Unipotent Hecke algebras: the structure, representation theory, and combinatorics

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

Last updated: 26 March 2015

This is an excerpt of the PhD thesis Unipotent Hecke algebras: the structure, representation theory, and combinatorics by F. Nathaniel Edgar Thiem.

A basis with multiplication in the G=GLn(𝔽q) case

Unipotent Hecke algebras

The group GLn(𝔽q)

Let G=GLn(𝔽q) be the general linear group over the finite field 𝔽q with q elements. Define the subgroups T={diagonal matrices}, N={monomial matrices}, W={permutation matrices},and U={(1*01)}, (4.1) where a monomial matrix is a matrix with exactly one nonzero entry in each row and column (N=WT). If necessary, specify the size of the matrices by a subscript such as Gn,Un,Wn, etc. If aGm and bGn are matrices, then let abGm+n be the block diagonal matrix ab= (a00b).

Let xij(t)U be the matrix with t in position (i,j), ones on the diagonal and zeroes elsewhere; write xi(t)=xi,i+1(t). Let hεi(t)T denote the diagonal matrix with t in the ith slot and ones elsewhere, and let siWN be the identity matrix with the ith and (i+1)st columns interchanged. That is, xi(t) = Idi-1 (1t01) Idn-i-1, hεi(t) = Idi-1(t) Idn-i, si = Idi-1 (0110) Idn-i-1, (4.2) where Idk is the k×k identity matrix. Then W=s1,s2,,sn-1, T=hεi(t)|1in,t𝔽q*, N=WT, U=xij(t)|1i<jn,t𝔽q, G=U,W,T. (4.3) The Chevalley group relations for G are (see also Section 3.3.1) xij(a) xrs(b) = xij(b) xrs(a) xis (δjrab) xrj (-δisab), (i,j)(r,s), (U1) xij(a) xij(b) = xij(a+b), (U2) si2 = 1, (N1) sisi+1si = si+1sisi+1 andsisj= sjsi, i-j>1, (N2) hεi(b) hεi(a) = hεi(ab), (N4) hεj(b) hεi(a) = hεj(b) hεi(a), (N5) srxij(t) = xsr(i)sr(j) (t)sr, (UN1) xij(a) hεr(t) = hεr(t) xij(t-δritδrja), (UN2) sixi(t)si = xi(t-1) sixi(-t) hεi(t) hεi+1 (-t-1), t0, (UN3) where δij is the Kronecker delta.

A pictorial version of GLn(𝔽q)

For the results that follow, it will be useful to view these elements of G as braid-like diagrams instead of matrices. Consider the following depictions of elements by diagrams with vertices, strands between the vertices, and various objects that slide around on the strands. View si as
i
,
(4.4)
hεi(t) as
1 1 t 1 1 i
,
(4.5)
xij(ab) as
a b i j
,
(4.6)
where each diagram has two rows of n vertices. Multiplication in G corresponds to the concatenation of two diagrams; for example, s2s1 is s2s1=
s1 s2
=
=
.
In the following Chevalley relations, curved strands indicate longer strands, so for example (UN1) indicates that a and b slide along the strands they are on (no matter how long). The Chevalley relations are
a a 1 b
=
a a a 1 b b
(U1)
1 1 a b
=
1 a+b
(U2)
=
(N1)
=
and
=
(N2)
a b
=
ab
(N4)
a b
=
a b
(N5)
a b
=
a b
(UN1)
a b r s
=
ar-1 bs r s
(UN2)
a b
=
b -a b-1 a-1 ab -(ab)-1
ab0.
(UN3)

The unipotent Hecke algebra μ

Fix a nontrivial group homomorphism ψ:𝔽q+*, and a map μ: {(i,j)|1i<jn} 𝔽q (i,j) μij withμij=0 forji+1. (4.7) Then ψμ: U * xij(t) ψ(μijt) (4.8) is a group homomorphism. Since μij=0 for all ji+1, write μ=(μ(1),μ(2),,μ(n-1),μ(n)), whereμ(i)= μi,i+1and μ(n)=0. (4.7)

The unipotent Hecke algebra μ of the triple (G,U,ψμ) is μ=EndG (IndUG(ψμ)) eμGeμ, whereeμ=1U uUψμ (u-1)u. (4.9)

The type of a linear character ψμ:U* is the composition ν such that μ(i)=0if and only if iν(with νas in Section 2.3.2).

