A Pair of Garside shadows

Let $P=Y\cap X^{0}$, where $Y$ is a Shi component. It suffices to show that for each $p_{0},p_{n}\in P$ there is $p\in P$ satisfying $p_{0}\succeq p\preceq p_{n}$. Let $(p_{0},p_{1},\ldots,p_{n})$ be the vertices of a geodesic edge-path $\pi$ in $X^{1}$ from $p_{0}$ to $p_{n}$, which lies in $Y$. Let $L=\max_{i=0}^{n}\ell(p_{i})$.

We will now modify $\pi$ and replace it by another embedded edge-path from $p_{0}$ to $p_{n}$ with vertices in $P$, so that there is no $p_{i}$ with $p_{i-1}\prec p_{i}\succ p_{i+1}$. Then we will be able to choose $p$ to be the smallest $p_{i}$ with respect to $\preceq$.

If $p_{i-1}\prec p_{i}\succ p_{i+1}$, then let $\mathcal{W}_{r},\mathcal{W}_{q}$ be the (intersecting) walls separating $p_{i}$ from $p_{i-1},p_{i+1}$, respectively. Moreover, if $r$ and $q$ do not commute, then $r,q$ are sharp-angled, with $\mathrm{id}$ in a geometric fundamental domain for $\langle r,q\rangle$. We claim that all the elements of the residue $R=\langle r,q\rangle(p_{i})$ lie in $P$.

Indeed, since $p_{i-1},p_{i+1}$ are both in $P$, we have that $\mathcal{W}_{r},\mathcal{W}_{q}\notin\mathcal{E}$. It remains to justify that each wall $\mathcal{W}^{\prime}\neq\mathcal{W}_{r},\mathcal{W}_{q}$ that is the translate of $\mathcal{W}_{r}$ or $\mathcal{W}_{q}$ under an element of $\langle r,q\rangle$ does not belong to $\mathcal{E}$. We can thus assume that $r$ and $q$ do not commute, since otherwise there is no such $\mathcal{W}^{\prime}$. Since $\mathcal{W}_{r}\notin\mathcal{E}$, there is a wall $\mathcal{U}$ separating $\mathrm{id}$ from $\mathcal{W}_{r}$. By Lemma 2.2, there is a wall $\mathcal{U}^{\prime}$ separating $\mathrm{id}$ from $\mathcal{W}^{\prime}$, justifying the claim.

We now replace the subpath $(p_{i-1},p_{i},p_{i+1})$ of $\pi$ by the second embedded edge-path with vertices in the residue $R$ from $p_{i-1}$ to $p_{i+1}$. Since all the elements of $R$ are $\prec p_{i}$ [Ronan_2009, Thm 2.9], this decreases the complexity of $\pi$ defined as the tuple $(n_{L},\ldots,n_{2},n_{1})$, where $n_{j}$ is the number of $p_{i}$ in $\pi$ with $\ell(p_{i})=j$, with lexicographic order. After possibly removing a subpath, we can assume that the new edge-path is embedded. After finitely many such modifications, we obtain the desired path. ∎

Let $k$ be the minimal number of distinct Shi components traversed by a geodesic edge-path $\gamma$ from $h$ to $g$. We proceed by induction on $k$, where for $k=1$ we have $m(g)=m(h)$. Suppose now $k>1$. If a neighbour $f$ of $h$ on $\gamma$ lies in the same Shi component as $h$, then we can replace $h$ by $f$. Thus we can assume that $f$ lies in a different Shi component than $h$. Consequently, the wall $\mathcal{W}_{r}$ separating $h$ from $f$ belongs to $\mathcal{E}$. Since $g\preceq f$, by the inductive assumption we have $m(g)\preceq m(f)$. Thus it suffices to prove $m(f)\preceq m(h)$.

In the first case, where for every neighbour $h^{\prime}$ of $h$ on a geodesic edge-path from $h$ to $\mathrm{id}$, the wall separating $h$ from $h^{\prime}$ belongs to $\mathcal{E}$, we have $h=m(h)$ and we are done. Otherwise, let $\mathcal{W}_{q}$ be such a wall separating $h$ from $h^{\prime}$ outside $\mathcal{E}$. If $r$ and $q$ do not commute, then $r,q$ are sharp-angled, with $\mathrm{id}$ in a geometric fundamental domain for $\langle r,q\rangle$. By Lemma 2.2, among the walls in $\langle r,q\rangle\{\mathcal{W}_{r},\mathcal{W}_{q}\}$ only $\mathcal{W}_{r}$ belongs to $\mathcal{E}$. Let $\bar{h},\bar{f}$ be the vertices opposite to $f,h$ in the residue $\langle r,q\rangle h$. We have $m(\bar{h})=m(h),m(\bar{f})=m(f)$. Replacing $h,f$ by $\bar{h},\bar{f}$, and possibly repeating this procedure finitely many times, we arrive at the first case. ∎

For any neighbour $h^{\prime}$ of $h$ on a geodesic edge-path from $h$ to $g$, the wall $\mathcal{W}$ separating $h$ from $h^{\prime}$ belongs to $\mathcal{E}$. Consequently, we also have $g^{-1}\mathcal{W}\in\mathcal{E}$, and so $g^{-1}h\in M$. ∎

The proof is identical to that of Theorem 1.1, with $P$ replaced by $Q$. The vertices of a geodesic edge-path $\pi$ in $X^{1}$ from $p_{0}$ to $p_{n}$ belong to $Q$ by Remark 3.1. We also make the following change in the proof of the claim that all the elements of $R=\langle r,q\rangle(p_{i})$ lie in $Q$. Namely, since $T=T(p^{-1}_{i})$ equals $T(p^{-1}_{i-1})$, we have $\mathcal{W}_{r}\notin\partial T$. Analogously we obtain $\mathcal{W}_{q}\notin\partial T$. If $r$ and $q$ do not commute, we have that $T$ is contained in a geometric fundamental domain for $\langle r,q\rangle$, and so we also have $\mathcal{W}^{\prime}\notin\partial T$ for any $\mathcal{W}^{\prime}$ that is a translate of $\mathcal{W}_{r}$ or $\mathcal{W}_{q}$ under an element of $\langle r,q\rangle$. This justifies the claim. ∎

The proof structure is similar to that of Lemma 2.3. We need to justify that for $g\preceq h$, we have $\mu(g)\preceq\mu(h)$, where we induct on the minimal number $k$ of distinct cone type components traversed by a geodesic edge-path $\gamma$ from $h$ to $g$. Suppose $k>1$, and let $Q=Q(T)$ be the cone type component containing $h$. If a neighbour $f$ of $h$ on $\gamma$ lies in $Q$, then we can replace $h$ by $f$. Thus we can assume $f\notin Q$. Consequently, the wall $\mathcal{W}_{r}$ separating $h$ from $f$ belongs to $\partial T$. Since $g\preceq f$, by the inductive assumption we have $\mu(g)\preceq\mu(f)$. Thus it suffices to prove $\mu(f)\preceq\mu(h)$.

If for every neighbour $h^{\prime}$ of $h$ on a geodesic edge-path from $h$ to $\mathrm{id}$, the wall separating $h$ from $h^{\prime}$ belongs to $\partial T$, we have $h=\mu(h)$ and we are done. Otherwise, let $\mathcal{W}_{q}$ be such a wall separating $h$ from $h^{\prime}$ outside $\partial T$. Let $\bar{h},\bar{f}$ be the vertices opposite to $f,h$ in the residue $\langle r,q\rangle h$, and let $f^{\prime}=rqh$. It suffices to prove $\mu(\bar{h})=\mu(h),\mu(\bar{f})=\mu(f)$. To justify $\mu(\bar{h})=\mu(h)$, or, equivalently, $\bar{h}\in Q$, it suffices to observe that among the walls in $\langle r,q\rangle\{\mathcal{W}_{r},\mathcal{W}_{q}\}$ only $\mathcal{W}_{r}$ belongs to $\partial T$: Indeed, if $r$ and $q$ do not commute, then $r,q$ are sharp-angled, with $T$ in the geometric fundamental domain $F$ for $\langle r,q\rangle$ containing $\mathrm{id}$.

It remains to justify $\mu(\bar{f})=\mu(f)$, or, equivalently, $T(\bar{f}^{-1})=\widetilde{T}$ for $\widetilde{T}=T(f^{-1})$. Since $\widetilde{T}\cap F=T$, to show, for example, $T(f^{\prime-1})=\widetilde{T}$, it suffices to show that the wall $\mathcal{W}=r\mathcal{W}_{q}$ does not belong to $\partial\widetilde{T}$.

Otherwise, let $b\in\widetilde{T}$ be adjacent to $\mathcal{W}$. Then $rb\in F$ is adjacent to $\mathcal{W}_{q}$, which is outside $\partial T$. Consequently, $rb\notin T$. Thus there is a wall $\mathcal{W}^{\prime}$ separating $\mathrm{id}$ from $h$ and $rb$. Note that $\mathcal{W}^{\prime}\neq\mathcal{W}_{r}$ and so $\mathcal{W}^{\prime}$ separates $\mathrm{id}$ from $f$. Since $\mathrm{id}$ lies on a geodesic edge-path from $f$ to $b$, we have that $\mathcal{W}^{\prime}$ does not separate $\mathrm{id}$ from $b$. Thus $r\mathcal{W}^{\prime}$ separates $r$ and $rb$ from $f,h,b$, and $\mathrm{id}$, since, again, $\mathrm{id}$ lies on a geodesic edge-path from $f$ to $b$.

Consider the distinct connected components $\Lambda_{1},\Lambda_{2},\Lambda_{3},\Lambda_{4}$ of $X^{1}\setminus(\mathcal{W}_{r}\cup r\mathcal{W}^{\prime})$ with $\mathrm{id}\in\Lambda_{1},b\in\Lambda_{2},r\in\Lambda_{3},rb\in\Lambda_{4}$. Since $\mathrm{id}$ and $r$ are interchanged by the reflection $r$ and they lie in the opposite connected components, we have $r\Lambda_{2}\subsetneq\Lambda_{1}$. On the other hand, since $b$ and $rb$ lie in the opposite connected components, we have $r\Lambda_{1}\subsetneq\Lambda_{2}$, which is a contradiction.

This proves that the wall $\mathcal{W}$ does not belong to $\partial\widetilde{T}$, and hence neither does any other wall in $\langle r,q\rangle\{\mathcal{W}_{r},\mathcal{W}_{q}\}$. Consequently $T(\bar{f}^{-1})=\widetilde{T}$, as desired. ∎