December 11th, 2007, 3:39 pm
I will make an informed guess. Suppose we model the failure of something (say a device or the death of a person) asa Poisson process with a time-dependent intensity r(t). Roughly, this means that at time t the prob of surving forthe next Delta_t is 1 - r(t) (Delta_t) + O(Delta_t^2). Also, it means that the probability of surving for the next instantis independent of having survived so far. Given this independence assumption, the probability of surving over (0,T) is P(survival) = [1 - r(t_1) (Delta_t)] x [ 1 - r(t_2) (Delta_t)] x ... [1 - r(t_N) (Delta_t] + O((Delta_t)^2)where Delta_t = T/N, and the {t_i} are a partition of (0,T).Now just exponentiate, use that log (1 - r(t_i) Delta_t ) = - r(t_i) Delta_t + O((Delta_t)^2), and collect together all the leading terms. You should be able to recognize that the leading term in the exponential is a sum that tends to an integral in the limt N -> infinity, Delta_t -> 0, N Delta_t = T. You should also be able to show that the remainingterms vanish in this limit. The result is that: P(survival) = exp(- int_0^T r(t) dt)regards,
Last edited by
Alan on December 10th, 2007, 11:00 pm, edited 1 time in total.