@#+#+#+#+analysis code? Need no prose. Contents skipped per instruction. Page number skipped. Body empty? Must output @body.
Introduction
W W W -algebras are now central objects in modern mathematics and mathematical physics. Since their introduction by Zamolodchikov [38 ] , they have found important applications to integrable partial differential equations and Drinfeld–Sokolov hierarchies [16 ] , integrable field theories [20 ] , and two-dimensional conformal field theory [15 ] . Later developments revealed that W W W -algebras also form an essential part of the correspondence between four-dimensional supersymmetric gauge theories and two-dimensional conformal field theories [1 ] .
One of the principal ways to construct W W W -algebras is Hamiltonian reduction, known in the affine setting as Drinfeld–Sokolov reduction [16 , 21 ] . Quantum Hamiltonian reduction also produces finite W W W -algebras associated with nilpotent elements and compatible good gradings. For g l N \mathfrak{gl}_N gl N , the relevant data can be encoded by pyramids [17 , 5 ] . In this work we consider the finite W W W -algebras for g l N \mathfrak{gl}_N gl N associated with a rectangular pyramid.
Let N = n p N=np N = n p . The algebra g = g l N \mathfrak{g}=\mathfrak{gl}_N g = gl N then has a natural n × n n \times n n × n block structure with blocks of size p × p p \times p p × p . We denote its matrix units by E a , b i , j E_{a,b}^{i,j} E a , b i , j , where a , b ∈ { 1 , … , n } a,b \in\{1,\ldots,n\} a , b ∈ { 1 , … , n } label the blocks and i , j ∈ { 1 , … , p } i,j \in\{1,\ldots,p\} i , j ∈ { 1 , … , p } label the entries inside a block. The subalgebra used in the reduction is
m n = span { E a , b i , j ∣ a > b } . \mathfrak{m}_n=\operatorname{span}\{E_{a,b}^{i,j}\mid a>b\}. m n = span { E a , b i , j ∣ a > b } .
and the character is ψ ( E a , b i , j ) = δ i , j δ a , b + 1 \psi(E_{a,b}^{i,j})=\delta_{i,j}\delta_{a,b+1} ψ ( E a , b i , j ) = δ i , j δ a , b + 1 . We introduce the left ideal
I ψ = U ( g l N ) ( m n − ψ ) . I_\psi=U(\mathfrak{gl}_N)(\mathfrak{m}_n-\psi). I ψ = U ( gl N ) ( m n − ψ ) .
and define the rectangular finite W W W -algebra by
W n ( g l p ) = ( U ( g l N ) / I ψ ) m n . \mathcal{W}_n(\mathfrak{gl}_p)=\left(U(\mathfrak{gl}_N)/I_\psi\right)^{\mathfrak{m}_n}. W n ( gl p ) = ( U ( gl N ) / I ψ ) m n .
A deep connection between these finite W W W -algebras and Yangians was established in [33 , 5 ] . In the rectangular case, Brundan and Kleshchev identify the finite W W W -algebra with a truncated Yangian.
Theorem 1.1 (Brundan–Kleshchev [5 ] ). The finite W W W -algebra W n ( g l p ) \mathcal{W}_n(\mathfrak{gl}_p) W n ( gl p ) is isomorphic to the level-n n n truncated Yangian Y ( n ) ( g l p ) Y^{(n)}(\mathfrak{gl}_p) Y ( n ) ( gl p ) .
Heuristically, one may view this result as passing from a one-site spin-chain object of rank n p np n p to an n n n -site generating matrix with rank-p p p symmetry by Hamiltonian reduction. This picture is only heuristic, but it helps explain why the Yangian coproduct and the factorized form of the generating matrix arise naturally.
Quasideterminants play a central role in the explicit construction of the generating matrix. De Sole, Kac, and Valeri [10 ] used them to describe the isomorphism with the truncated Yangian, building on the general theory developed in [27 ] . Unlike an ordinary determinant, a quasideterminant retains the order of multiplication of its noncommuting entries, making it well suited to constructions in noncommutative algebras.
More precisely, the corner quasideterminant of the L L L -operator, a matrix with entries in U ( g l N ) [ u ] U(\mathfrak{gl}_N)[u] U ( gl N ) [ u ] , is invariant under the action of the subalgebra used in the reduction and satisfies the RTT relation. Its image under reduction is the Brundan–Kleshchev generating matrix, so both the RTT relation and the generating property pass to the finite W W W -algebra. Thus the quasideterminant constructs the reduction invariant while carrying its Yangian structure through the reduction.
Another important tool in representation theory and mathematical physics is the differential realization of a Lie algebra on the big cell of its associated flag variety [29 , 3 ] :
ρ λ : U ( g ) ⟶ D ( N − ) , (1) \rho_\lambda: \mathcal{U}(\mathfrak{g}) \longrightarrow\mathcal{D}(N_-), \tag*{(1)} ρ λ : U ( g ) ⟶ D ( N − ) , (1)
where N − ⊂ G L N N_- \subset\mathrm{GL}_N N − ⊂ GL N is the lower unipotent subgroup, D ( N − ) \mathcal{D}(N_-) D ( N − ) is its algebra of polynomial differential operators, and λ ∈ h ∗ \lambda\in\mathfrak{h}^* λ ∈ h ∗ is a weight parameter. These realizations give explicit differential-operator models of representations and provide a natural setting in which to study reduction. The affine analogues of these realizations lead to Wakimoto realizations of affine Lie algebras [37 , 9 ] .
Together, these constructions lead to the central question of this paper:
What is the Yangian structure induced by quantum Hamiltonian reduction of the big-cell differential realization
A related but separate question is how modules and intertwiners behave under reduction. We plan to address it in the continuation of this work.
We answer the first question by proving that the Yangian action induced by Hamiltonian reduction of the big-cell differential realization of g l N \mathfrak{gl}_N gl N coincides with the coproduct action on a tensor product of n n n rank-p p p differential realizations. The key step is an explicit factorization of the reduced quasideterminant. This establishes compatibility between two constructions that arise from different perspectives: Hamiltonian reduction of the higher-rank differential realization and the coproduct realization of the Yangian on a tensor product of lower-rank differential realizations. The quasideterminant streamlines the calculations and makes transparent the identities and invariance properties needed for the finite W W W -algebra generators.
To state the result precisely, let N − ′ ⊂ G L p N'_- \subset\mathrm{GL}_p N − ′ ⊂ GL p be the lower unipotent subgroup. The reduced differential algebra is naturally identified with
ξ : D − → ∼ D ( N − ′ ) ⊗ n , (2) \xi: \mathcal{D}_- \xrightarrow{\sim} \mathcal{D}(N'_-)^{\otimes n}, \tag*{(2)} ξ : D − ∼ D ( N − ′ ) ⊗ n , (2)
where D − \mathcal{D}_- D − denotes the differential algebra obtained by Hamiltonian reduction of D ( N − ) \mathcal{D}(N_-) D ( N − ) . Via the Brundan–Kleshchev identification, this reduction endows D ( N − ′ ) ⊗ n \mathcal{D}(N'_-)^{\otimes n} D ( N − ′ ) ⊗ n with a Y ( g l p ) Y(\mathfrak{gl}_p) Y ( gl p ) -module structure. The generating matrix for this action is the Brundan–Kleshchev matrix evaluated in the reduced differential realization.
There is also a direct construction of a Yangian action on D ( N − ′ ) ⊗ n \mathcal{D}(N'_-)^{\otimes n} D ( N − ′ ) ⊗ n , using the coproduct together with evaluation homomorphisms. Write
λ = diag ( λ 1 , … , λ n ) , \lambda= \operatorname{diag}(\lambda_1,\ldots,\lambda_n), λ = diag ( λ 1 , … , λ n ) ,
where each λ a \lambda_a λ a is a diagonal p × p p \times p p × p matrix. These parameters define differential realizations
ρ λ a : U ( g l p ) ⟶ D ( N − ′ ) , \rho_{\lambda_a} : \mathcal{U}(\mathfrak{gl}_p) \longrightarrow\mathcal{D}(N'_-), ρ λ a : U ( gl p ) ⟶ D ( N − ′ ) ,
one in each tensor factor. Let E [ a ] E_{[a]} E [ a ] denote the generator matrix in the a a a th factor after applying the corresponding realization; we suppress the homomorphisms in operator formulas. In polynomial normalization, the associated generating matrix is
T ( u ) = ( u + E [ n ] ) ⋯ ( u + E [ 1 ] ) . T(u) = (u + E_{[n]}) \cdots(u + E_{[1]}). T ( u ) = ( u + E [ n ] ) ⋯ ( u + E [ 1 ] ) .
Theorem 1.2. After the identification by ξ \xi ξ , the Hamiltonian reduction of (1) and the Yangian action induced by the coproduct and the tensor product of differential realizations define the same Y ( g l p ) Y(\mathfrak{gl}_p) Y ( gl p ) -module structure on D ( N − ′ ) ⊗ n \mathcal{D}(N'_{-})^{\otimes n} D ( N − ′ ) ⊗ n .
The proof uses the invariance of the corner quasideterminant under certain block multiplications. These multiplications remove the off-diagonal coordinate factors and leave a matrix whose entries are already invariant under the reduction. The reduction can therefore be performed on these entries inside the quasideterminant, yielding the ordered product asserted in the theorem.
The factorization also motivates the representation-theoretic continuation of this work, in which we will study reduced intertwiners. Let W W W and W ′ W' W ′ be Whittaker modules obtained from suitable specializations of the differential realization. Write Wh ( W ) \operatorname{Wh}(W) Wh ( W ) for the subspace of Whittaker vectors in W W W , namely, vectors v v v satisfying
E a , b i , j v = ψ ( E a , b i , j ) v . E^{i,j}_{a,b}v=\psi(E^{i,j}_{a,b})v. E a , b i , j v = ψ ( E a , b i , j ) v .
Skryabin’s equivalence [36 ] gives an isomorphism between the corresponding intertwining spaces:
Hom g l N ( W , W ′ ) ≃ Hom W n ( g l p ) ( Wh ( W ) , Wh ( W ′ ) ) . \operatorname{Hom}_{\mathfrak{gl}_N}(W,W')\simeq\operatorname{Hom}_{W_n(\mathfrak{gl}_p)}\bigl(\operatorname{Wh}(W),\operatorname{Wh}(W')\bigr). Hom gl N ( W , W ′ ) ≃ Hom W n ( gl p ) ( Wh ( W ) , Wh ( W ′ ) ) .
The factorization established here gives an explicit functional model for the Whittaker subspaces in terms of functions on n n n copies of the lower-rank group N − ′ N'_{-} N − ′ . Upon reduction, natural intertwiners of the unreduced differential realizations induce intertwiners between these Whittaker spaces. These operators are closely related to the factorized R R R -matrices and Baxter Q Q Q -operators studied by Derkachov and Manashov [13 ] . We plan to develop this connection in the second part of the work.
The paper is organized as follows. In § 2 we introduce the Yangian and the truncated Yangian, recall the coproduct and evaluation homomorphisms, and review the tensor-product realization. In § 3 we introduce the corner quasideterminant, relate it to a block of the inverse matrix, and prove the RTT relation and commutation properties that make it an invariant for reduction. In § 4 we recall rectangular finite W W W -algebras, identify the reduced quasideterminant with the Brundan–Kleshchev generating matrix, and describe the Miura homomorphism.
