3 Monomial realizations of crystal bases

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[G^u,v] of the double Bruhat cellG^u,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[G^u,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[G^u,v] of double Bruhat cell G^u,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 G^u,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= (p_i,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 p_i,j = 1 if i < j and p_i,j = 0 if i > j, we have the following crystal graph of the crystal B(Λ₃), where Λ₃ is the 3^rdfundamental 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 {τ_j}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 G^u,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 =hs_i|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 onL^u,v :=N uN ∩B−vB− instead ofG^u,v since their difference is, indeed, only the factor from the torus part. We call L^u,v reduced double Bruhat cell. For the above u,v, there exists a birational mapx^L_i : (C^×)¹⁰−→^∼ L^u,v given by

x^L_i(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) =i^th











 . ..

t⁻¹ 0

1 t

. ..











 .

Therefore, by (1.2) and (1.3) we find

∆^L(6;i)(τ) := ∆(6;i)◦x^L_i

(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 onL^u,v ⊂G^u,v and the monomial realizations of crystals.

In this paper, we shall treat the caseG= SL_r+1(C),v=eand some specialu∈W =S_r+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 formG^u,e:=BuB∩B₋andL^u,e:=N uN∩B₋, whereB(resp.B−) is the subgroup of upper (resp. lower) triangular matrices inG= SL_r+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 B_u≤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 G^u,v,L^u,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= SL_r+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 groupS_r+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 G^u,v as follows G^u,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 cellL^u,v as follows L^u,v :=N uN ∩B−vB−⊂G^u,v.

As is similar to the caseG^u,v,L^u,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 =s_i1· · ·s_in be a reduced expression ofu ∈W,i₁, . . . , i_n ∈[1, r]. Then the finite sequence

i:= (i₁, . . . , i_n)

is called reduced word iforu.

In this paper, we treat (reduced) double Bruhat cells of the form G^u,e := BuB∩B− and L^u,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≤i_n≤r−m+ 1. Let i be a reduced word ofu:

i= (1, . . . , r

| {z }

1^stcycle

,1, . . . ,(r−1)

| {z }

2^ndcycle

, . . . ,1, . . . ,(r−m+ 2)

| {z }

(m−1)^thcycle

,1, . . . , i_n

| {z }

m^th 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 G^u,e and H×(C^×)^l(u), and between L^u,e and (C^×)^l(u). As in the previous section, we consider the case G:= SL_r+1(C). We set g:= Lie(G) with the Cartan decomposition g=n₋⊕h⊕n. Lete_i,f_i (i ∈ [1, r]) be the generators of n, n₋. For i ∈ [1, r] and t ∈ C, we set x_i(t) := exp(te_i), 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 expressx_i(t), y_i(t), as the following matrices

x_i(t) =i^th











 . ..

1 t 0 1

. ..













, y_i(t) =i^th











 . ..

1 0 t 1

. ..













. (2.3)

For a reduced word i= (i1, i2, . . . , in), we define a map x^G_i :H×Cⁿ→Gas x^G_i (a;t₁, . . . t_n) :=a·y_i1(t₁)y_i2(t₂)· · ·y_in(t_n).

Theorem 2.2 ([3, Theorem 1.2]). We setu∈W and its reduced word i as in (2.1) and (2.2).

The map x^G_i defined above can be restricted to a biregular isomorphism between H×(C^×)^l(u) and a Zariski open subset of G^u,e.

Next, fori∈[1, r] andt∈C^×, we define as follows α^∨_i(t) :=ϕ_i

t 0 0 t⁻¹

, x−i(t) :=y_i(t)α^∨_i(t⁻¹) =ϕ_i

t⁻¹ 0

1 t

. We can expressx−i(t) and α^∨_i(t) as the following matrices

x−i(t) =i^th











 . ..

t⁻¹ 0

1 t

. ..













, α^∨_i(t) = diag(1, . . . ,1,

i

∨

t,

i+1

∨

t⁻¹,1, . . . ,1). (2.4)

Fori= (i₁, . . . , i_n) (i₁, . . . , i_n∈[1, r]), we define a mapx^L_i :Cⁿ→Gas x^L_i(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 x^L_i defined above can be restricted to a biregular isomorphism between (C^×)^l(u) and a Zariski open subset of L^u,e.

Finally, we define a map ¯x^G_i :H×(C^×)ⁿ→G^u,v as

¯

x^G_i (a;t₁, . . . , t_n) =ax^L_i (t₁, . . . , t_n).

Proposition 2.4. In the above setting, the map x¯^G_i is a biregular isomorphism between H × (C^×)ⁿ and a Zariski open subset ofG^u,e.

Proof . We set l₀ := 0,l₁ :=r,l₂ :=r+ (r−1), . . . , l_m:=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(t_r)⁻¹

| {z }

1^stcycle

· · ·α^∨₁(t_lm−1+1)⁻¹· · ·α^∨_in(t_lm−1+in)⁻¹

| {z }

m^thcycle

,

τ_ls+j(t) = (t_ls+1+j−1t_ls+2+j−1· · ·t_lm−1+j−1)(t_ls+j+1t_ls+1+j+1· · ·t_lm−1+j+1)

t_ls+j(t_ls+1+j· · ·t_lm−1+j)² , (2.5) where in (2.5), if i does not include j (resp. j+ 1, j−1) in ζ^th cycle then we set t_lζ−1+j = 1 (resp. t_lζ−1+j+1= 1, t_lζ−1+j−1= 1). This is a biregular isomorphism.

Let us prove

¯

x^G_i (a;t₁, . . . , t_n) = x^G_i ◦φ

(a;t₁, . . . , t_n),

which implies that ¯x^G_i :H×(C^×)ⁿ→G^u,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)⁻¹y_j(t) =







y_i(c²t)α^∨_i(c)⁻¹ if i=j, yj(c⁻¹t)α^∨_i(c)⁻¹ if |i−j|= 1, y_j(t)α^∨_i(c)⁻¹ otherwise,

(2.6)

for 1≤i,j≤r and c, t∈C^×. On the other hand, we obtain

(x^G_i ◦φ)(a;t₁, . . . , t_n) =a×α^∨₁(t₁)⁻¹· · ·α^∨_r(t_r)⁻¹· · ·α^∨₁(t_lm−1+1)⁻¹· · ·α_i^∨_n(t_lm−1+in)⁻¹

×y₁(τ₁(t))y₂(τ₂(t))· · ·y_r(τ_r(t))· · ·y₁(τ_lm−1+1(t))· · ·y_in(τ_lm−1+in(t)). (2.7) For each sand j, let us move α^∨_j(t_ls+j)⁻¹, α^∨_j+1(t_ls+j+1)⁻¹, . . . , α^∨_in(t_lm−1+in)⁻¹ to the right of y_j(τ_ls+j(t)) by using the relations (2.6). For example,

α^∨_j(t_ls+j)⁻¹· · ·α^∨_j−1(t_lm−1+j−1)⁻¹α^∨_j(t_lm−1+j)⁻¹

×α^∨_j+1(t_lm−1+j+1)⁻¹· · ·α^∨_in(t_lm−1+in)⁻¹y_j(τ_ls+j(t))

=α^∨_j(t_ls+j)⁻¹· · ·y_j t²_lm−1+j

t_lm−1+j−1t_lm−1+j+1τ_ls+j(t)

!

α^∨_j−1(t_lm−1+j−1)⁻¹

×α^∨_j(t_lm−1+j)⁻¹α^∨_j+1(t_lm−1+j+1)⁻¹· · ·α^∨_in(t_lm−1+in)⁻¹. Repeating this argument, we have

=y_j

(t_ls+j· · ·t_lm−1+j)²

(t_ls+j−1· · ·t_lm−1+j−1)(t_ls+j+1· · ·t_lm−1+j+1)τ_ls+j(t)

α^∨_j(t_ls+j)⁻¹· · ·α^∨_in(t_lm−1+in)⁻¹. Note that ^{(tls+j...tlm−1+j)}

2

(t_ls+j−1···t_lm−1+j−1)(t_ls+j+1···t_lm−1+j+1)τ_ls+j(t) =t_ls+j. By (2.7), we have x^G_i ◦φ

(a;t₁, . . . , t_n) =a·y₁(t₁)α^∨₁(t₁)⁻¹· · ·y_r(t_r)α^∨_r(t_r)⁻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) = ¯x^G_i (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 ={α_i|i∈I} (resp. Π^∨ ={h_i|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) =hh_i, wt(b)i, (ii) wt( ˜eib) = wt(b) +αi, if ˜eib∈B, (iii) wt( ˜f_ib) = wt(b)−α_i, if ˜f_ib∈B,

(iv) ε_i( ˜e_ib) =ε_i(b)−1,ϕ_i( ˜e_ib) =ϕ_i(b) + 1 if ˜e_ib∈B, (v) ε_i( ˜f_ib) =ε_i(b) + 1,ϕ_i( ˜f_ib) =ϕ_i(b)−1 if ˜f_ib∈B, (vi) ˜f_ib=b⁰ ⇔b= ˜e_ib⁰, ifb,b⁰∈B,

(vii) ϕ_i(b) =−∞,b∈B,⇒e˜_ib= ˜f_ib= 0.

