Introduction
For a finite group G G G , the normal subgroups form a finite modular lattice N ( G ) \mathcal{N}(G) N ( G ) whose atoms are the minimal normal subgroups of G G G . Their product is the socle Soc ( G ) \operatorname{Soc}(G) Soc ( G ) , which is also the least essential normal subgroup: it meets every non-trivial normal subgroup, and every essential normal subgroup contains it. Thus the socle provides a natural point of departure for studying intersection properties in N ( G ) \mathcal{N}(G) N ( G ) . In this paper we identify a class of normal subgroups defined by such a property and show that, for finite groups, its structure is governed by the socle and by the geometry of minimal normal subgroups.
A normal subgroup E ⊴ G E\trianglelefteq G E ⊴ G is called an M \mathcal{M} M -subgroup if, for all normal subgroups N 1 , N 2 ⊴ G N_1,N_2\trianglelefteq G N 1 , N 2 ⊴ G ,
N 1 ∩ N 2 ≠ 1 , E ∩ N 1 ≠ 1 , E ∩ N 2 ≠ 1 ⟹ E ∩ N 1 ∩ N 2 ≠ 1. \begin{gathered}
N_1\cap N_2\ne1,\qquad E\cap N_1\ne1,\qquad E\cap N_2\ne1\\
\Longrightarrow E\cap N_1\cap N_2\ne1.
\end{gathered} N 1 ∩ N 2 = 1 , E ∩ N 1 = 1 , E ∩ N 2 = 1 ⟹ E ∩ N 1 ∩ N 2 = 1.
As we prove in Proposition 3.3 , this condition is equivalent to the analogous implication for any finite family of normal subgroups. We write N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) for the collection of M \mathcal{M} M -subgroups and call G G G M \mathcal{M} M -full when
2020 Mathematics Subject Classification. Primary 20D30; Secondary 20D15, 20C20, 20E22, 06C05.
Key words and phrases. M \mathcal{M} M -subgroup, μ \mu μ -element, essential subgroup, minimal normal subgroup, socle length, M \mathcal{M} M -depth, direct product, crown-based power, normal-subgroup lattice. N M ( G ) = N ( G ) \mathcal{N}^{\mathcal{M}}(G)=\mathcal{N}(G) N M ( G ) = N ( G ) . Our principal questions are to characterise these subgroups, determine when G G G is M \mathcal{M} M -full, and describe the resulting lattice for direct products and crown-based powers.
The intersection condition is motivated by the theory of M \mathcal{M} M -ideals in Banach spaces, initiated by Alfsen and Effros [1 ] (see also [14 ] ), and its later analogues for ring ideals and quantales. In particular, [3 ] introduced ring M \mathcal{M} M -ideals, while [6 ] studied μ \mu μ -elements of quantales. The dichotomy between essentiality and a relative uniformity condition in these settings motivates the group case. Our arguments use the modular normal-subgroup lattice directly, rather than applying the quantale results.
Let us now discuss our main results. We first show, for arbitrary groups, that a normal subgroup is an M \mathcal{M} M -subgroup if and only if it is essential or relatively uniform (Theorem 3.4 ); the latter means that any two non-trivial normal subgroups contained in it meet non-trivially. For finite groups, this becomes a classification in terms of minimal normal subgroups. Write MinNorm ( G ) \operatorname{MinNorm}(G) MinNorm ( G ) for their set and put supp ( N ) = { A ∈ MinNorm ( G ) : A ≤ N } \operatorname{supp}(N)=\{A\in\operatorname{MinNorm}(G): A\leq N\} supp ( N ) = { A ∈ MinNorm ( G ) : A ≤ N } for N ⊴ G N\unlhd G N ⊴ G .
Theorem A. Let G G G be finite and let E ⊴ G E\unlhd G E ⊴ G . Then E E E is an M \mathcal{M} M -subgroup if and only if
E = 1 , ∣ supp ( E ) ∣ = 1 , or Soc ( G ) ≤ E . E=1,\qquad|\operatorname{supp}(E)|=1,\qquad\text{or}\qquad\operatorname{Soc}(G)\leq E. E = 1 , ∣ supp ( E ) ∣ = 1 , or Soc ( G ) ≤ E .
The M \mathcal{M} M -subgroups are closed under intersections. Moreover, the least M \mathcal{M} M -subgroup containing a normal subgroup N N N is
cl M ( N ) = { N , ∣ supp ( N ) ∣ ≤ 1 , N Soc ( G ) , ∣ supp ( N ) ∣ ≥ 2. \operatorname{cl}_{\mathcal{M}}(N)=
\begin{cases}
N, & |\operatorname{supp}(N)|\leq1,\\
N\operatorname{Soc}(G), & |\operatorname{supp}(N)|\geq2.
\end{cases} cl M ( N ) = { N , N Soc ( G ) , ∣ supp ( N ) ∣ ≤ 1 , ∣ supp ( N ) ∣ ≥ 2.
Consequently they form a lattice whose join is E ∨ M F = cl M ( E F ) E\vee_{\mathcal{M}}F=\operatorname{cl}_{\mathcal{M}}(EF) E ∨ M F = cl M ( E F ) .
The classification extends to normally atomic groups (Theorem 3.7 ). In the finite case it also provides an explicit closure operator on N ( G ) \mathcal{N}(G) N ( G ) , and hence a lattice structure on N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) (Proposition 3.11 and Corollary 3.13 ). In particular, the normal subgroups excluded from N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) are precisely those containing more than one, but not all, of the minimal normal subgroups of G G G .
Let σ ( G ) \sigma(G) σ ( G ) denote the length of the interval [ 1 , Soc ( G ) ] N ( G ) [1,\operatorname{Soc}(G)]_{\mathcal{N}(G)} [ 1 , Soc ( G ) ] N ( G ) ; equivalently, it is the number of factors in a decomposition of the socle into a direct product of minimal normal subgroups. The criterion for M \mathcal{M} M -fullness is then numerical.
Theorem B. Let G G G be a non-trivial finite group and put S = Soc ( G ) S=\operatorname{Soc}(G) S = Soc ( G ) . The following are equivalent:
every normal subgroup of G G G is an M \mathcal{M} M -subgroup;
σ ( G ) ≤ 2 \sigma(G)\leq2 σ ( G ) ≤ 2 ;
one of the following holds: (a) G G G has one minimal normal subgroup; (b) G G G has exactly two minimal normal subgroups; (c) for some prime p p p and some irreducible F p G \mathbb{F}_pG F p G -module A A A , one has S ≅ A ⊕ A S\cong A\oplus A S ≅ A ⊕ A as G G G -modules.
In alternative (c), if D = End F p G ( A ) D=\operatorname{End}_{\mathbb{F}_pG}(A) D = End F p G ( A ) , then MinNorm ( G ) \operatorname{MinNorm}(G) MinNorm ( G ) is naturally the projective line P 1 ( D ) \mathbb{P}^{1}(D) P 1 ( D ) and has ∣ D ∣ + 1 |D|+1 ∣ D ∣ + 1 elements.
Thus the relevant invariant is the length of the socle, not the number of its minimal normal subgroups. For example, C p × C p C_p\times C_p C p × C p has p + 1 p+1 p + 1 minimal normal subgroups, but its socle has length two, so it is M \mathcal{M} M -full. More generally, the abelian isotypic components of the socle give rise to finite projective geometries (Proposition 4.4 ). The classification also has a lattice-theoretic formulation: for arbitrary G G G , M \mathcal{M} M -fullness is equivalent to the normal-subgroup lattice having uniform (Goldie) dimension at most two (Theorem 4.11 ).
Direct products and crown-based powers make these results concrete. The minimal normal subgroups of a direct product need not come from the factors: central elementary abelian subgroups also produce diagonal minimal normal subgroups. A crown-based power, introduced by Dalla Volta and Lucchini [8 ] , is an important construction in the structure and generation theory of finite groups; in particular, certain quotients G / R G ( B ) G/R_G(B) G / R G ( B ) attached to non-Frattini chief factors are crown-based powers [10 , 19 ] . Burness, Garonzi and Lucchini [4, Section 5] use crown theory to bound the intersection number of a finite group, which is defined using intersections of maximal subgroups. Our intersection condition, by contrast, concerns normal subgroups. We describe the atoms of the corresponding normal-subgroup lattices and determine M \mathcal{M} M -fullness.
Theorem C. Let G = G 1 × ⋯ × G t G=G_1\times\cdots\times G_t G = G 1 × ⋯ × G t be finite. For each prime p p p , set
V i , p = { z ∈ Z ( G i ) : z p = 1 } , V p = ∏ i = 1 t V i , p . V_{i,p}=\{z\in Z(G_i):z^p=1\},\qquad V_p=\prod_{i=1}^{t}V_{i,p}. V i , p = { z ∈ Z ( G i ) : z p = 1 } , V p = i = 1 ∏ t V i , p . The minimal normal subgroups of G G G are precisely the coordinate copies of the noncentral minimal normal subgroups of the factors, together with the one-dimensional F p \mathbb{F}_p F p -subspaces of the spaces V p V_p V p . If t ≥ 2 t\geq2 t ≥ 2 and all factors are non-trivial, then G G G is M \mathcal{M} M -full exactly when t = 2 t=2 t = 2 and both factors are monolithic.
If L L L is a finite monolithic primitive group with socle A A A and L k L_k L k is its crown-based power, then Soc ( L k ) = A k \operatorname{Soc}(L_k)=A^k Soc ( L k ) = A k . If A A A is non-abelian, the atoms are the k k k coordinate copies of A A A . If A A A is abelian, then A A A is an irreducible F p [ L / A ] \mathbb{F}_p[L/A] F p [ L / A ] -module for a prime p p p ; with D = End F p [ L / A ] ( A ) D=\operatorname{End}_{\mathbb{F}_p[L/A]}(A) D = End F p [ L / A ] ( A ) , the atoms are parametrized by P k − 1 ( D ) \mathbb{P}^{k-1}(D) P k − 1 ( D ) . In either case every normal subgroup of L k L_k L k is an M \mathcal{M} M -subgroup exactly when k ≤ 2 k\leq2 k ≤ 2 .
The comparability of every normal subgroup of a crown-based power L k L_k L k with its socle V = A k V=A^k V = A k is known by a lemma of Lucchini, Moscatiello and Spiga [18, Lemma 2.2]. Using this result, we identify the lower interval of N ( L k ) \mathcal{N}(L_k) N ( L k ) with a Boolean lattice or a subspace lattice and the upper interval with $\mathcal{N}(L/A) ({xref}Theorem 6.1). Our subsequent results concern the information retained by $\mathcal{N}^{\mathcal{M}}(L_k)$: we give a rigidity criterion comparing normal-subgroup lattices and $\mathcal{M}$-lattices ({xref}Theorem 6.5), classify the quotients by arbitrary normal subgroups contained in $V$ together with their automorphism orbits ({xref}Theorem 6.6), and determine precisely which quotient maps preserve $\mathcal{M}$-closure ({xref}Proposition 6.7). The description also gives enumeration formulas ({xref}Corollary 6.2) and applies to crowns associated with arbitrary finite groups ({xref}Corollary 6.10`).
Finally, the omission of normal subgroups from N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) can be measured by a chain-length invariant. Define the M \mathcal{M} M -depth md ( G ) \operatorname{md}(G) md ( G ) to be the maximum length of a chain from 1 1 1 to G G G in N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) , and let ℓ n ( G ) \ell_n(G) ℓ n ( G ) be the chief length of G G G .
Theorem D. Let G G G be a non-trivial finite group. If G G G has one minimal normal subgroup, then
md ( G ) = ℓ n ( G ) . \operatorname{md}(G)=\ell_n(G). md ( G ) = ℓ n ( G ) . If G G G has at least two minimal normal subgroups, then
md ( G ) = ℓ n ( G ) − σ ( G ) + 2. \operatorname{md}(G)=\ell_n(G)-\sigma(G)+2. md ( G ) = ℓ n ( G ) − σ ( G ) + 2. In particular, if G 1 , … , G t G_1,\ldots,G_t G 1 , … , G t are non-trivial finite groups and t ≥ 2 t\geq2 t ≥ 2 , then
md ( G 1 × ⋯ × G t ) = ∑ i = 1 t ( ℓ n ( G i ) − σ ( G i ) ) + 2. \operatorname{md}(G_1\times\cdots\times G_t)=\sum_{i=1}^{t}\bigl(\ell_n(G_i)-\sigma(G_i)\bigr)+2. md ( G 1 × ⋯ × G t ) = i = 1 ∑ t ( ℓ n ( G i ) − σ ( G i ) ) + 2.
In fact, every maximal chain of M \mathcal{M} M -subgroups has the same length. When G G G has at least two minimal normal subgroups, such a chain follows a chief series below a normal subgroup containing a unique minimal normal subgroup, crosses the socle in one step, and continues along a chief series above the socle (Corollary 8.3 ). Consequently, md ( G ) = ℓ n ( G ) \operatorname{md}(G)=\ell_n(G) md ( G ) = ℓ n ( G ) holds precisely for M \mathcal{M} M -full finite groups (Corollary 8.4 ). We also compute the depth for several standard families, including finite abelian and nilpotent groups, dihedral groups, and crown-based powers.
The organization of the paper is as follows. In § 2 , we fix notation. In § 3 , we prove the basic dichotomy and the classification of M \mathcal{M} M -subgroups. The purpose of Section 4 is to treat M \mathcal{M} M -fullness and its relation to the socle, whereas in § 5 and § 6 we address direct products and crown-based powers. In § 7 , we specialise the results to finite abelian groups, and finally, in § 8 , we study M \mathcal{M} M -depth.
Definitions and notation
Groups are arbitrary unless finiteness is stated explicitly. We begin with the lattice of normal subgroups and the standard lattice-theoretic vocabulary that we use for it.
Definition 2.1. For a group G G G , write
N ( G ) = { N ≤ G : N ⊴ G } \mathcal{N}(G)=\{N\leq G:N\trianglelefteq G\} N ( G ) = { N ≤ G : N ⊴ G } for the poset of normal subgroups ordered by inclusion, and write Sub ( G ) \operatorname{Sub}(G) Sub ( G ) for the poset of all subgroups. The least element of N ( G ) \mathcal{N}(G) N ( G ) is the trivial subgroup 1 1 1 and the greatest element is G G G . For H , K ⊴ G H,K\trianglelefteq G H , K ⊴ G , their meet is H ∩ K H\cap K H ∩ K and their join is the subgroup product H K = { h k : h ∈ H , k ∈ K } HK=\{hk:h\in H,\ k\in K\} H K = { hk : h ∈ H , k ∈ K } . For H ≤ K H\leq K H ≤ K in a poset P P P , the interval [ H , K ] P [H,K]_P [ H , K ] P is { X ∈ P : H ≤ X ≤ K } \{X\in P:H\leq X\leq K\} { X ∈ P : H ≤ X ≤ K } , and the principal filter generated by H H H is [ H , G ] P [H,G]_P [ H , G ] P when G G G is the greatest element. An atom of a poset with least element is a minimal element strictly larger than the least one. A lattice is modular if
x ∨ ( y ∧ z ) = ( x ∨ y ) ∧ z ( x ≤ z ) . x\vee(y\wedge z)=(x\vee y)\wedge z \qquad(x\leq z). x ∨ ( y ∧ z ) = ( x ∨ y ) ∧ z ( x ≤ z ) . A meet-subsemilattice is a subset closed under finite nonempty meets. The length of a finite chain X 0 < X 1 < ⋯ < X r X_0<X_1<\cdots<X_r X 0 < X 1 < ⋯ < X r is r r r ; the length of a finite interval is the maximum length of a chain joining its endpoints.
The group-theoretic notation is standard [15 , 20 ] ; we record the conventions that matter for computations, notably the order of factors in commutators.
Definition 2.2. For subgroups A , B ≤ G A,B\leq G A , B ≤ G , our commutator convention is [ a , b ] = a b a − 1 b − 1 [a,b]=aba^{-1}b^{-1} [ a , b ] = ab a − 1 b − 1 , and [ A , B ] [A,B] [ A , B ] denotes the subgroup generated by these commutators. We write Z ( G ) Z(G) Z ( G ) for the centre of G G G and C n C_n C n for the cyclic group of order n n n . As usual, an elementary abelian p p p -group is regarded as a vector space over F p \mathbb{F}_p F p . All other elementary group-theoretic terminology is standard.
The atoms of N ( G ) \mathcal{N}(G) N ( G ) are the minimal normal subgroups, and they are the central objects of this paper. The following notions are classical for finite groups [15 , 20 ] ; we state them for arbitrary groups, since some of our results hold in that generality.
Definition 2.3. A minimal normal subgroup of G G G is a non-trivial normal subgroup minimal under inclusion. Thus the minimal normal subgroups are exactly the atoms of N ( G ) \mathcal{N}(G) N ( G ) . Write MinNorm ( G ) \operatorname{MinNorm}(G) MinNorm ( G ) for their set and define the socle
Soc ( G ) = ⟨ A : A ∈ MinNorm ( G ) ⟩ . \operatorname{Soc}(G)=\langle A:A\in\operatorname{MinNorm}(G)\rangle. Soc ( G ) = ⟨ A : A ∈ MinNorm ( G )⟩ . For N ⊴ G N\trianglelefteq G N ⊴ G , define its atom support by
supp ( N ) = { A ∈ MinNorm ( G ) : A ≤ N } . \operatorname{supp}(N)=\{A\in\operatorname{MinNorm}(G):A\leq N\}. supp ( N ) = { A ∈ MinNorm ( G ) : A ≤ N } . The group G G G is normally atomic if every non-trivial normal subgroup contains a minimal normal subgroup. It is monolithic if it has a unique minimal normal subgroup.
The group Z \mathbb{Z} Z is not normally atomic, whereas all finite groups are. The Prüfer p p p -group Z ( p ∞ ) ≤ Q / Z \mathbb{Z}(p^\infty)\leq\mathbb{Q}/\mathbb{Z} Z ( p ∞ ) ≤ Q / Z is normally atomic and monolithic, with ⟨ 1 p + Z ⟩ \langle\frac{1}{p}+\mathbb{Z}\rangle ⟨ p 1 + Z ⟩ as its unique minimal normal subgroup.
A finite group G G G is monolithic if and only if the intersection of all non-trivial normal subgroups is non-trivial; equivalently, it is subdirectly irreducible in the sense of universal algebra [5 Chapter II, Section 8] . In general, a non-trivial group is subdirectly irreducible precisely when it has a least non-trivial normal subgroup. Thus the equivalence with monolithic, as defined above, fails for infinite groups. For example, in C p × Z C_p\times\mathbb{Z} C p × Z , where p p p is prime, C p × 0 C_p\times0 C p × 0 is the unique minimal normal subgroup, but the intersection of all non-trivial normal subgroups is trivial; of course 0 × Z 0\times\mathbb{Z} 0 × Z contains no minimal normal subgroup, so this group is not normally atomic.
The two halves of the dichotomy proved in § 3 are the following notions. Essential normal subgroups of arbitrary groups are studied in [17 ] ; relative uniformity is the group analogue of the relatively irreducible ideals of [3 ] and of the irreducible elements of [6 ] .
Definition 2.4. A normal subgroup E ⊴ G E \trianglelefteq G E ⊴ G is essential in G G G if E ∩ N ≠ 1 E \cap N \ne1 E ∩ N = 1 for every non-trivial N ⊴ G N \trianglelefteq G N ⊴ G . It is relatively uniform in G G G if any two non-trivial normal subgroups of G G G contained in E E E have non-trivial intersection.
The central definition of the paper transfers the ring M \mathcal{M} M -ideals of [3 ] and the μ \mu μ -elements of [6 ] to the lattice N ( G ) \mathcal{N}(G) N ( G ) .
Definition 2.5. A normal subgroup E ⊴ G E \trianglelefteq G E ⊴ G is an M \mathcal{M} M -subgroup if, for every finite family N 1 , … , N r N_1,\ldots,N_r N 1 , … , N r of non-trivial normal subgroups with r ≥ 2 r \ge2 r ≥ 2 satisfying
⋂ i = 1 r N i ≠ 1 and E ∩ N i ≠ 1 ( 1 ≤ i ≤ r ) , \bigcap_{i=1}^{r} N_i \ne1 \quad\text{and} \quad E \cap N_i \ne1 \quad(1 \le i \le r), i = 1 ⋂ r N i = 1 and E ∩ N i = 1 ( 1 ≤ i ≤ r ) , one has
E ∩ ⋂ i = 1 r N i ≠ 1. E \cap\bigcap_{i=1}^{r} N_i \ne1. E ∩ i = 1 ⋂ r N i = 1.
The trivial subgroup satisfies this condition vacuously. Write
N M ( G ) = { E ⊴ G : E is an M -subgroup } . \mathcal{N}^{\mathcal{M}}(G) = \{E \trianglelefteq G : E \text{ is an } \mathcal{M}\text{-subgroup}\}. N M ( G ) = { E ⊴ G : E is an M -subgroup } .
A group G G G is M \mathcal{M} M -full if N M ( G ) = N ( G ) \mathcal{N}^{\mathcal{M}}(G) = \mathcal{N}(G) N M ( G ) = N ( G ) .
Trivially, for every group G G G the subgroups 1 1 1 and G G G are M \mathcal{M} M -subgroups, and so is every minimal normal subgroup (for instance every normal subgroup of prime order), because two normal subgroups meeting it non-trivially both contain it. The first part of the following example computes N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) directly from the definition in a group with three minimal normal subgroups; the second part exhibits an infinite family of M \mathcal{M} M -full groups.
Example 2.6. (1) Let G = D 60 G = D_{60} G = D 60 be the dihedral group of the 30-gon, generated by r r r and s s s , satisfying r 30 = 1 = s 2 r^{30} = 1 = s^2 r 30 = 1 = s 2 and s r s = r − 1 srs = r^{-1} sr s = r − 1 . Every non-trivial normal subgroup of ⟨ r ⟩ \langle r\rangle ⟨ r ⟩ contains one of ⟨ r 15 ⟩ \langle r^{15}\rangle ⟨ r 15 ⟩ , ⟨ r 10 ⟩ \langle r^{10}\rangle ⟨ r 10 ⟩ , ⟨ r 6 ⟩ \langle r^6\rangle ⟨ r 6 ⟩ . Moreover, if a normal subgroup N N N is not contained in ⟨ r ⟩ \langle r\rangle ⟨ r ⟩ , then s r i ∈ N sr^i \in N s r i ∈ N for some 0 ≤ i < 30 0 \le i < 30 0 ≤ i < 30 . Since N N N is normal, r ( s r i ) r − 1 = s r i − 2 ∈ N r(sr^i)r^{-1} = sr^{i-2} \in N r ( s r i ) r − 1 = s r i − 2 ∈ N and we get
( s r i ) ( s r i − 2 ) − 1 = s r 2 s = r − 2 ∈ N . (sr^i)(sr^{i-2})^{-1} = sr^2s = r^{-2} \in N. ( s r i ) ( s r i − 2 ) − 1 = s r 2 s = r − 2 ∈ N .
Therefore ⟨ r 2 ⟩ ⊆ N \langle r^2\rangle\subseteq N ⟨ r 2 ⟩ ⊆ N and hence ⟨ r 10 ⟩ ⊆ N \langle r^{10}\rangle\subseteq N ⟨ r 10 ⟩ ⊆ N . Such a normal subgroup N N N either contains ⟨ r 2 , s ⟩ \langle r^2,s\rangle ⟨ r 2 , s ⟩ or ⟨ r 2 , r s ⟩ \langle r^2,rs\rangle ⟨ r 2 , r s ⟩ , according as i i i is even or odd. The group D 60 D_{60} D 60 has 11 normal subgroups
N ( D 60 ) = { 1 , ⟨ r 15 ⟩ , ⟨ r 10 ⟩ , ⟨ r 6 ⟩ , ⟨ r 5 ⟩ , ⟨ r 3 ⟩ , ⟨ r 2 ⟩ , ⟨ r 2 , s ⟩ , ⟨ r 2 , r s ⟩ , ⟨ r ⟩ , D 60 } . \mathcal{N}(D_{60}) = \{1,\langle r^{15}\rangle,\langle r^{10}\rangle,\langle r^6\rangle,\langle r^5\rangle,\langle r^3\rangle,\langle r^2\rangle,\langle r^2,s\rangle,\langle r^2,rs\rangle,\langle r\rangle,D_{60}\}. N ( D 60 ) = { 1 , ⟨ r 15 ⟩ , ⟨ r 10 ⟩ , ⟨ r 6 ⟩ , ⟨ r 5 ⟩ , ⟨ r 3 ⟩ , ⟨ r 2 ⟩ , ⟨ r 2 , s ⟩ , ⟨ r 2 , r s ⟩ , ⟨ r ⟩ , D 60 } .
The collection of M \mathcal{M} M -subgroups is precisely
N M ( D 60 ) = { 1 , ⟨ r 15 ⟩ , ⟨ r 10 ⟩ , ⟨ r 6 ⟩ , ⟨ r ⟩ , D 60 } . \mathcal{N}^{\mathcal{M}}(D_{60}) = \{1,\langle r^{15}\rangle,\langle r^{10}\rangle,\langle r^6\rangle,\langle r\rangle,D_{60}\}. N M ( D 60 ) = { 1 , ⟨ r 15 ⟩ , ⟨ r 10 ⟩ , ⟨ r 6 ⟩ , ⟨ r ⟩ , D 60 } .
The obvious M \mathcal{M} M -subgroups are { 1 , ⟨ r 15 ⟩ , D 60 } \{1,\langle r^{15}\rangle,D_{60}\} { 1 , ⟨ r 15 ⟩ , D 60 } . If E ∈ { ⟨ r 10 ⟩ , ⟨ r 6 ⟩ } E \in\{\langle r^{10}\rangle,\langle r^6\rangle\} E ∈ {⟨ r 10 ⟩ , ⟨ r 6 ⟩} , then for any normal subgroup N i N_i N i with E ∩ N i ≠ 1 E \cap N_i \ne1 E ∩ N i = 1 , E ∩ N i E \cap N_i E ∩ N i is a non-trivial subgroup of E E E , which has prime order, and so E ≤ N i E \le N_i E ≤ N i . Hence E ∩ ⋂ i N i = E ≠ 1 E \cap\bigcap_i N_i = E \ne1 E ∩ ⋂ i N i = E = 1 . Every non-trivial normal subgroup contains one of the three minimal normal subgroups ⟨ r 15 ⟩ \langle r^{15}\rangle ⟨ r 15 ⟩ , ⟨ r 10 ⟩ \langle r^{10}\rangle ⟨ r 10 ⟩ , ⟨ r 6 ⟩ \langle r^6\rangle ⟨ r 6 ⟩ , all of which lie in ⟨ r ⟩ \langle r\rangle ⟨ r ⟩ ; hence ⟨ r ⟩ \langle r\rangle ⟨ r ⟩ meets every non-trivial normal subgroup, that is, ⟨ r ⟩ \langle r\rangle ⟨ r ⟩ is essential, and an essential normal subgroup is clearly an M \mathcal{M} M -subgroup. Finally, each of ⟨ r 5 ⟩ \langle r^5\rangle ⟨ r 5 ⟩ , ⟨ r 3 ⟩ \langle r^3\rangle ⟨ r 3 ⟩ , ⟨ r 2 ⟩ \langle r^2\rangle ⟨ r 2 ⟩ , ⟨ r 2 , s ⟩ \langle r^2,s\rangle ⟨ r 2 , s ⟩ , ⟨ r 2 , r s ⟩ \langle r^2,rs\rangle ⟨ r 2 , r s ⟩ contains exactly two of the three minimal normal subgroups (the last three contain ⟨ r 10 ⟩ \langle r^{10}\rangle ⟨ r 10 ⟩ and ⟨ r 6 ⟩ \langle r^6\rangle ⟨ r 6 ⟩ but not ⟨ r 15 ⟩ \langle r^{15}\rangle ⟨ r 15 ⟩ ); if E E E is one of them and B B B is the minimal normal subgroup not contained in E E E , while A 1 , A 2 ≤ E A_1,A_2 \le E A 1 , A 2 ≤ E are the other two, then N 1 = A 1 B N_1 = A_1B N 1 = A 1 B and N 2 = A 2 B N_2 = A_2B N 2 = A 2 B satisfy N 1 ∩ N 2 = B ≠ 1 N_1 \cap N_2 = B \ne1 N 1 ∩ N 2 = B = 1 and E ∩ N i ≥ A i E \cap N_i \ge A_i E ∩ N i ≥ A i , but E ∩ N 1 ∩ N 2 = E ∩ B = 1 E \cap N_1 \cap N_2 = E \cap B = 1 E ∩ N 1 ∩ N 2 = E ∩ B = 1 . So none of these five is an M \mathcal{M} M -subgroup. For odd n ≥ 3 n \ge3 n ≥ 3 , a general formula for N M ( D 2 n ) \mathcal{N}^{\mathcal{M}}(D_{2n}) N M ( D 2 n ) can be found in the proof of Proposition 8.10 .
:::
(2) For n ≥ 2 n \ge2 n ≥ 2 , the group S n S_n S n is M \mathcal{M} M -full. Indeed, the lattice N ( S n ) \mathcal{N}(S_n) N ( S n ) is a chain:
N ( S 2 ) = { 1 < S 2 } , N ( S 3 ) = { 1 < A 3 < S 3 } , \mathcal{N}(S_2)=\{1<S_2\}, \qquad\mathcal{N}(S_3)=\{1<A_3<S_3\}, N ( S 2 ) = { 1 < S 2 } , N ( S 3 ) = { 1 < A 3 < S 3 } ,
N ( S 4 ) = { 1 < V 4 < A 4 < S 4 } , N ( S n ) = { 1 < A n < S n } ( n ≥ 5 ) , \mathcal{N}(S_4)=\{1<V_4<A_4<S_4\}, \qquad\mathcal{N}(S_n)=\{1<A_n<S_n\}\quad(n\ge5), N ( S 4 ) = { 1 < V 4 < A 4 < S 4 } , N ( S n ) = { 1 < A n < S n } ( n ≥ 5 ) ,
where V 4 = { 1 , ( 12 ) ( 34 ) , ( 13 ) ( 24 ) , ( 14 ) ( 23 ) } V_4=\{1,(12)(34),(13)(24),(14)(23)\} V 4 = { 1 , ( 12 ) ( 34 ) , ( 13 ) ( 24 ) , ( 14 ) ( 23 )} . For each case, S n S_n S n is monolithic and M \mathcal{M} M -fullness follows from Theorem 4.5 , case (4a).
Since G G G itself is always an M \mathcal{M} M -subgroup, every normal subgroup lies in at least one M \mathcal{M} M -subgroup, so the intersection of all M \mathcal{M} M -subgroups containing it is defined. This gives a closure operator, and it also leads us to name the pieces into which N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) decomposes.
Definition 2.7. For a group G G G and N ⊴ G N\trianglelefteq G N ⊴ G , define
cl M ( N ) = ⋂ { E ⊴ G : E is an M -subgroup and N ≤ E } . \operatorname{cl}_{\mathcal{M}}(N)=\bigcap\{E\trianglelefteq G:E\text{ is an }\mathcal{M}\text{-subgroup and }N\le E\}. cl M ( N ) = ⋂ { E ⊴ G : E is an M -subgroup and N ≤ E } .
When N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) is regarded as an ordered set, it is called the M \mathcal{M} M -poset; when it is a lattice, ∨ M \vee_{\mathcal{M}} ∨ M denotes its join. Put S = Soc ( G ) S=\operatorname{Soc}(G) S = Soc ( G ) and, for A ∈ MinNorm ( G ) A\in\operatorname{MinNorm}(G) A ∈ MinNorm ( G ) , define the A A A -branch
R A = { E ⊴ G : supp ( E ) = { A } } , E ( G ) = [ S , G ] N ( G ) . \mathcal{R}_A=\{E\trianglelefteq G:\operatorname{supp}(E)=\{A\}\}, \qquad\mathcal{E}(G)=[S,G]_{\mathcal{N}(G)}. R A = { E ⊴ G : supp ( E ) = { A }} , E ( G ) = [ S , G ] N ( G ) .
The numerical invariants we use are lengths in N ( G ) \mathcal{N}(G) N ( G ) . Chief series and the Jordan–Hölder theorem for them are standard [15 , 20 ] ; the length σ ( G ) \sigma(G) σ ( G ) of the socle as a G G G -group appears, for instance, in the study of generation of finite groups [7 ] .
Definition 2.8. ([16 Chapter 8, Section B] ) For a finite group G G G , a chief series is a chain
1 = G 0 < G 1 < ⋯ < G n = G 1=G_0<G_1<\cdots<G_n=G 1 = G 0 < G 1 < ⋯ < G n = G of normal subgroups of G G G such that no normal subgroup of G G G lies strictly between G i − 1 G_{i-1} G i − 1 and G i G_i G i . Its number n n n of factors is the chief length (or normal length) and is denoted ℓ n ( G ) \ell_n(G) ℓ n ( G ) . For N ⊴ G N\trianglelefteq G N ⊴ G , let λ G ( N ) \lambda_G(N) λ G ( N ) be the length of the interval [ 1 , N ] N ( G ) [1,N]_{\mathcal{N}(G)} [ 1 , N ] N ( G ) . Define the socle length
σ ( G ) = λ G ( Soc ( G ) ) . \sigma(G)=\lambda_G(\operatorname{Soc}(G)). σ ( G ) = λ G ( Soc ( G )) .
An M \mathcal{M} M -chain is a chain all of whose terms belong to N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) , and the M \mathcal{M} M -depth of G G G is
md ( G ) = max { r : 1 = E 0 < E 1 < ⋯ < E r = G , E i ∈ N M ( G ) } . \operatorname{md}(G)=\max\{r:1=E_0<E_1<\cdots<E_r=G,\ E_i\in\mathcal{N}^{\mathcal{M}}(G)\}. md ( G ) = max { r : 1 = E 0 < E 1 < ⋯ < E r = G , E i ∈ N M ( G )} .
Abelian minimal normal subgroups are irreducible modules, and the combinatorics of the socle is that of projective spaces over endomorphism rings. We fix the corresponding module-theoretic language.
Definition 2.9. For a prime p p p , F p \mathbb{F}_p F p denotes the field with p p p elements and F p G \mathbb{F}_pG F p G denotes the group algebra. A left module is simple (or irreducible ) if it is non-zero and has no proper non-zero submodule. For a ring R R R and a left R R R -module A A A , End R ( A ) \operatorname{End}_R(A) End R ( A ) denotes its endomorphism ring. A module is semisimple if it is a direct sum of simple submodules. A semisimple module is homogeneous if all of its simple summands are isomorphic; an isotypic decomposition is a direct-sum decomposition into homogeneous components belonging to pairwise nonisomorphic simple-module types. For a division ring D D D , P n − 1 ( D ) \mathbb{P}^{n-1}(D) P n − 1 ( D ) denotes the set of one-dimensional right D D D -subspaces of D n D^n D n ; in particular P 1 ( D ) \mathbb{P}^1(D) P 1 ( D ) is the projective line.
For direct products the central p p p -torsion of the factors plays a special role, and we introduce notation to count the minimal normal subgroups lying in a given normal subgroup.
Definition 2.10. For G = G 1 × ⋯ × G t G=G_1\times\cdots\times G_t G = G 1 × ⋯ × G t , let G i ∗ G_i^* G i ∗ be the i i i -th coordinate subgroup, and for A ≤ G i A\le G_i A ≤ G i let A ^ ≤ G i ∗ \widehat{A}\le G_i^* A ≤ G i ∗ be its coordinate copy. For each prime p p p , put
V i , p = { z ∈ Z ( G i ) : z p = 1 } , V p = ∏ i = 1 t V i , p . V_{i,p}=\{z\in Z(G_i):z^p=1\}, \qquad V_p=\prod_{i=1}^{t}V_{i,p}. V i , p = { z ∈ Z ( G i ) : z p = 1 } , V p = i = 1 ∏ t V i , p .
Each V p V_p V p is viewed as an F p \mathbb{F}_p F p -vector space, and a line in V p V_p V p means a one-dimensional F p \mathbb{F}_p F p -subspace. Such a line is coordinate if it is contained in the coordinate copy of some V i , p V_{i,p} V i , p , and it is non-coordinate otherwise. If H ⊴ G 1 × ⋯ × G t H \trianglelefteq G_1 \times\cdots\times G_t H ⊴ G 1 × ⋯ × G t , define
a ( H ) = # { ( i , A ) : A ∈ MinNorm ( G i ) , A ≰ Z ( G i ) , A ^ ≤ H } a(H)=\#\{(i,A): A\in\operatorname{MinNorm}(G_i),\ A\not\leq Z(G_i),\ \widehat{A}\leq H\} a ( H ) = # {( i , A ) : A ∈ MinNorm ( G i ) , A ≤ Z ( G i ) , A ≤ H }
and
d p ( H ) = dim F p ( H ∩ V p ) . d_p(H)=\dim_{\mathbb{F}_p}(H\cap V_p). d p ( H ) = dim F p ( H ∩ V p ) .
For a direct power X k X^k X k , X i X_i X i denotes the i i i -th coordinate copy of X X X .
Crown-based powers were introduced by Dalla Volta and Lucchini [8 ] , building on the crowns of Gaschütz [12 ] ; for primitive groups see [11 ] .
Definition 2.11. For a subgroup M ≤ L M\leq L M ≤ L , its core is
core L ( M ) = ⋂ x ∈ L x M x − 1 . \operatorname{core}_L(M)=\bigcap_{x\in L}xMx^{-1}. core L ( M ) = x ∈ L ⋂ x M x − 1 . The subgroup M M M is core-free if core L ( M ) = 1 \operatorname{core}_L(M)=1 core L ( M ) = 1 . A finite group L L L is primitive if it has a core-free maximal subgroup. If L L L is finite, monolithic, and primitive with unique minimal normal subgroup A = Soc ( L ) A=\operatorname{Soc}(L) A = Soc ( L ) , then for k ≥ 1 k\geq1 k ≥ 1 its crown-based power of size k k k is
L k = { ( x 1 , … , x k ) ∈ L k : x 1 A = ⋯ = x k A } . L_k=\{(x_1,\ldots,x_k)\in L^k:x_1A=\cdots=x_kA\}. L k = {( x 1 , … , x k ) ∈ L k : x 1 A = ⋯ = x k A } .
Throughout statements concerning L k L_k L k , write V = A k V=A^k V = A k and let A 1 , … , A k A_1,\ldots,A_k A 1 , … , A k be the coordinate copies of A A A in V V V .
Finally, for the counting results on abelian, nilpotent, cyclic and dihedral groups we use the following notation.
Definition 2.12. For a finite abelian group G G G , additive notation may be used. Write
G = ⨁ p ∣ ∣ G ∣ G p , G p ≅ C p λ p , 1 ⊕ ⋯ ⊕ C p λ p , r p , G=\bigoplus_{p\mid|G|}G_p,\qquad G_p\cong C_p^{\lambda_{p,1}}\oplus\cdots\oplus C_p^{\lambda_{p,r_p}}, G = p ∣ ∣ G ∣ ⨁ G p , G p ≅ C p λ p , 1 ⊕ ⋯ ⊕ C p λ p , r p , where G p G_p G p is the Sylow p p p -subgroup and λ p , 1 ≥ ⋯ ≥ λ p , r p ≥ 1 \lambda_{p,1}\geq\cdots\geq\lambda_{p,r_p}\geq1 λ p , 1 ≥ ⋯ ≥ λ p , r p ≥ 1 . Put
G p [ p ] = { x ∈ G p : p x = 0 } , u p ( k ) = ∑ i = 1 r p min { k , λ p , i } ( k ≥ 0 ) , G_p[p]=\{x\in G_p:px=0\},\qquad u_p(k)=\sum_{i=1}^{r_p}\min\{k,\lambda_{p,i}\}\quad(k\geq0), G p [ p ] = { x ∈ G p : p x = 0 } , u p ( k ) = i = 1 ∑ r p min { k , λ p , i } ( k ≥ 0 ) , with u p ( 0 ) = 0 u_p(0)=0 u p ( 0 ) = 0 . Let c p ( G ) c_p(G) c p ( G ) be the number of non-trivial cyclic p p p -subgroups of G G G , and set
r ( G ) = ∑ p ∣ ∣ G ∣ r p . r(G)=\sum_{p\mid|G|}r_p. r ( G ) = p ∣ ∣ G ∣ ∑ r p .
Definition 2.13. For a positive integer n n n , Ω ( n ) \Omega(n) Ω ( n ) is the number of prime factors of n n n counted with multiplicity and ω ( n ) \omega(n) ω ( n ) is the number of distinct prime factors. If G G G is a finite p p p -group, set
Ω 1 ( Z ( G ) ) = { z ∈ Z ( G ) : z p = 1 } . \Omega_1(Z(G))=\{z\in Z(G):z^p=1\}. Ω 1 ( Z ( G )) = { z ∈ Z ( G ) : z p = 1 } . If G G G is a finite nilpotent group and P p P_p P p is its Sylow p p p -subgroup, set
ρ ( G ) = ∑ p ∣ ∣ G ∣ dim F p { z ∈ Z ( P p ) : z p = 1 } . \rho(G)=\sum_{p\mid|G|}\dim_{\mathbb{F}_p}\{z\in Z(P_p):z^p=1\}. ρ ( G ) = p ∣ ∣ G ∣ ∑ dim F p { z ∈ Z ( P p ) : z p = 1 } . For n ≥ 3 n\geq3 n ≥ 3 , D 2 n D_{2n} D 2 n denotes the dihedral group of order 2 n 2n 2 n with presentation
D 2 n = ⟨ r , s : r n = s 2 = 1 , s r s = r − 1 ⟩ . D_{2n}=\langle r,s:r^n=s^2=1,\ srs=r^{-1}\rangle. D 2 n = ⟨ r , s : r n = s 2 = 1 , sr s = r − 1 ⟩ .
The dichotomy and the finite classification
In this section we prove that M \mathcal{M} M -subgroups are exactly the essential and the relatively uniform normal subgroups, and deduce Theorem A. Everything rests on the modularity of N ( G ) \mathcal{N}(G) N ( G ) and on its consequence for lengths.
The following lemma is well known (cf. [9 Chapter 4, Example 4.6(5)] ) and will be used throughout. We give a short proof to make the paper self-contained.
Lemma 3.1. If A , B , C ⊴ G A,B,C \trianglelefteq G A , B , C ⊴ G and A ≤ C A \leq C A ≤ C , then
A ( B ∩ C ) = A B ∩ C . A(B \cap C)=AB \cap C. A ( B ∩ C ) = A B ∩ C . In particular, N ( G ) \mathcal{N}(G) N ( G ) is a modular lattice.
Proof. The inclusion A ( B ∩ C ) ≤ A B ∩ C A(B \cap C) \leq AB \cap C A ( B ∩ C ) ≤ A B ∩ C is immediate. If a b ∈ A B ∩ C ab \in AB \cap C ab ∈ A B ∩ C , with a ∈ A a \in A a ∈ A and b ∈ B b \in B b ∈ B , then a ∈ C a \in C a ∈ C implies b = a − 1 ( a b ) ∈ B ∩ C b=a^{-1}(ab) \in B \cap C b = a − 1 ( ab ) ∈ B ∩ C . Thus a b ∈ A ( B ∩ C ) ab \in A(B \cap C) ab ∈ A ( B ∩ C ) , proving the identity and hence modularity. □
For finite G G G , modularity has a numerical consequence that we use repeatedly: lengths in N ( G ) \mathcal{N}(G) N ( G ) are additive along chief series and satisfy the rank identity of modular lattices [2 ] . In particular, every maximal chain in an interval of N ( G ) \mathcal{N}(G) N ( G ) has the same length (the Jordan–Hölder theorem for chief series).
Lemma 3.2. Let G G G be a finite group and let H , K ⊴ G H,K \trianglelefteq G H , K ⊴ G . Then
ℓ N ( G ) = λ G ( H ) + ℓ N ( G / H ) , \ell_{\mathcal{N}}(G)=\lambda_G(H)+\ell_{\mathcal{N}}(G/H), ℓ N ( G ) = λ G ( H ) + ℓ N ( G / H ) ,
λ G ( H K ) + λ G ( H ∩ K ) = λ G ( H ) + λ G ( K ) . \lambda_G(HK)+\lambda_G(H \cap K)=\lambda_G(H)+\lambda_G(K). λ G ( H K ) + λ G ( H ∩ K ) = λ G ( H ) + λ G ( K ) .
Proof. A maximal chain in [ 1 , H ] N ( G ) [1,H]_{\mathcal{N}(G)} [ 1 , H ] N ( G ) concatenated with the inverse image of a chief series of G / H G/H G / H is a chief series of G G G . The Jordan–Hölder theorem for chief series therefore gives (1). For (2), the correspondence (diamond) isomorphism [ H ∩ K , K ] N ( G ) ≅ [ H , H K ] N ( G ) [H \cap K,K]_{\mathcal{N}(G)} \cong[H,HK]_{\mathcal{N}(G)} [ H ∩ K , K ] N ( G ) ≅ [ H , H K ] N ( G ) gives
λ G ( K ) − λ G ( H ∩ K ) = λ G ( H K ) − λ G ( H ) . \lambda_G(K)-\lambda_G(H \cap K)=\lambda_G(HK)-\lambda_G(H). λ G ( K ) − λ G ( H ∩ K ) = λ G ( H K ) − λ G ( H ) .
Rearranging proves the claim. □
We first observe that in Definition 2.5 it suffices to test pairs of normal subgroups; the ring-theoretic analogue is proved in [3 ] , where it is also observed that this contrasts with Banach spaces: the Alfsen–Effros characterisation of M \mathcal{M} M -ideals by intersections of balls can be restricted to three balls, but not to two [1 , 14 ] .
Proposition 3.3. A normal subgroup E ⊴ G E \trianglelefteq G E ⊴ G is an M \mathcal{M} M -subgroup if and only if it satisfies the defining implication for pairs N 1 , N 2 N_1,N_2 N 1 , N 2 .
Proof. Necessity follows by taking r = 2 r=2 r = 2 . Conversely, assume the implication holds for pairs, and proceed by induction on r ≥ 2 r \geq2 r ≥ 2 . Given N 1 , … , N r N_1,\ldots,N_r N 1 , … , N r as in Definition 2.5, put N = ⋂ i = 1 r − 1 N i N=\bigcap_{i=1}^{r-1}N_i N = ⋂ i = 1 r − 1 N i . Then N ∩ N r ≠ 1 N \cap N_r \ne1 N ∩ N r = 1 ; by induction E ∩ N ≠ 1 E \cap N \ne1 E ∩ N = 1 , and by assumption E ∩ N r ≠ 1 E \cap N_r \ne1 E ∩ N r = 1 . Applying the pairwise implication to N N N and N r N_r N r yields E ∩ ⋂ i = 1 r N i ≠ 1 E \cap\bigcap_{i=1}^{r}N_i \ne1 E ∩ ⋂ i = 1 r N i = 1 . □
With pairs at hand, the M \mathcal{M} M -condition splits into two classical conditions. For rings this is the main theorem of [3 ] , and for modular quantales it is the characterisation of μ \mu μ -elements in [6 ] ; as N ( G ) \mathcal{N}(G) N ( G ) is in general not an integral commutative quantale (see the discussion before Theorem 4.11 ), we give the short direct proof.
Theorem 3.4. A normal subgroup E ⊴ G E \trianglelefteq G E ⊴ G is an M \mathcal{M} M -subgroup if and only if it is essential in G G G or relatively uniform in G G G .
Proof. Suppose E E E is an M \mathcal{M} M -subgroup that is neither essential nor relatively uniform. Then there are non-trivial N 1 , N 2 , N 3 ⊴ G N_1,N_2,N_3 \trianglelefteq G N 1 , N 2 , N 3 ⊴ G such that E ∩ N 1 = 1 E \cap N_1=1 E ∩ N 1 = 1 , N 2 , N 3 ≤ E N_2,N_3 \leq E N 2 , N 3 ≤ E and N 2 ∩ N 3 = 1 N_2 \cap N_3=1 N 2 ∩ N 3 = 1 . Now, by the modular law, for i = 2 , 3 i=2,3 i = 2 , 3 , we have
E ∩ N i N 1 = N i ( E ∩ N 1 ) = N i . E \cap N_iN_1=N_i(E \cap N_1)=N_i. E ∩ N i N 1 = N i ( E ∩ N 1 ) = N i .
Therefore, for the pairs of normal subgroups N 2 N 1 N_{2}N_{1} N 2 N 1 and N 3 N 1 N_{3}N_{1} N 3 N 1 , E E E fails to be an M \mathcal{M} M -subgroup: indeed N 2 N 1 ∩ N 3 N 1 ≥ N 1 ≠ 1 N_{2}N_{1} \cap N_{3}N_{1} \geq N_{1} \neq1 N 2 N 1 ∩ N 3 N 1 ≥ N 1 = 1 and E ∩ N i N 1 = N i ≠ 1 E \cap N_{i}N_{1} = N_{i} \neq1 E ∩ N i N 1 = N i = 1 , whereas E ∩ N 2 N 1 ∩ N 3 N 1 = N 2 ∩ N 3 = 1 E \cap N_{2}N_{1} \cap N_{3}N_{1} = N_{2} \cap N_{3} = 1 E ∩ N 2 N 1 ∩ N 3 N 1 = N 2 ∩ N 3 = 1 .
