Schur-Weyl dualities

Schur-Weyl dualities

Arun Ram
Department of Mathematics and Statistics
University of Melbourne
Parkville, VIC 3010 Australia
aram@unimelb.edu.au
and

Department of Mathematics
University of Wisconsin, Madison
Madison, WI 53706 USA
ram@math.wisc.edu

Last updates: 20 June 2010

Schur-Weyl dualities

Let n∈ ℤ >0 and let V be a vector space with basis v1 ,…, vn . Then the tensor product V ⊗k =V⊗V⊗…V k  factors   has basis   v i1 ⊗…⊗ v ik | 1≤ i1 ,…, ik ≤n . For k∈ ℤ >0 let V ⊗ k+ 1 2 = V ⊗k ⊗ vn ,  a subspace of   V ⊗ k+1 (which is isomorphic as a vector space to V ⊗k ). If b∈End V ⊗k let b i 1' ,…, i k' i1 ,…, ik ∈ℂ be the coefficients in the expansion b v i1 ⊗…⊗ v ik = ∑ 1≤ i 1' ,…, i k' ≤n b i 1' ,…, i k' i1 ,…, ik v i 1' ⊗…⊗ v i k' . For d∈ Ak and values i1 ,…, <k , i 1' ,… i k' ∈ 1…n define

d i1 ,…, ik i 1' ,…, i k' = 1, if   ir = is   when  r  and  s  are in the same block of  d, 0, otherwise.
For example, viewing d i1 ,…, ik i 1' ,…, i k' as the diagram d with vertices labeled by the values i1 ,…, ik and i 1' ,…, i k' , we have

i1 i3 i2 i4 i5 i6 i7 i8 i 1' i 3 ' i 2 ' i 4 ' i 5 ' i 6 ' i 7 ' i 8 ' = δ i1 i2 δ i1 i4 δ i1 i 2' δ i1 i 5 ' δ i5 i6 δ i5 i7 δ i5 i 3 ' δ i5 i 4 ' δ i5 i 6 ' δ i5 i 7 ' δ i8 i 8 ' .

It follows from (???) and (???) that for all d∈ Ak , Φk xd i 1' ,…, i k' i1 ,…, ik = 1, if   ir = is   iff  r  and  s  are in the same block of  d, 0, otherwise.

The group GL n ℂ acts on the vector spaces V and V ⊗k by g vi = ∑ j=1 n g ji vj ,  and  g v i1 ⊗ v i2 ⊗…⊗ v ik =g v i1 ⊗g v i2 ⊗…⊗g v ik , for g= g ji ∈ GL n ℂ . For any subgroup G⊆ GL n ℂ , End G V ⊗k = b∈End V ⊗k | bgv=gbv  for all  g∈G  and  v∈ V ⊗k .

Let n∈ ℤ >0 and let xd | d∈ Ak be the basis of ℂ Ak n defined in (???). Then the notation in (???) and (???) defines algebra homomorphisms Φk :ℂ Ak n →End V ⊗k  for  k∈ 1 2 ℤ >0 , giving a right action of the partition algebra ℂ Ak n on V ⊗k . View the symmetric group Sn as the subgroup of GL n ℂ of permutation matrices.

  1. Φk :ℂ Ak n →End V ⊗k has image im Φk = End Sn V ⊗k   and  ker Φk =ℂ-span xd | d  has more than  n  blocks , and
  2. Φ k+ 1 2 :ℂ A k+ 1 2 n →End V ⊗k has image im Φ k+ 1 2 = End S n-1 V ⊗k   and  ker Φ k+ 1 2 =ℂ-span xd | d  has more than  n  blocks .

Proof.
  1. As a subgroup of GLnℂ, Sn acts on V via its permutation representation and Sn acts on V ⊗k by σ v i1 ⊗…⊗ v ik = v σ i1 ⊗…⊗ v σ ik . Then b∈ End Sn V ⊗k iff σ -1 bσ=b (as endomorphisms on V ⊗k for all σ∈ Sn . Thus using the notation of (???), b∈ End Sn V ⊗k iff b i 1' ,…, i k' i1 ,…, ik = σ -1 bσ i 1' ,…, i k' i1 ,…, ik = b σ i 1' ,…,σ i k' σ i1 ,…,σ ik ,  for all  σ∈ Sn . It follows that the matrix entries of b are constant on the Sn -orbits of its matrix coordinates. These orbits decompose 1…k 1' … k' into subsets and thus correspond to set partitions d∈ Ak . Thus Φk xd has 1s in the matrix position corresponding to d and 0s elsewhere, and so b is a linear combination of Φk xd , d∈ Ak . Since xd , d∈ Ak , form a basis of ℂ Ak , im Φk = End Sn V ⊗k .