LetU_q(g) be the universal enveloping algebra associated with the Cartan matrixA in (3.1), and g = sl_r+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=MΛ_d,M ∈Z>0, we have

B⁺(MΛd) =B⁻(−MΛ_r−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= (p_j,i)j,i∈I, j6=i such that p_j,i =

(1 if j < i, 0 if i < j.

Second, for doubly-indexed variables {Y_s,i|i∈I,s∈Z}, we define the set of monomials Y :=







Y = Y

s∈Z, i∈I

Y_s,i^ζs,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

Y_s,i^ζs,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

A_s,i:=Y_s,iY_s+1,iY

j6=i

Y_s+p^aj,i

j,i,j =



















Ys,1Ys+1,1

Y_s,2 if i= 1, Y_s,iY_s+1,i

Ys,i+1Ys+1,i−1

if 2≤i≤r−1, Y_s,rY_s+1,r

Ys+1,r−1

if i=r,

(3.3)

and define the Kashiwara operators as follows f˜iY =

(A⁻¹_n

fi,iY if ϕ_i(Y)>0,

0 if ϕ_i(Y) = 0, ˜eiY =

(An_ei,iY if εi(Y)>0, 0 if εi(Y) = 0, where

n_fi := min





 n

ϕi(Y) =X

k≤n

ζ_k,i







, nei := max





 n

ϕi(Y) =X

k≤n

ζ_k,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 thenµ_Y =µ_Y⁰.

Remark 3.4. The actions of ˜e_i and ˜f_i on Y are determined by wt(Y)(h_i), ϕ_i(Y) and ε_i(Y), which are determined by the factors Y_s,i^±1, s ∈ Z. Thus, when we consider the actions of ˜e_i and ˜f_i, we need to see the factors{Y_s,i^±1}_s∈Z only.

Example 3.5. For λ = βΛd (resp. λ = −βΛ_d), β ∈ Z>0, d ∈ I, we can embed B⁺(λ) (resp. B⁻(λ)) inY as a crystal by

v_λ 7→Y_β+γ,dYβ−1+γ,d· · ·Y_1+γ,d,

resp.v_λ 7→ 1

Y_β+γ,dYβ−1+γ,d· · ·Y_1+γ,d

,

where v_λ is the highest (resp. lowest) weight vector of B⁺(λ) (resp. B⁻(λ)), and γ is an ar- bitrary integer. For Y⁺ := Y_β+γ,dYβ−1+γ,d· · ·Y_1+γ,d (resp. Y⁻ := _Y ¹

β+γ,dYβ−1+γ,d···Y_1+γ,d), µ_Y⁺ (resp. µ_Y⁻) denotes the embedding in Definition3.3. Then Y⁺ (resp. Y⁻) is the highest (resp.

lowest) weight vector in µ_Y⁺(B⁺(λ)) (resp. µ_Y⁻(B⁻(λ))).

We set l₀ := 0, l₁ := r, l₂ := r+ (r−1), . . . , l_s := 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 τ_l0+1 . . . τ_l0+r−1 τ_l0+r τ_l1+1 . . . τ_l1+r−1 τ_l2+1 τ_l2+2 . . .

Remark 3.6. In the above setting, the variables {Y_s,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 variablesτ_ls+0,τ_ls+r+1 (0≤s≤m−1) we understand

τ_ls+0 =τ_ls+r+1= 1.

