\documentclass[11pt]{amsart} \usepackage[T1]{fontenc} \usepackage[utf8]{inputenc} \usepackage{lmodern} \usepackage{amsmath,amssymb,amsthm,mathtools,mathrsfs} \usepackage{microtype} \usepackage[margin=1.1in]{geometry} \usepackage[shortlabels]{enumitem} \usepackage[colorlinks=true,linkcolor=blue,citecolor=blue,urlcolor=blue]{hyperref} \allowdisplaybreaks \newtheorem{theorem}{Theorem}[section] \newtheorem{proposition}[theorem]{Proposition} \newtheorem{lemma}[theorem]{Lemma} \newtheorem{corollary}[theorem]{Corollary} \theoremstyle{definition} \newtheorem{definition}[theorem]{Definition} \theoremstyle{remark} \newtheorem{remark}[theorem]{Remark} \newcommand{\I}{[0,1]} \newcommand{\N}{\mathbb N} \newcommand{\Nzero}{\mathbb N_0} \newcommand{\Fix}{\operatorname{Fix}} \newcommand{\Per}{\operatorname{Per}} \newcommand{\Ret}{\mathcal R} \newcommand{\id}{\mathrm{id}} \newcommand{\Lip}{\operatorname{Lip}} \newcommand{\dist}{\operatorname{dist}} \newcommand{\Crit}{\operatorname{Crit}} \newcommand{\CV}{\operatorname{CV}} \newcommand{\hent}{h_{\mathrm{top}}} \newcommand{\core}{\operatorname{core}} \newcommand{\RR}{\mathsf R} \DeclareMathOperator{\Int}{int} \title[Common periodic points of commuting interval maps]{Common periodic points of commuting interval maps} \author{Luke Hammer} \hypersetup{pdfauthor={\authors}} \thanks{OpenAI's ChatGPT models discovered and proved all three main theorems and formalized them in Lean~4. With minor human direction, GPT~5.6 found the proof of Theorem~\ref{thm:pm} in about two weeks and formalized it in about two days. Given that proof, GPT Astra~6.2 Ultra was asked to determine whether two commuting continuous maps of the unit interval must share a periodic point. With an average of six agents working concurrently, it proved Theorems~\ref{thm:returns} and~\ref{thm:omega} in approximately 24~hours. Theorem~\ref{thm:returns} gives a negative answer. The same model formalized both theorems in Lean~4 in a total of 30~minutes.} \date{} \begin{document} \begin{abstract} We construct pairs of commuting continuous self-maps of the compact interval with disjoint periodic-point sets. More precisely, every finite union of sets of positive multiples, including the empty union, is the set of integers $n$ for which two commuting $n$th iterates have a common fixed point. In the other direction, we show that if either generator has zero topological entropy, the generators act as mutually inverse cyclic permutations on a common finite orbit. We also establish a common fixed-point theorem for two commuting maps that are each monotone on a finite partition, with constant pieces allowed. \end{abstract} \maketitle \section{Introduction} Throughout, $I=\I$ and composition is written from right to left. For a continuous map $f:I\to I$ put $\Fix(f)=\{x:f(x)=x\}$ and $\Per(f)=\bigcup_{n\ge1}\Fix(f^n)$. For a commuting pair $f,g$ define its \emph{common return exponents} by \[ \Ret(f,g)=\{n\in\N:\Fix(f^n)\cap\Fix(g^n)\ne\varnothing\}. \] A point is periodic for both generators if and only if it belongs to $\Fix(f^n)\cap\Fix(g^n)$ for some common $n$; hence $\Per(f)\cap\Per(g)=\varnothing$ if and only if $\Ret(f,g)=\varnothing$. Necessarily $n\in\Ret(f,g)$ implies $kn\in\Ret(f,g)$ for every $k\ge1$. Boyce \cite{Boyce} and Huneke \cite{Huneke} constructed commuting interval maps without a common fixed point. This left open the possibility of a common periodic point, formulated explicitly by Alikhani--Koopaei \cite[Conjecture 1.1]{AK1998}, revisited in \cite{AK2003}, and included among the questions in \cite[Problem 9.4]{Canovas}. Our first theorem answers that question. \begin{theorem}[Prescribed common return exponents]\label{thm:returns} For every finite set $D\subset\N$, including $D=\varnothing$, there are commuting continuous maps $F,G:I\to I$ such that \[ \Ret(F,G)=\bigcup_{d\in D}d\N. \] In particular, there exist commuting continuous interval maps with $\Per(F)\cap\Per(G)=\varnothing$. \end{theorem} For a continuous map $f$, denote the $\omega$-limit set of $x$ by $\omega_f(x)$. The following is an independent positive result. \begin{theorem}[A common cycle in the fixed set of the composition] \label{thm:omega} Let $f,g:I\to I$ be commuting continuous maps, and set $K=\Fix(fg)$. Suppose that for each $x\in K$ with infinite $\omega_f(x)$ there exist disjoint closed intervals $L_x,R_x\subset I$ with \[ f(L_x)\subset R_x,\qquad f(R_x)\subset L_x,\qquad \omega_f(x)\subset L_x\cup R_x. \] Then $K$ contains a finite orbit on which $f$ and $g$ act as mutually inverse cyclic permutations. In particular, this conclusion holds if $\hent(f)=0$ or $\hent(g)=0$. \end{theorem} A complementary regularity theorem shows that counterexamples to a common fixed point cannot be finitely piecewise monotone in both generators. \begin{theorem}[The finite-piece fixed-point theorem] \label{thm:pm} Let $P,Q:I\to I$ be commuting continuous maps. If each is monotone (not necessarily strictly) on