{
  "id": "shadowing-lemma",
  "title": "Shadowing lemma",
  "lang": "en",
  "source_lang": "en",
  "summary": "The shadowing lemma states that approximate trajectories of a uniformly hyperbolic dynamical system can be followed closely by exact trajectories, with a possibly adjusted initial condition.",
  "category": "mathematics",
  "subcategory": "analysis",
  "type": "concept",
  "aliases": [
    "Pseudo-orbit tracing lemma",
    "Anosov–Bowen shadowing lemma"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/shadowing-lemma/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/shadowing-lemma/",
      "json": "https://aiwiki.page/zh/shadowing-lemma.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "The shadowing lemma is a theorem in dynamical systems stating that, under suitable hyperbolicity assumptions, every sufficiently accurate approximate trajectory remains uniformly close to a genuine trajectory. The approximate trajectory may make a small error at every iteration, yet one exact orbit follows the entire sequence. Its initial point generally differs from the starting point of the approximate trajectory. The lemma is fundamental to hyperbolic dynamics and helps explain how reliable orbit information can be extracted from computations in systems exhibiting chaos."
    },
    {
      "heading": "Pseudo-orbits and the shadowing property",
      "text": "Let \\((X,d)\\) be a metric space and let \\(f:X\\to X\\) be a continuous map. A sequence \\((x_n)\\) is a \\(\\delta\\)-pseudo-orbit if\n\n\\[d\\bigl(f(x_n),x_{n+1}\\bigr)<\\delta\\]\n\nat every relevant index. Thus, each step satisfies the evolution rule up to an error smaller than \\(\\delta\\).\n\nFor forward-time dynamics, the indices are \\(n\\geq 0\\). When \\(f\\) is a homeomorphism, one can also consider two-sided sequences indexed by all integers.\n\nAn exact orbit through \\(y\\) \\(\\varepsilon\\)-shadows the pseudo-orbit if\n\n\\[d\\bigl(f^n(y),x_n\\bigr)<\\varepsilon\\]\n\nfor every index under consideration. Here \\(f^n\\) denotes the \\(n\\)-fold iterate, with negative iterates defined using the inverse map when necessary.\n\nThe map has the shadowing property, also called the pseudo-orbit tracing property, if\n\n\\[\\forall \\varepsilon>0\\;\\exists\\delta>0 \\quad \\text{such that every \\(\\delta\\)-pseudo-orbit is \\(\\varepsilon\\)-shadowed by an exact orbit.}\\]\n\nThe order of these quantifiers matters. The permitted step error \\(\\delta\\) must work for every pseudo-orbit, independently of its duration. The definition itself does not require the shadowing orbit to be unique or require \\(\\delta\\) to depend linearly on \\(\\varepsilon\\).\n\nShadowing is therefore stronger than a fixed-time error estimate: a bound valid for a specified number of steps need not remain useful over arbitrarily long trajectories."
    },
    {
      "heading": "The hyperbolic shadowing lemma",
      "text": "A standard smooth version concerns a \\(C^1\\) diffeomorphism \\(f:M\\to M\\) of a smooth manifold. A compact invariant set \\(\\Lambda\\) is uniformly hyperbolic if its tangent spaces admit a continuous, derivative-invariant splitting\n\n\\[T_xM=E_x^s\\oplus E_x^u, \\qquad x\\in\\Lambda,\\]\n\nand there are constants \\(C\\geq1\\) and \\(0<\\lambda<1\\) such that, for every \\(n\\geq0\\),\n\n\\[\\|Df_x^n v^s\\|\\leq C\\lambda^n\\|v^s\\|, \\qquad v^s\\in E_x^s,\\]\n\nand\n\n\\[\\|Df_x^{-n}v^u\\|\\leq C\\lambda^n\\|v^u\\|, \\qquad v^u\\in E_x^u.\\]\n\nThe spaces \\(E_x^s\\) and \\(E_x^u\\) are the stable and unstable directions in the tangent space at \\(x\\).