I might have thought about this problem a little longer but am quite confused so said I would put this question to the good people here...

Consider a finite group $G$ or rather the algebra of functions on $G$, $\mathbb{C}^G=F(G)$ with multiplication in $F(G)$ defined pointwise. We have an involution on $F(G)$:


If we take the supremum norm on $F(G)$:

$$\|f\|_\infty=\max_{g\in G}|f(g)|,$$

we have a C*-algebra. The positive elements of $F(G)$ are those of the form $f^*f$ and so those that have positive coefficients with respect to the basis $\{\delta_s:s\in G\}$ are the positive functions.

Denote by $\mathbb{C}G$ the linear functionals on $F(G)$. This set is spanned by elements dual to the $\{\delta^s:s\in G\}$:

$$\delta^s(\delta_t)=\delta_{t,s},$$ where the latter $\delta$ is the Kronecker delta. It is not difficult to show that the positive linear functionals of $\mathbb{C}(G)$ are those with positive coefficients with respect to the $\{\delta^s:s\in G\}$ basis.

So for example, if $G=\{\mathbb{Z}_3,\,\oplus_3\}$ then $\displaystyle \nu=\frac{\delta^1+\delta^2}{2}\in\mathbb{C}G$ is a positive linear functional.

However one can also give $\mathbb{C}G$ the structure of a C*-algebra. Extend by linearity the convolution product $$(\delta^s,\delta^t)\mapsto \delta^{st};\qquad\delta^s\star\delta^t=\delta^{st},$$ to the whole of $\mathbb{C}G$. My understanding is that there are C*-norms on $\mathbb{C}G$ when equipped with the involution:


Now positivity in this C*-algebra is not the same as positivity as a linear functional. In particular it is not difficult to show that there is no $\mu\in\mathbb{C}\mathbb{Z}_3$ such that

$$\mu^*\star \mu=\frac{\delta^1+\delta^2}{2}=:\nu.$$

Therefore, as an element of the C*-algebra $\mathbb{C}\mathbb{Z}_3$, $\nu$ is not positive.

Is there an easy way of recognising if an element of the C*-algebra $\mathbb{C}G$ is positive?

Random thoughts... is it necessary for $\nu(\delta_e)\neq0$, symmetry plays a role? $\nu(\delta_s)=\nu(\delta_{s^{-1}})$... have we any positive elements of the algebra which are not positive functionals. Alternatively I need to look at self-adjoint --- $\nu(\delta_s)=\overline{\nu(\delta_{s^{-1}})}$ --- and positive spectrum... can I find the spectrum of an element of $\mathbb{C}G$?

Thank you.


Consider the space $\mathbb{C}G$ as a Hilbert space with orthonormal basis $e_g$. $G$ acts on it by left multiplication which are isometries. $$L_g \colon h \mapsto g h$$ Hence the adjoint of the map $L_g$ is $L_{g}^{-1} = L_{g^{-1}}$.

Now $\mathbb{C}G$ acts on $\mathbb{C}G$ by left multiplication so we can look at $\mathbb{C}G$ as an algebra of linear operators on the Hilberts spaced $\mathbb{C}G$. This algebra is closed under $*$, since $$\left(\sum a_g g\right)^* = \sum \bar a_g g^{-1}$$ so it's a $*$ subalgebra of $End(\mathbb{C}G)$, hence a $C^{*}$ algebra. Now, the positivity is preserved under the imbeddings of $C^{*}$ algebras. Now, a basic fact about $C^{*}$ algebras is that an element of $\mathbb{C}G$ is positive, if and only if its image in $End(\mathbb{C}G)$ is positive. So you consider the corresponding $|G|$ dimensional matrix check whether that one is positive.

In general, an element $\sum a_g g$ will have the corresponding matrix \begin{eqnarray} \sum a_g g &\mapsto& (x_{gh})_{g, h \in G}\\ x_{gh} &=& a_{g h^{-1}} \end{eqnarray} a so called group matrix.

Consider an element of the group algebra $\mathbb{C} ( \mathbb{Z}/3)$ $$\alpha = a_0 e_0 + a_1 e_1 + a_2 e_2$$ with corresponding matrix $$ \left( \begin{array}{ccc} a_0 & a_2 & a_1 \\ a_1 & a_0 & a_2\\ a_2 & a_1 & a_0 \end{array} \right ) $$ The element $\alpha$ and corresponding matrix are selfadjoint if and only if $a_2 = \bar a_1$ and $a_0 \in \mathbb{R}$. So take $\alpha = 1 \cdot e_1 + 1 \cdot e_2$. The corresponding matrix $$ \left( \begin{array}{ccc} 0 & 1 & 1 \\ 1 & 0 & 1\\ 1 & 1& 0 \end{array} \right ) $$ is not positive definite so neither is $\alpha$.

One can do this in general for other groups.

The idea: Map your $C^{*}$ algebra into a matrix algebra and test positivity there.

You can also map $\mathbb{C}G$ isomorphically to a product of matrix algebras ( representation of finite groups ) and test positivity on each component. Eg: for $G = \mathbb{Z}/n$, a self-adjoint matrix $(a_{m-n})$ is positive if and only if all the sums $\sum a_n \omega^{n}$ are positive, where $\omega$ is an $n-th$ root of $1$ ( these are the eigenvalues of this circular matrix).


Let's mention that every injective morphism of $C^{*}$ algebras preserves the norms. The way it goes with $\mathbb{C} G \subset End (\mathbb{C}G)$ that's clear Now let's assume that some $\alpha$ from $\mathbb{C} G$ has a square root in $End (\mathbb{C}G)$. Then that square root is the limit of a sequence of form $r_{n+1} = \frac{1}{2} ( \alpha + \frac{r_n}{\alpha})$, $r_0 = 1$, so that will also be in $\mathbb{C}G$.

  • $\begingroup$ Thank you very much...have you a reference for "an element of $A$ is positive if and only if its image in $\text{End}(A)$" is positive...also I have elements of $\mathbb{C}\mathbb{Z}_3$ that are self-adjoint but not positive such as can be seen in math.stackexchange.com/questions/1327302/… $\endgroup$ – JP McCarthy Jun 17 '15 at 6:28
  • $\begingroup$ @Jp McCarthy: No worries... Added some details. For $C^{*}$ algebras, perhaps Dixmier. $\endgroup$ – Orest Bucicovschi Jun 17 '15 at 7:00
  • $\begingroup$ Thank you... what about $\nu=(1/3,\frac16 (2+\sqrt{3}i),\frac{1}{6}(2-\sqrt{3}i))$. According to my calculations, and the answer to the linked answer, there is no $\mu\in\mathbb{C}\mathbb{Z}_3$ such that $\nu=\mu^*\mu$... does this not mean that $\nu$ is not positive (but self-adjoint yes)? $\endgroup$ – JP McCarthy Jun 17 '15 at 7:31
  • $\begingroup$ @Jp McCarthy: No, it's not positive, one of the sums involving the $3$-rd roots of $1$ is $-\frac{1}{2}$ ( the other are $1$ and $-\frac{1}{2}$). All three of them have to be positive. The nice thing is that the conditions for positivity in the case of abelian groups are linear in the coefficients. $\endgroup$ – Orest Bucicovschi Jun 17 '15 at 7:45
  • $\begingroup$ Lovely thank you. This gives me enough to go on to give you the bounty anyway. $\endgroup$ – JP McCarthy Jun 17 '15 at 8:00

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.