MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

One can show that the adjoint representation of $\mathrm{SU}(n)$, the image of the map $\mathrm{Ad}:\mathrm{SU}(n) \rightarrow \mathrm{Aut}(\mathrm{su(n)})\subset \mathrm{GL}(\mathrm{su}(n))$, is an $n^2-1$-dimensional lie subgroup of $\mathrm{SO}(n^2-1)$.

I would to know if any characterisations of this subgroup are known? In particular, I would like to understand what object, apart from the inner product, is preserved under the action of elements from the lie group $\mathrm{Im}(\mathrm{Ad})\subset \mathrm{SO}(n^2-1)$. Alternatively, I am looking to understand the action of $\mathrm{Im}(\mathrm{Ad})$ on $\mathbb{R}^{n^2-1}$.

I know that given a basis for $\mathrm{su}(n)$ one can construct a basis for the Lie algebra of $\mathrm{Im}(\mathrm{Ad})$ using the $\mathrm{ad}$ map and the structure constants for $\mathrm{su}(n)$ - However I cannot see how to use this to answer the question above?

In addition, I am aware that $\mathrm{Im}(\mathrm{Ad}) \simeq \mathrm{SU}(n)/\mathbb{Z}_n$ - However I also cannot see how to use this to help me answer the above question? In the $n=2$ case one can use the first isomorphism theorem along with the well know group homomorphism between $\mathrm{SU}(2)$ and $\mathrm{SO}(3)$, however for the general case it seems like this approach would require the construction of a group homomorphism $f:\mathrm{SU}(n)\rightarrow \mathrm{SO}(n^2-1)$ with kernel $\mathbb{Z}_n$? It seems unlikely one can construct such a homomorphism in general without the useful parameterizations of $\mathrm{SU}(2)$ and $\mathrm{SO}(3)$ one has for the case where $n=2$? Is there another way to exploit this quotient group structure?

share|cite|improve this question
Actually, your first sentence contains an error. The dimension of $\mathrm{Ad}\bigl(\mathrm{SU}(n)\bigr)$ is $n^2{-}1$, not $n$. – Robert Bryant Aug 26 '14 at 20:49
Edited - thanks! – Ryan Aug 29 '14 at 12:24

Here is a different characterization of the subgroup $\mathrm{Ad}\bigl(\mathrm{SU}(n)\bigr)\subset\mathrm{SO}(n^2{-}1)$ that works when $n>2$.

Define a skew-symmetric trilinear form $\kappa:{\frak{su}}(n)\times{\frak{su}}(n)\times{\frak{su}}(n)\to\mathbb{R}$ by the formula $$ \kappa(x,y,z)=\mathrm{tr}(xyz-xzy)\ \bigl(=\mathrm{tr}(yzx-yxz)=\mathrm{tr}(zxy-zyx)\bigr). $$ Then, for $n>2$, the subgroup $\mathrm{Ad}\bigl(\mathrm{SU}(n)\bigr)\subset\mathrm{GL}\bigl({\frak{su}}(n)\bigr)$ is the identity component of the subgroup of $\mathrm{GL}\bigl({\frak{su}}(n)\bigr)$ that preserves $\kappa$. (This doesn't work for $n=2$, of course. Also, you do have to restrict to the identity component because $\mathrm{SU}(n)$ has an outer automorphism (namely, conjugation) when $n>2$, and this preserves $\kappa$ as well; accordingly, the full $\kappa$-stabilizer has two components.)

In a certain sense, this is the 'simplest' characterization of $\mathrm{Ad}\bigl(\mathrm{SU}(n)\bigr)$ as a subgroup of $\mathrm{GL}\bigl({\frak{su}}(n)\bigr)$. What I mean is that, if you want to describe this subgroup as the set of linear transformations that preserve some set of algebraic objects on the vector space ${\frak{su}}(n)$, then $\kappa$ is the object that sits in the smallest representation of $\mathrm{GL}\bigl({\frak{su}}(n)\bigr)$ that does the job.

For example, if you want to define $\mathrm{Ad}\bigl(\mathrm{SU}(n)\bigr)$ as the subgroup that preserves the Lie bracket $[,]:{\frak{su}}(n)\times {\frak{su}}(n)\to {\frak{su}}(n)$, then you have to think of $[,]$ as an element of ${\frak{su}}(n)\otimes {\frak{su}}(n)^\ast\otimes {\frak{su}}(n)^\ast$ or, if you want to build in the skew-symmetry of the bracket, as an element of ${\frak{su}}(n)\otimes \Lambda^2({\frak{su}}(n)^\ast)$. Either way, these vector spaces have a much higher dimension than $\Lambda^3({\frak{su}}(n)^\ast)$, which is where $\kappa$ lives.

