Cluster Variables on Certain Double Bruhat Cells of Type (u, e) and Monomial Realizations
of Crystal Bases of Type A
?Yuki KANAKUBO and Toshiki NAKASHIMA
Division of Mathematics, Sophia University, Yonban-cho 4, Chiyoda-ku, Tokyo 102-0081, Japan
E-mail: j chi sen you ky@sophia.ac.jp,toshiki@sophia.ac.jp
Received October 01, 2014, in final form April 14, 2015; Published online April 23, 2015 http://dx.doi.org/10.3842/SIGMA.2015.033
Abstract. LetGbe a simply connected simple algebraic group overC,B andB− be two opposite Borel subgroups inGandW be the Weyl group. Foru,v∈W, it is known that the coordinate ringC[Gu,v] of the double Bruhat cellGu,v=BuB∩B−vB− is isomorphic to an upper cluster algebra ¯A(i)C and the generalized minors {∆(k;i)} are the cluster variables belonging to a given initial seed in C[Gu,v] [Berenstein A., Fomin S., Zelevinsky A.,Duke Math. J. 126(2005), 1–52]. In the caseG= SLr+1(C),v=eand some specialu∈W, we shall describe the generalized minors {∆(k;i)} as summations of monomial realizations of certain Demazure crystals.
Key words: cluster variables; double Bruhat cells; crystal bases; monomial realizations, generalized minors
2010 Mathematics Subject Classification: 13F60; 81R50; 17B37
1 Introduction
As is well-known that theory of cluster algebras has been initiated by S. Fomin and A. Zelevinsky in the study of product expressions by q-commuting elements for upper global bases (= dual canonical bases). Crystal bases are obtained from global bases considering the parameterq at 0.
Thus, we can guess that they should be deeply related each other at their origins.
LetGbe a simply connected simple algebraic group overCof rankr. LetB and B− be two opposite Borel subgroups in G, N ⊂B and N− ⊂B− their unipotent radicals, H := B∩B−
a maximal torus, and W the associated Weyl group. In [1], it is shown that for u, v ∈ W the coordinate ring C[Gu,v] of double Bruhat cell Gu,v := BuB∩B−vB− has the structure of an upper cluster algebra. The initial cluster variables of this upper cluster algebras are given as certain generalized minors on Gu,v.
In [11], the second author revealed the relations between some generalized minors and mono- mial realizations of crystal bases. A naive definition of monomial realizations of crystal bases is as follows (see Section 3 for the exact definitions): Let Y be the set of monomials in infinitely many variables (see Section 3, equation (3.2)). We shall define the crystal structures onY as- sociated with certain set of integersp= (pi,j)1≤i6=j≤r and a Cartan matrix. And we can obtain a crystal for an irreducible module as a connected component ofY. For example, for typeA4and pi,j = 1 if i < j and pi,j = 0 if i > j, we have the following crystal graph of the crystal B(Λ3), where Λ3 is the 3rdfundamental weight. The set of integerspgives the cyclic sequence of indices
?This paper is a contribution to the Special Issue on New Directions in Lie Theory. The full collection is available athttp://www.emis.de/journals/SIGMA/LieTheory2014.html
. . .123412341234. . . and we associate variables {τj}as follows
. . . 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 . . .
. . . τ−4 τ−3 τ−2 τ−1 τ1 τ2 τ3 τ4 τ5 τ6 τ7 τ8 τ9 τ10 . . .
Note that the skip between τ7 andτ8 orτ9 and τ10 means no corresponding variable appears in the following crystal graph
τ−2 τ−1τ2
τ3
τ2
τ4
τ−1τ5
τ6
τ3τ5
τ4τ6
τ5
τ7
τ−1
τ8
τ3
τ4τ8
τ6
τ7τ8
1 τ9 ,
3 // 4 //
2 4 //
1 4 // //
2
1
//
3
2
//3
1
(1.1)
where the highest weight monomialτ−2 has a weight Λ3 and the lowest weight monomial τ1
9 has a weight −Λ2.
As stated above, the initial cluster variables on Gu,v are expressed by generalized minors {∆(k;i)|1 ≤ k ≤l(u) +l(v)}, where i is the reduced expression of (u, v) ∈ W ×W. Now, as an example we consider the case G= SL5(C) as above. Let W =S5 =hsi|1≤i≤ 4i be the symmetric group and set u := s1s2s3s4s1s2s3s1s2s1, v := e, and set a reduced word i foru as i:= (1,2,3,4,1,2,3,1,2,1). We have for x= (xi,j)∈SL5(C)
∆(6;i)(x) =
x31 x32
x41 x42
=x31x42−x32x41. (1.2)
See Section 5for the detailed explanation.
