A simultaneous variant of the Hall-Paige theorem and applications

We clearly can assume that N < G. Let \{\beta_i\}_{i \leq k} be automorphisms of G. Since N is characteristic in G and each \beta_i \in \mathrm{Aut}(G), each \beta_i induces an automorphism \bar{\beta}_i \in \mathrm{Aut}(G/N) defined by

Because G/N satisfies \mathcal{P}_k, there exists a bijection \bar{\theta} : G/N \to G/N such that

is a bijection of G/N for all i=1, \dots , k.

Let T be a transversal of N in G. Since \bar{\theta} is a bijection of G/N, there exists a permutation \sigma of T such that

for all t \in T. We also note that every element g \in G is written uniquely as g= tn with t \in T and n \in N.

Now fix t \in T and observe that for every i =1, \dots, k, the map \alpha_{i,t}: G \to G defined as

is also an automorphism of G because \beta_i is. Furthermore, each restriction \alpha_{i,t} |_N lies in \mathrm{Aut}(N), because N is characteristic subgroup in G. Now N also satisfies \mathcal{P}_k hence we obtain a bijection \phi_t such that, for each i=1, \dots, k, the map

is a bijection of N.

We define f: G \to G as

for every t \in T and n \in N and we claim that f is a bijection of G. Indeed, assume

where (t, n) \neq (t',n'). Since T is a transversal of N and \phi_t ( n), \phi_{t'}(n') \in N, we must have \sigma(t) = \sigma(t'), which implies t=t'. Equation (?) then simplifies to \phi_t(n)= \phi_{t}(n'). As \phi_t is a bijection, it follows that n=n', contradicting our initial assumption and thus proving our claim.

To complete the proof, it remains to show that \beta_i(tn) f(tn) is a bijection of G, for every i=1, \dots, k. Assume g\neq g' are distinct elements of G, with g =tn and g'=t'n', so (t,n) \neq (t',n'). If

then \bar{\beta}_i(tN) \bar{\theta}(tN)= \bar{\beta}_i(t'N) \bar{\theta}(t'N). Hence tN= t'N as gN \to \bar{\beta}_i(gN)\bar{\theta}(gN) is a bijection of G/N and so t=t', as T is a transversal. Thus (?) implies that

since \beta_i is an automorphism of G. So \alpha_{i,t}(n)\phi_t(n)= \alpha_{i,t}(n')\phi_t(n') and thus n=n', as the map n \to \alpha_{i,t}(n)\phi_t(n) is a bijection. This leads to (t,n)= (t',n'), a contradiction that completes the proof of the proposition.

First assume that G \cong \mathbb Z_{p^e}^n is a homocyclic p-group. By Theorem 1, for each automorphism \beta_i there exists a matrix A_i \in R_G, where R_G=\mathrm{Mat}_n(\mathbf{Z}) in the homocyclic case, such that \beta_i=\psi(A_i) and A_i \bmod p is non-singular, for all i=1,\dots,k. We observe that it is enough to find a matrix B\in \mathrm{Mat}_n(\mathbf{Z})) such that both B\bmod p and (A_i +B )\bmod p are invertible for every i, as then f:=\psi(B) will be the required automorphism of G.

Define

and denote by \overline{ A_i} the reduction of A_i modulo p. The condition that A_i+B is invertible modulo p is equivalent to requiring that

for each i=1,\dots,k. (By T-\overline{ A_i} we denote the set consisting of the matrices X-\overline{ A_i } with X \in T.)

It is enough to prove that

We have

and, for each i,

Hence

Thus

Furthermore,

Since k\leq p-2, and the function x\mapsto x/(x+1) is increasing, it follows that \frac{p-2}{p-1}\geq \frac{k}{k+1}. Therefore \prod_{t=1}^n(1-p^{-t})>\frac{k}{k+1}. Together with (?), this gives

Hence the theorem holds in this homocyclic case.

Now suppose that

where 1 \leq e_1 < \cdots < e_r and n=m_1+\dots+m_r.

As earlier, for each automorphism \beta_i \in \mathrm{Aut}(G) there exists a matrix A_i \in \mathbf{R}_G such that \beta_i = \psi(A_i) and A_i \bmod p is nonsingular, for all i=1, \dots, k. Notice that, in view of the definition of R_G the reduced matrices \overline{A_i} are block upper triangular matrices of the form

where \overline {A}_i^j \in \mathrm{GL}_{m_j}(\mathbf{F}_p), for all j=1, \dots, r.

