Leeds-2016

 Leeds-2016

Canonical idempotents in multiplicity-free families of algebras

Bb90a2011740dd9ed01f00c96ba04259?s=128

Stephen Doty

July 19, 2016
Tweet

Transcript

  1. Canonical idempotents in multiplicity-free families of algebras S.R. Doty (Joint

    with Aaron Lauve and George H. Seelinger) Representation Theory and Physics Leeds University 19 July 2016
  2. Multiplicity-free families Let k be a field. All algebras A

    considered here are finite dimensional associative algebras over k, with a unit element 1. A partition of unity in A is a complete set of pairwise primitive orthogonal idempotents that sum to 1. We study the following two important (related) problems: Find a partition of unity in A. Find the (unique) partition of unity in its center Z(A). These are fundamental problems for any A. We solve them only in a special case, in which the given algebra A = An fits into a multiplicity-free family (MFF). The collection {An | n ≥ 0} is an MFF if each Ak is split semisimple over k, and: 1 A0 ∼ = k. 2 There is an embedding Ak ⊆ Ak+1 sending 1 to 1, for all k ≥ 0. 3 For any given irrep V of Ak, the multiplicities of its irreducible summands on restriction to Ak−1 are all ≤ 1.
  3. Examples of MFFs The motivating example is the family {kSn

    | n ≥ 0} of symmetric group algebras over a field k of characteristic 0. Other examples of MFFs include: The Iwahori–Hecke algebras {Hq(Sn) | n ≥ 0}. Certain families of other Weyl groups and their Iwahori–Hecke algebras (see [Ram 97]). The partition algebras. The Brauer algebras and Birman–Murakami–Wenzl (BMW) algebras. Etc, . . . Note that in the case of diagram algebras there are restrictions on the parameters as well as the underlying field in order to ensure semisimplicity.
  4. The branching graph Let {An | n ≥ 0} be

    an MFF. Write Irr(n) for a complete set of isomorphism classes of irreducible An-modules. Pick a representative V λ in each class λ ∈ Irr(n). The branching graph (Bratelli diagram) B is the directed graph with vertices and edges as follows: the vertices are the isomorphism classes n≥0 Irr(n); there is an edge µ → λ from the vertex µ to the vertex λ if and only if V µ is isomorphic to a direct summand of the restriction of V λ. Given λ ∈ Irr(n), let Tab(λ) denote the set of paths in the branching graph starting from the unique element ∅ ∈ Irr(0) and terminating at λ. An element of Tab(λ) has the form T : ∅ = λ0 → λ1 → λ2 → · · · → λn−1 → λn = λ. Set Tab(n) = λ∈Irr(n) Tab(λ). We say that a path T ∈ Tab(n) has length n. We sometimes write T → λ to indicate that T ∈ Tab(λ).
  5. The Gelfand–Tsetlin bases Let V be a given irreducible An-module.

    By the multiplicity-free branching assumption and Schur’s Lemma, the decomposition resAn−1 V = [W ]→[V ] W is canonical. Decomposing each W on the right hand side upon restriction to An−2 and continuing inductively all the way down to A0 ∼ = k, we obtain a canonical decomposition resA0 V = T VT into irreducible A0-modules, which are the 1-dimensional subspaces VT, where T runs over the set of T ∈ Tab(n) terminating in [V ] = λ. Choosing any 0 = vT ∈ VT for each T in Tab(λ), we get a basis {vT | T → λ} of each V λ, called the Gelfand–Tsetlin basis. We note that the choice of vT is uniquely determined only up to a scalar multiple.
  6. Wedderburn–Artin decomposition The standing assumption that k is a splitting

    field means that the Wedderburn–Artin decomposition of An, An = λ∈Irr(n) ε(λ)An ∼ = λ∈Irr(n) End k (V λ), expresses An as a direct sum of matrix algebras over k. In the isomorphism, the central idempotent ε(λ) ∈ An acts as the identity in End k (V λ) and zero in the other components, so {ε(λ) | λ ∈ Irr(n)} is the (unique) partition of unity of the center Z(An).
  7. The Gelfand–Tsetlin subalgebras Following [Okounkov–Vershik 96, 04], let Xn be

    the subalgebra of An generated by the centers Z(A1), . . . , Z(An). This is the Gelfand–Tsetlin subalgebra of An; it is a commutative subalgebra of An. Definition (Canonical projections) To each path T : ∅ = λ0 → λ1 → · · · → λn of length n in the branching graph, we associate a unique element εT := ε(λ1)ε(λ2) · · · ε(λn) of the Gelfand–Tsetlin subalgebra Xn. Given an irreducible module V ∼ = V λ for An and any T → λ, the element εT ∈ Xn ⊆ An is the projection mapping V onto VT. These projections are used to prove the following result. Proposition (Okounkov–Vershik) The Gelfand–Tsetlin algebra Xn is the algebra of all elements of An that act diagonally on the Gelfand–Tsetlin basis {vT : T ∈ Tab(λ)} for each irreducible An-module V λ (λ ∈ Irr(n)). In particular, the algebra Xn is a maximal commutative subalgebra of An.
  8. Canonical idempotents in MFFs The following did not appear explicitly

