4
$\begingroup$

In a class talking about $C^*$ algebra and (higher) index theory, I heard a theorem (related to Kaplansky, proved?), that is

Suppose $\Gamma$ is a group (admitting Haar measure if necessary) while $\mathbb{C} \Gamma := \widehat{C_c(\Gamma)}$ is the group $C^*$ algebra of $\Gamma$ ($\widehat{(-)}$ means completion with largest $C^*$-norm $\|f\|:= \sup_{\pi \, \text{a Repn of} \, L^1(\Gamma)} \|\pi(f)\|_{H_{\pi}}$).

Then if $a,b \in \mathbb{C} \Gamma, ab = 1$ one derives $ba = 1$.

I want to know if what he said is true. If true, how to prove?

The speaker also remarked that the $C^*$ algebra condition is necessary. There are counterexamples for usual group algebra $F\Gamma$ where $F$ a field and $\Gamma$ as above. How to find one?


I also have few knowledge about Kaplansky's work on noncommutative ring/module theory. And His Unit Conjecture goes

For $\mathbb{F}$ a field and $\Gamma$ a group, $a \in \mathbb{F} \Gamma^{\times}$ iff $a \in \Gamma$.

Any comments will be helpful.

$\endgroup$
3
  • $\begingroup$ You have written ${\mathbb C}\Gamma$ for the full group ${\rm C}^\ast$-algebra - are you sure that this is what the speaker meant? It seems odd to say that "the most astonishing thing is that the only way to prove it is operator algebra theory" if one was referring already to a ${\rm C}^\ast$-algebra ... $\endgroup$
    – Yemon Choi
    Commented Sep 16, 2023 at 3:09
  • $\begingroup$ Also, I think that you (or the speaker) may have been mistaken about the precise assumptions on $\Gamma$. You talk of $\Gamma$ having a Haar measure which suggests that you are allowing ${\rm C}^\ast$-algebras of not-necessarily discrete groups. But if $G$ is not discrete then ${\rm C}^\ast(G)$ does not have an identity element! $\endgroup$
    – Yemon Choi
    Commented Sep 16, 2023 at 3:14
  • $\begingroup$ @YemonChoi Thanks your correction. the Haar measure is about definition of Cstar norm. I should be more careful. I am not familiar with Kaplanskys work. If admitting, you can help modify this question. $\endgroup$
    – YOTAL
    Commented Sep 16, 2023 at 15:15

1 Answer 1

8
$\begingroup$

This is not accurate as far as discrete groups are concerned. First of all it is an open question, called Kaplansky's direct finiteness conjecture, whether every group ring over a field has the property that $ab=1$ implies $ba=1$ (i.e. is directly finite). The best result to date is due to Elek and Szabo who proved if $G$ is a sofic group (this includes Gardam's example from the unit conjecture), then $KG$ is directly finite for any field $K$. It is an open question whether every group is sofic.

It's also not true that $C^*$-algebras are needed in characteristic 0 although all known proofs to the best of my knowledge use at least some "analysis". Passman gave a proof (you can find it in his book The Algebraic Theory of Group Rings) using just basic epsilon-delta stuff inside of $\mathbb CG$ without going to the completion. The journal reference is Passman, D. S., Idempotents in group rings. Proc. Amer. Math. Soc. 28 (1971), 371–374.

To give some details, the key thing need to prove that $\mathbb CG$ is directly finite is to show that the trace of any nonzero idempotent is positive. The trace here is the coefficient of $1$. Kaplansky uses that the group von Neumann algebra has a faithful trace extending this trace (it is enough that the reduced $C^*$-algebra have such a trace). Then Kaplansky uses that in a $C^*$-algebra, every idempotent is equivalent to projection (i.e., $e=ab$ and $p=ba$ with $p$ a projection). Since a faithful trace is positive on nonzero projections, it must be positive on nonzero idempotents. The argument uses that in a $C^*$-algebra, $1+aa^*$ is always invertible (by spectral theory). The group algebra is a $*$-algebra, but it is not complete and $1+aa^*$ is not always (or even usually) invertible. Passman gives a direct proof that the trace of any nonzero idempotent of $\mathbb CG$ is positive via elementary epsilon-delta arguments.

Added. In fact if a group has a directly finite algebra over all finite fields, then it has a directly finite algebra over every field and hence for sofic groups, operator algebras and analysis are not needed at all (the proof for sofic groups is not analytic). If all groups are sofic, analysis is not needed.

The reduction to finite fields is standard. If $a,b\in KG$ with $ab=1$ but $ba\neq 1$, then there are only finitely many coefficients appearing in $a,b$ and so we can find a finitely generated (over $\mathbb Z$) subring $R$ of $K$ with $RG$ containing $a,b$. By a version of nullstellensatz, since $R$ is a finitely generated integral domain the intersection of the maximal ideals of $R$ is trivial and each quotient of $R$ by a maximal ideal is a finite field. Since $0\neq ab-ba$, we can find a finite field $F$ which is an image of $R$ in which the image of $ab-ba$ under $RG\to FG$ is nonzero. Then $FG$ is not directly finite.

$\endgroup$
6
  • $\begingroup$ I have never been quite convinced that the Passman proof is really avoiding "Hilbert space operator theory/operator algebras", since it does involve the $\ell^2$-norm. I wonder what goes wrong if one tries to use the $\ell^1$-norm? $\endgroup$
    – Yemon Choi
    Commented Sep 16, 2023 at 3:07
  • $\begingroup$ @YemonChoi, my memory is the proof uses both norms. Lemma 1v uses l1 crucially and it's important in the proof. But no operator theory or spectral theory is used. $\endgroup$ Commented Sep 16, 2023 at 14:31
  • 1
    $\begingroup$ @YemonChoi I accidentally deleted my comment while trying to add another. My paper arxiv.org/pdf/2207.11194.pdf axiomatize Passman's proof in Section 6 to *-algebras and the l1-norm is the more important norm. Basically if you have a faithful trace such that the norm induced by the trace plays nicely with the l1-norn then you can get idempotents have positive trace. For the group algebra the norm induced by the trace is the l2-norm but for inverse semigroups with weirdly constructed traces it need not be. Maybe you want to say any norm from a faithful trace is operator algebras. $\endgroup$ Commented Sep 16, 2023 at 15:56
  • $\begingroup$ @YemonChoi (ctd) but these traces are very kind of like Frobenius forms in algebra $\endgroup$ Commented Sep 16, 2023 at 15:57
  • 2
    $\begingroup$ @YemonChoi, I agree the idea is to approximate the orthogonal projection to the range of an idempotent for an inner product coming from a faithful trace. His proof doesn't need the completeness. The l1-norm still surfaces in a way as well. In any event he avoids taking completions and spectral theory. In my paper there seemed possible issues extending the trace to the inverse semigroup C*-algebra when the associated groupoid is non-Hausdorff and had noncompact unit space so I had to stay in the complex algebra $\endgroup$ Commented Sep 18, 2023 at 22:41

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .