So, the blog stats show that semisimple Lie algebras haven’t exactly been popular. Traffic has actually been unusually high, but people have been reading about the heat equation or Ricci curvature rather than Verma modules. Which is interesting, since I thought there was a dearth of analysts in the mathosphere. At MathOverflow, for instance, there have been a few complaints that everyone there is an algebraic geometer. Anyway, there wasn’t going to be that much more I would say about semisimple Lie algebras in the near future, so for the next few weeks I plan random and totally disconnected posts at varying levels (but loosely related to algebra or algebraic geometry, in general).
I learned a while back that there is a classification of the simple modules over the semidirect product between a group and an commutative algebra which works the same way as the (more specific case) between a group and an abelian group. The result for abelian groups rather than commutative rings appears in a lot of places, e.g. Serre’s Linear Representations of Finite Groups or Pavel Etingof’s notes. I couldn’t find a source for the more general result though. I wanted to work that out here, though I got a bit confused near the end, at which point I’ll toss out a bleg.
Let be a finite group acting on a finite-dimensional commutative algebra over an algebraically closed field of characteristic prime to the order of . Then, the irreducible representations of correspond to maximal ideals in , or equivalently (by Hilbert’s Nullstellensatz!) homomorphisms , called characters. In other words, acts on a 1-dimensional space via the character .
It is of interest to determine the irreducible representations of the semidirect product . This algebra is finite dimensional. So, all simple representations are finite-dimensional. Fix one, say . Then contains a simple -representation, i.e. there is a character and a vector such that
Now consider the -representation . It is in fact -stable, because
Here , of course, denotes acting on .
Let be the set of elements of that fix the character . Then the above calculation shows that fixes the “weight space” of with respect to . In other words, if satisfies the above condition (that ) then so does for . In particular, is a -submodule on which acts by scalar multiplication by . Let be an irreducible -representation.
Now there is a map of -representations, , so by Frobenius reciprocity, we get a map
It is surjective, since the image is nonzero and is irreducible. I claim in fact that it is injective. For this, it is sufficient to check that if are in different right-cosets, then . [Edit: This is not quite the case, cf. Darij’s comment below. -AM, 3/2/2010] This follows because an element in the intersection of the two is a weight for the characters , and do not lie in the same left coset modulo .
We have obtained most of:
Theorem 1An irreducible representation of can be obtained as follows. Fix a character , the stabilizer , an irreducible -representation which we let act on via , and setThese exhaust the irreducible representations of .
What remains to be seen is that any as defined above by such an induction process is indeed irreducible. This may be proved as follows. can be written as a direct sum
where acts by on . (Shades of the weight space decomposition.) Here ranges over the characters in the orbits of . Let they be enumerated as . I claim now that there is such that for ; this follows from the Chinese remainder theorem applied to the maximal ideals associated with the .
Anyway, choose some , which we write as for . We will prove that .
Since permutes the transitively, we find that itself contains a vector all of whose components in the are nonzero. So, replacing by this vector, we may assume each .
Then, if we multiply by (which was chosen as above), we have for all . In particular,
for all . If we invert since the vanderMonde determinant is nonzero, we find each . Since is -irreducible and generates , this implies that generates .
It is clear that changing the orbit of does not affect the module constructed above (up to isomorphism).
We have a uniqueness result as well:
Theorem 2 The representations associated to and are nonisomorphic when lie in different orbits of or if they are equal but .
The eigenvalues of the action of on (resp. ) are given by for in the orbit of (resp. ), which proves that unless are in the same orbit.
Now if but , we must show that . I’m getting stuck here—does anyone know what to do? [Edit- Answered in comments.]