Feller-Markov Setting Fundamentals

• Feller-Markov setting is a foundational framework in Markov process theory that employs Feller semigroups on C₀(E) to rigorously define stochastic processes.
• It integrates spatially inhomogeneous dynamics through state-dependent mixing of Lévy processes, ensuring unique process generation and convergence.
• The framework leverages the Hille–Yosida theorem and practical simulation schemes, making it essential for applications in physics, finance, and biology.

The Feller-Markov setting refers to a foundational framework in the modern theory of Markov processes, characterized by semigroups of operators acting on spaces of continuous functions with specified continuity and contractivity properties. This framework, initially designed to encompass and generalize diffusive processes like Brownian motion and spatially-homogeneous Lévy processes, has developed into a mathematical infrastructure capable of incorporating spatial inhomogeneities, nonlocal jump dynamics, nonlinearities, and a wide array of applications in stochastic modeling and analysis (Böttcher, 2010).

1. Feller Semigroups and Processes: Definition and Structure

A Feller process is defined on a locally compact, separable metric space $E$, with its Banach space $C_0(E)$ of real-valued continuous functions vanishing at infinity and the supremum norm $\| \cdot \|_\infty$ (Böttcher, 2010). A family $\{T_t\}_{t \geq 0}$ of linear operators on $C_0(E)$ is called a Feller semigroup if, for all $t, s \geq 0$ and $f \in C_0(E)$:

1. Positivity: $T_tf \geq 0$ whenever $f \geq 0$.
2. Contractivity: $\|T_tf\|_\infty \leq \|f\|_\infty$.
3. Semigroup: $T_0 = I$, $T_{t+s} = T_t T_s$.
4. Strong continuity: $\lim_{t \rightarrow 0} \|T_t f - f\|_\infty = 0$ for all $f \in C_0(E)$.

If there exists a Markov process $(X_t)_{t \geq 0}$ with càdlàg paths such that $T_tf(x) = \mathbb{E}^x[f(X_t)]$, this process is called a Feller process. The generator $A$ is specified on its domain $D(A) \subset C_0(E)$ by

$A f := \lim_{t \to 0^+} \frac{T_t f - f}{t}.$

The Hille–Yosida theorem states that a densely defined linear operator $A$ generates a strongly continuous contraction semigroup on $C_0(E)$ if and only if $A$ is dissipative and the range of $\lambda I - A$ is $C_0(E)$ for all $\lambda > 0$.

This characterization encompasses Brownian motion, Lévy processes, and a broad array of spatially inhomogeneous processes (Böttcher, 2010).

2. General Construction: State-Dependent Mixing of Lévy Processes

Empirical data in many applications suggests the necessity for spatially inhomogeneous Markov processes, leading to the construction of Feller processes via state-space dependent mixing of Lévy processes (Böttcher, 2010). For each $x \in E$, let $\psi_x(\xi)$ be the characteristic exponent of a Lévy process $L^{(x)}$, and define the symbol $q(x, \xi) = \psi_x(\xi)$.

The generator acts on $f \in C_c^\infty(\mathbb{R}^d)$ via a pseudo-differential representation: $$A f(x) = -\int_{\mathbb{R}^d} e^{i x \cdot \xi} q(x, \xi) \widehat{f}(\xi) d\xi,$$ with an equivalent integro-differential representation: $$A f(x) = b(x) \cdot \nabla f(x) + \frac{1}{2} \operatorname{Tr}(c(x) \nabla^2 f(x)) + \int_{y \neq 0} [f(x+y) - f(x) - 1_{|y| < 1} y \cdot \nabla f(x)] \nu(x, dy).$$ Here, $(b(x), c(x), \nu(x, \cdot))$ are the Lévy–Khintchine characteristics associated with $\psi_x$.

Existence and uniqueness are established under conditions such as uniform growth bound $|\psi_x(\xi)| \leq C(1 + |\xi|^2)$ and uniform Lipschitz continuity in $x$. The operator $A$ is shown to be well-defined and dissipative, and the range condition can be verified by perturbation or martingale problem methods. This ensures a unique (in law) Feller process is generated (Böttcher, 2010).

3. Simulation and Sample Path Generation Schemes

A practical simulation technique, critical for applications including Monte Carlo methods, is provided by the stepwise-mixing Euler-type scheme (Böttcher, 2010). Given a time step $h > 0$ and initial state $x_0$, an approximate path $(X_{nh})_{n=0,\dots,N}$ is generated as follows:

• For $n=0,1,\ldots,N-1$:
  • At state $x_n$, sample an increment $Z_n$ from the law of $L^{(x_n)}_h$.
  • Set $x_{n+1} = x_n + Z_n$.

The pseudocode implements this recursion, and under the standing regularity conditions the Euler scheme converges in distribution (in Skorokhod space) to the true Feller process as $h \to 0$ (Böttcher, 2010).

4. Illustrative Model Classes

The general Feller–Markov setting encompasses a spectrum of model behaviors via state-dependent symbols:

• Brownian–Poisson–Cauchy mixture: For Lipschitz $a_1,a_2,a_3$ with $a_1(x)+a_2(x)+a_3(x)=1$, set

$$\psi_x(\xi) = a_1(x) \cdot \frac{1}{2}|\xi|^2 + a_2(x)(1 - e^{-i\xi}) + a_3(x)|\xi|.$$

This interpolates between Brownian, Poisson, and Cauchy behavior depending on $x$.

• Symmetric stable-like: For Lipschitz $\alpha(x):E\to(0,2]$ bounded away from $0,2$, set $\psi_x(\xi)=|\xi|^{\alpha(x)}$.
• Normal-inverse-Gaussian-like: With smooth bounded $\alpha(x) > |\beta(x)| > 0, \delta(x) > 0$, let

$$\psi_x(\xi)=-i\mu(x)\xi + \delta(x)\left(\sqrt{\alpha(x)^2-(\beta(x)+i\xi)^2} - \sqrt{\alpha(x)^2-\beta(x)^2}\right).$$

Parameter choices (e.g., $\beta(x) = - \frac{1}{\pi} \arctan(x)$) can induce spatial mean-reversion.

These and other constructions provide a flexible modeling framework for spatially inhomogeneous Markov processes (Böttcher, 2010).

5. Analytic and Practical Implications

The Feller–Markov framework unifies diffusion, jump, and mixed processes under the scheme of strongly continuous contraction semigroups on $C_0(E)$. The analytic foundation (generators, semigroup approach, Hille–Yosida) ensures rigor, while the explicit simulation algorithm bridges to practical computation and numerical analysis. The combination of existence, uniqueness, and straightforward sample path generation makes the Feller–Markov setting robust for applications in physics, biology, finance, and beyond (Böttcher, 2010).

6. Limitations and Extensions

While many properties of Feller processes are well-understood, proving existence or providing explicit constructions can be highly nontrivial and technical. The presented construction method, grounded in state-dependent Lévy mixing, overcomes some of these obstacles. The Feller–Markov setting is flexible enough to accommodate spatial inhomogeneities—unlike classical Lévy processes—by permitting the defining symbol to depend on the state in a Lipschitz and bounded way. A plausible implication is that the setting could be generalized to time-inhomogeneous, nonlinear, or interacting systems, provided the generator's range condition and dissipativity can be verified at the analytic level (Böttcher, 2010).

---

In conclusion, the Feller–Markov setting offers a rigorous, analytic, and computationally effective paradigm for spatially inhomogeneous jump-diffusion processes, enabling both mathematical analysis and applied stochastic modeling across a range of domains (Böttcher, 2010).