Hostname: page-component-cb9f654ff-fg9bn Total loading time: 0 Render date: 2025-08-24T09:24:45.576Z Has data issue: false hasContentIssue false

The Schur–Agler class in infinitely many variables

Published online by Cambridge University Press:  14 July 2025

Greg Knese*
Affiliation:
Department of Mathematics, https://ror.org/01yc7t268 Washington University in St. Louis , One Brookings Drive, St. Louis, MO 63130, United States
Rights & Permissions [Opens in a new window]

Abstract

We define the Schur–Agler class in infinite variables to consist of functions whose restrictions to finite-dimensional polydisks belong to the Schur–Agler class. We show that a natural generalization of an Agler decomposition holds and the functions possess transfer function realizations that allow us to extend the functions to the unit ball of $\ell ^\infty $. We also give a Pick interpolation type theorem which displays a subtle difference with finitely many variables. Finally, we make a brief connection to Dirichlet series derived from the Schur–Agler class in infinite variables via the Bohr correspondence.

Information

Type
Article
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (https://creativecommons.org/licenses/by/4.0), which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
Copyright
© The Author(s), 2025. Published by Cambridge University Press on behalf of Canadian Mathematical Society

1 Introduction

This article is about establishing basic properties of the Schur–Agler class in infinitely many variables. To back up a bit, the Schur class in N variables, $\mathcal {S}_N$ , will refer to the set of analytic functions $f:\mathbb {D}^N \to \overline {\mathbb {D}}$ where $\mathbb {D}^N$ is the N-dimensional unit polydisk

$$\begin{align*}\mathbb{D}^N= \{z=(z_1,\dots, z_N) \in \mathbb{C}^N: \forall j, |z_j|<1\}. \end{align*}$$

The Schur class, $\mathcal {S}_{\infty }$ , in infinitely many variables will refer to holomorphic functions (meaning complex Fréchet differentiable) on $\mathrm{Ball}(\ell ^{\infty })$ that are bounded by one in supremum norm. (We review standard notations in Section 2.)

A remarkable result, attributed to Hilbert, is that if we are given Schur class functions $f_N \in \mathcal {S}_N$ for each N, and if for $N>M$ we have

$$\begin{align*}f_N(z_1,\dots, z_M,0,\dots, 0) \equiv f_M(z_1,\dots, z_M), \end{align*}$$

then there exists a holomorphic function $f: \mathrm{Ball}(c_0)\to \overline {\mathbb {D}}$ such that $f(z_1,\dots , z_N, 0, \dots ) = f_N(z_1,\dots , z_N)$ and such that f is continuous in the norm topology on $\mathrm{Ball}(c_0)$ . See [Reference Defant, García, Maestre and Sevilla-Peris15], Theorem 2.21 for details.

Even more, f has a homogeneous expansion

$$\begin{align*}f(z) = \sum_{m=0}^{\infty} P_m(z), \end{align*}$$

where each $P_m$ is an m-homogeneous form on $c_0$ (see [Reference Defant, García, Maestre and Sevilla-Peris15], Proposition 2.28). Davie-Gamelin [Reference Davie and Gamelin13] proved that f and its homogeneous expansion extends further to $\mathrm{Ball}(\ell ^{\infty })$ . This extension (called the Aron-Berner extension) is somewhat elaborate as it requires passing to the symmetric m-multilinear form associated with each $P_m$ , extending to $\ell ^{\infty }$ and then proving that the extended homogeneous expansion converges in $\mathrm{Ball}(\ell ^{\infty })$ .

Remark 1.1 An important subtlety in all of this theory is that, although we can form a Taylor series $\sum _{\alpha } f_{\alpha } z^{\alpha }$ associated with f that converges absolutely to f on the Hilbert multidisk $\mathbb {D}_2^{\infty } = \mathrm{Ball}(\ell ^{\infty })\cap \ell ^2$ , in general there will be points in $\mathrm{Ball}(c_0)$ where the Taylor series does not converge absolutely. See [Reference Defant, García, Maestre and Sevilla-Peris15], Proposition 4.6 and Theorem 10.1.

Remark 1.2 An important motivation in recent decades for the study of $\mathcal {S}_{\infty }$ is through its application to Dirichlet series. In particular, the space $H^{\infty }(\mathrm{Ball}(c_0))$ of bounded holomorphic functions on $\mathrm{Ball}(c_0)$ is isometrically isomorphic to the space $\mathscr {H}^{\infty }$ of Dirchlet series that converge and are bounded on the right half plane $\{z\in \mathbb {C}: \Re z>0\}$ . The isomorphism, called the the Bohr correspondence, is given by

$$\begin{align*}F \in H^{\infty}(\mathrm{Ball}(c_0)) \mapsto f(s) = F(p_1^{-s}, p_2^{-s},\dots) \in \mathscr{H}^{\infty} \end{align*}$$

where $p_1=2,p_2=3,\dots $ are the prime numbers. The norm in both cases refers to the supremum norm. The space $\mathscr {H}^{\infty }$ and its isomorphism with $H^{\infty }(\mathrm{Ball}(c_0))$ appears naturally in the study of dilation completeness problems on $L^2(0,1)$ as presented in Hedenmalm–Lindqvist–Seip [Reference Hedenmalm, Lindqvist and Seip17]. See also [Reference McCarthy19], [Reference Queffélec and Queffélec22], [Reference Dan and Guo12], [Reference Nikolski20], [Reference Nikolski21], and [Reference Defant, García, Maestre and Sevilla-Peris15]. $\diamond $

Returning to the main topic, the Schur–Agler class in N dimensions, $\mathcal {A}_N$ , consists of $f \in \mathcal {S}_N$ such that for any N-tuple $T=(T_1,\dots , T_N)$ of strictly contractive commuting operators on a Hilbert space we have

$$\begin{align*}\|f(T)\| \leq 1, \end{align*}$$

where $f(T)$ is defined using absolutely convergent power series. We will say Agler class for short. An inequality of von Neumann [Reference von Neumann24] proves that the Agler class in one variable coincides with the Schur class in one variable; $\mathcal {A}_1=\mathcal {S}_1$ . Andô’s dilation theorem [Reference Andô9] proves that the same relation holds in two variables, namely $\mathcal {A}_2=\mathcal {S}_2$ . Counterexamples first constructed by Varopoulos [Reference Varopoulos23] show that $\mathcal {A}_N\ne \mathcal {S}_N$ for $N>2$ . A basic motivation for studying the Agler class is that it can provide insights into the more classical spaces $\mathcal {S}_1,\mathcal {S}_2$ — see Agler-McCarthy-Young [Reference Agler, McCarthy and Young4], [Reference Agler, McCarthy and Young5], and [Reference Agler, McCarthy and Young7]. On the other hand, the Agler class is interesting more broadly: (1) for studying the operator theoretic problem of understanding the failure of von Neumann’s inequality in three or more variables and (2) for providing a large source of interesting and easily constructible examples of functions within $\mathcal {S}_N$ . Some recent papers on the Agler class are in [Reference Barik, Bhattacharjee and Das10], [Reference Bhowmik and Kumar11], [Reference Debnath and Sarkar14], and [Reference Kojin18].

Functions in the Agler class have a variety of useful properties. First, they possess an Agler decomposition and an associated interpolation theorem. An Agler decomposition is a formula of the form

(1.1) $$ \begin{align} 1- f(z)\overline{f(w)} = \sum_{j=1}^{N}(1-\bar{w}_j z_j) K_j(z,w), \end{align} $$

where $K_1,\dots , K_N$ are positive semi-definite kernels on $\mathbb {D}^N\times \mathbb {D}^N$ . The Agler-Pick interpolation theorem can be stated in the following form. Some standard references are [Reference Agler1], [Reference Agler and McCarthy2], [Reference Agler and McCarthy3].

Theorem 1.3 (Agler)

Given a finite subset $X \subset \mathbb {D}^N$ and a function $f:X \to \overline {\mathbb {D}}$ the following are equivalent:

  1. (1) There exists $\tilde {f} \in \mathcal {A}_N$ with $\left .\tilde {f}\right |{}_{X} = f$ .

  2. (2) There exist positive semi-definite functions $K_1,\dots , K_N$ on $X\times X$ such that for $z,w \in X$

    $$\begin{align*}1- f(z)\overline{f(w)} = \sum_{j=1}^{N} (1-z_j\bar{w}_j)K_j(z,w). \end{align*}$$
  3. (3) For every N-tuple $T=(T_1,\dots , T_N)$ of commuting, contractive, simultaneously diagonalizable matrices whose joint eigenspaces have dimension at most 1 and satisfy $\sigma (T) \subset X$ , we have $\|f(T)\|\leq 1$ .

