--- title: Piecewise-Canonical Homeomorphisms of Edge Shifts url: https://www.emergentmind.com/topics/piecewise-canonical-homeomorphisms-of-edge-shifts type: topic --- # Piecewise-Canonical Homeomorphisms of Edge Shifts Searching arXiv for the cited work and closely related papers on piecewise-canonical homeomorphisms of edge shifts. Piecewise-canonical homeomorphisms of edge shifts are homeomorphisms obtained by partitioning an edge-shift space into finitely many cylinder sets and acting on each part by a canonical prefix-replacement map. In the formulation developed for finite directed graphs $\Gamma$, the underlying space $S(\Gamma\mid Y)$ consists of infinite paths starting from a finite ordered multiset $Y$ of initial vertices, topologized by cylinder sets $C_w$ determined by finite prefixes $w$; when two prefixes $p,q$ have the same terminal vertex, there is a canonical homeomorphism $\Phi_{p,q}:C_p\to C_q$ given by $\Phi_{p,q}(p\alpha)=q\alpha$ [2508.02582]. A piecewise-canonical homeomorphism is then a global homeomorphism assembled from finitely many such canonical cylinderwise identifications. In parallel, the sliding-block-code analysis of local homeomorphisms and $*$-commutation for one-sided shifts supplies a classification blueprint: local homeomorphism behavior is controlled by weakly progressive block maps, $*$-commutation with the shift by regressive block maps, and the conjunction yields $k$-fold covering maps [1010.5739]. The combination of these viewpoints places piecewise-canonical homeomorphisms at the intersection of symbolic dynamics, Cantor dynamics, diagram groups, and the algebra of topological full groups. ## 1. Edge shifts, cylinders, and canonical prefix exchange Let $\Gamma=(V_\Gamma,E_\Gamma,i,t)$ be a finite directed graph, and let $Y$ be a finite ordered multiset over $V_\Gamma$. The edge-shift space considered in the recent literature is $$ S(\Gamma\mid Y):=\{y e_1 e_2\cdots \mid y\in Y,\ e_j\in E_\Gamma,\ i(e_1)=y,\ t(e_j)=i(e_{j+1})\text{ for }j\ge 1\}, $$ the set of infinite directed paths starting from the specified initial multiset [2508.02582]. Its finite-prefix language, including a trivial empty path at each $y\in Y$, is denoted $L(\Gamma\mid Y)$. For each $w\in L(\Gamma\mid Y)$, the associated cylinder $$ C_w=\{w\alpha\in S(\Gamma\mid Y)\} $$ is a basic clopen set, and these cylinders generate the topology, making $S(\Gamma\mid Y)$ a Cantor space plus possibly isolated points [2508.02582]. The fundamental local model is the canonical homeomorphism between cylinders of the same terminal color. If $p,q\in L(\Gamma\mid Y)$ satisfy $t(p)=t(q)$, then $$ \Phi_{p,q}:C_p\to C_q,\qquad \Phi_{p,q}(p\alpha)=q\alpha, $$ is a homeomorphism, often written in prefix-exchange notation as $\Phi_{p,q}:[p]\to[q]$ [2508.02582]. This is the basic “piece” from which piecewise-canonical maps are assembled. A homeomorphism $f:S(\Gamma\mid Y)\to S(\Gamma\mid Y)$ is piecewise-canonical if there exist finite cylinder partitions $\{C_{w_1},\dots,C_{w_k}\}$ and $\{C_{v_1},\dots,C_{v_k}\}$ of $S(\Gamma\mid Y)$ such that for each $i$ one has $$ f|_{C_{w_i}}=\Phi_{w_i,v_i}, $$ equivalently $f(w_i\alpha)=v_i\alpha$ on each piece [2508.02582]. The group of all such homeomorphisms is denoted $G(\Gamma\mid Y)$. This definition places the emphasis on prefix replacement rather than arbitrary local coding. The resulting maps are globally homeomorphisms by construction, but their local symbolic structure is still naturally compared with sliding-block codes, especially when one asks for compatibility with the shift or for classification via covering behavior. This suggests a synthesis between the graph-theoretic cylinder calculus of piecewise-canonical maps and the block-map criteria developed for one-sided shifts [1010.5739]. ## 2. Group structure, diagrams, and representative families Under composition, piecewise-canonical homeomorphisms form a group $G(\Gamma\mid Y)$ [2508.02582]. A broader groupoid $G(\Gamma)$ is obtained by allowing morphisms between spaces $S(\Gamma\mid Y)$ and $S(\Gamma\mid Y')$ for varying ordered multisets $Y,Y'$, with composition implemented by refining and de-refining cylinder partitions. In diagrammatic language, elements are encoded by forest-pair diagrams $(F_D,f,F_R)$ in which $F_D$ and $F_R$ are finite complete rooted subforests of the infinite rooted forest of prefixes $L(\Gamma\mid Y)$, and $f$ is a leaf-color-preserving bijection between them [2508.02582]. A basic structural fact is that every element factors, after reductions, as a split part, followed by a permutation part, followed by a merge part. Split diagrams subdivide cylinders into their children; permutation diagrams permute cylinders of equal color; merge diagrams invert splits [2508.02582]. Correspondingly, $G(\Gamma)$ is generated by the infinite set of all split diagrams together with all permutation diagrams [2508.02582]. Several standard groups occur as special cases. | Choice of $\Gamma$ and $Y$ | Resulting group | |---|---| | One vertex with $n$ loop-edges; $Y$ consists of $k$ copies of the vertex | Higman–Thompson group $V_{n,k}$ | | Disjoint union of $n$ half-lines | Houghton group $H_n$ | | Suitable graph with two loops plus an isolated loop | Thompson-like group $QV$ | These identifications show that piecewise-canonical homeomorphisms of edge shifts form a unifying framework for several diagrammatic and Cantor-dynamical groups [2508.02582]. In the irreducible case, and with $Y=V_\Gamma$ containing one copy of each vertex, Matui’s results imply that the associated topological full group is of type $F_\infty$ and that its commutator subgroup is simple and finitely generated [2508.02582]. The same source states more generally that for any $\Gamma,Y$, the commutator subgroup sits as a normal subgroup of index at most two and often is simple [2508.02582]. The forest-pair formalism also provides normal forms. Every element of $G(\Gamma\mid Y)$ has a unique reduced forest-pair diagram; any two representations differ by a sequence of expansions and reductions, either degenerate or regular [2508.02582]. When $\Gamma$ is irreducible, this yields finite generation and indeed finite presentability [2508.02582]. ## 3. Sliding-block codes, local homeomorphisms, and the weakly progressive criterion The symbolic-dynamical counterpart begins with finite alphabets $A,B$, a block map of memory $n\ge 1$, $$ \beta:A^n\to B, $$ and the induced sliding-block code $\phi=T_\beta:A^\mathbb{N}\to B^\mathbb{N}$ defined by $$ (\phi(x))_i=\beta(x_i x_{i+1}\cdots x_{i+n-1}),\qquad i=1,2,3,\dots $$ [1010.5739]. Such maps are continuous and commute with the one-sided shift $\sigma$ [1010.5739]. A continuous map $f:X\to Y$ is a local homeomorphism if every point has a neighborhood on which $f$ restricts to a homeomorphism onto an open image [1010.5739]. For sliding-block codes, the classical sufficient condition is progressiveness: a block map $\beta:A^n\to A$ is progressive if for every fixed prefix $w\in A^{n-1}$, the map $$ P_{\beta,w}:a\mapsto \beta(w\cdot a) $$ is a bijection of $A$ [1010.5739]. Exel–Renault’s theorem states that if $\beta$ is progressive, then $T_\beta$ is a local homeomorphism [1010.5739]. The key refinement in Willis’s work is that progressiveness is not necessary. The paper gives an explicit block map on $A=\{0,1,2,3\}$: \[ \beta(00)=0,\ \beta(01)=0,\ \beta(02)=1,\ \beta(03)=1, \] \[ \beta(10)=3,\ \beta(11)=3,\ \beta(12)=2,\ \beta(13)=2, \] \[ \beta(20)=2,\ \beta(21)=2,\ \beta(22)=3,\ \beta(23)=3, \] \[ \beta(30)=1,\ \beta(31)=1,\ \beta(32)=2,\ \beta(33)=0, \] for which no progressive block map defines the same sliding-block code, yet the induced map is a local homeomorphism; on the clopen sets $Z(0)\cup Z(1)$ and $Z(2)\cup Z(3)$ it is a homeomorphism onto the full shift [1010.5739]. This counterexample motivates the generalized notion of weak progressiveness. A block map $\beta:A^n\to A$ is weakly progressive of order $m\ge 