Let ν be a composition. Suppose χ and χ are two type ν linear characters of U. Then (G,U,χ) (G,U,χ).

Proof.

Note that for h=diag(h1,h2,,hn)T, hxij(t)h-1= xij(hihj-1t) for1i<jn,t 𝔽q. Thus, T normalizes Uij=xij(t)|t𝔽q and acts on linear characters of U by hχ(u)=χ (huh-1), forhT,uU. This action preserves the type of χ. The map IndUG(χ) Geχ Geχ IndUG(χ) geχ gheχ where χ=hχfor hT, is a G-module isomorphism, so (G,U,χ)(G,U,hχ). It therefore suffices to prove that T acts transitively on the linear characters of type μ.

By Proposition 3.2 (c), the number of T-orbits of type μ linear characters is Zμ(q-1)n-(μ) T = {hT|hxi(t)h-1=xi(t),μ(i)0}(q-1)n-(μ) (q-1)n = (q-1)(μ)(q-1)n-(μ) (q-1)n = 1. Therefore T acts transitively on the type μ linear characters of U.

Proposition 4.1 implies that given any fixed character ψ:𝔽q+*, it suffices to specify the type of the linear character ψμ. Note that the map (given by example) {Compositionsμ=(μ1,μ2,,μ)n} {μ=(μ(1),μ(2),,μ(n))μ(i){0,1}andμ(n)=0} μ=
1 0 1 1 1 1 0 1 1 0 1 1 1 0
(1,0,1,1,1,1,0,1,1,0,1,1,1,0)
(4.10)
is a bijection. In the following sections identify the compositions μ=(μ1,,μ) with the map μ=(μ(1),,μ(n)) using this bijection.

The classical examples of unipotent Hecke algebras are the Yokonuma algebra (1n) [Yok1969-2] and the Gelfand-Graev Hecke algebra (n) [Ste1967].

An indexing for the standard basis of μ

The group G has a double-coset decomposition G=vNUvU, so if Nμ = {vN|eμveμ0} = {vN|u,vuv-1Uimpliesψμ(u)=ψμ(vuv-1)} (4.11) then the isomorphism in (4.9) implies that the set {eμveμ|vNμ} is a basis for μ [CRe1981, Prop. 11.30].

Suppose aM(𝔽q[X]) is an × matrix with polynomial entries. Let d(aij) be the degree of the polynomial aij. Define the degree row sums and the degree column sums of a to be the compositions d(a)= ( d(a)1, d(a)2,, d(a) ) and d(a)= ( d(a)1, d(a)2,, d(a) ) , where d(a)i= j=1d (aij)and d(a)j =i=1d (aij). Let Mμ= { aM(μ) (𝔽q[X]) |d(a) =d(a)=μ ,aijmonic, aij(0) 0 } . (4.12) For example, ( X+1 11 X+2 ) (1+0+0+1=2) X+3 X3+2X+3 1 X+2 (1+3+0+1=5) 1 X2+4X+2 X+2 1 (0+2+1+0=3) 1 1 X2+3X+1 X2+2 (0+0+2+2=4) (1+1+0+0=2) (0+3+2+0=5) (0+0+1+2=3) (1+1+0+2=4) M(2,5,3,4). (*)

Suppose (f)=(a0+a1Xi1+a2Xi2++arXir+Xn)M(n) is a 1×1 matrix with a0,a1,,ar0. Let v(f)=
(f)
=w(n) ( w(i1)a0 w(i2-i1) a1w(n-ir) ar ) N,
(4.13)
where w(k)=
Wk,
and by convention v(1) is the empty strand (not a strand). For example, v(a+bX3+cX4+X6) is
(a+bX3+cX4+X6)
=
a a a b c c
=
a a a b c c
.

For each aMμ define a matrix vaN pictorially by va=
(a1) (a21) (a11) (a2) (a22) (a12) (a) (a2) (a1)
,
(4.14)
where aij is the (i,j)th entry of a, and the top vertex associated with aij goes to the bottom vertex associated with aji. Note that if aij=1, then both its strand and corresponding vertices vanish. The fact that aMμ guarantees that we are left with two rows of exactly n vertices.