Now, let us consider the generalized minors onLu,v :=N uN ∩B−vB− instead ofGu,v since their difference is, indeed, only the factor from the torus part. We call Lu,v reduced double Bruhat cell. For the above u,v, there exists a birational mapxLi : (C×)10−→∼ Lu,v given by
xLi(τ1, . . . , τ10)
=x−1(τ1)x−2(τ2)x−3(τ3)x−4(τ4)x−1(τ5)x−2(τ6)x−3(τ7)x−1(τ8)x−2(τ9)x−1(τ10)
=
1
τ1τ5τ8τ10 0 0 0 0
A τ1τ5τ8τ10
τ2τ6τ9
0 0 0
B C τ2τ6τ9
τ3τ7
0 0
D E τ3τ9
τ4
+τ6τ9 τ7
τ3τ7 τ4
0
1 τ10 τ9 τ7 τ4
, (1.3)
where
A= τ1τ5τ8
τ2τ6τ9 + τ1τ5
τ2τ6τ10 + τ1
τ2τ8τ10 + 1 τ5τ8τ10, B = τ2τ6
τ3τ7
+τ2τ8 τ3τ9
+τ5τ8 τ6τ9
+ τ2 τ3τ10
+ τ5 τ6τ10
+ 1
τ8τ10
, C = τ2τ6τ10
τ3τ7 +τ2τ8τ10
τ3τ9 +τ5τ8τ10 τ6τ9 , D= τ3
τ4
+ τ6 τ7
+τ8 τ9
+ 1 τ10
, E= τ3τ10 τ4
+τ6τ10 τ7
+τ8τ10 τ9
,
x−i(t) =ith
. ..
t−1 0
1 t
. ..
.
Therefore, by (1.2) and (1.3) we find
∆L(6;i)(τ) := ∆(6;i)◦xLi
(τ1, . . . , τ10) =
B C
D E
= τ2 τ4
+τ3τ5 τ4τ6
+τ5 τ7
+ τ3 τ4τ8
+ τ6 τ7τ8
+ 1 τ9
. (1.4)
Now, observing the crystal graph (1.1) and the Laurent polynomial (1.4), we realize that each term in (1.4) appears in (1.1) and they constitute so-called lower Demazure crystal associated with the element u≤6 ∈S5 [7].
Those facts motivate us to find a new linkage between the cluster variables onLu,v ⊂Gu,v and the monomial realizations of crystals.
In this paper, we shall treat the caseG= SLr+1(C),v=eand some specialu∈W =Sr+1. More precisely, we treat an element u∈W whose reduced word (Definition2.1) can be written as a left factor of the standard longest word (1,2,3, . . . , r,1,2,3, . . . ,(r−1), . . . ,1,2,1):
u=s1s2· · ·srs1· · ·sr−1· · ·s1· · ·sr−m+2s1· · ·sin,
wheren:=l(u) is the length ofuand 1≤in≤r−m+ 1. And we treat (reduced) double Bruhat cells of the formGu,e:=BuB∩B−andLu,e:=N uN∩B−, whereB(resp.B−) is the subgroup of upper (resp. lower) triangular matrices inG= SLr+1(C). Then generalized minors are a part of classical minors (Definition4.10). This case matches well to the Demazure crystals. In fact, we shall describe generalized minors in terms of summations over certain monomial realizations of Demazure crystals in the main result Theorem5.6. For example, (1.4) shows that the generalized minor ∆L(6;i)(τ) is described in terms of summation over certain monomial realization of the Demazure crystal Bu≤6− (−Λ2).
In forthcoming paper, we shall treat more general setting, like as, the Weyl group element v ∈ W is non-identity or type C. In these cases, the generalized minors are described also by monomial realizations of crystals.
2 Factorization theorem for type A
In this section, we shall introduce (reduced) double Bruhat cells Gu,v,Lu,v, and their properties in the case G= SLr+1(C), v =eand some special u∈W. In [2, 3], these properties had been proven for simply connected, connected, semisimple complex algebraic groups and arbitrary u, v∈W.
Forl∈Z>0, we set [1, l] :={1,2,3, . . . , l}.
2.1 Double Bruhat cells
LetG= SLr+1(C) be the simple complex algebraic group of type Ar,B andB−be two opposite Borel subgroups in G, that is, B (resp. B−) is the subgroup of upper (resp. lower) triangular matrices inG= SLr+1(C). LetN ⊂B andN−⊂B− be their unipotent radicals,H :=B∩B−
a maximal torus, and W := NormG(H)/H the Weyl group. In this case, Weyl group W is isomorphic to the symmetric groupSr+1.
We have two kinds of Bruhat decompositions ofG as follows
G= a
u∈W
BuB = a
u∈W
B−uB−.
Then, for u,v∈W, we define thedouble Bruhat cell Gu,v as follows Gu,v :=BuB∩B−vB−.
This is biregularly isomorphic to a Zariski open subset of an affine space of dimensionr+l(u) + l(v) [3, Theorem 1.1].
We also define thereduced double Bruhat cellLu,v as follows Lu,v :=N uN ∩B−vB−⊂Gu,v.
As is similar to the caseGu,v,Lu,v is biregularly isomorphic to a Zariski open subset of an affine space of dimension l(u) +l(v) [2, Proposition 4.4].
Definition 2.1. Letu =si1· · ·sin be a reduced expression ofu ∈W,i1, . . . , in ∈[1, r]. Then the finite sequence
i:= (i1, . . . , in)
is called reduced word iforu.