Conversely, if E E E is essential, then for any family { N i } i ∈ Λ \{N_{i}\}_{i \in\Lambda} { N i } i ∈ Λ of normal subgroups with ⋂ i ∈ Λ N i ≠ 1 \bigcap_{i \in\Lambda} N_{i} \neq1 ⋂ i ∈ Λ N i = 1 , we have E ∩ N i ≠ 1 E \cap N_{i} \neq1 E ∩ N i = 1 and E ∩ ⋂ i ∈ Λ N i ≠ 1 E \cap\bigcap_{i \in\Lambda} N_{i} \neq1 E ∩ ⋂ i ∈ Λ N i = 1 by the definition of essentiality. Moreover, if E E E is relatively uniform and N 1 , N 2 N_{1}, N_{2} N 1 , N 2 are as in Proposition 3.3 , then E ∩ N 1 E \cap N_{1} E ∩ N 1 and E ∩ N 2 E \cap N_{2} E ∩ N 2 are non-trivial normal subgroups contained in E E E , so E ∩ N 1 ∩ N 2 ≠ 1 E \cap N_{1} \cap N_{2} \neq1 E ∩ N 1 ∩ N 2 = 1 . Therefore, in both cases E E E is an M \mathcal{M} M -subgroup. □
Theorem 3.4 confirms that each M \mathcal{M} M -subgroup is either essential or relatively uniform. The next example shows that neither class contains the other, that they can overlap, and that a normal subgroup may lie in neither.
Example 3.5. In the group C p n C_{p^{n}} C p n the subgroups form a chain, so any two non-trivial subgroups meet non-trivially. Hence every non-trivial subgroup of C p n C_{p^{n}} C p n is both essential and relatively uniform. Now let G = C 30 G = C_{30} G = C 30 , and for any divisor d d d of 30 let C d C_{d} C d be its cyclic subgroup of order d d d . Clearly, C d 1 ∩ C d 2 = C gcd ( d 1 , d 2 ) C_{d_{1}} \cap C_{d_{2}} = C_{\gcd(d_{1},d_{2})} C d 1 ∩ C d 2 = C g c d ( d 1 , d 2 ) .
The subgroup C 2 C_{2} C 2 is relatively uniform but not essential, since C 2 ∩ C 3 = 1 C_{2} \cap C_{3} = 1 C 2 ∩ C 3 = 1 .
The whole group G G G is essential, but not relatively uniform, since C 2 , C 3 ≤ G C_{2}, C_{3} \leq G C 2 , C 3 ≤ G and C 2 ∩ C 3 = 1 C_{2} \cap C_{3} = 1 C 2 ∩ C 3 = 1 .
The subgroup C 6 C_{6} C 6 is neither essential nor relatively uniform, since C 6 ∩ C 5 = 1 C_{6} \cap C_{5} = 1 C 6 ∩ C 5 = 1 , and C 2 , C 3 ≤ C 6 C_{2}, C_{3} \leq C_{6} C 2 , C 3 ≤ C 6 with C 2 ∩ C 3 = 1 C_{2} \cap C_{3} = 1 C 2 ∩ C 3 = 1 . So C 6 C_{6} C 6 is not an M \mathcal{M} M -subgroup.
In normally atomic groups relative uniformity and essentiality can be read off from the minimal normal subgroups. We need the following elementary fact.
Lemma 3.6. Distinct minimal normal subgroups A , B A, B A , B of a group satisfy A ∩ B = 1 A \cap B = 1 A ∩ B = 1 and [ A , B ] = 1 [A,B] = 1 [ A , B ] = 1 .
Proof. The normal subgroup A ∩ B A \cap B A ∩ B is contained in both A A A and B B B , and minimality forces A ∩ B = 1 A \cap B = 1 A ∩ B = 1 . Furthermore, the subgroup [ A , B ] [A,B] [ A , B ] is generated by [ a , b ] [a,b] [ a , b ] for a ∈ A a \in A a ∈ A , b ∈ B b \in B b ∈ B , and each [ a , b ] ∈ A ∩ B [a,b] \in A \cap B [ a , b ] ∈ A ∩ B . Therefore [ A , B ] ≤ A ∩ B = 1 [A,B] \leq A \cap B = 1 [ A , B ] ≤ A ∩ B = 1 . □
Theorem 3.7. Let G G G be normally atomic and let E ⊴ G E \trianglelefteq G E ⊴ G . Then E E E is an M \mathcal{M} M -subgroup if and only if one of the following holds:
E = 1 E = 1 E = 1 ;
E E E contains exactly one minimal normal subgroup of G G G ;
E E E contains every minimal normal subgroup of G G G .
Proof. Let E ≠ 1 E \neq1 E = 1 be an M \mathcal{M} M -subgroup. As G G G is normally atomic, E E E contains at least one minimal normal subgroup, so if E E E does not contain two distinct ones, then (2) holds. Suppose the M \mathcal{M} M -subgroup E ≠ 1 E \neq1 E = 1 contains two distinct minimal normal subgroups N N N and N ′ N' N ′ . By Lemma 3.6 , N ∩ N ′ = 1 N \cap N' = 1 N ∩ N ′ = 1 . Then E E E is not relatively uniform, and so Theorem 3.4 forces E E E to be essential. Then E ∩ M ≠ 1 E \cap M \neq1 E ∩ M = 1 for every M ∈ MinNorm ( G ) M \in\operatorname{MinNorm}(G) M ∈ MinNorm ( G ) and so E ∩ M = M E \cap M = M E ∩ M = M , that is, M M M is a subgroup of E E E . Hence E E E contains every minimal normal subgroup of G G G .
Conversely, if E = 1 E = 1 E = 1 , then vacuously E E E is an M \mathcal{M} M -subgroup. Now, let E E E contain exactly one minimal normal subgroup, say N N N . Consider two non-trivial normal subgroups N 1 , N 2 ⊴ G N_{1}, N_{2} \trianglelefteq G N 1 , N 2 ⊴ G such that N 1 , N 2 ≤ E N_{1}, N_{2} \leq E N 1 , N 2 ≤ E . As G G G is normally atomic, there exist M 1 , M 2 ∈ MinNorm ( G ) M_{1}, M_{2} \in\operatorname{MinNorm}(G) M 1 , M 2 ∈ MinNorm ( G ) such that M 1 ≤ N 1 M_{1} \leq N_{1} M 1 ≤ N 1 and M 2 ≤ N 2 M_{2} \leq N_{2} M 2 ≤ N 2 . Thus, M 1 , M 2 ≤ E M_{1}, M_{2} \leq E M 1 , M 2 ≤ E . But E E E contains a unique minimal normal subgroup N N N . Therefore, M 1 = N = M 2 M_{1} = N = M_{2} M 1 = N = M 2 and hence N ≤ N 1 N \leq N_{1} N ≤ N 1 and N ≤ N 2 N \leq N_{2} N ≤ N 2 . Thus, N ≤ N 1 ∩ N 2 N \leq N_{1} \cap N_{2} N ≤ N 1 ∩ N 2 and hence N 1 ∩ N 2 ≠ 1 N_{1} \cap N_{2} \neq1 N 1 ∩ N 2 = 1 . Thus E E E is relatively uniform in G G G and so by Theorem 3.4 , E E E is an M \mathcal{M} M -subgroup. Now, suppose E E E contains every minimal normal subgroup of G G G . Now, consider a normal subgroup H ⊴ G H \trianglelefteq G H ⊴ G , where H ≠ 1 H \neq1 H = 1 . Now, as G G G is normally atomic, there exists K ∈ MinNorm ( G ) K \in\operatorname{MinNorm}(G) K ∈ MinNorm ( G ) such that K ≤ H K \leq H K ≤ H . But by hypothesis, K ≤ E K \leq E K ≤ E . So, K ≤ E ∩ H K \leq E \cap H K ≤ E ∩ H and hence E ∩ H ≠ 1 E \cap H \neq1 E ∩ H = 1 . Thus E E E is essential in G G G and so by Theorem 3.4 E E E is an M \mathcal{M} M -subgroup. □
Corollary 3.8. Let G G G be finite and let E ⊴ G E \trianglelefteq G E ⊴ G . Then E E E is an M \mathcal{M} M -subgroup if and only if
E = 1 , ∣ supp ( E ) ∣ = 1 , or Soc ( G ) ≤ E . E = 1,\qquad|\operatorname{supp}(E)| = 1,\qquad\text{or}\qquad\operatorname{Soc}(G) \leq E. E = 1 , ∣ supp ( E ) ∣ = 1 , or Soc ( G ) ≤ E .
Proof. As G G G is finite, it is normally atomic. Now the result follows immediately from Theorem 3.7 , as ∣ supp ( E ) ∣ = 1 |\operatorname{supp}(E)|=1 ∣ supp ( E ) ∣ = 1 means that E E E contains a unique minimal normal subgroup and Soc ( G ) ≤ E \operatorname{Soc}(G)\leq E Soc ( G ) ≤ E means that E E E contains every minimal normal subgroup of G G G , as Soc ( G ) \operatorname{Soc}(G) Soc ( G ) is the smallest subgroup containing all minimal normal subgroups of G G G . □
For a finite group G G G , Corollary 3.8 gives us a tool to check whether a subgroup is an M \mathcal{M} M -subgroup. In the next example, we use it to show that the centre and the derived subgroup of a group need not be M \mathcal{M} M -subgroups.
Example 3.9. (1) Let G = K 4 × S 3 G=K_{4}\times S_{3} G = K 4 × S 3 , where K 4 = { 1 , a , b , a b } ≅ C 2 × C 2 K_{4}=\{1,a,b,ab\}\cong C_{2}\times C_{2} K 4 = { 1 , a , b , ab } ≅ C 2 × C 2 . Then Z ( G ) = K 4 × 1 ≠ 1 Z(G)=K_{4}\times1\neq1 Z ( G ) = K 4 × 1 = 1 , ∣ supp ( Z ( G ) ) ∣ = 3 |\operatorname{supp}(Z(G))|=3 ∣ supp ( Z ( G )) ∣ = 3 (the three subgroups of order 2 2 2 of Z ( G ) Z(G) Z ( G ) are central of prime order, hence minimal normal), and Z ( G ) Z(G) Z ( G ) does not contain the minimal normal subgroup 1 × A 3 1\times A_{3} 1 × A 3 . Therefore, by Corollary 3.8, Z ( G ) Z(G) Z ( G ) is not an M \mathcal{M} M -subgroup.
(2) Take G = D 60 G=D_{60} G = D 60 (cf. Example 2.6 ), then for any 1 ≤ i ≤ 30 1\leq i\leq30 1 ≤ i ≤ 30 ,
[ r i , s ] = r i s r − i s = r 2 i , [r^{i},s]=r^{i}sr^{-i}s=r^{2i}, [ r i , s ] = r i s r − i s = r 2 i ,
and hence [ D 60 , D 60 ] ≥ ⟨ r 2 ⟩ [D_{60},D_{60}]\geq\langle r^{2}\rangle [ D 60 , D 60 ] ≥ ⟨ r 2 ⟩ ; as D 60 / ⟨ r 2 ⟩ D_{60}/\langle r^{2}\rangle D 60 / ⟨ r 2 ⟩ has order 4 4 4 and is therefore abelian, [ D 60 , D 60 ] = ⟨ r 2 ⟩ [D_{60},D_{60}]=\langle r^{2}\rangle [ D 60 , D 60 ] = ⟨ r 2 ⟩ . But ⟨ r 2 ⟩ \langle r^{2}\rangle ⟨ r 2 ⟩ is not an M \mathcal{M} M -subgroup, since it contains the two minimal normal subgroups ⟨ r 10 ⟩ \langle r^{10}\rangle ⟨ r 10 ⟩ and ⟨ r 6 ⟩ \langle r^{6}\rangle ⟨ r 6 ⟩ , while the minimal normal subgroup ⟨ r 15 ⟩ \langle r^{15}\rangle ⟨ r 15 ⟩ is not inside ⟨ r 2 ⟩ \langle r^{2}\rangle ⟨ r 2 ⟩ . Hence by Corollary 3.8 the derived subgroup of D 60 D_{60} D 60 is not an M \mathcal{M} M -subgroup.
We now turn to the closure system formed by M \mathcal{M} M -subgroups. Finite intersections are always harmless.
Proposition 3.10. The intersection of finitely many M \mathcal{M} M -subgroups of an arbitrary group is an M \mathcal{M} M -subgroup.
Proof. Let E 1 E_{1} E 1 , E 2 E_{2} E 2 be M \mathcal{M} M -subgroups and for r ≥ 2 r\geq2 r ≥ 2 , let N 1 , … , N r N_{1},\ldots,N_{r} N 1 , … , N r be non-trivial normal subgroups such that ⋂ i N i ≠ 1 \bigcap_{i}N_{i}\neq1 ⋂ i N i = 1 and E 1 ∩ E 2 ∩ N i ≠ 1 E_{1}\cap E_{2}\cap N_{i}\neq1 E 1 ∩ E 2 ∩ N i = 1 for all i i i . Since E 1 ∩ N i ≠ 1 E_{1}\cap N_{i}\neq1 E 1 ∩ N i = 1 and E 1 E_{1} E 1 is an M \mathcal{M} M -subgroup, then E 1 ∩ ⋂ i N i ≠ 1 E_{1}\cap\bigcap_{i}N_{i}\neq1 E 1 ∩ ⋂ i N i = 1 . Now each E 1 ∩ N i E_{1}\cap N_{i} E 1 ∩ N i meets E 2 E_{2} E 2 non-trivially, and the non-trivial normal subgroups E 1 ∩ N 1 , … , E 1 ∩ N r E_{1}\cap N_{1},\ldots,E_{1}\cap N_{r} E 1 ∩ N 1 , … , E 1 ∩ N r have non-trivial intersection E 1 ∩ ⋂ i N i E_{1}\cap\bigcap_{i}N_{i} E 1 ∩ ⋂ i N i ; then since E 2 E_{2} E 2 is an M \mathcal{M} M -subgroup, E 1 ∩ E 2 ∩ ⋂ i N i ≠ 1 E_{1}\cap E_{2}\cap\bigcap_{i}N_{i}\neq1 E 1 ∩ E 2 ∩ ⋂ i N i = 1 . Therefore E 1 ∩ E 2 E_{1}\cap E_{2} E 1 ∩ E 2 is an M \mathcal{M} M -subgroup, and the case of finitely many M \mathcal{M} M -subgroups follows by induction. □
Proposition 3.11. The map cl M : N ( G ) → N ( G ) \operatorname{cl}_{\mathcal{M}}:\mathcal{N}(G)\to\mathcal{N}(G) cl M : N ( G ) → N ( G ) is extensive, order preserving, and idempotent, and every M \mathcal{M} M -subgroup is a fixed point of cl M \operatorname{cl}_{\mathcal{M}} cl M . If G G G is finite, then the fixed points are exactly the M \mathcal{M} M -subgroups, and explicitly
cl M ( N ) = { N , N = 1 or ∣ supp ( N ) ∣ = 1 , N Soc ( G ) , ∣ supp ( N ) ∣ ≥ 2. \operatorname{cl}_{\mathcal{M}}(N)=
\begin{cases}
N, & N=1\text{ or }|\operatorname{supp}(N)|=1,\\
N\operatorname{Soc}(G), & |\operatorname{supp}(N)|\geq2.
\end{cases} cl M ( N ) = { N , N Soc ( G ) , N = 1 or ∣ supp ( N ) ∣ = 1 , ∣ supp ( N ) ∣ ≥ 2. Proof. Let N ⊴ G N\trianglelefteq G N ⊴ G . Now, if E ∈ N M ( G ) E\in\mathcal{N}^{\mathcal{M}}(G) E ∈ N M ( G ) such that N ≤ E N\leq E N ≤ E , then
N ≤ ⋂ { E ∈ N M ( G ) : N ≤ E } = cl M ( N ) N\leq\bigcap\{E\in\mathcal{N}^{\mathcal{M}}(G):N\leq E\}=\operatorname{cl}_{\mathcal{M}}(N) N ≤ ⋂ { E ∈ N M ( G ) : N ≤ E } = cl M ( N ) Hence, cl M \operatorname{cl}_{\mathcal{M}} cl M is extensive. Now suppose N , M ∈ N ( G ) N,M\in\mathcal{N}(G) N , M ∈ N ( G ) such that N ≤ M N\leq M N ≤ M . If E ∈ N M ( G ) E\in\mathcal{N}^{\mathcal{M}}(G) E ∈ N M ( G ) such that M ≤ E M\leq E M ≤ E , then N ≤ E N\leq E N ≤ E . Thus, { E ∈ N M ( G ) : M ≤ E } ⊆ { E ∈ N M ( G ) : N ≤ E } \{E\in\mathcal{N}^{\mathcal{M}}(G):M\leq E\}\subseteq\{E\in\mathcal{N}^{\mathcal{M}}(G):N\leq E\} { E ∈ N M ( G ) : M ≤ E } ⊆ { E ∈ N M ( G ) : N ≤ E } . As intersection reverses the order, we get
⋂ { E ∈ N M ( G ) : N ≤ E } ≤ ⋂ { E ∈ N M ( G ) : M ≤ E } \bigcap\{E\in\mathcal{N}^{\mathcal{M}}(G):N\leq E\}\leq\bigcap\{E\in\mathcal{N}^{\mathcal{M}}(G):M\leq E\} ⋂ { E ∈ N M ( G ) : N ≤ E } ≤ ⋂ { E ∈ N M ( G ) : M ≤ E } Thus, cl M ( N ) ≤ cl M ( M ) \operatorname{cl}_{\mathcal{M}}(N)\leq\operatorname{cl}_{\mathcal{M}}(M) cl M ( N ) ≤ cl M ( M ) . Hence, cl M \operatorname{cl}_{\mathcal{M}} cl M is order preserving.
Now let N ∈ N ( G ) N\in\mathcal{N}(G) N ∈ N ( G ) . Define, A ( N ) = { E ∈ N M ( G ) : N ≤ E } \mathcal{A}(N)=\{E\in\mathcal{N}^{\mathcal{M}}(G):N\leq E\} A ( N ) = { E ∈ N M ( G ) : N ≤ E } and C = cl M ( N ) \mathcal{C}=\operatorname{cl}_{\mathcal{M}}(N) C = cl M ( N ) . If H ∈ A ( N ) H\in\mathcal{A}(N) H ∈ A ( N ) then C ≤ H \mathcal{C}\leq H C ≤ H . Thus, H ∈ A ( C ) H\in\mathcal{A}(\mathcal{C}) H ∈ A ( C ) . Conversely if H ∈ A ( C ) H\in\mathcal{A}(\mathcal{C}) H ∈ A ( C ) , then N ≤ C ≤ H N\leq\mathcal{C}\leq H N ≤ C ≤ H , where the first inequality comes from extensiveness of cl M \operatorname{cl}_{\mathcal{M}} cl M . Thus, H ∈ A ( N ) H\in\mathcal{A}(N) H ∈ A ( N ) . Thus, A ( N ) = A ( C ) \mathcal{A}(N)=\mathcal{A}(\mathcal{C}) A ( N ) = A ( C ) . Thus,
cl M ( N ) = ⋂ A ( N ) = ⋂ A ( C ) = cl M ( cl M ( N ) ) = cl M 2 ( N ) \operatorname{cl}_{\mathcal{M}}(N)=\bigcap\mathcal{A}(N)=\bigcap\mathcal{A}(\mathcal{C})=\operatorname{cl}_{\mathcal{M}}(\operatorname{cl}_{\mathcal{M}}(N))=\operatorname{cl}_{\mathcal{M}}^{2}(N) cl M ( N ) = ⋂ A ( N ) = ⋂ A ( C ) = cl M ( cl M ( N )) = cl M 2 ( N ) Thus, cl M \operatorname{cl}_{\mathcal{M}} cl M is idempotent.
Now suppose, N N N is an M \mathcal{M} M -subgroup. Then N ∈ { E ∈ N M ( G ) : N ≤ E } N\in\{E\in\mathcal{N}^{\mathcal{M}}(G):N\leq E\} N ∈ { E ∈ N M ( G ) : N ≤ E } . Thus, cl M ( N ) ≤ N \operatorname{cl}_{\mathcal{M}}(N)\leq N cl M ( N ) ≤ N . But by extensiveness we get N ≤ cl M ( N ) N\leq\operatorname{cl}_{\mathcal{M}}(N) N ≤ cl M ( N ) . Thus, N = cl M ( N ) N=\operatorname{cl}_{\mathcal{M}}(N) N = cl M ( N ) .
Now let G G G be finite. Then cl M ( N ) \operatorname{cl}_{\mathcal{M}}(N) cl M ( N ) is the intersection of finitely many M \mathcal{M} M -subgroups, hence an M \mathcal{M} M -subgroup by Proposition 3.10. In particular, if cl M ( N ) = N \operatorname{cl}_{\mathcal{M}}(N)=N cl M ( N ) = N , then N N N is an M \mathcal{M} M -subgroup, so the fixed points are exactly the M \mathcal{M} M -subgroups. If N = 1 N=1 N = 1 or ∣ supp ( N ) ∣ = 1 |\operatorname{supp}(N)|=1 ∣ supp ( N ) ∣ = 1 , then N N N is an M \mathcal{M} M -subgroup by Corollary 3.8, and so cl M ( N ) = N \operatorname{cl}_{\mathcal{M}}(N)=N cl M ( N ) = N . Lastly, if ∣ supp ( N ) ∣ ≥ 2 |\operatorname{supp}(N)|\ge2 ∣ supp ( N ) ∣ ≥ 2 , every M \mathcal{M} M -subgroup E ≥ N E\ge N E ≥ N has ∣ supp ( E ) ∣ ≥ 2 |\operatorname{supp}(E)|\ge2 ∣ supp ( E ) ∣ ≥ 2 , so by Corollary 3.8 such an E E E must contain Soc ( G ) \operatorname{Soc}(G) Soc ( G ) and hence N Soc ( G ) ≤ E N\operatorname{Soc}(G)\le E N Soc ( G ) ≤ E . Clearly, N Soc ( G ) N\operatorname{Soc}(G) N Soc ( G ) is itself an M \mathcal{M} M -subgroup and the smallest M \mathcal{M} M -subgroup containing both N N N and Soc ( G ) \operatorname{Soc}(G) Soc ( G ) . □
Finiteness of the group is assumed in Corollary 3.8 and in the second part of Proposition 3.11, and Proposition 3.10 is about finitely many M \mathcal{M} M -subgroups. The next example shows that these assumptions cannot be dropped, even for finitely generated abelian groups.
Example 3.12. Let G = Z 3 G=\mathbb{Z}^{3} G = Z 3 . Consider the subgroups
P = Z × Z × 0 and H m = Z × Z × m Z ( m ≥ 1 ) . P=\mathbb{Z}\times\mathbb{Z}\times0 \qquad\text{and}\qquad H_m=\mathbb{Z}\times\mathbb{Z}\times m\mathbb{Z}\quad(m\ge1). P = Z × Z × 0 and H m = Z × Z × m Z ( m ≥ 1 ) . Each H m H_m H m is an essential subgroup. Indeed, let N N N be a non-zero subgroup and 0 ≠ x ∈ N 0\ne x\in N 0 = x ∈ N . Then m x ∈ N ∩ H m mx\in N\cap H_m m x ∈ N ∩ H m . So H m H_m H m is essential, and hence an M \mathcal{M} M -subgroup. Moreover, P P P is not an M \mathcal{M} M -subgroup. To this end, let N 1 = Z × 0 × Z N_1=\mathbb{Z}\times0\times\mathbb{Z} N 1 = Z × 0 × Z and N 2 = 0 × Z × Z N_2=0\times\mathbb{Z}\times\mathbb{Z} N 2 = 0 × Z × Z . Then
N 1 ∩ N 2 = 0 × 0 × Z ≠ 0 , P ∩ N 1 = Z × 0 × 0 ≠ 0 , P ∩ N 2 = 0 × Z × 0 ≠ 0 , N_1\cap N_2=0\times0\times\mathbb{Z}\ne0,\qquad P\cap N_1=\mathbb{Z}\times0\times0\ne0,\qquad P\cap N_2=0\times\mathbb{Z}\times0\ne0, N 1 ∩ N 2 = 0 × 0 × Z = 0 , P ∩ N 1 = Z × 0 × 0 = 0 , P ∩ N 2 = 0 × Z × 0 = 0 , but P ∩ N 1 ∩ N 2 = 0 P\cap N_1\cap N_2=0 P ∩ N 1 ∩ N 2 = 0 . We observe the following:
Clearly, ⋂ m ≥ 1 H m = P \bigcap_{m\ge1}H_m=P ⋂ m ≥ 1 H m = P . Therefore, an intersection of infinitely many M \mathcal{M} M -subgroups need not be an M \mathcal{M} M -subgroup. Therefore, Proposition 3.10 holds only for finitely many.
The only subgroups containing P P P are P P P and the H m H_m H m with m ≥ 1 m\ge1 m ≥ 1 . Indeed, if H H H is any subgroup containing P P P , then ( a , b , c ) ∈ H (a,b,c)\in H ( a , b , c ) ∈ H if and only if ( 0 , 0 , c ) ∈ H (0,0,c)\in H ( 0 , 0 , c ) ∈ H . Therefore { c ∣ ( 0 , 0 , c ) ∈ H } \{c\mid(0,0,c)\in H\} { c ∣ ( 0 , 0 , c ) ∈ H } is a subgroup of Z \mathbb{Z} Z and so it is of the form m Z m\mathbb{Z} m Z for some m ≥ 0 m\ge0 m ≥ 0 . Hence H = H m H=H_m H = H m for m > 0 m>0 m > 0 and H = P H=P H = P for m = 0 m=0 m = 0 . Now, each H m H_m H m with m ≥ 1 m\ge1 m ≥ 1 is an M \mathcal{M} M -subgroup containing P P P , and P P P is not, so by (1)
P ≤ cl M ( P ) ≤ ⋂ m ≥ 1 H m = P , P\le\operatorname{cl}_{\mathcal{M}}(P)\le\bigcap_{m\ge1}H_m=P, P ≤ cl M ( P ) ≤ m ≥ 1 ⋂ H m = P , that is, cl M ( P ) = P \operatorname{cl}_{\mathcal{M}}(P)=P cl M ( P ) = P . Thus P P P is a fixed point of the closure operator cl M \operatorname{cl}_{\mathcal{M}} cl M , but it is not an M \mathcal{M} M -subgroup. Therefore, the assertion of Proposition 3.11 that the fixed points are exactly the M \mathcal{M} M -subgroups needs G G G to be finite.
Let N N N be a non-trivial subgroup of G G G . If x ∈ N ∖ { 0 } x\in N\setminus\{0\} x ∈ N ∖ { 0 } , then ⟨ 2 x ⟩ \langle2x\rangle ⟨ 2 x ⟩ is a non-zero subgroup of N N N not containing x x x . Hence G G G has no minimal subgroups, so Soc ( G ) = 0 ≤ P \operatorname{Soc}(G)=0\le P Soc ( G ) = 0 ≤ P , so the condition Soc ( G ) ≤ E \operatorname{Soc}(G)\le E Soc ( G ) ≤ E of Corollary 3.8 holds for E = P E=P E = P , although P P P is not an M \mathcal{M} M -subgroup.
For finite groups we can now assemble Theorem A.
Corollary 3.13. For every finite group G G G , the ordered set N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) is a lattice. Its meet is ordinary intersection, and its join is
E ∨ M F = cl M ( E F ) . E\vee_{\mathcal{M}}F=\operatorname{cl}_{\mathcal{M}}(EF). E ∨ M F = cl M ( E F ) . Thus the M \mathcal{M} M -join agrees with the product whenever E F EF E F is an M \mathcal{M} M -subgroup, and otherwise it is E F Soc ( G ) EF\operatorname{Soc}(G) E F Soc ( G ) .
Proof. This follows as a consequence of Proposition 3.10 and Proposition 3.11: the meet of E , F ∈ N M ( G ) E,F\in\mathcal{N}^{\mathcal{M}}(G) E , F ∈ N M ( G ) in N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) is E ∩ F E\cap F E ∩ F , and cl M ( E F ) \operatorname{cl}_{\mathcal{M}}(EF) cl M ( E F ) is the least M \mathcal{M} M -subgroup containing E E E and F F F . If E F EF E F is not an M \mathcal{M} M -subgroup, then E F ≠ 1 EF\ne1 E F = 1 and ∣ supp ( E F ) ∣ ≠ 1 |\operatorname{supp}(EF)|\ne1 ∣ supp ( E F ) ∣ = 1 , so ∣ supp ( E F ) ∣ ≥ 2 |\operatorname{supp}(EF)|\ge2 ∣ supp ( E F ) ∣ ≥ 2 and cl M ( E F ) = E F Soc ( G ) \operatorname{cl}_{\mathcal{M}}(EF)=EF\operatorname{Soc}(G) cl M ( E F ) = E F Soc ( G ) .
The M \mathcal{M} M -poset and the no-loss problem
Passing from N ( G ) \mathcal{N}(G) N ( G ) to N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) discards the normal subgroups that contain at least two, but not all, minimal normal subgroups. In this section we show that N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) remembers the minimal normal subgroups, the socle and the filter above it, and we determine exactly when nothing is lost (Theorem B). In the rest of this section G G G is finite.
By Theorem A, the M \mathcal{M} M -poset of a finite group splits into 1 1 1 , the branches lying over the individual minimal normal subgroups, and the filter above the socle.
Proposition 4.1. Suppose that G G G is finite and has at least two minimal normal subgroups. Then
N M ( G ) = { 1 } ⊔ ∐ A ∈ MinNorm ( G ) R A ⊔ E ( G ) . \mathcal{N}^{\mathcal{M}}(G)=\{1\}\mathbin{\sqcup}\coprod_{A\in\operatorname{MinNorm}(G)}\mathcal{R}_{A}\mathbin{\sqcup}\mathcal{E}(G). N M ( G ) = { 1 } ⊔ A ∈ MinNorm ( G ) ∐ R A ⊔ E ( G ) . Each { 1 } ∪ R A \{1\}\cup\mathcal{R}_{A} { 1 } ∪ R A is closed under intersections, and E ( G ) \mathcal{E}(G) E ( G ) is a sublattice of N ( G ) \mathcal{N}(G) N ( G ) . If E ∈ R A E\in\mathcal{R}_{A} E ∈ R A and F ∈ R B F\in\mathcal{R}_{B} F ∈ R B with A ≠ B A\ne B A = B , then E F EF E F is an M \mathcal{M} M -subgroup if and only if Soc ( G ) ≤ E F \operatorname{Soc}(G)\le EF Soc ( G ) ≤ E F .
Proof. By Corollary 3.8 the M \mathcal{M} M -subgroups are 1 1 1 , the members of the R A \mathcal{R}_{A} R A , and the members of E ( G ) \mathcal{E}(G) E ( G ) , and they are disjoint. Clearly, R A ∩ R B = ∅ \mathcal{R}_{A}\cap\mathcal{R}_{B}=\varnothing R A ∩ R B = ∅ for A ≠ B A\ne B A = B , and each member of E ( G ) \mathcal{E}(G) E ( G ) contains S S S , hence at least two minimal normal subgroups. Since supp ( E ∩ F ) = supp ( E ) ∩ supp ( F ) \operatorname{supp}(E\cap F)=\operatorname{supp}(E)\cap\operatorname{supp}(F) supp ( E ∩ F ) = supp ( E ) ∩ supp ( F ) , the intersection of two members of R A \mathcal{R}_{A} R A lies in R A \mathcal{R}_{A} R A . Moreover, E ( G ) \mathcal{E}(G) E ( G ) is an interval of N ( G ) \mathcal{N}(G) N ( G ) , hence a sublattice. Finally, if E ∈ R A E\in\mathcal{R}_{A} E ∈ R A and F ∈ R B F\in\mathcal{R}_{B} F ∈ R B with A ≠ B A\ne B A = B , then A , B ≤ E F A,B\le EF A , B ≤ E F , so ∣ supp ( E F ) ∣ ≥ 2 \lvert\operatorname{supp}(EF)\rvert\ge2 ∣ supp ( E F )∣ ≥ 2 , and by Corollary 3.8, E F EF E F is an M \mathcal{M} M -subgroup if and only if Soc ( G ) ≤ E F \operatorname{Soc}(G)\le EF Soc ( G ) ≤ E F . □ \square □
The decomposition is visible inside the abstract poset N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) : the minimal normal subgroups, the socle and the branches can all be recovered from it. The last part of the next theorem answers the no-loss problem in terms of products of minimal normal subgroups.
Theorem 4.2. Let G G G be finite, put P = N M ( G ) P=\mathcal{N}^{\mathcal{M}}(G) P = N M ( G ) , ordered by inclusion, and put S = Soc ( G ) S=\operatorname{Soc}(G) S = Soc ( G ) .
(1) The atoms of P P P are exactly the minimal normal subgroups of G G G .
(2) If G G G has at least two atoms, then S S S is the least element of P P P lying above every atom of P P P .
(3) Under the same hypothesis, the interval [ S , G ] N ( G ) [S,G]_{\mathcal{N}(G)} [ S , G ] N ( G ) is the principal filter [ S , G ] P [S,G]_{P} [ S , G ] P , and the branch R A \mathcal{R}_{A} R A is recovered from P P P as
{ E ∈ P : A ≤ E and S ≰ E } . \{E\in P:A\le E\text{ and }S\nleq E\}. { E ∈ P : A ≤ E and S ≰ E } . (4) The following conditions are equivalent:
(a) P = N ( G ) P=\mathcal{N}(G) P = N ( G ) ;
(b) P P P is closed under products of its elements;
(c) A B = S AB=S A B = S for every two distinct A , B ∈ MinNorm ( G ) A,B\in\operatorname{MinNorm}(G) A , B ∈ MinNorm ( G ) .
Proof. (1) Each A ∈ MinNorm ( G ) A\in\operatorname{MinNorm}(G) A ∈ MinNorm ( G ) lies in P P P by Corollary 3.8, and no normal subgroup lies strictly between 1 1 1 and A A A . Therefore, A A A is an atom of P P P . Since G G G is normally atomic, an atom B B B of P P P contains some A ∈ MinNorm ( G ) ⊆ P A\in\operatorname{MinNorm}(G)\subseteq P A ∈ MinNorm ( G ) ⊆ P , forcing B = A B=A B = A .
(2) S ∈ P S\in P S ∈ P by Corollary 3.8, and by (1) an element of P P P lies above every atom if and only if it contains S S S .
(3) By Corollary 3.8 every normal subgroup containing S S S lies in P P P . Therefore, [ S , G ] N ( G ) = [ S , G ] P [S,G]_{\mathcal{N}(G)}=[S,G]_{P} [ S , G ] N ( G ) = [ S , G ] P . If E ∈ R A E\in\mathcal{R}_{A} E ∈ R A , then A ≤ E A\le E A ≤ E , and S ≰ E S\nleq E S ≰ E because E E E misses the other atoms. Conversely, if E ∈ P E\in P E ∈ P , A ≤ E A\le E A ≤ E and S ≰ E S\nleq E S ≰ E , then ∣ supp ( E ) ∣ = 1 \lvert\operatorname{supp}(E)\rvert=1 ∣ supp ( E )∣ = 1 by Corollary 3.8, and so supp ( E ) = { A } \operatorname{supp}(E)=\{A\} supp ( E ) = { A } .
(4) (a)⇒ \Rightarrow ⇒ (b) is clear.
(b)⇒ \Rightarrow ⇒ (c): for distinct A , B ∈ MinNorm ( G ) ⊆ P A,B\in\operatorname{MinNorm}(G)\subseteq P A , B ∈ MinNorm ( G ) ⊆ P we get A B ∈ P AB\in P A B ∈ P and ∣ supp ( A B ) ∣ ≥ 2 \lvert\operatorname{supp}(AB)\rvert\ge2 ∣ supp ( A B )∣ ≥ 2 , and hence S ≤ A B ≤ S S\le AB\le S S ≤ A B ≤ S by Corollary 3.8.
(c)⇒ \Rightarrow ⇒ (a): a normal subgroup containing two distinct minimal normal subgroups A , B A,B A , B contains A B = S AB=S A B = S , so every normal subgroup lies in P P P by Corollary 3.8. □ \square □
:::
Condition (c) of Theorem 4.2 (4) involves products of minimal normal subgroups, and to make it effective we need to know how many minimal normal subgroups the socle has. Abelian minimal normal subgroups are simple modules, and the simple submodules of a homogeneous semisimple module form a projective space. We record this classical fact with a proof, since the explicit parametrisation is used again in § 6 .
Lemma 4.3. Let R R R be a finite ring, let A A A be a simple left R R R -module, and put D = End R ( A ) D = \operatorname{End}_{R}(A) D = End R ( A ) . Then D D D is a finite division ring. For n ≥ 1 n \ge1 n ≥ 1 , the simple submodules of A n A^n A n are naturally parametrized by the one-dimensional right D D D -subspaces of D n D^n D n . Consequently, their number is
∣ D ∣ n − 1 ∣ D ∣ − 1 . \frac{|D|^n-1}{|D|-1}. ∣ D ∣ − 1 ∣ D ∣ n − 1 . For n = 2 n = 2 n = 2 , any two distinct simple submodules have direct sum A 2 A^2 A 2 .
Proof. By Schur’s Lemma, D D D is a division ring and every non-zero R R R -module homomorphism between simple R R R -modules is an isomorphism. If 0 ≠ a ∈ A 0 \ne a \in A 0 = a ∈ A , then R a Ra R a is a non-zero submodule of A A A , so A = R a A = Ra A = R a . As R R R is finite, A A A is finite, and hence D D D is finite. Here the product in D D D is the composition, and D n D^n D n is a right D D D -vector space via ( f 1 , … , f n ) g = ( f 1 g , … , f n g ) (f_1,\ldots,f_n)g = (f_1g,\ldots,f_ng) ( f 1 , … , f n ) g = ( f 1 g , … , f n g ) .
Let π i : A n → A \pi_i : A^n \to A π i : A n → A be the i i i -th projection. For f = ( f 1 , … , f n ) ∈ D n f = (f_1,\ldots,f_n) \in D^n f = ( f 1 , … , f n ) ∈ D n define Φ f ∈ Hom R ( A , A n ) \Phi_f \in\operatorname{Hom}_{R}(A,A^n) Φ f ∈ Hom R ( A , A n ) by Φ f ( a ) = ( f 1 ( a ) , … , f n ( a ) ) \Phi_f(a) = (f_1(a),\ldots,f_n(a)) Φ f ( a ) = ( f 1 ( a ) , … , f n ( a )) . As π i ∘ Φ f = f i \pi_i \circ\Phi_f = f_i π i ∘ Φ f = f i , the map f ↦ Φ f f \mapsto\Phi_f f ↦ Φ f is a bijection from D n D^n D n onto Hom R ( A , A n ) \operatorname{Hom}_{R}(A,A^n) Hom R ( A , A n ) , with inverse φ ↦ ( π 1 ∘ φ , … , π n ∘ φ ) \varphi\mapsto(\pi_1 \circ\varphi,\ldots,\pi_n \circ\varphi) φ ↦ ( π 1 ∘ φ , … , π n ∘ φ ) . Moreover, for g ∈ D g \in D g ∈ D we have
Φ f g = Φ f ∘ g . \Phi_{fg} = \Phi_f \circ g. Φ f g = Φ f ∘ g . Now let f ≠ 0 f \ne0 f = 0 . Then Φ f ≠ 0 \Phi_f \ne0 Φ f = 0 , so ker Φ f = { 0 } \ker\Phi_f = \{0\} ker Φ f = { 0 } as A A A is simple. Hence im Φ f ≅ A \operatorname{im}\Phi_f \cong A im Φ f ≅ A is a simple submodule of A n A^n A n . Conversely, let S S S be a simple submodule of A n A^n A n . As S ≠ { 0 } S \ne\{0\} S = { 0 } , π i ( S ) ≠ { 0 } \pi_i(S) \ne\{0\} π i ( S ) = { 0 } for some i i i , and by Schur’s Lemma π i ∣ S : S → A \pi_i|_S : S \to A π i ∣ S : S → A is an isomorphism. In particular, S ≅ A S \cong A S ≅ A . If ι : S ↪ A n \iota: S \hookrightarrow A^n ι : S ↪ A n is the inclusion, then ι ∘ ( π i ∣ S ) − 1 = Φ f \iota\circ(\pi_i|_S)^{-1} = \Phi_f ι ∘ ( π i ∣ S ) − 1 = Φ f for some 0 ≠ f ∈ D n 0 \ne f \in D^n 0 = f ∈ D n , and S = im Φ f S = \operatorname{im}\Phi_f S = im Φ f .
Now let 0 ≠ f , f ′ ∈ D n 0 \ne f, f' \in D^n 0 = f , f ′ ∈ D n . We claim that im Φ f = im Φ f ′ \operatorname{im}\Phi_f = \operatorname{im}\Phi_{f'} im Φ f = im Φ f ′ if and only if f ′ = f g f' = fg f ′ = f g for some g ∈ D × g \in D^\times g ∈ D × , that is, if and only if f D = f ′ D fD = f'D f D = f ′ D . If f ′ = f g f' = fg f ′ = f g with g ∈ D × g \in D^\times g ∈ D × , then g g g is onto and so im Φ f ′ = im ( Φ f ∘ g ) = im Φ f \operatorname{im}\Phi_{f'} = \operatorname{im}(\Phi_f \circ g) = \operatorname{im}\Phi_f im Φ f ′ = im ( Φ f ∘ g ) = im Φ f . Conversely, if im Φ f = im Φ f ′ = S \operatorname{im}\Phi_f = \operatorname{im}\Phi_{f'} = S im Φ f = im Φ f ′ = S , then Φ f , Φ f ′ : A → S \Phi_f,\Phi_{f'} : A \to S Φ f , Φ f ′ : A → S are isomorphisms. So g = Φ f ′ − 1 ∘ Φ f ∈ D × g = \Phi_{f'}^{-1} \circ\Phi_f \in D^\times g = Φ f ′ − 1 ∘ Φ f ∈ D × and Φ f g = Φ f ∘ g = Φ f ′ \Phi_{fg} = \Phi_f \circ g = \Phi_{f'} Φ f g = Φ f ∘ g = Φ f ′ , which gives f ′ = f g f' = fg f ′ = f g . Hence
f D ⟼ im Φ f fD \longmapsto\operatorname{im}\Phi_f f D ⟼ im Φ f is a well-defined bijection from P n − 1 ( D ) \mathbb{P}^{n-1}(D) P n − 1 ( D ) onto the set of simple submodules of A n A^n A n . Every non-zero f ∈ D n f \in D^n f ∈ D n lies in exactly one line, namely f D fD f D . Also g ↦ f g g \mapsto fg g ↦ f g is injective, since f g = f g ′ fg = fg' f g = f g ′ with g ≠ g ′ g \ne g' g = g ′ would give f = f ( g − g ′ ) ( g − g ′ ) − 1 = 0 f = f(g-g')(g-g')^{-1} = 0 f = f ( g − g ′ ) ( g − g ′ ) − 1 = 0 . So each line has exactly ∣ D ∣ − 1 |D|-1 ∣ D ∣ − 1 non-zero elements, and thus the number of simple submodules of A n A^n A n is
∣ P n − 1 ( D ) ∣ = ∣ D ∣ n − 1 ∣ D ∣ − 1 . |\mathbb{P}^{n-1}(D)| = \frac{|D|^n-1}{|D|-1}. ∣ P n − 1 ( D ) ∣ = ∣ D ∣ − 1 ∣ D ∣ n − 1 . Finally, let A 1 ≠ A 2 A_1 \ne A_2 A 1 = A 2 be simple submodules of A 2 A^2 A 2 . If A 1 ∩ A 2 ≠ { 0 } A_1 \cap A_2 \ne\{0\} A 1 ∩ A 2 = { 0 } , then A 1 ∩ A 2 = A 1 A_1 \cap A_2 = A_1 A 1 ∩ A 2 = A 1 as A 1 A_1 A 1 is simple, so A 1 ≤ A 2 A_1 \le A_2 A 1 ≤ A 2 and hence A 1 = A 2 A_1 = A_2 A 1 = A 2 as A 2 A_2 A 2 is simple, which is a contradiction. So A 1 ∩ A 2 = { 0 } A_1 \cap A_2 = \{0\} A 1 ∩ A 2 = { 0 } , and as A 1 ≅ A ≅ A 2 A_1 \cong A \cong A_2 A 1 ≅ A ≅ A 2 , we get
∣ A 1 + A 2 ∣ = ∣ A 1 ∣ ∣ A 2 ∣ = ∣ A ∣ 2 = ∣ A 2 ∣ . |A_1+A_2| = |A_1||A_2| = |A|^2 = |A^2|. ∣ A 1 + A 2 ∣ = ∣ A 1 ∣∣ A 2 ∣ = ∣ A ∣ 2 = ∣ A 2 ∣. Hence, A 1 ⊕ A 2 = A 2 A_1 \oplus A_2 = A^2 A 1 ⊕ A 2 = A 2 .
Applying Lemma 4.3 to the isotypic components of the abelian part of the socle, we obtain the complete geometry of the set of minimal normal subgroups of a finite group and an expression for the socle length.
Proposition 4.4. Let G G G be a non-trivial finite group. Let B 1 , … , B b B_1,\ldots,B_b B 1 , … , B b be its non-abelian minimal normal subgroups. For each prime p p p , let S p S_p S p be the subgroup generated by the abelian minimal normal subgroups that are p p p -groups. Then
Soc ( G ) = B 1 × ⋯ × B b × ∏ p S p . \operatorname{Soc}(G)=B_1\times\cdots\times B_b\times\prod_p S_p. Soc ( G ) = B 1 × ⋯ × B b × p ∏ S p . and each S p S_p S p is a semisimple F p G \mathbb{F}_pG F p G -module. Let A p , 1 , … , A p , m p A_{p,1},\ldots,A_{p,m_p} A p , 1 , … , A p , m p be representatives of the isomorphism classes of simple submodules of S p S_p S p , and let S p , j S_{p,j} S p , j be the sum of all simple submodules of S p S_p S p isomorphic to A p , j A_{p,j} A p , j . Then S p S_p S p has the isotypic decomposition
S p = ⨁ j = 1 m p S p , j , S p , j ≅ A p , j n p , j , S_p=\bigoplus_{j=1}^{m_p}S_{p,j},\qquad S_{p,j}\cong A_{p,j}^{n_{p,j}}, S p = j = 1 ⨁ m p S p , j , S p , j ≅ A p , j n p , j , where n p , j ≥ 1 n_{p,j}\geq1 n p , j ≥ 1 . Set
D p , j = End F p G ( A p , j ) . D_{p,j}=\operatorname{End}_{\mathbb{F}_pG}(A_{p,j}). D p , j = End F p G ( A p , j ) . Then:
the minimal normal subgroups of G G G are the groups B i B_i B i together with the simple submodules of the homogeneous components S p , j S_{p,j} S p , j ;
their number is
b + ∑ p ∑ j = 1 m p ∣ D p , j ∣ n p , j − 1 ∣ D p , j ∣ − 1 ; b+\sum_p\sum_{j=1}^{m_p}\frac{|D_{p,j}|^{n_{p,j}}-1}{|D_{p,j}|-1}; b + p ∑ j = 1 ∑ m p ∣ D p , j ∣ − 1 ∣ D p , j ∣ n p , j − 1 ; the socle length is
σ ( G ) = b + ∑ p ∑ j = 1 m p n p , j . \sigma(G)=b+\sum_p\sum_{j=1}^{m_p}n_{p,j}. σ ( G ) = b + p ∑ j = 1 ∑ m p n p , j .
Proof. Let N ∈ MinNorm ( G ) N\in\operatorname{MinNorm}(G) N ∈ MinNorm ( G ) be abelian and let p p p be a prime dividing ∣ N ∣ |N| ∣ N ∣ . The Sylow p p p -subgroup of N N N and { x ∈ N : x p = 1 } \{x\in N:x^p=1\} { x ∈ N : x p = 1 } are non-trivial characteristic subgroups of N N N , and hence normal in G G G . So both are equal to N N N , that is, N N N is an elementary abelian p p p -group. Thus every minimal normal subgroup of G G G is some B i B_i B i or lies in some S p S_p S p , and hence Soc ( G ) = ⋃ B i ⋯ B b \operatorname{Soc}(G)=\bigcup B_i\cdots B_b Soc ( G ) = ⋃ B i ⋯ B b , where U = ∏ p S p U=\prod_p S_p U = ∏ p S p .
By Lemma 3.6 , distinct minimal normal subgroups commute elementwise. So U U U is abelian and each S p S_p S p is an elementary abelian p p p -group. As the S p S_p S p have pairwise coprime orders, U = ∏ p S p U=\prod_p S_p U = ∏ p S p is a direct product. Now Z ( B i ) Z(B_i) Z ( B i ) is characteristic in B i B_i B i , hence normal in G G G , and Z ( B i ) ≠ B i Z(B_i)\ne B_i Z ( B i ) = B i as B i B_i B i is non-abelian. So Z ( B i ) = 1 Z(B_i)=1 Z ( B i ) = 1 . Again by Lemma 3.6 , U B 1 ⋯ B i − 1 UB_1\cdots B_{i-1} U B 1 ⋯ B i − 1 centralizes B i B_i B i , and so
B i ∩ U B 1 ⋯ B i − 1 ≤ Z ( B i ) = 1. B_i\cap UB_1\cdots B_{i-1}\leq Z(B_i)=1. B i ∩ U B 1 ⋯ B i − 1 ≤ Z ( B i ) = 1.