For example, if a is as in (*), then va is
(3+X) (1+X) (2+4X+X2) (3+2X+X3) (1+3X+X2) (2+X) (2+X2) (2+X) (2+X)
=
[3] [1] [2 4] [3 2 2] [1 3] [2] [2 2] [2] [2]
.

Viewed as a matrix, va=
μ1columns μ2columns μ3columns } } } μ1rows μ2rows μ3rows ( v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) 0 v(a11) v(a11) v(a11) v(a11) 0 v(a11) v(a11) 0 v(a11) v(a11) v(a13) v(a11) 0 v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a12) v(a11) 0 v(a11) 0 v(a11) 0 v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) 0 v(a11) 0 v(a11) 0 v(a11) 0 v(a11) 0 v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) 0 v(a11) 0 v(a11) 0 v(a11) v(a11) v(a11) 0 v(a11) 0 v(a11) 0 v(a11) v(a11) v(a23) 0 v(a11) 0 v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a22) 0 0 0 0 v(a11) 0 v(a11) v(a11) v(a11) v(a21) 0 0 0 0 0 0 0 v(a11) 0 v(a11) v(a11) v(a11) v(a11) v(a11) 0 0 0 0 v(a11) 0 0 v(a11) v(a11) 0 0 0 v(a11) v(a33) 0 0 v(a11) 0 v(a11) v(a11) v(a11) v(a11) v(a32) 0 0 0 0 0 0 v(a11) 0 v(a11) v(a11) v(a31) 0 0 0 0 0 0 0 0 0 0 v(a11) 0 v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) 0 0 0 0 0 0 0 0 0 v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) v(a11) )

Let Nμ be as in (4.11) and Mμ be as in (4.12). The map Mμ Nμ a va, given by (4.14) is a bijection.

Remark. When μ=(n) this theorem says that the map (f)v(f) of (4.13) is a bijection between M(n) and N(n).

Proof.

Using the remark following the theorem, it is straightforward to reconstruct a from va. Therefore the map is invertible, and it suffices to show (a) the map is well-defined (vaNμ), (b) the map is surjective. To show (a) and (b), we investigate the diagrams of elements in Nμ. Suppose vNμ. Let vi be the entry above theith vertex ofv, v(i) be the number of the bottom vertex connected to theith top vertex, so that {v1,v2,,vn} are the labels above the vertices of v and (v(1),v(2),,v(n)) is the permutation determined the diagram (and ignoring the labels vi). By (4.8), ψμ(xij(t))= { ψ(t), ifj=i+1and μ(i)=1, 1, otherwise. (A) Recall that vNμ if and only if u,vuv-1U implies ψμ(u)=ψμ(vuv-1). That is, vNμ if and only if for all 1i<jn such that v(i)<v(j), ψμ(xij(t)) = ψμ(vxij(t)v-1) = ψμ(xv(i)v(j)(vitvj-1)) = { ψ(vitvj-1), ifv(j)=v(i) +1andμ(v(i))=1, 1, otherwise. (B) Compare (A) and (B) to obtain that bNμ if and only if for all 1i<jn such that v(i)<v(j), (i) If μ(i)=1 and μ(v(i))=0, then ji+1, (ii) If μ(i)=0 and μ(v(i))=1, then v(j)v(i)+1, (iii) If μ(i)=μ(v(i))=1, then j=i+1 if and only if v(j)=v(i)+1, (iii) If μ(i)=μ(v(i))=1 and v(j)=v(i)+1, then vi=vi+1. We can visualize the implications of the conditions (i)−(iii) in the following way. Partition the vertices of vNμ by μ. For example, μ=(2,3,1) partitions v according to
v1 v2 v3 v4 v5 v6
.
Suppose the ith top vertex is not next to a dotted line and the v(i)th bottom vertex is immediately to the left of a dotted line. Then condition (i) implies that v(i+1)<v(i), so
vi vi+1
(I)
Similarly, condition (ii) implies
vi
(II)
and conditions (iii) and (iii) imply
vi vi+1
or
vi vi
(vi+1= viin the second).
(III)
In the case μ=(n) condition (III) implies that every vN(n) is of the form
a1 a1 a1 a2 a2 a2 ar ar ar
=w(n) ( a1w(i1) a2w(i2) arw(ir) ) ,
where (i1,i2,,ir)n, ai𝔽q*, and w(k)Wk is as in (4.13). In fact, this observation proves that the map (f)v(f) is a bijection between M(n) and N(n) (mentioned in the remark).

