What makes a process regenerative?

It has random times at which it restarts independently of its history, so the pieces between these regeneration epochs are independent and identically distributed cycles.

Why are regenerative processes useful in simulation?

Because the cycles are independent, averages over cycles behave like independent samples, allowing valid confidence intervals for steady-state quantities without assuming a particular distribution.

Regenerative Processes

A regenerative process contains random times at which it starts over independently of its past, breaking its evolution into independent and identically distributed cycles.

Definition

A regenerative process is a stochastic process possessing random regeneration epochs, forming a renewal process, such that the segments between consecutive epochs are independent and identically distributed, so the process probabilistically restarts at each epoch.

Scope

This topic covers regeneration epochs and cycles, the renewal-reward theorem expressing long-run averages as expected reward per cycle over expected cycle length, the existence of limiting time-stationary distributions, the regenerative method for steady-state simulation and confidence intervals, and the connection between regeneration and the renewal structure of Markov processes.

Core questions

What are regeneration epochs and how do they partition a process into independent cycles?
How does the renewal-reward theorem give long-run averages from a single cycle?
When does a regenerative process have a limiting distribution?
How is regeneration exploited for steady-state simulation and inference?

Key theories

Renewal-reward theorem: For a regenerative process the long-run average of a reward accrued over time equals the expected reward earned in one cycle divided by the expected length of a cycle, reducing time-average computations to a single regeneration cycle.
Limiting distribution of regenerative processes: When the cycle-length distribution is non-lattice and has finite mean, a regenerative process converges in distribution to a time-stationary law given by the expected occupation time per cycle, which establishes steady-state existence for many queues and Markov chains.

Clinical relevance

Regeneration provides a unifying way to prove steady-state results for queues, inventory systems, and Markov processes, and the regenerative method gives rigorous confidence intervals in discrete-event simulation by treating cycle averages as independent samples.

History

The regenerative viewpoint was articulated by Smith in the 1950s as an extension of renewal theory, and its application to steady-state simulation through the regenerative method was developed by Crane and Iglehart in the 1970s, becoming a standard tool in applied probability and performance analysis.

Key figures

Walter Smith
Soren Asmussen
Donald Iglehart

Seminal works

asmussen2003

Frequently asked questions

What makes a process regenerative?: It has random times at which it restarts independently of its history, so the pieces between these regeneration epochs are independent and identically distributed cycles.
Why are regenerative processes useful in simulation?: Because the cycles are independent, averages over cycles behave like independent samples, allowing valid confidence intervals for steady-state quantities without assuming a particular distribution.