Renewal Theory — Mathematics Knowledge System

Overview

Renewal theory generalizes the Poisson process by allowing inter-renewal times to have an arbitrary distribution F (with finite mean μ = E[T]) rather than just the exponential. A renewal process N(t) counts the number of renewals in [0,t], where renewal times Sₙ = T₁ + ... + Tₙ are partial sums of i.i.d. non-negative random variables. Key results include the elementary renewal theorem (E[N(t)]/t → 1/μ), Blackwell's theorem on renewal spacing, and the renewal reward theorem, which is a central tool for computing long-run averages in stochastic systems.

Intuition

Imagine replacing light bulbs as they burn out. Each bulb has lifetime drawn from some distribution F. The renewal process counts how many bulbs you have gone through by time t. The key insight is universality: regardless of the lifetime distribution, the long-run replacement rate is always 1/μ (one per mean lifetime). The renewal reward theorem extends this: if you earn a reward R during each cycle, the long-run reward rate is E[R]/E[T]. This is remarkably general and underpins analysis of Markov chains, queues, and reliability systems.

Formal Definition

Definition

Let T₁, T₂, ... be i.i.d. non-negative random variables with distribution F and mean μ < ∞. The renewal process and its key quantities are defined as follows.

S_n = T_1 + \cdots + T_n, \quad N(t) = \max\{n : S_n \le t\}

Renewal counting process

m(t) = \mathbb{E}[N(t)] = \sum_{n=1}^{\infty} F^{*n}(t)

Renewal function (F^{*n} = n-fold convolution of F)

m(t) = F(t) + \int_0^t m(t-s)\,dF(s)

Renewal equation

\frac{m(t)}{t} \to \frac{1}{\mu} \quad \text{as } t \to \infty

Elementary renewal theorem

Notation

Notation	Meaning
$N(t)$	Number of renewals in [0, t]
$m(t)$	Renewal function: E[N(t)]
$\mu$	Mean inter-renewal time E[T]
$F^{*n}$	n-fold convolution of F: distribution of Sₙ

Theorems

Theorem 1: Theorem 1

\text{ Let } R_n \text{ be the reward earned during the } n\text{ -th cycle, with } \mathbb{ E }[R_n] < \infty. \text{ Then the long-run average reward satisfies}\quad \frac{1}{t}\int_0^t R(s)\,ds \to \frac{\mathbb{E}[R]}{\mathbb{E}[T]} \quad \text{a.s. as } t \to \infty.

Theorem 2: Theorem 2

\text{If } F \text{ is non-lattice, then } m(t+h) - m(t) \to h/\mu \text{ as } t \to \infty \text{ for every } h > 0. \text{ If } F \text{ is lattice with span } d, \text{ then } P(\text{renewal at } nd) \to d/\mu.

Theorem 3: Theorem 3

\text{For a renewal process with finite mean inter-renewal time } \mu = \mathbb{E}[T]{:} \quad \frac{N(t)}{t} \to \frac{1}{\mu} \quad \text{a.s.} \quad \text{and} \quad \frac{m(t)}{t} \to \frac{1}{\mu} \quad \text{as } t \to \infty.

Worked Examples

1
Mean lifetime μ = (1+3)/2 = 2 years.
$\mu = \mathbb{E}[T] = \frac{1+3}{2} = 2$
2
By the elementary renewal theorem, E[N(t)] ≈ t/μ = 10/2 = 5.
$m(10) \approx \frac{t}{\mu} = \frac{10}{2} = 5$

✓ Answer

Approximately 5 replacements in 10 years.

Practice Problems

Mediumapplication

Light bulbs have exponential lifetimes with mean 100 hours. Approximately how many spares should you stock for a 1000-hour period so that stock-outs are rare?

Mediumfree response

Derive the renewal equation m(t) = F(t) + ∫₀ᵗ m(t-s) dF(s) by conditioning on the first renewal time T₁.

Common Mistakes

Common Mistake

Applying the elementary renewal theorem when the mean inter-renewal time is infinite

The theorem m(t)/t -> 1/mu requires mu = E[T] < infinity. For heavy-tailed distributions with infinite mean, the rate is 0.

Common Mistake

Forgetting the fresh-start property when deriving the renewal equation

After a renewal at time s, the process restarts independently: the future is a fresh copy of the original process, which gives the renewal equation via conditioning on T_1.

Common Mistake

Confusing the renewal function m(t) with the random variable N(t)

m(t) = E[N(t)] is a deterministic function; N(t) is a random variable. For large t, m(t) is approximately t/mu but N(t) fluctuates around this value.

Quiz

Historical Background

Renewal theory originated in industrial reliability and actuarial science during the early 20th century, motivated by questions about replacing worn-out components. The mathematical framework was developed by Feller, Blackwell, and Smith in the 1940s-1950s. Feller's 1941 paper on the law of large numbers for renewal processes and Blackwell's 1948 renewal theorem established the field's foundations. The renewal reward theorem, crucial for steady-state analysis, was formalized by Cox in the 1960s.

1941
Feller establishes the law of large numbers for renewal processes
William Feller
1948
Blackwell proves his renewal theorem on the asymptotic density of renewals
David Blackwell
1960s
Cox develops the renewal reward theorem for long-run average rewards
David Cox

Summary

A renewal process generalizes the Poisson process: inter-renewal times are i.i.d. with arbitrary distribution F and mean μ.
The renewal function m(t) = E[N(t)] satisfies the renewal equation m(t) = F(t) + ∫₀ᵗ m(t-s) dF(s).
The elementary renewal theorem: N(t)/t → 1/μ a.s.; the long-run renewal rate equals the reciprocal of the mean cycle length.
The renewal reward theorem: if rewards Rₙ are earned in each cycle, the long-run rate is E[R]/E[T], a central tool for steady-state analysis.

References

BookRoss, S. — Introduction to Probability Models, Chapter 7
BookDurrett, R. — Probability: Theory and Examples, 4th ed.
WebsiteWikipedia — Renewal theory