In § 5 we review the differential realization on the big cell, identify its reduced differential algebra with D ( N − ′ ) ⊗ n \mathcal{D}(N'_{-})^{\otimes n} D ( N − ′ ) ⊗ n , and compute the reduction of the quasideterminant to obtain the factorization in the main theorem. Appendix § 8 collects auxiliary calculations. Its first subsection gives a direct proof of the commutation relations, and the second provides detailed index calculations used in § 3 . Appendix § 9 proves the block proposition and explains the reduction in a concrete functional realization.
Preliminaries on the RTT Yangian
Let e i , j e_{i,j} e i , j , 1 ≤ i , j ≤ N 1\leq i,j\leq N 1 ≤ i , j ≤ N , be the standard matrix units in Mat N ( C ) \operatorname{Mat}_N(\mathbb{C}) Mat N ( C ) , and let E i , j E_{i,j} E i , j be the standard generators of the Lie algebra g l N \mathfrak{gl}_N gl N , with commutation relations
[ E i , j , E k , q ] = δ j , k E i , q − δ q , i E k , j . (3) [E_{i,j},E_{k,q}]=\delta_{j,k}E_{i,q}-\delta_{q,i}E_{k,j}. \tag*{(3)} [ E i , j , E k , q ] = δ j , k E i , q − δ q , i E k , j . (3)
Let P = ∑ i , j = 1 N e i , j ⊗ e j , i P=\sum_{i,j=1}^{N}e_{i,j}\otimes e_{j,i} P = ∑ i , j = 1 N e i , j ⊗ e j , i be the permutation operator on C N ⊗ C N \mathbb{C}^{N}\otimes\mathbb{C}^{N} C N ⊗ C N . The rational R R R -matrix
R ( u ) = u + P ∈ End ( C N ⊗ C N ) [ u ] (4) R(u)=u+P\in\operatorname{End}(\mathbb{C}^{N}\otimes\mathbb{C}^{N})[u] \tag*{(4)} R ( u ) = u + P ∈ End ( C N ⊗ C N ) [ u ] (4)
satisfies the Yang–Baxter equation
R 12 ( u ) R 13 ( u + v ) R 23 ( v ) = R 23 ( v ) R 13 ( u + v ) R 12 ( u ) (5) R_{12}(u)R_{13}(u+v)R_{23}(v)=R_{23}(v)R_{13}(u+v)R_{12}(u) \tag*{(5)} R 12 ( u ) R 13 ( u + v ) R 23 ( v ) = R 23 ( v ) R 13 ( u + v ) R 12 ( u ) (5)
where R i , j R_{i,j} R i , j indicates the two tensor factors on which R R R acts nontrivially.
Definition 2.1. The RTT Yangian Y ( g l N ) Y(\mathfrak{gl}_N) Y ( gl N ) is the unital associative algebra generated by
t i , j ( r ) , 1 ≤ i , j ≤ N , r ≥ 1 , t_{i,j}^{(r)},\qquad1\le i,j\le N,\quad r\ge1, t i , j ( r ) , 1 ≤ i , j ≤ N , r ≥ 1 , whose generating series
t i , j ( u ) = δ i , j + ∑ r ≥ 1 t i , j ( r ) u − r , T ( u ) = ∑ i , j = 1 N e i , j ⊗ t i , j ( u ) , (6) t_{i,j}(u)=\delta_{i,j}+\sum_{r\ge1}t_{i,j}^{(r)}u^{-r},\qquad T(u)=\sum_{i,j=1}^{N}e_{i,j}\otimes t_{i,j}(u), \tag*{(6)} t i , j ( u ) = δ i , j + r ≥ 1 ∑ t i , j ( r ) u − r , T ( u ) = i , j = 1 ∑ N e i , j ⊗ t i , j ( u ) , (6) satisfy the relations
( u − v ) t i , j ( u ) t k , q ( v ) + t k , j ( u ) t i , q ( v ) = ( u − v ) t k , q ( v ) t i , j ( u ) + t k , j ( v ) t i , q ( u ) (7) (u-v)t_{i,j}(u)t_{k,q}(v)+t_{k,j}(u)t_{i,q}(v)=(u-v)t_{k,q}(v)t_{i,j}(u)+t_{k,j}(v)t_{i,q}(u) \tag*{(7)} ( u − v ) t i , j ( u ) t k , q ( v ) + t k , j ( u ) t i , q ( v ) = ( u − v ) t k , q ( v ) t i , j ( u ) + t k , j ( v ) t i , q ( u ) (7) where T 1 ( u ) = T ( u ) ⊗ 1 N T_1(u)=T(u)\otimes1_N T 1 ( u ) = T ( u ) ⊗ 1 N and T 2 ( v ) = 1 N ⊗ T ( v ) T_2(v)=1_N\otimes T(v) T 2 ( v ) = 1 N ⊗ T ( v ) are N 2 × N 2 N^2\times N^2 N 2 × N 2 matrices with entries in Y ( g l N ) [ [ u − 1 , v − 1 ] ] Y(\mathfrak{gl}_N)[[u^{-1},v^{-1}]] Y ( gl N ) [[ u − 1 , v − 1 ]] .
We next recall the coproduct of the Yangian.
Proposition 2.2. There is an algebra homomorphism
Δ : Y ( g l N ) ⟶ Y ( g l N ) ⊗ Y ( g l N ) \Delta:Y(\mathfrak{gl}_N)\longrightarrow Y(\mathfrak{gl}_N)\otimes Y(\mathfrak{gl}_N) Δ : Y ( gl N ) ⟶ Y ( gl N ) ⊗ Y ( gl N ) determined by
Δ ( T ( u ) ) = T [ 1 ] ( u ) T [ 2 ] ( u ) , (8) \Delta(T(u))=T_{[1]}(u)T_{[2]}(u), \tag*{(8)} Δ ( T ( u )) = T [ 1 ] ( u ) T [ 2 ] ( u ) , (8) where T [ 1 ] ( u ) = T ( u ) ⊗ 1 T_{[1]}(u)=T(u)\otimes1 T [ 1 ] ( u ) = T ( u ) ⊗ 1 and T [ 2 ] ( u ) = 1 ⊗ T ( u ) T_{[2]}(u)=1\otimes T(u) T [ 2 ] ( u ) = 1 ⊗ T ( u ) are elements of Mat N ( C ) ⊗ Y ( g l N ) ⊗ 2 [ [ u − 1 ] ] \operatorname{Mat}_N(\mathbb{C})\otimes Y(\mathfrak{gl}_N)^{\otimes2}[[u^{-1}]] Mat N ( C ) ⊗ Y ( gl N ) ⊗ 2 [[ u − 1 ]] . Equivalently,
Δ ( t i , j ( u ) ) = ∑ k = 1 N t i , k ( u ) ⊗ t k , j ( u ) . \Delta(t_{i,j}(u))=\sum_{k=1}^{N}t_{i,k}(u)\otimes t_{k,j}(u). Δ ( t i , j ( u )) = k = 1 ∑ N t i , k ( u ) ⊗ t k , j ( u ) . The map Δ \Delta Δ is coassociative:
( Δ ⊗ 1 ) ∘ Δ = ( 1 ⊗ Δ ) ∘ Δ . (\Delta\otimes1)\circ\Delta=(1\otimes\Delta)\circ\Delta. ( Δ ⊗ 1 ) ∘ Δ = ( 1 ⊗ Δ ) ∘ Δ.
Multiplication of the monodromy matrix by a scalar power of u u u preserves the RTT relation. We denote by Y m ( g l N ) Y_m(\mathfrak{gl}_N) Y m ( gl N ) the corresponding polynomial normalization, in which T ( u ) = u m + O ( u m − 1 ) T(u)=u^m+O(u^{m-1}) T ( u ) = u m + O ( u m − 1 ) . The map
T ( u ) ⟼ u m T ( u ) (9) T(u)\longmapsto u^mT(u) \tag*{(9)} T ( u ) ⟼ u m T ( u ) (9)
identifies Y 0 ( g l N ) Y_0(\mathfrak{gl}_N) Y 0 ( gl N ) with Y m ( g l N ) Y_m(\mathfrak{gl}_N) Y m ( gl N ) . The coproduct is compatible with this normalization and defines
Δ m 1 , m 2 : Y m 1 + m 2 ( g l N ) ⟶ Y m 1 ( g l N ) ⊗ Y m 2 ( g l N ) . \Delta_{m_1,m_2}:Y_{m_1+m_2}(\mathfrak{gl}_N)\longrightarrow Y_{m_1}(\mathfrak{gl}_N)\otimes Y_{m_2}(\mathfrak{gl}_N). Δ m 1 , m 2 : Y m 1 + m 2 ( gl N ) ⟶ Y m 1 ( gl N ) ⊗ Y m 2 ( gl N ) .
Proposition 2.3. For every c ∈ C c \in\mathbb{C} c ∈ C , the assignment
ev c ( T ( u ) ) = 1 + 1 u − c ∑ i , j = 1 N e i , j ⊗ E j , i \operatorname{ev}_{c}(T(u))=1+\frac{1}{u-c}\sum_{i,j=1}^{N}e_{i,j}\otimes E_{j,i} ev c ( T ( u )) = 1 + u − c 1 i , j = 1 ∑ N e i , j ⊗ E j , i defines an evaluation homomorphism
ev c : Y ( g l N ) ⟶ U ( g l N ) . \operatorname{ev}_{c}:Y(\mathfrak{gl}_{N})\longrightarrow U(\mathfrak{gl}_{N}). ev c : Y ( gl N ) ⟶ U ( gl N ) . In the polynomial normalization, the same homomorphism is written as
ev c : Y 1 ( g l N ) ⟶ U ( g l N ) , ev c ( T ( u ) ) = L ( u − c ) , L ( u ) = u + ∑ i , j = 1 N e i , j ⊗ E j , i . \operatorname{ev}_{c}:Y_{1}(\mathfrak{gl}_{N})\longrightarrow U(\mathfrak{gl}_{N}),\qquad\operatorname{ev}_{c}(T(u))=L(u-c),\qquad L(u)=u+\sum_{i,j=1}^{N}e_{i,j}\otimes E_{j,i}. ev c : Y 1 ( gl N ) ⟶ U ( gl N ) , ev c ( T ( u )) = L ( u − c ) , L ( u ) = u + i , j = 1 ∑ N e i , j ⊗ E j , i . Thus every g l N \mathfrak{gl}_{N} gl N -module M M M becomes an evaluation Y 1 ( g l N ) Y_{1}(\mathfrak{gl}_{N}) Y 1 ( gl N ) -module, denoted by M ( c ) M(c) M ( c ) . The coproduct equips
M 1 ( z 1 ) ⊗ ⋯ ⊗ M m ( z m ) M_{1}(z_{1})\otimes\cdots\otimes M_{m}(z_{m}) M 1 ( z 1 ) ⊗ ⋯ ⊗ M m ( z m ) with a Y m ( g l N ) Y_{m}(\mathfrak{gl}_{N}) Y m ( gl N ) -module structure. Its monodromy matrix is
( ev z 1 ⊗ ⋯ ⊗ ev z m ) ∘ Δ 1 , … , 1 ( T ( u ) ) = L [ 1 ] ( u − z 1 ) ⋯ L [ m ] ( u − z m ) , (10) (\operatorname{ev}_{z_{1}}\otimes\cdots\otimes\operatorname{ev}_{z_{m}})\circ\Delta_{1,\ldots,1}(T(u))=L_{[1]}(u-z_{1})\cdots L_{[m]}(u-z_{m}), \tag*{(10)} ( ev z 1 ⊗ ⋯ ⊗ ev z m ) ∘ Δ 1 , … , 1 ( T ( u )) = L [ 1 ] ( u − z 1 ) ⋯ L [ m ] ( u − z m ) , (10) where Δ 1 , … , 1 \Delta_{1,\ldots,1} Δ 1 , … , 1 is the iterated coproduct and L [ i ] ( u ) L_{[i]}(u) L [ i ] ( u ) is the L L L -operator acting nontrivially only in the i i i -th tensor factor.
