The short version:

Can the theory of weights for SL(n,C) be explained concretely in terms of raising and lowering operators on spaces of polynomials?

A deleted question asked how to prove SL(3,C) acts irreducibly on the space of homogenous polynomials of fixed degree. I think it was deleted as a homework question, but it has never been a homework question at my university (at least not in the past decade), and so I thought it might be reasonable to try and work out these ideas in a setting that makes sense to me:

Let K be a field of characteristic 0 and let SL(n,K) be the group of n×n matrices with entries from K and determinant 1. Let V be the K-vector space with basis the monomials xi of total degree w, where i = (i1,i2,...,in) is a multi-index with n entries, all non-negative integers whose sum is w. V has dimension w+1. SL(n,K) operates on V in the one true manner, by substitution of variables. In particular, for j = 2,3,...,n, the group SL(n,K) contains the elementary substitutions Ej which takes xj to x1+xj and takes xk to xk when j≠k.

Define the lowering operator Lj as Ej−1, so it takes a polynomial f to the difference between Ej(f) and f itself. The L are called lowering operators because they lower the degree of f when considered as a polynomial in xj, though they hold the total degree constant. This can be checked by applying it to monomials. For any individual monomial, factor it as xja⋅xi where ij=0, then Lj(xja⋅xi) = (x1+xj)a.xi − xja⋅xi = ( (x1+xj)a−xja )⋅xi. The leading term, when considered as a polynomial in xj is then ax1xja−1xi, so it has degree a−1 in xj.

If W is a submodule of V containing some nonzero polynomial f, then it must have nonzero degree in one of the variables xj. If any of these j are greater than 1, then apply Lj to transfer the degree from j to 1. Repeat until one is left with a⋅x1w. Since each Lj takes W to W and W is closed under division by nonzero scalars, we must have that any non-empty SL(n,K) invariant subspace of V contains x1w. In fact, we only use that it is closed under the maximal unipotent subgroup generated by the Ej.

For instance, if n=3, then we can more simply write x1=x, x2=y, and x3=z. If w=4, and W contains xyyz+2xyzz, then apply L3 = Lz to get xyy(x+z)+2xy(x+z)(x+z) − xyyz+2xyzz = xyyx + 2xyxx + 4xyxz = 2xxxy + xxyy + 4xxyz. The total degree remained the same, but shifted from z to x, at the cost of increasing the leading coefficient. Applying Lz again, one gets the first two terms vanish, and one is left with 4xxy(x+z) − 4xxyz = 4xxxy. Applying Ly one gets the pure 4xxx(x+y)−4xxxy = 4xxxx.

Notice that there were multiple paths we could have taken to move from f to xw. I think these paths are basically what are called "weights", and the lowering operators are called (simple, positive) "roots".

Now we must recover all other monomials from this monomial, and so we define the raising operators. We begin with the elementary Fj in SL(n,K) defined by taking x1 to x1+xj and xk to xk for k≠j. The raising operator Rj is then Fj−1 which takes a polynomial f to the difference between Fj(f) and f itself. Rj sacrifices a degree in x1 to provide a degree in xj. Applying each Rj operator ij times to x1w we obtain a polynomial with nonzero xi term, but with a great many other terms that seem difficult to control.

How does one use the raising operator to recover the other monomials?

For instance, to show xxyz is in W, I imagine that we should apply Rz to xxxx to get (x+z)(x+z)(x+z)(x+z) − xxxx = 4xxxz + 6xxzz + 4xzzz + zzzz. But then, how does one isolate 4xxxz? Here is my current method, which seems overly complex compared to lowering: Applying Lz 3 times and subtract 60xxxx (already known to be in W) takes zzzz to a multiple of xxxz, but it does so by zero-ing out the other terms. Applying Ry to xxxz gives (x+y)(x+y)(x+y)z−xxxz = 3xxyz + 3xyyz + yyyz. Now apply Ly twice to zero out the first two terms and take yyyz to 12xxxz + 6xxyz. Since xxxz is known to be in W, we can remove it and rescale to get xxyz in W.

This random sort of zig-zag makes it hard to keep track of which "roots" we've been applying (that is, which raising and lowering operators). In the Lie algebra case, I thought things were a bit cleaner. At any rate, successfully applying these operators should have the side-effect of detecting in a very concrete way the "highest weight", if I understand correctly. However, I cannot yet check, since I cannot yet successfully keep track of how I am applying these operators.

Is there a clear relationship between the paths one takes using these operators and the theory of weights? For instance, is x1w a highest weight vector?