Item (1) is a way of phrasing an interpolation problem as an extension problem. Item (2) is a restriction of an Agler decomposition. (see Section 2 for the definition of positive semi-definite function.) Item (3) says that the function f needs to satisfy a particular type of matrix von Neumann inequality. Notice that in this case $f(T)$ can be defined by using the diagonalization of T, and the dimension of the space that the $T_j$ act on is at most $\# X$ . Item (3) is not stated explicitly in the literature, at least not in this form, but it is a known component of the Agler-Pick interpolation theorem. (For the skeptical reader, the approach in this article proves a generalization to infinite variables and does not directly rely on the finite variable theorem so one could pull a proof of the equivalence of (3) by simplifying certain proofs below.) We think item (3) is important to emphasize since it (conceptually) gives a way to check if interpolation is possible while item (2) is a useful conclusion when you know interpolation is possible. Item (3) is also the source of a subtlety in infinite variables.

A second key property of Agler class functions is that they possess a contractive transfer function realization formula, which means the following. There exists a contractive operator V acting on $\mathbb {C}\oplus \bigoplus _{j=1}^{N} \mathcal {H}_j$ where $\mathcal {H}_1,\dots , \mathcal {H}_N$ are Hilbert spaces such that when we write V in block form $V = \begin {pmatrix} A & B \\ C & D\end {pmatrix}$ we have

(1.2) $$ \begin{align} f(z) = A + B\Delta(z) (1-D\Delta(z))^{-1} C, \end{align} $$

where $\Delta (z) = \sum _{j=1}^{N} z_j P_j$ and each $P_j$ represents projection onto $\mathcal {H}_j$ within the direct sum $\bigoplus _{k=1}^{N} \mathcal {H}_k$ . It turns out that membership in the Agler class can be tested using generic matrices. Namely, analytic $f:\mathbb {D}^N\to \overline {\mathbb {D}}$ belongs to $\mathcal {A}_N$ if for every N-tuple T of commuting contractive simultaneously diagonalizable matrices with joint eigenspaces having dimension $1$ we have $\|f(T)\|\leq 1$ . With this reduction, defining $f(T)$ only requires the evaluation of f on the joint eigenvalues of T and not any regularity or absolute summability. This can be derived from item (3) in Theorem 1.3 or see [Reference Agler, McCarthy and Young6] (Theorem 6.1 therein) where something more general is proven.

Our goal is to prove that analogues of the Agler decomposition (1.1), the Agler-Pick interpolation theorem (Theorem 1.3), and the transfer function realization (1.2) hold in infinitely many variables. We also wish to make connections to Dirichlet series as in Remark 1.2. Remark 1.1 suggests that we cannot define $f(T)$ in the infinite variable setting using absolutely convergent series.

We would like to remark that this article is not the first mention of the Agler class in infinitely many variables—see [Reference Alpay and Levanony8], Section 7. However, one of our larger goals is to start with the simplest definition possible and deduce basic properties of the Agler class (such as its functional calculus) as well as to point out some subtleties of the theory. Here is what we imagine to be the simplest definition of the Agler class in infinitely many variables.

Definition 1.4 Given a function $f: \mathrm{Ball}(c_{00}) \to \mathbb {C}$ we say f is in the Agler class in infinite variables, $\mathcal {A}_{\infty }$ , if for every N, the restriction of f to $\mathbb {D}^N$ belongs to the Agler class in N variables, $\mathcal {A}_N$ .

Theorem 1.5 If $f \in \mathcal {A}_{\infty }$ , then f has a transfer function realization: there exists a contractive operator V acting on $\mathbb {C}\oplus \bigoplus _{j=1}^{\infty } \mathcal {H}_j$ where $\mathcal {H}_1,\mathcal {H}_2, \dots $ are Hilbert spaces such that when we write V in block form $V = \begin {pmatrix} A & B \\ C & D\end {pmatrix}$ we have

(1.3) $$ \begin{align} f(z) = A + B\Delta(z) (1-D\Delta(z))^{-1} C, \end{align} $$

where $\Delta (z) = \sum _{j=1}^{\infty } z_j P_j$ and each $P_j$ represents projection onto $\mathcal {H}_j$ within the direct sum $\bigoplus _{k=1}^{\infty } \mathcal {H}_k$ . Letting $f_N(z) = f(z_1,\dots , z_N,0 ,\dots )$ , we have the following extension of von Neumann’s inequality: for any tuple $T=(T_1,T_2,\dots )$ of commuting contractive operators (acting on a common Hilbert space) such that $\sup _{j} \|T_j\| <\infty $ we have that

$$\begin{align*}\lim_{N \to \infty} f_N(T) \end{align*}$$

converges in the strong operator topology to a natural definition of $f(T)$ as

$$\begin{align*}f(T) := (A\otimes I) + (B\otimes I) \Delta(T) (1-(D\otimes I)\Delta(T))^{-1} (C\otimes I), \end{align*}$$

where $\Delta (T) := \sum _{j=1}^{\infty } P_j \otimes T_j$ (also convergent in the strong operator topology).

The transfer function formula (1.3) for f makes sense for $z \in \mathrm{Ball}(\ell ^{\infty })$ and we have

$$\begin{align*}\lim_{N\to \infty} f_N(z) = f(z) \end{align*}$$

for $z \in \mathrm{Ball}(\ell ^{\infty })$ . Thus, the transfer function formula readily shows that Agler class functions extend to $\mathrm{Ball}(\ell ^{\infty })$ . Our proof relies on a Montel theorem in infinite variables from [Reference Defant, García, Maestre and Sevilla-Peris15] and does not use the intricate argument involving nets of points in $\mathrm{Ball}(c_0)$ as in Davie-Gamelin [Reference Davie and Gamelin13] for the Schur class case.

The transfer function formula (1.3) is essentially equivalent (via a standard argument) to an Agler decomposition

$$\begin{align*}1- f(z)\overline{f(w)} = \sum_{j=1}^{\infty}(1-\bar{w}_j z_j) K_j(z,w), \end{align*}$$

which we will show converges absolutely. Again, the $K_j$ are positive semi-definite kernels on $\mathrm{Ball}(\ell ^{\infty })\times \mathrm{Ball}(\ell ^{\infty })$ .

Remark 1.6 The paper Dritschel-McCullough [Reference Dritschel and McCullough16] discusses a version of the Agler class in infinite variables via an approach to interpolation and realization formulas using test functions. Their definition of the Agler class in infinitely many variables is more expansive and allows for, for instance, linear functionals on $\ell ^{\infty }$ that annihilate $c_0$ and do not have Agler decompositions in the sense of Theorem 1.5. This expanded Agler class has a more general type of Agler decomposition. See Proposition 5.4 of [Reference Dritschel and McCullough16]. Here is a simplified version of an example they present.

Let $\alpha = (a_n)_{n=1}^{\infty }$ be an increasing sequence of positive real numbers that converge to some $a \in (0,1)$ ; e.g. $a_n = \frac {1}{2} - \frac {1}{n+1}$ . By the Hahn–Banach theorem, there exists a linear functional $L \in (\ell ^{\infty })^*$ such that $L(\alpha ) = a$ and $\|L\| = 1$ . We claim $L \notin \mathcal {A}_\infty $ . If we had an Agler decomposition,

$$\begin{align*}1-L(z)\overline{L(w)} = \sum_{j=1}^{\infty}(1-z_j\bar{w}_j) K_j(z,w) \end{align*}$$

with each $K_j$ positive semi-definite on $\mathrm{Ball}(\ell ^{\infty })$ and $\sum _{j=1}^{\infty } K_j(z,z)<\infty $ , then inserting different combinations $z,w \in \{0,\alpha \}$ we have

$$\begin{align*}1 = \sum_{j=1}^{\infty} K_j(0,0), \qquad 1 = \sum_{j=1}^{\infty} K_j(\alpha,0), \qquad 1-a^2 = \sum_{j=1}^{\infty} (1-a_j^2)K_j(\alpha,\alpha). \end{align*}$$

By Cauchy–Schwarz,

$$\begin{align*}1 \leq \sum_{j=1}^{\infty} |K_j(\alpha,0)| \leq \left(\sum_{j=1}^{\infty} K_j(0,0)\right)^{1/2} \left(\sum_{j=1}^{\infty} K_j(\alpha, \alpha)\right)^{1/2} \end{align*}$$

so that $ 1\leq \sum _{j=1}^{\infty } K_j(\alpha , \alpha ). $ On the other hand, we have

$$\begin{align*}(1-a^2)(1-\sum_{j=1}^{\infty} K_j(\alpha,\alpha)) = \sum_{j=1}^{\infty}(a^2-a_j^2)K_j(\alpha,\alpha) \geq 0 \end{align*}$$

and therefore this must equal zero which would imply $K_j(\alpha ,\alpha ) = 0$ for all j. This is a contradiction.

Theorem 1.5 is adapted to functions that have the added continuity that makes them completely determined by their values on $c_{00} \subset c_0$ and so our definition rules out functions that are zero on all of $c_{00}$ . $\diamond $

Some aspects of interpolation in $\mathcal {A}_\infty $ are straightforward, however one important aspect has a subtlety.

