Variable-Order Stable Subordinator

• Variable-Order Stable Subordinator (VOSS) is a time-inhomogeneous process defined by a variable stability index α(t) that generalizes classical α-stable subordinators with non-stationary, independent increments.
• Its analytical characterization features a Laplace transform expressed as exp(-∫₀ᵗ u^(α(s)) ds) and an inhomogeneous Lévy measure that adapts to local variations in jump behavior.
• VOSS underpins advanced modeling applications including time-changed Poisson processes, anomalous diffusion in heterogeneous media, and financial models with heavy-tailed jump dynamics.

A variable-order stable subordinator (VOSS), also known as a multistable subordinator, is a real-valued, non-decreasing, right-continuous process characterized by independent but typically non-stationary increments, where the stable index $\alpha(\cdot): [0, \infty) \rightarrow (0,1)$ is a time-dependent function. These processes generalize classical $\alpha$-stable subordinators, allowing the heavy-tailed nature of jumps and their scaling properties to vary over time. The VOSS is fundamental in the study of time-inhomogeneous jump processes, fractional calculus with variable order, and stochastic modeling of anomalous transport in heterogeneous or evolving media. Recent work has established thorough analytical characterizations, sample path properties, limiting behaviors, and applications ranging from random time-changes in Poisson processes to financial modeling with heavy-tailed increments (Singh et al., 11 Jan 2026, Orsingher et al., 2015, Molchanov et al., 2014, Beghin et al., 2016).

1. Rigorous Definition and Laplace Transform

Let $\alpha(t) \in (0,1)$ be right-continuous (often assumed piecewise constant or continuous). The VOSS $S^{\alpha(\cdot)}(t)$ is defined via the following concatenation of independent classical stable subordinators (Definition 2.1 in (Singh et al., 11 Jan 2026)): For a finite partition $0 = t_0 < t_1 < \cdots < t_n = T$ with

$$\alpha(t) = \alpha_k, \quad t_{k-1} \leq t < t_k, \quad k=1,\dots,n,$$

the process is

$$S^{\alpha(\cdot)}(t) = \sum_{j=1}^{k-1} S^{\alpha_j}(t_j - t_{j-1}) + S^{\alpha_k}(t - t_{k-1}), \quad t \in [t_{k-1}, t_k).$$

Each $S^{\alpha_j}$ is a classical stable subordinator of index $\alpha_j$.

The Laplace transform is given by

$$\mathbb{E}[e^{-u S^{\alpha(\cdot)}(t)}] = \exp\left(-\int_0^t u^{\alpha(s)} ds\right), \qquad u \geq 0,$$

generalizing the classical formula $e^{-t u^{\alpha}}$ for constant index $\alpha$ (Singh et al., 11 Jan 2026, Orsingher et al., 2015, Molchanov et al., 2014). Increments over arbitrary intervals $[s, t]$ are independent, with Laplace transform

$$\mathbb{E}[e^{-u(S^{\alpha(\cdot)}(t) - S^{\alpha(\cdot)}(s))}] = \exp\left(-\int_s^t u^{\alpha(r)} dr\right).$$

2. Lévy Measure, Path Properties, and Local Scaling

The Lévy kernel for $S^{\alpha(\cdot)}(t)$ is inhomogeneous in time, with density

$$\nu(ds, dx) = \frac{\alpha(s)}{\Gamma(1-\alpha(s))} x^{-1-\alpha(s)} dx\, ds, \quad s \in [0, t],\, x > 0,$$

so the law of jump sizes adapts to local variations in $\alpha(s)$ (Singh et al., 11 Jan 2026, Orsingher et al., 2015, Molchanov et al., 2014, Beghin et al., 2016).

Sample-path properties include:

• Pure-jump, non-decreasing, càdlàg trajectories.
• Infinitely many small jumps on any interval for $\alpha(\cdot) < 1$.
• Independent increments, but increments are non-stationary except on intervals of constant $\alpha$.
• Lack of global self-similarity; however, for $t$ fixed with smooth $\alpha(\cdot)$, the process is locally $\alpha(t)$-self-similar: $$\frac{S^{\alpha(\cdot)}(t + r \tau) - S^{\alpha(\cdot)}(t)}{r^{1/\alpha(t)}} \Longrightarrow S^{\alpha(t)}(\tau) \text{ as } r \to 0^+.$$
• For $\alpha(t) \equiv \alpha$ (constant), one recovers the strictly stable subordinator with stationary increments and classical scaling (Singh et al., 11 Jan 2026, Molchanov et al., 2014).

3. Governing Equations and Analytical Characterization

Let $p(x, t)$ denote the probability density function of $S^{\alpha(\cdot)}(t)$. The transition law is expressed via Laplace inversion,

$$p(x, t) = \mathcal{L}^{-1}_{u \to x} \left[\exp\left(-\int_0^t u^{\alpha(s)} ds\right)\right](x),$$

but in general has no closed-form except for constant-index or special cases (Orsingher et al., 2015, Beghin et al., 2016).

The forward (Kolmogorov) equation is

$$\frac{\partial}{\partial t} p(x, t) = - D_x^{\alpha(t)} p(x, t), \qquad p(x, 0) = \delta_0(x),$$

where $D_x^{\alpha}$ denotes the one-sided Riemann–Liouville fractional derivative in $x$. In each constant-$\alpha$ block, this reduces to the space-fractional drift equation for stable subordinators (Singh et al., 11 Jan 2026, Orsingher et al., 2015, Beghin et al., 2016).

