Introduction
The study of orbit lengths on subsets goes back to Livingstone and Wagner [18 ] , who proved that a transitive group has at least as many orbits on k k k -sets as on ( k − 1 ) (k-1) ( k − 1 ) -sets. Siemons and Wagner [23 ] turned from the number of orbits to their lengths and introduced the property ⋆ k \star_k ⋆ k : a transitive group G ≤ Sym ( P ) G \le\operatorname{Sym}(\mathcal{P}) G ≤ Sym ( P ) has ⋆ k \star_k ⋆ k if there is a k k k -subset D \mathcal{D} D with ∣ D G ∣ > ∣ Σ G ∣ |\mathcal{D}^G| > |\Sigma^G| ∣ D G ∣ > ∣ Σ G ∣ for every ( k + 1 ) (k+1) ( k + 1 ) -subset Σ ⊃ D \Sigma\supset\mathcal{D} Σ ⊃ D . They proved that a k k k -subset D \mathcal{D} D with ∣ D G ∣ > ∣ Σ G ∣ |\mathcal{D}^G| > |\Sigma^G| ∣ D G ∣ > ∣ Σ G ∣ for every ( k + 1 ) (k+1) ( k + 1 ) -subset Σ ⊃ D \Sigma\supset\mathcal{D} Σ ⊃ D satisfies k + 1 ≥ ∣ D G Σ ∣ > ∣ Σ G D ∣ ≥ 1 k+1 \ge|\mathcal{D}^{G\Sigma}| > |\Sigma^{G\mathcal{D}}| \ge1 k + 1 ≥ ∣ D G Σ ∣ > ∣ Σ G D ∣ ≥ 1 , and that for k ≥ 2 k \ge2 k ≥ 2 either every 2 2 2 -subset occurs in some G G G -image of D \mathcal{D} D or G G G is imprimitive with blocks meeting D \mathcal{D} D in at most one point; for primitive G G G the case k = 2 k=2 k = 2 was classified completely, the only instance being PSL ( 2 , 5 ) \operatorname{PSL}(2,5) PSL ( 2 , 5 ) of degree 6. Related identities for the orbit lengths were added by Mnukhin [20 ] .
The case k = 3 k=3 k = 3 was analysed by Bradley [1 , 2 ] . In his doctoral thesis he proved that a 3 3 3 -homogeneous group satisfying ⋆ 3 \star_3 ⋆ 3 is PSL ( 2 , 7 ) \operatorname{PSL}(2,7) PSL ( 2 , 7 ) or PGL ( 2 , 7 ) \operatorname{PGL}(2,7) PGL ( 2 , 7 ) [1 Theorem 6.2.5] , excluded the small 2 2 2 -transitive non-3 3 3 -homogeneous socles PSL ( 2 , 11 ) \operatorname{PSL}(2,11) PSL ( 2 , 11 ) , PSL ( 2 , 8 ) \operatorname{PSL}(2,8) PSL ( 2 , 8 ) , Alt ( 7 ) \operatorname{Alt}(7) Alt ( 7 ) , HS \operatorname{HS} HS and Co 3 \operatorname{Co}_3 Co 3 [1 Proposition 6.2.6] , and conjectured that a primitive group of degree at least 8 satisfies ⋆ 3 \star_3 ⋆ 3 if and only if it is PSL ( 2 , 7 ) \operatorname{PSL}(2,7) PSL ( 2 , 7 ) or PGL ( 2 , 7 ) \operatorname{PGL}(2,7) PGL ( 2 , 7 ) [1 Conjecture 6.2.7] . Bradley’s reduction reaches the 3 3 3 -homogeneous groups and the small almost simple socles, but leaves open the 2 2 2 -transitive non-3 3 3 -homogeneous almost simple socles of his Table 6.2 — PSL ( 2 , q ) \operatorname{PSL}(2,q) PSL ( 2 , q ) with q ≡ 1 ( m o d 4 ) q \equiv1 \pmod{4} q ≡ 1 ( mod 4 ) , PSU ( 3 , q ) \operatorname{PSU}(3,q) PSU ( 3 , q ) , PSL ( d , q ) \operatorname{PSL}(d,q) PSL ( d , q ) with d > 2 d>2 d > 2 , Sz ( q ) \operatorname{Sz}(q) Sz ( q ) , Ree ( q ) \operatorname{Ree}(q) Ree ( q ) and PSp ( 2 d , 2 ) / PΓSp ( 2 d , 2 ) \operatorname{PSp}(2d,2)/\operatorname{P\Gamma Sp}(2d,2) PSp ( 2 d , 2 ) / PΓSp ( 2 d , 2 ) — and does not reach the 2 2 2 -homogeneous affine groups at all.
In this paper we close the two gaps left open by Bradley and thereby settle the conjecture in full, as an application of a structural theory of three-point orbits of 2-transitive groups developed in the process — the orbit-length hierarchies, the pointwise and setwise stabilizer chains, and the geometric realisations (the unital, the ovoid and the classical geometries). The proof is in two steps. First Lemma 2.5 shows that a primitive group satisfying ⋆ 3 \star_3 ⋆ 3 is 2-homogeneous: the reduction is purely orbit-theoretic, uses the rank-3 classification [14 , 16 , 17 ] , and is completed for the small degrees by the exhaustive census of Section 3. Second, every 2-homogeneous non-3-homogeneous group fails ⋆ 3 \star_3 ⋆ 3 : by Cameron’s classification [3 ] and Kantor’s classification of 2-homogeneous groups [13 ] the candidates are the 2-transitive families and the affine one-dimensional 2-homogeneous groups, and for each candidate the witness lemma 2.16 together with the four-point witnesses of Lemmas 2.18 and 2.29 produces a four-subset that violates the defining inequality. Only the 3-homogeneous case of Bradley remains, and it yields exactly PSL ( 2 , 7 ) \operatorname{PSL}(2,7) PSL ( 2 , 7 ) and PGL ( 2 , 7 ) \operatorname{PGL}(2,7) PGL ( 2 , 7 ) .
The central technical tool is a single numerical criterion. For a candidate triple Δ \Delta Δ and an external point w w w write K w = G Δ ∪ { w } K_w=G_{\Delta\cup\{w\}} K w = G Δ ∪ { w } , J w = G Δ , w J_w=G_{\Delta,w} J w = G Δ , w and r w = [ G Δ : J w ] r_w=[G_\Delta:J_w] r w = [ G Δ : J w ] , t w = [ K w : J w ] t_w=[K_w:J_w] t w = [ K w : J w ] ; the defining inequality fails for the quadruple Δ ∪ { w } \Delta\cup\{w\} Δ ∪ { w } exactly when r w ≥ t w r_w\ge t_w r w ≥ t w , i.e. when ∣ K w ∣ ≤ ∣ G Δ ∣ |K_w|\le|G_\Delta| ∣ K w ∣ ≤ ∣ G Δ ∣ . This criterion excludes the affine groups uniformly: for collinear triples in every characteristic and for affinely independent triples with p ≥ 5 p\ge5 p ≥ 5 the linear dependence of the four points has a unique coefficient, forcing t w = 1 t_w=1 t w = 1 ; for p = 3 p=3 p = 3 a two-dimensional localization reduces the problem to the fifty-five subgroups of GL ( 2 , 3 ) \operatorname{GL}(2,3) GL ( 2 , 3 ) , and for p = 2 p=2 p = 2 , d = 3 d=3 d = 3 an exhaustive check of the seventeen irreducible subgroups of GL ( 3 , 2 ) \operatorname{GL}(3,2) GL ( 3 , 2 ) verifies the criterion for every triple.
The precise statement is Corollary 2.30 below. The orbit-theoretic machinery used in the four-point arguments — regularity of the maximal triple orbit, pointwise stabilizers, the geometry of the unital and the ovoid — is organized here as a structural theory of three-point orbits; the constituent facts are classical or are read from the literature cited there, and a systematic development of the triple-orbit classification of the 2-transitive groups and of its applications to designs, graphs, codes and association schemes appears in the companion monograph Three-point orbits of 2-transitive groups and their combinatorial applications by the same author [5 ] . The present paper is self-contained.
Classifications of 2-transitive and 2-homogeneous groups
The input of the witness arguments is the classification of the finite 2-transitive groups, assembled from Hering [10 ] , Huppert [11 ] , Curtis–Kantor–Seitz [8 ] and the survey [4 ] :
Affine type. soc ( G ) = V \operatorname{soc}(G)=V soc ( G ) = V is elementary abelian of order p d p^d p d , and G = V ⋊ G 0 G=V\rtimes G_0 G = V ⋊ G 0 with G 0 ≤ GL ( d , p ) G_0\leq\operatorname{GL}(d,p) G 0 ≤ GL ( d , p ) transitive on the nonzero vectors (Hering’s theorem [10 ] ).
Almost simple type. soc ( G ) = T \operatorname{soc}(G)=T soc ( G ) = T is a nonabelian simple group with T ≤ G ≤ Aut ( T ) T\leq G\leq\operatorname{Aut}(T) T ≤ G ≤ Aut ( T ) acting 2-transitively: T = PSL ( n , q ) T=\operatorname{PSL}(n,q) T = PSL ( n , q ) (n = 2 n=2 n = 2 on the projective line, n ≥ 3 n\geq3 n ≥ 3 on projective space), Sp ( 2 n , 2 ) \operatorname{Sp}(2n,2) Sp ( 2 n , 2 ) (n ≥ 3 n\geq3 n ≥ 3 , on the two halves of the nonsingular quadric), PSU ( 3 , q ) \operatorname{PSU}(3,q) PSU ( 3 , q ) , Sz ( q ) \operatorname{Sz}(q) Sz ( q ) , Ree ( q ) \operatorname{Ree}(q) Ree ( q ) (i.e. 2 G 2 ( q ) {}^{2}G_{2}(q) 2 G 2 ( q ) ), Alt ( n ) \operatorname{Alt}(n) Alt ( n ) , or a sporadic group (the Mathieu groups together with HS \operatorname{HS} HS and Co 3 \operatorname{Co}_3 Co 3 ), with G G G contained in the corresponding automorphism group.
The 2-homogeneous groups
The 2-homogeneous groups that are not 2-transitive were classified by Kantor [13 ] : the only examples are the affine one-dimensional groups G ≤ AΓL ( 1 , q ) G\leq\operatorname{A\Gamma L}(1,q) G ≤ AΓL ( 1 , q ) with q q q odd.
Lemma 2.1 (Classification of the 2-homogeneous groups). Let G G G be a finite 2-homogeneous permutation group that is not 2-transitive. Then G G G is affine one-dimensional, G ≤ AΓL ( 1 , q ) G\leq\operatorname{A\Gamma L}(1,q) G ≤ AΓL ( 1 , q ) on v = q = p d v=q=p^d v = q = p d points with q q q odd, G G G is solvable, and the point stabilizer G 0 G_0 G 0 has odd order. In particular ∣ G ∣ |G| ∣ G ∣ is odd, so G G G contains no involution, and no almost simple example occurs.
Proof. The 2-homogeneous groups were classified by Kantor [13 ] ; the only ones that are not 2-transitive are the affine one-dimensional groups G ≤ AΓL ( 1 , q ) G\leq\operatorname{A\Gamma L}(1,q) G ≤ AΓL ( 1 , q ) with q q q odd. If G 0 G_0 G 0 contained an element of order 2, then together with 2-homogeneity this element would invert the action on the remaining points and make G G G 2-transitive [18 ] : an involution fixing one point of a pair acts transitively on the other, so transitivity on unordered pairs upgrades to transitivity on ordered pairs. Hence ∣ G 0 ∣ |G_0| ∣ G 0 ∣ is odd, and ∣ G ∣ = v ∣ G 0 ∣ |G|=v|G_0| ∣ G ∣ = v ∣ G 0 ∣ is odd because v = q v=q v = q is odd. Since ∣ G ∣ |G| ∣ G ∣ is odd, G G G is solvable [9 ] , and no almost simple (non-solvable) group is 2-homogeneous without being 2-transitive.
The affine three-point stabilizer
The affine computations use the following elementary description of the three-point stabilizer in an affine group.
Lemma 2.2 (The three-point stabilizer in affine groups). Let G = V ⋊ G 0 G=V\rtimes G_0 G = V ⋊ G 0 with V = F p d V=\mathbb{F}_p^d V = F p d and G 0 ≤ GL ( d , p ) G_0\leq\operatorname{GL}(d,p) G 0 ≤ GL ( d , p ) transitive on V ∖ { 0 } V\setminus\{0\} V ∖ { 0 } , and let O = { 0 , u , w } \mathcal{O}=\{0,u,w\} O = { 0 , u , w } be a three-point subset of V V V . An element ( v , g 0 ) ∈ G (v,g_0)\in G ( v , g 0 ) ∈ G lies in G O G_{\mathcal{O}} G O iff g 0 g_0 g 0 and v v v satisfy one of the following:
(T0) fixed point 0 0 0 , swap u ↔ w u\leftrightarrow w u ↔ w : v = 0 v=0 v = 0 , g 0 ∈ G 0 g_0\in G_0 g 0 ∈ G 0 , g 0 ( u ) = w g_0(u)=w g 0 ( u ) = w , g 0 ( w ) = u g_0(w)=u g 0 ( w ) = u (a transposition fixing 0 0 0 ; if additionally g 0 ( u ) = u g_0(u)=u g 0 ( u ) = u and g 0 ( w ) = w g_0(w)=w g 0 ( w ) = w , then g 0 ∈ G ( O ) g_0\in G_{(\mathcal{O})} g 0 ∈ G ( O ) );
(T1) fixed point u u u : v = w v=w v = w , g 0 ( w ) = − w g_0(w)=-w g 0 ( w ) = − w , g 0 ( u ) = u − w g_0(u)=u-w g 0 ( u ) = u − w ;
(T2) fixed point w w w : v = u v=u v = u , g 0 ( u ) = − u g_0(u)=-u g 0 ( u ) = − u , g 0 ( w ) = w − u g_0(w)=w-u g 0 ( w ) = w − u ;
(C) 3-cycle ( 0 ↦ u ↦ w ↦ 0 ) (0\mapsto u\mapsto w\mapsto0) ( 0 ↦ u ↦ w ↦ 0 ) : v = u v=u v = u , g 0 ( u ) = w − u g_0(u)=w-u g 0 ( u ) = w − u , g 0 ( w ) = − u g_0(w)=-u g 0 ( w ) = − u ;
(I) identity : v = 0 v=0 v = 0 , g 0 = id g_0=\operatorname{id} g 0 = id (fixes all three points).
In particular:
(1) The pointwise stabilizer is G ( O ) = { g 0 ∈ G 0 : g 0 ( u ) = u , g 0 ( w ) = w } G_{(\mathcal{O})}=\{g_0\in G_0:g_0(u)=u,\ g_0(w)=w\} G ( O ) = { g 0 ∈ G 0 : g 0 ( u ) = u , g 0 ( w ) = w } .
(2) A transposition fixing 0 0 0 exists iff G 0 G_0 G 0 contains an element swapping u u u and w w w ; a transposition fixing u u u exists iff G 0 G_0 G 0 contains an element with g 0 ( u ) = u − w g_0(u)=u-w g 0 ( u ) = u − w , g 0 ( w ) = − w g_0(w)=-w g 0 ( w ) = − w ; a 3-cycle exists iff G 0 G_0 G 0 contains an element with g 0 ( u ) = w − u g_0(u)=w-u g 0 ( u ) = w − u , g 0 ( w ) = − u g_0(w)=-u g 0 ( w ) = − u .
(3) If g 0 g_0 g 0 satisfies (C), then the affine element ( u , g 0 ) (u,g_0) ( u , g 0 ) permutes { 0 , u , w } \{0,u,w\} { 0 , u , w } as the 3-cycle ( 0 u w ) (0\,u\,w) ( 0 u w ) ; in particular g 0 3 g_0^3 g 0 3 acts as the identity on the plane ⟨ u , w ⟩ \langle u,w\rangle ⟨ u , w ⟩ .
Proof. ( v , g 0 ) ∈ G O (v,g_0)\in G_{\mathcal{O}} ( v , g 0 ) ∈ G O iff { v , g 0 ( u ) + v , g 0 ( w ) + v } = { 0 , u , w } \{v,g_0(u)+v,g_0(w)+v\}=\{0,u,w\} { v , g 0 ( u ) + v , g 0 ( w ) + v } = { 0 , u , w } . The point v v v must be one of 0 , u , w 0,u,w 0 , u , w (as g 0 g_0 g 0 is invertible, v ≠ g 0 ( u ) + v v\neq g_0(u)+v v = g 0 ( u ) + v etc.). The four cases are obtained by matching the three images to the three points in all six ways; (C) is the case v = u v=u v = u with g 0 ( u ) + u = w g_0(u)+u=w g 0 ( u ) + u = w , g 0 ( w ) + u = 0 g_0(w)+u=0 g 0 ( w ) + u = 0 , i.e. g 0 ( u ) = w − u g_0(u)=w-u g 0 ( u ) = w − u , g 0 ( w ) = − u g_0(w)=-u g 0 ( w ) = − u . The affine element ( u , g 0 ) (u,g_0) ( u , g 0 ) sends 0 ↦ u 0\mapsto u 0 ↦ u , u ↦ g 0 ( u ) + u = w u\mapsto g_0(u)+u=w u ↦ g 0 ( u ) + u = w and w ↦ g 0 ( w ) + u = 0 w\mapsto g_0(w)+u=0 w ↦ g 0 ( w ) + u = 0 , so it permutes { 0 , u , w } \{0,u,w\} { 0 , u , w } as the 3-cycle ( 0 u w ) (0\,u\,w) ( 0 u w ) ; on the plane π = ⟨ u , w ⟩ \pi=\langle u,w\rangle π = ⟨ u , w ⟩ the matrix of g 0 g_0 g 0 in the basis ( u , w ) (u,w) ( u , w ) is ( − 1 − 1 1 0 ) \begin{pmatrix}-1&-1\\1&0\end{pmatrix} ( − 1 1 − 1 0 ) ,
whose cube is the identity. (The action of g 0 g_0 g 0 on a complement of π \pi π is not constrained by (C), so g 0 g_0 g 0 may have order a multiple of 3; only the induced image on O \mathcal{O} O matters.)
History of the conjecture. The property ⋆ k \star_k ⋆ k records how the lengths of the orbits behave as the subset size grows, and its origin lies in the orbit-counting monotonicity of Livingstone and Wagner [18 ] and the orbit-length theory of Siemons and Wagner [23 ] ; the precise statements, together with the classification of k = 2 k=2 k = 2 and the orbit-length identities of Mnukhin [20 ] , are recalled in the Introduction. The analysis of the case k = 3 k=3 k = 3 rests on the candidate framework supplied by the k k k -homogeneous classification of Kantor [12 ] and the 2-homogeneous classification of Cameron [3 ] , and was carried out out by Bradley [1 , 2 ] in his doctoral thesis and a concise companion note [2 ] : he settled the 3-homogeneous case and excluded the small 2-transitive non-3-homogeneous socles PSL ( 2 , 11 ) \operatorname{PSL}(2,11) PSL ( 2 , 11 ) , PSL ( 2 , 8 ) \operatorname{PSL}(2,8) PSL ( 2 , 8 ) , Alt ( 7 ) \operatorname{Alt}(7) Alt ( 7 ) , H S HS H S and C o 3 Co_3 C o 3 (Proposition 6.2.6), reducing the non-3-homogeneous question to the 2-transitive almost-simple socles of Table 6.2 of [1 ] . Bradley’s conjecture, however, is stated for all primitive groups of degree ≥ 8 \ge8 ≥ 8 (the framework of [1 ] leaves the 2-homogeneous affine groups out of its reduction), and this paper treats the affine side as well (Lemma 2.29 ), so that the conjecture is closed completely.
The Siemons–Wagner property for triples and Bradley’s conjecture
Recall the defining property (from the Introduction): a transitive group G ≤ Sym ( P ) G \le\operatorname{Sym}(\mathcal{P}) G ≤ Sym ( P ) has ⋆ k \star_k ⋆ k if there is a k k k -subset Δ \Delta Δ with ∣ Δ G ∣ > ∣ Σ G ∣ |\Delta^G| > |\Sigma^G| ∣ Δ G ∣ > ∣ Σ G ∣ for every ( k + 1 ) (k+1) ( k + 1 ) -subset Σ ⊃ Δ \Sigma\supset\Delta Σ ⊃ Δ . What Bradley [1 ] left open in the k = 3 k=3 k = 3 case is exactly the 2-transitive non-3-homogeneous almost simple socles of Table 6.2 of [1 ] : PSL ( 2 , q ) \operatorname{PSL}(2,q) PSL ( 2 , q ) (q ≡ 1 ( m o d 4 ) q \equiv1 \pmod{4} q ≡ 1 ( mod 4 ) ), PSU ( 3 , q ) \operatorname{PSU}(3,q) PSU ( 3 , q ) , PSL ( d , q ) \operatorname{PSL}(d,q) PSL ( d , q ) (d > 2 d > 2 d > 2 ), Sz ( q ) \operatorname{Sz}(q) Sz ( q ) , Ree ( q ) \operatorname{Ree}(q) Ree ( q ) and PSp ( 2 d , 2 ) / PΓSp ( 2 d , 2 ) \operatorname{PSp}(2d,2)/\operatorname{P\Gamma Sp}(2d,2) PSp ( 2 d , 2 ) / PΓSp ( 2 d , 2 ) , together with the 2-homogeneous affine side (the 2-transitive affine families and the 2-homogeneous non-2-transitive groups of Section 2.1), which Bradley’s reduction did not reach. These are precisely the families whose triple-orbit and stabilizer data are determined in this paper, and the triple classification closes the case uniformly.
Lemma 2.4 (Transitive triple stabilizers force 2-homogeneity). Let G ≤ Sym ( P ) G \le\operatorname{Sym}(\mathcal{P}) G ≤ Sym ( P ) be a primitive group of degree at least 8, and let D ∈ ( P 3 ) \mathcal{D} \in\binom{\mathcal{P}}{3} D ∈ ( 3 P ) be a ⋆ 3 \star_3 ⋆ 3 witness, i.e. ∣ D G ∣ > ∣ Σ G ∣ |\mathcal{D}^G| > |\Sigma^G| ∣ D G ∣ > ∣ Σ G ∣ for every Σ ∈ ( P 4 ) \Sigma\in\binom{\mathcal{P}}{4} Σ ∈ ( 4 P ) with D ⊆ Σ \mathcal{D} \subseteq\Sigma D ⊆ Σ . If the setwise stabilizer G D G_{\mathcal{D}} G D acts transitively on D \mathcal{D} D , then G G G is 2-homogeneous. Consequently, if G G G is not 2-homogeneous, the induced image of G D G_{\mathcal{D}} G D on D \mathcal{D} D is trivial or generated by a transposition.
Proof. The structural conclusion of the Siemons–Wagner theorem applied to the witness D \mathcal{D} D [1, Thm. 6.1.6, p. 83; Lem. 6.2.1, p. 86] gives two alternatives: either every 2-subset of P \mathcal{P} P is contained in a G G G -image of D \mathcal{D} D , or G G G is imprimitive with a nontrivial system of imprimitivity whose blocks meet D \mathcal{D} D in at most one point. Since G G G is primitive, the second alternative is impossible. Thus, for every A ∈ ( P 2 ) A \in\binom{\mathcal{P}}{2} A ∈ ( 2 P ) there exists g ∈ G g \in G g ∈ G with A ⊆ D g A \subseteq\mathcal{D}^g A ⊆ D g , the pair-covering property following from this structural alternative and not from the numerical stabilizer-orbit inequality. The complementation map D → ( D 2 ) \mathcal{D} \to\binom{\mathcal{D}}{2} D → ( 2 D ) , x ↦ D ∖ { x } x \mapsto\mathcal{D} \setminus\{x\} x ↦ D ∖ { x } , is a G D G_{\mathcal{D}} G D -equivariant bijection, so transitivity of G D G_{\mathcal{D}} G D on D \mathcal{D} D gives transitivity on ( D 2 ) \binom{\mathcal{D}}{2} ( 2 D ) . Let A , B ∈ ( P 2 ) A,B \in\binom{\mathcal{P}}{2} A , B ∈ ( 2 P ) be arbitrary; by pair-covering choose g , h ∈ G g,h\in G g , h ∈ G with A ⊆ D g A\subseteq D^g A ⊆ D g and B ⊆ D h B\subseteq D^h B ⊆ D h , so that A g − 1 , B h − 1 ∈ ( D 2 ) A^{g^{-1}},B^{h^{-1}}\in\binom{D}{2} A g − 1 , B h − 1 ∈ ( 2 D ) . Transitivity of G D G_D G D on ( D 2 ) \binom{D}{2} ( 2 D ) yields k ∈ G D k\in G_D k ∈ G D with ( A g − 1 ) k = B h − 1 (A^{g^{-1}})^k=B^{h^{-1}} ( A g − 1 ) k = B h − 1 , hence A g − 1 k h = B A^{g^{-1}kh}=B A g − 1 k h = B ; so G G G is transitive on ( P 2 ) \binom{P}{2} ( 2 P ) , i.e. 2-homogeneous. The final assertion follows: if G G G is not 2-homogeneous, G D G_D G D cannot act transitively on D D D , so the induced image is an intransitive subgroup of Sym ( D ) ≅ S 3 \operatorname{Sym}(D)\cong S_3 Sym ( D ) ≅ S 3 , namely the trivial group or a subgroup generated by a transposition.
Lemma 2.5 (Primitive ⋆ 3 \star_3 ⋆ 3 -groups are 2-homogeneous). Let G ≤ Sym ( P ) G\leq\operatorname{Sym}(\mathcal{P}) G ≤ Sym ( P ) be a primitive group of degree at least 8. If G G G satisfies the Siemons–Wagner property ⋆ 3 \star_3 ⋆ 3 , then G G G is 2-homogeneous.
Proof. Suppose G G G is not 2-homogeneous; then G G G has at least two orbits on pairs, and the witness triple D = { 1 , 2 , 3 } D=\{1,2,3\} D = { 1 , 2 , 3 } lies in a maximal triple orbit [1 Prop. 6.2.3] . If σ 2 ( G ) = 3 \sigma_2(G)=3 σ 2 ( G ) = 3 the witness triple is a rainbow triple, whose structural constraints are recorded in Lemma 2.6 ; its exclusion is not covered by the present lemma. Since we are assuming that G G G is not 2-homogeneous, Lemma 2.4 implies that G D G_D G D is intransitive on D D D ; consequently the induced image of G D G_D G D on D D D is trivial or generated by a transposition (relabelling so that, in the latter case, the image is ⟨ ( 1 2 ) ⟩ \langle(1\,2)\rangle ⟨( 1 2 )⟩ , fixing 3). For every quadruple Σ = { 1 , 2 , 3 , 4 } ⊃ D \Sigma=\{1,2,3,4\}\supset D Σ = { 1 , 2 , 3 , 4 } ⊃ D the defining inequality of ⋆ 3 \star_3 ⋆ 3 reads
∣ G Σ ∣ > m , m : = ∣ G D ∣ . (1) |G_\Sigma|>m,\qquad m:=|G_D|. \tag*{(1)} ∣ G Σ ∣ > m , m := ∣ G D ∣. (1)
Step 1. The image of G Σ G_\Sigma G Σ on Σ \Sigma Σ contains no 3-cycle on { 1 , 2 , 3 } \{1,2,3\} { 1 , 2 , 3 } . A 3-cycle ( 1 2 3 ) (1\,2\,3) ( 1 2 3 ) or ( 1 3 2 ) (1\,3\,2) ( 1 3 2 ) preserves D D D setwise and acts transitively on it, so the image of G D G_D G D would contain C 3 C_3 C 3 , contradicting image ( G D ) ∈ { 1 , C 2 } \operatorname{image}(G_D)\in\{1,C_2\} image ( G D ) ∈ { 1 , C 2 } .
Step 2. G Σ G_\Sigma G Σ does not preserve D D D setwise. Otherwise G Σ ≤ G D G_\Sigma\leq G_D G Σ ≤ G D and ∣ G Σ ∣ ≤ m |G_\Sigma|\leq m ∣ G Σ ∣ ≤ m , contradicting (*). Hence some k ∈ G Σ k\in G_\Sigma k ∈ G Σ maps D D D onto D ′ = { 1 , 2 , 4 } D'=\{1,2,4\} D ′ = { 1 , 2 , 4 } (relabelling); as k k k fixes Σ \Sigma Σ setwise it sends 4 to 3, so
k ( D ′ ) = { k ( 1 ) , k ( 2 ) , 3 } . (2) k(D')=\{k(1),k(2),3\}. \tag*{(2)} k ( D ′ ) = { k ( 1 ) , k ( 2 ) , 3 } . (2)
Step 3. The pair-orbit patterns. The element k k k preserves the pair-orbits of G G G ; in particular the pairs k ( 12 ) , k ( 13 ) , k ( 23 ) k(12),k(13),k(23) k ( 12 ) , k ( 13 ) , k ( 23 ) are the three pairs of D ′ D' D ′ , each in the pair-orbit of its preimage. If { k ( 1 ) , k ( 2 ) } = { 1 , 2 } \{k(1),k(2)\}=\{1,2\} { k ( 1 ) , k ( 2 )} = { 1 , 2 } , then (†) gives k ( D ′ ) = D k(D')=D k ( D ′ ) = D ; if { k ( 1 ) , k ( 2 ) } = { 1 , 4 } \{k(1),k(2)\}=\{1,4\} { k ( 1 ) , k ( 2 )} = { 1 , 4 } or { 2 , 4 } \{2,4\} { 2 , 4 } , then k ( D ′ ) = { 1 , 3 , 4 } k(D')=\{1,3,4\} k ( D ′ ) = { 1 , 3 , 4 } or { 2 , 3 , 4 } \{2,3,4\} { 2 , 3 , 4 } , which lies in the orbit of D ′ D' D ′ and hence of D D D . In the non-triangular case ( 12 ∈ O 1 , 13 , 23 ∈ O 2 ) (12\in O_1, 13,23\in O_2) ( 12 ∈ O 1 , 13 , 23 ∈ O 2 ) the pair-orbit condition forces the image of k k k to be ( 3 4 ) (3\,4) ( 3 4 ) or ( 1 2 ) ( 3 4 ) (1\,2)(3\,4) ( 1 2 ) ( 3 4 ) : Step 1 excludes the 3-cycles ( 1 2 3 ) , ( 1 3 2 ) (1\,2\,3),(1\,3\,2) ( 1 2 3 ) , ( 1 3 2 ) ; a direct check against the forced pattern excludes the 4-cycles, the 3-cycles involving 4, and every transposition other than ( 3 4 ) (3\ 4) ( 3 4 ) , ( 1 2 ) ( 3 4 ) (1\ 2)(3\ 4) ( 1 2 ) ( 3 4 ) — e.g. a 4-cycle sends the pair 12 ∈ O 1 12 \in O_1 12 ∈ O 1 to a pair in O 2 O_2 O 2 or fixes none of 13 13 13 , 23 23 23 , and a transposition ( 1 3 ) (1\ 3) ( 1 3 ) sends 12 12 12 to 23 ∉ O 1 23 \notin O_1 23 ∈ / O 1 . Thus the image of G Σ G_\Sigma G Σ lies in the Klein group V 4 = ⟨ ( 1 2 ) , ( 3 4 ) ⟩ V_4=\langle(1\ 2),(3\ 4)\rangle V 4 = ⟨( 1 2 ) , ( 3 4 )⟩ .
Step 4. The surviving configurations. Steps 1–3 reduce a hypothetical counterexample to one of two explicit configurations, whose pair-orbit patterns are forced by k k k :
(C1) non-triangular witness: 12 ∈ O 1 12 \in O_1 12 ∈ O 1 and 13 , 23 ∈ O 2 13,23 \in O_2 13 , 23 ∈ O 2 . Since the image of k k k is ( 3 4 ) (3\ 4) ( 3 4 ) or ( 1 2 ) ( 3 4 ) (1\ 2)(3\ 4) ( 1 2 ) ( 3 4 ) by Step 3 and k k k preserves pair-orbits, k ( 13 ) k(13) k ( 13 ) is { 1 , 4 } \{1,4\} { 1 , 4 } or { 2 , 4 } \{2,4\} { 2 , 4 } and lies in O 2 O_2 O 2 , so 14 , 24 ∈ O 2 14,24 \in O_2 14 , 24 ∈ O 2 ; the image of G Σ G_\Sigma G Σ is contained in V 4 = ⟨ ( 1 2 ) , ( 3 4 ) ⟩ V_4=\langle(1\ 2),(3\ 4)\rangle V 4 = ⟨( 1 2 ) , ( 3 4 )⟩ .
(C2) triangular witness: 12 , 13 , 23 ∈ O 1 12,13,23 \in O_1 12 , 13 , 23 ∈ O 1 . This configuration does not survive: by the pair-covering property of Lemma 2.4 (which follows from the structural alternative of [1 Thm. 6.1.6, p. 83] , not from the numerical inequality) every 2-subset of P \mathcal{P} P lies in a G G G -image D g \mathcal{D}^g D g of the witness; the three pairs of D g \mathcal{D}^g D g all lie in O 1 O_1 O 1 because D \mathcal{D} D is triangular, so every pair lies in O 1 O_1 O 1 and σ 2 ( G ) = 1 \sigma_2(G)=1 σ 2 ( G ) = 1 , contradicting the standing hypothesis that G G G is not 2-homogeneous. (For completeness we record the structure forced by such a witness. For the quadruple Σ = { 1 , 2 , 3 , 4 } \Sigma=\{1,2,3,4\} Σ = { 1 , 2 , 3 , 4 } the element k k k of Step 2 maps D \mathcal{D} D onto one of { 1 , 2 , 4 } , { 1 , 3 , 4 } , { 2 , 3 , 4 } \{1,2,4\},\{1,3,4\},\{2,3,4\} { 1 , 2 , 4 } , { 1 , 3 , 4 } , { 2 , 3 , 4 } ; since k k k preserves the pair-orbits and the three pairs of D \mathcal{D} D all lie in O 1 O_1 O 1 , the three pairs of k ( D ) k(\mathcal{D}) k ( D ) lie in O 1 O_1 O 1 ; hence two of the three pairs 14 , 24 , 34 14,24,34 14 , 24 , 34 lie in O 1 O_1 O 1 . If 34 ∈ O 2 34 \in O_2 34 ∈ O 2 , then necessarily k ( D ) = { 1 , 2 , 4 } k(\mathcal{D})=\{1,2,4\} k ( D ) = { 1 , 2 , 4 } (the alternatives { 1 , 3 , 4 } \{1,3,4\} { 1 , 3 , 4 } or { 2 , 3 , 4 } \{2,3,4\} { 2 , 3 , 4 } would put 34 34 34 into O 1 O_1 O 1 ), so that 14 , 24 ∈ O 1 14,24 \in O_1 14 , 24 ∈ O 1 and { 1 , 2 , 4 } ∈ D G \{1,2,4\}\in\mathcal{D}^G { 1 , 2 , 4 } ∈ D G .
The odd-order case is impossible. Suppose ∣ G ∣ |G| ∣ G ∣ is odd; then so is ∣ G Σ ∣ |G_\Sigma| ∣ G Σ ∣ , and the permutation image of G Σ G_\Sigma G Σ on Σ \Sigma Σ is a subgroup of S 4 S_4 S 4 of odd order, hence trivial or cyclic of order 3. It is not trivial, since k ≠ 1 k\ne1 k = 1 . A 3-cycle preserving D \mathcal{D} D setwise (one of ( 1 2 3 ) , ( 1 3 2 ) (1\ 2\ 3),(1\ 3\ 2) ( 1 2 3 ) , ( 1 3 2 ) ) would put C 3 C_3 C 3 into the image of G D G_\mathcal{D} G D and give ∣ G D ∣ ≥ 3 ∣ G ( D ) ∣ ≥ 3 ∣ G ( Σ ) ∣ = ∣ G Σ ∣ |G_\mathcal{D}|\ge3|G(\mathcal{D})|\ge3|G(\Sigma)|=|G_\Sigma| ∣ G D ∣ ≥ 3∣ G ( D ) ∣ ≥ 3∣ G ( Σ ) ∣ = ∣ G Σ ∣ , contradicting (*). A 3-cycle of the type ( 1 2 4 ) (1\ 2\ 4) ( 1 2 4 ) fixes 3 and preserves the pair-orbits, so the pairs k ( 12 ) , k ( 13 ) , k ( 23 ) ∈ O 1 k(12),k(13),k(23)\in O_1 k ( 12 ) , k ( 13 ) , k ( 23 ) ∈ O 1 , being the three pairs of { 1 , 2 , 4 } \{1,2,4\} { 1 , 2 , 4 } , force 24 ∈ O 1 24\in O_1 24 ∈ O 1 and 34 ∈ O 1 34\in O_1 34 ∈ O 1 ; and a 3-cycle of the type ( 1 3 4 ) (1\ 3\ 4) ( 1 3 4 ) fixes 2 and forces 14 ∈ O 1 14\in O_1 14 ∈ O 1 and 34 ∈ O 1 34\in O_1 34 ∈ O 1 . In the first case 2 is O 1 O_1 O 1 -adjacent to 1, 3 and to every w w w (apply the argument to Σ = D ∪ { w } \Sigma=\mathcal{D}\cup\{w\} Σ = D ∪ { w } with the same relabelling), and in the second case the same holds for 1; either way O 1 O_1 O 1 contains all pairs and σ 2 ( G ) = 1 \sigma_2(G)=1 σ 2 ( G ) = 1 , contradicting the standing hypothesis. Hence a triangular witness does not exist when ∣ G ∣ |G| ∣ G ∣ is odd.
The even-order case. The excluded-position arguments of the family-level Propositions 2.9 –2.15 apply verbatim to the triangular triple: the fourth-point witnesses w w w constructed there (a common non-neighbour on the quadric, an exterior point off a line, a singular vector w = z + u ′ w=z+u' w = z + u ′ , or a subspace of defect ( m − 2 , m − 3 ) (m-2,m-3) ( m − 2 , m − 3 ) ) depend only on the span W = ⟨ x , y ⟩ W=\langle x,y\rangle W = ⟨ x , y ⟩ and the quadratic geometry, not on the pair type of D \mathcal{D} D , and the four-point inequality ∣ G D ∪ { w } ∣ ≤ ∣ G D ∣ |G_{\mathcal{D}\cup\{w\}}|\leq|G_{\mathcal{D}}| ∣ G D ∪ { w } ∣ ≤ ∣ G D ∣ holds for the triangular triple by the same excluded-position computation: the witness w w w constructed in each case depends only on the span W = ⟨ x , y ⟩ W=\langle x,y\rangle W = ⟨ x , y ⟩ and the quadratic geometry, not on the pair type of D \mathcal{D} D , and the stabilizer-order comparison applies directly to the triangular representatives, which are included in the exhaustive computations of Section 3. Thus (C2) is excluded in full generality, and every hypothetical counterexample is of the non-triangular type (C1) of Step 2, whose exclusion is carried out family-by-family in Propositions 2.9 , 2.10, 2.11, 2.12, 2.13, 2.14 and 2.15: these propositions exhibit, for every rank-3 group outside the exhaustive census of degree $8\leq v\leq 50 ({xref}Section 3, {xref}Tables 3 and 2), a fourth point $w$ with $|G_{\mathcal{D}\cup\{w\}}|\leq |G_{\mathcal{D}}|$ (the four-point inequality of {xref}Lemma 2.9`), so that the defining inequality ( ∗ ) (\ast) ( ∗ ) of ⋆ 3 \star_3 ⋆ 3 fails for that quadruple. The reduction of Steps 1–3 is independent of the degree, so the propositions and the census together close the lemma.
Lemma 2.6 (The structure of a rainbow witness). Let G ≤ Sym ( P ) G\leq\operatorname{Sym}(\mathcal{P}) G ≤ Sym ( P ) be a primitive group of degree n ≥ 8 n\geq8 n ≥ 8 satisfying ⋆ 3 \star_3 ⋆ 3 with a witness triple D = { 1 , 2 , 3 } \mathcal{D}=\{1,2,3\} D = { 1 , 2 , 3 } , and suppose σ 2 ( G ) = 3 \sigma_2(G)=3 σ 2 ( G ) = 3 with pair-orbits O 1 , O 2 , O 3 \mathcal{O}_1,\mathcal{O}_2,\mathcal{O}_3 O 1 , O 2 , O 3 such that, for a quadruple Σ = { 1 , 2 , 3 , 4 } \Sigma=\{1,2,3,4\} Σ = { 1 , 2 , 3 , 4 } ,
12 , 34 ∈ O 1 , 13 , 24 ∈ O 2 , 14 , 23 ∈ O 3 . ( ⋄ ) 12,34\in\mathcal{O}_1,\qquad13,24\in\mathcal{O}_2,\qquad14,23\in\mathcal{O}_3. \qquad(\diamond) 12 , 34 ∈ O 1 , 13 , 24 ∈ O 2 , 14 , 23 ∈ O 3 . ( ⋄ ) Set T : = D G T:=\mathcal{D}^G T := D G . Then the following structural constraints hold.
(i) Slice counts. For e ∈ O i e\in\mathcal{O}_i e ∈ O i ,
# { P ∈ T : e ⊂ P } = ∣ T ∣ ∣ O i ∣ , \#\{P\in T:e\subset P\}=\frac{|T|}{|\mathcal{O}_i|}, # { P ∈ T : e ⊂ P } = ∣ O i ∣ ∣ T ∣ , and ∑ S c T ( S ) = n − 3 \sum_S c_T(S)=n-3 ∑ S c T ( S ) = n − 3 , where S S S ranges over the quadruple orbits containing a D \mathcal{D} D -extension and c T ( S ) : = # { w : D ∪ { w } ∈ S } c_T(S):=\#\{w:\mathcal{D}\cup\{w\}\in S\} c T ( S ) := # { w : D ∪ { w } ∈ S } .
(ii) The stabilizer images. For every quadruple orbit S S S the image of the setwise stabilizer G Σ G_\Sigma G Σ on a quadruple Q ∈ S Q\in S Q ∈ S is either generated by a double transposition or is the Klein group V 4 V_4 V 4 .
(iii) The C 2 C_2 C 2 branch. If the image is ⟨ ( 1 2 ) ( 3 4 ) ⟩ \langle(1\ 2)(3\ 4)\rangle ⟨( 1 2 ) ( 3 4 )⟩ (or a conjugate), then exactly two triples of Q Q Q lie in T T T , c T ( S ) = 1 c_T(S)=1 c T ( S ) = 1 and ∣ S ∣ = ∣ T ∣ / 2 |S|=|T|/2 ∣ S ∣ = ∣ T ∣/2 ; the map P ↦ P ∗ P\mapsto P^* P ↦ P ∗ defined by P ∪ P ∗ = Q P\cup P^*=Q P ∪ P ∗ = Q is a fixed-point-free G G G -equivariant involution on T T T , and the two triples of the pair { P , P ∗ } \{P,P^*\} { P , P ∗ } share the edge of Q Q Q fixed by the double transposition, whence 2 ∣ O i ∣ ∣ ∣ T ∣ 2|O_i|\mid|T| 2∣ O i ∣ ∣ ∣ T ∣ for the corresponding i ∈ { 1 , 2 , 3 } i\in\{1,2,3\} i ∈ { 1 , 2 , 3 } .
(iv) The V 4 V_4 V 4 branch. If the image is V 4 V_4 V 4 , then all four triples of Q Q Q lie in T T T and c T ( S ) = u ∈ { 1 , 2 , 3 } c_T(S)=u\in\{1,2,3\} c T ( S ) = u ∈ { 1 , 2 , 3 } with 4 ∣ S ∣ = u ∣ T ∣ 4|S|=u|T| 4∣ S ∣ = u ∣ T ∣ .
(v) The rank bound. The stabilizer-image dichotomy of (ii) shows that 2 ∣ ∣ G ∣ 2\mid|G| 2 ∣ ∣ G ∣ : the image of G Σ G_\Sigma G Σ on a witness quadruple Σ \Sigma Σ has order 2 or 4, so ∣ G ∣ |G| ∣ G ∣ is even and G G G contains an involution. An involution swaps the endpoints of some pair, so at least one of the three pair-orbits is self-paired. Writing s s s for the number of self-paired pair-orbits, the point action of G G G has rank
1 + s + 2 ( 3 − s ) = 7 − s ≤ 6. 1+s+2(3-s)=7-s\le6. 1 + s + 2 ( 3 − s ) = 7 − s ≤ 6.
In particular the rank of a rainbow witness group is at most six; the three pair-orbits need not all be non-self-paired.
(vi) Valency and completion counts. Writing O A = O 1 O_A=O_1 O A = O 1 , O B = O 2 O_B=O_2 O B = O 2 , O C = O 3 O_C=O_3 O C = O 3 for the three colour-labelled pair-orbits of ⋄ \diamond ⋄ , let k i k_i k i be the valency of the graph O i O_i O i and let λ i \lambda_i λ i be the number of triples of T T T containing a given pair of O i O_i O i ; both are constant on O i O_i O i by G G G -transitivity. Independently of self-pairing,
k A + k B + k C = n − 1 , ∣ T ∣ = n k i 2 λ i , λ i = 2 ∣ T ∣ n k i , \begin{aligned}
k_A+k_B+k_C&=n-1,\qquad|T|=\frac{n k_i}{2}\lambda_i,\\
\lambda_i&=\frac{2|T|}{n k_i},
\end{aligned} k A + k B + k C λ i = n − 1 , ∣ T ∣ = 2 n k i λ i , = n k i 2∣ T ∣ ,
and the common-neighbour identity holds: for an O A O_A O A -pair { x , y } \{x,y\} { x , y } ,
λ A = ∣ Γ B ( x ) ∩ Γ C ( y ) ∣ + ∣ Γ C ( x ) ∩ Γ B ( y ) ∣ , \lambda_A=|\Gamma_B(x)\cap\Gamma_C(y)|+|\Gamma_C(x)\cap\Gamma_B(y)|, λ A = ∣ Γ B ( x ) ∩ Γ C ( y ) ∣ + ∣ Γ C ( x ) ∩ Γ B ( y ) ∣ ,
where Γ i ( x ) \Gamma_i(x) Γ i ( x ) denotes the neighbourhood of x x x in the graph O i O_i O i , and cyclically for B , C B,C B , C . With C i = { w ∉ D : e i ∪ { w } ∈ T } C_i=\{w\notin D:e_i\cup\{w\}\in T\} C i = { w ∈ / D : e i ∪ { w } ∈ T } for the O i O_i O i -edge e i e_i e i of D \mathcal{D} D one has ∣ C i ∣ = λ i − 1 |C_i|=\lambda_i-1 ∣ C i ∣ = λ i − 1 , C A ∪ C B ∪ C C = P ∖ D C_A\cup C_B\cup C_C=\mathcal{P}\setminus\mathcal{D} C A ∪ C B ∪ C C = P ∖ D , and, with u = ∣ { w ∉ D : w ∈ C A ∩ C B ∩ C C } ∣ u=|\{w\notin\mathcal{D}:w\in C_A\cap C_B\cap C_C\}| u = ∣ { w ∈ / D : w ∈ C A ∩ C B ∩ C C } ∣ .
λ A + λ B + λ C = n + 2 u . \lambda_A+\lambda_B+\lambda_C=n+2u. λ A + λ B + λ C = n + 2 u .
These identities are necessary and hold without any hypothesis on self-pairing; by themselves they do not exclude the self-paired cases.
(vii) Four-point orbits and stabilizer indices. For a quadruple orbit Q Q Q containing a triple of T T T put d Q = ∣ T ∩ ( Σ 3 ) ∣ d_Q=|T\cap\binom{\Sigma}{3}| d Q = ∣ T ∩ ( 3 Σ ) ∣ and b Q = ∣ { Σ ′ ∈ Q : D ⊆ Σ ′ } ∣ b_Q=|\{\Sigma'\in Q:\mathcal{D}\subseteq\Sigma'\}| b Q = ∣ { Σ ′ ∈ Q : D ⊆ Σ ′ } ∣ for Σ ∈ Q \Sigma\in Q Σ ∈ Q . Double counting gives ∣ Q ∣ d Q = ∣ T ∣ b Q |Q|d_Q=|T|b_Q ∣ Q ∣ d Q = ∣ T ∣ b Q . Writing H Σ H_\Sigma H Σ for the image of G Σ G_\Sigma G Σ on Σ = D ∪ { w } \Sigma=\mathcal{D}\cup\{w\} Σ = D ∪ { w } and r w = [ G D : G D ∩ G Σ ] r_w=[G_{\mathcal{D}}:G_{\mathcal{D}}\cap G_\Sigma] r w = [ G D : G D ∩ G Σ ] , the orbit-stabilizer identity ∣ G Σ ∣ ∣ G D ∣ = ∣ H Σ ∣ r w \frac{|G_\Sigma|}{|G_{\mathcal{D}}|}=\frac{|H_\Sigma|}{r_w} ∣ G D ∣ ∣ G Σ ∣ = r w ∣ H Σ ∣ and the witness inequality ∣ G Σ ∣ > ∣ G D ∣ |G_\Sigma|>|G_{\mathcal{D}}| ∣ G Σ ∣ > ∣ G D ∣ give ∣ H Σ ∣ > r w |H_\Sigma|>r_w ∣ H Σ ∣ > r w . The image H Σ H_\Sigma H Σ acts regularly on the set Q w = { P ∈ T : P ⊂ Σ } Q_w=\{P\in T:P\subset\Sigma\} Q w = { P ∈ T : P ⊂ Σ } not necessarily on all four triples of Σ \Sigma Σ , so d Q = ∣ Q w ∣ = ∣ H Σ ∣ ∈ { 2 , 4 } d_Q=|Q_w|=|H_\Sigma|\in\{2,4\} d Q = ∣ Q w ∣ = ∣ H Σ ∣ ∈ { 2 , 4 } and d Q > r w d_Q>r_w d Q > r w :
d Q = 2 ⇒ H Σ ≅ C 2 , r w = 1 ; d_Q=2\ \Rightarrow\ H_\Sigma\cong C_2,\ r_w=1; d Q = 2 ⇒ H Σ ≅ C 2 , r w = 1 ;
d Q = 4 ⇒ H Σ ≅ V 4 , r w ∈ { 1 , 2 , 3 } . d_Q=4\ \Rightarrow\ H_\Sigma\cong V_4,\ r_w\in\{1,2,3\}. d Q = 4 ⇒ H Σ ≅ V 4 , r w ∈ { 1 , 2 , 3 } .