every member of a finite interval partition, then \[ \Fix(P)\cap\Fix(Q)\ne\varnothing. \] \end{theorem} The strong-commutation version for strictly piecewise-monotone maps was established by Anu\v{s}i\'{c} and Mouron \cite[Corollary 5.24]{AnusicMouron}. Ordinary commutation is weaker. Theorem~\ref{thm:pm} covers ordinary commutation and permits constant branches. The contrast with Theorem~\ref{thm:returns} is sharp at the level of finite versus unrestricted complexity of the generators. The proof of Theorem~\ref{thm:returns} occupies Sections~\ref{sec:nonempty} and~\ref{sec:empty}: the first uses finite cell permutations, while the second constructs a compatible system of maps on increasingly fine partitions. The dynamical criterion in Theorem~\ref{thm:omega} is proved in Section~\ref{sec:omega}, and Section~\ref{sec:finitepiece} proves the piecewise-monotone fixed-point theorem using constant-slope factors and substitution dynamics. The \href{https://github.com/volcanrb/Commuting-Monotone-maps/blob/a57e201b306354a7fe37785ba5d1e1f96a8b1922/COMMON_PERIODIC_POINTS.md}{index of Lean proofs} links Theorems~\ref{thm:returns}--\ref{thm:pm} to their formal statements and proofs. \section{Prescribing nonempty common return exponents}\label{sec:nonempty} \begin{proof}[Proof of Theorem~\ref{thm:returns} when $D\ne\varnothing$] Set $m=\sum_{d\in D}d$. Partition $I$ into $2m$ equal closed cells, colored alternately $1,2$ (with disjoint interiors). For each color, choose a permutation $P$ of its $m$ cells having exactly one cycle of length $d$ for each $d\in D$. The two permutations together define a color-preserving permutation of all cells. For each cell $J$, choose an affine embedding $s_J:I\to\Int(J)$. For each cell $J$ choose a continuous decoder $A_J:J\to I$ satisfying \begin{equation}\label{eq:section-finite} A_Js_J=\id_I. \end{equation} At an endpoint $e$ of $J$ shared with the neighboring, opposite-color cell $K$, prescribe \begin{equation}\label{eq:end-finite} A_J(e)=s_{P(K)}(0). \end{equation} The values at an exterior endpoint may be prescribed arbitrarily. Interpolation on the two components of $J\setminus s_J(I)$, joining the endpoint marks to $0$ or $1$ at the appropriate end of the core, produces a decoder with the stated properties. For $i\in\{1,2\}$ define $F_i:I\to I$ on its own color-$i$ cells by $F_i|_J=A_J$. On a cell $K$ of the other color require \begin{equation}\label{eq:recursive-finite} F_i|_K=s_{P(K)}\,F_i\,A_K. \end{equation} Here and below, restrictions and identities are equalities of maps with the indicated domains. To solve \eqref{eq:recursive-finite}, consider the closed subset of $C(I,I)$ consisting of maps with the prescribed color-$i$ branches. It is nonempty: at each mixed endpoint, the opposite-color branch is assigned the same required endpoint value, so the remaining branches may be interpolated continuously. The update in \eqref{eq:recursive-finite} preserves these endpoint values. Indeed, at a junction between $J$ of color $i$ and $K$ of the other color, $A_K(e)=s_{P(J)}(0)$ and $F_i(s_{P(J)}(0))=0$ by \eqref{eq:section-finite}. Thus the updated value on $K$ is $s_{P(K)}(0)=A_J(e)$. The uniform distance between two updates is at most \[ q\|U-V\|_\infty,\qquad q:=\max_J\Lip(s_J)<1. \] Banach's theorem supplies the unique continuous $F_i$. If $J$ has color $i$ and $j\ne i$, then, on $J$, \[ F_jF_i=F_jA_J, \qquad F_iF_j=F_i s_{P(J)}F_jA_J=F_jA_J. \] This proves $F_1F_2=F_2F_1$ on every cell. Moreover, for a cell $J$ of color $i$ and $j\ne i$, \begin{equation}\label{eq:opposite-inclusion} F_j(J)\subset s_{P(J)}(I)\subset\Int P(J). \end{equation} If $x$ is fixed by both $F_1^n$ and $F_2^n$, choose a cell $J$ containing $x$ and let $j$ be its opposite color. Iterating \eqref{eq:opposite-inclusion} gives $x=F_j^n(x)\in\Int P^n(J)$. The point $x$ therefore lies in the interior of its original cell and $P^n(J)=J$. Thus the cycle length of $J$, which belongs to $D$, divides $n$. Conversely, the affine contraction $s_J$ has a unique fixed point $z_J\in\Int(J)$. By \eqref{eq:section-finite} the map of the same color fixes $z_J$. The opposite-color relation is \[ F_j s_J=s_{P(J)}F_j, \] so $F_j(z_J)$ is the unique fixed point of $s_{P(J)}$ and is therefore $z_{P(J)}$. Along a $P$-cycle of length $d$ this gives a common point of return exponent $d$. All multiples of $d$ are return exponents. The two inclusions prove the assertion. \end{proof} \section{Commuting maps without common periodic points} \label{sec:empty} The empty case cannot be obtained by omitting the cycles from the finite construction. Instead we arrange that a hypothetical common periodic point of exponent $n$ lifts, with the \emph{same} exponent, to grades where the required cycle lengths grow without bound. \subsection{Graded cells, successors, and sections}\label{subsec:cells} Let $\Nzero=\{0,1,2,\dots\}$, let $v=(a,b)\in\Nzero^2$, and set \[ I_v=I,\qquad M_v=5^a7^b,\qquad e_1=(1,0),\quad e_2=(0,1). \] The index $v$ is a grade: the spaces $I_v$ are separate copies of $I$. Put $\sigma(t)=\dist(t,2\mathbb Z)$; this is the triangular folding map $\mathbb R\to I$, with period $2$. At grade $v$ define the closed cells \begin{equation}\label{eq:cell-charts} J_{v,\ell,k}=\chi_{v,\ell,k}(I),\qquad \chi_{v,\ell,k}(u)= \sigma\!