD and CD block action types. The arguments below treat the general case λ > 1 \lambda> 1 λ > 1 .
The status of each reduced type is as follows. For the twisted wreath type, by Lemma 2.10 , a block-primitive non-trivial 2-design with a normal subgroup regular on points must be symmetric, so no non-symmetric block-primitive 2-design has TW point action. For the one-sided SD and CD types, the exclusion is stated below. For the product action type, the three-class analysis of Proposition 7.9 and the five-layer sieve of Proposition 7.11 exclude the product action configuration completely for the six socles with the seventeen listed primitive degrees and every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 , with arbitrary block stabilizers; combined with the enumeration of admissible primitive degrees from the degree tables of Dixon–Mortimer [24 ] this is the complete exclusion for non-symmetric designs.
Lemma 7.1 (The one-sided SD type with a proper coordinate projection, for every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 ). Let D \mathcal{D} D be a non-trivial 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design with 2 < k < v − 1 2 < k < v - 1 2 < k < v − 1 admitting a block-primitive, point-quasiprimitive automorphism group G G G whose block action is of simple diagonal type. Let N = Soc ( G ) = T ℓ N = \operatorname{Soc}(G) = T^\ell N = Soc ( G ) = T ℓ with T T T non-abelian simple, q = ∣ T ∣ q = \lvert T\rvert q = ∣ T ∣ and ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 , and suppose that for some j j j the projection π j ( K ) \pi_j(K) π j ( K ) of K = N ∩ G α K = N \cap G_\alpha K = N ∩ G α to the j j j -th factor is a proper subgroup of T T T . This hypothesis covers the case in which some coordinate of K K K is trivial, since then π j ( K ) = 1 \pi_j(K) = 1 π j ( K ) = 1 . Then no such D \mathcal{D} D exists.
Proof of Lemma 7.1. By Theorem 2.3 the group N N N is transitive on both P \mathcal{P} P and B \mathcal{B} B , the intersection N ∩ G B = Δ N \cap G^{\mathcal{B}} = \Delta N ∩ G B = Δ is a diagonal subgroup isomorphic to T T T , and b = q ℓ − 1 b = q^{\ell-1} b = q ℓ − 1 .
Choose j j j with H j : = π j ( K ) < T H_j := \pi_j(K) < T H j := π j ( K ) < T and put e = [ T : H j ] > 1 e = [T : H_j] > 1 e = [ T : H j ] > 1 . Since π j ( K ) ≤ H j \pi_j(K) \le H_j π j ( K ) ≤ H j , the rule θ ( n K ) = n j H j \theta(nK) = n_jH_j θ ( n K ) = n j H j defines a map θ : P = N / K → T / H j \theta: \mathcal{P} = N/K \to T/H_j θ : P = N / K → T / H j ; it is N N N -equivariant for the action of N N N on T / H j T/H_j T / H j through the j j j -th coordinate, and it is onto because π j ( N ) = T \pi_j(N) = T π j ( N ) = T . Every fibre θ − 1 ( t H j ) \theta^{-1}(tH_j) θ − 1 ( t H j ) therefore has size v / e v/e v / e .
The diagonal subgroup Δ ≅ T \Delta\cong T Δ ≅ T acts on T / H j T/H_j T / H j by left multiplication, hence transitively, and it preserves B 0 B_0 B 0 setwise. Hence the intersections B 0 ∩ θ − 1 ( t H j ) B_0 \cap\theta^{-1}(tH_j) B 0 ∩ θ − 1 ( t H j ) are carried bijectively onto one another by elements of Δ \Delta Δ and have a common cardinality; summing over the e e e cosets gives ∣ B 0 ∩ θ − 1 ( t H j ) ∣ = k / e \lvert B_0 \cap\theta^{-1}(tH_j)\rvert= k/e ∣ B 0 ∩ θ − 1 ( t H j )∣ = k / e for every t t t . Every block is an N N N -translate of B 0 B_0 B 0 and θ \theta θ is N N N -equivariant, so ∣ B ∩ θ − 1 ( t H j ) ∣ = k / e |B \cap\theta^{-1}(tH_j)| = k/e ∣ B ∩ θ − 1 ( t H j ) ∣ = k / e for every block B B B and every coset; in particular e ∣ k e \mid k e ∣ k .
Fix t t t and put D = θ − 1 ( t H j ) D=\theta^{-1}(tH_j) D = θ − 1 ( t H j ) , so that ∣ D ∣ = v / e |D|=v/e ∣ D ∣ = v / e and ∣ B ∩ D ∣ = k / e |B \cap D|=k/e ∣ B ∩ D ∣ = k / e for every block B B B . Counting pairs consisting of a block and two distinct points of its intersection with D D D gives
b ( k / e 2 ) = λ ( v / e 2 ) , b\binom{k/e}{2}=\lambda\binom{v/e}{2}, b ( 2 k / e ) = λ ( 2 v / e ) ,
that is
b k ( k − e ) = λ v ( v − e ) . bk(k-e)=\lambda v(v-e). bk ( k − e ) = λ v ( v − e ) .
The design equation reads b k ( k − 1 ) = λ v ( v − 1 ) bk(k-1)=\lambda v(v-1) bk ( k − 1 ) = λ v ( v − 1 ) , and all four factors are positive since λ > 0 \lambda>0 λ > 0 , k > 1 k>1 k > 1 and v > 1 v>1 v > 1 . Dividing the two identities gives ( k − e ) / ( k − 1 ) = ( v − e ) / ( v − 1 ) (k-e)/(k-1)=(v-e)/(v-1) ( k − e ) / ( k − 1 ) = ( v − e ) / ( v − 1 ) , hence ( e − 1 ) ( k − v ) = 0 (e-1)(k-v)=0 ( e − 1 ) ( k − v ) = 0 , so k = v k=v k = v because e > 1 e>1 e > 1 , contradicting k ≤ v − 2 k\leq v-2 k ≤ v − 2 . No division by k − e k-e k − e or by v − e v-e v − e occurs, so the identities are legitimate also when k = e k=e k = e : then ( k / e 2 ) = 0 \binom{k/e}{2}=0 ( 2 k / e ) = 0 while λ ( v / e 2 ) > 0 \lambda\binom{v/e}{2}>0 λ ( 2 v / e ) > 0 , because k ≤ v − 2 k\leq v-2 k ≤ v − 2 forces v / e > 1 v/e>1 v / e > 1 , and this is a contradiction in its own right.
Remark 7.2. The proof uses the block stabilizer only through the diagonal subgroup Δ ≅ T \Delta\cong T Δ ≅ T that it contains, and the point stabilizer only through one coordinate projection π j ( K ) \pi_j(K) π j ( K ) that is a proper subgroup of T T T , which turns θ \theta θ into an N N N -equivariant map N / K → T / H j N/K\to T/H_j N / K → T / H j all of whose fibres have size v / e v/e v / e . The uniformity of the intersections of the blocks with these fibres follows from Δ ≤ G B \Delta\leq G^{B} Δ ≤ G B , and a double count inside a single fibre then forces k = v k=v k = v . Neither Δ ∩ K = 1 \Delta\cap K=1 Δ ∩ K = 1 nor ∣ K ∣ ≤ q |K|\leq q ∣ K ∣ ≤ q is needed, no semiregularity of Δ \Delta Δ on P P P and no assertion about blocks other than B 0 B_0 B 0 is used, no arithmetic property of T T T and no subgroup enumeration enters, and ℓ \ell ℓ plays no role beyond ℓ ≥ 2 \ell\geq2 ℓ ≥ 2 . The case of a trivial coordinate is the special case H j = 1 H_j=1 H j = 1 , e = q e=q e = q of the present statement.
Lemma 7.3. (The one-sided SD type is excluded for ℓ = 3 \ell=3 ℓ = 3 .) Let D D D , G G G , N = T 3 N=T^3 N = T 3 , q = ∣ T ∣ q=|T| q = ∣ T ∣ and K = N ∩ G α K=N\cap G_\alpha K = N ∩ G α be as in Lemma 7.1 , with ℓ = 3 \ell=3 ℓ = 3 . Then no such D D D exists.
Proof of Lemma 7.3. By Lemma 7.1 we may assume that all three projections π i ( K ) \pi_i(K) π i ( K ) , i = 1 , 2 , 3 i=1,2,3 i = 1 , 2 , 3 , are onto T T T , since otherwise k = v k=v k = v contradicts k ≤ v − 2 k\leq v-2 k ≤ v − 2 . Put K 12 = π 12 ( K ) ≤ T 1 T 2 K_{12}=\pi_{12}(K)\leq T_1T_2 K 12 = π 12 ( K ) ≤ T 1 T 2 , whose projections to both factors are onto T T T . By Goursat’s theorem, such a subgroup is either T 1 T 2 T_1T_2 T 1 T 2 or the graph of an automorphism of T T T ; in particular ∣ K 12 ∣ ∈ { q 2 , q } |K_{12}|\in\{q^2,q\} ∣ K 12 ∣ ∈ { q 2 , q } . Let L = K ∩ ( 1 × 1 × T 3 ) ≤ T 3 L=K\cap(1\times1\times T_3)\leq T_3 L = K ∩ ( 1 × 1 × T 3 ) ≤ T 3 ; then ∣ L ∣ ∈ { 1 , q } |L|\in\{1,q\} ∣ L ∣ ∈ { 1 , q } and ∣ K ∣ = ∣ K 12 ∣ ∣ L ∣ |K|=|K_{12}||L| ∣ K ∣ = ∣ K 12 ∣∣ L ∣ by the first isomorphism theorem applied to π 12 \pi_{12} π 12 . If ∣ L ∣ = q |L|=q ∣ L ∣ = q , then 1 × 1 × T 3 ≤ K 1\times1\times T_3\leq K 1 × 1 × T 3 ≤ K . This subgroup is normal in N = T 3 N=T^3 N = T 3 , so it lies in core N ( K ) = ker ( N ↷ N / K ) \operatorname{core}_N(K)=\ker(N\curvearrowright N/K) core N ( K ) = ker ( N ↷ N / K ) , and N N N is not faithful on P \mathcal{P} P , contradicting N ≤ G ≤ Sym ( P ) N\leq G\leq\operatorname{Sym}(\mathcal{P}) N ≤ G ≤ Sym ( P ) . Hence ∣ L ∣ = 1 |L|=1 ∣ L ∣ = 1 in every remaining case.
If ∣ K 12 ∣ = q |K_{12}|=q ∣ K 12 ∣ = q and ∣ L ∣ = 1 |L|=1 ∣ L ∣ = 1 , then ∣ K ∣ = q |K|=q ∣ K ∣ = q and v = q 3 / q = q 2 = b v=q^3/q=q^2=b v = q 3 / q = q 2 = b by Theorem 2.3 ; the design is symmetric, which is excluded by the standing hypothesis recorded in § 2.1 .
If ∣ K 12 ∣ = q 2 |K_{12}|=q^2 ∣ K 12 ∣ = q 2 and ∣ L ∣ = 1 |L|=1 ∣ L ∣ = 1 , then π 12 \pi_{12} π 12 is a bijection and K = { ( a , b , φ ( a , b ) ) : a , b ∈ T } K=\{(a,b,\varphi(a,b)):a,b\in T\} K = {( a , b , φ ( a , b )) : a , b ∈ T } for a suitable map φ : T 2 → T \varphi:T^2\to T φ : T 2 → T . Since K K K is a subgroup, φ \varphi φ is a homomorphism. As T T T is non-abelian simple, the normal subgroups of T 2 T^2 T 2 are the products of normal subgroups of its two factors, so ker φ \ker\varphi ker φ is one of 1 1 1 , T 1 T_1 T 1 , T 2 T_2 T 2 , T 2 T^2 T 2 . Since π 3 ( K ) = T \pi_3(K)=T π 3 ( K ) = T the map φ \varphi φ is onto, and ∣ T 2 ∣ = q 2 > q = ∣ T ∣ |T^2|=q^2>q=|T| ∣ T 2 ∣ = q 2 > q = ∣ T ∣ , so ker φ \ker\varphi ker φ has order q q q and is therefore T 1 T_1 T 1 or T 2 T_2 T 2 , the cases ker φ = 1 \ker\varphi=1 ker φ = 1 and ker φ = T 2 \ker\varphi=T^2 ker φ = T 2 being excluded by the order and by the surjectivity respectively. In either remaining case φ \varphi φ is one of the two coordinate projections composed with an automorphism of T T T , so T 2 ≤ K T_2\leq K T 2 ≤ K or T 1 ≤ K T_1\leq K T 1 ≤ K , and faithfulness fails as above.
Remark 7.4. Consequently the one-sided SD type is excluded for ℓ = 3 \ell=3 ℓ = 3 and for every non-abelian simple T T T , with no arithmetic condition, no double-coset criterion and no restriction on T T T or on the block stabilizer beyond block primitivity. The two inputs are Lemma 7.1 and Goursat’s theorem for subgroups of T 1 T 2 T_1T_2 T 1 T 2 with both projections onto T T T ; the three possible orders q 3 q^3 q 3 , q 2 q^2 q 2 , q q q of K K K are removed respectively by v = 1 v=1 v = 1 , by faithfulness, and by the symmetry forced by v = b = q 2 v=b=q^2 v = b = q 2 .
Lemma 7.5 (The one-sided SD type with an all-surjective multi-part point stabilizer). Let D \mathcal{D} D be a non-trivial non-symmetric 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design with 2 < k < v − 2 2<k<v-2 2 < k < v − 2 admitting a block-primitive, point-quasiprimitive automorphism group G G G whose block action is of simple diagonal type, with N = Soc ( G ) = T ℓ N=\operatorname{Soc}(G)=T^\ell N = Soc ( G ) = T ℓ , T T T non-abelian simple, q = ∣ T ∣ q=|T| q = ∣ T ∣ and ℓ ≥ 2 \ell\geq2 ℓ ≥ 2 . Suppose that K = N ∩ G α K=N\cap G_\alpha K = N ∩ G α is a product of full diagonal subgroups on the parts S 1 , … , S s S_1,\ldots,S_s S 1 , … , S s of a partition S \mathcal{S} S of { 1 , … , ℓ } \{1,\ldots,\ell\} { 1 , … , ℓ } into s ≥ 2 s\geq2 s ≥ 2 parts, each of size at least 2, automorphism twists being permitted inside each part. Then no such D \mathcal{D} D exists.
Proof of Lemma 7.5 . By Theorem 2.3 the group N N N is transitive on P \mathcal{P} P and on B \mathcal{B} B , and Δ : = N ∩ G B \Delta:=N\cap G_B Δ := N ∩ G B is isomorphic to T T T ; hence v = ∣ P ∣ = ∣ N ∣ / ∣ K ∣ = q ℓ − s v=|\mathcal{P}|=|N|/|K|=q^{\ell-s} v = ∣ P ∣ = ∣ N ∣/∣ K ∣ = q ℓ − s and b = ∣ B ∣ = ∣ N ∣ / ∣ Δ ∣ = q ℓ − 1 b=|\mathcal{B}|=|N|/|\Delta|=q^{\ell-1} b = ∣ B ∣ = ∣ N ∣/∣Δ∣ = q ℓ − 1 . Write Δ = { ( β 1 ( a ) , … , β ℓ ( a ) ) : a ∈ T } \Delta= \{(\beta_1(a), \ldots, \beta_\ell(a)) : a \in T\} Δ = {( β 1 ( a ) , … , β ℓ ( a )) : a ∈ T } with homomorphisms β j : T → T \beta_j : T \to T β j : T → T (Proposition 7.15 , second step); at least one β j \beta_j β j is injective because ∣ Δ ∣ = q |\Delta| = q ∣Δ∣ = q . Put H = G α H = G_\alpha H = G α , so that K = N ∩ H ⊴ H K = N \cap H \unlhd H K = N ∩ H ⊴ H and G = N H G = NH G = N H because N N N is transitive on P \mathcal{P} P . Conjugation by an element of H H H normalises K K K and therefore permutes the parts S \mathcal{S} S , the partition being recovered from K K K as the support partition of its full-diagonal factors, while conjugation by an element of N N N fixes every coordinate; as G = N H G = NH G = N H , every g ∈ G g \in G g ∈ G acts on the ℓ \ell ℓ factors of N N N as
σ g ( x ) k = α k ( x π − 1 ( k ) ) , k = 1 , … , ℓ , \sigma_g(x)_k = \alpha_k\left(x_{\pi^{-1}(k)}\right), \qquad k = 1,\ldots,\ell, σ g ( x ) k = α k ( x π − 1 ( k ) ) , k = 1 , … , ℓ ,
where α k ∈ Aut ( T ) \alpha_k \in\operatorname{Aut}(T) α k ∈ Aut ( T ) and π \pi π is a permutation of { 1 , … , ℓ } \{1,\ldots,\ell\} { 1 , … , ℓ } preserving S \mathcal{S} S . Every automorphism of T ℓ T^\ell T ℓ has this form, since T T T is non-abelian simple and Aut ( T ℓ ) = Aut ( T ) ℓ ⋊ S ℓ \operatorname{Aut}(T^\ell) = \operatorname{Aut}(T)^\ell\rtimes S_\ell Aut ( T ℓ ) = Aut ( T ) ℓ ⋊ S ℓ . Since Δ = N ∩ M \Delta= N \cap M Δ = N ∩ M with M = G B M = G^B M = G B is normal in M M M , conjugation by m ∈ M m \in M m ∈ M stabilises Δ \Delta Δ : there is a homomorphism φ : T → T \varphi: T \to T φ : T → T with σ m ( β ( a ) ) = β ( φ ( a ) ) \sigma_m(\beta(a)) = \beta(\varphi(a)) σ m ( β ( a )) = β ( φ ( a )) for every a a a , that is,
α π ( i ) β i = β π ( i ) φ ( i = 1 , … , ℓ ) . \alpha_{\pi(i)}\beta_i = \beta_{\pi(i)}\varphi\qquad(i = 1,\ldots,\ell). α π ( i ) β i = β π ( i ) φ ( i = 1 , … , ℓ ) .
