Infinitesimal generator for non-Markov process

122 Views Asked by At

I have read that the infinitesimal generator for a Markov process $(X_t)_{t \geq 0}$ we define the generator $A$ by

$$Af(x) := \lim_{t \downarrow 0} \frac{\mathbb{E}^x(f(X_t))-f(x)}{t} $$

whenever the limit exists in $(C_{\infty},\|\cdot\|_{\infty})$. The definition implies that $$\mathbb E[Af(X)]=0$$ if $X$ is the stationary distribution.

I want to know why the definition is limited to Markov processes and not defined for a general stochastic process?

1

There are 1 best solutions below

0
On

Just my guess, it's because Markov property makes the infinitesimal generator identify the entire process. This is surprising because the infinitesimal generator is defined only at the neighborhood of $s=0$.

By the definition, the infinitesimal generator is similar to the derivative in calculus, i.e. we want to introduce some derivative-like notion of stochastic process. A straightforward approach to develop it is to define it for all time $s\ge0$. This may be tentatively defined for all $s\ge0$ as: $$ A_sf(x) = \lim_{t\to0} \frac{ \mathbb{E}\left\lbrack f(X_{s+t}) \middle\vert \mathcal{F}_{s} \right\rbrack - f(x)}{t}, $$ where $X_s = x$ and $\mathcal{F}_s$ is the filtration generated by $X_s$.

However, if $\{X_s\}_{s\ge0}$ has the Markov property (and it leads the time-homogeneity), we only need to define it at (the neighborhood of) $s=0$ with the initial value $x$. In fact, for all $s\ge0$, $$ \begin{align} A_s(x) &:= \lim_{t\to0} \frac{ \mathbb{E}\left\lbrack f(X_{s+t}) \middle\vert \mathcal{F}_{s} \right\rbrack - f(x)}{t} \\ &= \lim_{t\to0} \frac{ \mathbb{E}\left\lbrack f(X_{s+t}) \middle\vert X_s = x \right\rbrack - f(x)}{t} \quad (\because \text{Markov property}) \\ &= \lim_{t\to0} \frac{ \mathbb{E}\left\lbrack f(X_t) \middle\vert X_0 = x \right\rbrack - f(x)}{t} \quad (\because \text{time-homogeneity}) \\ &= A_0f(x). \end{align} $$

In other word, if $\{X_s\}_{s\ge0}$ has the Markov property, the derivative-like notion at $t=0$, so called the infinitesimal generator of $\{X_s\}_{s\ge0}$, determines, identifies or generates the entire process with initial value $x$.