---
title: Normally Hyperbolic Limit Tori
url: https://www.emergentmind.com/topics/normally-hyperbolic-limit-tori
type: topic
---

# Normally Hyperbolic Limit Tori

Searching arXiv for recent and foundational papers on normally hyperbolic limit tori.
Normally hyperbolic limit tori are compact invariant tori for which the contraction or expansion in directions transverse to the torus dominates the dynamics tangent to the torus. In non-autonomous \(T\)-periodic differential equations, they are naturally regarded as invariant manifolds in the extended phase space obtained by adjoining \(\dot t=1\), so that the time variable becomes an angular variable on \(\mathbb S^1\) [2208.10989]. In three-dimensional polynomial vector fields, limit tori are treated as the higher-dimensional analogue of limit cycles, and the normally hyperbolic ones are distinguished by their perturbative persistence and robustness [2507.17932]. Across recent work, the subject is organized around three closely related mechanisms: averaging criteria based on planar guiding systems, torus bifurcation via Neimark–Sacker or secondary Hopf theory, and degree-dependent constructions in polynomial dynamics.

## 1. Geometric definition and phase-space setting

The standard periodic-perturbation framework is the system
\[
\dot x=\sum_{i=1}^N \varepsilon^i F_i(t,x)+\varepsilon^{N+1}\widetilde F(t,x,\varepsilon), \qquad (t,x)\in \mathbb S^1\times D\times[0,\varepsilon_0],
\]
where \(D\subset\mathbb R^2\) is open and bounded, each \(F_i\) and \(\widetilde F\) is \(C^r\) with \(r\ge 2\), and all are \(T\)-periodic in \(t\) [2208.10989]. Because the system is periodic in time, the natural setting is the extended phase space obtained by adjoining \(\dot t=1\), so that \(t\) becomes an angular variable on \(\mathbb S^1\). In that setting, invariant tori of the original non-autonomous system are genuine invariant manifolds [2208.10989].

Normal hyperbolicity is expressed through a domination property between tangent and normal dynamics. In the diffeomorphism setting, a compact invariant submanifold \(N\) with splitting
\[
TM_{|N} = E^s \oplus E^u \oplus TN
\]
is normally hyperbolic if there exists \(0<\lambda<1\) such that
\[
\|T_x f_{|E_x^s}\| < \lambda, \qquad \|T_x f_{|E_x^u}^{-1}\| < \lambda,
\]
together with the domination inequalities
\[
\begin{cases}
\|T_x f_{|E_x^s}\| \cdot \|T_{f(x)} f_{|T_{f(x)}N}^{-1}\| < \lambda,\\[4pt]
\|T_x f_{|E_x^u}^{-1}\| \cdot \|T_{f^{-1}(x)} f_{|T_{f^{-1}(x)}N}\| < \lambda.
\end{cases}
\]
A strengthened \(r\)-normal hyperbolicity condition requires the analogous inequalities with tangential factors raised to powers \(1\le k\le r\) [1109.3280]. In the periodic-flow setting, the resulting torus is described as \(r\)-normally hyperbolic when normal contraction dominates tangent dynamics strongly enough to satisfy Fenichel’s criteria [2208.10989].

The term “limit torus” is used in several related but not identical senses. In the polynomial literature it denotes an isolated invariant torus in \(\mathbb R^3\), directly paralleling the notion of a limit cycle [2504.13832]. In periodic non-autonomous systems it is often the suspension of an invariant closed curve of a Poincaré or time-\(T\) map, hence an invariant torus in the extended phase space [2601.14247].

