{
  "id": "kam-theorem",
  "title": "Kolmogorov–Arnold–Moser Theorem",
  "lang": "en",
  "source_lang": "en",
  "summary": "The Kolmogorov–Arnold–Moser theorem establishes the persistence of many quasiperiodic motions in sufficiently small perturbations of integrable Hamiltonian systems.",
  "category": "mathematics",
  "subcategory": "analysis",
  "type": "concept",
  "aliases": [
    "KAM theorem",
    "Kolmogorov–Arnold–Moser theory",
    "KAM theory"
  ],
  "generated_at": "2026-11-06",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/kam-theorem/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/kam-theorem/",
      "json": "https://aiwiki.page/zh/kam-theorem.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "The Kolmogorov–Arnold–Moser theorem, usually abbreviated KAM theorem, is a foundational result in dynamical systems concerning the survival of regular motion under perturbation. In its classical setting, an integrable Hamiltonian system has families of invariant tori on which motion is quasiperiodic: several independent oscillations proceed together without exactly repeating. KAM theory shows that many of these tori persist when the equations are changed sufficiently little, provided that suitable arithmetic and nondegeneracy conditions hold.\n\nThe theorem explains why weakly interacting conservative systems need not become completely chaotic. Its surviving tori occupy positive phase-space measure, and under standard hypotheses their relative measure approaches full measure as the perturbation tends to zero. Nevertheless, resonances can produce gaps between them, and the theorem does not imply that every nearby trajectory remains regular or that all actions remain confined indefinitely.\n\nNamed after Andrey Kolmogorov, Vladimir Arnold, and Jürgen Moser, the theorem developed from work in the 1950s and 1960s into a broad theory encompassing Hamiltonian flows, area-preserving maps, celestial mechanics, and certain infinite-dimensional equations. Its central achievement is to overcome the “small-divisor problem,” in which attempts to calculate perturbed motions encounter denominators that can become arbitrarily small."
    },
    {
      "heading": "Historical origins and the stability problem",
      "text": "The background to KAM theory lies in the effort to understand the long-term behavior of systems in classical mechanics, especially planetary motion. A single planet moving under an idealized Newtonian two-body interaction follows an integrable model. With several planets, mutual gravitational perturbations complicate the motion. During the eighteenth and nineteenth centuries, mathematicians developed elaborate perturbation expansions intended to describe these effects through corrections to simpler motions.\n\nA persistent obstacle was the appearance of small divisors. If a system has basic frequencies \\(\\omega_1,\\ldots,\\omega_n\\), solving perturbative equations produces divisions by combinations\n\n\\[k_1\\omega_1+\\cdots+k_n\\omega_n, \\qquad k\\in\\mathbb Z^n.\\]\n\nEven when none of these combinations vanishes, some can be extremely small. Consequently, successively computed corrections need not become smaller. A formally consistent expansion may fail to converge, and finite-order accuracy does not by itself establish indefinite stability.\n\nHenri Poincaré’s work on the three-body problem revealed fundamental limitations of classical perturbation methods and the complicated geometry associated with nonintegrability. It did not establish that all regular motion disappears under perturbation. That distinction became crucial: failure of a general integrability scheme is compatible with the persistence of large sets of individual regular trajectories.\n\nIn 1954, Andrey Kolmogorov announced a theorem asserting the preservation of many quasiperiodic motions in analytic, nearly integrable Hamiltonian systems. His key methodological insight was a rapidly convergent iteration rather than an ordinary term-by-term expansion in the perturbation parameter. The corrections could be arranged so that the remaining error decreased approximately quadratically, fast enough to compensate for losses caused by small divisors.\n\nMoser’s 1962 invariant-curve theorem established a related persistence result for sufficiently differentiable area-preserving maps of an annulus. Arnold supplied a detailed analytic development of Kolmogorov’s approach in 1963 and extended the theory toward problems in celestial mechanics. These contributions addressed closely related, but not identical, mathematical settings. Accordingly, “the KAM theorem” now often denotes a family of persistence theorems rather than one statement with a single universal collection of assumptions.\n\nThe historical change was substantial. Stability could be proved not by finding enough conserved quantities for the entire perturbed system, but by constructing invariant structures on a large subset of its phase space. This made it possible to reconcile nonintegrability with extensive, rigorously established regular motion."