Equivalently, via ∣ Q ∣ d Q = ∣ T ∣ b Q |Q|d_Q=|T|b_Q ∣ Q ∣ d Q = ∣ T ∣ b Q and ∣ Q ∣ < ∣ T ∣ |Q|<|T| ∣ Q ∣ < ∣ T ∣ , b Q < d Q b_Q<d_Q b Q < d Q : b Q = 1 b_Q=1 b Q = 1 for d Q = 2 d_Q=2 d Q = 2 and b Q ∈ { 1 , 2 , 3 } b_Q\in\{1,2,3\} b Q ∈ { 1 , 2 , 3 } for d Q = 4 d_Q=4 d Q = 4 .
Proof. Let m ( S ) m(S) m ( S ) be the number of triples P ∈ T P\in T P ∈ T with P ⊂ Q P\subset Q P ⊂ Q for Q ∈ S Q\in S Q ∈ S . Since G G G is transitive on T T T and on S S S , the number of quadruples of S S S containing a fixed triple of T T T is constant, and double counting the incidences ( P , Q ) (P,Q) ( P , Q ) with P ∈ T P\in T P ∈ T , P ⊂ Q ∈ S P\subset Q\in S P ⊂ Q ∈ S gives
c T ( S ) ∣ T ∣ = ∣ S ∣ m ( S ) . c_T(S)|T|=|S|m(S). c T ( S ) ∣ T ∣ = ∣ S ∣ m ( S ) .
Summing ( ∗ ∗ ) (**) ( ∗ ∗ ) over the quadruple orbits S S S containing a triple of T T T yields ∑ S c T ( S ) = n − 3 \sum_S c_T(S)=n-3 ∑ S c T ( S ) = n − 3 , each triple of T T T having exactly n − 3 n-3 n − 3 extensions. For Q = Σ Q=\Sigma Q = Σ the four triples D \mathcal{D} D , { 1 , 2 , 4 } \{1,2,4\} { 1 , 2 , 4 } , { 1 , 3 , 4 } \{1,3,4\} { 1 , 3 , 4 } , { 2 , 3 , 4 } \{2,3,4\} { 2 , 3 , 4 } each carry one pair from each of O 1 , O 2 , O 3 O_1,O_2,O_3 O 1 , O 2 , O 3 by ( ♢ ) (\diamondsuit) ( ♢ ) , so all four are rainbow. An element of G Σ G_\Sigma G Σ preserving the pattern ( ♢ ) (\diamondsuit) ( ♢ ) permutes the three “opposite” pairs 12 , 34 ∈ O 1 12,34\in O_1 12 , 34 ∈ O 1 , 13 , 24 ∈ O 2 13,24\in O_2 13 , 24 ∈ O 2 and 14 , 23 ∈ O 3 14,23\in O_3 14 , 23 ∈ O 3 , which forces its image to lie in V 4 = ⟨ ( 1 2 ) ( 3 4 ) , ( 1 3 ) ( 2 4 ) ⟩ V_4=\langle(1\ 2)(3\ 4),(1\ 3)(2\ 4)\rangle V 4 = ⟨( 1 2 ) ( 3 4 ) , ( 1 3 ) ( 2 4 )⟩ ; transpositions and 3-cycles move an O 1 O_1 O 1 -pair into O 2 O_2 O 2 or O 3 O_3 O 3 and are excluded, as are the 4-cycles. Hence m ( S ) ∈ { 2 , 4 } m(S)\in\{2,4\} m ( S ) ∈ { 2 , 4 } . In the C 2 C_2 C 2 case the image of the generator is a double transposition k k k ; the second triple of Q Q Q in T T T is P ∗ : = k ( P ) P^{*}:=k(P) P ∗ := k ( P ) , the pair { P , P ∗ } \{P,P^{*}\} { P , P ∗ } sharing the edge of Q Q Q fixed by k k k , and ( ∗ ∗ ) (\ast\ast) ( ∗ ∗ ) with m ( S ) = 2 m(S)=2 m ( S ) = 2 reads c T ( S ) ∣ T ∣ = 2 ∣ S ∣ c_T(S)|T|=2|S| c T ( S ) ∣ T ∣ = 2∣ S ∣ ; the witness inequality ∣ S ∣ < ∣ T ∣ |S|<|T| ∣ S ∣ < ∣ T ∣ forces c T ( S ) = 1 c_T(S)=1 c T ( S ) = 1 and ∣ S ∣ = ∣ T ∣ / 2 |S|=|T|/2 ∣ S ∣ = ∣ T ∣/2 . The map P ↦ P ∗ P\mapsto P^{*} P ↦ P ∗ is G G G -equivariant, since G G G is transitive on S S S . The divisibility $2|O_i|\mid |T|$ follows from the slice count in (i) applied to the shared edge. In the V 4 V_4 V 4 case all four triples of Q Q Q lie in T T T , so ( ∗ ∗ ) (\ast\ast) ( ∗ ∗ ) reads c T ( S ) ∣ T ∣ = 4 ∣ S ∣ c_T(S)|T|=4|S| c T ( S ) ∣ T ∣ = 4∣ S ∣ , and the witness inequality gives c T ( S ) = u ∈ { 1 , 2 , 3 } c_T(S)=u\in\{1,2,3\} c T ( S ) = u ∈ { 1 , 2 , 3 } , i.e. 4 ∣ S ∣ = u ∣ T ∣ 4|S|=u|T| 4∣ S ∣ = u ∣ T ∣ . Finally, for the slice count: the assignment P ↦ P\mapsto P ↦ the unique pair of P P P in O i O_i O i is G G G -equivariant, so # { P ∈ T : e ⊂ P } \#\{P\in T:e\subset P\} # { P ∈ T : e ⊂ P } is constant for e ∈ O i e\in O_i e ∈ O i ; double counting { ( P , e ) : P ∈ T , e ⊂ P , e ∈ O i } \{(P,e):P\in T,\ e\subset P,\ e\in O_i\} {( P , e ) : P ∈ T , e ⊂ P , e ∈ O i } gives ∣ O i ∣ ⋅ # { P ∈ T : e ⊂ P } = ∣ T ∣ |O_i|\cdot\#\{P\in T:e\subset P\}=|T| ∣ O i ∣ ⋅ # { P ∈ T : e ⊂ P } = ∣ T ∣ , each P ∈ T P\in T P ∈ T having exactly one pair in O i O_i O i . Proof of (v)–(vii). For (v): by (ii) the image of G Σ G_\Sigma G Σ on a witness quadruple Σ \Sigma Σ has order 2 2 2 or 4 4 4 , so 2 ∣ ∣ G ∣ 2\mid|G| 2 ∣ ∣ G ∣ ; an involution of G G G swaps the endpoints of a pair lying in a self-paired pair-orbit, so s ≥ 1 s\ge1 s ≥ 1 and the rank formula 1 + s + 2 ( 3 − s ) = 7 − s 1+s+2(3-s)=7-s 1 + s + 2 ( 3 − s ) = 7 − s applies. For (vi): the sum of the valencies is n − 1 n-1 n − 1 ; counting the pairs of the triples of T T T gives ∣ T ∣ = n k i 2 λ i |T|=\frac{n k_i}{2}\lambda_i ∣ T ∣ = 2 n k i λ i ; the common-neighbour identity counts the completions of an O A O_A O A -pair { x , y } \{x,y\} { x , y } according as χ ( x w ) , χ ( y w ) \chi(xw),\chi(yw) χ ( x w ) , χ ( y w ) equals ( B , C ) (B,C) ( B , C ) or ( C , B ) (C,B) ( C , B ) ; the relations for the C i C_i C i and u u u follow by counting the points w ∉ D w\notin D w ∈ / D according to the H w H_w H w -orbit structure of the four triples of Σ = D ∪ { w } \Sigma=D\cup\{w\} Σ = D ∪ { w } —an image of order 2 2 2 produces m ( w ) = 1 m(w)=1 m ( w ) = 1 or 3 3 3 , and a V 4 V_4 V 4 image produces m ( w ) = 3 m(w)=3 m ( w ) = 3 . For (vii): the identity ∣ G Σ ∣ ∣ G D ∣ = ∣ H Σ ∣ r w \frac{|G_\Sigma|}{|G_D|}=\frac{|H_\Sigma|}{r_w} ∣ G D ∣ ∣ G Σ ∣ = r w ∣ H Σ ∣ follows from G D ∩ G Σ = G ( Σ ) G_D\cap G_\Sigma=G_{(\Sigma)} G D ∩ G Σ = G ( Σ ) and the orbit-stabilizer theorem; the witness inequality is ∣ H Σ ∣ > r w |H_\Sigma|>r_w ∣ H Σ ∣ > r w . To see that H Σ H_\Sigma H Σ is regular on Q w Q_w Q w , suppose h ∈ H Σ h\in H_\Sigma h ∈ H Σ fixes some P ∈ Q w P\in Q_w P ∈ Q w and lift it to h ~ ∈ G Σ \widetilde h\in G_\Sigma h ∈ G Σ . Since P P P is rainbow, h ~ \widetilde h h preserves three pairwise different pair-colours on P P P and therefore fixes P P P pointwise; as it also preserves Σ \Sigma Σ , it fixes the fourth point, hence fixes Σ \Sigma Σ pointwise and h = 1 h=1 h = 1 . Thus ∣ H Σ ∣ = ∣ Q w ∣ = d Q |H_\Sigma|=|Q_w|=d_Q ∣ H Σ ∣ = ∣ Q w ∣ = d Q , and a regular subgroup of S 4 S_4 S 4 acting on d Q d_Q d Q faces with d Q ∈ { 2 , 4 } i s d_Q\in\{2,4\} is d Q ∈ { 2 , 4 } i s C_2$ or V 4 V_4 V 4 : C 3 C_3 C 3 and C 4 C_4 C 4 would force all four faces of Σ \Sigma Σ to be rainbow, and the colour-preserving group of a rainbow K 4 K_4 K 4 is V 4 V_4 V 4 , which contains neither C 3 C_3 C 3 nor C 4 C_4 C 4 . For faces of Σ \Sigma Σ outside T T T , H Σ H_\Sigma H Σ may have nontrivial stabilizers—in the single-extension branch a transposition fixes the two non-T T T faces. The alternative form b Q < d Q b_Q<d_Q b Q < d Q is the double counting ∣ Q ∣ d Q = ∣ T ∣ b Q |Q|d_Q=|T|b_Q ∣ Q ∣ d Q = ∣ T ∣ b Q combined with the witness inequality ∣ Q ∣ < ∣ T ∣ |Q|<|T| ∣ Q ∣ < ∣ T ∣ , giving b Q = 1 b_Q=1 b Q = 1 for d Q = 2 d_Q=2 d Q = 2 and b Q ∈ { 1 , 2 , 3 } b_Q\in\{1,2,3\} b Q ∈ { 1 , 2 , 3 } for d Q = 4 d_Q=4 d Q = 4 .
Lemma 2.7 (Point-pair stabilizers and counting obstructions for a rainbow witness). Let G ≤ Sym ( P ) G\leq\operatorname{Sym}(\mathcal{P}) G ≤ Sym ( P ) be as in Lemma 2.6, and suppose in addition that G D = 1 G_D=1 G D = 1 . For i ∈ { 1 , 2 , 3 } i\in\{1,2,3\} i ∈ { 1 , 2 , 3 } let e i = { a , b } e_i=\{a,b\} e i = { a , b } be the edge of D D D of colour i i i and let $c
be the third point of D \mathcal{D} D ; put K i : = G a ∩ G b K_i := G_a \cap G_b K i := G a ∩ G b , the pointwise stabilizer of e i e_i e i , and let k i , λ i k_i,\lambda_i k i , λ i be the valency and the completion count of O i O_i O i as in Lemma 2.6(vi).
(i) Regularity of the point-pair stabilizers. The group K i K_i K i acts regularly on the orbit c K i c^{K_i} c K i . Indeed, if k ∈ K i k \in K_i k ∈ K i fixes c c c , then k k k fixes D \mathcal{D} D pointwise and k = 1 k=1 k = 1 (semi-regularity); the action on c K i c^{K_i} c K i is transitive by definition, so ∣ c K i ∣ = ∣ K i ∣ |c^{K_i}|=|K_i| ∣ c K i ∣ = ∣ K i ∣ .
(ii) Unique lifts of colour-preserving transpositions. For each w ∈ c K i w \in c^{K_i} w ∈ c K i there is a unique k ∈ K i k \in K_i k ∈ K i with k ( c ) = w k(c)=w k ( c ) = w : if k 1 ( c ) = k 2 ( c ) = w k_1(c)=k_2(c)=w k 1 ( c ) = k 2 ( c ) = w , then k 2 − 1 k 1 k_2^{-1}k_1 k 2 − 1 k 1 fixes a , b , c a,b,c a , b , c pointwise and lies in G D = 1 G_{\mathcal{D}}=1 G D = 1 . This k k k fixes a , b a,b a , b and preserves the colours of the edges a c ac a c and b c bc b c , so it is the unique lift in K i K_i K i of the colour-preserving transposition ( c w ) (c\ w) ( c w ) on Σ = D ∪ { w } \Sigma=\mathcal{D}\cup\{w\} Σ = D ∪ { w } .
(iii) Elementary abelian structure. For every k ∈ K i k \in K_i k ∈ K i one has k 2 = 1 k^2=1 k 2 = 1 : k 2 k^2 k 2 fixes a , b , c a,b,c a , b , c pointwise, hence lies in G D = 1 G_{\mathcal{D}}=1 G D = 1 . Thus K i = 1 K_i=1 K i = 1 or K i ≅ C 2 r K_i\cong C_2^r K i ≅ C 2 r (an elementary abelian 2-group).
(iv) Completion counts. With L i : = G { a , b } L_i:=G_{\{a,b\}} L i := G { a , b } the setwise stabilizer of the edge e i = { a , b } e_i=\{a,b\} e i = { a , b } and K i = G a ∩ G b K_i=G_a\cap G_b K i = G a ∩ G b the pointwise pair stabilizer, the total number λ i \lambda_i λ i of triples P ∈ T P\in T P ∈ T containing e i e_i e i decomposes over the suborbits of L i L_i L i on { P ∈ T : e i ⊆ P } \{P\in T:e_i\subseteq P\} { P ∈ T : e i ⊆ P } :
λ i = ∑ O ∈ O i ∣ L i ∣ ∣ L i ∩ G T O ∣ . \lambda_i=\sum_{O\in\mathcal{O}_i}\frac{|L_i|}{|L_i\cap G_{T_O}|}. λ i = O ∈ O i ∑ ∣ L i ∩ G T O ∣ ∣ L i ∣ .
where O i \mathcal{O}_i O i is the set of L i L_i L i -orbits on those triples and T O T_O T O is a representative of the orbit O O O . The orbit containing D \mathcal{D} D has length ∣ L i ∣ / ∣ G D ∣ |L_i|/|G_{\mathcal{D}}| ∣ L i ∣/∣ G D ∣ ; since L i / K i L_i/K_i L i / K i has index [ L i : K i ] = : ε i ∈ { 1 , 2 } [L_i:K_i]=:\varepsilon_i\in\{1,2\} [ L i : K i ] =: ε i ∈ { 1 , 2 } (an element swapping a a a and b b b ) and the K i K_i K i -orbit of c c c has length ∣ K i ∣ / ∣ K i , c ∣ = ∣ K i ∣ / m |K_i|/|K_{i,c}|=|K_i|/m ∣ K i ∣/∣ K i , c ∣ = ∣ K i ∣/ m , where m : = ∣ G D ∣ = ∣ G ( a , b , c ) ∣ m:=|G_{\mathcal{D}}|=|G_{(a,b,c)}| m := ∣ G D ∣ = ∣ G ( a , b , c ) ∣ is the pointwise stabilizer of D \mathcal{D} D ,
λ i ≥ ε i ∣ K i ∣ m , ε i = 2 ⟺ O i self-paired \lambda_i\ge\frac{\varepsilon_i|K_i|}{m},\qquad\varepsilon_i=2\Longleftrightarrow O_i\text{ self-paired} λ i ≥ m ε i ∣ K i ∣ , ε i = 2 ⟺ O i self-paired
*(self-paired for the ordered pair orbital: O i = O i ∗ O_i=O_i^* O i = O i ∗ ), with equality when L i L_i L i is transitive on the completion triples. In the case m = 1 m=1 m = 1 of Lemma 2.7 the orbit of D \mathcal{D} D has length ε i ∣ K i ∣ ∈ { ∣ K i ∣ , 2 ∣ K i ∣ } \varepsilon_i|K_i|\in\{|K_i|,2|K_i|\} ε i ∣ K i ∣ ∈ { ∣ K i ∣ , 2∣ K i ∣ } .
(v) The slice-count inequality. Double counting the points w ∉ D w \notin\mathcal{D} w ∈ / D according to the colour layers gives
∑ w ∉ D m ( w ) = ∑ i = 1 3 ( λ i − 1 ) , \sum_{w \notin\mathcal{D}} m(w)=\sum_{i=1}^{3}(\lambda_i-1), w ∈ / D ∑ m ( w ) = i = 1 ∑ 3 ( λ i − 1 ) ,
m ( w ) : = # { i : e i ∪ { w } ∈ T } . m(w):=\#\{i:e_i\cup\{w\}\in T\}. m ( w ) := # { i : e i ∪ { w } ∈ T } .
Every w ∉ D w \notin\mathcal{D} w ∈ / D satisfies m ( w ) ≥ 1 m(w)\geq1 m ( w ) ≥ 1 (by Lemma 2.6 (vii)), so a rainbow witness satisfies
λ 1 + λ 2 + λ 3 ≥ n . (3) \lambda_1+\lambda_2+\lambda_3\geq n. \tag*{(3)} λ 1 + λ 2 + λ 3 ≥ n . (3)
The orbit of D \mathcal{D} D among the completion triples has length ε i ∣ K i ∣ / m = 2 ∣ G α ∣ / ( k i m ) \varepsilon_i|K_i|/m=2|G_\alpha|/(k_i m) ε i ∣ K i ∣/ m = 2∣ G α ∣/ ( k i m ) ( Lemma 2.6 (vi) with (iv) above; here k i k_i k i denotes the valency of the undirected orbital graph on O i ∪ O i ∗ O_i\cup O_i^* O i ∪ O i ∗ , which equals the out-valency in the self-paired case and twice the out-valency otherwise), and the total completion count λ i \lambda_i λ i of (1) is at least this length. Each undirected orbital graph is regular and, G G G being primitive, connected; a connected regular graph of valency 1 1 1 is a perfect matching, whose blocks would form a non-trivial G G G -invariant partition, contrary to primitivity. Hence k i ≥ 2 k_i\geq2 k i ≥ 2 , and with k 1 + k 2 + k 3 = n − 1 k_1+k_2+k_3=n-1 k 1 + k 2 + k 3 = n − 1 this gives
∑ i 1 k i ≤ 1 + 1 n − 5 . \sum_i\frac{1}{k_i}\leq1+\frac{1}{n-5}. i ∑ k i 1 ≤ 1 + n − 5 1 .
In particular, whenever the completion triples form a single L i L_i L i -orbit (λ i = 2 ∣ G α ∣ / ( k i m ) \lambda_i=2|G_\alpha|/(k_i m) λ i = 2∣ G α ∣/ ( k i m ) ) a witness forces the estimate
∣ G α ∣ m ≥ n ( n − 5 ) 2 ( n − 4 ) . (4) \frac{|G_\alpha|}{m}\geq\frac{n(n-5)}{2(n-4)}. \tag*{(4)} m ∣ G α ∣ ≥ 2 ( n − 4 ) n ( n − 5 ) . (4)
which in the case m = 1 m=1 m = 1 reads ∣ G α ∣ ≥ n ( n − 5 ) / ( 2 ( n − 4 ) ) |G_\alpha|\geq n(n-5)/(2(n-4)) ∣ G α ∣ ≥ n ( n − 5 ) / ( 2 ( n − 4 )) . c : = ( λ 1 + λ 2 + λ 3 − n ) / 2 c:=(\lambda_1+\lambda_2+\lambda_3-n)/2 c := ( λ 1 + λ 2 + λ 3 − n ) /2 must be a non-negative integer, and for each i i i ,
f i : = λ i − c − ∣ K i ∣ ≥ 0. (5) f_i:=\lambda_i-c-|K_i|\geq0. \tag*{(5)} f i := λ i − c − ∣ K i ∣ ≥ 0. (5)
(vi) Non-triviality of the K i K_i K i . The hypothesis G D = 1 G_{\mathcal{D}}=1 G D = 1 alone does not force ∑ i ( ∣ K i ∣ − 1 ) > 0 \sum_i(|K_i|-1)>0 ∑ i ( ∣ K i ∣ − 1 ) > 0 ; it must be supplied by an actual single-extension transposition. Whenever a single-extension point w w w of colour i i i occurs, (ii) produces a unique non-trivial k ∈ K i k\in K_i k ∈ K i , so ∣ K i ∣ ≥ 2 |K_i|\geq2 ∣ K i ∣ ≥ 2 . Hence, if at least one single-extension transposition occurs, then ∑ i ( ∣ K i ∣ − 1 ) > 0 \sum_i(|K_i|-1)>0 ∑ i ( ∣ K i ∣ − 1 ) > 0 .
(vii) Fixed-point obstruction of a single extension. Let τ = ( 3 w ) \tau=(3w) τ = ( 3 w ) be the transposition realizing an A A A -type single extension (so w ∈ C 1 w\in C_1 w ∈ C 1 and r w = 1 ) r_w = 1) r w = 1 ) . Then τ ∈ G 1 , 2 \tau\in G_{1,2} τ ∈ G 1 , 2 and τ ( 3 ) = w \tau(3) = w τ ( 3 ) = w . If x ∉ { 1 , 2 } x \notin\{1,2\} x ∈ / { 1 , 2 } and τ ( x ) = x \tau(x) = x τ ( x ) = x , then { 1 , 2 , x } ∉ T \{1,2,x\} \notin T { 1 , 2 , x } ∈ / T : otherwise τ \tau τ fixes the triple { 1 , 2 , x } ∈ T \{1,2,x\} \in T { 1 , 2 , x } ∈ T pointwise, contradicting G D = 1 G_D = 1 G D = 1 . Hence
Fix ( τ ) ∖ { 1 , 2 } ⊆ { x : { 1 , 2 , x } ∉ T } . (6) \operatorname{Fix}(\tau) \setminus\{1,2\} \subseteq\{x : \{1,2,x\} \notin T\}. \tag*{(6)} Fix ( τ ) ∖ { 1 , 2 } ⊆ { x : { 1 , 2 , x } ∈ / T } . (6)
The B B B - and C C C -type transpositions ( 2 w ) (2\,w) ( 2 w ) , ( 1 w ) (1\,w) ( 1 w ) satisfy the analogous restrictions with { 1 , 3 } \{1,3\} { 1 , 3 } and { 2 , 3 } \{2,3\} { 2 , 3 } in place of { 1 , 2 } \{1,2\} { 1 , 2 } . Finally, the assignment w ↦ k w w \mapsto k_w w ↦ k w (the unique lift of (ii)) is injective, so the number of A A A -type single-extension points is at most # { τ ∈ K 1 : τ 2 = 1 , τ ( 3 ) ≠ 3 } \#\{\tau\in K_1 : \tau^2 = 1,\ \tau(3) \ne3\} # { τ ∈ K 1 : τ 2 = 1 , τ ( 3 ) = 3 } .
Proposition 2.8 (A rainbow witness contains a single-extension configuration). Let G ≤ Sym ( P ) G \leq\operatorname{Sym}(\mathcal{P}) G ≤ Sym ( P ) be primitive of degree n ≥ 8 n \geq8 n ≥ 8 satisfying ⋆ 3 \star_3 ⋆ 3 with a rainbow witness D = { 1 , 2 , 3 } D = \{1,2,3\} D = { 1 , 2 , 3 } and σ 2 ( G ) = 3 \sigma_2(G) = 3 σ 2 ( G ) = 3 as in Lemma 2.6, with colours χ ( 12 ) = 1 \chi(12) = 1 χ ( 12 ) = 1 , χ ( 13 ) = 2 \chi(13) = 2 χ ( 13 ) = 2 , χ ( 23 ) = 3 \chi(23) = 3 χ ( 23 ) = 3 . Then there exists w ∉ D w \notin D w ∈ / D whose colour vector ( χ ( 1 w ) , χ ( 2 w ) , χ ( 3 w ) ) (\chi(1w),\chi(2w),\chi(3w)) ( χ ( 1 w ) , χ ( 2 w ) , χ ( 3 w )) is not ( 3 , 2 , 1 ) (3,2,1) ( 3 , 2 , 1 ) ; equivalently, a rainbow witness necessarily contains a single-extension configuration in which at most one of the three extension triples { 1 , 2 , w } \{1,2,w\} { 1 , 2 , w } , { 1 , 3 , w } \{1,3,w\} { 1 , 3 , w } , { 2 , 3 , w } \{2,3,w\} { 2 , 3 , w } lies in T = D G T = D^G T = D G .
Proof. By (vii) of Lemma 2.6 every w ∉ D w \notin D w ∈ / D has d Q ≥ 2 d_Q \geq2 d Q ≥ 2 , so at least one extension triple lies in T T T . Suppose, for a contradiction, that every w w w has colour vector ( 3 , 2 , 1 ) (3,2,1) ( 3 , 2 , 1 ) ; that is, χ ( 1 w ) = 3 \chi(1w) = 3 χ ( 1 w ) = 3 , χ ( 2 w ) = 2 \chi(2w) = 2 χ ( 2 w ) = 2 , χ ( 3 w ) = 1 \chi(3w) = 1 χ ( 3 w ) = 1 for all w ∉ D w \notin D w ∈ / D . Then from the vertex 1 the only O 1 O_1 O 1 -neighbour is 2, so the valency k 1 k_1 k 1 of the pair-orbit O 1 O_1 O 1 equals 1; from the vertex 3 every w ∉ D w \notin D w ∈ / D satisfies χ ( 3 w ) = 1 \chi(3w) = 1 χ ( 3 w ) = 1 , i.e. 3 w ∈ O 1 3w \in O_1 3 w ∈ O 1 , so k 1 = n − 3 k_1 = n - 3 k 1 = n − 3 . Since G G G is transitive on points, the valency k 1 k_1 k 1 is independent of the base point, giving 1 = n − 3 1 = n - 3 1 = n − 3 , i.e. n = 4 n = 4 n = 4 , contradicting n ≥ 8 n \geq8 n ≥ 8 . Hence some w w w has a colour vector different from ( 3 , 2 , 1 ) (3,2,1) ( 3 , 2 , 1 ) , and by the colour table of Appendix Appendix A at most one of its extension triples is rainbow, hence lies in T T T .
Proposition 2.9 (The four-point criterion for rank-3 groups). Let G ≤ Sym ( P ) G \leq\operatorname{Sym}(\mathcal{P}) G ≤ Sym ( P ) be transitive with exactly two orbits on pairs (σ 2 = 2 \sigma_2 = 2 σ 2 = 2 ). Let D D D be any triple and suppose there is w ∉ D w \notin D w ∈ / D such that ∣ G D ∪ { w } ∣ ≤ ∣ G D ∣ \lvert G_{D\cup\{w\}}\rvert\leq\lvert G_D\rvert ∣ G D ∪ { w } ∣ ≤ ∣ G D ∣ . Then D D D is not a ⋆ 3 \star_3 ⋆ 3 -witness. In particular, if for every triple D D D such a w w w exists, then G G G does not satisfy ⋆ 3 \star_3 ⋆ 3 .
Proof. The defining inequality of ⋆ 3 \star_3 ⋆ 3 for the quadruple Σ : = D ∪ { w } \Sigma:= D \cup\{w\} Σ := D ∪ { w } reads ∣ G Σ ∣ > ∣ G D ∣ \lvert G_\Sigma\rvert> \lvert G_D\rvert ∣ G Σ ∣ > ∣ G D ∣ ; the hypothesis ∣ G Σ ∣ ≤ ∣ G D ∣ \lvert G_\Sigma\rvert\leq\lvert G_D\rvert ∣ G Σ ∣ ≤ ∣ G D ∣ contradicts it, so D D D is not a witness. The assertion (used in an earlier draft and reported by the referee) that any w w w moved by G D G_D G D forces a longer quadruple orbit is false in general: for $G = J_2 on 100 points (rank 3; Section 3) the 1-edge triple { 1 , 5 , 2 } \{1,5,2\} { 1 , 5 , 2 } has ∣ G D ∣ = 8 |G_D|=8 ∣ G D ∣ = 8 , the point 3 is moved (∣ 3 G D ∣ = 4 |3^{G_D}|=4 ∣ 3 G D ∣ = 4 ) and yet ∣ G D ∪ { 3 } ∣ = 4 < 8 |G_{D\cup\{3\}}|=4<8 ∣ G D ∪ { 3 } ∣ = 4 < 8 , giving the inequality of the lemma rather than its converse. The correct obstruction is therefore the four-point inequality above, and it is this inequality that is verified in Propositions 2.11–2.15. □
Proposition 2.10 (The forced structure of a (C1)-witness). Let G ≤ Sym ( P ) G \leq\operatorname{Sym}(\mathcal{P}) G ≤ Sym ( P ) be transitive of rank 3, with the two orbits on pairs O 1 O_1 O 1 and O 2 O_2 O 2 , and let D = { x , y , z } D=\{x,y,z\} D = { x , y , z } be a (C1)-type triple: x y ∈ O 1 xy \in O_1 x y ∈ O 1 , x z , y z ∈ O 2 xz,yz \in O_2 x z , y z ∈ O 2 (equivalently, x y xy x y is the unique O 1 O_1 O 1 -pair inside D D D ). Put N ‾ ( t ) : = { p ∈ P : t p ∈ O 2 } \overline{N}(t):=\{p \in\mathcal{P}:tp \in O_2\} N ( t ) := { p ∈ P : tp ∈ O 2 } for the non-neighbourhood of t t t with respect to O 1 O_1 O 1 .
(i) For every w ∈ N ‾ ( x ) ∩ N ‾ ( y ) ∖ D w \in\overline{N}(x) \cap\overline{N}(y) \setminus D w ∈ N ( x ) ∩ N ( y ) ∖ D the image of G D ∪ { w } G_{D\cup\{w\}} G D ∪ { w } on D ∪ { w } D\cup\{w\} D ∪ { w } lies in V 4 = ⟨ ( 1 2 ) , ( 3 4 ) ⟩ V_4=\langle(1\,2),(3\,4)\rangle V 4 = ⟨( 1 2 ) , ( 3 4 )⟩ , and the orbit of D D D under G D ∪ { w } G_{D\cup\{w\}} G D ∪ { w } has exactly two elements.
(ii) If D D D is a ⋆ 3 \star_3 ⋆ 3 -witness, then every point of N ‾ ( x ) ∩ N ‾ ( y ) ∖ D \overline{N}(x) \cap\overline{N}(y) \setminus D N ( x ) ∩ N ( y ) ∖ D is fixed by G D G_D G D .
Proof. For (i): an element of G D ∪ { w } G_{D\cup\{w\}} G D ∪ { w } preserves the four pairs { x y , x w , y w } \{xy,xw,yw\} { x y , x w , y w } and z w zw z w as sets, hence preserves the O 1 O_1 O 1 -edges of the quadruple; these are x y xy x y alone when z w ∈ O 2 zw \in O_2 z w ∈ O 2 , and the matching { x y , z w } \{xy,zw\} { x y , z w } when z w ∈ O 1 zw \in O_1 z w ∈ O 1 . A permutation of { x , y , z , w } \{x,y,z,w\} { x , y , z , w } preserving x y xy x y alone lies in ⟨ ( 1 2 ) ⟩ \langle(1\,2)\rangle ⟨( 1 2 )⟩ , and one preserving the matching { x y , z w } \{xy,zw\} { x y , z w } lies in ⟨ ( 1 2 ) , ( 3 4 ) ⟩ = V 4 \langle(1\,2),(3\,4)\rangle=V_4 ⟨( 1 2 ) , ( 3 4 )⟩ = V 4 ; moreover ⟨ ( 1 2 ) ( 3 4 ) ⟩ \langle(1\,2)(3\,4)\rangle ⟨( 1 2 ) ( 3 4 )⟩ swaps z , w z,w z , w and maps D = { x , y , z } D=\{x,y,z\} D = { x , y , z } to { x , y , w } \{x,y,w\} { x , y , w } , so the orbit of D D D under the image is { D , { x , y , w } } \{D,\{x,y,w\}\} { D , { x , y , w }} , of size 2. For (ii): under ⋆ 3 \star_3 ⋆ 3 the quadruple Σ w : = D ∪ { w } \Sigma_w:=D\cup\{w\} Σ w := D ∪ { w } satisfies ∣ G Σ w ∣ > ∣ G D ∣ |G_{\Sigma_w}|>|G_D| ∣ G Σ w ∣ > ∣ G D ∣ for every w ∈ N ‾ ( x ) ∩ N ‾ ( y ) ∖ D w\in\overline{N}(x)\cap\overline{N}(y)\setminus D w ∈ N ( x ) ∩ N ( y ) ∖ D . From (i), ∣ G Σ w ∩ G D ∣ = ∣ G Σ w ∣ / 2 |G_{\Sigma_w}\cap G_D|=|G_{\Sigma_w}|/2 ∣ G Σ w ∩ G D ∣ = ∣ G Σ w ∣/2 , since the orbit D G Σ w D^{G_{\Sigma_w}} D G Σ w has two elements; but G Σ w ∩ G D = ( G D ) w G_{\Sigma_w}\cap G_D=(G_D)_w G Σ w ∩ G D = ( G D ) w (an element preserving D D D and Σ w \Sigma_w Σ w fixes w w w ). Hence
∣ w G D ∣ = ∣ G D ∣ ∣ ( G D ) w ∣ = 2 ∣ G D ∣ ∣ G Σ w ∣ < 2 , |w^{G_D}|=\frac{|G_D|}{|(G_D)_w|}=\frac{2|G_D|}{|G_{\Sigma_w}|}<2, ∣ w G D ∣ = ∣ ( G D ) w ∣ ∣ G D ∣ = ∣ G Σ w ∣ 2∣ G D ∣ < 2 ,
so w G D = { w } w^{G_D}=\{w\} w G D = { w } . Thus G D G_D G D fixes every such w w w . □
By (ii), the verification of the four-point inequality for the (C1)-configuration is reduced to a single task: for every (C1)-type triple D D D exhibit w ∈ N ‾ ( x ) ∩ N ‾ ( y ) ∖ D w\in\overline{N}(x)\cap\overline{N}(y)\setminus D w ∈ N ( x ) ∩ N ( y ) ∖ D with ∣ w G D ∣ ≥ 2 |w^{G_D}|\geq2 ∣ w G D ∣ ≥ 2 . In each family below such a w w w is produced explicitly.
Proposition 2.11 (The nonsingular-point families of Ω ( 2 m + 1 , 3 ) \Omega(2m+1,3) Ω ( 2 m + 1 , 3 ) fail ⋆ 3 \star_3 ⋆ 3 ). Let m ≥ 2 m\geq2 m ≥ 2 , let V V V be a nondegenerate orthogonal space of dimension $2m+1
over F 3 \mathbb{F}_3 F 3 , and let ξ ∈ { ± } \xi\in\{ \pm\} ξ ∈ { ± } . Let G G G be Ω ( V ) \Omega(V) Ω ( V ) acting on the orbit E ξ E_\xi E ξ of nonsingular points of type ξ \xi ξ . Then G G G does not satisfy the Siemons–Wagner property ⋆ 3 \star_3 ⋆ 3 .
Proof. By Tong-Viet [25 Lemma 5.1] the action is primitive of rank 3, with parameters v = ∣ E ξ ∣ = 1 2 3 m ( 3 m + ξ ) v = |E_\xi| = \frac{1}{2}3^m(3^m+\xi) v = ∣ E ξ ∣ = 2 1 3 m ( 3 m + ξ ) , k = 1 2 3 m − 1 ( 3 m − ξ ) k = \frac{1}{2}3^{m-1}(3^m-\xi) k = 2 1 3 m − 1 ( 3 m − ξ ) , l = ( 3 m − ξ ) ( 3 m − 1 + ξ ) l = (3^m-\xi)(3^{m-1}+\xi) l = ( 3 m − ξ ) ( 3 m − 1 + ξ ) and λ = μ = 1 2 3 m − 1 ( 3 m − 1 − ξ ) \lambda= \mu= \frac{1}{2}3^{m-1}(3^{m-1}-\xi) λ = μ = 2 1 3 m − 1 ( 3 m − 1 − ξ ) ; in particular G G G has exactly two orbits on pairs, namely the orthogonal pairs ( O 1 ) (O_1) ( O 1 ) and the nonorthogonal pairs ( O 2 ) (O_2) ( O 2 ) . The triangular subcase (C2) of Lemma 2.5 is excluded combinatorially in the proof of that lemma, so it remains to exclude the (C1)-type triple D = { x , y , z } D = \{x,y,z\} D = { x , y , z } with x y ∈ O 1 xy \in O_1 x y ∈ O 1 and x z , y z ∈ O 2 xz,yz \in O_2 x z , y z ∈ O 2 . Put W = ⟨ x , y , z ⟩ W = \langle x,y,z\rangle W = ⟨ x , y , z ⟩ ; W W W is nondegenerate of dimension d ≤ 3 d \le3 d ≤ 3 (x ⊥ y x \perp y x ⊥ y with x , y x,y x , y nonsingular makes ⟨ x , y ⟩ \langle x,y\rangle ⟨ x , y ⟩ nondegenerate, and z ∉ x ⊥ ∪ y ⊥ z \notin x^\perp\cup y^\perp z ∈ / x ⊥ ∪ y ⊥ adds a nondegenerate direction). By Lemma 2.10(ii) it suffices to exhibit w ∈ N ‾ ( x ) ∩ N ‾ ( y ) ∖ D w \in\overline{N}(x) \cap\overline{N}(y) \setminus D w ∈ N ( x ) ∩ N ( y ) ∖ D with ∣ w G D ∣ ≥ 2 |w^{G_D}| \ge2 ∣ w G D ∣ ≥ 2 , where N ‾ ( t ) = E ξ ∖ ( t ⊥ ∪ { t } ) \overline{N}(t) = E_\xi\setminus(t^\perp\cup\{t\}) N ( t ) = E ξ ∖ ( t ⊥ ∪ { t }) . If m = 2 m = 2 m = 2 , then v = 1 2 3 2 ( 9 + ξ ) ≤ 45 ≤ 50 v = \frac{1}{2}3^2(9+\xi) \le45 \le50 v = 2 1 3 2 ( 9 + ξ ) ≤ 45 ≤ 50 , and the claim is part of the census of Section 3, Table 2 (v = 36 , 45 v = 36,45 v = 36 , 45 have no rank-3 group with ⋆ 3 \star_3 ⋆ 3 ). Assume m ≥ 3 m \ge3 m ≥ 3 . Since x , y x,y x , y form an O 1 O_1 O 1 -pair and the action is a strongly regular graph with parameters ( v , k , λ , μ ) (v,k,\lambda,\mu) ( v , k , λ , μ ) , the number of common non-neighbours of x , y x,y x , y is ∣ N ‾ ( x ) ∩ N ‾ ( y ) ∣ = v − 2 − 2 k + λ = 2 ( 3 2 m − 2 + 3 m − 1 ξ − 1 ) > 27 ≥ ∣ W ∩ E ξ ∣ |\overline{N}(x) \cap\overline{N}(y)| = v - 2 - 2k + \lambda= 2(3^{2m-2} + 3^{m-1}\xi- 1) > 27 \ge|W \cap E_\xi| ∣ N ( x ) ∩ N ( y ) ∣ = v − 2 − 2 k + λ = 2 ( 3 2 m − 2 + 3 m − 1 ξ − 1 ) > 27 ≥ ∣ W ∩ E ξ ∣ (the last bound uses ∣ W ∣ = 3 d ≤ 27 |W| = 3^d \le27 ∣ W ∣ = 3 d ≤ 27 ; for m = 3 m = 3 m = 3 , ξ = − 1 \xi= -1 ξ = − 1 the left hand side is 2 ( 81 − 9 − 1 ) = 142 2(81 - 9 - 1) = 142 2 ( 81 − 9 − 1 ) = 142 ). Hence there is w ∈ N ‾ ( x ) ∩ N ‾ ( y ) ∖ W w \in\overline{N}(x) \cap\overline{N}(y) \setminus W w ∈ N ( x ) ∩ N ( y ) ∖ W , and w ∉ W w \notin W w ∈ / W gives w ≠ x , y , z w \ne x,y,z w = x , y , z . Write w = u + u ′ w = u+u' w = u + u ′ with u ∈ W u \in W u ∈ W and 0 ≠ u ′ ∈ W ⊥ 0 \ne u' \in W^\perp 0 = u ′ ∈ W ⊥ . The setwise stabilizer G D G_D G D preserves W W W and W ⊥ W^\perp W ⊥ , and it contains the natural copy of Ω ( W ⊥ ) \Omega(W^\perp) Ω ( W ⊥ ) acting trivially on W W W (well defined since W W W is nondegenerate: every element of Ω ( W ⊥ ) \Omega(W^\perp) Ω ( W ⊥ ) extends to Ω ( V ) \Omega(V) Ω ( V ) by the identity on W W W ). Since Ω ( W ⊥ ) \Omega(W^\perp) Ω ( W ⊥ ) is transitive on the nonsingular points of each type in W ⊥ W^\perp W ⊥ and dim W ⊥ = 2 m + 1 − d ≥ 4 \dim W^\perp= 2m+1-d \ge4 dim W ⊥ = 2 m + 1 − d ≥ 4 , the orbit w G D w^{G_D} w G D contains u + ( u ′ ) Ω ( W ⊥ ) u + (u')^{\Omega(W^\perp)} u + ( u ′ ) Ω ( W ⊥ ) and has length at least 2. By Lemma 2.10(ii) the triple D D D is not a ⋆ 3 \star_3 ⋆ 3 -witness. Hence no triple can be a witness and G G G fails ⋆ 3 \star_3 ⋆ 3 .
Table 2. The ⋆ 3 \star_3 ⋆ 3 census for primitive groups of degree 31 ≤ v ≤ 50 31 \le v \le50 31 ≤ v ≤ 50 : the number of primitive groups, of rank-3 and of 2-homogeneous ones, and how many of each satisfy the Siemons–Wagner property ⋆ 3 \star_3 ⋆ 3 (none do; the rank-3 filter is justified by Lemma 2.5, Step 4 together with the rank-3 classification [14 , 16 , 17 ] ).
Proposition 2.12 (The line actions of PSL ( n , q ) \operatorname{PSL}(n,q) PSL ( n , q ) fail ⋆ 3 \star_3 ⋆ 3 ). Let n ≥ 4 n \ge4 n ≥ 4 , let G G G be PSL ( n , q ) \operatorname{PSL}(n,q) PSL ( n , q ) or one of its overgroups PGL ( n , q ) \operatorname{PGL}(n,q) PGL ( n , q ) , PΓL ( n , q ) \operatorname{P\Gamma L}(n,q) PΓL ( n , q ) , and let G G G act on the set L \mathcal{L} L of lines of PG ( n − 1 , q ) \operatorname{PG}(n-1,q) PG ( n − 1 , q ) . If ∣ L ∣ ≥ 8 |\mathcal{L}| \ge8 ∣ L ∣ ≥ 8 , then G G G does not satisfy ⋆ 3 \star_3 ⋆ 3 .
Proof. By Kantor–Liebler [14 Thm. 1.2] the action is primitive of rank 3: two lines are adjacent exactly when they meet, and σ 2 = 2 \sigma_2 = 2 σ 2 = 2 . The triangular subcase (C2) is excluded combinatorially in Lemma 2.5, so let D = { L 1 , L 2 , L 3 } D = \{L_1,L_2,L_3\} D = { L 1 , L 2 , L 3 } be a (C1)-type triple: L 1 L_1 L 1 meets L 2 L_2 L 2 ( O 1 ) (O_1) ( O 1 ) while L 1 , L 2 L_1,L_2 L 1 , L 2 are both skew to L 3 L_3 L 3 ( O 2 ) (O_2) ( O 2 ) . Put p = L 1 ∩ L 2 p = L_1 \cap L_2 p = L 1 ∩ L 2 and π = ⟨ L 1 , L 2 ⟩ \pi= \langle L_1,L_2\rangle π = ⟨ L 1 , L 2 ⟩ ; the setwise stabilizer G D G_D G D preserves p p p and π \pi π (preserving { L 1 , L 2 } \{L_1,L_2\} { L 1 , L 2 } as a set) and preserves L 3 L_3 L 3 . We produce $w \in N ‾ ( L 1 ) ∩ N ‾ ( L 2 ) ∖ D \overline{N}(L_1) \cap\overline{N}(L_2) \setminus\mathcal{D} N ( L 1 ) ∩ N ( L 2 ) ∖ D with ∣ w G D ∣ ≥ 2 |w^{G_D}| \ge2 ∣ w G D ∣ ≥ 2 , where N ‾ ( L ) = { M ∈ L : M ∩ L = ∅ } \overline{N}(L)=\{M\in\mathcal{L}:M\cap L=\varnothing\} N ( L ) = { M ∈ L : M ∩ L = ∅ } . First note that G D G_D G D moves some point of π ∖ ( L 1 ∪ L 2 ) \pi\setminus(L_1\cup L_2) π ∖ ( L 1 ∪ L 2 ) : if the image of G D G_D G D on π \pi π is nontrivial, any image element moves such a point (an element fixing every point outside L 1 ∪ L 2 L_1\cup L_2 L 1 ∪ L 2 together with L 1 , L 2 L_1,L_2 L 1 , L 2 as sets fixes three noncollinear points of π \pi π and is the identity); if the image is trivial, G D G_D G D preserves L 1 , L 2 L_1,L_2 L 1 , L 2 individually, and the same argument applies to the nontrivial part, or G D G_D G D acts trivially on π \pi π , in which case the faithful action of PGL ( n , q ) \operatorname{PGL}(n,q) PGL ( n , q ) on lines [14 § 2] shows that either some element moves a line q q q outside π \pi π or G D = 1 G_D=1 G D = 1 . In the first two cases pick p ′ ∈ π ∖ ( L 1 ∪ L 2 ) p'\in\pi\setminus(L_1\cup L_2) p ′ ∈ π ∖ ( L 1 ∪ L 2 ) moved by g ∈ G D g\in G_D g ∈ G D and a line q q q outside π \pi π ; then w = p ′ ∪ q w=p'\cup q w = p ′ ∪ q (the line through p ′ p' p ′ and q q q ) meets π \pi π only in p ′ p' p ′ , so w w w is skew to L 1 L_1 L 1 and L 2 L_2 L 2 , and g ( w ) = g ( p ′ ) ∪ g ( q ) ≠ w g(w)=g(p')\cup g(q)\ne w g ( w ) = g ( p ′ ) ∪ g ( q ) = w . In the third case pick p ′ ∈ π ∖ ( L 1 ∪ L 2 ) p'\in\pi\setminus(L_1\cup L_2) p ′ ∈ π ∖ ( L 1 ∪ L 2 ) and q ∉ π q\notin\pi q ∈ / π moved by G D G_D G D , so w = p ′ ∪ q w=p'\cup q w = p ′ ∪ q is moved; if G D = 1 G_D=1 G D = 1 , then a quadruple Σ = D ∪ { w } \Sigma=D\cup\{w\} Σ = D ∪ { w } with w w w in general position has ∣ G Σ ∣ = 1 |G_\Sigma|=1 ∣ G Σ ∣ = 1 : the image in S 4 S_4 S 4 preserves the single O 1 O_1 O 1 -edge { L 1 , L 2 } \{L_1,L_2\} { L 1 , L 2 } , and the transposition ( L 1 L 2 ) (L_1\ L_2) ( L 1 L 2 ) does not lift (an element swapping L 1 , L 2 L_1,L_2 L 1 , L 2 and fixing L 3 , w L_3,w L 3 , w would preserve their intersection points with π \pi π , forcing a symmetry that fails in general position). Thus ∣ G Σ ∣ ≤ ∣ G D ∣ |G_\Sigma|\le|G_D| ∣ G Σ ∣ ≤ ∣ G D ∣ in every case, and by Lemma 2.10 (for the moved w w w ) or by the four-point criterion (for G D = 1 G_D=1 G D = 1 ) the triple D D D is not a ⋆ 3 \star_3 ⋆ 3 -witness. This holds for every ( C 1 ) (C1) ( C 1 ) -type triple, so G G G fails ⋆ 3 \star_3 ⋆ 3 .
Proposition 2.13 (The singular- and totally-isotropic-point actions of the classical groups fail ⋆ 3 \star_3 ⋆ 3 ). Let ( V , h ) (V,h) ( V , h ) be a symplectic , orthogonal or hermitian space over a finite field, and let G G G be the isometry group Sp ( V ) \operatorname{Sp}(V) Sp ( V ) , Ω ( V ) \Omega(V) Ω ( V ) or SU ( V ) \operatorname{SU}(V) SU ( V ) (or an overgroup in Γ Sp ( V ) \Gamma\operatorname{Sp}(V) Γ Sp ( V ) , Γ O ( V ) \Gamma\operatorname{O}(V) Γ O ( V ) , Γ U ( V ) \Gamma\operatorname{U}(V) Γ U ( V ) ) acting on the set of singular points (isotropic 1-spaces) of V V V . In each of the rank-3 cases of Kantor–Liebler [14 Thm. 1.1(i)] with degree at least 8 and at most two orbits on pairs, G G G does not satisfy ⋆ 3 \star_3 ⋆ 3 .
