The following question looks simple, but the answer is not obvious for me:

Let $S$ be a $*$-algebra and $\left\Vert \cdot \right\Vert _{1}$, $\left\Vert \cdot \right\Vert _{2}$ $C^*$-norms on $S$ with $\left\Vert \cdot \right\Vert _{1} \leq \left\Vert \cdot \right\Vert _{2}$. Denote by $A$, $B$ the generated $C^*$-algebras. If a sequence $(x_n)_{n \in \mathbb{N}}$ is convergent in both norms with limit $0 \in A$ and $b\in B$, does this already imply $b=0$?

  • 2
    $\begingroup$ Robert Furber has given a nice commutative example, but it may also be worth noting a kind of canonical example. Take $S$ to be the convolution algebra of a (discrete) non-amenable group $G$, take $\Vert\cdot\Vert_1$ to be the norm induced by the left regular representation of $S$ acting on $\ell^2(G)$, and take $\Vert\cdot\Vert_2$ to be the full group $C^*$-norm $\endgroup$
    – Yemon Choi
    Oct 17, 2018 at 22:45

1 Answer 1


The answer is no (to the main question, not the title). Consider the *-algebra $S$ of *-polynomials generated by one variable $z$ such that $zz^* = z^*z$, i.e. the free commutative *-algebra on one generator. Each $a \in S$ can be considered to be a continuous function $\mathbb{C} \rightarrow \mathbb{C}$, so we can define $$ \|a\|_1 = \sup \{ a(x) \mid x \in [0,1] \} \\ \|a\|_2 = \sup \{ a(x) \mid x \in [0,2] \} $$ Since $[0,1] \subseteq [0,2]$, it is clear that $\|\cdot\|_1 \leq \|\cdot\|_2$. By the Stone-Weierstrass theorem, the completion of $S$ in $\|\cdot\|_1$ is $C([0,1])$ and the completion of $\|\cdot\|_2$ is $C([0,2])$.

So we can take a continuous function $b : [0,2] \rightarrow \mathbb{C}$ that vanishes on $[0,1]$ but takes a non-zero value in $(1,2]$, and a sequence $(a_i)_i$ in $S$ such that $a_i \to b$ with respect to $\|\cdot\|_2$. Then $$ \|a_i\|_1 = \|a_i-b\|_1 \leq \|a_i - b\|_2 \to 0 $$ So $(a_i)_i$ is convergent with respect to both norms and has limit $0$ in $\|\cdot\|_1$, but its limit in $\|\cdot\|_2$ is $b \neq 0$.

  • $\begingroup$ This is a nice answer! $\endgroup$
    – Nik Weaver
    Oct 17, 2018 at 15:45
  • $\begingroup$ @NikWeaver This is something I struggled a long time to convince myself about when I first learnt that completion does not preserve injectivity. $\endgroup$ Oct 17, 2018 at 15:46
  • $\begingroup$ @Robert do you mean injectivity of functions or of algebras? $\endgroup$
    – David Roberts
    Oct 17, 2018 at 20:45
  • 1
    $\begingroup$ @DavidRoberts Now you mention it, that is ambiguous. I meant functions. In this case it is that the completion functor applied to the identity map $(S,\|\cdot\|_2) \rightarrow (S,\|\cdot\|_1)$ produces a non-injective map $B \rightarrow A$ (in fact, it is the quotient map for the ideal of functions vanishing on $[0,1]$). $\endgroup$ Oct 17, 2018 at 20:54

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.