For example, ifi= 1 then τ_ls+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+1τ_l^a_r−2+p^2,1

2,1+2 =

τ_lr−2+1τ_lr−1+1

τ_lr−2+2 , we get

˜ e₁ 1

τlr−1+1

=Ar−2,1

1 τlr−1+1

= τ_lr−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

τ_lr−1+1 =Ar−4,3

τ_lr−3+2

τ_lr−3+3 = τ_lr−4+3

τ_lr−4+4, Ar−4,3 = τ_lr−4+3τ_lr−3+3 τ_lr−4+4τ_lr−3+2. Applying ˜e_i repeatedly, we obtain

˜

e_k· · ·e˜₂˜e₁ 1

τ_lr−1+1 =Ar−1−k,k˜ek−1· · ·˜e₂˜e₁ 1

τ_lr−1+1 = τlr−1−k+k

τ_{lr−1−k+k+1}, k= 1, . . . , r, where, Ar−1−k,k = _τ ^{τlr−1−k+kτlr−k+k}

lr−1−k+k+1τ_lr−k+k−1, τ_l−1+r = r, τ_l−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

τ_l^ζs,i

s+i, we defineν_Y(n) := P

s≤n

ζs,i. Forj∈Z>0, we set

Y = 1

τ_lq

1+iτ_lq

2+i· · ·τ_lqj+i, 0≤q1 < q2 <· · ·< qj ≤r−1.

First, let us calculatenei (3.4). We obtain wt(Y) =−jΛ_i and νY(n) = 0 for n <0,

νY(0) =νY(1) =· · ·=νY(q1−1) = 0, νY(q1) =νY(q1+ 1) =· · ·=νY(q2−1) =−1, ν_Y(q₂) =ν_Y(q₂+ 1) =· · ·=ν_Y(q₃−1) =−2, ν_Y(q₃) =· · ·=ν_Y(q₄−1) =−3, . . . . Thus, we get ϕ_i(Y) = max{ν_Y(n)|n∈Z}= 0 and

n_ei = max{n|ν_Y(n) = 0}=q₁−1.

Next, since wt(Y)(h_i) =−j, we have ε_i(Y) =ϕ_i(Y)−wt(Y)(h_i) =j >0. Therefore,

˜

eiY =Aq1−1,iY = τ_lq

1−1+i

τ_lq

1−1+i+1τ_lq

1+i−1τ_lq

2+i· · ·τ_lqj+i, Aq1−1,i = τ_lq

1−1+iτ_lq

1+i

τ_lq

1−1+i+1τ_lq

1+i−1

.

Similarly, for k= 1,2, . . . , j, we get

˜

e^k_iY =Aq_k−1,i· · ·Aq2−1,iAq1−1,iY =

k

Y

s=1

τ_lqk−1+i τ_lqk−1+i+1τ_lqk+i−1

! 1

τ_lqk+1+i· · ·τ_lqj+i.

3.2 Demazure crystal

For w ∈ W, let us define an upper Demazure crystal B_w⁺(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) :={u_λ}. Forw∈W, ifsiw < w,

B⁺_w(λ) :=f˜_i^kb|k≥0, b∈B_s⁺_iw(λ), ˜e_ib= 0 \ {0}.

Similarly, we define alower Demazure crystal B_w⁻(λ) as follows.

Definition 3.11. Let v_λ be the lowest weight vector of B⁻(λ). We set B_e⁻(λ) := {v_λ}. For w∈W, ifs_iw < w,

B⁻_w(λ) :=

˜

e^k_ib|k≥0, b∈B_s⁻_iw(λ), f˜ib= 0 \ {0}.

Theorem 3.12 ([7]). For w∈W, let w=s_i1· · ·s_in be an arbitrary reduced expression. Let u_λ (resp.v_λ⁰) be the highest (resp. lowest) weight vector of B⁺(λ) (resp. B⁻(λ⁰)). Then

B⁺_w(λ) =f˜_i^a(1)₁ · · ·f˜_i^a(n)_n uλ|a(1), . . . , a(n)∈Z≥0 \ {0}, B⁻_w(λ⁰) =

˜

e^a(1)_i1 · · ·˜e^a(n)_in v_λ⁰|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 µ_Y(B_w^±(λ)). Let us define the Demazure polynomial D_w^±[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= SL_r+1(C) (for example,G= Sp_2r(C)), we need to treat the case c(b) is not necessary equal to 1 for some b∈B_w⁻(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 x_n+1, . . . , x_n+m. We set ZP := Z[x^±1_n+1, . . . , x^±1_n+m]. Let K := {_h^g|g, h ∈ZP, h6= 0} be the field of fractions of ZP, and F :=K(x₁, . . . , x_n) 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 := (b_ij)1≤i,j≤n is sign skew symmetric.