Theorem 1.7 Let $X \subset \mathrm{Ball}(\ell ^{\infty })$ be a finite subset and let $f:X \to \overline {\mathbb {D}}$ be a function. Consider the following conditions.

  1. (1) There exists $\tilde {f} \in \mathcal {A}_\infty $ with $\left .\tilde {f}\right |{}_{X} = f$ .

  2. (2) There exist positive semi-definite functions $K_1, K_2, \dots $ on X such that for $z,w \in X$

    $$\begin{align*}1- f(z)\overline{f(w)} = \sum_{j=1}^{\infty} (1-z_j\bar{w}_j)K_j(z,w) \end{align*}$$
    and for all $z\in X$
    $$\begin{align*}\sum_{j=1}^{\infty} K_j(z,z) < \infty. \end{align*}$$
  3. (3) For every tuple $T=(T_1,T_2,\dots )$ of commuting, contractive, simultaneously diagonalizable matrices whose joint eigenspaces have dimension at most 1 and satisfy $\sigma (T) \subset X$ , we have $\|f(T)\|\leq 1$ .

Then,

  • (1) and (2) are equivalent and imply item (3).

  • Item (3) implies (1) and (2) when $X \subset \mathbb {D}_{2}^{\infty }$ .

In particular, (1), (2), and (3) are equivalent when $X \subset \mathbb {D}_{2}^{\infty }$ .

The subtlety alluded to above is that we only obtain a full generalization of an Agler-Pick interpolation theorem when our interpolation points lie in $\mathbb {D}_{2}^{\infty }$ and we do not know to what extent this condition can be removed.

Remark 1.8 Going back to Remark 1.6 and the example discussed there, the interpolation problem $0 \in \ell ^{\infty } \mapsto 0\in \mathbb {C}, \alpha \mapsto a$ cannot be solved within $\mathcal {A}_{\infty }$ , however, the associated function $f: \{0,\alpha \} \to \{0,a\}$ , $f(0)=0$ , $f(\alpha ) = a$ , satisfies the condition of item (3) above. Indeed, if we have commuting simultaneously diagonalizable contractive matrices $T_j$ with $\sigma (T_j) \subset \{0,a_j\}$ , then $f(T_1,T_2,\dots )$ will be contractive by continuity. Specifically, if $\vec {b}_0$ is the eigenvector for $0$ and $\vec {b}_1$ the eigenvector for $\alpha $ , then contractivity of $T_j$ means

$$\begin{align*}|T_j(\sum_{k=0,1} c_k \vec{b}_k)| \leq |\sum_{k=0,1} c_k \vec{b}_k| \end{align*}$$

for arbitrary $c_0,c_1\in \mathbb {C}$ . But $T_j(\sum _{k=0,1} c_k \vec {b}_k) = c_1 a_j \vec {b}_1$ and sending $j\to \infty $ we get

$$\begin{align*}|f(T) (\sum_{k=0,1} c_k \vec{b}_k)| \leq |\sum_{k=0,1} c_k \vec{b}_k|. \end{align*}$$

It seems that a complete relaxation of the condition $X\subset \mathbb {D}_{2}^{\infty }$ to $X \subset \mathrm{Ball}(\ell ^{\infty })$ would lead to the broader notion of Agler class constructed with the test function approach of [Reference Dritschel and McCullough16]. It would be interesting if the condition $X\subset \mathbb {D}_{2}^{\infty }$ could be relaxed to $X\subset \mathrm{Ball}(c_0)$ with a valid interpolation theorem in $\mathcal {A}_\infty $ . $\diamond $

Referring to Remark 1.2, it is of interest to understand the image of the map

$$\begin{align*}F \in \mathcal{A}_{\infty} \mapsto f(s) = F(p_1^{-s}, p_2^{-s},\dots) \in \mathscr{H}^{\infty} \end{align*}$$

into the space $\mathscr {H}^{\infty }$ of convergent and bounded Dirichlet series in the right half plane in $\mathbb {C}$ . We shall let $\mathscr {A}^{\infty }$ denote the image of the above map. (We caution that as we have defined things the functions in $\mathscr {A}^{\infty }$ are bounded by $1,$ whereas $\mathscr {H}^{\infty }$ is a Banach space of functions normed by supremum norm.) The following is basically a formality but worth pointing out. Let $\mathbb {C}_+ = \{z\in \mathbb {C}: \Re z>0\}$ denote the right half plane.

Theorem 1.9 Let $f \in \mathscr {H}^{\infty }$ . The following are equivalent:

  1. (1) $f\in \mathscr {A}^{\infty }$

  2. (2) There exist positive semi-definite kernels $K_1, K_2,\dots $ on $\mathbb {C}_+$ such that

    $$\begin{align*}1- f(s) \overline{f(w)} = \sum_{j=1}^{\infty} (1-p_j^{-(s+\bar{w})})K_j(s,w) \end{align*}$$
    and $\sum _{j=1}^{\infty } K_j(s,s)<\infty $ for each $s\in \mathbb {C}_+$ .
  3. (3) For every diagonalizable matrix M with one-dimensional eigenspaces, with $\sigma (M) \subset \mathbb {C}_+$ , and with the property $\|n^{-M}\|\leq 1$ for all $n \in \mathbb {N}$ , we have

    $$\begin{align*}\|f(M)\| \leq 1. \end{align*}$$

Again, $p_1,p_2,\dots $ are the prime numbers. It would be interesting if the matrices M in item (3) had a simpler description. Functions in $\mathscr {A}^{\infty }$ satisfy a special von Neumann inequality.

Theorem 1.10 Let $f \in \mathscr {A}^{\infty }$ . Suppose M is a bounded operator on a Hilbert space such that $\sigma (M) \subset \mathbb {C}_{+}$ and $\|n^{-M}\| \leq 1$ for every $n \in \mathbb {N}$ . Then,

$$\begin{align*}\|f(M)\| \leq 1 \end{align*}$$

where $n^{-M}$ and $f(M)$ are defined using the Riesz holomorphic functional calculus.

2 Notations and background

Several spaces of sequences will be of interest.

  • $\mathbb {N} = \{1,2,\dots \}$ .

  • $\ell ^{\infty } = \ell ^{\infty }(\mathbb {N}) = \{ z = (z_j)_{j\in \mathbb {N}} \in \mathbb {C}^{\mathbb {N}}: \|z\|_{\infty } := \sup _j |z_j| <\infty \}.$

  • $\ell ^2 = \ell ^2(\mathbb {N}) = \{ z= (z_j)_{j\in \mathbb {N}} : \sum _{j=1}^{\infty } |z_j|^2 <\infty \}.$

  • $c_0 = c_0(\mathbb {N}) = \{z\in \ell ^{\infty }: \lim _{j\to \infty } z_j = 0\}$ .

  • $c_{00} = c_{00}(\mathbb {N}) = \{ z \in \ell ^{\infty }: \exists N \in \mathbb {N}, z_j = 0 \text { for } j>N\}$ .

  • $\mathrm{Ball}(\ell ^{\infty }) = \{z\in \ell ^{\infty }: \|z\|_{\infty } <1\}$ denotes the open unit ball of $\ell ^{\infty }$ .

  • $\mathbb {D}^{\infty } = \mathbb {D}^{\mathbb {N}} = \{z \in \ell ^{\infty }: \forall j, |z_j| <1\}$ .

  • $\mathrm{Ball}(c_0) = \{z\in c_0: \|z\|_{\infty } <1\}$ .

  • $\mathrm{Ball}(c_{00}) = \{z\in c_{00}: \|z\|_{\infty } < 1\}$ .

  • We identify $\mathbb {D}^{N}$ with $\mathbb {D}^N \times \{(0,0,\dots )\} \subset \mathrm{Ball}(c_{00})$ .

  • The Hilbert multidisk is the set $\mathbb {D}_{2}^{\infty } := \ell ^2 \cap \mathrm{Ball}(\ell ^\infty )$ , namely, the set of sequences $(z_j)_{j\in \mathbb {N}}$ such that $\sup _j |z_j|<1$ and $\sum _{j} |z_j|^2 <\infty $ . Note that for $z,w \in \mathbb {D}_2^{\infty }$ the infinite product

    $$\begin{align*}\prod_{j=1}^{\infty} \frac{1}{1-\bar{w}_j z_j} \end{align*}$$
    converges absolutely.
  • We generally use standard modulus bars $|\cdot |$ for the modulus of complex numbers or vectors (in $\mathbb {C}^N$ or Hilbert space) while double bars $\|\cdot \|$ are reserved for operator norms or other norms as listed above.

