3
$\begingroup$

It is well known that the automorphism group of exceptional Jordan algebra $\mathcal{h}_{3}(\mathbb{O})$ is the exceptional Lie group $F_{4}$. I am trying to understand the automorphism group of the Jordan Algebra of Hermitian matrices $h_{3}(\mathbb{F})$, $\mathbb{F} = \mathbb{R}$, $\mathbb{C}$, $\mathbb{H}$. My suspicion is that at least the connected components are $\mathrm{SO}(3)$, $\mathrm{SU}(3)$ and $\mathrm{Sp}(3)$, respectively. My calculations, using the matrix model of projective spaces, result in

$$\mathbb{R}\mathrm{P}^{2} \simeq G_{\mathbb{R}} / \mathrm{SO}(2)$$ $$\mathbb{C}\mathrm{P}^{2} \simeq G_{\mathbb{C}} / \mathrm{SU}(2)$$ $$\mathbb{H}\mathrm{P}^{2} \simeq G_{\mathbb{H}} / \mathrm{Sp}(2)$$

where $G_{\mathbb{F}}$ are the desired groups, based on theorem 14.99 of Spinors and Calibration (F. Harvey). Of course, the above candidates lies in the desired groups by an action of the form $g(A) = gA\overline{g}^{\mathrm{t}}$. I am trying to compare with: $$ \mathbb{R}\mathrm{P}^{2} \simeq \mathrm{O}(3) / \mathrm{O}(2) \times Z_{2}$$ $$\mathbb{C}\mathrm{P}^{2} \simeq \mathrm{U}(3) / \mathrm{U}(2) \times U(1)$$ $$\mathbb{H}\mathrm{P}^{2} \simeq \mathrm{Sp}(3) / \mathrm{Sp}(2) \times \mathrm{Sp}(1)$$ References are welcome.

$\endgroup$
4
  • $\begingroup$ I also need this result, and have been fairly frustrated looking for references: almost all references talk about the Lie algebra and not the Lie group of symmetries. An answer can be extracted from Michael Orlitzky's example, but it's hard for me to believe that this was not written down before April 2023. $\endgroup$ May 8, 2023 at 17:40
  • 1
    $\begingroup$ Maybe much more elementary than what you are looking for, but I answered this specific question (at least for connected components and including the non-compact case) in my master's thesis: hdl.handle.net/1843/36039 (Theorems A.1 and A.2 in Appendix A). The proof, which is quite simple and has a geometric approach, is an adaptation of F. Harvey's proof for the exceptional case in the book Spinors and Calibrations. My mistake not having included this comment here before. $\endgroup$
    – Inácio
    May 9, 2023 at 19:36
  • $\begingroup$ It is worth to remark the following. My original motivation was to show that the Jordan algebra structure completely replaces the vector structure (compact and non-compact case), which allows us to construct the octonionic (projective or hyperbolic) plane. $\endgroup$
    – Inácio
    May 9, 2023 at 19:53
  • $\begingroup$ Thanks, I will look at that (and the other references you give)! Unfortunately for my application I need the other component for $h_3(\mathbb{C})$ too, which seems to have gone little-noticed in the literature. $\endgroup$ May 9, 2023 at 22:21

2 Answers 2

3
+50
$\begingroup$

In a simple Euclidean Jordan algebra, the Jordan-automorphism group of the algebra is the subgroup of the cone of squares' automorphism group that fixes the Jordan unit element. This is written down, for example, in Theorem 8 of Gowda, but the result goes all the way back to Vinberg, even for non-simple EJAs.

For $\mathbb{R}$ and $\mathbb{C}$, that's enough to squeeze the result out of Theorem 2 in Schneider. He gives you the operators that preserve the cones of squares (the PSD cones), namely,

$$ \operatorname{Aut}\left(\mathcal{H}^{n}_{+}\left(\mathbb{R}\right)\right) = \left\lbrace X \mapsto U^{T}XU \ \middle|\ U \in \mathbb{GL}_{n}\left(\mathbb{R}\right) \right\rbrace $$

and

$$ \begin{aligned} \operatorname{Aut}\left(\mathcal{H}^{n}_{+}\left(\mathbb{C}\right)\right) &= \left\lbrace X \mapsto U^{*}XU \ \middle|\ U \in \mathbb{GL}_{n}\left(\mathbb{C}\right) \right\rbrace\\ &\cup \left\lbrace X \mapsto U^{*}\overline{X}U \ \middle|\ U \in \mathbb{GL}_{n}\left(\mathbb{C}\right) \right\rbrace. \end{aligned} $$

Imposing the additional condition that these preserve the identity matrix,

$$ \begin{aligned} \operatorname{JAut}\left(\mathcal{H}^{n}\left(\mathbb{R}\right)\right) &= \left\lbrace X \mapsto U^{T}XU \ \middle|\ U \in \mathbb{R}^{n \times n}, U^{T}U = I \right\rbrace,\\ \operatorname{JAut}\left(\mathcal{H}^{n}\left(\mathbb{C}\right)\right) &= \left\lbrace X \mapsto U^{*}XU \ \middle|\ U \in \mathbb{C}^{n \times n}, U^{*}U = I \right\rbrace\\ &\cup \left\lbrace X \mapsto U^{*}\overline{X}U \ \middle|\ U \in \mathbb{C}^{n \times n}, U^{*}U = I \right\rbrace. \end{aligned} $$