As all these subgroups are normal in G G G , it follows that Soc ( G ) = B 1 × ⋯ × B b × ∏ p S p \operatorname{Soc}(G)=B_1\times\cdots\times B_b\times\prod_p S_p Soc ( G ) = B 1 × ⋯ × B b × ∏ p S p .
Now S p ⊴ G S_p\trianglelefteq G S p ⊴ G and S p S_p S p is elementary abelian, so G G G acts F p \mathbb{F}_p F p -linearly on S p S_p S p by conjugation, and S p S_p S p is an F p G \mathbb{F}_pG F p G -module. As every subgroup of S p S_p S p is an F p \mathbb{F}_p F p -subspace, the normal subgroups of G G G contained in S p S_p S p are exactly the F p G \mathbb{F}_pG F p G -submodules of S p S_p S p . In particular, the abelian minimal normal p p p -subgroups of G G G are exactly the simple submodules of S p S_p S p . As S p S_p S p is generated by them, S p S_p S p is a sum of simple submodules, and hence it is semisimple.
So let S p = T 1 ⊕ ⋯ ⊕ T s S_p=T_1\oplus\cdots\oplus T_s S p = T 1 ⊕ ⋯ ⊕ T s with simple submodules T k T_k T k . If T T T is a simple submodule of S p S_p S p with T ≅ A p , j T\cong A_{p,j} T ≅ A p , j , then by Schur’s Lemma the projection of T T T onto T k T_k T k is zero whenever T k ≇ A p , j T_k\not\cong A_{p,j} T k ≅ A p , j . So T T T lies in the sum of those T k T_k T k with T k ≅ A p , j T_k\cong A_{p,j} T k ≅ A p , j . Hence
S p , j = ⨁ T k ≅ A p , j T k ≅ A p , j n p , j , where n p , j = # { k : T k ≅ A p , j } . S_{p,j}=\bigoplus_{T_k\cong A_{p,j}}T_k\cong A_{p,j}^{n_{p,j}},\qquad\text{where }n_{p,j}=\#\{k:T_k\cong A_{p,j}\}. S p , j = T k ≅ A p , j ⨁ T k ≅ A p , j n p , j , where n p , j = # { k : T k ≅ A p , j } .
and n p , j ≥ 1 n_{p,j} \ge1 n p , j ≥ 1 , as some simple submodule of S p S_p S p is isomorphic to A p , j A_{p,j} A p , j . Also, each T k T_k T k is isomorphic to exactly one A p , j A_{p,j} A p , j , so S p = ⨁ j S p , j S_p = \bigoplus_j S_{p,j} S p = ⨁ j S p , j . By the definition of S p , j S_{p,j} S p , j and as this sum is direct, every simple submodule of S p S_p S p lies in exactly one S p , j S_{p,j} S p , j . This proves (1).
By Lemma 4.3 with R = F p G R = \mathbb{F}_pG R = F p G , the module S p , j ≅ A p , j n p , j S_{p,j} \cong A_{p,j}^{n_{p,j}} S p , j ≅ A p , j n p , j has exactly
∣ D p , j ∣ n p , j − 1 ∣ D p , j ∣ − 1 \frac{\lvert D_{p,j}\rvert^{n_{p,j}}-1}{\lvert D_{p,j}\rvert-1} ∣ D p , j ∣ − 1 ∣ D p , j ∣ n p , j − 1
simple submodules. Summing over all p p p and j j j and adding the b b b non-abelian ones gives (2). Finally, the B i B_i B i and the T k T_k T k (for all p p p ) exhibit Soc ( G ) \operatorname{Soc}(G) Soc ( G ) as a direct product X 1 × ⋯ × X t X_1 \times\cdots\times X_t X 1 × ⋯ × X t of t = b + ∑ p ∑ j n p , j t=b+\sum_p\sum_j n_{p,j} t = b + ∑ p ∑ j n p , j minimal normal subgroups. As X l ∩ X 1 ⋯ X l − 1 = 1 X_l \cap X_1\cdots X_{l-1}=1 X l ∩ X 1 ⋯ X l − 1 = 1 and λ G ( X l ) = 1 \lambda_G(X_l)=1 λ G ( X l ) = 1 , (2) of Lemma 3.2 gives
λ G ( X 1 ⋯ X l ) = λ G ( X 1 ⋯ X l − 1 ) + 1. \lambda_G(X_1\cdots X_l)=\lambda_G(X_1\cdots X_{l-1})+1. λ G ( X 1 ⋯ X l ) = λ G ( X 1 ⋯ X l − 1 ) + 1.
Hence σ ( G ) = λ G ( Soc ( G ) ) = t \sigma(G)=\lambda_G(\operatorname{Soc}(G))=t σ ( G ) = λ G ( Soc ( G )) = t , which is (3). □
We can now prove Theorem B. The point is that the combinatorial condition of Theorem 4.2 (4) is equivalent to the numerical condition σ ( G ) ≤ 2 \sigma(G)\le2 σ ( G ) ≤ 2 , and that Proposition 4.4 turns the latter into a short list of socle types.
Theorem 4.5. Let G G G be a non-trivial finite group and put S = Soc ( G ) S=\operatorname{Soc}(G) S = Soc ( G ) . The following are equivalent:
(1) G G G is M \mathcal{M} M -full;
(2) A B = S AB=S A B = S for every two distinct A , B ∈ MinNorm ( G ) A,B\in\operatorname{MinNorm}(G) A , B ∈ MinNorm ( G ) ;
(3) σ ( G ) ≤ 2 \sigma(G)\le2 σ ( G ) ≤ 2 ;
(4) one of the following mutually exclusive alternatives holds:
(a) G G G has exactly one minimal normal subgroup;
(b) G G G has exactly two minimal normal subgroups;
(c) G G G has at least three minimal normal subgroups, and there are a prime p p p and an irreducible F p G \mathbb{F}_pG F p G -module A A A such that S ≅ A ⊕ A S\cong A\oplus A S ≅ A ⊕ A as G G G -modules.
In case (4c), if D = End F p G ( A ) D=\operatorname{End}_{\mathbb{F}_pG}(A) D = End F p G ( A ) , then
MinNorm ( G ) ≅ P 1 ( D ) and ∣ MinNorm ( G ) ∣ = ∣ D ∣ + 1. \operatorname{MinNorm}(G)\cong\mathbb{P}^{1}(D) \qquad\text{and}\qquad\lvert\operatorname{MinNorm}(G)\rvert=\lvert D\rvert+1. MinNorm ( G ) ≅ P 1 ( D ) and ∣ MinNorm ( G )∣ = ∣ D ∣ + 1.
Proof. (1)⇔(2) By Theorem 4.2 (4).
(2)⇒(3) Assume that A B = S AB=S A B = S , for any two distinct A , B ∈ MinNorm ( G ) A,B\in\operatorname{MinNorm}(G) A , B ∈ MinNorm ( G ) . Suppose, σ ( G ) ≥ 3 \sigma(G)\ge3 σ ( G ) ≥ 3 . Then, there is a chain
where N , M ⊴ G N,M\trianglelefteq G N , M ⊴ G . As G G G is finite and hence normally atomic, there exists A ∈ MinNorm ( G ) A\in\operatorname{MinNorm}(G) A ∈ MinNorm ( G ) such that A ≤ N A\le N A ≤ N . As M < S M<S M < S and S S S is generated by the minimal normal subgroups of G G G , there exists B ∈ MinNorm ( G ) B\in\operatorname{MinNorm}(G) B ∈ MinNorm ( G ) such that B ≰ M B\nleq M B ≰ M . Then A ≠ B A\ne B A = B , as A ≤ N < M A\le N<M A ≤ N < M . So by our hypothesis, A B = S AB=S A B = S . Now, M ∩ B M\cap B M ∩ B is a normal subgroup of G G G properly contained in B B B , and hence M ∩ B = 1 M\cap B=1 M ∩ B = 1 . As A ≤ M A\le M A ≤ M , by Lemma 3.1 ,
M = A B ∩ M = A ( B ∩ M ) = A ≤ N , M=AB\cap M=A(B\cap M)=A\le N, M = A B ∩ M = A ( B ∩ M ) = A ≤ N ,
which is a contradiction as N < M N<M N < M . Thus, σ ( G ) ≤ 2 \sigma(G)\le2 σ ( G ) ≤ 2 .
(3)⇒(4) Assume σ ( G ) ≤ 2 \sigma(G)\le2 σ ( G ) ≤ 2 . As G ≠ { 1 } G\ne\{1\} G = { 1 } , S ≠ { 1 } S\ne\{1\} S = { 1 } and σ ( G ) ≥ 1 \sigma(G)\ge1 σ ( G ) ≥ 1 . We use the notation of Proposition 4.4 . First suppose n p , j = 1 n_{p,j}=1 n p , j = 1 for all p p p and j j j . Then each term
∣ D p , j ∣ n p , j − 1 ∣ D p , j ∣ − 1 (1) \frac{\lvert D_{p,j}\rvert^{n_{p,j}}-1}{\lvert D_{p,j}\rvert-1} \tag*{(1)} ∣ D p , j ∣ − 1 ∣ D p , j ∣ n p , j − 1 (1)
is 1 1 1 , and so by Proposition 4.4 (2),(3),
∣ MinNorm ( G ) ∣ = b + ∑ p ∑ j = 1 m p n p , j = σ ( G ) ∈ { 1 , 2 } . |\operatorname{MinNorm}(G)|=b+\sum_{p}\sum_{j=1}^{m_p}n_{p,j}=\sigma(G)\in\{1,2\}. ∣ MinNorm ( G ) ∣ = b + p ∑ j = 1 ∑ m p n p , j = σ ( G ) ∈ { 1 , 2 } .
So, G G G has exactly one or exactly two minimal normal subgroups. Now suppose n p , j ≥ 2 n_{p,j}\ge2 n p , j ≥ 2 for some p p p and j j j . As σ ( G ) = b + ∑ p ∑ j n p , j ≤ 2 \sigma(G)=b+\sum_p\sum_j n_{p,j}\le2 σ ( G ) = b + ∑ p ∑ j n p , j ≤ 2 , b = 0 b=0 b = 0 , n p , j = 2 n_{p,j}=2 n p , j = 2 , and S p , j S_{p,j} S p , j is the only homogeneous component. Thus, S = S p = S p , j ≅ A ⊕ A S=S_p=S_{p,j}\cong A\oplus A S = S p = S p , j ≅ A ⊕ A , where A = A p , j A=A_{p,j} A = A p , j is an irreducible F p G \mathbb{F}_pG F p G -module. By Proposition 4.4 (2), with D = End F p G ( A ) D=\operatorname{End}_{\mathbb{F}_pG}(A) D = End F p G ( A ) ,
∣ MinNorm ( G ) ∣ = ∣ D ∣ 2 − 1 ∣ D ∣ − 1 = ∣ D ∣ + 1 ≥ 3 |\operatorname{MinNorm}(G)|=\frac{|D|^2-1}{|D|-1}=|D|+1\ge3 ∣ MinNorm ( G ) ∣ = ∣ D ∣ − 1 ∣ D ∣ 2 − 1 = ∣ D ∣ + 1 ≥ 3
as ∣ D ∣ ≥ 2 |D|\ge2 ∣ D ∣ ≥ 2 . So (c) holds. Clearly, the alternatives (a), (b), (c) are mutually exclusive.
(4)⇒ \Rightarrow ⇒ (2) In case (a), there is nothing to prove. In case (b), if M 1 , M 2 M_1,M_2 M 1 , M 2 are the two minimal normal subgroups of G G G , then S = ⟨ M 1 , M 2 ⟩ = M 1 M 2 S=\langle M_1,M_2\rangle=M_1M_2 S = ⟨ M 1 , M 2 ⟩ = M 1 M 2 . In case (c), S ≅ A ⊕ A S\cong A\oplus A S ≅ A ⊕ A is an elementary abelian p p p -group and every subgroup of S S S is an F p \mathbb{F}_p F p -subspace. So the normal subgroups of G G G contained in S S S are exactly the F p G \mathbb{F}_pG F p G -submodules of S S S . As every minimal normal subgroup of G G G lies in S S S , the minimal normal subgroups of G G G are exactly the simple submodules of S ≅ A 2 S\cong A^2 S ≅ A 2 . Hence by Lemma 4.3 ,
MinNorm ( G ) ≅ P 1 ( D ) , ∣ MinNorm ( G ) ∣ = ∣ D ∣ + 1 , \operatorname{MinNorm}(G)\cong\mathbb{P}^{1}(D),\qquad|\operatorname{MinNorm}(G)|=|D|+1, MinNorm ( G ) ≅ P 1 ( D ) , ∣ MinNorm ( G ) ∣ = ∣ D ∣ + 1 ,
and any two distinct M 1 , M 2 ∈ MinNorm ( G ) M_1,M_2\in\operatorname{MinNorm}(G) M 1 , M 2 ∈ MinNorm ( G ) satisfy M 1 ⊕ M 2 = S M_1\oplus M_2=S M 1 ⊕ M 2 = S , that is, M 1 M 2 = S M_1M_2=S M 1 M 2 = S . □
Whether a group is M \mathcal{M} M -full depends only on its normal subgroups, not on its size or on its other subgroups. For example, for every n ≥ 2 n\ge2 n ≥ 2 , the group S n S_n S n is M \mathcal{M} M -full (Example 2.6 ); thus S 5 S_5 S 5 , which has 156 subgroups, is M \mathcal{M} M -full because it has only three normal subgroups. The following example shows that each of the three alternatives in Theorem 4.5 (4) actually occurs, already inside one family of groups.
Example 4.6. For n ≥ 2 n\ge2 n ≥ 2 let SL 2 ( Z n ) \operatorname{SL}_2(\mathbb{Z}_n) SL 2 ( Z n ) be the group of 2 × 2 2\times2 2 × 2 matrices over Z n \mathbb{Z}_n Z n with determinant 1 1 1 . The three cases of Theorem 4.5 (4) all occur.
For n = 3 , 5 n=3,5 n = 3 , 5 , the group SL 2 ( Z n ) \operatorname{SL}_2(\mathbb{Z}_n) SL 2 ( Z n ) is monolithic with minimal normal subgroup { I , − I } \{I,-I\} { I , − I } . Indeed, the normal subgroups of SL 2 ( Z 3 ) \operatorname{SL}_2(\mathbb{Z}_3) SL 2 ( Z 3 ) are 1 1 1 , { I , − I } \{I,-I\} { I , − I } , the quaternion subgroup of order 8 8 8 and SL 2 ( Z 3 ) \operatorname{SL}_2(\mathbb{Z}_3) SL 2 ( Z 3 ) , while those of SL 2 ( Z 5 ) \operatorname{SL}_2(\mathbb{Z}_5) SL 2 ( Z 5 ) are 1 1 1 , { I , − I } \{I,-I\} { I , − I } and SL 2 ( Z 5 ) \operatorname{SL}_2(\mathbb{Z}_5) SL 2 ( Z 5 ) , because SL 2 ( Z 5 ) / { I , − I } ≅ A 5 \operatorname{SL}_2(\mathbb{Z}_5)/\{I,-I\}\cong A_5 SL 2 ( Z 5 ) / { I , − I } ≅ A 5 is simple and SL 2 ( Z 5 ) \operatorname{SL}_2(\mathbb{Z}_5) SL 2 ( Z 5 ) is perfect. This is case (4a).
Clearly, SL 2 ( Z 6 ) ≅ SL 2 ( Z 2 ) × SL 2 ( Z 3 ) ≅ S 3 × SL 2 ( Z 3 ) \operatorname{SL}_2(\mathbb{Z}_6)\cong\operatorname{SL}_2(\mathbb{Z}_2)\times\operatorname{SL}_2(\mathbb{Z}_3)\cong S_3\times\operatorname{SL}_2(\mathbb{Z}_3) SL 2 ( Z 6 ) ≅ SL 2 ( Z 2 ) × SL 2 ( Z 3 ) ≅ S 3 × SL 2 ( Z 3 ) . By (1) and Theorem 5.1 ,
Soc ( S 3 × SL 2 ( Z 3 ) ) = A 3 × { I , − I } . \operatorname{Soc}(S_3\times\operatorname{SL}_2(\mathbb{Z}_3))=A_3\times\{I,-I\}. Soc ( S 3 × SL 2 ( Z 3 )) = A 3 × { I , − I } .
The only minimal normal subgroups are A 3 × 1 A_3\times1 A 3 × 1 and 1 × { I , − I } 1\times\{I,-I\} 1 × { I , − I } : the only non-central minimal normal subgroup of a factor is A 3 ⊴ S 3 A_3\trianglelefteq S_3 A 3 ⊴ S 3 , and, as Z ( S 3 ) = 1 Z(S_3)=1 Z ( S 3 ) = 1 and Z ( SL 2 ( Z 3 ) ) = { I , − I } Z(\operatorname{SL}_2(\mathbb{Z}_3))=\{I,-I\} Z ( SL 2 ( Z 3 )) = { I , − I } , the space V 2 V_2 V 2 is the line 1 × { I , − I } 1\times\{I,-I\} 1 × { I , − I } and V p = 1 V_p=1 V p = 1 for p ≠ 2 p\ne2 p = 2 . This is case (4b).
For n = 15 n=15 n = 15 , we have SL 2 ( Z 15 ) ≅ SL 2 ( Z 3 ) × SL 2 ( Z 5 ) \operatorname{SL}_2(\mathbb{Z}_{15})\cong\operatorname{SL}_2(\mathbb{Z}_3)\times\operatorname{SL}_2(\mathbb{Z}_5) SL 2 ( Z 15 ) ≅ SL 2 ( Z 3 ) × SL 2 ( Z 5 ) . So by (1) and Theorem 5.1 ,
Soc ( SL 2 ( Z 15 ) ) ≅ { I , − I } × { I , − I } ≅ C 2 × C 2 . \operatorname{Soc}(\operatorname{SL}_2(\mathbb{Z}_{15}))\cong\{I,-I\}\times\{I,-I\}\cong C_2\times C_2. Soc ( SL 2 ( Z 15 )) ≅ { I , − I } × { I , − I } ≅ C 2 × C 2 .
So the minimal normal subgroups are exactly its three subgroups of order 2 2 2 . Inside SL 2 ( Z 15 ) \operatorname{SL}_2(\mathbb{Z}_{15}) SL 2 ( Z 15 ) the minimal normal subgroups are generated by 4 I 4I 4 I , 11 I 11I 11 I and 14 I = − I 14I=-I 14 I = − I and they correspond to ( I , − I ) (I,-I) ( I , − I ) , ( − I , I ) (-I,I) ( − I , I ) and ( − I , − I ) (-I,-I) ( − I , − I ) , since 4 ≡ 1 ( m o d 3 ) 4\equiv1\pmod3 4 ≡ 1 ( mod 3 ) , 4 ≡ − 1 ( m o d 5 ) 4\equiv-1\pmod5 4 ≡ − 1 ( mod 5 ) , and 11 ≡ − 1 ( m o d 3 ) 11\equiv-1\pmod3 11 ≡ − 1 ( mod 3 ) , 11 ≡ 1 ( m o d 5 ) 11\equiv1\pmod5 11 ≡ 1 ( mod 5 ) . Here A = Z 2 A=\mathbb{Z}_2 A = Z 2 is the trivial module and D = Z 2 D=\mathbb{Z}_2 D = Z 2 , and ∣ MinNorm ( SL 2 ( Z 15 ) ) ∣ = 3 = ∣ D ∣ + 1 |\operatorname{MinNorm}(\operatorname{SL}_2(\mathbb{Z}_{15}))|=3=|D|+1 ∣ MinNorm ( SL 2 ( Z 15 )) ∣ = 3 = ∣ D ∣ + 1 . This is case (4c).
For p p p -groups all minimal normal subgroups are central of order p p p , and Theorem 4.5 becomes a condition on the rank of the centre.
Corollary 4.7. Let G G G be a non-trivial finite p p p -group and put
V = Ω 1 ( Z ( G ) ) = { z ∈ Z ( G ) : z p = 1 } . V=\Omega_1(Z(G))=\{z\in Z(G):z^p=1\}. V = Ω 1 ( Z ( G )) = { z ∈ Z ( G ) : z p = 1 } . Then the minimal normal subgroups of G G G are exactly the one-dimensional F p \mathbb{F}_p F p -subspaces of V V V , and Soc ( G ) = V \operatorname{Soc}(G)=V Soc ( G ) = V . Moreover, G G G is M \mathcal{M} M -full if and only if dim F p V ≤ 2 \dim_{\mathbb{F}_p}V\le2 dim F p V ≤ 2 .
Proof. As Z ( G ) Z(G) Z ( G ) is abelian, V V V is an F p \mathbb{F}_p F p -vector space. Let N ∈ MinNorm ( G ) N\in\operatorname{MinNorm}(G) N ∈ MinNorm ( G ) and let G ↷ N G\curvearrowright N G ↷ N by
g ⋅ n = g n g − 1 . g\cdot n=gng^{-1}. g ⋅ n = g n g − 1 .
Then, as G G G is a p p p -group, N N N is also a p p p -group and hence the class equation of this action gives:
∣ N ∣ = ∣ N G ∣ + ∑ i ∣ O i ∣ . |N|=|N^G|+\sum_i|\mathcal{O}_i|. ∣ N ∣ = ∣ N G ∣ + i ∑ ∣ O i ∣.
where p ∣ ∣ O i ∣ p\mid|\mathcal{O}_i| p ∣ ∣ O i ∣ for each i i i . So it immediately follows that p ∣ ∣ N G ∣ p\mid|N^G| p ∣ ∣ N G ∣ . But,
N G = { n ∈ N : g n g − 1 = n , ∀ g ∈ G } = N ∩ Z ( G ) . N^G=\{n\in N:gng^{-1}=n,\ \forall g\in G\}=N\cap Z(G). N G = { n ∈ N : g n g − 1 = n , ∀ g ∈ G } = N ∩ Z ( G ) .
Now, 1 ≠ N ∩ Z ( G ) ⊴ G 1\ne N\cap Z(G)\trianglelefteq G 1 = N ∩ Z ( G ) ⊴ G and N ∩ Z ( G ) ≤ N N\cap Z(G)\le N N ∩ Z ( G ) ≤ N . But, N ∈ MinNorm ( G ) N\in\operatorname{MinNorm}(G) N ∈ MinNorm ( G ) implies that N = N ∩ Z ( G ) N=N\cap Z(G) N = N ∩ Z ( G ) and hence, N ≤ Z ( G ) N\le Z(G) N ≤ Z ( G ) . So, every subgroup of N N N is normal in G G G . Now, if ∣ N ∣ > p |N|>p ∣ N ∣ > p , then there exists L ≤ N L\le N L ≤ N such that 1 < L < N 1<L<N 1 < L < N and ∣ L ∣ = p |L|=p ∣ L ∣ = p . But, L ⊴ G L\trianglelefteq G L ⊴ G — which is a contradiction as N ∈ MinNorm ( G ) N\in\operatorname{MinNorm}(G) N ∈ MinNorm ( G ) . So, ∣ N ∣ = p |N|=p ∣ N ∣ = p . Thus, every minimal normal subgroup of G G G is cyclic of order p p p and hence a one-dimensional subspace of V V V . So, N ≤ V N\le V N ≤ V , and we get
MinNorm ( G ) ⊆ { 1 -dimensional F p -subspaces of V } . \operatorname{MinNorm}(G)\subseteq\{1\text{-dimensional }\mathbb{F}_p\text{-subspaces of }V\}. MinNorm ( G ) ⊆ { 1 -dimensional F p -subspaces of V } .
Conversely, if L ≤ V L\le V L ≤ V is a one-dimensional subspace of V V V , then ∣ L ∣ = p |L|=p ∣ L ∣ = p . As L ≤ V ≤ Z ( G ) L\le V\le Z(G) L ≤ V ≤ Z ( G ) , we have L ⊴ G L\trianglelefteq G L ⊴ G . As a group of order p p p has no non-trivial proper subgroups, it follows that L ∈ MinNorm ( G ) L\in\operatorname{MinNorm}(G) L ∈ MinNorm ( G ) . Hence,
{ 1 -dimensional F p -subspaces of V } ⊆ MinNorm ( G ) . \{1\text{-dimensional }\mathbb{F}_p\text{-subspaces of }V\}\subseteq\operatorname{MinNorm}(G). { 1 -dimensional F p -subspaces of V } ⊆ MinNorm ( G ) .
Thus, the minimal normal subgroups of G G G are exactly the one-dimensional F p \mathbb{F}_p F p -subspaces of V V V .
Now, as M ∈ MinNorm ( G ) M\in\operatorname{MinNorm}(G) M ∈ MinNorm ( G ) implies that M ≤ V M\le V M ≤ V , it follows at once, Soc ( G ) ≤ V \operatorname{Soc}(G)\le V Soc ( G ) ≤ V . But, V V V is an F p \mathbb{F}_p F p -vector space, so it is generated by its one-dimensional F p \mathbb{F}_p F p -subspaces. Hence, V ≤ Soc ( G ) V\le\operatorname{Soc}(G) V ≤ Soc ( G ) . Thus, we get Soc ( G ) = V \operatorname{Soc}(G)=V Soc ( G ) = V .
Now, observe that the interval [ 1 , V ] N ( G ) [1,V]_{\mathcal{N}(G)} [ 1 , V ] N ( G ) is the interval of F p \mathbb{F}_p F p -subspaces of V V V . So,
[ 1 , V ] N ( G ) = SubSp F p ( V ) , [1,V]_{\mathcal{N}(G)}=\operatorname{SubSp}_{\mathbb{F}_p}(V), [ 1 , V ] N ( G ) = SubSp F p ( V ) ,
where SubSp F p ( V ) \operatorname{SubSp}_{\mathbb{F}_p}(V) SubSp F p ( V ) denotes the lattice of F p \mathbb{F}_p F p -subspaces of V V V . But, a maximal chain in SubSp F p ( V ) \operatorname{SubSp}_{\mathbb{F}_p}(V) SubSp F p ( V ) looks like
V 0 = { 0 } < V 1 < V 2 < ⋯ < V n = V , V_0=\{0\}<V_1<V_2<\cdots<V_n=V, V 0 = { 0 } < V 1 < V 2 < ⋯ < V n = V ,
where n = dim F p V n=\dim_{\mathbb{F}_p}V n = dim F p V , i.e., there are dim F p V \dim_{\mathbb{F}_p}V dim F p V number of inclusions. Hence, λ G ( V ) = dim F p V \lambda_G(V)=\dim_{\mathbb{F}_p}V λ G ( V ) = dim F p V . But, Soc ( G ) = V \operatorname{Soc}(G)=V Soc ( G ) = V . So, [ 1 , V ] N ( G ) = [ 1 , Soc ( G ) ] N ( G ) [1,V]_{\mathcal{N}(G)}=[1,\operatorname{Soc}(G)]_{\mathcal{N}(G)} [ 1 , V ] N ( G ) = [ 1 , Soc ( G ) ] N ( G ) , and thus
dim F p V = λ G ( V ) = λ G ( Soc ( G ) ) = σ ( G ) . \dim_{\mathbb{F}_p}V=\lambda_G(V)=\lambda_G(\operatorname{Soc}(G))=\sigma(G). dim F p V = λ G ( V ) = λ G ( Soc ( G )) = σ ( G ) .
But, by Theorem 4.5, we know that
G is M -full ⟺ σ ( G ) ≤ 2. G\text{ is }\mathcal{M}\text{-full}\Longleftrightarrow\sigma(G)\le2. G is M -full ⟺ σ ( G ) ≤ 2.
Thus, G G G is M \mathcal{M} M -full if and only if dim F p V ≤ 2 \dim_{\mathbb{F}_p}V\le2 dim F p V ≤ 2 .
Since a finite nilpotent group is the direct product of its Sylow subgroups, the same argument applies prime by prime.
Corollary 4.8. Let G G G be a non-trivial finite nilpotent group, let P p P_p P p denote its Sylow p p p -subgroup, and put
ρ ( G ) = ∑ p ∣ ∣ G ∣ dim F p { z ∈ Z ( P p ) : z p = 1 } . \rho(G)=\sum_{p\mid|G|}\dim_{\mathbb{F}_p}\{z\in Z(P_p):z^p=1\}. ρ ( G ) = p ∣ ∣ G ∣ ∑ dim F p { z ∈ Z ( P p ) : z p = 1 } . Then G G G is M \mathcal{M} M -full if and only if ρ ( G ) ≤ 2 \rho(G)\le2 ρ ( G ) ≤ 2 .
Proof. Let G G G be a finite nilpotent group. Then,
G = ∏ p ∣ ∣ G ∣ P p G=\prod_{p\mid|G|}P_p G = p ∣ ∣ G ∣ ∏ P p
where each P p P_p P p is the unique Sylow p p p -subgroup of G G G . Then,
Z ( G ) = Z ( ∏ p ∣ ∣ G ∣ P p ) = ∏ p ∣ ∣ G ∣ Z ( P p ) . Z(G)=Z\left(\prod_{p\mid|G|}P_p\right)=\prod_{p\mid|G|}Z(P_p). Z ( G ) = Z p ∣ ∣ G ∣ ∏ P p = p ∣ ∣ G ∣ ∏ Z ( P p ) .
Now, as each P p P_p P p is a finite p p p -group, Z ( P p ) Z(P_p) Z ( P p ) is a finite abelian p p p -group and hence, Ω 1 ( Z ( P p ) ) \Omega_1(Z(P_p)) Ω 1 ( Z ( P p )) is an elementary abelian p p p -group. Thus, for each p ∣ ∣ G ∣ p\mid|G| p ∣ ∣ G ∣ , Ω 1 ( Z ( P p ) ) \Omega_1(Z(P_p)) Ω 1 ( Z ( P p )) is an F p \mathbb{F}_p F p -vector space. Now, let N ∈ MinNorm ( G ) N\in\operatorname{MinNorm}(G) N ∈ MinNorm ( G ) . Then,
N = ∏ p ∣ ∣ G ∣ ( N ∩ P p ) , N=\prod_{p\mid|G|}(N\cap P_p), N = p ∣ ∣ G ∣ ∏ ( N ∩ P p ) ,
that is, every element of N N N is the product of its p p p -parts, which are powers of it and so lie in N ∩ P p N\cap P_p N ∩ P p , where N ∩ P p ⊴ G N\cap P_p\trianglelefteq G N ∩ P p ⊴ G . As N ≠ 1 N\ne1 N = 1 , there exists at least one N ∩ P p ≠ { 1 } N\cap P_p\ne\{1\} N ∩ P p = { 1 } . But, N ∩ P p ≤ N N\cap P_p\le N N ∩ P p ≤ N . So by minimality of N N N , we get N ∩ P p = N N\cap P_p=N N ∩ P p = N and thus, N ≤ P p N\le P_p N ≤ P p . As the other Sylow subgroups of G G G centralize P p P_p P p , every normal subgroup of P p P_p P p is normal in G G G , and so N N N is a minimal normal subgroup of the finite p p p -group P p P_p P p . So, by Corollary 4.7 we get
N ≤ Ω 1 ( Z ( P p ) ) and ∣ N ∣ = p . N\le\Omega_1(Z(P_p))\quad\text{and}\quad|N|=p. N ≤ Ω 1 ( Z ( P p )) and ∣ N ∣ = p .
Hence, N N N is a one-dimensional F p \mathbb{F}_p F p -subspace of Ω 1 ( Z ( P p ) ) \Omega_1(Z(P_p)) Ω 1 ( Z ( P p )) .
Conversely, suppose L ≤ Ω 1 ( Z ( P p ) ) L\le\Omega_1(Z(P_p)) L ≤ Ω 1 ( Z ( P p )) is a one-dimensional subspace. Then, ∣ L ∣ = p |L|=p ∣ L ∣ = p . Now,
L ≤ Ω 1 ( Z ( P p ) ) ≤ Z ( P p ) ≤ Z ( G ) L\le\Omega_1(Z(P_p))\le Z(P_p)\le Z(G) L ≤ Ω 1 ( Z ( P p )) ≤ Z ( P p ) ≤ Z ( G )
implies that L ⊴ G L\trianglelefteq G L ⊴ G and as ∣ L ∣ = p |L|=p ∣ L ∣ = p , it follows L ∈ MinNorm ( G ) L\in\operatorname{MinNorm}(G) L ∈ MinNorm ( G ) . So, we have a correspondence,
MinNorm ( G ) = ∏ p ∣ ∣ G ∣ P ( Ω 1 ( Z ( P p ) ) ) , \operatorname{MinNorm}(G)=\prod_{p\mid|G|}\mathbb{P}(\Omega_1(Z(P_p))), MinNorm ( G ) = p ∣ ∣ G ∣ ∏ P ( Ω 1 ( Z ( P p ))) ,
where, P ( Ω 1 ( Z ( P p ) ) ) \mathbb{P}(\Omega_1(Z(P_p))) P ( Ω 1 ( Z ( P p ))) is the collection of all one-dimensional F p \mathbb{F}_p F p -subspaces of Ω 1 ( Z ( P p ) ) \Omega_1(Z(P_p)) Ω 1 ( Z ( P p )) . Hence, we get
Soc ( G ) = ∏ p ∣ ∣ G ∣ Ω 1 ( Z ( P p ) ) . \operatorname{Soc}(G)=\prod_{p\mid|G|}\Omega_1(Z(P_p)). Soc ( G ) = p ∣ ∣ G ∣ ∏ Ω 1 ( Z ( P p )) .
As Ω 1 ( Z ( P p ) ) ≤ Z ( G ) \Omega_1(Z(P_p))\le Z(G) Ω 1 ( Z ( P p )) ≤ Z ( G ) , it follows that every subgroup of Ω 1 ( Z ( P p ) ) \Omega_1(Z(P_p)) Ω 1 ( Z ( P p )) is normal in G G G and thus like before we get
[ 1 , Ω 1 ( Z ( P p ) ) ] N ( G ) = SubSp F p ( Ω 1 ( Z ( P p ) ) ) . [1,\Omega_1(Z(P_p))]_{\mathcal{N}(G)}=\operatorname{SubSp}_{\mathbb{F}_p}(\Omega_1(Z(P_p))). [ 1 , Ω 1 ( Z ( P p )) ] N ( G ) = SubSp F p ( Ω 1 ( Z ( P p ))) .
Hence,
λ G ( Ω 1 ( Z ( P p ) ) ) = dim F p Ω 1 ( Z ( P p ) ) . \lambda_G(\Omega_1(Z(P_p)))=\dim_{\mathbb{F}_p}\Omega_1(Z(P_p)). λ G ( Ω 1 ( Z ( P p ))) = dim F p Ω 1 ( Z ( P p )) .
Now, Soc ( G ) = ∏ p ∣ ∣ G ∣ Ω 1 ( Z ( P p ) ) \operatorname{Soc}(G)=\prod_{p\mid|G|}\Omega_1(Z(P_p)) Soc ( G ) = ∏ p ∣ ∣ G ∣ Ω 1 ( Z ( P p )) and as each Ω 1 ( Z ( P p ) ) \Omega_1(Z(P_p)) Ω 1 ( Z ( P p )) has p p p -power order, these factors have pairwise coprime orders, and every subgroup of Soc ( G ) ≤ Z ( G ) \operatorname{Soc}(G)\le Z(G) Soc ( G ) ≤ Z ( G ) is normal in G G G , so we get
[ 1 , Soc ( G ) ] N ( G ) ≅ ∏ p ∣ ∣ G ∣ [ 1 , Ω 1 ( Z ( P p ) ) ] N ( G ) . [1,\operatorname{Soc}(G)]_{\mathcal{N}(G)}\cong\prod_{p\mid|G|}[1,\Omega_1(Z(P_p))]_{\mathcal{N}(G)}. [ 1 , Soc ( G ) ] N ( G ) ≅ p ∣ ∣ G ∣ ∏ [ 1 , Ω 1 ( Z ( P p )) ] N ( G ) .
Therefore, we get
σ ( G ) = λ G ( Soc ( G ) ) = ∑ p ∣ ∣ G ∣ λ G ( Ω 1 ( Z ( P p ) ) ) = ∑ p ∣ ∣ G ∣ dim F p Ω 1 ( Z ( P p ) ) = ∑ p ∣ ∣ G ∣ dim F p { z ∈ Z ( P p ) : z p = 1 } = ρ ( G ) . \begin{aligned}
\sigma(G) &= \lambda_G(\operatorname{Soc}(G)) \\
&= \sum_{p\mid|G|} \lambda_G(\Omega_1(Z(P_p))) \\
&= \sum_{p\mid|G|} \dim_{\mathbb{F}_p}\Omega_1(Z(P_p)) \\
&= \sum_{p\mid|G|} \dim_{\mathbb{F}_p}\{z\in Z(P_p):z^p=1\} \\
&= \rho(G).
\end{aligned} σ ( G ) = λ G ( Soc ( G )) = p ∣ ∣ G ∣ ∑ λ G ( Ω 1 ( Z ( P p ))) = p ∣ ∣ G ∣ ∑ dim F p Ω 1 ( Z ( P p )) = p ∣ ∣ G ∣ ∑ dim F p { z ∈ Z ( P p ) : z p = 1 } = ρ ( G ) .
By Theorem 4.5 , we know that
G is M -full ⟺ σ ( G ) ≤ 2. G\text{ is }\mathcal{M}\text{-full}\Longleftrightarrow\sigma(G)\le2. G is M -full ⟺ σ ( G ) ≤ 2.
Hence, we conclude that G G G is M \mathcal{M} M -full if and only if ρ ( G ) ≤ 2 \rho(G)\le2 ρ ( G ) ≤ 2 .
We close this section by showing that the equivalence (1) ⇔ \Leftrightarrow ⇔ (3) of Theorem 4.5 has a purely lattice-theoretic form, which holds for arbitrary groups.
Let G G G be a fixed group. Then N ( G ) \mathcal{N}(G) N ( G ) , the set of normal subgroups of G G G , is a complete modular lattice by Lemma 3.1 , with meet the intersection and join the subgroup generated. Recall that a quantale is a complete lattice Q Q Q together with an associative multiplication ⋅ \cdot ⋅ which distributes over arbitrary joins in both variables, that is,
a ⋅ ⋁ i b i = ⋁ i ( a ⋅ b i ) and ( ⋁ i a i ) ⋅ b = ⋁ i ( a i ⋅ b ) a\cdot\bigvee_i b_i=\bigvee_i(a\cdot b_i)
\qquad\text{and}\qquad
\left(\bigvee_i a_i\right)\cdot b=\bigvee_i(a_i\cdot b) a ⋅ i ⋁ b i = i ⋁ ( a ⋅ b i ) and ( i ⋁ a i ) ⋅ b = i ⋁ ( a i ⋅ b )
for all a , b , a i , b i ∈ Q a,b,a_i,b_i\in Q a , b , a i , b i ∈ Q . A quantale is commutative if ⋅ \cdot ⋅ is commutative, and integral if the top element of Q Q Q is an identity for ⋅ \cdot ⋅ . The μ \mu μ -elements of [6 ] are defined in integral commutative quantales, but their definition only uses the meet, and the notions below only use the meet and the join. So they make sense in N ( G ) \mathcal{N}(G) N ( G ) , although N ( G ) \mathcal{N}(G) N ( G ) is in general not an integral commutative quantale: the commutator [ N , M ] [N,M] [ N , M ] is not associative (for G = S 3 G=S_3 G = S 3 , [ [ G , G ] , A 3 ] = 1 ≠ A 3 = [ G , [ G , A 3 ] ] [[G,G],A_3]=1\ne A_3=[G,[G,A_3]] [[ G , G ] , A 3 ] = 1 = A 3 = [ G , [ G , A 3 ]] ), the product N M NM N M is just the join, for which G G G is not an identity, and ∩ \cap ∩ does not distribute over joins (for G = C 2 × C 2 G=C_2\times C_2 G = C 2 × C 2 ). In this language, the M \mathcal{M} M -subgroups of G G G are precisely the μ \mu μ -elements of N ( G ) \mathcal{N}(G) N ( G ) . Following [6 , 13 ] , we use the following notion.
Definition 4.9. A family of non-trivial normal subgroups N 1 , … , N k N_1,\ldots,N_k N 1 , … , N k of G G G is called independent (also called join-independent ) if, for every i i i ,
N i ∩ ⋂ j ≠ i N j = 1. N_i\cap\bigcap_{j\ne i}N_j=1. N i ∩ j = i ⋂ N j = 1.
As the lattice N ( G ) \mathcal{N}(G) N ( G ) is modular, following [13 ] , in the terminology of Goldie and dual Goldie dimensions, we make the following definition.
Definition 4.10. The uniform dimension of N ( G ) \mathcal{N}(G) N ( G ) is
udim N ( G ) ∈ { 0 , 1 , 2 , … } ∪ { ∞ } , \operatorname{udim}\mathcal{N}(G)\in\{0,1,2,\ldots\}\cup\{\infty\}, udim N ( G ) ∈ { 0 , 1 , 2 , … } ∪ { ∞ } , defined to be the supremum of the cardinalities of independent families of non-trivial normal subgroups of G G G .
We say that N ( G ) \mathcal{N}(G) N ( G ) is uniform if udim N ( G ) ≤ 1 \operatorname{udim}\mathcal{N}(G)\le1 udim N ( G ) ≤ 1 , that is, every non-trivial normal subgroup of G G G is essential in G G G .
Theorem 4.11. The following conditions are equivalent for a group G G G .
(1) G G G is M \mathcal{M} M -full, i.e., N M ( G ) = N ( G ) \mathcal{N}^{\mathcal{M}}(G)=\mathcal{N}(G) N M ( G ) = N ( G ) .
(2) There are no non-trivial normal subgroups A , B , C A,B,C A , B , C of G G G such that
A ∩ B = 1 and A B ∩ C = 1. A \cap B = 1 \qquad\text{and} \qquad AB \cap C = 1. A ∩ B = 1 and A B ∩ C = 1.
(3) G G G contains no internal direct product A × B × C A \times B \times C A × B × C of three non-trivial normal subgroups of G G G .
(4) udim N ( G ) ≤ 2 \operatorname{udim}_{\mathcal{N}}(G) \le2 udim N ( G ) ≤ 2 .
Proof. We begin with (1) ⇒ \Rightarrow ⇒ (2), arguing contrapositively. Suppose that A , B , C A,B,C A , B , C are non-trivial normal subgroups of G G G such that
A ∩ B = 1 and A B ∩ C = 1. A \cap B = 1 \qquad\text{and} \qquad AB \cap C = 1. A ∩ B = 1 and A B ∩ C = 1.
Put E = A B E = AB E = A B . Since A A A and B B B are normal, E E E is again a normal subgroup of G G G . Moreover,
A ≤ E , B ≤ E , E ∩ C = 1. A \le E, \qquad B \le E, \qquad E \cap C = 1. A ≤ E , B ≤ E , E ∩ C = 1.
Consider the normal subgroups A C AC A C and B C BC B C . Since C ≤ A C ∩ B C C \le AC \cap BC C ≤ A C ∩ B C , we have A C ∩ B C ≠ 1 AC \cap BC \ne1 A C ∩ B C = 1 . Also,
A ≤ A C ∩ E and B ≤ B C ∩ E , A \le AC \cap E \qquad\text{and} \qquad B \le BC \cap E, A ≤ A C ∩ E and B ≤ B C ∩ E ,
so A C ∩ E ≠ 1 AC \cap E \ne1 A C ∩ E = 1 and B C ∩ E ≠ 1 BC \cap E \ne1 B C ∩ E = 1 .
Finally, in fact A C ∩ B C = C AC \cap BC = C A C ∩ B C = C . Indeed, if x ∈ A C ∩ B C x \in AC \cap BC x ∈ A C ∩ B C , write x = a c = b c ′ x = ac = bc' x = a c = b c ′ with a ∈ A a \in A a ∈ A , b ∈ B b \in B b ∈ B , and c , c ′ ∈ C c,c' \in C c , c ′ ∈ C . Then
a − 1 b = c c ′ − 1 ∈ A B ∩ C = 1 , a^{-1}b = cc'^{-1} \in AB \cap C = 1, a − 1 b = c c ′ − 1 ∈ A B ∩ C = 1 ,
whence a = b ∈ A ∩ B = 1 a = b \in A \cap B = 1 a = b ∈ A ∩ B = 1 , and therefore x = c = c ′ ∈ C x = c = c' \in C x = c = c ′ ∈ C . Consequently,
E ∩ ( A C ∩ B C ) = E ∩ C = 1. E \cap(AC \cap BC) = E \cap C = 1. E ∩ ( A C ∩ B C ) = E ∩ C = 1.
Thus E E E is not an M \mathcal{M} M -subgroup, and hence G G G is not M \mathcal{M} M -full. This proves (1) ⇒ \Rightarrow ⇒ (2) by contraposition.
For (2) ⇒ \Rightarrow ⇒ (1), suppose that E ⊴ G E \trianglelefteq G E ⊴ G is not an M \mathcal{M} M -subgroup. Then by Theorem 3.4 , E E E is neither essential nor relatively uniform in G G G . So there are non-trivial normal subgroups A , B , C A,B,C A , B , C of G G G such that A , B ≤ E A,B \le E A , B ≤ E , A ∩ B = 1 A \cap B = 1 A ∩ B = 1 and E ∩ C = 1 E \cap C = 1 E ∩ C = 1 . As A B ≤ E AB \le E A B ≤ E , A B ∩ C ≤ E ∩ C = 1 AB \cap C \le E \cap C = 1 A B ∩ C ≤ E ∩ C = 1 , and hence (2) fails.
For (2) ⇔ \Leftrightarrow ⇔ (3), observe that for non-trivial normal subgroups A , B , C A,B,C A , B , C of G G G , the product A B C ABC A B C is the internal direct product A × B × C A \times B \times C A × B × C if and only if A ∩ B = 1 A \cap B = 1 A ∩ B = 1 and A B ∩ C = 1 AB \cap C = 1 A B ∩ C = 1 .
For (2) ⇔ \Leftrightarrow ⇔ (4), observe that a subfamily of an independent family is again independent. So udim N ( G ) ≥ 3 \operatorname{udim}_{\mathcal{N}}(G) \ge3 udim N ( G ) ≥ 3 if and only if there are three independent non-trivial normal subgroups A , B , C A,B,C A , B , C of G G G . Now if A , B , C A,B,C A , B , C are independent, then A ∩ B ≤ A ∩ B C = 1 A \cap B \le A \cap BC = 1 A ∩ B ≤ A ∩ B C = 1 and A B ∩ C = 1 AB \cap C = 1 A B ∩ C = 1 . Conversely, let A ∩ B = 1 A \cap B = 1 A ∩ B = 1 and A B ∩ C = 1 AB \cap C = 1 A B ∩ C = 1 . If x = b c ∈ A ∩ B C x = bc \in A \cap BC x = b c ∈ A ∩ B C with b ∈ B b \in B b ∈ B and c ∈ C c \in C c ∈ C , then c = b − 1 x ∈ A B ∩ C = 1 c = b^{-1}x \in AB \cap C = 1 c = b − 1 x ∈ A B ∩ C = 1 , and so x = b ∈ A ∩ B = 1 x = b \in A \cap B = 1 x = b ∈ A ∩ B = 1 . Thus A ∩ B C = 1 A \cap BC = 1 A ∩ B C = 1 , and similarly B ∩ A C = 1 B \cap AC = 1 B ∩ A C = 1 . So A , B , C A,B,C A , B , C are independent. Hence (2) holds if and only if udim N ( G ) ≤ 2 \operatorname{udim}_{\mathcal{N}}(G) \le2 udim N ( G ) ≤ 2 . □ \square □
Corollary 4.12. If G G G is finite, then udim N ( G ) = σ ( G ) \operatorname{udim}_{\mathcal{N}}(G) = \sigma(G) udim N ( G ) = σ ( G ) . Hence, a finite group is M \mathcal{M} M -full if and only if σ ( G ) ≤ 2 \sigma(G) \le2 σ ( G ) ≤ 2 .
Proof. If G = 1 G = 1 G = 1 , then both sides are 0. So let G ≠ 1 G \ne1 G = 1 . By Proposition 4.4 , Soc ( G ) = B 1 × ⋯ × B b × ∏ p S p \operatorname{Soc}(G) = B_1 \times\cdots\times B_b \times\prod_p S_p Soc ( G ) = B 1 × ⋯ × B b × ∏ p S p , where
S p = ⨁ j S p , j S_p = \bigoplus_j S_{p,j} S p = j ⨁ S p , j
and each S p , j ≅ A n p , j S_{p,j} \cong A^{n_{p,j}} S p , j ≅ A n p , j is a direct sum of n p , j n_{p,j} n p , j simple submodules, which are minimal normal subgroups of G G G by Proposition 4.4 (1). So by Proposition 4.4 (3), there are minimal normal subgroups N 1 , … , N σ ( G ) N_1,\ldots,N_{\sigma(G)} N 1 , … , N σ ( G ) of G G G such that
Soc ( G ) = N 1 × ⋯ × N σ ( G ) . \operatorname{Soc}(G) = N_1 \times\cdots\times N_{\sigma(G)}. Soc ( G ) = N 1 × ⋯ × N σ ( G ) .