Remark 2.1 We use the basics of positive semi-definite functions. Given a set X, a function $A:X\times X \to \mathbb {C}$ is positive semi-definite on X if for every finite subset $Y\subset X$ and every function $a:Y\to \mathbb {C}$ we have

$$\begin{align*}\sum_{z,w \in Y} a(z)\overline{a(w)} A(z,w) \geq 0. \end{align*}$$

We write $A(z,w) \succeq 0$ in this case. More generally, we write $A(z,w) \succeq B(z,w)$ if $A - B \succeq 0$ . We frequently use the Schur product theorem which say that if $A(z,w), B(z,w) \succeq 0$ then $A(z,w)B(z,w) \succeq 0$ . For a function $f:X\to \mathbb {C}$ , we let $f\otimes \bar {f}$ denote the function $(z,w) \mapsto f(z)\overline {f(w)}$ . Also, if $X \subset \mathbb {C}^{\mathbb {N}}$ , we let $Z_j \otimes \bar {Z}_j$ denote the function $(z,w) \mapsto z_j\bar {w}_j$ .

3 Proof of Theorem 1.5

Theorem 1.5 will be proven with three lemmas. The first lemma is our main advance while the other two are standard.

For the first lemma, let $\rho = (\rho _n)_{n\in \mathbb {N}} \in \mathbb {D}_{2}^{\infty }$ be a fixed square summable sequence of positive numbers. Let $\overline {\mathbb {D}}_{\rho } = \{z =(z_n)_{n\in \mathbb {N}}: |z_n|\leq \rho _n\}$ and let $\mathbb {D}_{\rho } = \{z \odot \rho : z \in \mathrm{Ball}(c_0)\}$ where $\rho \odot z = (\rho _jz_j)_{j\in \mathbb {N}}$ . (The notation is not entirely consistent but it is temporary.)

Lemma 3.1 Assume $f: \mathrm{Ball}(c_{00})\to \mathbb {C}$ is in the Agler class, $\mathcal {A}_{\infty }$ .

Then, for $j=0,1,2,\dots $ , there exist positive semi-definite kernels $K_j$ on $\mathbb {D}_{\rho }$ such that

$$\begin{align*}1- f(z)\overline{f(w)} = K_0(z,w) + \sum_{j=1}^{\infty}(1-z_j\bar{w}_j) K_j(z,w) \end{align*}$$

where the sum converges absolutely.

Note that we have introduced the term $K_0$ which is for convenience in the proof. This term can be absorbed into any of the other terms for instance as

$$\begin{align*}(1 - z_1 \bar{w}_1)\left( \frac{K_0(z,w)}{1-z_1\bar{w}_1} + K_1(z,w)\right). \end{align*}$$

Since $K_0$ will be positive semi-definite and since $\frac {1}{1-z_1\bar {w}_1}$ is positive semi-definite, the product is positive semi-definite.

Remark 3.2 In the proof, we will use the Montel theorem given in [Reference Defant, García, Maestre and Sevilla-Peris15] (Theorem 2.17). It states that for a separable normed vector space X, if we are given a sequence $(D_n)_{n\in \mathbb {N}}$ of $D_n \in H^\infty (\mathrm{Ball}(X))$ with uniformly bounded supremum norms, say $\|D_n\|_{\infty } \leq 1$ , then there exists a subsequence $(D_{n_j})_{j\in \mathbb {N}}$ that converges uniformly on compact subsets of $\mathrm{Ball}(X)$ to some $D \in H^{\infty }(\mathrm{Ball}(X))$ necessarily with $\|D\|_{\infty } \leq 1$ . We will apply this to $X=c_0$ using compact sets of the form $\overline {\mathbb {D}}_{\rho }$ defined above. $\diamond $

Proof of Lemma 3.1

For each N, let $f_N(z_1,z_2,\dots ) = f(z_1,\dots ,z_N, 0,\dots )$ . Since f restricted to $\mathbb {D}^N$ has an Agler decomposition, we can write

$$\begin{align*}1-f_N(z)\overline{f_N(w)} = \sum_{j=1}^{N}(1-\bar{w}_j z_j) K^N_j(z,w) \end{align*}$$

for positive semi-definite kernels $K_j^N$ that only depend on the first N variables $z_1,\dots , z_N,w_1,\dots , w_N$ , are analytic in z, and are anti-analytic in w.

Note that for $z,w\in \mathbb {D}_{2}^{\infty }$ , the product

$$\begin{align*}S(z,w) = \prod_{j=1}^{\infty} \frac{1}{1-\bar{w}_j z_j} \end{align*}$$

converges and is positive semi-definite. Note that

$$\begin{align*}S_n(z,w) = (1-\bar{w}_n z_n) S(z,w) = \prod_{j\ne n} \frac{1}{1-\bar{w}_j z_j} \end{align*}$$

is also positive semi-definite and $S, S_n \succeq 1$ using the partial order from Remark 2.1. Here ‘ $1$ ’ is the identically $1$ function.

Let $f_N\otimes \overline {f_N}$ denote the function $(z,w) \mapsto f_N(z)\overline {f_N(w)}$ . We have the positive semi-definite function inequality

(3.1) $$ \begin{align} S \succeq (1-f_N\otimes \overline{f_N})S = \sum_{j=1}^{N} K_j^{N} S_j \succeq \sum_{j=1}^{N} K_j^{N} \succeq \sum_{j=1}^{M} K_j^N \end{align} $$

for $M<N$ and by Cauchy–Schwarz we have for $z,w\in \mathbb {D}_{2}^{\infty }$

$$\begin{align*}S(z,z)^{1/2}S(w,w)^{1/2} \geq |K_j^N(z,w)|. \end{align*}$$

Recall $\rho \odot z = (\rho _jz_j)_{j\in \mathbb {N}}$ . For $z,w \in \mathrm{Ball}(c_0)$ , $K_j^N(\rho \odot z, \rho \odot \bar {w})$ is bounded and analytic in $\mathrm{Ball}(c_0)\times \mathrm{Ball}(c_0)$ with supremum norm bounded by $S(\rho ,\rho )$ . By the Montel theorem (see Remark 3.2), for each $j=1,2,\dots $ in succession there is a subsequence of $N\in \mathbb {N}$ such that

(3.2) $$ \begin{align} K_j^N(\rho\odot z, \rho \odot \bar{w}) \overset{N\to\infty}{\to} K_j(\rho\odot z, \rho \odot \bar{w}) \text{ uniformly on compact subsets of } \mathrm{Ball}(c_0). \end{align} $$

The limiting functions, $K_j$ , are necessarily positive semi-definite because this property is preserved under limits. By a standard diagonal argument, we can find a common subsequence of N such that for all j, (3.2) holds. Recall $\mathbb {D}_{\rho } = \{z \odot \rho : z \in \mathrm{Ball}(c_0)\}$ . By (3.1), for $z,w\in \mathbb {D}_{\rho }$ , $ S\succeq \sum _{j=1}^{M} K_j^N $ for $M<N$ and sending $N \to \infty $ we have $S \succeq \sum _{j=1}^{M} K_j$ . Finally, we can send $M\to \infty $ to obtain $ S \succeq \sum _{j=1}^{\infty } K_j $ with absolute convergence. Absolute convergence can be proven by looking at the diagonal $z=w$ first and then applying Cauchy–Schwarz. To finish the proof, we show

$$\begin{align*}K_0(z,w) := 1-f(z)\overline{f(w)} - \sum_{j=1}^{\infty}(1-z_j \bar{w}_j)K_j(z,w) \end{align*}$$

is positive semi-definite. Here we emphasize that f on $\mathbb {D}_{\rho }$ is defined via f’s holomorphic extension from $\mathrm{Ball}(c_{00})$ to $\mathrm{Ball}(c_0)$ as in the traditional Schur class of infinitely many variables.

Now, for $N<M$

$$\begin{align*}1-f_M(z)\overline{f_M(w)} - \sum_{j=1}^{N}(1-z_j \bar{w}_j)K_j^{M}(z,w)= \sum_{j=N+1}^{M}(1-z_j\bar{w}_j)K_j^{M}(z,w). \end{align*}$$

Let $Z_j \otimes \overline {Z_j}$ denote the function $(z,w) \mapsto z_j\bar {w}_j$ . Since $S \succeq K_j^{M}$ we see that

$$\begin{align*}1 - f_M\otimes \overline{f_M} - \sum_{j=1}^{N} (1 - Z_j \otimes \overline{Z_j}) K_j^M \succeq - \sum_{j=N+1}^{M} (Z_j \otimes \overline{Z_j}) S. \end{align*}$$

This inequality means that for any finite subset $Y \subset \mathbb {D}_{\rho }$ and any function $a:Y \to \mathbb {C}$

$$\begin{align*}\sum_{z,w \in Y} (1-f_M(z)\overline{f_M(w)} - \sum_{j=1}^{N}(1-z_j \bar{w}_j)K_j^{M}(z,w))a(z) \overline{a(w)} \geq -\sum_{z,w\in Y} \sum_{j=N+1}^{M} z_j\bar{w}_j S(z,w) a(z) \overline{a(w)}. \end{align*}$$

(See Remark 2.1.) Now,

$$\begin{align*}\sum_{z,w\in Y} \sum_{j=N+1}^{M} z_j\bar{w}_j S(z,w) a(z) \overline{a(w)} \leq S(\rho,\rho) \sum_{j=N+1}^{M}\rho_j^2 \left(\sum_{z\in Y} |a(z)|\right)^2 \leq S(\rho,\rho) \sum_{j=N+1}^{\infty}\rho_j^2 \left(\sum_{z\in Y} |a(z)|\right)^2. \end{align*}$$

Sending $M\to \infty $

$$\begin{align*}\sum_{z,w \in Y} (1-f(z)\overline{f(w)} - \sum_{j=1}^{N}(1-z_j \bar{w}_j)K_j(z,w))a(z) \overline{a(w)} \geq -S(\rho,\rho) \sum_{j=N+1}^{\infty}\rho_j^2 \left(\sum_{z\in Y} |a(z)|\right)^2 \end{align*}$$

and then sending $N\to \infty $

$$\begin{align*}\sum_{z,w \in Y} (1-f(z)\overline{f(w)} - \sum_{j=1}^{\infty}(1-z_j \bar{w}_j)K_j(z,w))a(z) \overline{a(w)} \geq 0 \end{align*}$$

which proves $K_0(z,w)$ is positive semi-definite.

