I am stuck in the following two exercises, which are about free groups:

  1. Let $F$ be a free group of rank $n\in\mathbb N$ and $n>1$. Prove that for each $m\in\mathbb N$ there is a subgroup of $F$ of rank $m$.
  2. Let $A$ and $B$ be groups. If there is an element $x\in A\ast B$ of order $k$, then in either $A$ or $B$ there is an element $y$ with $o(y)=k$, where $\ast$ stands for the free product of two groups and $o(\cdot)$ means the order of an element.

What I know about a free group is no more than its definition, some basic properties and the universal property. However, I tried for hours and made no progress, while I searched and found few sources that could help. I thus would like to ask how to solve these two problems. Any help is appreciated.

  • $\begingroup$ 1. is false when $n=1$. $\endgroup$ – Lord Shark the Unknown Nov 9 '17 at 6:41
  • $\begingroup$ @LordSharktheUnknown Yeah, I forgot to note this. Thanks! $\endgroup$ – josephz Nov 9 '17 at 10:19

I guess I have worked out a solution as follows.

1. If $F=F(S)$ with $|S|\geqslant 2$, then there are two elememts $a, b\in S$. The set $\{aa, bb, ab, a(ab)^n\colon n\in \mathbb N\}$ is a freely generating set of a subgroup infinite rank.

2. Suppose that $A=\langle S_A|R_A\rangle$ and $B=\langle S_B|R_B\rangle$. We write respectively $F(\cdot)$ for the freely generated group of a set, e.g. $F(S_A)$, and $N(\cdot)$ for the normal closure of a set in a group, e.g. $N(R_A\cup R_B)$. Then \begin{align} A\ast B=&\langle S_A\cup S_B| R_A\cup R_B\rangle\\ =&F(S_A\cup S_B)/N(R_A\cup R_B). \end{align} We write $\pi\colon F(S_A\cup S_B)\to A\ast B$ for the canonical projection. Suppose that $x=a_1b_1\cdots a_qb_q\in F(S_A\cup S_B)$ and $x^k\in N(R_A\cup R_B)$, i.e., $\pi(x^k)(=\pi(x)^k)=e\in A\ast B$. The case is trivial if each $a_i\in N(R_A)$ and each $b_i\in N(R_B)$. Thus it suffices to consider the case that there is at least one $a_i\notin N(R_A)$ or $b_i\notin N(R_B)$. Since $x^k\in N(R_A\cup R_B)$, from $$ x=a_1b_1a_1^{-1}a_1a_2b_2\cdots a_qb_qx^{k-1}\in N(R_A\cup R_B) $$ it follows that $a_1b_1a_1^{-1}\in N(R_A\cup R_B)$. Hence if $a_1\notin N(R_A)$, then $b_1\in N(R_B)$. Set $x_0=x$. Deleting $a_1$ (if $a_1\in N(R_A)$) or $b_1$ (if $b_1\in N(R_B)$), we arrive at (for example, if $b_1\in N(R_B)$) $x_1=a_1a_2b_2\cdots a_qb_q$. Note that $\pi(x_1)=\pi(x_0)\in A\ast B$. Carrying on this manner we can construct $x_2, x_3,\cdots$ such that $\pi(x_2)=\pi(x_1)=\pi(x_0)$, $\pi(x_3)=\pi(x_2)=\pi(x_0)$, $\cdots$. Therefore recursively we finally arrive at \begin{align} x_{q-2}=ba_qb_q\text{ for some $b\in F(S_B)\setminus N(R_B)$} \end{align} or \begin{align} x_{q-1}=ab_q\text{ for some $a\in F(S_A)\setminus N(R_A)$,} \end{align} which indicates \begin{align} x_{q-1}=bb_q \end{align} or \begin{align} x_q=a. \end{align} Either case shows that there is an $a\in A$ or $bb_q\in B$ of order $k$.

Moreover, from the proof there seems to be a stronger conclusion:

Corollary of 2. If $H\leqslant A\ast B$ is finite, then $H\hookrightarrow A$ or $B$ monomorphically.

Proof. We use induction on the cardinality of generating set of $H$. Let us write $\pi_A\colon F(S_A)\to A$ and $\pi_B\colon F(S_B)\to B$ for the canonical projections.