In this paper, we treat (reduced) double Bruhat cells of the form Gu,e := BuB∩B− and Lu,e := N uN ∩B−, where u ∈ W is an element whose reduced word can be written as a left factor of (1,2,3, . . . , r,1,2,3, . . . ,(r−1), . . . ,1,2,1):
u=s1s2· · ·srs1· · ·sr−1· · ·s1· · ·sr−m+2s1· · ·sin, (2.1) where n:=l(u) is the length ofu and 1≤in≤r−m+ 1. Let i be a reduced word ofu:
i= (1, . . . , r
| {z }
1stcycle
,1, . . . ,(r−1)
| {z }
2ndcycle
, . . . ,1, . . . ,(r−m+ 2)
| {z }
(m−1)thcycle
,1, . . . , in
| {z }
mth cycle
). (2.2)
Note that (1,2,3, . . . , r,1,2,3, . . . ,(r−1), . . . ,1,2,1) is a reduced word of the longest element inW.
2.2 Factorization theorem for type A
In this subsection, we shall introduce the isomorphisms between double Bruhat cell Gu,e and H×(C×)l(u), and between Lu,e and (C×)l(u). As in the previous section, we consider the case G:= SLr+1(C). We set g:= Lie(G) with the Cartan decomposition g=n−⊕h⊕n. Letei,fi (i ∈ [1, r]) be the generators of n, n−. For i ∈ [1, r] and t ∈ C, we set xi(t) := exp(tei), yi := exp(tfi). Let ϕi : SL2(C)→G be the canonical embedding corresponding to each simple rootαi. Then we have
xi(t) =ϕi
1 t 0 1
, yi(t) =ϕi
1 0 t 1
. We can expressxi(t), yi(t), as the following matrices
xi(t) =ith
. ..
1 t 0 1
. ..
, yi(t) =ith
. ..
1 0 t 1
. ..
. (2.3)
For a reduced word i= (i1, i2, . . . , in), we define a map xGi :H×Cn→Gas xGi (a;t1, . . . tn) :=a·yi1(t1)yi2(t2)· · ·yin(tn).
Theorem 2.2 ([3, Theorem 1.2]). We setu∈W and its reduced word i as in (2.1) and (2.2).
The map xGi defined above can be restricted to a biregular isomorphism between H×(C×)l(u) and a Zariski open subset of Gu,e.
Next, fori∈[1, r] andt∈C×, we define as follows α∨i(t) :=ϕi
t 0 0 t−1
, x−i(t) :=yi(t)α∨i(t−1) =ϕi
t−1 0
1 t
. We can expressx−i(t) and α∨i(t) as the following matrices
x−i(t) =ith
. ..
t−1 0
1 t
. ..
, α∨i(t) = diag(1, . . . ,1,
i
∨
t,
i+1
∨
t−1,1, . . . ,1). (2.4)
Fori= (i1, . . . , in) (i1, . . . , in∈[1, r]), we define a mapxLi :Cn→Gas xLi(t1, . . . , tn) :=x−i1(t1)· · ·x−in(tn).
We have the following theorem which is similar to the previous one.
Theorem 2.3([2, Proposition 4.5]). We setu∈W and its reduced wordi as in (2.1)and (2.2).
The map xLi defined above can be restricted to a biregular isomorphism between (C×)l(u) and a Zariski open subset of Lu,e.
Finally, we define a map ¯xGi :H×(C×)n→Gu,v as
¯
xGi (a;t1, . . . , tn) =axLi (t1, . . . , tn).
Proposition 2.4. In the above setting, the map x¯Gi is a biregular isomorphism between H × (C×)n and a Zariski open subset ofGu,e.
Proof . We set l0 := 0,l1 :=r,l2 :=r+ (r−1), . . . , lm:=r+ (r−1) +· · ·+ (r−m+ 1). We define a mapφ:H×(C×)n→H×(C×)n,t= (a;t1, . . . , tn)7→(a(t);τ1(t), . . . , τn(t)) as
a(t) =a·α∨1(t1)−1· · ·α∨r(tr)−1
| {z }
1stcycle
· · ·α∨1(tlm−1+1)−1· · ·α∨in(tlm−1+in)−1
| {z }
mthcycle
,
τls+j(t) = (tls+1+j−1tls+2+j−1· · ·tlm−1+j−1)(tls+j+1tls+1+j+1· · ·tlm−1+j+1)
tls+j(tls+1+j· · ·tlm−1+j)2 , (2.5) where in (2.5), if i does not include j (resp. j+ 1, j−1) in ζth cycle then we set tlζ−1+j = 1 (resp. tlζ−1+j+1= 1, tlζ−1+j−1= 1). This is a biregular isomorphism.
Let us prove
¯
xGi (a;t1, . . . , tn) = xGi ◦φ
(a;t1, . . . , tn),
which implies that ¯xGi :H×(C×)n→Gu,e is a biregular isomorphism by Theorem2.2.