For every fixed j, the argument for the homocyclic case provides an invertible matrix \overline {B}^j \in \mathrm{GL}_{m_j}(\mathbf{F}_p) such that the matrices \overline{A}_i^j + \overline{B}^j are all invertible and thus elements in \mathrm{GL}_{m_j}(\mathbf{F}_p).

Now define the block diagonal matrix

Clearly \overline B\in \mathrm{GL}_n(\mathbf{F}_p), because all its diagonal blocks are invertible.

Moreover, for every i=1,\dots,k, the matrix \overline A_i + \overline B is again block upper triangular, and its diagonal blocks are precisely \overline A_i^{j} + \overline B^{j}, for j=1,\dots,r. These blocks are invertible by construction. Hence

Choose a matrix B\in R_G whose reduction modulo p is \overline B. This is possible since \overline B is block diagonal, so it is compatible with the divisibility conditions defining R_G. By Theorem 1, the map f:=\psi(B) is an automorphism of G. Furthermore, for every i=1,\dots,k, the reduction of A_i + B modulo p is \overline A_i + \overline B which is non-singular. Therefore, again by Theorem 1, the map \beta_i +f =\psi(A_i + B ) is an automorphism of G, for every i=1,\dots,k.

Hence f has the desired properties, and the proof of the theorem is complete.

Let G = \mathbf{Z}_{p^e}^n for n \geq 2. Following the notation from the proof of Theorem 2, for each \beta_i (where i \leq p-1), there is a corresponding matrix A_i \in \mathrm{Mat}_n(\mathbf{Z}). We need to find a matrix B \in \mathrm{Mat}_n(\mathbf{Z}) such that \overline{B} and all \overline{A_i} + \overline{B} are invertible in \mathrm{Mat}_n(\mathbf{F}_p), for i=1, \ldots, p-1. Equivalently it is enough to show that

where T = \{X \in \mathrm{Mat}_n(\mathbf{F}_p) \mid \det X = 0\}. For i=1, \dots, p-1, let

and K_0= (T+ I_n) \cap \mathrm{GL}_n(\mathbf{F}_p). (We could have K_0 being one of the K_i for some i.) So to prove the proposition it suffices to show

We observe that |K_i|= |K_0| for every i, as the map S \to -S (\overline{A_i})^{-1} provides a bijection between K_i and K_0.

Let us note that

i.e. K_0 consists of those invertible matrices S \in \mathrm{GL}_n(\mathbf{F}_p) whose 1 is an eigenvalue.

To compute the cardinality of K_0 we look at the set

where \mathbf{F}_p^{n, *} are the non-zero vectors of \mathbf{F}_p^n. Note that \mathrm{GL}_n(\mathbf{F}_p) acts transitively on \mathbf{F}_p^{n, *} hence the Orbit-Stabilizer theorem implies that for any fixed v \in \mathbf{F}_p^{n,*}, the stabilizer \mathrm{Stab}_v in \mathrm{GL}_n(\mathbf{F}_p) has order equal to \frac{|\mathrm{GL}_n(\mathbf{F}_p)|}{ |\mathbf{F}_p^{n,*}|}, as the orbit of v is \mathbf{F}_p^{n, *}. Hence

and thus

On the other hand, for any fixed S \in K_0 if E_S is the eigenspace of S for the eigenvalue 1 then |R|= \sum_{S \in K_0} (|E_S|-1). But for any S \in K_0 its eigenspace for 1 has dimension at least 1 and thus |E_S| \geq p. One of those matrices of K_0 is I_n with E_{I_n}=\mathbf{F}_p^n. Hence

This, along with (?) implies

where the last inequality is strict as n \geq 2. So

This completes the proof of the proposition.

We first note that for p=2 the sum \sum_{x \in \mathbf{Z}_{2^n}} x equals the unique nonzero involution of \mathbf{Z}_{2^n}, that is the element 2^{n-1} \in \mathbf{Z}_{2^n} and so the lemma holds for p=2.

We may assume now that p is odd and we proceed by induction on n. The base case, n=1, is trivial:

Assume the hypothesis holds for values less than n. We will now prove the inductive step for n. To compute the sum, we first split the elements of \mathbf{Z}_{p^n} into two sets, the set of units of the multiplicative \mathbf{Z}_{p^n}^* and the multiples of p:

Applying the inductive hypothesis (for n-1)

the second part of the sum simplifies to:

for some integer r. Since p-1 \geq 2, we have \sum_{k=1}^{p^{n-1}} (kp)^{p-1} \equiv 0 \pmod{p^{n}} and thus

Now consider the endomorphism \phi: \mathbf{Z}_{p^n}^* \to \mathbf{Z}_{p^n}^* defined by \phi(x) = x^{p-1}. Its kernel is

Since \mathbf{Z}_{p^n}^* is cyclic (as p is odd), the order of the kernel is p-1. Given that |\mathbf{Z}_{p^n}^*|= (p-1)p^{n-1}, we know the size of its image |\mathrm{Im} (\phi)|=p^{n-1}. Thus, each element in \mathrm{Im} (\phi) has p-1 preimages in \mathbf{Z}_{p^n}^*.

