Compact Differences of Composition Operators
on Weighted Dirichlet Spaces

We let $\displaystyle\mathbb{D}$ denote the open unit disk in the complex plane, $\displaystyle\mathbb{D}=\left\{z\in\mathbb{C}:|z|<1\right\}$, and let $\displaystyle H(\mathbb{D})$ be the space of functions analytic on $\displaystyle\mathbb{D}$. The following classical spaces of analytic functions have received much attention in the study of composition operators. The Hardy space is defined by

where $\displaystyle d\theta$ is the Lebesgue arc-length measure on the unit circle. For $\displaystyle\beta>-1$, the standard weighted Bergman space is defined by

where $\displaystyle dA$ is the Lebesgue area measure normalized so that $\displaystyle A(\mathbb{D})=1$; the Dirichlet space is given by

Recall that a reproducing kernel Hilbert space $\displaystyle\mathcal{H}$ with inner product $\displaystyle\langle\cdot,\cdot\rangle_{\mathcal{H}}$ has the property that for each $\displaystyle w\in\mathbb{D}$, there is a unique function $\displaystyle K_{w}\in\mathcal{H}$ such that

For the Hardy and weighted Bergman spaces, the kernels have a similar form:

on $\displaystyle H^{2}$ and

on $\displaystyle A_{\beta}^{2}$ with $\displaystyle\beta>-1$. On the Dirichlet space the kernel takes on a more complicated form,

where $\displaystyle\log z$ denotes the principal branch of the logarithm.

Though often convenient from the computational point of view, presenting the norms for these spaces in terms of integrals obscures the relationship between the spaces, though it is somewhat revealed in the representations of the reproducing kernels. To make the relationship more explicit, we can consider the spaces with a series norm, (equal to the norm given above for the Hardy and Dirichlet spaces, but only equivalent to the Bergman norm):

With these characterizations, we see the obvious containment relationship $\displaystyle\mathcal{D}\subset H^{2}\subset A^{2},$ but more importantly it is apparent that there are other spaces that deserve consideration. One particular space that has received more attention as of late is $\displaystyle S^{2}$ which can be defined with a series norm or an equal integral norm,

While this space is also a reproducing kernel Hilbert space, one of the first difficulties encountered in this setting is that there is not a “nice” closed form for the reproducing kernel functions with respect to this norm. The reason for this is the fact that, on $\displaystyle S^{2}$, the kernel for evaluation at $\displaystyle w$ takes the form

however we cannot identify this sum as an elementary function; for more on this, see [9]. We will discuss how to overcome this obstacle shortly.

In general, for $\displaystyle\alpha\geq-1$ we define the weighted Dirichlet space

These are all Hilbert spaces and we see that $\displaystyle H^{2}=\mathcal{D}_{1}$ with equal norm; for $\displaystyle\beta>-1$, the standard weighted Bergman space $\displaystyle A_{\beta}^{2}=\mathcal{D}_{\beta+2}$ with an equivalent norm. Also, $\displaystyle\mathcal{D}=\mathcal{D}_{0}$ and $\displaystyle S^{2}=D_{-1}$ with equal norm. Moreover, if $\displaystyle-1\leq\alpha<\beta<\infty$, $\displaystyle\mathcal{D}_{\alpha}\subset\mathcal{D}_{\beta}$ with continuous inclusion and the analytic polynomials are dense in $\displaystyle\mathcal{D}_{\alpha}$.

As is the case with $\displaystyle S^{2}=D_{-1}$, there are no nice closed forms for the reproducing kernels for $\displaystyle\mathcal{D}_{\alpha}$ when $\displaystyle-1<\alpha<0$ which is a drawback since these kernels are quite useful in the study of composition operators. In particular, if $\displaystyle\mathcal{H}$ is a functional Hilbert space of functions on the disk and $\displaystyle C_{\varphi}$ is bounded on $\displaystyle\mathcal{H}$, then for $\displaystyle w\in\mathbb{D}$, we have

