7
$\begingroup$

Let $G$ be a group which is Hopfian and given a short exact sequence $1\to F \to H \to G \to 1$ with $F$ a finite normal subgroup of $H$. Is $H$ Hopfian?

$\endgroup$
4
  • 1
    $\begingroup$ I suspect you will need stronger assumptions on $G$, for example that any quotient of $G$ by one of its finite normal subgroups is Hopfian. $\endgroup$
    – ndkrempel
    Mar 22, 2011 at 19:57
  • $\begingroup$ @ ndkrempel: why the math symbols are not readable here? did you mean the group G or H on which i would need stronger assumptions? sorry for disturbing $\endgroup$
    – Poove
    Mar 22, 2011 at 21:03
  • 4
    $\begingroup$ @Poove: Must be a problem at your end. I meant conditions on G. Another possible condition would be to require G to be torsion-free, I think that's enough for the result to hold... It would be helpful if you gave more details about your particular G, if possible. $\endgroup$
    – ndkrempel
    Mar 22, 2011 at 23:07
  • $\begingroup$ @ ndkrempel: G has no normal subgroups isomorphic to a free abelian group of finite rank. (dont want to assume G is torsion free. $\endgroup$
    – Poove
    Mar 23, 2011 at 10:07

2 Answers 2

6
$\begingroup$

An example exists in this paper, p19-20. In fact they also construct a Hopfian group $H$ with finite (cyclic) normal subgroup $F$ and $H/F$ non-Hopfian.

$\endgroup$
5
  • $\begingroup$ @Mark, could you tell on what page the group is constructed? $\endgroup$ Mar 28, 2011 at 19:06
  • 1
    $\begingroup$ Take the Abels' group $A_n$ (pages 19,20). Then $A_n/Z$ is not Hopfian while there is a central cyclic subgroup there F such that $(A_n/Z)/F$ is Hopfian. $\endgroup$
    – user6976
    Mar 28, 2011 at 21:20
  • $\begingroup$ In fact, as I was told by Yves de Cornulier, the group is not a quotient of Abels' group $A_n$ by its center but the quotient of the analog $B_n$ of group $A_n$ over $\mathbb{F}_p[t,t^{-1}]$ over a central copy of $\mathbb{F}_p[t]$ (see 5.10 in the paper). Then the factor-group is not Hopfian, its factor by the (finite) subgroup generated by $t^{-2}$ is Hopfian. If one then kills $t^{-1}$ as well, one get a non-Hopfian group again. I hope he himself will explain here the Hopfian property of these groups. $\endgroup$
    – user6976
    Mar 29, 2011 at 4:05
  • $\begingroup$ That is a very nice example. But how does one show that killing $t^{-2}$ gives a Hopfian group? I doubt that this quotient is residually finite. $\endgroup$ Mar 30, 2011 at 14:24
  • $\begingroup$ @AndreasThom you're right. The modification by "BS." of my argument (see my answer) seems to fix the issue (in any case, $H$ will not be residually finite, because a finite-by-(residually finite) finitely generated group is always Hopfian). $\endgroup$
    – YCor
    Dec 3, 2016 at 20:27
6
$\begingroup$

It seems that the answer is no: there exists an exact sequence $$1\to F\to H\to Q\to 1$$ with $F$ finite (central), $H$ non-Hopfian, and $Q$ Hopfian (with in addition, $H$ finitely generated solvable).

Mark Sapir's answer refers to a group constructed here (see 5.10), which is Abels' group over the ring $\mathbf{F}_p[t,1/t]$, and which probably be used to provide a negative answer to the question. Define the group $G$ as the group of matrices

$$\left(\begin{array}{rrrr} 1 & u_{12} & u_{13} & u_{14}\newline 0 & d_{22} & u_{23} & u_{24}\newline 0 & 0 & d_{33} & u_{34}\newline 0 & 0 & 0 & 1\newline \end{array}\right) $$ where $u_{ij}\in\mathbf{F}_p[t,1/t]$, and $d_{ii}\in\mathbf{F}_p[t,1/t]^\times=\langle t\rangle\mathbf{F}_p^\times$. (Actually one can restrict to $d_{ii}\in\langle t\rangle$ but it's not important.) Let $e_{14}$ denote the map $x\mapsto\begin{pmatrix} 1&0&0&x\\0&1&0&0\\0&0&1&0\\0&0&0&1\end{pmatrix}$.

As suggested by user "BS.", let $N$ be the central subgroup $e_{14}(\mathbf{F}_p[t^2])$ and $M$ the larger central subgroup $e_{14}(\mathbf{F}_p[t^2]\oplus \mathbf{F}_pt^{-1})$. (In a the initial post, $M$ and $N$ were chosen as other central subgroups but unfortunately $G/M$ failed to be Hopfian). So we have the central exact sequence with finite kernel $$1\to M/N\to G/N\to G/M\to 1.$$ Conjugation by the diagonal matrix $(t^2,1,1,1)$ is an automorphism of $G$, which maps $N=e_{14}(\mathbf{F}_p[t^2])$ strictly into itself and hence $H=G/N$ is non-Hopfian.

I haven't completely checked but here are some guidelines to show the group $Q=G/M$ is Hopfian.

Write the original group (given by $4\times 4$ triangular matrices) as $G=D\ltimes U$ with $D=\mathbf{Z}^2$ and $U$ its unipotent part. Set $U^2=[U,U]$ and $U^3=[U,U^2]$, which is central and equal to $e_{14}(\mathbf{F}_p[t,1/t])$. Let $f$ be a surjective endomorphism of $Q$.

  1. check that the center of $G$ is precisely $U^3$. It follows that $f$ induces a surjective endomorphism of $G/U^3$. Since this group is linear, it is Hopfian so this is an automorphism of $G/U^3$.

  2. Describe the group of automorphisms of $G/U^2 = \mathbf{Z}^2\ltimes F_p[t,1/t]^3$. (It should be reasonably easy to describe).

  3. Deduce a description of the group of automorphisms of $G/U^3$, or at least describe how these automorphisms act on $U^2/U^3$, showing that modulo something, the coefficient $12$ is multiplied by a monomial $w\cdot t^a$ ($w\in \mathbf{F}_p^*$) and the coefficient $24$ is multiplied by $vt^b$. So, taking a commutator (which should kill the "modulo something"), we obtain that in the "action of $f$ on $G$", the coefficient $14$ should be multiplied by a nonzero monomial. This multiplication should stabilize $M$ so this is multiplication by a scalar in $\mathbf{F}_p^*$, which implies that f actually induces an automorphism of $Q$.

$\endgroup$
4
  • $\begingroup$ I just edited tex by inserting dollar signs etc. $\endgroup$ Apr 2, 2011 at 23:11
  • $\begingroup$ Sorry if I missed something, but doesn't conjugation by $diag(t^2,1,1,1)$ define an automorphism of $G$ that sends $M$ strictly inside itself (implying $G/M$ non-Hopfian)? $\endgroup$
    – BS.
    Jul 22, 2012 at 16:58
  • 1
    $\begingroup$ On the other hand, a similar idea might work : take $N=F_p[t^2]$ and $M=<N,t^{-1}$. Then $G/N$ is non-Hopfian, but since no non-trivial translation of $\mathbb{Z}$ sends $E={-1,0,2,4,...,2k,...}$ (strictly) inside itself, the conjugation counterexample breaks down and your argument might work. $\endgroup$
    – BS.
    Jul 22, 2012 at 17:41
  • $\begingroup$ @BS. thanks for the correction; this seems to fix the argument. $\endgroup$
    – YCor
    Dec 3, 2016 at 20:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.