    },
    {
      "heading": "Integrable Hamiltonians and quasiperiodic tori",
      "text": "The standard starting point is an integrable Hamiltonian expressed in action–angle variables\n\n\\[(I,\\theta)\\in D\\times\\mathbb T^n,\\]\n\nwhere \\(D\\) is an open subset of \\(\\mathbb R^n\\), \\(I=(I_1,\\ldots,I_n)\\) are actions, and\n\n\\[\\mathbb T^n=(\\mathbb R/2\\pi\\mathbb Z)^n\\]\n\nis the \\(n\\)-dimensional torus of angles. The actions label different invariant tori, while the angles identify positions on each torus. The Liouville–Arnold theorem provides these coordinates locally near a compact, connected, regular common level set of the requisite commuting first integrals.\n\nFor an integrable Hamiltonian \\(H_0(I)\\), Hamilton’s equations reduce to\n\n\\[\\dot I=0, \\qquad \\dot\\theta=\\omega(I), \\qquad \\omega(I)=\\nabla H_0(I).\\]\n\nThus the solution starting at \\((I_0,\\theta_0)\\) is\n\n\\[I(t)=I_0, \\qquad \\theta(t)=\\theta_0+\\omega(I_0)t \\pmod{2\\pi}.\\]\n\nEach set \\(I=I_0\\) is an invariant torus. The motion on it is a linear translation with a constant frequency vector.\n\nIf the components of \\(\\omega\\) are rationally independent, meaning\n\n\\[k\\cdot\\omega\\ne0 \\quad\\text{for every }k\\in\\mathbb Z^n\\setminus\\{0\\},\\]\n\neach trajectory is dense on the torus. This is quasiperiodic motion. It is distinct from periodic motion, which returns exactly after a finite period, and from chaotic motion. A quasiperiodic observable can contain many combination frequencies even though the underlying evolution is conjugate to a simple linear translation.\n\nA perturbation introduces angle dependence:\n\n\\[H_\\varepsilon(I,\\theta) = H_0(I)+\\varepsilon H_1(I,\\theta).\\]\n\nNow\n\n\\[\\dot I=-\\varepsilon\\,\\partial_\\theta H_1,\\]\n\nso the original sets \\(I=\\text{constant}\\) generally cease to be invariant. KAM theory seeks nearby, deformed tori rather than insisting that the original ones remain fixed. Their embeddings can vary with the angles, and their locations in the action variables usually shift.\n\nThe geometry is naturally expressed through symplectic geometry. In canonical coordinates, the symplectic form is\n\n\\[\\Omega=\\sum_{j=1}^{n}dI_j\\wedge d\\theta_j.\\]\n\nThe unperturbed invariant tori are Lagrangian: they have dimension \\(n\\), half the phase-space dimension, and the restriction of \\(\\Omega\\) to them vanishes. The classical persisting KAM tori retain this property.\n\nAction–angle coordinates are not generally available everywhere. Separatrices, singular invariant sets, and changes in the topology of level sets obstruct a single global description. Classical KAM statements are therefore usually formulated on regular action domains, often restricted away from their boundaries."
    },
    {
      "heading": "A standard persistence theorem and its hypotheses",
      "text": "A representative analytic KAM theorem considers\n\n\\[H_\\varepsilon(I,\\theta)=H_0(I)+\\varepsilon H_1(I,\\theta)\\]\n\non an action–angle domain. Both terms are assumed real analytic, with controlled analytic extensions to specified complex neighborhoods. The perturbation must be small in an appropriate analytic norm. Saying merely that \\(\\varepsilon\\) is numerically small is insufficient unless the size and domain of \\(H_1\\) have also been specified.\n\nThe classical Kolmogorov nondegeneracy condition is\n\n\\[\\det D^2H_0(I)\\ne0.\\]\n\nHere \\(D^2H_0\\) is the Hessian matrix. Since \\(\\omega(I)=\\nabla H_0(I)\\), this condition says that the frequency map is locally invertible. Nearby changes in frequency can therefore be produced by nearby changes in action. This freedom allows the proof to compensate for frequency changes introduced by the perturbation.\n\nFor a frequency vector satisfying a suitable Diophantine condition, and for a sufficiently small perturbation, the theorem constructs an embedding\n\n\\[K_\\varepsilon:\\mathbb T^n \\longrightarrow D\\times\\mathbb T^n\\]\n\nsatisfying the invariance equation\n\n\\[X_{H_\\varepsilon}\\bigl(K_\\varepsilon(\\varphi)\\bigr) = DK_\\varepsilon(\\varphi)\\,\\omega.