1. Preliminaries
Definition 1 (Hilbert Space Contraction).
A bounded linear operator T:H1→H2 between Hilbert spaces is called a contraction if
‖
Equivalently, \|T\| \leq 1.
Definition 2 (Stationary Process).
A stochastic process \{Y(t)\}_{t \in \mathbb{R}} is stationary if for any finite set of time points \{t_1,\ldots,t_n\} and any h \in \mathbb{R}, the joint distribution of
\{Y(t_1+h),\ldots,Y(t_n+h)\}
is identical to that of \{Y(t_1),\ldots,Y(t_n)\}.
Definition 3 (Stationary Dilation).
Given a non-stationary process X(t), a stationary dilation is a stationary process Y(s) together with a family of bounded operators \{\phi(t,\cdot)\}_{t \in \mathbb{R}} such that
X(t) = \int_{\mathbb{R}} \phi(t,s)Y(s)ds
where \phi(t,s) is a measurable function satisfying:
- \|\phi(t,\cdot)\|_{\infty} \leq 1 for all t
- The map t \mapsto \phi(t,\cdot) is strongly continuous
Remark.
The conditions on \phi(t,s) ensure that the integral is well-defined and the resulting process X(t) inherits appropriate regularity properties from Y(s).
2. Main Results
Proposition 1 (Properties of Scaling Function).
The scaling function \phi(t,s) in a stationary dilation satisfies:
- \|\phi(t,s)\| \leq 1 for all t,s \in \mathbb{R}
- For fixed t, s \mapsto \phi(t,s) is measurable
- For fixed s, t \mapsto \phi(t,s) is continuous
Theorem 1 (Equivalence of Containment).
For a non-stationary process X(t) and a stationary process Y(s), the following are equivalent:
- Y(s) is a stationary dilation of X(t)
- There exists a contractive mapping \Phi from the space generated by Y to the space generated by X such that X(t) = (\Phi Y)(t) for all t
Proof.
(1 \Rightarrow 2): Define \Phi by (\Phi Y)(t) = \int_{\mathbb{R}} \phi(t,s)Y(s)ds For any finite linear combination \sum_i \alpha_i Y(t_i): \begin{align*} \|\Phi(\sum_i \alpha_i Y(t_i))\|^2 &= \|\sum_i \alpha_i \int_{\mathbb{R}} \phi(t_i,s)Y(s)ds\|^2 \\ &\leq \|\sum_i \alpha_i Y(t_i)\|^2 \end{align*} where the inequality follows from the bound on \|\phi(t,s)\| and the Cauchy-Schwarz inequality. (2 \Rightarrow 1): The contractive mapping \Phi induces a family of operators \phi(t,s) via the Kernel theorem for Hilbert spaces. The stationarity of Y and the contractivity of \Phi ensure that these operators satisfy the required properties.
(1 \Rightarrow 2): Define \Phi by (\Phi Y)(t) = \int_{\mathbb{R}} \phi(t,s)Y(s)ds For any finite linear combination \sum_i \alpha_i Y(t_i): \begin{align*} \|\Phi(\sum_i \alpha_i Y(t_i))\|^2 &= \|\sum_i \alpha_i \int_{\mathbb{R}} \phi(t_i,s)Y(s)ds\|^2 \\ &\leq \|\sum_i \alpha_i Y(t_i)\|^2 \end{align*} where the inequality follows from the bound on \|\phi(t,s)\| and the Cauchy-Schwarz inequality. (2 \Rightarrow 1): The contractive mapping \Phi induces a family of operators \phi(t,s) via the Kernel theorem for Hilbert spaces. The stationarity of Y and the contractivity of \Phi ensure that these operators satisfy the required properties.
Lemma 1 (Minimal Dilation Property).
If Y(s) is a minimal stationary dilation of X(t), then the scaling function \phi(t,s) achieves the bound
\sup_{t,s} \|\phi(t,s)\| = 1
Proof.
If \sup_{t,s} \|\phi(t,s)\| < 1, we could construct a smaller dilation by scaling Y(s), contradicting minimality.
If \sup_{t,s} \|\phi(t,s)\| < 1, we could construct a smaller dilation by scaling Y(s), contradicting minimality.
3. Structure Theory
Theorem 2 (Sz.-Nagy Dilation).
For any contraction T on a Hilbert space H, there exists a minimal unitary dilation U on a larger space K \supseteq H such that:
T^n = P_H U^n|_H \quad \forall n \geq 0
where P_H is the orthogonal projection onto H.
Lemma 2 (Defect Operators).
For a contraction T, the defect operators defined by:
D_T = (I - T^*T)^{1/2}
D_{T^*} = (I - TT^*)^{1/2}
satisfy:
- \|D_T\| \leq 1 and \|D_{T^*}\| \leq 1
- D_T = 0 if and only if T is an isometry
- D_{T^*} = 0 if and only if T is a co-isometry
4. Convergence Properties
Theorem 3 (Strong Convergence).
For a contractive stationary dilation, the following limit exists in the strong operator topology:
\lim_{n \to \infty} T^n = P_{ker(I-T^*T)}
where P_{ker(I-T^*T)} is the orthogonal projection onto the kernel of I-T^*T.
Proof.
For any x in the Hilbert space:
For any x in the Hilbert space:
- The sequence \{\|T^n x\|\} is decreasing since T is a contraction
- It is bounded below by 0
- Therefore, \lim_{n \to \infty} \|T^n x\| exists
- The limit operator must be the projection onto the space of vectors x satisfying \|Tx\| = \|x\|
- This space is precisely ker(I-T^*T)
Corollary 1 (Asymptotic Behavior).
If T is a strict contraction (i.e., \|T\| < 1), then
\lim_{n \to \infty} T^n = 0
in the strong operator topology.
5. Partial Isometries: The Mathematical Scalpel
Definition 4 (Partial Isometry).
An operator A on a Hilbert space H is a partial isometry if A^*A is an orthogonal projection.
Remark (Geometric Intuition).
A partial isometry is like a mathematical scalpel that carves out a section of space:
- It acts as a perfect rigid motion (isometry) on a specific subspace
- It completely annihilates the rest of the space
Proposition 2 (Key Properties of Partial Isometries).
Let A be a partial isometry. Then:
- A is an isometry when restricted to (ker A)^\perp
- A(ker A)^\perp = ran A
- A^* is also a partial isometry
- AA^*A = A and A^*AA^* = A^*
Theorem 4 (Geometric Characterization).
For a partial isometry A:
A^*A = P_{(ker A)^\perp} \quad \text{and} \quad AA^* = P_{ran A}
where P_S denotes the orthogonal projection onto subspace S.
Proof.
The action of A can be decomposed as:
The action of A can be decomposed as:
- Project onto (ker A)^\perp (this is A^*A)
- Apply a perfect rigid motion to the projected space
Remark (The "Not So Partial" Nature).
Despite the name, there's nothing incomplete about a partial isometry. It performs a complete operation:
- It's a full isometry on its initial space ((ker A)^\perp)
- It perfectly maps this initial space onto its final space (ran A)
- It precisely annihilates everything else