Number of Schur covering groups

finite-groupsgroup-cohomologygroup-theoryreference-request

Let $G$ be a finite group with derived subgroup $G'$ and Schur multiplier $M(G)$. A Schur cover of $G$ is a group $H$ with a subgroup $Z\le H'\cap Z(H)$ such that $H/Z\cong G$ and $Z\cong M(G)$. Let $G/G'\cong \prod_{i}C_{n_i}$ and $M(G)\cong\prod_j C_{m_j}$. A theorem of Schur states that the number of non-isomorphic Schur covers of $G$ is at most $\prod_{i,j}\gcd(n_i,m_j)$. I have seen a proof in Karpilovsky's book "The Schur multiplier" (Thm 2.5.14), but I did not quite get through, because of some non-trivial cohomology statements. If I understand the strategy correctly, to every Schur cover of $G$ one assigns (injectively) a central extension $L$ of $G/G'$ by $W\cong M(G)$ such that $W\cap L'=1$.

  1. What is the most direct way of doing this?
  2. Is there an alternative way to prove Schur's theorem (other references)?

I'm particularly interested in the case $(|G/G'|,|M(G)|)=1$ where there is only one Schur cover.

  1. Is it easier to prove this special case? Preferably with as little cohomology as possible.

Best Answer

Note that in the general theory of Schur multipliers and covering groups, $G$ is not restricted to being a finite group.

For a covering group $H$ of $G$, choose elements $h_i \in H$ (for $i$ in some index set) such that their images in $H/H'$ generate $H/H'$. Since $H/Z(H)$ is the fixed group $G$, we can choose the $h_i$ to map onto specific elements of $G$ (i.e. independently of the choice of covering group $H$). Then, as I think you know, then values of the commutators $[h_i,h_j] \in H'$ do not depend on the choice of $H$. (In other words, the different covering groups of $G$ are all isoclinic.)

So, if $R$ is a set of relators of $H/H'$ that define $H/H'$ as an abelian group, then the isomorphism class of $H$ is determined by the values of $r_j \in R$ in $H'$, and these differ in different covering groups only by elements of $Z$. In other words, if $r_j = g_j$ in $H$, and $H_1$ is another cover of of $G$, then $r_j = g_jz_j$ in $H_1$, for some $z_j \in Z$.

(Of course, if $H/H'$ is finitely generated, then we can choose the $h_i$ such that their images generate the cyclic direct factors of $H/H'$, and then we can choose powers of $h_i$ as the relators of $H/H'$ as abelian group.)

Consider the abelian group extension $E$ of $Z$ by $H/H'$ defined by the relations $r_j = z$. This extension is split if and only if there exist $z_j' \in Z$ such that, when we replace the generators of $h_i$ of $H$ by $h_iz_j'$, the relations $r_j$ in $E$ become trivial in $E$, but since the $z_j$ are central in $H$, this is equivalent to the relations $r_j = g_jz_j$ in $H_1$ changing to $r_j = g_j$, in which case $H$ and $H_1$ are isomorphic.

So equal elements of ${\rm Ext}(H/H',Z)$ (the equivalence classes of abelian group extensions of $Z$ by $H/H'$) induce isomorphic covering groups, and hence the number of isomorphism classes of covering groups is at most $|{\rm Ext}(H/H',Z)|$ which, in your notation in the finite case, is equal to $\prod_{i,j} \gcd (n_i,m_j)$.

Of course this is an inequality, and it is possible for Schur covers to be isomorphic even when the corresponding elements of ${\rm Ext}(H/H',Z)$ are distinct. I believe that such isomorphisms are induced by automorphisms of $G$. Interesting examples are the symmetric groups $S_n$. We have $|M(S_n)| = 2$ for all $n \ge 4$, and, for $n \ne 6$, there are two isomorphism classes of covering groups, but for $n=6$ there is only one, with the exceptional outer automorphism of $S_6$ inducing an isomorphism between them.

Related Question