Note that since the diagrams va satisfy (I), (II) and (III), vaNμ, proving (a). On the other hand, (I), (II) and (III) imply that each vNμ must be of the form v=va for some aMμ, proving (b).

Remark. This bijection will prove useful in developing a generalization of the RSK correspondence in Chapter 5.

Let mμ= { aM(0) |row-sums and column-sums areμ } and for amμ, let (a)=Card {aij0|1i,j}.

Let μn. Then dim(μ)= amμ (q-1)(a) qn-(a)

Proof.

Theorem 4.2 gives dim(μ)= Mμ. We can obtain a matrix aMμ from a matrix amμ by selecting for each 1i,j(μ), a monic polynomial fij with nonzero constant term and degree aij. Conversely, every aMμ arises uniquely out of such a construction.

For a fixed amμ, the total number of ways to choose appropriate polynomials is 1i,jaij0 Card { f𝔽q[X]| fmonic,f(0)=0 andd(f)=aij } =(q-1)(a) qn-(a). The result follows by summing over all amμ.

Multiplication in μ

This section examines the implications of the unipotent Hecke algebra multiplication relations from Chapter 3 in the case G=GLn(𝔽q). The final goal is the algorithm given in Theorem 4.5.

Pictorial versions of eμveμ

Note that the map π: N = WT W wh w, forwW,hT, (4.15) is a surjective group homomorphism. Since WN, adjust the decomposition of (3.4) as follows. Let uN with π(u)=si1sir for r minimal. Then u has a unique decomposition u=u1u2uruT, whereuk=sik, uTT. (4.16) For t𝔽q, write uk(t)=sikxik(t).

Fix a composition μ. The decomposition U=1i<jn UijwhereUij =xij(t)|t𝔽q, implies eμ=1i<jn eij(1),where eij(k)=1q t𝔽qψ (-μijkt) xij(t). (4.17)

Pictorially, let uk as
ik k
,
(4.18)
uk(tk) as
ik k
,
(4.19)
eij(k) as
i j k
,
(4.20)
eμ as
μ(1) μ(2) μ(n-1)
.
(4.21)

Examples: 1. If u=u1u2u8uTN decomposes according to s3s1s2s3s1s4s2s3W with uT=diag(a,b,c,d,e), then u=
a b c d e 1 2 3 4 5 6 7 8
,u1(t1) u2(t2) u8(t8)uT=
a b c d e 1 2 3 4 5 6 7 8
and eμueμ=
a b c d e μ(1) μ(2) μ(3) μ(4) μ(1) μ(2) μ(3) μ(4) 1 2 3 4 5 6 7 8
.
2. If n=5, then (4.17) implies
μ(1) μ(2) μ(3) μ(4)
=
1 1 1 1 1 1 1 1 1 1
(eμ= e45(1) e35(1) e34(1) e25(1) e24(1) e23(1) e15(1) e14(1) e13(1) e12(1)).

The elements eij(k) also interact with U and N as follows (see also Section 3.3.1) sreij(k) sr = esr(i)sr(j) (k), (E1) eμveij(1) = eμv, vNμ,(πv) (i)<(πv) (j), (E2) eij(k) hεr(t) = hεr(t) eij(tδrit-δrjk), (E3) eμxij(t) = ψ(μijt) eμ=xij(t) eμ, (E4) or pictorially,
k
=
k
(E1)
k r s
=
krs-1 r s
(E3)
k a b
=ψ(kab)
k
and
k a b
=ψ(kab)
k
(E4)
and for vNμ,
1 v μ(i) μ(j-1)
=
v μ(i) μ(j-1)
(E2)

Suppose u=u1u2uruTNμ with uT=diag(h1,h2,,hn). Then using (4.17), (E3), (E1) and (E2), we can rewrite eμueμ as
μ(1) μ(2) μ(3) μ(n-1) h1 h2 h3 hn u1u2ur μ(1) μ(2) μ(3) μ(n-1)
=1qrt𝔽qr (ψfu)(t)
h1 h2 h3 hn u1(t1)u2(t2)ur(tr) μ(1) μ(2) μ(3) μ(n-1)
(4.22)
where fu𝔽q[y1,y2,,yr] is given by fu(y1,y2,,yr)=- μi1j1hi1 hj1-1y1- μi2j2hi2 hj2-1y2-- μirjrhir hjr-1yr, (4.23) for (ik,jk)=(a,b), if the kth crossing in u crosses the strands coming from the ath and bth top vertices.