## 2. Averaging-theoretic detection via guiding systems

A central detection mechanism is an averaging criterion that upgrades the classical averaging theorem for isolated periodic orbits to the torus level. After a near-identity \(T\)-periodic change of variables
\[
x=U(t,y,\varepsilon)=y+\sum_{i=1}^N \varepsilon^i u_i(t,y),
\]
the periodic system is transformed into
\[
\dot y=\sum_{i=1}^N \varepsilon^i g_i(y)+\varepsilon^{N+1}r_N(t,y,\varepsilon),
\]
where the \(g_i\) are higher-order averaged functions [2208.10989]. They are defined recursively by
\[
g_0(y)=0,\qquad g_i(y)=\frac{y_i(T,y)}{i!},
\]
with
\[
y_1(t,y)=\int_0^t F_1(s,y)\,ds,
\]
and, for \(i\ge 2\),
\[
y_i(t,y)=\int_0^t\left(i!F_i(s,y)+\sum_{j=1}^{i-1}\sum_{m=1}^j \frac{i!}{j!}\, F_{i-j}(s,y)\, B_{j,m}(y_1,\dots,y_{j-m+1})(s)\right)ds,
\]
where \(B_{j,m}\) are the partial Bell polynomials [2208.10989].

If \(\ell\) is the first index such that
\[
g_0=\cdots=g_{\ell-1}=0,\qquad g_\ell\neq 0,
\]
the reduced dynamics is governed by the guiding system
\[
\dot y=\frac1T g_\ell(y).
\]
The decisive point is that hyperbolic limit cycles of this planar guiding system play the role that simple zeros of averaged functions play in the classical theorem for periodic solutions [2208.10989].

The main criterion states that if \(g_0=\cdots=g_{\ell-1}=0\), \(g_\ell\neq 0\), and the guiding system has an attracting hyperbolic limit cycle \(\gamma\), then for all sufficiently small \(\varepsilon>0\) the original \(T\)-periodic differential equation has a \(T\)-periodic solution inside the torus and a normally hyperbolic attracting invariant torus in the extended phase space, which surrounds that periodic solution and converges to \(\gamma\times\mathbb S^1\) as \(\varepsilon\to 0\) [2208.10989]. In transformed coordinates, the corresponding theorem states that the torus is the graph of a \(C^r\) function of the angular variables and is \(r\)-normally hyperbolic [2208.10989].

This mechanism is proved by a continuation argument combined with Fenichel theory: one continues an invariant graph from the unperturbed torus, shows the family is open and closed in the parameter, and then verifies normal hyperbolicity through estimates on tangential and normal variational dynamics [2208.10989]. Conceptually, the averaging principle is thereby upgraded from detecting one-dimensional periodic solutions to detecting two-dimensional limit tori in the extended flow.

## 3. Torus bifurcation through Neimark–Sacker and secondary Hopf theory

A second major route to normally hyperbolic limit tori proceeds through bifurcation of periodic orbits. For a one-parameter family
\[
\dot x = F(x,\mu), \qquad (x,\mu)\in \Omega\times(-\mu_0,\mu_0),
\]
suppose that for \(\mu=0\) there is a \(T_0\)-periodic orbit and associated Poincaré map \(p\mapsto \Pi(p,\mu)\). In this setting, a secondary Hopf bifurcation means that the periodic orbit undergoes a Neimark–Sacker bifurcation in the Poincaré map: a pair of complex multipliers crosses the unit circle, an invariant closed curve \(S_\mu\) is created in the section, and its saturation
\[
T_\mu := \bigcup_{t\in\mathbb R}\Phi_t(S_\mu)
\]
is an invariant torus for the flow [2507.10721].

Under Hypothesis (S), the derivative of the Poincaré map at the bifurcating fixed point has one complex conjugate pair on the unit circle at \(\mu=0\), while the remaining multipliers stay off the unit circle and the nonresonance conditions \((\lambda_1(0))^q\neq 1\) for \(q=1,2,3,4\) hold, together with the crossing condition
\[
d:=\frac{d}{d\mu}\operatorname{Re}(\lambda_1(\mu))\Big|_{\mu=0}\neq 0
\]
[2507.10721]. The principal result is that the torus produced by the Neimark–Sacker bifurcation is normally hyperbolic. More precisely, it has exactly \(n_s\) stable directions in the subcritical case and \(n_s+1\) stable directions in the supercritical case [2507.10721]. The proof combines Chaperon’s result on normal hyperbolicity of the invariant circle in the Poincaré map with Fenichel’s thin-surface-of-section theorem [2507.10721].