Lemma 3.3 Let $X \subset \mathrm{Ball}(\ell ^{\infty })$ . Assume $f: X\to \mathbb {C}$ is a function such that for $j=0,1,2,\dots $ , there exist positive semi-definite kernels $K_j:X\times X \to \mathbb {C}$ on X such that

$$\begin{align*}1- f(z)\overline{f(w)} = \sum_{j=1}^{\infty}(1-z_j\bar{w}_j) K_j(z,w), \end{align*}$$

where the sum converges absolutely. Then, there exists a contractive operator V acting on $\mathbb {C}\oplus \bigoplus _{j=1}^{\infty } \mathcal {H}_j$ where $\mathcal {H}_1,\mathcal {H}_2, \dots $ are Hilbert spaces such that when we write V in block form $V = \begin {pmatrix} A & B \\ C & D\end {pmatrix}$ we have

$$\begin{align*}f(z) &= A + B\Delta(z) (1-D\Delta(z))^{-1} C\\K_j(z,w) &= C^*(I-\Delta(w)^* D^*)^{-1}P_j(I-D\Delta(z))^{-1} C, \end{align*}$$

where $\Delta (z) = \sum _{j=1}^{\infty } z_j P_j$ and each $P_j$ represents projection onto $\mathcal {H}_j$ within the direct sum $\bigoplus _{k=1}^{\infty } \mathcal {H}_k$ . With these formulas, f and $K_j$ extend to $\mathrm{Ball}(\ell ^{\infty })$ and there exists a positive semi-definite kernel $K_0(z,w)$ such that

$$\begin{align*}1- f(z)\overline{f(w)} = K_0(z,w) + \sum_{j=1}^{\infty}(1-z_j\bar{w}_j) K_j(z,w) \end{align*}$$

holds on $\mathrm{Ball}(\ell ^{\infty })\times \mathrm{Ball}(\ell ^{\infty })$ .

The proof is a standard lurking isometry argument (for those who know what that is) that we include for completeness (for those who do not).

Proof By the Moore–Aronszajn theorem on reproducing kernel Hilbert space, we can factor $K_j(z,w) = K_{j,z}^* K_{j,w}$ for $K_{j,z}$ an element of some Hilbert space $\mathcal {H}_j$ . We write $K_{j,z}^* K_{j,w}$ instead of $\langle K_{j,w}, K_{j,z} \rangle $ and view $K_{j,z}^*$ as an element of the dual $\mathcal {H}_j^*$ of $\mathcal {H}_j$ . The following map:

$$\begin{align*}\begin{pmatrix} 1 \\ z_1(K_{1,z})^* \\ z_2 (K_{2,z})^* \\ \vdots \end{pmatrix} \mapsto \begin{pmatrix} f(z) \\ (K_{1,z})^* \\ (K_{2,z})^* \\ \vdots \end{pmatrix} \end{align*}$$

initially defined for vectors indexed by $z\in X$ extends linearly and in a well-defined way to a contractive operator V from $\mathbb {C} \oplus \bigoplus _{j=1}^{\infty } \mathcal {H}_j^*$ to $\mathbb {C} \oplus \bigoplus _{j=1}^{\infty } \mathcal {H}_j^* $ . We write V in block form $ \begin {pmatrix} A & B \\ C & D \\ \end {pmatrix} $ with $A \in \mathbb {C} \cong \mathcal {B}(\mathbb {C},\mathbb {C})$ , $B \in \mathcal {B}(\bigoplus _{j=1}^{\infty } \mathcal {H}_j^*, \mathbb {C})$ , $C \in \mathcal {B}(\mathbb {C}, \bigoplus _{j=1}^{\infty } \mathcal {H}_j^*)$ , $D \in \mathcal {B}(\bigoplus _{j=1}^{\infty } \mathcal {H}_j^*)$ , using the notation $\mathcal {B}(X,Y)$ to denote the bounded linear operators from X to Y (as well as $\mathcal {B}(X)= \mathcal {B}(X,X)$ ). Let $\Delta (z) \in \mathcal {B}(\bigoplus _{j=1}^{\infty } \mathcal {H}_j^*)$ be the diagonal operator sending

$$\begin{align*}(h_j)_{j\in \mathbb{N}} \in \bigoplus_{j=1}^{\infty} \mathcal{H}_j^* \mapsto (z_j h_j)_{j\in\mathbb{N}} \in \bigoplus_{j=1}^{\infty} \mathcal{H}_j^*. \end{align*}$$

Let $F(z) := (K_{j,z}^*)_{j\in \mathbb {N}} \in \bigoplus _{j=1}^{\infty } \mathcal {H}_j^*$ . Then,

$$\begin{align*}V \begin{pmatrix} 1 \\ \Delta(z) F(z) \end{pmatrix} = \begin{pmatrix} f(z) \\ F(z) \end{pmatrix} \end{align*}$$

which implies

(3.3) $$ \begin{align} A + B \Delta(z) F(z) &= f(z) \end{align} $$
(3.4) $$ \begin{align} C+ D \Delta(z) F(z) &= F(z). \end{align} $$

Solving for $F(z)$ and then $f(z)$ we obtain

$$\begin{align*}\begin{aligned} F(z) &= (I-D\Delta(z))^{-1} C \\ f(z) &= A + B \Delta(z) (I-D \Delta(z))^{-1} C. \end{aligned} \end{align*}$$

Note the expressions on the right are defined as written for $z \in \mathrm{Ball}(\ell ^{\infty })$ . Evidently, $F(w)^*P_jF(z)$ extends $K_j(z,w)$ . Since V is contractive, we can factor ${I-V^*V = W^*W}$ for some operator W. Let

$$\begin{align*}G(z) = W \begin{pmatrix} 1 \\ \Delta(z) F(z) \end{pmatrix} \end{align*}$$

so that

$$\begin{align*}\begin{pmatrix} 1 \\ \Delta(w) F(w) \end{pmatrix}^* \begin{pmatrix} 1 \\ \Delta(z) F(z) \end{pmatrix} = \begin{pmatrix} f(w) \\ F(w) \end{pmatrix}^* \begin{pmatrix} f(z) \\ F(z) \end{pmatrix} + G(w)^*G(z). \end{align*}$$

This rearranges into

$$\begin{align*}1- f(z)\overline{f(w)} = F(w)^*(I-\Delta(w)^*\Delta(z))F(z) + G(w)^*G(z) \end{align*}$$

and since

$$\begin{align*}F(w)^*(I-\Delta(w)^*\Delta(z))F(z) = \sum_{j=1}^{\infty} (1-\bar{w}_jz_j) F(w)^*P_j F(z) \end{align*}$$

we have the desired extension of the Agler decomposition using $K_0(z,w) = G(w)^*G(z)$ .

Lemmas 3.1 and 3.3 establish most of Theorem 1.5. For the remaining part, we need a basic estimate on transfer function formulas in order to establish the full von Neumann inequality for the Agler class.