Example (continued). Suppose u=u1u2u8uTN decomposes according to s3s1s2s3s1s4s2s3W and uT=diag(a,b,c,d,e)T (as in Example 1 above). Consider the ordering eμ=
1 1 1 1 1 1 1 1 1 1
,
or eμ = e13(1) e23(1) e12(1) e45(1) e15(1) e25(1) e14(1) e35(1) e24(1) e34(1) . Then eμueμ = eμu e13(1) e23(1) e12(1) e45(1) e15(1) e25(1) e14(1) e35(1) e24(1) e34(1) = eμu e12(1) e45(1) e15(1) e25(1) e14(1) e35(1) e24(1) e34(1) (by (E2)) = eμu1u2u8 uT e12(1) e45(1) e15(1) e25(1) e14(1) e35(1) e24(1) e34(1) = eμu1u2u8 e12(ab) e45(de) e15(ae) e25(be) e14(ad) e35(ce) e24(bd) e34(cd) UT(by (E3)) = eμu1 e34(ab)u2 e12(de)u3 e23(ae)u4 e34(be)u5 e12(ad)u6 e45(ce)u7 e23(bd)u8 e34(cd)uT (by (E1)) = eμq8t𝔽q8 (ψf)(t)u1 x34(abt1)u2 x12(det2)u3 x23(aet3) u8x34(cdt8) uT, where by (4.17), f=-y1-y2-y8. Therefore, by renormalizing eμueμ = eμq8t𝔽q8 (ψfu)(t) u1x34(t1) u2x12(t2) u3x23(t3) u8x34(t8)uT = eμq8t𝔽q8 (ψfu)(t) u1(t1) u2(t2) u3(t3) u8(t8)uT, where fu=-ba-1y1-ed-1y2-dc-1y8. Pictorially, by sliding the eij(1) down along the strands until they get stuck, this computation gives eμueμ=
a b c d e μ(1) μ(2) μ(3) μ(4) μ(1) μ(2) μ(3) μ(4) 1 2 3 4 5 6 7 8
=1q8 t𝔽q8 (ψfu)(t)
a b c d e μ(1) μ(2) μ(3) μ(4) 1 2 3 4 5 6 7 8
(4.24)
where fu=-ba-1y1-ed-1y2-dc-1y8 (as in (4.23)), since (i1,j1)=(1,2), (i2,j2)=(4,5), (i3,j3)=(1,5), etc.

Relations for multiplying basis elements.

Let u=u1u2uruTNμ decompose according to a minimal expression in W as in (4.16). Let vNμ and use (N3) and (N4) to write uTv=w·diag(a1,a2,,an) for some w=π(v)W (see (4.7)). Then use (4.22) to write (eμueμ) (eμveμ)= 1qrt𝔽qr (𝓅fu)(t)
μ(1) μ(2) μ(3) μ(n-1) a1 a2 a3 an w u1(t1)u2(t2)ur(tr) μ(1) μ(2) μ(3) μ(n-1)
(4.25)
(This form corresponds to Ξ(u,uTv) of Corollary 3.9).

Example (continued). If u is as in (4.24) and v=s2s3s2s1s2·diag(f,g,h,i,j)N, then (eμueμveμ)= 1q8t𝔽q8 (ψfu)(t)
fd gc ah bi ej μ(1) μ(2) μ(3) μ(4) μ(1) μ(2) μ(3) μ(4) 1 2 3 4 5 6 7 8
.

Consider the crossing in (4.25) corresponding to ur(tr). There are two possibilities.

Case 1 the strands that cross at r do not cross again as they go up to the top of the diagram ((urw)>(w)),

Case 2 the strands that cross at r cross once on the way up to the top of the diagram ((urw)<(w)).

In the first case, eμueμveμ= 1qrt𝔽qr (ψfu)(t)
r μ(i) μ(j-1) a1 ai aj an w u1(t1)ur-1(tr-1) μ(1) μ(2) μ(3) μ(n-1)
,
so (UN1), (UN2) and (E4) imply eμueμveμ= 1qrt𝔽qr (ψf(+0)) (t)
μ(1) μ(2) μ(3) μ(n-1) a1 a2 a3 an urw u1(t1)ur(tr) μ(1) μ(2) μ(3) μ(n-1)
(4.26)
where f(+0)=fu+μijajai-1yr.