Let C = { j : β j is constant } C = \{j : \beta_j\text{ is constant}\} C = { j : β j is constant } , and write β j ≡ c j \beta_j \equiv c_j β j ≡ c j for j ∈ C j \in C j ∈ C . For i ∈ C i \in C i ∈ C the displayed relation gives α π ( i ) ( c i ) = β π ( i ) ( φ ( a ) ) \alpha_{\pi(i)}(c_i) = \beta_{\pi(i)}(\varphi(a)) α π ( i ) ( c i ) = β π ( i ) ( φ ( a )) , a value independent of a a a ; hence π ( i ) ∈ C \pi(i) \in C π ( i ) ∈ C and α π ( i ) ( c i ) = c π ( i ) \alpha_{\pi(i)}(c_i) = c_{\pi(i)} α π ( i ) ( c i ) = c π ( i ) . In particular π \pi π preserves C C C .
Two cases cover all possibilities.
At least two parts carry an injective β j \beta_j β j . For a part S S S put
D S = { x ∈ N : x j = β j ( a ) ( j ∈ S ) , x j = 1 ( j ∉ S ) , a ∈ T } , D_S = \{x \in N : x_j = \beta_j(a)\ (j \in S),\ x_j = 1\ (j \notin S),\ a \in T\}, D S = { x ∈ N : x j = β j ( a ) ( j ∈ S ) , x j = 1 ( j ∈ / S ) , a ∈ T } ,
a subgroup isomorphic to T T T if some β j \beta_j β j with j ∈ S j \in S j ∈ S is injective, and the group of the constant tuple otherwise; let E = ∏ S D S E = \prod_{S}D_S E = ∏ S D S be the product over the parts of S \mathcal{S} S , of order q s ′ q^{s'} q s ′ where s ′ s' s ′ is the number of parts carrying an injective β j \beta_j β j . The tuple β ( a ) \beta(a) β ( a ) restricts on each part S S S to the element of D S D_S D S with parameter a a a , so Δ ≤ E \Delta\leq E Δ ≤ E . If x ∈ E x \in E x ∈ E has parameter a S a_S a S on the part S S S , then for j ∈ S j \in S j ∈ S
σ m ( x ) π ( j ) = α π ( j ) ( x j ) = α π ( j ) ( β j ( a S ) ) = β π ( j ) ( φ ( a S ) ) , \sigma_m(x)_{\pi(j)} = \alpha_{\pi(j)}(x_j) = \alpha_{\pi(j)}(\beta_j(a_S)) = \beta_{\pi(j)}(\varphi(a_S)), σ m ( x ) π ( j ) = α π ( j ) ( x j ) = α π ( j ) ( β j ( a S )) = β π ( j ) ( φ ( a S )) ,
so σ m ( x ) \sigma_m(x) σ m ( x ) is the element of E E E with parameter φ ( a S ) \varphi(a_S) φ ( a S ) on the part π ( S ) \pi(S) π ( S ) . Hence M M M normalises E E E , and s ′ ≥ 2 s' \geq2 s ′ ≥ 2 gives ∣ E ∣ ≥ q 2 > q = ∣ Δ ∣ |E| \geq q^2 > q = |\Delta| ∣ E ∣ ≥ q 2 > q = ∣Δ∣ , so E ⊈ M E \not\subseteq M E ⊆ M while Δ ≤ E \Delta\leq E Δ ≤ E . At most one part carries an injective β j \beta_j β j . Then C C C contains all coordinates of the parts other than the one carrying an injective map, and it is non-empty because ∣ Δ ∣ = q > 1 |\Delta|=q>1 ∣Δ∣ = q > 1 ; such a part exists for the same reason. Put
E = { x ∈ N : x j = c j ( j ∈ C ) } , E=\{x\in N:x_j=c_j\ (j\in C)\}, E = { x ∈ N : x j = c j ( j ∈ C )} ,
so that ∣ E ∣ = q ℓ − ∣ C ∣ ≥ q ∣ S ∣ ≥ q 2 |E|=q^{\ell-|C|}\geq q^{|S|}\geq q^2 ∣ E ∣ = q ℓ − ∣ C ∣ ≥ q ∣ S ∣ ≥ q 2 , where S S S is that part and ∣ S ∣ ≥ 2 |S|\geq2 ∣ S ∣ ≥ 2 , and Δ ≤ E \Delta\leq E Δ ≤ E . For x ∈ E x\in E x ∈ E and j ∈ C j\in C j ∈ C the two relations above give σ m ( x ) π ( j ) = α π ( j ) ( c j ) = c π ( j ) \sigma_m(x)_{\pi(j)}=\alpha_{\pi(j)}(c_j)=c_{\pi(j)} σ m ( x ) π ( j ) = α π ( j ) ( c j ) = c π ( j ) , so σ m ( x ) ∈ E \sigma_m(x)\in E σ m ( x ) ∈ E and M M M normalises E E E ; and E ⊈ M E\nsubseteq M E ⊈ M , because E ∩ M ≤ N ∩ M = Δ E\cap M\leq N\cap M=\Delta E ∩ M ≤ N ∩ M = Δ while ∣ E ∣ > ∣ Δ ∣ |E|>|\Delta| ∣ E ∣ > ∣Δ∣ .
In either case E ≤ N E\leq N E ≤ N is normalised by M M M , satisfies Δ ≤ E \Delta\leq E Δ ≤ E , and is not contained in M M M . Moreover ( E M ) ∩ N = E (EM)\cap N=E ( E M ) ∩ N = E : an element e m ∈ N em\in N e m ∈ N with e ∈ E e\in E e ∈ E and m ∈ M m\in M m ∈ M gives m = e − 1 ( e m ) ∈ N ∩ M = Δ ≤ E m=e^{-1}(em)\in N\cap M=\Delta\leq E m = e − 1 ( e m ) ∈ N ∩ M = Δ ≤ E , hence e m ∈ E em\in E e m ∈ E . Therefore
the first inequality because E ⊈ M E\nsubseteq M E ⊈ M and the second because E M EM E M does not contain N N N . This contradicts the maximality of M M M in G G G , which is what block-primitivity asserts. The only case left is s = 1 s=1 s = 1 , where K K K is a full diagonal subgroup, v = q ℓ − 1 = b v=q^{\ell-1}=b v = q ℓ − 1 = b , and the design is symmetric.
Proposition 7.6 (The one-sided SD type is excluded for ℓ = 2 \ell=2 ℓ = 2 and the seven small simple groups). For one-sided simple diagonal type (block action of SD type, point action imprimitive quasiprimitive), Theorem 2.3 gives the unconditional necessary conditions
v ∣ q ℓ , v − 1 ∣ k ( k − 1 ) , 3 ≤ k ≤ v − 2 , v\mid q^\ell,\qquad v-1\mid k(k-1),\qquad3\leq k\leq v-2, v ∣ q ℓ , v − 1 ∣ k ( k − 1 ) , 3 ≤ k ≤ v − 2 , where q = ∣ T ∣ q=|T| q = ∣ T ∣ , ℓ \ell ℓ is the number of simple factors and v v v is the number of points. In addition, the block action of SD type forces the number of blocks b = q ℓ − 1 b=q^{\ell-1} b = q ℓ − 1 , so Fisher’s inequality b ≥ v b\geq v b ≥ v yields v ≤ q ℓ − 1 v\leq q^{\ell-1} v ≤ q ℓ − 1 ; and v = q v=q v = q is excluded because k ∣ v − 1 k\mid v-1 k ∣ v − 1 and 3 ≤ k ≤ v − 2 3\leq k\leq v-2 3 ≤ k ≤ v − 2 force k ∤ q k\nmid q k ∤ q whenever v = q v=q v = q (a direct contradiction in the single-orbit case; in the multiple-orbit case the same divisibility obstruction applies to each orbit summand). For the block size k k k one has the orbit constraints k ∣ q ∣ Out ( T ) k\mid q\mid\operatorname{Out}(T) k ∣ q ∣ Out ( T ) (single orbit) and k = ∑ i k i k=\sum_i k_i k = ∑ i k i with k i ∣ q ∣ Out ( T ) k_i\mid q\mid\operatorname{Out}(T) k i ∣ q ∣ Out ( T ) (multiple orbits), where the outer automorphism factor accounts for the action of the point stabilizer on the diagonal subgroup D ≅ T D\cong T D ≅ T of the block stabilizer; in particular for T = PSL 2 ( 8 ) T=\operatorname{PSL}_2(8) T = PSL 2 ( 8 ) one uses q ∣ Out ( T ) = 504 ⋅ 3 = 1512 q\mid\operatorname{Out}(T)=504\cdot3=1512 q ∣ Out ( T ) = 504 ⋅ 3 = 1512 (not 504).
We close the remaining cases completely for ℓ = 2 \ell= 2 ℓ = 2 and
T ∈ { A 5 , A 6 , P S L 2 ( 7 ) , P S L 2 ( 8 ) , P S L 2 ( 11 ) , P S L 2 ( 13 ) , M 11 } , T \in\{A_{5}, A_{6}, PSL_{2}(7), PSL_{2}(8), PSL_{2}(11), PSL_{2}(13), M_{11}\}, T ∈ { A 5 , A 6 , P S L 2 ( 7 ) , P S L 2 ( 8 ) , P S L 2 ( 11 ) , P S L 2 ( 13 ) , M 11 } ,
over the four global models G 0 = T 2 × ⟨ s ⟩ G_{0} = T^{2} \times\langle s\rangle G 0 = T 2 × ⟨ s ⟩ , G 1 = T 2 × ⟨ σ ^ 1 σ ^ 2 , s ⟩ G_{1} = T^{2} \times\langle\hat{\sigma}_{1}\hat{\sigma}_{2},s\rangle G 1 = T 2 × ⟨ σ ^ 1 σ ^ 2 , s ⟩ , G 2 = T 2 × ⟨ σ ^ 1 σ ^ 2 ⟩ G_{2} = T^{2} \times\langle\hat{\sigma}_{1}\hat{\sigma}_{2}\rangle G 2 = T 2 × ⟨ σ ^ 1 σ ^ 2 ⟩ and G 3 = T 2 × ( Out ( T ) 2 ⋊ S 2 ) G_{3} = T^{2} \times(\operatorname{Out}(T)^{2} \rtimes S_{2}) G 3 = T 2 × ( Out ( T ) 2 ⋊ S 2 ) (for T = M 11 T = M_{11} T = M 11 one has Out ( T ) = 1 \operatorname{Out}(T) = 1 Out ( T ) = 1 , so G 1 = G 2 = G 3 = G 0 G_{1} = G_{2} = G_{3} = G_{0} G 1 = G 2 = G 3 = G 0 ). The exclusion is effected in three layers:
Arithmetic and Fisher sieve. For each T T T , the divisibility conditions above, together with v ≤ q v \le q v ≤ q and the integrality of r = q ℓ − 1 k / v r = q^{\ell-1}k/v r = q ℓ − 1 k / v and λ = r ( k − 1 ) / ( v − 1 ) \lambda= r(k-1)/(v-1) λ = r ( k − 1 ) / ( v − 1 ) , leave only finitely many candidate pairs ( v , k ) (v,k) ( v , k ) ; for A 5 A_{5} A 5 and P S L 2 ( q ) PSL_{2}(q) P S L 2 ( q ) (q = 168 , 660 , 1092 q = 168, 660, 1092 q = 168 , 660 , 1092 ) the single-orbit and multiple-orbit arithmetic sieves leave only finitely many candidates. For v − 1 v - 1 v − 1 prime no candidate survives at all.
Subgroup-existence sieve. Any point stabilizer U U U of index v v v satisfies U ∩ N = K U \cap N = K U ∩ N = K with K K K of index e e e in N = T 2 N = T^{2} N = T 2 , where e ∈ { v , v / 2 } e \in\{v,v/2\} e ∈ { v , v /2 } according to the image of U U U in G 0 / N ≅ C 2 G_{0}/N \cong C_{2} G 0 / N ≅ C 2 . By Goursat’s lemma a subgroup of T 2 T^{2} T 2 is either a coordinate product H 1 × H 2 H_{1} \times H_{2} H 1 × H 2 or a diagonal-type subgroup of order ∣ H ∣ ≤ q |H| \le q ∣ H ∣ ≤ q ; since ∣ K ∣ = q 2 / e ≥ q 2 / v |K| = q^{2}/e \ge q^{2}/v ∣ K ∣ = q 2 / e ≥ q 2 / v and v ≤ q v \le q v ≤ q , one has ∣ K ∣ ≥ q > q |K| \ge q > q ∣ K ∣ ≥ q > q only for v < q v < q v < q ; in all surviving cases ∣ K ∣ > q |K| > q ∣ K ∣ > q , so K = H 1 × H 2 K = H_{1} \times H_{2} K = H 1 × H 2 is a coordinate product with ∣ H 1 ∣ ∣ H 2 ∣ = q 2 / e |H_{1}||H_{2}| = q^{2}/e ∣ H 1 ∣∣ H 2 ∣ = q 2 / e . Using the exact subgroup orders of each T T T , this eliminates all but finitely many v v v .
Construction check. For each surviving candidate U U U we verify [ G : U ] = v [G:U] = v [ G : U ] = v and k = [ G B : G B ∩ U ] k = [G_{B}:G_{B}\cap U] k = [ G B : G B ∩ U ] , and test quasiprimitivity of the point action by checking N G ( N 0 ) U = G N_{G}(N_{0})U = G N G ( N 0 ) U = G for every non-trivial normal subgroup N 0 ◃ G N_{0} \triangleleft G N 0 ◃ G . Every surviving candidate fails either integrality of λ \lambda λ or quasiprimitivity (typically k = v k = v k = v , i.e. the block meets every point, or N 0 U = U ≠ G N_{0}U = U \ne G N 0 U = U = G with N 0 = T × 1 ◃ G N_{0} = T \times1 \triangleleft G N 0 = T × 1 ◃ G ).