As another example, if you want to think of $\mathrm{Ad}\bigl(\mathrm{SU}(n)\bigr)$ as the subgroup that preserves the quadratic form $Q(x) = \mathrm{tr}(x^2)$ and the ($\mathbb{R}$-valued) cubic form $C(x) = i\,\mathrm{tr}(x^3)$, you are asking that it preserve an element of $S^2({\frak{su}}(n)^\ast)\oplus S^3({\frak{su}}(n)^\ast)$, and, again, this has considerably higher dimension than $\Lambda^3({\frak{su}}(n)^\ast)$. You can cut this down a little bit by noting that $C$ is harmonic with respect to $Q$, so that you could look at characterizing $\mathrm{Ad}\bigl(\mathrm{SU}(n)\bigr)$ as the subgroup of $\mathrm{O}\bigl({\frak{su}}(n),Q\bigr)$ that stabilizes $C\in S^3_0({\frak{su}}(n)^\ast,Q)$, but this still is more total equations than the equations that say that an element fixes $\kappa$.

I'm not saying that using $\kappa$ to define $\mathrm{Ad}\bigl(\mathrm{SU}(n)\bigr)$ is the most useful way to define it, I'm just saying that it's algebraically the least redundant.

For example, look at the case $n=3$: The codimension of $\mathrm{Ad}\bigl(\mathrm{SU}(3)\bigr)$ in $\mathrm{GL}\bigl({\frak{su}}(3)\bigr)\simeq \mathrm{GL}(8,\mathbb{R})$ is $64-8 = 56$, which is equal to the dimension of $\Lambda^3\bigl(\mathbb{R}^8\bigr)$, so it follows that the $\mathrm{GL}(8,\mathbb{R})$-orbit of $\kappa$ is actually open in $\Lambda^3\bigl(\mathbb{R}^8\bigr)$. Thus, the condition of fixing $\kappa$ is $56$ independent equations on an element of $\mathrm{GL}(8,\mathbb{R})$, which is the absolute minimum number of equations you need to cut out $\mathrm{Ad}\bigl(\mathrm{SU}(3)\bigr)$ as a submanifold.

share|cite|improve this answer

As your question already suggests, the "extra object preserved" is the Lie bracket. I.e., for any connected semisimple Lie group $G$, the image of $\mathrm{Ad}: G\to\mathrm{GL}(\mathfrak g)$ is precisely the identity component $\mathrm{Aut}(\mathfrak g)^0$ of $$ \mathrm{Aut}(\mathfrak g) = \{a\in\mathrm{GL}(\mathfrak g):a([X,Y])=[a(X),a(Y)] \text{ for all } X,Y\in\mathfrak g\}. $$

Proof. Clearly $\mathrm{Ad}$ maps $G$ into that identity component. To verify that it is onto it, it is enough to see that its derivative $\mathrm{ad}:\mathfrak{g}\to\mathrm{gl}(\mathfrak g)$ is onto the Lie algebra $$ \mathrm{aut}(\mathfrak g) = \{D\in\mathrm{gl}(\mathfrak g):D([X,Y])=[D(X),Y]+[X,D(Y)] \text{ for all } X,Y\in\mathfrak g\} $$ of $\mathrm{Aut}(\mathfrak g)$, or in other words, that every derivation $D$ of $\mathfrak g$ is inner: $D=\mathrm{ad}(Z)$ for some $Z\in\mathfrak g$. But that is precisely the content of Whitehead's Lemma (see any book on semisimple Lie algebras).

share|cite|improve this answer
Thanks, this is useful - I will have to put some thought into how to use this preservation of the lie-bracket to understand the action of a generic element of $\mathrm{Aut}(\mathrm{su}(n))$ - which is also an element of $\mathrm{SO}(n^2-1)$ - on $\mathbb{R}^{n^2-1}$. – Ryan Aug 21 '14 at 10:39

I am not sure if I am providing a characterization of the sort you seek. Nevertheless, the Lie group $SU(n)/\mathbb{Z}_n$ is called the projective special unitary group, $PSU(n)$. Note that this group has a natural action on $\mathbb{P}^{n-1}$. In fact, it is precisely the group of holomorphic isometries of $\mathbb{P}^{n-1}$.

share|cite|improve this answer
Thank you - do you know of a good reference I could use to read up on this group and its action? In particular I am not sure of the definition of a holomorphic isometry? – Ryan Aug 21 '14 at 9:39
The usual Hermitian inner product on $\mathbb{C}^n$ yields a Riemannian metric on $\mathbb{P}^{n-1}$. A holomorphic isometry of $\mathbb{P}^{n-1}$ is a biholomorphism of $\mathbb{P}^{n-1}$ with itself that preserves this metric. One quick reference for the group, its action on $\mathbb{P}^{n-1}$, and the metric is Einstein Manifolds by Arthur Besse, pages 178-181. – Peter Crooks Aug 21 '14 at 9:52
Thanks, I will have a look at this reference. However my initial impression is that this is not quite the characterisation I am looking for - I will investigate none the less! – Ryan Aug 21 '14 at 10:01

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.