---
title: Cherry Maps and Toral Dynamics
url: https://www.emergentmind.com/topics/cherry-maps
type: topic
---

# Cherry Maps and Toral Dynamics

Cherry maps are circle endomorphisms associated with Cherry flows on the torus. In the flat-interval formulation, they are order-preserving degree-one maps of \(S^1\) that collapse an open arc to a point and have controlled power-law singularities at the endpoints of that flat interval; in related work on structural transitions of Cherry flows, the relevant return maps are monotone but discontinuous circle maps with a gap. Across these formulations, Cherry maps serve as one-dimensional models for toral recurrence, rotation theory, non-wandering sets, and bifurcation geometry near singular return mechanisms [1203.6301], [1903.06968].

## 1. Definitions and model classes

Several papers study a class \(\mathcal L\) or \(\mathscr L\) of circle maps with a flat interval \(U=(a,b)\subset S^1\) such that \(f(U)\) is a single point, \(f\) is a diffeomorphism on the complement of \(U\), and the boundary of \(U\) carries prescribed power-law behavior. In the symmetric case the critical exponent is written \((\ell,\ell)\); in the asymmetric case it is written \((\ell_1,\ell_2)\). The regularity hypotheses vary across the literature: \(C^2\) weakly order preserving maps with a flat interval, \(C^3\) order-preserving circle maps with a flat piece, and \(C^\infty\) non-decreasing degree-one maps with a flat interval all appear as principal objects of study [1203.6301], [2107.06105], [1509.05994], [2508.12350].

The lift formalism is standard. A degree-one circle map is represented by a lift \(F:\mathbb R\to\mathbb R\) satisfying
\[
F(x+1)=F(x)+1,\qquad f(x)=F(x)\bmod 1.
\]
This framework accommodates continuous and discontinuous maps, as well as injective and non-injective cases. In the flat-interval setting, non-injectivity is concentrated on the collapsed interval \(U\); in the discontinuous setting, non-surjectivity produces a genuine gap in the image [1903.06968].

In Palmisano’s \(C^\infty\) construction, “flat” is literal: for the lift \(F\), one has
\[
F^{(m)}(x)=0 \quad \text{for all } m\in\mathbb N,\ x\in U.
\]
That construction shows that a flat interval is compatible with irrational rotation number and Denjoy-type behavior, so the class is much broader than smooth monotone circle diffeomorphisms [1509.05994].

## 2. Return maps of Cherry flows

A Cherry flow is described in the cited literature as a flow on \(\mathbb T^2\) with no closed trajectories and exactly two singularities, one sink and one saddle, both hyperbolic. Every Cherry flow admits a Poincaré global section \(S^1\), and the first return map on this section is order-preserving, constant on an interval \(U\), \(C^2\) away from \(U\), and near the endpoints of \(U\) behaves like a power law whose exponent is determined by the saddle eigenvalues. If the saddle has eigenvalues
\[
\lambda_1>0>\lambda_2,
\]
then the singularity exponent is
\[
\ell = \frac{|\lambda_2|}{\lambda_1}.
\]
The quasi-minimal set of the flow is locally homeomorphic to \(I\times \Omega\), where \(I\) is an interval and \(\Omega\) is the non-wandering set of the return map [1203.6301].

The return-map viewpoint also appears in a transition problem from Poincaré flows to Cherry flows. In that setting, a continuous monotone return map for a Poincaré flow becomes discontinuous when a parameter crosses a saddle-node mechanism. The local model is
\[
\dot\xi=\mu+\xi^2,\qquad \dot\eta=\lambda\eta,\qquad \lambda>0,
\]
followed by the quadratic coordinate change
\[
x=\xi+a\eta^2,\qquad y=\eta,\qquad a>0.
\]
This produces a local normal form that can be embedded into a global flow on the torus by patching with simple vector fields in three regions, allowing the transition from continuous Poincaré behavior to discontinuous Cherry behavior to be tracked explicitly [1903.06968].

These two return-map descriptions are compatible at the level of dynamical purpose rather than literal map class. The flat-interval theory emphasizes continuous non-invertible circle maps, whereas the discontinuity-creation theory emphasizes injective but discontinuous maps with gaps. Both are organized around how singular flow geometry is encoded by a degree-one circle map [1203.6301], [1903.06968].

## 3. Rotation number, quasi-minimal sets, and Denjoy phenomena

For a lift \(F\) of a Cherry map, the rotation number is defined by
\[
p(F)=\lim_{n\to\infty}\frac{F^n(x)-x}{n},
\]
independently of \(x\), and \(p(f):=[p(F)]\). An equivalent formulation used elsewhere is
\[
p(f) := \lim_{n\to\infty}\frac{F^n(0)}{n} \pmod 1.
\]
The irrational case is central throughout the literature. One paper states that a Cherry map has a periodic point if and only if its rotation number is rational; another studies exclusively the irrational case in order to analyze geometric scaling and Hausdorff dimension [1203.6301], [2205.12915].