Consider the set A \subseteq \mathbf{Z}_{p^n}^* defined as

whose cardinality is p^{n-1}. Notice that every element x \in \mathbf{Z}_{p^n}^* satisfies x^{p-1} \equiv 1 \pmod p. Hence \mathrm{Im} (\phi) \subseteq A and by cardinality \mathrm{Im} (\phi) = A. Hence

This completes the proof of the lemma.

Assume T(x)= a_0+a_1x+\dots +a_{p-1}x^{p-1} with a_i \in \mathbf{Z}_{p^n}. The conditions T(x_i)\equiv 0 \pmod{p^n} give the linear system

The determinant of the coefficient matrix is the Vandermonde determinant

Since the elements x_1,\dots,x_p are pairwise distinct modulo p, none of the differences x_j-x_i is divisible by p. Hence every x_j-x_i is a unit in \mathbf{Z}_{p^n}, and therefore the determinant is also a unit in \mathbf{Z}_{p^n}. Thus the Vandermonde matrix above is invertible over \mathbf{Z}_{p^n}. It follows that a_0=a_1=\cdots=a_{p-1}=0 in \mathbf{Z}_{p^n}. Therefore T is the zero polynomial in \mathbf{Z}_{p^n}[x].

For p=2 the statement is supported by the Hall-Paige theorem. For odd primes p, consider the automorphisms \beta_i \in \mathrm{Aut}(G) defined by

for a \in \mathbf{Z}_{p^n} and i=1, \dots , p-1. We assert that no bijection f: G \to G exists such that the maps a \mapsto ia+f(a) are bijections for all i=1, \dots,p-1. Assume, for contradiction, that such a bijection f: G \to G exists, meaning a \mapsto ia+f(a) is a bijection for every i=0,1,\dots,p-1.

We define

Then T(x) is a polynomial in \mathbf{Z}_{p^n}[x] of degree p-1.

Given that ia+f(a) covers all elements of \mathbf{Z}_{p^n}, for a \in \mathbf{Z}_{p^n}, it follows that T(i) = 0 \in \mathbf{Z}_{p^n}, for all i=1, \cdots,p-1. Also T(0) = 0 because f is a bijection of \mathbf{Z}_{p^n}. Applying Lemma 4 we conclude that T(x) must be the zero polynomial in \mathbf{Z}_{p^n}[x]. But, expanding the binomials in the definition of T(x), we see that the coefficient of x^{p-1} equals

As -p^{n-1} \not \equiv 0 \pmod{p^n} we reach a contradiction and thus proving the proposition.

Assume, for contradiction, that the abelian group G does not satisfy the proposition. Thus, G satisfies \mathcal{P}_k for some integer k, and its Sylow p-subgroup G_p is cyclic for a prime p \leq k+1. This implies k \geq p-1, so G satisfies \mathcal{P}_{p-1}. Let G = G_p \times H, where H is an abelian group of order m with \gcd(m,p)=1, and G_p \cong \mathbf{Z}_{p^n} with n \geq 1. Consider the automorphisms B_i \in \mathrm{Aut}(G) defined by

for i=1, \cdots , p-1. By assumption, there exists a bijection F: G \to G such that B_i + F is also a bijection of G for all i. We write

for the projection of G onto G_p composed with the isomorphism G_p \cong \mathbf{Z}_{p^n} (we write a for an element of G_p and its isomorphic element in \mathbf{Z}_{p^n}). (So \pi(B_i(a,h)) = ia.)

The fact that B_i + F is a bijection for every i, implies that the multiset \{\pi(B_i(g)+F(g))\mid g \in G\} picks every element of Z_{p^n} exactly m-times and the same holds for the multiset \{ \pi(F(g) ) \mid g \in G\}.

Similarly with the proof of ? we define

Then T(x) is a polynomial in \mathbf{Z}_{p^n}[x] of degree at most p-1. Furthermore

Hence T(i)= 0 for i=1, \cdots p-1. A similar argument shows that T(0)=0 because F is a bijection of G. The rest of the proof follows th elinbes of that in ?; T must be the zero polynomial by Lemma 4, while its leading coefficient is

As (m,p)=1 the leading coefficient is not zero contradicting the conclusion T=0. The proposition now follows.