Characterization of purely nondeterministic discrete parameter $L^2$-processes

30 Views Asked by At

In the characterization of purely nondeterministic processes, more precisely at proposition 2.2.7 of "Topics in Stochastic Processes", Ash, Gardner et al., after having defined, given $\{X(t), t \in T\}$ a stationary $L^2$ process, $$ \begin{aligned} H(X) & =L^2\{X(t), t \in T\} \\ H_t(X) & =L^2\{X(s), s \leq t\}, \quad t \in T \\ H_{-\infty}(X) & =\bigcap_{t \in T} H_t(X) . \end{aligned} $$ and having said that if $H_{-\infty}(X)=H(X)$ or, equivalently, $H_t(X)$ is the same for all $t$, we call $\{X(t), t \in T\}$ deterministic, nondeterministic if $H_{-\infty}(X) \varsubsetneqq H(X)$ and purely nondeterministic if $H_{-\infty}(X)=\{0\}$, the following is claimed:

Assume $\{X(n)\}$ purely nondeterministic. Write $H_n(X)=$ $H_{n-1}(X) \oplus D_n$; since $H_{n-1}(X) \cup\{a X(n): a \in C\}$ spans $H_n(X), \operatorname{dim} D_n \leq 1$. If $\operatorname{dim} D_n=0$, then $H_n(X)=H_{n-1}(X)$, so that $H_n(X)=H(X)$ for all $n$, and thus $\{X(n)\}$ is deterministic, a contradiction. Therefore $\operatorname{dim} D_n=1$ for all $n$.

It seems to me, however, that from this it is only possible to conclude that $\exists n: \text{dim }D_n=1$, and that we could indeed have $\text{dim }D_m=0$ for some m even if the process is nondeterministic. I figured that I am probably supposed to assume that $\{X(n)\}$ is stationary, so that I could say that $D_n=<X_{n+1}-P_{n}X_{n+1}>$ and $||X_{n+1}-P_{n}X_{n+1}||$ does not depend on $n$ by stationarity and is greater than $0$, otherwise the process would be deterministic, hence $\text{dim }D_n=1 \,\,\forall n$; I have however no idea on how it could be done without this assumption. Any hint would be greatly appreciated.