It seems to me that SL(n,K) should have other irreducible representations other than just these polynomial representations, just because the paths defined by the operators are so symmetric. I think for SL(2,K) this is basically all there is, perhaps allowing for field automorphisms to be applied first.

Can all of the finite dimensional irreducible representations of SL(n,K) be explained in terms of actions on polynomials by substitutions and field automorphisms?

Assuming that this stuff makes sense, it would be handy to try it out on a different weight lattice.

Is there a version of this sort of description for any of the other classical groups or for G2?

It would be nice if say the symplectic group operated by substitutions on some subspace of homogenous polynomials, possibly restricted in some way, and that one could then find the raising and lowering operators. Any group where the action is natural and is likely to work well in the setup of Dickson's Linear Groups is fine, but I suspect the symplectic group might be the one with the least complications.

  • $\begingroup$ For those following the problem: for SL2 and SL3 polynomials basically suffice, but for SL4 and up the action on the other sets of variables has lots of minors in them so I'm not yet happy calling them monomials. The messiness in the raising may be due to too harsh approximations, exp(x)=1+x and log(x)=x-1. The Ej and Fj should be modified to use adjacent indices, not always 1, otherwise raising cannot work. At any rate, the short answer is "yes", but perhaps it is a little too messy in the Lie group case, so that is why one only sees this for Lie algebras. $\endgroup$ – Jack Schmidt May 12 '10 at 17:49
  • 1
    $\begingroup$ If the monomials are called coordinates instead of variables, then I am comfortable calling them polynomials (in those coordinates). The "middle" ones for SL4 are the original Plücker coordinates; they represent 2-dim subspaces, and the action is the natural "move a subspace to its image under the matrix"; the formula for the matrix entries involves a bunch of minors, but in a simple way. The last challenge is finding motivated approximations to the exp(x∂y) that are close enough to keep the error terms under control; there are some characteristic free methods that might help here. $\endgroup$ – Jack Schmidt May 12 '10 at 18:09
  • $\begingroup$ So is this a (multipart) mathematical question for which you expect a mathematical answer? Or are you just ruminating in public? $\endgroup$ – Victor Protsak May 13 '10 at 20:47
  • $\begingroup$ Several mathematical physics texts would suffice as answers to the main question, though so far they concentrate on sl(n) or SU(n) rather than SL(n). Weyl's Classical Groups handles SL(n) nicely using polynomials since it emphasizes invariant theory, but does not emphasize Dirac's operators. I outlined their answers in the comments, but I expect a mathematical physicist will be able to find some treatment which emphasizes both the polynomials and the Dirac operators, which would be the perfect answer. $\endgroup$ – Jack Schmidt May 13 '10 at 22:11
  • $\begingroup$ Dirac operators? "Outlined their answers"? OK, I have no idea what you mean or want and I am not sure you know it, either. I have answered three specific questions you had (last 3 of the 5 highlighted sentences), and any good book on semisimple Lie groups/algebras would explain what weights are and how raising and lowering operators, which are specific elements of the Lie algebra, are used to describe and classify irreducible finite-dimensional representations (the first 2 highlighted sentences). If you can compress your thoughts into a single precise question, repost it. Good luck! $\endgroup$ – Victor Protsak May 14 '10 at 2:04

I am a bit reluctant to post a reply, since this is not even a homework question: you really ought to learn theory first from a textbook on representations of Lie algebras, and several of them have been published recently, instead of "discovering" it on your own and asking other people to check your understanding.

Answers (K has characteristic zero):

  1. The Lie algebra sl(n,K) and the algebraic group SL(n,K) act on the polynomial algebra in $n$ variables so that the generator $E_{ij}$ of the Lie algebra sl(n) acts by the differential operator $x_i\partial_j$. This representation is irreducible of highest weight (N,0,...,0)=$N\omega_1$ in standard enumeration, with the highest weight vector given by the monomial $x_1^N$.

  2. All simple finite-dimensional modules over a semisimple Lie algebra (equivalently, simple algebraic representations of a semisimple algebraic group) are classified in terms of their highest weights. For sl(n), a weight is an (n-1)-tuple of nonnegative integers and only very special n-tuples arise from the polynomials in n variables.

  3. For the classical Lie algebras sl(N), so(N), sp(N), the theory of standard monomials (due to Hodge, Pedoe, Laskshmibai, Musili, Sheshadri, Sundaram) gives an explicit realization of these modules on certain spaces of polynomials. For G2 Greg Kuperberg found a combinatorial analogue. The Littelmann path model provides a uniform description for all types that encompasses a lot of previous work including also Kashiwara's crystal bases.


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.