The outcome for each group is as follows: for A 5 A_{5} A 5 (q = 60 q = 60 q = 60 , Out ( T ) = C 2 \operatorname{Out}(T) = C_{2} Out ( T ) = C 2 ) all four models are empty. In G 2 = ( A 5 × A 5 ) ⋊ ⟨ τ ⟩ G_{2} = (A_{5} \times A_{5}) \rtimes\langle\tau\rangle G 2 = ( A 5 × A 5 ) ⋊ ⟨ τ ⟩ with τ = ( σ 1 , σ 2 ) \tau= (\sigma_{1},\sigma_{2}) τ = ( σ 1 , σ 2 ) , the candidate v = 15 v = 15 v = 15 has k = 15 k = 15 k = 15 ; here N = A 5 × 1 N = A_{5} \times1 N = A 5 × 1 is a non-trivial normal subgroup of G 2 G_{2} G 2 (as τ \tau τ normalizes both factors) with N ≤ U N \le U N ≤ U , so N U = U < G 2 NU = U < G_{2} N U = U < G 2 , N N N is not transitive on G 2 / U G_{2}/U G 2 / U ; indeed N N N fixes the base point U U U , and the action is not point-quasiprimitive. In G 0 G_{0} G 0 the candidate v = 36 v = 36 v = 36 has λ = 290 / 7 ∉ Z \lambda= 290/7 \notin\mathbb{Z} λ = 290/7 ∈ / Z . For P S L 2 ( 7 ) PSL_{2}(7) P S L 2 ( 7 ) , P S L 2 ( 11 ) PSL_{2}(11) P S L 2 ( 11 ) and P S L 2 ( 13 ) PSL_{2}(13) P S L 2 ( 13 ) all four models are empty (every surviving candidate has k = v k = v k = v and fails quasiprimitivity, or has no subgroup of the required order). For A 6 A_{6} A 6 (q = 360 q = 360 q = 360 , Out ( T ) = C 2 × C 2 \operatorname{Out}(T) = C_{2} \times C_{2} Out ( T ) = C 2 × C 2 , realized as PSL 2 ( 9 ) \operatorname{PSL}_{2}(9) PSL 2 ( 9 ) on the projective line) all four models are empty (G 3 G_{3} G 3 has no index-360 block stabilizer; the G 2 G_{2} G 2 and G 1 G_{1} G 1 survivors all have k = v k = v k = v with quasiprimitivity failure, or λ ∉ Z \lambda\notin\mathbb{Z} λ ∈ / Z ). For PSL 2 ( 8 ) \operatorname{PSL}_{2}(8) PSL 2 ( 8 ) (q = 504 q = 504 q = 504 , Out ( T ) = C 3 \operatorname{Out}(T) = C_{3} Out ( T ) = C 3 , field automorphism; note that PGL 2 ( 8 ) = PSL 2 ( 8 ) \operatorname{PGL}_{2}(8) = \operatorname{PSL}_{2}(8) PGL 2 ( 8 ) = PSL 2 ( 8 ) , the non-trivial outer extensions coming from field automorphisms, so the full extension is PΓL ( 2 , 8 ) \operatorname{P\Gamma L}(2, 8) PΓL ( 2 , 8 ) ) all four models are empty. Here the corrected single-orbit bound is k ∣ 1512 k \mid1512 k ∣ 1512 rather than k ∣ 504 k \mid504 k ∣ 504 ; the new factors of 1512 not dividing 504 are k = 27 d k = 27d k = 27 d with d ∣ 56 d \mid56 d ∣ 56 , of which k = 27 , 54 k = 27, 54 k = 27 , 54 lie in [ 3 , v − 2 ] [3, v - 2] [ 3 , v − 2 ] for v = 56 v = 56 v = 56 ; both fail integrality, since 55 ∤ 27 ⋅ 26 = 702 55 \nmid27 \cdot26 = 702 55 ∤ 27 ⋅ 26 = 702 and 55 ∤ 54 ⋅ 53 = 2862 55 \nmid54 \cdot53 = 2862 55 ∤ 54 ⋅ 53 = 2862 , and the same check for v = 28 , 36 , 63 v = 28, 36, 63 v = 28 , 36 , 63 yields no new integral candidates. The unique G 1 G_{1} G 1 survivor v = 56 v = 56 v = 56 has k = 56 = v k = 56 = v k = 56 = v and fails quasiprimitivity. For M 11 M_{11} M 11 (q = 7920 q = 7920 q = 7920 , Out ( T ) = 1 \operatorname{Out}(T) = 1 Out ( T ) = 1 ) the model G 0 G_{0} G 0 is empty: every U U U either contains a coordinate M 11 M_{11} M 11 (so that ⟨ M 11 × 1 , s ⟩ = G 0 \langle M_{11} \times1, s\rangle= G_{0} ⟨ M 11 × 1 , s ⟩ = G 0 forces U = G 0 U = G_{0} U = G 0 ), or lies in N N N (quasiprimitivity failure), or is one of the three explicit candidates v = 144 , 484 , 4356 v = 144, 484, 4356 v = 144 , 484 , 4356 with λ = 660 / 13 \lambda= 660/13 λ = 660/13 , λ = 360 / 23 \lambda= 360/23 λ = 360/23 , or U = G 0 U = G_{0} U = G 0 respectively. Consequently no non-trivial non-symmetric block-primitive 2-design of one-sided SD type exists for ℓ = 2 \ell= 2 ℓ = 2 and the seven small simple groups listed above, in any of the four global models. For every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 with a proper coordinate projection the case is settled for every non-abelian simple T T T by Lemma 7.1 , whose proof is a single double count and uses no restriction on T T T beyond ∣ T ∣ = q |T| = q ∣ T ∣ = q and no arithmetic input; and for ℓ = 2 \ell= 2 ℓ = 2 and ℓ = 3 \ell= 3 ℓ = 3 the all-surjective case is settled unconditionally as well, since Goursat’s theorem leaves only a full diagonal, which forces v = b v = b v = b , or a product of full factors, which forces v = 1 v = 1 v = 1 . The sieve below therefore strengthens rather than supplies the ℓ = 2 \ell= 2 ℓ = 2 exclusion, and the ℓ = 3 \ell= 3 ℓ = 3 branch is settled independently of it. At ℓ ≥ 4 \ell\ge4 ℓ ≥ 4 the all-surjective partitions with at least two parts, each of size ≥ 2 \ge2 ≥ 2 , lie outside the hypotheses of both arguments and are excluded by Lemma 7.5 , so the one-sided SD type is excluded for every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 and every non-abelian simple T T T . For ℓ ≥ 4 \ell\ge4 ℓ ≥ 4 the conclusion is obtained by the λ \lambda λ -consistency exclusion of Proposition 7.7 : every arithmetic candidate with a trivial coordinate fails the constancy-of-λ \lambda λ test, and the unique 2-homogeneous pair falls under the orbit construction of Lemma 4.2 .
Proposition 7.7 (The one-sided SD type with a trivial coordinate is excluded for every ℓ ≥ 4 ) . ∗ L e t \ell\ge4). *Let ℓ ≥ 4 ) . ∗ L e t T$ be one of the seven groups A 5 A_{5} A 5 , A 6 A_{6} A 6 , PSL 2 ( 7 ) \operatorname{PSL}_{2}(7) PSL 2 ( 7 ) , PSL 2 ( 8 ) \operatorname{PSL}_{2}(8) PSL 2 ( 8 ) , PSL 2 ( 11 ) \operatorname{PSL}_{2}(11) PSL 2 ( 11 ) , PSL 2 ( 13 ) \operatorname{PSL}_{2}(13) PSL 2 ( 13 ) , M 11 M_{11} M 11 and let S = T ℓ S = T^{\ell} S = T ℓ , ℓ ≥ 4 \ell\ge4 ℓ ≥ 4 , be the socle of a one-sided SD candidate whose point stabilizer admits at least two non-trivial projection indices e 3 , e 4 > 1 e_{3}, e_{4} > 1 e 3 , e 4 > 1 (and arbitrarily many further non-trivial
indices). Fix two such coordinates. In this model the point-pair orbits of S S S on the point set are classified by the double-coset system K 3 \ T / K 4 K_3\backslash T/K_4 K 3 \ T / K 4 , where K i < T K_i<T K i < T has index e i e_i e i ; write ndc ( e 3 , e 4 ) = ∣ K 3 \ T / K 4 ∣ \operatorname{ndc}(e_3,e_4)=|K_3\backslash T/K_4| ndc ( e 3 , e 4 ) = ∣ K 3 \ T / K 4 ∣ . Then:
If ndc ( e 3 , e 4 ) > 1 \operatorname{ndc}(e_3,e_4)>1 ndc ( e 3 , e 4 ) > 1 , no 2-design exists whose blocks are unions of orbits of the block stabilizer. Indeed, for a double-coset class w w w let d w d_w d w be the number of candidate blocks containing a pair of points of class w w w , and let e w , w ′ e_{w,w'} e w , w ′ be the number containing a pair of points of the two distinct classes w , w ′ w,w' w , w ′ ; the design condition λ ( x , y ) = λ 0 \lambda(x,y)=\lambda_0 λ ( x , y ) = λ 0 for every pair of points forces d w = λ 0 = e w , w ′ d_w=\lambda_0=e_{w,w'} d w = λ 0 = e w , w ′ for all w , w ′ w,w' w , w ′ . Since e w , w ′ ≤ d w e_{w,w'}\leq d_w e w , w ′ ≤ d w with equality only if every block containing class w w w contains every other class, constancy of λ \lambda λ forces every block to contain all classes, i.e. every block equals the whole point set, contradicting 2 ≤ k ≤ v − 2 2\leq k\leq v-2 2 ≤ k ≤ v − 2 .
If ndc ( e 3 , e 4 ) = 1 \operatorname{ndc}(e_3,e_4)=1 ndc ( e 3 , e 4 ) = 1 , then S S S is 2-homogeneous on the point set (a single point-pair orbit), and such candidates are covered by the 2-homogeneous orbit construction of Lemma 4.2 ; they therefore contribute no additional non-2-homogeneous example.
For all seven groups and every pair ( e 3 , e 4 ) (e_3,e_4) ( e 3 , e 4 ) arising from the arithmetic sieve of Theorem 2.3 , the GAP computation yields ndc ( e 3 , e 4 ) > 1 \operatorname{ndc}(e_3,e_4)>1 ndc ( e 3 , e 4 ) > 1 , with the single exception ( e 3 , e 4 ) = ( 11 , 720 ) (e_3,e_4)=(11,720) ( e 3 , e 4 ) = ( 11 , 720 ) for T = M 11 T=M_{11} T = M 11 , for which ndc = 1 \operatorname{ndc}=1 ndc = 1 (the 2-homogeneous case of (2); here K 3 ≅ M 10 K_3\cong M_{10} K 3 ≅ M 10 and K 4 ≅ C 11 K_4\cong C_{11} K 4 ≅ C 11 ). Consequently every one-sided SD candidate satisfying the hypothesis of two proper projection indices is excluded for every ℓ ≥ 4 \ell\geq4 ℓ ≥ 4 within the seven groups. By Lemma 7.1 , however, a single proper coordinate projection already forces k = v k=v k = v , so that hypothesis can never be met, and the arithmetic sieve of (1)–(2) is retained as the record of the parameter constraints rather than as the exclusion itself. The configurations that would survive at ℓ ≥ 4 \ell\geq4 ℓ ≥ 4 are the all-surjective partitions, excluded by Lemma 7.5 .
Lemma 7.8 (Prime ℓ \ell ℓ : theoretical exclusion). This lemma is stated within the flag-transitive framework, where the block-size formula (5) applies; it serves the symmetric (Menon) analysis, while the non-flag-transitive non-symmetric case is excluded independently by the orbit-union sieve of layers (i)–(ii) of Proposition 7.11. Let T T T be one of the six socles A 5 A_5 A 5 , A 6 A_6 A 6 , PSL 2 ( 7 ) \operatorname{PSL}_2(7) PSL 2 ( 7 ) , PSL 2 ( 8 ) \operatorname{PSL}_2(8) PSL 2 ( 8 ) , PSL 2 ( 11 ) \operatorname{PSL}_2(11) PSL 2 ( 11 ) , PSL 2 ( 13 ) \operatorname{PSL}_2(13) PSL 2 ( 13 ) , let ( m , n ) (m,n) ( m , n ) be a listed primitive degree pair with A < m A<m A < m , and let ℓ ≥ 23 \ell\geq23 ℓ ≥ 23 be prime. Suppose in addition
that the coordinate index j = ∣ C B ⋅ α ∣ j = \lvert C_B \cdot\alpha\rvert j = ∣ C B ⋅ α ∣ of (5) satisfies j ≤ ℓ j \le\ell j ≤ ℓ (this holds in particular when α \alpha α has at most two distinct coordinate values, one of them occurring ℓ − 1 \ell- 1 ℓ − 1 times; without this hypothesis the numerical sieve below is applied to every admissible j ∣ ℓ ! j \mid\ell! j ∣ ℓ ! ). Then no such flag-transitive candidate exists: the inequality of Lemma 2.17 fails at every prime ℓ ≥ 23 \ell\ge23 ℓ ≥ 23 , and for admissible j j j with ℓ < j ∣ ℓ ! \ell< j \mid\ell! ℓ < j ∣ ℓ ! the congruence A ℓ j ≡ 1 ( m o d Φ ℓ ( m ) ) A^{\ell}j \equiv1 \pmod{\Phi_{\ell}(m)} A ℓ j ≡ 1 ( mod Φ ℓ ( m )) has no solution among the divisors of ℓ ! \ell! ℓ ! whose prime divisors are at most ℓ \ell ℓ (verified by direct computation for every prime 23 ≤ ℓ ≤ 127 23 \le\ell\le127 23 ≤ ℓ ≤ 127 and every admissible pair); for prime ℓ > 127 \ell> 127 ℓ > 127 the same congruence check applies verbatim.
Proof. Every prime divisor of m − 1 m - 1 m − 1 is at most 13, so ℓ ∤ m − 1 \ell\nmid m - 1 ℓ ∤ m − 1 . By the cyclotomic factorization m ℓ − 1 = ( m − 1 ) Φ ℓ ( m ) m^{\ell} - 1 = (m - 1)\Phi_{\ell}(m) m ℓ − 1 = ( m − 1 ) Φ ℓ ( m ) and the LTE formula, v q ( m ℓ − 1 ) = v q ( m − 1 ) v_q(m^{\ell} - 1) = v_q(m - 1) v q ( m ℓ − 1 ) = v q ( m − 1 ) for every prime q ∣ m − 1 q \mid m - 1 q ∣ m − 1 , while every prime divisor q q q of Φ ℓ ( m ) \Phi_{\ell}(m) Φ ℓ ( m ) satisfies ord q ( m ) = ℓ \operatorname{ord}_q(m) = \ell ord q ( m ) = ℓ (the alternative q = ℓ q = \ell q = ℓ would force m ≡ 1 ( m o d ℓ ) m \equiv1 \pmod{\ell} m ≡ 1 ( mod ℓ ) , i.e. ℓ ∣ m − 1 \ell\mid m - 1 ℓ ∣ m − 1 , impossible). Hence q ≡ 1 ( m o d ℓ ) q \equiv1 \pmod{\ell} q ≡ 1 ( mod ℓ ) , so q > ℓ q > \ell q > ℓ for every q ∣ Φ ℓ ( m ) q \mid\Phi_{\ell}(m) q ∣ Φ ℓ ( m ) , and therefore
P ( ℓ ) : = ∏ q ≤ ℓ q v q ( m ℓ − 1 ) = m − 1 , M ′ ′ : = m ℓ − 1 P ( ℓ ) = Φ ℓ ( m ) . P(\ell) := \prod_{q \le\ell} q^{v_q(m^{\ell}-1)} = m - 1,\qquad M^{\prime\prime} := \frac{m^{\ell} - 1}{P(\ell)} = \Phi_{\ell}(m). P ( ℓ ) := q ≤ ℓ ∏ q v q ( m ℓ − 1 ) = m − 1 , M ′′ := P ( ℓ ) m ℓ − 1 = Φ ℓ ( m ) .
By Lemma 2.17 , a candidate with A < m A < m A < m and j ≤ ℓ j \le\ell j ≤ ℓ would force Φ ℓ ( m ) ≤ A ℓ − 1 \Phi_{\ell}(m) \le A^{\ell}\!-\!1 Φ ℓ ( m ) ≤ A ℓ − 1 , i.e. ( m / A ) ℓ ≤ ℓ ( m − 1 ) + o ( 1 ) (m/A)^{\ell} \le\ell(m - 1) + o(1) ( m / A ) ℓ ≤ ℓ ( m − 1 ) + o ( 1 ) . For every listed pair and every A < m A < m A < m this inequality fails for every prime $\ell \ge 23 (the only borderline case, A 6 ( 6 , 15 ) A_6(6, 15) A 6 ( 6 , 15 ) with A = 5 A = 5 A = 5 at ℓ = 23 \ell= 23 ℓ = 23 , is excluded by the finite check against (5) already contained in the sieve of Proposition 7.9 (3); and the ratio ( m / A ) ℓ / [ ℓ ( m − 1 ) ] (m/A)^{\ell}/[\ell(m - 1)] ( m / A ) ℓ / [ ℓ ( m − 1 )] is increasing in ℓ \ell ℓ once the inequality holds, with the minimal ratio m / A = 6 / 5 m/A = 6/5 m / A = 6/5 already failing at ℓ = 29 \ell= 29 ℓ = 29 ). Thus no candidate with j ≤ ℓ j \le\ell j ≤ ℓ exists for any prime ℓ ≥ 23 \ell\ge23 ℓ ≥ 23 ; for admissible j j j with ℓ < j ∣ ℓ ! \ell< j \mid\ell! ℓ < j ∣ ℓ ! the prime-divisor bound on j j j of (5) is applied in the numerical sieve of Proposition 7.9 (3), executed for every prime $\ell \le 127: for each listed pair the congruence A ℓ j ≡ 1 ( m o d Φ ℓ ( m ) ) A^{\ell}j \equiv1 \pmod{\Phi_{\ell}(m)} A ℓ j ≡ 1 ( mod Φ ℓ ( m )) has no solution among the divisors of ℓ ! \ell! ℓ ! whose prime divisors are at most ℓ \ell ℓ (verified by direct computation for every prime 23 ≤ ℓ ≤ 127 23 \le\ell\le127 23 ≤ ℓ ≤ 127 and every admissible pair), so these j j j are excluded as well. For prime ℓ > 127 \ell> 127 ℓ > 127 and j > ℓ j > \ell j > ℓ the same congruence check applies verbatim: by Lemma 2.17 the divisor Φ ℓ ( m ) = M ′ ′ \Phi_{\ell}(m) = M^{\prime\prime} Φ ℓ ( m ) = M ′′ of m ℓ − 1 m^{\ell} - 1 m ℓ − 1 still divides k − 1 = A ℓ j − 1 k - 1 = A^{\ell}j - 1 k − 1 = A ℓ j − 1 , and this congruence is the sole arithmetic ingredient of the PA sieve, the structural exclusions (the block-action types of Lemma 2.14 and Proposition 2.15 , the box model of Lemma 2.18 and the case j ≤ ℓ j \le\ell j ≤ ℓ settled above) not depending on the size of ℓ \ell ℓ ; the check is deterministic for every prime ℓ \ell ℓ , and it is executed here for every prime 23 ≤ ℓ ≤ 127 23 \le\ell\le127 23 ≤ ℓ ≤ 127 , matching the range of Proposition 7.9 (3).
