A prehistory problem is formulated for large occasional fluctuations in noise-driven systems. It has been studied theoretically and experimentally, thereby illuminating the concept of optimal paths and making it possible to visualize and investigate them. The prehistory probability distribution measured for a white-noise-driven system, taken as an example, is shown to be in agreement with the theory. PACS numbers: 05.40.+J, 02.50,+s, 05.20.-y Fluctuations in physical systems can often be viewed [1] as arising because of external noise. Under stationary conditions, a weak-noise-driven system fluctuates mostly about its attractor (or attractors, if several of them coexist). However, there is also a small probability that the system will be found at a position in phase space far from an attractor. It is just these large deviations from the average that are responsible for a number of interesting physical phenomena, e.g., for switching in a variety of multistable systems (including multimode lasers, passive optically bistable systems, and Josephson junctions) and large-angle scattering (in particular, that of light) in nearly homogeneous media.A convenient and powerful approach to the analysis of the tails of the probability density distribution p(x) (where the components of the vector x enumerate the dynamical variables of a system) for systems driven by Gaussian noise is based [2-8] on the method of optimal fluctuation [9]. This approach exploits the idea that the tails of p(x) must be formed by large occasional outbursts of noise fit) that push the system far from the attractor. The probabilities of such large outbursts are small, and the value of pixf) for a given remote Xf will actually be determined by the probability of the most probable outburst among those bringing the system to x/. This particular realization is just the optimal fluctuation for the given Xf. Because a realization (a path) of noise fit) results in a corresponding realization of the dynamical variable xO), there also exists an optimal path x op{ (t;Xf) along which the system arrives at x/, with an overwhelming probability. Although eminently reasonable and highly successful, such approaches have lacked a direct basis in experiment-the existence of optimal paths never having been demonstrated-and, to this extent, the use of the method of optimal fluctuation has amounted to an act of faith.In this Letter, we propose a new approach to the investigation of rare events in noise-driven systems, addressing ourselves directly to the question of how one of these events (i.e., the arrival of the system at Xf) comes to

No abstract