Let $G$ be a group generated by a set $X$, and $H$ a subgroup of $G$. A set $T$ of right coset representatives for $H$ in $G$ is called a right transversal. We always assume that $1\in T$. Then

$G=\bigsqcup_{t\in T}Ht$

For all $x\in G$, there exists $\overline{x}\in T$ such that $Hx = H\overline{x}$. The map $G\rightarrow T\ (x\mapsto \overline{x})$ is called a transversal function.

For all $h\in H$, there exist $x_1,\cdots, x_k \in X$ such that $h={x_1}^{\varepsilon_1}\cdots {x_k}^{\varepsilon_k}$. Where $\varepsilon_1,\cdots,\varepsilon_k\in \{-1,1\}$. Set

$t_i:= \left\{ \begin{array}{ll} 1 & (i=0),\\ \overline{t_{i-1}{x_i}^{\varepsilon_i}}=\overline{{x_1}^{\varepsilon_1} \cdots {x_i}^{\varepsilon_i}} & (i\geq 1) \end{array}\right.$

Then

$h = (t_0{x_1}^{\varepsilon_1}\overline{t_0{x_1}^{\varepsilon_1}}^{-1}) \cdots (t_{k-1}{x_k}^{\varepsilon_k}\overline{t_{k-1}{x_k}^{\varepsilon_k}}^{-1})t_k$

Since $t_k=\overline{h}=1$, we have

$h = (t_0{x_1}^{\varepsilon_1}\overline{t_0{x_1}^{\varepsilon_1}}^{-1}) \cdots (t_{k-1}{x_k}^{\varepsilon_k}\overline{t_{k-1}{x_k}^{\varepsilon_k}}^{-1})$

If $\varepsilon_i=-1$, then

$\begin{array}{rcl} t_{i-1}{x_i}^{\varepsilon_i}\overline{t_{i-1}{x_i}^{\varepsilon_i}}^{-1} &=& t_{i-1}{x_i}^{-1}\overline{t_{i-1}{x_i}^{-1}}^{-1}\\ &=& \left(\overline{t_{i-1}x_i^{-1}}x_it_{i-1}^{-1}\right)^{-1}\\ &=& \left(t_ix_i\overline{t_ix_i}^{-1}\right)^{-1} \end{array}$

Hence,

$H=\langle\ tx\overline{tx}^{-1}\ |\ t\in T,\ x\in X\rangle$


From now on, set

$F:=F(X)$the free group generated by $X$
$N:=\mathrm{NC}_{F(X)}(R)$the normal closure of $R$ in $F(X)$
$G=\langle X\ |\ R \rangle = F/N$
$H\leq G$
$\pi:F\rightarrow F/N=G$the natural projection
$E:=\pi^{-1}(H)$
$1 \in T\subseteq F$the set of right coset representatives for $E$ in $F$

In this case, the map

$ \begin{array}{ccc} E\backslash F & \rightarrow & H\backslash G\\ E w & \mapsto & H\pi(w) \end{array} $

is bijective. Let $Y$ be the set whose elements are indexed as follows:

$Y:=\{y_{t,x}\ |\ t\in T,\ x\in X,\ tx \neq \overline{tx}\}$

From now on,

we will assume that $T$ is prefix closed.

We will show that the group surjection

$\begin{array}{cccc} \varphi:& F(Y) & \rightarrow & E\\ & y_{t,x} & \mapsto & tx\overline{tx}^{-1} \end{array}$

is an isomorphism. We extend the definition as follows:

$\displaystyle{y_{t,\ x^{\varepsilon}}:= \left\{ \begin{array}{ll} y_{t,\ x} & (\varepsilon = 1,\ tx\neq \overline{tx})\\ y_{\overline{tx^{-1}},\ x}^{-1} & (\varepsilon = -1,\ tx^{-1}\neq \overline{tx^{-1}})\\ 1 & (tx^{\varepsilon}=\overline{tx^{\varepsilon}}) \end{array} \right.}$

Denote by $W(X)$ the free monoid generated by $X\sqcup X^{-1}$, and define a map $\sigma:T\times W(X)\rightarrow F(Y)$ by

$\sigma(t,w):= y_{t_0,\ x_1^{\varepsilon_1}}\cdots y_{t_{r-1},\ x_r^{\varepsilon_r}}$

where $w = x_1^{\varepsilon_1}\cdots x_r^{\varepsilon_r}\ (\varepsilon_1,\ \cdots,\ \varepsilon_r\in\{-1,\ 1\} )$, $t_i:=\overline{tx_1^{\varepsilon_1}\cdots x_i^{\varepsilon_i}}$.

It is easy to see that

$\begin{array}{rl} \sigma(t,x) = y_{t,x} & (t\in T,\ x\in X)\\ \sigma(t,uv)= \sigma(t,u)\sigma(\overline{tu},v) & (t\in T,\ u,v\in W(X)) \end{array}$

For $t\in T$, define

$\rho_t: F\rightarrow F(Y)$ via $\rho_t(w):=\sigma(t,w)$ for $w\in F$.

To show that $\rho_t$ is well-defined, it is enough to prove that

$\sigma(t,uxx^{-1}v)=\sigma(t,ux^{-1}xv) = \sigma(t,uv)$

Since $\rho_1(uv)=\rho_1(u)\rho_{\overline{u}}(v)\ (u,v\in F)$, $\rho_1|_E:E\rightarrow F(Y)$ is a group homomorphism.

We will show that

$\rho_1(tx\overline{tx}^{-1})=y_{t,x}\quad (t\in T,\ x\in X)$

Hence,

$\rho_1\varphi = id_E$

This implies that $\varphi:F(Y)\rightarrow E$ is an isomorphism of groups.

[Cor]

If $m:=|X|<\infty,\ n:=|T|<\infty$, then $E$ is a free group of rank $mn-n+1$.

Set

$S:=\{\rho_1(twt^{-1})\ |\ t\in T,\ w\in R\}$,

then the induced group homomorphism $\overline{\varphi}:\langle Y\ |\ S\rangle \rightarrow H$ is an isomorphism:

$\begin{xy} <0em,0em>*+{F(Y)}="fy", <7em,0em>*+{E}="e", <0em,-5em>*+{\langle Y\ |\ S\rangle}="h1", <7em,-5em>*+{H}="h", <3.5em,-2.5em>*+{\circlearrowleft}="c", "fy";"e" **@{-} ?>*@{>} ?<>(.5)*!/_0.6em/{\varphi}, "fy";"h1" **@{-} ?>*@{>>}, "e";"h" **@{-} ?>*@{>>}, "h1";"h" **@{-} ?>*@{>} ?<>(.5)*!/_0.6em/{\overline{\varphi}} \end{xy}$

For $t\in T,\ w\in R$,

$\rho_{\overline{tw}}(t^{-1})=\rho_1(t)^{-1}$

Hence

$\rho_1(twt^{-1})=\rho_1(t)\rho_t(w)\rho_{\overline{tw}}(t^{-1})=\rho_1(t)\rho_t(w)\rho_1(t)^{-1}$.

This shows that the set $S$ can be replaced by the follwing set:

$\{\rho_t(w)\ |\ t\in T,\ w\in R\}$.