Proposition 7.9 (PA type: exclusion for the six socles). The three-class exclusion below is stated within the flag-transitive framework, where the block-size formula (5) applies and serves the symmetric (Menon) analysis; the general non-flag-transitive non-symmetric case is excluded independently by the orbit-union sieve of layers (i)–(ii) of Proposition 7.11, using the orbit-union constraint in place of (5) . Let T T T be one of the six socles A 5 A_5 A 5 , A 6 A_6 A 6 , PSL 2 ( 7 ) \operatorname{PSL}_2(7) PSL 2 ( 7 ) , PSL 2 ( 8 ) \operatorname{PSL}_2(8) PSL 2 ( 8 ) , PSL 2 ( 11 ) \operatorname{PSL}_2(11) PSL 2 ( 11 ) , PSL 2 ( 13 ) \operatorname{PSL}_2(13) PSL 2 ( 13 ) and let ( m , n ) (m,n) ( m , n ) be a pair of primitive degrees of T T T with m < n m<n m < n . The primitive degrees of the six socles are the seventeen entries of Table 3 , and they form sixteen pairs ( m , n ) (m,n) ( m , n ) with m < n m<n m < n , which are exactly the pairs listed in (i)–(iii) below; here m = ∣ T : H ∣ m=|T:H| m = ∣ T : H ∣ is a point degree and n = ∣ T : K ∣ n=|T:K| n = ∣ T : K ∣ a block degree (H , K H,K H , K maximal subgroups of T T T ). In the standard product-action model of § 2.1 on P \mathcal{P} P and B = Δ ℓ \mathcal{B}=\Delta^\ell B = Δ ℓ (∣ Δ ∣ = m |\Delta|=m ∣Δ∣ = m ), a flag-transitive block-primitive 2-design of PA type has block size k = A ℓ ⋅ j k=A^\ell\cdot j k = A ℓ ⋅ j by (5) , where A = [ K : K ∩ H ] A=[K:K\cap H] A = [ K : K ∩ H ] and 1 ≤ j ∣ ℓ ! 1\le j\mid\ell! 1 ≤ j ∣ ℓ ! (every prime divisor of j j j is at most ℓ \ell ℓ ). Let A A A also denote the multiset of K K K -orbit sizes on Δ \Delta Δ (so that ∑ a ∈ A a = m \sum_{a\in A}a=m ∑ a ∈ A a = m and Σ 2 = ∑ a ∈ A a 2 \Sigma_2=\sum_{a\in A}a^2 Σ 2 = ∑ a ∈ A a 2 ), and put g = gcd ( m , n ) g=\gcd(m,n) g = g cd( m , n ) , M = m / g M=m/g M = m / g , N = n / g N=n/g N = n / g . The K K K -boxes of Lemma 2.18 have sizes ∏ i = 1 ℓ a i \prod_{i=1}^{\ell}a_i ∏ i = 1 ℓ a i (a i ∈ A a_i\in A a i ∈ A ), and the fiber double-counting
∣ O f i b e r ∣ = Σ 2 ℓ − m ℓ 2 , b ⋅ m f i b e r = λ ⋅ ∣ O f i b e r ∣ , b = n ℓ , |\mathcal{O}_{\mathrm{fiber}}|=\frac{\Sigma_2^\ell-m^\ell}{2},\qquad b\cdot m_{\mathrm{fiber}}=\lambda\cdot|\mathcal{O}_{\mathrm{fiber}}|,\qquad b=n^\ell, ∣ O fiber ∣ = 2 Σ 2 ℓ − m ℓ , b ⋅ m fiber = λ ⋅ ∣ O fiber ∣ , b = n ℓ , is exact for every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 . Then, for every listed pair ( T ; m , n ) (T;m,n) ( T ; m , n ) and every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 :
six pairs with A = m A=m A = m or a single K K K -orbit ( A 5 ( 5 , 6 ) , A 6 ( 6 , 10 ) , PSL 2 ( 7 ) ( 7 , 8 ) ; PSL 2 ( 8 ) ( 9 , 28 ) , PSL 2 ( 11 ) ( 11 , 12 ) , PSL 2 ( 11 ) ( 12 , 55 ) ) (A_5(5,6), A_6(6,10), \operatorname{PSL}_2(7)(7,8); \operatorname{PSL}_2(8)(9,28), \operatorname{PSL}_2(11)(11,12), \operatorname{PSL}_2(11)(12,55)) ( A 5 ( 5 , 6 ) , A 6 ( 6 , 10 ) , PSL 2 ( 7 ) ( 7 , 8 ) ; PSL 2 ( 8 ) ( 9 , 28 ) , PSL 2 ( 11 ) ( 11 , 12 ) , PSL 2 ( 11 ) ( 12 , 55 )) are excluded for every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 : k = A ℓ j ≥ m ℓ = v k=A^\ell j\ge m^\ell=v k = A ℓ j ≥ m ℓ = v violates 2 < k < v − 1 2<k<v-1 2 < k < v − 1 ;
three pairs with a unique K K K -box size and s = g s=g s = g ( A 5 ( 6 , 10 ) , PSL 2 ( 8 ) ( 28 , 36 ) ; PSL 2 ( 13 ) ( 14 , 78 ) ) (A_5(6,10), \operatorname{PSL}_2(8)(28,36); \operatorname{PSL}_2(13)(14,78)) ( A 5 ( 6 , 10 ) , PSL 2 ( 8 ) ( 28 , 36 ) ; PSL 2 ( 13 ) ( 14 , 78 )) are excluded for every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 : the fiber double-counting reduces, after the divisibility simplification, to the incompatible inequalities n t ≥ g ℓ nt\ge g^\ell n t ≥ g ℓ and n t < g ℓ nt<g^\ell n t < g ℓ for t = k / M ℓ t=k/M^\ell t = k / M ℓ ;
(iii) seven pairs with p ∣ g p \mid g p ∣ g and p ∤ Σ 2 p \nmid\Sigma_2 p ∤ Σ 2 (with p = 5 , 3 , 5 , 3 , 11 , 7 , 13 p = 5, 3, 5, 3, 11, 7, 13 p = 5 , 3 , 5 , 3 , 11 , 7 , 13 for A 5 ( 5 , 10 ) A_5(5, 10) A 5 ( 5 , 10 ) , A 6 ( 6 , 15 ) A_6(6, 15) A 6 ( 6 , 15 ) , A 6 ( 10 , 15 ) A_6(10, 15) A 6 ( 10 , 15 ) , P S L 2 ( 8 ) ( 9 , 36 ) PSL_2(8)(9, 36) P S L 2 ( 8 ) ( 9 , 36 ) , P S L 2 ( 11 ) ( 11 , 55 ) PSL_2(11)(11, 55) P S L 2 ( 11 ) ( 11 , 55 ) , P S L 2 ( 13 ) ( 14 , 91 ) PSL_2(13)(14, 91) P S L 2 ( 13 ) ( 14 , 91 ) , P S L 2 ( 13 ) ( 78 , 91 ) PSL_2(13)(78, 91) P S L 2 ( 13 ) ( 78 , 91 ) , respectively) are excluded as follows. Since v p ( ∣ O f i b e r ∣ ) = 0 v_p(|\mathcal{O}_{\mathrm{fiber}}|) = 0 v p ( ∣ O fiber ∣ ) = 0 and v p ( g ) = 1 v_p(g) = 1 v p ( g ) = 1 , the fiber double-counting forces p ∤ t p \nmid t p ∤ t and the unique candidate t ≡ t 0 ≡ M − ℓ ( m o d p ℓ ) t \equiv t_0 \equiv M^{-\ell} \pmod{p^\ell} t ≡ t 0 ≡ M − ℓ ( mod p ℓ ) with 1 ≤ t 0 < p ℓ < g ℓ 1 \leq t_0 < p^\ell< g^\ell 1 ≤ t 0 < p ℓ < g ℓ (t = k / M ℓ t = k/M^\ell t = k / M ℓ ); for the three pairs with M = 1 M = 1 M = 1 (A 5 ( 5 , 10 ) A_5(5, 10) A 5 ( 5 , 10 ) , P S L 2 ( 8 ) ( 9 , 36 ) PSL_2(8)(9, 36) P S L 2 ( 8 ) ( 9 , 36 ) , P S L 2 ( 11 ) ( 11 , 55 ) PSL_2(11)(11, 55) P S L 2 ( 11 ) ( 11 , 55 ) ) the unique candidate is t 0 = 1 t_0 = 1 t 0 = 1 , so k = 1 k = 1 k = 1 : degenerate, excluded for every ℓ ≥ 2 \ell\geq2 ℓ ≥ 2 . For the four pairs with M > 1 M > 1 M > 1 (A 6 ( 6 , 15 ) A_6(6, 15) A 6 ( 6 , 15 ) , A 6 ( 10 , 15 ) A_6(10, 15) A 6 ( 10 , 15 ) , P S L 2 ( 13 ) ( 14 , 91 ) PSL_2(13)(14, 91) P S L 2 ( 13 ) ( 14 , 91 ) , P S L 2 ( 13 ) ( 78 , 91 ) PSL_2(13)(78, 91) P S L 2 ( 13 ) ( 78 , 91 ) ), g g g is prime and c = ( k − 1 ) / p ℓ c = (k - 1)/p^\ell c = ( k − 1 ) / p ℓ is uniquely determined by c ≡ − p − ℓ ( m o d M ℓ ) c \equiv-p^{-\ell} \pmod{M^\ell} c ≡ − p − ℓ ( mod M ℓ ) ; the resulting k = M ℓ t 0 k = M^\ell t_0 k = M ℓ t 0 , r = N ℓ t 0 r = N^\ell t_0 r = N ℓ t 0 and λ = r ( k − 1 ) / ( m ℓ − 1 ) \lambda= r(k-1)/(m^\ell-1) λ = r ( k − 1 ) / ( m ℓ − 1 ) fail the λ \lambda λ -integrality or the fiber-counting divisibility and bounds checks. The unique candidate is excluded for every ℓ ≥ 2 \ell\geq2 ℓ ≥ 2 by the λ \lambda λ -integrality and the fiber-counting divisibility and bounds checks.
The prime-power congruence criterion (Lemma 2.16 ) applies as a further necessary condition in all cases.
Lemma 7.10 (The block-count alternatives for a product point action). Let D \mathcal{D} D be a non-trivial 2 2 2 -( v , k , λ ) (v,k,\lambda) ( v , k , λ ) design and G ≤ Aut ( D ) G \leq\operatorname{Aut}(\mathcal{D}) G ≤ Aut ( D ) transitive on points and primitive on blocks, with Soc ( G ) = T ℓ \operatorname{Soc}(G) = T^\ell Soc ( G ) = T ℓ of product action type on points of degree v = m ℓ v = m^\ell v = m ℓ (ℓ ≥ 2 \ell\geq2 ℓ ≥ 2 , T T T non-abelian simple, m m m a primitive degree of T T T ). Then the block action of G G G on B \mathcal{B} B is primitive, and one of the following exhaustive alternatives holds for the block count b b b :
(1) b = n ℓ b = n^\ell b = n ℓ with n = [ T : K ] ≥ m n = [T : K] \geq m n = [ T : K ] ≥ m the block-coordinate index (product action block action);
(2) b = ∣ T ∣ ℓ − s b = |T|^{\ell-s} b = ∣ T ∣ ℓ − s with 1 ≤ s ≤ ℓ − 1 1 \leq s \leq\ell- 1 1 ≤ s ≤ ℓ − 1 (simple or compound diagonal block action);
(3) b = ∣ T ∣ ℓ b = |T|^\ell b = ∣ T ∣ ℓ (twisted wreath block action);
(4) b = ∣ T ∣ b = |T| b = ∣ T ∣ (holomorph simple, ℓ = 2 \ell= 2 ℓ = 2 ) or b = ∣ T ∣ ℓ / 2 b = |T|^{\ell/2} b = ∣ T ∣ ℓ /2 (holomorph compound, ℓ \ell ℓ even);
(5) b = p a b = p^a b = p a , a a a prime power (affine block action);
(6) b ∣ ∣ T ∣ b \mid|T| b ∣ ∣ T ∣ and G B G^{\mathcal{B}} G B is almost simple with socle T T T (the image of N = T ℓ N = T^\ell N = T ℓ is a single factor).
Proof. The block action of G G G is primitive by hypothesis. The socle N = T ℓ N=T^{\ell} N = T ℓ is transitive on B \mathcal{B} B : by the primitivity of the block action the block stabilizer G β G_{\beta} G β is maximal in G G G , and N ≰ G β N\nleq G_{\beta} N ≰ G β (N N N is transitive on points, and the block action is not trivial), so N G β = G NG_{\beta}=G N G β = G and N B N^{\mathcal{B}} N B is transitive. Hence the minimal normal subgroup of the primitive block action G B G^{\mathcal{B}} G B is an image of N N N , and the O’Nan–Scott classification of primitive groups applies to G B G^{\mathcal{B}} G B . The correspondence with the standard O’Nan–Scott types is as follows (with M M M the socle of G B G^{\mathcal{B}} G B , an image of N = T ℓ N=T^{\ell} N = T ℓ , and ℓ ≥ 2 \ell\geq2 ℓ ≥ 2 ): (1) M ≅ T ℓ M\cong T^{\ell} M ≅ T ℓ acting in product action on Δ ℓ \Delta^{\ell} Δ ℓ , the product action (PA) type, block count b = n ℓ b=n^{\ell} b = n ℓ , n = [ T : K ] n=[T:K] n = [ T : K ] the block-coordinate index; (2) M M M a diagonal subgroup of T r ′ T^{r'} T r ′ (1 ≤ r ′ ≤ ℓ 1\leq r'\leq\ell 1 ≤ r ′ ≤ ℓ , the remaining ℓ − r ′ \ell-r' ℓ − r ′ coordinates acting freely), the diagonal type, simple diagonal (SD) for r ′ = ℓ r'=\ell r ′ = ℓ (socle a full diagonal subgroup) and compound diagonal (CD) for 1 ≤ r ′ < ℓ 1\leq r'<\ell 1 ≤ r ′ < ℓ , block count b = ∣ T ∣ ℓ − r ′ b=|T|^{\ell-r'} b = ∣ T ∣ ℓ − r ′ ; (3) M ≅ T ℓ M\cong T^{\ell} M ≅ T ℓ regular on blocks, the twisted wreath (TW) type, b = ∣ T ∣ ℓ b=|T|^{\ell} b = ∣ T ∣ ℓ ; (4) two commuting minimal normal subgroups M 1 ≅ M 2 M_{1}\cong M_{2} M 1 ≅ M 2 , the holomorph type: HS with M 1 ≅ M 2 ≅ T M_{1}\cong M_{2}\cong T M 1 ≅ M 2 ≅ T (so ℓ = 2 \ell=2 ℓ = 2 ) and block count b = ∣ T ∣ b=|T| b = ∣ T ∣ ; HC with M 1 ≅ M 2 ≅ T m M_{1}\cong M_{2}\cong T^{m} M 1 ≅ M 2 ≅ T m where m = ℓ / 2 m=\ell/2 m = ℓ /2 and ℓ \ell ℓ is even, so the ℓ \ell ℓ simple factors of N N N split into two minimal normal subgroups of T m T^{m} T m each, and b = ∣ T ∣ m = ∣ T ∣ ℓ / 2 b=|T|^{m}=|T|^{\ell/2} b = ∣ T ∣ m = ∣ T ∣ ℓ /2 ; (5) M M M elementary abelian, the affine (HA) type, b = p a b=p^{a} b = p a a prime power; (6) M ≅ T M\cong T M ≅ T a single non-abelian simple factor, the almost simple (AS) type, which occurs precisely when N N N acts on blocks through a single quotient (the other ℓ − 1 \ell-1 ℓ − 1 factors act trivially on blocks); since N N N is transitive on blocks, then b ∣ ∣ T ∣ b\mid|T| b ∣ ∣ T ∣ . The types are mutually exclusive by the O’Nan–Scott trichotomy: (5) is the only case with elementary abelian socle; (6) the only case with a single simple socle; (3) the only case with a regular non-abelian socle; (4) the only case with two minimal normal subgroups; and (2) is distinguished from (1) by M M M being a diagonal rather than a product subgroup (in (2) the coordinate stabilizers act through a diagonal subgroup, so M M M is not a direct product of transitive factors as in (1)). The displayed block counts follow from orbit–stabilizer as recorded. Alternatives (2)–(5) are excluded as follows. (2) Diagonal: the corresponding block sizes are impossible by the box-closure analysis of Lemma 2.14 (ℓ = 2 \ell=2 ℓ = 2 ) and Proposition 2.15 (ℓ ≥ 3 \ell\geq3 ℓ ≥ 3 ). (3) Twisted wreath: the point action of TW type has its socle N = T ℓ N=T^{\ell} N = T ℓ regular on points, so by the regular-normal-subgroup lemma (a block-primitive non-trivial 2-design with a normal subgroup regular on points is symmetric, cf. Lemma 2.10 ) the design is symmetric, contradicting the non-symmetric hypothesis; the regular subgroup to which the lemma applies is the socle of the point action. (4) Holomorph: the block action of holomorph type has two commuting transitive factors, M 1 ≅ M 2 ≅ T M_1 \cong M_2 \cong T M 1 ≅ M 2 ≅ T (HS, ℓ = 2 \ell= 2 ℓ = 2 ) or M 1 ≅ M 2 ≅ T m M_1 \cong M_2 \cong T^m M 1 ≅ M 2 ≅ T m with m = ℓ / 2 m = \ell/2 m = ℓ /2 (HC, ℓ \ell ℓ even; these are the two minimal normal subgroups; the ℓ \ell ℓ simple factors of N N N split between them, no individual simple factor being normal); two commuting transitive permutation groups are regular, so each factor is regular on blocks; each factor is also transitive on points (the holomorph point action has the same two factors), hence regular on points by Lemma 2.12 , and b = ∣ T ∣ m = v b = |T|^m = v b = ∣ T ∣ m = v (HS: m = 1 m = 1 m = 1 ; HC: m = ℓ / 2 m = \ell/2 m = ℓ /2 ), contradicting b ≠ v b \ne v b = v . (5) Affine: every quotient of T ℓ T^\ell T ℓ is a product of copies of the non-abelian simple group T T T , never an elementary abelian group, so the affine block action (whose socle is elementary abelian) cannot arise. Alternative (6) is excluded by the orbit-union sieve of Proposition 7.11 layers (i)–(ii): whether the block action is of type (1) or (6), every block B ∋ α B \ni\alpha B ∋ α is a union of G B , α = G B ∩ G α G_{B,\alpha} = G_B \cap G_\alpha G B , α = G B ∩ G α -orbits (equivalently of K ℓ K^\ell K ℓ -boxes, since K ℓ ≤ G B , α K^\ell\le G_{B,\alpha} K ℓ ≤ G B , α ), and the sieve eliminates all admissible ( k , b ) (k,b) ( k , b ) simultaneously; for the ten 2-homogeneous pairs G α G_\alpha G α is transitive on P \mathcal{P} P , so B = P B = \mathcal{P} B = P and k = v k = v k = v , contradiction, and for the seven remaining pairs the enumeration leaves no feasible block size. Hence b = n ℓ b = n^\ell b = n ℓ .
