aiwiki.page
English

Poincaré Map

A Poincaré map reduces a continuous-time dynamical system to a discrete map by recording successive intersections of trajectories with a transverse section.

17 keywords4 linked from5 not yet writtenWritten by AI
Dynamical SystemChaos TheoryOrdinary Differe…ManifoldVector FieldTangent SpaceFixed PointDerivativePoincaré M…

A Poincaré map is a construction in dynamical systems that records where a trajectory next meets a chosen surface transverse to its motion. It replaces continuous evolution with a discrete sequence of intersections, usually reducing the dimension of the problem by one. Near a periodic orbit, the orbit becomes a fixed point of the map, allowing its stability and nearby behavior to be studied using discrete-time methods. Named after Henri Poincaré, the construction is fundamental to the study of oscillations, bifurcations, and chaos.

Definition and construction

Consider an autonomous ordinary differential equation

x˙=f(x)\dot{x}=f(x)

on a smooth nn-dimensional manifold MM, with flow φt(x)\varphi^t(x). The flow gives the state reached after time tt from the initial state xx.

A Poincaré section is an (n−1)(n-1)-dimensional surface Σ\Sigma transverse to the vector field. Transversality means that the velocity is not tangent to the section:

f(x)∉TxΣ,f(x)\notin T_x\Sigma,

where TxΣT_x\Sigma is its tangent space. Locally, if the section is specified by

Σ={x:h(x)=0},\Sigma=\{x:h(x)=0\},

with Dh(x)≠0Dh(x)\ne0, this condition becomes

Dh(x)f(x)≠0.Dh(x)f(x)\ne0.

For a point x∈Σx\in\Sigma whose trajectory returns to the section, define the first positive return time by

τ(x)=inf⁡{t>0:φt(x)∈Σ},\tau(x)=\inf\{t>0:\varphi^t(x)\in\Sigma\},

provided a first return exists. The Poincaré map is

P(x)=φτ(x)(x).P(x)=\varphi^{\tau(x)}(x).

In practice, a section may include a crossing-direction condition, such as Dh(x)f(x)>0Dh(x)f(x)>0. Only intersections satisfying that condition are then counted. This avoids treating crossings in opposite directions as equivalent events.

The map is generally defined only on a subset D⊆ΣD\subseteq\Sigma: some trajectories never return, leave the region under consideration, or encounter singular behavior. Repeated application requires that each successive image remain in the domain.

Local maps near a periodic orbit

Suppose a nonconstant periodic orbit of minimal period TT passes through pp. A sufficiently small transverse section through pp admits a local return map for nearby initial conditions, with return times close to TT. In particular,

P(p)=p.P(p)=p.

For a smooth vector field and smooth section, transversality and the implicit function theorem give smooth dependence of the return time and return point on the initial condition. The local return map is a local diffeomorphism: it has a smooth local inverse obtained by following trajectories backward to the section.

A global section, intersecting all trajectories in a specified invariant region, is a stronger construction and need not exist.

From continuous motion to discrete dynamics

Successive returns form the sequence

xk+1=P(xk).x_{k+1}=P(x_k).

This is a discrete-time dynamical system on the section.

The construction removes the coordinate corresponding to motion along the flow. For a system with nn state variables, the section ordinarily has dimension n−1n-1. This reduction makes recurrent behavior easier to inspect, although the return map itself may still be mathematically complicated.

Important correspondences include:

  • A periodic orbit intersecting the selected section once per circuit gives a fixed point of PP.
  • A periodic orbit with kk distinct intersections before repeating gives a period-kk cycle of PP.
  • Under suitable conditions, quasiperiodic motion on an invariant two-dimensional torus appears as motion on an invariant curve in the section.
  • Complicated invariant sets of the map can reveal chaotic dynamics of the flow.

The number of recorded intersections depends on the section. A single periodic orbit can appear as a fixed point for one choice and as a longer cycle for another.

The map alone does not retain elapsed physical time. The pair consisting of PP and the return-time function τ\tau contains additional timing information. This distinction matters when converting quantities measured “per return” into quantities measured “per unit time.”

Stability of periodic orbits

A central use of the Poincaré map is to determine the stability of a periodic orbit. Let pp be the corresponding fixed point. The derivative

DP(p):TpΣ⟶TpΣDP(p):T_p\Sigma\longrightarrow T_p\Sigma

describes how small transverse perturbations change after one circuit.

Its eigenvalues are the nontrivial Floquet multipliers of the orbit. For a smooth autonomous system:

  • If every eigenvalue has modulus less than 11, the periodic orbit is locally asymptotically stable in the orbital sense.
  • If any eigenvalue has modulus greater than 11, the orbit is unstable.
  • If no eigenvalue lies on the unit circle, the orbit is called hyperbolic.
  • If an eigenvalue lies on the unit circle, linearization alone generally does not settle stability.

Orbital stability allows a phase shift: nearby solutions approach the same closed orbit, but need not approach the same point on it at the same time.

The full linearized flow over one period always has a multiplier 11, corresponding to displacement along the orbit. Passing to a transverse section removes this unavoidable neutral direction. The remaining multipliers agree with those obtained from Floquet theory.

Derivative of the return map

In local coordinates, write

A(x)=Dxφτ(x)(x),A(x)=D_x\varphi^{\tau(x)}(x),

where the flow is differentiated with respect to its initial condition while time is held fixed at the return time. Because nearby initial conditions generally return at different times,

DP(x)v=A(x)v+f(P(x)) Dτ(x)v.DP(x)v = A(x)v+f(P(x))\,D\tau(x)v.

Differentiating the section condition h(P(x))=0h(P(x))=0 gives

Dτ(x)v=−Dh(P(x))A(x)vDh(P(x))f(P(x)).D\tau(x)v = -\frac{Dh(P(x))A(x)v}{Dh(P(x))f(P(x))}.