1$ if for every $w\in A^{n-1}$ and every target block $v\in A^m$ beginning with $\beta(w\cdot a)$ for some $a$, there exists a unique length-$m$ extension $a\cdot u\in A^m$ such that the length-$m$ sliding-block output equals $v$ [1010.5739]. Proposition 3.10 asserts that if $\beta$ is weakly progressive, then for each $w\in A^{n-1}$ the restriction $$ \phi:Z(w)\to \bigcup_{a\in A} Z(\beta(w\cdot a)) $$ is bijective [1010.5739]. Theorem 3.11 then concludes that every sliding-block code arising from a weakly progressive block map is a local homeomorphism [1010.5739]. For piecewise-canonical edge-shift homeomorphisms, this criterion furnishes the local-homeomorphism side of the classification blueprint: whenever a cylinder restriction is modeled by a sliding-block code, weak progressiveness is the sufficient hypothesis ensuring that the piece is locally homeomorphic in the symbolic sense [1010.5739]. This suggests that piecewise-canonical maps compatible with shift dynamics should be locally constrained by weakly progressive block maps on each cylinder piece. ## 4. $*$-commutation, regressive block maps, and covering behavior Willis also isolates the correct symbolic criterion for a stronger compatibility with the shift. Two endomorphisms $S,T$ of a set $X$ are said to $*$-commute if 1. $S\circ T=T\circ S$, and 2. whenever $S(y)=T(z)$, there exists a unique $x\in X$ with $T(x)=y$ and $S(x)=z$ [1010.5739]. For sliding-block codes, the relevant pair is $\sigma$ and $T_\beta$. A block map $\beta:A^n\to A$ is regressive if for each fixed suffix $w\in A^{n-1}$, the map $$ R_{\beta,w}:a\mapsto \beta(a\cdot w) $$ is a bijection of $A$ [1010.5739]. Theorem 4.9 states that a sliding-block code $\phi=T_\beta$ $*$-commutes with the shift $\sigma$ if and only if $\beta$ is regressive [1010.5739]. The proof uses the bijectivity of the suffix maps to build the unique lift required in the definition of $*$-commutation, and conversely shows that the lifting property forces injectivity, hence bijectivity, of each $R_{\beta,w}$ on finite alphabets [1010.5739]. The combined local-homeomorphism and $*$-commutation hypotheses have a covering-theoretic consequence. A continuous surjection $p:E\to B$ is a $k$-fold covering if every $b\in B$ has a neighborhood $U$ such that $p^{-1}(U)$ is the disjoint union of $k$ open sets, each mapped homeomorphically onto $U$ [1010.5739]. Proposition 5.13 states that if $\phi=T_\beta$ is both a local homeomorphism and $*$-commutes with $\sigma$, then $\phi$ is a $k$-fold covering map for some $k$ [1010.5739]. The argument combines local homeomorphism with the constancy and finiteness of fiber cardinality implied by $*$-commutation [1010.5739]. Theorem 5.14 packages the conclusion into an equivalence: for $\phi=T_\beta:A^\mathbb{N}\to A^\mathbb{N}$, the following are equivalent: - $\phi$ is a local homeomorphism and $*$-commutes with $\sigma$; - $\beta$ is regressive and $\phi$ is a $k$-fold covering map, with $k=|\phi^{-1}(y)|$ [1010.5739]. This characterization is directly relevant to piecewise-canonical edge-shift homeomorphisms when one imposes global shift commutation. In that setting, each cylinderwise block-code piece should satisfy the regressive condition to obtain $*$-commutation and the weakly progressive condition to obtain local homeomorphism; the resulting pieces then exhibit uniform finite-sheeted covering behavior [1010.5739]. A plausible implication is that the symbolic rigidity expressed by regressiveness plays, on each piece, the same role as the canonical tail-preserving condition in the graph-theoretic definition. ## 5. Classification blueprint for piecewise-canonical shift-commuting homeomorphisms The provided classification blueprint adapts the one-sided-shift results to edge shifts. An edge shift $X_G$ is specified by a finite directed graph $G=(V,E)$, and a piecewise-canonical homeomorphism $\Psi$ of $X_G$ is, by definition, a global homeomorphism that on finitely many cylinder