In the second case, eμueμveμ= 1qrt𝔽qr (ψfu)(t)
r μ(i) μ(j-1) a1 ai aj an w u1(t1)ur-1(tr-1) μ(1) μ(2) μ(3) μ(n-1)
.
Use (UN3) and (N1) to split the sum into two parts corresponding to tr=0 and tr0, eμueμveμ = 1qrt𝔽qrtr=0 (ψf(-0))(t)
μ(i) μ(j-1) a1 ai aj an urw u1(t1)ur-1(tr-1) μ(1) μ(2) μ(3) μ(n-1)
+1qr t𝔽qrtr𝔽q* (ψfu)(t)
r 1 -1 tr tr-1 μ(i) μ(j-1) a1 aitr -ajtr-1 an w u1(t1)ur-1(tr-1) μ(1) μ(2) μ(3) μ(n-1)
where f(-0)=fu. Now use (UN1), (UN2), (U2), (U1) and (E4) on the second sum to get eμueμveμ = 1qrt𝔽qrtr=0 (ψf(-0))(t)
μ(1) μ(2) μ(3) μ(n-1) a1 a2 a3 an urw u1(t1)ur-1(tr-1) μ(1) μ(2) μ(3) μ(n-1)
+1qr t𝔽qrtr𝔽q* (ψf(1))(t)
μ(i) μ(j-1) a1 aitr -ajtr-1 an w u1(t1)ur-1(tr-1) μ(1) μ(2) μ(3) μ(n-1)
(4.27)
where f(1)=φr(fu)+μijajai-1yr-1, and φr(f) is defined by t𝔽qrtr𝔽q* (ψf)(t)
r 1 tr-1 u1(t1)ur-1(tr-1) μ(1) μ(2) μ(3) μ(n-1)
=t𝔽qrtr𝔽q* (ψφr(f))(t)
r u1(t1)ur-1(tr-1) μ(1) μ(2) μ(3) μ(n-1)
.
(*)

Remarks: (a) We could have applied these steps for any u, u, and v, so we can iterate the process with each sum. (b) The most complex step in these computations is determining φr. The following section will develop an efficient algorithm for computing the right-hand side of (*).

Computing φk via painting, paths and sinks.

Painting algorithm (uk). Suppose u=u1u2urN decomposes according to si1si2sikW (assume uT=1). Paint flows down strands (by gravity). Each step is illustrated with the example u=u1u2u8, decomposed according to s3s1s2s3s1s4s2s3W. (1) Paint the left [respectively right] strand exiting k red [blue] all the way to the bottom of the diagram.
1 2 3 4 5 6 7 8
wherekis 8 .
(2) For each crossing that the red [blue] strand passes through, paint the right [left] strand (if possible) red [blue] until that strand either reaches the bottom or crosses the blue [red] strand of (1).
1 2 3 4 5 6 7 8
1 2 3 4 5 6 7 8
1 2 3 4 5 6 7 8
(4.28)
Let uk u1(t1) u2(t2) uk(tk) painted according to the above algorithm (4.29)

Sinks and paths. The diagram uk has a crossed sink at j if j is a crossing between a red strand and a blue one, or
j
Note that since u is decomposed according to a minimal expression in U, there will be no crossings of the form
j
(since
j
would imply
j j
for somejj.)
The diagram uk has a bottom sink at j if a red strand enters jth bottom vertex and a blue strand enters the (j+1)st bottom vertex, or
j j+1
A red [respectively blue] path p from a sink s (either crossed or bottom) in uk is an increasing sequence j1<j2<< jl=k, such that in uk (a) jm is directly connected (no intervening crossings) to jm+1 by a red [blue] strand, (b) if s is a crossed sink, then j1=s, (b') if s is a bottom sink, then in a red path, the sth bottom vertex connects to the crossing j1 with a red strand. in a blue path, the (s+1)st bottom vertex connects to the crossing j1 with a blue strand. Let P=(uk,s)= {red paths fromsinuk} andP:(uk,s) ={blue paths fromsinuk} (4.30) The weight of a path p is wt(p)= { pswitchesstrands ati yi, ifpP= (uk,s), pswitchesstrands ati (-yi), ifpP: (uk,s). (4.31) Each sink s in uk (either crossed j or bottom j) has an associated polynomial gs𝔽q[y1,y2,,yk-1,yk-1] given by gs=pP=(uk,s)pP:(uk,s) wt(p)yk-1wt(p). (4.32)