Lemma 3.4 Suppose V is a contractive operator acting on $\mathbb {C}\oplus \bigoplus _{j=1}^{\infty } \mathcal {H}_j$ , where $\mathcal {H}_1,\mathcal {H}_2, \dots $ are Hilbert spaces. Writing V in block form $V = \begin {pmatrix} A & B \\ C & D\end {pmatrix}$ we define for $z \in \mathrm{Ball}(\ell ^{\infty })$

$$\begin{align*}f(z) = A + B\Delta(z) (1-D\Delta(z))^{-1} C \end{align*}$$

where $\Delta (z) = \sum _{j=1}^{\infty } z_j P_j$ and each $P_j$ represents projection onto $\mathcal {H}_j$ within the direct sum $\bigoplus _{k=1}^{\infty } \mathcal {H}_k$ . Then, for $z,w \in \mathrm{Ball}(\ell ^{\infty })$

$$\begin{align*}f(z) - f(w) = B(1-\Delta(z)D)^{-1}\Delta(z-w)(1-D\Delta(w))^{-1}C. \end{align*}$$

Proof

$$\begin{align*}\begin{aligned} f(z) - f(w) &= B(\Delta(z)(1-D\Delta(z))^{-1} - \Delta(w)(1-D\Delta(w))^{-1})C \\ &= B( \Delta(z)((1-D\Delta(z))^{-1} - (1-D\Delta(w))^{-1}) + \Delta(z-w) (1-D\Delta(w))^{-1})C\\ &=B(\Delta(z)(1-D\Delta(z))^{-1})(D\Delta(z-w))(1-D\Delta(w))^{-1} + \Delta(z-w)(1-D\Delta(w))^{-1})C\\ &= B(\Delta(z)(1-D\Delta(z))^{-1})D + 1)\Delta(z-w)(1-D\Delta(w))^{-1})C\\ &= B(1-\Delta(z)D)^{-1}\Delta(z-w)(1-D\Delta(w))^{-1}C. \end{aligned}\\[-34pt] \end{align*}$$

Now we finish the proof of Theorem 1.5. As written in Theorem 1.5, for an infinite tuple $T = (T_1,T_2,\dots )$ of operators on a common Hilbert space $\mathcal {H}$ satisfying

$$\begin{align*}\|T\|_{\infty} := \sup_j \|T_j\| < 1 \end{align*}$$

we define

$$\begin{align*}f(T) := (A\otimes I) + (B\otimes I) \Delta(T) (1-(D\otimes I)\Delta(T))^{-1} (C\otimes I) \end{align*}$$

where $\Delta (T) := \sum _{j=1}^{\infty } P_j \otimes T_j$ . Note that this definition does not require the operators $(T_j)_j$ to pairwise commute but if they do then each $T_j$ commutes with $f(T)$ since each $I\otimes T_j$ commutes with $\Delta (T)$ .

Since $(D\otimes I)\Delta (T)$ is strictly contractive, $f(T)$ equals the absolutely convergent sum

$$\begin{align*}(A\otimes I) + (B\otimes I) \Delta(T)\sum_{j=0}^{\infty} ((D\otimes I)\Delta(T))^{j} (C\otimes I) \end{align*}$$

and substituting $z\in \mathrm{Ball}(\ell ^{\infty })$ we obtain an absolutely convergent homogeneous expansion for $f(z)$

(3.5) $$ \begin{align} f(z) = A + B\Delta(z)\sum_{j=0}^{\infty} ((D\Delta(z))^{j} C. \end{align} $$

Before we finish the proof of Theorem 1.5 we make some clarifications about our functional calculus.

Remark 3.5 In finitely many variables we stated that our convention/definition for $f(T)$ is via an absolutely convergent power series expansion. Therefore, it should be pointed out that this new formulation of “ $f(T)$ ” using the transfer function formula matches the old one. All that really needs to be said is that when we insert $z=(z_1,\dots , z_N,0,\dots )$ into (3.5) the homogeneous terms $B\Delta (z)(D\Delta (z))^jC$ are homogeneous polynomials and since f is analytic on $\mathbb {D}^N$ , the monomial sum we obtain from expanding $B\Delta (z)(D\Delta (z))^jC$ is absolutely convergent in $\mathbb {D}^N$ . Thus, evaluating $f(T)$ at a finite tuple $T=(T_1,T_2,\dots , T_N,0,\dots )$ can either be evaluated using the transfer function formula or the absolutely convergent power series. $\diamond $

The proof of Lemma 3.4 extends directly to prove that for another such tuple S acting on the same Hilbert space $\mathcal {H}$ as T we have

$$\begin{align*}f(T) - f(S) = (B\otimes I)(1-\Delta(T)(D\otimes I))^{-1}\Delta(T-S)(1-(D\otimes I)\Delta(S))^{-1}(C\otimes I). \end{align*}$$

This implies the estimate that for $x \in \mathcal {H}$

$$\begin{align*}|(f(T)-f(S))x|^2 \leq \|(B\otimes I)(1-\Delta(T)(D\otimes I))^{-1}\|^2 |\Delta(T-S)(1-(D\otimes I)\Delta(S))^{-1}(C\otimes x)|^2. \end{align*}$$

Letting $T^{(N)} = (T_1,\dots , T_N,0,\dots )$ we have $f(T^{(N)}) = f_N(T)$ and

$$\begin{align*}\begin{aligned} |(f_N(T)-f(T))x|^2 &\leq (1-\|T\|_{\infty})^{-2} |\Delta(T^{(N)}-T)(1-(D\otimes I)\Delta(T))^{-1}(C\otimes x)|^2\\ &= (1-\|T\|_{\infty})^{-2} \sum_{j=N+1}^{\infty} |(P_j\otimes T_j)(1-(D\otimes I)\Delta(T))^{-1}(C\otimes x)|^2 \end{aligned} \end{align*}$$

and this goes to $0$ as $N\to \infty $ . Thus, $f_N(T) \to f(T)$ in the strong operator topology. This concludes the proof of Theorem 1.5.

4 Proof of Theorem 1.7

That (1) implies (2) follows from Lemmas 3.1 and 3.3. Proving (2) implies (1) is a standard lurking isometry argument that is somewhat similar to our proof of Theorem 1.5. Proving (1) implies (3) consists mostly of technicalities that we discuss next. The proof of (3) implies (1) is the main contribution below.

Regarding (1) implies (3), by definition, we can make sense of $\tilde {f}(T)$ for $\tilde {f} \in \mathcal {A}_\infty $ when $(\|T_j\|)_{j\in \mathbb {N}} \in \mathrm{Ball}(c_{00})$ and Theorem 1.5 lets us make sense of it when ${(\|T_j\|)_{j\in \mathbb {N}} \in \mathrm{Ball}(\ell ^{\infty })}$ . To prove $\|\tilde {f}(T)\|\leq 1$ when T consists of matrices that are commuting, contractive, simultaneously diagonalizable with joint spectrum $\sigma (T) \subset X$ and eigenspaces of dimension at most $1$ , we can give a continuity argument. First, $\|\tilde {f}(rT)\|\leq 1$ holds for $r<1$ because we will have

$$\begin{align*}\tilde{f}_N(rT) \to \tilde{f}(rT) \end{align*}$$

in the strong operator topology as $N\to \infty $ —as in previous sections $\tilde {f}_N$ refers to the restriction of $\tilde {f}$ to $\mathbb {D}^N$ . Note that $\tilde {f}_N(rT)$ is defined in terms of the absolutely convergent power series of $\tilde {f}_N$ . Next, we use the diagonalizability properties of T; let $\vec {b}(z)$ be the eigenvector associated with joint eigenvalue $z\in \sigma (T)$ . Then, for any function $a: \sigma (T) \to \mathbb {C}$ , we have

$$\begin{align*}| \tilde{f}(rT) \sum_{z\in \sigma(T)} a(z) \vec{b}(z)| = |\sum_{z\in \sigma(T)} \tilde{f}(rz) a(z) \vec{b}(z)| \leq |\sum_{z\in \sigma(T)} a(z) \vec{b}(z)|. \end{align*}$$

We can send $r\nearrow 1$ to conclude $\|\tilde {f}(T)\| \leq 1$ . This proves (1) implies (3).

Our main contribution is the proof of (3) implies (2) assuming the finite set X belongs to $\mathbb {D}_{2}^{\infty }$ . This is a modification of the finite variable cone separation argument; the main difference being Lemma 4.1 below. Consider the following cone of functions on $X\times X$

$$\begin{align*}\begin{aligned} \mathcal{C} = \left\{ (z,w) \right.&\mapsto \sum_{j=1}^{\infty} (1-z_j\bar{w}_j) A_j(z,w): \\ & A_1, A_2,\dots \text{ are positive semi-definite functions on } X; \\ &\left. \forall z\in X, \sum_{j=1}^{\infty} A_j(z,z)<\infty\right\}. \end{aligned} \end{align*}$$

Lemma 4.1 $\mathcal {C}$ is closed.

Proof Let

$$\begin{align*}C_n(z,w) = \sum_{j=1}^{\infty} (1-z_j\bar{w}_j) A_{n,j}(z,w) \end{align*}$$

define a sequence of functions in $\mathcal {C}$ that converges to the function $C:X\times X\to \mathbb {C}$ pointwise; in particular, each $A_{n,j}$ is positive semi-definite. We must show $C \in \mathcal {C}$ .

