Let $E$ be a Hilbert module over a $C^*$-algebra $A$. Let $T\colon E\to E$ be a densely defined, unbounded $A$-linear operator. (In particular, the initial domain of $T$ is an $A$-submodule of $E$.) If the operators $T\pm i$ each has dense range, then $T$ is an essentially self-adjoint, regular operator on $E$, and there is a continuous functional calculus

$$F_T\colon C_b(\mathbb{R})\to\mathcal{B}_A(E),$$

where $C_b(\mathbb{R})$ denotes the bounded continuous $\mathbb{C}$-valued functions on $\mathbb{R}$, and $\mathcal{B}_A(E)$ denotes the $C^*$-algebra of bounded adjointable $A$-linear operators on $E$.

Question: Suppose $D(T)$ is not an $A$-submodule of $E$, but only an $A_0$-submodule, where $A_0$ is a (metrically) dense $*$-subalgebra of $A$. Then, still assuming that $T\pm i$ each has dense range, is it possible to construct a continuous functional calculus for $T$?

Comment: It seems to me that even though in this situation $T$ is only $A_0$-linear, the operator $1+T^*T$ is injective, and that $(1+T^*T)^{-1}$ extends to an $A$-linear operator. Similarly, it seems possible to make sense of the operator $T(1+T^*T)^{-1/2}$ as being $A$-linear.

  • $\begingroup$ This is Woronowicz's "z transform", right? What source are you following for this material? $\endgroup$ – Matthew Daws Feb 4 at 9:47
  • 2
    $\begingroup$ @MatthewDaws I'm looking at Lance's book, which assumes that $D(T)$ is closed under the action of $A$. I think the operator $T(1+T^*T)^{-1/2}$ is the $z$-transform of $T$. I guess what I was speculating in the comment to the question was that even if $T$ is only $A_0$-linear, then it has an $A$-linear $z$-transform, so its functional calculus can be developed. $\endgroup$ – geometricK Feb 4 at 10:10

I think you need to be very careful about hypotheses. Looking in Lance's book, we have

Lemma 9.8: Suppose $t:E\rightarrow E$ is densely-defined and self-adjoint. Then $t$ is regular if and only if the operators $t\pm i$ are surjective.

(Lance's notation $t:E\rightarrow F$ means that $t$ is defined on $D(t)$ a submodule of $E$.) Notice here that $t$ needs to be self-adjoint, which you don't mention in your question. I am afraid I cannot find anything in Lance where $t\pm i$ merely have dense range is sufficient.

Suppose we have $t$ where $D(t)$ is only a dense $A_0$-submodule. What is $t^\ast$? Well, the definition is $$ D(t^\ast) = \{ y\in E : \exists\, z\in E, \ (tx|y) = (x|z) \ (x\in D(t) \} $$ with $t^\ast(y) = z$. Suppose $y\in D(t^\ast)$ with $t^\ast(y) = z$. For $a\in A$ we have $$ (x|za) = (x|z)a = (tx|y)a = (tx|ya) \qquad(x\in D(t), $$ just using $A$-linearity of the inner-product. So $ya\in D(t^\ast)$ and $t^\ast(ya) = za = t^\ast(y)a$. So $D(t^\ast)$ is an $A$-submodule, and $t^\ast$ is $A$-linear.

Thus, if $t=t^*$, then $D(t)$ is already an $A$-submodule! We could go back to the definition of $t$ being "regular", but that requires knowing that $1+t^*t$ has dense range. I suspect in any application, having the weaker hypothesis that $D(t)$ is only an $A_0$-submodule will not be an aid in showing $1+t^*t$ has dense range.

  • $\begingroup$ Thanks for the answer! I should have included that $t$ is symmetric in the hypotheses. Let me ponder this a little more, but I think you are right that knowing $D(t)$ is only an $A_0$-submodule won't help show $1+t^*t$ has dense range. However, if one happened to know that $1+t^*t$ has dense range, then this should be sufficient to define functional calculus of $t$, even in the $A_0$-submodule case. Here one won't expect to recover $t$ from its $z$-transform like in the usual case. I think the functional calculus formed in this way should just be the functional calculus of $t^{**}$. $\endgroup$ – geometricK Feb 4 at 11:29
  • $\begingroup$ Be careful with the difference between "symmetric" and "self-adjoint". We definitely need self-adjoint here. $\endgroup$ – Matthew Daws Feb 4 at 11:50
  • $\begingroup$ Right, I should have said closed and symmetric, which together with the density-of-range assumption implies self-adjoint. On the other hand, all symmetric operators are closable. $\endgroup$ – geometricK Feb 4 at 12:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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