To overcome this drawback, in many instances we will consider the linear functional for evaluation of the first derivative at a point in the disk; for a reference see [7] Theorem 2.16. If $\displaystyle\mathcal{H}=\mathcal{D}_{\alpha}$ for $\displaystyle\alpha\geq-1$, these functionals are bounded, and thus the Riesz Representation Theorem ([6] Theorem I.3.4) guarantees the existence of a function, denoted $\displaystyle K_{w}^{(1)}$ for $\displaystyle w\in\mathbb{D}$, such that

As with the point evaluation kernels, these kernels behave predictably under the action of the adjoint of a composition operator and it is easy to see that

In particular, we will employ these in the spaces $\displaystyle\mathcal{D}$ and $\displaystyle S^{2}$. On $\displaystyle\mathcal{D}$, we find that

whereas on $\displaystyle S^{2}$ we have

For an analytic self-map $\displaystyle\varphi$ of the unit disk, the angular derivative plays a key role in determining compactness of composition operators on many of the spaces in question. For $\displaystyle\zeta\in\partial\mathbb{D}$ and $\displaystyle M>1$, a nontangential approach region at $\displaystyle\zeta$ is defined by

and a function $\displaystyle f$ has a nontangential limit at $\displaystyle\zeta$ if $\displaystyle\lim_{z\rightarrow\zeta}f(z)$ exists in each nontangential region $\displaystyle\Gamma(\zeta,M)$. When a nontangential limit exists, we denote it by $\displaystyle\angle\lim_{z\rightarrow\zeta}f(z).$ Furthermore, if $\displaystyle\varphi$ is a self-map of the disk and $\displaystyle\zeta\in\partial\mathbb{D}$, then $\displaystyle\varphi$ has finite angular derivative at $\displaystyle\zeta$ if there exists $\displaystyle\eta\in\partial\mathbb{D}$ such that

exists as a (finite) complex value. One obvious implication of the existence of a finite angular derivative for $\displaystyle\varphi$ at $\displaystyle\zeta$ is that $\displaystyle\varphi$ has nontangential limit of modulus 1 at $\displaystyle\zeta.$ The Julia-Carathéodory Theorem provides several other implications; for a reference see [7] Theorem 2.44.

In characterizing compact differences, we will be interested in maps which have similar behavior on the boundary of $\displaystyle\mathbb{D}$. If $\displaystyle\varphi$ and $\displaystyle\psi$ are two self-maps of the unit disk both with finite angular derivative at $\displaystyle\zeta\in\partial\mathbb{D}$, then we say that the maps have the same first order data at $\displaystyle\zeta$ if $\displaystyle\varphi(\zeta)=\psi(\zeta)$ (as radial limits) and $\displaystyle\varphi^{\prime}(\zeta)=\psi^{\prime}(\zeta)$. If in addition $\displaystyle\varphi$ and $\displaystyle\psi$ are twice differentiable at $\displaystyle\zeta$ (meaning that if we consider $\displaystyle\varphi$ and $\displaystyle\psi$ as functions on $\displaystyle\mathbb{D}\cup\{\zeta\}$, then they are twice continuously differentiable) with $\displaystyle\varphi^{\prime\prime}(\zeta)=\psi^{\prime\prime}(\zeta)$, then we say that $\displaystyle\varphi$ and $\displaystyle\psi$ have the same second order data at $\displaystyle\zeta$.

Recall that a linear fractional map has the form $\displaystyle\varphi(z)=(az+b)/(cz+d)$; the condition that $\displaystyle ad-bc\neq 0$ is necessary and sufficient for such a $\displaystyle\varphi$ to be a univalent, nonconstant mapping of the Riemann sphere onto itself. Our focus here is on linear fractional self-maps of the disk and we point the reader to Chapter 0 of [20] for more information. With this narrowed focus we may assume that $\displaystyle d\neq 0$, in which case we can represent $\displaystyle\varphi$ in the form $\displaystyle\varphi(z)=(az+b)/(cz+1)$.

It is easy to see that every linear fractional self-map $\displaystyle\varphi$ of $\displaystyle\mathbb{D}$ will induce a bounded composition operator on the weighted Dirichlet spaces under consideration. This is due to the fact that the map $\displaystyle\varphi^{\prime}$ is continuous, and hence bounded, on $\displaystyle\overline{\mathbb{D}}$. Of particular interest here is the role of second order data. The following statement seems to be known but we were unable to find a proof in the literature.