\n\nOne common formulation of the lemma is:\n\nLet \\(\\Lambda\\) be a compact, locally maximal hyperbolic set for a \\(C^1\\) diffeomorphism. For every \\(\\varepsilon>0\\), there is a \\(\\delta>0\\) such that every two-sided \\(\\delta\\)-pseudo-orbit contained in \\(\\Lambda\\) is \\(\\varepsilon\\)-shadowed by an orbit contained in \\(\\Lambda\\).\n\nLocal maximality means that some neighborhood \\(U\\) satisfies\n\n\\[\\Lambda=\\bigcap_{n\\in\\mathbb Z}f^n(U).\\]\n\nThis hypothesis ensures that an exact orbit remaining sufficiently close to \\(\\Lambda\\) belongs to \\(\\Lambda\\). For a hyperbolic set that is not locally maximal, a corresponding ambient shadowing statement still holds, but the shadowing orbit need not lie in the set itself.\n\nAn Anosov diffeomorphism is a diffeomorphism for which the whole compact manifold is hyperbolic. Such maps satisfy the shadowing property globally.\n\nHyperbolic shadowing also admits a quantitative strengthening: for sufficiently small errors, constants \\(L>0\\) and \\(\\delta_0>0\\) can be chosen so that every \\(\\delta\\)-pseudo-orbit, with \\(\\delta<\\delta_0\\), is shadowed within distance \\(L\\delta\\). This is a form of Lipschitz shadowing."
    },
    {
      "heading": "Why hyperbolicity makes shadowing possible",
      "text": "Expansion does not prevent shadowing, because the exact orbit is allowed to start at a different point.\n\nIn a stable direction, errors are damped by forward iteration. In an unstable direction, errors are damped by backward iteration. An appropriate initial displacement balances these two effects, producing a uniformly bounded correction along the entire pseudo-orbit.\n\nFor example, consider the scalar map\n\n\\[f(x)=ax,\\qquad |a|>1,\\]\n\nand a two-sided pseudo-orbit satisfying\n\n\\[x_{n+1}=ax_n+e_n, \\qquad |e_n|\\leq\\delta.\\]\n\nTo obtain an exact orbit \\(y_n=x_n+v_n\\), the corrections must satisfy\n\n\\[v_{n+1}=av_n-e_n.\\]\n\nA bounded solution is\n\n\\[v_n=\\sum_{j=0}^{\\infty}a^{-j-1}e_{n+j},\\]\n\nso that\n\n\\[|v_n|\\leq\\frac{\\delta}{|a|-1}.\\]\n\nThe correction in an expanding direction depends on future errors. This is an existence construction, not a causal prediction procedure.\n\nFor nonlinear systems, a typical proof proceeds by:\n\n- Expressing nearby points in local coordinates along the pseudo-orbit.\n- Solving a linear difference equation using the stable–unstable splitting.\n- Controlling the nonlinear remainder through a contraction argument, often using the Banach fixed-point theorem.\n\nUniform hyperbolicity provides bounds that do not deteriorate with the length of the pseudo-orbit."
    },
    {
      "heading": "Uniqueness and expansivity",
      "text": "Existence of a shadowing orbit does not imply uniqueness for an arbitrary system. In hyperbolic dynamics, uniqueness at sufficiently small scales follows from expansivity.\n\nA homeomorphism is expansive on an invariant set if there is a constant \\(c>0\\) such that\n\n\\[d\\bigl(f^n(y),f^n(z)\\bigr)<c \\quad\\text{for every }n\\in\\mathbb Z\\]\n\nimplies \\(y=z\\).\n\nIf two orbits both \\(\\varepsilon\\)-shadow the same pseudo-orbit and \\(2\\varepsilon<c\\), then the triangle inequality makes their mutual distance smaller than \\(c\\) at every iterate. Expansivity forces their initial points to coincide.\n\nTwo-sided indexing is important here. Distinct points on a stable manifold can remain close for all future time, so forward shadowing alone need not identify a unique initial point."