Cherry maps with a flat interval are not generally conjugate to rigid rotations even when the rotation number is irrational. Palmisano proved that for any irrational number \(p\in[0,1)\) there exists a \(C^\infty\), non-decreasing circle map \(f\) of degree one and an arc \(U\) such that \(f(U)\) is a point, \(f\) has rotation number \(p\), and \(f\) is a Denjoy counterexample. In the same paper, for continuous non-decreasing degree-one circle maps with irrational rotation number, the following are equivalent: being a Denjoy counterexample, not having dense orbits, having a wandering interval, and the existence of an interval \(I\) with \(|f^n(I)|\to 0\) as \(n\to+\infty\) [1509.05994].

The discontinuous Cherry-flow return maps studied in the gap setting retain a well-defined rotation number but are not surjective. Because some orbit segments never return through a certain interval, the map has a gap, and irrational rotation number can coexist with non-dense orbits. The cited work describes this as a natural route to Denjoy-like behavior for injective discontinuous circle maps [1903.06968].

At the flow level, Palmisano used the flat-interval construction to obtain Cherry flows whose quasi-minimal set is an attractor in the sense of Milnor: its basin has strictly positive Lebesgue measure, and no proper closed subset has the same basin up to a null set. The resulting Cherry flow has the prescribed irrational rotation number, and the basin of attraction of the quasi-minimal set has non-empty interior [1509.05994].

## 4. Geometric scaling and Hausdorff dimension

A major theme in the theory of Cherry maps is the dichotomy between degenerate geometry and bounded geometry. In the symmetric \(C^2\) flat-interval setting, the relevant scale ratios are
\[
T_n := \frac{|(0,q_n)|}{|(q_n,q_{n-2})|},
\qquad
\alpha_n := \frac{|(-q_n,0)|}{|(0,q_n)|},
\]
where \(q_n\) are the continued-fraction denominators of the irrational rotation number. Degenerate geometry means \(T_n\to 0\); bounded geometry means that \(T_n\) stays bounded away from \(0\). The central theorem gives a sharp transition at \(\ell=2\): if \(\ell\le 2\), then \(T_n\to 0\) at least exponentially fast, whereas if the rotation number is of bounded type and \(\ell>2\), then \(T_n\) is bounded away from zero [1203.6301].

This geometric transition controls the Hausdorff dimension of the non-wandering set. If \(f\) has critical exponent \(\ell\in(1,2]\), then
\[
\dim_H(\Omega)=0.
\]
If \(\ell>2\) and the irrational rotation number is of bounded type, then
\[
\dim_H(\Omega)>0.
\]
For Cherry flows with saddle eigenvalues \(\lambda_1>0>\lambda_2\), the same threshold becomes \(|\lambda_2|>2\lambda_1\), and under bounded type one obtains
\[
\dim_H(Q)>1
\]
for the quasi-minimal set \(Q\) [1203.6301].

The asymmetric theory generalizes this picture to critical exponents \((\ell_1,\ell_2)\). For order-preserving \(C^3\) circle maps with a flat piece, irrational rotation number, and critical exponents \((\ell_1,\ell_2)\), the non-wandering set is written
\[
K_f= \mathcal{S}^1 \setminus \bigcup_{i=0}^\infty f^{-i}(U).
\]
Under the negative Schwarzian assumption (A1), the geometry is degenerate for
\[
(\ell_1,\ell_2)\in[1,2]^2,
\]
and for bounded-type rotation numbers it is bounded for
\[
(\ell_1,\ell_2)\in[2,\infty)^2\setminus\{(2,2)\}.
\]
When the rotation number is bi-periodic,
\[
p(f)=[abab\cdots],
\]
the geometry is bounded above a curve \(C_{\lambda_u<1}\) defined on \(]1,+\infty[^2\). The corresponding Hausdorff-dimension statement is parallel: \(\dim_H(K_f)=0\) in the degenerate regime and \(\dim_H(K_f)>0\) in the bounded regime [2107.06105].

The cited proofs rely on dynamical partitions, cross-ratio inequalities, and recursive estimates for scale ratios. A notable conclusion in the small-exponent regime is that the relevant subsequences of scaling ratios go to zero at least exponentially fast, and in the strictly intermediate case \(1<\ell_i<2\) at least double exponentially fast. This identifies the critical exponent \(2\) as the threshold between degenerate Cantor geometry and bounded Cantor geometry [2107.06105].

## 5. Discontinuity creation and the amended Arnold-tongue picture

In the discontinuous return-map formulation, the central question is how a Cherry-flow return map acquires a gap. The cited analysis shows that the natural mechanism is not an arbitrary jump but a singular transition controlled by the saddle-node structure of the flow. As \(\mu\to 0\), the map first develops a very steep region, then a discontinuity of finite size. More precisely: for \(\mu>0\), the map is continuous but becomes extremely steep near the would-be gap; at \(\mu=0\), the gap opens with finite size; for \(\mu<0\), the return map is discontinuous, with the slope blowing up at both ends of the jump [1903.06968].