Definition 2.4. In Y m ( g l N ) Y_{m}(\mathfrak{gl}_{N}) Y m ( gl N ) , we reindex the coefficients by writing
t i , j ( u ) = δ i , j u m + t i , j ( 1 − m ) u m − 1 + … + t i , j ( 0 ) + ∑ r > 0 t i , j ( r ) u − r . t_{i,j}(u)=\delta_{i,j}u^{m}+t_{i,j}^{(1-m)}u^{m-1}+\ldots+t_{i,j}^{(0)}+\sum_{r>0}t_{i,j}^{(r)}u^{-r}. t i , j ( u ) = δ i , j u m + t i , j ( 1 − m ) u m − 1 + … + t i , j ( 0 ) + r > 0 ∑ t i , j ( r ) u − r . For m ∈ N m\in\mathbb{N} m ∈ N , the level-m m m truncated Yangian is the quotient
Y ( m ) ( g l N ) = Y m ( g l N ) / ( t i , j ( r ) ∣ r > 0 , 1 ≤ i , j ≤ N ) . Y^{(m)}(\mathfrak{gl}_{N})=Y_{m}(\mathfrak{gl}_{N})/(t_{i,j}^{(r)}\mid r>0,\ 1\leq i,j\leq N). Y ( m ) ( gl N ) = Y m ( gl N ) / ( t i , j ( r ) ∣ r > 0 , 1 ≤ i , j ≤ N ) . This quotient is the Yangian of level m m m introduced by Cherednik [6 , 7 ] . The following tensor-product realization is appeared naturally in the framework of the Quantum Inverse Scattering Method [18 , 31 , 35 , 19 , 30 ] .
Proposition 2.5 (Cherednik [6 , 7 ] ; see also [5 ] ). For z 1 = … = z m = 0 z_{1}=\ldots=z_{m}=0 z 1 = … = z m = 0 , the homomorphism (10) descends to an injective homomorphism
Y ( m ) ( g l N ) ↪ U ( g l N ) ⊗ m . Y^{(m)}(\mathfrak{gl}_{N})\hookrightarrow U(\mathfrak{gl}_{N})^{\otimes m}. Y ( m ) ( gl N ) ↪ U ( gl N ) ⊗ m . Its image is generated by the matrix coefficients of the polynomial
T ( u ) = L [ 1 ] ( u ) L [ 2 ] ( u ) ⋯ L [ m ] ( u ) . (11) T(u)=L_{[1]}(u)L_{[2]}(u)\cdots L_{[m]}(u). \tag*{(11)} T ( u ) = L [ 1 ] ( u ) L [ 2 ] ( u ) ⋯ L [ m ] ( u ) . (11)
Quasideterminants and commutation relations
In this section, we collect the quasideterminant identities used below and prove the two properties of the corner quasideterminant that are essential for Hamiltonian reduction: its RTT relation and its commutativity with the blocks belonging to the reduction subalgebra. We keep the block calculations explicit, since the positions of the blocks are important in both arguments.
Basic notation
Fix positive integers n , p n,p n , p , set N = n p N=np N = n p , and let E r , s E_{r,s} E r , s denote the standard generators of g l N \mathfrak{gl}_N gl N . We use the block notation
E a , b i , j : = E ( a − 1 ) p + i , ( b − 1 ) p + j . E_{a,b}^{i,j}:=E_{(a-1)p+i,\,(b-1)p+j}. E a , b i , j := E ( a − 1 ) p + i , ( b − 1 ) p + j .
The lower indices record the n × n n \times n n × n block position, while the upper indices record the matrix entry inside a p × p p \times p p × p block. The commutation relations (3) can be written as
[ E a , b i , j , E c , d k , r ] = δ b , c δ j , k E a , d i , r − δ d , a δ r , i E c , b k , j . (12) [E_{a,b}^{i,j},E_{c,d}^{k,r}]=\delta_{b,c}\delta_{j,k}E_{a,d}^{i,r}-\delta_{d,a}\delta_{r,i}E_{c,b}^{k,j}. \tag*{(12)} [ E a , b i , j , E c , d k , r ] = δ b , c δ j , k E a , d i , r − δ d , a δ r , i E c , b k , j . (12)
Definition 3.1. Introduce the universal matrix
E = ∑ a , b = 1 n ∑ i , j = 1 p e ( a − 1 ) p + i , ( b − 1 ) p + j ⊗ E b , a j , i ∈ Mat N ( U ( g l N ) ) . (13) E=\sum_{a,b=1}^{n}\sum_{i,j=1}^{p}e_{(a-1)p+i,\,(b-1)p+j}\otimes E_{b,a}^{j,i}\in\operatorname{Mat}_N(U(\mathfrak{gl}_N)). \tag*{(13)} E = a , b = 1 ∑ n i , j = 1 ∑ p e ( a − 1 ) p + i , ( b − 1 ) p + j ⊗ E b , a j , i ∈ Mat N ( U ( gl N )) . (13)
The quantum L L L -operator is
L ( u ) = u + E ∈ Mat N ( U ( g l N ) [ u ] ) . L(u)=u+E\in\operatorname{Mat}_N(U(\mathfrak{gl}_N)[u]). L ( u ) = u + E ∈ Mat N ( U ( gl N ) [ u ]) .
We write the matrix blocks in boldface. For 1 ≤ a , b ≤ n 1\leq a,b\leq n 1 ≤ a , b ≤ n , set
E a , b = ∑ i , j = 1 p e ( a − 1 ) p + i , ( b − 1 ) p + j ⊗ E b , a j , i . (14) \mathbf{E}_{a,b}=\sum_{i,j=1}^{p}e_{(a-1)p+i,\,(b-1)p+j}\otimes E_{b,a}^{j,i}. \tag*{(14)} E a , b = i , j = 1 ∑ p e ( a − 1 ) p + i , ( b − 1 ) p + j ⊗ E b , a j , i . (14)
In n × n n\times n n × n block form,
L ( u ) = ( u + E 1 , 1 E 1 , 2 ⋯ E 1 , n E 2 , 1 u + E 2 , 2 ⋯ E 2 , n ⋮ ⋮ ⋱ ⋮ E n , 1 E n , 2 ⋯ u + E n , n ) , (15) L(u)=\begin{pmatrix}
u+\mathbf{E}_{1,1} & \mathbf{E}_{1,2} & \cdots& \mathbf{E}_{1,n}\\
\mathbf{E}_{2,1} & u+\mathbf{E}_{2,2} & \cdots& \mathbf{E}_{2,n}\\
\vdots& \vdots& \ddots& \vdots\\
\mathbf{E}_{n,1} & \mathbf{E}_{n,2} & \cdots& u+\mathbf{E}_{n,n}
\end{pmatrix}, \tag*{(15)} L ( u ) = u + E 1 , 1 E 2 , 1 ⋮ E n , 1 E 1 , 2 u + E 2 , 2 ⋮ E n , 2 ⋯ ⋯ ⋱ ⋯ E 1 , n E 2 , n ⋮ u + E n , n , (15)
where every block is p × p p\times p p × p and the scalar u u u on each diagonal block denotes scalar multiplication by the identity matrix.
Example. For N = 4 N = 4 N = 4 and n = p = 2 n = p = 2 n = p = 2 , the same L L L -operator can be written successively in ordinary matrix notation, indexed block notation, and block-matrix notation:
L ( u ) = ( u + E 1 , 1 E 2 , 1 E 3 , 1 E 4 , 1 E 1 , 2 u + E 2 , 2 E 3 , 2 E 4 , 2 E 1 , 3 E 2 , 3 u + E 3 , 3 E 4 , 3 E 1 , 4 E 2 , 4 E 3 , 4 u + E 4 , 4 ) = ( u + E 1 , 1 1 , 1 E 1 , 1 2 , 1 E 1 , 1 1 , 1 E 1 , 1 2 , 1 E 1 , 2 1 , 1 u + E 1 , 2 2 , 2 E 1 , 2 1 , 2 E 1 , 2 2 , 2 E 1 , 1 1 , 2 E 1 , 1 2 , 1 u + E 2 , 2 2 , 2 E 2 , 2 2 , 1 E 1 , 2 1 , 2 E 1 , 2 2 , 2 E 2 , 2 1 , 2 u + E 2 , 2 2 , 2 ) = ( u + E 1 , 1 E 1 , 2 E 2 , 1 u + E 2 , 2 ) . L(u) =
\begin{pmatrix}
u + E_{1,1} & E_{2,1} & E_{3,1} & E_{4,1} \\
E_{1,2} & u + E_{2,2} & E_{3,2} & E_{4,2} \\
E_{1,3} & E_{2,3} & u + E_{3,3} & E_{4,3} \\
E_{1,4} & E_{2,4} & E_{3,4} & u + E_{4,4}
\end{pmatrix}
=
\begin{pmatrix}
u + E_{1,1}^{1,1} & E_{1,1}^{2,1} & E_{1,1}^{1,1} & E_{1,1}^{2,1} \\
E_{1,2}^{1,1} & u + E_{1,2}^{2,2} & E_{1,2}^{1,2} & E_{1,2}^{2,2} \\
E_{1,1}^{1,2} & E_{1,1}^{2,1} & u + E_{2,2}^{2,2} & E_{2,2}^{2,1} \\
E_{1,2}^{1,2} & E_{1,2}^{2,2} & E_{2,2}^{1,2} & u + E_{2,2}^{2,2}
\end{pmatrix}
=
\begin{pmatrix}
u + E_{1,1} & E_{1,2} \\
E_{2,1} & u + E_{2,2}
\end{pmatrix}. L ( u ) = u + E 1 , 1 E 1 , 2 E 1 , 3 E 1 , 4 E 2 , 1 u + E 2 , 2 E 2 , 3 E 2 , 4 E 3 , 1 E 3 , 2 u + E 3 , 3 E 3 , 4 E 4 , 1 E 4 , 2 E 4 , 3 u + E 4 , 4 = u + E 1 , 1 1 , 1 E 1 , 2 1 , 1 E 1 , 1 1 , 2 E 1 , 2 1 , 2 E 1 , 1 2 , 1 u + E 1 , 2 2 , 2 E 1 , 1 2 , 1 E 1 , 2 2 , 2 E 1 , 1 1 , 1 E 1 , 2 1 , 2 u + E 2 , 2 2 , 2 E 2 , 2 1 , 2 E 1 , 1 2 , 1 E 1 , 2 2 , 2 E 2 , 2 2 , 1 u + E 2 , 2 2 , 2 = ( u + E 1 , 1 E 2 , 1 E 1 , 2 u + E 2 , 2 ) .