Let $\delta :X\times X \to \mathbb {C}$ denote the function $\delta (z,w) = 1$ if $z=w$ and $\delta (z,w) = 0$ if $z\ne w$ . There necessarily exist constants $c_1, c_2$ such that for all n, $c_1 \delta \succeq C_n \succeq c_2 \delta .$ This looks more familiar when we view our functions on $X\times X$ as matrices. Since $X \subset \mathbb {D}_2^{\infty }$ we can define

$$\begin{align*}S(z,w) = \prod_{j=1}^{\infty} \frac{1}{1-z_j \bar{w}_j} \text{ and } S_j(z,w) = (1-z_j\bar{w}_j)S(z,w) = \prod_{i\ne j} \frac{1}{1-z_i \bar{w}_i}; \end{align*}$$

which are positive semi-definite and satisfy $S, S_j \succeq 1$ , with ‘ $1$ ’ representing the identically $1$ function. Some parts of what follow are similar to the proof of Lemma 3.1. Now, for any $i \in \mathbb {N}$

$$\begin{align*}c_1S \succeq C_n S = \sum_{j=1}^{\infty} S_j A_{n,j} \succeq \sum_{j=1}^{\infty} A_{n,j} \succeq A_{n,i} \succeq 0 \end{align*}$$

which shows the matrices $A_{n,i}$ are uniformly bounded. Let $c_3>0$ satisfy $c_3 \delta \succeq c_1 S$ .

Using a diagonal argument we can select a subsequence such that each $A_{n,i}$ converges to a positive semi-definite matrix $A_i$ as $n\to \infty $ . Also, for each N we have $ C_n S \succeq \sum _{j=1}^{N} S_j A_{n,j} \succeq \sum _{j=1}^{N} A_{n,j}. $ Sending $n \to \infty $ we have $ C S \succeq \sum _{j=1}^{N} S_j A_j \succeq \sum _{j=1}^{N} A_j $ and sending $N\to \infty $ we have $ C S \succeq \sum _{j=1}^{\infty } S_j A_j \succeq \sum _{j=1}^{\infty } A_j $ where the sums converge absolutely. Next, recalling $Z_j\otimes \bar {Z}_j$ denotes the function $(z,w) \mapsto z_j\bar {w}_j$ we have

$$\begin{align*}\begin{aligned} C_n &- \sum_{j=1}^{N} (1-Z_{j}\otimes \bar{Z}_j)A_{n,j} \\ &= \sum_{j=N+1}^{\infty} (1-Z_j\otimes \bar{Z}_j)A_{n,j}\\ &\succeq -c_3 \sum_{j=N+1}^{\infty} (Z_{j}\otimes \bar{Z}_j) \delta\\ &\succeq -c_3 \max\{\sum_{j=N+1}^{\infty} |z_j|^2: z \in X\} \delta. \end{aligned} \end{align*}$$

The last inequality amounts to the fact that for any function $a:X\to \mathbb {C}$ we have

$$\begin{align*}\begin{aligned} &\sum_{z,w\in X} \sum_{j=N+1}^{\infty} (z_j\bar{w}_j) \delta(z,w) a(z) \overline{a(w)} \\ &\quad = \sum_{z\in X}\sum_{j=N+1}^{\infty} |z_j|^2 |a(z)|^2 \\ &\quad\leq \max\{\sum_{j=N+1}^{\infty} |z_j|^2: z\in X\} \sum_{z\in X}|a(z)|^2. \end{aligned} \end{align*}$$

Setting $M_N = \max \{\sum _{j=N+1}^{\infty } |z_j|^2: z\in X\}$ we have $M_N \to 0$ since $X \subset \ell ^2$ . Sending $n \to \infty $ we have

$$\begin{align*}C - \sum_{j=1}^{N} (1-Z_j\otimes \bar{Z}_j)A_j \succeq -c_3 M_N \delta \end{align*}$$

and finally sending $N\to \infty $ we have

$$\begin{align*}A_0 := C - \sum_{j=1}^{\infty} (1-Z_j\otimes \bar{Z}_j)A_j \succeq 0. \end{align*}$$

Finally, $A_0$ can be absorbed into any other term, say $A_1$ as $\tilde {A}_1 = A_1 + \frac {1}{1-Z_1\otimes \bar {Z}_1} A_0$ to see that

$$\begin{align*}C = (1-Z_1\otimes \bar{Z}_1)\tilde{A}_1+ \sum_{j=2}^{\infty} (1-Z_j\otimes \bar{Z}_j)A_j \end{align*}$$

is of the desired form.

What follows is now standard. Suppose $f:X \to \overline {\mathbb {D}}$ is a function with the assumed property in item (3). Suppose the function on $X\times X$ , $F= 1 - f\otimes \bar {f}$ is not in the cone $\mathcal {C}$ . By the Hahn–Banach hyperplane separation theorem, there exists a function ${B: X\times X \to \mathbb {C}}$ with $B(z,w) = \overline {B(w,z)}$ such that

$$\begin{align*}\sum_{z,w \in X} F(z,w) B(z,w) <0 \text{ and for all } C \in \mathcal{C} \text{ we have } \sum_{z,w\in X} C(z,w) B(z,w) \geq 0. \end{align*}$$

The second condition implies $B \succeq 0$ by setting $C = (1-Z_1\otimes \bar {Z}_1) \frac { a\otimes \bar {a} }{1-Z_1\otimes \bar {Z}_1}$ for an arbitrary function $a:X\to \mathbb {C}$ and observing

$$\begin{align*}\sum_{z,w\in X} C(z,w) B(z,w) = \sum_{z,w} a(z)\overline{a(w)} B(z,w) \geq 0. \end{align*}$$

Next, we factor $B(w,z) = \overline {B(z,w)} = \vec {b}(z)^* \vec {b}(w)$ for vectors $\vec {b}(z) \in \mathbb {C}^r$ where r is the rank of the matrix $(B(z,w))_{z,w\in X}$ .

Choosing now $C_j = (1-Z_j\otimes \bar {Z}_j) (a\otimes \bar {a})$ for an arbitrary $a:X\to \mathbb {C}$

$$\begin{align*}\begin{aligned} 0&\leq \sum_{z,w \in X} (1-z_j\bar{w}_j)a(z) \overline{a(w)}B(z,w)\\ &= \sum_{z,w\in X} (1-z_j\bar{w}_j) a(z) \overline{a(w)} \vec{b}(z)^t \overline{\vec{b}(w)}\\ &= {\left|\sum_{w\in X} a(w) \vec{b}(w)\right|}^2 - {\left|\sum_{w\in X} w_j a(w) \vec{b}(w)\right|}^2 \\ \end{aligned} \end{align*}$$

which proves that maps $T_j: \sum _{w\in X} a(w) \vec {b}(w) \mapsto \sum _{w\in X} w_j a(w) \vec {b}(w)$ are well-defined, contractive, diagonalizable. Setting $T=(T_1,T_2,\dots )$ , we have assumed that $f(T)$ is contractive which means for all functions $a:X\to \mathbb {C}$

$$\begin{align*}{\left|\sum_{w\in X} a(w) \vec{b}(w)\right|}^2 - {\left|\sum_{w\in X} f(w) a(w) \vec{b}(w)\right|}^2 \geq 0 \end{align*}$$

and this means

$$\begin{align*}\sum_{z,w\in X} (1-f(z)\overline{f(w)}) B(z,w) a(z) \overline{a(w)} \geq 0 \end{align*}$$

and this means

$$\begin{align*}FB = (1- f\otimes \bar{f})B \succeq 0 \end{align*}$$

(recall $F = 1-f\otimes f$ ) and in particular

$$\begin{align*}\sum_{z,w\in X} F(z,w) B(z,w) \geq 0 \end{align*}$$

which is a contradiction. This proves $F \in \mathcal {C}$ or more precisely, there exist positive semi-definite matrices $A_1, A_2,\dots $ with rows and columns indexed by S such that

$$\begin{align*}1-f(z)\overline{f(w)} = \sum_{j=1}^{\infty} (1-z_j\bar{w}_j) A_j(z,w) \end{align*}$$

and $\sum _{j=1}^{\infty } A_j(z,z) < \infty $ . This proves that (2) implies (3) as well as the full proof of Theorem 1.7.

5 Proof of Theorem 1.9

The equivalence of (1) and (2) is straightforward based on previous results. The implication (1) implies (3) follows from Theorem 1.7.

To prove (3) implies (2), we note that our function $f:\mathbb {C}_+\to \overline {\mathbb {D}}$ gives rise to a function $F: X \to \overline {\mathbb {D}}$ where

$$\begin{align*}X = \{\pi(s) := (p_1^{-s},p_2^{-s},\dots): s\in \mathbb{C}_+\} \subset \mathrm{Ball}(c_0) \end{align*}$$

and $F(\pi (s)) = f(s)$ . We can apply the interpolation result of Theorem 1.7 as follows. Let $Y \subset \{s \in \mathbb {C}: \Re s>1/2\}$ be a countable set with a limit point in $\{s \in \mathbb {C}: \Re s>1/2\}$ . We want a limit point so that it is a determining set for holomorphic functions and we want $\Re s> 1/2$ so that $\pi (s) \in \mathbb {D}_{2}^{\infty }$ . Let $\bigcup _{n=1}^{\infty } Y_n = Y$ be an increasing union of finite sets. Set $X_n = \pi (Y_n), X_{\infty } = \pi (Y)$ .