\left(\frac{2k+(\ell+u)/4}{M_v}\right), \quad 0\le\ell\le3,\quad k\in\mathbb Z/M_v\mathbb Z. \end{equation} Each $\chi_{v,\ell,k}$ is an affine homeomorphism onto its cell, possibly reversing orientation. These $4M_v$ cells tile $I_v$ with disjoint interiors. For example, $\sigma(M_vx)$ maps every cell homeomorphically onto the $\ell$th quarter of $I$. Color $\ell=0,2$ by $1$, and $\ell=1,3$ by $2$. The distinguished quarter for color $1$ is $\ell=2$; for color $2$ it is $\ell=1$. For a color-$i$ cell $J$ set \begin{equation}\label{eq:graded-sections} s_J:I_{v+e_i}\longrightarrow J,\qquad s_J(t)=\chi_J\!\left(\frac{1+t}{3}\right). \end{equation} Its image is the \emph{core} $\core J$; the other two closed thirds are the \emph{shoulders}. The core has length $|J|/3$, and in particular \begin{equation}\label{eq:small-section} \Lip(s_J)=\frac{1}{12M_v}\le\frac1{12}. \end{equation} For $w\in\Nzero^2$, the projection $\pi_{v,w}$ of a cell of grade $v+w$ to grade $v$ keeps $\ell$ and reduces $k$ modulo $M_v$. If $d=M_{v+w}/M_v$ (always odd), then the standard fold \begin{equation}\label{eq:standard-fold} \varphi_{v,w}:I_{v+w}\to I_v,\qquad \varphi_{v,w}(x)=\sigma(dx) \end{equation} maps each fine cell, its core, and its shoulders onto the projected coarse cell and corresponding pieces, preserving the parameter $u$ in \eqref{eq:cell-charts}. This follows from $\sigma(d\,\sigma(t))=\sigma(dt)$ for odd positive $d$. For a grade $v=(a,b)$ express $k$ in Chinese-remainder coordinates \[ k_5=k\pmod{5^a},\qquad k_7=k\pmod{7^b}. \] Define a color-preserving \emph{successor} $P_v$ as follows: \begin{align}\label{eq:successors} (2,k_5,k_7)&\longmapsto(2,k_5+1,0)&&\text{(color 1)},\\ (1,k_5,k_7)&\longmapsto(1,0,k_7+1)&&\text{(color 2)},\nonumber \end{align} where active coordinates are reduced modulo $5^a$ or $7^b$, respectively. Every other cell goes to the distinguished cell of its own color with residue zero. The only cycles of $P_v$ have length $5^a$ in color $1$ and $7^b$ in color $2$; all other cells enter a cycle in one step. Directly from the coordinate rules, \begin{equation}\label{eq:Pprojection} \pi_{v,w}P_{v+w}=P_v\pi_{v,w}. \end{equation} We also need a lift of a successor to a neighboring grade. If $K$ is a cell of color $j\ne i$ at grade $v$, denote by $\widehat P_{v,i}(K)$ the distinguished color-$j$ cell at grade $v+e_i$ whose active coordinate is the active coordinate of $P_v(K)$ and whose other coordinate is zero. Since increasing the $i$th coordinate does not change the modulus active for color $j$, this definition is unambiguous. It satisfies \begin{equation}\label{eq:Phatprojection} \pi\widehat P(K)=\widehat P(\pi K) \end{equation} for compatible reductions (the grades are understood). The three combinatorial identities \eqref{eq:Pprojection}, \eqref{eq:Phatprojection}, and preservation of the color are the only properties of the successor labels used later. \subsection{Decoders and explicit shoulder models}\label{subsec:decoders} A cell $J$ of color $i$ at grade $v$ will have a continuous decoder \begin{equation}\label{eq:decoder} A_J:J\longrightarrow I_{v+e_i},\qquad A_Js_J=\id_{I_{v+e_i}}. \end{equation} We first prescribe the endpoint values. If an endpoint $e$ of $J$ is shared with an opposite-color cell $K$, set \begin{equation}\label{eq:decoder-mixed} A_J(e)=s_{\widehat P_{v,i}(K)}(0). \end{equation} At a junction of two same-color cells, and at a global endpoint of $I_v$, set $A_J(e)=s_{K_*}(0)$, where $K_*$ is the distinguished opposite-color cell of residue zero at grade $v+e_i$. The two cells meeting at a same-color junction have the same nondistinguished quarter and hence receive equal marks. Thus adjacent decoders have compatible endpoint marks whenever such compatibility will be needed. The interpolation on each shoulder is chosen in a specific way. Suppose its outer endpoint is marked by $s_H(0)$ for a cell $H$ in the decoder target. Write the mark in unfolded coordinates: \begin{equation}\label{eq:beta} \beta=\frac{2k_H+(\ell_H+1/3)/4}{M_{v+e_i}}, \qquad s_H(0)=\sigma(\beta). \end{equation} Choose the displayed real representative of $\beta$. Let $E\in\{0,1\}$ be the decoder value required at the adjacent end of the core (by \eqref{eq:decoder}, this is $0$ or $1$). Let $B$ be the least integer $B\ge\beta$ satisfying $\sigma(B)=1-E$. Parametrize the physical shoulder affinely by $[\beta,B+1]$, sending $\beta$ to the outer endpoint and $B+1$ to the core endpoint. In this parameter define \begin{equation}\label{eq:shoulder-decoder} A_J(t)=\sigma(t). \end{equation} The value at $\beta$ is the prescribed