Quasideterminants, the inverse matrix, and the row determinant
We recall the definition of a quasideterminant [25 , 26 , 27 ] and the formulas needed below. Let A A A be a matrix with entries in a possibly noncommutative ring A \mathcal{A} A :
A = ( a 1 , 1 a 1 , 2 ⋯ a 1 , n ⋮ ⋮ ⋱ ⋮ a n − 1 , 1 a n − 1 , 2 ⋯ a n − 1 , n a n , 1 a n , 2 ⋯ a n , n ) . (16) A =
\begin{pmatrix}
a_{1,1} & a_{1,2} & \cdots& a_{1,n} \\
\vdots& \vdots& \ddots& \vdots\\
a_{n-1,1} & a_{n-1,2} & \cdots& a_{n-1,n} \\
a_{n,1} & a_{n,2} & \cdots& a_{n,n}
\end{pmatrix}. \tag*{(16)} A = a 1 , 1 ⋮ a n − 1 , 1 a n , 1 a 1 , 2 ⋮ a n − 1 , 2 a n , 2 ⋯ ⋱ ⋯ ⋯ a 1 , n ⋮ a n − 1 , n a n , n . (16)
Definition 3.2. The corner quasideterminant of A A A with respect to the entry a n , 1 a_{n,1} a n , 1 is
∣ A ∣ n , 1 = a n , 1 − ( a n , 2 ⋯ a n , n ) ( a 1 , 2 ⋯ a 1 , n ⋮ ⋱ ⋮ a n − 1 , 2 ⋯ a n − 1 , n ) − 1 ( a 1 , 1 ⋮ a n − 1 , 1 ) . (17) |A|_{n,1} = a_{n,1} - \begin{pmatrix} a_{n,2} & \cdots& a_{n,n} \end{pmatrix}
\begin{pmatrix}
a_{1,2} & \cdots& a_{1,n} \\
\vdots& \ddots& \vdots\\
a_{n-1,2} & \cdots& a_{n-1,n}
\end{pmatrix}^{-1}
\begin{pmatrix}
a_{1,1} \\
\vdots\\
a_{n-1,1}
\end{pmatrix}. \tag*{(17)} ∣ A ∣ n , 1 = a n , 1 − ( a n , 2 ⋯ a n , n ) a 1 , 2 ⋮ a n − 1 , 2 ⋯ ⋱ ⋯ a 1 , n ⋮ a n − 1 , n − 1 a 1 , 1 ⋮ a n − 1 , 1 . (17)
provided that the displayed ( n − 1 ) × ( n − 1 ) (n - 1) \times(n - 1) ( n − 1 ) × ( n − 1 ) submatrix is invertible in A \mathcal{A} A .
Proposition 3.3 ([27 ] ). Assume that A − 1 A^{-1} A − 1 and ∣ A ∣ n , 1 − 1 |A|_{n,1}^{-1} ∣ A ∣ n , 1 − 1 exist over A \mathcal{A} A . Then
∣ A ∣ n , 1 − 1 = ( A − 1 ) 1 , n . (18) |A|_{n,1}^{-1} = (A^{-1})_{1,n}. \tag*{(18)} ∣ A ∣ n , 1 − 1 = ( A − 1 ) 1 , n . (18)
Example. For n = 2 n = 2 n = 2 , the corner quasideterminant is
A = ( a 1 , 1 a 1 , 2 a 2 , 1 a 2 , 2 ) ; ∣ A ∣ 2 , 1 = a 2 , 1 − a 2 , 2 a 1 , 2 − 1 a 1 , 1 . A =
\begin{pmatrix}
a_{1,1} & a_{1,2} \\
a_{2,1} & a_{2,2}
\end{pmatrix}
\ ;\quad
|A|_{2,1} = a_{2,1} - a_{2,2}a_{1,2}^{-1}a_{1,1}. A = ( a 1 , 1 a 2 , 1 a 1 , 2 a 2 , 2 ) ; ∣ A ∣ 2 , 1 = a 2 , 1 − a 2 , 2 a 1 , 2 − 1 a 1 , 1 .
If, in addition, A A A and all the quasideterminants appearing below are invertible over A \mathcal{A} A , then
A − 1 = ( ( a 1 , 1 − a 1 , 2 a 2 , 2 − 1 a 2 , 1 ) − 1 ( a 2 , 1 − a 2 , 2 a 1 , 2 − 1 a 1 , 1 ) − 1 ( a 1 , 2 − a 1 , 1 a 2 , 1 − 1 a 2 , 2 ) − 1 ( a 2 , 2 − a 2 , 1 a 1 , 1 − 1 a 1 , 2 ) − 1 ) . A^{-1} =
\begin{pmatrix}
(a_{1,1} - a_{1,2}a_{2,2}^{-1}a_{2,1})^{-1} & (a_{2,1} - a_{2,2}a_{1,2}^{-1}a_{1,1})^{-1} \\
(a_{1,2} - a_{1,1}a_{2,1}^{-1}a_{2,2})^{-1} & (a_{2,2} - a_{2,1}a_{1,1}^{-1}a_{1,2})^{-1}
\end{pmatrix}. A − 1 = ( ( a 1 , 1 − a 1 , 2 a 2 , 2 − 1 a 2 , 1 ) − 1 ( a 1 , 2 − a 1 , 1 a 2 , 1 − 1 a 2 , 2 ) − 1 ( a 2 , 1 − a 2 , 2 a 1 , 2 − 1 a 1 , 1 ) − 1 ( a 2 , 2 − a 2 , 1 a 1 , 1 − 1 a 1 , 2 ) − 1 ) .
For the general relation between the entries of the inverse matrix and quasideterminants, see [27 ] .
Proposition 3.4. Suppose that the displayed quasideterminants are defined, that the upper-left ( n − 1 ) × ( n − 1 ) (n-1) \times(n-1) ( n − 1 ) × ( n − 1 ) block of X X X is invertible, and that the lower-right ( n − 1 ) × ( n − 1 ) (n-1) \times(n-1) ( n − 1 ) × ( n − 1 ) block of Y Y Y is invertible. The corner quasideterminant is invariant under left and right multiplication by matrices of the following forms:
∣ A ∣ n , 1 = ∣ X A Y ∣ n , 1 (19) |A|_{n,1}=|XAY|_{n,1} \tag*{(19)} ∣ A ∣ n , 1 = ∣ X A Y ∣ n , 1 (19) where
X = ( x 1 , 1 ⋯ x 1 , n − 1 0 ⋮ ⋱ ⋮ ⋮ x n − 1 , 1 ⋯ x n − 1 , n − 1 0 x n , 1 ⋯ x n , n − 1 1 ) ; Y = ( 1 0 ⋯ 0 y 2 , 1 y 2 , 2 ⋯ y 2 , n ⋮ ⋮ ⋱ ⋮ y n , 1 y n , 2 ⋯ y n , n ) . (20) X=\begin{pmatrix}
x_{1,1} & \cdots& x_{1,n-1} & 0\\
\vdots& \ddots& \vdots& \vdots\\
x_{n-1,1} & \cdots& x_{n-1,n-1} & 0\\
x_{n,1} & \cdots& x_{n,n-1} & 1
\end{pmatrix};
\quad
Y=\begin{pmatrix}
1 & 0 & \cdots& 0\\
y_{2,1} & y_{2,2} & \cdots& y_{2,n}\\
\vdots& \vdots& \ddots& \vdots\\
y_{n,1} & y_{n,2} & \cdots& y_{n,n}
\end{pmatrix}. \tag*{(20)} X = x 1 , 1 ⋮ x n − 1 , 1 x n , 1 ⋯ ⋱ ⋯ ⋯ x 1 , n − 1 ⋮ x n − 1 , n − 1 x n , n − 1 0 ⋮ 0 1 ; Y = 1 y 2 , 1 ⋮ y n , 1 0 y 2 , 2 ⋮ y n , 2 ⋯ ⋯ ⋱ ⋯ 0 y 2 , n ⋮ y n , n . (20)
Proof. We write out the calculation for n = 3 n=3 n = 3 : the same block cancellation proves the statement for arbitrary n n n . For left multiplication, the product X A XA X A has the block form
X A = ( x 1 , 1 x 1 , 2 x 2 , 1 x 2 , 2 x 3 , 1 x 3 , 2 ) ( 0 0 1 ) ( a 1 , 1 a 1 , 2 a 1 , 3 a 2 , 1 a 2 , 2 a 2 , 3 a 3 , 1 a 3 , 2 a 3 , 3 ) = ( ( x 1 , 1 x 1 , 2 x 2 , 1 x 2 , 2 ) ( a 1 , 1 a 2 , 1 ) ( x 1 , 1 x 1 , 2 x 2 , 1 x 2 , 2 ) ( a 1 , 2 a 1 , 3 a 2 , 2 a 2 , 3 ) a 3 , 1 + ( x 3 , 1 x 3 , 2 ) ( a 1 , 1 a 2 , 1 ) ( a 3 , 2 a 3 , 3 ) + ( x 3 , 1 x 3 , 2 ) ( a 1 , 2 a 1 , 3 a 2 , 2 a 2 , 3 ) ) . \begin{aligned}
XA=\begin{pmatrix}
x_{1,1} & x_{1,2}\\
x_{2,1} & x_{2,2}\\
x_{3,1} & x_{3,2}
\end{pmatrix}
\begin{pmatrix}
0\\
0\\
1
\end{pmatrix}
\begin{pmatrix}
a_{1,1} & a_{1,2} & a_{1,3}\\
a_{2,1} & a_{2,2} & a_{2,3}\\
a_{3,1} & a_{3,2} & a_{3,3}
\end{pmatrix} \\
&=\begin{pmatrix}
\begin{pmatrix}x_{1,1} & x_{1,2}\\ x_{2,1} & x_{2,2}\end{pmatrix}
\begin{pmatrix}a_{1,1}\\ a_{2,1}\end{pmatrix}
&
\begin{pmatrix}x_{1,1} & x_{1,2}\\ x_{2,1} & x_{2,2}\end{pmatrix}
\begin{pmatrix}a_{1,2} & a_{1,3}\\ a_{2,2} & a_{2,3}\end{pmatrix}\\
a_{3,1}+(x_{3,1}\;\;x_{3,2})\begin{pmatrix}a_{1,1}\\ a_{2,1}\end{pmatrix}
&
(a_{3,2}\;\;a_{3,3})+(x_{3,1}\;\;x_{3,2})
\begin{pmatrix}a_{1,2} & a_{1,3}\\ a_{2,2} & a_{2,3}\end{pmatrix}
\end{pmatrix}.
\end{aligned} X A = x 1 , 1 x 2 , 1 x 3 , 1 x 1 , 2 x 2 , 2 x 3 , 2 0 0 1 a 1 , 1 a 2 , 1 a 3 , 1 a 1 , 2 a 2 , 2 a 3 , 2 a 1 , 3 a 2 , 3 a 3 , 3 = ( x 1 , 1 x 2 , 1 x 1 , 2 x 2 , 2 ) ( a 1 , 1 a 2 , 1 ) a 3 , 1 + ( x 3 , 1 x 3 , 2 ) ( a 1 , 1 a 2 , 1 ) ( x 1 , 1 x 2 , 1 x 1 , 2 x 2 , 2 ) ( a 1 , 2 a 2 , 2 a 1 , 3 a 2 , 3 ) ( a 3 , 2 a 3 , 3 ) + ( x 3 , 1 x 3 , 2 ) ( a 1 , 2 a 2 , 2 a 1 , 3 a 2 , 3 ) .
