14
$\begingroup$

$\DeclareMathOperator\GL{GL}\DeclareMathOperator\Sym{Sym}$Let $\rho$ denote the irreducible algebraic representation of $\GL_n(\mathbb{C})$ with the highest weight $(2,2,\underset{n-2}{\underbrace{0,\dots,0}})$.

Let $k\leq n/2$ be a non-negative integer. How to decompose into irreducible representations the representation $\Sym^k(\rho)$?

More specifically, I am interested whether $\Sym^k(\rho)$ contains the representation with the highest weight $(\underset{2k}{\underbrace{2,\dots,2}},\underset{n-2k}{\underbrace{0,\dots,0}})$, and if yes, whether the mutiplicity is equal to one.

A a side remark, the representation $\rho$ has a geometric interpretation important for me: it is the space of curvature tensors, namely the curvature tensor of any Riemannian metric on $\mathbb{R}^n$ lies in $\rho$.

$\endgroup$
5
  • 2
    $\begingroup$ I would suggest to add "plethysm" tag. $\endgroup$
    – Sasha
    Commented Oct 3, 2012 at 18:11
  • 1
    $\begingroup$ computing a few examples in SAGE suggests that the answer is definitely yes. $\endgroup$ Commented Oct 3, 2012 at 18:21
  • $\begingroup$ @Dan Petersen: This sounds encouraging, many thanks. Unfortunately I do not know SAGE. $\endgroup$
    – asv
    Commented Oct 3, 2012 at 18:53
  • $\begingroup$ I would be interested to learn more about the connection with curvature tensors. Does anyone have a reference? $\endgroup$ Commented Oct 4, 2012 at 9:24
  • 1
    $\begingroup$ @Mark Wildon: some discussion of the connection can be found in \S 10.3 of "Symmetry, Representations, and Invariants" by Goodman and Wallach. Another place is "Einstein manifolds" by Besse, Ch. 1, paragraphs G,H. $\endgroup$
    – asv
    Commented Oct 4, 2012 at 11:56

2 Answers 2

19
$\begingroup$

The plethysm $\mathrm{Sym}^k \rho$ contains the irreducible representation with highest weight $(2,\ldots,2,0,\ldots,0)$ exactly once. It looks like a tricky problem to say much about its other irreducible constituents.

Let $\Delta^\lambda$ denote the Schur functor corresponding to the partition $\lambda$, and let $E$ be an $n$-dimensional complex vector space. Using symmetric polynomials (or other methods) one finds

$$\mathrm{Sym}^2 (\mathrm{Sym}^2 E) = \Delta^{(2,2)}E \oplus \mathrm{Sym}^4 E.$$

Therefore

$$ \mathrm{Sym}^k \mathrm{Sym}^2 \mathrm{Sym}^2 E \cong \sum_{r=0}^k \mathrm{Sym}^r (\Delta^{(2,2)}E) \otimes \mathrm{Sym}^{k-r} (\mathrm{Sym}^4 E) .$$

The irreducible representations contained in the $r$th summand are labelled by partitions with at most $2r+(k-r) = k+r$ parts. So to show that $\mathrm{Sym}^k(\Delta^{(2,2)}(E))$ contains $\Delta^{(2^{2k})}E$, it suffices to show that $\Delta^{(2^{2k})}E$ appears in $\mathrm{Sym}^k \mathrm{Sym}^2 \mathrm{Sym}^2 E$.

Let $U = \mathrm{Sym}^2 E$. There is a canonical surjection

$$ \mathrm{Sym}^k (\mathrm{Sym}^2 U ) \rightarrow \mathrm{Sym}^{2k} U. $$

given by mapping $(u_1u_1')\ldots (u_ku_k') \in \mathrm{Sym}^k (\mathrm{Sym}^2 U )$ to $u_1u_1'\ldots u_ku_k' \in \mathrm{Sym}^{2k} U$. Therefore $\mathrm{Sym}^k (\mathrm{Sym}^2 U )$ contains $ \mathrm{Sym}^{2k} U = \mathrm{Sym}^{2k} (\mathrm{Sym}^2 E)$. It is well known that

$$ \mathrm{Sym}^{2k} (\mathrm{Sym}^2 E) = \sum_{\lambda} \Delta^{2\lambda}(E) $$

where the sum is over all partitions $\lambda$ of $2k$ and $2(\lambda_1,\ldots,\lambda_m) = (2\lambda_1,\ldots, 2\lambda_m)$. Taking $\lambda = (1^{2k})$ we see that $\Delta^{(2^{2k})}E$ appears.