Fix n and let $T = (T_1, T_2,\dots )$ be an infinite tuple of contractive, commuting matrices with $\sigma (T) \subset X_n$ and one-dimensional joint eigenspaces. The eigenvalues of $T_1$ are of the form $p_1^{-s} = 2^{-s}$ for $s \in Y_n$ .

We can take $M = -\log T_1/\log 2$ using the principal log and then $T_j = p_j^{-M}$ for all j. By assumption, $\|T_j\|= \|p_j^{-M}\| \leq 1$ and this extends to all natural numbers by factoring into primes: $\|n^{-M}\| \leq 1$ for all $n\in \mathbb {N}$ . By assumption (3), $\|f(M)\| = \|F(T)\| \leq 1$ . Therefore, by the implication (3) $\implies $ (1) in Theorem 1.7, there exists $G_n \in \mathcal {A}_{\infty }$ such that $\left .G_n\right |{}_{X_n} = \left .F\right |{}_{X_n}$ . In particular, $g_n(s) := G_n(\pi (s)) \in \mathscr {A}^{\infty }$ and agrees with $f_n$ on $Y_n$ .

By the Montel Theorem of Remark 3.2, since $G_n \in \mathcal {A}_{\infty } \subset \mathcal {S}_{\infty }$ , there is a subsequence of $n\in \mathbb {N}$ such that $G_n \to G \in \mathcal {S}_{\infty }$ uniformly on compact subsets of $\mathrm{Ball}(c_0)$ . The limit G necessarily agrees with F on $X_{\infty }$ . Since membership in $\mathcal {A}_{\infty }$ can be tested by checking whether G restricted to $\mathbb {D}^N$ belongs to $\mathcal {A}_N$ by Theorem 1.5, and since this in turn can be tested by examining G on finite subsets of $\mathbb {D}^N$ , we see that the limit G belongs to $\mathcal {A}_{\infty }$ since each $G_n$ belongs to $\mathcal {A}_{\infty }$ . Finally, $g = G\circ \pi \in \mathscr {A}^{\infty }$ and agrees with f on the set of uniqueness Y, so we see that $g=f \in \mathscr {A}^{\infty }$ .

6 Proof of Theorem 1.10

Given $f \in \mathscr {A}^{\infty }$ , let $F \in \mathcal {A}_{\infty }$ be the Bohr lift of f namely, $f(s) = F(p_1^{-s},p_2^{-s},\dots )$ . Again, forming $F_N(z_1,z_2,\dots ) = F(z_1,\dots , z_N,0,\dots )$ , we see that the functions $f_N(s) := F_N(p_1^{-s},p_2^{-s},\dots )$ converge uniformly to f on the half planes $\{s\in \mathbb {C}: \Re s \geq \epsilon \}$ for each $\epsilon>0$ . The power series for $F_N$ converges absolutely on $\mathbb {D}^N$ and therefore the Dirichlet series for $f_N$ converges to $f_N$ absolutely on $\mathbb {C}_{+}$ . Because of this the holomorphic functional calculus evaluation $f_N(M)$ agrees with $F_N(p_1^{-M},p_2^{-M},\dots )$ . Since $F_N \in \mathcal {A}_{\infty }$ ,

$$\begin{align*}\|f_N(M)\| = \|F_N(p_1^{-M},p_2^{-M},\dots)\| \leq 1. \end{align*}$$

Since $f_N$ converges uniformly to f on a half-plane containing $\sigma (M)$ , $f_N(M) \to f(M)$ in operator norm and therefore $\|f(M)\| \leq 1$ as desired. This completes the proof.

Acknowledgments

Thank you to the referee for a thoughtful report and for pointing out several typos. Thank you to Professor Daniel Alpay for bringing the reference [Reference Alpay and Levanony8] to our attention.

Footnotes

This research was partially supported by the NSF grant DMS-2247702.

References

Agler, J., On the representation of certain holomorphic functions defined on a polydisc . Topics inoperator theory: Ernst D. Hellingermemorial volume, Oper. Theory Adv. Appl., vol. 48, Birkhäuser, Basel, 1990, pp. 4766.Google Scholar
Agler, J. and McCarthy, J. E., Nevanlinna-Pick interpolation on the bidisk . J. Reine Angew. Math. 506(1999), 191204.Google Scholar
Agler, J. and McCarthy, J. E., Pick interpolation and Hilbert function spaces, volume 44 of Graduate Studies in Mathematics, American Mathematical Society, Providence, RI, 2002.Google Scholar
Agler, J., McCarthy, J. E., and Young, N. J., A Carathéodory theorem for the bidisk via Hilbert space methods . Math. Ann. 352(2012), no. 3, 581624.Google Scholar
Agler, J., McCarthy, J. E., and Young, N. J., Operator monotone functions and Löwner functions of several variables . Ann. Math. (2) 176(2012), no. 3, 17831826.Google Scholar
Agler, J., McCarthy, J. E., and Young, N. J., On the representation of holomorphic functions on polyhedra . Michigan Math. J. 62(2013), no. 4, 675689.Google Scholar
Agler, J., McCarthy, J. E., and Young, N., Operator analysis—Hilbert space methods in complex analysis, volume 219 of Cambridge Tracts in Mathematics, Cambridge University Press, Cambridge, 2020.Google Scholar
Alpay, D. and Levanony, D., Rational functions associated with the white noise space and related topics . Potential Anal. 29(2008), no. 2, 195220.Google Scholar
Andô, T., On a pair of commutative contractions . Acta Sci. Math. (Szeged) 24(1963), 8890.Google Scholar
Barik, S., Bhattacharjee, M., and Das, B. K., Commutant lifting in the Schur–Agler class . J. Oper. Theory 91(2024), no. 2, 399419.Google Scholar
Bhowmik, M. and Kumar, P., Factorization of functions in the Schur-Agler class related to test functions . Proc. Amer. Math. Soc. 152(2024), no. 9, 39914001.Google Scholar
Dan, H. and Guo, K., The periodic dilation completeness problem: Cyclic vectors in the Hardy space over the infinite-dimensional polydisk . J. Lond. Math. Soc. (2) 103(2021), no. 1, 134.Google Scholar
Davie, A. M. and Gamelin, T. W., A theorem on polynomial-star approximation . Proc. Amer. Math. Soc. 106(1989), no. 2, 351356.Google Scholar
Debnath, R. and Sarkar, J., Factorizations of Schur functions . Complex Anal. Oper. Theory 15(2021), no. 3, Paper No. 49. 31.Google Scholar
Defant, A., García, D., Maestre, M., and Sevilla-Peris, P., Dirichlet series and holomorphic functions in high dimensions, volume 37 of New Mathematical Monographs, Cambridge University Press, Cambridge, 2019.Google Scholar
Dritschel, M. A. and McCullough, S., Test functions, kernels, realizations and interpolation . Operator theory, structured matrices, and dilations, Theta Ser. Adv. Math., vol. 7, Theta, Bucharest, 2007, pp. 153179. MR2389623.Google Scholar
Hedenmalm, H., Lindqvist, P., and Seip, K., A Hilbert space of Dirichlet series and systems of dilated functions in ${l}^2\left(0,1\right)$ . Duke Math. J. 86(1997), no. 1, 137.Google Scholar
Kojin, K., Some relations between Schwarz-Pick inequality and von Neumann’s inequality . Complex Anal. Oper. Theory 18(2024), no. 4, Paper No. 95. 16.Google Scholar
McCarthy, J. E., Hilbert spaces of Dirichlet series and their multipliers . Trans. Amer. Math. Soc. 356(2004), no. 3, 881893.Google Scholar
Nikolski, N., In a shadow of the RH: Cyclic vectors of Hardy spaces on the Hilbert multidisc . Ann. Inst. Fourier (Grenoble) 62(2012), no. 5, 16011626.Google Scholar
Nikolski, N., Hardy spaces, volume 179 of Cambridge Studies in Advanced Mathematics, Cambridge University Press, Cambridge, French edition 2019.Google Scholar
Queffélec, H. and Queffélec, M., Diophantine approximation and Dirichlet series, volume 2 of Harish-Chandra Research Institute Lecture Notes, Hindustan Book Agency, New Delhi, 2013.Google Scholar
Varopoulos, N. T., On an inequality of von Neumann and an application of the metric theory of tensor products to operators theory . J. Funct. Anal. 16(1974), 83100.Google Scholar
von Neumann, J., Eine spektraltheorie für allgemeine operatoren eines unitären raumes . Math. Nachr. 4(1951), 258281.Google Scholar