Proof. The rank-3 singular-point cases are Sp ( 2 m , q ) \operatorname{Sp}(2m,q) Sp ( 2 m , q ) on PG ( 2 m − 1 , q ) \operatorname{PG}(2m-1,q) PG ( 2 m − 1 , q ) (every 1-space is isotropic, and two points are adjacent exactly when orthogonal), Ω ± ( 2 m , q ) \Omega^{\pm}(2m,q) Ω ± ( 2 m , q ) and Ω ( 2 m − 1 , q ) \Omega(2m-1,q) Ω ( 2 m − 1 , q ) on the isotropic points, and SU ( m , q ) \operatorname{SU}(m,q) SU ( m , q ) , m ≥ 3 m\ge3 m ≥ 3 , on the isotropic points (m = 2 m=2 m = 2 is 2-transitive, hence 2-homogeneous and irrelevant; the small degrees are covered by Table 3 , Table 2 ). The triangular subcase (C2) is excluded combinatorially in Lemma 2.5, so let D = { x , y , z } D=\{x,y,z\} D = { x , y , z } be a (C1)-type triple: x y ∈ O 1 xy\in O_1 x y ∈ O 1 (orthogonal) and x z , y z ∈ O 2 xz,yz\in O_2 x z , y z ∈ O 2 (nonorthogonal). Put W = ⟨ x , y , z ⟩ W=\langle x,y,z\rangle W = ⟨ x , y , z ⟩ ; G D G_D G D preserves W W W and W ⊥ W^\perp W ⊥ . The general construction. Suppose W ⊥ W^\perp W ⊥ contains a nonzero isotropic vector u ′ u' u ′ with the property that G D G_D G D moves it. Then w : = z + u ′ w:=z+u' w := z + u ′ satisfies ( w , x ) = ( z , x ) ≠ 0 (w,x)=(z,x)\ne0 ( w , x ) = ( z , x ) = 0 and ( w , y ) = ( z , y ) ≠ 0 (w,y)=(z,y)\ne0 ( w , y ) = ( z , y ) = 0 , so w ∈ N ‾ ( x ) ∩ N ‾ ( y ) ∖ D w\in\overline{N}(x)\cap\overline{N}(y)\setminus D w ∈ N ( x ) ∩ N ( y ) ∖ D ; moreover w w w is isotropic ( ( w , w ) = ( z , z ) + ( u ′ , u ′ ) = 0 ) ((w,w)=(z,z)+(u',u')=0) (( w , w ) = ( z , z ) + ( u ′ , u ′ ) = 0 ) and moved by G D G_D G D , and by Lemma 2.10(ii) D \mathcal{D} D is not a witness. This applies whenever the isometry group of W ⊥ W^\perp W ⊥ has a nontrivial orbit on the isotropic points of W ⊥ W^\perp W ⊥ , in particular when W ⊥ W^\perp W ⊥ is nondegenerate of dimension at least 2 and contains isotropic points: the symplectic case always (every 1-space is isotropic); the orthogonal case unless W ⊥ W^\perp W ⊥ is of split type over odd q q q in dimension 2, where one replaces u ′ u' u ′ by a singular point in the other Weyl orbit (which is still moved by G D G_D G D ); and the hermitian case for m ≥ 5 m \ge5 m ≥ 5 (dim W ⊥ = m − d ≥ m − 3 ≥ 2 \dim W^\perp=m-d \ge m-3 \ge2 dim W ⊥ = m − d ≥ m − 3 ≥ 2 ). For the degenerate W W W one uses the isotropic points of W ⊥ W^\perp W ⊥ contained in the radical, which exist and are moved by the isometry group of W ⊥ W^\perp W ⊥ . This covers Sp ( 2 m , q ) \operatorname{Sp}(2m,q) Sp ( 2 m , q ) with m ≥ 3 m \ge3 m ≥ 3 , all orthogonal cases, and SU ( m , q ) \operatorname{SU}(m,q) SU ( m , q ) with m ≥ 5 m \ge5 m ≥ 5 . The low-dimensional cases Sp ( 4 , q ) \operatorname{Sp}(4,q) Sp ( 4 , q ) , q ≥ 4 q \ge4 q ≥ 4 , and SU ( 4 , q ) \operatorname{SU}(4,q) SU ( 4 , q ) , q ≥ 3 q \ge3 q ≥ 3 . (Sp ( 4 , 2 ) \operatorname{Sp}(4,2) Sp ( 4 , 2 ) , Sp ( 4 , 3 ) \operatorname{Sp}(4,3) Sp ( 4 , 3 ) and SU ( 4 , 2 ) \operatorname{SU}(4,2) SU ( 4 , 2 ) have degrees 15, 40, 40 ≤ 50 \le50 ≤ 50 and lie in the census). By (ii) of Lemma 2.10 a witness forces G D G_D G D to fix every point of N ‾ ( x ) ∩ N ‾ ( y ) \overline{N}(x) \cap\overline{N}(y) N ( x ) ∩ N ( y ) . If the image of G D G_D G D on W W W is nontrivial, an image element g g g either moves z z z — then w : = z + u ′ w := z+u' w := z + u ′ with 0 ≠ u ′ ∈ W ⊥ 0 \ne u' \in W^\perp 0 = u ′ ∈ W ⊥ isotropic is moved (contradiction) — or swaps x , y x,y x , y and fixes z z z ; in the latter case the point w ′ : = z + α x + β y + u ′ w' := z+\alpha x+\beta y+u' w ′ := z + α x + β y + u ′ with α ≠ β \alpha\ne\beta α = β and ( α , β ) (\alpha,\beta) ( α , β ) chosen so that ( w ′ , x ) ( w ′ , y ) ≠ 0 (w',x)(w',y) \ne0 ( w ′ , x ) ( w ′ , y ) = 0 is moved by g g g (a contradiction; such α , β \alpha,\beta α , β exist because q ≥ 4 q \ge4 q ≥ 4 resp. q ≥ 3 q \ge3 q ≥ 3 gives at least three values in the field, and the isotropy of w ′ w' w ′ is automatic in the symplectic case and free for the hermitian case since ( u ′ , u ′ ) = 0 (u',u')=0 ( u ′ , u ′ ) = 0 ). If the image is trivial and W W W is 3-dimensional nonsingular, then W ⊥ W^\perp W ⊥ is 1-dimensional and G D = 1 G_D=1 G D = 1 ; choosing w ∈ N ‾ ( x ) ∩ N ‾ ( y ) w \in\overline{N}(x) \cap\overline{N}(y) w ∈ N ( x ) ∩ N ( y ) in general position (z , w z,w z , w linearly independent over W W W ) the quadruple Σ = D ∪ { w } \Sigma=\mathcal{D}\cup\{w\} Σ = D ∪ { w } has ∣ G Σ ∣ = 1 |G_\Sigma|=1 ∣ G Σ ∣ = 1 : its image in S 4 S_4 S 4 preserves the single O 1 O_1 O 1 -edge { x , y } \{x,y\} { x , y } , and the transposition ( x y ) (x\,y) ( x y ) does not lift, since an isometry swapping x , y x,y x , y and fixing z , w z,w z , w would force ( z , x ) = ( z , y ) (z,x)=(z,y) ( z , x ) = ( z , y ) and ( w , x ) = ( w , y ) (w,x)=(w,y) ( w , x ) = ( w , y ) , which fails in general position. Thus ∣ G Σ ∣ = 1 = ∣ G D ∣ |G_\Sigma|=1=|G_D| ∣ G Σ ∣ = 1 = ∣ G D ∣ and D \mathcal{D} D is not a witness. If W W W is 2-dimensional or degenerate, the general construction above applies (W ⊥ W^\perp W ⊥ has dimension at least 2 and its isometry group moves isotropic points). Hence in all cases no triple is a ⋆ 3 \star_3 ⋆ 3 -witness and G G G fails ⋆ 3 \star_3 ⋆ 3 .
Table 3. The ⋆ 3 \star_{3} ⋆ 3 census for primitive groups of degree 8 ≤ v ≤ 30 8 \le v \le30 8 ≤ v ≤ 30 : the number of primitive groups, of non-2-homogeneous and 2-homogeneous ones, and how many of each satisfy the Siemons–Wagner property ⋆ 3 \star_{3} ⋆ 3 . The two groups with ⋆ 3 \star_{3} ⋆ 3 are PSL( 2 , 7 ) (2, 7) ( 2 , 7 ) and PGL( 2 , 7 ) (2, 7) ( 2 , 7 ) , both of degree 8.
Proposition 2.14 (The maximal totally isotropic subspace actions fail ⋆ 3 \star_3 ⋆ 3 ). Let ( V , h ) (V,h) ( V , h ) be one of the spaces of Kantor–Liebler [14 Thm. 1.1(ii)] : Sp ( 4 , q ) \operatorname{Sp}(4,q) Sp ( 4 , q ) , SU ( 4 , q ) \operatorname{SU}(4,q) SU ( 4 , q ) , SU ( 5 , q ) \operatorname{SU}(5,q) SU ( 5 , q ) , Ω − ( 6 , q ) \Omega^-(6,q) Ω − ( 6 , q ) , Ω + ( 8 , q ) \Omega^+(8,q) Ω + ( 8 , q ) or Ω + ( 10 , q ) \Omega^+(10,q) Ω + ( 10 , q ) acting on the set M \mathcal{M} M of maximal totally isotropic subspaces. Then G G G does not satisfy ⋆ 3 \star_3 ⋆ 3 .
Proof. Two maximal totally isotropic subspaces are adjacent exactly when they meet in dimension m − 1 m-1 m − 1 (m m m the Witt index), so G G G is of rank 3. The triangular subcase (C2) is excluded combinatorially in Lemma 2.5, so let D = { W 1 , W 2 , W 3 } \mathcal{D}=\{W_1,W_2,W_3\} D = { W 1 , W 2 , W 3 } be a ( C 1 ) (C1) ( C 1 ) -type triple: dim ( W 1 ∩ W 2 ) = m − 1 \dim(W_1\cap W_2)=m-1 dim ( W 1 ∩ W 2 ) = m − 1 ( O 1 ) (O_1) ( O 1 ) and dim ( W 1 ∩ W 3 ) , dim ( W 2 ∩ W 3 ) ≤ m − 2 \dim(W_1\cap W_3),\dim(W_2\cap W_3)\le m-2 dim ( W 1 ∩ W 3 ) , dim ( W 2 ∩ W 3 ) ≤ m − 2 ( O 2 ) (O_2) ( O 2 ) . We produce a maximal totally isotropic subspace w w w with w ∈ N ‾ ( W 1 ) ∩ N ‾ ( W 2 ) ∖ D w\in\overline{N}(W_1)\cap\overline{N}(W_2)\setminus\mathcal{D} w ∈ N ( W 1 ) ∩ N ( W 2 ) ∖ D (i.e. dim ( w ∩ W 1 ) , dim ( w ∩ W 2 ) ≤ m − 2 \dim(w\cap W_1),\dim(w\cap W_2)\le m-2 dim ( w ∩ W 1 ) , dim ( w ∩ W 2 ) ≤ m − 2 ) and ∣ G D ∣ ≥ 2 |G_{\mathcal{D}}|\ge2 ∣ G D ∣ ≥ 2 . If the image of G D G_{\mathcal{D}} G D on { W 1 , W 2 , W 3 } \{W_1,W_2,W_3\} { W 1 , W 2 , W 3 } is nontrivial, take an image element g g g ; since the image lies in C 2 C_2 C 2 , either g g g swaps W 1 , W 2 W_1,W_2 W 1 , W 2 — then any maximal totally isotropic w w w with ( dim ( w ∩ W 1 ) , dim ( w ∩ W 2 ) ) = ( m − 2 , m − 3 ) (\dim(w\cap W_1),\dim(w\cap W_2))=(m-2,m-3) ( dim ( w ∩ W 1 ) , dim ( w ∩ W 2 )) = ( m − 2 , m − 3 ) is moved by g g g , because g ( w ) g(w) g ( w ) meets W 1 W_1 W 1 in dimension m − 3 m-3 m − 3 — or g g g fixes each W i W_i W i , and g g g acts nontrivially on the intersection lattice and moves some maximal totally isotropic w w w with prescribed intersection dimensions ( m − 2 , m − 3 ) (m-2,m-3) ( m − 2 , m − 3 ) (such w w w exist in abundance by the standard counting of totally isotropic subspaces meeting a given one in a prescribed dimension; the smallest degrees are in the census). If the image is trivial, G D G_{\mathcal{D}} G D preserves each W i W_i W i ; the kernel moves some ( m − 2 ) (m-2) ( m − 2 ) -dimensional totally isotropic subspace U ≤ W 1 U\le W_1 U ≤ W 1 (it preserves the two hyperplanes W 1 ∩ W 2 W_1\cap W_2 W 1 ∩ W 2 and W 1 ∩ W 3 W_1\cap W_3 W 1 ∩ W 3 of W 1 W_1 W 1 , and a nontrivial element of it moves a suitable ( m − 2 ) (m-2) ( m − 2 ) -subspace), and extending U U U to a maximal totally isotropic w w w in general position so that dim ( w ∩ W 2 ) ≤ m − 2 \dim(w\cap W_2)\le m-2 dim ( w ∩ W 2 ) ≤ m − 2 gives a moved w w w . If G D = 1 G_{\mathcal{D}}=1 G D = 1 , then a quadruple Σ = D ∪ { w } \Sigma=\mathcal{D}\cup\{w\} Σ = D ∪ { w } with w w w in general position has ∣ G Σ ∣ = 1 : i t s i m a g e i n |G_\Sigma|=1: its image in ∣ G Σ ∣ = 1 : i t s ima g e in S_4$ preserves the single O 1 O_1 O 1 -edge { W 1 , W 2 } \{W_1,W_2\} { W 1 , W 2 } , and the transposition ( W 1 W 2 ) (W_1\,W_2) ( W 1 W 2 ) does not lift, since an element swapping W 1 , W 2 W_1,W_2 W 1 , W 2 and fixing W 3 , w W_3,w W 3 , w would preserve the intersection dimensions ( dim ( ⋅ ∩ W 3 ) , dim ( ⋅ ∩ w ) ) (\dim(\cdot\cap W_3),\dim(\cdot\cap w)) ( dim ( ⋅ ∩ W 3 ) , dim ( ⋅ ∩ w )) symmetrically, which fails in general position. Thus ∣ G Σ ∣ ≤ ∣ G D ∣ |G_\Sigma|\le|G_{\mathcal{D}}| ∣ G Σ ∣ ≤ ∣ G D ∣ in every case, and by Lemma 2.10 (for the moved w w w ) or by the four-point criterion (for G D = 1 G_{\mathcal{D}}=1 G D = 1 ) the triple D \mathcal{D} D is not a ⋆ 3 \star_3 ⋆ 3 -witness. This holds for every ( C 1 ) (C1) ( C 1 ) -type triple, so G G G fails ⋆ 3 \star_3 ⋆ 3 .
Proposition 2.15. The sporadic rank-3 representations fail ⋆ 3 \star_3 ⋆ 3 . Let G G G be a sporadic simple group in a primitive rank-3 representation (the sporadic representatives of the rank-3 classification [14 , 16 , 17 ] ; the rank is checked against the permutation tables of the ATLAS [6 ] ). Then G G G does not satisfy ⋆ 3 \star_3 ⋆ 3 .
Proof. The sporadic primitive rank-3 representations of degree at least 8 are, with the suborbit data of the ATLAS [6 ] : M 22 M_{22} M 22 on 77 points (rank 3; suborbits 1, 16, 60; point stabilizer 2 3 : A 6 2^3\!:\!A_6 2 3 : A 6 ), M 23 M_{23} M 23 on 253a points (rank 3; 1, 42, 210; point stabilizer L 3 ( 4 ) : 2 2 L_3(4)\!:\!2^2 L 3 ( 4 ) : 2 2 ), M 24 M_{24} M 24 on 276 points (rank 3; 1, 44, 231; point stabilizer M 22 : 2 M_{22}\!:\!2 M 22 : 2 ), J 2 J_2 J 2 on 100 points (rank 3; 1, 36, 63; point stabilizer U 3 ( 3 ) U_3(3) U 3 ( 3 ) ), F i 22 \mathrm{Fi}_{22} Fi 22 on 3510 points (rank 3; 1, 693, 2816), and the sporadic rank-3 graphs of the Higman–Sims, Hall–Janko, U 4 ( 3 ) U_4(3) U 4 ( 3 ) , McLaughlin, Suzuki and Conway families as listed in [22 ] (Table 2); the primitive rank-2 representations (C o 3 \mathrm{Co}_3 Co 3 on 276 points, suborbits 1, 275; and the 2-transitive ones) are 2-homogeneous and irrelevant here, and the rank-5 representations of H e He H e (on 2058) and J 1 J_1 J 1 (on 266) and the rank-4 representation M 24 M_{24} M 24 on 759 points are not rank 3. The triangular subcase (C2) of Lemma 2.5 is excluded combinatorially in the proof of that lemma, so only the (C1)-type witness remains. For J 2 J_2 J 2 on 100 points the exclusion is verified in GAP with the two standard generators of the Atlas (the script _R_J2_4pt.g, Section 3): for the 1-edge triple D = { 1 , 5 , 2 } \mathcal{D}=\{1,5,2\} D = { 1 , 5 , 2 } (12 , 15 ∈ O 1 12,15\in O_1 12 , 15 ∈ O 1 , 25 ∈ O 2 25\in O_2 25 ∈ O 2 ) one has ∣ G D ∣ = 8 |G_{\mathcal{D}}|=8 ∣ G D ∣ = 8 and the orbits of G D G_{\mathcal{D}} G D on the remaining 97 points have lengths 4, 8, 8, 8, 4, 8, 8, 8, 4, 8, 8, 2, 8, 8, 2, 1; for the point w = 3 w=3 w = 3 (∣ G D 3 ∣ = 4 |G_{\mathcal{D}}^3|=4 ∣ G D 3 ∣ = 4 ) the quadruple Σ = D ∪ { 3 } \Sigma=\mathcal{D}\cup\{3\} Σ = D ∪ { 3 } satisfies ∣ G Σ ∣ = 4 ≤ 8 = ∣ G D ∣ |G_{\Sigma}|=4\leq8=|G_{\mathcal{D}}| ∣ G Σ ∣ = 4 ≤ 8 = ∣ G D ∣ , so D \mathcal{D} D is not a witness. The same computation gives ∣ G D ∣ = 72 |G_{\mathcal{D}}|=72 ∣ G D ∣ = 72 for a triangular triple and ∣ G D ∣ = 8 |G_{\mathcal{D}}|=8 ∣ G D ∣ = 8 for an independent triple, with the four-point inequality verified for representatives of the surviving pair types; hence J 2 J_2 J 2 fails ⋆ 3 \star_3 ⋆ 3 . For the remaining sporadic rank-3 representations the suborbit data quoted from the ATLAS settle the rank, and the strongly regular parameters give an independent certificate: each of the graphs is SRG ( v , k , λ , μ ) \operatorname{SRG}(v,k,\lambda,\mu) SRG ( v , k , λ , μ ) with parameters recorded in [22 Table 2] , and the number v − 2 − 2 k + λ v-2-2k+\lambda v − 2 − 2 k + λ of common non-neighbours of an adjacent pair is at least 20 in every row, so a common non-neighbour w w w as in the J 2 J_2 J 2 computation always exists. The four-point inequality on such w w w is verified in GAP with the Atlas standard generators [6 ] (scripts _R_sporadic2.g, _R_sporadic4.g, _R_sporadic5.g; Section 3): with T T T a triple of smallest stabilizer one has ∣ G T ∣ = 16 |G_T|=16 ∣ G T ∣ = 16 , ∣ G T ∪ { w } ∣ = 2 |G_{T\cup\{w\}}|=2 ∣ G T ∪ { w } ∣ = 2 (M 22 M_{22} M 22 , degree 77), ∣ G T ∣ = 8 |G_T|=8 ∣ G T ∣ = 8 , ∣ G T ∪ { w } ∣ = 2 |G_{T\cup\{w\}}|=2 ∣ G T ∪ { w } ∣ = 2 (M 23 M_{23} M 23 , 253), ∣ G T ∣ = 144 |G_T|=144 ∣ G T ∣ = 144 , ∣ G T ∪ { w } ∣ = 12 |G_{T\cup\{w\}}|=12 ∣ G T ∪ { w } ∣ = 12 (M 24 M_{24} M 24 , 276), ∣ G T ∣ = 576 |G_T|=576 ∣ G T ∣ = 576 , ∣ G T ∪ { w } ∣ = 72 |G_{T\cup\{w\}}|=72 ∣ G T ∪ { w } ∣ = 72 (Higman–Sims, 100), ∣ G T ∣ = 720 |G_T|=720 ∣ G T ∣ = 720 , ∣ G T ∪ { w } ∣ = 20 |G_{T\cup\{w\}}|=20 ∣ G T ∪ { w } ∣ = 20 (McL, 275), ∣ G T ∣ = 1080 |G_T|=1080 ∣ G T ∣ = 1080 , ∣ G T ∪ { w } ∣ = 360 |G_{T\cup\{w\}}|=360 ∣ G T ∪ { w } ∣ = 360 (Suz, 1782), ∣ G T ∣ = 23040 |G_T|=23040 ∣ G T ∣ = 23040 , ∣ G T ∪ { w } ∣ = 3072 |G_{T\cup\{w\}}|=3072 ∣ G T ∪ { w } ∣ = 3072 (F i 22 Fi_{22} F i 22 , 3510), ∣ G T ∣ = 46080 |G_T|=46080 ∣ G T ∣ = 46080 , ∣ G T ∪ { w } ∣ ∈ { 512 , 1440 } |G_{T\cup\{w\}}|\in\{512,1440\} ∣ G T ∪ { w } ∣ ∈ { 512 , 1440 } (C o 2 Co_2 C o 2 , 2300), so in every row the four-point inequality holds and the triple is not a witness. The H e He H e representation on 2058 points has σ 2 = 4 \sigma_2=4 σ 2 = 4 (rank 5) and is outside the rank-3 case; the classical unitary row U 4 ( 3 ) U_4(3) U 4 ( 3 ) on 112 points has suborbits 1, 45, 66 and is covered by the same four-point computation (record listed with the scripts). Consequently every sporadic rank-3 representation fails ⋆ 3 \star_3 ⋆ 3 .
Lemma 2.16 (A witness for ⋆ 3 \star_3 ⋆ 3 ). Let G ≤ Sym ( P ) G\leq\operatorname{Sym}(\mathcal{P}) G ≤ Sym ( P ) be primitive with ∣ P ∣ ≥ 8 |\mathcal{P}|\geq8 ∣ P ∣ ≥ 8 , not 3-homogeneous, and let O \mathcal{O} O be a representative of a maximal triple orbit. If there is a four-subset Σ ⊃ O \Sigma\supset\mathcal{O} Σ ⊃ O with ∣ Σ ∣ = 4 |\Sigma|=4 ∣Σ∣ = 4 and ∣ G Σ ∣ ≤ ∣ G O ∣ |G_{\Sigma}|\leq|G_{\mathcal{O}}| ∣ G Σ ∣ ≤ ∣ G O ∣ , then G G G does not satisfy ⋆ 3 \star_3 ⋆ 3 . Explicitly: ∃ O \exists\mathcal{O} ∃ O in a maximal triple orbit and ∃ Σ ⊃ O \exists\Sigma\supset\mathcal{O} ∃Σ ⊃ O , ∣ Σ ∣ = 4 |\Sigma|=4 ∣Σ∣ = 4 , with ∣ G Σ ∣ ≤ ∣ G O ∣ ⇒ G |G_{\Sigma}|\leq|G_{\mathcal{O}}|\Rightarrow G ∣ G Σ ∣ ≤ ∣ G O ∣ ⇒ G fails ⋆ 3 \star_3 ⋆ 3 .
Proof. Under ⋆ 3 \star_3 ⋆ 3 — whose hypotheses include primitivity and degree at least 8, in exact agreement with [1 Prop. 6.2.3] — the witness triple lies in a maximal triple orbit, and ∣ G O ∣ < ∣ G Σ ∣ |G_O| < |G_\Sigma| ∣ G O ∣ < ∣ G Σ ∣ holds for every four-subset Σ ⊃ O \Sigma\supset O Σ ⊃ O ; a four-subset with ∣ G Σ ∣ ≤ ∣ G O ∣ |G_\Sigma| \le|G_O| ∣ G Σ ∣ ≤ ∣ G O ∣ contradicts this.
Lemma 2.17 (Transmission of four-point witnesses to overgroups). Let S ⊴ G ≤ Sym ( Ω ) S \trianglelefteq G \le\operatorname{Sym}(\Omega) S ⊴ G ≤ Sym ( Ω ) with G = S ⋉ H G = S \ltimes H G = S ⋉ H (equivalently G = S H G = SH G = S H , S ∩ H = 1 S \cap H = 1 S ∩ H = 1 ), and suppose that H H H leaves every S S S -orbit on the three-subsets of Ω \Omega Ω setwise invariant. Let O O O be a representative of a maximal S S S -triple orbit. If there is β ∈ Ω ∖ O \beta\in\Omega\setminus O β ∈ Ω ∖ O with ∣ S O ∪ { β } ∣ ≤ ∣ S O ∣ |S_{O \cup\{\beta\}}| \le|S_O| ∣ S O ∪ { β } ∣ ≤ ∣ S O ∣ , then ∣ G O ∪ { β } ∣ ≤ ∣ G O ∣ |G_{O \cup\{\beta\}}| \le|G_O| ∣ G O ∪ { β } ∣ ≤ ∣ G O ∣ , so G G G does not satisfy ⋆ 3 \star_3 ⋆ 3 .
Proof. Since H H H fixes the S S S -orbit of O O O setwise, the G G G -orbit of O O O coincides with its S S S -orbit — in particular O O O is automatically a maximal G G G -triple orbit representative whenever it is a maximal S S S -triple orbit representative — hence ∣ G O ∣ = ∣ G ∣ / ∣ O G ∣ = ∣ G ∣ / ∣ O S ∣ = ∣ G ∣ ⋅ ∣ S O ∣ / ∣ S ∣ = ∣ H ∣ ⋅ ∣ S O ∣ |G_O| = |G|/|O^G| = |G|/|O^S| = |G| \cdot|S_O|/|S| = |H| \cdot|S_O| ∣ G O ∣ = ∣ G ∣/∣ O G ∣ = ∣ G ∣/∣ O S ∣ = ∣ G ∣ ⋅ ∣ S O ∣/∣ S ∣ = ∣ H ∣ ⋅ ∣ S O ∣ . For Σ = O ∪ { β } \Sigma= O \cup\{\beta\} Σ = O ∪ { β } the G G G -orbit Σ G \Sigma^G Σ G contains the S S S -orbit Σ S \Sigma^S Σ S , so ∣ G Σ ∣ = ∣ G ∣ / ∣ Σ G ∣ ≤ ∣ G ∣ / ∣ Σ S ∣ = ∣ H ∣ ⋅ ∣ S Σ ∣ |G_\Sigma| = |G|/|\Sigma^G| \le|G|/|\Sigma^S| = |H| \cdot|S_\Sigma| ∣ G Σ ∣ = ∣ G ∣/∣ Σ G ∣ ≤ ∣ G ∣/∣ Σ S ∣ = ∣ H ∣ ⋅ ∣ S Σ ∣ . The hypothesis gives ∣ S Σ ∣ ≤ ∣ S O ∣ |S_\Sigma| \le|S_O| ∣ S Σ ∣ ≤ ∣ S O ∣ , whence ∣ G Σ ∣ ≤ ∣ H ∣ ⋅ ∣ S O ∣ = ∣ G O ∣ |G_\Sigma| \le|H| \cdot|S_O| = |G_O| ∣ G Σ ∣ ≤ ∣ H ∣ ⋅ ∣ S O ∣ = ∣ G O ∣ , and ⋆ 3 \star_3 ⋆ 3 fails by Lemma 2.16 .
Lemma 2.18 (Four-point witnesses in the almost simple families). Let S S S be one of the 2-transitive almost simple socles that are not 3-homogeneous, namely PSL ( 2 , q ) \operatorname{PSL}(2,q) PSL ( 2 , q ) (q ≡ 1 ( m o d 4 ) q \equiv1 \pmod{4} q ≡ 1 ( mod 4 ) ), PSL ( n , q ) \operatorname{PSL}(n,q) PSL ( n , q ) (n ≥ 3 n \ge3 n ≥ 3 ), PSU ( 3 , q ) \operatorname{PSU}(3,q) PSU ( 3 , q ) , Sz ( q ) \operatorname{Sz}(q) Sz ( q ) , Ree ( q ) \operatorname{Ree}(q) Ree ( q ) or PSp ( 2 n , 2 ) / PΓSp ( 2 n , 2 ) \operatorname{PSp}(2n,2)/\operatorname{P\Gamma Sp}(2n,2) PSp ( 2 n , 2 ) / PΓSp ( 2 n , 2 ) , and let O O O be a representative of a maximal triple orbit. Then there is β ∈ Ω ∖ O \beta\in\Omega\setminus O β ∈ Ω ∖ O such that ∣ S O ∪ { β } ∣ ≤ ∣ S O ∣ |S_{O \cup\{\beta\}}| \le|S_O| ∣ S O ∪ { β } ∣ ≤ ∣ S O ∣ ; in each family the admissible β \beta β form the complement of an explicitly excluded finite set (points and proper subspaces) of size bounded by an explicit B ( q ) < ∣ Ω ∣ − 3 B(q) < |\Omega|-3 B ( q ) < ∣Ω∣ − 3 for q ≥ q 0 q \ge q_0 q ≥ q 0 (the bounds B ( q ) B(q) B ( q ) and the thresholds are recorded in the proof), and the parameters below q 0 q_0 q 0 (all of degree ≤ 200 \le200 ≤ 200 ) are confirmed by the computations of Section 3.
Proof. For each family we exhibit β \beta β and the excluded positions; throughout Σ = O ∪ { β } \Sigma= O \cup\{\beta\} Σ = O ∪ { β } , and g g g denotes an element of S Σ S_\Sigma S Σ .
PSL ( 2 , q ) \operatorname{PSL}(2,q) PSL ( 2 , q ) , q ≡ 1 ( m o d 4 ) q \equiv1 \pmod{4} q ≡ 1 ( mod 4 ) : O = { 0 , 1 , ∞ } O = \{0,1,\infty\} O = { 0 , 1 , ∞ } , ∣ S O ∣ = 6 |S_O| = 6 ∣ S O ∣ = 6 (the determinants are squares since − 1 -1 − 1 is a square; GAP confirms ∣ S O ∣ = 6 |S_O| = 6 ∣ S O ∣ = 6 for q = 5 , 9 , 13 , 17 q = 5, 9, 13, 17 q = 5 , 9 , 13 , 17 ). A Möbius transformation is determined by the image of three points. Exclude from β = x ∈ F q ∖ { 0 , 1 , ∞ } \beta= x \in\mathbb{F}_q \setminus\{0,1,\infty\} β = x ∈ F q ∖ { 0 , 1 , ∞ } the fixed points of the six transformations preserving { 0 , 1 , ∞ } \{0,1,\infty\} { 0 , 1 , ∞ } and, for every three-subset T ⊂ Σ T \subset\Sigma T ⊂ Σ with T ∋ x T \ni x T ∋ x , the solutions of g T ( x ) ∈ { 0 , 1 , ∞ } g_T(x) \in\{0,1,\infty\} g T ( x ) ∈ { 0 , 1 , ∞ } , where g T g_T g T is the unique transformation with g T ( O ) = T g_T(O) = T g T ( O ) = T ; these are finitely many algebraic conditions. If g ( O ) = O g(O) = O g ( O ) = O then g ∈ S O g \in\mathcal{S}_O g ∈ S O and g ( x ) = x g(x)=x g ( x ) = x , so g = 1 g=1 g = 1 by the exclusion; if g ( O ) = T ≠ O g(O)=T\ne O g ( O ) = T = O then T T T contains x x x , g = g T g=g_T g = g T is uniquely determined and g ( x ) ∈ Σ ∖ T ⊆ { 0 , 1 , ∞ } g(x)\in\Sigma\setminus T\subseteq\{0,1,\infty\} g ( x ) ∈ Σ ∖ T ⊆ { 0 , 1 , ∞ } is excluded. Hence S Σ = 1 < 6 = ∣ S O ∣ \mathcal{S}_\Sigma=1<6=|\mathcal{S}_O| S Σ = 1 < 6 = ∣ S O ∣ . The excluded positions number at most 5 ⋅ 2 + 9 = 19 < q − 2 = ∣ Ω ∣ − 3 5\cdot2+9=19<q-2=|\Omega|-3 5 ⋅ 2 + 9 = 19 < q − 2 = ∣Ω∣ − 3 for q ≥ 21 q\ge21 q ≥ 21 (the five non-identity transformations of S O \mathcal{S}_O S O have at most two fixed points each, and the three three-subsets T ∋ x T\ni x T ∋ x give at most three excluded solutions each); the degrees q + 1 ≤ 200 q+1\le200 q + 1 ≤ 200 are covered by the census.
PSL ( n , q ) \operatorname{PSL}(n,q) PSL ( n , q ) , n ≥ 3 n\ge3 n ≥ 3 : O = { [ e 1 ] , [ e 2 ] , [ e 3 ] } O=\{[e_1],[e_2],[e_3]\} O = {[ e 1 ] , [ e 2 ] , [ e 3 ]} non-collinear (the maximal orbit, [8 ] ). Take β = [ v ] \beta=[v] β = [ v ] in totally regular position: all coordinates of v v v nonzero and pairwise distinct ratios. An element of S Σ \mathcal{S}_\Sigma S Σ preserving O O O setwise is a permutation matrix composed with a diagonal scaling modulo scalars, whose fixed points lie on coordinate directions or repeated-eigenvalue subspaces; none but the identity fixes [ v ] [v] [ v ] in totally regular position. An element mapping some [ e i ] [e_i] [ e i ] onto [ v ] [v] [ v ] , or [ v ] [v] [ v ] onto some [ e i ] [e_i] [ e i ] , forces v v v to lie in one of the finitely many diagonal-scaling families h − 1 g − 1 ( e i ) h^{-1}g^{-1}(e_i) h − 1 g − 1 ( e i ) (h h h ranging over the pointwise stabilizer of O O O , g g g over the finitely many three-point-image maps), excluded. Hence S Σ ≤ S O \mathcal{S}_\Sigma\le\mathcal{S}_O S Σ ≤ S O (indeed S Σ = 1 \mathcal{S}_\Sigma=1 S Σ = 1 ). The excluded positions are contained in the union of the n n n coordinate hyperplanes and the finitely many repeated-eigenvalue subspaces of the diagonal scalings, numbering at most n ( q n − 1 − 1 ) / ( q − 1 ) + c n q n − 2 < ∣ Ω ∣ − 3 n(q^{n-1}-1)/(q-1)+c_nq^{n-2}<|\Omega|-3 n ( q n − 1 − 1 ) / ( q − 1 ) + c n q n − 2 < ∣Ω∣ − 3 for q > n + c n q>n+c_n q > n + c n (the constant c n c_n c n counts the diagonal-scaling families; explicitly c n ≤ n 2 c_n\le n^2 c n ≤ n 2 ), so a totally regular β \beta β exists whenever the degree exceeds 200; the complementary parameters lie in the census.
PSU ( 3 , q ) \operatorname{PSU}(3,q) PSU ( 3 , q ) : O O O a non-collinear triple, ∣ S O ∣ ≤ 3 |\mathcal{S}_O|\le3 ∣ S O ∣ ≤ 3 (trivial pointwise stabilizer and image 1 or C 3 C_3 C 3 by the classical unital geometry, [19 ] ). The killers are supplied uniformly by Proposition 2.20 below: for every non-collinear triple Δ \Delta Δ there is a secant point s ∈ ( B ∩ O ) ∖ ( P ∪ Q ) s\in(B\cap O)\setminus(P\cup Q) s ∈ ( B ∩ O ) ∖ ( P ∪ Q ) on the block B = P Q B=PQ B = P Q with ∣ G Δ ∪ { s } ∣ = 1 ≤ ∣ G Δ ∣ |G_{\Delta\cup\{s\}}|=1\le|G_\Delta| ∣ G Δ ∪ { s } ∣ = 1 ≤ ∣ G Δ ∣ , and if ∣ G Δ ∣ = 3 |G_\Delta|=3 ∣ G Δ ∣ = 3 then ∣ G Δ ∪ { s } ∣ = 1 < 3 = ∣ G Δ ∣ |G_{\Delta\cup\{s\}}|=1<3=|G_\Delta| ∣ G Δ ∪ { s } ∣ = 1 < 3 = ∣ G Δ ∣ .
Proposition 2.19 (Killers for collinear unitary triples). Let G = PSU ( 3 , q ) G=\operatorname{PSU}(3,q) G = PSU ( 3 , q ) , q ≥ 3 q\ge3 q ≥ 3 , act on the Hermitian unital O O O with ∣ O ∣ = q 3 + 1 |O|=q^3+1 ∣ O ∣ = q 3 + 1 , and put d = gcd ( 3 , q + 1 ) d=\gcd(3,q+1) d = g cd( 3 , q + 1 ) . Let Δ \Delta Δ be a collinear 3-subset, contained in the secant block B B B . Then every s ∈ O ∖ B s\in O\setminus B s ∈ O ∖ B satisfies
∣ G Δ ∪ { s } ∣ ≤ 6 < 6 ( q + 1 ) d = ∣ G Δ ∣ . |G_{\Delta\cup\{s\}}|\le6<\frac{6(q+1)}{d}=|G_\Delta|. ∣ G Δ ∪ { s } ∣ ≤ 6 < d 6 ( q + 1 ) = ∣ G Δ ∣. Consequently, no collinear triple is a witness for ⋆ 3 \star_3 ⋆ 3 .
Proof. With the Hermitian form h ( v , w ) = v 1 w ‾ 3 + v 2 w ‾ 2 + v 3 w ‾ 1 h(v,w)=v_1\overline{w}_3+v_2\overline{w}_2+v_3\overline{w}_1 h ( v , w ) = v 1 w 3 + v 2 w 2 + v 3 w 1 we may take P = [ 1 : 0 : 0 ] P=[1:0:0] P = [ 1 : 0 : 0 ] , Q = [ 0 : 0 : 1 ] Q=[0:0:1] Q = [ 0 : 0 : 1 ] , B = { y = 0 } B=\{y=0\} B = { y = 0 } , and R = [ r : 0 : 1 ] ∈ B ∩ O R=[r:0:1]\in B\cap\mathcal{O} R = [ r : 0 : 1 ] ∈ B ∩ O with r ≠ 0 r\ne0 r = 0 and r + r ‾ = 0 r+\overline{r}=0 r + r = 0 , so that R ∈ O ∖ { P , Q } R\in\mathcal{O}\setminus\{P,Q\} R ∈ O ∖ { P , Q } ; write Δ = { P , Q , R } \Delta=\{P,Q,R\} Δ = { P , Q , R } .
Structure of the pointwise stabilizer. A unitary transformation fixing the points P P P and Q Q Q preserves the nondegenerate plane ⟨ e 1 , e 3 ⟩ \langle e_1,e_3\rangle ⟨ e 1 , e 3 ⟩ (the secant B B B ) and hence also its orthogonal complement ⟨ e 2 ⟩ \langle e_2\rangle ⟨ e 2 ⟩ ; its matrix is therefore diagonal. Fixing R = [ r : 0 : 1 ] R=[r:0:1] R = [ r : 0 : 1 ] forces the two eigenvalues on ⟨ e 1 , e 3 ⟩ \langle e_1,e_3\rangle ⟨ e 1 , e 3 ⟩ to agree, so the elements of the pointwise stabilizer K = G ( Δ ) K=G_{(\Delta)} K = G ( Δ ) are the images of k λ = diag ( λ , λ − 2 , λ ) k_\lambda=\operatorname{diag}(\lambda,\lambda^{-2},\lambda) k λ = diag ( λ , λ − 2 , λ ) with λ q + 1 = 1 \lambda^{q+1}=1 λ q + 1 = 1 , modulo the centre of SU ( 3 , q ) \operatorname{SU}(3,q) SU ( 3 , q ) . Hence
K ≅ Z ( q + 1 ) / d , ∣ G Δ ∣ = 6 ( q + 1 ) / d , K\cong Z_{(q+1)/d},\qquad|G_\Delta|=6(q+1)/d, K ≅ Z ( q + 1 ) / d , ∣ G Δ ∣ = 6 ( q + 1 ) / d ,
where the S 3 S_3 S 3 in G Δ G_\Delta G Δ permutes the three points of Δ \Delta Δ .
Semiregularity off the block. For s = [ x : y : 1 ] ∈ O ∖ B s=[x:y:1]\in\mathcal{O}\setminus B s = [ x : y : 1 ] ∈ O ∖ B (y ≠ 0 y\ne0 y = 0 , x + x ‾ + y y ‾ = 0 x+\overline{x}+y\overline{y}=0 x + x + y y = 0 ) one has k λ ( s ) = ( λ x , λ − 2 y , λ ) = λ ⋅ [ x : λ − 3 y : 1 ] k_\lambda(s)=(\lambda x,\lambda^{-2}y,\lambda)=\lambda\cdot[x:\lambda^{-3}y:1] k λ ( s ) = ( λ x , λ − 2 y , λ ) = λ ⋅ [ x : λ − 3 y : 1 ] up to scalars, so k λ ( s ) = s k_\lambda(s)=s k λ ( s ) = s iff λ 3 = 1 \lambda^3=1 λ 3 = 1 , iff k λ k_\lambda k λ is central and trivial in G G G . Thus K K K acts semiregularly on O ∖ B \mathcal{O}\setminus B O ∖ B : every non-identity element moves every off-block point. Since Σ s = Δ ∪ { s } \Sigma_s=\Delta\cup\{s\} Σ s = Δ ∪ { s } contains the unique collinear triple Δ \Delta Δ , we have G Σ s = G ( Δ , s ) G_{\Sigma_s}=G_{(\Delta,s)} G Σ s = G ( Δ , s ) and G ( Δ , s ) ∩ K = 1 G_{(\Delta,s)}\cap K=1 G ( Δ , s ) ∩ K = 1 ; the image of G Σ s G_{\Sigma_s} G Σ s in the permutation action on Δ \Delta Δ embeds into S 3 S_3 S 3 , so ∣ G Σ s ∣ ≤ 6 < 6 ( q + 1 ) / d = ∣ G Δ ∣ |G_{\Sigma_s}|\le6<6(q+1)/d=|G_\Delta| ∣ G Σ s ∣ ≤ 6 < 6 ( q + 1 ) / d = ∣ G Δ ∣ for q ≥ 3 q\ge3 q ≥ 3 , as claimed. □
The G G G -orbit of Δ \Delta Δ is the whole collinear triple orbit: G G G is transitive on secant blocks, and G Δ G_\Delta G Δ is transitive on the q − 2 q-2 q − 2 remaining points of B ∩ O B\cap\mathcal{O} B ∩ O . The claims K ≅ Z ( q + 1 ) / d K\cong Z_{(q+1)/d} K ≅ Z ( q + 1 ) / d and k λ ( s ) = [ x : λ − 3 y : 1 ] k_\lambda(s)=[x:\lambda^{-3}y:1] k λ ( s ) = [ x : λ − 3 y : 1 ] are verified for q = 3 , 4 , 5 , 7 , 8 q=3,4,5,7,8 q = 3 , 4 , 5 , 7 , 8 in GAP 4.16: the generator of K K K moves all 24, 60, 120, 336, 504 off-block points respectively.
Proposition 2.20 (Killers for non-collinear unitary triples). *Let G = PSU ( 3 , q ) G=\operatorname{PSU}(3,q) G = PSU ( 3 , q ) , q ≥ 3 q\ge3 q ≥ 3 , act on the Hermitian unital O \mathcal{O} O , and let Δ = { P , Q , R } \Delta=\{P,Q,R\} Δ = { P , Q , R } be a non-collinear triple, R ∉ B : = P Q R\notin B:=PQ R ∈ / B := P Q . Then there exists a secant point s ∈ ( B ∩ O ) ∖ { P , Q } s\in(B\cap\mathcal{O})\setminus\{P,Q\} s ∈ ( B ∩ O ) ∖ { P , Q } with
G Δ ∪ { s } = 1. G_{\Delta\cup\{s\}}=1. G Δ ∪ { s } = 1. Consequently, if ∣ G Δ ∣ = 3 |G_\Delta|=3 ∣ G Δ ∣ = 3 then ∣ G Δ ∪ { s } ∣ = 1 < 3 = ∣ G Δ ∣ |G_{\Delta\cup\{s\}}|=1<3=|G_\Delta| ∣ G Δ ∪ { s } ∣ = 1 < 3 = ∣ G Δ ∣ , so no non-collinear triple with three-point stabilizer of order 3 is a witness for ⋆ 3 \star_3 ⋆ 3 ; and if ∣ G Δ ∣ = 1 |G_\Delta|=1 ∣ G Δ ∣ = 1 then ∣ G Δ ∪ { s } ∣ = 1 = ∣ G Δ ∣ |G_{\Delta\cup\{s\}}|=1=|G_\Delta| ∣ G Δ ∪ { s } ∣ = 1 = ∣ G Δ ∣ , the four-point orbit has the same length as Δ G \Delta^G Δ G , and again Δ \Delta Δ is not a witness.
Proof. We first record a semiregularity lemma for the secant–off-block stabilizer. Let H = G B , R H=G_{B,R} H = G B , R , the stabilizer of the pair consisting of the secant block B B B and the off-block point R R R .
Semiregularity on the block. Let W W W be the two-dimensional nondegenerate Hermitian subspace defining B B B , and E = W ⊥ E=W^{\perp} E = W ⊥ the one-dimensional anisotropic subspace. Write R = [ u + α e ] R=[u+\alpha e] R = [ u + α e ] with u ∈ W u\in W u ∈ W , e ∈ E e\in E e ∈ E and α ≠ 0 \alpha\ne0 α = 0 . Since R R R is a singular point and e e e is anisotropic, h W ( u , u ) = − N ( α ) h ( e , e ) ≠ 0 h_W(u,u)=-N(\alpha)h(e,e)\ne0 h W ( u , u ) = − N ( α ) h ( e , e ) = 0 , so the point [ u ] ∈ PG ( W ) [u]\in\operatorname{PG}(W) [ u ] ∈ PG ( W ) is anisotropic. The block stabilizer induces the surjection π : G B ⟶ PGL ( 2 , q ) \pi:G_B\longrightarrow\operatorname{PGL}(2,q) π : G B ⟶ PGL ( 2 , q ) , whose kernel K B K_B K B is the cyclic group of order ( q + 1 ) / d (q+1)/d ( q + 1 ) / d fixing B B B pointwise, d = ( 3 , q + 1 ) d=(3,q+1) d = ( 3 , q + 1 ) ; in the coordinates of Proposition 2.19 its elements act on O ∖ B \mathcal{O}\setminus B O ∖ B by [ x : y : z ] ↦ [ x : λ − 3 y : z ] [x:y:z]\mapsto[x:\lambda^{-3}y:z] [ x : y : z ] ↦ [ x : λ − 3 y : z ] up to scalars, hence K B K_B K B is semiregular on O ∖ B \mathcal{O}\setminus B O ∖ B and K B ∩ G R = 1 K_B\cap G_R=1 K B ∩ G R = 1 . Thus π \pi π is injective on H H H . Every h ∈ H h\in H h ∈ H preserves W W W and E E E and fixes the projective point [ u + α e ] [u+\alpha e] [ u + α e ] , so its image fixes the anisotropic point [ u ] [u] [ u ] in PG ( W ) \operatorname{PG}(W) PG ( W ) . In PGL ( 2 , q ) \operatorname{PGL}(2,q) PGL ( 2 , q ) the stabilizer of an anisotropic point is a nonsplit torus of order q + 1 q+1 q + 1 , semiregular on the q + 1 q+1 q + 1 singular points of B ∩ O B\cap\mathcal{O} B ∩ O . Hence H H H is semiregular on B ∩ O B\cap\mathcal{O} B ∩ O as well.
Choosing the candidate. Put T s = { P , Q , s } T_s=\{P,Q,s\} T s = { P , Q , s } and Σ s = T s ∪ { R } \Sigma_s=T_s\cup\{R\} Σ s = T s ∪ { R } . The unique collinear three-subset of Σ s \Sigma_s Σ s is T s T_s T s , so every element of G Σ s G_{\Sigma_s} G Σ s preserves T s T_s T s , hence its span B B B , and fixes the remaining point R R R of T s ∪ { R } T_s\cup\{R\} T s ∪ { R } ; conversely any element of H T s H_{T_s} H T s preserves T s T_s T s and fixes R R R , hence preserves Σ s \Sigma_s Σ s . Therefore G Σ s = H T s G_{\Sigma_s}=H_{T_s} G Σ s = H T s , where H = G B , R H=G_{B,R} H = G B , R . By the semiregularity above, a non-identity element fixing the three-subset T s T_s T s has no fixed point on T s T_s T s and cannot induce a transposition (a transposition leaves one point fixed); it therefore acts as a 3-cycle, and its cube fixes T s T_s T s pointwise, hence is trivial. Such T s T_s T s must be an orbit of the unique subgroup of order 3 of H H H (H H H is a subgroup of the cyclic torus, so if 3 ∣ ∣ H ∣ 3\mid|H| 3 ∣ ∣ H ∣ the subgroup H 3 H_3 H 3 of order 3 is unique; if 3 ∤ ∣ H ∣ 3\nmid|H| 3 ∤ ∣ H ∣ there is no bad candidate). A bad candidate can occur only when P , Q P,Q P , Q lie in the same H 3 H_3 H 3 -orbit, in which case it is the third point of that orbit; hence, given P , Q P,Q P , Q , there is at most one bad s s s . The candidate set ( B ∩ O ) ∖ { P , Q } (B\cap\mathcal{O})\setminus\{P,Q\} ( B ∩ O ) ∖ { P , Q } has size q − 1 ≥ 2 q-1\ge2 q − 1 ≥ 2 ; choosing s s s outside the (possibly empty) bad set gives H T s = 1 H_{T_s}=1 H T s = 1 , whence G Σ s = 1 G_{\Sigma_s}=1 G Σ s = 1 , as claimed.
The claims are verified in GAP 4.16 for q = 3 , 4 , 5 q=3,4,5 q = 3 , 4 , 5 : for every non-collinear triple orbit (∣ G Δ ∣ = 6 , 3 |G_{\Delta}|=6,3 ∣ G Δ ∣ = 6 , 3 for q = 3 q=3 q = 3 ; ∣ G Δ ∣ = 3 |G_{\Delta}|=3 ∣ G Δ ∣ = 3 for q = 4 q=4 q = 4 ; ∣ G Δ ∣ = 6 , 1 |G_{\Delta}|=6,1 ∣ G Δ ∣ = 6 , 1 for q = 5 q=5 q = 5 ) the chosen secant candidate satisfies ∣ G Σ s ∣ = 1 |G_{\Sigma_s}|=1 ∣ G Σ s ∣ = 1 (< ∣ G Δ ∣ <|G_{\Delta}| < ∣ G Δ ∣ in the orders-6, 3 cases, = ∣ G Δ ∣ = 1 =|G_{\Delta}|=1 = ∣ G Δ ∣ = 1 in the order-1 cases).