In the defining formula, the upper-left block of X X X cancels between the inverse of the deleted submatrix and the first-column vector. Substitution into the definition of the corner quasideterminant therefore gives
∣ X A ∣ 3 , 1 = a 3 , 1 + ( x 3 , 1 x 3 , 2 ) ( a 1 , 1 a 2 , 1 ) |XA|_{3,1}=a_{3,1}+(x_{3,1}\;\;x_{3,2})
\begin{pmatrix}a_{1,1}\\ a_{2,1}\end{pmatrix} ∣ X A ∣ 3 , 1 = a 3 , 1 + ( x 3 , 1 x 3 , 2 ) ( a 1 , 1 a 2 , 1 )
− ( ( a 3 , 2 a 3 , 3 ) + ( x 3 , 1 x 3 , 2 ) ( a 1 , 2 a 1 , 3 a 2 , 2 a 2 , 3 ) ) ( a 1 , 2 a 1 , 3 a 2 , 2 a 2 , 3 ) − 1 ( a 1 , 1 a 2 , 1 ) = ∣ A ∣ 3 , 1 . \qquad-\left((a_{3,2}\;\;a_{3,3})+(x_{3,1}\;\;x_{3,2})
\begin{pmatrix}a_{1,2} & a_{1,3}\\ a_{2,2} & a_{2,3}\end{pmatrix}\right)
\begin{pmatrix}a_{1,2} & a_{1,3}\\ a_{2,2} & a_{2,3}\end{pmatrix}^{-1}
\begin{pmatrix}a_{1,1}\\ a_{2,1}\end{pmatrix}
=|A|_{3,1}. − ( ( a 3 , 2 a 3 , 3 ) + ( x 3 , 1 x 3 , 2 ) ( a 1 , 2 a 2 , 2 a 1 , 3 a 2 , 3 ) ) ( a 1 , 2 a 2 , 2 a 1 , 3 a 2 , 3 ) − 1 ( a 1 , 1 a 2 , 1 ) = ∣ A ∣ 3 , 1 .
Similarly, the product A Y AY A Y has the form
A Y = ( a 1 , 1 a 2 , 1 a 3 , 1 ) ( a 1 , 2 a 1 , 3 a 2 , 2 a 2 , 3 a 3 , 2 a 3 , 3 ) ( 1 ( 0 0 ) ( y 2 , 1 y 3 , 1 ) ( y 2 , 2 y 2 , 3 y 3 , 2 y 3 , 3 ) ) = ( ( a 1 , 1 a 2 , 1 ) + ( a 1 , 2 a 1 , 3 ) ( y 2 , 1 y 3 , 1 ) ( a 1 , 2 a 1 , 3 ) ( y 2 , 2 y 2 , 3 y 3 , 2 y 3 , 3 ) a 3 , 1 + ( a 3 , 2 a 3 , 3 ) ( y 2 , 1 y 3 , 1 ) ( a 3 , 2 a 3 , 3 ) ( y 2 , 2 y 2 , 3 y 3 , 2 y 3 , 3 ) ) . \begin{aligned}
AY=
\begin{pmatrix}
a_{1,1}\\
a_{2,1}\\
a_{3,1}
\end{pmatrix}
\begin{pmatrix}
a_{1,2} & a_{1,3}\\
a_{2,2} & a_{2,3}\\
a_{3,2} & a_{3,3}
\end{pmatrix}
\begin{pmatrix}
1 & \begin{pmatrix}0 & 0\end{pmatrix}\\
\begin{pmatrix}y_{2,1}\\ y_{3,1}\end{pmatrix} & \begin{pmatrix}y_{2,2} & y_{2,3}\\ y_{3,2} & y_{3,3}\end{pmatrix}
\end{pmatrix} \\
&=\begin{pmatrix}
\begin{pmatrix}a_{1,1}\\ a_{2,1}\end{pmatrix}
+\begin{pmatrix}a_{1,2} & a_{1,3}\end{pmatrix}
\begin{pmatrix}y_{2,1}\\ y_{3,1}\end{pmatrix}
&
\begin{pmatrix}a_{1,2} & a_{1,3}\end{pmatrix}
\begin{pmatrix}y_{2,2} & y_{2,3}\\ y_{3,2} & y_{3,3}\end{pmatrix}\\
a_{3,1}+(a_{3,2}\;\;a_{3,3})
\begin{pmatrix}y_{2,1}\\ y_{3,1}\end{pmatrix}
&
(a_{3,2}\;\;a_{3,3})
\begin{pmatrix}y_{2,2} & y_{2,3}\\ y_{3,2} & y_{3,3}\end{pmatrix}
\end{pmatrix}.
\end{aligned} A Y = a 1 , 1 a 2 , 1 a 3 , 1 a 1 , 2 a 2 , 2 a 3 , 2 a 1 , 3 a 2 , 3 a 3 , 3 1 ( y 2 , 1 y 3 , 1 ) ( 0 0 ) ( y 2 , 2 y 3 , 2 y 2 , 3 y 3 , 3 ) = ( a 1 , 1 a 2 , 1 ) + ( a 1 , 2 a 1 , 3 ) ( y 2 , 1 y 3 , 1 ) a 3 , 1 + ( a 3 , 2 a 3 , 3 ) ( y 2 , 1 y 3 , 1 ) ( a 1 , 2 a 1 , 3 ) ( y 2 , 2 y 3 , 2 y 2 , 3 y 3 , 3 ) ( a 3 , 2 a 3 , 3 ) ( y 2 , 2 y 3 , 2 y 2 , 3 y 3 , 3 ) .
Here the lower-right block of Y Y Y cancels between the last-row vector and the inverse of the deleted submatrix. Thus substitution into the defining formula gives
∣ A Y ∣ 3 , 1 = a 3 , 1 + ( a 3 , 2 a 3 , 3 ) ( y 2 , 1 y 3 , 1 ) − ( a 3 , 2 a 3 , 3 ) ( a 1 , 2 a 1 , 3 a 2 , 2 a 2 , 3 ) − 1 ( ( a 1 , 1 a 2 , 1 ) + ( a 1 , 2 a 1 , 3 a 2 , 2 a 2 , 3 ) ( y 2 , 1 y 3 , 1 ) ) = ∣ A ∣ 3 , 1 . \lvert AY\rvert_{3,1}=a_{3,1}+(a_{3,2}\quad a_{3,3})\begin{pmatrix}y_{2,1}\\y_{3,1}\end{pmatrix}-(a_{3,2}\quad a_{3,3})\begin{pmatrix}a_{1,2}&a_{1,3}\\a_{2,2}&a_{2,3}\end{pmatrix}^{-1}\left(\begin{pmatrix}a_{1,1}\\a_{2,1}\end{pmatrix}+\begin{pmatrix}a_{1,2}&a_{1,3}\\a_{2,2}&a_{2,3}\end{pmatrix}\begin{pmatrix}y_{2,1}\\y_{3,1}\end{pmatrix}\right)=\lvert A\rvert_{3,1}. ∣ A Y ∣ 3 , 1 = a 3 , 1 + ( a 3 , 2 a 3 , 3 ) ( y 2 , 1 y 3 , 1 ) − ( a 3 , 2 a 3 , 3 ) ( a 1 , 2 a 2 , 2 a 1 , 3 a 2 , 3 ) − 1 ( ( a 1 , 1 a 2 , 1 ) + ( a 1 , 2 a 2 , 2 a 1 , 3 a 2 , 3 ) ( y 2 , 1 y 3 , 1 ) ) = ∣ A ∣ 3 , 1 .
We next record explicit expressions for quasideterminants of two special classes of matrices.
Definition 3.5. For a matrix A = ( a a , b ) a , b = 1 n A=(a_{a,b})_{a,b=1}^{n} A = ( a a , b ) a , b = 1 n with entries in a unital associative algebra, define
rdet A = ∑ σ ∈ S n sgn ( σ ) a n , σ ( n ) ⋯ a 1 , σ ( 1 ) . (21) \operatorname{rdet} A=\sum_{\sigma\in S_n}\operatorname{sgn}(\sigma)a_{n,\sigma(n)}\cdots a_{1,\sigma(1)}. \tag*{(21)} rdet A = σ ∈ S n ∑ sgn ( σ ) a n , σ ( n ) ⋯ a 1 , σ ( 1 ) . (21)
For the following special matrices, the corner quasideterminant can be computed explicitly.
Proposition 3.6.