Proposition 7.11 (The product action type is completely excluded for the six socles and the listed primitive degree pairs, for symmetric and non-symmetric 2-designs alike, for every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 ). *Let D \mathcal{D} D be a non-trivial 2 2 2 -( v , k , λ ) (v, k, \lambda) ( v , k , λ ) design, and let G ≤ Aut ( D ) G \le\operatorname{Aut}(\mathcal{D}) G ≤ Aut ( D ) act transitively on points and primitively on blocks, with socle N = T ℓ N = T^\ell N = T ℓ of product action type on points of degree v = m ℓ v = m^\ell v = m ℓ , where m m m is a primitive degree of T T T and ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 . Both the symmetric case (b = v b = v b = v ) and the non-symmetric case (b ≠ v b \ne v b = v ) are covered: the non-symmetric product-action configuration is excluded by combining the orbit-wise criterion and the integrality constraints of steps (i)–(ii) below with the block-action exclusion of Lemma 2.14 and Proposition 2.15 , and the symmetric product-action configuration in steps (iii)–(v). For T T T one of the six socles A 5 A_5 A 5 , A 6 A_6 A 6 , P S L 2 ( 7 ) PSL_2(7) P S L 2 ( 7 ) , P S L 2 ( 8 ) PSL_2(8) P S L 2 ( 8 ) , P S L 2 ( 11 ) PSL_2(11) P S L 2 ( 11 ) , P S L 2 ( 13 ) PSL_2(13) P S L 2 ( 13 ) and m m m a listed primitive degree (the seventeen primitive degrees of Table 3 ), no such design exists for any ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 and any block stabilizer. For non-symmetric block-primitive 2-designs this is the complete exclusion for every non-abelian simple socle: by Proposition 3.4 the point action is primitive, and by the classification of finite primitive permutation groups together with the standard degree tables of Dixon–Mortimer [24 ] (derived from the minimal-degree bounds of Liebeck–Saxl [32 ] ), an exhaustive reduction over the admissible primitive de-
:::`
grees shows that a non-2 2 2 -homogeneous point action has socle one of the six groups above with one of the listed degree pairs; the 2 2 2 -homogeneous pairs are excluded by the general argument of step (ii).
Proof. We record the five-layer sieve whose layers are mutually independent and each terminates in a definite conclusion. Every layer is structural: it involves only the suborbit sizes of the base primitive action of T T T and the arithmetic of the parameters ( v , k , λ ) (v,k,\lambda) ( v , k , λ ) , hence applies verbatim for every non-abelian simple socle and every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 . The complete execution of the sieve for the six socles A 5 A_5 A 5 , A 6 A_6 A 6 , P S L 2 ( 7 ) PSL_2(7) P S L 2 ( 7 ) , P S L 2 ( 8 ) PSL_2(8) P S L 2 ( 8 ) , P S L 2 ( 11 ) PSL_2(11) P S L 2 ( 11 ) , P S L 2 ( 13 ) PSL_2(13) P S L 2 ( 13 ) and 2 ≤ ℓ ≤ 9 2 \le\ell\le9 2 ≤ ℓ ≤ 9 is recorded; for non-symmetric block-primitive 2 2 2 -designs this execution is complete in socle: the point action is primitive (Proposition 3.4 ), and the reduction over the admissible primitive degrees (from the degree tables of Dixon–Mortimer [24 ] , built on the minimal-degree bounds of Liebeck–Saxl [32 ] ) lists precisely the seventeen primitive degrees of Table 3, all within the six socles, while the 2 2 2 -homogeneous pairs are excluded by the general argument of step (ii).
(i) Orbit-wise Cameron–Praeger criterion for an arbitrary block stabilizer. The orbit form (5) of the Cameron–Praeger criterion (Theorem 2.19 ) depends only on the sizes of the G G G -orbits on unordered pairs; for a fixed point α \alpha α of a block B B B these are the G α G_\alpha G α -orbits on P ∖ { α } \mathcal{P} \setminus\{\alpha\} P ∖ { α } , of sizes c s i d c s_i^d c s i d with 1 ≤ c ≤ ( ℓ d ) 1 \le c \le\binom{\ell}{d} 1 ≤ c ≤ ( d ℓ ) , where s i s_i s i runs over the non-trivial suborbits of ( T , m ) (T,m) ( T , m ) and d ∈ { 1 , … , ℓ } d \in\{1,\ldots,\ell\} d ∈ { 1 , … , ℓ } . The criterion therefore applies verbatim to arbitrary block stabilizers L = N ∩ G β L=N\cap G_\beta L = N ∩ G β , box, subdirect, or of any other structure, since no information about L L L enters the condition. For the six socles and 2 ≤ ℓ ≤ 9 2 \le\ell\le9 2 ≤ ℓ ≤ 9 the criterion (for every box type ( i , d ) (i,d) ( i , d ) there must exist c c c with ( v − 1 ) / gcd ( v − 1 , c s i d ) (v-1)/\gcd(v-1,c s_i^d) ( v − 1 ) / g cd( v − 1 , c s i d ) dividing k ( k − 1 ) k(k-1) k ( k − 1 ) ) admits only finitely many parameter candidates, and every candidate that survives it is killed in the later layers; for socles outside the six and ℓ ≥ 10 \ell\ge10 ℓ ≥ 10 the same orbit-wise condition applies verbatim, since only the suborbit sizes of the base action enter; the failure for all c c c in one box type kills the candidate regardless of the structure of L L L . Two points about the range of c c c are immediate: the lower bound c ≥ 1 c \ge1 c ≥ 1 holds because the criterion ratio k ( k − 1 ) / ( v ( v − 1 ) ) k(k-1)/(v(v-1)) k ( k − 1 ) / ( v ( v − 1 )) is positive, so every pair orbit Q i Q_i Q i contributes q i > 0 q_i > 0 q i > 0 in-block pairs and hence meets the block B B B ; and the upper bound c ≤ ( ℓ d ) c \le\binom{\ell}{d} c ≤ ( d ℓ ) is the number of d d d -coordinate boxes of a G α G_\alpha G α -orbit in the box decomposition, an orbit-level fact independent of L L L .
(ii) Pell-orbit closure of the divisibility criterion. The orbit-wise criterion of step (i) reduces the admissible block sizes to the pairs ( g , h ) (g,h) ( g , h ) with g = g = g = gcd ( m ℓ − 1 , c s d ) \gcd(m^\ell-1,cs^d) g cd( m ℓ − 1 , c s d ) and 1 ≤ h ≤ g − 1 1 \le h \le g-1 1 ≤ h ≤ g − 1 , where s = s i s=s_i s = s i is a non-trivial suborbit length of ( T , m ) (T,m) ( T , m ) ; for such a pair the divisibility A c , s , d ∣ k ( k − 1 ) A_{c,s,d}\mid k(k-1) A c , s , d ∣ k ( k − 1 ) with A c , s , d = ( m ℓ − 1 ) / gcd ( m ℓ − 1 , c s d ) A_{c,s,d}=(m^\ell-1)/\gcd(m^\ell-1,cs^d) A c , s , d = ( m ℓ − 1 ) / g cd( m ℓ − 1 , c s d ) is equivalent, by the discriminant identity
g Δ − ( g − h ) ( 2 m ℓ − 1 ) 2 = h , Δ = 1 + 4 ( g − h ) m ℓ ( m ℓ − 1 ) g , g\Delta-(g-h)(2m^\ell-1)^2=h,\qquad\Delta=1+\frac{4(g-h)m^\ell(m^\ell-1)}{g}, g Δ − ( g − h ) ( 2 m ℓ − 1 ) 2 = h , Δ = 1 + g 4 ( g − h ) m ℓ ( m ℓ − 1 ) ,
to the solvability of the Pell-type equation
X 2 − D U 2 = N , D = g ( g − h ) , N = g h , X = g V , U = 2 m ℓ − 1 , X^2-DU^2=N,\qquad D=g(g-h),\qquad N=gh,\qquad X=gV,\qquad U=2m^\ell-1, X 2 − D U 2 = N , D = g ( g − h ) , N = g h , X = g V , U = 2 m ℓ − 1 ,
i.e. the discriminant Δ \Delta Δ is a square of an integer precisely when such a solution exists (Lemma 7.12), with X = g V X=gV X = g V and U = 2 m ℓ − 1 U=2m^\ell-1 U = 2 m ℓ − 1 . For T = P S L 2 ( q ) T=PSL_2(q) T = P S L 2 ( q ) with q q q arbitrary, the census gap_p1a_suborbit_multiset.log records 41 suborbit keys ( q , m , s ) (q,m,s) ( q , m , s ) and 42,878 candidate pairs over 2 ≤ ℓ ≤ 300 2\le\ell\le300 2 ≤ ℓ ≤ 300 , of which exactly 331 pairs (with g ≥ 2 g\ge2 g ≥ 2 ) form the open set of the criterion. The closure of this open set is stated in Lemma 7.12 and recorded in Table 14 : the 15 pairs with D D D a square are closed by the bounded degenerate hyperbola ( X − a U ) ( X + a U ) = N (X-aU)(X+aU)=N ( X − a U ) ( X + a U ) = N (U ≤ N / ( 2 a ) ≤ 820 U\le N/(2a)\le820 U ≤ N / ( 2 a ) ≤ 820 ), the 76 pairs for which Δ \Delta Δ is a non-residue modulo some prime are closed by the quadratic obstruction of step (i), and the remaining 240 pairs by the Pell-orbit argument; in particular 331 = 15 + 76 + 240 and no pair survives. The closures are independent of ℓ \ell ℓ and of the structure of the block stabilizer L L L , since only the suborbit sizes of the base action enter.
Table 14. Suborbit keys ( q , m , s ) (q,m,s) ( q , m , s ) of T = P S L 2 ( q ) T = PSL_2(q) T = P S L 2 ( q ) from the suborbit census, and the maximal absolute orbit representative ∣ U ∣ \lvert U\rvert ∣ U ∣ among the solutions of X 2 − D U 2 = N X^2-DU^2=N X 2 − D U 2 = N in the window U ≤ 1000 U \le1000 U ≤ 1000 over the open pairs ( g , h ) (g,h) ( g , h ) of Lemma 7.12; the representative distribution over the 316 non-square open pairs is 258 one-digit, 46 two-digit and 12 three-digit values, the global maximum 681 occurring at the key ( 32 , 528 , 31 ) (32,528,31) ( 32 , 528 , 31 ) . An entry 0 records that every solution in the window has U = 0 U=0 U = 0 , that is, lies on the trivial branch X 2 = N X^2=N X 2 = N of the equation; it is not a placeholder for an absent set of open pairs. The only orbit outside this window is listed in Lemma 7.12 and closed there by the periodic reduction modulo m ℓ m^{\ell} m ℓ (Appendix § 6.2 ).
Lemma 7.12 (Pell-orbit closure of the divisibility criterion for P S L 2 ( q ) PSL_2(q) P S L 2 ( q ) ). Let T = P S L 2 ( q ) T=PSL_2(q) T = P S L 2 ( q ) act primitively on Δ \Delta Δ of degree m m m with a non-trivial suborbit of length s s s , let ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 , v = m ℓ v=m^\ell v = m ℓ , and put A c , s , d = ( m ℓ − 1 ) / gcd ( m ℓ − 1 , c s d ) A_{c,s,d}=(m^\ell-1)/\gcd(m^\ell-1,cs^d) A c , s , d = ( m ℓ − 1 ) / g cd( m ℓ − 1 , c s d ) , 1 ≤ c ≤ ( ℓ d ) 1\le c\le\binom{\ell}{d} 1 ≤ c ≤ ( d ℓ ) . For every pair ( g , h ) (g,h) ( g , h ) with g = gcd ( m ℓ − 1 , c s d ) g=\gcd(m^\ell-1,cs^d) g = g cd( m ℓ − 1 , c s d ) and 1 ≤ h ≤ g − 1 1\le h\le g-1 1 ≤ h ≤ g − 1 , the divisibility A c , s , d ∣ k ( k − 1 ) A_{c,s,d}\mid k(k-1) A c , s , d ∣ k ( k − 1 ) holds for some admissible block size k k k if and only if the Pell-type equation
X 2 − D U 2 = N , D = g ( g − h ) , N = g h , X = g V , U = 2 m ℓ − 1 , X^2-DU^2=N,\qquad D=g(g-h),\qquad N=gh,\qquad X=gV,\qquad U=2m^\ell-1, X 2 − D U 2 = N , D = g ( g − h ) , N = g h , X = g V , U = 2 m ℓ − 1 , *has a solution in non-negative integers. If D D D is a non-square, every solution lies on one of finitely many orbits ( X 0 + U 0 D ) ( x 0 + y 0 D ) k (X_0+U_0\sqrt{D})(x_0+y_0\sqrt{D})^k ( X 0 + U 0 D ) ( x 0 + y 0 D ) k , k ∈ Z k\in\mathbb{Z} k ∈ Z , generated by the minimal unit x 0 + y 0 D x_0+y_0\sqrt{D} x 0 + y 0 D of Q ( D ) \mathbb{Q}(\sqrt{D}) Q ( D ) (Perron–Nagell). For the 41 keys of the census, the exhaustive enumeration of the window U ≤ 1000 U\le1000 U ≤ 1000 locates at least one orbit per open pair, with maximal absolute representative ∣ U ∣ ≤ 681 |U|\le681 ∣ U ∣ ≤ 681 (Table 14 ); an exact orbit audit (script pa_107_basic_audit.py)
shows that the only open pair carrying a further orbit outside this window is ( 32 , 528 , 31 ) (32, 528, 31) ( 32 , 528 , 31 ) with g = 31 g = 31 g = 31 , h = 12 h = 12 h = 12 , D = 589 D = 589 D = 589 , N = 372 N = 372 N = 372 , whose additional orbit has minimal representative ∣ U ∣ = 604,499 |U| = 604{,}499 ∣ U ∣ = 604 , 499 . For each orbit found in the window, the reduction modulo m e m^e m e , e = 1 , 2 e = 1, 2 e = 1 , 2 , is a periodic sequence in the finite ring Z / m e Z \mathbb{Z}/m^e\mathbb{Z} Z / m e Z , and no period contains the candidate class U ≡ − 1 ( m o d m e ) U \equiv-1 \pmod{m^e} U ≡ − 1 ( mod m e ) ; hence no window orbit contains a solution with U = 2 m ℓ − 1 U = 2m^\ell- 1 U = 2 m ℓ − 1 , for any ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 . The single orbit outside the window, of minimal representative ∣ U ∣ = 604,499 |U| = 604{,}499 ∣ U ∣ = 604 , 499 for the key ( 32 , 528 , 31 ) (32, 528, 31) ( 32 , 528 , 31 ) (g = 31 g = 31 g = 31 , h = 12 h = 12 h = 12 , D = 589 D = 589 D = 589 , N = 372 N = 372 N = 372 ), is closed in the same way: its periodic reductions modulo 528 e 528^e 52 8 e , e = 1 , 2 e = 1, 2 e = 1 , 2 , contain no pair ( X , U ) (X,U) ( X , U ) with U ≡ − 1 ( m o d 528 e ) U \equiv-1 \pmod{528^e} U ≡ − 1 ( mod 52 8 e ) and g ∣ X g \mid X g ∣ X (Appendix § 6.2 ). Since every admissible candidate has U = 2 m ℓ − 1 ≡ − 1 ( m o d m e ) U = 2m^\ell- 1 \equiv-1 \pmod{m^e} U = 2 m ℓ − 1 ≡ − 1 ( mod m e ) and X = g V X = gV X = g V divisible by g g g , no orbit, inside or outside the window, contains a candidate value U = 2 m ℓ − 1 U = 2m^\ell- 1 U = 2 m ℓ − 1 for any ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 ; the per-ℓ \ell ℓ direct check X 2 = N + D ( 2 m ℓ − 1 ) 2 X^2 = N + D(2m^\ell- 1)^2 X 2 = N + D ( 2 m ℓ − 1 ) 2 over 2 ≤ ℓ ≤ 300 2 \le\ell\le300 2 ≤ ℓ ≤ 300 (Appendix § 6.2 ) independently re-confirms the window range. If D D D is a square, the equation factors as ( X − a U ) ( X + a U ) = N (X-aU)(X+aU) = N ( X − a U ) ( X + a U ) = N with a = D a = \sqrt{D} a = D , forcing U ≤ N / ( 2 a ) ≤ 820 U \le N/(2a) \le820 U ≤ N / ( 2 a ) ≤ 820 ; the enumeration U ≤ 1000 U \le1000 U ≤ 1000 yields no solution of the form U = 2 m ℓ − 1 U = 2m^\ell- 1 U = 2 m ℓ − 1 . The equivalence is purely arithmetic: it decides whether the divisibility A c , s , d ∣ k ( k − 1 ) A_{c,s,d} \mid k(k-1) A c , s , d ∣ k ( k − 1 ) admits a solution with 3 ≤ k ≤ v − 2 3 \le k \le v-2 3 ≤ k ≤ v − 2 , and it does not assert the existence of a block of that size. The existence of a block is decided separately, by the orbit-union condition of step (ii) of Proposition 7.9 , which is where the admissible sizes are matched against the K K K -box unions.