mark, the value at $B+1$ is $E$, and the final unit interval $[B,B+1]$ maps once across $I$. Together with the inverse section on the core, this defines $A_J$ continuously on the whole cell. Here is a useful intrinsic form of the shoulder construction. Define \begin{equation}\label{eq:q} q_{\beta,B}:[\beta,\infty)\longrightarrow[\beta,B+1] \end{equation} by fixing $[\beta,B+1]$ and, for $t\ge B$, selecting the unique point of $[B,B+1]$ with the same $\sigma$-value as $t$. This is well-defined, is $1$-Lipschitz, and satisfies \begin{equation}\label{eq:qproperties} \sigma q_{\beta,B}=\sigma,\qquad q_{\beta,B}^{-1}(u)=\{u\}\quad(\beta\le u0$. The fixed set of a piecewise-monotone map need not have finitely many components. For example, if $C_0$ is the middle-thirds Cantor set, then the increasing interval map $x\mapsto x+\tfrac12\dist(x,C_0)$ fixes exactly $C_0$. Consequently, the proof cannot rely on the number of components of individual fixed-point sets (compare \cite{Canovas}). Instead, we first obtain a simultaneous constant-slope factor, then construct an expanding path from a minimal counterexample. Finiteness of its critical orbits leads to two substitutions, whose rigidity yields a common fixed point. The argument draws on the constant-slope semiconjugacy theorem \cite{AlsedaMisiurewicz}, fixed-point and entropy results in \cite{BF}, and Durand's theorem on substitutions \cite{Durand}. \subsection{Minimal invariant intervals and a simultaneous constant-slope factor}\label{subsec:factor} Assume, toward a contradiction, that $P,Q$ satisfy the hypotheses of Theorem~\ref{thm:pm} but have no common fixed point. A common forward-invariant closed interval means an interval $J$ with $P(J)\subset J$ and $Q(J)\subset J$. The set of its ordered endpoint pairs is closed in the compact square $I^2$. Minimize its length. The minimizer is nondegenerate, since a degenerate one would be a common fixed point. Its images under $P$ and $Q$ are again common forward-invariant intervals by commutativity. Minimality of the length implies that both maps carry it \emph{onto} itself. After an affine change of variable, we may therefore assume that $P,Q:I\to I$ are surjective and there is no proper nonempty common forward-invariant interval. We call this last property \emph{minimality of the pair}; it is not minimality in the usual pointwise dynamical sense. Let $H=PQ$. By \cite[Theorem 1]{BF}, a pair without a common fixed point satisfies $\hent(H)\ge\log4>0$. Put \[ \lambda=e^{\hent(H)}>1. \] We recall how one obtains a semiconjugacy that works simultaneously for $P$ and $Q$, rather than merely for $H$. Let $\mathscr M$ denote the real vector space of finite signed nonatomic Borel measures on $I$. For a continuous piecewise-monotone map $T$ define its branchwise pullback $L_T\nu$ by prescribing, on an interval $[x,y]$ lying in one monotonicity branch, \begin{equation}\label{eq:measure-pullback} (L_T\nu)([x,y])=\nu(T([x,y])). \end{equation} This defines a measure first on each branch via its continuous distribution function; adjoining branches causes no ambiguity because the measures have no atoms. Refining a branch partition does not alter the result. Constant branches contribute zero. The chain rule on common refinements gives \begin{equation}\label{eq:transfer-chain} L_{ST}=L_TL_S. \end{equation} In particular, $L_P,L_Q,L_H$ commute, and $L_PL_Q=L_QL_P=L_H$. The constant-slope semiconjugacy theorem \cite{AlsedaMisiurewicz} supplies a nondecreasing surjection $\theta_0:I\to I$ such that on every monotonicity branch of $H$, \[ |\theta_0(Hy)-\theta_0(Hx)| =\lambda|\theta_0(y)-\theta_0(x)|. \] The nonatomic probability measure determined by its continuous distribution function is an eigenmeasure of $L_H$ with eigenvalue $\lambda$. Thus the space \[ E=\ker(L_H-\lambda\id)\subset\mathscr M \] contains a probability measure. \begin{lemma}[Finite-dimensional eigenmeasure space] \label{lem:finiteE} The space $E$ is finite-dimensional. Its set of positive probability measures is a nonempty compact convex set in the norm $\|F_\nu\|_\infty$, where $F_\nu(x)=\nu([0,x])$. \end{lemma} \begin{proof} Choose a monotonicity partition $0=t_0<\cdots1$, respectively. The factor pair still has no proper nonempty common forward-invariant interval and no common fixed point. \end{proposition} \begin{proof} By \eqref{eq:transfer-chain}, $L_P$ and $L_Q$ preserve $E$, and their product on $E$ is $\lambda\id$. On the compact convex set of probability measures in $E$, consider \[ \mathcal T(\nu)=\frac{L_P\nu}{(L_P\nu)(I)}. \] The denominator is at least $1$: the images of the monotonicity branches cover $I$ by surjectivity, and $L_P\nu(I)$ is the sum of their nonnegative masses. Thus $\mathcal T$ is continuous and maps the compact convex set into itself. Brouwer's fixed-point theorem (on its finite-dimensional affine hull) yields a probability measure $\mu$ and $C>0$ such that \[ L_P\mu=C\mu,\qquad L_Q\mu=D\mu, \qquad