First, we can verify the following relations by the explicit forms (2.3), (2.4) and direct calcu- lations:
α∨i(c)−1yj(t) =
yi(c2t)α∨i(c)−1 if i=j, yj(c−1t)α∨i(c)−1 if |i−j|= 1, yj(t)α∨i(c)−1 otherwise,
(2.6)
for 1≤i,j≤r and c, t∈C×. On the other hand, we obtain
(xGi ◦φ)(a;t1, . . . , tn) =a×α∨1(t1)−1· · ·α∨r(tr)−1· · ·α∨1(tlm−1+1)−1· · ·αi∨n(tlm−1+in)−1
×y1(τ1(t))y2(τ2(t))· · ·yr(τr(t))· · ·y1(τlm−1+1(t))· · ·yin(τlm−1+in(t)). (2.7) For each sand j, let us move α∨j(tls+j)−1, α∨j+1(tls+j+1)−1, . . . , α∨in(tlm−1+in)−1 to the right of yj(τls+j(t)) by using the relations (2.6). For example,
α∨j(tls+j)−1· · ·α∨j−1(tlm−1+j−1)−1α∨j(tlm−1+j)−1
×α∨j+1(tlm−1+j+1)−1· · ·α∨in(tlm−1+in)−1yj(τls+j(t))
=α∨j(tls+j)−1· · ·yj t2lm−1+j
tlm−1+j−1tlm−1+j+1τls+j(t)
!
α∨j−1(tlm−1+j−1)−1
×α∨j(tlm−1+j)−1α∨j+1(tlm−1+j+1)−1· · ·α∨in(tlm−1+in)−1. Repeating this argument, we have
=yj
(tls+j· · ·tlm−1+j)2
(tls+j−1· · ·tlm−1+j−1)(tls+j+1· · ·tlm−1+j+1)τls+j(t)
α∨j(tls+j)−1· · ·α∨in(tlm−1+in)−1. Note that (tls+j...tlm−1+j)
2
(tls+j−1···tlm−1+j−1)(tls+j+1···tlm−1+j+1)τls+j(t) =tls+j. By (2.7), we have xGi ◦φ
(a;t1, . . . , tn) =a·y1(t1)α∨1(t1)−1· · ·yr(tr)α∨r(tr)−1· · ·
×y1(tlm−1+1)α∨1(tlm−1+1)−1· · ·yin(tlm−1+in)α∨in(tlm−1+in)−1
=a·x−1(t1)· · ·x−r(tr)· · ·x−1(tlm−1+1)· · ·x−in(tlm−1+in) = ¯xGi (a;t1, . . . , tn).
3 Monomial realizations of crystal bases
In this section, we review the monomial realizations of crystals [6,8,10]. LetI :={1,2, . . . , r}
be a finite index set.
3.1 Monomial realizations of crystal bases for type A
Definition 3.1. LetA= (aij)i,j∈I be the Cartan matrix of type Ar: A= (aij)i,j∈I is defined as
aij =
2 if i=j,
−1 if |i−j|= 1, 0 otherwise.
(3.1)
Let Π ={αi|i∈I} (resp. Π∨ ={hi|i∈I}) be the set of simple roots (resp. co-roots), and P be the weight lattice. A crystal associated with the Cartan matrix A is a set B together with the maps wt :B →P, ˜ei, ˜fi :B∪ {0} →B∪ {0} and εi,ϕi :B →Z∪ {−∞},i∈I, satisfying the following properties: Forb∈B,i∈I,
(i) ϕi(b)−εi(b) =hhi, wt(b)i, (ii) wt( ˜eib) = wt(b) +αi, if ˜eib∈B, (iii) wt( ˜fib) = wt(b)−αi, if ˜fib∈B,
(iv) εi( ˜eib) =εi(b)−1,ϕi( ˜eib) =ϕi(b) + 1 if ˜eib∈B, (v) εi( ˜fib) =εi(b) + 1,ϕi( ˜fib) =ϕi(b)−1 if ˜fib∈B, (vi) ˜fib=b0 ⇔b= ˜eib0, ifb,b0∈B,
(vii) ϕi(b) =−∞,b∈B,⇒e˜ib= ˜fib= 0.
LetUq(g) be the universal enveloping algebra associated with the Cartan matrixA in (3.1), and g = slr+1(C). Let B+(λ) (resp. B−(λ)) be the crystal base of the Uq(g)-highest (resp.
lowest) weight module [5, 9]. Note that B+(λ) = B−(w0λ), where w0 is the longest element of W. In particular, in the caseλ=MΛd,M ∈Z>0, we have
B+(MΛd) =B−(−MΛr−d+1).
Let us introduce monomial realizations which realize each element of B±(λ) as a certain Laurent monomial.
First, we define a set of integersp= (pj,i)j,i∈I, j6=i such that pj,i =
(1 if j < i, 0 if i < j.
Second, for doubly-indexed variables {Ys,i|i∈I,s∈Z}, we define the set of monomials Y :=
Y = Y
s∈Z, i∈I
Ys,iζs,i
ζs,i ∈Z, ζs,i= 0 except for finitely many (s, i)
. (3.2)
Finally, we define maps wt :Y →P,εi,ϕi :Y →Z,i∈I. For Y = Q
s∈Z, i∈I
Ys,iζs,i∈ Y,
wt(Y) :=X
i,s
ζs,iΛi, ϕi(Y) := max
X
k≤s
ζk,i|s∈Z
, εi(Y) :=ϕi(Y)−wt(Y)(hi).