The factors in a direct product are independent, that is, N i ∩ ⋂ j ≠ i N j = 1 N_i \cap\bigcap_{j \ne i} N_j = 1 N i ∩ ⋂ j = i N j = 1 for every i i i . Thus { N 1 , … , N σ ( G ) } \{N_1,\ldots,N_{\sigma(G)}\} { N 1 , … , N σ ( G ) } is an independent family of non-trivial normal subgroups of G G G , and the inequality
udim N ( G ) ≥ σ ( G ) \operatorname{udim}_{\mathcal{N}}(G) \ge\sigma(G) udim N ( G ) ≥ σ ( G )
is immediate.
Conversely, let N 1 , … , N k N_1,\ldots,N_k N 1 , … , N k be independent non-trivial normal subgroups of G G G , so that N i ∩ ⋂ j ≠ i N j = 1 N_i \cap\bigcap_{j \ne i} N_j = 1 N i ∩ ⋂ j = i N j = 1 for every i i i . For each i i i , we choose a minimal normal subgroup M i M_i M i of G G G with M i ≤ N i M_i \le N_i M i ≤ N i ; this is possible because $G is finite and N i ≠ 1 N_i \ne1 N i = 1 (for instance, take M i M_i M i to be a non-trivial normal subgroup of G G G of least order contained in N i N_i N i ). Since M 1 ⋯ M i − 1 ≤ N 1 ⋯ N i − 1 ≤ ∏ j ≠ i N j M_1\cdots M_{i-1} \le N_1\cdots N_{i-1} \le\prod_{j\ne i}N_j M 1 ⋯ M i − 1 ≤ N 1 ⋯ N i − 1 ≤ ∏ j = i N j , we clearly have
M i ∩ M 1 ⋯ M i − 1 ≤ N i ∩ ∏ j ≠ i N j = 1. M_i \cap M_1\cdots M_{i-1} \le N_i \cap\prod_{j\ne i}N_j = 1. M i ∩ M 1 ⋯ M i − 1 ≤ N i ∩ j = i ∏ N j = 1.
As M i ≠ 1 M_i \ne1 M i = 1 , it follows that M i ≰ M 1 ⋯ M i − 1 M_i \nleq M_1\cdots M_{i-1} M i ≰ M 1 ⋯ M i − 1 , and hence M 1 ⋯ M i − 1 < M 1 ⋯ M i M_1\cdots M_{i-1} < M_1\cdots M_i M 1 ⋯ M i − 1 < M 1 ⋯ M i . Moreover, each M i M_i M i , being a minimal normal subgroup of G G G , is contained in Soc ( G ) \operatorname{Soc}(G) Soc ( G ) . Therefore
1 < M 1 < M 1 M 2 < ⋯ < M 1 M 2 ⋯ M k ≤ Soc ( G ) 1 < M_1 < M_1M_2 < \cdots< M_1M_2\cdots M_k \le\operatorname{Soc}(G) 1 < M 1 < M 1 M 2 < ⋯ < M 1 M 2 ⋯ M k ≤ Soc ( G )
is a chain of length k k k in [ 1 , Soc ( G ) ] N ( G ) [1,\operatorname{Soc}(G)]_{\mathcal{N}(G)} [ 1 , Soc ( G ) ] N ( G ) , and so k ≤ λ G ( Soc ( G ) ) = σ ( G ) k \le\lambda_G(\operatorname{Soc}(G)) = \sigma(G) k ≤ λ G ( Soc ( G )) = σ ( G ) . As { N 1 , … , N k } \{N_1,\ldots,N_k\} { N 1 , … , N k } was an arbitrary independent family of non-trivial normal subgroups of G G G , we obtain
udim N ( G ) ≤ σ ( G ) . \operatorname{udim}_{\mathcal{N}}(G) \le\sigma(G). udim N ( G ) ≤ σ ( G ) .
Hence udim N ( G ) = σ ( G ) \operatorname{udim}_{\mathcal{N}}(G) = \sigma(G) udim N ( G ) = σ ( G ) . Since a group is M \mathcal{M} M -full if and only if udim N ( G ) ≤ 2 \operatorname{udim}_{\mathcal{N}}(G) \le2 udim N ( G ) ≤ 2 (cf. Theorem 4.11 ), the last assertion follows. □
Direct products and diagonal atoms
In a direct product the normal subgroups are not, in general, products of normal subgroups of the factors, and the minimal normal subgroups need not be coordinate subgroups. The deviation is caused entirely by central elements of prime order, which produce diagonal minimal normal subgroups. The following theorem, the first part of Theorem C, makes this precise.
Theorem 5.1. Let G = G 1 × ⋯ × G t G=G_1\times\cdots\times G_t G = G 1 × ⋯ × G t be a finite direct product. Its minimal normal subgroups are precisely
the coordinate subgroups A ^ \widehat{A} A , where A ∈ MinNorm ( G i ) A\in\operatorname{MinNorm}(G_i) A ∈ MinNorm ( G i ) and A ≰ Z ( G i ) A\nleq Z(G_i) A ≰ Z ( G i ) ;
the one-dimensional F p \mathbb{F}_p F p -subspaces of V p V_p V p , as p p p ranges over the prime divisors of ∣ G ∣ |G| ∣ G ∣ .
Consequently,
Soc ( G ) = Soc ( G 1 ) × ⋯ × Soc ( G t ) . \operatorname{Soc}(G)=\operatorname{Soc}(G_1)\times\cdots\times\operatorname{Soc}(G_t). Soc ( G ) = Soc ( G 1 ) × ⋯ × Soc ( G t ) . A line in V p V_p V p is non-coordinate precisely when its projection is non-zero in at least two factors.
Proof. Suppose, N ∈ MinNorm ( G ) N\in\operatorname{MinNorm}(G) N ∈ MinNorm ( G ) . Then, we have two cases:
Case 1: [ N , G i ∗ ] ≠ 1 [N,G_i^*]\ne1 [ N , G i ∗ ] = 1 for some i i i . As, N ⊴ G N\trianglelefteq G N ⊴ G and G i ∗ ⊴ G G_i^*\trianglelefteq G G i ∗ ⊴ G , so [ N , G i ∗ ] ≤ N ∩ G i ∗ [N,G_i^*]\le N\cap G_i^* [ N , G i ∗ ] ≤ N ∩ G i ∗ . Hence, N ∩ G i ∗ ≠ 1 N\cap G_i^*\ne1 N ∩ G i ∗ = 1 . But N ∩ G i ∗ ⊴ G N\cap G_i^*\trianglelefteq G N ∩ G i ∗ ⊴ G , and so by minimality of N N N , we get N ∩ G i ∗ = N N\cap G_i^*=N N ∩ G i ∗ = N , and hence N ≤ G i ∗ N\le G_i^* N ≤ G i ∗ . Now, let A ≤ G i A\le G_i A ≤ G i such that N = A ^ N=\widehat{A} N = A . For any g i ∈ G i g_i\in G_i g i ∈ G i and a ∈ A a\in A a ∈ A , we have
g i ^ a ^ g i ^ − 1 ∈ N = A ^ \widehat{g_i}\widehat{a}\widehat{g_i}^{-1}\in N=\widehat{A} g i a g i − 1 ∈ N = A
as N ⊴ G N\trianglelefteq G N ⊴ G . Thus, g i a g i − 1 ∈ A g_iag_i^{-1}\in A g i a g i − 1 ∈ A . Hence, A ⊴ G i A\trianglelefteq G_i A ⊴ G i . Now, let 1 ≠ C ⊴ G i 1\ne C\trianglelefteq G_i 1 = C ⊴ G i such that C ≤ A C\le A C ≤ A . Then, C ^ ⊴ G \widehat{C}\trianglelefteq G C ⊴ G , as C ⊴ G i C\trianglelefteq G_i C ⊴ G i and C ^ \widehat{C} C is centralized by G j ∗ G_j^* G j ∗ for j ≠ i j\ne i j = i , and C ^ ≤ A ^ = N \widehat{C}\le\widehat{A}=N C ≤ A = N . So, by minimality of N N N , we get C ^ = N = A ^ \widehat{C}=N=\widehat{A} C = N = A and hence, C = A C=A C = A . Thus, A ∈ MinNorm ( G i ) A\in\operatorname{MinNorm}(G_i) A ∈ MinNorm ( G i ) . As [ N , G i ∗ ] ≠ 1 [N,G_i^*]\ne1 [ N , G i ∗ ] = 1 , it implies that [ A ^ , G i ∗ ] ≠ 1 [\widehat{A},G_i^*]\ne1 [ A , G i ∗ ] = 1 and hence [ A , G i ] ≠ 1 [A,G_i]\ne1 [ A , G i ] = 1 . So, A ≰ Z ( G i ) A\nleq Z(G_i) A ≰ Z ( G i ) .
Case 2: [ N , G i ∗ ] = 1 [N,G_i^*]=1 [ N , G i ∗ ] = 1 for all i i i . As G = G 1 ∗ G 2 ∗ ⋯ G t ∗ G=G_1^*G_2^*\cdots G_t^* G = G 1 ∗ G 2 ∗ ⋯ G t ∗ and [ N , G i ∗ ] = 1 [N,G_i^*]=1 [ N , G i ∗ ] = 1 for all i i i , it follows that [ N , G ] = 1 [N,G]=1 [ N , G ] = 1 . Hence, N ≤ Z ( G ) N\le Z(G) N ≤ Z ( G ) . So, N N N is a finite abelian group. As N N N is minimal normal, it follows that ∣ N ∣ = p |N|=p ∣ N ∣ = p for some prime p p p . Hence, N N N is a one-dimensional F p \mathbb{F}_p F p -vector space. Now, Z ( G ) = ∏ i = 1 t Z ( G i ) Z(G)=\prod_{i=1}^{t}Z(G_i) Z ( G ) = ∏ i = 1 t Z ( G i ) and N ≤ Z ( G ) N\le Z(G) N ≤ Z ( G ) , ∣ N ∣ = p |N|=p ∣ N ∣ = p . So, it follows that
N ≤ ∏ i = 1 t V i , p = V p . N\le\prod_{i=1}^{t}V_{i,p}=V_p. N ≤ i = 1 ∏ t V i , p = V p .
So, N N N is a one-dimensional F p \mathbb{F}_p F p -subspace of V p V_p V p .
Conversely, suppose that A ∈ MinNorm ( G i ) A \in\operatorname{MinNorm}(G_i) A ∈ MinNorm ( G i ) such that A ≰ Z ( G i ) A \nleq Z(G_i) A ≰ Z ( G i ) . Then, A ≤ G i A \leq G_i A ≤ G i and hence A ^ ≤ G i ∗ \hat{A} \leq G_i^* A ^ ≤ G i ∗ . Now, let g ∈ G g \in G g ∈ G and a ^ ∈ A ^ \hat{a} \in\hat{A} a ^ ∈ A ^ . Then, as A ⊴ G i A \trianglelefteq G_i A ⊴ G i , we get g a ^ g − 1 ∈ A ^ g\hat{a}g^{-1} \in\hat{A} g a ^ g − 1 ∈ A ^ . Thus, A ^ ⊴ G \hat{A} \trianglelefteq G A ^ ⊴ G . Now, let 1 ≠ N ⊴ G 1 \ne N \trianglelefteq G 1 = N ⊴ G be a normal subgroup such that N ≤ A ^ N \leq\hat{A} N ≤ A ^ . Then N = C ^ N = \hat{C} N = C ^ for some 1 ≠ C ≤ A 1 \ne C \leq A 1 = C ≤ A , and C ⊴ G i C \trianglelefteq G_i C ⊴ G i as N ⊴ G N \trianglelefteq G N ⊴ G . But as A ∈ MinNorm ( G i ) A \in\operatorname{MinNorm}(G_i) A ∈ MinNorm ( G i ) , we must have C = A C = A C = A and hence, N = A ^ N = \hat{A} N = A ^ . Thus, A ^ ∈ MinNorm ( G ) \hat{A} \in\operatorname{MinNorm}(G) A ^ ∈ MinNorm ( G ) .
Now suppose L ≤ V p L \leq V_p L ≤ V p is a one-dimensional F p \mathbb{F}_p F p -subspace. Then, ∣ L ∣ = p |L| = p ∣ L ∣ = p and L ≤ V p ≤ Z ( G ) L \leq V_p \leq Z(G) L ≤ V p ≤ Z ( G ) . So, L ⊴ G L \trianglelefteq G L ⊴ G , and as a group of prime order, it has no proper non-trivial subgroups, it follows that L ∈ MinNorm ( G ) L \in\operatorname{MinNorm}(G) L ∈ MinNorm ( G ) . Thus, the minimal normal subgroups of G G G are precisely the coordinate subgroups A ^ \hat{A} A ^ , where A ∈ MinNorm ( G i ) A \in\operatorname{MinNorm}(G_i) A ∈ MinNorm ( G i ) and A ≰ Z ( G i ) A \nleq Z(G_i) A ≰ Z ( G i ) along with the one-dimensional F p \mathbb{F}_p F p -subspaces of V p V_p V p , where p p p varies over the prime divisors of ∣ G ∣ |G| ∣ G ∣ .
Now, for each i i i , we have
Soc ( G i ) = ⟨ A : A ∈ MinNorm ( G i ) ⟩ . \operatorname{Soc}(G_i)=\langle A:A\in\operatorname{MinNorm}(G_i)\rangle. Soc ( G i ) = ⟨ A : A ∈ MinNorm ( G i )⟩ .
So, the coordinate copy of Soc ( G i ) \operatorname{Soc}(G_i) Soc ( G i ) , denoted Soc ∗ ( G i ) \operatorname{Soc}^*(G_i) Soc ∗ ( G i ) , is generated by the coordinate minimal normal subgroups, A ^ \hat{A} A ^ where A ^ ∈ MinNorm ( G i ∗ ) \hat{A} \in\operatorname{MinNorm}(G_i^*) A ^ ∈ MinNorm ( G i ∗ ) . But, we just now proved that every A ∈ MinNorm ( G i ) A \in\operatorname{MinNorm}(G_i) A ∈ MinNorm ( G i ) gives us A ^ ∈ MinNorm ( G ) \hat{A} \in\operatorname{MinNorm}(G) A ^ ∈ MinNorm ( G ) , and the argument above does not use A ≰ Z ( G i ) A \nleq Z(G_i) A ≰ Z ( G i ) . Therefore,
Soc ∗ ( G 1 ) Soc ∗ ( G 2 ) ⋯ Soc ∗ ( G t ) ≤ Soc ( G ) . \operatorname{Soc}^*(G_1)\operatorname{Soc}^*(G_2)\cdots\operatorname{Soc}^*(G_t)\leq\operatorname{Soc}(G). Soc ∗ ( G 1 ) Soc ∗ ( G 2 ) ⋯ Soc ∗ ( G t ) ≤ Soc ( G ) .
As Soc ∗ ( G i ) \operatorname{Soc}^*(G_i) Soc ∗ ( G i ) is the canonical subgroup, we hence get
Soc ( G 1 ) × ⋯ × Soc ( G t ) ≤ Soc ( G ) . \operatorname{Soc}(G_1)\times\cdots\times\operatorname{Soc}(G_t)\leq\operatorname{Soc}(G). Soc ( G 1 ) × ⋯ × Soc ( G t ) ≤ Soc ( G ) .
Now, for any N ∈ MinNorm ( G ) N \in\operatorname{MinNorm}(G) N ∈ MinNorm ( G ) , either N = A ^ N = \hat{A} N = A ^ for some A ∈ MinNorm ( G i ) A \in\operatorname{MinNorm}(G_i) A ∈ MinNorm ( G i ) , A ≰ Z ( G i ) A \nleq Z(G_i) A ≰ Z ( G i ) or N ≤ V p N \leq V_p N ≤ V p is a one-dimensional F p \mathbb{F}_p F p -subspace of V p V_p V p . Now, if N = A ^ N = \hat{A} N = A ^ , then N ∈ MinNorm ( G i ∗ ) N \in\operatorname{MinNorm}(G_i^*) N ∈ MinNorm ( G i ∗ ) . So, if we canonically identify, G i G_i G i with G i ∗ G_i^* G i ∗ , we then get N ∈ MinNorm ( G i ) N \in\operatorname{MinNorm}(G_i) N ∈ MinNorm ( G i ) , and hence N ≤ Soc ∗ ( G i ) ≤ ∏ j = 1 t Soc ( G j ) N \leq\operatorname{Soc}^*(G_i) \leq\prod_{j=1}^{t}\operatorname{Soc}(G_j) N ≤ Soc ∗ ( G i ) ≤ ∏ j = 1 t Soc ( G j ) . Now suppose, N ≤ V p N \leq V_p N ≤ V p is a one-dimensional F p \mathbb{F}_p F p -subspace. Recall that,
V p = ∏ i = 1 t V i , p . V_p=\prod_{i=1}^{t}V_{i,p}. V p = i = 1 ∏ t V i , p .
So, for any 1 ≠ x ∈ V i , p 1 \ne x \in V_{i,p} 1 = x ∈ V i , p , ⟨ x ⟩ \langle x\rangle ⟨ x ⟩ has order p p p . Also, as ⟨ x ⟩ ≤ V i , p ≤ Z ( G i ) \langle x\rangle\leq V_{i,p} \leq Z(G_i) ⟨ x ⟩ ≤ V i , p ≤ Z ( G i ) , we get ⟨ x ⟩ ⊴ G i \langle x\rangle\trianglelefteq G_i ⟨ x ⟩ ⊴ G i . So, ⟨ x ⟩ ∈ MinNorm ( G i ) \langle x\rangle\in\operatorname{MinNorm}(G_i) ⟨ x ⟩ ∈ MinNorm ( G i ) . Hence, we get ⟨ x ⟩ ≤ Soc ( G i ) \langle x\rangle\leq\operatorname{Soc}(G_i) ⟨ x ⟩ ≤ Soc ( G i ) and thus x ∈ Soc ( G i ) x \in\operatorname{Soc}(G_i) x ∈ Soc ( G i ) . So, V i , p ≤ Soc ( G i ) V_{i,p} \leq\operatorname{Soc}(G_i) V i , p ≤ Soc ( G i ) . Hence,
N ≤ V p ≤ ∏ i = 1 t Soc ( G i ) . N\leq V_p\leq\prod_{i=1}^{t}\operatorname{Soc}(G_i). N ≤ V p ≤ i = 1 ∏ t Soc ( G i ) .
As N N N was an arbitrary minimal normal subgroup of G G G , it follows then
Soc ( G ) ≤ Soc ( G 1 ) × ⋯ × Soc ( G t ) . \operatorname{Soc}(G)\leq\operatorname{Soc}(G_1)\times\cdots\times\operatorname{Soc}(G_t). Soc ( G ) ≤ Soc ( G 1 ) × ⋯ × Soc ( G t ) .
Thus, Soc ( G ) = Soc ( G 1 ) × ⋯ × Soc ( G t ) \operatorname{Soc}(G)=\operatorname{Soc}(G_1)\times\cdots\times\operatorname{Soc}(G_t) Soc ( G ) = Soc ( G 1 ) × ⋯ × Soc ( G t ) . Now, let L L L be a line in V p V_p V p . Then, L L L is a one-dimensional F p \mathbb{F}_p F p -subspace of V p V_p V p . As V p = ∏ i = 1 t V i , p V_p=\prod_{i=1}^{t}V_{i,p} V p = ∏ i = 1 t V i , p , we have the projections:
π i : V p → V i , p \pi_i:V_p\to V_{i,p} π i : V p → V i , p
( v 1 , … , v t ) ⟼ v i . (v_1,\ldots,v_t)\longmapsto v_i. ( v 1 , … , v t ) ⟼ v i .
Now, suppose, 0 ≠ v = ( v 1 , … , v t ) ∈ L 0 \ne v=(v_1,\ldots,v_t)\in L 0 = v = ( v 1 , … , v t ) ∈ L . As L = F p v L=\mathbb{F}_p v L = F p v , so
L ≤ V ^ i , p ⟺ π i ( v ) ≠ 0 and π j ( v ) = 0 for all j ≠ i . L\leq\widehat{V}_{i,p}\Longleftrightarrow\pi_i(v)\ne0\text{ and }\pi_j(v)=0\text{ for all }j\ne i. L ≤ V i , p ⟺ π i ( v ) = 0 and π j ( v ) = 0 for all j = i .
Thus, L L L is coordinate in V p V_p V p if and only if exactly one of its projections is non-zero. Hence, as v ≠ 0 v\ne0 v = 0 , we get that L L L is non-coordinate in V p V_p V p if and only if at least two distinct projections are non-zero. □
As a consequence, the number of minimal normal subgroups below a given normal subgroup of a direct product, and hence its M \mathcal{M} M -status, can be computed from the factors and from linear algebra over the prime fields.
Corollary 5.2. Let G 1 , … , G t G_1,\ldots,G_t G 1 , … , G t be finite groups, put G = G 1 × ⋯ × G t G=G_1\times\cdots\times G_t G = G 1 × ⋯ × G t , and let H ⊴ G H\trianglelefteq G H ⊴ G . Define
a ( H ) = # { ( i , A ) : A ∈ MinNorm ( G i ) , A ≰ Z ( G i ) , A ^ ≤ H } a(H)=\#\{(i,A):A\in\operatorname{MinNorm}(G_i),\ A\nleq Z(G_i),\ \widehat{A}\leq H\} a ( H ) = # {( i , A ) : A ∈ MinNorm ( G i ) , A ≰ Z ( G i ) , A ≤ H } and
d p ( H ) = dim F p ( H ∩ V p ) . d_p(H)=\dim_{\mathbb{F}_p}(H\cap V_p). d p ( H ) = dim F p ( H ∩ V p ) . Then the number of minimal normal subgroups of G G G contained in H H H is
a ( H ) + ∑ p p d p ( H ) − 1 p − 1 . a(H)+\sum_p\frac{p^{d_p(H)}-1}{p-1}. a ( H ) + p ∑ p − 1 p d p ( H ) − 1 . In particular, H H H is an M \mathcal{M} M -subgroup if and only if H = 1 H=1 H = 1 , the displayed number is 1 1 1 , or Soc ( G 1 ) × ⋯ × Soc ( G t ) ≤ H \operatorname{Soc}(G_1)\times\cdots\times\operatorname{Soc}(G_t)\leq H Soc ( G 1 ) × ⋯ × Soc ( G t ) ≤ H .
Proof. Suppose, N ∈ MinNorm ( G ) N\in\operatorname{MinNorm}(G) N ∈ MinNorm ( G ) such that N ≤ H N\leq H N ≤ H . Then by Theorem 5.1 , N = A ^ N=\widehat{A} N = A for some A ∈ MinNorm ( G i ) A\in\operatorname{MinNorm}(G_i) A ∈ MinNorm ( G i ) such that A ≰ Z ( G i ) A\nleq Z(G_i) A ≰ Z ( G i ) or N ≤ V p N\leq V_p N ≤ V p is a one-dimensional F p \mathbb{F}_p F p -subspace. Now, if N = A ^ N=\widehat{A} N = A , then ( i , A ) ∈ { ( i , A ) : A ∈ MinNorm ( G i ) , A ≰ Z ( G i ) , A ^ ≤ H } (i,A)\in\{(i,A):A\in\operatorname{MinNorm}(G_i),\ A\nleq Z(G_i),\ \widehat{A}\leq H\} ( i , A ) ∈ {( i , A ) : A ∈ MinNorm ( G i ) , A ≰ Z ( G i ) , A ≤ H } .
Now, if N ≤ V p N\leq V_p N ≤ V p is a one-dimensional F p \mathbb{F}_p F p -subspace then, N ≤ H ∩ V p N\leq H\cap V_p N ≤ H ∩ V p is a one-dimensional F p \mathbb{F}_p F p -subspace as H ∩ V p H\cap V_p H ∩ V p is an F p \mathbb{F}_p F p -subspace of V p V_p V p . Conversely, by Theorem 5.1 , every such A ^ \widehat{A} A and every one-dimensional F p \mathbb{F}_p F p -subspace of H ∩ V p H\cap V_p H ∩ V p is a minimal normal subgroup of G G G contained in H H H . Moreover, distinct pairs ( i , A ) (i,A) ( i , A ) give distinct A ^ \widehat{A} A , as A ^ ≤ G i ∗ \widehat{A}\leq G_i^* A ≤ G i ∗ and G i ∗ ∩ G j ∗ = 1 G_i^*\cap G_j^*=1 G i ∗ ∩ G j ∗ = 1 for i ≠ j i\neq j i = j . Also, no A ^ \widehat{A} A lies in any V p V_p V p , as A ^ ≰ Z ( G ) \widehat{A}\nleq Z(G) A ≰ Z ( G ) while V p ≤ Z ( G ) V_p\leq Z(G) V p ≤ Z ( G ) , and the subspaces of V p V_p V p for different primes p p p are different. So, no minimal normal subgroup is counted twice. Now, the size of the set { ( i , A ) : A ∈ MinNorm ( G i ) , A ≰ Z ( G i ) , A ^ ≤ H } \{(i,A):A\in\operatorname{MinNorm}(G_i),\ A\nleq Z(G_i),\ \widehat{A}\leq H\} {( i , A ) : A ∈ MinNorm ( G i ) , A ≰ Z ( G i ) , A ≤ H } is given as:
a ( H ) = # { ( i , A ) : A ∈ MinNorm ( G i ) , A ≰ Z ( G i ) , A ^ ≤ H } . a(H)=\#\{(i,A):A\in\operatorname{MinNorm}(G_i),\ A\nleq Z(G_i),\ \widehat{A}\leq H\}. a ( H ) = # {( i , A ) : A ∈ MinNorm ( G i ) , A ≰ Z ( G i ) , A ≤ H } .
Also, the number of one-dimensional F p \mathbb{F}_p F p -subspaces of H ∩ V p H\cap V_p H ∩ V p is given by,
p dim F p ( H ∩ V p ) − 1 p − 1 = p d p ( H ) − 1 p − 1 . \frac{p^{\dim_{\mathbb{F}_p}(H\cap V_p)}-1}{p-1}=\frac{p^{d_p(H)}-1}{p-1}. p − 1 p d i m F p ( H ∩ V p ) − 1 = p − 1 p d p ( H ) − 1 .
Hence, the total number of minimal normal subgroups of G G G contained in H H H is
a ( H ) + ∑ p p d p ( H ) − 1 p − 1 , a(H)+\sum_p\frac{p^{d_p(H)}-1}{p-1}, a ( H ) + p ∑ p − 1 p d p ( H ) − 1 ,
where p p p ranges over all prime divisors of ∣ G ∣ |G| ∣ G ∣ .
Now, as G G G is finite, by Corollary 3.8, we have
H is an M -subgroup ⟺ H = 1 , ∣ supp ( H ) ∣ = 1 , Soc ( G ) ≤ H . H\text{ is an }\mathcal{M}\text{-subgroup}\Longleftrightarrow H=1,\ |\operatorname{supp}(H)|=1,\ \operatorname{Soc}(G)\leq H. H is an M -subgroup ⟺ H = 1 , ∣ supp ( H ) ∣ = 1 , Soc ( G ) ≤ H .
But,
∣ supp ( H ) ∣ = # { N ∈ MinNorm ( G ) : N ≤ H } = a ( H ) + ∑ p p d p ( H ) − 1 p − 1 . |\operatorname{supp}(H)|=\#\{N\in\operatorname{MinNorm}(G):N\leq H\}=a(H)+\sum_p\frac{p^{d_p(H)}-1}{p-1}. ∣ supp ( H ) ∣ = # { N ∈ MinNorm ( G ) : N ≤ H } = a ( H ) + p ∑ p − 1 p d p ( H ) − 1 .
Also, by Theorem 5.1 we have Soc ( G ) = Soc ( G 1 ) × ⋯ × Soc ( G t ) \operatorname{Soc}(G)=\operatorname{Soc}(G_1)\times\cdots\times\operatorname{Soc}(G_t) Soc ( G ) = Soc ( G 1 ) × ⋯ × Soc ( G t ) . Hence, H H H is an M \mathcal{M} M -subgroup if and only if
H = 1 , a ( H ) + ∑ p p d p ( H ) − 1 p − 1 = 1 or Soc ( G 1 ) × ⋯ × Soc ( G t ) ≤ H . H=1,\ a(H)+\sum_p\frac{p^{d_p(H)}-1}{p-1}=1\text{ or }\operatorname{Soc}(G_1)\times\cdots\times\operatorname{Soc}(G_t)\leq H. H = 1 , a ( H ) + p ∑ p − 1 p d p ( H ) − 1 = 1 or Soc ( G 1 ) × ⋯ × Soc ( G t ) ≤ H .
Combining Theorem 5.1 with Theorem 4.5 we obtain the second assertion of Theorem C: a non-trivial direct product is M \mathcal{M} M -full only in the smallest possible case.
Corollary 5.3. Let G 1 , … , G t G_1,\ldots,G_t G 1 , … , G t be non-trivial finite groups. If t ≥ 2 t \ge2 t ≥ 2 , then G 1 × ⋯ × G t G_1 \times\cdots\times G_t G 1 × ⋯ × G t is M \mathcal{M} M -full if and only if t = 2 t=2 t = 2 and both G 1 G_1 G 1 and G 2 G_2 G 2 are monolithic.
Proof. Suppose, G = G 1 × ⋯ × G t G=G_1 \times\cdots\times G_t G = G 1 × ⋯ × G t is M \mathcal{M} M -full. Then, by Theorem 4.5 , σ ( G ) ≤ 2 \sigma(G) \le2 σ ( G ) ≤ 2 . Also, by Theorem 5.1 , we have
Soc ( G ) = Soc ( G 1 ) × ⋯ × Soc ( G t ) . \operatorname{Soc}(G)=\operatorname{Soc}(G_1)\times\cdots\times\operatorname{Soc}(G_t). Soc ( G ) = Soc ( G 1 ) × ⋯ × Soc ( G t ) .
Now, for all i ∈ { 1 , … , t } i \in\{1,\ldots,t\} i ∈ { 1 , … , t } , let A i ∈ MinNorm ( G i ) A_i \in\operatorname{MinNorm}(G_i) A i ∈ MinNorm ( G i ) . Then, A i ^ ≤ G i ∗ \widehat{A_i} \le G_i^* A i ≤ G i ∗ and A i ^ ⊴ G \widehat{A_i} \trianglelefteq G A i ⊴ G . Now, suppose without loss of generality i < j i<j i < j . If A i ^ = { ( 1 , … , a i , … , 1 ) : a i ∈ A i } \widehat{A_i}=\{(1,\ldots,a_i,\ldots,1):a_i\in A_i\} A i = {( 1 , … , a i , … , 1 ) : a i ∈ A i } and A j ^ = { ( 1 , … , a j , … , 1 ) : a j ∈ A j } \widehat{A_j}=\{(1,\ldots,a_j,\ldots,1):a_j\in A_j\} A j = {( 1 , … , a j , … , 1 ) : a j ∈ A j } , then,
( 1 , … , a i , … , 1 ) ( 1 , … , a j , … , 1 ) = ( 1 , … , a i , … , a j , … , 1 ) = ( 1 , … , a j , … , 1 ) ( 1 , … , a i , … , 1 ) . (1,\ldots,a_i,\ldots,1)(1,\ldots,a_j,\ldots,1)=(1,\ldots,a_i,\ldots,a_j,\ldots,1)=(1,\ldots,a_j,\ldots,1)(1,\ldots,a_i,\ldots,1). ( 1 , … , a i , … , 1 ) ( 1 , … , a j , … , 1 ) = ( 1 , … , a i , … , a j , … , 1 ) = ( 1 , … , a j , … , 1 ) ( 1 , … , a i , … , 1 ) .
So, each distinct coordinate copy commutes. Hence, the subgroups, A 1 ^ \widehat{A_1} A 1 , A 1 ^ A 2 ^ \widehat{A_1}\widehat{A_2} A 1 A 2 , … \ldots … , A 1 ^ A 2 ^ ⋯ A t ^ \widehat{A_1}\widehat{A_2}\cdots\widehat{A_t} A 1 A 2 ⋯ A t give us a chain,
1 < A 1 ^ < A 1 ^ A 2 ^ < ⋯ < A 1 ^ A 2 ^ ⋯ A t ^ ≤ Soc ( G ) . 1<\widehat{A_1}<\widehat{A_1}\widehat{A_2}<\cdots<\widehat{A_1}\widehat{A_2}\cdots\widehat{A_t}\le\operatorname{Soc}(G). 1 < A 1 < A 1 A 2 < ⋯ < A 1 A 2 ⋯ A t ≤ Soc ( G ) .
This is because A i ≤ Soc ( G i ) A_i \le\operatorname{Soc}(G_i) A i ≤ Soc ( G i ) . So identifying with their coordinate copies we get, A i ^ ≤ Soc ∗ ( G i ) \widehat{A_i}\le\operatorname{Soc}^*(G_i) A i ≤ Soc ∗ ( G i ) . So, A i ^ ≤ Soc ( G ) \widehat{A_i}\le\operatorname{Soc}(G) A i ≤ Soc ( G ) . Each inclusion is strict, as 1 ≠ A k + 1 ^ ≤ G k + 1 ∗ 1\ne\widehat{A_{k+1}}\le G_{k+1}^* 1 = A k + 1 ≤ G k + 1 ∗ , while A 1 ^ ⋯ A k ^ ≤ G 1 ∗ ⋯ G k ∗ \widehat{A_1}\cdots\widehat{A_k}\le G_1^*\cdots G_k^* A 1 ⋯ A k ≤ G 1 ∗ ⋯ G k ∗ meets G k + 1 ∗ G_{k+1}^* G k + 1 ∗ trivially. Thus, σ ( G ) ≥ t \sigma(G)\ge t σ ( G ) ≥ t . But, σ ( G ) ≤ 2 \sigma(G)\le2 σ ( G ) ≤ 2 and so t ≤ 2 t\le2 t ≤ 2 . As t ≥ 2 t\ge2 t ≥ 2 , it follows t = 2 t=2 t = 2 . Hence, G = G 1 × G 2 G=G_1\times G_2 G = G 1 × G 2 and so by Theorem 5.1 Soc ( G ) = Soc ( G 1 ) × Soc ( G 2 ) \operatorname{Soc}(G)=\operatorname{Soc}(G_1)\times\operatorname{Soc}(G_2) Soc ( G ) = Soc ( G 1 ) × Soc ( G 2 ) . Now, suppose, G 1 G_1 G 1 is not monolithic. Then, there exist distinct A , A ′ ∈ MinNorm ( G 1 ) A,A'\in\operatorname{MinNorm}(G_1) A , A ′ ∈ MinNorm ( G 1 ) . As G 2 ≠ 1 G_2\ne1 G 2 = 1 , let B ∈ MinNorm ( G 2 ) B\in\operatorname{MinNorm}(G_2) B ∈ MinNorm ( G 2 ) . Now, A ^ , A ′ ^ , B ^ \widehat{A},\widehat{A'},\widehat{B} A , A ′ , B are three distinct normal subgroups of G G G . As A ∩ A ′ = 1 A\cap A'=1 A ∩ A ′ = 1 by Lemma 3.6 , and 1 ≠ B ^ ≤ G 2 ∗ 1\ne\widehat{B}\le G_2^* 1 = B ≤ G 2 ∗ while A ^ A ′ ^ ≤ G 1 ∗ \widehat{A}\widehat{A'}\le G_1^* A A ′ ≤ G 1 ∗ , the inclusions below are strict. So we have a chain,
1 < A ^ < A ^ A ′ ^ < A ^ A ′ ^ B ^ ≤ Soc ( G ) . 1<\widehat{A}<\widehat{A}\widehat{A'}<\widehat{A}\widehat{A'}\widehat{B}\le\operatorname{Soc}(G). 1 < A < A A ′ < A A ′ B ≤ Soc ( G ) .
This means σ ( G ) ≥ 3 \sigma(G)\ge3 σ ( G ) ≥ 3 — which is a contradiction, as σ ( G ) ≤ 2 \sigma(G)\le2 σ ( G ) ≤ 2 . Thus, G 1 G_1 G 1 must be monolithic. Similarly, G 2 G_2 G 2 is also monolithic.
Conversely, suppose G = G 1 × G 2 G=G_1\times G_2 G = G 1 × G 2 and G 1 , G 2 G_1,G_2 G 1 , G 2 are monolithic. So by Theorem 5.1 we have Soc ( G ) = Soc ( G 1 ) × Soc ( G 2 ) \operatorname{Soc}(G)=\operatorname{Soc}(G_1)\times\operatorname{Soc}(G_2) Soc ( G ) = Soc ( G 1 ) × Soc ( G 2 ) . Now we claim the following
Claim: If Soc ( G ) = Soc ( G 1 ) × Soc ( G 2 ) \operatorname{Soc}(G)=\operatorname{Soc}(G_1)\times\operatorname{Soc}(G_2) Soc ( G ) = Soc ( G 1 ) × Soc ( G 2 ) then σ ( G ) = σ ( G 1 ) + σ ( G 2 ) \sigma(G)=\sigma(G_1)+\sigma(G_2) σ ( G ) = σ ( G 1 ) + σ ( G 2 ) .
Now by Proposition 4.4 we have:
Soc ( G 1 ) = B 1 , 1 × ⋯ × B 1 , r 1 × ∏ p S 1 , p and Soc ( G 2 ) = B 2 , 1 × ⋯ × B 2 , r 2 × ∏ p S 2 , p , \operatorname{Soc}(G_1)=B_{1,1}\times\cdots\times B_{1,r_1}\times\prod_p S_{1,p}
\quad\text{and}\quad
\operatorname{Soc}(G_2)=B_{2,1}\times\cdots\times B_{2,r_2}\times\prod_p S_{2,p}, Soc ( G 1 ) = B 1 , 1 × ⋯ × B 1 , r 1 × p ∏ S 1 , p and Soc ( G 2 ) = B 2 , 1 × ⋯ × B 2 , r 2 × p ∏ S 2 , p ,
where, B 1 , 1 , … , B 1 , r 1 B_{1,1},\ldots,B_{1,r_1} B 1 , 1 , … , B 1 , r 1 and B 2 , 1 , … , B 2 , r 2 B_{2,1},\ldots,B_{2,r_2} B 2 , 1 , … , B 2 , r 2 are non-abelian minimal normal subgroups of G 1 G_1 G 1 and G 2 G_2 G 2 respectively, and S i , p S_{i,p} S i , p are the subgroups generated by the abelian minimal normal p p p -subgroups. Then, we get:
Soc ( G ) = B 1 , 1 × ⋯ × B 1 , r 1 × B 2 , 1 × ⋯ × B 2 , r 2 × ∏ p ( S 1 , p × S 2 , p ) . \operatorname{Soc}(G)=B_{1,1}\times\cdots\times B_{1,r_1}\times B_{2,1}\times\cdots\times B_{2,r_2}\times\prod_p(S_{1,p}\times S_{2,p}). Soc ( G ) = B 1 , 1 × ⋯ × B 1 , r 1 × B 2 , 1 × ⋯ × B 2 , r 2 × p ∏ ( S 1 , p × S 2 , p ) .
Suppose, S p = S 1 , p × S 2 , p S_p=S_{1,p}\times S_{2,p} S p = S 1 , p × S 2 , p . By Theorem 5.1 , the non-abelian minimal normal subgroups of G G G are the B i , k B_{i,k} B i , k , and the abelian minimal normal p p p -subgroups of G G G lie in S p S_p S p and generate it. So, S p S_p S p is the subgroup of Proposition 4.4 for G G G . Then, as F p G \mathbb{F}_pG F p G -modules, S p = S 1 , p ⊕ S 2 , p S_p=S_{1,p}\oplus S_{2,p} S p = S 1 , p ⊕ S 2 , p . By Proposition 4.4 , each S i , p S_{i,p} S i , p is a semisimple F p G i \mathbb{F}_pG_i F p G i -module. So, we can consider an isotypic decomposition for both of them as:
S 1 , p ≅ ⨁ α ∈ I 1 X α n α and S 2 , p ≅ ⨁ β ∈ I 2 Y β m β . S_{1,p}\cong\bigoplus_{\alpha\in I_1}X_\alpha^{n_\alpha}
\quad\text{and}\quad
S_{2,p}\cong\bigoplus_{\beta\in I_2}Y_\beta^{m_\beta}. S 1 , p ≅ α ∈ I 1 ⨁ X α n α and S 2 , p ≅ β ∈ I 2 ⨁ Y β m β .
here and below, these are isomorphisms of modules, and I 1 , I 2 I_1,I_2 I 1 , I 2 (and the exponents r , s r,s r , s introduced below) depend on p p p , where X α X_\alpha X α ’s are pairwise non-isomorphic simple F p G 1 \mathbb{F}_pG_1 F p G 1 -modules and Y β Y_\beta Y β ’s are pairwise non-isomorphic simple F p G 2 \mathbb{F}_pG_2 F p G 2 -modules. Now, when each X α X_\alpha X α and each Y β Y_\beta Y β is considered as an F p G \mathbb{F}_pG F p G -module, G i G_i G i acts on them via the projections π 1 \pi_1 π 1 and π 2 \pi_2 π 2 . So, if we consider x ∈ X α x\in X_\alpha x ∈ X α and ( 1 , g 2 ) ∈ G 2 ∗ (1,g_2)\in G_2^* ( 1 , g 2 ) ∈ G 2 ∗ , then ( 1 , g 2 ) ⋅ x = π 1 ( 1 , g 2 ) ⋅ x = x (1,g_2)\cdot x=\pi_1(1,g_2)\cdot x=x ( 1 , g 2 ) ⋅ x = π 1 ( 1 , g 2 ) ⋅ x = x . Similarly, for y ∈ Y β y\in Y_\beta y ∈ Y β and ( g 1 , 1 ) ∈ G 1 ∗ (g_1,1)\in G_1^* ( g 1 , 1 ) ∈ G 1 ∗ , ( g 1 , 1 ) ⋅ y = y (g_1,1)\cdot y=y ( g 1 , 1 ) ⋅ y = y . So, G 2 G_2 G 2 acts trivially on each X α X_\alpha X α and G 1 G_1 G 1 acts trivially on each Y β Y_\beta Y β . Now, suppose, X α ≅ Y β X_\alpha\cong Y_\beta X α ≅ Y β as F p G \mathbb{F}_pG F p G -modules. Suppose, f : X α → ≅ Y β f:X_\alpha\xrightarrow{\cong} Y_\beta f : X α ≅ Y β . Then, for any g ∈ G 1 g \in G_1 g ∈ G 1 and x ∈ X α x \in X_\alpha x ∈ X α , f ( g ⋅ x ) = g ⋅ f ( x ) f(g\cdot x)=g\cdot f(x) f ( g ⋅ x ) = g ⋅ f ( x ) . But f ( x ) ∈ Y β f(x)\in Y_\beta f ( x ) ∈ Y β and G 1 G_1 G 1 acts trivially on Y β Y_\beta Y β , so
f ( g ⋅ x ) = g ⋅ f ( x ) = f ( x ) . f(g\cdot x)=g\cdot f(x)=f(x). f ( g ⋅ x ) = g ⋅ f ( x ) = f ( x ) .
As f f f is an isomorphism, and hence is injective, it follows that g ⋅ x = x g\cdot x=x g ⋅ x = x . Thus, G 1 G_1 G 1 also acts trivially on X α X_\alpha X α . So, G G G acts trivially on X α X_\alpha X α and hence X α X_\alpha X α is the trivial F p G \mathbb{F}_pG F p G -module. So, Y β Y_\beta Y β is also trivial F p G \mathbb{F}_pG F p G -module. So, X α ≅ Y β X_\alpha\cong Y_\beta X α ≅ Y β if and only if they are the same trivial F p G \mathbb{F}_pG F p G -module. So, we get
S 1 , p = ( ⨁ α ∈ I 1 ′ X α n α ) ⊕ T r and S 2 , p = ( ⨁ β ∈ I 2 ′ Y β m β ) ⊕ T s . S_{1,p}=\left(\bigoplus_{\alpha\in I'_1}X_\alpha^{n_\alpha}\right)\oplus T^r
\quad\text{and}\quad
S_{2,p}=\left(\bigoplus_{\beta\in I'_2}Y_\beta^{m_\beta}\right)\oplus T^s. S 1 , p = α ∈ I 1 ′ ⨁ X α n α ⊕ T r and S 2 , p = β ∈ I 2 ′ ⨁ Y β m β ⊕ T s .
as F p G \mathbb{F}_pG F p G -modules, where I 1 ′ I'_1 I 1 ′ and I 2 ′ I'_2 I 2 ′ are the index sets for which X α X_\alpha X α and Y β Y_\beta Y β are non-trivial respectively. So, we get
S p = S 1 , p ⊕ S 2 , p = ( ⨁ α ∈ I 1 ′ X α n α ) ⊕ ( ⨁ β ∈ I 2 ′ Y β m β ) ⊕ T r + s . S_p=S_{1,p}\oplus S_{2,p}=
\left(\bigoplus_{\alpha\in I'_1}X_\alpha^{n_\alpha}\right)
\oplus
\left(\bigoplus_{\beta\in I'_2}Y_\beta^{m_\beta}\right)
\oplus T^{r+s}. S p = S 1 , p ⊕ S 2 , p = α ∈ I 1 ′ ⨁ X α n α ⊕ β ∈ I 2 ′ ⨁ Y β m β ⊕ T r + s .
Now, by Proposition 4.4 , we have
σ ( G 1 ) = r 1 + ∑ p ( ∑ α ∈ I 1 ′ n α + r ) and σ ( G 2 ) = r 2 + ∑ p ( ∑ β ∈ I 2 ′ m β + s ) . \sigma(G_1)=r_1+\sum_p\left(\sum_{\alpha\in I'_1}n_\alpha+r\right)
\quad\text{and}\quad
\sigma(G_2)=r_2+\sum_p\left(\sum_{\beta\in I'_2}m_\beta+s\right). σ ( G 1 ) = r 1 + p ∑ α ∈ I 1 ′ ∑ n α + r and σ ( G 2 ) = r 2 + p ∑ β ∈ I 2 ′ ∑ m β + s .
But, ∑ α ∈ I 1 ′ n α + r = ∑ α ∈ I 1 n α \sum_{\alpha\in I'_1}n_\alpha+r=\sum_{\alpha\in I_1}n_\alpha ∑ α ∈ I 1 ′ n α + r = ∑ α ∈ I 1 n α and ∑ β ∈ I 2 ′ m β + s = ∑ β ∈ I 2 m β \sum_{\beta\in I'_2}m_\beta+s=\sum_{\beta\in I_2}m_\beta ∑ β ∈ I 2 ′ m β + s = ∑ β ∈ I 2 m β . So,
σ ( G 1 ) = r 1 + ∑ p ∑ α ∈ I 1 n α and σ ( G 2 ) = r 2 + ∑ p ∑ β ∈ I 2 m β . \sigma(G_1)=r_1+\sum_p\sum_{\alpha\in I_1}n_\alpha
\quad\text{and}\quad
\sigma(G_2)=r_2+\sum_p\sum_{\beta\in I_2}m_\beta. σ ( G 1 ) = r 1 + p ∑ α ∈ I 1 ∑ n α and σ ( G 2 ) = r 2 + p ∑ β ∈ I 2 ∑ m β .
Finally then we get
σ ( G 1 ) + σ ( G 2 ) = r 1 + ∑ p ∑ α ∈ I 1 n α + r 2 + ∑ p ∑ β ∈ I 2 m β = r 1 + r 2 + ∑ p ( ∑ α ∈ I 1 n α + ∑ β ∈ I 2 m β ) = σ ( G ) . \begin{aligned}
\sigma(G_1)+\sigma(G_2)=r_1+\sum_p\sum_{\alpha\in I_1}n_\alpha+r_2+\sum_p\sum_{\beta\in I_2}m_\beta\\
&=r_1+r_2+\sum_p\left(\sum_{\alpha\in I_1}n_\alpha+\sum_{\beta\in I_2}m_\beta\right) \\
&=\sigma(G).
\end{aligned} σ ( G 1 ) + σ ( G 2 ) = r 1 + p ∑ α ∈ I 1 ∑ n α + r 2 + p ∑ β ∈ I 2 ∑ m β = r 1 + r 2 + p ∑ α ∈ I 1 ∑ n α + β ∈ I 2 ∑ m β = σ ( G ) .
The last equality again holds by Proposition 4.4 .
Now, as G 1 G_1 G 1 and G 2 G_2 G 2 are monolithic, it follows that σ ( G 1 ) = 1 = σ ( G 2 ) \sigma(G_1)=1=\sigma(G_2) σ ( G 1 ) = 1 = σ ( G 2 ) . Hence, by our claim, it follows that σ ( G ) = 2 \sigma(G)=2 σ ( G ) = 2 . So, by Theorem 4.5 , it follows then, G = G 1 × G 2 G=G_1\times G_2 G = G 1 × G 2 is M \mathcal{M} M -full. □
Corollary 5.3 allows exactly two monolithic factors. The next example shows that two cannot be replaced by three.