Sz ( q ) \operatorname{Sz}(q) Sz ( q ) : the maximal triple orbit is regular, ∣ S O ∣ = 1 |S_{\mathcal{O}}|=1 ∣ S O ∣ = 1 [24 ] ; cf. the class data of [1 ] . On the ovoid the maps g g g realizing a prescribed three-subset image g ( O ) = T g(\mathcal{O})=T g ( O ) = T are finitely many, and g ( β ) ∈ Σ g(\beta)\in\Sigma g ( β ) ∈ Σ excludes finitely many β \beta β ; choosing β \beta β outside them gives S Σ = 1 = ∣ S O ∣ S_{\Sigma}=1=|S_{\mathcal{O}}| S Σ = 1 = ∣ S O ∣ (confirmed for Sz ( 8 ) \operatorname{Sz}(8) Sz ( 8 ) ). The excluded positions number at most 3 ⋅ 6 + 12 = 30 < q 2 − 2 3\cdot6+12=30<q^2-2 3 ⋅ 6 + 12 = 30 < q 2 − 2 for q ≥ 8 q\ge8 q ≥ 8 (six three-subset images, three image points each); the degree q 2 + 1 q^2+1 q 2 + 1 is ≤ 200 \leq200 ≤ 200 precisely for Sz(8), in the census.
Ree(q q q ), q = 3 2 m + 1 q=3^{2m+1} q = 3 2 m + 1 : let Ω \Omega Ω be the point set of the Ree unital, v = ∣ Ω ∣ = q 3 + 1 v=|\Omega|=q^3+1 v = ∣Ω∣ = q 3 + 1 and ∣ S ∣ = q 3 ( q 3 + 1 ) ( q − 1 ) |S|=q^3(q^3+1)(q-1) ∣ S ∣ = q 3 ( q 3 + 1 ) ( q − 1 ) . The action is 2-transitive, and the blocks of the unital are the fixed-point sets of the involutions: for every involution j ∈ S j\in S j ∈ S the set B j = Fix Ω ( j ) B_j=\operatorname{Fix}_{\Omega}(j) B j = Fix Ω ( j ) is a block with ∣ B j ∣ = q + 1 |B_j|=q+1 ∣ B j ∣ = q + 1 , and C S ( j ) ≅ C 2 × PSL ( 2 , q ) C_S(j)\cong C_2\times\operatorname{PSL}(2,q) C S ( j ) ≅ C 2 × PSL ( 2 , q ) [21 , 15 ] . No non-identity element fixes three points apart from the involutions, so for every three-subset D D D the pointwise stabilizer satisfies S ( D ) ≤ C 2 S_{(D)}\leq C_2 S ( D ) ≤ C 2 , with S ( D ) ≠ 1 S_{(D)}\neq1 S ( D ) = 1 iff D ⊆ B j D\subseteq B_j D ⊆ B j for a unique involution j j j , in which case S ( D ) = ⟨ j ⟩ S_{(D)}=\langle j\rangle S ( D ) = ⟨ j ⟩ .
The three-subset orbits are parameterized by the two-point stabilizer H = S α , β H=S_{\alpha,\beta} H = S α , β , which is cyclic of order q − 1 q-1 q − 1 . Take coordinates B j ≅ PG ( 1 , q ) B_j\cong\operatorname{PG}(1,q) B j ≅ PG ( 1 , q ) with α = ∞ \alpha=\infty α = ∞ , β = 0 \beta=0 β = 0 , and let ν ∈ F q × \nu\in\mathbb{F}_q^{\times} ν ∈ F q × be a non-square: the two suborbits γ + = 1 \gamma_+=1 γ + = 1 , γ − = ν \gamma_-=\nu γ − = ν of H H H inside B j B_j B j form a single S 3 S_3 S 3 -orbit { H γ + , H γ − } \{H\gamma_+,H\gamma_-\} { H γ + , H γ − } of size 2, corresponding to the unique block-contained triple orbit, with r T = ∣ H γ ∣ = 2 r_T=|H\gamma|=2 r T = ∣ H γ ∣ = 2 and L T = Stab S 3 ( T ) ≅ C 3 L_T=\operatorname{Stab}_{S_3}(T)\cong C_3 L T = Stab S 3 ( T ) ≅ C 3 ; every external suborbit H γ H\gamma H γ , γ ∈ Ω ∖ B j \gamma\in\Omega\setminus B_j γ ∈ Ω ∖ B j , has trivial point stabilizer H γ = 1 H_\gamma=1 H γ = 1 , and the external representatives R = { γ i : i ∈ I } \mathcal{R}=\{\gamma_i:i\in I\} R = { γ i : i ∈ I } , ∣ I ∣ = q ( q + 1 ) |I|=q(q+1) ∣ I ∣ = q ( q + 1 ) , parameterize the remaining orbits via H γ i H\gamma_i H γ i . For a three-subset D γ = { α , β , γ } D_\gamma=\{\alpha,\beta,\gamma\} D γ = { α , β , γ } the induced action of the setwise stabilizer gives the exact sequence 1 ⟶ S ( D γ ) ⟶ S { D γ } ⟶ L T ⟶ 1 1\longrightarrow S_{(D_\gamma)}\longrightarrow S_{\{D_\gamma\}}\longrightarrow L_T\longrightarrow1 1 ⟶ S ( D γ ) ⟶ S { D γ } ⟶ L T ⟶ 1 , whence ∣ S { D γ } ∣ = r T ∣ L T ∣ |S_{\{D_\gamma\}}|=r_T|L_T| ∣ S { D γ } ∣ = r T ∣ L T ∣ and ∣ S D γ ∣ = ∣ S ∣ / ( r T ∣ T ∣ ) |S^{D_\gamma}|=|S|/(r_T|T|) ∣ S D γ ∣ = ∣ S ∣/ ( r T ∣ T ∣ ) ; the external orbits are classified by the S 3 S_3 S 3 -orbit size s = ∣ T S 3 ∣ s=|T^{S_3}| s = ∣ T S 3 ∣ of H γ i H\gamma_i H γ i on R \mathcal{R} R , with ∣ L T ∣ = 6 / s |L_T|=6/s ∣ L T ∣ = 6/ s . Note that S { D } S_{\{D\}} S { D } is the setwise stabilizer of the triple; the pointwise stabilizer S ( D ) S_{(D)} S ( D ) is trivial on the external orbits, but S { D } S_{\{D\}} S { D } need not be.
A three-subset D γ D_\gamma D γ is a witness exactly when ∣ S { Σ γ , x } ∣ > ∣ S { D γ } ∣ = r T ∣ L T ∣ |S_{\{\Sigma_{\gamma,x}\}}|>|S_{\{D_\gamma\}}|=r_T|L_T| ∣ S { Σ γ , x } ∣ > ∣ S { D γ } ∣ = r T ∣ L T ∣ for every x ∉ D γ x\notin D_\gamma x ∈ / D γ , where Σ γ , x = D γ ∪ { x } \Sigma_{\gamma,x}=D_\gamma\cup\{x\} Σ γ , x = D γ ∪ { x } ; hence D γ D_\gamma D γ is excluded once a complement x x x satisfies ∣ S { Σ γ , x } ∣ ≤ r T ∣ L T ∣ |S_{\{\Sigma_{\gamma,x}\}}|\leq r_T|L_T| ∣ S { Σ γ , x } ∣ ≤ r T ∣ L T ∣ . For the extension A = S ⋊ C 3 A=S\rtimes C_3 A = S ⋊ C 3 (3 ∣ 2 m + 1 3\mid2m+1 3 ∣ 2 m + 1 ) put π : A → A / S ≅ C 3 \pi:A\to A/S\cong C_3 π : A → A / S ≅ C 3 and e Y = ∣ π ( A { Y } ) ∣ e_Y=|\pi(A_{\{Y\}})| e Y = ∣ π ( A { Y } ) ∣ : e Y = 3 e_Y=3 e Y = 3 if the field automorphism preserves the S S S -orbit of Y Y Y , and e Y = 1 e_Y=1 e Y = 1 if it cyclically fuses it with two others, so the certificate inequality becomes e X ∣ S { Σ γ , x } ∣ ≤ e D γ r T ∣ L T ∣ e_X|S_{\{\Sigma_{\gamma,x}\}}|\leq e_{D_\gamma}r_T|L_T| e X ∣ S { Σ γ , x } ∣ ≤ e D γ r T ∣ L T ∣ .
For a fixed q q q the exclusion is verified by the finite certificate: enumerate the three-subset orbit representatives D D D , compute ∣ S { D } ∣ |S_{\{D\}}| ∣ S { D } ∣ and r T ∣ L T ∣ r_T|L_T| r T ∣ L T ∣ , and test one representative point x x x per orbit of S { D } S_{\{D\}} S { D } on Ω ∖ D \Omega\setminus D Ω ∖ D , checking the inequality above. GAP confirms this for 2 G 2 ( 3 ) ≅ PΓL ( 2 , 8 ) {}^2G_2(3)\cong\operatorname{P\Gamma L}(2,8) 2 G 2 ( 3 ) ≅ PΓL ( 2 , 8 ) (v = 28 v=28 v = 28 ): six three-subset orbits of lengths 756, 504, 756, 252, 756, 252 (∑ = 3276 = ( 28 3 ) \sum=3276=\binom{28}{3} ∑ = 3276 = ( 3 28 ) ), with setwise stabilizers ∣ S { D } ∣ = 2 , 3 , 2 , 6 , 2 , 6 |S_{\{D\}}|=2,3,2,6,2,6 ∣ S { D } ∣ = 2 , 3 , 2 , 6 , 2 , 6 and pointwise stabilizers ∣ S ( D ) ∣ = 1 , 1 , 1 , 1 , 1 , 2 |S_{(D)}|=1,1,1,1,1,2 ∣ S ( D ) ∣ = 1 , 1 , 1 , 1 , 1 , 2 , matching ∣ S { D } ∣ = r T ∣ L T ∣ |S_{\{D\}}|=r_T|L_T| ∣ S { D } ∣ = r T ∣ L T ∣ (one block contained orbit with r T = r_T= r T = 2 2 2 , ∣ L T ∣ = 3 \lvert L_T\rvert= 3 ∣ L T ∣ = 3 ; five external orbits with r T = 1 r_T = 1 r T = 1 and ∣ L T ∣ = 2 , 3 , 2 , 6 , 2 \lvert L_T\rvert= 2, 3, 2, 6, 2 ∣ L T ∣ = 2 , 3 , 2 , 6 , 2 , i.e. S 3 S_3 S 3 -orbit sizes s = 3 , 2 , 3 , 1 , 3 s = 3, 2, 3, 1, 3 s = 3 , 2 , 3 , 1 , 3 , so n 1 = n 2 = 1 n_1 = n_2 = 1 n 1 = n 2 = 1 , n 3 = 3 n_3 = 3 n 3 = 3 , n 6 = 0 n_6 = 0 n 6 = 0 and n 1 + 2 n 2 + 3 n 3 + 6 n 6 = 12 = q ( q + 1 ) n_1 + 2n_2 + 3n_3 + 6n_6 = 12 = q(q + 1) n 1 + 2 n 2 + 3 n 3 + 6 n 6 = 12 = q ( q + 1 ) , σ 3 ( S ) = 1 + n 1 + n 2 + n 3 + n 6 = 6 = ( q + 3 ) 2 / 6 \sigma_3(S) = 1 + n_1 + n_2 + n_3 + n_6 = 6 = (q + 3)^2/6 σ 3 ( S ) = 1 + n 1 + n 2 + n 3 + n 6 = 6 = ( q + 3 ) 2 /6 ); moreover min x ∣ S { γ , x } ∣ = 1 \min_x \lvert S^{\{\gamma,x\}}\rvert= 1 min x ∣ S { γ , x } ∣ = 1 in every orbit, so no witness exists.
The S 3 S_3 S 3 -orbit decomposition of the external parameter set. We now record the explicit decomposition which closes the parametrization above; the proof is self-contained and yields, in particular, the count σ 3 ( S ) = ( q + 3 ) 2 / 6 \sigma_3(S) = (q + 3)^2/6 σ 3 ( S ) = ( q + 3 ) 2 /6 in agreement with the Bradley class data [1 ] .
Proposition 2.21 (External S 3 S_3 S 3 -orbit decomposition). Let S = Ree ( q ) S = \operatorname{Ree}(q) S = Ree ( q ) , q = 3 2 m + 1 > 3 q = 3^{2m+1} > 3 q = 3 2 m + 1 > 3 , act on the Ree unital Ω \Omega Ω with ∣ Ω ∣ = q 3 + 1 \lvert\Omega\rvert= q^3 + 1 ∣ Ω ∣ = q 3 + 1 . Fix an ordered pair ( α , β ) (\alpha,\beta) ( α , β ) , let B B B be the unique block through it, put H = S α , β H = S_{\alpha,\beta} H = S α , β , and let R \mathcal{R} R be the set of external suborbit representatives, i.e. R = H \ ( Ω ∖ B ) \mathcal{R} = H\backslash(\Omega\setminus B) R = H \ ( Ω ∖ B ) . Then H ≅ C q − 1 H \cong C_{q-1} H ≅ C q − 1 and ∣ R ∣ = q ( q + 1 ) \lvert\mathcal{R}\rvert= q(q + 1) ∣ R ∣ = q ( q + 1 ) . Let n s n_s n s be the number of S 3 S_3 S 3 -orbits of size s ∈ { 1 , 2 , 3 , 6 } s \in\{1,2,3,6\} s ∈ { 1 , 2 , 3 , 6 } on R \mathcal{R} R under the coordinate permutation. Then
n 1 = 1 , n 2 = q − 1 2 , n 3 = q , n 6 = q ( q − 3 ) 6 . (7) \begin{aligned}
n_1 &= 1,\qquad n_2 = \frac{q-1}{2},\qquad n_3 = q,\\
n_6 &= \frac{q(q-3)}{6}. \tag*{(7)}
\end{aligned} n 1 n 6 = 1 , n 2 = 2 q − 1 , n 3 = q , = 6 q ( q − 3 ) . (7) with the corresponding external three-subset setwise stabilizers and orbit lengths
Table 4.
In particular the pointwise stabilizer of a non-block-contained triple is trivial while the setwise stabilizer need not be. Adding the unique block-contained triple orbit,
σ 3 ( S ) = ( q + 3 ) 2 6 . (8) \sigma_3(S) = \frac{(q+3)^2}{6}. \tag*{(8)} σ 3 ( S ) = 6 ( q + 3 ) 2 . (8) Proof. Let F 2 F_2 F 2 (resp. F 3 F_3 F 3 ) be the number of parameters of R \mathcal{R} R fixed by a prescribed transposition (resp. a prescribed 3-cycle) of S 3 S_3 S 3 . For the four orbit types the fixed-point contributions are: size 1 — every transposition and every 3-cycle fix the point; size 2 — no transposition fixes a point, every 3-cycle fixes both points; size 3 — the prescribed transposition fixes one point,
a 3-cycle fixes none; size 6 — none. Hence
F 2 = n 1 + n 3 , F 3 = n 1 + 2 n 2 . (9) F_{2}=n_{1}+n_{3}, \qquad F_{3}=n_{1}+2n_{2}. \tag*{(9)} F 2 = n 1 + n 3 , F 3 = n 1 + 2 n 2 . (9)
1. F 2 = q + 1 F_{2}=q+1 F 2 = q + 1 . Let j ∈ S j\in S j ∈ S be an involution. Its cycle type on Ω \Omega Ω is 1 q + 1 2 ( q 3 − q ) / 2 1^{q+1}2^{(q^{3}-q)/2} 1 q + 1 2 ( q 3 − q ) /2 , so the three-subsets consisting of one fixed point and one 2-cycle number ( q + 1 ) ( q 3 − q ) / 2 (q+1)(q^{3}-q)/2 ( q + 1 ) ( q 3 − q ) /2 . None of them is contained in a unital block: otherwise j j j would stabilize that block and induce a transposition on the triple, while the induced action of the block-contained triple stabilizer is C 3 C_{3} C 3 , which has no transposition. Involutions are in bijection with unital blocks, and b = ( q 3 + 1 2 ) / ( q + 1 2 ) = q 2 ( q 2 − q + 1 ) b=\binom{q^{3}+1}{2}/\binom{q+1}{2}=q^{2}(q^{2}-q+1) b = ( 2 q 3 + 1 ) / ( 2 q + 1 ) = q 2 ( q 2 − q + 1 ) , so the “involution–external triple” incidence number is
I 2 = b ( q + 1 ) ( q 3 − q ) 2 . (10) I_{2}=b\frac{(q+1)(q^{3}-q)}{2}. \tag*{(10)} I 2 = b 2 ( q + 1 ) ( q 3 − q ) . (10)
If the induced image of the external triple stabilizer is C 2 C_{2} C 2 the triple contributes one incidence, and if it is S 3 S_{3} S 3 it contributes three; by the orbit–stabilizer formula this gives
I 2 = ∣ S ∣ 2 ( n 1 + n 3 ) = ∣ S ∣ 2 F 2 . (11) I_{2}=\frac{|S|}{2}(n_{1}+n_{3})=\frac{|S|}{2}F_{2}. \tag*{(11)} I 2 = 2 ∣ S ∣ ( n 1 + n 3 ) = 2 ∣ S ∣ F 2 . (11)
With ∣ S ∣ = q 3 ( q 3 + 1 ) ( q − 1 ) |S|=q^{3}(q^{3}+1)(q-1) ∣ S ∣ = q 3 ( q 3 + 1 ) ( q − 1 ) and (4) we obtain F 2 = q + 1 F_{2}=q+1 F 2 = q + 1 .
2. F 3 = q F_{3}=q F 3 = q . The elements of order 3 of S S S fall into the three classes X , T , T − 1 X,T,T^{-1} X , T , T − 1 , all of cycle type 1 3 q 3 / 3 1\,3^{q^{3}/3} 1 3 q 3 /3 on Ω \Omega Ω , with centralizer orders q 3 q^{3} q 3 , 2 q 2 2q^{2} 2 q 2 , 2 q 2 2q^{2} 2 q 2 . Their total number is
∣ S ∣ q 3 + 2 ∣ S ∣ 2 q 2 = ∣ S ∣ ( q + 1 ) q 3 . (12) \frac{|S|}{q^{3}}+2\frac{|S|}{2q^{2}}=\frac{|S|(q+1)}{q^{3}}. \tag*{(12)} q 3 ∣ S ∣ + 2 2 q 2 ∣ S ∣ = q 3 ∣ S ∣ ( q + 1 ) . (12)
Each element of order 3 stabilizes exactly q 3 / 3 q^{3}/3 q 3 /3 three-subsets, its 3-cycle supports, so the “order-3 element–triple” incidence number is ∣ S ∣ ( q + 1 ) / 3 |S|(q+1)/3 ∣ S ∣ ( q + 1 ) /3 . The block-contained triples number b ( q + 1 3 ) = ∣ S ∣ / 6 b\binom{q+1}{3}=|S|/6 b ( 3 q + 1 ) = ∣ S ∣/6 , each with stabilizer C 6 C_{6} C 6 containing two elements of order 3, contributing ∣ S ∣ / 3 |S|/3 ∣ S ∣/3 incidences; subtracting gives the external incidence number
I 3 e x t = ∣ S ∣ q 3 . (13) I_{3}^{\mathrm{ext}}=\frac{|S|q}{3}. \tag*{(13)} I 3 ext = 3 ∣ S ∣ q . (13)
If the induced image of the external triple stabilizer contains C 3 C_{3} C 3 , then the image is C 3 C_{3} C 3 or S 3 S_{3} S 3 , contributing two elements of order 3 each; therefore I 3 e x t = ∣ S ∣ 3 ( n 1 + 2 n 2 ) = ∣ S ∣ 3 F 3 I_{3}^{\mathrm{ext}}=\frac{|S|}{3}(n_{1}+2n_{2})=\frac{|S|}{3}F_{3} I 3 ext = 3 ∣ S ∣ ( n 1 + 2 n 2 ) = 3 ∣ S ∣ F 3 , and comparison with (7) yields F 3 = q F_{3}=q F 3 = q .
3. n 1 = 1 n_1=1 n 1 = 1 . Fix ξ ∈ Ω \xi\in\Omega ξ ∈ Ω and use the standard Ree point stabilizer B ξ = S ξ = P ⋊ C B_\xi=S_\xi=P\rtimes C B ξ = S ξ = P ⋊ C with ∣ P ∣ = q 3 |P|=q^3 ∣ P ∣ = q 3 and C ≅ C q − 1 C\cong C_{q-1} C ≅ C q − 1 , where P P P is the normal Sylow 3-subgroup and Z ( P ) ≅ ( F q , + ) Z(P)\cong(\mathbb{F}_q,+) Z ( P ) ≅ ( F q , + ) , ∣ Z ( P ) ∣ = q |Z(P)|=q ∣ Z ( P ) ∣ = q . In the Ree root-group coordinates write the central root elements as z ( c ) z(c) z ( c ) , c ∈ F q c\in\mathbb{F}_q c ∈ F q ; if t t t is the unique involution of C C C , then
t z ( c ) t − 1 = z ( − c ) , (14) t z(c)t^{-1}=z(-c), \tag*{(14)} t z ( c ) t − 1 = z ( − c ) , (14)
a relation that has to be checked from the Ree root-group representation chosen (it does not follow from the element conjugacy class table alone). Since P P P centralizes Z ( P ) Z(P) Z ( P ) and every involution of B ξ B_\xi B ξ has image t t t in B ξ / P B_\xi/P B ξ / P , (8) implies that every involution fixing ξ \xi ξ inverts every subgroup of order 3 of Z ( P ) Z(P) Z ( P ) . Moreover only the X-class elements of order 3 are conjugate to their inverses in S S S ; T T T and T − 1 T^{-1} T − 1 lie in distinct S S S -classes. Hence if the setwise stabilizer of an external triple is S 3 S_3 S 3 , its C 3 C_3 C 3 must come from Z ( P ) Z(P) Z ( P ) .
Now Z ( P ) Z(P) Z ( P ) has ( q − 1 ) / 2 (q-1)/2 ( q − 1 ) /2 subgroups of order 3; the number of blocks through ξ \xi ξ is q 2 q^2 q 2 , so B ξ B_\xi B ξ contains q 2 q^2 q 2 involutions. Every pair ( Q , j ) (Q,j) ( Q , j ) with Q ≤ Z ( P ) Q\leq Z(P) Q ≤ Z ( P ) , ∣ Q ∣ = 3 |Q|=3 ∣ Q ∣ = 3 and j ∈ B ξ j\in B_\xi j ∈ B ξ an involution generates an S 3 S_3 S 3 subgroup K = ⟨ Q , j ⟩ K=\langle Q,j\rangle K = ⟨ Q , j ⟩ . Each such K K K has ξ \xi ξ as its unique global fixed point (the non-identity elements of Q Q Q fix only ξ \xi ξ ); the fixed-point sets of the three involutions of K K K pairwise meet in ξ \xi ξ and have q q q points each, so K K K has 3 q 3q 3 q points with stabilizer exactly C 2 C_2 C 2 , forming q q q orbits of size 3. These q q q triple orbits are all external: if one of them lay in a block, the unique block involution would fix it pointwise, contradicting the fact that the three points have three distinct C 2 C_2 C 2 stabilizers in K K K . Their setwise stabilizer is exactly K K K , since the external triple setwise stabilizer has trivial pointwise stabilizer and its induced image contains S 3 S_3 S 3 . For fixed ξ \xi ξ the number of pairs ( Q , j ) (Q,j) ( Q , j ) is q − 1 2 ⋅ q 2 \frac{q-1}{2}\cdot q^2 2 q − 1 ⋅ q 2 ; each generated S 3 S_3 S 3 subgroup is counted three times by its three involutions and has a unique fixed point, so the total number of such S 3 S_3 S 3 subgroups is
( q 3 + 1 ) q 2 ( q − 1 ) 6 . (15) \frac{(q^3+1)q^2(q-1)}{6}. \tag*{(15)} 6 ( q 3 + 1 ) q 2 ( q − 1 ) . (15)
Each of them corresponds to q q q external triples with orbit length ∣ S ∣ / 6 |S|/6 ∣ S ∣/6 , whence
n 1 = q ⋅ ( q 3 + 1 ) q 2 ( q − 1 ) / 6 ∣ S ∣ / 6 = 1. (16) n_1=\frac{q\cdot(q^3+1)q^2(q-1)/6}{|S|/6}=1. \tag*{(16)} n 1 = ∣ S ∣/6 q ⋅ ( q 3 + 1 ) q 2 ( q − 1 ) /6 = 1. (16)
4. The remaining n s n_s n s . From (3), F 2 = q + 1 F_2=q+1 F 2 = q + 1 , F 3 = q F_3=q F 3 = q and n 1 = 1 n_1=1 n 1 = 1 we get
n 3 = q , n 2 = q − 1 2 . (17) n_3=q,\qquad n_2=\frac{q-1}{2}. \tag*{(17)} n 3 = q , n 2 = 2 q − 1 . (17)
Burnside’s lemma gives the number of external orbits
n 1 + n 2 + n 3 + n 6 = ∣ R ∣ + 3 F 2 + 2 F 3 6 = q ( q + 1 ) + 3 ( q + 1 ) + 2 q 6 = q 2 + 6 q + 3 6 , (18) n_1+n_2+n_3+n_6
= \frac{|\mathcal{R}|+3F_2+2F_3}{6}
= \frac{q(q+1)+3(q+1)+2q}{6}
= \frac{q^2+6q+3}{6}, \tag*{(18)} n 1 + n 2 + n 3 + n 6 = 6 ∣ R ∣ + 3 F 2 + 2 F 3 = 6 q ( q + 1 ) + 3 ( q + 1 ) + 2 q = 6 q 2 + 6 q + 3 , (18)
hence n 6 = q ( q − 3 ) / 6 n_6=q(q-3)/6 n 6 = q ( q − 3 ) /6 , completing the proof.
Adding the block-contained orbit (one extra S 3 S_3 S 3 -orbit of size 2) to (18) gives σ 3 ( S ) = q 2 + 6 q + 3 6 + 1 = ( q + 3 ) 2 6 \sigma_3(S)=\frac{q^2+6q+3}{6}+1=\frac{(q+3)^2}{6} σ 3 ( S ) = 6 q 2 + 6 q + 3 + 1 = 6 ( q + 3 ) 2 , as claimed in (2). For q = 3 q=3 q = 3 the values n 1 = 1 n_1=1 n 1 = 1 , n 2 = 1 n_2=1 n 2 = 1 , n 3 = 3 n_3=3 n 3 = 3 , n 6 = 0 n_6=0 n 6 = 0 agree with the GAP certificate above.
The decomposition (1) determines all external triple setwise-stabilizer types and orbit numbers. What it does not settle is the existence, for every S S S -orbit representative D D D , of a complement x ∉ D x\notin D x ∈ / D satisfying
e D ∪ { x } ∣ S D ∪ { x } ∣ ≤ e D ∣ S D ∣ , (19) e_{D\cup\{x\}}|S_{D\cup\{x\}}|\leq e_D|S_D|, \tag*{(19)} e D ∪ { x } ∣ S D ∪ { x } ∣ ≤ e D ∣ S D ∣ , (19)
which is precisely the Ree extension certificate proposition ; in the extension A = S ⋊ C 3 A=S\rtimes C_3 A = S ⋊ C 3 the tuple ( n 1 , n 2 , n 3 , n 6 ) (n_1,n_2,n_3,n_6) ( n 1 , n 2 , n 3 , n 6 ) does not determine how the field automorphism δ \delta δ acts on the S S S -orbits, and the numbers e D e_D e D , e D ∪ { x } e_{D\cup\{x\}} e D ∪ { x } have to be pinned down orbit by orbit. We therefore record (1) as a proved orbit-decomposition theorem, and the extension certificate (19) is established below: for general q > 3 q>3 q > 3 by the bad-point bound of Proposition 2.22 , and for the tested parameter q = 3 q=3 q = 3 by the explicit certificate of Proposition 2.23 .
For general q > 3 q>3 q > 3 the existence of the complement point x x x satisfying (19) is now settled for the core s = 6 s=6 s = 6 case by the general bad-point bound of Proposition 2.22 ; the S 3 S_3 S 3 -orbit decomposition of the external parameter set is completed in Proposition 2.21 , and for the tested parameter q = 3 q=3 q = 3 the finite stabilizer computation of Proposition 2.23 supplies an explicit certificate for every S S S -orbit on 3-subsets, so that no witness quadruple arises on the socle level.
Proposition 2.22 (Bad-point bound for the external orbits and existence of good complements). *Let S = Ree ( q ) S=\operatorname{Ree}(q) S = Ree ( q ) act on the Ree unital Ω \Omega Ω with ∣ Ω ∣ = q 3 + 1 |\Omega|=q^3+1 ∣Ω∣ = q 3 + 1 , and let D ⊂ Ω D\subset\Omega D ⊂ Ω be a 3-subset of external S 3 S_3 S 3 -type s ∈ { 1 , 2 , 3 , 6 } s\in\{1,2,3,6\} s ∈ { 1 , 2 , 3 , 6 } , so that
∣ S { D } ∣ = 6 , 3 , 2 , 1 |S_{\{D\}}|=6,3,2,1 ∣ S { D } ∣ = 6 , 3 , 2 , 1 respectively by Proposition 2.21 . Define the bad points of D D D by
B D f u l l = { x ∈ Ω ∖ D : ∣ S { D ∪ { x } } ∣ > 1 } . (20) B_D^{\mathrm{full}}=\{x\in\Omega\setminus D:|S_{\{D\cup\{x\}\}}|>1\}. \tag*{(20)} B D full = { x ∈ Ω ∖ D : ∣ S { D ∪ { x }} ∣ > 1 } . (20)
(a) For s = 6 s=6 s = 6 (S { D } = 1 S_{\{D\}}=1 S { D } = 1 ) every x ∈ B D f u l l x\in B_D^{\mathrm{full}} x ∈ B D full has a witness g ∈ S { D ∪ { x } } ∖ { 1 } g\in S_{\{D\cup\{x\}\}}\setminus\{1\} g ∈ S { D ∪ { x }} ∖ { 1 } with ∣ D g ∩ D ∣ = 2 |D^g\cap D|=2 ∣ D g ∩ D ∣ = 2 . Indeed, if ∣ D g ∩ D ∣ = 3 |D^g\cap D|=3 ∣ D g ∩ D ∣ = 3 then g ( D ) = D g(D)=D g ( D ) = D , whence g ∈ S { D } = { 1 } g\in S_{\{D\}}=\{1\} g ∈ S { D } = { 1 } , contradiction.
(b) For every external D D D of any type the set B D f u l l B_D^{\mathrm{full}} B D full is covered by the candidate set T ( D ) T(D) T ( D ) obtained from the 18 transmission conditions (choose the source point d ∈ D d\in D d ∈ D , 3 choices, and an ordered pair ( u ′ , v ′ ) (u',v') ( u ′ , v ′ ) of distinct points of D D D for the images of the remaining two points, 6 choices): for fixed ( d ; u ′ , v ′ ) (d;u',v') ( d ; u ′ , v ′ ) , with { u , v } = D ∖ { d } \{u,v\}=D\setminus\{d\} { u , v } = D ∖ { d } , any g g g sending d d d outside D D D and u ↦ u ′ , v ↦ v ′ u\mapsto u',v\mapsto v' u ↦ u ′ , v ↦ v ′ lies in the coset t ⋅ S { ( u ′ , v ′ ) } t\cdot S_{\{(u',v')\}} t ⋅ S {( u ′ , v ′ )} of the pointwise stabilizer of ( u ′ , v ′ ) (u',v') ( u ′ , v ′ ) , of size ≤ q − 1 \leq q-1 ≤ q − 1 . Hence
∣ B D f u l l ∣ ≤ ∣ T ( D ) ∣ ≤ 18 ( q − 1 ) . (21) |B_D^{\mathrm{full}}|\leq|T(D)|\leq18(q-1). \tag*{(21)} ∣ B D full ∣ ≤ ∣ T ( D ) ∣ ≤ 18 ( q − 1 ) . (21)
(c) Since ∣ Ω ∖ D ∣ = q 3 − 2 |\Omega\setminus D|=q^3-2 ∣Ω ∖ D ∣ = q 3 − 2 and $18(q-1)<q^3-2$ for all q ≥ 27 q\geq27 q ≥ 27 (e.g. 19681 > 468 19681>468 19681 > 468 for q = 27 q=27 q = 27 ; in general q 3 − 18 q + 16 > 0 q^3-18q+16>0 q 3 − 18 q + 16 > 0 for q ≥ 4 q\geq4 q ≥ 4 ), the complement set has a point x ∉ T ( D ) x\notin T(D) x ∈ / T ( D ) , hence x ∉ B D f u l l x\notin B_D^{\mathrm{full}} x ∈ / B D full and ∣ S { D ∪ { x } } ∣ = 1 |S_{\{D\cup\{x\}\}}|=1 ∣ S { D ∪ { x }} ∣ = 1 . By the orbit–stabilizer identity ∣ Σ x S ∣ / ∣ D S ∣ = ∣ S { D } ∣ / ∣ S { Σ x } ∣ = s ≥ 1 |\Sigma_x^S|/|D^S|=|S_{\{D\}}|/|S_{\{\Sigma_x\}}|=s\geq1 ∣ Σ x S ∣/∣ D S ∣ = ∣ S { D } ∣/∣ S { Σ x } ∣ = s ≥ 1 with Σ x = D ∪ { x } \Sigma_x=D\cup\{x\} Σ x = D ∪ { x } , this x x x is a certificate point for D D D in the sense of Proposition 2.23 (a); in particular no witness quadruple is built on $D.
The precise distribution of ∣ B D f u l l ∣ |B_D^{\mathrm{full}}| ∣ B D full ∣ and of the stabilizer spectra of the bad points, for the tested parameter q = 27 q=27 q = 27 , is a finite computation recorded in the verification section; it confirms (15) with maximum ∣ B D f u l l ∣ = 81 |B_D^{\mathrm{full}}|=81 ∣ B D full ∣ = 81 among the 108 external s = 6 s=6 s = 6 orbits, and separates the orbits into the four spectrum types 81 ⋅ C 2 81\cdot C_2 81 ⋅ C 2 , 78 ⋅ C 2 + 1 ⋅ C 3 + 1 ⋅ V 4 78\cdot C_2+1\cdot C_3+1\cdot V_4 78 ⋅ C 2 + 1 ⋅ C 3 + 1 ⋅ V 4 , 78 ⋅ C 2 + 1 ⋅ V 4 78\cdot C_2+1\cdot V_4 78 ⋅ C 2 + 1 ⋅ V 4 and 72 ⋅ C 2 + 3 ⋅ V 4 72\cdot C_2+3\cdot V_4 72 ⋅ C 2 + 3 ⋅ V 4 .
Proposition 2.23 (Ree extension certificate: computational framework and tested parameters). Let S = Ree ( q ) S=\operatorname{Ree}(q) S = Ree ( q ) act on the Ree unital Ω \Omega Ω with ∣ Ω ∣ = q 3 + 1 |\Omega|=q^3+1 ∣Ω∣ = q 3 + 1 , and let K D = S { D } K_D=S_{\{D\}} K D = S { D } denote the setwise stabilizer of a 3-subset D D D .
(a) By Proposition 2.21 one has ∣ K D ∣ = r D |K_D|=r_D ∣ K D ∣ = r D , where r D = 6 , 3 , 2 , 1 r_D=6,3,2,1 r D = 6 , 3 , 2 , 1 according as the external S 3 S_3 S 3 -type of D D D is s = 1 , 2 , 3 , 6 s=1,2,3,6 s = 1 , 2 , 3 , 6 . For Σ x = D ∪ { x } \Sigma_x=D\cup\{x\} Σ x = D ∪ { x } , x ∈ Ω ∖ D x\in\Omega\setminus D x ∈ Ω ∖ D , the orbit–stabilizer theorem gives ∣ D S ∣ / ∣ Σ x S ∣ = ∣ K D ∣ / ∣ S { Σ x } ∣ |D^S|/|\Sigma_x^S|=|K_D|/|S_{\{\Sigma_x\}}| ∣ D S ∣/∣ Σ x S ∣ = ∣ K D ∣/∣ S { Σ x } ∣ , so a certificate for D D D is a point x ∉ D x\notin D x ∈ / D with ∣ S { Σ x } ∣ ≤ ∣ S { D } ∣ |S_{\{\Sigma_x\}}|\leq|S_{\{D\}}| ∣ S { Σ x } ∣ ≤ ∣ S { D } ∣ ; equivalently ∣ Σ x S ∣ ≥ ∣ D S ∣ |\Sigma_x^S|\geq|D^S| ∣ Σ x S ∣ ≥ ∣ D S ∣ , i.e. no witness quadruple is built on D D D .
(b) For q = 3 q = 3 q = 3 (S ≅ PGL ( 2 , 8 ) S \cong\operatorname{PGL}(2,8) S ≅ PGL ( 2 , 8 ) , ∣ Ω ∣ = 28 |\Omega| = 28 ∣Ω∣ = 28 ) the framework of (a), implemented in GAP 4.16, returns a certificate for one representative of each of the six S S S -orbits on 3-subsets; see Table 1. The two block-contained orbits (∣ S { D } ∣ = 6 |S_{\{D\}}| = 6 ∣ S { D } ∣ = 6 ) admit a complement point with trivial four-point setwise stabilizer, and the external s = 3 s = 3 s = 3 and s = 2 s = 2 s = 2 orbits admit one with ∣ S { Σ x } ∣ = 2 |S_{\{\Sigma_x\}}| = 2 ∣ S { Σ x } ∣ = 2 , matching the bounds of (a); in particular no witness quadruple occurs for q = 3 q = 3 q = 3 .
(c) For the extension A = S ⋊ C 3 A = S \rtimes C_3 A = S ⋊ C 3 , ∣ C 3 ∣ = 3 |C_3| = 3 ∣ C 3 ∣ = 3 , put e Y = ∣ A { Y } ∣ / ∣ S { Y } ∣ ∈ { 1 , 3 } e_Y = |A_{\{Y\}}|/|S_{\{Y\}}| \in\{1,3\} e Y = ∣ A { Y } ∣/∣ S { Y } ∣ ∈ { 1 , 3 } for a subset Y Y Y . A certificate for the pair ( A , D ) (A,D) ( A , D ) is a point x x x with e Σ x ∣ S { Σ x } ∣ ≤ e D ∣ S { D } ∣ e_{\Sigma_x}|S_{\{\Sigma_x\}}| \le e_D|S_{\{D\}}| e Σ x ∣ S { Σ x } ∣ ≤ e D ∣ S { D } ∣ ; orbit representatives of the complement are then taken with respect to the A D A_D A D -orbits rather than the S D S_D S D -orbits, so that no representative of the same A A A -orbit is missed.
Lemma 2.24 (Extension stabilizer comparison). Let S = Ree ( q ) S = \operatorname{Ree}(q) S = Ree ( q ) act on the Ree unital Ω \Omega Ω , q = 3 2 m + 1 q = 3^{2m+1} q = 3 2 m + 1 , and let G = S ⋊ C d G = S \rtimes C_d G = S ⋊ C d with d ∣ 2 m + 1 d \mid2m+1 d ∣ 2 m + 1 (d = 3 d = 3 d = 3 for q = 27 q = 27 q = 27 , d = 5 d = 5 d = 5 for q = 243 q = 243 q = 243 ). For a subset Y ⊆ Ω Y \subseteq\Omega Y ⊆ Ω put e Y = [ G { Y } : S { Y } ] ∈ { 1 , d } e_Y = [G_{\{Y\}} : S_{\{Y\}}] \in\{1,d\} e Y = [ G { Y } : S { Y } ] ∈ { 1 , d } . Let D D D be an external triple with ∣ S { D } ∣ = s ∈ { 1 , 2 , 3 , 6 } |S_{\{D\}}| = s \in\{1,2,3,6\} ∣ S { D } ∣ = s ∈ { 1 , 2 , 3 , 6 } and let x ∉ D x \notin D x ∈ / D be a certificate point with S { Σ } = 1 S_{\{\Sigma\}} = 1 S { Σ } = 1 , Σ = D ∪ { x } \Sigma= D \cup\{x\} Σ = D ∪ { x } . Then ∣ G { Σ } ∣ ≤ ∣ G { D } ∣ |G_{\{\Sigma\}}| \le|G_{\{D\}}| ∣ G { Σ } ∣ ≤ ∣ G { D } ∣ ; more precisely ∣ G { Σ } ∣ = e Σ |G_{\{\Sigma\}}| = e_\Sigma ∣ G { Σ } ∣ = e Σ and ∣ G { D } ∣ = s e D |G_{\{D\}}| = s e_D ∣ G { D } ∣ = s e D , and the implication e Σ = d ⇒ e D = d e_\Sigma= d \Rightarrow e_D = d e Σ = d ⇒ e D = d holds, so that ∣ G { Σ } ∣ / ∣ G { D } ∣ ≤ 1 |G_{\{\Sigma\}}|/|G_{\{D\}}| \le1 ∣ G { Σ } ∣/∣ G { D } ∣ ≤ 1 .
Proof. Since S { Σ } = 1 S_{\{\Sigma\}} = 1 S { Σ } = 1 , every element of G { Σ } G_{\{\Sigma\}} G { Σ } has a unique decomposition g i σ g^i\sigma g i σ (σ \sigma σ the field automorphism of order d d d ) and e Σ = [ G { Σ } : 1 ] = ∣ G { Σ } ∣ e_\Sigma= [G_{\{\Sigma\}} : 1] = |G_{\{\Sigma\}}| e Σ = [ G { Σ } : 1 ] = ∣ G { Σ } ∣ ; similarly ∣ G { D } ∣ = ∣ S { D } ∣ e D = s e D |G_{\{D\}}| = |S_{\{D\}}|e_D = s e_D ∣ G { D } ∣ = ∣ S { D } ∣ e D = s e D . Assume e Σ = d e_\Sigma= d e Σ = d , and pick g ∈ G { Σ } ∖ S g \in G_{\{\Sigma\}} \setminus S g ∈ G { Σ } ∖ S . Then g d ∈ S { Σ } = 1 g^d \in S_{\{\Sigma\}} = 1 g d ∈ S { Σ } = 1 and g ≠ 1 g \ne1 g = 1 , so ∣ g ∣ = d |g| = d ∣ g ∣ = d . The permutation induced by g g g on the four points of Σ \Sigma Σ lies in S 4 S_4 S 4 and has order dividing d d d . Case d = 5 d = 5 d = 5 : S 4 S_4 S 4 has no element of order 5, so g g g fixes Σ \Sigma Σ pointwise, in particular D D D pointwise, whence g ∈ G { D } ∖ S g \in G_{\{D\}} \setminus S g ∈ G { D } ∖ S and e D = d e_D = d e D = d . Case d = 3 d = 3 d = 3 : the induced permutation is a 3-cycle, which fixes the fourth point of Σ \Sigma Σ and cycles the other three; on a three-subset D D D of Σ \Sigma Σ it either cycles D D D onto itself (3-cycle within D D D ) or swaps two points of D D D and fixes the third (3-cycle through the fourth point), and in both cases g ( D ) = D g(D) = D g ( D ) = D , whence again g ∈ G { D } ∖ S g \in G_{\{D\}} \setminus S g ∈ G { D } ∖ S and e D = d e_D = d e D = d . In both cases ∣ G { Σ } ∣ = d ≤ s d = ∣ G { D } ∣ |G_{\{\Sigma\}}| = d \le s d = |G_{\{D\}}| ∣ G { Σ } ∣ = d ≤ s d = ∣ G { D } ∣ since s ≥ 1 s \ge1 s ≥ 1 . If instead e Σ = 1 e_\Sigma= 1 e Σ = 1 then ∣ G { Σ } ∣ = 1 ≤ s e D |G_{\{\Sigma\}}| = 1 \le s e_D ∣ G { Σ } ∣ = 1 ≤ s e D trivially. □
Corollary 2.25 (Certificates survive field automorphisms). In the setting of Lemma 2.24, any S S S -level certificate point x x x (with S { Σ } = 1 S_{\{\Sigma\}} = 1 S { Σ } = 1 ) for an external triple D D D of any type s ∈ { 1 , 2 , 3 , 6 } s \in\{1,2,3,6\} s ∈ { 1 , 2 , 3 , 6 } is a G G G -level certificate: the orbit–stabilizer
Table 1 (PDF p. 34). Ree extension certificates for q = 3 q=3 q = 3 (S ≅ PGL ( 2 , 8 ) S\cong\operatorname{PGL}(2,8) S ≅ PGL ( 2 , 8 ) on 28 points; x x x the complement point found by GAP; the last column records the order of the induced image of S { Σ } S_{\{\Sigma\}} S { Σ } on the four points, as a subgroup of S 4 S_4 S 4 )
*formula applied in G G G gives ∣ Σ G ∣ ≥ ∣ D G ∣ |\Sigma^G|\geq|D^G| ∣ Σ G ∣ ≥ ∣ D G ∣ , so no witness quadruple is built on D D D in G G G . Indeed the four possibilities for ( e D , e Σ ) (e_D,e_\Sigma) ( e D , e Σ ) are
Table 5.
so every S S S -level certificate ∣ S { Σ } ∣ ≤ ∣ S { D } ∣ |S_{\{\Sigma\}}|\leq|S_{\{D\}}| ∣ S { Σ } ∣ ≤ ∣ S { D } ∣ lifts automatically: ∣ G { Σ } ∣ = e Σ ∣ S { Σ } ∣ ≤ e D ∣ S { D } ∣ = ∣ G { D } ∣ |G_{\{\Sigma\}}|=e_\Sigma|S_{\{\Sigma\}}|\leq e_D|S_{\{D\}}|=|G_{\{D\}}| ∣ G { Σ } ∣ = e Σ ∣ S { Σ } ∣ ≤ e D ∣ S { D } ∣ = ∣ G { D } ∣ . In particular, for the s = 1 , 2 , 3 s=1,2,3 s = 1 , 2 , 3 certificates of Proposition 2.22 (with S { Σ } = 1 S_{\{\Sigma\}}=1 S { Σ } = 1 ) no fourth point has to be discarded for turning bad under e D = 1 / e Σ = 5 e_D=1/e_\Sigma=5 e D = 1/ e Σ = 5 ; the comparison is independent of subfield descent or of the number of σ \sigma σ -fixed three-point orbits. Consequently the Ree( q ) (q) ( q ) family, including the extensions G = S ⋊ C d G=S\rtimes C_d G = S ⋊ C d , d ∣ 2 m + 1 d\mid2m+1 d ∣ 2 m + 1 (d = 3 d=3 d = 3 for q = 27 q=27 q = 27 , d = 5 d=5 d = 5 for q = 243 q=243 q = 243 ), admits no witness for ⋆ 3 \star_3 ⋆ 3 for every q = 3 2 m + 1 ≥ 27 q=3^{2m+1}\geq27 q = 3 2 m + 1 ≥ 27 .
The induced image of S { Σ } S_{\{\Sigma\}} S { Σ } on the four points is recorded as a subgroup of S 4 S_4 S 4 ; this diagnostic exposes errors in the coordinates or in the setwise-stabilizer computation. To summarize the boundary between theory and computation: the S 3 S_3 S 3 -decomposition of the external parameter set is given by Proposition 2.21 ; the existence of a certificate point for general q q q is established for the core s = 6 s=6 s = 6 case by the bad-point bound Proposition 2.22 ; for each tested parameter q q q the Ree extension certificate is verified by the finite stabilizer computation described in Proposition 2.23 , which covers the remaining external types s = 1 , 2 , 3 s=1,2,3 s = 1 , 2 , 3 as well.
For the tested parameter q = 27 q = 27 q = 27 the geometric signatures of the 108 external s = 6 s = 6 s = 6 orbits are recorded in the next proposition; they are the finite computations that separate the four stabilizer-spectrum types recorded in the verification section and that give the classification object for general q q q .
Proposition 2.26 (Geometric signatures of the s = 6 s = 6 s = 6 orbits for q = 27 q = 27 q = 27 ). Let S = 2 G 2 ( 27 ) S = {}^{2}G_{2}(27) S = 2 G 2 ( 27 ) act on the Ree unital Ω \Omega Ω with ∣ Ω ∣ = 19684 \lvert\Omega\rvert= 19684 ∣ Ω ∣ = 19684 , and let D = { d 1 , d 2 , d 3 } ⊂ Ω D = \{d_{1},d_{2},d_{3}\} \subset\Omega D = { d 1 , d 2 , d 3 } ⊂ Ω be a 3-subset of external S 3 S_{3} S 3 -type s = 6 s = 6 s = 6 . For each pair { d i , d j } \{d_{i},d_{j}\} { d i , d j } write U i j U_{ij} U ij for the unique Ree unital block through d i , d j d_{i},d_{j} d i , d j (∣ U i j ∣ = 28 \lvert U_{ij}\rvert= 28 ∣ U ij ∣ = 28 ). The following hold for every one of the 108 s = 6 s = 6 s = 6 orbits:
(a) Common three-point base. D D D is non-collinear, ∣ S { D } ∣ = 1 \lvert S_{\{D\}}\rvert= 1 ∣ S { D } ∣ = 1 , the image of S { D } S_{\{D\}} S { D } on D D D is trivial, and ∣ D S ∣ = ∣ S ∣ \lvert D^{S}\rvert= \lvert S\rvert ∣ D S ∣ = ∣ S ∣ .