We set
As,i:=Ys,iYs+1,iY
j6=i
Ys+paj,i
j,i,j =
Ys,1Ys+1,1
Ys,2 if i= 1, Ys,iYs+1,i
Ys,i+1Ys+1,i−1
if 2≤i≤r−1, Ys,rYs+1,r
Ys+1,r−1
if i=r,
(3.3)
and define the Kashiwara operators as follows f˜iY =
(A−1n
fi,iY if ϕi(Y)>0,
0 if ϕi(Y) = 0, ˜eiY =
(Anei,iY if εi(Y)>0, 0 if εi(Y) = 0, where
nfi := min
n
ϕi(Y) =X
k≤n
ζk,i
, nei := max
n
ϕi(Y) =X
k≤n
ζk,i
. (3.4)
Then the following theorem holds:
Theorem 3.2 ([8,10]).
(i) For the set p= (pj,i) as above, (Y,wt, ϕi, εi,f˜i,e˜i)i∈I is a crystal. When we emphasize p, we write Y as Y(p).
(ii) If a monomial Y ∈ Y(p) satisfies εi(Y) = 0 (resp. ϕi(Y) = 0) for all i ∈ I, then the connected component containing Y is isomorphic to B+(wt(Y)) (resp.B−(wt(Y))).
Definition 3.3. Let Y ∈ Y(p) be a monomial and let B be the unique connected component in Y(p) including Y. Suppose that λ is the highest (resp. lowest) weight of B. We denote the embedding
µY : B+(λ),→B⊂ Y(p) (resp. µY : B−(λ),→B).
Note that ifY and Y0 are in the same component thenµY =µY0.
Remark 3.4. The actions of ˜ei and ˜fi on Y are determined by wt(Y)(hi), ϕi(Y) and εi(Y), which are determined by the factors Ys,i±1, s ∈ Z. Thus, when we consider the actions of ˜ei and ˜fi, we need to see the factors{Ys,i±1}s∈Z only.
Example 3.5. For λ = βΛd (resp. λ = −βΛd), β ∈ Z>0, d ∈ I, we can embed B+(λ) (resp. B−(λ)) inY as a crystal by
vλ 7→Yβ+γ,dYβ−1+γ,d· · ·Y1+γ,d,
resp.vλ 7→ 1
Yβ+γ,dYβ−1+γ,d· · ·Y1+γ,d
,
where vλ is the highest (resp. lowest) weight vector of B+(λ) (resp. B−(λ)), and γ is an ar- bitrary integer. For Y+ := Yβ+γ,dYβ−1+γ,d· · ·Y1+γ,d (resp. Y− := Y 1
β+γ,dYβ−1+γ,d···Y1+γ,d), µY+ (resp. µY−) denotes the embedding in Definition3.3. Then Y+ (resp. Y−) is the highest (resp.
lowest) weight vector in µY+(B+(λ)) (resp. µY−(B−(λ))).
We set l0 := 0, l1 := r, l2 := r+ (r−1), . . . , ls := r+ (r−1) +· · ·+ (r −s+ 1), . . ., lr :=r+ (r−1) +· · ·+ 2 + 1 and changing the variablesYs,j toτls+j, 1≤j≤r−s. Fors <0, we transform the variables Ys,j toτ−(r+1−j), 1≤j≤r,
. . . r 1 . . . r−1 r 1 . . . r−1 r 1 2 . . .
. . . τ−1 τl0+1 . . . τl0+r−1 τl0+r τl1+1 . . . τl1+r−1 τl2+1 τl2+2 . . .
Remark 3.6. In the above setting, the variables {Ys,j|r −s < j} do not correspond to any variables in τ. As we have seen in (1.1), these variables do not appear in the crystal base which we treat in this paper. In other words, we only need variables associated with
j= (1, . . . , r,1, . . . , r−1, . . . ,1,2,1),
which coincides with a specific reduced word of the longest element of W. Remark 3.7. For the variablesτls+0,τls+r+1 (0≤s≤m−1) we understand
τls+0 =τls+r+1= 1.
For example, ifi= 1 then τls+i−1 = 1.
Example 3.8. Let us consider the action of ˜e1 on the monomial τ 1
lr−1+1. Following the method in Section 3.1, we have wt(τ 1
lr−1+1) = −Λ1, ϕ1(τ 1
lr−1+1) = 0, ε1(τ 1
lr−1+1) = ϕ1(τ 1
lr−1+1) − wt(τ 1
lr−1+1)(h1) = 1, and ne1 =r−2. Thus, since we have Ar−2,1 =τlr−2+1τlr−1+1τlar−2+p2,1
2,1+2 =
τlr−2+1τlr−1+1
τlr−2+2 , we get
˜ e1 1
τlr−1+1
=Ar−2,1
1 τlr−1+1
= τlr−2+1 τlr−2+2
.