Example 5.4. The group S 3 × S 3 S_3\times S_3 S 3 × S 3 is M \mathcal{M} M -full by Corollary 5.3 , since S 3 S_3 S 3 is monolithic with minimal normal subgroup A 3 A_3 A 3 . But S 3 × S 3 × S 3 S_3\times S_3\times S_3 S 3 × S 3 × S 3 is not. Let A 3 ( i ) A_3^{(i)} A 3 ( i ) denote A 3 A_3 A 3 placed in the i i i -th coordinate of S 3 × S 3 × S 3 S_3\times S_3\times S_3 S 3 × S 3 × S 3 . Since A 3 A_3 A 3 is not central in S 3 S_3 S 3 , Theorem 5.1 shows that A 3 ( 1 ) , A 3 ( 2 ) , A 3 ( 3 ) A_3^{(1)}, A_3^{(2)}, A_3^{(3)} A 3 ( 1 ) , A 3 ( 2 ) , A 3 ( 3 ) are exactly the minimal normal subgroups of S 3 × S 3 × S 3 S_3\times S_3\times S_3 S 3 × S 3 × S 3 . The normal subgroup A 3 × A 3 × 1 A_3\times A_3\times1 A 3 × A 3 × 1 contains A 3 ( 1 ) A_3^{(1)} A 3 ( 1 ) and A 3 ( 2 ) A_3^{(2)} A 3 ( 2 ) but not A 3 ( 3 ) A_3^{(3)} A 3 ( 3 ) , so it is not an M \mathcal{M} M -subgroup by Corollary 3.8.
The same phenomenon occurs for groups of continuous maps into a finite group, where the factors correspond to the pieces of a partition of the underlying space into open connected subsets.
Example 5.5. Let X X X be a non-empty topological space and G G G a non-trivial finite group with the discrete topology. Write C ( X , G ) C(X,G) C ( X , G ) for the group of continuous maps X → G X \to G X → G , with pointwise multiplication. Then C ( X , G ) C(X,G) C ( X , G ) is M \mathcal{M} M -full if and only if either
X X X is connected and G G G is M \mathcal{M} M -full; or
X X X is the disjoint union of two non-empty open connected subsets and G G G is monolithic.
Indeed, in case (i) every continuous map is constant, so C ( X , G ) ≅ G C(X,G) \cong G C ( X , G ) ≅ G . In case (ii), restriction to the two connected components gives C ( X , G ) ≅ G × G C(X,G) \cong G \times G C ( X , G ) ≅ G × G , which is M \mathcal{M} M -full by Corollary 5.3 .
Conversely, suppose C ( X , G ) C(X,G) C ( X , G ) is M \mathcal{M} M -full and X X X is disconnected. If X X X admitted a partition into three non-empty clopen subsets U 1 , U 2 , U 3 U_1,U_2,U_3 U 1 , U 2 , U 3 , let K i K_i K i be the normal subgroup of continuous maps that are trivial on U i U_i U i . Choose a nonidentity g ∈ G g \in G g ∈ G and define e i e_i e i to be g g g on U i U_i U i and 1 1 1 elsewhere. Then
e 3 ∈ K 1 ∩ K 2 ∖ { 1 } , e 1 ∈ K 3 ∩ K 2 ∖ { 1 } , e 2 ∈ K 3 ∩ K 1 ∖ { 1 } . e_3 \in K_1 \cap K_2 \setminus\{1\}, \qquad e_1 \in K_3 \cap K_2 \setminus\{1\}, \qquad e_2 \in K_3 \cap K_1 \setminus\{1\}. e 3 ∈ K 1 ∩ K 2 ∖ { 1 } , e 1 ∈ K 3 ∩ K 2 ∖ { 1 } , e 2 ∈ K 3 ∩ K 1 ∖ { 1 } .
whereas K 1 ∩ K 2 ∩ K 3 = 1 K_1 \cap K_2 \cap K_3 = 1 K 1 ∩ K 2 ∩ K 3 = 1 . Thus K 3 K_3 K 3 would fail the M \mathcal{M} M -condition. Hence no such partition exists. Any decomposition X = U ⊔ V X = U \sqcup V X = U ⊔ V into two non-empty clopen subsets must therefore have U U U and V V V connected, for otherwise one could split a disconnected part and obtain a partition into three. Consequently C ( X , G ) ≅ G × G C(X,G) \cong G \times G C ( X , G ) ≅ G × G , and Corollary 5.3 forces G G G to be monolithic. The connected case reduces to G G G itself.
We end this section with the two extreme kinds of characteristically simple groups, where the M \mathcal{M} M -lattice can be written down completely; they will reappear as the crown-based powers with trivial top in § 6 .
Proposition 5.6. Let k ≥ 1 k \ge1 k ≥ 1 . Then the following statements hold.
(1) If T T T is a non-abelian finite simple group, then every normal subgroup of T k T^k T k is a product of coordinate factors. Hence
N M ( T k ) = { { 1 , T } , k = 1 , { 1 , T 1 , … , T k , T k } , k ≥ 2. \mathcal{N}^{\mathcal{M}}(T^k) =
\begin{cases}
\{1,T\}, & k=1,\\
\{1,T_1,\ldots,T_k,T^k\}, & k \ge2.
\end{cases} N M ( T k ) = { { 1 , T } , { 1 , T 1 , … , T k , T k } , k = 1 , k ≥ 2. where T i T_i T i is the i i i -th coordinate factor.
(2) For the elementary abelian group C p k C_p^k C p k , the M \mathcal{M} M -subgroups are 0 0 0 , all one-dimensional subspaces, and the whole space. For k ≤ 2 k \le2 k ≤ 2 this is the entire subgroup lattice; for k ≥ 3 k \ge3 k ≥ 3 it is a proper meet-subsemilattice.
Proof. For (1), let T k = T 1 × ⋯ × T k T^k = T_1 \times\cdots\times T_k T k = T 1 × ⋯ × T k , where T i T_i T i is the i i i -th coordinate subgroup. Then, T i ⊲ T k T_i \lhd T^k T i ⊲ T k . Now suppose N ∈ N ( T k ) N \in\mathcal{N}(T^k) N ∈ N ( T k ) . Then, as N ⊲ T k N \lhd T^k N ⊲ T k and T i ⊲ T k T_i \lhd T^k T i ⊲ T k we have N ∩ T i ⊲ T k N \cap T_i \lhd T^k N ∩ T i ⊲ T k . In particular, N ∩ T i ⊲ T i N \cap T_i \lhd T_i N ∩ T i ⊲ T i . But, T i ≅ T T_i \cong T T i ≅ T . So
N ∩ T i = 1 or N ∩ T i = T i . N \cap T_i = 1 \quad\text{or}\quad N \cap T_i = T_i. N ∩ T i = 1 or N ∩ T i = T i .
Define, I = { i : N ∩ T i = T i } I = \{i : N \cap T_i = T_i\} I = { i : N ∩ T i = T i } . Then,
∏ i ∈ I T i ≤ N . \prod_{i \in I} T_i \le N. i ∈ I ∏ T i ≤ N .
Let n = ( t 1 , … , t k ) ∈ N n = (t_1,\ldots,t_k) \in N n = ( t 1 , … , t k ) ∈ N and suppose t i ≠ 1 t_i \ne1 t i = 1 . As T T T is non-abelian and simple, we get Z ( T ) = 1 Z(T) = 1 Z ( T ) = 1 . Then, there exists x i ∈ T i x_i \in T_i x i ∈ T i such that [ t i , x i ] ≠ 1 [t_i,x_i] \ne1 [ t i , x i ] = 1 . As N ⊲ T k N \lhd T^k N ⊲ T k , it follows that [ n , x i ] ∈ N [n,x_i] \in N [ n , x i ] ∈ N . So, [ n , x i ] = ( 1 , … , [ t i , x i ] , … , 1 ) ∈ T i [n,x_i] = (1,\ldots,[t_i,x_i],\ldots,1) \in T_i [ n , x i ] = ( 1 , … , [ t i , x i ] , … , 1 ) ∈ T i and thus 1 ≠ [ n , x i ] ∈ N ∩ T i 1 \ne[n,x_i] \in N \cap T_i 1 = [ n , x i ] ∈ N ∩ T i . Now, since T i ≅ T T_i \cong T T i ≅ T is simple, so N ∩ T i = T i N \cap T_i = T_i N ∩ T i = T i and so i ∈ I i \in I i ∈ I . Therefore,
n ∈ ∏ i ∈ I T i and N ≤ ∏ i ∈ I T i . n \in\prod_{i \in I} T_i \quad\text{and}\quad N \le\prod_{i \in I} T_i. n ∈ i ∈ I ∏ T i and N ≤ i ∈ I ∏ T i .
So, N = ∏ i ∈ I T i N = \prod_{i \in I} T_i N = ∏ i ∈ I T i .
Conversely, for any I ⊆ { 1 , … , k } I \subseteq\{1,\ldots,k\} I ⊆ { 1 , … , k } we have
∏ i ∈ I T i ⊲ T k . \prod_{i \in I} T_i \lhd T^k. i ∈ I ∏ T i ⊲ T k .
So,
N ( T k ) = { ∏ i ∈ I T i : I ⊆ { 1 , … , k } } . \mathcal{N}(T^k) = \left\{\prod_{i \in I} T_i : I \subseteq\{1,\ldots,k\}\right\}. N ( T k ) = { i ∈ I ∏ T i : I ⊆ { 1 , … , k } } .
Now, as T i ≅ T T_i \cong T T i ≅ T is non-abelian and simple, MinNorm ( T k ) = { T 1 , … , T k } \operatorname{MinNorm}(T^k)=\{T_1,\ldots,T_k\} MinNorm ( T k ) = { T 1 , … , T k } . So when k = 1 k=1 k = 1 , T 1 = T T^1=T T 1 = T and hence N ( T 1 ) = N ( T ) = { 1 , T } \mathcal{N}(T^1)=\mathcal{N}(T)=\{1,T\} N ( T 1 ) = N ( T ) = { 1 , T } . For any E ⊴ T 1 = T E\trianglelefteq T^1=T E ⊴ T 1 = T , E = 1 E=1 E = 1 or E = T E=T E = T . But by Corollary 3.8,
E ∈ N M ( T 1 ) ⟺ E = 1 or ∣ supp ( E ) ∣ = 1 or Soc ( T 1 ) ≤ E . E\in\mathcal{N}^{\mathcal{M}}(T^1)\Longleftrightarrow E=1\quad\text{or}\quad|\operatorname{supp}(E)|=1\quad\text{or}\quad\operatorname{Soc}(T^1)\leq E. E ∈ N M ( T 1 ) ⟺ E = 1 or ∣ supp ( E ) ∣ = 1 or Soc ( T 1 ) ≤ E .
So, if E = 1 E=1 E = 1 , then E ∈ N M ( T 1 ) E\in\mathcal{N}^{\mathcal{M}}(T^1) E ∈ N M ( T 1 ) . If E = T E=T E = T , then as MinNorm ( T 1 ) = MinNorm ( T ) = { T } \operatorname{MinNorm}(T^1)=\operatorname{MinNorm}(T)=\{T\} MinNorm ( T 1 ) = MinNorm ( T ) = { T } , Soc ( T 1 ) = T ≤ E \operatorname{Soc}(T^1)=T\leq E Soc ( T 1 ) = T ≤ E . Thus, E ∈ N M ( T 1 ) E\in\mathcal{N}^{\mathcal{M}}(T^1) E ∈ N M ( T 1 ) . Hence,
N M ( T 1 ) = { 1 , T } . \mathcal{N}^{\mathcal{M}}(T^1)=\{1,T\}. N M ( T 1 ) = { 1 , T } .
For k ≥ 2 k\geq2 k ≥ 2 , let E ⊴ T k E\trianglelefteq T^k E ⊴ T k . Then, as we showed either E = 1 E=1 E = 1 or E = ∏ i ∈ I T i E=\prod_{i\in I}T_i E = ∏ i ∈ I T i for some I ⊆ { 1 , … , k } I\subseteq\{1,\ldots,k\} I ⊆ { 1 , … , k } . If E = 1 E=1 E = 1 , then trivially E ∈ N M ( T k ) E\in\mathcal{N}^{\mathcal{M}}(T^k) E ∈ N M ( T k ) . If E = ∏ i ∈ I T i E=\prod_{i\in I}T_i E = ∏ i ∈ I T i , then we have two cases:
Case 1: ∣ I ∣ = 1 |I|=1 ∣ I ∣ = 1 . Then, E = T i E=T_i E = T i and hence supp ( E ) = { T i } \operatorname{supp}(E)=\{T_i\} supp ( E ) = { T i } . So, E ∈ N M ( T k ) E\in\mathcal{N}^{\mathcal{M}}(T^k) E ∈ N M ( T k ) , by Corollary 3.8.
Case 2: ∣ I ∣ ≥ 2 |I|\geq2 ∣ I ∣ ≥ 2 . As Soc ( T k ) = Soc ( T 1 ) × ⋯ × Soc ( T k ) \operatorname{Soc}(T^k)=\operatorname{Soc}(T_1)\times\cdots\times\operatorname{Soc}(T_k) Soc ( T k ) = Soc ( T 1 ) × ⋯ × Soc ( T k ) , and each T i ≅ T T_i\cong T T i ≅ T , we have,
Soc ( T k ) = T 1 × ⋯ × T k = T k . \operatorname{Soc}(T^k)=T_1\times\cdots\times T_k=T^k. Soc ( T k ) = T 1 × ⋯ × T k = T k .
Since, ∣ I ∣ ≥ 2 |I|\geq2 ∣ I ∣ ≥ 2 , so E E E contains at least two distinct minimal normal subgroups, say T i T_i T i and T j T_j T j . But, if I ≠ { 1 , … , k } I\neq\{1,\ldots,k\} I = { 1 , … , k } , then
E = ∏ i ∈ I T i < T 1 T 2 ⋯ T k = Soc ( T k ) E=\prod_{i\in I}T_i<T_1T_2\cdots T_k=\operatorname{Soc}(T^k) E = i ∈ I ∏ T i < T 1 T 2 ⋯ T k = Soc ( T k )
and ∣ supp ( E ) ∣ ≥ 2 |\operatorname{supp}(E)|\geq2 ∣ supp ( E ) ∣ ≥ 2 . Therefore, E E E is not an M \mathcal{M} M -subgroup. But when I = { 1 , … , k } I=\{1,\ldots,k\} I = { 1 , … , k } , then E = T 1 ⋯ T k = T k = Soc ( T k ) E=T_1\cdots T_k=T^k=\operatorname{Soc}(T^k) E = T 1 ⋯ T k = T k = Soc ( T k ) . Hence, E ∈ N M ( T k ) E\in\mathcal{N}^{\mathcal{M}}(T^k) E ∈ N M ( T k ) .
So for k ≥ 2 k\geq2 k ≥ 2 ,
N M ( T k ) = { 1 , T 1 , … , T k , T k } . \mathcal{N}^{\mathcal{M}}(T^k)=\{1,T_1,\ldots,T_k,T^k\}. N M ( T k ) = { 1 , T 1 , … , T k , T k } .
For (2), let G = C p k G=C_p^k G = C p k . Then, G G G is an elementary abelian p p p -group. So, G G G is an F p \mathbb{F}_p F p -vector space. Consequently, every subgroup of G G G is normal and hence E ⊴ G E\trianglelefteq G E ⊴ G implies that E E E is an F p \mathbb{F}_p F p -subspace. By Corollary 4.7 , Soc ( G ) = Ω 1 ( Z ( G ) ) \operatorname{Soc}(G)=\Omega_1(Z(G)) Soc ( G ) = Ω 1 ( Z ( G )) . But G G G is abelian and is a non-trivial finite p p p -group. So Z ( G ) = G Z(G)=G Z ( G ) = G and hence, Soc ( G ) = Ω 1 ( G ) = G \operatorname{Soc}(G)=\Omega_1(G)=G Soc ( G ) = Ω 1 ( G ) = G . Now, if E ⊴ G E\trianglelefteq G E ⊴ G and E = 0 E=0 E = 0 , then clearly E ∈ N M ( G ) E\in\mathcal{N}^{\mathcal{M}}(G) E ∈ N M ( G ) . If E ≠ 0 E\neq0 E = 0 , then let dim F p E = d \dim_{\mathbb{F}_p}E=d dim F p E = d . By Corollary 4.7 , the minimal normal subgroups of G G G are all the one-dimensional F p \mathbb{F}_p F p -subspaces of G G G . So, by Corollary 5.2 , the number of minimal normal subgroups contained in E E E is
p d − 1 p − 1 . \frac{p^d-1}{p-1}. p − 1 p d − 1 .
As E ≠ 0 E\neq0 E = 0 ,
E ∈ N M ( G ) ⟺ ∣ supp ( E ) ∣ = 1 or Soc ( G ) ≤ E . E\in\mathcal{N}^{\mathcal{M}}(G)\Longleftrightarrow|\operatorname{supp}(E)|=1\quad\text{or}\quad\operatorname{Soc}(G)\leq E. E ∈ N M ( G ) ⟺ ∣ supp ( E ) ∣ = 1 or Soc ( G ) ≤ E .
But, Soc ( G ) ≤ E \operatorname{Soc}(G)\leq E Soc ( G ) ≤ E if and only if E = G E=G E = G and ∣ supp ( E ) ∣ = 1 |\operatorname{supp}(E)|=1 ∣ supp ( E ) ∣ = 1 if and only if p d − 1 p − 1 = 1 \frac{p^d-1}{p-1}=1 p − 1 p d − 1 = 1 if and only if d = 1 d=1 d = 1 if and only if dim F p E = 1 \dim_{\mathbb{F}_p}E=1 dim F p E = 1 . So,
E ∈ N M ( G ) ⟺ E = 0 or dim F p E = 1 or E = G . E\in\mathcal{N}^{\mathcal{M}}(G)\Longleftrightarrow E=0\quad\text{or}\quad\dim_{\mathbb{F}_p}E=1\quad\text{or}\quad E=G. E ∈ N M ( G ) ⟺ E = 0 or dim F p E = 1 or E = G .
Now, as G = C p k G=C_p^k G = C p k is abelian, N ( G ) = Sub ( G ) \mathcal{N}(G)=\operatorname{Sub}(G) N ( G ) = Sub ( G ) . If k = 1 k=1 k = 1 , then G = C p G=C_p G = C p and hence N ( G ) = { 0 , C p } \mathcal{N}(G)=\{0,C_p\} N ( G ) = { 0 , C p } . So clearly, N M ( G ) = N ( G ) \mathcal{N}^{\mathcal{M}}(G)=\mathcal{N}(G) N M ( G ) = N ( G ) . So, G G G is M \mathcal{M} M -full and Sub ( G ) = N M ( G ) \operatorname{Sub}(G)=\mathcal{N}^{\mathcal{M}}(G) Sub ( G ) = N M ( G ) . If k = 2 k=2 k = 2 , then G = C p 2 G=C_p^2 G = C p 2 . But C p C_p C p is monolithic and so by Corollary 5.3 , N M ( G ) = N ( G ) = Sub ( G ) \mathcal{N}^{\mathcal{M}}(G)=\mathcal{N}(G)=\operatorname{Sub}(G) N M ( G ) = N ( G ) = Sub ( G ) . So for k ≤ 2 k\leq2 k ≤ 2 , the M \mathcal{M} M -subgroup lattice is precisely the entire subgroup lattice of G G G .
If k ≥ 3 k\geq3 k ≥ 3 , then we already have
N M ( G ) = { 0 } ⊔ { L ≤ C p k : dim F p L = 1 } ⊔ { C p k } . \mathcal{N}^{\mathcal{M}}(G)=\{0\}\sqcup\{L\leq C_p^k:\dim_{\mathbb{F}_p}L=1\}\sqcup\{C_p^k\}. N M ( G ) = { 0 } ⊔ { L ≤ C p k : dim F p L = 1 } ⊔ { C p k } .
As k ≥ 3 k \ge3 k ≥ 3 , there exists a 2-dimensional F p \mathbb{F}_p F p -subspace E E E of C p k C_p^k C p k . So, E ∉ N M ( G ) E \notin\mathcal{N}^{\mathcal{M}}(G) E ∈ / N M ( G ) but E ⊴ G E \trianglelefteq G E ⊴ G and hence E ∈ N ( G ) = Sub ( G ) E \in\mathcal{N}(G) = \operatorname{Sub}(G) E ∈ N ( G ) = Sub ( G ) . So, N M ( G ) ⊊ Sub ( G ) \mathcal{N}^{\mathcal{M}}(G) \subsetneq\operatorname{Sub}(G) N M ( G ) ⊊ Sub ( G ) . By Proposition 3.10, for any two E , F ∈ N M ( G ) E,F \in\mathcal{N}^{\mathcal{M}}(G) E , F ∈ N M ( G )
E ∧ F = E ∩ F ∈ N M ( G ) . E \wedge F = E \cap F \in\mathcal{N}^{\mathcal{M}}(G). E ∧ F = E ∩ F ∈ N M ( G ) .
So, N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) is a proper meet-subsemilattice of Sub ( G ) \operatorname{Sub}(G) Sub ( G ) .
Crown-based powers: normal lattices and quotients
Let L L L be a finite monolithic primitive group, let A = Soc ( L ) A = \operatorname{Soc}(L) A = Soc ( L ) , and put H = L / A H = L/A H = L / A . For k ≥ 1 k \ge1 k ≥ 1 , the crown-based power of L L L of size k k k is
L k = { ( x 1 , … , x k ) ∈ L k : x 1 A = ⋯ = x k A } . L_k = \{(x_1,\ldots,x_k) \in L^k : x_1A = \cdots= x_kA\}. L k = {( x 1 , … , x k ) ∈ L k : x 1 A = ⋯ = x k A } .
Write V = A k V = A^k V = A k . Then V ⊴ L k V \trianglelefteq L_k V ⊴ L k and L k / V ≅ H L_k/V \cong H L k / V ≅ H . Crown-based powers were introduced in [8 ] to describe the finite groups that need more generators than any proper quotient, and through the isomorphism G / R G ( B ) ≅ ( L B ) δ G ( B ) G/R_G(B) \cong(L_B)^{\delta_G(B)} G / R G ( B ) ≅ ( L B ) δ G ( B ) of Detomi and Lucchini [10 ] they describe the crowns of an arbitrary finite group (Corollary 6.10). This crown-quotient construction is also used in the study of maximal-subgroup intersection numbers [4 Section 5.1] ; see [19 ] for a survey. It is known that Soc ( L k ) = A k \operatorname{Soc}(L_k) = A^k Soc ( L k ) = A k is a direct product of k k k minimal normal subgroups [8 ] . Furthermore, Lucchini, Moscatiello and Spiga [18 Lemma 2.2] proved that every normal subgroup of L k L_k L k is comparable with V V V . We recall the proof of this fact and combine it with the standard module-theoretic description of V V V to obtain an explicit presentation of the whole normal-subgroup lattice, which is the starting point for our results on M \mathcal{M} M -lattices, rigidity and quotients.
Theorem 6.1. Let L L L , A A A , H H H , L k L_k L k , and V = A k V = A^k V = A k be as above.
(1) [18 Lemma 2.2] . Every normal subgroup of L k L_k L k is comparable with V V V :
N ⊴ L k ⟹ N ≤ V or V ≤ N . N \trianglelefteq L_k \quad\Longrightarrow\quad N \le V \text{ or } V \le N. N ⊴ L k ⟹ N ≤ V or V ≤ N . Consequently,
N ( L k ) = [ 1 , V ] N ( L k ) ∪ [ V , L k ] N ( L k ) , [ 1 , V ] N ( L k ) ∩ [ V , L k ] N ( L k ) = { V } . \mathcal{N}(L_k) = [1,V]_{\mathcal{N}(L_k)} \cup[V,L_k]_{\mathcal{N}(L_k)}, \qquad[1,V]_{\mathcal{N}(L_k)} \cap[V,L_k]_{\mathcal{N}(L_k)} = \{V\}. N ( L k ) = [ 1 , V ] N ( L k ) ∪ [ V , L k ] N ( L k ) , [ 1 , V ] N ( L k ) ∩ [ V , L k ] N ( L k ) = { V } . (2) The quotient map induces a lattice isomorphism
[ V , L k ] N ( L k ) ⟶ N ( H ) , N ⟼ N / V . [V,L_k]_{\mathcal{N}(L_k)} \longrightarrow\mathcal{N}(H), \qquad N \longmapsto N/V. [ V , L k ] N ( L k ) ⟶ N ( H ) , N ⟼ N / V . (3) If A A A is non-abelian, then
I ⟼ A I : = ∏ i ∈ I A i I \longmapsto A_I := \prod_{i \in I} A_i I ⟼ A I := i ∈ I ∏ A i is a lattice isomorphism from the Boolean lattice of subsets of { 1 , … , k } \{1,\ldots,k\} { 1 , … , k } onto [ 1 , V ] N ( L k ) [1,V]_{\mathcal{N}(L_k)} [ 1 , V ] N ( L k ) . Here A i A_i A i is the i i i -th coordinate copy of A A A and A ∅ = 1 A_{\varnothing} = 1 A ∅ = 1 .
(4) If A A A is abelian, then A A A is an irreducible faithful F p H \mathbb{F}_pH F p H -module for some prime p p p . With
D = End F p H ( A ) , D = \operatorname{End}_{\mathbb{F}_pH}(A), D = End F p H ( A ) , the map
K ⟼ Hom F p H ( A , K ) K \longmapsto\operatorname{Hom}_{\mathbb{F}_pH}(A,K) K ⟼ Hom F p H ( A , K ) identifies [ 1 , V ] N ( L k ) [1,V]_{\mathcal{N}(L_k)} [ 1 , V ] N ( L k ) with the lattice of D D D -subspaces of D k D^k D k .
In particular, Soc ( L k ) = V \operatorname{Soc}(L_k) = V Soc ( L k ) = V . Its minimal normal subgroups are the coordinate factors A 1 , … , A k A_1,\ldots,A_k A 1 , … , A k when A A A is non-abelian, and are naturally parametrized by P k − 1 ( D ) \mathbb{P}^{k-1}(D) P k − 1 ( D ) when A A A is abelian.
Proof. (1) For completeness we give the argument behind the cited comparability lemma. Let C = C L ( A ) = { x ∈ L : x a x − 1 = a for every a ∈ A } C=C_L(A)=\{x\in L: xax^{-1}=a\text{ for every }a\in A\} C = C L ( A ) = { x ∈ L : x a x − 1 = a for every a ∈ A } . Then C ⊴ L C\unlhd L C ⊴ L . We first show that C ≤ A C\le A C ≤ A . If A A A is non-abelian, then C ∩ A = Z ( A ) = 1 C\cap A=Z(A)=1 C ∩ A = Z ( A ) = 1 , as Z ( A ) Z(A) Z ( A ) is characteristic in A A A , hence normal in L L L , and Z ( A ) ≠ A Z(A)\ne A Z ( A ) = A . As L L L is monolithic, every non-trivial normal subgroup of L L L contains A A A , and so C = 1 C=1 C = 1 . Now suppose A A A is abelian, so that A ≤ C A\le C A ≤ C . Let M M M be a maximal subgroup of L L L with core L ( M ) = 1 \operatorname{core}_L(M)=1 core L ( M ) = 1 . If A ≤ M A\le M A ≤ M , then A ≤ core L ( M ) = 1 A\le\operatorname{core}_L(M)=1 A ≤ core L ( M ) = 1 , which is a contradiction. So A ≰ M A\nleq M A ≰ M , and as M M M is maximal, L = A M L=AM L = A M . Now C ∩ M C\cap M C ∩ M is normalized by M M M and centralized by A A A . So C ∩ M ⊴ A M = L C\cap M\unlhd AM=L C ∩ M ⊴ A M = L , and hence C ∩ M ≤ core L ( M ) = 1 C\cap M\le\operatorname{core}_L(M)=1 C ∩ M ≤ core L ( M ) = 1 . If x ∈ C x\in C x ∈ C , then x = a m x=am x = am for some a ∈ A a\in A a ∈ A and m ∈ M m\in M m ∈ M , and m = a − 1 x ∈ C ∩ M = 1 m=a^{-1}x\in C\cap M=1 m = a − 1 x ∈ C ∩ M = 1 . Thus, x = a ∈ A x=a\in A x = a ∈ A . So in both cases C ≤ A C\le A C ≤ A , that is,
x ∈ L ∖ A ⟹ [ x , a ] ≠ 1 for some a ∈ A . x\in L\setminus A\Longrightarrow[x,a]\ne1\text{ for some }a\in A. x ∈ L ∖ A ⟹ [ x , a ] = 1 for some a ∈ A .
Now, let N ⊴ L k N\unlhd L_k N ⊴ L k with N ≰ V N\nleq V N ≰ V , and let n = ( n 1 , … , n k ) ∈ N ∖ V n=(n_1,\ldots,n_k)\in N\setminus V n = ( n 1 , … , n k ) ∈ N ∖ V . As n 1 A = ⋯ = n k A n_1A=\cdots=n_kA n 1 A = ⋯ = n k A and n ∉ A k n\notin A^k n ∈ / A k , n j ∉ A n_j\notin A n j ∈ / A for every j j j . Fix j j j . Then [ n j , a ] ≠ 1 [n_j,a]\ne1 [ n j , a ] = 1 for some a ∈ A a\in A a ∈ A . Let a ^ j = ( 1 , … , a , … , 1 ) ∈ A j \hat a_j=(1,\ldots,a,\ldots,1)\in A_j a ^ j = ( 1 , … , a , … , 1 ) ∈ A j , with a a a in the j j j -th place. As N ⊴ L k N\unlhd L_k N ⊴ L k , so
1 ≠ [ n , a ^ j ] = ( 1 , … , [ n j , a ] , … , 1 ) ∈ N ∩ A j . 1\ne[n,\hat a_j]=(1,\ldots,[n_j,a],\ldots,1)\in N\cap A_j. 1 = [ n , a ^ j ] = ( 1 , … , [ n j , a ] , … , 1 ) ∈ N ∩ A j .
Let K ≤ A K\le A K ≤ A be the set of j j j -th coordinates of the elements of N ∩ A j N\cap A_j N ∩ A j . As N ∩ A j ⊴ L k N\cap A_j\unlhd L_k N ∩ A j ⊴ L k and ( l , … , l ) ∈ L k (l,\ldots,l)\in L_k ( l , … , l ) ∈ L k for every l ∈ L l\in L l ∈ L , K ⊴ L K\unlhd L K ⊴ L . As K ≠ 1 K\ne1 K = 1 and A A A is minimal normal in L L L , K = A K=A K = A , that is, A j ≤ N A_j\le N A j ≤ N . As j j j was arbitrary, V ≤ N V\le N V ≤ N . The rest of (1) follows at once.
(2) The map L k → H L_k\to H L k → H , ( x 1 , … , x k ) ↦ x 1 A (x_1,\ldots,x_k)\mapsto x_1A ( x 1 , … , x k ) ↦ x 1 A is a surjective homomorphism with kernel V V V . So, by the correspondence theorem, N ↦ N / V N\mapsto N/V N ↦ N / V is a bijection from [ V , L k ] N ( L k ) [V,L_k]_{\mathcal{N}(L_k)} [ V , L k ] N ( L k ) onto N ( L k / V ) ≅ N ( H ) \mathcal{N}(L_k/V)\cong\mathcal{N}(H) N ( L k / V ) ≅ N ( H ) , which preserves inclusion in both directions. Hence, it is a lattice isomorphism.
(3) Clearly, each A I ⊴ L k A_I\unlhd L_k A I ⊴ L k , and A I ≤ A J A_I\le A_J A I ≤ A J if and only if I ⊆ J I\subseteq J I ⊆ J . Now let N ⊴ L k N\unlhd L_k N ⊴ L k with N ≤ V N\le V N ≤ V , and put I = { i : A i ≤ N } I=\{i:A_i\le N\} I = { i : A i ≤ N } . Then A I ≤ N A_I\le N A I ≤ N . Let n = ( a 1 , … , a k ) ∈ N n=(a_1,\ldots,a_k)\in N n = ( a 1 , … , a k ) ∈ N and i ∉ I i\notin I i ∈ / I . If a i ≠ 1 a_i\ne1 a i = 1 , then as Z ( A ) = 1 Z(A)=1 Z ( A ) = 1 , [ a i , a ] ≠ 1 [a_i,a]\ne1 [ a i , a ] = 1 for some a ∈ A a\in A a ∈ A . Then, as in (1), 1 ≠ [ n , a ^ i ] ∈ N ∩ A i 1\ne[n,\hat a_i]\in N\cap A_i 1 = [ n , a ^ i ] ∈ N ∩ A i and so A i ≤ N A_i\le N A i ≤ N , which is a contradiction. So a i = 1 a_i=1 a i = 1 for all i ∉ I i\notin I i ∈ / I , and hence N ≤ A I N\le A_I N ≤ A I . Thus, N = A I N=A_I N = A I . So I ↦ A I I\mapsto A_I I ↦ A I is a bijection which preserves inclusion in both directions, and hence it is a lattice isomorphism.
(4) As A A A is abelian, the Sylow subgroups of A A A and { a ∈ A : a p = 1 } \{a\in A:a^p=1\} { a ∈ A : a p = 1 } are characteristic in A A A , and hence normal in L L L . So, by minimality of A A A , A A A is an elementary abelian p p p -group for some prime p p p . Now L L L acts on A A A by conjugation, and A A A acts trivially as A A A is abelian. So A A A is an F p H \mathbb{F}_pH F p H -module. Its submodules are exactly the normal subgroups of L L L contained in A A A , so A A A is irreducible. It is faithful, as 1 A ∈ H 1A\in H 1 A ∈ H acts trivially only if l ∈ C ≤ A l\in C\le A l ∈ C ≤ A . Moreover, for ( l i ) ∈ L k (l_i)\in L_k ( l i ) ∈ L k and ( a i ) ∈ V (a_i)\in V ( a i ) ∈ V we have l i a i l i − 1 = l i a i l i − 1 l_i a_i l_i^{-1}=l_i a_i l_i^{-1} l i a i l i − 1 = l i a i l i − 1 , as l i ∈ l 1 A l_i\in l_1A l i ∈ l 1 A and A A A is abelian. So, V ≅ A k V\cong A^k V ≅ A k with H H H acting diagonally, and [ 1 , V ] N ( L k ) [1,V]_{\mathcal{N}(L_k)} [ 1 , V ] N ( L k ) is the lattice of F p H \mathbb{F}_pH F p H -submodules of A k A^k A k .
For x ∈ D k x\in D^k x ∈ D k , let Φ x : A → A k \Phi_x:A\to A^k Φ x : A → A k be as in the proof of Lemma 4.3 . Then x ↦ Φ x x\mapsto\Phi_x x ↦ Φ x is a bijection from D k D^k D k onto Hom F p H ( A , A k ) \operatorname{Hom}_{\mathbb{F}_pH}(A,A^k) Hom F p H ( A , A k ) , Φ x + y = Φ x + Φ y \Phi_{x+y}=\Phi_x+\Phi_y Φ x + y = Φ x + Φ y and Φ x d = Φ x ∘ d \Phi_{xd}=\Phi_x\circ d Φ x d = Φ x ∘ d for d ∈ D d\in D d ∈ D . For a submodule K ≤ A k K\le A^k K ≤ A k and a subspace W ≤ D k W\le D^k W ≤ D k put
W ( K ) = { x ∈ D k : im Φ x ≤ K } and K ( W ) = ∑ x ∈ W im Φ x . W(K)=\{x\in D^k:\operatorname{im}\Phi_x\le K\}\qquad\text{and}\qquad K(W)=\sum_{x\in W}\operatorname{im}\Phi_x. W ( K ) = { x ∈ D k : im Φ x ≤ K } and K ( W ) = x ∈ W ∑ im Φ x .
So W ( K ) W(K) W ( K ) corresponds to Hom F p H ( A , K ) \operatorname{Hom}_{\mathbb{F}_pH}(A,K) Hom F p H ( A , K ) , and it is a subspace of D k D^k D k , as im Φ x + y ≤ im Φ x + im Φ y \operatorname{im}\Phi_{x+y}\le\operatorname{im}\Phi_x+\operatorname{im}\Phi_y im Φ x + y ≤ im Φ x + im Φ y and im Φ x d ≤ im Φ x \operatorname{im}\Phi_{xd}\le\operatorname{im}\Phi_x im Φ x d ≤ im Φ x . Clearly, both maps preserve inclusion. As A k A^k A k is semisimple, so is K K K , and hence K K K is a sum of simple submodules. By the proof of Lemma 4.3 , each of them is im Φ x \operatorname{im}\Phi_x im Φ x for some x ∈ D k x\in D^k x ∈ D k , and then x ∈ W ( K ) x\in W(K) x ∈ W ( K ) . Hence, K ( W ( K ) ) = K K(W(K))=K K ( W ( K )) = K .
Conversely, clearly W ⊆ W ( K ( W ) ) W\subseteq W(K(W)) W ⊆ W ( K ( W )) . Let b 1 , … , b r b_1,\ldots,b_r b 1 , … , b r be a basis of W W W and define F : A r → K ( W ) F:A^r\to K(W) F : A r → K ( W ) by F ( a 1 , … , a r ) = ∑ i Φ b i ( a i ) F(a_1,\ldots,a_r)=\sum_i\Phi_{b_i}(a_i) F ( a 1 , … , a r ) = ∑ i Φ b i ( a i ) . Then F F F is a surjective homomorphism, as Φ x = ∑ i Φ b i ∘ d i \Phi_x=\sum_i\Phi_{b_i}\circ d_i Φ x = ∑ i Φ b i ∘ d i for x = b 1 d 1 + ⋯ + b r d r ∈ W x=b_1d_1+\cdots+b_rd_r\in W x = b 1 d 1 + ⋯ + b r d r ∈ W . As A r A^r A r is semisimple, there exists a homomorphism s : K ( W ) → A r s:K(W)\to A^r s : K ( W ) → A r with F ∘ s = id F\circ s=\operatorname{id} F ∘ s = id . Now let x ∈ W ( K ( W ) ) x\in W(K(W)) x ∈ W ( K ( W )) . Then s ∘ Φ x ∈ Hom F p H ( A , A r ) s\circ\Phi_x\in\operatorname{Hom}_{\mathbb{F}_pH}(A,A^r) s ∘ Φ x ∈ Hom F p H ( A , A r ) , so s ∘ Φ x = ( c 1 , … , c r ) s\circ\Phi_x=(c_1,\ldots,c_r) s ∘ Φ x = ( c 1 , … , c r ) for some c i ∈ D c_i\in D c i ∈ D . Hence,
Φ x = F ∘ s ∘ Φ x = ∑ i Φ b i ∘ c i = Φ b 1 c 1 + ⋯ + b r c r \Phi_x=F\circ s\circ\Phi_x=\sum_i\Phi_{b_i}\circ c_i=\Phi_{b_1c_1+\cdots+b_rc_r} Φ x = F ∘ s ∘ Φ x = i ∑ Φ b i ∘ c i = Φ b 1 c 1 + ⋯ + b r c r
and so x = b 1 c 1 + ⋯ + b r c r ∈ W x=b_1c_1+\cdots+b_rc_r\in W x = b 1 c 1 + ⋯ + b r c r ∈ W . Thus, W ( K ( W ) ) = W W(K(W))=W W ( K ( W )) = W . Therefore, K ↦ W ( K ) K\mapsto W(K) K ↦ W ( K ) and W ↦ K ( W ) W\mapsto K(W) W ↦ K ( W ) are mutually inverse bijections which preserve inclusion, and so K ↦ Hom F p H ( A , K ) K\mapsto\operatorname{Hom}_{\mathbb{F}_pH}(A,K) K ↦ Hom F p H ( A , K ) is a lattice isomorphism onto the lattice of D k D^k D k -subspaces of D k D^k D k .
Finally, if N ∈ MinNorm ( L k ) N\in\operatorname{MinNorm}(L_k) N ∈ MinNorm ( L k ) , then by (1), N ≤ V N\le V N ≤ V or V ≤ N V\le N V ≤ N , and in the latter case N = V N=V N = V . So MinNorm ( L k ) \operatorname{MinNorm}(L_k) MinNorm ( L k ) is the set of atoms of [ 1 , V ] N ( L k ) [1,V]_{\mathcal{N}(L_k)} [ 1 , V ] N ( L k ) . These are the A i A_i A i when A A A is non-abelian, by (3), and they correspond to the one-dimensional subspaces of D k D^k D k , that is, to P k − 1 ( D ) \mathbb{P}^{k-1}(D) P k − 1 ( D ) , when A A A is abelian, by (4). In both cases each A i A_i A i is an atom (in the abelian case A i ≅ A A_i\cong A A i ≅ A is a simple submodule), and V = A 1 ⋯ A k V=A_1\cdots A_k V = A 1 ⋯ A k . Hence, Soc ( L k ) = V \operatorname{Soc}(L_k)=V Soc ( L k ) = V . □
Theorem 6.1 turns the lower part of the normal lattice into an explicit finite geometry. Here [ k r ] q \left[\begin{smallmatrix}k\\r\end{smallmatrix}\right]_q [ k r ] q denotes the number of r r r -dimensional subspaces of F q k \mathbb{F}_q^k F q k when 0 ≤ r ≤ k 0\le r\le k 0 ≤ r ≤ k , and is zero otherwise.
Corollary 6.2. Assume k ≥ 2 k\ge2 k ≥ 2 , and, in the abelian case, put q = ∣ D ∣ q=|D| q = ∣ D ∣ . Set
a k = { k , A non-abelian , [ k 1 ] q = q k − 1 q − 1 , A abelian . a_k=
\begin{cases}
k, & A\text{ non-abelian},\\
\left[\begin{matrix}k\\1\end{matrix}\right]_q=\dfrac{q^k-1}{q-1}, & A\text{ abelian}.
\end{cases} a k = ⎩ ⎨ ⎧ k , [ k 1 ] q = q − 1 q k − 1 , A non-abelian , A abelian . Then
N M ( L k ) = { 1 } ⊔ MinNorm ( L k ) ⊔ [ V , L k ] N ( L k ) . \mathcal{N}^{\mathcal{M}}(L_k)=\{1\}\sqcup\operatorname{MinNorm}(L_k)\sqcup[V,L_k]_{\mathcal{N}(L_k)}. N M ( L k ) = { 1 } ⊔ MinNorm ( L k ) ⊔ [ V , L k ] N ( L k ) . Equivalently, the M \mathcal{M} M -lattice is obtained by placing a k a_k a k incomparable atoms between 1 1 1 and V V V and then attaching a copy of N ( H ) \mathcal{N}(H) N ( H ) above V V V . In particular,
∣ N M ( L k ) ∣ = ∣ N ( H ) ∣ + a k + 1. |\mathcal{N}^{\mathcal{M}}(L_k)|=|\mathcal{N}(H)|+a_k+1. ∣ N M ( L k ) ∣ = ∣ N ( H ) ∣ + a k + 1. Moreover,
∣ N ( L k ) ∣ = { ∣ N ( H ) ∣ + 2 k − 1 , A non-abelian , ∣ N ( H ) ∣ + ∑ r = 0 k [ k r ] q − 1 , A abelian , |\mathcal{N}(L_k)|=
\begin{cases}
|\mathcal{N}(H)|+2^k-1, & A\text{ non-abelian},\\
|\mathcal{N}(H)|+\sum_{r=0}^{k}\left[\begin{matrix}k\\r\end{matrix}\right]_q-1, & A\text{ abelian},
\end{cases} ∣ N ( L k ) ∣ = ⎩ ⎨ ⎧ ∣ N ( H ) ∣ + 2 k − 1 , ∣ N ( H ) ∣ + ∑ r = 0 k [ k r ] q − 1 , A non-abelian , A abelian , and the number of normal subgroups discarded by the M \mathcal{M} M -closure system is
∣ N ( L k ) ∖ N M ( L k ) ∣ = { 2 k − k − 2 , A non-abelian , ∑ r = 2 k − 1 [ k r ] q , A abelian . |\mathcal{N}(L_k)\setminus\mathcal{N}^{\mathcal{M}}(L_k)|=
\begin{cases}
2^k-k-2, & A\text{ non-abelian},\\
\sum_{r=2}^{k-1}\left[\begin{matrix}k\\r\end{matrix}\right]_q, & A\text{ abelian}.
\end{cases} ∣ N ( L k ) ∖ N M ( L k ) ∣ = ⎩ ⎨ ⎧ 2 k − k − 2 , ∑ r = 2 k − 1 [ k r ] q , A non-abelian , A abelian . For every K ⊴ L k K\unlhd L_k K ⊴ L k with K ≤ V K\le V K ≤ V , the closure is
cl M ( K ) = { K , λ L k ( K ) ∈ { 0 , 1 , k } , V , 2 ≤ λ L k ( K ) ≤ k − 1. \operatorname{cl}_{\mathcal{M}}(K)=
\begin{cases}
K, & \lambda_{L_k}(K)\in\{0,1,k\},\\
V, & 2\le\lambda_{L_k}(K)\le k-1.
\end{cases} cl M ( K ) = { K , V , λ L k ( K ) ∈ { 0 , 1 , k } , 2 ≤ λ L k ( K ) ≤ k − 1. The displayed loss vanishes exactly when k = 2 k=2 k = 2 .
Proof. Let E ⊴ L k E\unlhd L_k E ⊴ L k . By Theorem 6.1 , either E ≤ V E\le V E ≤ V or V ≤ E V\le E V ≤ E . If V ≤ E V\le E V ≤ E , then as V = Soc ( L k ) V=\operatorname{Soc}(L_k) V = Soc ( L k ) , E ∈ N M ( L k ) E\in\mathcal{N}^{\mathcal{M}}(L_k) E ∈ N M ( L k ) by Corollary 3.8. Now suppose E ≤ V E\le V E ≤ V . By Theorem 6.1 , [ 1 , V ] N ( L k ) [1,V]_{\mathcal{N}(L_k)} [ 1 , V ] N ( L k ) is isomorphic to the Boolean lattice of subsets of { 1 , … , k } \{1,\ldots,k\} { 1 , … , k } if A A A is non-abelian, and to the lattice of D D D -subspaces of D k D^k D k if A A A is abelian. So λ L k ( E ) = ∣ I ∣ \lambda_{L_k}(E)=|I| λ L k ( E ) = ∣ I ∣ if E = A I E=A_I E = A I , and λ L k ( E ) = r \lambda_{L_k}(E)=r λ L k ( E ) = r if E E E corresponds to an r r r -dimensional subspace. Moreover, the minimal normal subgroups contained in E E E are the atoms below E E E , so
∣ supp ( E ) ∣ = { λ L k ( E ) , A non-abelian , [ r 1 ] q , A abelian . |\operatorname{supp}(E)|=
\begin{cases}
\lambda_{L_k}(E), & A\text{ non-abelian},\\
\left[\begin{matrix}r\\1\end{matrix}\right]_q, & A\text{ abelian}.
\end{cases} ∣ supp ( E ) ∣ = ⎩ ⎨ ⎧ λ L k ( E ) , [ r 1 ] q , A non-abelian , A abelian .
Hence, ∣ supp ( E ) ∣ = 1 |\operatorname{supp}(E)|=1 ∣ supp ( E ) ∣ = 1 if and only if λ L k ( E ) = 1 \lambda_{L_k}(E)=1 λ L k ( E ) = 1 , and V ≤ E V\leq E V ≤ E if and only if λ L k ( E ) = k \lambda_{L_k}(E)=k λ L k ( E ) = k . So, by Corollary 3.8,
E ∈ N M ( L k ) ⟺ λ L k ( E ) ∈ { 0 , 1 , k } ⟺ E = 1 , E ∈ MinNorm ( L k ) or E = V . E\in\mathcal{N}^{\mathcal{M}}(L_k)\Longleftrightarrow\lambda_{L_k}(E)\in\{0,1,k\}\Longleftrightarrow E=1,\ E\in\operatorname{MinNorm}(L_k)\text{ or }E=V. E ∈ N M ( L k ) ⟺ λ L k ( E ) ∈ { 0 , 1 , k } ⟺ E = 1 , E ∈ MinNorm ( L k ) or E = V .
As k ≥ 2 k\geq2 k ≥ 2 , V V V is not minimal normal, and so the union below is disjoint. Thus,
\mathcal{N}^{\mathcal{M}}(L_k)=\{1\}\mathbin{\mathaccent\cdot\cup}\operatorname{MinNorm}(L_k)\mathbin{\mathaccent\cdot\cup}[V,L_k]_{\mathcal{N}}(L_k).