    If d has more than n blocks, then by (???) the matrix entry Φ k xd i 1' ,…, i k' i1 ,…, ik =0 for all indices i1 ,…, ik , i 1' ,…, i k' , since we need a distinct ij ∈ 1…n for each block of d. Thus xd ∈ker Φk . If d has ≤n blocks, then we can find an index set i1 ,…, ik , i 1' ,…, i k' with Φ k xd i 1' ,…, i k' i1 ,…, ik =1 simply by choosing a distinct index from 1…n for each block of d. Thus, if d has ≤n blocks the xd ∉ker Φk , and so ker Φk =ℂ-span xd | d  has more than  n  blocks .

  2. The vector space V ⊗k ⊗ vn ⊆ V ⊗ k+1 is a submodule both for ℂ A k+ 1 2 ⊆ℂ A k+1 and ℂ S n-1 ⊆ℂ Sn . If σ∈ S n-1 , then σ v i1 ⊗…⊗ v ik ⊗ vn = v σ i1 ⊗…⊗ v σ ik ⊗ vn . Then as above b∈ End S n-1 V ⊗k iff b i 1' ,…, i k' ,n i1 ,…, ik ,n = b σ i 1' ,…,σ i k' ,n σ i1 ,…,σ ik ,n ,  for all  σ∈ S n-1 . The S n-1 orbits of the matrix coordinates of b correspond to set partitions d∈ A k+ 1 2 ; that is vertices i k+1 and i k+1 ' must be in the same block of d. The same argument as part (a) can be used to show that ker Φ k+ 1 2 is the span of xd with d∈ A k+ 1 2 having more than n blocks. We always choose the index n for the block containing k+1 and k+1 ' . □

The multiplication in ℂ Ak n is, in terms of the basis xd , x d1 x d2 = 0, if   d1 ∘ d2   match exactly in the middle, ∑d cd xd , if   d1 ∘ d2   match exactly in the middle, where the sum is over all coarsenings of d1 ∘ d2 obtained by merging a top horizontal block and a bottom horizontal block and cd = n- d d1 ∘ d2 ,where   l r =l l-1 … l-r+1 , d is the number of blocks of d, and d1 ∘ d2 is the number of internal blocks of d1 ∘ d2 .

Proof.

Let n>>k so that ℂ Ak n ≅ End Sn V ⊗k . Then v i1 ⊗…⊗ v ik x d1 x d2 = ∑ i 1' ,…, i k'
i 1" ,…, i k" x d1 i 1' ,…, i k' i1 ,…, ik x d2 i 1" ,…, i k" i 1' ,…, i k' and ∑ i 1' ,…, i k'
i 1" ,…, i k" x d1 i 1' ,…, i k' i1 ,…, ik x d2 i 1" ,…, i k" i 1' ,…, i k' = ∑ interior   blocks 1= n- d n- d +1 … n- d - l-1 , where the sum is over labeled interior blocks of d1 ⊗ d2 such that the labels on these blocks are distinct and do not lie in i1 … ik i 1' … i k' . □

For k∈ ℤ >0 the following result is due to Tanabe [Ta, Lemma 2.1].

  1. Let T be the subgroup of GLnℂ of diagonal matrices in GLnℂ and let N= G ℂ* ,1,n be the normaliser of T if GLnℂ. Then Φk ℂ A k,∞,1 n = EndN V ⊗k   and  ker Φk ∩ℂ A k,∞,1 n =?
  2. Then Φk ℂ Sk = End 𝔤𝔩 n V ⊗k   and  ker Φk ∩ℂ Sk =the ideal generated by   ∑ w∈ Sk det w w.

Let L n-1 = G r,1,n × ℤ/rℤ ∩ G r,p,n . Then im Φk = End G r,p,n V ⊗k   and  ker Φk =ℂ-span xd | d  has more than  n  blocks , im Φ k+ 1 2 = End L n-1 V ⊗k   and  ker Φ k+ 1 2 =ℂ-span xd | d  has more than  n  blocks .

Proof.

