{
  "id": "phase-portrait",
  "title": "Phase Portrait",
  "lang": "en",
  "source_lang": "en",
  "summary": "A phase portrait depicts trajectories in a dynamical system’s state space, revealing equilibria, stability, oscillations, and the qualitative organization of motion.",
  "category": "mathematics",
  "subcategory": "analysis",
  "type": "concept",
  "aliases": [
    "Phase-space portrait"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/phase-portrait/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/phase-portrait/",
      "json": "https://aiwiki.page/zh/phase-portrait.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "A phase portrait is a geometric representation of the trajectories of a dynamical system in its phase space, the space whose coordinates specify the system’s state. Instead of plotting each variable against time, it shows how the variables evolve together from different initial conditions. Arrows usually indicate the direction of increasing time. Phase portraits are especially useful for systems with two state variables, where they reveal equilibrium points, attraction and repulsion, periodic motion, and boundaries between different kinds of behavior."
    },
    {
      "heading": "Mathematical setting",
      "text": "For an autonomous system of ordinary differential equations,\n\n\\[\\dot{\\mathbf{x}}=\\mathbf{f}(\\mathbf{x}), \\qquad \\mathbf{x}\\in U\\subseteq\\mathbb{R}^{n},\\]\n\nthe state is \\(\\mathbf{x}\\), and the dot denotes differentiation with respect to time. “Autonomous” means that the right-hand side does not depend explicitly on time. The function \\(\\mathbf{f}\\) defines a vector field: at each state, it specifies the instantaneous direction and rate of change.\n\nA solution starting at \\(\\mathbf{x}_0\\) traces a trajectory\n\n\\[\\{\\mathbf{x}(t;\\mathbf{x}_0):t\\in I\\},\\]\n\nwhere \\(I\\) is its interval of existence. A phase portrait displays a representative collection of these trajectories, often together with equilibria and other invariant sets. An invariant set is one that trajectories starting within it do not leave.\n\nFor a planar system,\n\n\\[\\dot{x}=f(x,y),\\qquad \\dot{y}=g(x,y),\\]\n\nthe portrait lies in the \\(xy\\)-plane. The coordinates need not be spatial positions: they may represent position and velocity, two population sizes, or electrical quantities.\n\nA trajectory as a geometric curve does not, by itself, specify how quickly the system moves along it. Multiplying the vector field by a strictly positive scalar function preserves the oriented trajectories wherever that function is defined, while changing their time parametrization. Consequently, phase portraits primarily describe qualitative motion rather than precise timing.\n\nUniqueness and nonintersection\n\nIf \\(\\mathbf{f}\\) satisfies suitable local regularity conditions, such as local Lipschitz continuity, a state determines a unique local solution. Distinct trajectories of an autonomous system therefore cannot cross at the same state: a crossing would give two different evolutions from one initial condition.\n\nThis statement requires several qualifications:\n\n- Two solution curves may describe the same orbit with different choices of starting time.\n- A periodic trajectory returns to an earlier state, forming a closed orbit rather than a transverse self-intersection.\n- A nonconstant trajectory may approach an equilibrium asymptotically without reaching it in finite time.\n- A projection of a higher-dimensional trajectory may cross itself even though the full trajectory does not."
    },
    {
      "heading": "Reading a planar phase portrait",
      "text": "Equilibria and stability\n\nAn equilibrium \\(\\mathbf{x}_*\\) satisfies\n\n\\[\\mathbf{f}(\\mathbf{x}_*)=\\mathbf{0}.\\]\n\nIt is a fixed point of the evolution: a system placed exactly there remains there.\n\nAn equilibrium is stable in the sense of Lyapunov if trajectories starting sufficiently close remain close for all future time. It is asymptotically stable if it is stable and nearby trajectories also converge to it. An unstable equilibrium fails the stability condition.\n\nThese distinctions matter visually. Nearby trajectories may circulate indefinitely around a stable equilibrium without converging to it, whereas trajectories near an asymptotically stable equilibrium approach it. A portrait should therefore distinguish “remaining nearby” from “being attracted.”\n\nNullclines\n\nFor a planar system, the nullclines are the sets\n\n\\[f(x,y)=0 \\qquad\\text{and}\\qquad g(x,y)=0.\\]\n\nOn an \\(x\\)-nullcline, the horizontal component of velocity vanishes, so the vector field is vertical unless both components vanish. On a \\(y\\)-nullcline, it is horizontal. Intersections of the two nullcline sets are equilibria.\n\nNullclines help divide the plane into regions where the signs of \\(\\dot{x}\\) and \\(\\dot{y}\\) are known. They are generally not trajectories: solutions usually cross them. Their geometry provides a framework for sketching motion, but does not alone determine the complete portrait.\n\nAway from points where \\(f=0\\), trajectories also satisfy\n\n\\[\\frac{dy}{dx}=\\frac{g(x,y)}{f(x,y)}.\\]\n\nThis slope equation describes their geometric shape but loses the direction of time unless additional information is retained.\n\nPeriodic orbits and separatrices\n\nA nonconstant periodic solution produces a closed trajectory. A limit cycle is an isolated periodic orbit: there is a neighborhood containing no other periodic orbit. It may attract nearby trajectories, repel them, or attract from one side and repel from the other.\n\nA separatrix separates regions with qualitatively different motion. Important examples are trajectories approaching or leaving a saddle equilibrium. Separatrices may form boundaries between basins of attraction, where a basin consists of initial states converging to a particular attracting set.\n\nNot every closed trajectory is a limit cycle. A family of nested periodic orbits, such as those around an ideal linear center, contains no isolated member."