Now, by Theorem 6.1 , ∣ [ V , L k ] N ( L k ) ∣ = ∣ N ( H ) ∣ |[V,L_k]_{\mathcal{N}}(L_k)|=|\mathcal{N}(H)| ∣ [ V , L k ] N ( L k ) ∣ = ∣ N ( H ) ∣ , and ∣ [ 1 , V ] N ( L k ) ∣ |[1,V]_{\mathcal{N}}(L_k)| ∣ [ 1 , V ] N ( L k ) ∣ is 2 k 2^k 2 k or ∑ r = 0 k ( k r ) q \sum_{r=0}^{k}\binom{k}{r}_{q} ∑ r = 0 k ( r k ) q . Also, L k L_k L k has exactly a k a_k a k minimal normal subgroups, namely the k k k factors A i A_i A i , or the ones corresponding to the ( k 1 ) q \binom{k}{1}_{q} ( 1 k ) q one-dimensional subspaces of D k D^k D k . As the two intervals meet only in V V V , so
∣ N ( L k ) ∣ = ∣ [ 1 , V ] N ( L k ) ∣ + ∣ N ( H ) ∣ − 1 and ∣ N M ( L k ) ∣ = ∣ N ( H ) ∣ + a k + 1 , |\mathcal{N}(L_k)|=|[1,V]_{\mathcal{N}}(L_k)|+|\mathcal{N}(H)|-1
\quad\text{and}\quad
|\mathcal{N}^{\mathcal{M}}(L_k)|=|\mathcal{N}(H)|+a_k+1, ∣ N ( L k ) ∣ = ∣ [ 1 , V ] N ( L k ) ∣ + ∣ N ( H ) ∣ − 1 and ∣ N M ( L k ) ∣ = ∣ N ( H ) ∣ + a k + 1 ,
which are the formulas in the statement. Subtracting, and using ( k 0 ) q = ( k k ) q = 1 \binom{k}{0}_{q}=\binom{k}{k}_{q}=1 ( 0 k ) q = ( k k ) q = 1 , we get
∣ N ( L k ) ∖ N M ( L k ) ∣ = { 2 k − k − 2 , A non-abelian , ∑ r = 0 k ( k r ) q − 1 − ( k 1 ) q − 1 = ∑ r = 2 k − 1 ( k r ) q , A abelian . |\mathcal{N}(L_k)\setminus\mathcal{N}^{\mathcal{M}}(L_k)|=
\begin{cases}
2^k-k-2,& A\text{ non-abelian},\\
\sum_{r=0}^{k}\binom{k}{r}_{q}-1-\binom{k}{1}_{q}-1
=\sum_{r=2}^{k-1}\binom{k}{r}_{q},& A\text{ abelian}.
\end{cases} ∣ N ( L k ) ∖ N M ( L k ) ∣ = { 2 k − k − 2 , ∑ r = 0 k ( r k ) q − 1 − ( 1 k ) q − 1 = ∑ r = 2 k − 1 ( r k ) q , A non-abelian , A abelian .
Now let K ⊴ L k K\unlhd L_k K ⊴ L k with K ≤ V K\leq V K ≤ V . If K K K is an M \mathcal{M} M -subgroup, then cl M ( K ) = K \operatorname{cl}_{\mathcal{M}}(K)=K cl M ( K ) = K . Otherwise K ≠ 1 K\neq1 K = 1 and ∣ supp ( K ) ∣ ≠ 1 |\operatorname{supp}(K)|\neq1 ∣ supp ( K ) ∣ = 1 , so ∣ supp ( K ) ∣ ≥ 2 |\operatorname{supp}(K)|\geq2 ∣ supp ( K ) ∣ ≥ 2 , and by Proposition 3.11, cl M ( K ) = K Soc ( L k ) = K V = V \operatorname{cl}_{\mathcal{M}}(K)=K\operatorname{Soc}(L_k)=KV=V cl M ( K ) = K Soc ( L k ) = K V = V . As K K K is an M \mathcal{M} M -subgroup if and only if λ L k ( K ) ∈ { 0 , 1 , k } \lambda_{L_k}(K)\in\{0,1,k\} λ L k ( K ) ∈ { 0 , 1 , k } , this gives the formula for cl M ( K ) \operatorname{cl}_{\mathcal{M}}(K) cl M ( K ) .
Finally, if k = 2 k=2 k = 2 , then 2 k − k − 2 = 0 2^k-k-2=0 2 k − k − 2 = 0 and ∑ r = 2 1 ( 2 r ) q = 0 \sum_{r=2}^{1}\binom{2}{r}_{q}=0 ∑ r = 2 1 ( r 2 ) q = 0 . If k ≥ 3 k\geq3 k ≥ 3 , then 2 k − k − 2 ≥ 3 2^k-k-2\geq3 2 k − k − 2 ≥ 3 and
∑ r = 2 k − 1 ( k r ) q ≥ ( k 2 ) q ≥ 1. \sum_{r=2}^{k-1}\binom{k}{r}_{q}\geq\binom{k}{2}_{q}\geq1. r = 2 ∑ k − 1 ( r k ) q ≥ ( 2 k ) q ≥ 1.
So the loss vanishes exactly when k = 2 k=2 k = 2 . □
In particular, we recover the crown-based part of Theorem C.
Corollary 6.3. For every k ≥ 1 k\geq1 k ≥ 1 , the crown-based power L k L_k L k is M \mathcal{M} M -full if and only if k ≤ 2 k\leq2 k ≤ 2 .
Proof. If k = 1 k=1 k = 1 , then L k = L L^k=L L k = L and thus L 1 = L L_1=L L 1 = L . So as L L L is finite and monolithic, it follows by Theorem 4.5 (4a) that L 1 L_1 L 1 is M \mathcal{M} M -full. For k ≥ 2 k\geq2 k ≥ 2 , by Corollary 6.2 we have N ( L k ) ∖ N M ( L k ) = ∅ \mathcal{N}(L_k)\setminus\mathcal{N}^{\mathcal{M}}(L_k)=\varnothing N ( L k ) ∖ N M ( L k ) = ∅ exactly when k = 2 k=2 k = 2 . So, L k L_k L k is M \mathcal{M} M -full when k = 2 k=2 k = 2 and since it is also M \mathcal{M} M -full when k = 1 k=1 k = 1 , so we get, L k L_k L k is M \mathcal{M} M -full when k ≤ 2 k\leq2 k ≤ 2 . Also, for k ≥ 3 k\geq3 k ≥ 3 we have N ( L k ) ∖ N M ( L k ) ≠ ∅ \mathcal{N}(L_k)\setminus\mathcal{N}^{\mathcal{M}}(L_k)\neq\varnothing N ( L k ) ∖ N M ( L k ) = ∅ , so L k L_k L k is not M \mathcal{M} M -full. Hence, L k L_k L k is M \mathcal{M} M -full if and only if k ≤ 2 k\leq2 k ≤ 2 . □
The complete lattice description also determines all maximal M \mathcal{M} M -chains, not only their maximum length.
Proposition 6.4. Assume k ≥ 2 k\geq2 k ≥ 2 . The lattice N M ( L k ) \mathcal{N}^{\mathcal{M}}(L_k) N M ( L k ) is graded and modular. Its rank function is
rk M ( E ) = { 0 , E = 1 , 1 , E ∈ MinNorm ( L k ) , 2 + λ H ( E / V ) , V ≤ E . \operatorname{rk}_{\mathcal{M}}(E)=
\begin{cases}
0,&E=1,\\
1,&E\in\operatorname{MinNorm}(L_k),\\
2+\lambda_H(E/V),&V\leq E.
\end{cases} rk M ( E ) = ⎩ ⎨ ⎧ 0 , 1 , 2 + λ H ( E / V ) , E = 1 , E ∈ MinNorm ( L k ) , V ≤ E . Every maximal M \mathcal{M} M -chain has the form
1 < W < V = E 0 < E 1 < ⋯ < E s = L k , 1<W<V=E_0<E_1<\cdots<E_s=L_k, 1 < W < V = E 0 < E 1 < ⋯ < E s = L k , where W ∈ MinNorm ( L k ) W\in\operatorname{MinNorm}(L_k) W ∈ MinNorm ( L k ) and
1 = E 0 / V < E 1 / V < ⋯ < E s / V = H 1=E_0/V<E_1/V<\cdots<E_s/V=H 1 = E 0 / V < E 1 / V < ⋯ < E s / V = H is a chief series of H H H . Hence every maximal M \mathcal{M} M -chain has length ℓ n ( H ) + 2 \ell_n(H)+2 ℓ n ( H ) + 2 . If c ( H ) c(H) c ( H ) denotes the number of chief series of H H H , with c ( 1 ) = 1 c(1)=1 c ( 1 ) = 1 , then L k L_k L k has exactly a k c ( H ) a_kc(H) a k c ( H ) maximal M \mathcal{M} M -chains.
Furthermore, N M ( L k ) \mathcal{N}^{\mathcal{M}}(L_k) N M ( L k ) is distributive if and only if A A A is non-abelian, k = 2 k=2 k = 2 , and N ( H ) \mathcal{N}(H) N ( H ) is distributive.
Proof. By Corollary 6.2 we know that
N M ( L k ) = { 1 } ⊔ MinNorm ( L k ) ⊔ [ V , L k ] N ( L k ) . \mathcal{N}^{\mathcal{M}}(L_k)=\{1\}\sqcup\operatorname{MinNorm}(L_k)\sqcup[V,L_k]_{\mathcal{N}(L_k)}. N M ( L k ) = { 1 } ⊔ MinNorm ( L k ) ⊔ [ V , L k ] N ( L k ) .
So we define, N 1 = { 1 } ⊔ MinNorm ( L k ) ∪ { V } \mathcal{N}_1=\{1\}\sqcup\operatorname{MinNorm}(L_k)\cup\{V\} N 1 = { 1 } ⊔ MinNorm ( L k ) ∪ { V } and N 2 = [ V , L k ] N ( L k ) \mathcal{N}_2=[V,L_k]_{\mathcal{N}(L_k)} N 2 = [ V , L k ] N ( L k ) . Then, N M ( L k ) = N 1 ∪ N 2 \mathcal{N}^{\mathcal{M}}(L_k)=\mathcal{N}_1\cup\mathcal{N}_2 N M ( L k ) = N 1 ∪ N 2 and N 1 ∩ N 2 = { V } \mathcal{N}_1\cap\mathcal{N}_2=\{V\} N 1 ∩ N 2 = { V } . Moreover, for any X ∈ N 1 ∖ { V } X\in\mathcal{N}_1\setminus\{V\} X ∈ N 1 ∖ { V } and Y ∈ N 2 ∖ { V } Y\in\mathcal{N}_2\setminus\{V\} Y ∈ N 2 ∖ { V } , we have
Thus, N M ( L k ) = N 1 ⊕ N 2 \mathcal{N}^{\mathcal{M}}(L_k)=\mathcal{N}_1\oplus\mathcal{N}_2 N M ( L k ) = N 1 ⊕ N 2 , the vertical sum of N 1 \mathcal{N}_1 N 1 and N 2 \mathcal{N}_2 N 2 , that is, N 1 \mathcal{N}_1 N 1 is placed below N 2 \mathcal{N}_2 N 2 and the top V V V of N 1 \mathcal{N}_1 N 1 is identified with the bottom V V V of N 2 \mathcal{N}_2 N 2 . Now for any M ≠ N ∈ MinNorm ( L k ) M\ne N\in\operatorname{MinNorm}(L_k) M = N ∈ MinNorm ( L k ) we have M ∩ N = 1 M\cap N=1 M ∩ N = 1 and M ∨ M N = cl M ( M N ) = V M\vee_{\mathcal{M}}N=\operatorname{cl}_{\mathcal{M}}(MN)=V M ∨ M N = cl M ( M N ) = V by Corollary 6.2 , although the product M N MN M N itself is smaller than V V V when k ≥ 3 k\ge3 k ≥ 3 . So, N 1 \mathcal{N}_1 N 1 is a lattice of height 2 and hence N 1 \mathcal{N}_1 N 1 is modular. Further by Theorem 6.1 , [ V , L k ] N ( L k ) ≅ N ( H ) [V,L_k]_{\mathcal{N}(L_k)}\cong\mathcal{N}(H) [ V , L k ] N ( L k ) ≅ N ( H ) and by Lemma 3.1 , N ( H ) \mathcal{N}(H) N ( H ) is modular. Also, N 2 \mathcal{N}_2 N 2 is a sublattice of N M ( L k ) \mathcal{N}^{\mathcal{M}}(L_k) N M ( L k ) , as cl M ( E F ) = E F \operatorname{cl}_{\mathcal{M}}(EF)=EF cl M ( E F ) = E F for E , F ∈ N 2 E,F\in\mathcal{N}_2 E , F ∈ N 2 . As we are gluing the bottom of N 2 \mathcal{N}_2 N 2 with the top of N 1 \mathcal{N}_1 N 1 and both N 1 , N 2 \mathcal{N}_1,\mathcal{N}_2 N 1 , N 2 are modular, it follows that N M ( L k ) \mathcal{N}^{\mathcal{M}}(L_k) N M ( L k ) is also modular.
Now consider V ≤ E ⊴ L k V\le E\unlhd L_k V ≤ E ⊴ L k . As [ V , L k ] N ( L k ) ≅ N ( H ) [V,L_k]_{\mathcal{N}(L_k)}\cong\mathcal{N}(H) [ V , L k ] N ( L k ) ≅ N ( H ) , if we restrict this isomorphism, we get [ V , E ] N ( L k ) ≅ [ 1 , E / V ] N ( H ) [V,E]_{\mathcal{N}(L_k)}\cong[1,E/V]_{\mathcal{N}(H)} [ V , E ] N ( L k ) ≅ [ 1 , E / V ] N ( H ) . So a chain,
V = E 0 < E 1 < ⋯ < E r = E . V=E_0<E_1<\cdots<E_r=E. V = E 0 < E 1 < ⋯ < E r = E .
corresponds to
1 = E 0 / V < E 1 / V < ⋯ < E r / V = E / V . 1=E_0/V<E_1/V<\cdots<E_r/V=E/V. 1 = E 0 / V < E 1 / V < ⋯ < E r / V = E / V .
Thus, λ H ( E / V ) \lambda_H(E/V) λ H ( E / V ) is the length of a maximal chain from V V V to E E E . Now, for any E ∈ N M ( L k ) E\in\mathcal{N}^{\mathcal{M}}(L_k) E ∈ N M ( L k ) , if E = 1 E=1 E = 1 then rk M ( E ) = 0 \operatorname{rk}_{\mathcal{M}}(E)=0 rk M ( E ) = 0 . If E ∈ MinNorm ( L k ) E\in\operatorname{MinNorm}(L_k) E ∈ MinNorm ( L k ) , then a maximal chain looks like 1 < E 1<E 1 < E . Hence, rk M ( E ) = 1 \operatorname{rk}_{\mathcal{M}}(E)=1 rk M ( E ) = 1 . If E = V E=V E = V , then a maximal chain is 1 < M < E = V 1<M<E=V 1 < M < E = V , where M ∈ MinNorm ( L k ) M\in\operatorname{MinNorm}(L_k) M ∈ MinNorm ( L k ) and so rk M ( E ) = 2 \operatorname{rk}_{\mathcal{M}}(E)=2 rk M ( E ) = 2 . If V < E V<E V < E , then a maximal chain from 1 1 1 to E E E looks like
1 < M < V = E 0 < E 1 < E 2 < ⋯ < E r = E , 1<M<V=E_0<E_1<E_2<\cdots<E_r=E, 1 < M < V = E 0 < E 1 < E 2 < ⋯ < E r = E ,
and hence, rk M ( E ) = 2 + λ H ( E / V ) \operatorname{rk}_{\mathcal{M}}(E)=2+\lambda_H(E/V) rk M ( E ) = 2 + λ H ( E / V ) . Thus the rank function increases by one across every cover and hence N M ( L k ) \mathcal{N}^{\mathcal{M}}(L_k) N M ( L k ) is graded. Now, observe that a maximal M \mathcal{M} M -chain must first pass through the chain 1 < W < V 1<W<V 1 < W < V , where W ∈ MinNorm ( L k ) W\in\operatorname{MinNorm}(L_k) W ∈ MinNorm ( L k ) . After that, it follows a maximal chain in [ V , L k ] N ( L k ) [V,L_k]_{\mathcal{N}(L_k)} [ V , L k ] N ( L k )
V = E 0 < E 1 < ⋯ < E s = L k , V=E_0<E_1<\cdots<E_s=L_k, V = E 0 < E 1 < ⋯ < E s = L k ,
which as we showed corresponds to a chief series,
1 = E 0 / V < E 1 / V < ⋯ < E s / V = L k / V ≅ H 1=E_0/V<E_1/V<\cdots<E_s/V=L_k/V\cong H 1 = E 0 / V < E 1 / V < ⋯ < E s / V = L k / V ≅ H
in N ( H ) \mathcal{N}(H) N ( H ) , due to the isomorphism [ V , L k ] N ( L k ) ≅ N ( H ) [V,L_k]_{\mathcal{N}(L_k)}\cong\mathcal{N}(H) [ V , L k ] N ( L k ) ≅ N ( H ) in Theorem 6.1 . Thus, if ℓ n ( H ) \ell_n(H) ℓ n ( H ) denotes the length of a chief series in H H H , the length of a maximal M \mathcal{M} M -chain in L k L_k L k is 2 + ℓ n ( H ) 2+\ell_n(H) 2 + ℓ n ( H ) . Furthermore, if a k a_k a k denotes the number of atoms of N M ( L k ) \mathcal{N}^{\mathcal{M}}(L_k) N M ( L k ) as in Corollary 6.2 , then due to our two independent choices: first a W ∈ MinNorm ( L k ) W\in\operatorname{MinNorm}(L_k) W ∈ MinNorm ( L k ) and then a chief series of H H H , the number of maximal M \mathcal{M} M -chains that L k L_k L k has is a k c ( H ) a_kc(H) a k c ( H ) .
Now suppose N M ( L k ) \mathcal{N}^{\mathcal{M}}(L_k) N M ( L k ) is distributive and a k = ∣ MinNorm ( L k ) ∣ a_k=|\operatorname{MinNorm}(L_k)| a k = ∣ MinNorm ( L k ) ∣ . If N , M , K N,M,K N , M , K are three atoms such that their pairwise join is V V V , then, M ∧ N = 1 M\wedge N=1 M ∧ N = 1 and M ∧ K = 1 M\wedge K=1 M ∧ K = 1 and thus,
M = M ∧ V = M ∧ ( N ∨ K ) = ( M ∧ N ) ∨ ( M ∧ K ) = 1 , M=M\wedge V=M\wedge(N\vee K)=(M\wedge N)\vee(M\wedge K)=1, M = M ∧ V = M ∧ ( N ∨ K ) = ( M ∧ N ) ∨ ( M ∧ K ) = 1 ,
which is a contradiction as M M M is an atom. So, this tells us that a k ≤ 2 a_k\le2 a k ≤ 2 . Now if A A A is abelian, then by Corollary 6.2 we have
a k = [ k 1 ] q = q k − 1 q − 1 , a_k=\genfrac{[}{]}{0pt}{}{k}{1}_q=\frac{q^k-1}{q-1}, a k = [ 1 k ] q = q − 1 q k − 1 ,
where q = ∣ D ∣ q=|D| q = ∣ D ∣ . As k ≥ 2 k\ge2 k ≥ 2 , q k − 1 ≥ q 2 − 1 q^{k}-1\ge q^{2}-1 q k − 1 ≥ q 2 − 1 and hence a k ≥ q + 1 ≥ 3 a_k\ge q+1\ge3 a k ≥ q + 1 ≥ 3 as q ≥ 2 q\ge2 q ≥ 2 . So, A A A cannot be abelian. Then, A A A must be non-abelian and a k = k a_k=k a k = k . But k ≥ 2 k\ge2 k ≥ 2 and k = a k ≤ 2 k=a_k\le2 k = a k ≤ 2 gives us k = 2 k=2 k = 2 . Also as every normal subgroup containing V V V is an M \mathcal{M} M -subgroup so
[ V , L k ] N M ( L k ) = [ V , L k ] N ( L k ) ≅ N ( H ) . [V,L_k]_{\mathcal{N}^{\mathcal{M}}(L_k)}=[V,L_k]_{\mathcal{N}(L_k)}\cong\mathcal{N}(H). [ V , L k ] N M ( L k ) = [ V , L k ] N ( L k ) ≅ N ( H ) .
As N M ( L k ) \mathcal{N}^{\mathcal{M}}(L_k) N M ( L k ) is distributive, [ V , L k ] N M ( L k ) [V,L_k]_{\mathcal{N}^{\mathcal{M}}(L_k)} [ V , L k ] N M ( L k ) is also distributive and hence N ( H ) \mathcal{N}(H) N ( H ) is distributive. Conversely suppose A A A is non-abelian, k = 2 k=2 k = 2 and N ( H ) \mathcal{N}(H) N ( H ) is distributive. Then, both the interval below V V V and the interval above V V V are distributive. But joining the top of the lower interval and the bottom of the above interval preserves distributivity and gives us N M ( L k ) \mathcal{N}^{\mathcal{M}}(L_k) N M ( L k ) . Hence, N M ( L k ) \mathcal{N}^{\mathcal{M}}(L_k) N M ( L k ) is distributive. □
The full normal lattice and the M \mathcal{M} M -lattice retain different data about a crown-based power. The following isomorphism criteria make this comparison precise; notably, the M \mathcal{M} M -lattice need not determine either the crown multiplicity or the field appearing in the abelian socle.
Theorem 6.5. Let L L L and L ′ L' L ′ be finite monolithic primitive groups with socles A A A and B B B , respectively. Put
H = L / A , H ′ = L ′ / B , H=L/A,\qquad H'=L'/B, H = L / A , H ′ = L ′ / B , and let L k L_k L k and L m ′ L'_m L m ′ be crown-based powers with k , m ≥ 2 k,m\ge2 k , m ≥ 2 . Define
a ( L , k ) = ∣ MinNorm ( L k ) ∣ = { k , A non-abelian , q k − 1 q − 1 , A abelian , a(L,k)=|\operatorname{MinNorm}(L_k)|=
\begin{cases}
k, & A\ \text{non-abelian},\\
\frac{q^k-1}{q-1}, & A\ \text{abelian},
\end{cases} a ( L , k ) = ∣ MinNorm ( L k ) ∣ = { k , q − 1 q k − 1 , A non-abelian , A abelian , where q = ∣ End F p H ( A ) ∣ q=|\operatorname{End}_{\mathbb{F}_pH}(A)| q = ∣ End F p H ( A ) ∣ in the abelian case, and define a ( L ′ , m ) a(L',m) a ( L ′ , m ) analogously.
(1) The normal lattices N ( L k ) \mathcal{N}(L_k) N ( L k ) and N ( L m ′ ) \mathcal{N}(L'_m) N ( L m ′ ) are isomorphic if and only if all of the following hold:
(a) k = m k=m k = m ;
(b) A A A and B B B are either both abelian or both non-abelian;
(c) N ( H ) ≅ N ( H ′ ) \mathcal{N}(H)\cong\mathcal{N}(H') N ( H ) ≅ N ( H ′ ) ;
(d) in the abelian case,
∣ End F p H ( A ) ∣ = ∣ End F p ′ H ′ ( B ) ∣ , |\operatorname{End}_{\mathbb{F}_pH}(A)|=|\operatorname{End}_{\mathbb{F}_{p'}H'}(B)|, ∣ End F p H ( A ) ∣ = ∣ End F p ′ H ′ ( B ) ∣ , where B B B is an irreducible F p ′ H ′ \mathbb{F}_{p'}H' F p ′ H ′ -module.
(2) The M \mathcal{M} M -lattices N M ( L k ) \mathcal{N}^{\mathcal{M}}(L_k) N M ( L k ) and N M ( L m ′ ) \mathcal{N}^{\mathcal{M}}(L'_m) N M ( L m ′ ) are isomorphic if and only if
a ( L , k ) = a ( L ′ , m ) and N ( H ) ≅ N ( H ′ ) . a(L,k)=a(L',m)\qquad\text{and}\qquad\mathcal{N}(H)\cong\mathcal{N}(H'). a ( L , k ) = a ( L ′ , m ) and N ( H ) ≅ N ( H ′ ) .
Thus the full normal lattice determines the crown multiplicity, the abelian or non-abelian nature of the socle, and the order of the endomorphism field in the abelian case. None of these data is determined in general by the M \mathcal{M} M -lattice. Indeed, for every non-abelian finite simple group T T T ,
N M ( T 3 ) ≅ N M ( C 2 2 ) , N ( T 3 ) ≇ N ( C 2 2 ) , \mathcal{N}^{\mathcal{M}}(T^3)\cong\mathcal{N}^{\mathcal{M}}(C_2^2),\qquad
\mathcal{N}(T^3)\not\cong\mathcal{N}(C_2^2), N M ( T 3 ) ≅ N M ( C 2 2 ) , N ( T 3 ) ≅ N ( C 2 2 ) ,
and one also has
N M ( C 2 5 ) ≅ N M ( C 3 5 ) , N ( C 2 5 ) ≇ N ( C 3 5 ) . \mathcal{N}^{\mathcal{M}}(C_2^5)\cong\mathcal{N}^{\mathcal{M}}(C_3^5),\qquad
\mathcal{N}(C_2^5)\not\cong\mathcal{N}(C_3^5). N M ( C 2 5 ) ≅ N M ( C 3 5 ) , N ( C 2 5 ) ≅ N ( C 3 5 ) .
Proof. For (1), first suppose that Φ : N ( L k ) → N ( L m ′ ) \Phi:\mathcal{N}(L_k)\to\mathcal{N}(L'_m) Φ : N ( L k ) → N ( L m ′ ) is an isomorphism. Let V = A k V=A^k V = A k and V ′ = B m V'=B^m V ′ = B m . Then,
V = ⋁ MinNorm ( L k ) and V ′ = ⋁ MinNorm ( L m ′ ) . V=\bigvee\operatorname{MinNorm}(L_k)\qquad\text{and}\qquad V'=\bigvee\operatorname{MinNorm}(L'_m). V = ⋁ MinNorm ( L k ) and V ′ = ⋁ MinNorm ( L m ′ ) .
So we must have Φ ( V ) = V ′ \Phi(V)=V' Φ ( V ) = V ′ . So by Theorem 6.1 , we get [ 1 , V ] N ( L k ) ≅ [ 1 , V ′ ] N ( L m ′ ) [1,V]_{\mathcal{N}(L_k)}\cong[1,V']_{\mathcal{N}(L'_m)} [ 1 , V ] N ( L k ) ≅ [ 1 , V ′ ] N ( L m ′ ) and k = λ L k ( V ) = λ L m ′ ( V ′ ) = m k=\lambda_{L_k}(V)=\lambda_{L'_m}(V')=m k = λ L k ( V ) = λ L m ′ ( V ′ ) = m . Now, if A A A is non-abelian then [ 1 , V ] N ( L k ) [1,V]_{\mathcal{N}(L_k)} [ 1 , V ] N ( L k ) is distributive as it is isomorphic to the Boolean lattice P ( { 1 , … , k } ) \mathcal{P}(\{1,\ldots,k\}) P ({ 1 , … , k }) . If A A A is abelian, then [ 1 , V ] N ( L k ) [1,V]_{\mathcal{N}(L_k)} [ 1 , V ] N ( L k ) is not distributive, as Sub D ( D k ) \operatorname{Sub}_D(D^k) Sub D ( D k ) contains three distinct lines ℓ 1 , ℓ 2 , ℓ 3 \ell_1,\ell_2,\ell_3 ℓ 1 , ℓ 2 , ℓ 3 lying in one plane P P P (as k ≥ 2 k\ge2 k ≥ 2 and ∣ D ∣ + 1 ≥ 3 |D|+1\ge3 ∣ D ∣ + 1 ≥ 3 ), and then
ℓ 1 ∧ ( ℓ 2 ∨ ℓ 3 ) = ℓ 1 ∧ P = ℓ 1 ≠ 0 = ( ℓ 1 ∧ ℓ 2 ) ∨ ( ℓ 1 ∧ ℓ 3 ) . \ell_1\wedge(\ell_2\vee\ell_3)=\ell_1\wedge P=\ell_1\ne0=(\ell_1\wedge\ell_2)\vee(\ell_1\wedge\ell_3). ℓ 1 ∧ ( ℓ 2 ∨ ℓ 3 ) = ℓ 1 ∧ P = ℓ 1 = 0 = ( ℓ 1 ∧ ℓ 2 ) ∨ ( ℓ 1 ∧ ℓ 3 ) .
As [ 1 , V ] N ( L k ) ≅ [ 1 , V ′ ] N ( L m ′ ) [1,V]_{\mathcal{N}(L_k)} \cong[1,V']_{\mathcal{N}(L'_m)} [ 1 , V ] N ( L k ) ≅ [ 1 , V ′ ] N ( L m ′ ) , either both A A A and B B B are abelian or both are non-abelian. Since we also have [ V , L k ] N ( L k ) ≅ N ( H ) [V,L_k]_{\mathcal{N}(L_k)} \cong\mathcal{N}(H) [ V , L k ] N ( L k ) ≅ N ( H ) and [ V ′ , L m ′ ] N ( L m ′ ) ≅ N ( H ′ ) [V',L'_m]_{\mathcal{N}(L'_m)} \cong\mathcal{N}(H') [ V ′ , L m ′ ] N ( L m ′ ) ≅ N ( H ′ ) , so by our assumption it follows that N ( H ) ≅ N ( H ′ ) \mathcal{N}(H) \cong\mathcal{N}(H') N ( H ) ≅ N ( H ′ ) . Now if both A A A and B B B are abelian, then [ 1 , V ] N ( L k ) ≅ [ 1 , V ′ ] N ( L m ′ ) [1,V]_{\mathcal{N}(L_k)} \cong[1,V']_{\mathcal{N}(L'_m)} [ 1 , V ] N ( L k ) ≅ [ 1 , V ′ ] N ( L m ′ ) so they contain the same number of atoms and hence,
1 + ∣ D ∣ + ⋯ + ∣ D ∣ k − 1 = ( k 1 ) ∣ D ∣ = ( m 1 ) ∣ D ′ ∣ = 1 + ∣ D ′ ∣ + ⋯ + ∣ D ′ ∣ m − 1 , 1+|D|+\cdots+|D|^{k-1}=\binom{k}{1}_{|D|}=\binom{m}{1}_{|D'|}=1+|D'|+\cdots+|D'|^{m-1}, 1 + ∣ D ∣ + ⋯ + ∣ D ∣ k − 1 = ( 1 k ) ∣ D ∣ = ( 1 m ) ∣ D ′ ∣ = 1 + ∣ D ′ ∣ + ⋯ + ∣ D ′ ∣ m − 1 ,
where D = End F p H ( A ) D=\operatorname{End}_{\mathbb{F}_p H}(A) D = End F p H ( A ) and D ′ = End F p ′ H ′ ( B ) D'=\operatorname{End}_{\mathbb{F}_{p'} H'}(B) D ′ = End F p ′ H ′ ( B ) . As k = m ≥ 2 k=m\ge2 k = m ≥ 2 , the sum 1 + q + ⋯ + q k − 1 1+q+\cdots+q^{k-1} 1 + q + ⋯ + q k − 1 is strictly increasing in q q q , and hence ∣ D ∣ = ∣ D ′ ∣ |D|=|D'| ∣ D ∣ = ∣ D ′ ∣ . So, ∣ End F p H ( A ) ∣ = ∣ End F p ′ H ′ ( B ) ∣ |\operatorname{End}_{\mathbb{F}_p H}(A)|=|\operatorname{End}_{\mathbb{F}_{p'} H'}(B)| ∣ End F p H ( A ) ∣ = ∣ End F p ′ H ′ ( B ) ∣ .
Conversely, assume that conditions ( a ) (a) ( a ) through ( d ) (d) ( d ) hold. Then clearly [ V , L k ] N ( L k ) ≅ [ V ′ , L m ′ ] N ( L m ′ ) [V,L_k]_{\mathcal{N}(L_k)} \cong[V',L'_m]_{\mathcal{N}(L'_m)} [ V , L k ] N ( L k ) ≅ [ V ′ , L m ′ ] N ( L m ′ ) . Now if both A A A and B B B are non-abelian then
[ 1 , V ] N ( L k ) ≅ P ( { 1 , ⋯ , k } ) and [ 1 , V ′ ] N ( L m ′ ) ≅ P ( { 1 , ⋯ , m } ) . [1,V]_{\mathcal{N}(L_k)} \cong\mathcal{P}(\{1,\cdots,k\}) \quad\text{and}\quad[1,V']_{\mathcal{N}(L'_m)} \cong\mathcal{P}(\{1,\cdots,m\}). [ 1 , V ] N ( L k ) ≅ P ({ 1 , ⋯ , k }) and [ 1 , V ′ ] N ( L m ′ ) ≅ P ({ 1 , ⋯ , m }) .
But k = m k=m k = m and hence [ 1 , V ] N ( L k ) ≅ [ 1 , V ′ ] N ( L m ′ ) [1,V]_{\mathcal{N}(L_k)} \cong[1,V']_{\mathcal{N}(L'_m)} [ 1 , V ] N ( L k ) ≅ [ 1 , V ′ ] N ( L m ′ ) . Now if both A A A and B B B are abelian then,
[ 1 , V ] N ( L k ) ≅ Sub D ( D k ) and [ 1 , V ′ ] N ( L m ′ ) ≅ Sub D ′ ( D ′ k ) . [1,V]_{\mathcal{N}(L_k)} \cong\operatorname{Sub}_{D}(D^k) \quad\text{and}\quad[1,V']_{\mathcal{N}(L'_m)} \cong\operatorname{Sub}_{D'}(D'^k). [ 1 , V ] N ( L k ) ≅ Sub D ( D k ) and [ 1 , V ′ ] N ( L m ′ ) ≅ Sub D ′ ( D ′ k ) .
But k = m k=m k = m and ∣ D ∣ = ∣ D ′ ∣ = q |D|=|D'|=q ∣ D ∣ = ∣ D ′ ∣ = q (say). By Wedderburn’s little theorem, the finite division rings D D D and D ′ D' D ′ are fields. As, up to isomorphism, there is a unique field with q q q elements, it follows that
Let φ : D → D ′ \varphi:D\to D' φ : D → D ′ be a field isomorphism. As k = m k=m k = m , this field isomorphism induces a bijection φ ~ : D k → D ′ k \widetilde{\varphi}:D^k\to D'^k φ : D k → D ′ k given by ( d 1 , ⋯ , d k ) ↦ ( φ ( d 1 ) , ⋯ , φ ( d k ) ) (d_1,\cdots,d_k)\mapsto(\varphi(d_1),\cdots,\varphi(d_k)) ( d 1 , ⋯ , d k ) ↦ ( φ ( d 1 ) , ⋯ , φ ( d k )) . Now let U ≤ D k U\leq D^k U ≤ D k be a D D D -subspace and u = ( u 1 , ⋯ , u k ) ∈ U u=(u_1,\cdots,u_k)\in U u = ( u 1 , ⋯ , u k ) ∈ U . Then, if φ ( d ) = d ′ \varphi(d)=d' φ ( d ) = d ′ we have,
φ ~ ( u d ) = ( φ ( u 1 ) φ ( d ) , ⋯ , φ ( u k ) φ ( d ) ) = ( φ ( u 1 ) d ′ , ⋯ , φ ( u k ) d ′ ) = φ ~ ( u ) d ′ . \widetilde{\varphi}(ud)=(\varphi(u_1)\varphi(d),\cdots,\varphi(u_k)\varphi(d))=(\varphi(u_1)d',\cdots,\varphi(u_k)d')=\widetilde{\varphi}(u)d'. φ ( u d ) = ( φ ( u 1 ) φ ( d ) , ⋯ , φ ( u k ) φ ( d )) = ( φ ( u 1 ) d ′ , ⋯ , φ ( u k ) d ′ ) = φ ( u ) d ′ .
As u d ∈ U ud\in U u d ∈ U it follows that φ ~ ( u ) d ′ ∈ φ ~ ( U ) \widetilde{\varphi}(u)d'\in\widetilde{\varphi}(U) φ ( u ) d ′ ∈ φ ( U ) . As φ \varphi φ preserves addition, φ ~ \widetilde{\varphi} φ also preserves addition and hence φ ~ ( U ) \widetilde{\varphi}(U) φ ( U ) is a D ′ D' D ′ -subspace of D ′ k D'^k D ′ k . So, we obtain a map ψ : Sub D ( D k ) → Sub D ′ ( D ′ k ) \psi:\operatorname{Sub}_{D}(D^k)\to\operatorname{Sub}_{D'}(D'^k) ψ : Sub D ( D k ) → Sub D ′ ( D ′ k ) defined as U ↦ φ ~ ( U ) U\mapsto\widetilde{\varphi}(U) U ↦ φ ( U ) . This is an isomorphism as sums and intersections are preserved, and its inverse is ψ − 1 : U ′ ↦ φ ~ − 1 ( U ′ ) \psi^{-1}:U'\mapsto\widetilde{\varphi}^{-1}(U') ψ − 1 : U ′ ↦ φ − 1 ( U ′ ) . Thus,
[ 1 , V ] N ( L k ) ≅ [ 1 , V ′ ] N ( L m ′ ) . [1,V]_{\mathcal{N}(L_k)}\cong[1,V']_{\mathcal{N}(L'_m)}. [ 1 , V ] N ( L k ) ≅ [ 1 , V ′ ] N ( L m ′ ) .
Now, both the lower and upper intervals of N ( L k ) \mathcal{N}(L_k) N ( L k ) and N ( L m ′ ) \mathcal{N}(L'_m) N ( L m ′ ) are isomorphic. Both isomorphisms send V V V to V ′ V' V ′ , and by Theorem 6.1 (1) every normal subgroup of L k L_k L k (resp. L m ′ L'_m L m ′ ) is comparable with V V V (resp. V ′ V' V ′ ). So the two isomorphisms glue to a bijection N ( L k ) → N ( L m ′ ) \mathcal{N}(L_k)\to\mathcal{N}(L'_m) N ( L k ) → N ( L m ′ ) which preserves inclusion in both directions, and hence it is a lattice isomorphism.
For (2), first suppose that N M ( L k ) ≅ N M ( L m ′ ) \mathcal{N}^{\mathcal{M}}(L_k)\cong\mathcal{N}^{\mathcal{M}}(L'_m) N M ( L k ) ≅ N M ( L m ′ ) . As V V V is the join of the atoms of N M ( L k ) \mathcal{N}^{\mathcal{M}}(L_k) N M ( L k ) , and similarly for V ′ V' V ′ , the isomorphism maps V V V to V ′ V' V ′ . Then by Corollary 6.2 we have
[ V , L k ] N M ( L k ) ≅ [ V ′ , L m ′ ] N M ( L m ′ ) [V,L_k]_{\mathcal{N}^{\mathcal{M}}(L_k)}\cong[V',L'_m]_{\mathcal{N}^{\mathcal{M}}(L'_m)} [ V , L k ] N M ( L k ) ≅ [ V ′ , L m ′ ] N M ( L m ′ )
and hence N ( H ) ≅ N ( H ′ ) \mathcal{N}(H)\cong\mathcal{N}(H') N ( H ) ≅ N ( H ′ ) . Also, the intervals below V V V and V ′ V' V ′ are isomorphic, implying they have the same number of atoms. So, a ( L , k ) = a ( L ′ , m ) a(L,k)=a(L',m) a ( L , k ) = a ( L ′ , m ) . Conversely if N ( H ) ≅ N ( H ′ ) \mathcal{N}(H)\cong\mathcal{N}(H') N ( H ) ≅ N ( H ′ ) and a ( L , k ) = a ( L ′ , m ) a(L,k)=a(L',m) a ( L , k ) = a ( L ′ , m ) then [ V , L k ] N M ( L k ) ≅ [ V ′ , L m ′ ] N M ( L m ′ ) [V,L_k]_{\mathcal{N}^{\mathcal{M}}(L_k)}\cong[V',L'_m]_{\mathcal{N}^{\mathcal{M}}(L'_m)} [ V , L k ] N M ( L k ) ≅ [ V ′ , L m ′ ] N M ( L m ′ ) . Also, as the interval below V V V and V ′ V' V ′ is determined completely by the number of atoms, it follows that [ 1 , V ] N M ( L k ) ≅ [ 1 , V ′ ] N M ( L m ′ ) [1,V]_{\mathcal{N}^{\mathcal{M}}(L_k)}\cong[1,V']_{\mathcal{N}^{\mathcal{M}}(L'_m)} [ 1 , V ] N M ( L k ) ≅ [ 1 , V ′ ] N M ( L m ′ ) . So if we glue these two isomorphisms at the common socle, we get an extended isomorphism to N M ( L k ) ≅ N M ( L m ′ ) \mathcal{N}^{\mathcal{M}}(L_k)\cong\mathcal{N}^{\mathcal{M}}(L'_m) N M ( L k ) ≅ N M ( L m ′ ) .
Now let T T T be a non-abelian finite simple group. Then T T T is monolithic with Soc ( T ) = T \operatorname{Soc}(T)=T Soc ( T ) = T and H = T / T = 1 H=T/T=1 H = T / T = 1 . So the crown-based power of T T T of size 3 is T 3 T^3 T 3 . As T T T is non-abelian, we get the normal subgroup interval below the socle is [ 1 , T 3 ] N ( T 3 ) ≅ P ( { 1 , 2 , 3 } ) [1,T^3]_{\mathcal{N}(T^3)}\cong\mathcal{P}(\{1,2,3\}) [ 1 , T 3 ] N ( T 3 ) ≅ P ({ 1 , 2 , 3 }) . So T 3 T^3 T 3 has exactly three atoms. Now observe that C 2 2 ≅ F 2 2 C_2^2\cong\mathbb{F}_2^2 C 2 2 ≅ F 2 2 . So its subgroups are precisely its subspaces, and hence the three one-dimensional subspaces are precisely the three atoms. So, by (2), as a ( T , 3 ) = 3 = a ( C 2 , 2 ) a(T,3)=3=a(C_2,2) a ( T , 3 ) = 3 = a ( C 2 , 2 ) and H = 1 H=1 H = 1 in both cases,
N M ( T 3 ) ≅ N M ( C 2 2 ) . \mathcal{N}^{\mathcal{M}}(T^3)\cong\mathcal{N}^{\mathcal{M}}(C_2^2). N M ( T 3 ) ≅ N M ( C 2 2 ) .
However, the full normal lattice of T 3 T^3 T 3 contains intermediate normal subgroups: T 1 T 2 , T 2 T 3 , T 1 T 3 T_1T_2,T_2T_3,T_1T_3 T 1 T 2 , T 2 T 3 , T 1 T 3 that are not M \mathcal{M} M -subgroups. As [ 1 , V ] N ( L k ) ≅ P ( { 1 , 2 , 3 } ) [1,V]_{\mathcal{N}(L_k)}\cong\mathcal{P}(\{1,2,3\}) [ 1 , V ] N ( L k ) ≅ P ({ 1 , 2 , 3 }) , ∣ N ( T 3 ) ∣ = 8 |\mathcal{N}(T^3)|=8 ∣ N ( T 3 ) ∣ = 8 . But in C 2 2 C_2^2 C 2 2 , as the dimension is 2, there is no subspace strictly between a line and the whole space. Hence, ∣ N ( C 2 2 ) ∣ = 1 + 3 + 1 = 5 |\mathcal{N}(C_2^2)|=1+3+1=5 ∣ N ( C 2 2 ) ∣ = 1 + 3 + 1 = 5 . So,
N ( T 3 ) ≇ N ( C 2 2 ) . \mathcal{N}(T^3)\not\cong\mathcal{N}(C_2^2). N ( T 3 ) ≅ N ( C 2 2 ) .
Now consider the groups C 2 C_2 C 2 and C 5 C_5 C 5 . Observe that Soc ( C 2 ) = C 2 \operatorname{Soc}(C_2)=C_2 Soc ( C 2 ) = C 2 and Soc ( C 5 ) = C 5 \operatorname{Soc}(C_5)=C_5 Soc ( C 5 ) = C 5 . So, it follows that
H = C 2 / Soc ( C 2 ) = 1 and H ′ = C 5 / Soc ( C 5 ) = 1. H=C_2/\operatorname{Soc}(C_2)=1 \quad\text{and}\quad H'=C_5/\operatorname{Soc}(C_5)=1. H = C 2 / Soc ( C 2 ) = 1 and H ′ = C 5 / Soc ( C 5 ) = 1.
Moreover, ( C 2 ) 5 = C 2 5 (C_2)_5=C_2^5 ( C 2 ) 5 = C 2 5 and ( C 5 ) 3 = C 5 3 (C_5)_3=C_5^3 ( C 5 ) 3 = C 5 3 . So both the socles of C 2 5 C_2^5 C 2 5 and C 5 3 C_5^3 C 5 3 are abelian. Now, the 1-dimensional subspaces of F 2 5 \mathbb{F}_2^5 F 2 5 are:
[ 5 1 ] 2 = 2 5 − 1 2 − 1 = 31 \begin{bmatrix}5\\1\end{bmatrix}_2=\frac{2^5-1}{2-1}=31 [ 5 1 ] 2 = 2 − 1 2 5 − 1 = 31
and the 1-dimensional subspaces of F 5 3 \mathbb{F}_5^3 F 5 3 are
[ 3 1 ] 5 = 5 3 − 1 5 − 1 = 31. \begin{bmatrix}3\\1\end{bmatrix}_5=\frac{5^3-1}{5-1}=31. [ 3 1 ] 5 = 5 − 1 5 3 − 1 = 31.
Thus the M \mathcal{M} M -lattice of both C 2 5 C_2^5 C 2 5 and C 5 3 C_5^3 C 5 3 has height 2 with the same number of atoms. Hence,
N M ( C 2 5 ) ≅ N M ( C 5 3 ) . \mathcal{N}^{\mathcal{M}}(C_2^5)\cong\mathcal{N}^{\mathcal{M}}(C_5^3). N M ( C 2 5 ) ≅ N M ( C 5 3 ) .
However, the normal subgroup interval below the socle of C 2 5 C_2^5 C 2 5 contains proper non-zero subspaces of dimensions 1, 2, 3 and 4, whereas the normal subgroup interval below the socle of C 5 3 C_5^3 C 5 3 contains proper non-zero subspaces of dimensions 1 and 2 only. Thus, N ( C 2 5 ) \mathcal{N}(C_2^5) N ( C 2 5 ) has a height of 5 and N ( C 5 3 ) \mathcal{N}(C_5^3) N ( C 5 3 ) has a height of 3. Hence
N ( C 2 5 ) ≇ N ( C 5 3 ) . \mathcal{N}(C_2^5)\not\cong\mathcal{N}(C_5^3). N ( C 2 5 ) ≅ N ( C 5 3 ) .
The quotient of a crown-based power by a coordinate minimal normal subgroup is naturally the crown-based power with one fewer coordinate, a standard feature of the construction [8 ] . We extend this observation to every normal subgroup contained in the socle, and also describe the automorphism orbits of such kernels and their enumerating polynomial. The answer retains the Boolean–Grassmann dichotomy of the lower normal lattice.
Theorem 6.6. Let K ⊴ L k K\trianglelefteq L_k K ⊴ L k with K ≤ V K\leq V K ≤ V , put
r = λ L k ( K ) , m = k − r , r=\lambda_{L_k}(K),\qquad m=k-r, r = λ L k ( K ) , m = k − r , and, in the abelian case, put q = ∣ D ∣ q=|D| q = ∣ D ∣ .
(1) If m = 0 m=0 m = 0 , then L k / K ≅ H L_k/K\cong H L k / K ≅ H . If m ≥ 1 m\geq1 m ≥ 1 , then
L k / K ≅ L m . L_k/K\cong L_m. L k / K ≅ L m . (2) For each r ∈ { 0 , … , k } r\in\{0,\ldots,k\} r ∈ { 0 , … , k } , the normal subgroups K ≤ V K\leq V K ≤ V of length r r r form a single orbit under Aut ( L k ) \operatorname{Aut}(L_k) Aut ( L k ) .
(3) The number of kernels K ≤ V K\leq V K ≤ V for which λ L k / K ( V / K ) = m \lambda_{L_k/K}(V/K)=m λ L k / K ( V / K ) = m is
{ ( k m ) , A non-abelian , [ k m ] q , A abelian . \begin{cases}
\binom{k}{m}, & \text{A non-abelian},\\
\begin{bmatrix}k\\m\end{bmatrix}_q, & \text{A abelian}.
\end{cases} ⎩ ⎨ ⎧ ( m k ) , [ k m ] q , A non-abelian , A abelian .
Equivalently, the socle-quotient enumerator
Ψ L , k ( t ) : = ∑ K ⊲ L k K ≤ V t λ L k / K ( V / K ) \Psi_{L,k}(t) := \sum_{\substack{K\lhd L_k\\K\leq V}} t^{\lambda_{L_k/K}(V/K)} Ψ L , k ( t ) := K ⊲ L k K ≤ V ∑ t λ L k / K ( V / K )
is
Ψ L , k ( t ) = { ( 1 + t ) k , A non-abelian , ∑ m = 0 k ( k m ) q t m , A abelian . \Psi_{L,k}(t) =
\begin{cases}
(1+t)^k, & A\text{ non-abelian},\\
\sum_{m=0}^{k} \binom{k}{m}_{q} t^m, & A\text{ abelian}.
\end{cases} Ψ L , k ( t ) = { ( 1 + t ) k , ∑ m = 0 k ( m k ) q t m , A non-abelian , A abelian .
(4) Every homomorphic image of L k L_k L k is isomorphic either to L m L_m L m for some 1 ≤ m ≤ k 1 \leq m \leq k 1 ≤ m ≤ k , or to a quotient of H H H .
In particular, the normal subgroups contained in V V V form exactly k + 1 k+1 k + 1 Aut ( L k ) \operatorname{Aut}(L_k) Aut ( L k ) -orbits, and their quotients have the k + 1 k+1 k + 1 distinct isomorphism types
H , L 1 , … , L k . H,L_1,\ldots,L_k. H , L 1 , … , L k .