\\]\n\nHere \\(X_{H_\\varepsilon}\\) is the Hamiltonian vector field. The equation states that the flow on the embedded torus is conjugate to\n\n\\[\\dot\\varphi=\\omega.\\]\n\nThe torus lies near the corresponding unperturbed torus, and its motion remains quasiperiodic.\n\nIn a frequency-preserving formulation, the perturbed torus retains a prescribed frequency vector, while its position in action space changes. Other formulations describe a family of tori whose frequencies are relabeled or slightly shifted. These statements are compatible, but their parameter conventions differ.\n\nFor \\(n\\ge2\\), the surviving tori typically form a family indexed by a Cantor-like subset of the original action domain. The family need not fill an open region, yet its union has positive Lebesgue measure in phase space. On suitable compact subdomains, the relative measure of the excluded set tends to zero as the perturbation tends to zero, with the Diophantine cutoff chosen appropriately.\n\nThe smallness threshold depends on the dimension, analytic widths, nondegeneracy bounds, arithmetic constants, and perturbation norm. There is no universal numerical threshold valid for all systems. Furthermore, this theorem is sufficient rather than necessary: a torus that fails one of its hypotheses may nevertheless exist, but its persistence is not established by this particular statement."
    },
    {
      "heading": "Diophantine frequencies and small divisors",
      "text": "The arithmetic condition most commonly used in classical KAM theory is\n\n\\[|k\\cdot\\omega| \\ge \\frac{\\gamma}{|k|^\\tau} \\qquad \\text{for every }k\\in\\mathbb Z^n\\setminus\\{0\\},\\]\n\nwhere \\(\\gamma>0\\), \\(\\tau>n-1\\), and \\(|k|\\) is a fixed norm on integer vectors. Such an \\(\\omega\\) is called a Diophantine frequency vector. The condition gives a quantitative lower bound on the nonzero divisors that occur in the proof.\n\nThis is stronger than rational independence. Rational independence excludes exact resonances, but does not prevent extremely accurate approximate resonances. Some vectors have integer combinations approaching zero faster than any prescribed inverse power of \\(|k|\\). For these vectors, the usual Diophantine estimates are unavailable. The distinction connects KAM theory with Diophantine approximation and number theory.\n\nThe analytic difficulty can be seen in the cohomological equation\n\n\\[\\omega\\cdot\\partial_\\theta u(\\theta)=f(\\theta),\\]\n\nwhere \\(f\\) has zero average. Expanding \\(f\\) and \\(u\\) in Fourier series gives\n\n\\[u_k=\\frac{f_k}{i\\,k\\cdot\\omega}, \\qquad k\\ne0.\\]\n\nEven a smooth \\(f\\) can produce a poorly controlled solution if the denominators become too small. The zero Fourier mode must be treated separately because differentiation annihilates constants.\n\nFor analytic \\(f\\), the coefficients \\(f_k\\) decay exponentially with \\(|k|\\). A Diophantine lower bound allows the polynomial growth introduced by division to be absorbed by a small reduction in the analytic domain. The resulting estimate typically includes negative powers of the lost analytic width and of \\(\\gamma\\). In finitely differentiable settings, the same division costs derivatives, making smoothing procedures necessary.\n\nThe arithmetic condition is restrictive locally but large in measure globally. On a bounded frequency region, for a fixed exponent \\(\\tau>n-1\\), the measure removed by excluding resonance strips can be bounded by a constant times \\(\\gamma\\). Almost every frequency vector satisfies such a condition for some positive \\(\\gamma\\). Nevertheless, exact resonance hyperplanes are dense in the relevant multidimensional sense: every frequency neighborhood encounters some resonance.\n\nFor two-frequency systems, the arithmetic can often be expressed through the ratio of the frequencies. Continued fractions then provide a practical language for assessing rational approximation. Ratios with bounded continued-fraction coefficients, including the golden ratio, are especially resistant to rational approximation. This is one reason golden-mean rotation numbers play a prominent role in studies of invariant-circle breakup."