Then a pair Σ = (x,B˜) is called aseed,x a cluster andx₁, . . . , x_n 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\ {x_k})∪ {x⁰_k},

where x⁰_k is the new cluster variable defined by the exchange relation xkx⁰_k= Y

1≤i≤n+m, b_ki>0

x^b_i^ki + Y

1≤i≤n+m, b_ki<0

x^−b_i ^ki.

Definition 4.3. Let A= (aij), A⁰ = (a⁰_ij) be two matrices of the same size. We say thatA⁰ is obtained from A by the matrix mutation in direction k, and denoteA⁰=µ_k(A) if

a⁰_ij =







−a_ij if i=k or j=k,

aij+|a_ik|a_kj+a_ik|a_kj|

2 otherwise.

ForA,A⁰, if there exists a finite sequence (k1, . . . , ks),ki∈[1, n], such thatA⁰ =µ_k1· · ·µ_ks(A), we say Ais mutation equivalent to A⁰, and denote A∼=A⁰.

Next proposition can be easily verified by the definition ofµ_k:

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[x_n+1, . . . , x_n+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^±1_n

. 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 =S_r+1 be its Weyl group. We set u∈W and its reduced word ias in (2.1) and (2.2):

u=s1s2· · ·sr

| {z }

1^stcycle

s1· · ·sr−1

| {z }

2^ndcycle

· · ·s1· · ·sr−m+2

| {z }

(m−1)^thcycle

s1· · ·sin

| {z }

m^th cycle

, (4.1)

i= (1, . . . , r

| {z }

1^stcycle

,1, . . . ,(r−1)

| {z }

2^ndcycle

, . . . ,1, . . . ,(r−m+ 2)

| {z }

(m−1)^thcycle

,1, . . . , in

| {z }

m^th cycle

). (4.2)

We shall constitute the upper cluster algebra A(i) from i. Let i_k, k ∈ [1, l(u)], be the k^th 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 i_k =i_l}.

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 b_kl of ˜B(i) is determined as follows

b_kl=







−sgn((k−l)·i_p) 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)⊗CandF_C:=F ⊗C. It is known that the coordinate ringC[G^u,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 i_k be the k^th index of i (4.2) from the left, and we suppose that it belongs to the m^0th cycle. We set

u≤k =u≤k(i) :=s₁s₂· · ·s_r

| {z }

1^stcycle

s₁· · ·sr−1

| {z }

2^ndcycle

· · ·s₁· · ·s_ik

| {z }

m^0thcycle

. (4.3)

Fork∈[−1,−r], we setu≤k:=eandi_k:=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,|i_k|]) (resp. [1,|i_k|]).

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(G^u,e) [3, Theorem 1.12]. Then, we have the following theorem.

Theorem 4.11 ([1, Theorem 2.10]). The isomorphism of fields ϕ :F_C → C(G^u,e) defined by ϕ(x_k) = ∆(k;i),k∈[−1,−r]∪[1, l(u)], restricts to an isomorphism of algebrasA(i)¯ _C→C[G^u,e].

5 Generalized minors and crystals

In the rest of the paper, we consider the case G = SL_r+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 }

1^stcycle

,1, . . . ,(r−1)

| {z }

2^ndcycle

, . . . ,1, . . . ,(r−m+ 2)

| {z }

(m−1)^thcycle

,1, . . . , i_n

| {z }

m^th 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 thek^th index of i from the left, and belong to m^0th cycle. As we shall show in Lemma 5.4, we may assume i_n=i_k.

By Theorem 4.11, we can regard C[G^u,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 onG^u,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¯^G_i (a,t),

∆^L(k;i)(τ) := ∆(k;i)◦x^L_i (τ), where t= (t1, . . . , t_l(u)), τ = (τ1, . . . , τ_l(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(t_r)

| {z }

1^stcycle

x−1(t_l1+1)· · ·x_−(r−1)(t_l1+r−1)

| {z }

2^ndcycle

· · ·

×x−1(t_lm−1+1)· · ·x−in(t_lm−1+in)

| {z }

m^th cycle

,

resp.

x−1(τ1)x−2(τ2)· · ·x−r(τr)

| {z }

1^stcycle

x−1(τl1+1)· · ·x−(r−1)(τl1+r−1)

| {z }

2^ndcycle

· · ·

×x−1(τ_lm−1+1)· · ·x−in(τ_lm−1+in)

| {z }

m^thcycle

. (5.3)