For a block matrix whose first upper diagonal consists of identity blocks and whose blocks above the first upper diagonal are zero, one has
∣ ( a 1 , 1 1 0 ⋯ 0 a 2 , 1 a 2 , 2 1 ⋯ 0 a 3 , 1 a 3 , 2 a 3 , 3 ⋯ 0 ⋮ ⋮ ⋮ ⋱ ⋮ a n , 1 a n , 2 a n , 3 ⋯ a n , n ) ∣ n , 1 = ( − 1 ) 1 − n rdet ( a 1 , 1 1 0 ⋯ 0 a 2 , 1 a 2 , 2 1 ⋯ 0 a 3 , 1 a 3 , 2 a 3 , 3 ⋯ 0 ⋮ ⋮ ⋮ ⋱ ⋮ a n , 1 a n , 2 a n , 3 ⋯ a n , n ) . (22) \left\lvert\begin{pmatrix}
a_{1,1}&1&0&\cdots&0\\
a_{2,1}&a_{2,2}&1&\cdots&0\\
a_{3,1}&a_{3,2}&a_{3,3}&\cdots&0\\
\vdots&\vdots&\vdots&\ddots&\vdots\\
a_{n,1}&a_{n,2}&a_{n,3}&\cdots&a_{n,n}
\end{pmatrix}\right\rvert_{n,1}=(-1)^{1-n}\operatorname{rdet}\begin{pmatrix}
a_{1,1}&1&0&\cdots&0\\
a_{2,1}&a_{2,2}&1&\cdots&0\\
a_{3,1}&a_{3,2}&a_{3,3}&\cdots&0\\
\vdots&\vdots&\vdots&\ddots&\vdots\\
a_{n,1}&a_{n,2}&a_{n,3}&\cdots&a_{n,n}
\end{pmatrix}. \tag*{(22)} a 1 , 1 a 2 , 1 a 3 , 1 ⋮ a n , 1 1 a 2 , 2 a 3 , 2 ⋮ a n , 2 0 1 a 3 , 3 ⋮ a n , 3 ⋯ ⋯ ⋯ ⋱ ⋯ 0 0 0 ⋮ a n , n n , 1 = ( − 1 ) 1 − n rdet a 1 , 1 a 2 , 1 a 3 , 1 ⋮ a n , 1 1 a 2 , 2 a 3 , 2 ⋮ a n , 2 0 1 a 3 , 3 ⋮ a n , 3 ⋯ ⋯ ⋯ ⋱ ⋯ 0 0 0 ⋮ a n , n . (22) If b i ∈ A × b_i\in A^{\times} b i ∈ A × for 1 ≤ i < n 1\le i<n 1 ≤ i < n , then
∣ ( a 1 b 1 0 ⋯ 0 0 a 2 b 2 ⋯ 0 0 0 a 3 ⋱ 0 ⋮ ⋮ ⋱ ⋱ b n − 1 0 0 ⋯ 0 a n ) ∣ n , 1 = ( − 1 ) 1 − n a n b n − 1 − 1 a n − 1 b n − 2 − 1 ⋯ a 2 b 1 − 1 a 1 . (23) \left\lvert\begin{pmatrix}
a_1&b_1&0&\cdots&0\\
0&a_2&b_2&\cdots&0\\
0&0&a_3&\ddots&0\\
\vdots&\vdots&\ddots&\ddots&b_{n-1}\\
0&0&\cdots&0&a_n
\end{pmatrix}\right\rvert_{n,1}=(-1)^{1-n}a_n b_{n-1}^{-1}a_{n-1}b_{n-2}^{-1}\cdots a_2b_1^{-1}a_1. \tag*{(23)} a 1 0 0 ⋮ 0 b 1 a 2 0 ⋮ 0 0 b 2 a 3 ⋱ ⋯ ⋯ ⋯ ⋱ ⋱ 0 0 0 0 b n − 1 a n n , 1 = ( − 1 ) 1 − n a n b n − 1 − 1 a n − 1 b n − 2 − 1 ⋯ a 2 b 1 − 1 a 1 . (23)
Proof. For the first assertion, the general identity is standard; see [27 ] . For illustration, let A A A be the displayed block matrix with n = 3 n=3 n = 3 , identity blocks on the first upper diagonal, and zero blocks above that diagonal. Then the defining formula gives
∣ A ∣ 3 , 1 = a 3 , 1 − ( a 3 , 2 a 3 , 3 ) ( 1 0 a 2 , 2 1 ) − 1 ( a 1 , 1 a 2 , 1 ) = a 3 , 1 + a 3 , 3 a 2 , 2 a 1 , 1 − a 3 , 2 a 1 , 1 − a 3 , 3 a 2 , 1 . \lvert A\rvert_{3,1}=a_{3,1}-(a_{3,2}\quad a_{3,3})\begin{pmatrix}1&0\\a_{2,2}&1\end{pmatrix}^{-1}\begin{pmatrix}a_{1,1}\\a_{2,1}\end{pmatrix}=a_{3,1}+a_{3,3}a_{2,2}a_{1,1}-a_{3,2}a_{1,1}-a_{3,3}a_{2,1}. ∣ A ∣ 3 , 1 = a 3 , 1 − ( a 3 , 2 a 3 , 3 ) ( 1 a 2 , 2 0 1 ) − 1 ( a 1 , 1 a 2 , 1 ) = a 3 , 1 + a 3 , 3 a 2 , 2 a 1 , 1 − a 3 , 2 a 1 , 1 − a 3 , 3 a 2 , 1 .
This expression coincides with the row determinant
rdet ( a 1 , 1 1 0 a 2 , 1 a 2 , 2 1 a 3 , 1 a 3 , 2 a 3 , 3 ) = a 3 , 3 a 2 , 2 a 1 , 1 + a 3 , 1 − a 3 , 2 a 1 , 1 − a 3 , 3 a 2 , 1 . \operatorname{rdet}\begin{pmatrix} a_{1,1} & 1 & 0 \\ a_{2,1} & a_{2,2} & 1 \\ a_{3,1} & a_{3,2} & a_{3,3} \end{pmatrix}=a_{3,3}a_{2,2}a_{1,1}+a_{3,1}-a_{3,2}a_{1,1}-a_{3,3}a_{2,1}. rdet a 1 , 1 a 2 , 1 a 3 , 1 1 a 2 , 2 a 3 , 2 0 1 a 3 , 3 = a 3 , 3 a 2 , 2 a 1 , 1 + a 3 , 1 − a 3 , 2 a 1 , 1 − a 3 , 3 a 2 , 1 .
The general statement follows by the same expansion.
For the second assertion, use Proposition 3.4 and the factorization
( a 1 b 1 0 ⋯ 0 0 a 2 b 2 ⋯ 0 0 0 a 3 ⋱ 0 ⋮ ⋮ ⋱ ⋱ b n − 1 0 0 ⋯ 0 a n ) = ( b 1 0 0 ⋯ 0 0 b 2 0 ⋯ 0 0 0 b 3 ⋱ 0 ⋮ ⋮ ⋱ ⋱ 0 0 0 ⋯ 0 1 ) ( b 1 − 1 a 1 1 0 ⋯ 0 0 b 2 − 1 a 2 1 ⋯ 0 0 0 b 3 − 1 a 3 ⋱ 0 ⋮ ⋮ ⋱ ⋱ 1 0 0 ⋯ 0 a n ) . \begin{pmatrix} a_1 & b_1 & 0 & \cdots& 0 \\ 0 & a_2 & b_2 & \cdots& 0 \\ 0 & 0 & a_3 & \ddots& 0 \\ \vdots& \vdots& \ddots& \ddots& b_{n-1} \\ 0 & 0 & \cdots& 0 & a_n \end{pmatrix}
=
\begin{pmatrix} b_1 & 0 & 0 & \cdots& 0 \\ 0 & b_2 & 0 & \cdots& 0 \\ 0 & 0 & b_3 & \ddots& 0 \\ \vdots& \vdots& \ddots& \ddots& 0 \\ 0 & 0 & \cdots& 0 & 1 \end{pmatrix}
\begin{pmatrix} b_1^{-1}a_1 & 1 & 0 & \cdots& 0 \\ 0 & b_2^{-1}a_2 & 1 & \cdots& 0 \\ 0 & 0 & b_3^{-1}a_3 & \ddots& 0 \\ \vdots& \vdots& \ddots& \ddots& 1 \\ 0 & 0 & \cdots& 0 & a_n \end{pmatrix}. a 1 0 0 ⋮ 0 b 1 a 2 0 ⋮ 0 0 b 2 a 3 ⋱ ⋯ ⋯ ⋯ ⋱ ⋱ 0 0 0 0 b n − 1 a n = b 1 0 0 ⋮ 0 0 b 2 0 ⋮ 0 0 0 b 3 ⋱ ⋯ ⋯ ⋯ ⋱ ⋱ 0 0 0 0 0 1 b 1 − 1 a 1 0 0 ⋮ 0 1 b 2 − 1 a 2 0 ⋮ 0 0 1 b 3 − 1 a 3 ⋱ ⋯ ⋯ ⋯ ⋱ ⋱ 0 0 0 0 1 a n .
The left factor does not change the corner quasideterminant. Applying the first assertion to the right factor and evaluating its row determinant gives the stated product.
Commutation relations
We now regard L ( u ) L(u) L ( u ) as an n × n n \times n n × n matrix whose entries are p × p p \times p p × p matrices over U ( g l N ) ( ( u − 1 ) ) \mathcal{U}(\mathfrak{gl}_N)((u^{-1})) U ( gl N ) (( u − 1 )) . In this subsection only, it is convenient to retain the notation
L a , b ( u ) = u δ a , b + E a , b , L a , b i ( u ) = ( L a , b ( u ) ) i , j . L_{a,b}(u)=u\delta_{a,b}+E_{a,b},\qquad L^i_{a,b}(u)=(L_{a,b}(u))_{i,j}. L a , b ( u ) = u δ a , b + E a , b , L a , b i ( u ) = ( L a , b ( u ) ) i , j .
Definition 3.7. Suppose that the block submatrix obtained by deleting the last block row and the first block column is invertible in a suitable overring. The corner quasideterminant of the block matrix L ( u ) L(u) L ( u ) is
T ( u ) = ∣ L ( u ) ∣ n , 1 = L n , 1 ( u ) − ( L n , 2 ( u ) ⋯ L n , n ( u ) ) ( L 1 , 2 ( u ) ⋯ L 1 , n ( u ) ⋮ ⋱ ⋮ L n − 1 , 2 ( u ) ⋯ L n − 1 , n ( u ) ) − 1 ( L 1 , 1 ( u ) ⋮ L n − 1 , 1 ( u ) ) . (24) T(u)=|L(u)|_{n,1}=L_{n,1}(u)-\bigl(L_{n,2}(u)\ \cdots\ L_{n,n}(u)\bigr)
\begin{pmatrix}
L_{1,2}(u) & \cdots& L_{1,n}(u)\\
\vdots& \ddots& \vdots\\
L_{n-1,2}(u) & \cdots& L_{n-1,n}(u)
\end{pmatrix}^{-1}
\begin{pmatrix}
L_{1,1}(u)\\
\vdots\\
L_{n-1,1}(u)
\end{pmatrix}. \tag*{(24)} T ( u ) = ∣ L ( u ) ∣ n , 1 = L n , 1 ( u ) − ( L n , 2 ( u ) ⋯ L n , n ( u ) ) L 1 , 2 ( u ) ⋮ L n − 1 , 2 ( u ) ⋯ ⋱ ⋯ L 1 , n ( u ) ⋮ L n − 1 , n ( u ) − 1 L 1 , 1 ( u ) ⋮ L n − 1 , 1 ( u ) . (24)
The matrix L ( u ) = u + E L(u)=u+E L ( u ) = u + E is invertible as a formal Laurent series:
L ( u ) − 1 = u − 1 ∑ r ≥ 0 ( − u − 1 E ) r ∈ Mat N ( U ( g l N ) ( ( u − 1 ) ) ) . L(u)^{-1}=u^{-1}\sum_{r\geq0}(-u^{-1}E)^r\in\operatorname{Mat}_N\bigl(\mathcal{U}(\mathfrak{gl}_N)((u^{-1}))\bigr). L ( u ) − 1 = u − 1 r ≥ 0 ∑ ( − u − 1 E ) r ∈ Mat N ( U ( gl N ) (( u − 1 )) ) .
Using the same n × n n \times n n × n block decomposition as in (15) , write
L ( u ) − 1 = ( M 1 , 1 ( u ) M 1 , 2 ( u ) ⋯ M 1 , n ( u ) M 2 , 1 ( u ) M 2 , 2 ( u ) ⋯ M 2 , n ( u ) ⋮ ⋮ ⋱ ⋮ M n , 1 ( u ) M n , 2 ( u ) ⋯ M n , n ( u ) ) . (25) L(u)^{-1}=
\begin{pmatrix}
M_{1,1}(u) & M_{1,2}(u) & \cdots& M_{1,n}(u)\\
M_{2,1}(u) & M_{2,2}(u) & \cdots& M_{2,n}(u)\\
\vdots& \vdots& \ddots& \vdots\\
M_{n,1}(u) & M_{n,2}(u) & \cdots& M_{n,n}(u)
\end{pmatrix}. \tag*{(25)} L ( u ) − 1 = M 1 , 1 ( u ) M 2 , 1 ( u ) ⋮ M n , 1 ( u ) M 1 , 2 ( u ) M 2 , 2 ( u ) ⋮ M n , 2 ( u ) ⋯ ⋯ ⋱ ⋯ M 1 , n ( u ) M 2 , n ( u ) ⋮ M n , n ( u ) . (25)
where each p × p p \times p p × p block is written explicitly as
M a , b ( u ) = ∑ i , j = 1 p e i , j ⊗ M a , b i , j ( u ) . M_{a,b}(u)=\sum_{i,j=1}^{p}e_{i,j}\otimes M_{a,b}^{i,j}(u). M a , b ( u ) = i , j = 1 ∑ p e i , j ⊗ M a , b i , j ( u ) .