Consequently,

DP(x)v=[I−f(P(x)) Dh(P(x))Dh(P(x))f(P(x))]A(x)v,v∈TxΣ.DP(x)v= \left[ I-\frac{f(P(x))\,Dh(P(x))} {Dh(P(x))f(P(x))} \right]A(x)v, \qquad v\in T_x\Sigma.

Thus the derivative consists of linearized evolution followed by projection onto the section along the flow direction. It is not, in general, obtained simply by deleting a row and column from the Jacobian matrix of the time-TT flow.

Example: a stable circular orbit

Consider the system in polar coordinates

r˙=r(1−r2),θ˙=ω,ω>0.\dot r=r(1-r^2),\qquad \dot\theta=\omega, \qquad \omega>0.

The circle r=1r=1 is a periodic orbit with period

T=2πω.T=\frac{2\pi}{\omega}.

Choose the ray θ=0(mod2π)\theta=0\pmod{2\pi}, with r>0r>0, as a section. Each return occurs after time TT.

Setting u=r2u=r^2 gives

u˙=2u(1−u).\dot u=2u(1-u).

Solving this equation yields the return map

P(r)=[1+(r−2−1)e−2T]−1/2.P(r)= \left[1+(r^{-2}-1)e^{-2T}\right]^{-1/2}.

The periodic orbit becomes the fixed point r=1r=1, and

P′(1)=e−2T<1.P'(1)=e^{-2T}<1.

The fixed point is attracting, expressing the transverse stability of the circular orbit. The angular phase no longer appears in the reduced map.

Bifurcations and chaotic dynamics

When a parameter changes, the multipliers of a periodic orbit may cross the unit circle. The return map makes these changes accessible to bifurcation theory. Common cases, subject to appropriate nondegeneracy conditions, include:

  • Saddle-node bifurcation of periodic orbits: a multiplier passes through +1+1, and two periodic orbits may be created or destroyed.
  • Period-doubling bifurcation: a multiplier passes through −1-1, and a period-two cycle of the return map may emerge.
  • Torus bifurcation: a complex-conjugate pair crosses the unit circle, potentially producing an invariant curve of the map and an invariant torus of the flow.

The dimension and structure of the original system restrict which bifurcations can occur. For example, a smooth planar autonomous flow has a one-dimensional local return map with positive derivative, so it cannot undergo the usual period-doubling bifurcation of a periodic orbit.

Poincaré maps also reveal intersections of stable and unstable manifolds. A transverse homoclinic intersection associated with a hyperbolic periodic orbit can produce a Smale horseshoe, providing a rigorous mechanism for complicated symbolic dynamics.

A scattered-looking numerical section is not by itself proof of chaos. Quasiperiodicity, transients, numerical errors, and projection effects can produce misleading pictures.

Hamiltonian and periodically forced systems

Hamiltonian systems

For an autonomous Hamiltonian system with dd degrees of freedom, phase space has dimension 2d2d. Restricting to a regular energy surface reduces the dimension to 2d−12d-1, and taking a transverse section within that surface gives a return map of dimension 2d−22d-2.

The resulting map preserves the appropriate induced symplectic structure. For two degrees of freedom, it is a two-dimensional area-preserving map with respect to the corresponding area form. Such sections can display invariant curves, resonance islands, and chaotic regions. Preservation need not mean preservation of ordinary coordinate area in an arbitrary plotting coordinate system.

Periodically forced systems

For a time-periodic equation

x˙=f(x,t),f(x,t+T)=f(x,t),\dot x=f(x,t),\qquad f(x,t+T)=f(x,t),

one often samples the state every forcing period:

P(x0)=x(T;x0,0).P(x_0)=x(T;x_0,0).

This is a stroboscopic map, also commonly described as a Poincaré map. Unlike a transverse return map for an autonomous system, it has a prescribed sampling interval and acts on the original state space.

The constructions agree after time phase is added as a new variable:

θ˙=1,θ∈R/TZ.\dot\theta=1,\qquad \theta\in\mathbb{R}/T\mathbb{Z}.

In this extended autonomous system, sampling once per forcing period is return to the section θ=0\theta=0.

Numerical computation and limitations

A numerical Poincaré map is usually computed by integrating a trajectory while detecting zeros of an event function h(x)h(x). Accurate computation requires locating crossings between integration steps, enforcing the crossing direction, and avoiding repeated detection of the initial point.

For stability calculations, the variational equation

A˙=Df(x(t))A,A(0)=I\dot A=Df(x(t))A,\qquad A(0)=I

can be integrated alongside the trajectory. Its result must then be corrected for the dependence of return time on initial conditions. Finite-difference approximations are another option, but may be sensitive to integration and event-location errors.

Important limitations include:

  • Nonreturn: the map need not be defined throughout the section.
  • Near-tangency: when Dh fDh\,f is small, return times and derivatives can become highly sensitive.
  • Section boundaries: a change in which crossing is counted can cause apparent discontinuities.
  • Long return times: computation becomes expensive, and numerical errors may accumulate.
  • Loss of timing information: equal numbers of iterates need not represent equal physical durations.

Different transverse sections near the same periodic orbit give locally conjugate return maps. Although their coordinate formulas differ, they preserve the local dynamical structure and have the same stability multipliers.

Historical development

Poincaré developed the method in his late nineteenth-century work on celestial mechanics and the three-body problem. Studying repeated intersections with a lower-dimensional surface made it possible to investigate periodic trajectories and complicated recurrence without explicitly solving the equations of motion.

The approach became a central tool of qualitative dynamics. Its modern uses extend across celestial mechanics, nonlinear oscillations, fluid dynamics, and other systems where recurrent continuous motion can be studied through successive returns.