Exponential distribution
An exponential distribution is a continuous probability model whose unshifted form describes nonnegative lifetimes or waiting times with constant hazard.
For the unshifted model with rate lambda > 0 and elapsed time t >= 0, S(t) = exp(-lambda t), F(t) = 1 - exp(-lambda t), and h(t) = lambda. Lambda has inverse-time units; lambda t is dimensionless. Rate lambda and scale 1/lambda are different parameter conventions. [1]
The mean is 1/lambda, but the median is ln(2)/lambda. Substituting the mean into the CDF gives 1 - exp(-1), not one half. Neither summary promises an individual service life. A shifted model additionally specifies a location parameter and a different time origin. [2]
For a >= 0 and u >= 0, the unshifted model gives P(T > a + u | T > a) = S(a + u)/S(a) = exp(-lambda u). This derived memoryless property concerns additional lifetime conditional on survival; it is not evidence that devices avoid wear. Two-parameter Weibull shape one gives this exponential case. [1] [3]
A homogeneous Poisson process requires independent, identically distributed exponential interarrival times. Exponential individual lifetimes alone establish neither independence between devices nor a Poisson fleet-replacement stream. State the event, population and clock before transferring a rate. These are model distinctions, not a fitted storage-fleet result. [4]
For a fixed initial cohort of N independent items with common exponential rate lambda, each counted only at its first failure and never replaced, the failure count by t is binomial with probability 1 - exp(-lambda t). Its expected count is N(1 - exp(-lambda t)), so its expected event rate is N lambda exp(-lambda t), not a constant. This is an analytic deduction from the lifetime CDF under the stated cohort assumptions, not a fitted fleet model. [1]
Relations
- related
- Weibull distribution, Survival function, Hazard rate, Mean residual life, Poisson rate model, Interarrival time