(b) Block membership. No bad point lies on U 12 ∪ U 13 ∪ U 23 U_{12}\cup U_{13}\cup U_{23} U 12 ∪ U 13 ∪ U 23 : the normalized block-membership vector ϵ ( x ) = ( 1 x ∈ U 12 , 1 x ∈ U 13 , 1 x ∈ U 23 ) \epsilon(x) = (1_{x\in U_{12}},1_{x\in U_{13}},1_{x\in U_{23}}) ϵ ( x ) = ( 1 x ∈ U 12 , 1 x ∈ U 13 , 1 x ∈ U 23 ) equals ( 0 , 0 , 0 ) (0,0,0) ( 0 , 0 , 0 ) for all x ∈ B D f u l l x \in B_{D}^{\mathrm{full}} x ∈ B D full , and ∣ U 12 ∩ U 13 ∣ = ∣ U 12 ∩ U 23 ∣ = ∣ U 13 ∩ U 23 ∣ = 1 \lvert U_{12}\cap U_{13}\rvert= \lvert U_{12}\cap U_{23}\rvert= \lvert U_{13}\cap U_{23}\rvert= 1 ∣ U 12 ∩ U 13 ∣ = ∣ U 12 ∩ U 23 ∣ = ∣ U 13 ∩ U 23 ∣ = 1 with empty triple intersection.
(c) Fourth-point expansion signature. With b C 3 = ∣ { x ∉ D : ∣ S { D ∪ { x } } ∣ = 3 } ∣ b_{C_{3}} = \lvert\{x \notin D : \lvert S_{\{D\cup\{x\}\}}\rvert= 3\}\rvert b C 3 = ∣{ x ∈ / D : ∣ S { D ∪ { x }} ∣ = 3 }∣ and b V 4 = ∣ { x ∉ D : S { D ∪ { x } } ≅ V 4 } ∣ b_{V_{4}} = \lvert\{x \notin D : S_{\{D\cup\{x\}\}} \cong V_{4}\}\rvert b V 4 = ∣{ x ∈ / D : S { D ∪ { x }} ≅ V 4 }∣ (counted over B D f u l l B_{D}^{\mathrm{full}} B D full of Definition (14)), the 108 orbits split into four types with signatures
Table 6.
(d) Witness statistics. For the 8568 bad points in total the witness counts (number of transmission witnesses per bad point) are ( w 1 , w 2 , w 3 ) = ( 8430 , 12 , 126 ) (w_{1},w_{2},w_{3}) = (8430, 12, 126) ( w 1 , w 2 , w 3 ) = ( 8430 , 12 , 126 ) : every bad point of types II–IV has either one witness (in almost all cases a single involution; in a few classes an element of order exceeding 4) or three; the twelve type-I points of C 3 C_{3} C 3 -type have two witnesses, the two non-identity elements of the C 3 C_{3} C 3 -stabilizer, of order 3, each fixing one point of D D D and cycling the other two with x x x ; the V 4 V_{4} V 4 -type points have three witnesses, the three involutions of the V 4 V_{4} V 4 -stabilizer, each with no fixed point on D ∪ { x } D \cup\{x\} D ∪ { x } . The aggregate involution and
order-3 witness counts are ord 2 = 972 + 2916 + 3888 + 972 = 8748 \operatorname{ord}_2=972+2916+3888+972=8748 ord 2 = 972 + 2916 + 3888 + 972 = 8748 and ord 3 = 24 \operatorname{ord}_3=24 ord 3 = 24 , with no order-4 witnesses.
(e) Block-configuration criterion for C 3 C_3 C 3 -bad points. Order the six blocks through the six pairs of D ∪ { x } D\cup\{x\} D ∪ { x } as ( U 12 , U 13 , U 1 x , U 23 , U 2 x , U 3 x ) (U_{12},U_{13},U_{1x},U_{23},U_{2x},U_{3x}) ( U 12 , U 13 , U 1 x , U 23 , U 2 x , U 3 x ) . A point x ∉ D x\notin D x ∈ / D with D ∪ { x } D\cup\{x\} D ∪ { x } in general position is C 3 C_3 C 3 -bad precisely when S { D ∪ { x } } S_{\{D\cup\{x\}\}} S { D ∪ { x }} contains an element of order 3 whose induced permutation on these six blocks is [ 2 , 3 , 1 , 6 , 4 , 5 ] = ( 123 ) ( 465 ) [2,3,1,6,4,5]=(123)(465) [ 2 , 3 , 1 , 6 , 4 , 5 ] = ( 123 ) ( 465 ) : equivalently there is σ ∈ S { D ∪ { x } } \sigma\in S_{\{D\cup\{x\}\}} σ ∈ S { D ∪ { x }} with σ 3 = 1 \sigma^3=1 σ 3 = 1 , σ ( d 1 ) = d 1 \sigma(d_1)=d_1 σ ( d 1 ) = d 1 , and σ \sigma σ cycling { d 2 , d 3 , x } \{d_2,d_3,x\} { d 2 , d 3 , x } . The first 3-cycle consists of the three blocks through d 1 d_1 d 1 and the second of the complementary three blocks. Counting with the fixed orientation σ ( d 2 ) = d 3 \sigma(d_2)=d_3 σ ( d 2 ) = d 3 (so that σ \sigma σ and σ − 1 \sigma^{-1} σ − 1 are not double-counted),
b C 3 ( q ; D ) = # { x ∉ D : ∃ σ ∈ S , σ ( d 1 ) = d 1 , σ 3 = 1 , σ cycles { d 2 , d 3 , x } } . b_{C_3}(q;D)=\#\{x\notin D:\ \exists\,\sigma\in S,\ \sigma(d_1)=d_1,\ \sigma^3=1,\ \sigma\text{ cycles }\{d_2,d_3,x\}\}. b C 3 ( q ; D ) = # { x ∈ / D : ∃ σ ∈ S , σ ( d 1 ) = d 1 , σ 3 = 1 , σ cycles { d 2 , d 3 , x }} .
In terms of the permutation representation ρ Σ \rho_\Sigma ρ Σ on the ordered six blocks and the fixed orientation σ ( d 2 ) = d 3 \sigma(d_2)=d_3 σ ( d 2 ) = d 3 , with A ( D ) = { x ∉ D : D ∪ { x } in general position } \mathcal{A}(D)=\{x\notin D:D\cup\{x\}\text{ in general position}\} A ( D ) = { x ∈ / D : D ∪ { x } in general position } , this equals
b C 3 ( q ; D ) = ∑ x ∈ A ( D ) ε C 3 ( x ) = # { x ∈ A ( D ) : γ ∈ ρ Σ ( S { D ∪ { x } } ) } . \begin{aligned}
b_{C_3}(q;D)&=\sum_{x\in\mathcal{A}(D)}\varepsilon_{C_3}(x)\\
&=\#\{x\in\mathcal{A}(D):\ \gamma\in\rho_\Sigma(S_{\{D\cup\{x\}\}})\}.
\end{aligned} b C 3 ( q ; D ) = x ∈ A ( D ) ∑ ε C 3 ( x ) = # { x ∈ A ( D ) : γ ∈ ρ Σ ( S { D ∪ { x }} )} .
where γ = ( 123 ) ( 465 ) \gamma=(123)(465) γ = ( 123 ) ( 465 ) is the block permutation of the table in part (f); equivalently ε C 3 ( x ) = 1 \varepsilon_{C_3}(x)=1 ε C 3 ( x ) = 1 precisely when some σ ∈ S \sigma\in S σ ∈ S satisfies σ ( d 1 ) = d 1 \sigma(d_1)=d_1 σ ( d 1 ) = d 1 , σ ( d 2 ) = d 3 \sigma(d_2)=d_3 σ ( d 2 ) = d 3 , σ ( d 3 ) = x \sigma(d_3)=x σ ( d 3 ) = x ( σ \sigma σ realizes ( d 2 d 3 x ) (d_2\,d_3\,x) ( d 2 d 3 x ) on the four points and ( 123 ) ( 465 ) (123)(465) ( 123 ) ( 465 ) on the six blocks). Each admissible x x x is counted once, and the inverse block permutation γ − 1 \gamma^{-1} γ − 1 is not counted a second time (it belongs to the same C 3 C_3 C 3 ). In the generic case S { D } = 1 S_{\{D\}}=1 S { D } = 1 (s = 6 s=6 s = 6 ) the containment condition is already an exact count: since ( S { Σ } ) x ≤ S { D } = 1 (S_{\{\Sigma\}})_x\leq S_{\{D\}}=1 ( S { Σ } ) x ≤ S { D } = 1 and the action on the four points is faithful, ∣ S { Σ } ∣ = ∣ x S { Σ } ∣ ≤ 4 |S_{\{\Sigma\}}|=|x^{S_{\{\Sigma\}}}|\leq4 ∣ S { Σ } ∣ = ∣ x S { Σ } ∣ ≤ 4 ; a stabilizer whose block image contains γ \gamma γ has order divisible by 3 and at most 4, hence ∣ S { Σ } ∣ = 3 |S_{\{\Sigma\}}|=3 ∣ S { Σ } ∣ = 3 and the block image is exactly ⟨ γ ⟩ \langle\gamma\rangle ⟨ γ ⟩ . The four-point action of γ \gamma γ is d 1 ↦ d 1 d_1\mapsto d_1 d 1 ↦ d 1 , d 2 ↦ d 3 d_2\mapsto d_3 d 2 ↦ d 3 , d 3 ↦ x d_3\mapsto x d 3 ↦ x , x ↦ d 2 x\mapsto d_2 x ↦ d 2 , and γ − 1 = ( 132 ) ( 456 ) \gamma^{-1}=(132)(456) γ − 1 = ( 132 ) ( 456 ) is the other generator of the same C 3 C_3 C 3 , so each admissible x x x is counted once. For q = 27 q=27 q = 27 this criterion is verified for all 108 s = 6 s=6 s = 6 orbits: b C 3 ( q ; D ) ∈ { 0 , 1 } b_{C_3}(q;D)\in\{0,1\} b C 3 ( q ; D ) ∈ { 0 , 1 } , and b C 3 = 1 b_{C_3}=1 b C 3 = 1 exactly for the twelve type-I classes.
Block-configuration criterion for V 4 V_{4} V 4 -bad points
For a general-position Σ = D ∪ { x } \Sigma= D \cup\{x\} Σ = D ∪ { x } the three double transpositions τ 12 , 3 x \tau_{12,3x} τ 12 , 3 x , τ 13 , 2 x \tau_{13,2x} τ 13 , 2 x , τ 1 x , 23 \tau_{1x,23} τ 1 x , 23 on the four points induce on the ordered six blocks the permutations
Table 7.
each involution fixes its pair of disjoint blocks and swaps the remaining four blocks in two transpositions; the three block permutations commute pairwise and the product of the first two is the third, so together with the identity they form a V 4 V_{4} V 4 . A point x ∉ D x \notin D x ∈ / D with D ∪ { x } D \cup\{x\} D ∪ { x } in general position is V 4 V_{4} V 4 -bad precisely when S { D ∪ { x } } S_{\{D\cup\{x\}\}} S { D ∪ { x }} realizes the three double transpositions (equivalently, two distinct ones). In the generic case S { D } = 1 S_{\{D\}} = 1 S { D } = 1 (s = 6 s = 6 s = 6 ) containment is again exact: with Σ = D ∪ { x } \Sigma= D \cup\{x\} Σ = D ∪ { x } one has ( S { Σ } ) x ≤ S { D } = 1 (S_{\{\Sigma\}})_{x} \le S_{\{D\}} = 1 ( S { Σ } ) x ≤ S { D } = 1 and a faithful four-point action, so ∣ S { Σ } ∣ = ∣ x S { Σ } ∣ ≤ 4 |S_{\{\Sigma\}}| = |x^{S_{\{\Sigma\}}}| \le4 ∣ S { Σ } ∣ = ∣ x S { Σ } ∣ ≤ 4 . A stabilizer whose block image contains the two commuting double transpositions a = ( 25 ) ( 34 ) a = (25)(34) a = ( 25 ) ( 34 ) and b = ( 16 ) ( 34 ) b = (16)(34) b = ( 16 ) ( 34 ) then has order at least 4; the order is exactly 4, and the block image is exactly V 4 = { 1 , a , b , a b } V_{4} = \{1,a,b,ab\} V 4 = { 1 , a , b , ab } with a b = ( 16 ) ( 25 ) ab = (16)(25) ab = ( 16 ) ( 25 ) . Counting x x x with the fixed ordered representative D = [ 1 , 2 , 32 ] D = [1,2,32] D = [ 1 , 2 , 32 ] , each admissible x x x is counted once, not once per involution. For q = 27 q = 27 q = 27 the computed block permutations match the three arrays above for all 108 s = 6 s = 6 s = 6 orbits, and b V 4 ( q ; D ) ∈ { 0 , 1 , 3 } b_{V_{4}}(q;D) \in\{0,1,3\} b V 4 ( q ; D ) ∈ { 0 , 1 , 3 } with b V 4 = 1 , 3 b_{V_{4}} = 1,3 b V 4 = 1 , 3 exactly for the type-I, type-III and type-IV classes respectively. For general q q q the exact ordered count is
b V 4 ( q ; D ) = # { x ∈ A ( D ) : a , b ∈ ρ D ∪ { x } ( S { D ∪ { x } } ) } , b_{V_{4}}(q;D) = \#\{x \in A(D) : a,b \in\rho_{D\cup\{x\}}(S_{\{D\cup\{x\}\}})\}, b V 4 ( q ; D ) = # { x ∈ A ( D ) : a , b ∈ ρ D ∪ { x } ( S { D ∪ { x }} )} ,
where A ( D ) = { x ∉ D : D ∪ { x } in general position } A(D) = \{x \notin D : D \cup\{x\}\text{ in general position}\} A ( D ) = { x ∈ / D : D ∪ { x } in general position } and ρ Σ \rho_{\Sigma} ρ Σ is the permutation representation on the ordered six blocks; since ∣ S { D } ∣ = 1 |S_{\{D\}}| = 1 ∣ S { D } ∣ = 1 part (a) the representation ρ Σ \rho_{\Sigma} ρ Σ is faithful on S { D ∪ { x } } S_{\{D\cup\{x\}\}} S { D ∪ { x }} , so by the order bound above the containment a , b ∈ ρ Σ ( S { D ∪ { x } } ) a,b \in\rho_{\Sigma}(S_{\{D\cup\{x\}\}}) a , b ∈ ρ Σ ( S { D ∪ { x }} ) already forces the image to be exactly V 4 V_{4} V 4 (no larger stabilizer needs to be excluded).
Consequently the classification object for general q q q is the common non-collinear regular three-point base of external type s = 6 s = 6 s = 6 together with the fourth-point expansion signature ( b C 3 ( q ) , b V 4 ( q ) ) (b_{C_{3}}(q),b_{V_{4}}(q)) ( b C 3 ( q ) , b V 4 ( q )) and the witness distribution: b C 3 b_{C_{3}} b C 3 separates type I, and b V 4 ∈ { 0 , 1 , 3 } b_{V_{4}} \in\{0,1,3\} b V 4 ∈ { 0 , 1 , 3 } separates types II, III, IV.
PSp ( 2 n , 2 ) / PΓSp ( 2 n , 2 ) \operatorname{PSp}(2n, 2)/\operatorname{P\Gamma Sp}(2n, 2) PSp ( 2 n , 2 ) / PΓSp ( 2 n , 2 ) : as in the PSL ( n , q ) \operatorname{PSL}(n, q) PSL ( n , q ) case, with β \beta β in totally regular position on the half-quadric (not on any subspace through the reference points, not on any fixed-point subspace of the stabilizer). The excluded positions are finite in number and each lies in a proper subspace of the ambient space; moreover the rank-3 parameters of the half-quadric graphs [14 Thm. 1.1] give at least 2 2 n − 2 − 2 n − 1 2^{2n-2}-2^{n-1} 2 2 n − 2 − 2 n − 1 common non-neighbours of two adjacent points, exceeding the size of any fixed-point subspace of the stabilizer, so a totally regular β \beta β exists for n ≥ 4 n \ge4 n ≥ 4 . An element of S Σ S_{\Sigma} S Σ preserving O \mathcal{O} O setwise is a permutation composed with a diagonal scaling modulo scalars and fixes [ β ] [\beta] [ β ] only if it is trivial; an element mapping some reference point onto [ β ] [\beta] [ β ] , or [ β ] [\beta] [ β ] onto a reference point, forces β \beta β into one of the finitely many excluded families. Hence S Σ ≤ S O S_{\Sigma} \le S_{\mathcal{O}} S Σ ≤ S O (indeed S Σ = 1 S_{\Sigma} = 1 S Σ = 1 ); the rows n = 3 , 4 n = 3, 4 n = 3 , 4 (degrees 28 , 36 , 120 , 136 ≤ 200 28, 36, 120, 136 \le200 28 , 36 , 120 , 136 ≤ 200 ) are confirmed by the explicit Sp ( 6 , 2 ) \operatorname{Sp}(6, 2) Sp ( 6 , 2 ) computation and the census.
Theorem 2.27 (Killers for the symplectic quadratic-form actions). Let n ≥ 3 n \ge3 n ≥ 3 , and let G = Sp ( 2 n , 2 ) G = \operatorname{Sp}(2n, 2) G = Sp ( 2 n , 2 ) act on either Arf class of quadratic forms on V = F 2 2 n V = \mathbb{F}_2^{2n} V = F 2 2 n with a fixed nondegenerate polar form. For every 3-subset Δ \Delta Δ there exists a 4-subset Σ ⊃ Δ \Sigma\supset\Delta Σ ⊃ Δ with ∣ G Σ ∣ < ∣ G Δ ∣ |G_{\Sigma}| < |G_{\Delta}| ∣ G Σ ∣ < ∣ G Δ ∣ ; consequently neither of these two actions satisfies ⋆ 3 \star_3 ⋆ 3 .
Proof. Write Ω ε \Omega_{\varepsilon} Ω ε for the set of quadratic forms in the given Arf class with polar form B B B , and fix Q ∈ Ω ε Q \in\Omega_{\varepsilon} Q ∈ Ω ε . Every Q a ( x ) = Q ( x ) + B ( x , a ) Q_a(x) = Q(x) + B(x, a) Q a ( x ) = Q ( x ) + B ( x , a ) lies in Ω ε \Omega_{\varepsilon} Ω ε iff Q ( a ) = 0 Q(a) = 0 Q ( a ) = 0 , so a generic triple is Δ = { Q , Q a , Q b } \Delta= \{Q, Q_a, Q_b\} Δ = { Q , Q a , Q b } with Q ( a ) = Q ( b ) = 0 Q(a) = Q(b) = 0 Q ( a ) = Q ( b ) = 0 , a , b ≠ 0 a, b \ne0 a , b = 0 , a ≠ b a \ne b a = b . Put δ = B ( a , b ) \delta= B(a, b) δ = B ( a , b ) and K = G ( Δ ) K = G_{(\Delta)} K = G ( Δ ) . The image of G Δ G_{\Delta} G Δ on Δ \Delta Δ is S 3 S_3 S 3 , so ∣ G Δ ∣ = 6 ∣ K ∣ |G_{\Delta}| = 6|K| ∣ G Δ ∣ = 6∣ K ∣ .
We construct c c c and v v v such that
Q ( c ) = 0 , B ( a , c ) = 0 , B ( b , c ) = 1 , c ∉ ⟨ a , b ⟩ , and Q ( v ) = 1 , B ( a , v ) = B ( b , v ) = 0 , B ( c , v ) = 1. \begin{aligned}
Q(c) &= 0, & B(a,c) &= 0, & B(b,c) &= 1, & c &\notin\langle a,b\rangle,\\
\text{and}\quad Q(v) &= 1, & B(a,v) &= B(b,v) = 0,\\
B(c,v) &= 1.
\end{aligned} Q ( c ) and Q ( v ) B ( c , v ) = 0 , = 1 , = 1. B ( a , c ) B ( a , v ) = 0 , = B ( b , v ) = 0 , B ( b , c ) = 1 , c ∈ / ⟨ a , b ⟩ ,
If δ = 0 \delta= 0 δ = 0 then W = ⟨ a , b ⟩ W = \langle a, b\rangle W = ⟨ a , b ⟩ is totally singular; by nondegeneracy of B B B choose t t t with B ( a , t ) = 0 B(a, t) = 0 B ( a , t ) = 0 , B ( b , t ) = 1 B(b, t) = 1 B ( b , t ) = 1 , and set c = t c = t c = t if Q ( t ) = 0 Q(t) = 0 Q ( t ) = 0 and c = t + b c = t + b c = t + b if Q ( t ) = 1 Q(t) = 1 Q ( t ) = 1 . The space W ⊥ / W W^{\perp}/W W ⊥ / W has dimension 2 n − 4 ≥ 2 2n - 4 \ge2 2 n − 4 ≥ 2 and contains a vector of quadratic value 1; lifting to v ∈ W ⊥ v \in W^{\perp} v ∈ W ⊥ (replacing v v v by v + b v + b v + b if necessary) achieves Q ( v ) = 1 Q(v) = 1 Q ( v ) = 1 and B ( c , v ) = 1 B(c, v) = 1 B ( c , v ) = 1 . If δ = 1 \delta= 1 δ = 1 then W W W is a nondegenerate hyperbolic plane and W ⊥ W^{\perp} W ⊥ has dimension 2 n − 2 ≥ 4 2n - 2 \ge4 2 n − 2 ≥ 4 ; pick a nonzero singular u ∈ W ⊥ u \in W^{\perp} u ∈ W ⊥ and set c = a + u c = a + u c = a + u , then choose y ∈ W ⊥ y \in W^{\perp} y ∈ W ⊥ with B ( u , y ) = 1 B(u, y) = 1 B ( u , y ) = 1 and take v ∈ { y , y + u } v \in\{y, y + u\} v ∈ { y , y + u } with Q ( v ) = 1 Q(v) = 1 Q ( v ) = 1 ; again Q ( v ) = 1 Q(v) = 1 Q ( v ) = 1 , v ∈ W ⊥ v \in W^{\perp} v ∈ W ⊥ , B ( c , v ) = 1 B(c, v) = 1 B ( c , v ) = 1 . The orthogonal transvection t v ( x ) = x + B ( x , v ) v t_v(x)=x+B(x,v)v t v ( x ) = x + B ( x , v ) v preserves Q Q Q , fixes a a a and b b b , and moves c c c ; hence t v ∈ K t_v\in K t v ∈ K and Q c t v ≠ Q c Q_{c}^{t_v}\ne Q_c Q c t v = Q c . For Σ = { Q , Q a , Q b , Q c } \Sigma=\{Q,Q_a,Q_b,Q_c\} Σ = { Q , Q a , Q b , Q c } and L = G ( Σ ) L=G_{(\Sigma)} L = G ( Σ ) this gives L < K L<K L < K and ∣ L ∣ ≤ ∣ K ∣ / 2 |L|\le|K|/2 ∣ L ∣ ≤ ∣ K ∣/2 . The four 3-subsets of Σ \Sigma Σ carry the labels δ , 0 , 1 , δ + 1 \delta,0,1,\delta+1 δ , 0 , 1 , δ + 1 — exactly two labels 0 0 0 and two labels 1 1 1 — and G Σ G_\Sigma G Σ preserves the two classes, so its induced image lies in S 2 × S 2 S_2\times S_2 S 2 × S 2 , of order at most 4. Hence
∣ G Σ ∣ ≤ 4 ∣ L ∣ ≤ 2 ∣ K ∣ < 6 ∣ K ∣ = ∣ G Δ ∣ . |G_\Sigma|\le4|L|\le2|K|<6|K|=|G_\Delta|. ∣ G Σ ∣ ≤ 4∣ L ∣ ≤ 2∣ K ∣ < 6∣ K ∣ = ∣ G Δ ∣.
The claims ∣ G Δ ∣ = 6 ∣ K ∣ |G_\Delta|=6|K| ∣ G Δ ∣ = 6∣ K ∣ and t v ∈ K t_v\in K t v ∈ K are verified in GAP 4.16 for n = 3 , 4 n=3,4 n = 3 , 4 (degrees 28, 36 and 120, 136): every triple has a strict killer in both Arf classes.
By Lemmas 2.16 and 2.18 , ⋆ 3 \star_3 ⋆ 3 fails for every 2-transitive non-3-homogeneous almost simple socle of degree at least 8; the affine side is treated in Lemma 2.29 , and the closure of Conjecture 6.2.7 of [1 ] is completed in Corollary 2.30 .
Lemma 2.28 (Stabilizers in the natural module of G 2 ′ ( q ) G'_2(q) G 2 ′ ( q ) ). Let q = 2 f ≥ 4 q=2^f\ge4 q = 2 f ≥ 4 , let H = G 2 ′ ( q ) = G 2 ( q ) H=G'_2(q)=G_2(q) H = G 2 ′ ( q ) = G 2 ( q ) act on its natural six-dimensional module V = F q 6 V=\mathbb{F}_q^6 V = F q 6 and let h h h be the symplectic form inherited from Sp ( 6 , q ) \operatorname{Sp}(6,q) Sp ( 6 , q ) . For u ≠ 0 u\ne0 u = 0 and v ∉ ⟨ u ⟩ v\notin\langle u\rangle v ∈ / ⟨ u ⟩ write N ( u , v ) = H ( u , v ) N(u,v)=H_{(u,v)} N ( u , v ) = H ( u , v ) for the pointwise stabilizer of the pair.
(1) ∣ H u ∣ = q 6 ( q 2 − 1 ) |H_u|=q^6(q^2-1) ∣ H u ∣ = q 6 ( q 2 − 1 ) , and the H u H_u H u -orbits on V ∖ ⟨ u ⟩ V\setminus\langle u\rangle V ∖ ⟨ u ⟩ are given by the following table [7 ] :
Table 8 (PDF p. 39).
(2) For a hyperbolic pair (h ( u , v ) ≠ 0 h(u,v)\ne0 h ( u , v ) = 0 ) the pointwise stabilizer N ( u , v ) N(u,v) N ( u , v ) is isomorphic to SL ( 2 , q ) \operatorname{SL}(2,q) SL ( 2 , q ) ; it is the pointwise stabilizer of a hyperbolic plane in the standard A 1 A_1 A 1 -subsystem of G 2 ( q ) G_2(q) G 2 ( q ) . Under the N ( u , v ) N(u,v) N ( u , v ) -module decomposition
V = W ⊕ W ⊥ , W = ⟨ u , v ⟩ F q , V=W\oplus W^\perp,\qquad W=\langle u,v\rangle_{\mathbb{F}_q}, V = W ⊕ W ⊥ , W = ⟨ u , v ⟩ F q ,
N ( u , v ) N(u,v) N ( u , v ) acts trivially on W W W and as the simple 4-dimensional module L ( 3 ) L(3) L ( 3 ) on W ⊥ W^\perp W ⊥ ; in particular Fix V ( N ( u , v ) ) = W \operatorname{Fix}_V(N(u,v))=W Fix V ( N ( u , v )) = W .
::: (3) For a totally isotropic pair (h ( u , v ) = 0 h(u,v)=0 h ( u , v ) = 0 ) the pointwise stabilizer N ( u , v ) N(u,v) N ( u , v ) contains a root subgroup U α U_{\alpha} U α of order q q q fixing W W W pointwise: its fixed-point space H α = Fix V ( U α ) H_{\alpha}=\operatorname{Fix}_{V}(U_{\alpha}) H α = Fix V ( U α ) is a proper subspace containing W W W , and every point of V ∖ H α V\setminus H_{\alpha} V ∖ H α is moved by U α U_{\alpha} U α in orbits of length q q q .
Proof. (1) The stabilizer of a non-zero vector is H u ≅ q 5 : SL ( 2 , q ) H_{u}\cong q^{5}\!:\operatorname{SL}(2,q) H u ≅ q 5 : SL ( 2 , q ) and the three displayed orbit lengths on V ∖ ⟨ u ⟩ V\setminus\langle u\rangle V ∖ ⟨ u ⟩ are q 5 q^{5} q 5 , q ( q 2 − 1 ) q(q^{2}-1) q ( q 2 − 1 ) and q 3 ( q 2 − 1 ) q^{3}(q^{2}-1) q 3 ( q 2 − 1 ) [7 the natural-module orbit calculation for $G_{2}(2^{n})] ; the stabilizer orders follow from the orbit–stabilizer theorem and ∣ H u ∣ = q 6 ( q 2 − 1 ) |H_{u}|=q^{6}(q^{2}-1) ∣ H u ∣ = q 6 ( q 2 − 1 ) . (2) The pointwise stabilizer of the hyperbolic pair is the derived group of the pointwise stabilizer of the hyperbolic plane W W W ; in the A 1 A_{1} A 1 -subsystem picture it is the SL ( 2 , q ) \operatorname{SL}(2,q) SL ( 2 , q ) acting as L ( 1 ) ⊕ L ( 3 ) L(1)\oplus L(3) L ( 1 ) ⊕ L ( 3 ) on V = W ⊕ W ⊥ V=W\oplus W^{\perp} V = W ⊕ W ⊥ , where L ( 1 ) L(1) L ( 1 ) is the trivial module on W W W and L ( 3 ) L(3) L ( 3 ) the simple 4-dimensional F q SL ( 2 , q ) \mathbb{F}_{q}\operatorname{SL}(2,q) F q SL ( 2 , q ) -module [7 , 15 ] . Since L ( 3 ) L(3) L ( 3 ) has no non-zero trivial submodule, the fixed-point space is exactly W W W . (3) A root subgroup of G 2 ( q ) G_{2}(q) G 2 ( q ) has order q q q and fixes pointwise the subspace spanned by the weight spaces on which it acts trivially; since G 2 ( q ) G_{2}(q) G 2 ( q ) acts faithfully on V V V , this fixed space H α H_{\alpha} H α is a proper subspace, and it contains W W W because U α U_{\alpha} U α fixes W W W pointwise [7 ] . A point outside H α H_{\alpha} H α is not fixed by U α U_{\alpha} U α , and its U α U_{\alpha} U α -orbit has length a power of q q q dividing ∣ U α ∣ = q |U_{\alpha}|=q ∣ U α ∣ = q , hence exactly q q q .
Lemma 2.29 (Four-point witnesses in the affine groups). Let G = V ⋊ G 0 G=V\rtimes G_{0} G = V ⋊ G 0 be a 2-homogeneous affine group (V = F p d V=\mathbb{F}_{p}^{d} V = F p d , G 0 ≤ GL ( d , p ) G_{0}\leq\operatorname{GL}(d,p) G 0 ≤ GL ( d , p ) transitive on V ∖ { 0 } V\setminus\{0\} V ∖ { 0 } ), O = { 0 , u , v } \mathcal{O}=\{0,u,v\} O = { 0 , u , v } with u , v u,v u , v linearly independent, and m = ∣ G O ∣ m=|G_{\mathcal{O}}| m = ∣ G O ∣ . Then there is w ∈ V w\in V w ∈ V with ∣ G O ∪ { w } ∣ ≤ m |G_{\mathcal{O}\cup\{w\}}|\leq m ∣ G O ∪ { w } ∣ ≤ m .
Proof. Write the affine element as g ( x ) = A x + b g(x)=Ax+b g ( x ) = A x + b , A ∈ G 0 A\in G_{0} A ∈ G 0 , b ∈ V b\in V b ∈ V , and put Σ = O ∪ { w } \Sigma=\mathcal{O}\cup\{w\} Σ = O ∪ { w } . Let G ( O ) = { A ∈ G 0 : A ( u ) = u , A ( v ) = v } G_{(\mathcal{O})}=\{A\in G_{0}:A(u)=u,\ A(v)=v\} G ( O ) = { A ∈ G 0 : A ( u ) = u , A ( v ) = v } be the pointwise stabilizer of the triple and let e = ∣ G O ∣ ≤ 6 e=|G_{\mathcal{O}}|\leq6 e = ∣ G O ∣ ≤ 6 be the order of the permutation image on O \mathcal{O} O , so that m = ∣ G O ∣ = e ∣ G ( O ) ∣ m=|G_{\mathcal{O}}|=e|G_{(\mathcal{O})}| m = ∣ G O ∣ = e ∣ G ( O ) ∣ (Lemma 2.2(1)).
(1) The counting bound. By the orbit–stabilizer theorem applied to the action of the pointwise stabilizer on V V V ,
∣ G ( Σ ) ∣ = ∣ G ( O ) ∩ G ( w ) ∣ = ∣ G ( O ) ∣ / ℓ , ℓ = ∣ G ( O ) ⋅ w ∣ . |G_{(\Sigma)}|=|G_{(\mathcal{O})}\cap G_{(w)}|=|G_{(\mathcal{O})}|/\ell,\qquad
\ell=|G_{(\mathcal{O})}\cdot w|. ∣ G ( Σ ) ∣ = ∣ G ( O ) ∩ G ( w ) ∣ = ∣ G ( O ) ∣/ ℓ , ℓ = ∣ G ( O ) ⋅ w ∣.
so that
∣ G Σ ∣ ≤ ∣ Sym ( Σ ) ∣ ⋅ ∣ G ( Σ ) ∣ = 24 ∣ G ( O ) ∣ / ℓ , |G_{\Sigma}|\leq|\operatorname{Sym}(\Sigma)|\cdot|G_{(\Sigma)}|=24|G_{(\mathcal{O})}|/\ell, ∣ G Σ ∣ ≤ ∣ Sym ( Σ ) ∣ ⋅ ∣ G ( Σ ) ∣ = 24∣ G ( O ) ∣/ ℓ ,
and ∣ G Σ ∣ ≤ m = e ∣ G ( O ) ∣ |G_{\Sigma}|\leq m=e|G_{(\mathcal{O})}| ∣ G Σ ∣ ≤ m = e ∣ G ( O ) ∣ follows as soon as ℓ ≥ 24 / e \ell\geq24/e ℓ ≥ 24/ e . In (2)–(6) we produce such w w w in every family; the finitely many rows where the excluded set may cover all of V V V are confirmed by the GAP census of Section 3 (degree ≤ 200 \le200 ≤ 200 ); the dedicated exhaustive computations reported there confirm the remaining rows.
Collinear triples are excluded uniformly in every affine row of dimension d ≥ 2 d \ge2 d ≥ 2 . For O = { 0 , u , λ u } \mathcal{O} = \{0,u,\lambda u\} O = { 0 , u , λ u } with λ ∉ { 0 , 1 } \lambda\notin\{0,1\} λ ∈ / { 0 , 1 } , the pointwise stabilizer G ( O ) G_{(\mathcal{O})} G ( O ) fixes the line ⟨ u ⟩ \langle u\rangle ⟨ u ⟩ pointwise, and in each of the families AGL ( d , q ) \operatorname{AGL}(d,q) AGL ( d , q ) , ASL ( d , q ) \operatorname{ASL}(d,q) ASL ( d , q ) , Sp ( 2 m , q ) \operatorname{Sp}(2m,q) Sp ( 2 m , q ) and G 2 ′ ( q ) G'_2(q) G 2 ′ ( q ) it contains the root group
U u = { x ↦ x + c ℓ ( x ) u : c ∈ F q } , ℓ ∈ V ∗ , ℓ ( u ) = 0 , ℓ ( w ) ≠ 0 , U_u = \{x \mapsto x + c\ell(x)u : c \in\mathbb{F}_q\}, \qquad\ell\in V^*, \quad\ell(u) = 0, \quad\ell(w) \ne0, U u = { x ↦ x + c ℓ ( x ) u : c ∈ F q } , ℓ ∈ V ∗ , ℓ ( u ) = 0 , ℓ ( w ) = 0 ,
of order q q q (a shear group in GL / SL \operatorname{GL}/\operatorname{SL} GL / SL , a transvection group in Sp \operatorname{Sp} Sp , its analogue in G 2 ′ ( q ) G'_2(q) G 2 ′ ( q ) ), where ℓ \ell ℓ is a linear functional vanishing on ⟨ u ⟩ \langle u\rangle ⟨ u ⟩ and non-zero at w w w . Taking w ∉ ⟨ u ⟩ w \notin\langle u\rangle w ∈ / ⟨ u ⟩ with ℓ ( w ) ≠ 0 \ell(w) \ne0 ℓ ( w ) = 0 (all but finitely many w w w ) each element of U u U_u U u fixes 0 , u , λ u 0,u,\lambda u 0 , u , λ u pointwise and moves w w w into q q q distinct points, so ℓ ≥ q \ell\ge q ℓ ≥ q , and the orbit of the four-point set is at least q ≥ 2 q \ge2 q ≥ 2 times the orbit of the triple: every collinear triple fails the strict inequality of ⋆ 3 \star_3 ⋆ 3 . The 1-dimensional rows are handled in (7) below.
(2) The n = 2 n = 2 n = 2 rows ASL ( 2 , q ) \operatorname{ASL}(2,q) ASL ( 2 , q ) and AGL ( 2 , q ) \operatorname{AGL}(2,q) AGL ( 2 , q ) . Let O = { 0 , u , λ u } \mathcal{O} = \{0,u,\lambda u\} O = { 0 , u , λ u } be collinear. By Remark 2.3 the pointwise stabilizer contains the shear group U u U_u U u of order q q q . Take w w w off the line ⟨ u ⟩ \langle u\rangle ⟨ u ⟩ : since w w w is the unique point of Σ = O ∪ { w } \Sigma= \mathcal{O}\cup\{w\} Σ = O ∪ { w } outside that line, G Σ = G O ∩ G w G_\Sigma= G_{\mathcal{O}} \cap G_w G Σ = G O ∩ G w , and by the orbit–stabilizer theorem
∣ ( O ∪ { w } ) G ∣ ∣ O G ∣ = ∣ G O ∣ ∣ G Σ ∣ = ∣ w G O ∣ ≥ ∣ w U u ∣ = q , \frac{|(\mathcal{O}\cup\{w\})^G|}{|\mathcal{O}^G|} = \frac{|G_{\mathcal{O}}|}{|G_\Sigma|} = |w^{G_{\mathcal{O}}}| \ge|w^{U_u}| = q, ∣ O G ∣ ∣ ( O ∪ { w } ) G ∣ = ∣ G Σ ∣ ∣ G O ∣ = ∣ w G O ∣ ≥ ∣ w U u ∣ = q ,
so the orbit of the four-point set is at least q ≥ 2 q \ge2 q ≥ 2 times as long as the orbit of the triple; in particular every collinear triple fails the strict inequality required by ⋆ 3 \star_3 ⋆ 3 (cf. Lemma 2.16 ). For a non-collinear triple O = { 0 , u , v } \mathcal{O} = \{0,u,v\} O = { 0 , u , v } the pointwise stabilizer is trivial and m = ∣ G O ∣ = e m = |G_{\mathcal{O}}| = e m = ∣ G O ∣ = e equals the order of the permutation image: e = 3 e = 3 e = 3 for ASL ( 2 , q ) \operatorname{ASL}(2,q) ASL ( 2 , q ) with q q q odd (the 3-cycle ( 0 − 1 1 − 1 ) \begin{pmatrix}0 & -1 \\ 1 & -1\end{pmatrix} ( 0 1 − 1 − 1 ) has determinant 1 1 1 , a transposition has determinant − 1 -1 − 1 ), and e = 6 e = 6 e = 6 for AGL ( 2 , q ) \operatorname{AGL}(2,q) AGL ( 2 , q ) and for ASL ( 2 , q ) \operatorname{ASL}(2,q) ASL ( 2 , q ) with q q q even (where − 1 = 1 -1 = 1 − 1 = 1 ). For q q q odd take w = − u − v w = -u-v w = − u − v : fixing u u u and v v v fixes the basis, so the pointwise stabilizer of Σ \Sigma Σ is trivial, and a direct check in the basis ( u , v ) (u,v) ( u , v ) shows that no element of G Σ G_\Sigma G Σ moves 0 0 0 (the image set { b , A u + b , A v + b , − A u − A v + b } \{b,Au+b,Av+b,-Au-Av+b\} { b , A u + b , A v + b , − A u − A v + b } can equal Σ \Sigma Σ for b ∈ { u , v , − u − v } b \in\{u,v,-u-v\} b ∈ { u , v , − u − v } only if 4 ( u + v ) = 0 4(u+v) = 0 4 ( u + v ) = 0 resp. 4 u = 0 4u = 0 4 u = 0 , impossible in odd characteristic), so the image on Σ \Sigma Σ is C 3 C_3 C 3 resp. S 3 S_3 S 3 , giving ∣ G Σ ∣ = 3 = m |G_\Sigma| = 3 = m ∣ G Σ ∣ = 3 = m resp. 6 = m 6 = m 6 = m . For q q q even (the rows ASL( 2 , q ) (2,q) ( 2 , q ) = ASp( 2 , q ) (2,q) ( 2 , q ) and AGL( 2 , q ) (2,q) ( 2 , q ) , q ≥ 4 q \ge4 q ≥ 4 ) take w = u + t v w=u+tv w = u + t v with t ∉ { 0 , 1 } t \notin\{0,1\} t ∈ / { 0 , 1 } : fixing u , v u,v u , v fixes the basis, so the pointwise stabilizer of Σ \Sigma Σ is trivial. An element g ( x ) = A x + b g(x)=Ax+b g ( x ) = A x + b of G Σ G_{\Sigma} G Σ with b = 0 b=0 b = 0 is linear and permutes the three points { u , v , u + t v } \{u,v,u+tv\} { u , v , u + t v } ; the transformations u ↦ u + t v u\mapsto u+tv u ↦ u + t v , v ↦ v v\mapsto v v ↦ v and u ↔ v u\leftrightarrow v u ↔ v have linear parts of determinant 1 1 1 in characteristic 2 2 2 , so the image contains the full symmetric group on { u , v , u + t v } \{u,v,u+tv\} { u , v , u + t v } . If b ≠ 0 b\ne0 b = 0 then b ∈ { u , v , u + t v } b\in\{u,v,u+tv\} b ∈ { u , v , u + t v } , and the sum of the three images of u , v , u + t v u,v,u+tv u , v , u + t v under g g g (the linear parts cancel) equals 3 u = u 3u=u 3 u = u , 3 v = v 3v=v 3 v = v resp. 3 ( u + t v ) = u + t v 3(u+tv)=u+tv 3 ( u + t v ) = u + t v in characteristic 2 2 2 , while the corresponding sum over the complement of b b b in Σ \Sigma Σ equals u + ( 1 + t ) v u+(1+t)v u + ( 1 + t ) v , t v tv t v resp. u + v u+v u + v , forcing t = 1 t=1 t = 1 in each case; hence no element of G Σ G_{\Sigma} G Σ moves 0 0 0 for t ≠ 1 t\ne1 t = 1 . Consequently ∣ G Σ ∣ = 6 = m |G_{\Sigma}|=6=m ∣ G Σ ∣ = 6 = m in the rows ASL( 2 , q ) (2,q) ( 2 , q ) and AGL( 2 , q ) (2,q) ( 2 , q ) with q q q even. The exceptional group ASL( 2 , 2 ) ≅ (2,2)\cong ( 2 , 2 ) ≅ AGL( 2 , 2 ) ≅ S 4 (2,2)\cong S_4 ( 2 , 2 ) ≅ S 4 has degree 4 4 4 , outside the range ∣ P ∣ ≥ 8 |\mathcal{P}|\ge8 ∣ P ∣ ≥ 8 of Corollary 2.30 (there the three-point stabilizer has order 6 6 6 and the four-point stabilizer 24 24 24 , so the four-point orbit is one quarter of the three-point orbit). Thus in every n = 2 n=2 n = 2 row there is w w w with ∣ G O ∪ { w } ∣ ≤ m |G_{O\cup\{w\}}|\le m ∣ G O ∪ { w } ∣ ≤ m .
(3) The ASL( n , q ) (n,q) ( n , q ) rows (n ≥ 3 n\ge3 n ≥ 3 ). On a non-collinear triple the image contains the 3-cycle (C) of Lemma 2.2 (the linear part has determinant 1 1 1 in the basis ( u , v ) (u,v) ( u , v ) ), so e ≥ 3 e\ge3 e ≥ 3 ; for n ≥ 4 n\ge4 n ≥ 4 it also contains the transposition ( u v ) (u\,v) ( u v ) (a base permutation whose determinant is absorbed in a complementary block), hence e = 6 e=6 e = 6 . For n ≥ 4 n\ge4 n ≥ 4 take w w w in a complement W W W of ⟨ u , v ⟩ \langle u,v\rangle ⟨ u , v ⟩ : the pointwise stabilizer fixes u , v u,v u , v and acts on W W W through SL( n − 2 , q ) (n-2,q) ( n − 2 , q ) , which is transitive on W ∖ { 0 } W\setminus\{0\} W ∖ { 0 } , so ℓ ≥ q n − 2 − 1 ≥ 4 \ell\ge q^{n-2}-1\ge4 ℓ ≥ q n − 2 − 1 ≥ 4 (q ≥ 3 q\ge3 q ≥ 3 ; the rows q = 2 q=2 q = 2 , n = 4 n=4 n = 4 and n = 3 n=3 n = 3 , q ≤ 3 q\le3 q ≤ 3 have degrees 16 , 8 , 27 ≤ 200 16,8,27\le200 16 , 8 , 27 ≤ 200 ). For n = 3 n=3 n = 3 and q ≥ 4 q\ge4 q ≥ 4 take w = e 3 w=e_3 w = e 3 : the pointwise stabilizer of the four points is trivial, and the permutation image on Σ \Sigma Σ is contained in the alternating group A 4 A_4 A 4 : writing A A A for the linear part of an element and π \pi π for its permutation of the four points, a direct check in the basis ( e 1 , e 2 , e 3 ) (e_1,e_2,e_3) ( e 1 , e 2 , e 3 ) gives det A = sgn ( π ) \det A=\operatorname{sgn}(\pi) det A = sgn ( π ) , and det A = 1 \det A=1 det A = 1 in SL( 3 , q ) (3,q) ( 3 , q ) forces sgn ( π ) = 1 \operatorname{sgn}(\pi)=1 sgn ( π ) = 1 , so the image lies in A 4 A_4 A 4 and G Σ 2 ≤ A 4 G_{\Sigma}^2\le A_4 G Σ 2 ≤ A 4 , hence ∣ G Σ ∣ ≤ 12 |G_{\Sigma}|\le12 ∣ G Σ ∣ ≤ 12 . The pointwise stabilizer of u , v u,v u , v consists of the matrices ( 1 0 a 0 1 b 0 0 1 ) \begin{pmatrix}1&0&a\\0&1&b\\0&0&1\end{pmatrix} 1 0 0 0 1 0 a b 1 , a , b ∈ F q a,b\in\mathbb{F}_q a , b ∈ F q , so ∣ G ( O ) ∣ = q 2 |G_{(O)}|=q^2 ∣ G ( O ) ∣ = q 2 , m = e ∣ G ( O ) ∣ = 3 q 2 m=e|G_{(O)}|=3q^2 m = e ∣ G ( O ) ∣ = 3 q 2 , and ∣ G Σ ∣ ≤ 12 ≤ 3 q 2 = m |G_{\Sigma}|\le12\le3q^2=m ∣ G Σ ∣ ≤ 12 ≤ 3 q 2 = m for q ≥ 4 q\ge4 q ≥ 4 ; the rows ( 3 , 2 ) (3,2) ( 3 , 2 ) , ( 3 , 3 ) (3,3) ( 3 , 3 ) have degrees 8 , 27 ≤ 200 8,27\le200 8 , 27 ≤ 200 and are covered by the census of Section 3.
(4) The Sp( 2 m , q ) (2m,q) ( 2 m , q ) rows. These affine rows are 2-transitive for every q q q : Sp( 2 m , q ) (2m,q) ( 2 m , q ) acts transitively on V ∖ { 0 } V\setminus\{0\} V ∖ { 0 } , so G = V ⋊ G=V\rtimes G = V ⋊ Sp( V ) (V) ( V ) is 2-transitive, hence 2-homogeneous. We first consider q q q even: here − 1 = 1 -1=1 − 1 = 1 , so the transposition (T0) of Lemma 2.2 exists and e = 6 e=6 e = 6 . The argument depends on the type of the plane U = ⟨ u , v ⟩ U=\langle u,v\rangle U = ⟨ u , v ⟩ . If U U U is non-degenerate (i.e. β ( u , v ) ≠ 0 \beta(u,v)\ne0 β ( u , v ) = 0 , also the only possible type for q q q odd), take w w w in the orthogonal complement W = U ⊥ W=U^\perp W = U ⊥ : the pointwise stabilizer of u , v u,v u , v acts on W W W through Sp ( 2 m − 2 , q ) \operatorname{Sp}(2m-2,q) Sp ( 2 m − 2 , q ) , which is transitive on W ∖ { 0 } W\setminus\{0\} W ∖ { 0 } , so ℓ ≥ q 2 m − 2 − 1 \ell\ge q^{2m-2}-1 ℓ ≥ q 2 m − 2 − 1 ; for q q q even this is ≥ 4 \ge4 ≥ 4 (q ≥ 4 q\ge4 q ≥ 4 ; the row q = 2 q=2 q = 2 , m = 2 m=2 m = 2 has degree 16 ≤ 200 16\le200 16 ≤ 200 ), and for q q q odd, where e = 3 e=3 e = 3 on a non-degenerate plane, the required threshold is 8 and q 2 m − 2 − 1 ≥ 3 2 − 1 = 8 q^{2m-2}-1\ge3^2-1=8 q 2 m − 2 − 1 ≥ 3 2 − 1 = 8 . If U U U is totally isotropic (forced only for q q q even, β ( u , v ) = 0 \beta(u,v)=0 β ( u , v ) = 0 , U ⊆ U ⊥ U\subseteq U^\perp U ⊆ U ⊥ ) the complement argument does not apply, and instead take w ∈ X = { x ∈ V : β ( u , x ) = 1 , β ( v , x ) = 0 } w\in X=\{x\in V:\beta(u,x)=1,\ \beta(v,x)=0\} w ∈ X = { x ∈ V : β ( u , x ) = 1 , β ( v , x ) = 0 } : the two independent affine conditions cut out a hyperplane of codimension 2, so ∣ X ∣ = q 2 m − 2 |X|=q^{2m-2} ∣ X ∣ = q 2 m − 2 . The pointwise stabilizer H H H of u , v u,v u , v preserves X X X and is transitive on it: with a Witt decomposition V = U ⊕ W ⊕ U ′ V=U\oplus W\oplus U' V = U ⊕ W ⊕ U ′ (W W W a non-degenerate symplectic space of dimension 2 m − 4 2m-4 2 m − 4 ), H H H induces Sp ( W ) ≅ Sp ( 2 m − 4 , q ) \operatorname{Sp}(W)\cong\operatorname{Sp}(2m-4,q) Sp ( W ) ≅ Sp ( 2 m − 4 , q ) on the quotient classes, its nilpotent part realizes the translations x ↦ x + a x\mapsto x+a x ↦ x + a (a ∈ U a\in U a ∈ U ) inside each class, and Sp ( W ) \operatorname{Sp}(W) Sp ( W ) is transitive on W ∖ { 0 } W\setminus\{0\} W ∖ { 0 } ; hence ℓ = ∣ H ⋅ w ∣ = ∣ X ∣ = q 2 m − 2 ≥ 4 \ell=|H\cdot w|=|X|=q^{2m-2}\ge4 ℓ = ∣ H ⋅ w ∣ = ∣ X ∣ = q 2 m − 2 ≥ 4 for every m ≥ 2 m\ge2 m ≥ 2 (equality for q = 2 q=2 q = 2 , and ℓ ≥ 16 \ell\ge16 ℓ ≥ 16 for q ≥ 4 q\ge4 q ≥ 4 ), so no small-row census is needed in this subcase.
