9
$\begingroup$

Let $F_n$ be a free group of rank $n\ge 2$, and $F_n\rightarrow G$ a surjection with $G$ finite. Let $R$ be the kernel. From this, we get an action of $G$ on the abelianization $R/R'$ (a free abelian group of rank $|G|(n-1)+1$).

What is the structure of $R/R'$ as a $G$-representation? I seem to remember reading somewhere that this is just $\mathbb{Z}[G]^{n-1}\oplus\mathbb{Z}$. Is this right? I'm looking for a reference or a proof.

$\endgroup$
2
  • 9
    $\begingroup$ Here’s a cute proof with $\mathbb{C}$ coefficients. Let $X$ be the bouquet of circles and $X’$ the corresponding cover, on which $G$ acts freely by deck transformations. You are asking for $H_1(X’)$ as a $G$-representation. By the Lefschetz Fixed Point theorem, every non-trivial element has trace 1, while the trace of the identity follows from the Nielsen—Schreier formula. Now the result follows by character theory, because we just computed the character of the representation and it matches the one you state in the question. $\endgroup$
    – HJRW
    Commented Feb 22 at 7:16
  • 1
    $\begingroup$ Over Z, I suspect this might not work for the Klein 4-group with its usual 2 generators. I think the copy of Z doesn't have a complement. $\endgroup$ Commented Feb 22 at 19:53

3 Answers 3

11
$\begingroup$

Your memory is correct, at least if you replace $\mathbb{Z}$ with a field $k$ of characteristic $0$. This is a theorem of Gaschütz. See

W. Gaschütz, Über modulare Darstellungen endlicher Gruppen, die von freien Gruppen induziert werden, Math. Z. 60 (1954), 274–286.

One way to prove it is to think of your free group as the fundamental groups of a graph $X$ with one vertex and $n$ loops. Then $R$ is the fundamental group of a regular cover $Y$ of $X$ with deck group $G$. Since the deck group acts freely on $Y$ its cellular chain complex is a chain complex of $G$-modules that can be identified with

$$0 \rightarrow k[G]^n \rightarrow k[G] \rightarrow 0.$$

When you take homology you know you get $k$ in degree $0$, and the theorem now follows from semisimplicity.

$\endgroup$
2
  • $\begingroup$ Do you know if anything can be said about the integral structure? Presumably the integral structure depends on the choice of surjection $F_n\to G$? Are there situations where the relation module is known to be integrally $\mathbb{Z}[G]^{n-1}\oplus\mathbb{Z}$? $\endgroup$ Commented Feb 22 at 20:22
  • $\begingroup$ I don't know the literature about integral results very well, but I'm certain that it depends on the surjection. For cyclic $G$ it is not hard to check that it is integrally $\mathbb{Z}[G]^{n-1} \oplus \mathbb{Z}$. $\endgroup$ Commented Feb 22 at 21:53
6
$\begingroup$

The question has been answered over fields of characteristic $0$ but not over $\mathbb Z$. as originally asked. It turns out that the statement is never true for $G$ noncyclic. It is proved in Lemma 3.4 of Free presentations and relation modules of finite groups by J. S. Williams that if $R/[R,R]\cong A\oplus P$ with $P$ projective and $A\neq 0$ and if $G$ is noncyclic, then $A$ is a faithful $G$-module. In particular, we cannot have $R/[R,R]\cong \mathbb Z\oplus \mathbb ZG^{n-1}$. Of course, if $n=1$ and $G$ is cyclic, then obviously $R/[R,R]\cong \mathbb Z$. The proof uses the decomposition over $\mathbb Q$ and some group cohomology to deduce that $\mathbb Q\otimes A$ must contain at least one of the $\mathbb QG$ copies if $G$ is not cyclic.

$\endgroup$
1
$\begingroup$

I think we may be able to extend the result to fields if the characteristic of the field $F$ does not divide the order of the group $G$.

