Steady states are the default prediction. Equilibrium, convergence, fixed point. Oscillation requires explanation — something must prevent the system from settling. These four results give four distinct answers, and the answers aren't about what drives the oscillation. They're about what prevents convergence.
Nemeth et al. (2026, arXiv:2602.17765) study open quantum systems at the boundary between a topological and a trivial phase. The Lindblad dynamics can be mapped to a hopping problem on a two-dimensional operator lattice, and in the topological phase, the winding number of the operator-space Hamiltonian is non-trivial. This winding number forbids convergence to a steady state. The system must oscillate — not because anything is driving it, but because the topology of the operator space has no fixed point. The oscillation is a topological boundary time crystal: it exists at the interface, it persists regardless of initial conditions, and no local perturbation can remove it because the obstruction is global.
Caupin et al. (2026, arXiv:2602.18785) observe phase flipping in confined fluids. In volumes small enough that the free energy barrier between “bubble present” and “bubble absent” is comparable to thermal fluctuations, the system oscillates between states rather than settling into phase coexistence. At macroscopic scale, phase coexistence is a stable equilibrium. At nanoscale, it's an oscillation. The system oscillates because it's too small to stabilize — the steady state requires more degrees of freedom than the system possesses. Convergence is a property of size, not dynamics.
Provata et al. (2026, arXiv:2602.18198) study FitzHugh-Nagumo networks with Hebb-Oja adaptive coupling. When the coupling rules evolve on a timescale much slower than the individual oscillators, the network spontaneously produces traveling waves, synchronized states, and chimera states — spatial coexistence of order and disorder. The oscillation here is maintained by the adaptation itself. The coupling strengths change, the network topology shifts, and the system never reaches a fixed point because the landscape it's searching keeps being modified by the search. The steady state doesn't exist because pursuing it changes the target.
Voits and Schwarz (2026, arXiv:2602.18265) prove that first-passage time distributions in large Markov networks converge to exactly two universal forms: exponential (generic) or delta-function (deterministic). Everything in between — including oscillatory or peaked distributions — exists only at finite network size. Intermediate dynamics, including oscillation, require specific structural features that prevent the network from reaching either universal limit. In a large generic network, you get exponential (memoryless). Oscillation at large scale requires structure that maintains memory against the thermodynamic pressure toward forgetfulness.
Four mechanisms, one question: why doesn't the system converge? Nemeth: it can't. Topology forbids it. Caupin: it's too small. The steady state needs more room. Provata: it's chasing itself. Adaptation moves the target. Voits-Schwarz: it has structure. Structure maintains memory. The first is a hard topological obstruction — the oscillation is protected by a winding number and no smooth deformation can remove it. The second is a finite-size effect — the oscillation would stop if the system were larger. The third is a self-referential loop — the oscillation is the adaptation observing its own effects. The fourth is a structural privilege — the oscillation survives because specific network features resist generic convergence. These are not the same thing. Topological protection is robust to perturbation; finite-size oscillation is destroyed by growth; adaptive oscillation is maintained by the very process that could in principle converge; structural oscillation is fragile to the specific features being disrupted. The practical consequence: when you observe oscillation in a system, the mechanism that prevents convergence determines everything about the oscillation's fate. If it's topological, no parameter tuning will eliminate it. If it's finite-size, scaling up will. If it's adaptive, stopping the adaptation will. If it's structural, disrupting the right feature will. Same oscillation, different causes, different futures.