Similarly, we have
˜ e2e˜1 1
τlr−1+1
=Ar−3,2
τlr−2+1
τlr−2+2
= τlr−3+2
τlr−3+3
, Ar−3,2 = τlr−3+2τlr−2+2
τlr−3+3τlr−2+1
,
˜ e3e˜2˜e1
1
τlr−1+1 =Ar−4,3
τlr−3+2
τlr−3+3 = τlr−4+3
τlr−4+4, Ar−4,3 = τlr−4+3τlr−3+3 τlr−4+4τlr−3+2. Applying ˜ei repeatedly, we obtain
˜
ek· · ·e˜2˜e1 1
τlr−1+1 =Ar−1−k,k˜ek−1· · ·˜e2˜e1 1
τlr−1+1 = τlr−1−k+k
τlr−1−k+k+1, k= 1, . . . , r, where, Ar−1−k,k = τ τlr−1−k+kτlr−k+k
lr−1−k+k+1τlr−k+k−1, τl−1+r = r, τl−1+r+1 := 1. For i ∈ I, we have ϕi(τ 1
lr−1+1) = 0. Hence ˜fi(τ 1
lr−1+1) = 0.
Example 3.9. For a giveni∈I and Y = Q
s∈Z
τlζs,i
s+i, we defineνY(n) := P
s≤n
ζs,i. Forj∈Z>0, we set
Y = 1
τlq
1+iτlq
2+i· · ·τlqj+i, 0≤q1 < q2 <· · ·< qj ≤r−1.
First, let us calculatenei (3.4). We obtain wt(Y) =−jΛi and νY(n) = 0 for n <0,
νY(0) =νY(1) =· · ·=νY(q1−1) = 0, νY(q1) =νY(q1+ 1) =· · ·=νY(q2−1) =−1, νY(q2) =νY(q2+ 1) =· · ·=νY(q3−1) =−2, νY(q3) =· · ·=νY(q4−1) =−3, . . . . Thus, we get ϕi(Y) = max{νY(n)|n∈Z}= 0 and
nei = max{n|νY(n) = 0}=q1−1.
Next, since wt(Y)(hi) =−j, we have εi(Y) =ϕi(Y)−wt(Y)(hi) =j >0. Therefore,
˜
eiY =Aq1−1,iY = τlq
1−1+i
τlq
1−1+i+1τlq
1+i−1τlq
2+i· · ·τlqj+i, Aq1−1,i = τlq
1−1+iτlq
1+i
τlq
1−1+i+1τlq
1+i−1
.
Similarly, for k= 1,2, . . . , j, we get
˜
ekiY =Aqk−1,i· · ·Aq2−1,iAq1−1,iY =
k
Y
s=1
τlqk−1+i τlqk−1+i+1τlqk+i−1
! 1
τlqk+1+i· · ·τlqj+i.
3.2 Demazure crystal
For w ∈ W, let us define an upper Demazure crystal Bw+(λ). This is a subset of the crystal B+(λ) defined as follows.
Definition 3.10. Let uλ be the highest weight vector of B+(λ). For the identity element e of W, we setB+e(λ) :={uλ}. Forw∈W, ifsiw < w,
B+w(λ) :=f˜ikb|k≥0, b∈Bs+iw(λ), ˜eib= 0 \ {0}.
Similarly, we define alower Demazure crystal Bw−(λ) as follows.
Definition 3.11. Let vλ be the lowest weight vector of B−(λ). We set Be−(λ) := {vλ}. For w∈W, ifsiw < w,
B−w(λ) :=
˜
ekib|k≥0, b∈Bs−iw(λ), f˜ib= 0 \ {0}.
Theorem 3.12 ([7]). For w∈W, let w=si1· · ·sin be an arbitrary reduced expression. Let uλ (resp.vλ0) be the highest (resp. lowest) weight vector of B+(λ) (resp. B−(λ0)). Then
B+w(λ) =f˜ia(1)1 · · ·f˜ia(n)n uλ|a(1), . . . , a(n)∈Z≥0 \ {0}, B−w(λ0) =
˜
ea(1)i1 · · ·˜ea(n)in vλ0|a(1), . . . , a(n)∈Z≥0 \ {0}.
LetP+ be the set of dominant weights. We setP−:=−P+.
Definition 3.13. Let Y(p) be the monomial realization of crystal associated with p = (pi,j).
Suppose that Y ∈ Y(p) be a highest (resp. lowest) monomial with a weight λ∈ P±. Thus, Y is included in µY(Bw±(λ)). Let us define the Demazure polynomial Dw±[λ, Y;C] associated with a monomialY,w∈W and coefficientsC= (c(b))b∈B±
w(λ) (c(b)∈Z>0), D±w[λ, Y;C] := X
b∈B±w(λ)
c(b)µY(b).
Remark 3.14. In this paper, we only treat the case that the coefficients c(b) are equal to 1 for allb∈B−w(λ) (see Theorem5.6). But when G6= SLr+1(C) (for example,G= Sp2r(C)), we need to treat the case c(b) is not necessary equal to 1 for some b∈Bw−(λ). Therefore, we need non-trivial coefficients c(b)∈Z>0 in Definition3.13.