(5) The G 2 ′ ( q ) G'_2(q) G 2 ′ ( q ) row. For the affine row q 6 : G 2 ′ ( q ) q^6:G'_2(q) q 6 : G 2 ′ ( q ) with q = 2 f ≥ 4 q=2^f\ge4 q = 2 f ≥ 4 , where G 2 ′ ( q ) G'_2(q) G 2 ′ ( q ) denotes the derived subgroup of G 2 ( q ) G_2(q) G 2 ( q ) acting faithfully on the 6-dimensional module V = F q 6 V=\mathbb{F}_q^6 V = F q 6 (the smallest faithful module in even characteristic; G 2 ′ ( q ) = G 2 ( q ) G'_2(q)=G_2(q) G 2 ′ ( q ) = G 2 ( q ) for q ≥ 4 q\ge4 q ≥ 4 ), let O = { 0 , u , v } \mathcal{O}=\{0,u,v\} O = { 0 , u , v } with u , v u,v u , v linearly independent and N = G 2 ′ ( q ) ( O ) N=G'_2(q)(\mathcal{O}) N = G 2 ′ ( q ) ( O ) . If h ( u , v ) = 0 h(u,v)=0 h ( u , v ) = 0 (totally isotropic) then, by Lemma 2.283, the pointwise stabilizer N N N contains a root subgroup U α U_\alpha U α of order q q q fixing W W W pointwise, with fixed-point space H α ⊇ W H_\alpha\supseteq W H α ⊇ W a proper subspace; hence for w ∈ V ∖ H α w\in V\setminus H_\alpha w ∈ V ∖ H α (non-empty for q ≥ 4 q\ge4 q ≥ 4 ) one has ℓ = ∣ w N ∣ ≥ ∣ w U α ∣ = q ≥ 4 \ell=|w^N|\ge|w^{U_\alpha}|=q\ge4 ℓ = ∣ w N ∣ ≥ ∣ w U α ∣ = q ≥ 4 . The triple stabilizer induces Sym ( 3 ) \operatorname{Sym}(3) Sym ( 3 ) on O \mathcal{O} O (Lemma 2.281), so e = 6 e=6 e = 6 , and with ∣ N ∣ ∈ { q 5 , q 3 } |N|\in\{q^5,q^3\} ∣ N ∣ ∈ { q 5 , q 3 } ∣ G Σ ∣ ≤ 24 ∣ N ∣ / ℓ ≤ 24 ∣ N ∣ / q ≤ 6 ∣ N ∣ = m |G_\Sigma|\le24|N|/\ell\le24|N|/q\le6|N|=m ∣ G Σ ∣ ≤ 24∣ N ∣/ ℓ ≤ 24∣ N ∣/ q ≤ 6∣ N ∣ = m . If h ( u , v ) ≠ 0 h(u,v)\ne0 h ( u , v ) = 0 (hyperbolic), the maximal orbits are of this type: by Lemma 2.281 the totally isotropic triples have stabilizers 6 q 5 6q^5 6 q 5 or 6 q 3 6q^3 6 q 3 , strictly larger than 6 q ( q 2 − 1 ) 6q(q^2-1) 6 q ( q 2 − 1 ) (6 q 3 > 6 q ( q 2 − 1 ) 6q^3>6q(q^2-1) 6 q 3 > 6 q ( q 2 − 1 ) for q ≥ 2 q\ge2 q ≥ 2 ). For a hyperbolic triple Lemma 2.282 gives N ≅ SL ( 2 , q ) N\cong\operatorname{SL}(2,q) N ≅ SL ( 2 , q ) with ∣ N ∣ = q ( q 2 − 1 ) |N|=q(q^2-1) ∣ N ∣ = q ( q 2 − 1 ) and Fix V ( N ) = W \operatorname{Fix}_V(N)=W Fix V ( N ) = W . Since SL ( 2 , q ) \operatorname{SL}(2,q) SL ( 2 , q ) has no subgroup of index 2 or 3 for q ≥ 4 q\ge4 q ≥ 4 (an index-2 or index-3 subgroup would give a non-trivial permutation representation SL ( 2 , q ) → Sym ( 2 ) \operatorname{SL}(2,q)\to\operatorname{Sym}(2) SL ( 2 , q ) → Sym ( 2 ) or SL ( 2 , q ) → Sym ( 3 ) \operatorname{SL}(2,q)\to\operatorname{Sym}(3) SL ( 2 , q ) → Sym ( 3 ) ; SL ( 2 , q ) \operatorname{SL}(2,q) SL ( 2 , q ) is simple for q ≥ 4 q\ge4 q ≥ 4 in characteristic 2, and SL ( 2 , 4 ) ≅ Alt ( 5 ) \operatorname{SL}(2,4)\cong\operatorname{Alt}(5) SL ( 2 , 4 ) ≅ Alt ( 5 ) ), every w ∉ W w\notin W w ∈ / W has orbit length ℓ = ∣ w N ∣ = [ N : N w ] \ell=|w^N|=[N:N_w] ℓ = ∣ w N ∣ = [ N : N w ] equal to neither 1 (w ∉ W = Fix V ( N ) w\notin W=\operatorname{Fix}_V(N) w ∈ / W = Fix V ( N ) ) nor 2, 3 (no subgroup of index 2, 3); hence ℓ ≥ 4 \ell\ge4 ℓ ≥ 4 . The threshold of part (1) is ℓ ≥ 24 / e = 4 \ell\ge24/e=4 ℓ ≥ 24/ e = 4 (since e = 6 e=6 e = 6 on a hyperbolic triple), so ∣ G Σ ∣ ≤ 24 ∣ N ∣ / ℓ ≤ 6 ∣ N ∣ = m |G_\Sigma|\le24|N|/\ell\le6|N|=m ∣ G Σ ∣ ≤ 24∣ N ∣/ ℓ ≤ 6∣ N ∣ = m . The row q = 2 q=2 q = 2 has degree 2 6 = 64 ≤ 200 2^{6}=64\leq200 2 6 = 64 ≤ 200 and is covered by the census of Section 3.
(6) Rows handled by direct stabilizer computation. For the nearfield rows of degree 11 2 11^{2} 1 1 2 , 23 2 23^{2} 2 3 2 , 29 2 29^{2} 2 9 2 , 59 2 59^{2} 5 9 2 [10 ] (the exponent denotes the degree, i.e. the nearfield plane of order n 2 n^{2} n 2 ) and for the row 3 6 3^{6} 3 6 : SL ( 2 , 13 ) \operatorname{SL}(2,13) SL ( 2 , 13 ) , the exhaustive GAP computations of Section 3 directly enumerate, for a representative O = { 0 , u , v } \mathcal{O}=\{0,u,v\} O = { 0 , u , v } of a maximal triple orbit, the stabilizer G Σ G_{\Sigma} G Σ of every four-point set Σ = O ∪ { w } \Sigma=\mathcal{O}\cup\{w\} Σ = O ∪ { w } with w ∉ O w\notin\mathcal{O} w ∈ / O and confirm ∣ G Σ ∣ ≤ m = ∣ G O ∣ |G_{\Sigma}|\leq m=|G_{\mathcal{O}}| ∣ G Σ ∣ ≤ m = ∣ G O ∣ in every case; no appeal to the pointwise stabilizer of a pair is made.
(7) The 1-dimensional 2-homogeneous rows. These are the 2-homogeneous non-2-transitive affine groups of § 2.1 : G = V ⋊ G 0 ≤ AΓL ( 1 , q ) G=V\rtimes G_{0}\leq\operatorname{A\Gamma L}(1,q) G = V ⋊ G 0 ≤ AΓL ( 1 , q ) with ∣ G 0 ∣ |G_{0}| ∣ G 0 ∣ odd and q q q odd (Lemma 2.1). The row q = 3 q=3 q = 3 has degree $3$ and admits no fourth point; it is excluded here and is handled separately in Corollary 2.30 . First let G ≤ AGL ( 1 , q ) G\leq\operatorname{AGL}(1,q) G ≤ AGL ( 1 , q ) be linear. The stabilizer of a triple has odd order, hence its permutation image is trivial or a 3 3 3 -cycle. If char F q ≠ 3 \operatorname{char}\mathbb{F}_{q}\neq3 char F q = 3 , the triples whose stabilizer acts as a 3 3 3 -cycle are at most q ( q − 1 ) / 3 q(q-1)/3 q ( q − 1 ) /3 in number, fewer than ( q 3 ) \binom{q}{3} ( 3 q ) for q > 4 q>4 q > 4 , so some triple O \mathcal{O} O is regular: ∣ G O ∣ = 1 = m |G_{\mathcal{O}}|=1=m ∣ G O ∣ = 1 = m . For such O \mathcal{O} O at most six points w w w can make ∣ G Σ ∣ > 1 |G_{\Sigma}|>1 ∣ G Σ ∣ > 1 : a 3 3 3 -cycle on Σ \Sigma Σ fixes one of the three points of O \mathcal{O} O (it cannot fix w w w , else O \mathcal{O} O itself would be a 3 3 3 -cycle orbit) and its multiplier is one of the two primitive cube roots, so each of the 3 ⋅ 2 3\cdot2 3 ⋅ 2 pairs gives at most one excluded w w w . Since q ≥ 11 q\geq11 q ≥ 11 gives q − 3 > 6 q-3>6 q − 3 > 6 , a suitable w w w exists and ∣ G Σ ∣ = 1 = m |G_{\Sigma}|=1=m ∣ G Σ ∣ = 1 = m . The degree q = 5 q=5 q = 5 does not occur here: by the classification [13 ] the degree of a 2-homogeneous non-2-transitive affine group satisfies q ≡ 3 ( m o d 4 ) q\equiv3\pmod4 q ≡ 3 ( mod 4 ) [18 ] , while 5 ≡ 1 ( m o d 4 ) 5\equiv1\pmod4 5 ≡ 1 ( mod 4 ) . The row q = 7 q=7 q = 7 is checked directly: G 0 = { 1 , 2 , 4 } ≤ F 7 × G_{0}=\{1,2,4\}\leq\mathbb{F}_{7}^{\times} G 0 = { 1 , 2 , 4 } ≤ F 7 × , and for O = { 0 , 1 , 2 } \mathcal{O}=\{0,1,2\} O = { 0 , 1 , 2 } , w = 3 w=3 w = 3 , neither O \mathcal{O} O nor Σ \Sigma Σ is a 3 3 3 -cycle orbit or a fixed point with a 3 3 3 -cycle orbit, so both stabilizers are trivial. If char F q = 3 \operatorname{char}\mathbb{F}_{q}=3 char F q = 3 , an element of order 3 3 3 is a translation and has no fixed point, so it cannot stabilize a four-point set and every w w w works. Finally, in the semilinear rows (G 0 G_{0} G 0 contains a non-trivial field automorphism) the same argument applies for q ≥ 27 q\geq27 q ≥ 27 (the least semilinear degree; the degrees q = 25 , 49 q=25,49 q = 25 , 49 do not occur since 25 , 49 ≡ 1 ( m o d 4 ) 25,49\equiv1\pmod4 25 , 49 ≡ 1 ( mod 4 ) , and q = 7 q=7 q = 7 has no non-trivial automorphism). An element of order 3 3 3 in AΓL ( 1 , q ) \operatorname{A\Gamma L}(1,q) AΓL ( 1 , q ) has the form g ( x ) = a x σ + b g(x)=ax^{\sigma}+b g ( x ) = a x σ + b with field automorphism part σ \sigma σ of order 1 1 1 or 3 3 3 . Normalize the triple as O t = { 0 , 1 , t } \mathcal{O}_{t}=\{0,1,t\} O t = { 0 , 1 , t } with t ∈ F q ∖ { 0 , 1 } t\in\mathbb{F}_{q}\setminus\{0,1\} t ∈ F q ∖ { 0 , 1 } satisfying the forbidden-root condition ( ∗ ) (*) ( ∗ ) : t t t is not a root of ( t − 1 ) t σ = − 1 (t-1)t^{\sigma}=-1 ( t − 1 ) t σ = − 1 for any automorphism part σ \sigma σ of order $3, and t t t lies outside the proper fixed field of every such σ \sigma σ . (The existence of such t t t is proved below.) If g g g permutes O t \mathcal{O}_{t} O t as a 3 3 3 -cycle then, up to the direction of the cycle, g ( 0 ) = 1 g(0)=1 g ( 0 ) = 1 , g ( 1 ) = t g(1)=t g ( 1 ) = t , g ( t ) = 0 g(t)=0 g ( t ) = 0 , i.e. b = 1 , b=1, b = 1 , a = t − 1 a = t - 1 a = t − 1 and ( t − 1 ) t σ = − 1 (t - 1)t^{\sigma} = -1 ( t − 1 ) t σ = − 1 (the reverse direction gives the conjugate equation t σ + 1 = t − 1 t^{\sigma+1} = t - 1 t σ + 1 = t − 1 ). For σ = 1 \sigma= 1 σ = 1 this is t 2 − t + 1 = 0 t^2 - t + 1 = 0 t 2 − t + 1 = 0 , whose two roots are the primitive cube roots of unity. For σ ≠ 1 \sigma\ne1 σ = 1 of order d σ ≥ 3 d_{\sigma} \ge3 d σ ≥ 3 , writing σ ( x ) = x p e σ \sigma(x) = x^{p^{e_{\sigma}}} σ ( x ) = x p e σ with e σ = f / d σ e_{\sigma} = f/d_{\sigma} e σ = f / d σ , the equation ( t − 1 ) t σ = − 1 (t - 1)t^{\sigma} = -1 ( t − 1 ) t σ = − 1 has degree at most p e σ + 1 ≤ q 1 / d σ + 1 p^{e_{\sigma}} + 1 \le q^{1/d_{\sigma}} + 1 p e σ + 1 ≤ q 1/ d σ + 1 over F q \mathbb{F}_q F q , hence at most q 1 / d σ + 1 q^{1/d_{\sigma}} + 1 q 1/ d σ + 1 roots t ∈ F q ∖ { 0 , 1 } t \in\mathbb{F}_q \setminus\{0,1\} t ∈ F q ∖ { 0 , 1 } ; the parameters in the proper fixed field contribute at most q 1 / d σ − 1 q^{1/d_{\sigma}} - 1 q 1/ d σ − 1 further values each. Altogether the number of parameters t t t violating ( ∗ ) (*) ( ∗ ) is at most
2 + ∑ σ ≠ 1 ( ( q 1 / d σ + 1 ) + ( q 1 / d σ − 1 ) ) = 2 + 2 ∑ σ ≠ 1 q 1 / d σ < q − 2 , 2 + \sum_{\sigma\ne1} \left((q^{1/d_{\sigma}} + 1) + (q^{1/d_{\sigma}} - 1)\right) = 2 + 2\sum_{\sigma\ne1}q^{1/d_{\sigma}} < q - 2, 2 + σ = 1 ∑ ( ( q 1/ d σ + 1 ) + ( q 1/ d σ − 1 ) ) = 2 + 2 σ = 1 ∑ q 1/ d σ < q − 2 ,
so a parameter satisfying ( ∗ ) (*) ( ∗ ) exists for every q ≥ 27 q \ge27 q ≥ 27 (q = p f q = p^f q = p f : if f = 3 f = 3 f = 3 then the left side is 4 p + 2 < p 3 − 2 4p + 2 < p^3 - 2 4 p + 2 < p 3 − 2 for p ≥ 3 p \ge3 p ≥ 3 ; if f ≥ 5 f \ge5 f ≥ 5 then 2 ( f − 1 ) q 1 / 3 ≤ 2 ( f − 1 ) 3 f / 3 < 3 f − 4 ≤ q − 4 2(f - 1)q^{1/3} \le2(f - 1)3^{f/3} < 3^f - 4 \le q - 4 2 ( f − 1 ) q 1/3 ≤ 2 ( f − 1 ) 3 f /3 < 3 f − 4 ≤ q − 4 , in particular f q 1 / 3 < q − 3 fq^{1/3} < q - 3 f q 1/3 < q − 3 , and the borderline degrees are also covered by the census of Section 3). Hence some t t t gives a regular triple O t \mathcal{O}_t O t with ∣ G O t ∣ = 1 = m |G_{\mathcal{O}_t}| = 1 = m ∣ G O t ∣ = 1 = m . For such O \mathcal{O} O , a 3-cycle on Σ = O ∪ { w } \Sigma= \mathcal{O} \cup\{w\} Σ = O ∪ { w } is determined by its fixed point (one of the three points of O \mathcal{O} O ), its multiplier (one of the two primitive cube roots) and its automorphism part (identity or one of the two order-3 automorphisms), hence at most 3 ⋅ 2 ⋅ 3 = 18 3 \cdot2 \cdot3 = 18 3 ⋅ 2 ⋅ 3 = 18 points w w w are excluded; since q ≥ 27 q \ge27 q ≥ 27 gives q − 3 ≥ 24 > 18 q - 3 \ge24 > 18 q − 3 ≥ 24 > 18 , a suitable w w w exists and ∣ G Σ ∣ = 1 = m |G_{\Sigma}| = 1 = m ∣ G Σ ∣ = 1 = m . For the full group AΓL ( 1 , q ) \operatorname{A\Gamma L}(1,q) AΓL ( 1 , q ) itself — the 2-transitive branch d = f d = f d = f and the intermediate 2-homogeneous non-2-transitive groups with 2 ≤ d < f 2 \le d < f 2 ≤ d < f — the exhaustive GAP computations of Section 3 verify the failure of ⋆ 3 \star_3 ⋆ 3 directly: for every prime power q = p f > 50 q = p^f > 50 q = p f > 50 with q ≤ 2401 q \le2401 q ≤ 2401 and every divisor d ∣ f d \mid f d ∣ f , the subgroup G d = V ⋊ ( F q × ⋊ C d ) G_d = V \rtimes(\mathbb{F}_q^{\times} \rtimes C_d) G d = V ⋊ ( F q × ⋊ C d ) of AΓL ( 1 , q ) \operatorname{A\Gamma L}(1,q) AΓL ( 1 , q ) was constructed in GAP, and for a representative O \mathcal{O} O of a maximal triple orbit the inequality ∣ G Σ ∣ ≤ ∣ G O ∣ = m |G_{\Sigma}| \le|G_{\mathcal{O}}| = m ∣ G Σ ∣ ≤ ∣ G O ∣ = m was confirmed for some four-subset Σ = O ∪ { w } \Sigma= \mathcal{O} \cup\{w\} Σ = O ∪ { w } (the minimal computed values of ∣ G Σ ∣ |G_{\Sigma}| ∣ G Σ ∣ ranged over 1 , 2 , 3 , 4 , 6 , 8 , 9 1,2,3,4,6,8,9 1 , 2 , 3 , 4 , 6 , 8 , 9 , always at most m m m ; the degrees ≤ 50 \le50 ≤ 50 are covered by the census of Section 3). This covers the 2-transitive rows d = f d = f d = f including q = 2 f q = 2^f q = 2 f and all intermediate rows; the normal row d = 1 d = 1 d = 1 is covered by (1). Beyond q = 2401 q = 2401 q = 2401 the analytic arguments above apply, the excluded points being bounded in the sum form
∣ ⋃ σ ≠ 1 Fix ( σ ) ∣ ≤ ∑ σ ≠ 1 q 1 / ord ( σ ) : \left|\bigcup_{\sigma\ne1}\operatorname{Fix}(\sigma)\right| \le\sum_{\sigma\ne1}q^{1/\operatorname{ord}(\sigma)}: σ = 1 ⋃ Fix ( σ ) ≤ σ = 1 ∑ q 1/ ord ( σ ) :
for d = 2 d = 2 d = 2 there is no automorphism part of order 3 and the excluded points number at most 3 ⋅ 2 = 6 3 \cdot2 = 6 3 ⋅ 2 = 6 , against q − 3 ≥ 24 q - 3 \ge24 q − 3 ≥ 24 ; for d ≥ 3 d \ge3 d ≥ 3 the bound 3 ⋅ 2 ⋅ 3 = 18 3 \cdot2 \cdot3 = 18 3 ⋅ 2 ⋅ 3 = 18 applies, against q − 3 ≥ 24 > 18 q - 3 \ge24 > 18 q − 3 ≥ 24 > 18 . Thus every 1-dimensional row has a witnessing w w w .
Corollary 2.30 (Bradley’s Conjecture 6.2.7). Let G G G be a primitive permutation group on P \mathcal{P} P with ∣ P ∣ ≥ 8 |\mathcal{P}| \ge8 ∣ P ∣ ≥ 8 . Then G G G satisfies the S̆iemons–Wagner property ⋆ 3 \star_3 ⋆ 3 if and only if G ≅ PSL ( 2 , 7 ) G \cong\operatorname{PSL}(2,7) G ≅ PSL ( 2 , 7 ) or G ≅ PGL ( 2 , 7 ) G \cong\operatorname{PGL}(2,7) G ≅ PGL ( 2 , 7 ) .
Proof. If G G G is 3-homogeneous and satisfies ⋆ 3 \star_{3} ⋆ 3 , Bradley’s Theorem 6.2.5 [1 ] gives G ≅ PSL ( 2 , 7 ) G \cong\operatorname{PSL}(2,7) G ≅ PSL ( 2 , 7 ) or PGL ( 2 , 7 ) \operatorname{PGL}(2,7) PGL ( 2 , 7 ) . Suppose instead that G G G is not 3-homogeneous. By Lemma 2.5, a primitive group satisfying ⋆ 3 \star_{3} ⋆ 3 is 2-homogeneous; by Cameron’s classification of 2-homogeneous groups [3 ] the socle of G G G is then either affine or almost simple. If G G G is affine (whether 2-transitive or merely 2-homogeneous), let O \mathcal{O} O be any maximal triple orbit. By 2-homogeneity translate a representative of O \mathcal{O} O to { 0 , u , v } \{0,u,v\} { 0 , u , v } with u , v u,v u , v linearly independent (u , v u,v u , v independent since O \mathcal{O} O is maximal: in every family below a collinear triple has a strictly larger stabilizer, hence a strictly smaller orbit; in the G 2 ′ ( q ) G_{2}^{\prime}(q) G 2 ′ ( q ) row the totally isotropic triples likewise have stabilizers 6 q 5 6q^{5} 6 q 5 or 6 q 3 6q^{3} 6 q 3 , strictly larger than the hyperbolic value 6 q ( q 2 − 1 ) 6q(q^{2}-1) 6 q ( q 2 − 1 ) , by Lemma 2.28 (1), so the maximal orbit is hyperbolic). Lemma 2.29 applies to every such triple (its proof, parts (1)–(6), is uniform in the triple and does not depend on the orbit O \mathcal{O} O , so no transitivity hypothesis on G O G_{\mathcal{O}} G O is needed; in particular the assumption of Bradley’s Lemma 6.2.1 is not used): there is a fourth point w w w with ∣ G { 0 , u , v , w } ∣ ≤ ∣ G O ∣ \lvert G_{\{0,u,v,w\}}\rvert\leq\lvert G_{\mathcal{O}}\rvert ∣ G { 0 , u , v , w } ∣ ≤ ∣ G O ∣ , and pulling w w w back by the translating element gives a certificate point for O \mathcal{O} O itself. This covers every maximal orbit, hence every possible witness triple. (the lemma covers the 2-transitive affine families, namely AGL ( 1 , q ) \operatorname{AGL}(1,q) AGL ( 1 , q ) , ASL ( n , q ) \operatorname{ASL}(n,q) ASL ( n , q ) , ASp ( 2 m , q ) \operatorname{ASp}(2m,q) ASp ( 2 m , q ) , the nearfield rows and the exceptional Hering rows, as well as the 2-homogeneous non-2-transitive groups of § 2.1 ; the small rows of degree ≤ 200 \leq200 ≤ 200 are confirmed by the GAP census of Section 3). By Lemma 2.16 the group does not satisfy ⋆ 3 \star_{3} ⋆ 3 . If instead G G G is almost simple, then by § 2.1 every 2-homogeneous non-2-transitive group is affine, so G G G is in fact 2-transitive and its socle is one of the non-3-homogeneous families listed above, namely the small socles PSL ( 2 , 11 ) \operatorname{PSL}(2,11) PSL ( 2 , 11 ) , PSL ( 2 , 8 ) \operatorname{PSL}(2,8) PSL ( 2 , 8 ) , Alt ( 7 ) \operatorname{Alt}(7) Alt ( 7 ) , HS \operatorname{HS} HS , Co 3 \operatorname{Co}_{3} Co 3 having been excluded by [1 Prop. 6.2.6] ; for each of these socles Lemma 2.18 supplies a four-point witness for S S S ; the automorphism overgroup H = G / S H=G/S H = G / S leaves every S S S -orbit on triples setwise invariant (field automorphisms preserve the square-class and the conic/coset orbits, diagonal scalars act pointwise-fixedly, and for PSp ( 2 n , 2 ) \operatorname{PSp}(2n,2) PSp ( 2 n , 2 ) no diagonal extension exists), so Lemma 2.17 transmits the witness to G G G , and Lemma 2.16 shows ⋆ 3 \star_{3} ⋆ 3 fails. Hence no non-3-homogeneous primitive group satisfies ⋆ 3 \star_{3} ⋆ 3 ; G G G is 3-homogeneous and the claim follows.
Computational verification and the ⋆ 3 \star_3 ⋆ 3 census
To close Lemma 2.5 we record the exhaustive verification of the Siemons–Wagner property on primitive groups of small degree. The GAP script star3_census.g enumerates every primitive group of degree 8 ≤ v ≤ 30 8 \le v \le30 8 ≤ v ≤ 30 from the PRIMGRP library, computes the number σ 2 \sigma_2 σ 2 of orbits on pairs (the group is 2-homogeneous exactly when σ 2 = 1 \sigma_2 = 1 σ 2 = 1 ), and tests the defining inequality of ⋆ 3 \star_3 ⋆ 3 on every triple and every four-subset. The outcome is Table 3 : among the 209 primitive groups in the range, 68 fail 2-homogeneity and none of them satisfies ⋆ 3 \star_3 ⋆ 3 ; among the 141 2-homogeneous groups exactly two satisfy ⋆ 3 \star_3 ⋆ 3 , both of degree 8, namely PSL( 2 , 7 ) ≅ (2, 7) \cong ( 2 , 7 ) ≅ PSL( 3 , 2 ) (3, 2) ( 3 , 2 ) and PGL( 2 , 7 ) ≅ (2, 7) \cong ( 2 , 7 ) ≅ PSL( 3 , 2 ) : 2 (3, 2):2 ( 3 , 2 ) : 2 , in agreement with Bradley’s Theorem 6.2.5 in the 3-homogeneous case.
Beyond degree 30 the full quadruple enumeration of star3_census.g is too expensive, so the script R_rank3.g exploits the reduction of Lemma 2.5, Step 4 : by the rank-3 classification [14 , 16 , 17 ] , a primitive group failing 2-homogeneity and satisfying ⋆ 3 \star_3 ⋆ 3 would have to arise from one of the (C1)/(C2) configurations with at most two orbits on pairs. For every primitive group of degree 31 ≤ v ≤ 50 31 \le v \le50 31 ≤ v ≤ 50 the script therefore first tests σ 2 \sigma_2 σ 2 (rank 3 means exactly two orbits on pairs) and, for each rank-3 group and each 2-homogeneous group, verifies the defining inequality of ⋆ 3 \star_3 ⋆ 3 on every triple (four-subsets are handled by orbit representatives of G O G_O G O on the remaining points, not by enumeration). The outcome: in the range 31 ≤ v ≤ 50 31 \le v \le50 31 ≤ v ≤ 50 there are 176 primitive groups, of which 42 have rank 3 and 82 are 2-homogeneous, and none of the 42 rank-3 groups nor any of the 82 2-homogeneous groups satisfies ⋆ 3 \star_3 ⋆ 3 (per-degree data: v = 35 v = 35 v = 35 : 2 rank-3; v = 36 v = 36 v = 36 : 15 rank-3, 3 2-homogeneous; v = 37 v = 37 v = 37 : 2 rank-3; v = 40 v = 40 v = 40 : 4 rank-3; v = 41 v = 41 v = 41 : 1 rank-3; v = 45 v = 45 v = 45 : 4 rank-3; v = 49 v = 49 v = 49 : 12 rank-3, 11 2-homogeneous; v = 50 v = 50 v = 50 : 2 rank-3; the remaining degrees have no rank-3 groups). Combined with Table 3 , no primitive group of degree 8 ≤ v ≤ 50 8 \le v \le50 8 ≤ v ≤ 50 satisfies ⋆ 3 \star_3 ⋆ 3 except the two degree-8 groups PSL( 2 , 7 ) (2, 7) ( 2 , 7 ) , PGL( 2 , 7 ) (2, 7) ( 2 , 7 ) .
The rainbow case (σ 2 = 3 \sigma_2 = 3 σ 2 = 3 ). For the exclusion of a rainbow witness (Lemma 2.6 ) we verified the counting obstructions of Lemma 2.7 exhaustively on the primitive groups of degree 8 ≤ v ≤ 100 8 \le v \le100 8 ≤ v ≤ 100 with σ 2 = 3 \sigma_2 = 3 σ 2 = 3 . The scripts verify_sol_lemma.g and verify_sol_lemma_51100.g construct a representative D D D of a largest triple orbit, compute λ i \lambda_i λ i directly as the number of triples of T = D G T = D^G T = D G containing a fixed pair of the orbit O i O_i O i , and test the conditions (1)–(3) of Lemma 2.7 together with the elementary-abelian test of (iii). Among the 101 groups with σ 2 = 3 \sigma_2 = 3 σ 2 = 3 in this range (30 of degree ≤ 50 \le50 ≤ 50 and 71 of degree 51 ≤ v ≤ 100 51 \leq v \leq100 51 ≤ v ≤ 100 ), 99 fail the inequality (1) immediately; the two remaining groups 64 i 60 64\,i60 64 i 60 and 64 i 61 64\,i61 64 i 61 (∣ G ∣ = 322560 |G| = 322560 ∣ G ∣ = 322560 and 161280 161280 161280 ; λ = ( 20 , 24 , 24 ) \lambda= (20,24,24) λ = ( 20 , 24 , 24 ) and ( 24 , 24 , 20 ) (24,24,20) ( 24 , 24 , 20 ) ) pass (1)–(2) but violate the elementary-abelian test of (iii): their point-pair stabilizers K i K_i K i are not elementary abelian 2-groups and the condition (3) fails. Hence no primitive group of degree 8 ≤ v ≤ 100 8 \leq v \leq100 8 ≤ v ≤ 100 with σ 2 = 3 \sigma_2 = 3 σ 2 = 3 admits a rainbow witness. The intersection numbers R i = p j k i + p k j i R_i = p^i_{jk} + p^i_{kj} R i = p j k i + p k j i (computed from the directed two-point orbits, so that O i O_i O i and O i ∗ O_i^* O i ∗ are not conflated) satisfy λ i ≤ R i \lambda_i \leq R_i λ i ≤ R i for all 101 groups, and ∑ i R i < v \sum_i R_i < v ∑ i R i < v for 41 of them; the two residual groups 64 i 60 64\,i60 64 i 60 and 64 i 61 64\,i61 64 i 61 have ∑ i R i = 88 > 64 \sum_i R_i = 88 > 64 ∑ i R i = 88 > 64 and are excluded by the stabilizer structure above. These computations are finite-range certificates: they support, but do not by themselves constitute, a proof for unrestricted degree. For the groups with G D ≠ 1 G_D \neq1 G D = 1 (m > 1 m > 1 m > 1 ) the stronger obstruction u ≥ ∣ { w : r w ≥ 2 } ∣ u \geq|\{w : r_w \geq2\}| u ≥ ∣ { w : r w ≥ 2 } ∣ of Lemma 2.6 (vii) was verified to fail in every case (64 i 60 64\,i60 64 i 60 : u = 7 u = 7 u = 7 yet r w ≥ 2 r_w \geq2 r w ≥ 2 for all 61 points w ∉ D w \notin D w ∈ / D ; 52 i 1 52\,i1 52 i 1 : u = 2 u = 2 u = 2 yet r w = 2 r_w = 2 r w = 2 for all 49 points; and similarly for 56 i 6 56\,i6 56 i 6 , 81 i 138 81\,i138 81 i 138 etc.); and for the 26 groups of degree ≤ 50 \leq50 ≤ 50 with G D = 1 G_D = 1 G D = 1 the fixed-point obstruction (F) of Lemma 2.7 (vii) was confirmed: no involution τ ∈ K i \tau\in K_i τ ∈ K i fixes a point x x x with { a , b , x } ∈ T \{a,b,x\} \in T { a , b , x } ∈ T . The scripts and full output are kept with the submission package.
The family-wide closure of the two configurations (C1) and (C2) of Lemma 2.5, Step 4 is carried out by the four-point criterion Lemma 2.9: every rank-3 group that fails 2-homogeneity must, for every triple D D D , admit a fourth point w w w with ∣ G D ∪ { w } ∣ ≤ ∣ G D ∣ |G_{D\cup\{w\}}| \leq|G_D| ∣ G D ∪ { w } ∣ ≤ ∣ G D ∣ (the four-point inequality). The triangular subcase (C2) is excluded combinatorially inside Lemma 2.5 (the witness triple is then the unique triple of its pair type in the quadruple, whence G D ∪ { w } ≤ G D G_{D\cup\{w\}} \leq G_D G D ∪ { w } ≤ G D ); for the remaining (C1)-type witness the inequality is verified in closed form for the orthogonal nonsingular-point families Ω ( 2 m + 1 , 3 ) \Omega(2m+1,3) Ω ( 2 m + 1 , 3 ) and the projective-line actions of PSL ( n , q ) \operatorname{PSL}(n,q) PSL ( n , q ) , n ≥ 4 n \geq4 n ≥ 4 , in Propositions 2.11 and 2.12, for the singular-point and maximal-subspace actions in Propositions 2.13 and 2.14, and for the sporadic rank-3 representations in Proposition 2.15 (with the complete J 2 J_2 J 2 -computation and the Atlas rank checks quoted there). These cover all families of the rank-3 classification [14, Thm. 1.1, 1.2],[16 , 17 ] ; the remaining small degrees of that classification are part of the census above (all classical rank-3 actions of degree v ≤ 50 v \leq50 v ≤ 50 ). For each of the 68 non-2-homogeneous groups of degree 8 ≤ v ≤ 30 8 \leq v \leq30 8 ≤ v ≤ 30 the script displays a witnessing quadruple: a triple in a maximal triple orbit and a four-subset Σ \Sigma Σ containing it with ∣ Σ G ∣ ≥ ∣ D G ∣ |\Sigma^G| \geq|D^G| ∣ Σ G ∣ ≥ ∣ D G ∣ , so that the defining inequality of ⋆ 3 \star_3 ⋆ 3 fails (for A 5 A_5 A 5 of degree 10 and its overgroup S 5 S_5 S 5 , the maximal triple { 1 , 2 , 10 } \{1,2,10\} { 1 , 2 , 10 } has orbit length 30 while the quadruple
{ 1 , 2 , 3 , 10 } \{1, 2, 3, 10\} { 1 , 2 , 3 , 10 } has orbit length 60; for A 7 \mathrm{A}_7 A 7 of degree 21 the maximal triple { 1 , 2 , 16 } \{1, 2, 16\} { 1 , 2 , 16 } has orbit length 630 and the quadruple { 1 , 2 , 3 , 16 } \{1, 2, 3, 16\} { 1 , 2 , 3 , 16 } has orbit length 1260; similar witnesses are printed for every group in the range). The computations are documented in the GAP scripts and outputs distributed with this paper; the outputs record the GAP and PRIMGRP versions used.
Conclusion
Bradley’s conjecture is thereby settled: a primitive group of degree at least 8 satisfies ⋆ 3 \star_3 ⋆ 3 exactly when it is PSL ( 2 , 7 ) \operatorname{PSL}(2,7) PSL ( 2 , 7 ) or PGL ( 2 , 7 ) \operatorname{PGL}(2,7) PGL ( 2 , 7 ) (Corollary 2.30 ). The proof rests on three mechanisms, each of independent interest. (1) The reduction of Lemma 2.5: a primitive ⋆ 3 \star_{3} ⋆ 3 -group is 2-homogeneous. Its proof combines the rank-3 classification [14 , 16 , 17 ] with the exhaustive census of Section 3; we expect the same reduction, and the same census, to be useful for the higher orders ⋆ k \star_{k} ⋆ k , k ≥ 4 k \ge4 k ≥ 4 , where no classification is in sight. (2) The witness lemma 2.16 and the four-point witnesses of Lemmas 2.18 and 2.29 : these turn the defining inequality of ⋆ 3 \star_{3} ⋆ 3 into a finite stabilizer comparison in every 2-homogeneous family. (3) The 2-homogeneous classification [3 , 13 ] , which confines the candidates to the affine and almost simple families treated here.
Two natural questions remain open. First, the analogue of the Siemons– Wagner theorem for k ≥ 4 k \ge4 k ≥ 4 : which primitive groups satisfy ⋆ k \star_k ⋆ k ? The case k = 2 k = 2 k = 2 is the single group PSL ( 2 , 5 ) \operatorname{PSL}(2,5) PSL ( 2 , 5 ) of degree 6 [23 ] , and the present paper settles k = 3 k = 3 k = 3 ; for k ≥ 4 k \ge4 k ≥ 4 even the 3-homogeneous analogue of Bradley’s Theorem 6.2.5 is not known. Second, the role of primitivity: the imprimitive case was only partially classified by Siemons and Wagner [23 ] , and its k = 3 k = 3 k = 3 analogue is open.
Acknowledgements
I have made extensive use of the AI tool GPT6-Astra: it helped me draft and polish the exposition, write and debug the GAP scripts on which the census of Section 3 rests, and stress-test and refine parts of the argument (in particular the orbit-theoretic reductions and the classification-based exclusions) during the development of this work. When this work had long been stalled, it was GPT6-Astra that opened a new line of thought and helped break the impasse. In every case the output was read, checked and verified by myself before it entered the manuscript, and I take full responsibility for all of the mathematics presented here.
The colour table of a quadruple extension
Throughout, G ≤ Sym ( P ) G \le\operatorname{Sym}(\mathcal{P}) G ≤ Sym ( P ) has σ 2 ( G ) = 3 \sigma_2(G) = 3 σ 2 ( G ) = 3 with pair-orbits O 1 O_1 O 1 , O 2 O_2 O 2 , O 3 O_3 O 3 , and D = { 1 , 2 , 3 } \mathcal{D} = \{1,2,3\} D = { 1 , 2 , 3 } is a rainbow triple with χ ( 12 ) = 1 \chi(12) = 1 χ ( 12 ) = 1 , χ ( 13 ) = 2 \chi(13) = 2 χ ( 13 ) = 2 , χ ( 23 ) = 3 \chi(23) = 3 χ ( 23 ) = 3 . For w ∉ D w \notin\mathcal{D} w ∈ / D put
x = χ ( 1 w ) , y = χ ( 2 w ) , z = χ ( 3 w ) . x = \chi(1w), \qquad y = \chi(2w), \qquad z = \chi(3w). x = χ ( 1 w ) , y = χ ( 2 w ) , z = χ ( 3 w ) .
An extension triple is rainbow exactly as follows (a direct check of the three colours):
{ 1 , 2 , w } rainbow ⟺ ( x , y ) = ( 2 , 3 ) or ( 3 , 2 ) , \{1,2,w\}\text{ rainbow} \Longleftrightarrow(x,y) = (2,3)\text{ or }(3,2), { 1 , 2 , w } rainbow ⟺ ( x , y ) = ( 2 , 3 ) or ( 3 , 2 ) ,
{ 1 , 3 , w } rainbow ⟺ ( x , z ) = ( 1 , 3 ) or ( 3 , 1 ) , \{1,3,w\}\text{ rainbow} \Longleftrightarrow(x,z) = (1,3)\text{ or }(3,1), { 1 , 3 , w } rainbow ⟺ ( x , z ) = ( 1 , 3 ) or ( 3 , 1 ) ,
{ 2 , 3 , w } rainbow ⟺ ( y , z ) = ( 1 , 2 ) or ( 2 , 1 ) . \{2,3,w\}\text{ rainbow} \Longleftrightarrow(y,z) = (1,2)\text{ or }(2,1). { 2 , 3 , w } rainbow ⟺ ( y , z ) = ( 1 , 2 ) or ( 2 , 1 ) .
The full list of colour vectors ( x , y , z ) ∈ { 1 , 2 , 3 } 3 (x,y,z) \in\{1,2,3\}^3 ( x , y , z ) ∈ { 1 , 2 , 3 } 3 , displayed by x x x , with rows y y y and columns z z z , and entry the set of rainbow extension triples among A = A = A = { 1 , 2 , w } \{1,2,w\} { 1 , 2 , w } , B = { 1 , 3 , w } B=\{1,3,w\} B = { 1 , 3 , w } , C = { 2 , 3 , w } C=\{2,3,w\} C = { 2 , 3 , w } :
x = 1 : z = 1 z = 2 z = 3 y = 1 ∅ C B y = 2 C ∅ B y = 3 ∅ ∅ B \begin{array}{c|ccc}
x=1: & z=1 & z=2 & z=3\\
\hline
y=1 & \varnothing& C & B\\
y=2 & C & \varnothing& B\\
y=3 & \varnothing& \varnothing& B
\end{array} x = 1 : y = 1 y = 2 y = 3 z = 1 ∅ C ∅ z = 2 C ∅ ∅ z = 3 B B B
x = 2 : z = 1 z = 2 z = 3 y = 1 ∅ C ∅ y = 2 C ∅ ∅ y = 3 A A A \begin{array}{c|ccc}
x=2: & z=1 & z=2 & z=3\\
\hline
y=1 & \varnothing& C & \varnothing\\
y=2 & C & \varnothing& \varnothing\\
y=3 & A & A & A
\end{array} x = 2 : y = 1 y = 2 y = 3 z = 1 ∅ C A z = 2 C ∅ A z = 3 ∅ ∅ A
x = 3 : z = 1 z = 2 z = 3 y = 1 B C ∅ y = 2 A B C A A y = 3 B ∅ ∅ \begin{array}{c|ccc}
x=3: & z=1 & z=2 & z=3\\
\hline
y=1 & B & C & \varnothing\\
y=2 & ABC & A & A\\
y=3 & B & \varnothing& \varnothing
\end{array} x = 3 : y = 1 y = 2 y = 3 z = 1 B A B C B z = 2 C A ∅ z = 3 ∅ A ∅
In particular the unique vector with three rainbow extensions is ( x , y , z ) = ( 3 , 2 , 1 ) (x,y,z)=(3,2,1) ( x , y , z ) = ( 3 , 2 , 1 ) , for which the four faces of D ∪ { w } \mathcal{D}\cup\{w\} D ∪ { w } form a rainbow K 4 K_4 K 4 with 12 , 34 ∈ O 1 12,34\in O_1 12 , 34 ∈ O 1 , 13 , 24 ∈ O 2 13,24\in O_2 13 , 24 ∈ O 2 , 14 , 23 ∈ O 3 14,23\in O_3 14 , 23 ∈ O 3 ; every other vector has at most one rainbow extension. The colour-preserving group of the rainbow K 4 K_4 K 4 is V 4 = ⟨ ( 1 2 ) ( 3 4 ) , ( 1 3 ) ( 2 4 ) ⟩ = { 1 , ( 1 2 ) ( 3 4 ) , ( 1 3 ) ( 2 4 ) , ( 1 4 ) ( 2 3 ) } V_4=\langle(1\,2)(3\,4),(1\,3)(2\,4)\rangle=\{1,(1\,2)(3\,4),(1\,3)(2\,4),(1\,4)(2\,3)\} V 4 = ⟨( 1 2 ) ( 3 4 ) , ( 1 3 ) ( 2 4 )⟩ = { 1 , ( 1 2 ) ( 3 4 ) , ( 1 3 ) ( 2 4 ) , ( 1 4 ) ( 2 3 )} , acting regularly on the four faces.
The affine families: a uniform four-point criterion
The counting criterion of Lemma 2.29 treats the 2 2 2 -homogeneous affine rows uniformly; the present appendix develops the same exclusion by elementary localizations — collinear triples (Theorem Appendix B.2), affinely independent triples with p ≥ 5 p\geq5 p ≥ 5 (Theorem Appendix B.3), the p = 3 p=3 p = 3 localization (Theorem Appendix B.6) and the exhaustive case p = 2 p=2 p = 2 , d = 3 d=3 d = 3 — and records the structural statements needed for the framework of Remark Appendix B.22.
Throughout this appendix G = V ⋊ H ≤ AGL ( d , p ) G=V\rtimes H\leq\operatorname{AGL}(d,p) G = V ⋊ H ≤ AGL ( d , p ) is an affine permutation group on the vector space V = F p d V=\mathbb{F}_p^d V = F p d , H ≤ GL ( d , p ) H\leq\operatorname{GL}(d,p) H ≤ GL ( d , p ) , and Δ = { 0 , u , v } ⊂ V \Delta=\{0,u,v\}\subset V Δ = { 0 , u , v } ⊂ V is a three-point subset. For w ∉ Δ w\notin\Delta w ∈ / Δ put
K w = G Δ ∪ { w } , J w = G Δ , w = G Δ ∩ K w , K_w=G_{\Delta\cup\{w\}},\qquad J_w=G_{\Delta,w}=G_\Delta\cap K_w, K w = G Δ ∪ { w } , J w = G Δ , w = G Δ ∩ K w ,
r w = [ G Δ : J w ] , t w = [ K w : J w ] . r_w=[G_\Delta:J_w],\qquad t_w=[K_w:J_w]. r w = [ G Δ : J w ] , t w = [ K w : J w ] .
Since J w = G Δ ∩ K w J_w=G_\Delta\cap K_w J w = G Δ ∩ K w and t w ≤ 4 t_w\le4 t w ≤ 4 (the permutation image of K w K_w K w on the four points Δ ∪ { w } \Delta\cup\{w\} Δ ∪ { w } lies in S 4 S_4 S 4 ), we have
∣ K w ∣ = t w ∣ J w ∣ ≤ r w ∣ J w ∣ = ∣ G Δ ∣ ⟺ r w ≥ t w . (22) |K_w|=t_w|J_w|\le r_w|J_w|=|G_\Delta|\Longleftrightarrow r_w\ge t_w. \tag*{(22)} ∣ K w ∣ = t w ∣ J w ∣ ≤ r w ∣ J w ∣ = ∣ G Δ ∣ ⟺ r w ≥ t w . (22)
By the witness lemma 2.16, Δ \Delta Δ is not a ⋆ 3 \star_3 ⋆ 3 -witness as soon as some w ∉ Δ w\notin\Delta w ∈ / Δ satisfies (B.1); in particular r w ≥ 4 r_w\ge4 r w ≥ 4 always suffices, because t w ≤ 4 t_w\le4 t w ≤ 4 .
Lemma Appendix B.1 (Orbit–ratio criterion). Let G ≤ Sym ( V ) G\le\operatorname{Sym}(V) G ≤ Sym ( V ) be affine primitive on V = F p d V=\mathbb{F}_p^d V = F p d as above, let Δ = { 0 , u , v } \Delta=\{0,u,v\} Δ = { 0 , u , v } be a three-point subset and w ∉ Δ w\notin\Delta w ∈ / Δ ; write S = Δ ∪ { w } S=\Delta\cup\{w\} S = Δ ∪ { w } , K = G Δ K=G_\Delta K = G Δ and C = G S C=G_S C = G S . Then
(i) J : = K ∩ C = K w = G Δ , w J:=K\cap C=K_w=G_{\Delta,w} J := K ∩ C = K w = G Δ , w , the stabilizer in K K K of the point w w w ;
(ii) with t ( w ) : = ∣ w K ∣ = ∣ K : J ∣ t(w):=|w^K|=|K:J| t ( w ) := ∣ w K ∣ = ∣ K : J ∣ and r ( w ) : = ∣ Δ C ∣ = ∣ C : J ∣ r(w):=|\Delta^C|=|C:J| r ( w ) := ∣ Δ C ∣ = ∣ C : J ∣ one has ∣ C ∣ / ∣ K ∣ = r ( w ) / t ( w ) |C|/|K|=r(w)/t(w) ∣ C ∣/∣ K ∣ = r ( w ) / t ( w ) , and in particular ∣ C ∣ ≤ ∣ K ∣ ⟺ r ( w ) ≤ t ( w ) |C|\le|K|\Longleftrightarrow r(w)\le t(w) ∣ C ∣ ≤ ∣ K ∣ ⟺ r ( w ) ≤ t ( w ) ;
(iii) r ( w ) ∈ { 1 , 2 , 3 , 4 } r(w)\in\{1,2,3,4\} r ( w ) ∈ { 1 , 2 , 3 , 4 } : r ( w ) r(w) r ( w ) is the size of the orbit of the three-subset Δ \Delta Δ under the induced action of C C C on the four three-subsets of S S S .
Consequently, by the witness lemma 2.16, Δ \Delta Δ fails to be a ⋆ 3 \star_3 ⋆ 3 -witness as soon as some w ∉ Δ w\notin\Delta w ∈ / Δ satisfies r ( w ) ≤ t ( w ) r(w)\le t(w) r ( w ) ≤ t ( w ) , while Δ \Delta Δ is a witness exactly when t ( w ) < r ( w ) ≤ 4 t(w)<r(w)\le4 t ( w ) < r ( w ) ≤ 4 for every w ∉ Δ w\notin\Delta w ∈ / Δ .
Proof. (i) If g ∈ K ∩ C g\in K\cap C g ∈ K ∩ C , then g g g stabilizes Δ \Delta Δ and S S S , hence fixes the unique point w ∈ S ∖ Δ w\in S\setminus\Delta w ∈ S ∖ Δ , so g ∈ K w g\in K_w g ∈ K w ; conversely g ∈ K w g\in K_w g ∈ K w stabilizes Δ \Delta Δ and fixes w w w , hence stabilizes S S S , so g ∈ K ∩ C g\in K\cap C g ∈ K ∩ C . (ii) By the orbit–stabilizer theorem t ( w ) = ∣ w K ∣ = ∣ K : K w ∣ = ∣ K : J ∣ t(w)=|w^K|=|K:K_w|=|K:J| t ( w ) = ∣ w K ∣ = ∣ K : K w ∣ = ∣ K : J ∣ , and since C Δ = C ∩ G Δ = C ∩ K = J C_\Delta=C\cap G_\Delta=C\cap K=J C Δ = C ∩ G Δ = C ∩ K = J , r ( w ) = ∣ Δ C ∣ = ∣ C : C Δ ∣ = ∣ C : J ∣ r(w)=|\Delta^C|=|C:C_\Delta|=|C:J| r ( w ) = ∣ Δ C ∣ = ∣ C : C Δ ∣ = ∣ C : J ∣ ; dividing gives ∣ C ∣ / ∣ K ∣ = ∣ C : J ∣ / ∣ K : J ∣ = r ( w ) / t ( w ) |C|/|K|=|C:J|/|K:J|=r(w)/t(w) ∣ C ∣/∣ K ∣ = ∣ C : J ∣/∣ K : J ∣ = r ( w ) / t ( w ) . (iii) C C C acts on the four three-subsets of S S S , and Δ C \Delta^C Δ C is one of its orbits, so r ( w ) ≤ 4 r(w)\le4 r ( w ) ≤ 4 . The final sentence restates (B.1) together with Lemma 2.16 and the transitivity of G G G on points, which gives ∣ Δ G ∣ = ∣ G ∣ / ∣ K ∣ |\Delta^G|=|G|/|K| ∣ Δ G ∣ = ∣ G ∣/∣ K ∣ and ∣ S G ∣ = ∣ G ∣ / ∣ C ∣ |S^G|=|G|/|C| ∣ S G ∣ = ∣ G ∣/∣ C ∣ .
Collinear triples in every characteristic
Assume { 0 , u , v } \{0,u,v\} { 0 , u , v } is contained in a line, so v = λ u v=\lambda u v = λ u for some λ ∈ F p × ∖ { 1 } \lambda\in\mathbb{F}_p^\times\setminus\{1\} λ ∈ F p × ∖ { 1 } . Since d ≥ 2 d\ge2 d ≥ 2 there exists w ∉ ⟨ u ⟩ w\notin\langle u\rangle w ∈ / ⟨ u ⟩ . Then Δ \Delta Δ is the unique collinear three-point subset of Δ ∪ { w } \Delta\cup\{w\} Δ ∪ { w } : any affine line meets Δ ∪ { w } \Delta\cup\{w\} Δ ∪ { w } in at most the three points 0 , u , λ u 0,u,\lambda u 0 , u , λ u . Hence every g ∈ K w = G Δ ∪ { w } g\in K_w=G_{\Delta\cup\{w\}} g ∈ K w = G Δ ∪ { w } preserves Δ \Delta Δ setwise and fixes w w w , so K w ≤ G Δ K_w\le G_\Delta K w ≤ G Δ and t w = 1 t_w=1 t w = 1 . By (B.1), Δ \Delta Δ is not a witness.
Theorem Appendix B.2 (Collinear triples). If d ≥ 2 d\ge2 d ≥ 2 and Δ \Delta Δ is collinear, then Δ \Delta Δ is not a ⋆ 3 \star_3 ⋆ 3 -witness for any affine group G = V ⋊ H ≤ AGL ( d , p ) G=V\rtimes H\le\operatorname{AGL}(d,p) G = V ⋊ H ≤ AGL ( d , p ) .
Affinely independent triples with p ≥ 5 p \ge5 p ≥ 5
Assume now that 0 , u , v 0,u,v 0 , u , v are affinely independent, so u , v u,v u , v are linearly independent, and p ≥ 5 p \ge5 p ≥ 5 . Choose
w = u + v 3 ∈ V ∖ { 0 , u , v } . w = \frac{u+v}{3} \in V \setminus\{0,u,v\}. w = 3 u + v ∈ V ∖ { 0 , u , v } .
The four points 0 , u , v , w 0,u,v,w 0 , u , v , w have the unique affine dependence
0 + u + v − 3 w = 0 , 0+u+v-3w=0, 0 + u + v − 3 w = 0 ,
whose coefficient vector ( 1 , 1 , 1 , − 3 ) (1,1,1,-3) ( 1 , 1 , 1 , − 3 ) has − 3 ≠ 1 -3 \ne1 − 3 = 1 and no two equal coefficients. Any g ∈ K w g \in K_w g ∈ K w permutes the four points and therefore preserves the relation, hence fixes w w w ; consequently K w ≤ G Δ K_w \le G_\Delta K w ≤ G Δ and t w = 1 t_w=1 t w = 1 .
Theorem Appendix B.3 (Independent triples, p ≥ 5 p \ge5 p ≥ 5 ). If p ≥ 5 p \ge5 p ≥ 5 , d ≥ 2 d \ge2 d ≥ 2 and Δ = { 0 , u , v } \Delta=\{0,u,v\} Δ = { 0 , u , v } is affinely independent, then Δ \Delta Δ is not a ⋆ 3 \star_3 ⋆ 3 -witness for any affine group G ≤ AGL ( d , p ) G \le\operatorname{AGL}(d,p) G ≤ AGL ( d , p ) . In particular no affine group over F p \mathbb{F}_p F p , p ≥ 5 p \ge5 p ≥ 5 , satisfies ⋆ 3 \star_3 ⋆ 3 .
The case p = 3 p=3 p = 3 : localization to GL ( 2 , 3 ) \operatorname{GL}(2,3) GL ( 2 , 3 )
Assume p = 3 p=3 p = 3 , d ≥ 2 d \ge2 d ≥ 2 and Δ = { 0 , u , v } \Delta=\{0,u,v\} Δ = { 0 , u , v } is affinely independent. Normalize u = e 1 u=e_1 u = e 1 , v = e 2 v=e_2 v = e 2 ; put W = ⟨ e 1 , e 2 ⟩ ≅ F 3 2 W=\langle e_1,e_2\rangle\cong\mathbb{F}_3^2 W = ⟨ e 1 , e 2 ⟩ ≅ F 3 2 . For candidate points it suffices to consider w ∈ W ∖ Δ w \in W\setminus\Delta w ∈ W ∖ Δ : the six points of W ∖ Δ W\setminus\Delta W ∖ Δ are the six candidate points of the two-point stabilizer picture, and G Δ G_\Delta G Δ acts on W W W through its linear part. More precisely, the local action of G Δ G_\Delta G Δ on W W W is the affine group V 0 ⋊ L ≤ AGL ( 2 , 3 ) V_0\rtimes L \le\operatorname{AGL}(2,3) V 0 ⋊ L ≤ AGL ( 2 , 3 ) , V 0 = F 3 2 V_0=\mathbb{F}_3^2 V 0 = F 3 2 , L ≤ GL ( 2 , 3 ) L \le\operatorname{GL}(2,3) L ≤ GL ( 2 , 3 ) ; an element of G Δ G_\Delta G Δ stabilizing the plane W W W setwise induces such a local transformation, and the full stabilizer G Δ G_\Delta G Δ may have additional complementary part, which only enlarges r w r_w r w for the candidates w ∈ W w \in W w ∈ W .
By an exhaustive computation over the fifty-five subgroups of GL ( 2 , 3 ) \operatorname{GL}(2,3) GL ( 2 , 3 ) we obtain the following.
Lemma Appendix B.5 (The local criterion for p = 3 p=3 p = 3 ). Let L ≤ G L ( 2 , 3 ) L \le\mathrm{GL}(2,3) L ≤ GL ( 2 , 3 ) and let V 0 ⋊ L ≤ A G L ( 2 , 3 ) V_0 \rtimes L \le\mathrm{AGL}(2,3) V 0 ⋊ L ≤ AGL ( 2 , 3 ) act on V 0 V_0 V 0 with Δ 0 = { 0 , e 1 , e 2 } \Delta_0=\{0,e_1,e_2\} Δ 0 = { 0 , e 1 , e 2 } . If L L L is irreducible on V 0 V_0 V 0 , then among the six candidate points w ∈ V 0 ∖ Δ 0 w \in V_0 \setminus\Delta_0 w ∈ V 0 ∖ Δ 0 there is w w w with ∣ ( V 0 ⋊ L ) Δ 0 ∪ { w } ∣ ≤ ∣ ( V 0 ⋊ L ) Δ 0 ∣ \lvert(V_0 \rtimes L)_{\Delta_0\cup\{w\}}\rvert\le\lvert(V_0 \rtimes L)_{\Delta_0}\rvert ∣( V 0 ⋊ L ) Δ 0 ∪ { w } ∣ ≤ ∣( V 0 ⋊ L ) Δ 0 ∣ . The only subgroups for which this fails are the three subgroups of order 6 6 6 of upper/lower triangular type, for which the affine group V 0 ⋊ L V_0 \rtimes L V 0 ⋊ L is imprimitive on V 0 V_0 V 0 .
Proof sketch. The fifty-five subgroups of G L ( 2 , 3 ) \mathrm{GL}(2,3) GL ( 2 , 3 ) were enumerated and for each of them the stabilizers ∣ ( V 0 ⋊ L ) Δ 0 ∣ \lvert(V_0 \rtimes L)_{\Delta_0}\rvert ∣( V 0 ⋊ L ) Δ 0 ∣ and ∣ ( V 0 ⋊ L ) Δ 0 ∪ { w } ∣ \lvert(V_0 \rtimes L)_{\Delta_0\cup\{w\}}\rvert ∣( V 0 ⋊ L ) Δ 0 ∪ { w } ∣ , w ∈ V 0 ∖ Δ 0 w \in V_0 \setminus\Delta_0 w ∈ V 0 ∖ Δ 0 , were computed (GAP). For every irreducible L L L the inequality holds for at least one candidate; the three failing subgroups have order 6 6 6 , ∣ ( V 0 ⋊ L ) Δ 0 ∣ = 1 \lvert(V_0 \rtimes L)_{\Delta_0}\rvert=1 ∣( V 0 ⋊ L ) Δ 0 ∣ = 1 and min w ∣ ( V 0 ⋊ L ) Δ 0 ∪ { w } ∣ = 2 \min_w\lvert(V_0 \rtimes L)_{\Delta_0\cup\{w\}}\rvert=2 min w ∣( V 0 ⋊ L ) Δ 0 ∪ { w } ∣ = 2 , and each corresponding V 0 ⋊ L V_0 \rtimes L V 0 ⋊ L is imprimitive. □
Theorem Appendix B.6 (p = 3 p=3 p = 3 affine groups). Let G = V ⋊ H ≤ A G L ( d , 3 ) G=V \rtimes H \le\mathrm{AGL}(d,3) G = V ⋊ H ≤ AGL ( d , 3 ) be an affine primitive group, d ≥ 2 d \ge2 d ≥ 2 . Then no three-point subset is a ⋆ 3 \star_3 ⋆ 3 -witness; in particular G G G does not satisfy ⋆ 3 \star_3 ⋆ 3 .
Proof. For a collinear triple the claim is Theorem Appendix B.2. For an affinely independent triple, normalize to Δ = { 0 , e 1 , e 2 } \Delta=\{0,e_1,e_2\} Δ = { 0 , e 1 , e 2 } and let L L L be the image of H H H in G L ( 2 , 3 ) \mathrm{GL}(2,3) GL ( 2 , 3 ) induced on W = ⟨ e 1 , e 2 ⟩ W=\langle e_1,e_2\rangle W = ⟨ e 1 , e 2 ⟩ . If L L L is irreducible, Lemma Appendix B.5 supplies w ∈ W ∖ Δ w \in W \setminus\Delta w ∈ W ∖ Δ with the required inequality, and since the restriction K w → K w ∣ W K_w \to K_w|_W K w → K w ∣ W has image in ( V 0 ⋊ L ) Δ 0 ∪ { w } (V_0 \rtimes L)_{\Delta_0\cup\{w\}} ( V 0 ⋊ L ) Δ 0 ∪ { w } while G Δ → G Δ ∣ W G_\Delta\to G_\Delta|_W G Δ → G Δ ∣ W has image ( V 0 ⋊ L ) Δ 0 (V_0 \rtimes L)_{\Delta_0} ( V 0 ⋊ L ) Δ 0 , the inequality lifts to G G G . If L L L is reducible, then G G G has a proper system of imprimitivity: H H H preserves a 1 1 1 - or 2 2 2 -dimensional subspace (in characteristic 3 3 3 the only reducible subgroups of G L ( 2 , 3 ) \mathrm{GL}(2,3) GL ( 2 , 3 ) are of upper/lower triangular type, corresponding to a preserved line), contradicting primitivity. Hence the irreducible case applies to every primitive G G G . □
The case p = 2 p=2 p = 2 , d = 3 d=3 d = 3 : exhaustive verification
In characteristic 2 2 2 the normalization w = ( u + v ) / 3 w=(u+v)/3 w = ( u + v ) /3 is meaningless and the four-point dependence 0 + u + v + w = 0 0+u+v+w=0 0 + u + v + w = 0 has all coefficients equal to 1 1 1 , so a four-point stabilizer may carry a nontrivial S 4 S_4 S 4 -image. For d = 3 d=3 d = 3 the affine space has eight points and the full verification is finite.
Lemma Appendix B.7 (d = 3 d=3 d = 3 , p = 2 p=2 p = 2 ). Let H ≤ G L ( 3 , 2 ) H \le\mathrm{GL}(3,2) H ≤ GL ( 3 , 2 ) be irreducible on V = F 2 3 V=\mathbb{F}_2^3 V = F 2 3 and let G = V ⋊ H G=V \rtimes H G = V ⋊ H . Then for every affinely independent triple Δ = { 0 , u , v } \Delta=\{0,u,v\} Δ = { 0 , u , v } there exists w ∉ Δ w \notin\Delta w ∈ / Δ (in particular w ∈ V ∖ ⟨ u , v ⟩ w \in V \setminus\langle u,v\rangle w ∈ V ∖ ⟨ u , v ⟩ ) with ∣ G Δ ∪ { w } ∣ ≤ ∣ G Δ ∣ \lvert G_{\Delta\cup\{w\}}\rvert\le\lvert G_\Delta\rvert ∣ G Δ ∪ { w } ∣ ≤ ∣ G Δ ∣ .
Proof sketch. The seventeen irreducible subgroups of GL ( 3 , 2 ) \operatorname{GL}(3,2) GL ( 3 , 2 ) were enumerated (they are the subgroups without invariant 1- or 2-dimensional subspaces, i.e. exactly the linear parts of primitive affine groups of degree 8). For each of them the stabilizers of Δ = { 0 , e 1 , e 2 } \Delta= \{0,e_1,e_2\} Δ = { 0 , e 1 , e 2 } and of Δ ∪ { w } \Delta\cup\{w\} Δ ∪ { w } , w w w ranging over all five external points (including the parallel point e 1 + e 2 e_1+e_2 e 1 + e 2 of W = ⟨ e 1 , e 2 ⟩ W = \langle e_1,e_2\rangle W = ⟨ e 1 , e 2 ⟩ ), were computed (GAP); in every case some external w w w satisfies the inequality. Since all triples are equivalent under GL ( 3 , 2 ) \operatorname{GL}(3,2) GL ( 3 , 2 ) (any two independent vectors can be mapped to e 1 , e 2 e_1,e_2 e 1 , e 2 ), the check covers every triple.
Remark Appendix § 6 .8 (The role of irreducibility). The enumeration also shows why the invariant-subspace condition is necessary: the reducible subgroups A 4 A_4 A 4 and S 4 S_4 S 4 of GL ( 3 , 2 ) \operatorname{GL}(3,2) GL ( 3 , 2 ) that preserve a 2-dimensional subspace fail the criterion for Δ = { 0 , e 1 , e 2 } \Delta= \{0,e_1,e_2\} Δ = { 0 , e 1 , e 2 } (all five external w w w give ∣ G Δ ∪ { w } ∣ > ∣ G Δ ∣ |G_{\Delta\cup\{w\}}| > |G_\Delta| ∣ G Δ ∪ { w } ∣ > ∣ G Δ ∣ , e.g. ∣ G Δ ∣ = 2 |G_\Delta| = 2 ∣ G Δ ∣ = 2 and min w ∣ G Δ ∪ { w } ∣ = 3 \min_w |G_{\Delta\cup\{w\}}| = 3 min w ∣ G Δ ∪ { w } ∣ = 3 ), but the corresponding affine groups V ⋊ A 4 V \rtimes A_4 V ⋊ A 4 , V ⋊ S 4 V \rtimes S_4 V ⋊ S 4 are imprimitive on V V V and hence outside the statement of the conjecture.
For p = 2 p = 2 p = 2 , d ≥ 4 d \ge4 d ≥ 4 the argument follows the same pattern as the S 4 S_4 S 4 -image analysis of Section Appendix B.4: for w ∉ W = ⟨ u , v ⟩ w \notin W = \langle u,v\rangle w ∈ / W = ⟨ u , v ⟩ the four points { 0 , u , v , w } \{0,u,v,w\} { 0 , u , v , w } are affinely independent and the restriction to the affine span U = ⟨ u , v , w ⟩ U = \langle u,v,w\rangle U = ⟨ u , v , w ⟩ carries the S 4 S_4 S 4 -image. If the action of H H H on the quotient V / U V/U V / U has a nontrivial irreducible constituent, the orbit of w w w under G Δ G_\Delta G Δ inside U ⊥ U^\perp U ⊥ -directions grows, and a standard counting argument yields r w ≥ 4 r_w \ge4 r w ≥ 4 , which suffices by (B.1). The remaining degenerate cases (small image on V / U V/U V / U ) reduce to the three-dimensional problem of Lemma Appendix B.7. This high-dimensional amplification is recorded as a lemma whose full proof is completed in the final section.
The Singer-type lemma and the structure of the global obstruction
The following lemma isolates a high-dimensional branch in which every off-plane candidate passes by the equality case t ( w ) = r ( w ) = 1 t(w) = r(w) = 1 t ( w ) = r ( w ) = 1 , i.e. G Δ ∪ { w } = G Δ G_{\Delta\cup\{w\}} = G_\Delta G Δ ∪ { w } = G Δ (a complete exclusion certificate).
Lemma Appendix B.9. Let G = V ⋊ H ≤ AGL ( d , 2 ) G = V \rtimes H \le\operatorname{AGL}(d,2) G = V ⋊ H ≤ AGL ( d , 2 ) with V = F 2 d V = \mathbb{F}_2^d V = F 2 d , and let Δ = { 0 , u , v } ⊂ V \Delta= \{0,u,v\} \subset V Δ = { 0 , u , v } ⊂ V be a three-point set. Suppose that
Then for every w ∉ W : = ⟨ u , v ⟩ w \notin W := \langle u,v\rangle w ∈ / W := ⟨ u , v ⟩ one has G Δ ∪ { w } = 1 G_{\Delta\cup\{w\}}=1 G Δ ∪ { w } = 1 . Consequently every off-plane candidate satisfies t ( w ) = r ( w ) = 1 t(w)=r(w)=1 t ( w ) = r ( w ) = 1 , i.e. ∣ G Δ ∪ { w } ∣ = ∣ G Δ ∣ = 1 |G_{\Delta\cup\{w\}}|=|G_\Delta|=1 ∣ G Δ ∪ { w } ∣ = ∣ G Δ ∣ = 1 .
Proof. Put S = Δ ∪ { w } S=\Delta\cup\{w\} S = Δ ∪ { w } and C = G S C=G_S C = G S . The action of C C C on S S S is faithful: any element fixing S S S pointwise fixes Δ \Delta Δ pointwise, and G Δ = 1 G_\Delta=1 G Δ = 1 , so C ↪ Sym ( S ) ≅ S 4 C\hookrightarrow\operatorname{Sym}(S)\cong S_4 C ↪ Sym ( S ) ≅ S 4 and ∣ C ∣ ∣ 24 |C|\mid24 ∣ C ∣ ∣ 24 . Next, C C C contains no nontrivial translation: a translation x ↦ x + a x\mapsto x+a x ↦ x + a preserving the four-point set S S S forces S S S to be the union of two translation pairs, so ∑ x ∈ S x = 0 \sum_{x\in S}x=0 ∑ x ∈ S x = 0 ; but here u , v , w u,v,w u , v , w are linearly independent, so ∑ x ∈ S x = u + v + w ≠ 0 \sum_{x\in S}x=u+v+w\ne0 ∑ x ∈ S x = u + v + w = 0 . Hence C ∩ V = 1 C\cap V=1 C ∩ V = 1 and the projection G → H G\to H G → H is injective on C C C , so ∣ C ∣ |C| ∣ C ∣ divides ∣ H ∣ |H| ∣ H ∣ . Since gcd ( ∣ H ∣ , 6 ) = 1 \gcd(|H|,6)=1 g cd( ∣ H ∣ , 6 ) = 1 and ∣ C ∣ ∣ gcd ( 24 , ∣ H ∣ ) |C|\mid\gcd(24,|H|) ∣ C ∣ ∣ g cd( 24 , ∣ H ∣ ) , we get C = 1 C=1 C = 1 . Thus G Δ ∪ { w } = 1 G_{\Delta\cup\{w\}}=1 G Δ ∪ { w } = 1 .
Corollary Appendix B.10. Let d ≥ 3 d\ge3 d ≥ 3 be odd and let H ≤ GL ( d , 2 ) H\le\operatorname{GL}(d,2) H ≤ GL ( d , 2 ) be a Singer cycle C 2 d − 1 C_{2^d-1} C 2 d − 1 , or let d = p ≥ 5 d=p\ge5 d = p ≥ 5 be a prime and let H = GL ( 1 , 2 p ) ⋊ C p H=\operatorname{GL}(1,2^p)\rtimes C_p H = GL ( 1 , 2 p ) ⋊ C p be the full Singer normalizer. Then for every three-point set Δ = { 0 , u , v } \Delta=\{0,u,v\} Δ = { 0 , u , v } the stabilizer G Δ G_\Delta G Δ is trivial, the unique failing candidate is the plane completion point w = u + v w=u+v w = u + v (with t ( w ) = 1 t(w)=1 t ( w ) = 1 , r ( w ) = 4 r(w)=4 r ( w ) = 4 ), and all remaining candidates satisfy t ( w ) = r ( w ) = 1 t(w)=r(w)=1 t ( w ) = r ( w ) = 1 .
Proof. Identify V = F 2 d V=\mathbb{F}_{2^d} V = F 2 d and write H = F 2 d × H=\mathbb{F}_{2^d}^{\times} H = F 2 d × for the Singer cycle. A Singer cycle acts regularly on the nonzero vectors, so H ( W ) = 1 H_{(W)}=1 H ( W ) = 1 for W = ⟨ u , v ⟩ W=\langle u,v\rangle W = ⟨ u , v ⟩ : an element fixing two linearly independent vectors pointwise is the identity. Since H W ↪ GL ( 2 , 2 ) H_W\hookrightarrow\operatorname{GL}(2,2) H W ↪ GL ( 2 , 2 ) , we have ∣ H W ∣ ∣ 6 |H_W|\mid6 ∣ H W ∣ ∣ 6 , and ∣ H W ∣ ∣ ∣ H ∣ = 2 d − 1 |H_W|\mid|H|=2^d-1 ∣ H W ∣ ∣ ∣ H ∣ = 2 d − 1 . For odd d d d the numbers 2 d − 1 2^d-1 2 d − 1 and 6 6 6 are coprime (3 ∤ 2 d − 1 3\nmid2^d-1 3 ∤ 2 d − 1 and 2 ∤ ∣ H ∣ 2\nmid|H| 2 ∤ ∣ H ∣ ), so H W = 1 H_W=1 H W = 1 and G Δ ≅ H W = 1 G_\Delta\cong H_W=1 G Δ ≅ H W = 1 .
For the normalizer H = Γ L ( 1 , 2 p ) H=\mathop{\rm\Gamma L}(1,2^p) H = ΓL ( 1 , 2 p ) , p p p a prime with p ≥ 5 p\ge5 p ≥ 5 , we have gcd ( ∣ H ∣ , 6 ) = 1 \gcd(|H|,6)=1 g cd( ∣ H ∣ , 6 ) = 1 . An element of H H H is h ( x ) = a x 2 i h(x)=ax^{2^i} h ( x ) = a x 2 i for some a ≠ 0 a\ne0 a = 0 and 0 ≤ i < p 0\le i<p 0 ≤ i < p . If i = 0 i=0 i = 0 and a ≠ 1 a\ne1 a = 1 there is no nonzero fixed vector. If i ≠ 0 i\ne0 i = 0 , a nonzero fixed vector satisfies x 2 i − 1 = a − 1 x^{2^i-1}=a^{-1} x 2 i − 1 = a − 1 , which has exactly one nonzero solution because gcd ( 2 i − 1 , 2 p − 1 ) = 2 gcd ( i , p ) − 1 = 1 \gcd(2^i-1,2^p-1)=2^{\gcd(i,p)}-1=1 g cd( 2 i − 1 , 2 p − 1 ) = 2 g c d ( i , p ) − 1 = 1 . Hence a nonidentity element fixes at most one nonzero vector, so H ( W ) = 1 H_{(W)}=1 H ( W ) = 1 , and then H W / H ( W ) ↪ GL ( 2 , 2 ) H_W/H_{(W)}\hookrightarrow\operatorname{GL}(2,2) H W / H ( W ) ↪ GL ( 2 , 2 ) together with gcd ( ∣ H ∣ , 6 ) = 1 \gcd(|H|,6)=1 g cd( ∣ H ∣ , 6 ) = 1 forces H W = 1 H_W=1 H W = 1 , i.e. G Δ = 1 G_\Delta=1 G Δ = 1 .
Lemma Appendix B.9 applies in both cases: every off-plane candidate satisfies t = r = 1 t=r=1 t = r = 1 . For w = z : = u + v w=z:=u+v w = z := u + v the four-point set S = Δ ∪ { z } S=\Delta\cup\{z\} S = Δ ∪ { z } equals the 2-plane W W W ; the translation group T W T_W T W (of order 4) preserves W W W , and since G Δ ≅ H W = 1 G_\Delta\cong H_W=1 G Δ ≅ H W = 1 one has G W = T W ⋊ H W G_W=T_W\rtimes H_W G W = T W ⋊ H W with ∣ H W ∣ = 1 |H_W|=1 ∣ H W ∣ = 1 , so G W = T W G_W=T_W G W = T W , ∣ G W ∣ = 4 |G_W|=4 ∣ G W ∣ = 4 , and t = 1 t=1 t = 1 , r = 4 r=4 r = 4 . No other candidate can fail. GAP checks for d = 5 d=5 d = 5 (C 31 C_{31} C 31 , 15 three-point orbits) and d = 7 d=7 d = 7 (C 127 C_{127} C 127 , 63 orbits) confirm this pattern as an implementation check.
Proposition Appendix B.11. Assume that for a three-point set Δ = { 0 , u , v } \Delta= \{0,u,v\} Δ = { 0 , u , v } every candidate w ∉ Δ w \notin\Delta w ∈ / Δ fails, i.e. t ( w ) < r ( w ) ≤ 4 t(w) < r(w) \le4 t ( w ) < r ( w ) ≤ 4 for all w w w . Let K = G Δ K = G_{\Delta} K = G Δ , W = ⟨ u , v ⟩ W = \langle u,v\rangle W = ⟨ u , v ⟩ , z = u + v z = u+v z = u + v . Then every orbit of K K K on V V V has length at most 3, so K ↪ ∏ i Sym ( O i ) K \hookrightarrow\prod_i \operatorname{Sym}(O_i) K ↪ ∏ i Sym ( O i ) with ∣ O i ∣ ≤ 3 \lvert O_i\rvert\le3 ∣ O i ∣ ≤ 3 . In particular ∣ K ∣ = 2 α 3 β \lvert K\rvert= 2^{\alpha}3^{\beta} ∣ K ∣ = 2 α 3 β for some α , β ≥ 0 \alpha,\beta\ge0 α , β ≥ 0 , every element of K K K has order dividing 6, and the Sylow 2- and 3-subgroups of K K K are elementary abelian. Moreover K ≅ H W K \cong H_W K ≅ H W , K ( Δ ) ≅ H ( W ) K_{(\Delta)} \cong H_{(W)} K ( Δ ) ≅ H ( W ) and H W / H ( W ) ↪ GL ( 2 , 2 ) ≅ S 3 H_W/H_{(W)} \hookrightarrow\operatorname{GL}(2,2) \cong S_3 H W / H ( W ) ↪ GL ( 2 , 2 ) ≅ S 3 ; the action on Δ \Delta Δ only shows K / K ( Δ ) ≤ S 3 K/K_{(\Delta)} \le S_3 K / K ( Δ ) ≤ S 3 , and the pointwise stabilizer K ( Δ ) K_{(\Delta)} K ( Δ ) is not controlled by this argument.
Proof. If t ( w ) < r ( w ) ≤ 4 t(w) < r(w) \le4 t ( w ) < r ( w ) ≤ 4 for every w ∉ Δ w \notin\Delta w ∈ / Δ , then t ( w ) ≤ 3 t(w) \le3 t ( w ) ≤ 3 for all w ∉ Δ w \notin\Delta w ∈ / Δ , so every orbit of K K K on V ∖ Δ V \setminus\Delta V ∖ Δ has length at most 3; the orbits on Δ \Delta Δ have length at most 3 as well. Since K K K acts faithfully on V V V , the permutation representation embeds K K K into ∏ i Sym ( O i ) \prod_i \operatorname{Sym}(O_i) ∏ i Sym ( O i ) over its orbits, with ∣ O i ∣ ≤ 3 \lvert O_i\rvert\le3 ∣ O i ∣ ≤ 3 , yielding the prime-divisor, order and Sylow claims. For the affine statement, translate the origin to z z z ; the elements of K K K preserving the pair { 0 , z } \{0,z\} { 0 , z } form H W H_W H W , the pointwise part is H ( W ) H_{(W)} H ( W ) , and the quotient embeds into GL ( W ) ≅ S 3 \operatorname{GL}(W) \cong S_3 GL ( W ) ≅ S 3 .
Lemma Appendix B.13 (Fixed-point equality criterion). Let G G G be an affine primitive group on V = F 2 d V = \mathbb{F}_2^d V = F 2 d as in Appendix Appendix B, let Δ = { 0 , u , v } \Delta= \{0,u,v\} Δ = { 0 , u , v } be a three-point set with W = ⟨ u , v ⟩ W = \langle u,v\rangle W = ⟨ u , v ⟩ , put K = G Δ = H W K = G_{\Delta} = H_W K = G Δ = H W and K ( Δ ) = H ( W ) = 1 K_{(\Delta)} = H_{(W)} = 1 K ( Δ ) = H ( W ) = 1 , and let X = Fix V ( K ) ∖ W X = \operatorname{Fix}_V(K) \setminus W X = Fix V ( K ) ∖ W . For w ∈ X w \in X w ∈ X one has t ( w ) = 1 t(w)=1 t ( w ) = 1 , and with C w = G Δ ∪ { w } C_w = G_{\Delta\cup\{w\}} C w = G Δ ∪ { w } ,
K ≤ C w , r ( w ) = [ C w : K ] , K \le C_w,\qquad r(w) = [C_w:K], K ≤ C w , r ( w ) = [ C w : K ] , r ( w ) = 1 ⟺ C w = K ⟺ no g ∈ G ∖ K stabilises Δ ∪ { w } setwise . r(w)=1 \Longleftrightarrow C_w=K \Longleftrightarrow\text{no }g\in G\setminus K\text{ stabilises }\Delta\cup\{w\}\text{ setwise}. r ( w ) = 1 ⟺ C w = K ⟺ no g ∈ G ∖ K stabilises Δ ∪ { w } setwise .
Since H ( W ) = 1 H_{(W)}=1 H ( W ) = 1 , C w C_w C w acts faithfully on S = W ∪ { w } S=W\cup\{w\} S = W ∪ { w } , so C w ↪ Sym ( S ) ≅ S 4 C_w\hookrightarrow\operatorname{Sym}(S)\cong S_4 C w ↪ Sym ( S ) ≅ S 4 and K ↪ S Δ ≅ S 3 K\hookrightarrow S_\Delta\cong S_3 K ↪ S Δ ≅ S 3 ; in particular C w ≤ S 4 C_w\leq S_4 C w ≤ S 4 with K = ( C w ) w K=(C_w)_w K = ( C w ) w and r ( w ) = [ C w : K ] r(w)=[C_w:K] r ( w ) = [ C w : K ] . The possible strict overgroups of K K K in S 4 S_4 S 4 are
K = C 2 : { V 4 , S 3 , D 8 , A 4 , S 4 } , K = C 3 : { S 3 , A 4 , S 4 } , K = S 3 : { S 4 } . \begin{aligned}
K=C_2 &: \{V_4,S_3,D_8,A_4,S_4\},\\
K=C_3 &: \{S_3,A_4,S_4\},\\
K=S_3 &: \{S_4\}.
\end{aligned} K = C 2 K = C 3 K = S 3 : { V 4 , S 3 , D 8 , A 4 , S 4 } , : { S 3 , A 4 , S 4 } , : { S 4 } .
The following sufficient conditions imply C w = K C_w=K C w = K for some w ∈ X w\in X w ∈ X :
(A) (normaliser) N G ( K ) w = K N_G(K)_w=K N G ( K ) w = K for some w ∈ X w\in X w ∈ X and every g ∈ C w g\in C_w g ∈ C w normalises K K K ;
(B) (excluding four-point extensions) for some w ∈ X w\in X w ∈ X the induced image π ( C w ) ≤ S 4 \pi(C_w)\leq S_4 π ( C w ) ≤ S 4 lies in no L ∈ E K = { L ≤ S 4 : K < L } L\in E_K=\{L\leq S_4:K<L\} L ∈ E K = { L ≤ S 4 : K < L } ;
(C) (fixed-point covering estimate) X ∖ ⋃ L > K X L ≠ ∅ X\setminus\bigcup_{L>K}X_L\ne\varnothing X ∖ ⋃ L > K X L = ∅ , where X L = { w ∈ X : π ( C w ) ⊇ L } X_L=\{w\in X:\pi(C_w)\supseteq L\} X L = { w ∈ X : π ( C w ) ⊇ L } .
Proof. For w ∈ X w\in X w ∈ X , every element of K K K fixes w w w , so the orbit w G Δ w^{G_\Delta} w G Δ has length t ( w ) = 1 t(w)=1 t ( w ) = 1 ; clearly K ≤ C w K\leq C_w K ≤ C w and r ( w ) = [ C w : K ] r(w)=[C_w:K] r ( w ) = [ C w : K ] . The equivalence r ( w ) = 1 ⟺ C w = K r(w)=1\Longleftrightarrow C_w=K r ( w ) = 1 ⟺ C w = K is immediate. As H ( W ) = K ( Δ ) = 1 H_{(W)}=K_{(\Delta)}=1 H ( W ) = K ( Δ ) = 1 , any element fixing the four points of S S S pointwise is trivial, so C w ↪ Sym ( S ) C_w\hookrightarrow\operatorname{Sym}(S) C w ↪ Sym ( S ) is faithful. Since K K K fixes w w w and acts faithfully on Δ \Delta Δ , K ↪ S Δ ≅ S 3 K\hookrightarrow S_\Delta\cong S_3 K ↪ S Δ ≅ S 3 . The table of strict overgroups in S 4 S_4 S 4 follows from the subgroup lattice of S 4 S_4 S 4 . For (A), C w ≤ N G ( K ) w = K C_w\leq N_G(K)_w=K C w ≤ N G ( K ) w = K forces C w = K C_w=K C w = K . For (B), the image π ( C w ) \pi(C_w) π ( C w ) must then be K K K itself, so C w = K C_w=K C w = K . For (C), a point w ∈ X ∖ ⋃ L > K X L w\in X\setminus\bigcup_{L>K}X_L w ∈ X ∖ ⋃ L > K X L has π ( C w ) \pi(C_w) π ( C w ) contained in no strict overgroup of K K K and hence π ( C w ) = K \pi(C_w)=K π ( C w ) = K , so C w = K C_w=K C w = K .
Lemma Appendix B.14 (Affine fixed-point data, d = 4 d=4 d = 4 ). Let G G G be a primitive affine group with V = F 2 4 V=\mathbb{F}_2^4 V = F 2 4 and socle GL ( 4 , 2 ) \operatorname{GL}(4,2) GL ( 4 , 2 ) (the 20 primitive irreducible classes). Among the 113 three-point orbits, the 39 triples Δ \Delta Δ with H ( W ) = 1 H_{(W)}=1 H ( W ) = 1 and K = H W ≠ 1 K=H_W\ne1 K = H W = 1 satisfy:
31 of them have X = Fix V ( K ) ∖ W ≠ ∅ X=\operatorname{Fix}_V(K)\setminus W\ne\varnothing X = Fix V ( K ) ∖ W = ∅ , and every such triple has at least one w ∈ X w\in X w ∈ X with r ( w ) = 1 r(w)=1 r ( w ) = 1 , i.e. G Δ ∪ { w } = G Δ G_{\Delta\cup\{w\}}=G_\Delta G Δ ∪ { w } = G Δ ;
the remaining 8 have K ≅ S 3 K\cong S_3 K ≅ S 3 and Fix V ( K ) = ⟨ 0 ⟩ \operatorname{Fix}_V(K)=\langle0\rangle Fix V ( K ) = ⟨ 0 ⟩ (so X = ∅ X=\varnothing X = ∅ ); they pass through a large-orbit candidate with t ( w ) ≥ 4 t(w)\geq4 t ( w ) ≥ 4 ;
for every candidate with r ( w ) > 1 r(w) > 1 r ( w ) > 1 , the induced four-point image is π ( C w ) ≅ S 3 \pi(C_w) \cong S_3 π ( C w ) ≅ S 3 ; no V 4 V_4 V 4 , D 8 D_8 D 8 , A 4 A_4 A 4 or S 4 S_4 S 4 images occur.
In particular all three conditions of Lemma Appendix B.13 hold in d = 4 d = 4 d = 4 , and the small-orbit layer for d = 4 d = 4 d = 4 is closed.
Proof. Computed with GAP (version 4.16.1), using the natural permutation action of the 20 primitive irreducible classes of GL ( 4 , 2 ) \operatorname{GL}(4,2) GL ( 4 , 2 ) on the 16 points of V V V ; the stabilizers K = G Δ K = G_\Delta K = G Δ , the fixed-point sets Fix V ( K ) \operatorname{Fix}_V(K) Fix V ( K ) , and the four-point stabilizers C w C_w C w were computed as point stabilizers of set stabilizers, and the images π ( C w ) \pi(C_w) π ( C w ) by Action on the four points. The counts listed above are the exact output; in particular every nonempty X X X contains an equality candidate. □ \square □
Lemma Appendix B.15 (High-dimensional fixed-point cases). Let d ≥ 4 d \ge4 d ≥ 4 and K ≤ GL ( V ) K \le\operatorname{GL}(V) K ≤ GL ( V ) , V = F 2 d V = \mathbb{F}_2^d V = F 2 d , preserve a two-dimensional subspace W ≤ V W \le V W ≤ V and act faithfully on W W W . If K ≠ 1 K \ne1 K = 1 then K ≅ C 2 K \cong C_2 K ≅ C 2 , C 3 C_3 C 3 or S 3 S_3 S 3 , and with X = Fix V ( K ) ∖ W X = \operatorname{Fix}_V(K) \setminus W X = Fix V ( K ) ∖ W one has:
(a) if K ≅ C 2 K \cong C_2 K ≅ C 2 , then X ≠ ∅ X \ne\varnothing X = ∅ ; more precisely dim Fix V ( K ) ≥ ⌈ d / 2 ⌉ ≥ 2 \dim\operatorname{Fix}_V(K) \ge\lceil d/2 \rceil\ge2 dim Fix V ( K ) ≥ ⌈ d /2 ⌉ ≥ 2 , while dim W K = 1 \dim W^K = 1 dim W K = 1 ;
(b) if K ≅ C 3 K \cong C_3 K ≅ C 3 or S 3 S_3 S 3 and X = ∅ X = \varnothing X = ∅ , then d d d is even;
(c) if K ≅ S 3 K \cong S_3 K ≅ S 3 and X = ∅ X = \varnothing X = ∅ , then there exists w ∈ V ∖ W w \in V \setminus W w ∈ V ∖ W with K w = 1 K_w = 1 K w = 1 , hence t ( w ) = ∣ w K ∣ = 6 ≥ r ( w ) ≤ 4 t(w) = |w^K| = 6 \ge r(w) \le4 t ( w ) = ∣ w K ∣ = 6 ≥ r ( w ) ≤ 4 ;
(d) if K ≅ C 3 K \cong C_3 K ≅ C 3 and X = ∅ X = \varnothing X = ∅ , then V K = 0 V^K = 0 V K = 0 and for every w ∈ V ∖ W w \in V \setminus W w ∈ V ∖ W one has K w = 1 K_w = 1 K w = 1 , t ( w ) = 3 t(w) = 3 t ( w ) = 3 , so the candidate w w w passes exactly when r ( w ) ≤ 3 r(w) \le3 r ( w ) ≤ 3 ; the only failure is ∣ G Δ ∪ { w } ∣ = 4 |G_{\Delta\cup\{w\}}| = 4 ∣ G Δ ∪ { w } ∣ = 4 , i.e. a four-point stabilizer of order 4 (abstractly C 4 C_4 C 4 or V 4 V_4 V 4 ).
For w ∈ X w \in X w ∈ X with t ( w ) = 1 t(w) = 1 t ( w ) = 1 , the admissible strict extensions C w = G Δ ∪ { w } C_w = G_{\Delta\cup\{w\}} C w = G Δ ∪ { w } in the faithful four-point action are given by the marked-subgroup table
Table 9.
where the stabilizer of the marked point is exactly K K K ; and the structure of the X ≠ ∅ X \ne\varnothing X = ∅ case is not a consequence of X ≠ ∅ X \ne\varnothing X = ∅ alone, but requires a coverage argument (a structural proof that # { w ∈ X : C w = K } > 0 \#\{w \in X : C_w = K\} > 0 # { w ∈ X : C w = K } > 0 , or an exact count over the marked extension types).
Proof. (a) A non-trivial involution s s s satisfies ( s − I ) 2 = 0 (s-I)^2=0 ( s − I ) 2 = 0 in characteristic 2, so dim ker ( s − I ) ≥ d − dim ℑ ( s − I ) ≥ d − ⌊ d / 2 ⌋ = ⌈ d / 2 ⌉ \dim\ker(s-I) \ge d-\dim\Im(s-I) \ge d-\lfloor d/2\rfloor=\lceil d/2\rceil dim ker ( s − I ) ≥ d − dim ℑ ( s − I ) ≥ d − ⌊ d /2 ⌋ = ⌈ d /2 ⌉ ; since W K W^K W K is the one-dimensional fixed line of W W W , the fixed space Fix V ( K ) \operatorname{Fix}_V(K) Fix V ( K ) is not contained in W W W . (b) The natural two-dimensional K K K -modules over F 2 \mathbb{F}_2 F 2 have no non-zero fixed vector; if V K ≠ 0 V^K \ne0 V K = 0 a fixed vector lies outside W W W , so X ≠ ∅ X \ne\varnothing X = ∅ , contrary to assumption, hence V K = 0 V^K=0 V K = 0 . For C 3 C_3 C 3 the non-trivial irreducible F 2 C 3 \mathbb{F}_2C_3 F 2 C 3 -modules have dimension 2, so d d d is even. For S 3 S_3 S 3 with P = C 3 ◃ S 3 P=C_3\triangleleft S_3 P = C 3 ◃ S 3 : if V P ≠ 0 V^P\ne0 V P = 0 , an involution has a non-zero fixed vector on V P V^P V P , producing an S 3 S_3 S 3 -fixed vector, contradiction; hence V V V is a sum of two-dimensional non-trivial C 3 C_3 C 3 -modules, and d d d is even. (c) With X = ∅ X=\varnothing X = ∅ one has V K = 0 V^K=0 V K = 0 ; the natural two-dimensional S 3 S_3 S 3 -module W W W is projective (its restriction to a Sylow 2-subgroup is the regular module), so W W W is a K K K -invariant direct summand: V = W ⊕ U V=W\oplus U V = W ⊕ U with U ≠ 0 U\ne0 U = 0 (d ≥ 4 d\ge4 d ≥ 4 ). Pick x ∈ W ∖ { 0 } x\in W\setminus\{0\} x ∈ W ∖ { 0 } ; then Stab K ( x ) = ⟨ s ⟩ \operatorname{Stab}_K(x)=\langle s\rangle Stab K ( x ) = ⟨ s ⟩ of order 2. Since U C 3 = 0 U^{C_3}=0 U C 3 = 0 and s s s is not the identity on U U U (otherwise s p s = p − 1 sps=p^{-1} s p s = p − 1 would force p p p to be the identity on U U U ), there is y ∈ U y\in U y ∈ U with s y ≠ y sy\ne y sy = y . For w = x + y ∉ W w=x+y\notin W w = x + y ∈ / W one has K w = 1 K_w=1 K w = 1 , so t ( w ) = ∣ w K ∣ = ∣ K ∣ = 6 ≥ r ( w ) ≤ 4 t(w)=|w^K|=|K|=6\ge r(w)\le4 t ( w ) = ∣ w K ∣ = ∣ K ∣ = 6 ≥ r ( w ) ≤ 4 by Lemma Appendix B.1. (d) With V K = 0 V^K=0 V K = 0 , for every w ∉ W w\notin W w ∈ / W the stabilizer K w K_w K w fixes the four-point set Δ ∪ { w } \Delta\cup\{w\} Δ ∪ { w } pointwise in its K K K -action, so K w = 1 K_w=1 K w = 1 and t ( w ) = 3 t(w)=3 t ( w ) = 3 ; the candidate passes iff r ( w ) = [ G Δ ∪ { w } : 1 ] ≤ 3 r(w)=[G_{\Delta\cup\{w\}}:1]\le3 r ( w ) = [ G Δ ∪ { w } : 1 ] ≤ 3 , and the unique failure is ∣ G Δ ∪ { w } ∣ = 4 |G_{\Delta\cup\{w\}}|=4 ∣ G Δ ∪ { w } ∣ = 4 . The marked-subgroup table follows from the subgroup lattice of S 4 S_4 S 4 together with the fixed marked point.
V H W = 0 V^{H_W}=0 V H W = 0 . The precise target for (d) is therefore to show
∃ w ∉ W : ∣ G Δ ∪ { w } ∣ ≤ 3 , \exists w \notin W:\quad|G_{\Delta\cup\{w\}}|\le3, ∃ w ∈ / W : ∣ G Δ ∪ { w } ∣ ≤ 3 ,
a single such candidate suffices even if some candidates do carry regular C 4 C_4 C 4 or V 4 V_4 V 4 stabilizers; the d = 6 d=6 d = 6 computation should record the two failure types separately and check whether they can cover all of V ∖ W V\setminus W V ∖ W .
Proposition Appendix B.17 (The d = 6 d=6 d = 6 affine primitive verification). Let G = V ⋊ H ≤ AGL ( 6 , 2 ) G=V\rtimes H\le\operatorname{AGL}(6,2) G = V ⋊ H ≤ AGL ( 6 , 2 ) range over the 64 affine-type primitive groups of degree 64 contained in the GAP database PrimitiveGroup ( 64 , ∗ ) \operatorname{PrimitiveGroup}(64,*) PrimitiveGroup ( 64 , ∗ ) (74 primitive groups in total, 64 of affine type with socle C 2 6 C_2^6 C 2 6 ), with H ≤ GL ( 6 , 2 ) H\le\operatorname{GL}(6,2) H ≤ GL ( 6 , 2 ) the corresponding linear part. For each G G G -orbit representative Δ ∈ ( V 3 ) \Delta\in\binom{V}{3} Δ ∈ ( 3 V ) , put K = G Δ K=G_\Delta K = G Δ , and for w ∉ Δ w\notin\Delta w ∈ / Δ write t ( w ) = ∣ w K ∣ t(w)=|w^K| t ( w ) = ∣ w K ∣ and r ( w ) = [ G Δ ∪ { w } : K w ] r(w)=[G_{\Delta\cup\{w\}}:K_w] r ( w ) = [ G Δ ∪ { w } : K w ] with K w = K ∩ G Δ ∪ { w } K_w=K\cap G_{\Delta\cup\{w\}} K w = K ∩ G Δ ∪ { w } . Then there exists w ∉ Δ w\notin\Delta w ∈ / Δ with r ( w ) ≤ t ( w ) r(w)\le t(w) r ( w ) ≤ t ( w ) . Consequently no three-point set Δ \Delta Δ satisfies r ( w ) > t ( w ) r(w)>t(w) r ( w ) > t ( w ) for all w ∉ Δ w\notin\Delta w ∈ / Δ ; in particular none of these groups realises the witness configuration of the Siemons–Wagner property. The computation is exhaustive over all three-point orbit representatives, and for each representative over all w ∈ V ∖ Δ w\in V\setminus\Delta w ∈ V ∖ Δ ; the orbit counts are verified by
∑ Δ ∣ G : G Δ ∣ = ( 64 3 ) . \sum_{\Delta}|G:G_\Delta|=\binom{64}{3}. Δ ∑ ∣ G : G Δ ∣ = ( 3 64 ) .
and the affine reduction must show that these strict inequalities cannot cover all of V ∖ W V \setminus W V ∖ W (not that the plane-completion point is vetoed: the near-miss groups of Example Appendix B.19 have only the plane-completion channel with r > t r > t r > t , but they are large-orbit configurations, t ≥ 4 t \ge4 t ≥ 4 occurring off W W W , hence they do not enter the small-orbit branch above).