Proof. The equivalence is the completion of the square recorded above: g Δ − ( g − h ) ( 2 m ℓ − 1 ) 2 = h g\Delta- (g-h)(2m^\ell- 1)^2 = h g Δ − ( g − h ) ( 2 m ℓ − 1 ) 2 = h follows from Δ = 1 + 4 ( g − h ) m ℓ ( m ℓ − 1 ) / g \Delta= 1 + 4(g-h)m^\ell(m^\ell- 1)/g Δ = 1 + 4 ( g − h ) m ℓ ( m ℓ − 1 ) / g , and multiplying g V 2 − ( g − h ) U 2 = h gV^2 - (g-h)U^2 = h g V 2 − ( g − h ) U 2 = h by g g g gives X 2 − D U 2 = N X^2 - DU^2 = N X 2 − D U 2 = N with X = g V X = gV X = g V , D = g ( g − h ) D = g(g-h) D = g ( g − h ) , N = g h N = gh N = g h . Conversely a solution ( X , U ) (X,U) ( X , U ) of the Pell equation with U = 2 m ℓ − 1 U = 2m^\ell- 1 U = 2 m ℓ − 1 and g ∣ X g \mid X g ∣ X gives V = X / g V = X/g V = X / g and restores Δ = g V 2 \Delta= gV^2 Δ = g V 2 a square; the block size k k k is then determined by the orbit-union equation of step (ii). The orbit structure is Perron–Nagell (each class of solutions modulo the unit group lies on a unique two-sided orbit); the bounded representatives and the periodic reductions are verified by exhaustive enumeration (the finite checks of Appendix § 6.2 ; Table 14 ), and the single orbit lying outside the window U ≤ 1000 U \le1000 U ≤ 1000 is closed without enumerating it: for the key ( 32 , 528 , 31 ) (32, 528, 31) ( 32 , 528 , 31 ) with g = 31 g = 31 g = 31 , h = 12 h = 12 h = 12 (D = 589 D = 589 D = 589 , N = 372 N = 372 N = 372 ) the exact orbit audit identifies its minimal representative ∣ U ∣ = 604,499 |U| = 604{,}499 ∣ U ∣ = 604 , 499 , and the periodic reductions of this orbit modulo 528 e 528^e 52 8 e , e = 1 , 2 e = 1, 2 e = 1 , 2 , contain no pair ( X , U ) (X,U) ( X , U ) with U ≡ − 1 ( m o d 528 e ) U \equiv-1 \pmod{528^e} U ≡ − 1 ( mod 52 8 e ) and g ∣ X g \mid X g ∣ X (Appendix § 6.2 ); since every candidate has U ≡ − 1 ( m o d m e ) U \equiv-1 \pmod{m^e} U ≡ − 1 ( mod m e ) and g ∣ X g \mid X g ∣ X , this closes the outside orbit for all ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 without any upper bound on ℓ \ell ℓ . The per-ℓ \ell ℓ direct check over 2 ≤ ℓ ≤ 300 2 \le\ell\le300 2 ≤ ℓ ≤ 300 (Appendix § 6.2 ) independently re-confirms the window range.
(ii) Integrality, Fisher, and the block-count divisibility. For the remaining parameter triples the admissible block sizes k k k (3 ≤ k ≤ v − 2 3 \le k \le v-2 3 ≤ k ≤ v − 2 ) are constrained by the integrality conditions
r = b k v ∈ Z , λ = r ( k − 1 ) v − 1 = b k ( k − 1 ) v ( v − 1 ) ∈ Z , r = \frac{bk}{v} \in\mathbb{Z}, \qquad\lambda= \frac{r(k-1)}{v-1} = \frac{bk(k-1)}{v(v-1)} \in\mathbb{Z}, r = v bk ∈ Z , λ = v − 1 r ( k − 1 ) = v ( v − 1 ) bk ( k − 1 ) ∈ Z ,
by the Fisher inequality b ≥ v b \ge v b ≥ v , and by the block-primitivity fact that N = T ℓ N = T^{\ell} N = T ℓ is transitive on blocks (the argument of Proposition 2.24 : N G β = G N G_{\beta} = G N G β = G by maximality), whence b = [ N : N β ] ∣ ∣ T ∣ ℓ b = [N : N_{\beta}] \mid|T|^{\ell} b = [ N : N β ] ∣ ∣ T ∣ ℓ . The block-count structure enters this layer through Lemma 7.10: the block action of G G G on B \mathcal{B} B is primitive, and its block count b b b is one of the six alternatives (1)–(6) recorded there. Alternatives (2)–(5) are excluded as stated in the lemma (diagonal by Lemma 2.14 and Proposition 2.15 ; twisted wreath by the regular-normal-subgroup lemma; holomorph by the commuting-transitive-regular argument giving v = b v=b v = b ; affine by the non-abelian structure of T ℓ T^{\ell} T ℓ ), so the only remaining counts are the product-action value b = n ℓ b=n^{\ell} b = n ℓ and, a fortiori bounded by it, the almost simple value b ∣ ∣ T ∣ b\mid|T| b ∣ ∣ T ∣ of (6). This conclusion is structural: it does not rest on the divisibility condition b ∣ ∣ T ∣ ℓ b\mid|T|^{\ell} b ∣ ∣ T ∣ ℓ alone (which admits many non-n ℓ n^{\ell} n ℓ divisors), and it is the intermediate statement requested in the review: the hypotheses used are precisely the primitivity of the block action, the transitivity of N = T ℓ N=T^{\ell} N = T ℓ on blocks, and the O’Nan–Scott classification applied to the block action. The block size enters the sieve through the orbit-union constraint, which replaces the flag-transitive formula k = A ℓ j k=A^{\ell}j k = A ℓ j of (5) in the non-symmetric case: the block B B B is G B G_B G B -invariant, so fixing α ∈ B \alpha\in B α ∈ B the block is a union of G B , α = G B ∩ G α G_{B,\alpha}=G_B\cap G_{\alpha} G B , α = G B ∩ G α -orbits; since K ℓ ≤ G B , α K^{\ell}\le G_{B,\alpha} K ℓ ≤ G B , α , every G B , α G_{B,\alpha} G B , α -orbit is contained in a K K K -box or a C B C_B C B -amalgam of K K K -boxes (Lemma 2.18 ), so k k k is a sum of K K K -box sizes (equivalently of lengths of G B , α G_{B,\alpha} G B , α -orbits). The sieve then runs over the finitely many such sums, imposing the integrality of r = b k / v = n ℓ k / m ℓ r=bk/v=n^{\ell}k/m^{\ell} r = bk / v = n ℓ k / m ℓ and of λ = b k ( k − 1 ) / ( v ( v − 1 ) ) \lambda=bk(k-1)/(v(v-1)) λ = bk ( k − 1 ) / ( v ( v − 1 )) above together with the orbit-wise Cameron–Praeger condition of step (i). Two remarks close the layer. First, the point action is primitive: non-symmetric block-primitive designs are point-primitive by Proposition 3.4 , so H H H is maximal in T T T (the product action on points is primitive if and only if H H H is maximal) and the coordinate stabilizers are conjugate. Second, for the ten 2-homogeneous pairs the stabilizer H H H is transitive on Δ \Delta Δ , hence G α = H ℓ × C α G_{\alpha}=H^{\ell}\times C_{\alpha} G α = H ℓ × C α is transitive on P \mathcal{P} P , so B = P B=\mathcal{P} B = P and k = v k=v k = v , contradicting 2 ≤ k ≤ v − 2 2\le k\le v-2 2 ≤ k ≤ v − 2 ; for the seven non-2-homogeneous pairs every union of K K K -boxes containing the α \alpha α -box (a superset of every admissible block, since K ℓ ≤ G B , α K^{\ell}\le G_{B,\alpha} K ℓ ≤ G B , α ) is enumerated directly (GAP 4.16.1, script gap_pa_layer2_dp.g): for all seventeen primitive degrees, all block-coordinate indices n n n with n ℓ ≥ m ℓ n^{\ell} \ge m^{\ell} n ℓ ≥ m ℓ , and all admissible ℓ \ell ℓ , the integrality sieve over the K K K -box unions (r = n ℓ k / m ℓ ∈ Z r = n^{\ell}k/m^{\ell} \in\mathbb{Z} r = n ℓ k / m ℓ ∈ Z , λ ∈ Z \lambda\in\mathbb{Z} λ ∈ Z , the orbit-wise Cameron–Praeger condition of step (i), and Fisher b ≥ v b \ge v b ≥ v ) leaves only five block-size candidates, all with ℓ = 2 \ell= 2 ℓ = 2 : ( A 5 , 5 , 10 ) (A_{5},5,10) ( A 5 , 5 , 10 ) with k = 4 , 9 , 12 k = 4,9,12 k = 4 , 9 , 12 , ( A 6 , 15 , 15 ) (A_{6},15,15) ( A 6 , 15 , 15 ) with k = 64 k = 64 k = 64 , and ( P S L 2 ( 13 ) , 14 , 91 ) (PSL_{2}(13),14,91) ( P S L 2 ( 13 ) , 14 , 91 ) with k = 36 k = 36 k = 36 . For each of these five candidates the exhaustive enumeration of the corresponding K K K -box unions containing the α \alpha α -box (GAP 4.16.1, script gap_pa_layer2_check.g; the enumeration covers every relative conjugate position of K K K in each coordinate; for the three socles the conjugates of K K K yield a single distinct orbit decomposition on Δ \Delta Δ , verified over 10, 15 and 91 conjugates respectively, so one representative suffices) produces only blocks whose pair count is not constant over the G G G -pair orbits. The k = 4 k = 4 k = 4 branch for ( A 5 , 5 , 10 ) (A_{5},5,10) ( A 5 , 5 , 10 ) admits a short argument in the spirit of the orbit-wise Cameron–Praeger condition, without any search: the point action of degree 5 has point stabilizer H ≅ A 4 H \cong A_{4} H ≅ A 4 , and the block-coordinate stabilizer K ≅ S 3 K \cong S_{3} K ≅ S 3 of index 10 has K K K -orbits of sizes 2, 3 on Δ \Delta Δ (GAP-verified); GAP records the orbit of α \alpha α in each coordinate; for ( A 5 , 5 , 10 ) (A_{5},5,10) ( A 5 , 5 , 10 ) both coordinates of α \alpha α lie in the 2-orbit, so the α \alpha α -box (the K 2 K^{2} K 2 -orbit of α \alpha α ) is the unique 2 × 2 2 \times2 2 × 2 grid of size 4; its in-box pair allocation (4, 2) over the two relevant pair orbits fails the balanced allocation (2, 4) required by the design equations (the orbit-wise Cameron–Praeger criterion), so it cannot be a block. The remaining exclusions are recorded relative to the fixed point α \alpha α : every admissible block is K ℓ K^{\ell} K ℓ -invariant, hence contains the whole α \alpha α -box, so k k k is the α \alpha α -box size plus a sum of the other K K K -box sizes. For ( A 5 , 5 , 10 ) (A_{5},5,10) ( A 5 , 5 , 10 ) the α \alpha α -box has size 4 and the other box sizes are 4, 6, 6, 9; unions of sizes 9 and 12 do exist, and the enumeration below finds 36 respectively 48 of them, all failing the λ \lambda λ -constancy test. For ( A 6 , 15 , 15 ) (A_{6},15,15) ( A 6 , 15 , 15 ) GAP records α \alpha α in the 1-orbit in both coordinates (α \alpha α -box of size 1, other box sizes 1, 8, 6, 64, 48, 36); the target 64 − 1 = 63 64 - 1 = 63 64 − 1 = 63 is attained, for instance 48 + 8 + 6 + 1 = 63 48 + 8 + 6 + 1 = 63 48 + 8 + 6 + 1 = 63 , so unions of size 64 do exist, and the enumeration below finds 256 of them, all failing (GAP-verified, script gap_pa_layer2_check.g). For ( P S L 2 ( 13 ) , 14 , 91 ) (PSL_{2}(13),14,91) ( P S L 2 ( 13 ) , 14 , 91 ) GAP records α \alpha α in the 6-orbit in both coordinates, so the α \alpha α -box has size 6 ⋅ 6 = 36 = k 6 \cdot6 = 36 = k 6 ⋅ 6 = 36 = k , and its pair count is not constant. The enumeration of the box unions is exhaustive in the conjugates: for every conjugate K t 1 K^{t_{1}} K t 1 of K K K in the first coordinate and every conjugate K t 2 K^{t_{2}} K t 2 of K K K in the second coordinate, in full, over the 10 × 10 10 \times10 10 × 10 , 15 × 15 15 \times15 15 × 15 and 91 × 91 91 \times91 91 × 91 ordered conjugate pairs respectively, the script gap_pa_final_enum3.g builds the K t 1 × K t 2 K^{t_{1}} \times K^{t_{2}} K t 1 × K t 2 -boxes on Δ 2 \Delta^{2} Δ 2 , takes the α \alpha α -box, enumerates every union of the remaining boxes containing it, and for every candidate of size k k k verifies the full G G G -pair-orbit coverage: the number λ ( 1 , y ) \lambda(1,y) λ ( 1 , y ) of blocks through the pair { 1 , y } \{1,y\} { 1 , y } is computed by the coset formula λ ( 1 , y ) = ∑ b ∈ B ∣ s b − 1 B ∩ G α y ∣ \lambda(1,y)=\sum_{b\in B}|s_b^{-1}B\cap G_{\alpha y}| λ ( 1 , y ) = ∑ b ∈ B ∣ s b − 1 B ∩ G α y ∣ (s b ∈ G s_b\in G s b ∈ G with s b ⋅ 1 = b s_b\cdot1=b s b ⋅ 1 = b ) and is compared with the design value on every representative y y y of the G α G_\alpha G α -orbits on P ∖ { α } P\setminus\{\alpha\} P ∖ { α } . The outcome is a complete zero survivor count in every case: ( A 5 , 5 , 10 ) (A_5,5,10) ( A 5 , 5 , 10 ) yields 16, 36 and 48 candidates of sizes 4, 9, 12 respectively, all failing the λ \lambda λ -constancy test; ( A 6 , 15 , 15 ) (A_6,15,15) ( A 6 , 15 , 15 ) yields 256 candidates of size 64, all failing; ( P S L 2 ( 13 ) , 14 , 91 ) (PSL_2(13),14,91) ( P S L 2 ( 13 ) , 14 , 91 ) yields 12,168 candidates of size 36, all failing. The observation that the conjugates of K K K produce a single distinct orbit decomposition on Δ \Delta Δ (verified over 10, 15 and 91 conjugates) is recorded only as a consistency check; it is not used to reduce the enumeration, which is an exhaustive disjoint enumeration over the finite box collections. Hence no admissible block survives. Together with the block-count structure above, this excludes the product action type completely for non-symmetric designs; this conclusion is independent of the block-action type (1) or (6), since the orbit-union sieve constrains k k k directly. These killed alternatives are recorded once in Lemma 7.10; the layers (iii)–(v) below eliminate the surviving symmetric (Menon) candidates, where the flag-transitive formula (5) applies. Hence no non-symmetric product-action configuration survives.
(iii) Exact Bruck–Ryser–Chowla test. For the surviving symmetric candidates b = v b=v b = v the exact Hasse–Minkowski local-global test is applied to the ternary quadratic form ⟨ n , σ λ , − 1 ⟩ \langle n,\sigma\lambda,-1\rangle ⟨ n , σ λ , − 1 ⟩ , σ = ( − 1 ) ( v − 1 ) / 2 \sigma=(-1)^{(v-1)/2} σ = ( − 1 ) ( v − 1 ) /2 : a good prime is automatically soluble iff ( n / p ) = 1 (n/p)=1 ( n / p ) = 1 or ( − σ λ / p ) = 1 (-\sigma\lambda/p)=1 ( − σ λ / p ) = 1 , so whenever n n n and − σ λ -\sigma\lambda − σ λ are both non-squares, Chebotarev provides a good prime with both non-residues and the candidate is excluded; the bad-prime set { 2 } ∪ Primes ( n λ ) \{2\}\cup\operatorname{Primes}(n\lambda) { 2 } ∪ Primes ( nλ ) is checked locally when n n n or − σ λ -\sigma\lambda − σ λ is a square. The test excludes every candidate except a list of symmetric (b = v b=v b = v ) survivors with n = k − λ n=k-\lambda n = k − λ a square: the Menon parameter pairs ( k = 2 u 2 ± u ) (k=2u^2\pm u) ( k = 2 u 2 ± u ) and one non-Menon pair.