In the case r=p=1 this is a result of Jones [Jo] and Martin [Ma] (see [HR], Theorem ???). A direct computation (which we will not do here) shows that if d∈ Ak rpn then xd commutes with each of the generators t1p , s1 , s2 ,…, sn of G r,p,n . Thus im Φk ⊆ End G r,p,n V ⊗k . If a∈ End G r,p,n V ⊗k then a∈ End Sn V ⊗k and so by the Jones-Martin theorem, a= ∑ d∈ Ak cd xd ,for some   cd ∈ℂ. We shall show that if d∉ Ak rpn then cd =0. Let d∈ Ak and let vt = v i1 ⊗…⊗ v ik   and   vb = v i 1' ⊗…⊗ v i k' be such that ir = is iff r and s are in the same block of d. Then cd =a vt | vb , the coefficient of the basis element vb in the expansion of a vt . Choose a block B of d and let l∈B. Then cd = a vt | vb = t il -p a t il p | vb = ξ pκ B cd . Hence cd =0 unless κ B =0 modr/p . Now choose a pair of distinct blocks B1 and B2 in d. Let l∈ B1 and m∈ B2 . Then cd = a vt | vb = t im t il -1 a t il t im -1 vt | vb = ξ κ B1 -κ B2 cd . So cd =0 unles κ B1 =κ B2 =0 modr . The same argument with vt = v i1 ⊗…⊗ v ik ⊗ vn   and   vb = v i 1' ⊗…⊗ v i k' ⊗ vn applies to establish case (b). □

If d∈ Bk is a diagram choose a labeling of the blocks of d from ????? with 1,2,…,k by marking one vertex in each block. An element σ∈ Sk permutes the marked vertices to produce a new diagram σd∈ Bk .

PICTURE INSERT HERE

Suppose n+1≤k. For a given element d∈ Bk , the element of ℂ Bk n given by ∑ σ∈ S n+1 -1 l σ σd, depends on the choice of the labeling of d, but the set ∑ σ∈ S n+1 -1 l σ σd| d∈ Bk does not???????/

If k>n let S n+1 be the subgroup of Sk which fixes n+2,…,k.

(Schur-Weyl) Let n∈ ℤ >0 .

  1. Φk :ℂ Sk →End V ⊗k has im Φk= End GLnℂ V ⊗k and ker Φk is the ideal of ℂ Sk generated by ∑ σ∈ S n+1 -1 l σ σd| d∈ Sk   has more than  n  blocks .
  2. Φk :ℂ Bk →End V ⊗k has im Φk= End Onℂ V ⊗k and ker Φk is the ideal generated by ∑ σ∈ S n+1 -1 l σ σd| d∈ Sk   has more than  n  blocks .
Need to define action of σ on d.

Define linear maps ε 1 2 :End V ⊗k →End V ⊗k ε 1 2 :End V ⊗k →End V ⊗ k-1 and ε1 :End V ⊗k →End V ⊗ k-1 by ε 1 2 b i 1' ,…, i k' i1 ,…, ik = b i 1' ,…, i k' i1 ,…, ik δ ik i k' ε 1 2 b i 1' ,…, i k-1 ' i1 ,…, i k-1 = ∑ j,l=1 n b i 1' ,…, i k-1 ' ,l i1 ,…, i k-1 ,j and ε 1 b i 1' ,…, i k-1 ' i1 ,…, i k-1 = ∑ j=1 n b i 1' ,…, i k-1 ' ,j i1 ,…, i k-1 ,j . Then ε1 = ε 1 2 ∘ ε 1 2   and  Tr b = ε1k b ,for  b∈End V ⊗k . The relation between the maps ε 1 2 , ε 1 2 and ε1 in (???) and the maps ε 1 2 , ε 1 2 and ε1 in (???) is given by Φ k- 1 2 ε 1 2 b = ε 1 2 Φk b | V ⊗ k- 1 2 , Φ k-1 ε 1 2 b = 1 n ε 1 2 Φk b and Φ k-1 ε1 b = ε1 Φk b , where, on the right hand side of the middle inequality, b is viewed as an element of ℂ Ak via the natural inclusion ℂ A k- 1 2 n ⊆ℂ Ak n . Then, for k∈ ℤ >0 , Tr Φk b = tr k b ,  and  Tr Φ k- 1 2 b = 1 n tr k- 1 2 b .

References [PLACEHOLDER]

[BG] A. Braverman and D. Gaitsgory, Crystals via the affine Grassmanian, Duke Math. J. 107 no. 3, (2001), 561-575; arXiv:math/9909077v2, MR1828302 (2002e:20083)

page history