Crisis-induced intermittency

Crisis-induced intermittency is the dynamical regime that follows a boundary crisis in which a chaotic attractor has been destroyed by collision with a basin of attraction boundary. After the crisis, trajectories that previously settled into sustained chaotic motion now spend long episodes in the vicinity of the destroyed attractor's remnants — exhibiting nearly periodic, laminar behavior — before being abruptly ejected into a new dynamical regime. The laminar episodes are punctuated by chaotic bursts of increasing duration and intensity, creating a temporal structure that is neither fully ordered nor fully chaotic but an alternation between the two. The phenomenon was first identified and named by Celso Grebogi, Edward Ott, and James Yorke in 1982, and it has since been recognized as one of the signature post-crisis behaviors in dissipative dynamical systems.

The Mechanism

The mechanism of crisis-induced intermittency is geometrically simple but dynamically profound. Before the crisis, the chaotic attractor exists within a bounded region of phase space, surrounded by a basin of attraction whose boundary is typically fractal. As a control parameter is varied, the attractor grows and eventually collides with the basin boundary — an unstable periodic orbit or saddle point that marks the edge of the basin. At the moment of collision, the attractor and its basin vanish simultaneously. The system loses its chaotic home.

But the geometry of phase space does not disappear cleanly. The unstable periodic orbit that formed the basin boundary remains in the phase space, and trajectories launched near the former attractor still approach this orbit before diverging. During the approach, the trajectory shadows the ghost of the destroyed attractor, producing behavior that looks almost like the pre-crisis chaos — nearly periodic, with low-amplitude fluctuations. Eventually the trajectory reaches a region where the local manifold structure forces it away from the ghost, and it is ejected into the post-crisis regime: a distant attractor, divergence to infinity, or some other qualitatively different behavior.

The duration of the laminar episodes follows a characteristic scaling law. Near the crisis parameter value, the mean length of the laminar phase scales as a power law: ⟨τ⟩ ∝ |p − p_c|^{−γ}, where p is the control parameter, p_c is its critical value, and γ is a critical exponent that depends on the dimensionality and the local eigenvalues of the collision orbit. This power-law scaling is the dynamical signature of the crisis: the closer the system is to the critical parameter, the longer it lingers in the ghost regime, and the more the post-crisis dynamics resembles the pre-crisis chaos.

Crisis-Induced Intermittency and Other Forms of Intermittency

Crisis-induced intermittency is distinct from the three classical forms of intermittency identified by Yves Pomeau and Paul Manneville in 1980 — Type I, Type II, and Type III — which arise from saddle-node, Hopf, and inverse period-doubling bifurcations respectively. In Pomeau-Manneville intermittency, the laminar phases are associated with passage near a marginally stable fixed point or periodic orbit that has not yet disappeared; the system is on the verge of a bifurcation but has not crossed it. In crisis-induced intermittency, the orbit has already been destroyed. The ghost is not a real attractor; it is a phantom that haunts the phase space because the local geometry near the collision point still channels trajectories toward it for a finite time.

This distinction is philosophically significant. Pomeau-Manneville intermittency is a story about systems on the brink of change: they almost bifurcate, they hesitate, they eventually commit. Crisis-induced intermittency is a story about systems that have already changed: they are in the post-crisis regime, they know the old attractor is gone, but they cannot stop returning to its memory. The ghost does not exist; it is remembered by the geometry of the phase space. This is intermittency as mourning, as the inability of a dynamical system to fully abandon its former self.

Manifestations Across Domains

Fluid dynamics. In Rayleigh–Bénard convection, as the Rayleigh number is increased past certain thresholds, laminar convection rolls can give way to chaotic motion through a sequence of crises. The resulting regime often exhibits crisis-induced intermittency: long episodes of regular convection punctuated by bursts of turbulent activity, as the flow repeatedly attempts to reorganize into its pre-crisis pattern before being ejected into a new attractor. The pattern is not random but structured by the ghost of the destroyed convection state.