  • $H=\langle \tilde a\rangle$ is cyclic. Then the case is reduced to 2.

  • $H=\langle \tilde a,\tilde b\rangle$. Let $x_{\tilde a, 0}, x_{\tilde b, 0}\in F(S_A\cup S_B)$ such that $\pi(x_{\tilde a, 0})=\tilde a$, and $\pi(x_{\tilde b, 0})=\tilde b$. Suppose that $$ x_{\tilde a, 0}=a_1b_1\cdots a_qb_q $$ and $$ x_{\tilde b, 0}=c_1d_1\cdots c_rd_r, $$ where $a_i,c_i\in F(S_A)$, $b_i,d_i\in F(S_B)$. Then by 2., $x_{\tilde a,0}$ and $x_{\tilde b, 0}$ can be reduced to $x_{\tilde a,q}$ and $x_{\tilde b,r}$ respectively, satisfying $x_{\tilde a,q}\in F(S_A)\text{ or }F(S_B)$ and $x_{\tilde b, r}\in F(S_A)\text{ or }F(S_B)$. We assert that if it is not the case that $x_{\tilde a, q}, x_{\tilde b,r}\in F(S_A)$ or $x_{\tilde a,q}, x_{\tilde b,r}\in F(S_B)$, then either $x_{\tilde a,q}\in N(R_A)$ or $x_{\tilde b,r}\in N(R_B)$. Indeed, if, for example, $x_{\tilde a,q}\in F(S_A)$ and $x_{\tilde b,r}\in F(S_B)$, since $\tilde a\tilde b$ is of finite order, and $\pi(x_{\tilde a\tilde b, 0})=\tilde a\tilde b$ where $x_{\tilde a\tilde b,0}=x_{\tilde a,q}x_{\tilde b,r}$, we can see from the proof of 2., that $$ x_{\tilde a\tilde b,1}\in F(S_A)\text{ or }F(S_B), $$ i.e., $x_{\tilde a,q}\in N(R_A)$ or $x_{\tilde b,r}\in N(R_B)$. Note that this case is reduced to that $H$ is cyclic. Thus if $\tilde a, \tilde b\neq e\in A\ast B$, then $x_{\tilde a,q}$ and $x_{\tilde b, r}$ both lie in $F(S_A)$ or $F(S_B)$, say in $F(S_A)$. Therefore $$ H\cong\langle \pi_A(x_{\tilde a,q}),\pi_A(x_{\tilde b,r})\rangle\leqslant A. $$

  • Assume the proposition for $n-1$. Then if $H=\langle \tilde a_1,\cdots,\tilde a_n\rangle \leqslant A\ast B$ with each $\tilde a_i$ of finite order, we have $G=\langle \tilde a_1,\cdots,\tilde a_{n-1}\rangle\hookrightarrow A\text{ or }B$, say $G\hookrightarrow A$, and $K=\langle \tilde a_{n-1},\tilde a_n\rangle\hookrightarrow A\text{ or }B$. By $x_{\tilde a_{n-1},q_{n-1}}\in F(S_A)$, and the deduction of the case $H=\langle \tilde a, \tilde b\rangle$, it follows that either $x_{\tilde a_n,q_n}\in N(R_B)$, or $x_{\tilde a_n,q_n}\in F(S_A)$. Thence $H\hookrightarrow A$.

By induction, each finite $H\leqslant A\ast B$ is isomorphic to a subgroup of either $A$ or $B$.

Remark. It should be noted that in the proof of this corollary, the fact is used that if $x_1,\cdots,x_n\in F(S_A)$, then \begin{align} G\triangleq &\langle\pi(x_1),\cdots,\pi(x_n)\rangle\\ \cong&\langle\pi_A(x_1),\cdots,\pi_A(x_n)\rangle\triangleq G_A. \end{align} Indeed, let us define \begin{align} \psi\colon G_A&\to G\\ \pi_A(x)&\mapsto\pi(x). \end{align} By $N(R_A)\subset N(R_A\cup R_B)$ it follows that $\psi$ is well-defined and epimorphic. For each $x\in \langle x_1,\cdots,x_n\rangle$, by definition we have $x\in N(R_A\cup R_B)$ iff $x\in N(R_A)$. Thus $\psi$ is injective, and thus is isomorphic.


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.