The same logic enters periodic perturbation problems through the stroboscopic map in extended phase space. For
\[
\dot x = \sum_{i=1}^k \varepsilon^i F_i(t,x,\mu) + \varepsilon^{k+1}\widetilde F(t,x,\mu),
\]
with each \(F_i\) and \(\widetilde F\) \(T\)-periodic in \(t\), the Poincaré map on \(\Sigma_0=\{0\}\times D\) has the expansion
\[
\Pi_\varepsilon(x,\mu) = x+\sum_{i=1}^k \varepsilon^i g_i(x,\mu)+\varepsilon^{k+1}R(x,\mu,\varepsilon),
\]
where \(g_i\) are averaged functions [2507.10721]. If the first nonvanishing averaged function \(g_l\) has a Hopf point, if the crossing condition \(\alpha'(0)\neq 0\) holds, and if the Lyapunov coefficient expansion
\[
\ell_1^\varepsilon = \varepsilon^l\ell_{1,l}+\varepsilon^{l+1}\ell_{1,l+1}+\cdots+\varepsilon^k\ell_{1,k}+\mathcal O(\varepsilon^{k+1})
\]
has a first nonzero coefficient \(\ell_{1,j^*}\), then there is either a periodic orbit with no invariant torus or a periodic orbit together with a unique invariant torus surrounding it, according to the sign of \(\alpha'(0)\ell_{1,j^*}(\mu-\mu(\varepsilon))\) [2507.10721]. The strengthened conclusion is that this torus is normally hyperbolic, repelling if \(\ell_{1,j^*}>0\), and attracting if \(\ell_{1,j^*}<0\) [2507.10721].

## 4. Polynomial vector fields and Hilbert-type counting problems

The three-dimensional analogue of Hilbert’s 16th problem asks for the maximal number of isolated invariant tori in polynomial vector fields of degree \(m\). One formulation introduces
\[
N(m)=\sup\{\tau(P,Q,R): \deg P,\deg Q,\deg R\le m\},
\]
and the normally hyperbolic refinement
\[
N_h(m)=\sup\{\tau_h(P,Q,R): \deg P,\deg Q,\deg R\le m\},
\]
where \(\tau_h(P,Q,R)\) counts normally hyperbolic limit tori [2212.12006]. The fundamental constructive mechanism starts with a planar system
\[
\dot x=P(x,y), \qquad \dot y=Q(x,y)
\]
having \(H\) hyperbolic limit cycles and embeds it into the three-dimensional polynomial system
\[
\begin{cases}
\dot x=-y,\\[2mm]
\dot y=x+y\,P(x^2+y^2,z),\\[2mm]
\dot z=2y^2Q(x^2+y^2,z).
\end{cases}
\]
For sufficiently small perturbation parameter \(\varepsilon>0\) in the averaging framework, each planar hyperbolic limit cycle \(\gamma\) generates a torus
\[
T_0^\gamma=\{(u(t)\cos\theta,\;u(t)\sin\theta,\;v(t)):\ (t,\theta)\in[0,T)\times[0,2\pi)\},
\]
and there is a nearby normally hyperbolic invariant torus \(T_\varepsilon^\gamma\) with the same stability type as \(\gamma\) [2212.12006]. If \(P,Q\) are polynomials of degree \(k\), the resulting three-dimensional system is polynomial of degree \(2k+2\) [2212.12006].

This yields the lower bound
\[
N_h(m)\ge \underline{H}_h\!\left(\left[\frac m2\right]-1\right) \qquad\text{for all }m\ge 2,
\]
where \(\underline{H}_h(k)\) is any known lower bound for the maximal number of hyperbolic limit cycles of planar polynomial vector fields of degree \(k\) [2212.12006]. Using a methodology due to Christopher & Lloyd, the same work obtains
\[
\liminf_{m\to\infty}\frac{N_h(m)}{m^3}\ge \frac1{128},
\]
so \(N_h(m)\), and therefore \(N(m)\), grows at least on the order of \(m^3/128\) [2212.12006].