sets acts by sliding-block codes, equivalently by finite-memory maps on edges [1010.5739]. According to this blueprint, each cylinder restriction $\Psi|_{Z(w)}$ arising from a block map $\beta_w$ must satisfy the following conditions [1010.5739]: 1. to be locally a homeomorphism, $\beta_w$ must be weakly progressive; 2. to $*$-commute with $\sigma$, $\beta_w$ must be regressive; 3. therefore each piece is a $k_w$-fold covering for some $k_w$; 4. the compatibility conditions on overlaps of the partition cylinders force all $k_w$ to be the same $k$, making $\Psi$ a global $k$-fold covering of $X_G$ by itself. In the notation given there, $$ X_G=\bigsqcup_{i=1}^N Z(w_i),\qquad \Psi|_{Z(w_i)}=T_{\beta_i}, $$ with each $\beta_i$ regressive and weakly progressive [1010.5739]. The covering index $k=|T_{\beta_i}^{-1}(x)|$ is constant, and on each piece the map is injective onto a full shift-invariant open set [1010.5739]. Reassembling the pieces yields a global homeomorphism that is a regular $k$-fold covering of $X_G$ by itself, and conversely every $k$-fold covering arising from a regressive block map is piecewise-canonical in this sense [1010.5739]. The stated conclusion is a complete classification: piecewise-canonical $\sigma$-commuting homeomorphisms of an edge shift $X_G$ are exactly the regressive/weakly progressive sliding-block codes that assemble to a global $k$-fold covering [1010.5739]. Because this blueprint is presented as a synthesized “final classification” in the supplied account, it should be read with some care. The graph-theoretic definition of piecewise-canonical homeomorphism in Tarocchi’s work is stated directly in terms of canonical prefix exchange, whereas the classification blueprint phrases the pieces as sliding-block codes on cylinders [2508.02582; 1010.5739]. This suggests an identification between canonical cylinder maps and finite-memory edge substitutions on suitable partitions. A plausible implication is that the blueprint is best understood as a symbolic-dynamical normal form for those piecewise-canonical maps that also commute with the shift. ## 6. Conjugacy, strand diagrams, and the loops semigroup Tarocchi’s principal theorem is algorithmic: for every finite graph $\Gamma$ and every initial multiset $Y$, the group $G(\Gamma\mid Y)$ has solvable conjugacy problem [2508.02582]. The proof adapts strand diagrams, introduced by Belk and Matucci for Thompson’s groups $F$, $T$, and $V$, to the setting of edge shifts and supplements them with a commutative semigroup of loops [2508.02582]. A $\Gamma$-strand diagram is a finite directed acyclic graph with vertices colored by $V_\Gamma$ and local combinatorics matching the branching structure of the path forest $F(\Gamma\mid Y)$ [2508.02582]. Sources are roots of split carets, sinks are leaves of merge carets, splits and merges respect color-ordering constraints, and degenerate vertices correspond to isolated unary loops in $\Gamma$ [2508.02582]. Every piecewise-canonical map between $S(\Gamma\mid Y)$ and $S(\Gamma\mid Y')$ is represented uniquely, up to local reductions of types $0,1,2$, by such a diagram with source colors given by $Y$ and sink colors by $Y'$ [2508.02582]. To test conjugacy of two elements represented by reduced strand diagrams $F$ and $G$, one closes the diagrams by identifying sources with sinks in order, obtaining closed strand diagrams $C(F)$ and $C(G)$ [2508.02582]. One then works with several classes of transformations: - similarities, namely base-line shifts through splits and merges, and base-line permutations; - reductions and inverse reductions of types $0,1,2$; - a new type $3$ reduction [2508.02582]. The type $3$ reduction is defined as follows. If a closed diagram contains a consecutive block of $k\times d$ base-points $$ (y_{i+1},\dots,y_{i+kd}) $$ which split into $d$ loops of winding number $k$ in colors $(c_1,\dots,c_d)$, where $c_1,\dots,c_d$ are exactly the terminal colors $t(e_j)$ of the $k$ edges $e_1<\cdots