Neural dynamics. Crisis-induced intermittency has been proposed as a model for certain seizure dynamics in epilepsy. In some models, the transition from normal brain activity to seizure corresponds to a crisis in which a low-dimensional chaotic attractor — representing the normal dynamical repertoire of the cortex — collides with a basin boundary. The seizure itself is the post-crisis regime, but the pre-ictal period (the phase before the seizure) exhibits intermittent bursts of near-normal activity interrupted by increasing pathological discharges — the neural equivalent of crisis-induced intermittency. The ghost of normal activity persists until the seizure fully takes over.

Climate dynamics. Paleoclimate records show abrupt transitions between distinct climate states — glacial and interglacial, or different modes of ocean circulation — that may be modeled as crises in the Earth's climate attractor. The transitions are not instantaneous; they are preceded by intermittency in which the old climate state reasserts itself repeatedly before the final shift. The Dansgaard-Oeschger events, in which Greenland temperature oscillates between cold and warm states with increasing amplitude before a permanent shift, have been interpreted as crisis-induced intermittency in a coupled ocean-atmosphere system.

The Critical Exponent and Its Universality

The scaling exponent γ in crisis-induced intermittency is not universal across all systems; it depends on the dimension of the phase space and the eigenvalues of the unstable periodic orbit involved in the crisis. However, for certain classes of crises — particularly those involving two-dimensional dissipative maps and three-dimensional flows — the exponent takes on universal values that can be computed analytically. For a two-dimensional dissipative map with a boundary crisis involving a saddle periodic orbit with eigenvalues λ_u > 1 and λ_s < 1, the exponent is γ = −ln λ_s / ln λ_u, a formula that connects the global scaling of laminar durations to the local linearization of the collision orbit.

This formula is remarkable because it links a macroscopic observable — the mean duration of laminar episodes — to a microscopic property of the phase space geometry. It is an example of the kind of exponent relation that appears throughout critical phenomena, from the Feigenbaum constant in period-doubling cascades to the critical exponents in phase transitions. Crisis-induced intermittency is, in this sense, a critical phenomenon: the crisis point is a critical point in parameter space, and the power-law scaling of laminar durations is the critical divergence.

The Systems Reading

From a systems perspective, crisis-induced intermittency reveals something fundamental about the aftermath of structural change. When a system loses a stable operating regime, it does not immediately settle into a new one. It oscillates between the old and the new, unable to fully commit to either. The ghost of the old regime persists not because the system is nostalgic but because the system's geometry — its feedback loops, its network topology, its basin structure — still channels behavior toward the destroyed attractor. The transition is not a switch; it is a prolonged, painful, intermittently reversed process.

This has implications for how we think about regime shifts in any complex system. A financial market that has undergone a structural break does not immediately find a new equilibrium; it exhibits volatility clustering in which the old correlations reassert themselves before the new regime takes hold. An ecosystem that has lost a keystone species does not immediately reorganize; it shows transient recovery of the old community structure before the new stable state emerges. A political system that has undergone a crisis of legitimacy does not immediately settle into a new order; it oscillates between the old institutional forms and the new, with each oscillation progressively weakening the old and strengthening the new.

The formal structure of crisis-induced intermittency — the power-law scaling of laminar durations, the ghost attractor, the eventual ejection — provides a mathematical template for understanding these post-crisis oscillations. It tells us that the duration of the transitional phase is not arbitrary; it is determined by the system's internal geometry and by how close the parameter change was to the critical threshold. Near-critical changes produce long, painful transitions. Far-from-critical changes produce abrupt, clean breaks. The worst transitions are not the sudden ones; they are the ones that linger.

Crisis-induced intermittency is the mathematical form of a system that cannot let go. It tells us that destruction is not the end of a story but the beginning of a prolonged, oscillatory aftermath — and that the ghost of what was destroyed will haunt the dynamics until the system's geometry itself is rewritten.