Subsequent work sharpened the low-degree picture near monodromic singularities. For three-dimensional polynomial vector fields of degree \(m\), improved lower bounds were obtained through Hopf-zero and nilpotent-zero constructions:
  
| Degree \(m\) | Lower bound for \(N_h(m)\) |
|---|---:|
| 2 | 3 |
| 3 | 5 |
| 4 | 7 |
| 5 | 13 |

These are stated as the best available lower bounds in that work [2507.17932]. The same paper emphasizes that nilpotent-zero singularities often yield better lower bounds for limit tori than Hopf-zero ones because the averaged guiding system can have larger degree even when the original degree is fixed [2507.17932].

A separate monotonicity result shows that, if \(N_h(m)<\infty\), then
\[
N_h(m+1)\ge N_h(m)+1.
\]
The proof preserves \(N_h(m)\) existing normally hyperbolic limit tori by Fenichel persistence and creates one additional torus through a torus bifurcation near a Hopf-zero equilibrium [2504.13832]. This provides a strict increase property for \(N_h(m)\) analogous to the strict increase of the planar Hilbert number established by Santana and Gasull [2504.13832].

## 5. Representative constructions and explicit families

The averaging criterion has been implemented in several explicit families. In a family of jerk differential equations
\[
\dddot x=-\dot x+\varepsilon^{N-1}P(x,\dot x,\ddot x)+\varepsilon^N Q(x,\dot x,\ddot x)+\varepsilon^{N+1}R(x,\dot x,\ddot x,\varepsilon), \qquad N\ge 3,
\]
a change to variables \((r,z,\theta)\) produces a \(2\pi\)-periodic planar system in standard form. Under hypotheses \(H1\) and \(H2\), the guiding system is
\[
\dot r=r\big(6+z(1+2z)-r(11+(-6+r)r+z^2)\big),\qquad \dot z=r(2-r).
\]
After a change of variables, this guiding system has the attracting hyperbolic limit cycle \(\gamma=\mathbb S^1+(2,0)\), so the averaging theorem yields a normally hyperbolic attracting invariant torus in \((x,\dot x,\ddot x)\)-space [2208.10989]. In the explicit example, the limiting torus is
\[
\mathbb T=\{(\sqrt r\sin\theta-z,\sqrt r\cos\theta,-\sqrt r\sin\theta):(r,z)\in \mathbb S^1+(2,0),\ \theta\in[0,2\pi]\},
\]
and the choice \(P(x,\dot x,\ddot x)=-\dot x^3\) satisfies the hypotheses as well [2208.10989].

Near monodromic singular lines, the perturbative family
\[
X=X_0+\varepsilon X_1(\mu)+\varepsilon^2 X_2(\mu), \qquad X_0=-y\,\partial_x+x^{2n-1}\,\partial_y,
\]
is studied after the generalized polar change
\[
x=r\,\mathrm{Cs}\,\theta,\qquad y=r^n\,\mathrm{Sn}\,\theta,\qquad z=w,
\]
where \(\mathrm{Cs}\) and \(\mathrm{Sn}\) solve
\[
u'=-v,\qquad v'=u^{2n-1}.
\]
The first nonzero averaged function defines a planar guiding system
\[
(r',w')=\mathbf g_\ell(r,w;\mu)=\frac{1}{T}\mathbf f_\ell(r,w;\mu),
\]
and each hyperbolic limit cycle of that guiding system lifts to a normally hyperbolic limit torus in the original three-dimensional flow [2507.17932]. This construction is used to produce multiple nested normally hyperbolic limit tori near Hopf-zero and nilpotent-zero singularities [2507.17932].

An analytically different example appears in non-smooth dynamics. For non-autonomous \(T\)-periodic piecewise smooth systems, the time-\(T\) map is expanded as a smooth near-identity map
\[
\mathcal P_T(x;\alpha,\varepsilon) = x+\varepsilon^r\Delta_r(x;\alpha)+\cdots+\varepsilon^k\Delta_k(x;\alpha) +\mathcal O(\varepsilon^{k+1}),
\]
and a Neimark–Sacker bifurcation of this map produces a normally hyperbolic invariant closed curve, hence a limit torus in the suspension flow [2601.14247]. In the three-dimensional piecewise linear family treated there, the first nonzero Lyapunov coefficient term is
\[
\ell_{1,2}
=
-\frac{(\pi^2-6)b}{3\pi(8+9\pi^2)},
\]
so \(b>0\) gives an attracting torus and \(b<0\) gives a repelling torus [2601.14247]. The same work states that these tori persist under small perturbations [2601.14247].