    in [Okounkov–Vershik], but it is implicit in the setup. (See also [Goodman–Graber 11].) Proposition Let {An | n ≥ 0} be a MFF. For any n, the set {εT | T ∈ Tab(n)} of canonical projections is a partition of unity in An. It is also a k-basis for the Gelfand–Tsetlin subalgebra Xn. Suppose that we have elements Jk ∈ Xk for all k ∈ N. The sequence of inclusions A1 ⊂ · · · ⊂ An−1 ⊂ An induces a corresponding sequence of inclusions X1 ⊂ · · · ⊂ Xn−1 ⊂ Xn among the Gelfand–Tsetlin subalgebras. So J1, . . . , Jn ∈ Xn. Writing each Jk in terms of the idempotent basis of Xn, we obtain Jk = T∈Tab(n) cT(k) εT, where cT(k) ∈ k. This defines an n-tuple cT := (cT(1), . . . , cT(n)) ∈ kn for each T ∈ Tab(n) that we call the T-content of the operators J1, . . . , Jn.
  9. Generalized Jucys–Murphy sequences A given sequence (Jk ∈ Xk)k∈N is

    a generalized Jucys–Murphy sequence (GJM-sequence) in the MFF {An | n ≥ 0} if Xn = J1, . . . , Jn , for all n ∈ N. If the field k is infinite then a GJM-sequence must exist. We can check whether or not a given sequence is a GJM-sequence by means of the content eigenvalues cT = (cT(1), . . . , cT(n)) of J1, . . . , Jn. (Note: cT(k) is the eigenvalue of Jk on the Gelfand–Tsetlin basis element vT.) To be precise, Proposition For a sequence (Jk ∈ Xk)k∈N in a MFF, the following are equivalent. 1 J1, . . . , Jn = Xn. 2 For all S, T ∈ Tab(n), S = T ⇐⇒ cS = cT. This is proved by a “Lagrange interpolation” argument. (Such arguments have appeared before, in work of [Murphy], [Diaconis–Greene], [Ram], etc.)
  10. Recursive formula for canonical projections If T ∈ Tab(n) is

    a path of length n, let T be the path of length n − 1 obtained by omitting its last edge and vertex. Proposition Let (Jk)k∈N be a GJM-sequence in the multiplicity-free family {Ak | k ≥ 0}. Suppose that S, T ∈ Tab(n). Then 1 cT(k) = c T (k) for all k < n. 2 If S = T but S = T then cS(n) = cT(n). This implies that the following “Lagrange interpolation” polynomial is well-defined (see [Garsia] in symmetric groups case): PT(Jn) := S∈Tab(n) S=T, S=T Jn−cS(n) cT(n)−cS(n) . Theorem (Recursion for εT ) If (Jn)n∈N is a GJM-sequence then εT = ε T PT(Jn), for any T ∈ Tab(n).
  11. Jucys–Murphy sequences If (Jk ∈ Xk)k∈N is a GJM-sequence such

    that zn = J1 + · · · + Jn ∈ Z(An), for all n ∈ N then we call it a Jucys–Murphy sequence (JM-sequence). Such sequences always exist in any MFF, if the field k is infinite. Given a path T → λ that terminates in λ we write type(T) = λ. Proposition Assume that (Jn)n∈N is a JM-sequence in a multiplicity-free family. For any T ∈ Tab(n) the polynomial PT(Jn) depends only on type(T), type(T). If λ = type(T) and µ = type(T) then we write Pλ µ (Jn) := PT(Jn). The proposition says this is well-defined. Theorem (Recursion for central idempotents) Assume that (Jn)n∈N is a JM-sequence in a multiplicity-free family. For any λ ∈ Irr(n), we have ε(λ) = µ ε(µ) · Pλ µ (Jn), where µ varies over the set of immediate predecessors of λ in the branching graph B.
  12. Synopsis Suppose {An | n ≥ 0} is an MFF.