(iv) Block-action type and the prime-divisor sieve. The block action of G G G is primitive with socle N = T ℓ N=T^\ell N = T ℓ . An O’Nan–Scott analysis of this action excludes the affine type (v = m ℓ v=m^\ell v = m ℓ is not a prime power for the listed degrees), the twisted wreath type (∣ T ∣ r = v |T|^r=v ∣ T ∣ r = v forces m m m to be a prime power, contradicted by the listed primitive degrees), the simple/compound diagonal types (∣ T ∣ ℓ − s = v |T|^{\ell-s}=v ∣ T ∣ ℓ − s = v forces m s = ∣ T ∣ m^s=|T| m s = ∣ T ∣ , contradicted by the primitive degree tables), and the almost simple type (the simple factor of the block socle would have order ≤ v \leq v ≤ v , impossible since ∣ T ∣ ℓ > v |T|^\ell>v ∣ T ∣ ℓ > v for ℓ ≥ 2 \ell\geq2 ℓ ≥ 2 and the listed degrees). Hence the block action is itself of product action (box) type. As B ⊇ α B\supseteq\alpha B ⊇ α is a union of K K K -boxes (since K ℓ ≤ G B , α K^\ell\leq G_{B,\alpha} K ℓ ≤ G B , α ), each K K K -box has size dividing n ℓ n^\ell n ℓ with n ∣ ∣ T ∣ n\mid|T| n ∣ ∣ T ∣ , but a sum of such sizes need not have all its prime divisors in Primes ( ∣ T ∣ ) \operatorname{Primes}(|T|) Primes ( ∣ T ∣ ) ; the primes of k k k are therefore computed directly for each surviving candidate and compared with Primes(∣ T ∣ |T| ∣ T ∣ ). This direct comparison excludes the non-Menon pair, whose block size carries primes outside Primes(∣ T ∣ |T| ∣ T ∣ ), and leaves exactly the two Menon candidates
( v , k , λ ) = ( 100 , 45 , 20 ) and ( 784 , 378 , 182 ) , (v,k,\lambda)=(100,45,20)\quad\text{and}\quad(784,378,182), ( v , k , λ ) = ( 100 , 45 , 20 ) and ( 784 , 378 , 182 ) ,
with T = A 5 T=A_5 T = A 5 , m = 10 m=10 m = 10 , ℓ = 2 \ell=2 ℓ = 2 and T = P S L 2 ( 8 ) T=PSL_2(8) T = P S L 2 ( 8 ) , m = 28 m=28 m = 28 , ℓ = 2 \ell=2 ℓ = 2 , respectively.
(v) Box search for the two Menon candidates. Both parameter pairs are known to be realized by symmetric designs (2-(100 , 45 , 20 100,45,20 100 , 45 , 20 ) is the Menon design of the regular Hadamard matrix of order 100, studied e.g. in [8 ] ; 2-(784 , 378 , 182 784,378,182 784 , 378 , 182 ) was constructed in [19 ] with automorphism group Frob 29 ⋅ 14 × Z 13 \operatorname{Frob}_{29\cdot14}\times\mathbb{Z}_{13} Frob 29 ⋅ 14 × Z 13 , in particular not of product action type). The question is whether a block-primitive automorphism of box type exists: G = A 5 2 : C 2 G=A_5^2\mathbin{:}C_2 G = A 5 2 : C 2 acting on 100 points, or G = P S L 2 ( 8 ) 2 : C 2 G=PSL_2(8)^2\mathbin{:}C_2 G = P S L 2 ( 8 ) 2 : C 2 acting on 784 points (product action with coordinate swap). With the degree-10 (respectively degree-28, suborbits [ 1 , 9 , 9 , 9 ] [1,9,9,9] [ 1 , 9 , 9 , 9 ] ) primitive actions embedded in S 10 S_{10} S 10 (respectively S 28 S_{28} S 28 ), the box groups have orders 7200 and 508032, are primitive on points, and satisfy ∣ G α ∣ = 72 |G_\alpha|=72 ∣ G α ∣ = 72 and 648 respectively (Schreier generators). The G α G_\alpha G α -orbit lengths are { 12 , 6 , 36 , 36 , 9 } \{12,6,36,36,9\} { 12 , 6 , 36 , 36 , 9 } on 100 points and { 18 3 , 81 3 , 162 3 } \{18^3,81^3,162^3\} { 1 8 3 , 8 1 3 , 16 2 3 } on 784 points. An exhaustive GAP 4.16.1 computation (scripts gap_pa_layer5_menon.g and gap_pa_layer5_orbit_union.g) runs two independent searches. (1) General block stabilizers, in the K K K -box language: every block B ∋ α B\ni\alpha B ∋ α is a union of K K K -boxes (since K ℓ ≤ G B , α K^\ell\leq G_{B,\alpha} K ℓ ≤ G B , α ), so all unions of K K K -boxes containing the α \alpha α -box are enumerated, over every block-coordinate index n n n and, exhaustively in the conjugates, every ordered pair ( K t 1 , K t 2 ) (K^{t_1},K^{t_2}) ( K t 1 , K t 2 ) of conjugates in the two coordinates (script gap_pa_final_enum3.g; 10 × 10 10\times10 10 × 10 ordered pairs for ( A 5 , 10 ) (A_5,10) ( A 5 , 10 ) , and 28 × 28 28\times28 28 × 28 and 36 × 36 36\times36 36 × 36 for ( P S L 2 ( 8 ) , 28 ) (PSL_2(8),28) ( P S L 2 ( 8 ) , 28 ) with n = 28 , 36 n=28,36 n = 28 , 36 respectively); every candidate of the required size is then checked for constant λ ( 1 , y ) \lambda(1,y) λ ( 1 , y ) over the G α G_\alpha G α -pair-orbit representatives by the coset formula. The outcome is a complete zero survivor count: for ( A 5 , 10 ) (A_5,10) ( A 5 , 10 ) the 540 candidates of size 45 all fail the λ \lambda λ -constancy test, for ( P S L 2 ( 8 ) , 28 ) (PSL_2(8),28) ( P S L 2 ( 8 ) , 28 ) with n = 28 n=28 n = 28 the 47,628 candidates of size 378 all fail, and with n = 36 n=36 n = 36 there are no candidates of size 378 at all: the α \alpha α -box is the 7 × 7 7\times7 7 × 7 grid of size 49, and 378 − 49 = 329 378-49=329 378 − 49 = 329 is not a sum of the other box sizes (all multiples of 7), so a size-378 box union cannot even be formed. The single orbit decomposition of K K K on Δ \Delta Δ (one over the 10, 28 and 36 conjugates respectively) is recorded as a consistency check only. (2) The symmetric screening branch: a G α G_\alpha G α -invariant block would force G α ≤ G B G_\alpha\leq G_B G α ≤ G B and hence b ≤ v b\leq v b ≤ v , so by Fisher’s inequality b = v b=v b = v (the Menon case); the unions of G α G_{\alpha} G α -orbits of size k k k are enumerated separately: a union containing α \alpha α has size 1 + ∑ i ℓ i 1+\sum_i \ell_i 1 + ∑ i ℓ i over the non-trivial orbits, and for ( A 5 , 10 ) (A_5,10) ( A 5 , 10 ) the orbit lengths 12 , 6 , 36 , 36 , 9 12,6,36,36,9 12 , 6 , 36 , 36 , 9 are all multiples of 3 3 3 whereas 45 − 1 = 44 ≡ 2 ( m o d 3 ) 45-1=44\equiv2\pmod{3} 45 − 1 = 44 ≡ 2 ( mod 3 ) , while for ( P S L 2 ( 8 ) , 28 ) (PSL_2(8),28) ( P S L 2 ( 8 ) , 28 ) the orbit lengths 18 , 81 , 162 18,81,162 18 , 81 , 162 are all multiples of 9 9 9 whereas 378 − 1 = 377 ≡ 8 ( m o d 9 ) 378-1=377\equiv8\pmod{9} 378 − 1 = 377 ≡ 8 ( mod 9 ) ; in both cases no union of G α G_{\alpha} G α -orbits containing α \alpha α has the required size. Hence no block-primitive 2-design of product-action box type exists for either candidate. Since G G G is transitive on points, every orbit of G G G on unordered pairs has a representative ( 1 , y ) (1,y) ( 1 , y ) with y y y in a G α G_{\alpha} G α -orbit; checking the pair count λ ( 1 , y ) \lambda(1,y) λ ( 1 , y ) on these representatives confirms the exclusion:
(a) 2 2 2 -( 100 , 45 , 20 ) (100,45,20) ( 100 , 45 , 20 ) : the 540 candidates of size 45 that are unions of K K K -boxes containing the α \alpha α -box all fail the λ \lambda λ -constancy test, and by 44 ≡ 2 ( m o d 3 ) 44\equiv2\pmod{3} 44 ≡ 2 ( mod 3 ) no union of G α G_{\alpha} G α -orbits containing α \alpha α has size 45;
(b) 2 2 2 -( 784 , 378 , 182 ) (784,378,182) ( 784 , 378 , 182 ) : the 47,628 candidates of size 378 that are unions of K K K -boxes containing the α \alpha α -box all fail the λ \lambda λ -constancy test, and by 377 ≡ 8 ( m o d 9 ) 377\equiv8\pmod{9} 377 ≡ 8 ( mod 9 ) no union of G α G_{\alpha} G α -orbits containing α \alpha α has size 378.
Hence neither Menon candidate admits a block-primitive automorphism group of box type; the product action type is completely excluded. For ( 100 , 45 , 20 ) (100,45,20) ( 100 , 45 , 20 ) this conclusion is independently confirmed by the complete classification of primitive symmetric designs with at most 255 points of Braić, Golemac, Mandić and Vučićíć [7 ] : the four non-affine primitive groups at degree 100 for these parameters are ruled out there by the non-existence of a difference set in their regular subgroups, so every point-primitive automorphism group of a 2 2 2 -( 100 , 45 , 20 ) (100,45,20) ( 100 , 45 , 20 ) design is affine. □
No product-action configuration survives. The exclusion of Proposition 7.11 for the six socles and the listed primitive degree pairs is explicit and machine-checked: the box framework, the fiber identity, the orbit-wise Cameron–Praeger criterion, the block-count divisibility, the Bruck–Ryser–Chowla test and the prime-divisor sieve are structural and involve no information specific to ℓ \ell ℓ ; the non-2 2 2 -homogeneous pairs are precisely the seven among the seventeen primitive degrees of Table 3 , the 2 2 2 -homogeneous pairs are excluded by the transitivity argument of step (ii), and the reduction over the degree tables of Dixon–Mortimer [24 ] (derived from the minimal-degree bounds of Liebeck–Saxl [32 ] ) identifies the six socles as the complete set of possibilities for non-symmetric block-primitive point actions. Together this closes the product action type for every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 , with arbitrary block stabilizers.
Proposition 7.14 (The CD type is excluded). We exclude CD blocks in the minimal n = 6 n = 6 n = 6 model in full. (i) Single-orbit blocks: for every k = r s k = rs k = r s with r > r ′ r > r' r > r ′ this case is excluded in full by divisibility. Indeed k g = [ M 0 : H 0 ∩ M g 0 ] k_g = [M_0 : H_0 \cap M_g^0] k g = [ M 0 : H 0 ∩ M g 0 ] divides ∣ M 0 ∣ = q r ′ |M_0| = q^{r'} ∣ M 0 ∣ = q r ′ , and the design equations give λ = q r − r ′ k g ( k g − 1 ) / ( q r ( s − 1 ) − 1 ) \lambda= q^{r-r'}k_g(k_g-1)/(q^{r(s-1)}-1) λ = q r − r ′ k g ( k g − 1 ) / ( q r ( s − 1 ) − 1 ) , so q r ( s − 1 ) − 1 ∣ q r − r ′ k g ( k g − 1 ) q^{r(s-1)} - 1 \mid q^{r-r'}k_g(k_g - 1) q r ( s − 1 ) − 1 ∣ q r − r ′ k g ( k g − 1 ) . As gcd ( q r ′ , q r ( s − 1 ) − 1 ) = 1 \gcd(q^{r'}, q^{r(s-1)} - 1) = 1 g cd( q r ′ , q r ( s − 1 ) − 1 ) = 1 and k g ∣ q r ′ k_g \mid q^{r'} k g ∣ q r ′ , we have gcd ( k g , q r ( s − 1 ) − 1 ) = 1 \gcd(k_g,q^{r(s-1)} - 1) = 1 g cd( k g , q r ( s − 1 ) − 1 ) = 1 ; hence q r ( s − 1 ) − 1 ∣ k g − 1 q^{r(s-1)} - 1 \mid k_g - 1 q r ( s − 1 ) − 1 ∣ k g − 1 , impossible for k g > 1 k_g > 1 k g > 1 because k g − 1 < q r ( s − 1 ) − 1 k_g - 1 < q^{r(s-1)} - 1 k g − 1 < q r ( s − 1 ) − 1 . Thus k g = 1 k_g = 1 k g = 1 , the trivial block, excluded. (ii) Multiple-orbit blocks: the multi-orbit CD case is completely eliminated. After imposing the D B D_B D B -orbit condition, the K B K_B K B -fusion condition, the outer-automorphism fusion condition, the subset-sum condition and the design-parameter integrality conditions, the only remaining candidate is ( T , a , c ) = ( A 5 , 15 , 10 ) (T,a,c) = (A_5,15,10) ( T , a , c ) = ( A 5 , 15 , 10 ) , with v = 15 ⋅ 60 3 = 3 , 240 , 000 v = 15 \cdot60^3 = 3,240,000 v = 15 ⋅ 6 0 3 = 3 , 240 , 000 , b = 10 ⋅ 60 4 = 129 , 600 , 000 b = 10 \cdot60^4 = 129,600,000 b = 10 ⋅ 6 0 4 = 129 , 600 , 000 and k = 1 , 619 , 100 k = 1,619,100 k = 1 , 619 , 100 . For this candidate, an explicit construction of the block B B B gives
Stab G 0 ( B ) = D B × ( S 3 ≀ S 2 ) , G 0 = A 5 6 ⋊ S 6 , ∣ G B ∣ = 259 , 200 = ∣ G 0 ∣ / b . \operatorname{Stab}_{G_0}(B) = D_B \times(S_3 \wr S_2), \qquad G_0 = A_5^6 \rtimes S_6, \qquad|G_B| = 259,200 = |G_0|/b. Stab G 0 ( B ) = D B × ( S 3 ≀ S 2 ) , G 0 = A 5 6 ⋊ S 6 , ∣ G B ∣ = 259 , 200 = ∣ G 0 ∣/ b . The equality is verified by computing the coordinate projection (Step A: the projection of G B G_B G B onto S 6 S_6 S 6 is exactly S 3 ≀ S 2 S_3 \wr S_2 S 3 ≀ S 2 , the subgroup preserving the two layer sets of sizes $9$ and 6 6 6 ) and the socle stabilizer (Step B: $|\operatorname{Stab}_{A_5^6}(\mathcal{O}_{28}) =
3600 = ∣ D B ∣ 3600 = |D_B| 3600 = ∣ D B ∣ for the selected D B D_B D B -orbit O 28 \mathcal{O}_{28} O 28 of length 1200 on the 6-layer part). However,
D B ⋊ ( S 3 ≀ S 2 ) < ( A 5 3 × A 5 3 ) ⋊ ( S 3 ≀ S 2 ) < A 6 5 ⋊ S 6 , D_B \rtimes(S_3 \wr S_2) < (A_5^3 \times A_5^3) \rtimes(S_3 \wr S_2) < A_6^5 \rtimes S_6, D B ⋊ ( S 3 ≀ S 2 ) < ( A 5 3 × A 5 3 ) ⋊ ( S 3 ≀ S 2 ) < A 6 5 ⋊ S 6 ,
so G B G_B G B is not maximal in G 0 G_0 G 0 and the induced block action is imprimitive; hence this candidate does not yield a CD-type block-primitive 2-design. In the diagonal outer-automorphism extension G 1 = A 5 6 ⋊ ( C 2 × S 6 ) G_1 = A_5^6 \rtimes(C_2 \times S_6) G 1 = A 5 6 ⋊ ( C 2 × S 6 ) , the global outer automorphism φ \varphi φ (conjugation by a transposition of the five points) preserves the selected orbit O 28 \mathcal{O}_{28} O 28 and hence the whole block, so Stab G 1 ( B ) ≠ G B \operatorname{Stab}_{G_1}(B) \ne G_B Stab G 1 ( B ) = G B ; the prescribed block stabilizer is again not obtained. Consequently the multi-orbit CD case is empty in the two global models G 0 , G 1 G_0,G_1 G 0 , G 1 together with the six small simple groups listed above; the screening conditions of (ii) are purely structural and are imposed in the same way in every model, and the maximality obstruction applies unchanged. For λ = 1 \lambda= 1 λ = 1 , CD block type is excluded by the linear space classification. Together with (i), the CD block type is excluded in full.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Data availability
The complete records of the computational sieves used in Propositions 7.6 , 7.7 and 7.14 are available from the author.
Acknowledgements: The authors acknowledge the use of GPT-6 Astra, a large language model, for assistance with language editing and proofreading during the preparation of this manuscript. All mathematical content and conclusions are entirely the author’ own, and the author take full responsibility for the paper.