## 6. Persistence theory, broader frameworks, and common distinctions

The persistence of normally hyperbolic limit tori is part of the general persistence theory of normally hyperbolic invariant manifolds. For a \(C^1\)-diffeomorphism \(f\) leaving invariant a closed \(C^1\)-submanifold \(N\), if \(f\) is normally hyperbolic at \(N\), then there exists a \(C^1\)-neighborhood \(\mathcal U\) of \(f\) such that every \(g\in\mathcal U\) leaves invariant and is normally hyperbolic at a \(C^1\)-submanifold \(N_g\), diffeomorphic and \(C^1\)-close to \(N\); moreover, \(N_g\) is unique and uniformly locally maximal [1109.3280]. If \(f\) is \(C^r\) and \(N\) is \(r\)-normally hyperbolic, the persisting manifold inherits \(C^r\) regularity [1109.3280]. This is the structural reason normally hyperbolic limit tori are treated as robust dynamical objects.

Beyond finite-dimensional smooth flows, invariant-manifold theory has been extended to partially normally hyperbolic invariant manifolds for generalized dynamics in Banach spaces, including correspondences that may be non-smooth, non-Lipschitz, or non-mapping. In that framework, whiskered tori are modeled by embeddings \(K:\mathbb T^d\to X\) with trichotomy
\[
T_{K(\theta)}X = X^s_\theta \oplus X^c_\theta \oplus X^u_\theta,
\]
and the torus is fully normally hyperbolic when \(\dim X^c_\theta=d\) [1905.09764]. The theory yields local center-stable, center-unstable, and center manifolds near the torus, together with higher regularity under appropriate spectral gap conditions [1905.09764].

A distinct but related context is semiclassical normally hyperbolic trapping. There the trapped set \(K^\delta\) is assumed to be a smooth symplectic submanifold of \(T^*X\) with invariant transverse bundles \(E^\pm_\rho\) satisfying
\[
\forall v\in E_\rho^\pm,\ \forall t>0,\qquad
\|d\varphi_{\mp t}(\rho)v\|_{\varphi_{\mp t}(\rho)}
\le C e^{-\lambda t}\|v\|_\rho .
\]
This framework is broad enough to include torus-like trapped sets if they are smooth, symplectic, and transversely hyperbolic, but it is not a torus-specific theorem [1302.4483]. Its emphasis is on resonance-free strips and exponential decay of correlations rather than torus detection [1302.4483].

A recurrent distinction in the literature concerns invariant tori that are robust for reasons other than normal hyperbolicity. In Beltrami fields, one can prescribe invariant tori, including knotted or linked ones, and obtain almost full-measure sets of invariant tori together with many hyperbolic periodic orbits inside each solid torus. However, that work explicitly does not claim normal hyperbolicity of the invariant tori themselves in the strong dynamical-systems sense; the relevant mechanism is KAM-type persistence for most invariant tori together with a Melnikov argument for hyperbolic periodic orbits [1909.07448]. The contrast is instructive: quasiperiodic or structural robustness of invariant tori does not by itself imply normal hyperbolicity.

Taken together, these developments show that normally hyperbolic limit tori occupy a precise intersection of averaging theory, bifurcation theory, persistence theory, and polynomial dynamics. The subject has moved from existence criteria for a single torus in extended phase space to explicit low-degree multiplicity results, strict monotonicity statements for \(N_h(m)\), and analytic detection in non-smooth systems, while retaining a common organizing principle: a hyperbolic object in a reduced or transverse dynamics is lifted to a robust invariant torus of the full system [2208.10989].

Source: https://www.emergentmind.com/topics/normally-hyperbolic-limit-tori