Alternating Renewal Process

• The alternating renewal process is a two-state stochastic model that alternates between distinct phases with independent sojourn time distributions.
• Renewal equations employing Laplace transforms allow precise analysis of cycle counts, occupation times, and first-passage problems within the process.
• Applications span queueing, reliability, environmental science, and finance, addressing both finite-mean and heavy-tailed, non-ergodic regimes.

An alternating renewal process is a two-state stochastic process in which the system alternates between distinct phases, each phase having independent and identically distributed (iid) sojourn times, with alternation governed by a renewal structure. This framework generalizes the classical renewal process by allowing for separate, potentially non-identical sojourn time distributions for each state, and arises in diverse domains including queueing, reliability, point process theory, first-passage diffusion, and stochastic modeling with regime-switching structure (Akimoto, 2023, Starreveld et al., 2016, Du et al., 2024, Crescenzo et al., 2021).

1. Formal Definition and Structure

In an alternating renewal process, the state space is partitioned as $E = A \cup B$, $A \cap B = \emptyset$, with the process $X(t)$ sojourning in $A$ and $B$ alternately. Let $D_1$ denote the first sojourn (residence) time in $A$, and $U_1$ the subsequent sojourn in $B$:

$$D_1 = \inf\{ t > 0 : X(t) \notin A \}, \qquad U_1 = \inf\{ t > D_1 : X(t) \in A \} - D_1$$

Successive cycles $A \cap B = \emptyset$0 for $A \cap B = \emptyset$1 are iid random vectors. If $A \cap B = \emptyset$2 and $A \cap B = \emptyset$3 are independent sequences, the process is called a classical alternating renewal process; otherwise, dependence between $A \cap B = \emptyset$4 and $A \cap B = \emptyset$5 is allowed (Starreveld et al., 2016, Akimoto, 2023).

The total number of completed cycles by time $A \cap B = \emptyset$6 is:

$A \cap B = \emptyset$7

Sojourn times in each phase may follow general distributions, including those with infinite means, heavy tails, or deterministic durations.

2. Renewal Equations and Transform Methods

Key statistical quantities—such as the renewal function, occupation times, and state probabilities—are accessible via Laplace or double Laplace–Stieltjes transforms, reflecting the renewal structure.

The Laplace transform of the mean number of renewals $A \cap B = \emptyset$8 for the ordinary process with residence time Laplace transforms $A \cap B = \emptyset$9 is:

$X(t)$0

In equilibrium (both means $X(t)$1), this simplifies to:

$X(t)$2

Occupation times in one state, e.g., the total time in $X(t)$3 up to $X(t)$4, follow:

$X(t)$5

With the double Laplace–Stieltjes transform ($X(t)$6):

$X(t)$7

where $X(t)$8 are Laplace transforms of sojourn and cycle lengths (Starreveld et al., 2016).

The general state probabilities and higher occupation time moments similarly admit transform representations, which can be numerically inverted for specific interval distributions.

3. Generalizations and Statistical Behavior

When one or both sojourn distributions exhibit heavy tails ($X(t)$9), the process reveals rich non-ergodic behavior. Define $A$0:

• If $A$1 (all means finite): The mean number of renewals grows linearly, $A$2, and the variance scales linearly or sublinearly depending on the decay exponent.
• If $A$3: Sublinear growth $A$4 and fluctuations demonstrate distributional limit theorems: the rescaled renewal count $A$5 converges to the Mittag–Leffler law; occupation time $A$6 converges in distribution to the generalized arcsine (Lamperti) law (Akimoto, 2023). These properties are characteristic of aging, i.e., statistics depend on the process’s "age."

Covariances, correlations, and two-point functions can be expressed via Laplace transforms, with exponential or algebraic decay depending on tail exponents.

4. On/Off (Doubly Stochastic Poisson) Process and Markov Modulations

Alternating renewal processes naturally arise in piecewise-constant intensity models. Consider a counting process with an instantaneous event rate switching between “on” ($A$7) and “off” ($A$8), governed by a semi-Markov process $A$9 with iid sojourns: “on” periods have Exp(k) length, “off” periods have law $B$0. The associated inter-renewal time distribution has Laplace–Stieltjes transform:

$B$1

This exactly characterizes renewal processes with an alternating on/off intensity representation. Special cases include two-state Markov-modulated Poisson processes (MMPPs). The full distribution and moments of the cycle and interarrival can be written explicitly in terms of the parameters $B$2, geometric and exponential random variables (Du et al., 2024).

5. Occupation Time Theory, Limit Theorems, and Large Deviations

Occupation time analysis centers on the functional $B$3:

• Mean: $B$4
• Variance: $B$5 with

$B$6

• Central Limit Theorem: As $B$7, the normalized occupation time converges to a normal distribution if variances are finite (Starreveld et al., 2016).
• Large Deviations: The process admits a large deviation principle with explicit rate function:

$B$8

These results encapsulate a unified view from renewal theory to non-Gaussian, heavy-tailed, and non-ergodic regimes (Akimoto, 2023, Starreveld et al., 2016).

6. First-Passage Problems and Simulation Methods

Alternating renewal processes are used to modulate parameters in diffusions, notably to study first-passage times (FPT) in systems with regime switching. Di Crescenzo et al. construct a model where the drift and variance of a Brownian motion alternate between two values according to a two-state renewal process (Crescenzo et al., 2021). The FPT law satisfies coupled renewal-type integral equations for the FPT density, from which:

• Lower and upper bounds are derived for the density and cumulative distribution.
• A Monte Carlo simulation algorithm samples the FPT via acceptance–rejection, conditioning on Wiener first-passage laws and the current regime. The method exploits the independent-increment structure of Brownian motion alongside the renewal structure of alternations.

Applications in environmental science model rewetting/drying (sojourns with inverse-Gaussian law) and in finance model asset prices with alternating bullish-bearish durations (sojourns with generalized Pareto law), both requiring accurate FPT statistics under regime-switching noise (Crescenzo et al., 2021).

7. Applications and Connections to Lévy Processes and Beyond

Alternating renewal processes underpin a variety of stochastic models:

• Lévy Processes Reflected at the Infimum: The time a reflected Lévy process spends below a threshold is encoded as occupation time in “on/stable” ($B$9) vs. “off/transient” ($D_1$0) states, each sojourn giving the $D_1$1 cycle. Explicit expressions for joint transforms and occupation times are provided via scale functions (Starreveld et al., 2016).
• Queueing, Reliability, and Point Processes: Alternating renewal models capture systems with alternating operational/failure or busy/idle phases.
• Generalizations: The framework also includes cases with dependent sojourns, heavy tails, and infinite mean durations, leading to anomalous fluctuation regimes and aging phenomena (Akimoto, 2023).

Applications range from stochastic networks, atmospheric sciences, to mathematical finance, wherever switching dynamics or alternating environmental conditions are fundamental.

---

References:

• (Akimoto, 2023) – K. Nakamura and S. Toyabe, "Statistics of the number of renewals, occupation times and correlation in ordinary, equilibrium and aging alternating renewal processes"
• (Starreveld et al., 2016) – N. Starreveld, R. Bekker and M. Mandjes, "Occupation times of alternating renewal processes with Lévy applications"
• (Du et al., 2024) – S. L. Hager, "Renewal Processes Represented as Doubly Stochastic Poisson Processes"
• (Crescenzo et al., 2021) – A. Di Crescenzo, E. Di Nardo, and L. M. Ricciardi, "Simulation of first-passage times for alternating Brownian motions"