1
$\begingroup$

There may well be an answer to this question in a simpler category than that of finite dimensional quantum groups and in that case this question is more suitable to math.stack and I apologise in advance if this is the case.

Consider a finite dimensional quantum group $A=F(\mathbb{G})$ with a Haar state $h:A\rightarrow\mathbb{C}$. In the finite dimensional case (and in more generality in fact), this allows us to define '$p$-norms' on $A$ via

$$\|a\|_p:=\sqrt[p]{h((a^*a)^{p/2})}.$$

In particular, $\|a\|_1=h((a^*a)^{1/2})$ and $\|a\|_2=\sqrt{h(a^*a)}$. There is a Cauchy-Schwarz Inequality:

$$\|ab\|_1\leq\|a\|_2\|b\|_2.$$

We can define a dual quantum group $\hat{A}$ via the map $\mathcal{F}:A\mapsto A'$, $\mathcal{F}(a)(b)=h(ba)$. In this, finite dimensional case, $\hat{A}=A'$, and the multiplication in $\hat{A}$ is given by the convolution:

$$\nu\star\mu=(\nu\otimes\mu)\Delta,$$

the Haar state $\hat{h}:\hat{A}\rightarrow\mathbb{C}$ is given by

$$\hat{h}(\mathcal{F}(a))=\varepsilon(a),$$

and the involution is

$$\nu^*(a)=\overline{\nu(S(a)^*)}.$$

I am interested in finding bounds for

$$\|\nu\|_1=\hat{h}((\nu^*\nu)^{1/2}).$$

Via $\varepsilon\star \nu=\nu$, we have the following upper bound on $\|\nu\|_1$:

$$\|\nu\|_1=\|\varepsilon\star \nu\|_1\leq\|\varepsilon\|_2\|\nu\|_2=\sqrt{\dim A}\cdot\|\nu\|_2.$$

For my application, I have a way of calculating and bounding $\|\nu\|_2$ above but I am also interested in bounding below:

$$?\leq \|\nu\|_1\leq \sqrt{\dim A}\|\nu\|_2.$$

In the classical case where $G$ is a finite group, a 1-norm on $F(G)$ might be given by

$$\|f\|_1'=\sum_{t\in G}|f(t)|,$$

and this fits quite well into this framework:

$$\|f\|_1=h\left((f^*f)^{1/2}\right)=h(|f|)=\frac{1}{|G|}\sum_{t\in G}|f(t)|.$$

Now I am more interesting in looking at $\|\cdot\|_1:\hat{A}\rightarrow[0,\infty)$. Now classically one might define the 1-norm on $\mathbb{C}G\supset M_p(G)$ as

$$\|\nu\|_1'=\sum_{t\in G}|\nu(\delta_t)|.$$

An advantage of working with this norm is that we have

$$\|\nu\|_1'=\sup_{\underset{f\in F(G)}{\|f\|_{\infty}\leq1}}|\nu(f)|,$$

so that we can generate lower bounds by looking at test functions $\phi\in F(G)^1$ and so we have

$$|\nu(\phi)|\leq \|\nu\|_1'\leq \sqrt{|G|}\|\nu\|_2'.$$

The problem with using $\|\cdot\|_1:\hat{A}\rightarrow[0,\infty)$ via the Haar state $\hat{h}$:

$$\|\nu\|_1=\hat{h}((\nu^*\nu)^{1/2}),$$

is that even in the classical case I don't quite have something like

$$\|\nu\|_1=\frac{1}{|G|}\sum_{t\in G}|\nu(\delta_t)|.$$

When we are in $A=F(G)$ the involution is simply

$$f^*(s)=\overline{f(s)},$$ and with pointwise multiplication and positivity in the C*-algebra equivalent to positivity of the coefficients, we have

$$(f^*f)^{1/2}(s)=|f(s)|,$$

so the 1-norm works quite nicely in there.

Things are more complicated however in $\hat{A}=\mathbb{C}G$ (as is alluded to in this question of mine). In general, even for symmetric probability measures, we don't have

$$(\nu^*\nu)^{1/2}=\nu.$$

Now in the classical case I can just use $\|\cdot\|_1'$ and the ordinary C-S to get my upper bounds. However in the truly non-commutative case I want to use $\|\cdot\|_1$... if I can get myself some lower bounds! Otherwise I can just use and bound $\|\cdot \|_2$ above.

In the not-necessarily-commutative, quantum group case, is there a way to generate lower bounds on $\|\nu\|_1=\hat{h}((\nu^*\nu)^{1/2})$ via 'test functions': $$\|\nu\|_1\geq\sup_{s\in S}F(s,\nu)?$$ Perhaps $F(s,\nu)$ involves convolving $\nu$ with some element of $s\in S\subset\hat{A}$ with a small norm or maybe something like hitting $\nu$ with an element of $s\in S\subset A$ with small norm: $\nu(s)$.

Thank you for your help.

$\endgroup$
  • $\begingroup$ I find your question somewhat unclear. Are you merely asking for a way to express the noncommutative L^1-norm given by a tracial state as a supremum using the natural pairing? (Your definition of the 1-norm seems to be the usual one if the Haar state is tracial, as it is for Kac examples, but I am not sure it is the correct definition in the non-tracial case) $\endgroup$ – Yemon Choi Jun 27 '15 at 13:58
  • $\begingroup$ Yes to expressing it as a supremum --- or even greater than a supremum. You can assume Kac (and if this isn't enough tracial also). $\endgroup$ – JP McCarthy Jun 27 '15 at 14:01
  • 1
    $\begingroup$ Hmm, well perhaps I have misunderstood your question, but if $\tau$ is a faithful normal trace on a von Neumann algebra $M$, then IIRC $\tau((x^*x)^{1/2})$ is equal to the supremum of $|\tau(xy)|$ as $y$ runs over all elements in unit ball of M $\endgroup$ – Yemon Choi Jun 27 '15 at 14:11
  • $\begingroup$ Well cf. my first paragraph... have you got a reference or proof of this? $\endgroup$ – JP McCarthy Jun 27 '15 at 16:58
  • 1
    $\begingroup$ OK, I've cobbled something together. Apologies for any earlier confusion $\endgroup$ – Yemon Choi Jun 29 '15 at 11:55
1
+50
$\begingroup$

In a more general setting than that of the original question: suppose we have a faithful normal state $h$ on a von Neumann algebra $M$. Suppose furthermore that $h$ is tracial, meaning that $h(xy)=h(yx)$ for all $x,y\in M$. (Warning! there are important examples of compact quantum groups where the Haar state is faithful but not tracial.)

In this setting we may define $\Vert x\Vert_{L^1(M,h)}$ to be $h(|x|)$. With this definition it is not clear that we have a norm; however, it is known that one has $$ h(|x|) = \sup\{ | h(xy) | \colon y \in M, \Vert y\Vert_M \leq 1 \} \tag{$*$}$$ and the proof has been given by Martin Argerami on MathStackExchange. The motivating example to keep in mind is the commutative case $M=L^\infty[0,1]$ with usual ess.sup norm and usual weak-star topology, with $h(x) = \int_0^1 x(t)\,dt$.

Historical note: if one looks at Definition 3.2 of I. Segal's paper

MR0054864 (14,991f) I. E. Segal, A non-commutative extension of abstract integration. Ann. of Math. (2) 57 (1953). 401–457

one sees that Segal took the RHS of $(*)$ as the definition of the $L^1$-norm on $M$ given by $h$. He then, in part (d) of Corollary 10.1, shows that the RHS of $(*)$ is equal to the LHS of $(*)$.

$\endgroup$

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.