    },
    {
      "heading": "Linear systems and local classification",
      "text": "For a planar linear system\n\n\\[\\dot{\\mathbf{x}}=A\\mathbf{x},\\]\n\nthe eigenvalues and eigenvectors of the real \\(2\\times2\\) matrix \\(A\\) determine the basic portrait near the origin.\n\nEigenvalue pattern | Typical phase portrait\nTwo negative real eigenvalues | Stable node\nTwo positive real eigenvalues | Unstable node\nReal eigenvalues of opposite signs | Saddle\nComplex conjugates with negative real part | Stable spiral, also called a stable focus\nComplex conjugates with positive real part | Unstable spiral or focus\nNonzero purely imaginary conjugate eigenvalues | Center, with nested closed trajectories\n\nRepeated nonzero real eigenvalues give nodal cases whose geometry depends on whether there are two independent eigenvectors. Zero eigenvalues produce additional degenerate cases, including possible lines of equilibria.\n\nThe trace \\(\\tau=\\operatorname{tr}(A)\\) and determinant \\(\\Delta=\\det(A)\\) provide a convenient classification through\n\n\\[\\lambda^2-\\tau\\lambda+\\Delta=0.\\]\n\nIn particular:\n\n- \\(\\Delta<0\\) gives a saddle.\n- \\(\\Delta>0\\) and \\(\\tau^2-4\\Delta>0\\) give distinct real eigenvalues and a node.\n- \\(\\Delta>0\\) and \\(\\tau^2-4\\Delta<0\\) give a spiral when \\(\\tau\\ne0\\), and a center when \\(\\tau=0\\).\n\nNonlinear systems\n\nNear an equilibrium of a differentiable nonlinear system, write \\(\\mathbf{x}=\\mathbf{x}_*+\\boldsymbol{\\xi}\\). To first order,\n\n\\[\\dot{\\boldsymbol{\\xi}} =J\\boldsymbol{\\xi}, \\qquad J=D\\mathbf{f}(\\mathbf{x}_*),\\]\n\nwhere \\(J\\) is the Jacobian matrix.\n\nIf every eigenvalue of \\(J\\) has nonzero real part, the equilibrium is called hyperbolic. Under the usual smoothness assumptions, its local orbit structure is topologically equivalent to that of the linearized system. This gives a rigorous basis for identifying nonlinear nodes, saddles, and spirals from the Jacobian.\n\nIf an eigenvalue has zero real part, linearization may be inconclusive. Purely imaginary eigenvalues, for example, do not establish that a nonlinear equilibrium is a center; nonlinear terms may instead cause attraction or repulsion. Likewise, a local classification does not determine the global arrangement of trajectories."
    },
    {
      "heading": "Representative examples",
      "text": "Harmonic oscillator\n\nThe undamped oscillator\n\n\\[\\ddot{x}+\\omega^2x=0, \\qquad \\omega>0,\\]\n\nbecomes\n\n\\[\\dot{x}=v,\\qquad \\dot{v}=-\\omega^2x.\\]\n\nIts trajectories satisfy\n\n\\[v^2+\\omega^2x^2=C.\\]\n\nFor \\(C>0\\), these are ellipses surrounding a center. They represent periodic motion, and none is a limit cycle because neighboring trajectories are also closed.\n\nAdding linear damping,\n\n\\[\\dot{v}=-\\omega^2x-cv, \\qquad c>0,\\]\n\nmakes the origin asymptotically stable. For \\(0<c<2\\omega\\), trajectories spiral inward; for \\(c>2\\omega\\), they approach as a stable node. The critical case \\(c=2\\omega\\) has a repeated negative eigenvalue.\n\nA system with a limit cycle\n\nConsider\n\n\\[\\dot{x}=x(1-x^2-y^2)-y, \\qquad \\dot{y}=y(1-x^2-y^2)+x.\\]\n\nIn polar coordinates, away from the origin,\n\n\\[\\dot{r}=r(1-r^2),\\qquad \\dot{\\theta}=1.\\]\n\nTrajectories with \\(0<r<1\\) move outward, while those with \\(r>1\\) move inward. The circle \\(r=1\\) is therefore a stable limit cycle, and the origin is unstable.\n\nThe pendulum\n\nFor an ideal pendulum,\n\n\\[\\dot{\\theta}=\\omega,\\qquad \\dot{\\omega}=-\\frac{g}{\\ell}\\sin\\theta,\\]\n\nsmall oscillations correspond to closed trajectories around the downward equilibrium. Separatrices through the upright saddle equilibria divide oscillatory motion from continuous rotation.\n\nBecause angles differing by \\(2\\pi\\) represent the same configuration, the natural phase space is a cylinder. A planar plot using an unrestricted angle coordinate repeats the same physical configurations, illustrating why the topology of phase space matters."
    },
    {
      "heading": "Global behavior and bifurcations",
      "text": "Phase portraits organize motion beyond a single equilibrium. They can show several attractors, their basins, connecting trajectories, and escape to infinity.\n\nFor planar continuous-time systems, the Poincaré–Bendixson theorem places strong restrictions on long-term behavior. In a standard formulation, a forward trajectory confined to a compact subset of the domain, whose omega-limit set contains no equilibrium, has a periodic orbit as its omega-limit set. The omega-limit set consists of accumulation points approached at arbitrarily late times.\n\nThis is one reason planar autonomous differential equations behave differently from higher-dimensional systems. The familiar strange attractors associated with chaos do not occur in smooth autonomous flows on the plane. Periodically forced systems, discrete-time maps, and higher-dimensional flows are not subject to the same planar restriction.\n\nAs parameters vary, a portrait may change qualitatively through a bifurcation. Equilibria may appear or disappear, change stability, or give rise to periodic orbits. A Hopf bifurcation, under appropriate nondegeneracy conditions, involves a pair of complex eigenvalues crossing the imaginary axis and the emergence or disappearance of a nearby periodic orbit. Comparing portraits across parameter values helps distinguish a qualitative transition from a mere change in trajectory shape."
    },
    {
      "heading": "Construction and limitations",
      "text": "A practical construction combines analytical and numerical methods:\n\n- Identify the state space and any constraints.\n- Find equilibria and nullclines.\n- Classify equilibria using linearization where applicable.\n- Examine invariant sets, conserved quantities, and monotone functions.\n- Integrate representative trajectories from different initial conditions.\n- Check suspected periodic orbits and basin boundaries with additional analysis.\n\nNumerical analysis is often essential, but a plot alone is not a proof. Finite integration times may make slowly converging trajectories look periodic. Integration errors can cause artificial drift, and coarse sampling may miss narrow basins or unstable trajectories. Normalized direction-field arrows display orientation clearly but suppress speed information.\n\nFor nonautonomous equations, trajectories can cross in the displayed state plane because the vector field depends on time. Adding time, or a forcing phase for periodic forcing, restores an autonomous description in a larger space. For discrete-time systems, phase-space plots usually show successive states rather than continuous flow curves, and planar-flow nonintersection rules do not apply in the same way.\n\nThe geometric approach developed through nineteenth-century qualitative differential-equation theory, particularly the work of Henri Poincaré. It remains central in classical mechanics, nonlinear oscillation theory, population dynamics, and control theory, where the organization of possible motion may be more informative than an explicit solution formula."
    }
  ],
  "links": [
    "dynamical-system",
    "phase-space",
    "ordinary-differential-equation",
    "vector-field",
    "lipschitz-continuity",
    "fixed-point",
    "eigenvalues-and-eigenvectors",
    "matrix",
    "matrix-trace",
    "determinant",
    "jacobian-matrix",
    "polar-coordinates",
    "topology",
    "chaos-theory",
    "bifurcation",
    "numerical-analysis",
    "classical-mechanics"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}