Adiabatic invariants in mechanics

For an oscillator with a smoothly varying positive frequency, a useful local small parameter is the magnitude of its frequency rate divided by frequency squared. It compares the fractional parameter change over a typical oscillation with unity, up to the usual two-pi convention in the cycle duration. The condition should hold throughout the interval, not merely at its endpoints. Abrupt switching, sharp transients, or a frequency approaching zero can defeat an apparently long overall adjustment time.

Averaging over the fast phase suppresses the leading net action drift under suitable regularity and scale-separation assumptions. The action can still have small phase-dependent oscillations, and accumulated errors depend on the time interval and smoothness of the protocol. Calling J an adiabatic invariant therefore does not assert that its derivative is exactly zero at every instant. State both the approximation and the orbit family when using it to predict a measurable change.

Keep mass fixed and increase the oscillator frequency slowly. At approximately fixed J, energy increases in proportion to frequency, while amplitude decreases as its inverse square root. There is no contradiction: higher frequency raises the stiffness, so a smaller displacement can store more energy. The work comes from changing the parameter. Conservation of energy applies to an appropriately enlarged closed system, not automatically to the oscillator subsystem whose Hamiltonian depends explicitly on time.

The momentum amplitude behaves oppositely to the position amplitude: it grows as the square root of frequency. Thus the phase-space ellipse becomes narrower and taller while keeping its enclosed area approximately fixed. This picture links adiabatic invariance to canonical area, but it must not be used as a proof that every area-preserving deformation keeps each orbit's action fixed. The result relies on slow evolution within the specified regular family and the averaging over its rapid phase.

If frequency changes instantaneously, position and momentum can remain continuous while the Hamiltonian value jumps. The new action depends on the phase at which the change occurred. At a position maximum, raising stiffness changes the potential energy immediately; at a zero crossing, the same change initially leaves the kinetic energy unchanged. That phase dependence is precisely what slow averaging avoids. An animation slider that rebuilds a solution with fresh initial data is neither of these physical protocols unless it explicitly models them.

Near a separatrix, an orbit's period can become arbitrarily long. A parameter variation that was slow compared with ordinary oscillations is then no longer slow compared with that orbit, even if the external adjustment rate remains small in laboratory units. Action changes at separatrix crossings require their own analysis. This limitation becomes important in resonance capture, slowly driven nonlinear systems, and magnetic confinement; a familiar invariant should not be extrapolated into a region where its underlying fast cycle has disappeared.