I am reading Agler and McCarthy's Pick Interpolation and Hilbert Function spaces. In Chapter 9, the authors ask to observe that weakly separated in a Reproducing kernel hilbert space implies separated in the Gleason metric. This is not immediate to me however.

Here are the definitions:

Fix a Hilbert function space $\mathcal H_k$ on $X$. The sequence $\Lambda = \{ \lambda_i \}_{i=1}^{\infty}$ in $X$ is weakly separated if there exist a constant $\varepsilon > 0$ such that whenever $i\ne j$, there is some function $\varphi _{ij}$ in the closed unit ball of the multiplier algebra $\mathcal H _k$ such that $\varphi _{ij} (\lambda_i) = \varepsilon$ and $\varphi _{ij} (\lambda _j) = 0$.

Also, the sequence $\Lambda$ in $X$ is said to be $d$-separated if there is some constant $\varepsilon > 0$ such that for $i\ne j$ we have that $d (\lambda _i , \lambda _j) > \varepsilon$. Note that the (pseudo)metric $d$ on $X$ is given by \begin{align*} d(\lambda _1 , \lambda_2) = \sqrt{ 1- \frac{\lvert k(\lambda_1 , \lambda _2)\rvert ^2}{k(\lambda_1 , \lambda_1 )k(\lambda_2 , \lambda_2 )}} \end{align*} where $\lambda_1 , \lambda_2 \in X$.

Here's my attempt: Let $\Lambda = \{ \lambda_i \}_{i=1}^{\infty}$ in $X$ be weakly separated. It suffices to show that \begin{align*} \sup \left\{ \frac{\lvert k(\lambda_i , \lambda _j)\rvert ^2}{k(\lambda_i , \lambda_i )k(\lambda_j , \lambda_j )} : \lambda_i \ne \lambda_j \right\} <1 \end{align*} in order to show that $\Lambda$ is $d$-separated. Since $\Lambda$ is weakly separated, there is some $\varepsilon_0 > 0$ and for each $i\ne j$, there is a function $\varphi_{ij}$ such that in the closed unit ball of the multiplier algebra $\mathcal H _k$ such that $\varphi _{ij} (\lambda_i) = \varepsilon_0$ and $\varphi _{ij} (\lambda _j) = 0$. Note that $\varepsilon_0 \le 1$ since $\varepsilon_0 =\lvert \varphi_{ij} (\lambda_i)\rvert \le \lVert \varphi_{ij} \rVert_{\infty} \le \lVert \varphi \rVert _{\mathcal M (\mathcal H)} \le 1$.

Now, note that for $i \ne j$, we have that \begin{align*} \varepsilon_0 ^2 \lvert k(\lambda_i , \lambda_j) \rvert &=\lvert{ \langle \epsilon_0 k_{ \lambda_j}, \epsilon_0 k_{ \lambda_i} \rangle}\rvert \\ &= \lvert \langle \varphi_{ji}(\lambda_j ) k_{\lambda_j}, \varphi_{ij}(\lambda_i ) k_{\lambda_i} \rangle \rvert \\ &= \lvert \langle {M_{\varphi_{ji}}}^* k_{\lambda_j}, {M_{\varphi_{ij}}}^* k_{\lambda_i} \rangle \rvert \\ &\le \lVert k_{\lambda _j} \rVert \lVert k_{\lambda _i} \rVert. \end{align*} This implies that \begin{align*} \frac{\lvert k(\lambda_i , \lambda _j)\rvert ^2}{k(\lambda_i , \lambda_i )k(\lambda_j , \lambda_j )} \le \frac{1}{\varepsilon_0 ^4}. \end{align*}

But $\frac{1}{\varepsilon _0 ^4} \ge 1$ so I do not get what I desire. I also realise that I have failed to use that fact that $\varphi_{ij} (\lambda_j) = 0$. Directions on proving this will be appreciated! $\ddot \smile$


1 Answer 1


If a sequence is weakly separated, i.e. there exists a multiplier $\varphi_{ij}$ of multiplier norm at most one such that $\varphi_{ij}(\lambda_i)=\varepsilon, \varphi_{ij}(\lambda_j)=0$, then necessarily the $2\times2$ Pick matrix associated to this $2$-point interpolation problem is positive semidefinite; \begin{equation*} \begin{bmatrix} (1-\varepsilon^2)k(\lambda_i,\lambda_i) & k(\lambda_i,\lambda_j)\\ k(\lambda_i,\lambda_j) & k(\lambda_j,\lambda_j) \end{bmatrix} \geq0. \end{equation*} In particular its determinant has to be non negative, that is \begin{equation*} (1-\varepsilon^2)k(\lambda_i,\lambda_i)k(\lambda_j,\lambda_j)-|k(\lambda_i,\lambda_j)|^2 \geq0 \end{equation*} which after rearranging the terms gives $d$-separation.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.