3
$\begingroup$

I have a $C^*$-algebra $\mathcal{A}$, and would like to make use of the spectral order $\preceq$ coming from (the self-adjoint part of) its enveloping von Neumann algebra $\mathcal{A}^{**}$.

I am most interested in checking that the spectral join/meet of finite subsets of $\mathcal{A}^\text{sa}$ are contained in $\mathcal{A}^\text{sa}$. It suffices to check for pairs only.

Given self-adjoint $a, b \in \mathcal{A}^{**}$, the spectral join $a \vee b$ turns out to be the self-adjoint element of $\mathcal{A}^{**}$ whose spectral projections satisfy $$\chi_{(\lambda, \infty)}(a \vee b) = \chi_{(\lambda, \infty)}(a) \vee \chi_{(\lambda, \infty)}(b),$$ where $\chi_S$ denotes the indicator function of a subset $S \subset \mathbb{R}$, and the $\vee$ on the right is the standard join for projections of a von Neumann algebra.

If we restrict $a$ and $b$ to be elements of $\mathcal{A}$, then the two projections on the right above are open (in the sense described by Akemann in The general Stone-Weierstrass problem, 1969), and hence so is the one on the left (which is necessary for inclusion of $a \vee b$ in $\mathcal{A}$). However, I am not sure how to guarantee that $a \vee b \in \mathcal{A}$.

Would anyone have a reference or proof relevant to my objective? If it happens that $a \vee b$ is not an element of $\mathcal{A}$, might $a$ and $b$ still have a larger join in the smaller poset $(\mathcal{A}, \preceq|_{\mathcal{A}})$? I am fairly new to the operator algebra world, so I apologise if this is a standard result that I just haven't been able to track down in the literature.

$\endgroup$

1 Answer 1

4
$\begingroup$

Sorry, the answer is no. In general the spectral join of two positive operators $a$ and $b$ is the strong operator limit of $(a^k + b^k)^{1/k}$ as $k \to \infty$ (Corollary 10 of this old old paper of mine). But this need not belong to the C${}^*$-algebra generated by $a$ and $b$. For a counterexample, work in $l^\infty(\mathbb{N}, M_2)$ acting on the $l^2$ direct sum $\mathbb{C}^2 \oplus \mathbb{C}^2 \oplus \cdots$ and let $p$ and $q$ be the projections with $p(n) = \left[\begin{matrix}1&0\cr 0&0\end{matrix}\right]$ and $q(n) = \left[\begin{matrix}\cos^2(1/n)&\cos(1/n)\sin(1/n)\cr \cos(1/n)\sin(1/n)&\sin^2(1/n)\end{matrix}\right]$ for all $n$. The spectral join of two projections is just their join as projections, i.e., the projection onto the join of their ranges, and the join of the ranges of $p$ and $q$ contains each $\mathbb{C}^2$ summand, so it is the whole space and thus $p \vee q$ is the identity. But every element $r$ of the C*-algebra generated by $p$ and $q$ has $(2,2)$ entries going to zero as $n \to \infty$ (since this is true of any monomial in $p$ and $q$), so this C${}^*$-algebra doesn't contain the identity.

Any self-adjoint operator that is spectrally larger than the identity has null spectral projection for the interval $(-\infty, 1/2]$ and is therefore invertible, so it can't belong to the C${}^*$ algebra generated by $p$ and $q$. So that falsifies your final question too.

$\endgroup$
2
  • $\begingroup$ Hi Nik, thanks for the very comprehensive answer! Thanks also for linking your old paper, I had been trying to find different explicit formulae for various kinds of non-commutative "join", so it's helpful to know that one works for the spectral order. $\endgroup$
    – Sean
    Jun 26, 2022 at 23:32
  • $\begingroup$ No problem. It was fun to look at my old paper and relearn some of this ... $\endgroup$
    – Nik Weaver
    Jun 26, 2022 at 23:57

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.