4 Cluster algebras and generalized minors
In this section, we shall review the notions of cluster algebras. For all definitions in this section, see, e.g., [1,4].
We set [1, l] :={1,2, . . . , l}and [−1,−l] :={−1,−2, . . . ,−l}forl∈Z>0. Forn, m∈Z>0, let x1, . . . , xn, xn+1, . . . , xn+m be variables and P be a free multiplicative abelian group generated by xn+1, . . . , xn+m. We set ZP := Z[x±1n+1, . . . , x±1n+m]. Let K := {hg|g, h ∈ZP, h6= 0} be the field of fractions of ZP, and F :=K(x1, . . . , xn) be the field of rational functions.
4.1 Cluster algebras of geometric type
Definition 4.1. We set n-tuple of variables x = (x1, . . . , xn). Let ˜B = (bij)1≤i≤n,1≤j≤n+m
be n×(n+m) integer matrix whose principal part B := (bij)1≤i,j≤n is sign skew symmetric.
Then a pair Σ = (x,B˜) is called aseed,x a cluster andx1, . . . , xn cluster variables. For a seed Σ = (x,B˜), principal partB of ˜B is called theexchange matrix.
Definition 4.2. For a seed Σ = (x,B˜ = (bij)), an adjacent cluster in direction k ∈ [1, n] is defined by
xk= (x\ {xk})∪ {x0k},
where x0k is the new cluster variable defined by the exchange relation xkx0k= Y
1≤i≤n+m, bki>0
xbiki + Y
1≤i≤n+m, bki<0
x−bi ki.
Definition 4.3. Let A= (aij), A0 = (a0ij) be two matrices of the same size. We say thatA0 is obtained from A by the matrix mutation in direction k, and denoteA0=µk(A) if
a0ij =
−aij if i=k or j=k,
aij+|aik|akj+aik|akj|
2 otherwise.
ForA,A0, if there exists a finite sequence (k1, . . . , ks),ki∈[1, n], such thatA0 =µk1· · ·µks(A), we say Ais mutation equivalent to A0, and denote A∼=A0.
Next proposition can be easily verified by the definition ofµk:
Proposition 4.4([4, Proposition 3.6]). LetAbe a skew symmetrizable matrix. Then any matrix that is mutation equivalent to A is sign skew symmetric.
For a seed Σ = (x,B˜), we say that the seed Σ0= (x0,B˜0) is adjacent to Σ ifx0 is adjacent tox in directionkand ˜B0=µk( ˜B). Two seeds Σ and Σ0 are mutation equivalent if one of them can be obtained from another seed by a sequence of pairwise adjacent seeds and we denote Σ∼Σ0.
Now let us define a cluster algebra of geometric type.
Definition 4.5. Let ˜B be a skew symmetrizable matrix, and Σ = (x,B) a seed.˜ We set A := Z[xn+1, . . . , xn+m]. The cluster algebra (of geometric type) A =A(Σ) over A associated with seed Σ is defined as the A-subalgebra of F generated by all cluster variables in all seeds which are mutation equivalent to Σ.
For a seed Σ, we defineZP-subalgebraU(Σ) of F by U(Σ) :=ZP
x±1
∩ZP x±11
∩ · · · ∩ZP x±1n
. Here,ZP[x±1] is the Laurent polynomial ring inx.
Definition 4.6. Let Σ0 = (x,B) be a seed such that ˜˜ B is skew symmetrizable. We define an upper cluster algebra A = A(Σ0) as the intersection of the subalgebras U(Σ) for all seeds Σ∼Σ0.
Following the inclusion relation holds [1]:
A(Σ)⊂ A(Σ).
4.2 Cluster algebras on double Bruhat cells of type A
As in Section 2, let G= SLr+1(C) be the simple algebraic group of type Ar and W =Sr+1 be its Weyl group. We set u∈W and its reduced word ias in (2.1) and (2.2):
u=s1s2· · ·sr
| {z }
1stcycle
s1· · ·sr−1
| {z }
2ndcycle
· · ·s1· · ·sr−m+2
| {z }
(m−1)thcycle
s1· · ·sin
| {z }
mth cycle
, (4.1)
i= (1, . . . , r
| {z }
1stcycle
,1, . . . ,(r−1)
| {z }
2ndcycle
, . . . ,1, . . . ,(r−m+ 2)
| {z }
(m−1)thcycle
,1, . . . , in
| {z }
mth cycle
). (4.2)
We shall constitute the upper cluster algebra A(i) from i. Let ik, k ∈ [1, l(u)], be the kth index of i from the left.
At first, we define a sete(i) as
e(i) := [−1,−r]∪ {k|there exist some l > ksuch that ik =il}.
Next, let us define a matrix ˜B= ˜B(i).
Definition 4.7. Let ˜B(i) be an integer matrix with rows labeled by all the indices in [−1,−r]∪ [1, l(u)] and columns labeled by all the indices in e(i). For k∈ [−1,−r]∪[1, l(u)] and l∈e(i), an entry bkl of ˜B(i) is determined as follows
bkl=
−sgn((k−l)·ip) if p=q,
−sgn((k−l)·ip·a|ik||il|) if p < q and sgn(ip·iq)(k−l)(k+−l+)>0,
0 otherwise.