The backward equation for test function expectations $u(t) = \mathbb{E}_x[\phi(S^{\alpha(\cdot)}(t))]$ is governed by the non-stationary variable-order fractional generator: $$\frac{\partial}{\partial t} u(t) = -(-\tfrac{d}{dx})^{\alpha(t)} u(t),$$ with fractional powers defined via functional calculus.

For the semigroup theory, the two-parameter propagator $G(s, t)$ (propagating between $s$ and $t$) satisfies

$$\partial_t G(s, t)u = -(-A)^{\alpha(t)} G(s, t)u, \qquad G(s, s) = I,$$

for a $C_0$-semigroup $(T_w)_{w \geq 0}$ with generator $A$ (Orsingher et al., 2015).

4. Hitting Times, Inverse Processes, and Asymptotics

Let $$\tau_y = \inf\{t \geq 0: S^{\alpha(\cdot)}(t) > y\}$$ be the first passage (hitting) time of level $y>0$. In the constant-index case, the law of $\tau_y$ is the classical Mittag–Leffler distribution. Globally, for variable-order,

$$\mathbb{E}[e^{-s \tau_y}] = \exp\left(-\int_0^y s^{\alpha(u)} du\right),$$

which encodes the non-stationary scaling via the index function $\alpha(\cdot)$ (Singh et al., 11 Jan 2026).

The inverse process $E(t) = \inf\{s \geq 0: S^{\alpha(\cdot)}(s) > t\}$ admits a density $l(s, t)$ given by

$$l(s, t) = \int_0^t p(s, y) \frac{(t - y)^{-\alpha(s)}}{\Gamma(1-\alpha(s))} dy,$$

and solves a variable-order evolution equation featuring Riemann–Liouville derivatives and additional non-stationary convolution operators (Orsingher et al., 2015).

5. Time-Change Constructions and Applications

The VOSS underlies several classes of time-changed stochastic processes:

• Space-Fractional Poisson Process / Generalized Space-Fractional Poisson Process via VOSS (GSFPP-VO):

$\{N(S^{\alpha(\cdot)}(t))\}_{t \geq 0}$

is formed by time-changing a homogeneous Poisson process with a VOSS, producing a counting process with state probabilities satisfying variable-order fractional difference-differential equations (Singh et al., 11 Jan 2026, Beghin et al., 2016).

The probability generating function for GSFPP-VO is

$$\mathbb{E}[u^{N(S^{\alpha(\cdot)}(t))}] = \exp\left(-\int_0^t \lambda^{\alpha(s)} (1-u)^{\alpha(s)} ds\right)$$

for rate parameter $\lambda > 0$ (Singh et al., 11 Jan 2026).

• Multifractional Poisson Process: Defined by subordinating a Poisson process with the inverse VOSS, generalizing time-fractional Poisson processes to non-homogeneous evolution (Molchanov et al., 2014, Beghin et al., 2016).

Other applications include:

• Modeling anomalous diffusion in media with spatial or temporal heterogeneity, with $\alpha(t)$ representing variable diffusion exponents.
• Reliability theory and shock models with time-dependent heavy-tailed damage increments.
• Financial models involving activity time with variable jump-index distribution in subordinated asset price dynamics (Singh et al., 11 Jan 2026, Beghin et al., 2016).

6. Approximation Methods and Numerical Schemes

Construction of the VOSS can be approached via random sum approximations and continuous-time random walk (CTRW) limits:

• Series-scheme approximation: Approximating $S^{\alpha(\cdot)}(t)$ by sums of independent heavy-tailed random variables with index $\alpha(k/n)$ and scales chosen via regularly varying functions yields weak convergence to the VOSS in Skorokhod topology (Molchanov et al., 2014).
• Compound Poisson approximation: Simulating VOSS via compound Poisson processes with time-varying laws is justified and useful for numerical applications (Orsingher et al., 2015).

These schemes enable practical computation, simulation, and empirical analysis of variable-order effects in stochastic systems.

7. Special Cases and Connection to Classical Stable Subordinators

When $\alpha(t) \equiv \alpha$ constant, $S^{\alpha(\cdot)}(t)$ reduces to the classical strictly $\alpha$-stable subordinator:

• Stationary, independent increments.
• Exact self-similarity: $$S^{\alpha}(ct) \stackrel{d}{=} c^{1/\alpha}S^{\alpha}(t)$$.
• Well-known Laplace exponent, densities, scaling behaviors, and hitting-time distributions (Singh et al., 11 Jan 2026, Molchanov et al., 2014).

The variable-order case loses global self-similarity and stationarity, but locally retains stable-like characteristics dependent on the instantaneous index $\alpha(t)$.

Property	Variable-Order Stable Subordinator	Classical Stable Subordinator
Increment stationarity	Non-stationary (unless $\alpha(t)$ constant)	Stationary
Self-similarity	Local, with order $\alpha(t)$	Global ($c^{1/\alpha}$ scaling)
Laplace transform	$\exp\left(-\int_0^t u^{\alpha(s)}ds\right)$	$e^{-t u^\alpha}$
Lévy measure	Time-inhomogeneous	Homogeneous in time

• "Generalized Space-Fractional Poisson Process via Variable-Order Stable Subordinator" (Singh et al., 11 Jan 2026)
• "Time-inhomogeneous jump processes and variable order operators" (Orsingher et al., 2015)
• "Multifractional Poisson process, multistable subordinator and related limit theorems" (Molchanov et al., 2014)
• "Time-inhomogeneous fractional Poisson processes defined by the multistable subordinator" (Beghin et al., 2016)