    },
    {
      "heading": "The rapidly convergent proof mechanism",
      "text": "The central proof problem is to remove angle dependence without allowing accumulated small-divisor losses to overwhelm the construction. An ordinary perturbation series attempts to organize the answer by powers of \\(\\varepsilon\\). KAM proofs instead repeatedly solve a linearized correction problem, obtaining a sequence of increasingly accurate invariant tori or normal forms.\n\nIn a normal-form approach, a near-identity canonical transformation is chosen to cancel the principal unwanted terms. Schematically, its generating function \\(S\\) solves a homological equation\n\n\\[\\omega\\cdot\\partial_\\theta S = f-\\langle f\\rangle,\\]\n\nwhere \\(\\langle f\\rangle\\) denotes angular average. The nonzero Fourier modes can be solved using the Diophantine estimates. The average cannot be eliminated this way; it contributes to changes in the integrable part or to translations of the torus.\n\nNondegeneracy enters when these average changes must be controlled. If a correction alters the frequencies, the invertibility of \\(D^2H_0\\) permits an action adjustment that restores the intended frequency. This separates two obstacles: small divisors govern oscillatory terms, while nondegeneracy governs the solvability of the finite-dimensional average equations.\n\nAfter a successful step, the new error is roughly quadratic in the old one:\n\n\\[E_{j+1}\\le C_jE_j^2.\\]\n\nThe constants \\(C_j\\) may grow because each step loses some analytic width. The widths are chosen so that their total loss remains finite, while the errors decrease sufficiently rapidly. With suitable initial smallness, quadratic improvement dominates the deteriorating constants.\n\nThis resembles Newton’s method, but in a space of functions where inverses are not uniformly bounded in a single norm. Solving the linearized equation typically loses regularity or analytic width. Consequently, a direct appeal to a simple contraction theorem does not capture the main difficulty.\n\nModern proofs often work directly with the embedding equation. Starting from an approximate embedding \\(K\\), they calculate its defect\n\n\\[E(\\varphi) = X_H(K(\\varphi))-DK(\\varphi)\\omega\\]\n\nand construct a correction \\(\\Delta K\\). The revised embedding has a much smaller defect. Symplectic identities simplify the linearized equations and help establish the required solvability conditions.\n\nFor Hamiltonians with only finitely many derivatives, smoothing replaces analyticity as the main control mechanism. Such methods are related to Nash–Moser iteration, which compensates for derivative loss by combining smoothing with rapid convergence. Required differentiability depends on the theorem’s dimension and formulation. Analytic and smooth KAM results therefore express the same persistence principle through different technical estimates, not through an identical regularity hypothesis."
    },
    {
      "heading": "Resonance, mixed dynamics, and barriers to transport",
      "text": "KAM theory predicts a phase space containing both regular and potentially irregular regions. A resonance is specified by\n\n\\[k\\cdot\\omega(I)=0\\]\n\nfor a nonzero integer vector \\(k\\). Near such a surface, a corresponding combination of angles evolves slowly. Perturbation terms involving that slow angle can accumulate rather than averaging away, producing dynamics that differ qualitatively from those on a nonresonant torus.\n\nIn two-dimensional area-preserving maps, resonant regions commonly contain chains of elliptic and hyperbolic periodic points. Near an elliptic point, smaller invariant curves may survive; near hyperbolic points, intersections of stable and unstable manifolds can create complicated orbit patterns. These structures underlie the familiar coexistence of islands and chaotic layers in chaos theory. KAM theory establishes the regular tori under its hypotheses, but does not by itself classify every orbit in the intervening gaps.\n\nThe effect of a surviving torus on transport depends critically on dimension. An autonomous Hamiltonian system with \\(n\\) degrees of freedom has a \\(2n\\)-dimensional phase space and, at regular energies, a \\((2n-1)\\)-dimensional energy surface. A KAM torus has dimension \\(n\\), hence codimension \\(n-1\\) within that energy surface.\n\nFor \\(n=2\\), the tori have codimension one on the energy surface. In standard action–angle regions they act as barriers: a trajectory cannot cross an invariant torus. Equivalently, a rotational invariant circle in an annulus can separate the annulus into distinct regions. Global separation nevertheless depends on the topology of the ambient space, not on dimension alone.\n\nFor \\(n\\ge3\\), an individual invariant torus generally does not separate the energy surface. Trajectories outside the tori may move around them through resonance regions. This opens the possibility of Arnold diffusion, slow changes in actions associated with transport through a resonance network. Arnold constructed an influential example in 1964. The existence and rate of such transport require additional arguments; they do not follow merely from gaps in the KAM set.\n\nA complementary result is Nekhoroshev theory. Under suitable steepness and regularity assumptions, it bounds action drift for very long but finite times, often exponentially long in an inverse power of perturbation size. KAM gives eternal quasiperiodic motion on selected invariant tori, whereas Nekhoroshev estimates can control a broader collection of initial conditions without asserting that their motions are quasiperiodic."