Proposition 4.8 ([1, Proposition 2.6]). The matrix B(i)˜ is skew symmetrizable.
By Proposition 4.4, Definition 4.6 and Proposition 4.8, we can construct the upper cluster algebra from ˜B(i):
Definition 4.9. We denote this upper cluster algebra byA(i).
Now, we set ¯A(i)C:= ¯A(i)⊗CandFC:=F ⊗C. It is known that the coordinate ringC[Gu,e] of the double Bruhat cell is isomorphic to ¯A(i)C (Theorem4.11). To describe this isomorphism explicitly, we need generalized minors. Fork ∈ [1, l(u)], let ik be the kth index of i (4.2) from the left, and we suppose that it belongs to the m0th cycle. We set
u≤k =u≤k(i) :=s1s2· · ·sr
| {z }
1stcycle
s1· · ·sr−1
| {z }
2ndcycle
· · ·s1· · ·sik
| {z }
m0thcycle
. (4.3)
Fork∈[−1,−r], we setu≤k:=eandik:=k. In the caseG= SLr+1(C), the generalized minors are nothing but the ordinary minors of a matrix:
Definition 4.10 ([1]). For x ∈ G = SLr+1(C) and k ∈ [−1,−r]∪[1, l(u)], we define the generalized minor ∆(k;i)(x) as the minor of x whose rows (resp. columns) are labeled by the elements of the set u≤k([1,|ik|]) (resp. [1,|ik|]).
Finally, we set
F(i) :={∆(k;i)|k∈[−1,−r]∪[1, l(u)]}.
It is known that the set F(i) is an algebraically independent generating set for the field of rational functions C(Gu,e) [3, Theorem 1.12]. Then, we have the following theorem.
Theorem 4.11 ([1, Theorem 2.10]). The isomorphism of fields ϕ :FC → C(Gu,e) defined by ϕ(xk) = ∆(k;i),k∈[−1,−r]∪[1, l(u)], restricts to an isomorphism of algebrasA(i)¯ C→C[Gu,e].
5 Generalized minors and crystals
In the rest of the paper, we consider the case G = SLr+1(C), and let u ∈ W and its reduced word ias in (4.1) and (4.2):
u=s1s2· · ·srs1· · ·sr−1· · ·s1· · ·sr−m+2s1· · ·sin, (5.1) i= (1, . . . , r
| {z }
1stcycle
,1, . . . ,(r−1)
| {z }
2ndcycle
, . . . ,1, . . . ,(r−m+ 2)
| {z }
(m−1)thcycle
,1, . . . , in
| {z }
mth cycle
), (5.2)
that is,i is the left factor of (1,2,3, . . . , r,1,2,3, . . . ,(r−1), . . . ,1,2,1). Letik be thekth index of i from the left, and belong to m0th cycle. As we shall show in Lemma 5.4, we may assume in=ik.
By Theorem 4.11, we can regard C[Gu,e] as an upper cluster algebra and {∆(k;i)} as its cluster variables belonging to a given initial seed. Each ∆(k;i) is a regular function onGu,e. On the other hand, by Proposition 2.4 (resp. Theorem 2.3), we can consider ∆(k;i) as a function on H×(C×)l(u) (resp. (C×)l(u)). Then we change the variables of {∆(k;i)} as follows:
Definition 5.1. Fora∈H and t,τ ∈(C×)l(u) we set
∆G(k;i)(a,t) := ∆(k;i)◦x¯Gi (a,t),
∆L(k;i)(τ) := ∆(k;i)◦xLi (τ), where t= (t1, . . . , tl(u)), τ = (τ1, . . . , τl(u)).
We will describe the function ∆L(k;i)(τ) by using monomial realizations of Demazure crys- tals.
5.1 Generalized minor ∆G(k; i)(a,t)
In this subsection, we shall prove that ∆G(k;i)(a,t) is immediately obtained from ∆L(k;i):
Proposition 5.2. We set d:=ik. For a= diag(a1, . . . , ar+1)∈H,
∆G(k;i)(a,t) =am0+1· · ·am0+d∆L(k;i)(t).
This proposition follows from the following lemma:
Lemma 5.3. In the above setting, ∆G(k;i)(a,t) (resp.∆L(k;i)(τ)) is given as a minor whose row are labeled by the set {m0+ 1, . . . , m0+d} and column are labeled by the set {1, . . . , d} of the matrix
a x−1(t1)x−2(t2)· · ·x−r(tr)
| {z }
1stcycle
x−1(tl1+1)· · ·x−(r−1)(tl1+r−1)
| {z }
2ndcycle
· · ·
×x−1(tlm−1+1)· · ·x−in(tlm−1+in)
| {z }
mth cycle
,
resp.
x−1(τ1)x−2(τ2)· · ·x−r(τr)
| {z }
1stcycle
x−1(τl1+1)· · ·x−(r−1)(τl1+r−1)
| {z }
2ndcycle
· · ·
×x−1(τlm−1+1)· · ·x−in(τlm−1+in)
| {z }
mthcycle
. (5.3)