Thus
∑ b = 1 n ∑ j = 1 p L a , b i , j ( u ) M b , c j , k ( u ) = δ a , c δ i , k , ∑ b = 1 n ∑ j = 1 p M a , b i , j ( u ) L b , c j , k ( u ) = δ a , c δ i , k . (26) \sum_{b=1}^{n}\sum_{j=1}^{p}L_{a,b}^{i,j}(u)M_{b,c}^{j,k}(u)=\delta_{a,c}\delta_{i,k},\qquad\sum_{b=1}^{n}\sum_{j=1}^{p}M_{a,b}^{i,j}(u)L_{b,c}^{j,k}(u)=\delta_{a,c}\delta_{i,k}. \tag*{(26)} b = 1 ∑ n j = 1 ∑ p L a , b i , j ( u ) M b , c j , k ( u ) = δ a , c δ i , k , b = 1 ∑ n j = 1 ∑ p M a , b i , j ( u ) L b , c j , k ( u ) = δ a , c δ i , k . (26)
Example. When n = 2 n=2 n = 2 , the block matrices are
L ( u ) = ( L 1 , 1 ( u ) L 1 , 2 ( u ) L 2 , 1 ( u ) L 2 , 2 ( u ) ) , L ( u ) − 1 = ( M 1 , 1 ( u ) M 1 , 2 ( u ) M 2 , 1 ( u ) M 2 , 2 ( u ) ) . L(u)=\begin{pmatrix}L_{1,1}(u)&L_{1,2}(u)\\L_{2,1}(u)&L_{2,2}(u)\end{pmatrix},\qquad L(u)^{-1}=\begin{pmatrix}M_{1,1}(u)&M_{1,2}(u)\\M_{2,1}(u)&M_{2,2}(u)\end{pmatrix}. L ( u ) = ( L 1 , 1 ( u ) L 2 , 1 ( u ) L 1 , 2 ( u ) L 2 , 2 ( u ) ) , L ( u ) − 1 = ( M 1 , 1 ( u ) M 2 , 1 ( u ) M 1 , 2 ( u ) M 2 , 2 ( u ) ) .
The relations L ( u ) L ( u ) − 1 = L ( u ) − 1 L ( u ) = 1 L(u)L(u)^{-1}=L(u)^{-1}L(u)=1 L ( u ) L ( u ) − 1 = L ( u ) − 1 L ( u ) = 1 become
( L 1 , 1 ( u ) L 1 , 2 ( u ) L 2 , 1 ( u ) L 2 , 2 ( u ) ) ( M 1 , 1 ( u ) M 1 , 2 ( u ) M 2 , 1 ( u ) M 2 , 2 ( u ) ) = ( M 1 , 1 ( u ) M 1 , 2 ( u ) M 2 , 1 ( u ) M 2 , 2 ( u ) ) ( L 1 , 1 ( u ) L 1 , 2 ( u ) L 2 , 1 ( u ) L 2 , 2 ( u ) ) = ( 1 p 0 0 1 p ) . \begin{pmatrix}L_{1,1}(u)&L_{1,2}(u)\\L_{2,1}(u)&L_{2,2}(u)\end{pmatrix}\begin{pmatrix}M_{1,1}(u)&M_{1,2}(u)\\M_{2,1}(u)&M_{2,2}(u)\end{pmatrix}=\begin{pmatrix}M_{1,1}(u)&M_{1,2}(u)\\M_{2,1}(u)&M_{2,2}(u)\end{pmatrix}\begin{pmatrix}L_{1,1}(u)&L_{1,2}(u)\\L_{2,1}(u)&L_{2,2}(u)\end{pmatrix}=\begin{pmatrix}1_p&0\\0&1_p\end{pmatrix}. ( L 1 , 1 ( u ) L 2 , 1 ( u ) L 1 , 2 ( u ) L 2 , 2 ( u ) ) ( M 1 , 1 ( u ) M 2 , 1 ( u ) M 1 , 2 ( u ) M 2 , 2 ( u ) ) = ( M 1 , 1 ( u ) M 2 , 1 ( u ) M 1 , 2 ( u ) M 2 , 2 ( u ) ) ( L 1 , 1 ( u ) L 2 , 1 ( u ) L 1 , 2 ( u ) L 2 , 2 ( u ) ) = ( 1 p 0 0 1 p ) .
Whenever the corner quasideterminant T ( u ) = ∣ L ( u ) ∣ n , 1 T(u)=|L(u)|_{n,1} T ( u ) = ∣ L ( u ) ∣ n , 1 is defined in a localized ring A \mathcal{A} A containing U ( g l N ( ( u − 1 ) ) ) U(\mathfrak{gl}_N((u^{-1}))) U ( gl N (( u − 1 ))) , Proposition 3.3 gives
M 1 , n ( u ) = T ( u ) − 1 . (27) M_{1,n}(u)=T(u)^{-1}. \tag*{(27)} M 1 , n ( u ) = T ( u ) − 1 . (27)
Recall that the commutation relations among the entries of L ( u ) L(u) L ( u ) can be written in RTT form (7) as
( u − v ) L a , b k , i ( u ) L c , d s , j ( v ) + L c , b s , i ( u ) L a , d k , j ( v ) = ( u − v ) L c , d s , j ( v ) L a , b k , i ( u ) + L c , b s , i ( v ) L a , d k , j ( u ) . (28) (u-v)L_{a,b}^{k,i}(u)L_{c,d}^{s,j}(v)+L_{c,b}^{s,i}(u)L_{a,d}^{k,j}(v)=(u-v)L_{c,d}^{s,j}(v)L_{a,b}^{k,i}(u)+L_{c,b}^{s,i}(v)L_{a,d}^{k,j}(u). \tag*{(28)} ( u − v ) L a , b k , i ( u ) L c , d s , j ( v ) + L c , b s , i ( u ) L a , d k , j ( v ) = ( u − v ) L c , d s , j ( v ) L a , b k , i ( u ) + L c , b s , i ( v ) L a , d k , j ( u ) . (28)
We use this identity to establish the two properties required for Hamiltonian reduction.
Proposition 3.8. For 1 ≤ a ≤ n − 1 1\leq a\leq n-1 1 ≤ a ≤ n − 1 and 2 ≤ b ≤ n 2\leq b\leq n 2 ≤ b ≤ n , every entry of L a , b ( u ) L_{a,b}(u) L a , b ( u ) commutes with every entry of M 1 , n ( v ) M_{1,n}(v) M 1 , n ( v ) :
[ L a , b i , j ( u ) , M 1 , n ( v ) ] = 0 , (29) [L_{a,b}^{i,j}(u),M_{1,n}(v)]=0, \tag*{(29)} [ L a , b i , j ( u ) , M 1 , n ( v )] = 0 , (29) where the commutator is understood entrywise. Consequently, whenever T ( v ) T(v) T ( v ) is defined and invertible in a common overring,
[ L a , b i , j ( u ) , T ( v ) ] = 0. (30) [L_{a,b}^{i,j}(u),T(v)]=0. \tag*{(30)} [ L a , b i , j ( u ) , T ( v )] = 0. (30)
Proof. Multiply (28) by the entries of L ( v ) − 1 L(v)^{-1} L ( v ) − 1 and use (26) . The resulting identity is
( u − v ) M e , f q , m ( v ) L a , b k , i ( u ) + ∑ c = 1 n ∑ s = 1 p M e , c q , s ( v ) L c , b s , i ( u ) δ k , m δ a , f = ( u − v ) L a , b k , i ( u ) M e , f q , m ( v ) + ∑ d = 1 n ∑ j = 1 p δ q , i δ e , b L a , d k , j ( u ) M d , f j , m ( v ) . (31) \begin{aligned}
(u-v)M_{e,f}^{q,m}(v)L_{a,b}^{k,i}(u)+\sum_{c=1}^{n}\sum_{s=1}^{p}M_{e,c}^{q,s}(v)L_{c,b}^{s,i}(u)\delta_{k,m}\delta_{a,f} \\
&=(u-v)L_{a,b}^{k,i}(u)M_{e,f}^{q,m}(v)+\sum_{d=1}^{n}\sum_{j=1}^{p}\delta_{q,i}\delta_{e,b}L_{a,d}^{k,j}(u)M_{d,f}^{j,m}(v). \tag*{(31)}
\end{aligned} ( u − v ) M e , f q , m ( v ) L a , b k , i ( u ) + c = 1 ∑ n s = 1 ∑ p M e , c q , s ( v ) L c , b s , i ( u ) δ k , m δ a , f = ( u − v ) L a , b k , i ( u ) M e , f q , m ( v ) + d = 1 ∑ n j = 1 ∑ p δ q , i δ e , b L a , d k , j ( u ) M d , f j , m ( v ) . (31)
Set e = 1 e=1 e = 1 and f = n f=n f = n . If a ≤ n − 1 a \le n-1 a ≤ n − 1 , then δ a , f = 0 \delta_{a,f}=0 δ a , f = 0 ; if b ≥ 2 b \ge2 b ≥ 2 , then δ e , b = 0 \delta_{e,b}=0 δ e , b = 0 . Hence both terms containing Kronecker symbols vanish, and the relation above reduces to
M 1 , n q , m ( v ) L a , b k , i ( u ) = L a , b k , i ( u ) M 1 , n q , m ( v ) . M_{1,n}^{q,m}(v)L_{a,b}^{k,i}(u)=L_{a,b}^{k,i}(u)M_{1,n}^{q,m}(v). M 1 , n q , m ( v ) L a , b k , i ( u ) = L a , b k , i ( u ) M 1 , n q , m ( v ) .
Finally, (27) and the general rule
[ A , B − 1 ] = − B − 1 [ A , B ] B − 1 [A,B^{-1}]=-B^{-1}[A,B]B^{-1} [ A , B − 1 ] = − B − 1 [ A , B ] B − 1
give (30) .