    },
    {
      "heading": "Variants for maps, fixed energies, and lower-dimensional tori",
      "text": "For maps, a central example is an integrable area-preserving twist map of an annulus:\n\n\\[I'=I, \\qquad \\theta'=\\theta+\\Omega(I)\\pmod{2\\pi},\\]\n\nwith\n\n\\[\\Omega'(I)\\ne0.\\]\n\nThe twist condition means that neighboring invariant circles rotate at different rates. Suitable sufficiently small perturbations preserve many circles whose rotation numbers satisfy appropriate arithmetic conditions. Moser’s invariant-curve theorem is a foundational result of this kind.\n\nThe standard map provides a widely studied concrete model:\n\n\\[p_{m+1}=p_m+K\\sin q_m, \\qquad q_{m+1}=q_m+p_{m+1}\\pmod{2\\pi}.\\]\n\nWhen \\(K=0\\), every circle \\(p=\\text{constant}\\) is invariant. For sufficiently small \\(|K|\\), many rotational invariant circles survive. As \\(|K|\\) increases, particular circles may break up, and the global organization of transport changes. The small-perturbation theorem alone does not locate the final breakup of any specified circle.\n\nFor autonomous flows, an important alternative is isoenergetic KAM theory, which seeks invariant tori on a prescribed energy surface. A standard nondegeneracy condition is\n\n\\[\\det \\begin{pmatrix} D^2H_0(I)&\\nabla H_0(I)\\\\ \\nabla H_0(I)^{T}&0 \\end{pmatrix} \\ne0.\\]\n\nThis condition controls the variation of frequency ratios along an energy surface rather than demanding unrestricted local invertibility of the full frequency map. The resulting theorem can preserve the energy while allowing an overall rescaling of the frequency vector.\n\nOther variants address integrable Hamiltonians for which the classical Hessian determinant vanishes. Such degeneracy occurs naturally in mechanical problems with symmetries or exceptionally simple unperturbed motions. Persistence may still follow after reduction, averaging, or replacement of Kolmogorov nondegeneracy by a weaker condition. Failure of the classical determinant condition is therefore not equivalent to the destruction of quasiperiodicity.\n\nLower-dimensional invariant tori require further distinctions. A torus of dimension \\(d<n\\) has transverse directions whose dynamics matter independently of its tangential frequencies. For normally elliptic tori, additional small divisors involve normal oscillation frequencies as well as integer combinations of tangential frequencies. Extra nonresonance assumptions and parameters are commonly needed.\n\nNormally hyperbolic tori behave differently: sufficiently strong transverse contraction and expansion can yield persistence through hyperbolic invariant-manifold theory rather than classical KAM estimates. KAM methods may still be needed to control quasiperiodic motion on the persisting manifold. The broader theory thus separates persistence of an invariant geometric object from preservation of a particular dynamical behavior within it."