    The family {εT | T ∈ Tab(n)} is a partition of unity of An and {ε(λ) | λ ∈ Irr(n)} is the partition of unity of Z(An). Knowing one family is equivalent to knowing the other. Let (Jk)k∈N be a JM-sequence. Let Pλ µ (Jk) ∈ Xk be the polynomial defined above, for each edge µ → λ (λ ∈ Irr(k)) in the branching graph B. Then: For any path T ∈ Tab(n), the idempotent εT is the product of all the polynomials along the edges of T. For any λ ∈ Irr(n), the central idempotent ε(λ) is the sum of all such products, taken over all the paths terminating in λ. This extends combinatorial approach to representations from symmetric groups to all MFFs, using paths in the branching graph B in place of tableaux. Remark. JM-sequences are not unique.
  13. EXAMPLE: symmetric group algebras Assume k has characteristic zero. Then

    {kSn | n ≥ 0} is a MFF. The sequence of JM-elements is defined by setting Jk := (1, k) + (2, k) + · · · + (k − 1, k) (the formal sum of all transpositions moving the point k). Irreps of Sn are indexed by partitions of n, and paths in branching graph by standard tableaux. Young found certain combinatorial bases of the irreps (and kSn itself) called the “seminormal form”. This leads to a family of pairwise orthogonal primitive idempotents that we call the seminormal idempotents. [Thrall 41] gave a recursive formula for the seminormal idempotents (in terms of Young symmetrizers). We show Theorem The seminormal idempotents described by Thrall’s recursion coincide with the canonical idempotents {εT | T a standard tableau}. This was stated without proof in [Murphy 83].
  14. Branching graph for {kSn | n ≥ 0} ∅ .

    . . . . . . . .
  15. Tables of polynomials Pλ µ (Jk ) for kSn λ|µ

    Pλ µ 2 | 1 1 2 (J2 + 1) 12 | 1 −1 2 (J2 − 1) 3 | 2 1 3 (J3 + 1) 21 | 2 −1 3 (J3 − 2) 21 | 12 1 3 (J3 + 2) 13 | 12 −1 3 (J3 − 1) λ|µ Pλ µ 4 | 3 1 4 (J4 + 1) 31 | 3 −1 4 (J4 − 3) 31 | 21 1 8 (J4 + 2)J4 22 | 21 −1 4 (J4 + 2)(J4 − 2) 212 | 21 1 8 (J4 − 2)J4 212 | 13 1 4 (J4 + 3) 14 | 13 −1 4 (J4 − 1)
  16. EXAMPLE: Brauer algebras Suppose that k has characteristic zero. Then

    the Brauer algebra Bn(δ) is semisimple (and split) over k provided the parameter δ ∈ k is not an integer [Wenzl 88]. Then {Bn(δ) | n ≥ 0} is an MFF. A sequence of JM-elements [Nazarov 96, etc] is defined by Jk = (1, k) + (2, k) + · · · + (k − 1, k) − (1, k) − (2, k) − · · · − (k − 1, k). Here (i, j) is the pure contraction corresponding to a transposition (i, j). Irreps of Bn(δ) are indexed by partitions of n − 2l, for 0 ≤ 2l ≤ n. Paths T in the branching graph correspond to the up-down tableaux in the literature.
  17. Branching graph for Brauer algebras {Bn (δ) | n ≥

    0} ∅ ∅ ∅ . . . ... ...
  18. Tables of Pλ µ (Jk ) for Bn (δ) λ|µ

    Pλ µ ∅ | 1 (J2+1)(J2−1) (δ−2)δ 2 | 1 (J2+δ−1)(J2+1) 2 δ 12 | 1 −(J2+δ−1)(J2−1) 2 (δ−2) λ|µ Pλ µ 1 | ∅ 1 1 | 2 (J3+1)(J3−2) (δ+2)(δ−1) 1 | 12 (J3+2)(J3−1) (δ−1)(δ−4) 3 | 2 (J3+δ)(J3+1) 3 (δ+2) 21 | 2 −(J3+δ)(J3−2) 3 (δ−1) 21 | 12 (J3+δ−2)(J3+2) 3 (δ−1) 13 | 12 −(J3+δ−2)(J3−1) 3 (δ−4)
  19. Tables of Pλ µ (Jk ) for Bn (δ) λ|µ

    Pλ µ ∅ | 1 (J4+1)(J4−1) (δ−2)δ 2 | 1 (J4+δ−1)(J4+1) 2 δ 2 | 3 (J4+1)(J4−3) (δ+4)δ 2 | 21 −(J4+δ)(J4+2)(J4−2)J4 2 (δ−2)(δ−4)δ 12 | 1 −(J4+δ−1)(J4−1) 2 (δ−2) 12 | 21 (J4+δ−2)(J4+2)(J4−2)J4 2 (δ+2)(δ−2)δ 12 | 13 (J4+3)(J4−1) (δ−2)(δ−6) λ|µ Pλ µ 4 | 3 (J4+δ+1)(J4+1) 4 (δ+4) 31 | 3 −(J4+δ+1)(J4−3) 4 δ 31 | 21 (J4+δ)(J4+δ−2)(J4+2)J4 8 (δ+2)δ 22 | 21 −(J4+δ)(J4+δ−2)(J4+2)(J4−2) 4 (δ−2)δ 212 | 21 (J4+δ)(J4+δ−2)(J4−2)J4 8 (δ−2)(δ−4) 212 | 13 (J4+δ−3)(J4+3) 4 (δ−2) 14 | 13 −(J4+δ−3)(J4−1) 4 (δ−6)