[Paper Review] Partially Isometric Dilations of Noncommuting $N$-tuples of Operators

Michael T. Jury, David W. Kribs|ArXiv.org|Sep 24, 2003

Advanced Operator Algebra Research7 references4 citations

TL;DR

This paper introduces a novel dilation theory for noncommuting $n$-tuples of operators, constructing a unique minimal joint dilation to partial isometries via a directed graph framework. The key contribution is a 'finest' dilation that captures the joint operator behavior more precisely than the classical Frazho-Bunce-Popescu dilation, with a complete partially ordered structure of all minimal dilations.

ABSTRACT

Given a row contraction of operators on Hilbert space and a family of projections on the space which stabilize the operators, we show there is a unique minimal joint dilation to a row contraction of partial isometries which satisfy natural relations. For a fixed row contraction the set of all dilations forms a partially ordered set with a largest and smallest element. A key technical device in our analysis is a connection with directed graphs. We use a Wold Decomposition for partial isometries to describe the models for these dilations, and discuss how the basic properties of a dilation depend on the row contraction.

Motivation & Objective

• To develop a dilation theory for noncommuting $n$-tuples of operators that better reflects their joint structure than the classical Frazho-Bunce-Popescu (FBP) dilation.
• To identify a unique minimal joint dilation of a row contraction to partial isometries satisfying natural algebraic and geometric relations.
• To characterize the set of all minimal dilations as a partially ordered set of directed graphs with a largest (finest) and smallest (FBP-like) element.
• To show that the new dilation avoids the classification difficulties of the Cuntz algebra representations inherent in the FBP dilation by using Cuntz-Krieger graph $\mathrm{C}^*$-algebras.

Proposed method

• Use a family of projections that stabilize the row contraction $T = (T_1, \dots, T_n)$, ensuring mutual commutativity and compatibility with the operator relations.
• Construct a directed graph $G_{\mathcal{P}}$ from the initial and final projections of the dilated partial isometries, where vertices correspond to distinct initial projections and edges to the operators $S_i$.
• Apply the Wold decomposition for families of partial isometries to model the dilation structure, separating the pure and unitary-like parts.
• Define a partial order on minimal dilations via graph deformation: $G_{{\mathcal{P}}_1} \leq G_{{\mathcal{P}}_2}$ if each projection in ${{\mathcal{P}}_1}$ is a sum of projections in ${{\mathcal{P}}_2}$.
• Define the join operation on dilations using the meet of projections: ${\mathcal{P}}_1 \vee {\mathcal{P}}_2 = \{P_1 \wedge P_2 : P_i \in {\mathcal{P}}_i\}$, ensuring a well-defined lattice structure.
• Construct the 'finest' dilation via ${{\mathcal{P}}_0} = \bigwedge_{\alpha} \{P_\alpha : P_\alpha \in {\mathcal{P}}_\alpha\}$, the infimum over all stabilizing projection families, yielding the maximal graph in the order.

Experimental results

Research questions

• RQ1 Can a minimal joint dilation of a noncommuting $n$-tuple of operators be constructed using partial isometries that reflect the joint operator structure more precisely than the FBP dilation?
• RQ2 How is the set of all minimal dilations for a fixed row contraction structured, and does it admit a maximal or minimal element?
• RQ3 What is the role of directed graphs in modeling the structure of minimal partially isometric dilations of operator tuples?
• RQ4 In what cases does the new dilation avoid the representation-theoretic obstructions of the Cuntz algebra that plague the FBP dilation?
• RQ5 Can the finest dilation be explicitly constructed as the supremum of all minimal dilation graphs via a projection meet operation?

Key findings

• For any row contraction $T = (T_1, \dots, T_n)$, there exists a unique minimal joint dilation to an $n$-tuple of partial isometries satisfying the relations $(\dagger)$, which encode orthogonal ranges, compatible initial projections, and full support.
• The set of all minimal dilations forms a partially ordered set of directed graphs, with the smallest element corresponding to the classical Sz.-Nagy or Frazho-Bunce-Popescu dilation (single vertex with $n$ loops).
• The largest element in this poset is the 'finest' dilation, corresponding to the projection family ${{\mathcal{P}}_0} = \bigwedge_{\alpha} \{P_\alpha : P_\alpha \in {\mathcal{P}}_\alpha\}$, which captures the maximal structural detail of $T$.
• Every other minimal dilation corresponds to a deformation of the finest dilation’s graph, obtained by identifying vertices (i.e., summing projections in the family).
• The finest dilation avoids the classification issues of the FBP dilation by realizing the unitary part as a representation of a Cuntz-Krieger graph $\mathrm{C}^*$-algebra, which is type I for many graphs, enabling complete classification.
• The construction is canonical and independent of the choice of stabilizing projection family, with the finest dilation uniquely determined by the operator tuple $T$.