I am looking for a reference for the following

Fact 1: if $A$ and $B$ are finitely generated subgroups of infinite index in a finitely generated free group $F$ then there exists $f \in F$ such that $fAf^{-1} \cap B=\{1\}$.

I know several proofs for this, but surely this should be classical, so a reference would be desirable. Modulo a known result of W. Neumann (Neumann, Walter D. `On intersections of finitely generated subgroups of free groups.' Groups—Canberra 1989, 161–170, Lecture Notes in Math., 1456, Springer, Berlin, 1990), this fact would immediately follow from the following

Fact 2: with the assumptions of Fact 1, $F$ cannot be covered by finitely many double cosets of the form AgB, $g \in F$.

I proved a generalization of Fact 2 for hyperbolic groups some time ago, but I do not know of an explicit (classical) reference for it.

  • $\begingroup$ I do not know any references, but I think Fact 1 can be strengthened: $A$ and $B^f$ generate their free product for some $f$. $\endgroup$ Apr 1 '13 at 14:29
  • 1
    $\begingroup$ It seems like something that Karrass and Solitar might have proved in their 69 paper on finitely generated subgroups of a free group. It can be proved easily using Stallings graphs. $\endgroup$ Apr 1 '13 at 14:54
  • $\begingroup$ Yes, Stallings core graphs would give a short proof of both Fact 1 and the statement Anton Klyachko mentions above. I could not find this in the 1969 paper of Karrass-Solitar, though. $\endgroup$ Apr 1 '13 at 15:04
  • $\begingroup$ If you like to avoid mentioning graphs, you may recall that any f.g. subgroup is just a free factor of a finite-index subgroup. This reduces the problem to the case where $A$ is a free factor of $F$... $\endgroup$ Apr 1 '13 at 15:24
  • $\begingroup$ Anton, I do not really see how $A$ being a free factor helps here... $\endgroup$ Apr 1 '13 at 15:26

Here is the proof of Fact 1 which uses only facts that were known already to Klein and Poincare (or at least to Dehn and Nielsen):

  1. Every finitely generated free group can be realized as a discrete group of isometries $F$ of the hyperbolic plane $H^2$, so that the quotient $H^2/F$ has finite area but noncompact; thus, $F$ contains parabolic elements.

  2. If $A\subset F$ is a finitely generated infinite index subgroup then the limit set of $A$ is a proper subset of the unit circle. Indeed, being finitely generated Fuchsian group, $A$ is geometrically finite, i.e., the quotient of the convex hull of its limit set by $A$ has finite area. (This is the only mildly nontrivial ingredient in the entire proof.) In our setting, this would mean that covering map $H^2/A\to H^2/F$ is between two surfaces of finite area and so has to be finite. This contradicts the assumption that $|F:A|=\infty$. Thus, the limit set $L(A)$ of $A$ has empty interior in $S^1$.

  3. Fixed points of parabolic elements of $F$ are dense in $S^1$.

  4. Now, suppose that $A, B$ are finitely generated subgroups of infinite index. By combining 2 and 3 we can find a parabolic element $g\in F$ whose fixed point $p$ is not in the union of the limit sets $L(A) \cup L(B)$ of the groups $A$ and $B$. (Similarly, one can use hyperbolic elements of $F$ since their fixed pairs are dense in $S^1\times S^1$.) Since $g^n, n\to infinity$ converges to $p$ uniformly on compacts away from $p$, there exists $n$ so that $$ g^p(L(A))\cap L(B)=\emptyset. $$ It follows that for $f=g^n$ the groups $fAf^{-1}, B$ have trivial intersection (since fixed points of infinite order elements of a Fuchsian group belong to its limit set).

  • $\begingroup$ Thanks, Misha. I have thought about a similar argument that works more generally for infinite index quasiconvex subgroups in a hyperbolic group. $\endgroup$ Apr 1 '13 at 22:04
  • 1
    $\begingroup$ Ashot: Of course, this is what the general hyperbolic groups argument is modelled on. My point is that this staff is ancient. $\endgroup$
    – Misha
    Apr 1 '13 at 22:16

I'm sorry I don't have a reference but here is a constructive proof of Fact 1 just using basic algebraic topology: Let $F$ be generated by $a_1$ and $a_2$ (the higher rank case is an easy generalization). Let $B$ be the figure eight with basepoint $b_0$. Let $X$ and $Y$ be covers corresponding to $A$ and $B$ with base points $x_0$ and $y_0$. Since $A$ and $B$ are finitely generated and of infinite-index, there exists infinite graphs that are isomorphic, with labelings, to the images of the lifts of all cyclically reduced words which start with $a_1$ (and similarly $a_1^{-1}$) in the universal cover. Call $I_X$ this infinite graph corresponding to $a_1$ (including the initial $a_1$ edge). Call $I_Y$ the infinite graph corresponding to $a_1^{-1}$ in $Y$ but do not include the edge $a_1^{-1}$ in $I_Y$ (so it does not include the initial $a_1^{-1}$ edge).

Take the graph $X$ and remove the infinite piece $I_X$ to obtain $X_0$. Take $Y$ and remove the infinite piece $I_Y$ to obtain $Y_0$. Glue $X_0$ to $Y_0$ in the obvious place. This gives a new graph $Z$ which is a cover of $B$. A path from $x_0$ to $y_0$ in $Z$ is an $f$ you want.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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