Proposition 7.15 (Reduction of the 2 + 2 2+2 2 + 2 residual). Let T T T be a non-abelian simple group, q = ∣ T ∣ q=|T| q = ∣ T ∣ , N = T 4 N=T^4 N = T 4 , let K = N ∩ G α = { ( a , φ ( a ) , c , ψ ( c ) ) : a , c ∈ T } K=N\cap G_{\alpha}=\{(a,\varphi(a),c,\psi(c)):a,c\in T\} K = N ∩ G α = {( a , φ ( a ) , c , ψ ( c )) : a , c ∈ T } be a point stabilizer of the configuration of the one-sided SD type, so that v = q 2 v=q^2 v = q 2 and b = q 3 b=q^3 b = q 3 , and suppose that the block stabilizer Δ = N ∩ G B \Delta=N\cap G_B Δ = N ∩ G B of a block B B B is isomorphic to T T T . Then with λ \lambda λ the parameter of D \mathcal{D} D there exist q q q subsets D 1 , … , D q D_1,\ldots,D_q D 1 , … , D q of a subgroup S J ≅ T 2 S_J\cong T^2 S J ≅ T 2 of N N N with ∣ D i ∣ = k |D_i|=k ∣ D i ∣ = k such that
∑ i = 1 q N i ( g ) = λ ( g ≠ 1 ) , ∑ i = 1 q N i ( 1 ) = q k , \sum_{i=1}^{q}N_i(g)=\lambda\quad(g\ne1),\qquad\sum_{i=1}^{q}N_i(1)=qk, i = 1 ∑ q N i ( g ) = λ ( g = 1 ) , i = 1 ∑ q N i ( 1 ) = q k , where N i ( g ) = # { ( d 1 , d 2 ) ∈ D i 2 : d 1 − 1 d 2 = g } N_i(g)=\#\{(d_1,d_2)\in D_i^2:d_1^{-1}d_2=g\} N i ( g ) = # {( d 1 , d 2 ) ∈ D i 2 : d 1 − 1 d 2 = g } , that is, D \mathcal{D} D yields a ( q 2 ; k , … , k ; λ ) (q^2;k,\ldots,k;\lambda) ( q 2 ; k , … , k ; λ ) difference system in T 2 T^2 T 2 , equivalently a T 2 T^2 T 2 -developed 2 2 2 -( q 2 , k , λ ) (q^2,k,\lambda) ( q 2 , k , λ ) design with q 3 q^3 q 3 blocks. Conversely, q q q subsets D i D_i D i of a point-regular subgroup S ≅ T 2 S\cong T^2 S ≅ T 2 with these difference conditions develop into a 2 2 2 -( q 2 , k , λ ) (q^2,k,\lambda) ( q 2 , k , λ ) design with q 3 q^3 q 3 blocks on which S S S acts point-regularly; such a design arises inside the present configuration if and only if in addition its block family is invariant under the action induced by N N N on S S S .
Proof. The arguments are those of the preceding discussion. First, S J ∩ K = 1 S_J\cap K=1 S J ∩ K = 1 by construction, so the subgroup S J S_J S J of N N N acts point-regularly on P = N / K P=N/K P = N / K and ∣ S J ∣ = v |S_J|=v ∣ S J ∣ = v . Second, Δ \Delta Δ is a twisted diagonal Δ = { ( β 1 ( a ) , … , β 4 ( a ) ) : a ∈ T } \Delta=\{(\beta_1(a),\ldots,\beta_4(a)):a\in T\} Δ = {( β 1 ( a ) , … , β 4 ( a )) : a ∈ T } whose coordinate maps β i : T → T \beta_i:T\to T β i : T → T are homomorphisms, hence automorphisms or the constant map 1 1 1 ; since ∣ Δ ∣ = q |\Delta|=q ∣Δ∣ = q at least one β i 0 \beta_{i_0} β i 0 is injective. Third, choose J J J to contain one coordinate of each of the two parts and the index i 0 i_0 i 0 , and let S J = { x : x k = 1 , k ∈ J } S_J=\{x:x_k=1,\ k\in J\} S J = { x : x k = 1 , k ∈ J } . If x ∈ S J ∩ g Δ g − 1 x\in S_J\cap g\Delta g^{-1} x ∈ S J ∩ g Δ g − 1 then x i 0 = g i 0 β i 0 ( a ) g i 0 − 1 x_{i_0}=g_{i_0}\beta_{i_0}(a)g_{i_0}^{-1} x i 0 = g i 0 β i 0 ( a ) g i 0 − 1 , and x i 0 = 1 x_{i_0}=1 x i 0 = 1 because i 0 ∈ J i_0\in J i 0 ∈ J ; injectivity of β i 0 \beta_{i_0} β i 0 then gives a = 1 a=1 a = 1 and x = 1 x=1 x = 1 . Hence S J ∩ G B = 1 S_J\cap G_B=1 S J ∩ G B = 1 for every block B B B , every orbit of S J S_J S J on the block set has length v v v , and b = q 3 = q v b=q^3=qv b = q 3 = q v gives exactly q q q orbits, with representatives B 1 , … , B q B_1,\ldots,B_q B 1 , … , B q . Put D i = { x ∈ S J : x α ∈ B i } D_i=\{x\in S_J:x\alpha\in B_i\} D i = { x ∈ S J : xα ∈ B i } , where α \alpha α is the base point, so that ∣ D i ∣ = k |D_i|=k ∣ D i ∣ = k and the map x ↦ x α x\mapsto x\alpha x ↦ xα is a bijection of S J S_J S J onto P P P . A block x D i xD_i x D i contains the points x d 1 α xd_1\alpha x d 1 α and x d 2 α xd_2\alpha x d 2 α with d 1 ≠ d 2 d_1\ne d_2 d 1 = d 2 precisely when d 1 − 1 d 2 = g d_1^{-1}d_2=g d 1 − 1 d 2 = g for the element g g g determined by the two points, and each pair ( d 1 , d 2 ) (d_1,d_2) ( d 1 , d 2 ) determines x x x ; therefore the number of blocks through two distinct points at distance g g g is ∑ i N i ( g ) \sum_iN_i(g) ∑ i N i ( g ) , and the design condition is exactly ∑ i N i ( g ) = λ \sum_iN_i(g)=\lambda ∑ i N i ( g ) = λ for g ≠ 1 g\ne1 g = 1 , while ∑ i N i ( 1 ) = q k \sum_iN_i(1)=qk ∑ i N i ( 1 ) = q k counts the ordered pairs inside a block. The converse is the standard development of a difference system, and the last assertion records that N N N acts on P P P with K K K as point stabilizer and on blocks with stabilizer Δ \Delta Δ , which is an additional condition on the developed design and not a consequence of the difference conditions alone.
Concluding remarks
In this paper, we have completed the structural reduction of non-symmetric block-primitive 2-designs: every O’Nan–Scott type other than almost simple is either excluded or reduced to an explicit algebraic feasibility problem, and we have classified the almost simple case up to the two families of block stabilizers named in Theorem 4.6 . The symmetric case (b = v b=v b = v ) is not part of the classification of this paper: it is isolated in the main theorem as a separate class, and by duality block-primitive symmetric designs coincide with point-primitive symmetric designs, a classical subject covered by an extensive literature; hence it plays no role in the reduction and no reader should take the designs listed here to include symmetric ones. Our main contributions are:
Main structural lemma. We prove that block-primitivity implies point-quasiprimitivity for non-trivial non-symmetric 2-designs, using the incidence matrix rank argument. This is a fundamental result that bridges the gap between block-side and point-side primitivity.
Systematic exclusion of the non-AS O’Nan–Scott types. We strictly exclude the HA, HC, HS (under non-conjugate stabilizers), two-sided
S D 2 SD_2 S D 2 , and TW types, and further exclude the one-sided SD type for every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 and every non-abelian simple T T T (the double count of Lemma 7.1 , Goursat’s theorem, and the subgroup chain of Lemma 7.5 ), as well as the compound diagonal CD type for n = 6 n = 6 n = 6 and the six small simple groups, the arithmetic sieve of Proposition 7.6 and the λ \lambda λ -consistency exclusion of Proposition 7.7 being retained as the parameter record for the one-sided type. Each exclusion uses a different technique: symmetry forcing via the regular normal subgroup lemma (HA, TW, S D 2 SD_2 S D 2 ), centralizer obstruction (HC), permutation-module obstruction (HS), Fisher-type arithmetic with subgroup-existence and construction checks (SD, CD).
Reduction of the remaining types. We reduce the PA and AS types to well-defined algebraic problems: algebraic feasibility with a character-inclusion necessary condition for PA, and quadruple ( T , G , H , M ) (T, G, H, M) ( T , G , H , M ) analysis for AS, respectively. For the PA type we obtain the complete exclusion for the six socles with the seventeen listed primitive degrees and every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 with arbitrary block stabilizers (Proposition 7.9 and Proposition 7.11); this is complete for non-symmetric designs by the reduction of Proposition 3.4 together with the degree tables of Dixon–Mortimer [24 ] . (The one-sided SD and CD types are excluded within the framework described above; see § 2.1 .)
Complete point-imprimitive exclusion. No non-trivial non-symmetric block-primitive 2-design has a point-imprimitive automorphism group (Proposition 3.4 ): an intransitive normal subgroup would force every block to be a union of its orbits, contradicting the 2-design condition, while a quasiprimitive imprimitive point action would be of almost simple, twisted wreath, simple diagonal, compound diagonal or product action type (the affine, holomorph simple and holomorph compound types are always primitive), all excluded (TW by Lemma 2.10 ; SD and CD in § 2.1 ; PA by Proposition 7.11 for the point-primitive product action and by Lemma 3.11 for the point-imprimitive product action; AS by the point-imprimitive exclusions of § 3.1 ); the geometric Grassmann-type candidates also fail the constant-coverage test. Consequently every non-symmetric block-primitive 2-design is point-primitive, and since block-primitivity coincides with point-primitivity for symmetric designs by duality, every block-primitive combinatorial
design is point-primitive, settling the long-standing problem recorded in the Introduction (Theorem 1.1 ).
Point-primitive AS classification. We have also classified all point-primitive, non-2 2 2 -homogeneous non-symmetric block-primitive 2 2 2 -designs with almost simple socle, across all families of finite simple groups, the even-characteristic P S L 2 ( q ) PSL_2(q) P S L 2 ( q ) subfield block stabilizers being reduced in Theorem 4.6 to the class equations of Proposition 4.34 :
Classical Lie type: all designs are geometric, from Grassmann or polar space configurations.
Sporadic groups: the non-2 2 2 -homogeneous actions of J 2 J_2 J 2 and H S HS H S on 100 points yield the two designs 2 2 2 -( ( ( 100,56,560) ) ) and 2 2 2 -( ( ( 100,64,1680) ) ) ; the 2 2 2 -homogeneous actions of C o 3 Co_3 C o 3 and H S HS H S are covered by the orbit construction of Lemma 4.2 .
Exceptional Lie type: only parabolic chain designs, plus special Suzuki and Ree cases.
Our results provide a classification of all almost simple (AS) type non-symmetric block-primitive 2 2 2 -designs, which are necessarily point-primitive by Proposition 3.4 , complete, the even-characteristic P S L 2 ( q ) PSL_2(q) P S L 2 ( q ) subfield block stabilizers being reduced in Theorem 4.6 to the class equations of Proposition 4.34 . Every O’Nan–Scott type other than almost simple is excluded (HA, HC, HS, two-sided S D 2 SD_2 S D 2 , TW; the one-sided SD type for every ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 and every non-abelian simple T T T ; the compound diagonal CD type for n = 6 n=6 n = 6 and the six small simple groups; and the product action type, see Proposition 7.9 and Proposition 7.11).
Finally, we record why the grid-imprimitive constructions of the Delandtsheer–Doyen line, although point-imprimitive and block-transitive, do not contradict the classification above, and in fact illustrate it. In the setting of [1 ] the point set is a product P = ∏ i ∈ I Z e i P=\prod_{i\in I}\mathbb{Z}_{e_i} P = ∏ i ∈ I Z e i , the group is G = ∏ i ∈ I Sym ( Z e i ) G=\prod_{i\in I}\operatorname{Sym}(\mathbb{Z}_{e_i}) G = ∏ i ∈ I Sym ( Z e i ) , and the block set is the orbit of a single block B B B . For any proper nonempty J ⊂ I J\subset I J ⊂ I put x B J ( δ ) = ∣ B ∩ π J − 1 ( δ ) ∣ x_B^J(\delta)=|B\cap\pi_J^{-1}(\delta)| x B J ( δ ) = ∣ B ∩ π J − 1 ( δ ) ∣ and let supp J ( B ) \operatorname{supp}_J(B) supp J ( B ) be its support. Since x g ( B ) J ( g ‾ δ ) = x B J ( δ ) x_{g(B)}^J(\overline{g}\delta)=x_B^J(\delta) x g ( B ) J ( g δ ) = x B J ( δ ) for the induced permutation g ‾ \overline{g} g of E J = ∏ i ∈ J Z e i E_J=\prod_{i\in J}\mathbb{Z}_{e_i} E J = ∏ i ∈ J Z e i , the fibres of supp J \operatorname{supp}_J supp J form a G G G -invariant partition of the block set. For the infinite family of [1 ] one has k = p 2 s − 1 + 1 k=p^{2^{s-1}}+1 k = p 2 s − 1 + 1 while ∏ i < s e i = p 2 s − 1 + p 2 s − 2 + 1 \prod_{i<s}e_i=p^{2^{s-1}}+p^{2^{s-2}}+1 ∏ i < s e i = p 2 s − 1 + p 2 s − 2 + 1 , so for J = { 1 , … , s − 1 } J=\{1,\ldots,s-1\} J = { 1 , … , s − 1 } the support satisfies ∣ supp J ( B ) ∣ ≤ k < ∣ E J ∣ |\operatorname{supp}_J(B)|\le k<|E_J| ∣ supp J ( B ) ∣ ≤ k < ∣ E J ∣ and is never all of E J E_J E J ; and since e s ∤ k e_s\nmid k e s ∤ k (the gcd of the two divides p 2 s − 2 p^{2^{s-2}} p 2 s − 2 while e s ≡ 1 ( m o d p ) e_s \equiv1 \pmod{p} e s ≡ 1 ( mod p ) , so it equals 1), some permutation of the last coordinate preserves that support while moving B B B : if π s ( B ) ≠ E s \pi_s(B) \ne E_s π s ( B ) = E s one takes a cell outside π s ( B ) \pi_s(B) π s ( B ) , and otherwise two s s s -slices differ, since coinciding slices would give e s ∣ k e_s \mid k e s ∣ k . Hence the induced action on blocks is imprimitive for every member of the family, so the whole grid line consists of designs that are point-imprimitive and block-imprimitive simultaneously, exactly as Proposition 3.4 requires.
Finite checks
Several steps of the argument use a finite verification. This appendix states what each verification covers.
(1.1) (Design conditions.) Fix a primitive action G / H G/H G / H and a maximal subgroup M M M with ∣ H ∣ > ∣ M ∣ |H| > |M| ∣ H ∣ > ∣ M ∣ . A candidate block is a union B B B of M M M -orbits on G / H G/H G / H . For each such union the check computes k = ∣ B ∣ k = |B| k = ∣ B ∣ , b = [ G : M ] b = [G : M] b = [ G : M ] , the orbits of G G G on unordered pairs of points of G / H G/H G / H , and the covering numbers λ ( x , y ) \lambda(x,y) λ ( x , y ) on each of these orbits. A design occurs exactly when b k = v r bk = vr bk = v r for an integer r r r and all covering numbers coincide. The enumeration is exact; no sampling is involved.
(2.2) (P S L 2 ( q ) PSL_2(q) P S L 2 ( q ) .) For odd q ≤ 127 q \le127 q ≤ 127 the maximal subgroups of the five listed types and all subfield subgroups are tested against the point stabilizers of Cases 1–4 of § 3.1 ; for even q q q the two dihedral subgroups D 2 ( q − 1 ) D_{2(q-1)} D 2 ( q − 1 ) and D 2 ( q + 1 ) D_{2(q+1)} D 2 ( q + 1 ) are tested against the block stabilizers of § 3.2.3 , every admissible union of orbits of one of them being checked for the constancy of the pair counts at q = 8 q = 8 q = 8 and q = 32 q = 32 q = 32 , while at q = 16 q = 16 q = 16 and q = 64 q = 64 q = 64 the arithmetic sieve of Lemma 4.8 together with Theorem 4.26 and Corollary 4.27 settles the dihedral case without any enumeration, and the same holds at every even degree with ( q − 2 ) / 2 (q - 2)/2 ( q − 2 ) /2 prime. The outcome is the two block-primitive designs at q = 8 q = 8 q = 8 , the two designs at q = 32 q = 32 q = 32 and no design at q = 16 q = 16 q = 16 ; at q = 8 q = 8 q = 8 the subfield unions which are 2-designs are imprimitive in their block action, and a subfield block stabilizer of type S 3 S_3 S 3 is not admissible at any even degree, since S 3 S_3 S 3 is not a maximal subgroup for odd f f f ; at q = 64 q = 64 q = 64 the two dihedral block stabilizers give nothing, while of the subfield block stabilizers of that degree P S L 2 ( 8 ) PSL_2(8) P S L 2 ( 8 ) is excluded by Fisher’s inequality (its index 520 is smaller than v = 2016 v = 2016 v = 2016 ), and A 5 A_5 A 5 is excluded by Proposition 4.42 ; every even degree beyond 64 lies outside the range of the present enumeration.
(3.3) (Classical groups.) The parabolic and non-parabolic subgroup chains of § 3.1 are tested for the constancy of λ \lambda λ ; the outcome is the one stated in 4.11.
(4.4) (Sporadic groups.) The maximal subgroups recorded in the ATLAS are tested as in item 1; the outcome is the table of § 3.1 .
(5.5) (Module identity.) The identity M 1 G − H 1 G = χ W {}^{1}_{M}G-{}^{1}_{H}G=\chi_W M 1 G − H 1 G = χ W used in Case 2 for q = 3 q=3 q = 3 (mod 4) is verified on the natural module W W W .
The scripts are available from the author on request.