Example (continued). If u=u1u2u8 decomposes according to s3s1s2s3s1s4s2s3 (as in (4.28)), then P=(u8,4)= {
1 3 5 7 8
,
1 4 7 8
}
P:(u8,4) = {
6 8
}
with wt(1<3<5<7<8)=y5, wt(1<4<7<8)=y1y7, and wt(6<8)=1. The corresponding polynomial is g4=y5y8-1 +y1y7y8-1. (4.33)

Let u=u1u2ur and φr be as in (4.27) and (*); suppose ur is painted as above. Then φr(f)=f |{yjyj-gj|ja crossed sink} +ja bottomsink μ(j)gj.

Proof.

In the painting,
marks a strand travelled by
a
and
marks a strand travelled by
a
.
Substitutions due to crossed sinks correspond to the normalizations in relation (U2), and the sum over bottom sinks comes from applications of relation (E4).

Example (continued). Recall u=s3s1s2s3s1s4s2s3. Then u8 at 2, 3, and 4. The only bottom sink is at 4. Therefore, φ8(f)=f |y4y4-g4y3y3-g3y2y2-g2+ μ(4)g4=f | y4y4+y7y8-1y6 y3y3+y5y8-1y6 y2y2+y2y8-1y6 + μ(4) (y5y8-1+y1y7y8-1). (for example, g4 was computed in (4.33)).

A multiplication algorithm

(The algorithm). Let G=GLn(𝔽q) and u,vNμ. An algorithm for multiplying eμueμ and eμveμ is (1) Decompose u=u1u2uruT according to some minimal expression in W (as in (4.16)). (2) Put eμueμveμ into the form specified by (4.25), with uTv=w·diag(a1,a2,,an) (w=π(v)W). (3) Complete the following (a) If (urw)>(w), then apply relation (4.26). (b) If (urw)<(w), then apply relation (4.27), using (u1u2ur)r and Lemma 4.4 to compute φr. (4) If r>1, then reapply (3) to each sum with rr-1 and with (a) wurw, after using (3a) or using (3b), in the first sum, (b) ww, after using (3b), in the second sum. (5) Set all diagrams not in Nμ to zero.

Sample computation. Suppose n=3 and μ(i)=1 for all 1i3 (i.e.. the Gelfand-Graev case). Then Nμ= {
a a a
,
a a b
,
a b b
,
a b c
|a,b,c𝔽q* }
Suppose u=
a b c
andv=
d e e
.

1. Theorem 4.5 (1): Let u=u1u2u3uTNμ decompose according to s2s1s2W, with uT=diag(a,b,c).

2. Theorem 4.5 (2): By (4.25) ( eμ
a b c
eμ )
( eμ
d e e
eμ )
=1q3 t𝔽q3 (ψfu)(t)
1 2 3 1 1 1 1 cd ae be
with uTv=s2s1·diag(cd,ae,be) (so w=s2s1), and fu=-bay1-cby3 (as in (4.23)).

3. Theorem 4.5 (3b): Since (u3w)<(w), paint u1(t1)u2(t2)u3(t3) to get (u1u2u3)3 (as in (4.29)), =1q3t𝔽q3 (ψfu)(t)
1 2 3 1 1 1 1 cd ae be
Now apply (4.27), =1q3t𝔽q3t3=0 (ψf(-0))(t)
1 2 1 1 1 1 cd ae be
+1q3t𝔽q3t3𝔽q* (ψf(1))(t)
1 2 1 1 1 1 cdt3 ae -bet3-1
where f(-0)=-bay1-cby3 and by Lemma 4.4, f(1)=φ3(fu) +μ13becdy3-1 =-bay1+bay2 y3-1-cdy3+ y3-1.