CD=\lambda. \] Let $\theta(x)=\mu([0,x])$. It is a continuous nondecreasing surjection. A fiber of $\theta$ is a closed interval of zero $\mu$-measure. Split such a fiber at the finitely many turns of $P$. By the eigenmeasure identity, the image under $P$ of each resulting subinterval has zero $\mu$-measure. Their union is an interval, so $P$ sends the entire fiber into a fiber of $\theta$. The same argument applies to $Q$. Both maps consequently descend to continuous, commuting maps $\overline P,\overline Q$ under $\theta$. The branchwise increment identity \eqref{eq:measure-pullback} gives \[ |\theta(Py)-\theta(Px)|=C|\theta(y)-\theta(x)| \] on each monotonicity branch of $P$, with the analogous formula for $Q$ and $D$. Thus the descended maps are piecewise-linear of constant slope magnitudes $C,D$. They are surjective, since $P,Q$ and $\theta$ are. If a closed interval is forward invariant for both factor maps, its full preimage under $\theta$ is a common forward-invariant closed interval upstairs. Minimality therefore passes to the factor. A common fixed point of the factor pair would be a proper common invariant singleton, which is impossible. Finally, a surjective interval map of constant slope magnitude $c$ has total variation at least $1$, hence $c\ge1$. If $c=1$, it is either the identity or the reflection $x\mapsto1-x$. Each commutes with an interval map only when there is a common fixed point: for the identity take any fixed point of the other generator, while for the reflection the point $1/2$ is fixed by every commuting map. Hence $C,D>1$. \end{proof} \subsection{Minimal factors and invariant return intervals} \label{subsec:minfactors} We shall repeatedly use the following elementary property of monotone semiconjugacies. \begin{lemma}[Fixed points pass through monotone factors] \label{lem:fixed-factor} Suppose $\pi:I\to I$ is continuous, nondecreasing and surjective, $\pi T=\overline T\pi$, and $T,\overline T$ are continuous interval maps. Then \begin{equation}\label{eq:fixfactor} \Fix(\overline T)=\pi(\Fix(T)). \end{equation} \end{lemma} \begin{proof} The inclusion from right to left is immediate. If $y\in\Fix(\overline T)$, the fiber $\pi^{-1}(y)$ is a nonempty closed interval, and $T$ carries it into itself. The intermediate value theorem gives a fixed point of $T$ in that fiber. \end{proof} Take a minimal constant-slope counterexample $(P_0,Q_0)$ from Proposition~\ref{prop:simultaneous}. Let \[ \nu_0=\#\Fix(P_0Q_0),\qquad L=(2\nu_0)!. \] The number $\nu_0$ is finite and positive: $P_0Q_0$ is piecewise linear of constant slope magnitude greater than $1$ and, as an interval self-map, has a fixed point. Consider all minimal constant-slope counterexample pairs obtained as nondecreasing factors of $(P_0,Q_0)$. Their numbers $\#\Fix(P^{2L})$ are finite positive integers, so one can choose a pair $(P,Q)$ minimizing this number. Write \[ A=P^L,\qquad B=Q^L. \] Any further nondecreasing factor, followed if necessary by Proposition~\ref{prop:simultaneous}, is among the pairs compared in this minimization. \begin{lemma}[No common return interval]\label{lem:noABinterval} There is no proper nondegenerate closed interval $J\subset I$ with $A(J)=J=B(J)$. \end{lemma} \begin{proof} Suppose such a $J$ exists, and consider the finite family of intervals \[ \mathscr J=\{P^iQ^j(J):0\le i,jx$ to the left of $z$, and $T(x)1$, so $A,B$ have slope magnitudes \[ a=C^L,\qquad b=D^L, \] respectively. The finite set $K=\Fix(PQ)$ has at least two points: if $K$ consisted of one point, commutativity would force both $P$ and $Q$ to fix it. By Lemma~\ref{lem:fixed-factor} and the choice of the original factor, $\#K\le\nu_0$. Choose consecutive points of $K$. On the open gap between them, $PQ(x)-x$ has constant sign. If this sign is positive, choose the left endpoint $p$ and set $\delta=+1$; otherwise choose the right endpoint and set $\delta=-1$. Because $PQ$ has constant slope magnitude $CD>1$, locally on the indicated side, \begin{equation}\label{eq:Hlocal} PQ(p+\delta t)=p+\delta CD\,t \qquad(0\le t\ll1). \end{equation} We briefly explain why large equal powers of the two maps fix this \emph{one-sided direction}, and not merely the base point. Each generator permutes the points of $K$, with the other as inverse, and is locally strictly monotone on either side of every point of $K$. It therefore acts on the finite set of one-sided germs based at $K$. On the subset of germs fixed by $PQ$, the induced actions of $P,Q$ are inverse permutations. The selected germ in \eqref{eq:Hlocal} belongs to this subset. There are at most $2\nu_0$ such germs, so $L=(2\nu_0)!$ fixes every germ in the subset under both generators. In particular, \begin{equation}\label{eq:ABlocal} A(p+\delta t)=p+\delta a t,\qquad B(p+\delta t)=p+\delta b t \qquad(0\le t\ll1). \end{equation} Set $R=AB$ and $s=ab$. For $t\ge0$ define \begin{equation}\label{eq:expandingray} \Psi(t)=R^n\!