Home Statistical Dictionary Renewal Process

Renewal Process

A counting process, like the Poisson process, but generalized to allow the gaps between consecutive events to follow any probability distribution, not just the exponential distribution.

Stochastic ProcessesIntermediateFree

In Plain English

A renewal process asks what happens if you take the Poisson process's basic structure, counting events as they arrive one after another, but drop the requirement that the waiting time between events has to be exponentially distributed. Light bulbs are the classic teaching example: a bulb burns out, gets instantly 'renewed' (replaced), burns out again, and so on, and there's no reason a bulb's lifetime has to follow the memoryless exponential distribution specifically, it might follow a Weibull distribution reflecting wear-and-tear instead. Every time a renewal happens, the process probabilistically resets completely, this fresh-start property, called regeneration, is what gives the model its name and its mathematical tractability.

Definition

A renewal process is a counting process ${N (t)}_{t \geq 0}$ built from i.i.d. non-negative interarrival times $T_{1}, T_{2}, \dots$ with common distribution $F$ (not necessarily exponential), where $N (t) = max {n : S_{n} \leq t}$ and $S_{n} = T_{1} + \dots + T_{n}$ is the time of the $n$ -th renewal. When $F$ is exponential, a renewal process reduces exactly to a Poisson process; more general choices of $F$ (gamma, Weibull, or any non-negative distribution) produce renewal processes with correlated, non-Poisson increment structure. Because each renewal 'restarts the clock' with an identical i.i.d. future, a renewal process is the prototypical example of a regenerative process. Key asymptotic results, collectively the renewal theorems, describe the long-run behavior of $N (t)$ and the expected number of renewals in an interval as $t \to \infty$ .

Formula

N (t) = max {n : S_{n} \leq t}, S_{n} = T_{1} + T_{2} + \dots + T_{n}

Notation

T_{i}

—The $i$-th interarrival (interrenewal) time, i.i.d. from a general distribution $F$

S_{n}

—Time of the $n$-th renewal, the sum of the first $n$ interarrival times

N (t)

—Number of renewals that have occurred by time $t$

Properties

Losing the exponential distribution's memorylessness is the whole point of the generalization, and the cost: $N (t)$ is no longer Markov in continuous time for a general renewal process (only the sequence of renewal epochs itself has a clean recursive structure), which is exactly the tradeoff a semi-Markov process makes deliberately at the level of an entire state space.
Reliability engineering and maintenance planning are the classic application domains, modeling equipment replacement or repair cycles where failure times follow a realistic distribution (Weibull, gamma) rather than being forced into the exponential distribution's constant-hazard assumption.
The renewal-reward theorem extends the basic counting framework by attaching a random 'reward' to each renewal cycle (e.g. cost incurred, revenue earned), giving a powerful and widely used tool for computing long-run average rates in queueing systems, inventory models, and maintenance-cost analysis.

At a Glance

Generalizesthe Poisson process, by allowing any interarrival distribution, not just exponential

Poisson process as special casearises exactly when interarrival times are exponential

Regenerationeach renewal probabilistically restarts the process from scratch

Classic examplelight bulb replacement, with lifetimes following any chosen distribution

Last updated September 8, 2026← Back to the Dictionary