Proof. (1) If m = 0 m=0 m = 0 , then K = V K=V K = V , as K ≤ V K\leq V K ≤ V and λ L k ( K ) = k = λ L k ( V ) \lambda_{L_k}(K)=k=\lambda_{L_k}(V) λ L k ( K ) = k = λ L k ( V ) . So L k / K = L k / V ≅ H L_k/K=L_k/V\cong H L k / K = L k / V ≅ H . Now let m ≥ 1 m\geq1 m ≥ 1 . First assume A A A is non-abelian. By Theorem 6.1 , K = A I K=A_I K = A I for a unique subset I ⊆ { 1 , … , k } I\subseteq\{1,\ldots,k\} I ⊆ { 1 , … , k } and r = ∣ I ∣ r=|I| r = ∣ I ∣ . Let J = { 1 , … , k } ∖ I J=\{1,\ldots,k\}\setminus I J = { 1 , … , k } ∖ I , so ∣ J ∣ = m |J|=m ∣ J ∣ = m , and let π J : L k → L J \pi_J:L_k\to L^J π J : L k → L J be the projection onto the J J J -coordinates. As g 1 A = ⋯ = g k A g_1A=\cdots=g_kA g 1 A = ⋯ = g k A for every ( g 1 , … , g k ) ∈ L k (g_1,\ldots,g_k)\in L_k ( g 1 , … , g k ) ∈ L k , and as any ( g j ) j ∈ J (g_j)_{j\in J} ( g j ) j ∈ J with all g j A g_jA g j A equal extends to an element of L k L_k L k by putting g i = g j 0 g_i=g_{j_0} g i = g j 0 for i ∈ I i\in I i ∈ I , for a fixed j 0 ∈ J j_0\in J j 0 ∈ J , relabelling the coordinates in J J J as 1 1 1 to m m m we get
im π J = { ( g j ) j ∈ J : g j A is independent of j } ≅ L m . \operatorname{im}\pi_J=\{(g_j)_{j\in J}:g_jA\text{ is independent of }j\}\cong L_m. im π J = {( g j ) j ∈ J : g j A is independent of j } ≅ L m .
If ( g 1 , … , g k ) ∈ ker π J (g_1,\ldots,g_k)\in\ker\pi_J ( g 1 , … , g k ) ∈ ker π J , then g j = 1 g_j=1 g j = 1 for j ∈ J j\in J j ∈ J , and so g i A = g j A = A g_iA=g_jA=A g i A = g j A = A , i.e., g i ∈ A g_i\in A g i ∈ A for all i i i . Thus ker π J ≤ A I \ker\pi_J\leq A_I ker π J ≤ A I , and clearly A I ≤ ker π J A_I\leq\ker\pi_J A I ≤ ker π J . So ker π J = K \ker\pi_J=K ker π J = K and L k / K ≅ L m L_k/K\cong L_m L k / K ≅ L m .
Now assume that A A A is abelian. Recall that in the proof of Theorem 6.1 (1), we had shown that L = A M L=AM L = A M with A ∩ M = 1 A\cap M=1 A ∩ M = 1 (as A ≤ C A\leq C A ≤ C and C ∩ M = 1 C\cap M=1 C ∩ M = 1 there), where M M M is a core-free maximal subgroup of L L L . So M ≅ L / A = H M\cong L/A=H M ≅ L / A = H and L = A ⋊ H L=A\rtimes H L = A ⋊ H . Identifying H H H with its diagonal copy in L k L^k L k , we get L k = A k ⋊ H L_k=A^k\rtimes H L k = A k ⋊ H , with H H H acting diagonally on A k A^k A k . As K ⊲ L k K\lhd L_k K ⊲ L k , K K K is H H H -invariant and
L k / K ≅ ( A k / K ) ⋊ H . L_k/K\cong(A^k/K)\rtimes H. L k / K ≅ ( A k / K ) ⋊ H .
As A k A^k A k is semisimple, A k = K ⊕ W A^k=K\oplus W A k = K ⊕ W for some submodule W ≅ A k / K W\cong A^k/K W ≅ A k / K , and K K K , W W W are semisimple. As all composition factors of A k A^k A k are isomorphic to A A A , and K K K , W W W have composition lengths r r r and k − r k-r k − r , K ≅ A r K\cong A^r K ≅ A r and A k / K ≅ A k − r = A m A^k/K\cong A^{k-r}=A^m A k / K ≅ A k − r = A m as F p H \mathbb{F}_pH F p H -modules. Now if T : A k / K → A m T:A^k/K\to A^m T : A k / K → A m is an F p H \mathbb{F}_pH F p H -isomorphism, then T ~ : ( x , h ) ↦ ( T ( x ) , h ) \widetilde{T}:(x,h)\mapsto(T(x),h) T : ( x , h ) ↦ ( T ( x ) , h ) is a bijection from ( A k / K ) ⋊ H (A^k/K)\rtimes H ( A k / K ) ⋊ H onto A m ⋊ H A^m\rtimes H A m ⋊ H , and as T ( h ⋅ x ) = h ⋅ T ( x ) T(h\cdot x)=h\cdot T(x) T ( h ⋅ x ) = h ⋅ T ( x ) ,
T ~ ( ( x 1 , h 1 ) ( x 2 , h 2 ) ) = ( T ( x 1 ) + h 1 ⋅ T ( x 2 ) , h 1 h 2 ) = T ~ ( ( x 1 , h 1 ) ) T ~ ( ( x 2 , h 2 ) ) . \widetilde{T}((x_1,h_1)(x_2,h_2))=(T(x_1)+h_1\cdot T(x_2),h_1h_2)=\widetilde{T}((x_1,h_1))\widetilde{T}((x_2,h_2)). T (( x 1 , h 1 ) ( x 2 , h 2 )) = ( T ( x 1 ) + h 1 ⋅ T ( x 2 ) , h 1 h 2 ) = T (( x 1 , h 1 )) T (( x 2 , h 2 )) .
Thus we get
L k / K ≅ ( A k / K ) ⋊ H ≅ A m ⋊ H ≅ L m . L_k/K\cong(A^k/K)\rtimes H\cong A^m\rtimes H\cong L_m. L k / K ≅ ( A k / K ) ⋊ H ≅ A m ⋊ H ≅ L m .
(2) First assume A A A is non-abelian. Every σ ∈ S k \sigma\in S_k σ ∈ S k induces an automorphism
φ σ : ( g 1 , … , g k ) ↦ ( g σ − 1 ( 1 ) , … , g σ − 1 ( k ) ) \varphi_\sigma:(g_1,\ldots,g_k)\mapsto(g_{\sigma^{-1}(1)},\ldots,g_{\sigma^{-1}(k)}) φ σ : ( g 1 , … , g k ) ↦ ( g σ − 1 ( 1 ) , … , g σ − 1 ( k ) )
of L k L_k L k , and φ σ ( A I ) = A σ ( I ) \varphi_\sigma(A_I)=A_{\sigma(I)} φ σ ( A I ) = A σ ( I ) . So if K 1 = A I K_1=A_I K 1 = A I and K 2 = A J K_2=A_J K 2 = A J with ∣ I ∣ = ∣ J ∣ = r |I|=|J|=r ∣ I ∣ = ∣ J ∣ = r , then choosing σ ∈ S k \sigma\in S_k σ ∈ S k with σ ( I ) = J \sigma(I)=J σ ( I ) = J , we get φ σ ( K 1 ) = K 2 \varphi_\sigma(K_1)=K_2 φ σ ( K 1 ) = K 2 .
Now suppose A A A is abelian. Any H H H -endomorphism of A k A^k A k can be represented by a k × k k\times k k × k -matrix over D D D , where the ( i , j ) (i,j) ( i , j ) -entry is the H H H -map from the j j j th copy of A A A to the i i i th copy of A A A . So
Aut H ( A k ) ≅ GL k ( D ) . \operatorname{Aut}_H(A^k)\cong\operatorname{GL}_k(D). Aut H ( A k ) ≅ GL k ( D ) .
As Hom F p H ( A , T ( K ) ) = T ∘ Hom F p H ( A , K ) \operatorname{Hom}_{\mathbb{F}_pH}(A,T(K))=T\circ\operatorname{Hom}_{\mathbb{F}_pH}(A,K) Hom F p H ( A , T ( K )) = T ∘ Hom F p H ( A , K ) , under the identification of Theorem 6.1 (4), each T ∈ Aut H ( A k ) T\in\operatorname{Aut}_H(A^k) T ∈ Aut H ( A k ) acts on the D D D -subspaces of D k D^k D k through its matrix. Now let K , K ′ ≤ V K,K'\leq V K , K ′ ≤ V be normal subgroups of length r r r , and let U , U ′ U,U' U , U ′ be the corresponding r r r -dimensional D D D -subspaces of D k D^k D k . Extending bases { u 1 , … , u r } \{u_1,\ldots,u_r\} { u 1 , … , u r } of U U U and { u 1 ′ , … , u r ′ } \{u'_1,\ldots,u'_r\} { u 1 ′ , … , u r ′ } of U ′ U' U ′ to bases of D k D^k D k , we get a D D D -linear automorphism of D k D^k D k with u i ↦ u i ′ u_i \mapsto u'_i u i ↦ u i ′ for 1 ≤ i ≤ r 1 \le i \le r 1 ≤ i ≤ r , which maps U U U onto U ′ U' U ′ . So there exists T ∈ Aut H ( A k ) T \in\operatorname{Aut}_H(A^k) T ∈ Aut H ( A k ) with T ( K ) = K ′ T(K)=K' T ( K ) = K ′ . By the same computation as in (1), T ~ : ( l , h ) ↦ ( T ( l ) , h ) \widetilde{T}:(l,h)\mapsto(T(l),h) T : ( l , h ) ↦ ( T ( l ) , h ) is an automorphism of L k = A k ⋊ H L_k=A^k\rtimes H L k = A k ⋊ H , and T ~ ( K ) = K ′ \widetilde{T}(K)=K' T ( K ) = K ′ . So in both cases, all normal subgroups K ≤ V K \le V K ≤ V of length r r r form a single orbit under Aut ( L k ) \operatorname{Aut}(L_k) Aut ( L k ) .
(3) By the correspondence theorem, [ 1 , V / K ] N ( L k / K ) ≅ [ K , V ] N ( L k ) [1,V/K]_{\mathcal{N}(L_k/K)} \cong[K,V]_{\mathcal{N}(L_k)} [ 1 , V / K ] N ( L k / K ) ≅ [ K , V ] N ( L k ) , and as N ( L k ) \mathcal{N}(L_k) N ( L k ) is modular, so
λ L k / K ( V / K ) = λ L k ( V ) − λ L k ( K ) = k − r . \lambda_{L_k/K}(V/K)=\lambda_{L_k}(V)-\lambda_{L_k}(K)=k-r. λ L k / K ( V / K ) = λ L k ( V ) − λ L k ( K ) = k − r .
So the kernels K ≤ V K \le V K ≤ V with λ L k / K ( V / K ) = m \lambda_{L_k/K}(V/K)=m λ L k / K ( V / K ) = m are precisely the normal subgroups K ≤ V K \le V K ≤ V with λ L k ( K ) = k − m \lambda_{L_k}(K)=k-m λ L k ( K ) = k − m . If A A A is non-abelian, then by Theorem 6.1 these are the A I A_I A I with ∣ I ∣ = k − m |I|=k-m ∣ I ∣ = k − m , and so their number is ( k k − m ) = ( k m ) \binom{k}{k-m}=\binom{k}{m} ( k − m k ) = ( m k ) . If A A A is abelian, then they correspond to the ( k − m ) (k-m) ( k − m ) -dimensional D D D -subspaces of D k D^k D k , and so their number is [ k k − m ] q = [ k m ] q \left[\begin{smallmatrix}k\\k-m\end{smallmatrix}\right]_q=\left[\begin{smallmatrix}k\\m\end{smallmatrix}\right]_q [ k k − m ] q = [ k m ] q , by the symmetry of the Gaussian binomial coefficients. Hence,
Ψ L , k ( t ) = ∑ m = 0 k # { K ⊴ L k : K ≤ V , λ L k / K ( V / K ) = m } t m = { ∑ m = 0 k ( k m ) t m = ( 1 + t ) k , A is non-abelian , ∑ m = 0 k [ k m ] q t m , A is abelian . \begin{aligned}
\Psi_{L,k}(t)=\sum_{m=0}^{k}\#\{K\trianglelefteq L_k:K\le V,\ \lambda_{L_k/K}(V/K)=m\}t^m \\
&=\begin{cases}
\sum_{m=0}^{k}\binom{k}{m}t^m=(1+t)^k,& A\text{ is non-abelian},\\
\sum_{m=0}^{k}\left[\begin{matrix}k\\m\end{matrix}\right]_q t^m,& A\text{ is abelian}.
\end{cases}
\end{aligned} Ψ L , k ( t ) = m = 0 ∑ k # { K ⊴ L k : K ≤ V , λ L k / K ( V / K ) = m } t m = ⎩ ⎨ ⎧ ∑ m = 0 k ( m k ) t m = ( 1 + t ) k , ∑ m = 0 k [ k m ] q t m , A is non-abelian , A is abelian .
(4) Let π : L k ↠ Q \pi:L_k\twoheadrightarrow Q π : L k ↠ Q be a surjective homomorphism and N = ker π N=\ker\pi N = ker π , so Q ≅ L k / N Q\cong L_k/N Q ≅ L k / N . By Theorem 6.1 , either N ≤ V N\le V N ≤ V or V ≤ N V\le N V ≤ N . If N ≤ V N\le V N ≤ V , then by (1), Q ≅ H Q\cong H Q ≅ H or Q ≅ L m Q\cong L_m Q ≅ L m with m = k − λ L k ( N ) ≥ 1 m=k-\lambda_{L_k}(N)\ge1 m = k − λ L k ( N ) ≥ 1 . If V ≤ N V\le N V ≤ N , then by the correspondence theorem N / V ⊴ L k / V ≅ H N/V\trianglelefteq L_k/V\cong H N / V ⊴ L k / V ≅ H , and hence
Q ≅ L k / N ≅ ( L k / V ) / ( N / V ) Q\cong L_k/N\cong(L_k/V)/(N/V) Q ≅ L k / N ≅ ( L k / V ) / ( N / V )
is a quotient of H H H .
Finally, every automorphism of L k L_k L k induces an automorphism of N ( L k ) \mathcal{N}(L_k) N ( L k ) , and so it preserves λ L k \lambda_{L_k} λ L k . Also, A 1 ⋯ A r ≤ V A_1\cdots A_r\le V A 1 ⋯ A r ≤ V is a normal subgroup of length r r r for each r ∈ { 0 , … , k } r\in\{0,\ldots,k\} r ∈ { 0 , … , k } . So by (2), the normal subgroups contained in V V V form exactly k + 1 k+1 k + 1 Aut ( L k ) \operatorname{Aut}(L_k) Aut ( L k ) -orbits. As A ≠ 1 A\ne1 A = 1 , for m ≠ n m\ne n m = n we have
∣ L m ∣ = ∣ A ∣ m ∣ H ∣ ≠ ∣ A ∣ n ∣ H ∣ = ∣ L n ∣ , |L_m|=|A|^m|H|\ne|A|^n|H|=|L_n|, ∣ L m ∣ = ∣ A ∣ m ∣ H ∣ = ∣ A ∣ n ∣ H ∣ = ∣ L n ∣ ,
and ∣ L m ∣ > ∣ H ∣ |L_m|>|H| ∣ L m ∣ > ∣ H ∣ for m ≥ 1 m\ge1 m ≥ 1 . So by (1), the quotients coming from the k + 1 k+1 k + 1 orbits are pairwise non-isomorphic, and they are exactly H , L 1 , … , L k H,L_1,\ldots,L_k H , L 1 , … , L k . □
For a finite group X X X and K ⊴ X K\trianglelefteq X K ⊴ X , we call the quotient map π K : X → X / K \pi_K:X\to X/K π K : X → X / K M \mathcal{M} M -closure compatible if
π K ( cl M X ( N ) ) = cl M X / K ( N / K ) ( K ≤ N ⊴ X ) . \pi_K\left(\operatorname{cl}_{\mathcal{M}}^X(N)\right)=\operatorname{cl}_{\mathcal{M}}^{X/K}(N/K)\qquad(K\le N\trianglelefteq X). π K ( cl M X ( N ) ) = cl M X / K ( N / K ) ( K ≤ N ⊴ X ) .
Proposition 6.7. Let K ⊴ L k K\trianglelefteq L_k K ⊴ L k . The quotient map π K \pi_K π K is M \mathcal{M} M -closure compatible if and only if one of the following holds:
K = 1 K=1 K = 1 ;
k = 2 k=2 k = 2 and K ∈ MinNorm ( L 2 ) K\in\operatorname{MinNorm}(L_2) K ∈ MinNorm ( L 2 ) ;
V ≤ K V\le K V ≤ K and
H / ( K / V ) ≅ L k / K H/(K/V)\cong L_k/K H / ( K / V ) ≅ L k / K is M \mathcal{M} M -full.
Thus, for k ≥ 3 k\ge3 k ≥ 3 , no quotient by a non-trivial proper subgroup of the socle is M \mathcal{M} M -closure compatible.
Proof. First assume π K \pi_K π K is M \mathcal{M} M -closure compatible. Taking N = K N=K N = K , we get
π K ( cl M L k ( K ) ) = cl M L k / K ( 1 ) = 1. \pi_K(\operatorname{cl}_{\mathcal{M}}^{L_k}(K))=\operatorname{cl}_{\mathcal{M}}^{L_k/K}(1)=1. π K ( cl M L k ( K )) = cl M L k / K ( 1 ) = 1.
As ker π K = K \ker\pi_K=K ker π K = K , cl M L k ( K ) ≤ K \operatorname{cl}_{\mathcal{M}}^{L_k}(K)\leq K cl M L k ( K ) ≤ K . But by definition K ≤ cl M L k ( K ) K\leq\operatorname{cl}_{\mathcal{M}}^{L_k}(K) K ≤ cl M L k ( K ) . Thus, K = cl M L k ( K ) K=\operatorname{cl}_{\mathcal{M}}^{L_k}(K) K = cl M L k ( K ) and hence K K K is an M \mathcal{M} M -subgroup. By Theorem 6.1 , either K < V K<V K < V or V ≤ K V\leq K V ≤ K .
Now let K < V K<V K < V . If k = 1 k=1 k = 1 , then V = A V=A V = A is the only minimal normal subgroup of L 1 = L L_1=L L 1 = L , and so K = 1 K=1 K = 1 . Suppose k ≥ 2 k\geq2 k ≥ 2 . Then by Corollary 6.2 , K = 1 K=1 K = 1 or K ∈ MinNorm ( L k ) K\in\operatorname{MinNorm}(L_k) K ∈ MinNorm ( L k ) . So let K ∈ MinNorm ( L k ) K\in\operatorname{MinNorm}(L_k) K ∈ MinNorm ( L k ) and suppose k ≥ 3 k\geq3 k ≥ 3 . Choose M ∈ MinNorm ( L k ) M\in\operatorname{MinNorm}(L_k) M ∈ MinNorm ( L k ) with M ≠ K M\ne K M = K and put N = M K N=MK N = M K . If A A A is non-abelian, then K = A i K=A_i K = A i and M = A j M=A_j M = A j for some i ≠ j i\ne j i = j , and so N = A { i , j } N=A_{\{i,j\}} N = A { i , j } . If A A A is abelian, then K K K and M M M correspond to distinct 1 1 1 -dimensional D D D -subspaces of D k D^k D k , and so N N N corresponds to a 2 2 2 -dimensional subspace. So in both cases λ L k ( N ) = 2 ≤ k − 1 \lambda_{L_k}(N)=2\leq k-1 λ L k ( N ) = 2 ≤ k − 1 . Hence N < V N<V N < V , and by Corollary 6.2 , cl M L k ( N ) = V \operatorname{cl}_{\mathcal{M}}^{L_k}(N)=V cl M L k ( N ) = V . Also, as λ L k ( K ) = 1 \lambda_{L_k}(K)=1 λ L k ( K ) = 1 and N ( L k ) \mathcal{N}(L_k) N ( L k ) is modular by Lemma 3.1 , there is no normal subgroup of L k L_k L k strictly between K K K and N N N . Thus, N / K ∈ MinNorm ( L k / K ) N/K\in\operatorname{MinNorm}(L_k/K) N / K ∈ MinNorm ( L k / K ) and hence cl M L k / K ( N / K ) = N / K \operatorname{cl}_{\mathcal{M}}^{L_k/K}(N/K)=N/K cl M L k / K ( N / K ) = N / K . So,
V / K = π K ( cl M L k ( N ) ) = cl M L k / K ( N / K ) = N / K . V/K=\pi_K(\operatorname{cl}_{\mathcal{M}}^{L_k}(N))=\operatorname{cl}_{\mathcal{M}}^{L_k/K}(N/K)=N/K. V / K = π K ( cl M L k ( N )) = cl M L k / K ( N / K ) = N / K .
By the correspondence theorem we get V = N V=N V = N , which is a contradiction as N < V N<V N < V . Thus, k = 2 k=2 k = 2 and K ∈ MinNorm ( L 2 ) K\in\operatorname{MinNorm}(L_2) K ∈ MinNorm ( L 2 ) .
Now let V ≤ K V\leq K V ≤ K , and let N / K ⊴ L k / K N/K\unlhd L_k/K N / K ⊴ L k / K with K ≤ N ⊴ L k K\leq N\unlhd L_k K ≤ N ⊴ L k . Then V ≤ N V\leq N V ≤ N , so N N N contains the socle and hence is an M \mathcal{M} M -subgroup. Therefore,
cl M L k / K ( N / K ) = π K ( cl M L k ( N ) ) = π K ( N ) = N / K . \operatorname{cl}_{\mathcal{M}}^{L_k/K}(N/K)=\pi_K(\operatorname{cl}_{\mathcal{M}}^{L_k}(N))=\pi_K(N)=N/K. cl M L k / K ( N / K ) = π K ( cl M L k ( N )) = π K ( N ) = N / K .
Thus L k / K L_k/K L k / K is M \mathcal{M} M -full, and
L k / K ≅ ( L k / V ) / ( K / V ) ≅ H / ( K / V ) . L_k/K\cong(L_k/V)/(K/V)\cong H/(K/V). L k / K ≅ ( L k / V ) / ( K / V ) ≅ H / ( K / V ) .
Conversely, if K = 1 K=1 K = 1 , then π 1 \pi_1 π 1 is just the identification of L k L_k L k with L k / 1 L_k/1 L k /1 , and so it is M \mathcal{M} M -closure compatible. Now assume (2) or (3) holds, and let K ≤ N ⊴ L k K\leq N\unlhd L_k K ≤ N ⊴ L k . In case (2), L 2 L_2 L 2 is M \mathcal{M} M -full by Corollary 6.3 , and L 2 / K ≅ L 1 L_2/K\cong L_1 L 2 / K ≅ L 1 by Theorem 6.6 , which is again M \mathcal{M} M -full by Corollary 6.3 . In case (3), V ≤ N V\leq N V ≤ N , so N N N is an M \mathcal{M} M -subgroup, and L k / K L_k/K L k / K is M \mathcal{M} M -full by our assumption. So in both cases cl M L k ( N ) = N \operatorname{cl}_{\mathcal{M}}^{L_k}(N)=N cl M L k ( N ) = N and cl M L k / K ( N / K ) = N / K \operatorname{cl}_{\mathcal{M}}^{L_k/K}(N/K)=N/K cl M L k / K ( N / K ) = N / K . Hence,
π K ( cl M L k ( N ) ) = π K ( N ) = N / K = cl M L k / K ( N / K ) \pi_K(\operatorname{cl}_{\mathcal{M}}^{L_k}(N))=\pi_K(N)=N/K=\operatorname{cl}_{\mathcal{M}}^{L_k/K}(N/K) π K ( cl M L k ( N )) = π K ( N ) = N / K = cl M L k / K ( N / K )
and the compatibility holds. Finally, if k ≥ 3 k\geq3 k ≥ 3 and 1 < K < V 1<K<V 1 < K < V , then none of (1)–(3) holds, and so π K \pi_K π K is not M \mathcal{M} M -closure compatible. □
Definition 6.8. A finite group X X X is said to be quotient-M \mathcal{M} M -full if every homomorphic image of X X X is M \mathcal{M} M -full.
The quotient classification makes this property completely recursive for crown-based powers.
Corollary 6.9. In the abelian case, put q = ∣ D ∣ q=|D| q = ∣ D ∣ . Let K ⊴ L k K\unlhd L_k K ⊴ L k with K < V K<V K < V , and put m = λ L k / K ( V / K ) m=\lambda_{L_k/K}(V/K) m = λ L k / K ( V / K ) . Then
L k / K is M -full ⟺ m ≤ 2. L_k/K\text{ is }\mathcal{M}\text{-full}\quad\Longleftrightarrow\quad m\leq2. L k / K is M -full ⟺ m ≤ 2.
The number of such kernels K < V K<V K < V is
{ k + ( k 2 ) , A non-abelian , [ k 1 ] q + [ k 2 ] q , A abelian . \begin{cases}
k+\binom{k}{2}, & A\text{ non-abelian},\\
\left[\begin{matrix}k\\1\end{matrix}\right]_q+\left[\begin{matrix}k\\2\end{matrix}\right]_q, & A\text{ abelian}.
\end{cases} ⎩ ⎨ ⎧ k + ( 2 k ) , [ k 1 ] q + [ k 2 ] q , A non-abelian , A abelian .
The kernel K = V K=V K = V gives one additional M \mathcal{M} M -full quotient exactly when H H H is M \mathcal{M} M -full. Finally,
L k is quotient- M -full ⟺ k ≤ 2 and H is quotient- M -full . L_k\text{ is quotient-}\mathcal{M}\text{-full}\quad\Longleftrightarrow\quad k\leq2\text{ and }H\text{ is quotient-}\mathcal{M}\text{-full}. L k is quotient- M -full ⟺ k ≤ 2 and H is quotient- M -full .
Proof. As K < V K < V K < V , m = λ L k / K ( V / K ) ≥ 1 m = \lambda_{L_k/K}(V/K) \ge1 m = λ L k / K ( V / K ) ≥ 1 . Also, by the proof of Theorem 6.6 (3), m = k − λ L k ( K ) m = k - \lambda_{L_k}(K) m = k − λ L k ( K ) . Now,
L k / K is M -full ⟺ L m is M -full ⟺ m ≤ 2 , L_k/K \text{ is } \mathcal{M}\text{-full} \Longleftrightarrow L_m \text{ is } \mathcal{M}\text{-full} \Longleftrightarrow m \le2, L k / K is M -full ⟺ L m is M -full ⟺ m ≤ 2 ,
where the first equivalence follows from Theorem 6.6 (1), as L k / K ≅ L m L_k/K \cong L_m L k / K ≅ L m , and the second follows from Corollary 6.3 . So the kernels K < V K < V K < V with L k / K L_k/K L k / K M \mathcal{M} M -full are precisely those with m ∈ { 1 , 2 } m \in\{1,2\} m ∈ { 1 , 2 } , and by Theorem 6.6 (3) their number is
{ ( k 1 ) + ( k 2 ) = k + ( k 2 ) , A is non-abelian , ( k 1 ) q + ( k 2 ) q , A is abelian . \begin{cases}
\binom{k}{1}+\binom{k}{2}=k+\binom{k}{2}, & A\text{ is non-abelian},\\
\binom{k}{1}_{q}+\binom{k}{2}_{q}, & A\text{ is abelian}.
\end{cases} { ( 1 k ) + ( 2 k ) = k + ( 2 k ) , ( 1 k ) q + ( 2 k ) q , A is non-abelian , A is abelian .
When K = V K = V K = V , L k / K = L k / V ≅ H L_k/K = L_k/V \cong H L k / K = L k / V ≅ H , and so L k / V L_k/V L k / V is M \mathcal{M} M -full if and only if H H H is M \mathcal{M} M -full. This gives exactly one additional M \mathcal{M} M -full quotient in our count.
Now suppose L k L_k L k is quotient-M \mathcal{M} M -full. Then in particular L k L_k L k is M \mathcal{M} M -full, and so k ≤ 2 k \le2 k ≤ 2 by Corollary 6.3 . As H H H is a homomorphic image of L k L_k L k , every quotient of H H H is also a quotient of L k L_k L k , and hence H H H is also quotient-M \mathcal{M} M -full.
Conversely, assume k ≤ 2 k \le2 k ≤ 2 and H H H is quotient-M \mathcal{M} M -full. By Theorem 6.6 (4), every homomorphic image of L k L_k L k is isomorphic either to L m L_m L m for some 1 ≤ m ≤ k ≤ 2 1 \le m \le k \le2 1 ≤ m ≤ k ≤ 2 , which is M \mathcal{M} M -full by Corollary 6.3 , or to a quotient of H H H , which is M \mathcal{M} M -full as H H H is quotient-M \mathcal{M} M -full. Thus, L k L_k L k is quotient-M \mathcal{M} M -full.
Everything proved in this section applies to the crowns of an arbitrary finite group. We use the notation of Detomi and Lucchini [10 ] .
Corollary 6.10. Let G G G be a finite group, let B B B be a non-Frattini chief factor of G G G , and let L B L_B L B be the associated monolithic primitive group. Write
A B = Soc ( L B ) , H B = L B / A B , A_B = \operatorname{Soc}(L_B), \qquad H_B = L_B/A_B, A B = Soc ( L B ) , H B = L B / A B , and let δ G ( B ) \delta_G(B) δ G ( B ) and R G ( B ) R_G(B) R G ( B ) be the crown multiplicity and crown kernel, respectively (see [10 Definitions 6 and 8] ). Define I G ( B ) I_G(B) I G ( B ) by
I G ( B ) / R G ( B ) = Soc ( G / R G ( B ) ) . I_G(B)/R_G(B) = \operatorname{Soc}(G/R_G(B)). I G ( B ) / R G ( B ) = Soc ( G / R G ( B )) . Then δ G ( B ) ≥ 1 \delta_G(B) \ge1 δ G ( B ) ≥ 1 and I G ( B ) / R G ( B ) I_G(B)/R_G(B) I G ( B ) / R G ( B ) is the B B B -crown by [10 Proposition 7] , and, by [10 Proposition 9] (see also [19 Section 5] ),
G / R G ( B ) ≅ ( L B ) δ G ( B ) . G/R_G(B) \cong(L_B)^{\delta_G(B)}. G / R G ( B ) ≅ ( L B ) δ G ( B ) . Consequently:
(1) the complete normal-subgroup lattice of G / R G ( B ) G/R_G(B) G / R G ( B ) is the vertical gluing of N ( H B ) \mathcal{N}(H_B) N ( H B ) with a Boolean lattice of rank δ G ( B ) \delta_G(B) δ G ( B ) when B B B is non-abelian, and with the subspace lattice of D B δ G ( B ) D_B^{\delta_G(B)} D B δ G ( B ) when B B B is abelian, with N ( H B ) \mathcal{N}(H_B) N ( H B ) on top and the two glued at I G ( B ) / R G ( B ) I_G(B)/R_G(B) I G ( B ) / R G ( B ) , where A B A_B A B is then an F p H B \mathbb{F}_pH_B F p H B -module for a prime p p p and
D B = End F p H B ( A B ) ; D_B = \operatorname{End}_{\mathbb{F}_pH_B}(A_B); D B = End F p H B ( A B ) ; (2)
G / R G ( B ) is M -full ⟺ δ G ( B ) ≤ 2 ; G/R_G(B) \text{ is } \mathcal{M}\text{-full} \Longleftrightarrow\delta_G(B) \le2; G / R G ( B ) is M -full ⟺ δ G ( B ) ≤ 2 ; (3)
G / R G ( B ) is quotient- M -full ⟺ δ G ( B ) ≤ 2 and H B is quotient- M -full ; G/R_G(B) \text{ is quotient-}\mathcal{M}\text{-full} \Longleftrightarrow\delta_G(B) \le2 \text{ and } H_B \text{ is quotient-}\mathcal{M}\text{-full}; G / R G ( B ) is quotient- M -full ⟺ δ G ( B ) ≤ 2 and H B is quotient- M -full ; (4) if R G ( B ) ≤ K ⊴ G R_G(B) \le K \unlhd G R G ( B ) ≤ K ⊴ G , then the quotient map
G / R G ( B ) ⟶ G / K , G/R_G(B) \longrightarrow G/K, G / R G ( B ) ⟶ G / K , identifying ( G / R G ( B ) ) / ( K / R G ( B ) ) (G/R_G(B))/(K/R_G(B)) ( G / R G ( B )) / ( K / R G ( B )) with G / K G/K G / K , is M \mathcal{M} M -closure compatible if and only if one of the following holds:
(a) K = R G ( B ) K = R_G(B) K = R G ( B ) ;
(b) δ G ( B ) = 2 \delta_G(B) = 2 δ G ( B ) = 2 and K / R G ( B ) K/R_G(B) K / R G ( B ) is minimal normal in G / R G ( B ) G/R_G(B) G / R G ( B ) ;
(c) I G ( B ) ≤ K I_G(B) \le K I G ( B ) ≤ K and G / K G/K G / K is M \mathcal{M} M -full;
(5) the number of normal subgroups K K K with
R G ( B ) ≤ K < I G ( B ) R_G(B) \le K < I_G(B) R G ( B ) ≤ K < I G ( B )
for which G / K G/K G / K is M \mathcal{M} M -full equals
{ δ G ( B ) + ( δ G ( B ) 2 ) , B non-abelian , ( δ G ( B ) 1 ) q B + ( δ G ( B ) 2 ) q B , B abelian , \begin{cases}
\delta_G(B)+\binom{\delta_G(B)}{2}, & B\text{ non-abelian},\\
\binom{\delta_G(B)}{1}_{q_B}+\binom{\delta_G(B)}{2}_{q_B}, & B\text{ abelian},
\end{cases} { δ G ( B ) + ( 2 δ G ( B ) ) , ( 1 δ G ( B ) ) q B + ( 2 δ G ( B ) ) q B , B non-abelian , B abelian ,
where q B = ∣ D B ∣ q_B=|D_B| q B = ∣ D B ∣ in the abelian case.
(6) if δ G ( B ) ≥ 2 \delta_G(B)\ge2 δ G ( B ) ≥ 2 , then the abstract lattice N ( G / R G ( B ) ) \mathcal{N}(G/R_G(B)) N ( G / R G ( B )) determines δ G ( B ) \delta_G(B) δ G ( B ) , detects whether B B B is abelian, and, in the abelian case, determines q B q_B q B .
Proof. For (1) through (6), apply Theorem 6.1 , Corollaries 6.2 , 6.3 , and 6.9, Proposition 6.7 , and Theorem 6.5 , with L = L B L=L_B L = L B and k = δ G ( B ) k=\delta_G(B) k = δ G ( B ) .
Finite abelian groups
For finite abelian groups Theorem A takes a particularly transparent form: apart from 0 0 0 and the subgroups above the socle, the M \mathcal{M} M -subgroups are the non-trivial cyclic p p p -subgroups. This allows us to count them. For cyclic groups C n C_n C n the classification may be compared with the description of ring M \mathcal{M} M -ideals of Z n \mathbb{Z}_n Z n in [3 ] .
Theorem 7.1. Let G G G be a finite abelian group and let H ≤ G H\le G H ≤ G . Then H H H is an M \mathcal{M} M -subgroup if and only if one of the following holds:
H = 0 H=0 H = 0 ;
H H H is a non-trivial cyclic p p p -group for some prime p p p ;
Soc ( G ) ≤ H \operatorname{Soc}(G)\le H Soc ( G ) ≤ H .
Proof. Let N ⊲ G N\lhd G N ⊲ G be a minimal normal subgroup and 0 ≠ x ∈ N 0\ne x\in N 0 = x ∈ N . As G G G is abelian, ⟨ x ⟩ ⊲ G \langle x\rangle\lhd G ⟨ x ⟩ ⊲ G , and by minimality of N N N , N = ⟨ x ⟩ N=\langle x\rangle N = ⟨ x ⟩ is cyclic. If ∣ N ∣ = n |N|=n ∣ N ∣ = n is not a prime, then there exists d ∈ N d\in N d ∈ N with d ∣ n d\mid n d ∣ n and 1 < d < n 1<d<n 1 < d < n , and the subgroup of N N N of order d d d is normal in G G G , which contradicts the minimality of N N N . Hence ∣ N ∣ = p |N|=p ∣ N ∣ = p for some prime p p p . Conversely, a subgroup of G G G of prime order is normal and has no non-trivial proper subgroups, and so it is minimal normal. Thus the minimal normal subgroups of G G G are precisely its subgroups of prime order.
Now let H ≠ 0 H\ne0 H = 0 with ∣ supp ( H ) ∣ = 1 |\operatorname{supp}(H)|=1 ∣ supp ( H ) ∣ = 1 . If x , y ∈ H x,y\in H x , y ∈ H with ord ( x ) = p ≠ q = ord ( y ) \operatorname{ord}(x)=p\ne q=\operatorname{ord}(y) ord ( x ) = p = q = ord ( y ) for distinct primes p , q p,q p , q , such x , y x,y x , y exist by Cauchy’s theorem if p q ∣ ∣ H ∣ pq\mid|H| pq ∣ ∣ H ∣ , then H H H would contain two minimal normal subgroups, contradicting ∣ supp ( H ) ∣ = 1 |\operatorname{supp}(H)|=1 ∣ supp ( H ) ∣ = 1 . So H H H is a p p p -group. If H H H is not cyclic, then as H H H is a finite abelian p p p -group we have
H ≅ C p α 1 ⊕ ⋯ ⊕ C p α n H\cong C_{p^{\alpha_1}}\oplus\cdots\oplus C_{p^{\alpha_n}} H ≅ C p α 1 ⊕ ⋯ ⊕ C p α n
with n ≥ 2 n\ge2 n ≥ 2 . Let x i x_i x i be a generator of the i i i th summand, and put
x = ( p α 1 − 1 x 1 , 0 , … , 0 ) and y = ( 0 , p α 2 − 1 x 2 , 0 , … , 0 ) . x=(p^{\alpha_1-1}x_1,0,\ldots,0)\quad\text{and}\quad y=(0,p^{\alpha_2-1}x_2,0,\ldots,0). x = ( p α 1 − 1 x 1 , 0 , … , 0 ) and y = ( 0 , p α 2 − 1 x 2 , 0 , … , 0 ) .
Then ord ( x ) = p = ord ( y ) \operatorname{ord}(x)=p=\operatorname{ord}(y) ord ( x ) = p = ord ( y ) and ⟨ x ⟩ ∩ ⟨ y ⟩ = 0 \langle x\rangle\cap\langle y\rangle=0 ⟨ x ⟩ ∩ ⟨ y ⟩ = 0 , so ⟨ x ⟩ \langle x\rangle ⟨ x ⟩ and ⟨ y ⟩ \langle y\rangle ⟨ y ⟩ are distinct minimal normal subgroups of G G G contained in H H H . This again contradicts ∣ supp ( H ) ∣ = 1 |\operatorname{supp}(H)|=1 ∣ supp ( H ) ∣ = 1 . Thus, H H H is cyclic. Conversely, if H H H is a non-trivial cyclic p p p -group, then H H H has a unique subgroup of order p p p , and hence ∣ supp ( H ) ∣ = 1 |\operatorname{supp}(H)|=1 ∣ supp ( H ) ∣ = 1 . Thus, for H ≠ 0 H\ne0 H = 0 ,
∣ supp ( H ) ∣ = 1 ⟺ H is a cyclic p -group . |\operatorname{supp}(H)|=1\Longleftrightarrow H\text{ is a cyclic }p\text{-group}. ∣ supp ( H ) ∣ = 1 ⟺ H is a cyclic p -group .
Now by Corollary 3.8, H ∈ N M ( G ) H\in\mathcal{N}^{\mathcal{M}}(G) H ∈ N M ( G ) if and only if H = 0 H=0 H = 0 , or ∣ supp ( H ) ∣ = 1 |\operatorname{supp}(H)|=1 ∣ supp ( H ) ∣ = 1 , or Soc ( G ) ≤ H \operatorname{Soc}(G)\le H Soc ( G ) ≤ H . So H H H is an M \mathcal{M} M -subgroup of G G G if and only if either H = 0 H=0 H = 0 or H H H is a non-trivial cyclic p p p -group or Soc ( G ) ≤ H \operatorname{Soc}(G)\le H Soc ( G ) ≤ H .
Consequently ∣ N M ( G ) ∣ |\mathcal{N}^{\mathcal{M}}(G)| ∣ N M ( G ) ∣ is determined by the number of cyclic p p p -subgroups, which can be read off from the invariants of G G G .
Proposition 7.2. Let G G G be a non-trivial finite abelian group, with notation as in Definition 2.12 . For every prime p p p dividing ∣ G ∣ |G| ∣ G ∣ ,
c p ( G ) = ∑ k = 1 λ p , 1 p u p ( k ) − p u p ( k − 1 ) p k − 1 ( p − 1 ) . c_p(G)=\sum_{k=1}^{\lambda_{p,1}}\frac{p^{u_p(k)}-p^{u_p(k-1)}}{p^{k-1}(p-1)}. c p ( G ) = k = 1 ∑ λ p , 1 p k − 1 ( p − 1 ) p u p ( k ) − p u p ( k − 1 ) . If G ≅ C p a G\cong C_{p^a} G ≅ C p a , then ∣ M M ( G ) ∣ = a + 1 |\mathcal{M}^{\mathcal{M}}(G)|=a+1 ∣ M M ( G ) ∣ = a + 1 . For every other non-trivial finite abelian group,
∣ M M ( G ) ∣ = 1 + ∣ Sub ( G / Soc ( G ) ) ∣ + ∑ p ∣ ∣ G ∣ c p ( G ) . |\mathcal{M}^{\mathcal{M}}(G)|=1+|\operatorname{Sub}(G/\operatorname{Soc}(G))|+\sum_{p\mid|G|}c_p(G). ∣ M M ( G ) ∣ = 1 + ∣ Sub ( G / Soc ( G )) ∣ + p ∣ ∣ G ∣ ∑ c p ( G ) .
Proof. Following the notation in Definition 2.12 , let G p ≅ C p λ p , 1 ⊕ ⋯ ⊕ C p λ p , r p G_p\cong C_{p^{\lambda_{p,1}}}\oplus\cdots\oplus C_{p^{\lambda_{p,r_p}}} G p ≅ C p λ p , 1 ⊕ ⋯ ⊕ C p λ p , r p and put G p [ p k ] = { x ∈ G p : p k x = 0 } G_p[p^k]=\{x\in G_p:p^kx=0\} G p [ p k ] = { x ∈ G p : p k x = 0 } . The condition p k x = 0 p^kx=0 p k x = 0 is checked coordinate-wise, and the elements x i ∈ C p λ p , i x_i\in C_{p^{\lambda_{p,i}}} x i ∈ C p λ p , i with p k x i = 0 p^kx_i=0 p k x i = 0 form the whole of C p λ p , i C_{p^{\lambda_{p,i}}} C p λ p , i if k ≥ λ p , i k\ge\lambda_{p,i} k ≥ λ p , i , and the unique subgroup of order p k p^k p k if k < λ p , i k<\lambda_{p,i} k < λ p , i . Thus,
∣ G p [ p k ] ∣ = ∏ i = 1 r p p min { k , λ p , i } = p ∑ i = 1 r p min { k , λ p , i } = p u p ( k ) . |G_p[p^k]|=\prod_{i=1}^{r_p}p^{\min\{k,\lambda_{p,i}\}}=p^{\sum_{i=1}^{r_p}\min\{k,\lambda_{p,i}\}}=p^{u_p(k)}. ∣ G p [ p k ] ∣ = i = 1 ∏ r p p m i n { k , λ p , i } = p ∑ i = 1 r p m i n { k , λ p , i } = p u p ( k ) .
As G p G_p G p is a p p p -group, p k x = 0 p^kx=0 p k x = 0 if and only if ord ( x ) ∣ p k \operatorname{ord}(x)\mid p^k ord ( x ) ∣ p k . So for k ≥ 1 k\ge1 k ≥ 1 ,
x ∈ G p [ p k ] ∖ G p [ p k − 1 ] ⟺ ord ( x ) ∣ p k and ord ( x ) ∤ p k − 1 ⟺ ord ( x ) = p k . x\in G_p[p^k]\setminus G_p[p^{k-1}]\Longleftrightarrow\operatorname{ord}(x)\mid p^k\text{ and }\operatorname{ord}(x)\nmid p^{k-1}\Longleftrightarrow\operatorname{ord}(x)=p^k. x ∈ G p [ p k ] ∖ G p [ p k − 1 ] ⟺ ord ( x ) ∣ p k and ord ( x ) ∤ p k − 1 ⟺ ord ( x ) = p k .
Thus, ∣ { x ∈ G p : ord ( x ) = p k } ∣ = p u p ( k ) − p u p ( k − 1 ) |\{x\in G_p:\operatorname{ord}(x)=p^k\}|=p^{u_p(k)}-p^{u_p(k-1)} ∣ { x ∈ G p : ord ( x ) = p k } ∣ = p u p ( k ) − p u p ( k − 1 ) .
Every cyclic p p p -subgroup of G G G lies in G p G_p G p , and as the largest order of an element of G p G_p G p is p λ p , 1 p^{\lambda_{p,1}} p λ p , 1 , the non-trivial ones have orders p , p 2 , ⋯ , p λ p , 1 p,p^2,\cdots,p^{\lambda_{p,1}} p , p 2 , ⋯ , p λ p , 1 . Let c p , k ( G ) c_{p,k}(G) c p , k ( G ) be the number of cyclic subgroups of G p G_p G p of order p k p^k p k . Each of them has φ ( p k ) = p k − 1 ( p − 1 ) \varphi(p^k)=p^{k-1}(p-1) φ ( p k ) = p k − 1 ( p − 1 ) generators, and each element of order p k p^k p k generates exactly one of them. So c p , k ( G ) φ ( p k ) = p u p ( k ) − p u p ( k − 1 ) c_{p,k}(G)\varphi(p^k)=p^{u_p(k)}-p^{u_p(k-1)} c p , k ( G ) φ ( p k ) = p u p ( k ) − p u p ( k − 1 ) , and hence
c p ( G ) = ∑ k = 1 λ p , 1 c p , k ( G ) = ∑ k = 1 λ p , 1 p u p ( k ) − p u p ( k − 1 ) p k − 1 ( p − 1 ) . c_p(G)=\sum_{k=1}^{\lambda_{p,1}}c_{p,k}(G)=\sum_{k=1}^{\lambda_{p,1}}\frac{p^{u_p(k)}-p^{u_p(k-1)}}{p^{k-1}(p-1)}. c p ( G ) = k = 1 ∑ λ p , 1 c p , k ( G ) = k = 1 ∑ λ p , 1 p k − 1 ( p − 1 ) p u p ( k ) − p u p ( k − 1 ) .
Now if G ≅ C p a G\cong C_{p^a} G ≅ C p a , then u p ( k ) = min { k , a } u_p(k)=\min\{k,a\} u p ( k ) = min { k , a } , and so
c p ( G ) = ∑ k = 1 a p k − p k − 1 p k − 1 ( p − 1 ) = ∑ k = 1 a 1 = a . c_p(G)=\sum_{k=1}^{a}\frac{p^k-p^{k-1}}{p^{k-1}(p-1)}=\sum_{k=1}^{a}1=a. c p ( G ) = k = 1 ∑ a p k − 1 ( p − 1 ) p k − p k − 1 = k = 1 ∑ a 1 = a .
As every subgroup of C p a C_{p^a} C p a is either 0 or a non-trivial cyclic p p p -group, by Theorem 7.1 ,
∣ M M ( G ) ∣ = 1 + c p ( G ) = a + 1. |\mathcal{M}^{\mathcal{M}}(G)|=1+c_p(G)=a+1. ∣ M M ( G ) ∣ = 1 + c p ( G ) = a + 1.
For any other non-trivial finite abelian group G G G , by the correspondence theorem, the subgroups of G G G containing Soc ( G ) \operatorname{Soc}(G) Soc ( G ) are in bijection with Sub ( G / Soc ( G ) ) \operatorname{Sub}(G/\operatorname{Soc}(G)) Sub ( G / Soc ( G )) . Moreover, no non-trivial cyclic p p p -subgroup C C C of G G G contains Soc ( G ) \operatorname{Soc}(G) Soc ( G ) . Indeed, if Soc ( G ) ≤ C \operatorname{Soc}(G)\le C Soc ( G ) ≤ C , then Soc ( G ) \operatorname{Soc}(G) Soc ( G ) is a p p p -group, and so by Cauchy’s theorem G G G is a p p p -group. Then Soc ( G ) = G p [ p ] ≤ C \operatorname{Soc}(G)=G_p[p]\le C Soc ( G ) = G p [ p ] ≤ C is cyclic of order p u p ( 1 ) = p r p p^{u_p(1)}=p^{r_p} p u p ( 1 ) = p r p . So r p = 1 r_p=1 r p = 1 and G ≅ C p λ p , 1 G\cong C_{p^{\lambda_{p,1}}} G ≅ C p λ p , 1 , which is a contradiction. Also, Soc ( G ) ≠ 0 \operatorname{Soc}(G)\ne0 Soc ( G ) = 0 as G ≠ 0 G\ne0 G = 0 . So the three types of M \mathcal{M} M -subgroups in Theorem 7.1 are pairwise disjoint, and hence
∣ M M ( G ) ∣ = 1 + ∑ p ∣ ∣ G ∣ c p ( G ) + ∣ Sub ( G / Soc ( G ) ) ∣ . |\mathcal{M}^{\mathcal{M}}(G)|=1+\sum_{p\mid|G|}c_p(G)+|\operatorname{Sub}(G/\operatorname{Soc}(G))|. ∣ M M ( G ) ∣ = 1 + p ∣ ∣ G ∣ ∑ c p ( G ) + ∣ Sub ( G / Soc ( G )) ∣.