The asymptotics quantify this transition. Writing
\[
\mu=\sigma^2,\qquad \sigma>0,
\]
the local return time satisfies
\[
T\approx \frac{\pi}{\sigma}+O(1),
\]
so \(T\to\infty\) as \(\mu\downarrow 0\). In a very small neighborhood of the critical initial condition, the map stretches by a factor \(s\) with
\[
\log s \approx \frac{\lambda\pi}{\sigma}.
\]
Thus the slope becomes enormous as the bifurcation is approached from the Poincaré side, but only on an exponentially narrow set of initial conditions. For \(\mu<0\), the map has a finite jump, and near the discontinuity the leading non-constant term behaves like
\[
C_1|y|^\alpha,
\]
with \(\alpha<1\), so the derivative diverges at the edge of the gap. The paper describes this as a square-root singularity mechanism; in Cherry flows the singularity appears on both sides of the gap, unlike threshold systems where it is typically one-sided [1903.06968].

This local singularity changes the organization of phase locking. In smooth monotone circle maps without gaps, the classical Arnold tongue picture has tongue boundaries formed by saddle-node bifurcations. Once a gap is present, the boundary of a tongue may instead be formed partly by border collisions with the endpoints of the gap. The cited work distinguishes type I border collisions, where the derivative is infinite at the gap endpoint, from type II border collisions, where the derivative is finite. For monotone maps with a gap and a square-root singularity, periodic orbits in a phase-locked region can be organized by sequences such as
\[
\text{border collision} \to \text{border collision} \to \text{saddle-node}
\]
or
\[
\text{saddle-node} \to \text{border collision} \to \text{border collision} \to \text{saddle-node}.
\]
This is the precise sense in which the Arnold tongue picture is amended once gaps are present [1903.06968].

## 6. Conjugacy, bi-Lagrangian structures, and associated connections

A separate line of work places Cherry maps inside a symplectic and bi-Lagrangian framework. On \(\mathbb T^2\), a pair of transversal vector fields endowed with a symplectic form defines a bi-Lagrangian structure \((\omega,\mathcal F_1,\mathcal F_2)\). This viewpoint motivates constructions in which dynamics on \(\mathbb T^2\) are transported to tangent and cotangent bundles and related back to Cherry maps. In particular, a Cherry vector field \(X\) on \(T^2\simeq S^1\times[0,1]\), with flow \(\Phi_X^t\), determines a return-type map
\[
f_X(x)=\Phi_X^{t(x)}(x),
\]
where \(t(x)\) is the minimal positive time carrying \(x\in S^1\times\{0\}\) to \(S^1\times\{1\}\). The cited proposition states that \(f_X\) belongs to the class \(\mathcal L\) [2205.12915].

The same papers show that diffeomorphism actions on vector fields induce conjugacy actions on Cherry maps. The push-forward
\[
(\varphi,X)\mapsto \varphi_*X
\]
induces
\[
(\phi,f)\mapsto \phi\circ f\circ \phi^{-1}.
\]
More specifically, if \(X\) generates \(f\), then \(\varphi_*X\) generates \(\phi\circ f\circ \phi^{-1}\). For irrational rotation number, one cited paper states that the conjugacy orbit of a map in \(\mathcal L\) is determined by the rotation number [2205.12915].

Pairs of Cherry vector fields can also be combined to generate new Cherry maps with asymmetric critical exponents. If \(X_1\) and \(X_2\) generate \(f_1,f_2\in\mathcal L\) with flat pieces \(U_1=(a_1,b_1)\), \(U_2=(a_2,b_2)\), with \(f_1(U_1)=f_2(U_2)\) and
\[
a_1<a_2\le b_1<b_2,
\]
then the piecewise-defined map belongs to \(\mathcal L\), has flat piece \(U=(a_1,b_2)\), and has critical exponents \((\ell_1,\ell_2)\). This provides an explicit vector-field mechanism for producing left-right asymmetry in a Cherry map [2205.12915].

The more recent bi-Lagrangian extension develops this correspondence further. For a bi-Lagrangian manifold \((M,\omega,\mathcal F_1,\mathcal F_2)\), the tangent lift
\[
(TM,\omega^c,\mathcal F_1^c,\mathcal F_2^c)
\]
is again bi-Lagrangian, with Hess connection satisfying
\[
\nabla^c_{X^c}Y^c=(\nabla_XY)^c.
\]
On the cotangent bundle, the conormal foliations define
\[
(T^*M,d\theta,N^*\mathcal F_1,N^*\mathcal F_2).
\]
In the Cherry-map setting, for a pair \((X,Y)\) of transversal Cherry vector fields with the same singular set, the complement of the singularities carries a symplectic form and a bi-Lagrangian structure, hence a Hess connection \({}^{X,Y}\nabla\). The paper then defines the linear connection associated to the Cherry map as the common restriction
\[
{}^{X,Y}\nabla_{\mid \mathbb S^1},
\]
provided these restrictions agree for all generating pairs. This suggests a program in which a Cherry map is studied not only through its one-dimensional dynamics but also through a connection induced by the geometry of the generating flows [2508.12350].

Source: https://www.emergentmind.com/topics/cherry-maps