[Paper Review] Every sum system is divisible

Masaki Izumi|ArXiv.org|Aug 12, 2007

Mathematical Dynamics and Fractals6 references3 citations

TL;DR

This paper proves that every sum system is divisible, which, combined with prior results, establishes that all product systems arising from sum systems—and generalized CCR flows—are either of type I or type III. The proof hinges on analyzing the domains of generators of $C_0$-semigroups in generalized CCR flows, showing that real and imaginary addits form subspaces with manageable topologies, thereby enabling a general type I criterion.

ABSTRACT

We show that every sum system is divisible. Combined with B. V. R. Bhat and R. Srinivasan's result, this shows that every product system arising from a sum system (and every generalized CCR flow) is either of type I or type III. A necessarily and sufficient condition for such a product system to be of type I is obtained.

Motivation & Objective

• To establish that every sum system is divisible, resolving a key gap in the classification of product systems arising from sum systems.
• To extend the type I/type III dichotomy—previously known only for divisible sum systems—to all sum systems.
• To provide a general type I criterion for product systems arising from sum systems by analyzing the topological structure of real and imaginary addits.
• To strengthen the connection between generalized CCR flows and sum systems by proving divisibility universally.

Proposed method

• Analyzes the domains of generators of $C_0$-semigroups associated with generalized CCR flows derived from a given sum system.
• Uses the graph inner product on $D(A^*)$ to define the orthogonal complement $\mathcal{K}$ of $D(B)$, where $B$ is the generator of the perturbed semigroup.
• Applies spectral analysis by considering $s > C = \lim_{t\to\infty} \frac{\log\|T_t\|}{t}$, showing that $s$ in the resolvent set of $B$ implies $B$ has no eigenvalues in $s > C$ unless $\dim \mathcal{K} = \dim K_{\mathbb{R}}$.
• Constructs a map $M(s): D(A^*) \to K_{\mathbb{R}}$ via $M(s)f = \int_0^\infty e^{-sx}f(x)dx - s\int_0^\infty e^{-sx}f'(x)dx$ to relate $M(s)p = 0$ to eigenvalue conditions.
• Uses contradiction: if $\dim \mathcal{K} > \dim K_{\mathbb{R}}$, then $p \in \mathcal{K} \setminus \{0\}$ with $M(s)p = 0$ leads to a differential equation $p'' = p$, whose $L^2$-solutions are exponential, contradicting $M(s)p = 0$.
• Concludes $\dim \mathcal{K} = \dim K_{\mathbb{R}}$, proving the index of the perturbation pair is $\dim K_{\mathbb{R}}$, and hence every sum system is divisible.

Experimental results

Research questions

• RQ1 Is every sum system divisible, given that divisibility was previously required for the type I/type III dichotomy?
• RQ2 Can the type I criterion for generalized CCR flows be extended to all sum systems, not just those of finite index?
• RQ3 What is the relationship between the topological structure of real and imaginary addits and the type classification of the resulting product system?
• RQ4 Do non-isomorphic sum systems give rise to cocycle conjugate generalized CCR flows when the product system is of type III?

Key findings

• Every sum system is divisible, as shown by proving $\dim \mathcal{K} = \dim K_{\mathbb{R}}$, where $\mathcal{K}$ is the orthogonal complement of $D(B)$ in $D(A^*)$ with respect to the graph inner product.
• The index of any perturbation pair $(\{S_t\}, \{T_t\})$ of $C_0$-semigroups is equal to $\dim K_{\mathbb{R}}$, confirming the generalization of the index formula beyond the scalar case.
• The type I/type III dichotomy for product systems arising from sum systems now holds universally, since all sum systems are divisible.
• A generalized CCR flow is of type I if and only if the associated sum system is divisible and satisfies the type I criterion, now valid in full generality.
• When the resulting product system is of type III, two sum systems constructed from the same generalized CCR flow may not be isomorphic, even though they yield cocycle conjugate flows.
• The proof establishes that the spaces of real and imaginary addits admit subspaces with manageable topologies, enabling a full generalization of the type I criterion.