An analogous result based on the same sort of inertia theorem holds over the quaternions, $\mathbb{H}$. The book by Rodman contains enough of the spectral theory to show that,

$$ \operatorname{Aut}\left(\mathcal{H}^{n}_{+}\left(\mathbb{H}\right)\right) = \left\lbrace X \mapsto U^{*}XU \ \middle|\ U \in \mathbb{GL}_{n}\left(\mathbb{H}\right) \right\rbrace, $$

from which it follows that

$$ \operatorname{JAut}\left(\mathcal{H}^{n}\left(\mathbb{H}\right)\right) = \left\lbrace X \mapsto U^{*}XU \ \middle|\ U \in \mathbb{H}^{n \times n}, U^{*}U = I \right\rbrace. $$

I've just posted a preprint where I work out all of these details. This method has the advantage that you wind up knowing the cone automorphisms as well, but there is a more direct approach that can be used to verify the results. Theorem 6.5 in Huang characterizes the matrix Jordan-automorphism groups over $\mathbb{R}$, $\mathbb{C}$, and $\mathbb{H}$, assuming that you know the linear automorphism groups of $\mathbb{R}$, $\mathbb{C}$, and $\mathbb{H}$ as algebras over $\mathbb{R}$. Here again Rodman comes to the rescue with the prerequisite result for $\mathbb{H}$. In any case, we eventually deduce the same Jordan-automorphism groups that we do using the stabilizer subgroup approach.

$\endgroup$
1
  • $\begingroup$ So to specialize to $n=3$ translate into the notation for Lie groups that the question used, your answer is that $G_{\mathbb{R}} = PO(3) = SO(3)$, $G_{\mathbb{C}} = PU(3) \rtimes \mathbb{Z}/2$, and $G_{\mathbb{H}} = PSp(3)$? $\endgroup$ May 8, 2023 at 12:50
1
$\begingroup$

To expand on Michael Orlitzky's answer, unpacking the notation the claim is that, with $h_n(R)$ the $n\times n$ Jordan algebra of Hermitian matrices is

$$ \begin{aligned} \mathrm{Aut}(h_n(\mathbb{R})) &= PO(n) (= SO(n) \text{ if $n$ is odd})\\ \mathrm{Aut}(h_n(\mathbb{C})) &= PU(n) \rtimes \mathbb{Z}/2\mathbb{Z} = (SU(n) / \mathbb{Z}/n\mathbb{Z}) \rtimes \mathbb{Z}/2\mathbb{Z}\\ \mathrm{Aut}(h_n(\mathbb{H})) &= PSp(n) \end{aligned} $$

The projectivizations come from the way the action is defined, the center automatically acts trivially. The most interesting case is for $h_n(\mathbb{C})$, where the group is not connected; the $\mathbb{Z}/2\mathbb{Z}$ acts by the automorphism of the Dynkin diagram (as it exchanges the defining representation and its dual/conjugate representation).

This doesn't look compatible with the expectation in the original question that $\mathbb{C}P^{n-1} = \mathrm{Aut}(h_n(\mathbb{C}))/SU(n-1)$, I would need to see more of the argument to see what might be going wrong.


Update: For the cases of $h_n(\mathbb{R})$ and $h_n(\mathbb{H})$, this is proved by Kalisch (Theorem 6):

Kalisch, G. K., On special Jordan algebras, Trans. Am. Math. Soc. 61, 482-494 (1947). ZBL0032.25003.

The paper also says something about the $h_n(\mathbb{C})$ case, but I don't think it computes the automorphism group. This is also treated by Jacobson:

Jacobson, Nathan, Isomorphisms of Jordan rings, Am. J. Math. 70, 317-326 (1948). ZBL0039.02801.

Jacobson, Nathan, Some groups of transformations defined by Jordan algebras. I, J. Reine Angew. Math. 201, 178-195 (1959). ZBL0084.03601.

The 1948 Jacobson paper is rather inexplicit (referring to automorphisms of the matrix algebra that commute with the involution), and one could easily miss the central extension in the $h_n(\mathbb{C})$ case (which is the $A_{II}$ case in his notation). The 1959 Jacobson paper is way more general and more explicit, perhaps too general (and doesn't state succinctly the result above).

$\endgroup$
6
  • $\begingroup$ A modern reference is Yokota: arxiv.org/pdf/0902.0431.pdf. Some discussion about this is contained in Lie Groups in the Foundations of Geometry (p.165) by Freudenthal. I think this latter has a more geometric approach. $\endgroup$
    – Inácio
    May 9, 2023 at 19:44
  • $\begingroup$ As far as I can tell, Yokota is almost exclusively interested in the cases that give exceptional groups, right? Does that paper consider the other cases in your original question? $\endgroup$ May 9, 2023 at 22:24
  • 1
    $\begingroup$ On further investigation, Theorem 3 of the 1947 Jacobson paper appears to be wrong, it omits the non-identity component of automorphisms of $h_n(\mathbb{C})$ (coming from an anti-automorphism of the matrix algebra rather than an automorphism). $\endgroup$ May 9, 2023 at 23:15
  • 1
    $\begingroup$ Yes, Yokota considers. Check Proposition 2.11.1 and 2.12.1. More specifically, he obtains the complete automorphism group in the complex case. It seems to me that the embeddings of the classical cases, although expected, require some algebraic technicalities to be described. $\endgroup$
    – Inácio
    May 10, 2023 at 1:57
  • $\begingroup$ Just a naive question: do you have an explicit expression for these actions? I would expect the other connected component of $G_{\mathbb{C}}$ to have a correspondence with the anti-holomorphic isometries of the complex planes through this action. $\endgroup$
    – Inácio
    May 10, 2023 at 2:21

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.