The M \mathcal{M} M -full abelian groups are those of rank at most 2, in the following sense.
Corollary 7.3. Let G G G be a non-trivial finite abelian group and put r ( G ) = ∑ p r p r(G)=\sum_p r_p r ( G ) = ∑ p r p . Then G G G is M \mathcal{M} M -full if and only if r ( G ) ≤ 2 r(G)\le2 r ( G ) ≤ 2 .
Proof. Following Definition 2.12 and collecting the cyclic summands of all the G p G_p G p together, we get
G ≅ ⨁ p ∣ ∣ G ∣ ⨁ i = 1 r p C p λ p , i . G \cong\bigoplus_{p\mid|G|}\bigoplus_{i=1}^{r_p} C_{p^{\lambda_{p,i}}}. G ≅ p ∣ ∣ G ∣ ⨁ i = 1 ⨁ r p C p λ p , i .
So G G G is a direct sum of exactly r ( G ) r(G) r ( G ) nontrivial cyclic p p p -groups. By the fundamental theorem of finite cyclic groups, each C p λ p , i C_{p^{\lambda_{p,i}}} C p λ p , i is monolithic.
If r ( G ) = 1 r(G)=1 r ( G ) = 1 , then G ≅ C p a G\cong C_{p^a} G ≅ C p a for some prime p p p and some a ≥ 1 a\geq1 a ≥ 1 . So every non-zero subgroup of G G G is a cyclic p p p -group, and by Theorem 7.1 we get M ( G ) = N ( G ) \mathcal{M}(G)=\mathcal{N}(G) M ( G ) = N ( G ) . So G G G is M \mathcal{M} M -full. If r ( G ) ≥ 2 r(G)\geq2 r ( G ) ≥ 2 , then as each C p λ p , i C_{p^{\lambda_{p,i}}} C p λ p , i is monolithic, by Corollary 5.3 ,
G is M -full ⟺ r ( G ) = 2. G\text{ is }\mathcal{M}\text{-full}\Longleftrightarrow r(G)=2. G is M -full ⟺ r ( G ) = 2.
Combining the two cases, G G G is M \mathcal{M} M -full if and only if r ( G ) ≤ 2 r(G)\leq2 r ( G ) ≤ 2 . □
M-depth
A chief series of G G G is a maximal chain in N ( G ) \mathcal{N}(G) N ( G ) , and its length ℓ n ( G ) \ell_n(G) ℓ n ( G ) is an invariant of $G. In this section we show that the same holds for N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) : all maximal M \mathcal{M} M -chains of a finite group have the same length md ( G ) \operatorname{md}(G) md ( G ) , which we compute (Theorem D). The key observation is that a normal subgroup over a single minimal normal subgroup meets the socle only in that minimal normal subgroup, so a maximal M \mathcal{M} M -chain must cross the socle in one jump.
Lemma 8.1. Let G G G be a non-trivial finite group, let A ∈ MinNorm ( G ) A\in\operatorname{MinNorm}(G) A ∈ MinNorm ( G ) , and put S = Soc ( G ) S=\operatorname{Soc}(G) S = Soc ( G ) . There are minimal normal subgroups
A = A 1 , A 2 , … , A σ ( G ) A=A_1,A_2,\ldots,A_{\sigma(G)} A = A 1 , A 2 , … , A σ ( G ) such that
S = A 1 × A 2 × ⋯ × A σ ( G ) . S=A_1\times A_2\times\cdots\times A_{\sigma(G)}. S = A 1 × A 2 × ⋯ × A σ ( G ) . Consequently, if T ⊴ G T\unlhd G T ⊴ G contains exactly the atom A A A , then
and every G G G -normal subgroup in the interval [ 1 , T ] [1,T] [ 1 , T ] is an M \mathcal{M} M -subgroup.
Proof. Let N 1 = A = A 1 N_1=A=A_1 N 1 = A = A 1 . Then, λ G ( N 1 ) = 1 \lambda_G(N_1)=1 λ G ( N 1 ) = 1 . Suppose we have chosen A = A 1 , A 2 , … , A i ∈ MinNorm ( G ) A=A_1,A_2,\ldots,A_i\in\operatorname{MinNorm}(G) A = A 1 , A 2 , … , A i ∈ MinNorm ( G ) such that N i = A 1 × ⋯ × A i < Soc ( G ) N_i=A_1\times\cdots\times A_i<\operatorname{Soc}(G) N i = A 1 × ⋯ × A i < Soc ( G ) and λ G ( N i ) = i \lambda_G(N_i)=i λ G ( N i ) = i . As
Soc ( G ) = ⋁ { N : N ∈ MinNorm ( G ) } , \operatorname{Soc}(G)=\bigvee\{N:N\in\operatorname{MinNorm}(G)\}, Soc ( G ) = ⋁ { N : N ∈ MinNorm ( G )} ,
so there exists A i + 1 ∈ MinNorm ( G ) A_{i+1}\in\operatorname{MinNorm}(G) A i + 1 ∈ MinNorm ( G ) such that A i + 1 ≰ N i A_{i+1}\nleq N_i A i + 1 ≰ N i . Now N i ∩ A i + 1 ⊴ G N_i\cap A_{i+1}\unlhd G N i ∩ A i + 1 ⊴ G , and as A i + 1 A_{i+1} A i + 1 is minimal normal, this forces N i ∩ A i + 1 = 1 N_i\cap A_{i+1}=1 N i ∩ A i + 1 = 1 . Hence, [ N i , A i + 1 ] ≤ N i ∩ A i + 1 = 1 [N_i,A_{i+1}]\leq N_i\cap A_{i+1}=1 [ N i , A i + 1 ] ≤ N i ∩ A i + 1 = 1 , and therefore N i + 1 = N i A i + 1 = N i × A i + 1 N_{i+1}=N_iA_{i+1}=N_i\times A_{i+1} N i + 1 = N i A i + 1 = N i × A i + 1 . By Lemma 3.2 , we get
λ G ( N i + 1 ) = λ G ( N i ) + 1 = i + 1 , \lambda_G(N_{i+1})=\lambda_G(N_i)+1=i+1, λ G ( N i + 1 ) = λ G ( N i ) + 1 = i + 1 ,
since λ G ( A i + 1 ) = 1 \lambda_G(A_{i+1})=1 λ G ( A i + 1 ) = 1 and λ G ( N i ∩ A i + 1 ) = 0 \lambda_G(N_i\cap A_{i+1})=0 λ G ( N i ∩ A i + 1 ) = 0 . As G G G is finite, there exists a natural number k k k such that N k = Soc ( G ) N_k=\operatorname{Soc}(G) N k = Soc ( G ) . So, σ ( G ) = λ G ( N k ) = k \sigma(G)=\lambda_G(N_k)=k σ ( G ) = λ G ( N k ) = k . Thus, we have minimal normal subgroups A = A 1 , A 2 , … , A σ ( G ) A=A_1,A_2,\ldots,A_{\sigma(G)} A = A 1 , A 2 , … , A σ ( G ) such that
S = A 1 × ⋯ × A σ ( G ) . S=A_1\times\cdots\times A_{\sigma(G)}. S = A 1 × ⋯ × A σ ( G ) .
Now suppose T ⊴ G T\unlhd G T ⊴ G contains exactly the atom A A A . Then A ≤ T ∩ S A\leq T\cap S A ≤ T ∩ S . Put C = A 2 × ⋯ × A σ ( G ) C=A_2\times\cdots\times A_{\sigma(G)} C = A 2 × ⋯ × A σ ( G ) , so that S = A C S=AC S = A C and A ∩ C = 1 A\cap C=1 A ∩ C = 1 . Then by Lemma 3.1 ,
T ∩ S = A C ∩ ( T ∩ S ) = A ( C ∩ T ∩ S ) = A ( C ∩ T ) . T\cap S=AC\cap(T\cap S)=A(C\cap T\cap S)=A(C\cap T). T ∩ S = A C ∩ ( T ∩ S ) = A ( C ∩ T ∩ S ) = A ( C ∩ T ) .
If C ∩ T ≠ 1 C\cap T\neq1 C ∩ T = 1 , then as C ∩ T ⊴ G C\cap T\unlhd G C ∩ T ⊴ G , there exists B ∈ MinNorm ( G ) B\in\operatorname{MinNorm}(G) B ∈ MinNorm ( G ) with B ≤ C ∩ T B\leq C\cap T B ≤ C ∩ T . Then B ≤ T B\leq T B ≤ T , and B ≠ A B\neq A B = A as A ∩ C = 1 A\cap C=1 A ∩ C = 1 . This contradicts the fact that T T T contains exactly the atom A A A . Thus, C ∩ T = 1 C\cap T=1 C ∩ T = 1 and hence T ∩ S = A T\cap S=A T ∩ S = A . Furthermore, let 1 ≠ M ∈ [ 1 , T ] N ( G ) 1 \ne M \in[1,T]_{\mathcal{N}(G)} 1 = M ∈ [ 1 , T ] N ( G ) . As G G G is finite, there exists K ∈ MinNorm ( G ) K \in\operatorname{MinNorm}(G) K ∈ MinNorm ( G ) such that K ≤ M ≤ T K \le M \le T K ≤ M ≤ T . As T T T contains only the atom A A A , it follows that K = A K=A K = A , and hence ∣ supp ( M ) ∣ = 1 |\operatorname{supp}(M)|=1 ∣ supp ( M ) ∣ = 1 . Thus, M ∈ N M ( G ) M \in\mathcal{N}^{\mathcal{M}}(G) M ∈ N M ( G ) by Corollary 3.8. Also, 1 ∈ N M ( G ) 1 \in\mathcal{N}^{\mathcal{M}}(G) 1 ∈ N M ( G ) trivially. Thus, every G G G -normal subgroup in the interval [ 1 , T ] [1,T] [ 1 , T ] is an M \mathcal{M} M -subgroup.
We can now prove Theorem D.
Theorem 8.2. Let G G G be a non-trivial finite group and put S = Soc ( G ) S=\operatorname{Soc}(G) S = Soc ( G ) .
(1) If G G G has one minimal normal subgroup, then
md ( G ) = ℓ n ( G ) . \operatorname{md}(G)=\ell_n(G). md ( G ) = ℓ n ( G ) . (2) If G G G has at least two minimal normal subgroups, then
md ( G ) = ℓ n ( G ) − σ ( G ) + 2. \operatorname{md}(G)=\ell_n(G)-\sigma(G)+2. md ( G ) = ℓ n ( G ) − σ ( G ) + 2.
Proof. For (1), observe that since G G G has one minimal normal subgroup, every non-trivial normal subgroup of G G G contains it, and so by Corollary 3.8 it follows that G G G is M \mathcal{M} M -full. Now if
1 = E 0 < E 1 < ⋯ < E r = G 1=E_0<E_1<\cdots<E_r=G 1 = E 0 < E 1 < ⋯ < E r = G
is a chief series of G G G , then E i ∈ N ( G ) E_i\in\mathcal{N}(G) E i ∈ N ( G ) and as G G G is M \mathcal{M} M -full so E i ∈ N M ( G ) E_i\in\mathcal{N}^{\mathcal{M}}(G) E i ∈ N M ( G ) for all i i i . Conversely, every maximal M \mathcal{M} M -chain is a maximal chain of N ( G ) \mathcal{N}(G) N ( G ) and hence is a chief series of G G G . Thus, ℓ n ( G ) = md ( G ) \ell_n(G)=\operatorname{md}(G) ℓ n ( G ) = md ( G ) .
For (2), consider a maximal M \mathcal{M} M -chain,
1 = E 0 < E 1 < ⋯ < E r = G . 1=E_0<E_1<\cdots<E_r=G. 1 = E 0 < E 1 < ⋯ < E r = G .
Let E j E_j E j be the first element of this chain containing S S S , and put T = E j − 1 T=E_{j-1} T = E j − 1 . Then S ≰ T S\not\leq T S ≤ T . Also T ≠ 1 T\ne1 T = 1 . Indeed, if T = 1 T=1 T = 1 , then for any A ∈ MinNorm ( G ) A\in\operatorname{MinNorm}(G) A ∈ MinNorm ( G ) we have 1 < A < S ≤ E 1 1<A<S\leq E_1 1 < A < S ≤ E 1 , as G G G has at least two minimal normal subgroups, and A ∈ N M ( G ) A\in\mathcal{N}^{\mathcal{M}}(G) A ∈ N M ( G ) . This contradicts the maximality of the chain. As T ∈ N M ( G ) T\in\mathcal{N}^{\mathcal{M}}(G) T ∈ N M ( G ) , T ≠ 1 T\ne1 T = 1 and S ≰ T S\not\leq T S ≤ T , by Corollary 3.8, ∣ supp ( T ) ∣ = 1 |\operatorname{supp}(T)|=1 ∣ supp ( T ) ∣ = 1 . Let supp ( T ) = { A } \operatorname{supp}(T)=\{A\} supp ( T ) = { A } . Then, A ∈ MinNorm ( G ) A\in\operatorname{MinNorm}(G) A ∈ MinNorm ( G ) and by Lemma 8.1 , we get T ∩ S = A T\cap S=A T ∩ S = A . Now T S ∈ N M ( G ) TS\in\mathcal{N}^{\mathcal{M}}(G) T S ∈ N M ( G ) as S ≤ T S S\leq TS S ≤ T S . Thus we have
T < T S ≤ E j . T<TS\leq E_j. T < T S ≤ E j .
So maximality of the chain forces T S = E j TS=E_j T S = E j . Thus, our chain becomes
1 = E 0 < E 1 < ⋯ < T < T S < ⋯ < G . 1=E_0<E_1<\cdots<T<TS<\cdots<G. 1 = E 0 < E 1 < ⋯ < T < T S < ⋯ < G .
By Lemma 8.1 , [ 1 , T ] N ( G ) = [ 1 , T ] N M ( G ) [1,T]_{\mathcal{N}(G)}=[1,T]_{\mathcal{N}^{\mathcal{M}}(G)} [ 1 , T ] N ( G ) = [ 1 , T ] N M ( G ) , and every normal subgroup containing the socle is an M \mathcal{M} M -subgroup. So the parts of the chain from 1 1 1 to T T T and from T S TS T S to G G G cannot be refined in N ( G ) \mathcal{N}(G) N ( G ) , and as N ( G ) \mathcal{N}(G) N ( G ) is modular by Lemma 3.1 , they have lengths λ G ( T ) \lambda_G(T) λ G ( T ) and λ G ( G ) − λ G ( T S ) \lambda_G(G)-\lambda_G(TS) λ G ( G ) − λ G ( T S ) . Thus, we get
r = λ G ( T ) + 1 + λ G ( G ) − λ G ( T S ) . r=\lambda_G(T)+1+\lambda_G(G)-\lambda_G(TS). r = λ G ( T ) + 1 + λ G ( G ) − λ G ( T S ) .
Now by Lemma 3.2 , as λ G ( T ∩ S ) = λ G ( A ) = 1 \lambda_G(T\cap S)=\lambda_G(A)=1 λ G ( T ∩ S ) = λ G ( A ) = 1 and λ G ( S ) = σ ( G ) \lambda_G(S)=\sigma(G) λ G ( S ) = σ ( G ) , we get
λ G ( T S ) + 1 = λ G ( T ) + σ ( G ) . \lambda_G(TS)+1=\lambda_G(T)+\sigma(G). λ G ( T S ) + 1 = λ G ( T ) + σ ( G ) .
Substituting this value back, we get r = ℓ n ( G ) − σ ( G ) + 2 r=\ell_n(G)-\sigma(G)+2 r = ℓ n ( G ) − σ ( G ) + 2 , since λ G ( G ) = ℓ n ( G ) \lambda_G(G)=\ell_n(G) λ G ( G ) = ℓ n ( G ) . So every maximal M \mathcal{M} M -chain has length ℓ n ( G ) − σ ( G ) + 2 \ell_n(G)-\sigma(G)+2 ℓ n ( G ) − σ ( G ) + 2 , and as an M \mathcal{M} M -chain of maximum length is maximal, md ( G ) = ℓ n ( G ) − σ ( G ) + 2 \operatorname{md}(G)=\ell_n(G)-\sigma(G)+2 md ( G ) = ℓ n ( G ) − σ ( G ) + 2 .
The proof of Theorem 8.2 describes all chains of maximal length.
Corollary 8.3. Let G G G be a finite group with at least two minimal normal subgroups and put S = Soc ( G ) S=\operatorname{Soc}(G) S = Soc ( G ) . Let T ⊴ G T\trianglelefteq G T ⊴ G contain exactly one minimal normal subgroup. Concatenate a maximal chain in [ 1 , T ] N ( G ) [1,T]_{\mathcal{N}(G)} [ 1 , T ] N ( G ) , the step
and a maximal chain in [ T S , G ] N ( G ) [TS,G]_{\mathcal{N}(G)} [ T S , G ] N ( G ) . The resulting chain consists of M \mathcal{M} M -subgroups and has length md ( G ) \operatorname{md}(G) md ( G ) . Conversely, every M \mathcal{M} M -chain of length md ( G ) \operatorname{md}(G) md ( G ) arises in this way, with T T T its last nonessential non-trivial term.
Proof. The argument follows precisely as in Theorem 8.2 (2). Indeed, let supp ( T ) = { A } \operatorname{supp}(T)=\{A\} supp ( T ) = { A } . As G G G has at least two minimal normal subgroups, S ≰ T S\nleq T S ≰ T , and so T < T S T<TS T < T S . By Lemma 8.1 , T ∩ S = A T\cap S=A T ∩ S = A and the terms of the first chain are M \mathcal{M} M -subgroups, and the terms of the last chain contain S S S , so they are M \mathcal{M} M -subgroups too. As N ( G ) \mathcal{N}(G) N ( G ) is modular, the resulting chain has length λ G ( T ) + 1 + ℓ n ( G ) − λ G ( T S ) \lambda_G(T)+1+\ell_n(G)-\lambda_G(TS) λ G ( T ) + 1 + ℓ n ( G ) − λ G ( T S ) , which is
ℓ n ( G ) − σ ( G ) + 2 = md ( G ) \ell_n(G)-\sigma(G)+2=\operatorname{md}(G) ℓ n ( G ) − σ ( G ) + 2 = md ( G ) by Lemma 3.2 and Theorem 8.2 . Conversely, an M \mathcal{M} M -chain of length md ( G ) \operatorname{md}(G) md ( G ) is maximal, and so by the proof of Theorem 8.2 (2) it has the above form with T = E j − 1 T=E_{j-1} T = E j − 1 . Finally, as every non-trivial normal subgroup of G G G contains a minimal normal subgroup, a normal subgroup of G G G is essential if and only if it contains S S S . So T = E j − 1 T=E_{j-1} T = E j − 1 is the last non-essential non-trivial term of the chain. □
In particular, the M \mathcal{M} M -depth detects M \mathcal{M} M -fullness.
Corollary 8.4. For a non-trivial finite group G G G ,
md ( G ) = ℓ n ( G ) ⟺ G is M -full . \operatorname{md}(G)=\ell_n(G)\quad\Longleftrightarrow\quad G\text{ is }\mathcal{M}\text{-full}. md ( G ) = ℓ n ( G ) ⟺ G is M -full . Proof. First suppose md ( G ) = ℓ n ( G ) \operatorname{md}(G)=\ell_n(G) md ( G ) = ℓ n ( G ) . Now we have two cases. Firstly, if G G G contains a unique minimal normal subgroup. Then, every non-trivial N ∈ N ( G ) N\in\mathcal{N}(G) N ∈ N ( G ) contains exactly one minimal normal subgroup, and hence N ( G ) = N M ( G ) \mathcal{N}(G)=\mathcal{N}^{\mathcal{M}}(G) N ( G ) = N M ( G ) by Corollary 3.8. Secondly, if G G G contains at least two minimal normal subgroups. Then by Theorem 8.2 , we have
md ( G ) = ℓ n ( G ) − σ ( G ) + 2. \operatorname{md}(G)=\ell_n(G)-\sigma(G)+2. md ( G ) = ℓ n ( G ) − σ ( G ) + 2. As md ( G ) = ℓ n ( G ) \operatorname{md}(G)=\ell_n(G) md ( G ) = ℓ n ( G ) , it follows that σ ( G ) = 2 \sigma(G)=2 σ ( G ) = 2 . By Theorem 4.5 , G G G is M \mathcal{M} M -full. Conversely, suppose G G G is M \mathcal{M} M -full. If G G G has a unique minimal normal subgroup then by Theorem 8.2 , md ( G ) = ℓ n ( G ) \operatorname{md}(G)=\ell_n(G) md ( G ) = ℓ n ( G ) . Now suppose G G G has at least two minimal normal subgroups. Then, md ( G ) = ℓ n ( G ) − σ ( G ) + 2 \operatorname{md}(G)=\ell_n(G)-\sigma(G)+2 md ( G ) = ℓ n ( G ) − σ ( G ) + 2 . As G G G is M \mathcal{M} M -full, by Theorem 4.5 , σ ( G ) ≤ 2 \sigma(G)\leq2 σ ( G ) ≤ 2 . But σ ( G ) ≠ 1 \sigma(G)\ne1 σ ( G ) = 1 as G G G has at least two minimal normal subgroups. Hence, σ ( G ) = 2 \sigma(G)=2 σ ( G ) = 2 . Thus, md ( G ) = ℓ n ( G ) \operatorname{md}(G)=\ell_n(G) md ( G ) = ℓ n ( G ) . □
Since chief length and socle length are additive over direct products, we obtain the product formula of Theorem D.
Corollary 8.5. Let G 1 , … , G t G_1,\ldots,G_t G 1 , … , G t be non-trivial finite groups and let t ≥ 2 t\geq2 t ≥ 2 . Then
md ( G 1 × ⋯ × G t ) = ∑ i = 1 t ( ℓ n ( G i ) − σ ( G i ) ) + 2. \operatorname{md}(G_1\times\cdots\times G_t)=\sum_{i=1}^{t}\bigl(\ell_n(G_i)-\sigma(G_i)\bigr)+2. md ( G 1 × ⋯ × G t ) = i = 1 ∑ t ( ℓ n ( G i ) − σ ( G i ) ) + 2. In particular, if all G i G_i G i are monolithic, then
md ( G 1 × ⋯ × G t ) = ∑ i = 1 t ℓ n ( G i ) − t + 2. \operatorname{md}(G_1\times\cdots\times G_t)=\sum_{i=1}^{t}\ell_n(G_i)-t+2. md ( G 1 × ⋯ × G t ) = i = 1 ∑ t ℓ n ( G i ) − t + 2. Proof. Let G = G 1 × ⋯ × G t G=G_1\times\cdots\times G_t G = G 1 × ⋯ × G t , and for X ≤ G i X\leq G_i X ≤ G i let X ^ \widehat{X} X be its coordinate copy in G G G . As G j ∗ G_j^* G j ∗ centralizes G i ∗ G_i^* G i ∗ for j ≠ i j\ne i j = i , X ^ \widehat{X} X is normal in G G G if and only if X ⊴ G i X\trianglelefteq G_i X ⊴ G i . So X ↦ X ^ X\mapsto\widehat{X} X ↦ X is a lattice isomorphism from N ( G i ) \mathcal{N}(G_i) N ( G i ) onto [ 1 , G i ∗ ] N ( G ) [1,G_i^*]_{\mathcal{N}(G)} [ 1 , G i ∗ ] N ( G ) , and hence
λ G ( X ^ ) = λ G i ( X ) for all X ≤ G i . \lambda_G(\widehat{X})=\lambda_{G_i}(X)\quad\text{for all }X\leq G_i. λ G ( X ) = λ G i ( X ) for all X ≤ G i .
In particular, λ G ( G 1 ∗ ) = ℓ n ( G 1 ) \lambda_G(G_1^*)=\ell_n(G_1) λ G ( G 1 ∗ ) = ℓ n ( G 1 ) . As G / G 1 ∗ ≅ G 2 × ⋯ × G t G/G_1^* \cong G_2 \times\cdots\times G_t G / G 1 ∗ ≅ G 2 × ⋯ × G t , by Lemma 3.2 ,
ℓ n ( G ) = λ G ( G 1 ∗ ) + ℓ n ( G / G 1 ∗ ) = ℓ n ( G 1 ) + ℓ n ( G 2 × ⋯ × G t ) . \ell_n(G)=\lambda_G(G_1^*)+\ell_n(G/G_1^*)=\ell_n(G_1)+\ell_n(G_2\times\cdots\times G_t). ℓ n ( G ) = λ G ( G 1 ∗ ) + ℓ n ( G / G 1 ∗ ) = ℓ n ( G 1 ) + ℓ n ( G 2 × ⋯ × G t ) .
Hence by induction on t t t , ℓ n ( G ) = ∑ i = 1 t ℓ n ( G i ) \ell_n(G)=\sum_{i=1}^{t}\ell_n(G_i) ℓ n ( G ) = ∑ i = 1 t ℓ n ( G i ) . Also, by Theorem 5.1 , Soc ( G ) = Soc ( G 1 ‾ ) ⋯ Soc ( G t ‾ ) \operatorname{Soc}(G)=\operatorname{Soc}(\overline{G_1})\cdots\operatorname{Soc}(\overline{G_t}) Soc ( G ) = Soc ( G 1 ) ⋯ Soc ( G t ) , and each Soc ( G i ‾ ) \operatorname{Soc}(\overline{G_i}) Soc ( G i ) intersects Soc ( G 1 ‾ ) ⋯ Soc ( G i − 1 ‾ ) \operatorname{Soc}(\overline{G_1})\cdots\operatorname{Soc}(\overline{G_{i-1}}) Soc ( G 1 ) ⋯ Soc ( G i − 1 ) trivially. So applying Lemma 3.2 (2) repeatedly, we get
σ ( G ) = ∑ i = 1 t λ G ( Soc ( G i ‾ ) ) = ∑ i = 1 t σ ( G i ) ≥ t ≥ 2 \sigma(G)=\sum_{i=1}^{t}\lambda_G(\operatorname{Soc}(\overline{G_i}))=\sum_{i=1}^{t}\sigma(G_i)\geq t\geq2 σ ( G ) = i = 1 ∑ t λ G ( Soc ( G i )) = i = 1 ∑ t σ ( G i ) ≥ t ≥ 2
as σ ( G i ) ≥ 1 \sigma(G_i)\geq1 σ ( G i ) ≥ 1 for each i i i . Thus G G G has at least two minimal normal subgroups. Hence by Theorem 8.2 , we have
md ( G ) = ℓ n ( G ) − σ ( G ) + 2 = ∑ i = 1 t ( ℓ n ( G i ) − σ ( G i ) ) + 2. \operatorname{md}(G)=\ell_n(G)-\sigma(G)+2=\sum_{i=1}^{t}\bigl(\ell_n(G_i)-\sigma(G_i)\bigr)+2. md ( G ) = ℓ n ( G ) − σ ( G ) + 2 = i = 1 ∑ t ( ℓ n ( G i ) − σ ( G i ) ) + 2.
Now if each G i G_i G i is monolithic, then σ ( G i ) = 1 \sigma(G_i)=1 σ ( G i ) = 1 for all i i i . Hence md ( G ) = ∑ i = 1 t ℓ n ( G i ) − t + 2 \operatorname{md}(G)=\sum_{i=1}^{t}\ell_n(G_i)-t+2 md ( G ) = ∑ i = 1 t ℓ n ( G i ) − t + 2 .
For crown-based powers the M \mathcal{M} M -depth no longer depends on the crown multiplicity, once this is at least 2.
Corollary 8.6. Let L L L be finite monolithic primitive with socle A A A , and let L k L_k L k be its crown-based power. Then
md ( L k ) = { ℓ n ( L / A ) + 1 , k = 1 , ℓ n ( L / A ) + 2 , k ≥ 2. \operatorname{md}(L_k)=
\begin{cases}
\ell_n(L/A)+1, & k=1,\\
\ell_n(L/A)+2, & k\geq2.
\end{cases} md ( L k ) = { ℓ n ( L / A ) + 1 , ℓ n ( L / A ) + 2 , k = 1 , k ≥ 2. Proof. Let V = Soc ( L k ) V=\operatorname{Soc}(L_k) V = Soc ( L k ) . Now if k = 1 k=1 k = 1 , then L 1 = L L_1=L L 1 = L and V = A V=A V = A . By Corollary 6.3 , L 1 L_1 L 1 is M \mathcal{M} M -full. Hence by Corollary 8.4 , we get md ( L 1 ) = ℓ n ( L 1 ) \operatorname{md}(L_1)=\ell_n(L_1) md ( L 1 ) = ℓ n ( L 1 ) . By Lemma 3.2 , we have ℓ n ( L ) = λ L ( A ) + ℓ n ( L / A ) \ell_n(L)=\lambda_L(A)+\ell_n(L/A) ℓ n ( L ) = λ L ( A ) + ℓ n ( L / A ) . As A A A is minimal normal in this case, we have λ L ( A ) = 1 \lambda_L(A)=1 λ L ( A ) = 1 . Thus, md ( L 1 ) = ℓ n ( L / A ) + 1 \operatorname{md}(L_1)=\ell_n(L/A)+1 md ( L 1 ) = ℓ n ( L / A ) + 1 .
For k ≥ 2 k\geq2 k ≥ 2 , by Proposition 6.4 , every maximal M \mathcal{M} M -chain of L k L_k L k has length ℓ n ( L / A ) + 2 \ell_n(L/A)+2 ℓ n ( L / A ) + 2 . Thus, md ( L k ) = ℓ n ( L / A ) + 2 \operatorname{md}(L_k)=\ell_n(L/A)+2 md ( L k ) = ℓ n ( L / A ) + 2 .
We now evaluate md for several classical families. For nilpotent groups all chief factors have prime order, so ℓ n ( G ) = Ω ( ∣ G ∣ ) \ell_n(G)=\Omega(|G|) ℓ n ( G ) = Ω ( ∣ G ∣ ) .
Corollary 8.7. Let G G G be a non-trivial finite nilpotent group. Then
md ( G ) = { Ω ( ∣ G ∣ ) , ρ ( G ) = 1 , Ω ( ∣ G ∣ ) − ρ ( G ) + 2 , ρ ( G ) ≥ 2. \operatorname{md}(G)=
\begin{cases}
\Omega(|G|), & \rho(G)=1,\\
\Omega(|G|)-\rho(G)+2, & \rho(G)\geq2.
\end{cases} md ( G ) = { Ω ( ∣ G ∣ ) , Ω ( ∣ G ∣ ) − ρ ( G ) + 2 , ρ ( G ) = 1 , ρ ( G ) ≥ 2. Proof. Let
1 = G 0 < G 1 < ⋯ < G r = G 1=G_0<G_1<\cdots<G_r=G 1 = G 0 < G 1 < ⋯ < G r = G be a chief series of G G G . So G i / G i − 1 G_i/G_{i-1} G i / G i − 1 is a minimal normal subgroup of G / G i − 1 G/G_{i-1} G / G i − 1 . As G G G is finite nilpotent, G / G i − 1 G/G_{i-1} G / G i − 1 is also finite nilpotent. By the proof of Corollary 4.8 we get [ G i : G i − 1 ] = p i [G_i:G_{i-1}]=p_i [ G i : G i − 1 ] = p i for some prime p i p_i p i , for all i i i . Hence,
∣ G ∣ = ∏ i = 1 r [ G i : G i − 1 ] = p 1 p 2 ⋯ p r , |G|=\prod_{i=1}^{r}[G_i:G_{i-1}]=p_1p_2\cdots p_r, ∣ G ∣ = i = 1 ∏ r [ G i : G i − 1 ] = p 1 p 2 ⋯ p r , and so ℓ n ( G ) = r = Ω ( ∣ G ∣ ) \ell_n(G)=r=\Omega(|G|) ℓ n ( G ) = r = Ω ( ∣ G ∣ ) .
Now suppose ρ ( G ) = 1 \rho(G)=1 ρ ( G ) = 1 . Then by Corollary 4.8 we have that G G G is M \mathcal{M} M -full. So by Corollary 8.4 , md ( G ) = ℓ n ( G ) = Ω ( ∣ G ∣ ) \operatorname{md}(G)=\ell_n(G)=\Omega(|G|) md ( G ) = ℓ n ( G ) = Ω ( ∣ G ∣ ) . Again, by the proof of Corollary 4.8 we have σ ( G ) = ρ ( G ) \sigma(G)=\rho(G) σ ( G ) = ρ ( G ) . Thus for ρ ( G ) ≥ 2 \rho(G)\geq2 ρ ( G ) ≥ 2 we have σ ( G ) ≥ 2 \sigma(G)\geq2 σ ( G ) ≥ 2 . Hence G G G has at least two minimal normal subgroups. So by Theorem 8.2 ,
md ( G ) = ℓ n ( G ) − σ ( G ) + 2 = Ω ( ∣ G ∣ ) − ρ ( G ) + 2. \operatorname{md}(G)=\ell_n(G)-\sigma(G)+2=\Omega(|G|)-\rho(G)+2. md ( G ) = ℓ n ( G ) − σ ( G ) + 2 = Ω ( ∣ G ∣ ) − ρ ( G ) + 2.
For abelian groups ρ ( G ) \rho(G) ρ ( G ) is the rank r ( G ) r(G) r ( G ) , and we obtain the following.
Theorem 8.8. Let G G G be a non-trivial finite abelian group.
(1) If G G G is a cyclic p p p -group, then md ( G ) = Ω ( ∣ G ∣ ) \operatorname{md}(G)=\Omega(|G|) md ( G ) = Ω ( ∣ G ∣ ) .
(2) Otherwise,
md ( G ) = Ω ( ∣ G ∣ ) − r ( G ) + 2. \operatorname{md}(G)=\Omega(|G|)-r(G)+2. md ( G ) = Ω ( ∣ G ∣ ) − r ( G ) + 2.
Proof. As G G G is abelian, it is nilpotent and its Sylow p p p -subgroup G p G_p G p satisfies Z ( G p ) = G p Z(G_p)=G_p Z ( G p ) = G p . Also, as each cyclic summand of G p G_p G p has a unique subgroup of order p p p , G p [ p ] ≅ C p r p G_p[p]\cong C_p^{r_p} G p [ p ] ≅ C p r p . Hence,
ρ ( G ) = ∑ p ∣ ∣ G ∣ dim F p G p [ p ] = ∑ p ∣ ∣ G ∣ r p = r ( G ) . \rho(G)=\sum_{p\mid|G|}\dim_{\mathbb{F}_p}G_p[p]=\sum_{p\mid|G|}r_p=r(G). ρ ( G ) = p ∣ ∣ G ∣ ∑ dim F p G p [ p ] = p ∣ ∣ G ∣ ∑ r p = r ( G ) .
Now r ( G ) = 1 r(G)=1 r ( G ) = 1 if and only if G G G is a cyclic p p p -group. So the result follows from Corollary 8.7. □
For cyclic groups the formula involves only the prime factorisation of the order.
Corollary 8.9. Let n ≥ 2 n\ge2 n ≥ 2 and write n = p 1 a 1 ⋯ p s a s n=p_1^{a_1}\cdots p_s^{a_s} n = p 1 a 1 ⋯ p s a s with distinct primes p 1 , … , p s p_1,\ldots,p_s p 1 , … , p s . Then
md ( C n ) = { a 1 , s = 1 , Ω ( n ) − ω ( n ) + 2 , s ≥ 2. \operatorname{md}(C_n)=
\begin{cases}
a_1, & s=1,\\
\Omega(n)-\omega(n)+2, & s\ge2.
\end{cases} md ( C n ) = { a 1 , Ω ( n ) − ω ( n ) + 2 , s = 1 , s ≥ 2. Proof. First let s = 1 s=1 s = 1 . Then C n = C p 1 a 1 C_n=C_{p_1^{a_1}} C n = C p 1 a 1 is a cyclic p 1 p_1 p 1 -group. So by Theorem 8.8 , md ( C n ) = Ω ( p 1 a 1 ) = a 1 \operatorname{md}(C_n)=\Omega(p_1^{a_1})=a_1 md ( C n ) = Ω ( p 1 a 1 ) = a 1 . Now let s ≥ 2 s\ge2 s ≥ 2 . By the Chinese Remainder Theorem,
C n ≅ C p 1 a 1 ⊕ ⋯ ⊕ C p s a s . C_n\cong C_{p_1^{a_1}}\oplus\cdots\oplus C_{p_s^{a_s}}. C n ≅ C p 1 a 1 ⊕ ⋯ ⊕ C p s a s . So the Sylow p i p_i p i -subgroup of C n C_n C n is cyclic, i.e., r p i = 1 r_{p_i}=1 r p i = 1 for all i i i , and hence r ( C n ) = s = ω ( n ) ≥ 2 r(C_n)=s=\omega(n)\ge2 r ( C n ) = s = ω ( n ) ≥ 2 . In particular, C n C_n C n is not a cyclic p p p -group, and so by Theorem 8.8 , md ( C n ) = Ω ( n ) − ω ( n ) + 2 \operatorname{md}(C_n)=\Omega(n)-\omega(n)+2 md ( C n ) = Ω ( n ) − ω ( n ) + 2 . □
Non-nilpotent examples can be handled in the same way when the normal structure is known. For dihedral groups of order 2 n 2n 2 n with n n n odd, all non-trivial proper normal subgroups lie in the rotation subgroup.
Proposition 8.10. Let D 2 n D_{2n} D 2 n be the dihedral group of Definition 2.13 , with odd n ≥ 3 n\ge3 n ≥ 3 . Then
md ( D 2 n ) = md ( C n ) + 1. \operatorname{md}(D_{2n})=\operatorname{md}(C_n)+1. md ( D 2 n ) = md ( C n ) + 1. Proof. Recall from Definition 2.13 that
D 2 n = ⟨ r , s : r n = s 2 = 1 , s r s = r − 1 ⟩ . D_{2n}=\langle r,s:r^n=s^2=1,\ srs=r^{-1}\rangle. D 2 n = ⟨ r , s : r n = s 2 = 1 , sr s = r − 1 ⟩ . Suppose R = ⟨ r ⟩ R=\langle r\rangle R = ⟨ r ⟩ . Then R ≅ C n R\cong C_n R ≅ C n , and every subgroup of R R R is ⟨ r d ⟩ \langle r^d\rangle ⟨ r d ⟩ for some d ∣ n d\mid n d ∣ n . As
r ⟨ r d ⟩ r − 1 = ⟨ r d ⟩ and s ⟨ r d ⟩ s − 1 = ⟨ r − d ⟩ = ⟨ r d ⟩ , r\langle r^d\rangle r^{-1}=\langle r^d\rangle\quad\text{and}\quad s\langle r^d\rangle s^{-1}=\langle r^{-d}\rangle=\langle r^d\rangle, r ⟨ r d ⟩ r − 1 = ⟨ r d ⟩ and s ⟨ r d ⟩ s − 1 = ⟨ r − d ⟩ = ⟨ r d ⟩ , so every subgroup of R R R is normal in D 2 n D_{2n} D 2 n . Conversely, let N ⊴ D 2 n N\trianglelefteq D_{2n} N ⊴ D 2 n . If N N N contains no reflection, then N ≤ R N\le R N ≤ R . Otherwise, let s r i ∈ N sr^i\in N s r i ∈ N for some i i i . As r s = s r − 1 rs=sr^{-1} r s = s r − 1 , we get r ( s r i ) r − 1 = s r i − 2 ∈ N r(sr^i)r^{-1}=sr^{i-2}\in N r ( s r i ) r − 1 = s r i − 2 ∈ N , and so ( s r i − 2 ) ( s r i ) = r 2 ∈ N (sr^{i-2})(sr^i)=r^2\in N ( s r i − 2 ) ( s r i ) = r 2 ∈ N . As n ≥ 3 n\ge3 n ≥ 3 is odd, ord ( r 2 ) = n \operatorname{ord}(r^2)=n ord ( r 2 ) = n , and hence R = ⟨ r 2 ⟩ ≤ N R=\langle r^2\rangle\le N R = ⟨ r 2 ⟩ ≤ N . Then s = ( s r i ) r − i ∈ N s=(sr^i)r^{-i}\in N s = ( s r i ) r − i ∈ N , and thus N = D 2 n N=D_{2n} N = D 2 n . Hence,
N ( D 2 n ) = Sub ( C n ) ⊔ { D 2 n } . \mathcal{N}(D_{2n})=\operatorname{Sub}(C_n)\sqcup\{D_{2n}\}. N ( D 2 n ) = Sub ( C n ) ⊔ { D 2 n } . This also implies that the minimal normal subgroups of D 2 n D_{2n} D 2 n coincide with those of C n C_n C n , and so Soc ( D 2 n ) = Soc ( C n ) \operatorname{Soc}(D_{2n})=\operatorname{Soc}(C_n) Soc ( D 2 n ) = Soc ( C n ) . So for E ∈ Sub ( C n ) E\in\operatorname{Sub}(C_n) E ∈ Sub ( C n ) , we have supp D 2 n ( E ) = supp C n ( E ) \operatorname{supp}_{D_{2n}}(E)=\operatorname{supp}_{C_n}(E) supp D 2 n ( E ) = supp C n ( E ) , and thus by Corollary 3.8, E ∈ N M ( D 2 n ) ⟺ E ∈ N M ( C n ) E\in\mathcal{N}^{\mathcal{M}}(D_{2n})\Longleftrightarrow E\in\mathcal{N}^{\mathcal{M}}(C_n) E ∈ N M ( D 2 n ) ⟺ E ∈ N M ( C n ) . Also Soc ( D 2 n ) ≤ D 2 n \operatorname{Soc}(D_{2n})\le D_{2n} Soc ( D 2 n ) ≤ D 2 n , and so D 2 n ∈ N M ( D 2 n ) D_{2n}\in\mathcal{N}^{\mathcal{M}}(D_{2n}) D 2 n ∈ N M ( D 2 n ) . Thus we get,
N M ( D 2 n ) = N M ( C n ) ⊔ { D 2 n } . \mathcal{N}^{\mathcal{M}}(D_{2n})=\mathcal{N}^{\mathcal{M}}(C_n)\sqcup\{D_{2n}\}. N M ( D 2 n ) = N M ( C n ) ⊔ { D 2 n } . So every maximal M \mathcal{M} M -chain of D 2 n D_{2n} D 2 n looks like
1 = G 0 < G 1 < ⋯ < G r = C n < D 2 n , 1=G_0<G_1<\cdots<G_r=C_n<D_{2n}, 1 = G 0 < G 1 < ⋯ < G r = C n < D 2 n , where 1 = G 0 < G 1 < ⋯ < G r = C n 1=G_0<G_1<\cdots<G_r=C_n 1 = G 0 < G 1 < ⋯ < G r = C n is a maximal M \mathcal{M} M -chain of C n C_n C n . Thus, we get md ( D 2 n ) = md ( C n ) + 1 \operatorname{md}(D_{2n})=\operatorname{md}(C_n)+1 md ( D 2 n ) = md ( C n ) + 1 . □
Finally, in powers of simple groups the M \mathcal{M} M -depth is as small as it can be.
Corollary 8.11. If T T T is a non-abelian finite simple group and k ≥ 1 k \ge1 k ≥ 1 , then
md ( T k ) = { 1 , k = 1 , 2 , k ≥ 2. \operatorname{md}(T^k) =
\begin{cases}
1, & k=1,\\
2, & k \ge2.
\end{cases} md ( T k ) = { 1 , 2 , k = 1 , k ≥ 2. The same formula holds for C p k C_p^k C p k .
Proof. We have by Proposition 5.6 (1),
MinNorm ( T k ) = { T 1 , … , T k } , \operatorname{MinNorm}(T^k)=\{T_1,\ldots,T_k\}, MinNorm ( T k ) = { T 1 , … , T k } , where T i T_i T i is the i i i th coordinate factor. Moreover,
N M ( T k ) = { { 1 , T } , k = 1 , { 1 , T 1 , … , T k , T k } , k ≥ 2. \mathcal{N}^{\mathcal{M}}(T^k)=
\begin{cases}
\{1,T\}, & k=1,\\
\{1,T_1,\ldots,T_k,T^k\}, & k \ge2.
\end{cases} N M ( T k ) = { { 1 , T } , { 1 , T 1 , … , T k , T k } , k = 1 , k ≥ 2. Thus, when k = 1 k=1 k = 1 , md ( T k ) = 1 \operatorname{md}(T^k)=1 md ( T k ) = 1 . When k ≥ 2 k \ge2 k ≥ 2 , the M \mathcal{M} M -subgroups other than 1 1 1 and T k T^k T k are T 1 , … , T k T_1,\ldots,T_k T 1 , … , T k , which are pairwise incomparable. So an M \mathcal{M} M -chain contains at most one T i T_i T i , and as 1 < T 1 < T k 1<T_1<T^k 1 < T 1 < T k is an M \mathcal{M} M -chain, we have md ( T k ) = 2 \operatorname{md}(T^k)=2 md ( T k ) = 2 .
Again by Proposition 5.6 (2), we have
N M ( C p k ) = { 0 } ∪ { U ≤ C p k : dim F p ( U ) = 1 } ∪ { C p k } . \mathcal{N}^{\mathcal{M}}(C_p^k)=\{0\}\cup\{U\le C_p^k:\dim_{\mathbb{F}_p}(U)=1\}\cup\{C_p^k\}. N M ( C p k ) = { 0 } ∪ { U ≤ C p k : dim F p ( U ) = 1 } ∪ { C p k } . So if k = 1 k=1 k = 1 , then md ( C p ) = ℓ n ( C p ) = 1 \operatorname{md}(C_p)=\ell_n(C_p)=1 md ( C p ) = ℓ n ( C p ) = 1 . If k ≥ 2 k \ge2 k ≥ 2 , then in the same way, the one-dimensional subspaces are pairwise incomparable, and 0 < U < C p k 0<U<C_p^k 0 < U < C p k for every one-dimensional subspace U U U . Hence, md ( C p k ) = 2 \operatorname{md}(C_p^k)=2 md ( C p k ) = 2 . Thus, the same formula, for T k T^k T k , holds for C p k C_p^k C p k too. □
Concluding remarks. The results above show that, for finite groups, the M \mathcal{M} M -condition is a condition on the socle, and that the defect N ( G ) ∖ N M ( G ) \mathcal{N}(G)\setminus\mathcal{N}^{\mathcal{M}}(G) N ( G ) ∖ N M ( G ) is measured by the socle length σ ( G ) \sigma(G) σ ( G ) and by the geometry of MinNorm ( G ) \operatorname{MinNorm}(G) MinNorm ( G ) . We mention two directions suggested by this picture.
First, Theorems 3.4 and 4.11 hold for arbitrary groups, and Theorem 3.7 for all normally atomic groups, whereas Example 3.12 shows that the closure-theoretic part of Theorem A fails without atomicity. It would be interesting to describe N M ( G ) \mathcal{N}^{\mathcal{M}}(G) N M ( G ) for classes of infinite groups that are not normally atomic, such as finitely generated nilpotent or residually finite groups, where essential normal subgroups are abundant but minimal normal subgroups may not exist.
Second, Corollary 6.10 only uses the quotients G / R G ( B ) G/R_G(B) G / R G ( B ) . Since crowns are defined for profinite groups as well, one may ask for the analogue of Theorems B and D for the lattice of closed normal subgroups of a profinite group, where σ \sigma σ and ℓ n \ell_n ℓ n may be infinite.
Declarations
Competing interests. The authors declare that they have no competing financial or non-financial interests that are directly or indirectly related to this work.
Data availability. No datasets were generated or analyzed during the present study. [A. NAG] DEPARTMENT OF PURE MATHEMATICS, UNIVERSITY OF CALCUTTA, 35 BALLYGUNGE CIRCULAR ROAD, KOLKATA, WEST BENGAL, 700019, INDIA
Email address: aneesh89884@gmail.com URL: ORCID: https://orcid.org/0009-0007-3915-5316
[P. BISWAS] DEPARTMENT OF MATHEMATICS, JADAVPUR UNIVERSITY, 188 RAJA SUBODH CHANDRA MALLICK ROAD, JADAVPUR, KOLKATA, WEST BENGAL 700032, INDIA Email address: pronayb.math.rs@jadavpuruniversity.in, pronaybiswas1729@gmail.com URL: ORCID: https://orcid.org/0000-0008-2710-1748
[A. GOSWAMI] DEPARTMENT OF MATHEMATICS AND APPLIED MATHEMATICS, UNIVERSITY OF JOHANNESBURG, P.O. BOX 524, AUCKLAND PARK 2006, SOUTH AFRICA
NATIONAL INSTITUTE FOR THEORETICAL AND COMPUTATIONAL SCIENCES (NITHECS), SOUTH AFRICA Email address: agoswami@uj.ac.za URL: ORCID: https://orcid.org/0000-0003-4829-0847