\left(p+\frac{\delta t}{s^n}\right) \end{equation} with any sufficiently large integer $n$. This is independent of $n$: increasing $n$ by one first applies $R$ within the neighborhood where $R(p+\delta u)=p+\delta su$, and then applies the previous iterate. The map $\Psi:[0,\infty)\to I$ is continuous, locally piecewise linear and locally of unit speed. Commutativity and \eqref{eq:ABlocal} give the exact scaling identities \begin{equation}\label{eq:scale} \Psi(at)=A\Psi(t),\qquad \Psi(bt)=B\Psi(t)\qquad(t\ge0). \end{equation} The image closure $J=\overline{\Psi([0,\infty))}$ is a nondegenerate compact interval, since $\Psi$ is locally of unit speed. Both $A$ and $B$ map $J$ onto itself by \eqref{eq:scale}, because multiplication of the ray parameter by $a$ or $b$ is onto $[0,\infty)$. Lemma~\ref{lem:noABinterval} therefore gives \begin{equation}\label{eq:Psi-onto} \overline{\Psi([0,\infty))}=I. \end{equation} \begin{lemma}[Independence of the two slopes] \label{lem:independent} The numbers $a,b$ are multiplicatively independent: there are no positive integers $m,n$ with $a^m=b^n$. \end{lemma} \begin{proof} If $a^m=b^n$, then \eqref{eq:scale} implies $A^m\Psi(t)=B^n\Psi(t)$ for every $t$. By \eqref{eq:Psi-onto} and continuity, $A^m=B^n$ on $I$. Consequently $P^{Lm}=Q^{Ln}$. We recall the standard inverse-limit obstruction to an equality of positive powers in a commuting pair without a common fixed point. Suppose $P^i=Q^j$ with $i,j>0$, and form \[ X=\varprojlim(I,PQ). \] Because $PQ$ is onto, this is an arc-like continuum: finite coordinate projections give interval approximations with fibers of arbitrarily small diameter in the product metric. The coordinatewise lifts $S,T$ of $P,Q$ commute on $X$ and satisfy $ST=TS=\text{the invertible inverse-limit shift}$; thus each of $S,T$ is a homeomorphism. We also have $S^i=T^j$. Write $(i,j)=d(u,v)$ with $u,v$ coprime and choose $r,s\in\mathbb Z$ satisfying $rv+su=1$. Define \[ V=S^uT^{-v},\qquad U=S^rT^s. \] Then $V^d=\id$, $U$ commutes with the finite cyclic group $\langle V\rangle$, and \[ S=U^vV^s,\qquad T=U^uV^{-r}. \] The finite-group fixed-point theorem for commuting maps on arc-like continua \cite[Theorem 3]{BF} supplies a point of $X$ fixed by $U$ and $V$. It is fixed by both $S$ and $T$. Its zeroth coordinate is a common fixed point of $P,Q$, contrary to assumption. Therefore no positive powers of $P$ and $Q$ can agree, and $a,b$ are independent. \end{proof} \subsection{Finite critical orbits and two substitutions} \label{subsec:substitution} For a surjective piecewise-linear map $T:I\to I$ with nonzero slope magnitude, let $\Crit(T)$ consist of its turning points and the two endpoints of $I$, and put $\CV(T)=T(\Crit(T))$. These sets are finite. \begin{lemma}[Critical values under composition] \label{lem:CV} For surjective piecewise-linear maps $S,T$ with nonzero slope magnitudes, \begin{equation}\label{eq:CVformula} \CV(ST)=\CV(S)\cup S(\CV(T)). \end{equation} \end{lemma} \begin{proof} Every turn of $ST$ is either a turn of $T$, whose image under the composition lies in $S(\CV(T))$, or a preimage under $T$ of a turn of $S$, whose image lies in $\CV(S)$. Endpoint values also lie in the displayed union. Conversely, a turn of $T$ survives the composition: near such a point the image of $T$ approaches its turning value from one side, on which $S$ is strictly monotone, so $ST$ still turns there. Every interior turning point of $S$ has a crossing preimage under $T$, because $T$ is surjective and strictly piecewise monotone; such a crossing produces a turn of $ST$. Finally, every preimage of an endpoint of $I$ under $T$ is a turn or a domain endpoint, which accounts for the endpoint critical values of $S$. This proves both inclusions. \end{proof} Let \[ X_m=\bigcup_{0\le i0$ be the geometric length of the underlying interval of a directed edge $e$. Each substitution changes geometric length by its map's constant slope magnitude. Consequently the strictly positive vector $(\ell_e)$ is a right eigenvector of each row-source incidence matrix, with eigenvalue $a$, respectively $b$. A nonnegative matrix with a strictly positive eigenvector has that eigenvalue as its spectral radius. Furthermore the geometric length of $\tau_A^n(e)$ is $a^n\ell_e$, and similarly for $B$; thus both substitutions grow on every occurring letter. Each nonnegative integral incidence matrix has an irreducible diagonal block attaining the spectral radius. Taking a positive power that removes the period of such a block makes the corresponding diagonal block primitive. Hence a positive power of $a$, respectively $b$, is the Perron root of a primitive integral matrix, and is a Perron number. \end{proof} Choose powers $r,s\ge1$ for which $a^r,b^s$ are Perron numbers. They are still multiplicatively independent by Lemma~\ref{lem:independent}. The word $\mathbf w$ is fixed by $\tau_A^r$ and $\tau_B^s$, whose incidence spectral radii are $a^r$ and $b^s$. Durand's theorem \cite{Durand} for