Proposed Theorem: Let $F_n$ be a free group of rank $n \geq 2$ and let $G$ be a finite group. Suppose $\varphi: F_n \rightarrow G$ is surjective, and let $R = \ker(\varphi)$. If the characteristic of the field $F$ does not divide the order of $G$, then the abelianization of $R$, denoted $R/R'$ (where $R'$ is the commutator subgroup of $R$), is isomorphic as a $G$-module to $F[G]^{n-1} \oplus F$.

Proof Sketch: We first note that since $\varphi$ is surjective, $R$ is a normal subgroup of $F_n$. By the Nielsen-Schreier Theorem, $R$ is also free, and its rank is $1 + |G|(n - 1)$ due to the index of $R$ in $F_n$ being equal to $|G|$.

The group $G$ acts on $F_n$, and hence on $R$, by conjugation. This action is well-defined on the cosets of $R/R'$ and satisfies the module axioms, making $R/R'$ a $G$-module.

By Maschke's theorem, the assumption on the characteristic of $F$ ensures that the group algebra $F[G]$ is semisimple, which means that any $F[G]$-module, in particular $R/R'$, can be decomposed into a direct sum of simple modules.

When considering the cohomology of free groups, we take into account a resolution of $F_n$ that, when restricted to $R$, implies that $H^0(R, F)$ is isomorphic to $F$ and $H^q(R, F)$ is trivial for $q \geq 1$.

The Lyndon-Hochschild-Serre spectral sequence arises from the group extension $1 \rightarrow R \rightarrow F_n \rightarrow G \rightarrow 1$ and simplifies due to the trivial higher cohomology of $R$, leading to a short exact sequence involving $H^0(G, F)$, $H^0(G, F[G])$, and $H^0(G, I)$, where $I$ is the augmentation ideal in $F[G]$.

Finally, the elements of the augmentation ideal $I$ in $F[G]$ have coefficients that sum to zero. The abelianization $R/R'$ interacts with this ideal in a way that allows us to identify parts of $R/R'$ with non-trivial $G$ action. The short exact sequence mentioned earlier helps us analyze this interaction, revealing that the image of the first map is the $F$ summand arising from the trivial action, and the cokernel, $H^0(G, I)$, corresponds to elements of $R/R'$ whose images under the $G$ action lie in the augmentation ideal. These elements form the basis of copies of $F[G]$ within $R/R'$, completing the isomorphism.

$\endgroup$
5
  • $\begingroup$ A naive but genuine question from someone not doing much representation theory, but: isn't $R/R'$ only a $\mathbb{Z}G$-module? Or should one read $R/R'$ as shorthand for $F\otimes (R/R')$? Or perhaps there is a meaning to "isomorphic as $G$-module" that I've just not seen before? $\endgroup$
    – Max Horn
    Commented Feb 23 at 7:42
  • $\begingroup$ @MaxHorn: As a starting point, $R/R’$ is naturally a module over the integers $\mathbb{Z}$, with a $G$-action derived from conjugation. This makes it a $\mathbb{Z}G$-module, where $\mathbb{Z}G$ is the group ring of $G$ over the integers. To consider $R/R’$ in the context of a field $F$ (especially when discussing characteristics not dividing the order of $G$), we often look at $F \otimes_{\mathbb{Z}} (R/R’)$. This operation turns $R/R’$ into an $F[G]$-module $\endgroup$
    – user522465
    Commented Feb 23 at 10:17
  • 1
    $\begingroup$ If the characteristic of $k$ does not divide the order of the group, then the proof in my answer works just fine. All that is needed for it is that representations of $G$ over $k$ are semisimple. $\endgroup$ Commented Feb 23 at 12:42
  • $\begingroup$ In fact you just need $G$ is invertible in the ring $k$. Then a $kG$-module is projective iff it is projective over $k$, which is the case for the chain complex and the homology. $\endgroup$ Commented Feb 23 at 18:59
  • $\begingroup$ Well I'm not so certain because I guess I need to be able to cancel a copy of ZG $\endgroup$ Commented Feb 23 at 19:21

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .