------------------------------------------------------------------------------- B O S T O N U N I V E R S I T Y Computer Science Department C O L L O Q U I U M Thursday December 22, 1994 3:00pm (Coffee served at 2:30pm) Seminar Room / MCS 135 ------------------------------------------------------------------------------- On Tails in 1-Dimensional Harnesses Andrei Toom Incarnate Word College As it is typical of 1-dimensional random processes with local interaction, components, indexed by integer numbers, interact in a local uniform way. In the present case components' states are real random variables a_s^t, where s is in Z and t=0,1,2,3,..., and the interaction is linear. A symmetric random i.i.d. `noise' \nu is added to every component at every time step. Two parameters of any symmetric random variable \xi are used: power decay or P-decay P-decay(\xi) = \lim_{x\to\infty} (-\log_x Prob(\xi > x)) and exponential decay or E-decay \hbox{E-decay}(\xi) = \lim_{x\to\infty} \log_x(-\ln Prob(\xi > x)). (For example, the P-decay of a Cauchy distribution equals 1 and the E-decay of a normal distribution equals 2.) We prove that the limits (as t goes to infinity) of the differences a_s^t - a_{s-1}^t display at least three different modes of behavior depending on the noise: a) if P-decay(\nu) < 3/2, these differences diverge and if P-decay}(\nu) > 3/2, they converge; b) if 1 < E-decay}(\nu) \leq 3, these differences converge to a distribution, whose E-decay equals E-decay}(\nu); however, if 3 \leq E-decay(\nu), they converge to a distribution, whose E-decay equals 3. Host: Prof. Peter Gacs ------------------------------------------------------------------------------- The Spring colloquium series will begin on January 25, 1995. For colloquium info, including directions, see http://cs-www.bu.edu/colloquium For more information contact Prof. Mark Crovella -------------------------------------------------------------------------------