sequences substitutive with respect to multiplicatively independent Perron numbers implies that $\mathbf w$ is ultimately periodic. Because its letters record \emph{directed} edges, whose geometric lengths are fixed, the path $\Psi$ is eventually periodic in its real arclength parameter. \subsection{Rigidity of the periodic tail} \label{subsec:rigidity} Let $\psi:\mathbb R\to I$ be the periodic extension of the tail of $\Psi$, and let $\tau>0$ be its least period. It is nonconstant and locally piecewise linear of unit speed. Its image $J$ is a nondegenerate interval; by \eqref{eq:scale}, $A(J)=J=B(J)$, since the image of every sufficiently late tail is exactly $J$. Lemma~\ref{lem:noABinterval} gives \begin{equation}\label{eq:psi-surj} \psi(\mathbb R)=I. \end{equation} For all sufficiently large $t$, the scaling identities read $\psi(at)=A\psi(t)$ and $\psi(bt)=B\psi(t)$. Since $\psi(t+\tau)=\psi(t)$, the numbers $a\tau$ and $b\tau$ are eventual periods of $\psi$, and hence global periods. They are integer multiples of its least period, so \begin{equation}\label{eq:abinteger} a,b\in\N,\qquad a,b>1. \end{equation} Now translate a given parameter by a sufficiently large integer multiple of $\tau$. Because $a,b$ are integers, this translation leaves both sides of each scaling identity unchanged. We obtain \begin{equation}\label{eq:global-scale} \psi(at)=A\psi(t),\qquad \psi(bt)=B\psi(t)\qquad(t\in\mathbb R). \end{equation} We use a rigidity principle for periodic unit-speed paths. A proof is included because it is the step that turns the eventual periodicity into global circle symmetries. \begin{lemma}[Rigidity of expanding periodic paths] \label{lem:rigidity} Let $u,v:\mathbb R\to I$ be nonconstant periodic, locally piecewise-linear unit-speed paths. Suppose that, for a continuous map $T:I\to I$ and $r>1$, \[ u(rt)=T(u(t)),\qquad v(rt)=T(v(t)) \quad\text{for every }t\in\mathbb R. \] If $v(t)=u(\epsilon t+c)$ on some nonempty open interval, with $\epsilon\in\{\pm1\}$, then the same identity holds globally with possibly a different translation constant. In particular, if $u,v$ are surjective, they agree globally up to a translation and possibly reversal of the parameter. \end{lemma} \begin{proof} The scaling identities propagate the local agreement: for each $n\ge0$, \begin{equation}\label{eq:rigidity-dilate} v(t)=u(\epsilon t+r^nc) \quad\text{on }r^nJ, \end{equation} where $J$ is the original open interval. Choose the midpoint $t_n$ of $r^nJ$. The intervals $r^nJ$ have length tending to infinity. Pass to a subsequence for which $t_n$ modulo a period of $v$, and $\epsilon t_n+r^nc$ modulo a period of $u$, converge to phases $x$ and $y$. For any fixed $h\in\mathbb R$, $t_n+h\in r^nJ$ for all sufficiently large $n$. Letting $n\to\infty$ in \eqref{eq:rigidity-dilate} therefore gives \[ v(x+h)=u(y+\epsilon h)\qquad(h\in\mathbb R). \] This is a global parameter isometry. If both paths are surjective, choose a common value outside the finite sets of turning values of the paths over one period. Each path crosses this value with slope $+1$ or $-1$ at some regular parameter, so local agreement under a parameter isometry exists. Apply the first assertion. \end{proof} \begin{lemma}[Integer slopes and affine circle lifts] \label{lem:circlelifts} The original slope magnitudes $C,D$ are positive integers. With the parameter circle normalized to $\mathbb R/\mathbb Z$, there are integers $k,\ell$ with $|k|=C$, $|\ell|=D$ and constants $\alpha,\beta\in\mathbb R/\mathbb Z$ such that \begin{equation}\label{eq:circlelifts} P\psi(t)=\psi(kt+\alpha),\qquad Q\psi(t)=\psi(\ell t+\beta). \end{equation} The lifts commute modulo $1$. \end{lemma} \begin{proof} Consider $u(t)=\psi(t)$ and $v(t)=P\psi(t/C)$. Both are surjective periodic locally unit-speed paths: away from finitely many turning parameters, the slope magnitude of $v$ is $C\cdot(1/C)=1$. By commutativity of $P$ and $A$, and by \eqref{eq:global-scale}, \[ v(at)=P\psi(at/C)=PA\psi(t/C)=Av(t). \] Lemma~\ref{lem:rigidity} gives $v(t)=\psi(\epsilon t+c)$ on all of $\mathbb R$. Because $C\tau$ is a period of $v$, and $v$ has the same period group $\tau\mathbb Z$ as $\psi$, it follows that $C\tau\in\tau\mathbb Z$; hence $C$ is an integer. Rescaling the parameter yields $P\psi(t)=\psi(\epsilon C t+c)$. The same argument with $Q$ and $B$ shows that $D$ is an integer and gives the second identity in \eqref{eq:circlelifts}. For the normalized circle, commutativity downstairs implies that \[ \psi(k\ell t+k\beta+\alpha) =\psi(k\ell t+\ell\alpha+\beta) \] for every $t$. Thus the two constant offsets differ by a period of $\psi$, proving that the affine lifts commute modulo $1$. \end{proof} We finish the proof of the finite-piece theorem. Because $\psi$ is surjective and locally unit-speed, it has a strict local maximum, at which its two one-sided slopes are $+1$ and $-1$. Thus $\psi$ has local reflection symmetry about that parameter. Applying