    },
    {
      "heading": "Celestial mechanics, oscillators, and infinite-dimensional equations",
      "text": "Celestial mechanics supplied the original motivation for KAM theory, but direct application is subtle. The Kepler Hamiltonian has substantial degeneracy: its frequencies do not vary independently with all the actions. Planetary problems also possess rotational symmetries, several perturbation scales, and resonances. A valid application must handle these features through appropriate reductions and normal forms rather than simply declaring planetary masses to be small.\n\nArnold’s work established the relevance of KAM methods to planetary stability, and subsequent developments produced increasingly general planetary persistence results. Their conclusions concern positive-measure sets of quasiperiodic motions under explicit assumptions, such as sufficiently small planetary mass ratios and restrictions on orbital configurations. They do not assert that every initial condition in a planetary system is stable.\n\nIn particular, a planetary KAM theorem is not an unconditional proof that the actual Solar System is stable forever. The mathematical model, parameter bounds, selected initial conditions, and omitted physical effects all matter. The theorem’s rigorous content is the existence and persistence of specified invariant motions in the stated model.\n\nWeakly coupled nonlinear oscillators offer another important setting. At zero coupling, each oscillator can contribute an action and an amplitude-dependent frequency. If the collective frequency map satisfies the relevant nondegeneracy conditions, small coupling leaves many quasiperiodic states intact. Purely harmonic oscillators have amplitude-independent frequencies and therefore fall outside the simplest twist hypothesis, though more specialized persistence results may apply.\n\nKAM ideas also enter the study of magnetic field lines. In suitable toroidal geometries, a field-line return map can be represented as an area-preserving map. Invariant circles of that map correspond to magnetic surfaces, while resonant structures can disrupt such surfaces. This provides a geometric connection between invariant-torus theory and magnetic confinement, without replacing the additional physical analysis required for plasma dynamics.\n\nInfinite-dimensional KAM theory concerns certain Hamiltonian partial differential equations, including nonlinear wave and Schrödinger equations. Their linearized dynamics have infinitely many normal modes. Constructing a quasiperiodic solution requires controlling interactions between finitely many chosen tangential frequencies and infinitely many transverse frequencies.\n\nThe infinite number of modes introduces spectral and regularity difficulties absent from finite-dimensional systems. Results depend on features such as eigenvalue growth, boundary conditions, spatial dimension, external parameters, and the derivative structure of the nonlinearity. The existence of a finite-dimensional family of quasiperiodic solutions does not imply that the full equation is integrable or that all nearby solutions share the same behavior."
    },
    {
      "heading": "Computation, torus breakup, and limits of applicability",
      "text": "Numerical investigation of KAM tori often starts from sampled trajectories, Fourier spectra, or direct solution of an invariance equation. A quasiperiodic trajectory ideally has discrete spectral lines at integer combinations of its fundamental frequencies. However, a finite observation window cannot by itself establish exact quasiperiodicity. Slowly drifting trajectories or chaotic trajectories that remain near regular regions for long periods can look quasiperiodic over the available integration time.\n\nThe parameterization method instead represents an invariant torus directly. For a map \\(F\\), it seeks an embedding satisfying\n\n\\[F(K(\\varphi))=K(\\varphi+\\omega).\\]\n\nFourier discretization turns this functional equation into a computational problem. Iterative corrections can then exploit the same cohomological equations and small-divisor structure that occur in an analytic proof.\n\nA posteriori KAM theorems provide a bridge between approximate computation and rigorous existence. Rather than assuming that the entire Hamiltonian is close to an explicitly integrable model, they begin with an approximately invariant torus. If its residual is sufficiently small relative to verified arithmetic, nondegeneracy, and regularity bounds, an exact nearby invariant torus exists. With controlled numerical error, such a theorem can certify a computed object. A small residual alone is insufficient without the accompanying bounds.\n\nAs perturbations increase, the breakup of invariant circles becomes a separate subject from small-perturbation persistence. In the standard map, the breakup associated with the golden-mean rotational invariant circle is conventionally located near\n\n\\[K\\approx0.971635\\]\n\nin the normalization used above. This value belongs to a particular model and rotation number; it is not a universal KAM threshold.\n\nJohn Greene’s residue criterion, introduced in 1979, studies periodic orbits whose rotation numbers approximate that of a proposed invariant circle. Their stability data provide a powerful diagnostic for persistence and breakup. The criterion’s numerical success and its rigorous mathematical status must be distinguished: its full scope is not supplied by the classical KAM theorem.\n\nOpen and active directions include sharper arithmetic hypotheses, lower differentiability requirements, quantitative thresholds for concrete systems, degenerate problems, infinite-dimensional equations, and the interaction between surviving tori and transport outside them. Critical breakup involves scaling behavior and geometric structures that are not determined by a theorem designed for sufficiently small perturbations. Likewise, positive measure of regular motions does not specify the behavior of every complementary trajectory or establish the long-term statistical properties of a system."
    }
  ],
  "links": [
    "dynamical-system",
    "hamiltonian-mechanics",
    "classical-mechanics",
    "andrey-kolmogorov",
    "hessian-matrix",
    "vector-field",
    "lebesgue-measure",
    "number-theory",
    "fourier-series",
    "continued-fraction",
    "golden-ratio",
    "newtons-method",
    "banach-fixed-point-theorem",
    "chaos-theory",
    "partial-differential-equation"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}