    },
    {
      "heading": "Applications",
      "text": "Numerical trajectories\n\nA computed trajectory can be regarded as a pseudo-orbit when its one-step errors are bounded. Those errors may include roundoff and discretization errors from numerical methods.\n\nThe lemma then guarantees a nearby exact orbit, provided the relevant hypotheses and error bounds hold. It does not generally guarantee that the computation follows the exact orbit through the originally prescribed initial point. In a chaotic system, these two conclusions can differ substantially.\n\nNor does a qualitative theorem automatically certify a particular computation. Such a certification requires bounds on the errors, verification of the applicable dynamical assumptions, and control of the region in which the computed sequence lies.\n\nPeriodic orbits\n\nSuppose a pseudo-orbit repeats with period \\(N\\). If it has a unique sufficiently close shadowing point \\(y\\), then \\(f^N(y)\\) shadows the same sequence. Uniqueness gives\n\n\\[f^N(y)=y.\\]\n\nThus periodic pseudo-orbits can yield genuine periodic points, whose least periods divide \\(N\\).\n\nThis is closely related to closing lemmas for hyperbolic dynamics, which convert sufficiently close returns into periodic orbits. Shadowing and closing lemmas have distinct formulations: the former traces approximate trajectories, while the latter concerns closing nearly recurrent orbit segments.\n\nSymbolic dynamics and structural stability\n\nShadowing helps turn allowable symbolic itineraries into exact orbits. Together with local product structure and Markov partitions, it supports the description of hyperbolic systems by symbolic dynamics.\n\nIt is also central to structural stability. If a map \\(g\\) is uniformly close to \\(f\\), an orbit of \\(g\\) forms a pseudo-orbit for \\(f\\). Shadowing associates it with an exact \\(f\\)-orbit. Under suitable hypotheses, uniqueness and continuity turn this association into a conjugacy, or in weaker settings a semiconjugacy."
    },
    {
      "heading": "Variants and limitations",
      "text": "Continuous-time systems\n\nFor flows generated by ordinary differential equations, shadowing generally requires a reparametrization of time. A typical comparison has the form\n\n\\[d\\bigl(\\varphi_{h(t)}(y),x(t)\\bigr)<\\varepsilon,\\]\n\nwhere \\(h\\) is an increasing time change, with restrictions depending on the theorem.\n\nThis adjustment accounts for the direction tangent to a regular flow orbit, which is neither uniformly contracting nor uniformly expanding. Small errors in speed can accumulate into a large time displacement even when the trajectories remain geometrically close.\n\nFailure without suitable dynamical structure\n\nShadowing is not a consequence of continuity or compactness alone. Consider the identity map on the interval \\([0,1]\\). Every exact orbit is constant, but a pseudo-orbit can move from \\(0\\) to \\(1\\) in arbitrarily small steps and then remain at \\(1\\). No constant orbit stays within a fixed distance \\(\\varepsilon<1/2\\) of the entire sequence.\n\nThe obstruction is the accumulation of errors in a neutral direction. Nevertheless, hyperbolicity is sufficient rather than necessary for shadowing: some nonhyperbolic systems also possess the property.\n\nWeaker and stronger tracing properties\n\nRelated notions include:\n\n- Limit shadowing: step errors tending to zero are traced by an orbit whose tracing error also tends to zero.\n- Average shadowing: errors are controlled through averages rather than at every step.\n- Finite shadowing: tracing is formulated for finite pseudo-orbits.\n- Lipschitz shadowing: the tracing distance is bounded by a constant multiple of the step error.\n\nThese properties are not interchangeable; their relationships depend on assumptions such as compactness, invertibility, and expansivity."
    },
    {
      "heading": "Historical context",
      "text": "The classical hyperbolic shadowing lemma is associated with the work of Dmitri Anosov and Rufus Bowen in the development of uniformly hyperbolic dynamics. It became a standard tool for studying Anosov systems, hyperbolic invariant sets, periodic orbits, and symbolic descriptions of chaotic dynamics. Later work developed quantitative shadowing, computer-assisted orbit validation, and tracing properties beyond uniform hyperbolicity."
    }
  ],
  "links": [
    "dynamical-system",
    "chaos-theory",
    "metric-space",
    "continuous-function",
    "homeomorphism",
    "manifold",
    "tangent-space",
    "banach-fixed-point-theorem",
    "triangle-inequality",
    "floating-point-arithmetic",
    "numerical-analysis",
    "ordinary-differential-equation"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}