4. Theorem 4.5 (4): Set r2 with wurw=s1 in the first sum and ww in the second sum.

5. Theorem 4.5 (3a) (3b): In the first sum, (u2s1)<(s1), so paint u1(t1)u2(t2) to get (u1u2)2. In the second sum, (u2s2s1)>(s2s1), so apply (4.26), =1q3t𝔽q3t3=0 (ψf(-0))(t)
1 2 1 1 1 1 cd ae be
+1q3t𝔽q3t3𝔽q* (ψf(+0,1))(t)
1 1 1 1 1 cdt3 ae -bet3-1
where f(+0,1)=-bay1+bay2y3-1-cdy3+y3-1-μ(3)bay2y3-1=-bay1-cby3+y3-1. Now apply (4.27) to the first sum, = 1q3t𝔽q3t2=t3=0 (ψf(-0,-0))(t)
1 1 1 1 1 cd ae be
+1q3t𝔽q3t2𝔽q*,t3=0 (ψf(1,-0)) (t)
1 1 1 1 1 cdt2 -aet2-1 be
+1q3t𝔽q3t3𝔽q* (ψf(+0,1))(t)
1 1 1 1 1 cdt3 ae -bet3-1
where f(-0,-0)=-bay1-cby3 and f(1,-0)=φ(f(-0))+μ(1)aecdy2-1=-bay1-y2-1y1+aecdy2-1.

6. Theorem 4.5 (4): Set r=1 with wu2s1=1 in the first sum, ws1 in the second sum, and ws1s2s1 in the third sum.

7. Theorem 4.5 (3a) (3a) (3b): In the first sum (s21)>(1), so apply (4.26); in the second sum (s2s1)>(s1), so apply (4.26); in the third sum, (s2s1s2s1)<(s1s2s1), so paint u1(t1) to get u11, = 1q3t𝔽q3t2=t3=0 (ψf(+0,-0,-0))(t)
1 1 1 1 cd ae be
+1q3t𝔽q3t2𝔽q*,t3=0 (ψf(+0,1,-0))(t)
1 1 1 1 cdt2 -aet2-1 be
+1q3t𝔽q3t3𝔽q* (ψf(+0,1))(t)
1 1 1 1 1 cdt3 ae -bet3-1
where f(+0,-0,-0)=-bay1-cby3+bay1=-cby3 and f(+0,1,-0)=-bay1+aecdy2-1-y2-1y1. Now apply (4.27) to the third sum = 1q3t𝔽q3t2=t3=0 (ψf(+0,-0,-0))(t)
1 1 1 1 cd ae be
+1q3t𝔽q3t2𝔽q*,t3=0 (ψf(+0,1,-0))(t)
1 1 1 1 cdt2 -aet2-1 be
+1q3t𝔽q3t1=0,t3𝔽q* (ψf(-0,+0,1))(t)
1 1 1 1 cdt3 ae -bet3-1
+1q3t𝔽q3t1,t3𝔽q* (ψf(1,+0,1))(t)
1 1 1 1 cdt1t3 -aet1-1 -bet3-1
where f(-0,+0,1)=-cby3+y3-1 and f(1,+0,1)= φ1(f(+0,1)) +μ(1)aecdy1-1 y3-1=-bay1- cby3+y3-1+ y1-1+aecd y1-1y3-1.

8. Theorem 4.5 (5): The first sum contains no elements of Nμ, so set it to zero. The second sum contains elements of Nμ when be=-aet2-1, so set t2=-ab. The third sum contains elements of Nμ when cdt3=ae, so set t3=aecd. All the terms in the fourth sum are basis elements. = 0+1q3t𝔽q3t2=-ab,t3=0 (ψf(+0,1,-0))(t)
1 1 1 1 -ab-1cd be be
+1q3t𝔽q3t1=0,t3=aecd (ψf(-0,+0,1))(t)
1 1 1 1 ae ae -a-1bcd
+1q3t𝔽q3t1,t3𝔽q* (ψf(1,+0,1))(t)
1 1 1 1 cdt1t3 -aet1-1 -bet3-1
= 1q2ψ(-becd)
1 1 1 1 -ab-1cd be be
+1q2ψ(-aebd+cdae)
1 1 1 1 ae ae -a-1bcd
+1q2t1,t3𝔽q* ψ ( -bat1 -cbt3 +t3-1 +t1-1 +aecd t1-1 t3-1 )
1 1 1 1 cdt1t3 -aet1-1 -bet3-1
.

Notes and References

This is an excerpt of the PhD thesis Unipotent Hecke algebras: the structure, representation theory, and combinatorics by F. Nathaniel Edgar Thiem.

page history