Lemma~\ref{lem:rigidity} to the two paths $u=v=\psi$, with the common dilation $a$ and map $T=A$, promotes this local reflection to a global reflection $\rho$ of the parameter circle satisfying \begin{equation}\label{eq:reflection} \psi\rho=\psi. \end{equation} The reflection is unique modulo periods: two distinct reflection symmetries would compose to a nontrivial translation symmetry, contradicting minimality of the period. For either affine lift $L(t)=kt+\alpha$ of \eqref{eq:circlelifts}, there is a reflection $\rho'$ with $L\rho=\rho'L$. Since \[ \psi\rho' L=\psi L\rho=P\psi\rho=P\psi=\psi L \] and $L$ is onto the circle, $\psi\rho'=\psi$. Uniqueness forces $\rho'=\rho$. Thus both affine lifts commute with the reflection. Choose its fixed point as the origin of the unit circle, so $\rho(t)=-t$. A lift $L(t)=kt+\alpha$ commutes with $\rho$ if and only if $2\alpha=0\pmod1$, and therefore \begin{equation}\label{eq:half-offsets} \alpha,\beta\in\{0,\tfrac12\},\qquad (k-1)\beta=(\ell-1)\alpha\pmod1. \end{equation} Consider the three reflection orbits on the circle \[ \{0\},\qquad\{\tfrac14,\tfrac34\},\qquad\{\tfrac12\}. \] At least one is invariant under both lifts. Here is the complete parity check. If $\alpha=\beta=0$, both preserve $\{0\}$. If $\alpha=0,\beta=1/2$, \eqref{eq:half-offsets} requires $k$ odd. The first lift then preserves all three orbits; the second preserves the middle orbit when $\ell$ is odd and preserves $\{1/2\}$ when $\ell$ is even. The case $\alpha=1/2,\beta=0$ is symmetric. If both offsets are $1/2$, the congruence requires $k,\ell$ to have the same parity: both preserve the middle orbit if they are odd, and both preserve $\{1/2\}$ if they are even. The $\psi$-image of each such reflection orbit is a single point, by \eqref{eq:reflection}. If that orbit is preserved by both affine lifts, its image is fixed by $P$ and $Q$ by \eqref{eq:circlelifts}. This contradicts the assumption that the factor pair has no common fixed point, completing the proof of Theorem~\ref{thm:pm}. \hfill\qedsymbol \begin{thebibliography}{99} \bibitem{AK1998} A.~Alikhani--Koopaei, \emph{On common fixed and periodic points of commuting functions}, Internat. J. Math. Math. Sci. \textbf{21} (1998), 269--276. \href{https://doi.org/10.1155/S0161171298000386}{doi:10.1155/S0161171298000386}. \bibitem{AK2003} A.~Alikhani--Koopaei, \emph{On common fixed points, periodic points, and recurrent points of continuous functions}, Internat. J. Math. Math. Sci. \textbf{2003} (2003), no.~39, 2465--2473. \href{https://doi.org/10.1155/S0161171203205366}{doi:10.1155/S0161171203205366}. \bibitem{AlsedaMisiurewicz} Ll.~Alsed\`a and M.~Misiurewicz, \emph{Semiconjugacy to a map of a constant slope}, Discrete Contin. Dyn. Syst. Ser. B \textbf{20} (2015), 3403--3413. \href{https://doi.org/10.3934/dcdsb.2015.20.3403}{doi:10.3934/dcdsb.2015.20.3403}. \bibitem{AnusicMouron} A.~Anu\v{s}i\'{c} and C.~Mouron, \emph{Strongly commuting interval maps}, Fund. Math. \textbf{257} (2022), 39--68. \href{https://doi.org/10.4064/fm993-11-2021}{doi:10.4064/fm993-11-2021}. \bibitem{BF} S.~A.~Bogatyi and O.~D.~Frolkina, \emph{A common fixed point of commuting mappings of a tree}, Vestnik Moskov. Univ. Ser.~1 Mat. Mekh. (2002), no.~6, 3--10; English translation in Moscow Univ. Math. Bull. \textbf{57} (2002), no.~6, 1--8. \url{https://www.mathnet.ru/eng/vmumm1423}. \bibitem{Boyce} W.~M.~Boyce, \emph{Commuting functions with no common fixed point}, Trans. Amer. Math. Soc. \textbf{137} (1969), 77--92. \href{https://doi.org/10.1090/S0002-9947-1969-0236331-5}{doi:10.1090/S0002-9947-1969-0236331-5}. \bibitem{Canovas} J.~S.~C\'anovas, \emph{Combining the dynamics of discrete dynamical systems}, in \emph{Frontiers in the Study of Chaotic Dynamical Systems with Open Problems}, E.~Zeraoulia and J.~C.~Sprott, eds., World Scientific, 2011, Chapter~9. \bibitem{CanovasLinero} J.~S.~C\'anovas and A.~Linero, \emph{On topological entropy of commuting interval maps}, Nonlinear Anal. \textbf{51} (2002), no.~7, 1159--1165. \href{https://doi.org/10.1016/S0362-546X(01)00885-9}{doi:10.1016/S0362-546X(01)00885-9}. \bibitem{Dold} A.~Dold, \emph{Fixed point index and fixed point theorem for Euclidean neighborhood retracts}, Topology \textbf{4} (1965), 1--8. \bibitem{Durand} F.~Durand, \emph{Cobham's theorem for substitutions}, J. Eur. Math. Soc. \textbf{13} (2011), 1799--1814. \href{https://doi.org/10.4171/JEMS/294}{doi:10.4171/JEMS/294}. \bibitem{Huneke} J.~P.~Huneke, \emph{On common fixed points of commuting continuous functions on an interval}, Trans. Amer. Math. Soc. \textbf{139} (1969), 371--381. \href{https://doi.org/10.1090/S0002-9947-1969-0237724-2}{doi:10.1090/S0002-9947-1969-0237724-2}. \bibitem{Ruette} S.~Ruette, \emph{Chaos on the Interval}, University Lecture Series, vol.~67, American Mathematical Society, Providence, 2017. \href{https://doi.org/10.1090/ulect/067}{doi:10.1090/ulect/067}. \end{thebibliography} \end{document}