Let $\displaystyle(X_{0},\|\cdot\|_{0})$ and $\displaystyle(X_{1},\|\cdot\|_{1})$ be a compatible pair of Banach spaces in the sense of Calderón (see [5]). Both $\displaystyle X_{0}$ and $\displaystyle X_{1}$ may be continuously embedded in the complex topological vector space $\displaystyle X_{0}+X_{1}$ when equipped with the norm

In addition, the space $\displaystyle X_{0}\cap X_{1}$, with norm

maps continuously into $\displaystyle X_{0}$ and $\displaystyle X_{1}$. In this note we will further assume that the space $\displaystyle X_{0}\cap X_{1}$ is dense in both $\displaystyle X_{0}$ and $\displaystyle X_{1}$ and define the interpolation algebra $\displaystyle\mathcal{I}[X_{0},X_{1}]$ to be the set of all linear operators $\displaystyle T:X_{0}\cap X_{1}\rightarrow X_{0}\cap X_{1}$ that are both 0-continuous and 1-continuous. The interpolation algebra defined above first appeared in the $\displaystyle L^{p}$-space setting in [2]; for properties and applications to the study of spectra, see [1], [10], [16], and [17].

For a Banach space $\displaystyle\mathcal{Y}$, we let $\displaystyle\mathcal{B}(\mathcal{Y})$ denote the set of all bounded operators on $\displaystyle\mathcal{Y}$. By continuity any operator $\displaystyle T\in\mathcal{I}[X_{0},X_{1}]$ induces a unique operator $\displaystyle T_{i}\in\mathcal{B}(X_{i})$, $\displaystyle i=0,1$. For $\displaystyle t\in(0,1)$, let $\displaystyle X_{t}=[X_{0},X_{1}]_{t}$ be the interpolation space obtained via Calderón’s method of complex interpolation; it follows then that $\displaystyle X_{0}\cap X_{1}$ is dense in $\displaystyle X_{t}$ and $\displaystyle T$ also induces a unique operator $\displaystyle T_{t}\in\mathcal{B}(X_{t})$ satisfying

To apply interpolation techniques to our study, we first verify that the weighted Dirichlet spaces can be interpreted as interpolation spaces. One can see this by considering these spaces as weighted $\displaystyle\ell^{2}$-spaces or by considering the techniques developed in [13]; a direct proof of this nature can be found in [16].

One appealing property of working with composition operators on these spaces in the interpolation setting is the nested behavior mentioned earlier. Specifically, $\displaystyle\mathcal{D}_{\alpha}\subset\mathcal{D}_{\beta}$, for $\displaystyle-1<\alpha<\beta$, with continuous inclusion, so $\displaystyle\mathcal{D}_{\alpha}=\mathcal{D}_{\alpha}\cap\mathcal{D}_{\beta}$. Combining this with the fact that the analytic polynomials are dense in each weighted Dirichlet space implies that $\displaystyle\mathcal{D}_{\alpha}\cap\mathcal{D}_{\beta}$ is dense in $\displaystyle\mathcal{D}_{\alpha}$ and $\displaystyle\mathcal{D}_{\beta}$. Moreover, MacCluer and Shapiro showed in [12] that boundedness of $\displaystyle C_{\varphi}$ on $\displaystyle\mathcal{D}_{\alpha}$ implies boundedness on $\displaystyle\mathcal{D}_{\beta}$ when $\displaystyle-1<\alpha<\beta.$ Thus, for our purposes, it suffices to know that $\displaystyle C_{\varphi}$ is bounded on the single endpoint space $\displaystyle\mathcal{D}_{\alpha}$. Furthermore, the fact that we are defining our operators on a dense subset of each space implies that for $\displaystyle t\in(0,1)$ the interpolated operator satisfies $\displaystyle(C_{\varphi})_{t}=C_{\varphi}$.

When using interpolated operators, the goal is to determine properties of the operator on an interpolation space $\displaystyle X_{t}$ based on properties of the operator on the endpoint spaces, or to extrapolate properties from one interpolation space to the other interpolation spaces and/or the endpoint spaces. The result that we will make use of here is Cwikel’s compactness result which extrapolates compactness on one interpolation space to the other interpolation spaces, but not necessarily to the endpoint spaces.