continuous dependence on initial conditions

Continuous dependence asks a question about robustness: if you nudge the starting point a little, does the whole solution move only a little? It would be unsettling if a hair's-width change in the initial value produced a wildly different curve on every interval; we want small causes to have proportionally small effects, at least over a fixed finite stretch of time. This is the property that lets a model be trusted when the initial data are only known approximately, which is always.

Under the same hypotheses that give existence and uniqueness — a Lipschitz condition on f — this holds in a precise, quantitative form. If two solutions start a distance d apart, the gap between them at later 'time' x is bounded by d times e^(L|x - x0|), where L is the Lipschitz constant. The exponential factor is the honest part: the curves may drift apart, but no faster than this controlled exponential rate, so on any fixed interval the final gap stays small when the starting gap is small. This estimate is essentially Gronwall's inequality in action, and the same argument extends to dependence on parameters inside the equation, not just on the initial value.

The honest caveat lives in that exponential. Continuous dependence is a statement about a fixed finite interval; it does not promise that nearby trajectories stay near forever. Over long times the e^(L|x|) bound can be enormous, and in chaotic systems tiny initial differences really do amplify until predictions diverge completely — that is sensitive dependence, the engine of chaos. Continuous dependence and sensitive dependence are not in conflict: the solution depends continuously on its data at each fixed horizon, yet the constant of continuity can blow up as the horizon grows.