Example Appendix B.19 (The d = 6 d = 6 d = 6 near-miss groups). In the notation of Proposition Appendix B.17 the groups i = 41 , 43 , 44 , 45 , 46 i = 41, 43, 44, 45, 46 i = 41 , 43 , 44 , 45 , 46 of PrimitiveGroup(64,*) each admit a three-point set whose only candidate with r ( w ) > t ( w ) r(w) > t(w) r ( w ) > t ( w ) is the plane-completion point z z z ; the data are ∣ G 41 ∣ = 1290157424640 |G_{41}| = 1290157424640 ∣ G 41 ∣ = 1290157424640 , ∣ K ∣ = 30965760 |K| = 30965760 ∣ K ∣ = 30965760 , ∣ G 43 ∣ = 23224320 |G_{43}| = 23224320 ∣ G 43 ∣ = 23224320 , ∣ K ∣ = 17280 |K| = 17280 ∣ K ∣ = 17280 , ∣ G 44 ∣ = 7741440 |G_{44}| = 7741440 ∣ G 44 ∣ = 7741440 , ∣ K ∣ = 5760 |K| = 5760 ∣ K ∣ = 5760 , ∣ G 45 ∣ = 11612160 |G_{45}| = 11612160 ∣ G 45 ∣ = 11612160 , ∣ K ∣ = 8640 |K| = 8640 ∣ K ∣ = 8640 , ∣ G 46 ∣ = 3870720 |G_{46}| = 3870720 ∣ G 46 ∣ = 3870720 , ∣ K ∣ = 2880 |K| = 2880 ∣ K ∣ = 2880 , in each case K ∩ V = 1 K \cap V = 1 K ∩ V = 1 , t ( z ) = 1 t(z) = 1 t ( z ) = 1 , r ( z ) = 4 r(z) = 4 r ( z ) = 4 , ∣ G Δ ∪ { z } ∣ = 4 ∣ K ∣ |G_{\Delta\cup\{z\}}| = 4|K| ∣ G Δ ∪ { z } ∣ = 4∣ K ∣ , K z = K K_z = K K z = K , and Fix V ( K ) ∖ W = ∅ \operatorname{Fix}_{V}(K) \setminus W = \varnothing Fix V ( K ) ∖ W = ∅ ; the permutation image of G Δ ∪ { z } / G ( Δ ∪ { z } ) G_{\Delta\cup\{z\}}/G_{(\Delta\cup\{z\})} G Δ ∪ { z } / G ( Δ ∪ { z }) on the four-point set is S 4 S_4 S 4 (for i = 41 , 43 , 44 i = 41, 43, 44 i = 41 , 43 , 44 , where K / N ≅ S 3 K/N \cong S_3 K / N ≅ S 3 ) or A 4 A_4 A 4 (for i = 45 , 46 i = 45, 46 i = 45 , 46 , where K / N ≅ C 3 K/N \cong C_3 K / N ≅ C 3 ), with N = K ( W ) N = K(W) N = K ( W ) the (large) pointwise kernel on W W W . This is the exact analogue of the Singer pattern of Corollary Appendix B.10, with K K K nontrivial; the point is that these configurations are of large-orbit type, so the small-orbit classification above does not apply to them.
representative of the small-orbit branch is vetoed by the fixed-point criterion, and no three-point set of an affine group of degree 64 satisfies the witness inequalities.
Remark Appendix B.21 (The covering table for the regular-extension cases). For the five residual representatives with ∣ K ∣ = 1 |K| = 1 ∣ K ∣ = 1 from Appendix § 6 .20 (groups i = 17 , 18 , 25 , 26 i = 17,18,25,26 i = 17 , 18 , 25 , 26 of PrimitiveGroup(64, ·)), the table below records, for each four-point cycle type λ ∈ { ( 2 ) , ( 2 , 2 ) , ( 3 ) , ( 4 ) } \lambda\in\{(2),(2,2),(3),(4)\} λ ∈ {( 2 ) , ( 2 , 2 ) , ( 3 ) , ( 4 )} , the number of points w ∈ V ∖ W w \in V \setminus W w ∈ V ∖ W for which ρ w ( C w ) \rho_w(C_w) ρ w ( C w ) contains an element of type λ \lambda λ , together with the number of uncovered points (with C w = 1 C_w = 1 C w = 1 ). By the faithfulness established in Remark Appendix B.22a there is no distinction between C w ≠ 1 C_w \ne1 C w = 1 and ρ w ( C w ) ≠ 1 \rho_w(C_w) \ne1 ρ w ( C w ) = 1 here.
Table 11.
In particular the covering is far from saturated (≤ 17 \le17 ≤ 17 of 60 points), no 4-cycle type occurs and the 3-cycle type occurs exactly once in each case; explicit uncovered points include w = 2 , 3 , 6 w = 2, 3, 6 w = 2 , 3 , 6 . These data are numerical certificates for Appendix § 6 .22c§ 1 in dimension 6; they do not replace a general argument, and in particular the vanishing of type (4) must not be extrapolated to d ≥ 7 d \ge7 d ≥ 7 without proof.
Remark Appendix B.22 (A reduction framework for the affine case). The computation of Appendix § 6 .20 is numerical evidence only; the general exclusion needed for the main theorem is completed in Lemma 2.29 , and the framework below refines it into structural statements of independent interest. Guided by it, we record the following reduction framework, valid for any affine group G = V ⋊ H ≤ AGL ( d , 2 ) G = V \rtimes H \le\operatorname{AGL}(d,2) G = V ⋊ H ≤ AGL ( d , 2 ) with V ≅ F 2 d V \cong\mathbb{F}_2^d V ≅ F 2 d and Δ = { 0 , u , v } \Delta= \{0,u,v\} Δ = { 0 , u , v } , W = span ( Δ ) W = \operatorname{span}(\Delta) W = span ( Δ ) .
(a) Fixed-point veto as a covering criterion. Let K = G Δ K = G_\Delta K = G Δ and X = Fix V ( K ) ∖ W X = \operatorname{Fix}_V(K) \setminus W X = Fix V ( K ) ∖ W . For w ∈ X w \in X w ∈ X we have K ≤ C w : = G Δ ∪ { w } K \le C_w := G_{\Delta\cup\{w\}} K ≤ C w := G Δ ∪ { w } , t ( w ) = 1 t(w) = 1 t ( w ) = 1 and r ( w ) = [ C w : K ] r(w) = [C_w : K] r ( w ) = [ C w : K ] ; hence r ( w ) = 1 r(w) = 1 r ( w ) = 1 if and only if C w = K C_w = K C w = K . The fixed-point veto therefore holds provided X X X is not covered by the sets { x ∈ X : g ∈ G Δ ∪ { x } } \{x \in X : g \in G_{\Delta\cup\{x\}}\} { x ∈ X : g ∈ G Δ ∪ { x } } for g ∈ G ∖ K g \in G \setminus K g ∈ G ∖ K . A caveat is essential: one cannot check only the quotient G W / K G_W/K G W / K , because an element g ∈ C w g \in C_w g ∈ C w may map w w w into W W W while mapping some point of W W W to w w w , so g g g need not lie in G W G_W G W ; the “extra stabiliser” must be checked inside the full four-point stabiliser C w C_w C w . In the case K = 1 K=1 K = 1 the four-point action is faithful: the pointwise kernel of ρ w : C w → Sym ( S w ) \rho_w:C_w\to\operatorname{Sym}(S_w) ρ w : C w → Sym ( S w ) , S w = Δ ∪ { w } S_w=\Delta\cup\{w\} S w = Δ ∪ { w } , fixes Δ \Delta Δ pointwise and hence lies in G Δ = 1 G_\Delta=1 G Δ = 1 ; thus C w ↪ S 4 C_w\hookrightarrow S_4 C w ↪ S 4 , and a point w ∉ W w\notin W w ∈ / W is a bad point (C w ≠ 1 C_w\ne1 C w = 1 ) if and only if ρ w ( C w ) ≠ 1 \rho_w(C_w)\ne1 ρ w ( C w ) = 1 . With Δ = { 0 , a , b } \Delta=\{0,a,b\} Δ = { 0 , a , b } and W = ⟨ a , b ⟩ W=\langle a,b\rangle W = ⟨ a , b ⟩ , bad points satisfy the covering identity
V ∖ W = ⋃ 1 ≠ g ∈ G B g , B g ⊆ g ( Δ ) ∖ W , ∣ B g ∣ ≤ 3 , V\setminus W=\bigcup_{1\ne g\in G}B_g,\qquad B_g\subseteq g(\Delta)\setminus W,\qquad|B_g|\le3, V ∖ W = 1 = g ∈ G ⋃ B g , B g ⊆ g ( Δ ) ∖ W , ∣ B g ∣ ≤ 3 ,
so that g ∈ C w g\in C_w g ∈ C w for w ∉ W w\notin W w ∈ / W forces w ∈ g ( Δ ) ∖ W w\in g(\Delta)\setminus W w ∈ g ( Δ ) ∖ W (since g ( Δ ) ≠ Δ g(\Delta)\ne\Delta g ( Δ ) = Δ when g ≠ 1 g\ne1 g = 1 , K = 1 K=1 K = 1 ). The regular-extension statement ⋃ 1 ≠ g ∈ G B g ≠ V ∖ W \bigcup_{1\ne g\in G}B_g\ne V\setminus W ⋃ 1 = g ∈ G B g = V ∖ W is the content of (c1) below; the identity itself is unconditional. The affine span of S w S_w S w is the 3-dimensional space U w = ⟨ a , b , w ⟩ U_w=\langle a,b,w\rangle U w = ⟨ a , b , w ⟩ , and g ∣ S w g|_{S_w} g ∣ S w is determined by the four affine equations A x i + c = x π ( i ) Ax_i+c=x_{\pi(i)} A x i + c = x π ( i ) (g : x ↦ A x + c g:x\mapsto Ax+c g : x ↦ A x + c , A ∈ H A\in H A ∈ H , c ∈ V c\in V c ∈ V , π = ρ w ( g ) \pi=\rho_w(g) π = ρ w ( g ) ); hence the quantity to control is not ∣ H ∣ |H| ∣ H ∣ but the number of extensions of prescribed affine restrictions on 3-spaces. The local extension numbers are
e π ( w ) = ∣ { g ∈ G : g ( S w ) = S w , ρ w ( g ) = π } ∣ . e_\pi(w)=\left|\{g\in G:g(S_w)=S_w,\ \rho_w(g)=\pi\}\right|. e π ( w ) = ∣ { g ∈ G : g ( S w ) = S w , ρ w ( g ) = π } ∣ .
(b) Three-way split. With N = K ( W ) N=K_{(W)} N = K ( W ) and K / N ≤ GL ( W ) ≅ S 3 K/N\le\operatorname{GL}(W)\cong S_3 K / N ≤ GL ( W ) ≅ S 3 one distinguishes
(I) N = K ≠ 1 N=K\ne1 N = K = 1 ;
(II) 1 < N < K 1<N<K 1 < N < K (with K / N ≅ C 2 K/N\cong C_2 K / N ≅ C 2 or C 3 C_3 C 3 );
(III) N = 1 N=1 N = 1 .
For (I), if X ≠ ∅ X\ne\varnothing X = ∅ and the fixed-point veto of (a) applies, the witness is vetoed by a point of X X X with r = 1 r=1 r = 1 . Case (II) is neither covered by Appendix § 6 .15 nor by the pure large-kernel case: one must quotient by N N N and analyse K ˉ = K / N ≤ S 3 \bar K=K/N\le S_3 K ˉ = K / N ≤ S 3 together with the action of N N N on V / W V/W V / W ; the expected conclusion is that, absent a fixed-point veto, a K K K -orbit on V ∖ W V\setminus W V ∖ W of length at least 4 arises. Case (III) is the faithful case of Appendix § 6 .15 (a)–(d).
(c) The small-kernel branch splits into two lemmas. The residual 22 cases of Appendix § 6 .20 all satisfy ∣ K ∣ ≤ 3 |K|\le3 ∣ K ∣ ≤ 3 and K ∩ V = 1 K\cap V=1 K ∩ V = 1 . The 2-homogeneous rows needed for the main theorem are already excluded by Lemma 2.29 , and the two statements below are recorded for the general framework.
(c1) Regular-extension lemma (K = 1 K=1 K = 1 ). If K = G Δ = 1 K=G_\Delta=1 K = G Δ = 1 then there exists w ∉ W w\notin W w ∈ / W with C w = 1 C_w=1 C w = 1 (i.e. G Δ ∪ { w } = 1 G_{\Delta\cup\{w\}}=1 G Δ ∪ { w } = 1 ), except possibly for a finite list of four-point covering types. Note that this is not a fixed-point statement: the four-point stabiliser C w C_w C w is a set stabiliser, and an element of C w C_w C w may permute Δ \Delta Δ with w w w , so C w ⊈ G Δ C_w \not\subseteq G_\Delta C w ⊆ G Δ in general; the lemma asserts that the nontrivial four-point set stabilisers do not cover all of V ∖ W V \setminus W V ∖ W . Guided by the covering table Appendix § 6 .21, split the bad points according to the cycle type of ρ w ( C w ) \rho_w(C_w) ρ w ( C w ) :
B ( λ ) = { w ∈ V ∖ W : C w ≠ 1 , ρ w ( C w ) contains an element of type λ } , B^{(\lambda)} = \{w \in V \setminus W : C_w \neq1,\ \rho_w(C_w)\ \text{contains an element of type }\lambda\}, B ( λ ) = { w ∈ V ∖ W : C w = 1 , ρ w ( C w ) contains an element of type λ } ,
λ ∈ { ( 2 ) , ( 2 , 2 ) , ( 3 ) , ( 4 ) } \lambda\in\{(2),(2,2),(3),(4)\} λ ∈ {( 2 ) , ( 2 , 2 ) , ( 3 ) , ( 4 )} ; the faithfulness of (a) makes C w ≠ 1 C_w \neq1 C w = 1 and ρ w ( C w ) ≠ 1 \rho_w(C_w) \neq1 ρ w ( C w ) = 1 equivalent here. In the five dimension-6 cases all four-point types occurring are (2) and (2, 2) with exactly one (3) and no (4); the vanishing of the (4)-type is a numerical phenomenon and must not be extrapolated to d ≥ 7 d \ge7 d ≥ 7 .
(c1′) The four-cycle residual case. If g = ( c , A ) ∈ C w g = (c,A) \in C_w g = ( c , A ) ∈ C w , A ∈ H A \in H A ∈ H , c ∈ V c \in V c ∈ V , induces a 4-cycle on S w S_w S w , then the affine span U w = ⟨ a , b , w ⟩ U_w = \langle a,b,w\rangle U w = ⟨ a , b , w ⟩ satisfies A ( U w ) = U w A(U_w) = U_w A ( U w ) = U w and, writing A = I + N A = I + N A = I + N with N N N nilpotent, the restriction A ∣ U w A|_{U_w} A ∣ U w is conjugate to the Jordan block J 3 ( 1 ) J_3(1) J 3 ( 1 ) (N N N nilpotent of index 3; over F 2 \mathbb{F}_2 F 2 this means precisely that A A A has a 3-cycle on the nonzero vectors of U w U_w U w ). Such g g g need not fix Δ \Delta Δ (g ∉ G Δ g \notin G_\Delta g ∈ / G Δ ), so neither K = 1 K = 1 K = 1 nor irreducibility of H H H excludes this case: an irreducible linear group may well contain an element with Jordan type J 3 ( 1 ) J_3(1) J 3 ( 1 ) on a 3-space. The relevant local set is
E 4 ( U ) = { A ∈ H : A ( U ) = U , A ∣ U ∼ J 3 ( 1 ) } , E_4(U) = \{A \in H : A(U) = U,\ A|_U \sim J_3(1)\}, E 4 ( U ) = { A ∈ H : A ( U ) = U , A ∣ U ∼ J 3 ( 1 )} ,
and B ( 4 ) B^{(4)} B ( 4 ) is decided in two steps: whether E 4 ( U ) ≠ ∅ E_4(U) \neq\varnothing E 4 ( U ) = ∅ for some 3-space U U U , and whether the corresponding affine coset c + im ( I − A ) c + \operatorname{im}(I-A) c + im ( I − A ) solves the four-point image equations. A general exclusion of the (4)-type therefore requires E 4 ( U ) = ∅ E_4(U) = \varnothing E 4 ( U ) = ∅ for all relevant U U U , or a bound ∣ B ( 4 ) ∣ < 2 d − 4 |B^{(4)}| < 2^d - 4 ∣ B ( 4 ) ∣ < 2 d − 4 , or a separate elimination by H H H -type; in particular the four-cycle case is a genuine residual case, and B ( 4 ) = ∅ B^{(4)} = \varnothing B ( 4 ) = ∅ in dimension 6 is recorded as a numerical certificate only.
(c1′′) Transvection normal form for the (2)- and (2,2)-types. If g = ( c , A ) ∈ C w g = (c,A) \in C_w g = ( c , A ) ∈ C w (A ∈ H A \in H A ∈ H , c ∈ V c \in V c ∈ V ) induces a transposition or a double transposition on S w S_w S w , then g 2 g^2 g 2 fixes S w S_w S w pointwise, hence g 2 ∈ G ( S w ) ≤ G Δ = K = 1 g^2 \in G_{(S_w)} \le G_\Delta= K = 1 g 2 ∈ G ( S w ) ≤ G Δ = K = 1 , so g g g is an involution. Writing N = A − I = A + I N = A - I = A + I N = A − I = A + I (characteristic 2), the conditions A 2 = I A^2 = I A 2 = I and ( I + A ) c = 0 (I + A)c = 0 ( I + A ) c = 0 become N 2 = 0 N^2 = 0 N 2 = 0 and N c = 0 Nc = 0 N c = 0 ; on the 3-space U w = ⟨ a , b , w ⟩ U_w = \langle a,b,w\rangle U w = ⟨ a , b , w ⟩ the restriction is A ∣ U w = I + N w A|_{U_w} = I + N_w A ∣ U w = I + N w with N w 2 = 0 N_w^2 = 0 N w 2 = 0 and rank N w = 1 N_w = 1 N w = 1 , i.e. a rank-one transvection type. The two types differ only in the affine coset: for a transposition g g g has a fixed point and c ∈ im ( N w ) c \in\operatorname{im}(N_w) c ∈ im ( N w ) (after translating the fixed point to 0), while for a double transposition c ∉ im ( N w ) c \notin\operatorname{im}(N_w) c ∈ / im ( N w ) ; in both cases, with π = ρ w ( g ) \pi= \rho_w(g) π = ρ w ( g ) , the point w w w is determined by the affine equations g ( x i ) = x π ( i ) g(x_i) = x_{\pi(i)} g ( x i ) = x π ( i ) (x i ∈ S w x_i \in S_w x i ∈ S w ), and for a fixed linear part A A A and a fixed pattern π \pi π there is at most one candidate w = A x + y w = Ax+y w = A x + y for the corresponding transposition pattern. Define
n π ( A ) = # { w ∈ V ∖ W : ∃ c ∈ V , ( c , A ) ∣ S w = π } , n_{\pi}(A) = \#\{w \in V \setminus W : \exists c \in V,\ (c,A)|_{S_w} = \pi\}, n π ( A ) = # { w ∈ V ∖ W : ∃ c ∈ V , ( c , A ) ∣ S w = π } ,
I ( 2 ) = ∑ π ( 2 ) ∑ A ∈ H n π ( A ) , I^{(2)} = \sum_{\pi\ (2)} \sum_{A \in H} n_{\pi}(A), I ( 2 ) = π ( 2 ) ∑ A ∈ H ∑ n π ( A ) ,
I ( 2 , 2 ) = ∑ π ( 2 , 2 ) ∑ A ∈ H n π ( A ) . I^{(2,2)} = \sum_{\pi\ (2,2)} \sum_{A \in H} n_{\pi}(A). I ( 2 , 2 ) = π ( 2 , 2 ) ∑ A ∈ H ∑ n π ( A ) .
and the local transvection set
T l o c ( a , b ) = { A ∈ H : A 2 = I , N = A + I , N 2 = 0 , A satisfies the three-point constraint of some ( 2 ) / ( 2 , 2 ) pattern } . T_{\mathrm{loc}}(a,b) = \{A \in H : A^2 = I,\ N = A + I,\ N^2 = 0,\ A\text{ satisfies the three-point constraint of some }(2)/(2,2)\text{ pattern}\}. T loc ( a , b ) = { A ∈ H : A 2 = I , N = A + I , N 2 = 0 , A satisfies the three-point constraint of some ( 2 ) / ( 2 , 2 ) pattern } .
Then ∣ B ( 2 ) ∪ B ( 2 , 2 ) ∣ ≤ I ( 2 ) + I ( 2 , 2 ) ≤ 9 ∣ T l o c ( a , b ) ∣ |B^{(2)} \cup B^{(2,2)}| \leq I^{(2)} + I^{(2,2)} \leq9|T_{\mathrm{loc}}(a,b)| ∣ B ( 2 ) ∪ B ( 2 , 2 ) ∣ ≤ I ( 2 ) + I ( 2 , 2 ) ≤ 9∣ T loc ( a , b ) ∣ (a coarse bound), and a sufficient counting statement, to be proved for each H H H -type, is
I ( 2 ) + I ( 2 , 2 ) < 2 d − 4 , I^{(2)} + I^{(2,2)} < 2^d - 4, I ( 2 ) + I ( 2 , 2 ) < 2 d − 4 ,
for example via ∣ T l o c ( a , b ) ∣ ≤ α 2 d + O ( 1 ) |T_{\mathrm{loc}}(a,b)| \leq\alpha2^d + O(1) ∣ T loc ( a , b ) ∣ ≤ α 2 d + O ( 1 ) with 9 α < 1 9\alpha< 1 9 α < 1 . The data of Appendix B.21 give I ( 2 ) + I ( 2 , 2 ) ≤ 17 I^{(2)} + I^{(2,2)} \leq17 I ( 2 ) + I ( 2 , 2 ) ≤ 17 in dimension 6, far below 2 6 − 4 = 60 2^6 - 4 = 60 2 6 − 4 = 60 .
(c1‴) Class and functional versions of the counting bound. Two corrections refine the count. First, the four-point action only forces A 2 = I A^2 = I A 2 = I , N 2 = 0 N^2 = 0 N 2 = 0 and rank ( N ∣ U w ) = 1 \operatorname{rank}(N|_{U_w}) = 1 rank ( N ∣ U w ) = 1 ; the global rank r = rank N r = \operatorname{rank}N r = rank N may be larger than 1, so counting global rank-one transvections via R H ( u ) R_H(u) R H ( u ) is insufficient—one must count over all involution classes I ( H ) I(H) I ( H ) of H H H and impose the local condition rank ( N ∣ U w ) = 1 \operatorname{rank}(N|_{U_w}) = 1 rank ( N ∣ U w ) = 1 inside each class. Second, when K = 1 K = 1 K = 1 , a pattern π = ρ w ( g ) \pi= \rho_w(g) π = ρ w ( g ) fixing Δ \Delta Δ setwise forces g ∈ G Δ = 1 g \in G_\Delta= 1 g ∈ G Δ = 1 ; hence of the six transpositions only the three involving w w w occur, while all three double transpositions occur, and for a fixed g g g and an admissible pattern π \pi π the point w w w is unique (π ( w ) ∈ Δ \pi(w) \in\Delta π ( w ) ∈ Δ , so w = g − 1 ( π ( w ) ) w = g^{-1}(\pi(w)) w = g − 1 ( π ( w )) ). For P i n v P_{\mathrm{inv}} P inv the six admissible patterns, the class-intersection formula for arbitrary representations is
∣ B ( 2 ) ∪ B ( 2 , 2 ) ∣ ≤ ∑ π ∈ P i n v ∑ C ∈ I ( H ) m π ( C ; Δ ) , |B^{(2)} \cup B^{(2,2)}| \leq\sum_{\pi\in P_{\mathrm{inv}}} \sum_{C \in I(H)} m_{\pi}(C;\Delta), ∣ B ( 2 ) ∪ B ( 2 , 2 ) ∣ ≤ π ∈ P inv ∑ C ∈ I ( H ) ∑ m π ( C ; Δ ) ,
∣ P i n v ∣ = 6 , |P_{\mathrm{inv}}| = 6, ∣ P inv ∣ = 6 ,
where m π ( C ; Δ ) m_{\pi}(C;\Delta) m π ( C ; Δ ) counts the elements of class C C C realising pattern π \pi π together with their unique w w w ; the goal ∑ π ∈ P i n v ∑ C ∈ I ( H ) m π ( C ; Δ ) < 2 d − 4 \sum_{\pi\in P_{\mathrm{inv}}}\sum_{C\in I(H)}m_{\pi}(C;\Delta)<2^d-4 ∑ π ∈ P inv ∑ C ∈ I ( H ) m π ( C ; Δ ) < 2 d − 4 then takes the place of the coarser β < 1 / 9 \beta<1/9 β < 1/9 hypothesis. One notes that ambient classes may split into several H H H -classes, and a local rank-one restriction may come from a global involution of rank r > 1 r>1 r > 1 ; a bound “number of classes times per-class contribution” is valid only after the intersection number is proved class by class.
( c 1 ′ ′ ′ ′ ) (c1'''' ) ( c 1 ′′′′ ) The Γ L 1 \Gamma L_1 Γ L 1 subfamily. For H ≤ GL 1 ( 2 d ) H\leq\operatorname{GL}_1(2^d) H ≤ GL 1 ( 2 d ) , elements are x ↦ a x 2 i x\mapsto ax^{2^i} x ↦ a x 2 i ; such an element is an involution only if 2 i ≡ 0 ( m o d d ) 2i\equiv0\pmod d 2 i ≡ 0 ( mod d ) . For odd d d d only i = 0 i=0 i = 0 occurs (the identity); for even d d d the nonidentity possibility is i = d / 2 i=d/2 i = d /2 with a 2 d / 2 + 1 = 1 a^{2^{d/2}+1}=1 a 2 d /2 + 1 = 1 , giving at most 2 d / 2 + 1 2^{d/2}+1 2 d /2 + 1 such elements. Since each admissible pattern yields at most one w w w , the coarse bound
∣ B ( 2 ) ∪ B ( 2 , 2 ) ∣ ≤ 6 ( 2 d / 2 + 1 ) \left|B^{(2)}\cup B^{(2,2)}\right|\leq6(2^{d/2}+1) B ( 2 ) ∪ B ( 2 , 2 ) ≤ 6 ( 2 d /2 + 1 )
holds, and for even d ≥ 8 d\geq8 d ≥ 8 this is already < 2 d − 4 <2^d-4 < 2 d − 4 ; the odd-d d d case has no nontrivial involutions. This settles the semilinear-rank-one subfamily within (c1).
( c 1 ′ ′ ′ ′ ′ ) (c1''''') ( c 1 ′′′′′ ) A functional parametrisation (coarser version). For reference, a parameter-free version parametrises global rank-one transvections by
R H ( u ) = { φ ∈ V ∗ : φ ( u ) = 0 , I + u φ ∈ H } , R_H(u)=\{\varphi\in V^*: \varphi(u)=0,\ I+u\varphi\in H\}, R H ( u ) = { φ ∈ V ∗ : φ ( u ) = 0 , I + u φ ∈ H } ,
R H ( u ; s , a ) = { φ ∈ R H ( u ) : φ ( s ) = 0 , φ ( a ) = 1 } , R_H(u;s,a)=\{\varphi\in R_H(u): \varphi(s)=0,\ \varphi(a)=1\}, R H ( u ; s , a ) = { φ ∈ R H ( u ) : φ ( s ) = 0 , φ ( a ) = 1 } ,
so that for the pattern ( 0 w ) ( a b ) (0w)(ab) ( 0 w ) ( ab ) , with w = s + u w=s+u w = s + u , one has ∣ B π ∣ ≤ ∑ u ∉ W ∣ R H ( u ; s , a ) ∣ |B_\pi|\leq\sum_{u\notin W}|R_H(u;s,a)| ∣ B π ∣ ≤ ∑ u ∈ / W ∣ R H ( u ; s , a ) ∣ , distinct u u u giving distinct w w w ; a bound ∑ u ∉ W ∣ R H ( u ; s , a ) ∣ ≤ α 2 d + O ( 1 ) \sum_{u\notin W}|R_H(u;s,a)|\leq\alpha2^d+O(1) ∑ u ∈ / W ∣ R H ( u ; s , a ) ∣ ≤ α 2 d + O ( 1 ) for all admissible patterns with 6 α < 1 6\alpha<1 6 α < 1 suffices and avoids concentration phenomena inside a single class. The high-dimensional input may therefore be stated as a single sparsity hypothesis:
(Sparse)
∑ π ∈ P i n v ∣ T π ( H d ; Δ ) ∣ ≤ β 2 d + O ( 1 ) ( β < 1 / 6 , d → ∞ ) , \sum_{\pi\in P_{\mathrm{inv}}}|T_\pi(H_d;\Delta)|
\leq\beta2^d+O(1)\qquad(\beta<1/6,\ d\to\infty), π ∈ P inv ∑ ∣ T π ( H d ; Δ ) ∣ ≤ β 2 d + O ( 1 ) ( β < 1/6 , d → ∞ ) ,
for each irreducible family H d H_d H d , controlling the sum over the six admissible patterns directly rather than the coarser ∣ T l o c ∣ |T_{\mathrm{loc}}| ∣ T loc ∣ ; in view of ( c 1 ′ ′ ′ ) (c1''') ( c 1 ′′′ ) this functional version is subsumed by the class-intersection formula, which additionally covers local rank-one restrictions coming from global involutions of rank > 1 >1 > 1 .
(c2) Fixed-point lemma (K = C 2 K = C_2 K = C 2 or C 3 C_3 C 3 ). If K = C 2 K = C_2 K = C 2 or K = C 3 K = C_3 K = C 3 , X = Fix V ( K ) ∖ W ≠ ∅ X = \operatorname{Fix}_V(K) \setminus W \ne\varnothing X = Fix V ( K ) ∖ W = ∅ and there is no K K K -orbit on V ∖ W V \setminus W V ∖ W of length at least 4, then there exists w ∈ X w \in X w ∈ X with C w = G Δ ∪ { w } = K C_w = G_{\Delta\cup\{w\}} = K C w = G Δ ∪ { w } = K ; here one genuinely uses the definition of X X X and the structure of C w / K ∈ { C 2 , C 3 , S 3 } C_w/K \in\{C_2, C_3, S_3\} C w / K ∈ { C 2 , C 3 , S 3 } , together with the small-orbit hypothesis, to rule out the possibility that these subgroups cover all of X X X .
(d) Statement of the framework. Combining (a)–(c) with Appendix B.15, the affine exclusion reduces to the following chain: an orbit of length t ≥ 4 t \ge4 t ≥ 4 on V ∖ W V \setminus W V ∖ W vetoes directly; otherwise, in the small-orbit branch, the case N = K ≠ 1 N = K \ne1 N = K = 1 with X ≠ ∅ X \ne\varnothing X = ∅ is handled by the fixed-point veto, the case 1 < N < K 1 < N < K 1 < N < K by an intermediate-kernel lemma, and the case N = 1 N = 1 N = 1 by Appendix B.15 (a)–(d). The present paper records this framework and the computational verification of dimension 6; the individual steps — the regular-extension counting bound, the intermediate-kernel threshold application and the fixed-point lemma toward the semilinear lift problem — are statements of independent interest and are not needed for the main theorem: the affine exclusion required by Corollary 2.30 concerns only the 2-homogeneous rows, which are excluded uniformly by Lemma 2.29 .
Proposition Appendix B.23 (Natural classical groups fail the K = 1 K = 1 K = 1 case). Let V = F 2 d V = \mathbb{F}_2^d V = F 2 d and let H H H be either GL ( d , 2 ) \operatorname{GL}(d,2) GL ( d , 2 ) acting naturally on V V V , or Sp ( 2 m , 2 ) \operatorname{Sp}(2m,2) Sp ( 2 m , 2 ) acting naturally on V = F 2 2 m V = \mathbb{F}_2^{2m} V = F 2 2 m with its nondegenerate alternating form, d = 2 m d = 2m d = 2 m . If d ≥ 4 d \ge4 d ≥ 4 then for every Δ = { 0 , a , b } \Delta= \{0,a,b\} Δ = { 0 , a , b } with a , b a,b a , b linearly independent one has G Δ ≠ 1 G_\Delta\ne1 G Δ = 1 ; in fact G Δ G_\Delta G Δ contains a nontrivial rank-one transvection ( 0 , A ) (0,A) ( 0 , A ) . Consequently for d ≥ 8 d \ge8 d ≥ 8 the affine groups F 2 d ⋊ GL ( d , 2 ) \mathbb{F}_2^d \rtimes\operatorname{GL}(d,2) F 2 d ⋊ GL ( d , 2 ) and F 2 2 m ⋊ Sp ( 2 m , 2 ) \mathbb{F}_2^{2m} \rtimes\operatorname{Sp}(2m,2) F 2 2 m ⋊ Sp ( 2 m , 2 ) are excluded at the entry to the K = 1 K = 1 K = 1 case (c1) of Appendix B.22: there is no need to verify the counting bound (Sparse) for them.
Proof. For H = GL ( d , 2 ) H = \operatorname{GL}(d,2) H = GL ( d , 2 ) choose 0 ≠ φ ∈ V ∗ 0 \ne\varphi\in V^* 0 = φ ∈ V ∗ with φ ( a ) = φ ( b ) = 0 \varphi(a) = \varphi(b) = 0 φ ( a ) = φ ( b ) = 0 , and (since d ≥ 4 d \ge4 d ≥ 4 ) u ∉ ⟨ a , b ⟩ u \notin\langle a,b\rangle u ∈ / ⟨ a , b ⟩ with φ ( u ) = 0 \varphi(u) = 0 φ ( u ) = 0 . Put N = u φ N = u\varphi N = u φ and A = I + N A = I + N A = I + N . Then N 2 = u φ ( u ) φ = 0 N^2 = u\varphi(u)\varphi= 0 N 2 = u φ ( u ) φ = 0 , so A 2 = I A^2 = I A 2 = I ; and A A A fixes 0 , a , b 0,a,b 0 , a , b pointwise while A ≠ I A \ne I A = I . Hence ( 0 , A ) ∈ G Δ ∖ { 1 } (0,A) \in G_\Delta\setminus\{1\} ( 0 , A ) ∈ G Δ ∖ { 1 } . For H = Sp ( 2 m , 2 ) H = \operatorname{Sp}(2m,2) H = Sp ( 2 m , 2 ) let ⟨ , ⟩ \langle\ ,\ \rangle ⟨ , ⟩ be the alternating form and W = ⟨ a , b ⟩ W = \langle a,b\rangle W = ⟨ a , b ⟩ ; then dim W ⊥ = d − 2 \dim W^\perp= d - 2 dim W ⊥ = d − 2 , so for d ≥ 4 d \ge4 d ≥ 4 there is 0 ≠ u ∈ W ⊥ 0 \ne u \in W^\perp 0 = u ∈ W ⊥ . The symplectic transvection t u ( x ) = x + ⟨ x , u ⟩ u t_u(x) = x + \langle x,u\rangle u t u ( x ) = x + ⟨ x , u ⟩ u fixes a , b , 0 a,b,0 a , b , 0 (by ⟨ a , u ⟩ = ⟨ b , u ⟩ = 0 \langle a,u\rangle= \langle b,u\rangle= 0 ⟨ a , u ⟩ = ⟨ b , u ⟩ = 0 ) and satisfies t u ≠ I t_u \ne I t u = I ; hence ( 0 , t u ) ∈ G Δ ∖ { 1 } (0,t_u) \in G_\Delta\setminus\{1\} ( 0 , t u ) ∈ G Δ ∖ { 1 } . The rank-one statement is clear in both cases (A A A is a transvection, t u t_u t u a symplectic transvection). ∎
Proposition Appendix B.25 (The exterior-square module fails the K = 1 K=1 K = 1 case). Let n ≥ 5 n\ge5 n ≥ 5 , E = F 2 n E=\mathbb{F}_2^n E = F 2 n , V = ⋀ 2 E V=\bigwedge^2 E V = ⋀ 2 E , D = dim F 2 V = ( n 2 ) D=\dim_{\mathbb{F}_2}V=\binom{n}{2} D = dim F 2 V = ( 2 n ) , and let H H H be the image of SL ( n , 2 ) = GL ( n , 2 ) \operatorname{SL}(n,2)=\operatorname{GL}(n,2) SL ( n , 2 ) = GL ( n , 2 ) on V V V . Then for any linearly independent a , b ∈ V a,b\in V a , b ∈ V one has ( V ⋊ H ) { 0 , a , b } ≠ 1 (V\rtimes H)_{\{0,a,b\}}\ne1 ( V ⋊ H ) { 0 , a , b } = 1 . Hence the full exterior-square module is excluded at the entry to the K = 1 K=1 K = 1 case; no class-intersection count is needed for it. (For proper subgroups H ′ < H H'<H H ′ < H the order comparison below may fail; the two-point stabiliser and the local class intersections must then be checked separately.)
Proof. The exterior-square action is faithful: if A ∈ GL ( n , 2 ) A\in\operatorname{GL}(n,2) A ∈ GL ( n , 2 ) acts trivially on ⋀ 2 E \bigwedge^2 E ⋀ 2 E , then it preserves every coordinate plane ⟨ e i , e j ⟩ \langle e_i,e_j\rangle ⟨ e i , e j ⟩ ; intersecting the planes shows A A A preserves every coordinate line, and as F 2 × = { 1 } \mathbb{F}_2^\times=\{1\} F 2 × = { 1 } we get A = I A=I A = I . (In characteristic 2 a nonzero decomposable vector e i ∧ e j e_i\wedge e_j e i ∧ e j determines the plane.) On the other hand H H H acts on the set of ordered linearly independent pairs of V V V , of size ( 2 D − 1 ) ( 2 D − 2 ) < 2 2 D = 2 n ( n − 1 ) (2^D-1)(2^D-2)<2^{2D}=2^{n(n-1)} ( 2 D − 1 ) ( 2 D − 2 ) < 2 2 D = 2 n ( n − 1 ) , while
∣ H ∣ = ∣ GL ( n , 2 ) ∣ = 2 n 2 ∏ i = 1 n ( 1 − 2 − i ) > 2 n 2 − 2 > 2 n ( n − 1 ) ( n ≥ 5 ) , \begin{aligned}
|H|=|\operatorname{GL}(n,2)|=2^{n^2}\prod_{i=1}^{n}(1-2^{-i}) \\
&>2^{n^2-2}>2^{n(n-1)}\qquad(n\ge5),
\end{aligned} ∣ H ∣ = ∣ GL ( n , 2 ) ∣ = 2 n 2 i = 1 ∏ n ( 1 − 2 − i ) > 2 n 2 − 2 > 2 n ( n − 1 ) ( n ≥ 5 ) ,
the product bound following from ∏ i = 2 n ( 1 − 2 − i ) ≥ 1 − ∑ i = 2 n 2 − i > 1 2 \prod_{i=2}^{n}(1-2^{-i})\ge1-\sum_{i=2}^{n}2^{-i}>\frac12 ∏ i = 2 n ( 1 − 2 − i ) ≥ 1 − ∑ i = 2 n 2 − i > 2 1 together with the i = 1 i=1 i = 1 factor 1 2 \frac12 2 1 . Hence every two-point stabiliser is nontrivial; such a stabiliser fixes 0 , a , b 0,a,b 0 , a , b pointwise, so K ≠ 1 K\ne1 K = 1 .
Proposition Appendix B.26 (Scalar-restricted natural modules fail the K = 1 K=1 K = 1 case). Let q = 2 f q=2^f q = 2 f , V = Res F 2 F q ( F q m ) V=\operatorname{Res}_{\mathbb{F}_2}^{\mathbb{F}_q}(\mathbb{F}_q^m) V = Res F 2 F q ( F q m ) (m ≥ 2 m\ge2 m ≥ 2 ), and H ≥ SL ( m , q ) H\ge\operatorname{SL}(m,q) H ≥ SL ( m , q ) . If m ≥ 3 m\ge3 m ≥ 3 , or m = 2 m=2 m = 2 and a , b a,b a , b are F q \mathbb{F}_q F q -dependent, then for any F 2 \mathbb{F}_2 F 2 -linearly independent a , b ∈ V a,b\in V a , b ∈ V one has ( V ⋊ H ) { 0 , a , b } ≠ 1 (V\rtimes H)_{\{0,a,b\}}\ne1 ( V ⋊ H ) { 0 , a , b } = 1 : a nonzero F q \mathbb{F}_q F q -linear map fixing a , b a,b a , b gives an element ( 0 , s ) ∈ G Δ ∖ { 1 } (0,s)\in G_\Delta\setminus\{1\} ( 0 , s ) ∈ G Δ ∖ { 1 } , hence K ≠ 1 K\ne1 K = 1 . The remaining case
m = 2 m=2 m = 2 with a , b a,b a , b F q \mathbb{F}_q F q -independent must be treated by the class-intersection count with H a , b = 1 H_{a,b}=1 H a , b = 1 (or its two-point stabiliser); likewise, when H ≤ ΓL m ( 2 f ) H\leq\operatorname{ΓL}_m(2^f) H ≤ ΓL m ( 2 f ) does not contain SL ( m , 2 f ) \operatorname{SL}(m,2^f) SL ( m , 2 f ) , one partitions the involutions by F q \mathbb{F}_q F q -linear part and field-automorphism coset and solves the three-point image equations, retaining elements of global F 2 \mathbb{F}_2 F 2 -rank > 1 >1 > 1 with local rank one on U w U_w U w .
Proposition Appendix B.27. (The Mathieu case M 11 M_{11} M 11 on the augmentation module). Let H = M 11 H=M_{11} H = M 11 act sharply 4-transitively on Ω \Omega Ω , ∣ Ω ∣ = 11 |\Omega|=11 ∣Ω∣ = 11 , and let
V = { x ∈ F 2 Ω : ∑ ω ∈ Ω x ω = 0 } , V=\left\{x\in\mathbb{F}_2^\Omega:\sum_{\omega\in\Omega}x_\omega=0\right\}, V = { x ∈ F 2 Ω : ω ∈ Ω ∑ x ω = 0 } , dim F 2 V = 10 , \dim_{\mathbb{F}_2}V=10, dim F 2 V = 10 , be the augmentation submodule of the natural 11-point permutation module (its irreducibility as an F 2 H \mathbb{F}_2H F 2 H -module is standard). Then there exist a , b ∈ V a,b\in V a , b ∈ V such that K = 1 K=1 K = 1 for Δ = { 0 , a , b } \Delta=\{0,a,b\} Δ = { 0 , a , b } , and for this Δ \Delta Δ the two- and four-cycle bad points satisfy
∣ B ( 2 ) ∪ B ( 2 , 2 ) ∣ ≤ 6 ⋅ 165 = 990 < 1020 = 2 10 − 4 = ∣ V ∖ W ∣ . \begin{aligned}
|B_{(2)}\cup B_{(2,2)}|&\leq6\cdot165=990<1020\\
&=2^{10}-4=|V\setminus W|.
\end{aligned} ∣ B ( 2 ) ∪ B ( 2 , 2 ) ∣ ≤ 6 ⋅ 165 = 990 < 1020 = 2 10 − 4 = ∣ V ∖ W ∣. Thus the K = 1 K=1 K = 1 case (c1) holds for this nonclassical irreducible module without a class-by-class count of m π ( C ; Δ ) m_\pi(C;\Delta) m π ( C ; Δ ) : at least 30 points of V ∖ W V\setminus W V ∖ W are left uncovered.
Proof. Identify vectors of V V V with even-sized subsets of Ω \Omega Ω (symmetric difference as addition). Fix an ordered partition Ω = X ⊔ P ⊔ R ⊔ Q \Omega=X\sqcup P\sqcup R\sqcup Q Ω = X ⊔ P ⊔ R ⊔ Q with sizes 1, 3, 5, 2; there are
( 11 1 ) ( 10 3 ) ( 7 5 ) = 27720 \binom{11}{1}\binom{10}{3}\binom{7}{5}=27720 ( 1 11 ) ( 3 10 ) ( 5 7 ) = 27720
such partitions. Among the standard cycle shapes of M 11 M_{11} M 11 on 11 points only the involution shape 1 3 2 4 1^3 2^4 1 3 2 4 can preserve such a partition (any other nontrivial cycle shape contains a cycle that cannot be placed into a cell of size at most 5, or cannot be assigned). An involution of shape 1 3 2 4 1^3 2^4 1 3 2 4 preserves at most 3 ⋅ 4 ⋅ 3 ⋅ 2 = 72 3\cdot4\cdot3\cdot2=72 3 ⋅ 4 ⋅ 3 ⋅ 2 = 72 partitions (choose the fixed point of the singleton cell X X X , a 2-cycle of Q Q Q , a remaining 2-cycle assigned to P P P , and then the two remaining fixed points assigned to P , R P,R P , R ). Since M 11 M_{11} M 11 has 165 involutions, all nontrivial elements together preserve at most 165 ⋅ 72 = 11880 < 27720 165\cdot72=11880<27720 165 ⋅ 72 = 11880 < 27720 partitions; hence some partition X ⊔ P ⊔ R ⊔ Q X\sqcup P\sqcup R\sqcup Q X ⊔ P ⊔ R ⊔ Q has trivial setwise stabiliser.
Set A = X ∪ P A=X\cup P A = X ∪ P , B = X ∪ R B=X\cup R B = X ∪ R , a = 1 A a=1_A a = 1 A , b = 1 B b=1_B b = 1 B . Then a , b a,b a , b have weights 4, 6, and a + b = 1 P ∪ R a+b=1_{P\cup R} a + b = 1 P ∪ R has weight 8. If h ∈ H h\in H h ∈ H fixes a , b a,b a , b , then h h h preserves A ∩ B = X A\cap B=X A ∩ B = X , A ∖ B = P A\setminus B=P A ∖ B = P , B ∖ A = R B\setminus A=R B ∖ A = R and Ω ∖ ( A ∪ B ) = Q \Omega\setminus(A\cup B)=Q Ω ∖ ( A ∪ B ) = Q , hence the partition, so h = 1 h = 1 h = 1 . Any H H H -element preserves Hamming weight; the three edge difference vectors a a a , b b b , a + b a+b a + b have pairwise distinct weights 4, 6, 8, so any affine transformation preserving { 0 , a , b } \{0,a,b\} { 0 , a , b } must fix these three points pointwise, whence its translation part is 0 and its linear part fixes a a a , b b b ; it is therefore trivial. Thus K = 1 K = 1 K = 1 .
For the covering bound, M 11 M_{11} M 11 has 165 involutions. For each nontrivial admissible marking π \pi π (there are six, as ∣ P i n v ∣ = 6 \lvert P_{\mathrm{inv}}\rvert= 6 ∣ P inv ∣ = 6 in the K = 1 K = 1 K = 1 case), once the linear part A A A is fixed the marking determines one image point and hence the affine translation part, so at most one bad point w w w per involution and marking: at most 6 ⋅ 165 = 990 6 \cdot165 = 990 6 ⋅ 165 = 990 bad points in total. A bad affine involution with trivial linear part would be a pure translation; on the invariant four-point set S w = { 0 , a , b , w } S_w = \{0,a,b,w\} S w = { 0 , a , b , w } it pairs the points into two translation-paired pairs, so the four points sum to zero: a + b + w = 0 a+b+w = 0 a + b + w = 0 , i.e. w = a + b ∈ W = ⟨ a , b ⟩ w = a+b \in W = \langle a,b\rangle w = a + b ∈ W = ⟨ a , b ⟩ , contradicting w ∉ W w \notin W w ∈ / W . Hence no pure translations contribute, and ∣ B ( 2 ) ∪ B ( 2 , 2 ) ∣ ≤ 990 < 1020 \lvert B_{(2)} \cup B_{(2,2)}\rvert\le990 < 1020 ∣ B ( 2 ) ∪ B ( 2 , 2 ) ∣ ≤ 990 < 1020 .
Proposition Appendix B.28 (The Mathieu case M 23 M_{23} M 23 on the Golay quotient module fails K = 1 K = 1 K = 1 ). Let C 24 C_{24} C 24 be the binary extended Golay code , 1 ∈ C 24 \mathbf{1} \in C_{24} 1 ∈ C 24 the all-one vector, and let M 23 M_{23} M 23 fix coordinate 24. Put
V = C 24 / ⟨ 1 ⟩ , dim F 2 V = 11 , V = C_{24}/\langle\mathbf{1}\rangle,\qquad\dim_{\mathbb{F}_2} V = 11, V = C 24 / ⟨ 1 ⟩ , dim F 2 V = 11 , with H = M 23 H = M_{23} H = M 23 acting faithfully (the action is nontrivial and M 23 M_{23} M 23 is simple). Then for every linearly independent a a a , b ∈ V b \in V b ∈ V the two-point stabiliser H a , b H_{a,b} H a , b is nontrivial; in particular K ≠ 1 K \ne1 K = 1 . Hence the Golay quotient module is excluded at the entry to the K = 1 K = 1 K = 1 case (c1): no class-intersection count is needed for it.
Proof. The nonzero vectors of V V V form three orbits, represented by codewords of weights 8, 12 and 16; the weight-16 vectors are octad complements. Computer enumeration (Remark Appendix B.29) computes min b ∣ H a , b ∣ \min_b \lvert H_{a,b}\rvert min b ∣ H a , b ∣ : the minimum is 10 for weight-12 representatives and 24 for weight-8 representatives. Thus H a , b ≠ 1 H_{a,b} \ne1 H a , b = 1 for all pairs, and the affine stabiliser K K K contains nontrivial linear elements ( 0 , h ) (0,h) ( 0 , h ) .
M 23 M_{23} M 23 on V V V were computed; the resulting group has order 10200960 = ∣ M 23 ∣ 10200960 = \lvert M_{23}\rvert 10200960 = ∣ M 23 ∣ , confirming faithfulness. For each of the weight-8 and weight-12 representatives a a a all 2046 admissible b b b were enumerated and ∣ H a , b ∣ \lvert H_{a,b}\rvert ∣ H a , b ∣ computed; the minima are 24 and 10 respectively (the size values observed for weight 12 are 10, 24, 48, 72, 120, 360). In contrast with Proposition Appendix B.27, the vectors of the Golay quotient module are not fixed-weight sets ([ x ] = [ x + 1 ] [x] = [x + 1] [ x ] = [ x + 1 ] in the quotient), so the M 11 M_{11} M 11 block-partition argument cannot be imported; the enumeration above is the certificate.