It remains to show that the multiplicity of $\Delta^{(2^{2k})}E$ in $\mathrm{Sym}^k (\Delta^{(2,2)}E)$ is $1$. We work over $\mathbb{C}$, so there is a chain of inclusions

$$ \mathrm{Sym}^k (\Delta^{(2,2)}(E)) \subseteq \mathrm{Sym}^k (\mathrm{Sym}^2 E \otimes \mathrm{Sym}^2 E) \subseteq (\mathrm{Sym}^2 E)^{\otimes 2k}.$$

By the Littlewood–Richardson rule (or the easier Young's rule), the multiplicity of $\Delta^{(2^k)}E$ in the right-hand side is $1$.

$\endgroup$
2
  • $\begingroup$ This is nice. $\,$ $\endgroup$ Commented Oct 4, 2012 at 6:55
  • $\begingroup$ That's a great answer. $\endgroup$ Commented Oct 4, 2012 at 13:11
2
$\begingroup$

Here's another answer to the question about multiplicities which can be generalized to other partitions. First, for any partition $\lambda$, let $S_\lambda$ denote the irreducible representation (Schur functor) with highest weight $\lambda$. I'll explain how to get the following:

Theorem: Let $\lambda$ be a partition of even size and let $\mu$ be the result of scaling all column lengths of $\lambda$ by $k$. Then $Sym^k(S_\lambda({\bf C}^n))$ contains $S_\mu({\bf C}^n)$ with multiplicity 1 if $\mu$ has at most $n$ rows (otherwise $S_\mu({\bf C}^n)=0$).

For $\lambda = (2,2)$, scaling all column lengths by $k$ just means the partition $(2,2,2,\dots,2)$ ($2k$ instances of 2).

I'm going to assume basic fluency between symmetric functions and Schur functors.

Then $S_\lambda \otimes S_\mu$ always contains $S_{\lambda + \mu}$ with multiplicity 1: the tensor product of highest weight vectors gives the unique vector (up to scalar multiple) of this weight and it's also a highest weight vector (easy check). In particular, $S_{\lambda}^{\otimes k}$ contains $S_{k\lambda}$ with multiplicity 1. From the product description, the highest weight vector is symmetric so it also belongs to $Sym^k(S_\lambda)$.

Next, we use the involution $\omega$ on symmetric functions. One reference Section I.2 of Macdonald, Symmetric Functions and Hall Polynomials. This has the property that $\omega s_\lambda = s_{\lambda^T}$ where $\lambda^T$ is the transpose partition. It also behaves well with respect to plethysm (there are several kinds, here I refer to the one that corresponds to composition of Schur functors):

If $f$ and $g$ are homogeneous, then

$\omega (f \circ g) = \omega f \circ \omega g$ if $\deg g$ is odd, and

$\omega (f \circ g) = f \circ (\omega g)$ if $\deg g$ is even.

Reference: Macdonald, Example I.8.1(a).

In particular, this means that if $|\lambda|$ is even, then

$\omega (s_k \circ s_\lambda) = s_k \circ s_{\lambda^T}$.

Translating this back to representations:

The right side is $Sym^k(S_{\lambda^T})$ and hence contains $S_{k\lambda^T}$ with multiplicity 1. In particular, $\omega$ is an involution so $Sym^k(S_\lambda)$ contains $S_{(k\lambda^T)^T}$.

We can simplify further: $(k\lambda^T)^T$ is obtained from $\lambda$ by scaling all column lengths by $k$.

$\endgroup$

You must log in to answer this question.

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