Let $N\in\mathbb{N}$ and suppose $A$ is a unital C$^{*}$-algebra with the property that each irreducible representation of $A$ has dimension $N$. I.e., $A$ is $N$-homogeneous.

Suppose we have an element $a\in A$ such that $\varphi(a)$ is invertible in $M_{N}(\mathbb{C})$ for each irreducible representation $\varphi$ of $A$. Does it necessarily follow that $a$ will be invertible in $A$?

This is certainly true if $A$ is an algebra of the form $C(X,M_{N}(\mathbb{C}))$ (since the irreducible representations are equivalent to point evaluations), but what about more general homogeneous algebras?


If $A$ is separable, then any representation is (approximately unitarily equivalent to) a direct sum of irreps. So take a faithful representation $\pi:A\to B(H)$, then $\pi\sim\bigoplus_j\pi_j$. Then \begin{align} \sigma(a)&=\sigma(\pi(a))=\sigma(\bigoplus_j\pi_j(a))=\{\lambda:\ \bigoplus_j\pi_j(a)-\lambda I\ \text{ is not invertible }\}\\ \ \\ &=\{\lambda:\ \exists j, \pi_j(a-\lambda I)\ \text{ is not invertible }\}\\ \ \\ &=\bigcup_j\sigma(\pi_j(a)). \end{align} As $\pi_j(a)$ is invertible for all $j$, $0\not\in\sigma(a)$, and $a$ is invertible.

The fact that $\sigma(\pi(a))=\sigma(\bigoplus_j\pi_j(a))$ depends on the fact that approximate unitary equivalence preserves the spectrum. This in turn depends on the fact that approximate unitary equivalence preserves invertibility, since $a=\lim_nu_nbu_n^*$, then $a-\lambda I=\lim_nu_n(b-\lambda I)u_n^*$.

Now, let us show that if $a=\lim_n u_nbu_n^*$ and $b$ is invertible, then $a$ is invertible (which in turn, as mentioned, shows that approximate unitary equivalence preserves the spectrum). Indeed, note that $a^*a=\lim_n u_nb^*b u_n$ and that $a$ is invertible if and only if both $a^*a$ and $aa^*$ are invertible, so it is enough to check that approximate unitary equivalence preserves invertibility of positives. But this is easy: if $b\geq0$ and invertible, there exists $c>0$ with $b-cI\geq0$. But then $a-cI\geq0$, and so $a$ is invertible.

  • $\begingroup$ Assuming $A$ is simple, is it true that we can only conclude that $\pi$ is approximately unitarily equivalent to a direct sum of irreducibles? If so wouldn't this only imply that $\pi(a)$ is in the closure of the invertibles in $\pi(A)$? Is it possible that since $A$ is $N$-homogeneous, we don't have to worry about the approximately unitarily equivalent part, and that, in fact, $\pi$ does decompose as a direct sum of irreducibles? $\endgroup$ – ervx Oct 27 '18 at 18:03
  • $\begingroup$ @ervx: you are right that I need to address that in the answer (just done it). But it doesn't change things, as approximate unitary equivalence preserves invertibility. $\endgroup$ – Martin Argerami Oct 27 '18 at 20:14
  • $\begingroup$ Thank you very much for taking the time to clarify this. $\endgroup$ – ervx Oct 30 '18 at 12:57

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.