Proposition 3.9. The entries of the block M 1 , n ( u ) M_{1,n}(u) M 1 , n ( u ) satisfy the opposite R T T RTT R T T relation
( u − v ) M 1 , n i , j ( u ) M 1 , n q , m ( v ) + M 1 , n i , m ( u ) M 1 , n q , j ( v ) = ( u − v ) M 1 , n q , m ( v ) M 1 , n i , j ( u ) + M 1 , n i , m ( v ) M 1 , n q , j ( u ) . (32) \begin{aligned}
(u-v)M_{1,n}^{i,j}(u)M_{1,n}^{q,m}(v)+M_{1,n}^{i,m}(u)M_{1,n}^{q,j}(v)\\
=(u-v)M_{1,n}^{q,m}(v)M_{1,n}^{i,j}(u)+M_{1,n}^{i,m}(v)M_{1,n}^{q,j}(u). \tag*{(32)}
\end{aligned} ( u − v ) M 1 , n i , j ( u ) M 1 , n q , m ( v ) + M 1 , n i , m ( u ) M 1 , n q , j ( v ) = ( u − v ) M 1 , n q , m ( v ) M 1 , n i , j ( u ) + M 1 , n i , m ( v ) M 1 , n q , j ( u ) . (32) Equivalently, in matrix form,
( M 1 , n ( u ) ) 1 ( M 1 , n ( v ) ) 2 R ( u − v ) = R ( u − v ) ( M 1 , n ( v ) ) 2 ( M 1 , n ( u ) ) 1 . (33) (M_{1,n}(u))_1(M_{1,n}(v))_2R(u-v)=R(u-v)(M_{1,n}(v))_2(M_{1,n}(u))_1. \tag*{(33)} ( M 1 , n ( u ) ) 1 ( M 1 , n ( v ) ) 2 R ( u − v ) = R ( u − v ) ( M 1 , n ( v ) ) 2 ( M 1 , n ( u ) ) 1 . (33) Proof. Multiply (31) by the entries of L ( u ) − 1 L(u)^{-1} L ( u ) − 1 and use (26) once more. This gives
( u − v ) M c , h i , j ( u ) M e , f q , m ( v ) + M c , f i , m ( u ) M e , h q , j ( v ) = ( u − v ) M e , f q , m ( v ) M c , h i , j ( u ) + M c , f i , m ( v ) M e , h q , j ( u ) . (34) \begin{aligned}
(u-v)M_{c,h}^{i,j}(u)M_{e,f}^{q,m}(v)+M_{c,f}^{i,m}(u)M_{e,h}^{q,j}(v)\\
=(u-v)M_{e,f}^{q,m}(v)M_{c,h}^{i,j}(u)+M_{c,f}^{i,m}(v)M_{e,h}^{q,j}(u). \tag*{(34)}
\end{aligned} ( u − v ) M c , h i , j ( u ) M e , f q , m ( v ) + M c , f i , m ( u ) M e , h q , j ( v ) = ( u − v ) M e , f q , m ( v ) M c , h i , j ( u ) + M c , f i , m ( v ) M e , h q , j ( u ) . (34) Taking c = e = 1 c=e=1 c = e = 1 and h = f = n h=f=n h = f = n gives (32) . Relations of this type are equivalent to the matrix RTT relation with the rational R R R -matrix.
Explicit computations are written out in Appendix § 8 .§ 1 .
Corollary 3.10. Whenever T ( u ) T(u) T ( u ) is defined and invertible in a common overring, it satisfies the R T T RTT R T T relation
R ( u − v ) T 1 ( u ) T 2 ( v ) = T 2 ( v ) T 1 ( u ) R ( u − v ) . (35) R(u-v)T_1(u)T_2(v)=T_2(v)T_1(u)R(u-v). \tag*{(35)} R ( u − v ) T 1 ( u ) T 2 ( v ) = T 2 ( v ) T 1 ( u ) R ( u − v ) . (35) Proof. By (27) , the opposite RTT relation (33) is the corresponding relation for T ( u ) − 1 T(u)^{-1} T ( u ) − 1 . Taking inverses and multiplying by R ( u − v ) R(u-v) R ( u − v ) on the left and right gives (35) .
Remark 3.11. The formal inverse L ( u ) − 1 L(u)^{-1} L ( u ) − 1 is sufficient for the two propositions above. The reduction argument in § 4.3 shows why the result is independent of the choice of compatible localization.
Reduction and W-algebras
In the first subsection, we recall the basic definitions and results of Brundan–Kleshchev [5 ] on finite W-algebras, building on the earlier rectangular construction of Ragoucy–Sorba [33 ] . These works provide explicit formulas for the invariants, although the origin of these formulas, especially that of the shifts, remains unclear.
The finite W-algebra
Define the following subalgebras of g l N \mathfrak{gl}_N gl N directly in terms of the matrix units:
m n = span { E a , b i , j ∣ a > b } , m n + = span { E a , b i , j ∣ a < b } , l n = span { E a , a i , j ∣ 1 ≤ a ≤ n } , p n = span { E a , b i , j ∣ a ≤ b } = l n ⊕ m n + . (36) \begin{aligned}
\mathfrak{m}_n &= \operatorname{span}\{E^{i,j}_{a,b}\mid a>b\}, \qquad&\mathfrak{m}_n^+ &= \operatorname{span}\{E^{i,j}_{a,b}\mid a<b\},\\
\mathfrak{l}_n &= \operatorname{span}\{E^{i,j}_{a,a}\mid1\leq a\leq n\}, \qquad&\mathfrak{p}_n &= \operatorname{span}\{E^{i,j}_{a,b}\mid a\leq b\}=\mathfrak{l}_n\oplus\mathfrak{m}_n^+. \tag*{(36)}
\end{aligned} m n l n = span { E a , b i , j ∣ a > b } , = span { E a , a i , j ∣ 1 ≤ a ≤ n } , m n + p n = span { E a , b i , j ∣ a < b } , = span { E a , b i , j ∣ a ≤ b } = l n ⊕ m n + . (36)
Here and below, the internal indices i , j i,j i , j range from 1 to p p p in each displayed span. Thus m n \mathfrak{m}_n m n and m n + \mathfrak{m}_n^+ m n + are the strictly lower- and upper-block-triangular subalgebras, l n \mathfrak{l}_n l n is the block-diagonal subalgebra, and p n \mathfrak{p}_n p n is the upper-block-triangular parabolic subalgebra.
Define a character ψ : m n → C \psi:\mathfrak{m}_n\to\mathbb{C} ψ : m n → C by
ψ ( E a , b i , j ) = δ i , j δ a , b + 1 , a > b . (37) \psi(E^{i,j}_{a,b})=\delta_{i,j}\delta_{a,b+1},\qquad a>b. \tag*{(37)} ψ ( E a , b i , j ) = δ i , j δ a , b + 1 , a > b . (37)
Let I ψ = U ( g l N ) ( m n − ψ ) I_\psi=U(\mathfrak{gl}_N)(\mathfrak{m}_n-\psi) I ψ = U ( gl N ) ( m n − ψ ) be the corresponding left ideal. Set Q ψ = U ( g l N ) / I ψ Q_\psi=U(\mathfrak{gl}_N)/I_\psi Q ψ = U ( gl N ) / I ψ , 1 ‾ = 1 + I ψ \overline{1}=1+I_\psi 1 = 1 + I ψ .
Definition 4.1. Define the rectangular finite W W W -algebra is
W n ( g l p ) = Q ψ m n , (38) W_n(\mathfrak{gl}_p)=Q_\psi^{\mathfrak{m}_n}, \tag*{(38)} W n ( gl p ) = Q ψ m n , (38) where m n \mathfrak{m}_n m n acts on Q ψ Q_\psi Q ψ by the adjoint action. Explicitly,
Q ψ m n = { f ‾ 1 ∈ Q ψ ∣ [ x , f ‾ ] 1 ‾ = 0 for every x ∈ m n } . Q_\psi^{\mathfrak{m}_n}=\{\overline{f}1\in Q_\psi\mid[x,\overline{f}]\overline{1}=0\text{ for every }x\in\mathfrak{m}_n\}. Q ψ m n = { f 1 ∈ Q ψ ∣ [ x , f ] 1 = 0 for every x ∈ m n } .
The PBW decomposition associated with g l N = p n ⊕ m n \mathfrak{gl}_N=\mathfrak{p}_n\oplus\mathfrak{m}_n gl N = p n ⊕ m n gives a vector-space isomorphism
U ( p n ) ⊗ U ( m n ) → ∼ U ( g l N ) . U(\mathfrak{p}_n)\otimes U(\mathfrak{m}_n)\xrightarrow{\sim}U(\mathfrak{gl}_N). U ( p n ) ⊗ U ( m n ) ∼ U ( gl N ) .
Applying the character ψ \psi ψ to the second tensor factor defines the projection
pr ψ : U ( g l N ) ⟶ U ( p n ) . \operatorname{pr}_\psi:U(\mathfrak{gl}_N)\longrightarrow U(\mathfrak{p}_n). pr ψ : U ( gl N ) ⟶ U ( p n ) .
Its kernel is I ψ I_\psi I ψ , so it induces an isomorphism Q ψ ≃ U ( p n ) Q_\psi\simeq U(\mathfrak{p}_n) Q ψ ≃ U ( p n ) of left U ( p n ) U(\mathfrak{p}_n) U ( p n ) -modules. Under this identification, the finite W W W -algebra is realized as a subalgebra of U ( p n ) U(\mathfrak{p}_n) U ( p n ) , giving an injective algebra homomorphism
i : W n ( g l p ) ↪ U ( p n ) . i:W_n(\mathfrak{gl}_p)\hookrightarrow U(\mathfrak{p}_n). i : W n ( gl p ) ↪ U ( p n ) .
The assignments E a , a i , j ↦ E [ a ] i , j E^{i,j}_{a,a}\mapsto E^{i,j}_{[a]} E a , a i , j ↦ E [ a ] i , j induce a canonical algebra isomorphism
κ : U ( l n ) → ∼ U ( g l p ) ⊗ n . (39) \kappa:U(\mathfrak{l}_n)\xrightarrow{\sim}U(\mathfrak{gl}_p)^{\otimes n}. \tag*{(39)} κ : U ( l n ) ∼ U ( gl p ) ⊗ n . (39)
We use different notation on the two sides of this isomorphism. Set
E a = E a , a = ∑ i , j = 1 p e i , j ⊗ E a , a j , i ∈ Mat p ( C ) ⊗ U ( l n ) \mathbf{E}_a=E_{a,a}=\sum_{i,j=1}^{p}e_{i,j}\otimes E^{j,i}_{a,a}\in\operatorname{Mat}_p(\mathbb{C})\otimes U(\mathfrak{l}_n) E a = E a , a = i , j = 1 ∑ p e i , j ⊗ E a , a j , i ∈ Mat p ( C ) ⊗ U ( l n )
and
E [ a ] = ∑ i , j = 1 p e i , j ⊗ E [ a ] j , i ∈ Mat p ( C ) ⊗ U ( g l p ) ⊗ n . E_{[a]}=\sum_{i,j=1}^{p}e_{i,j}\otimes E^{j,i}_{[a]}\in\operatorname{Mat}_p(\mathbb{C})\otimes U(\mathfrak{gl}_p)^{\otimes n}. E [ a ] = i , j = 1 ∑ p e i , j ⊗ E [ a ] j , i ∈ Mat p ( C ) ⊗ U ( gl p ) ⊗ n .
Thus κ ( E a ) = E [ a ] \kappa(\mathbf{E}_a)=E_{[a]} κ ( E a ) = E [ a ] , where unbracketed block labels refer to diagonal blocks in g l N \mathfrak{gl}_N gl N and square-bracketed subscripts label tensor factors.
Brundan–Kleshchev generating matrix
We now construct the corresponding generating matrix in the finite W W W -algebra, using the row determinant defined in (21) . In the quotient by I ψ I_\psi I ψ , the strictly upper-block entries of (15) satisfy
E a , a + 1 = 1 p , E a , b = 0 ( b > a + 1 ) . E_{a,a+1}=1_p,\qquad E_{a,b}=0\quad(b>a+1). E a , a + 1 = 1 p , E a , b = 0 ( b > a + 1 ) .
We consider
L r e d ( u ) = ( u − p ( n − 1 ) + E 1 , 1 1 p ⋯ 0 E 2 , 1 u − p ( n − 2 ) + E 2 , 2 ⋱ ⋮ ⋮ ⋱ ⋱ 1 p E n , 1 ⋯ E n , n − 1 u + E n , n ) (40) L^{\mathrm{red}}(u)=\begin{pmatrix}
u-p(n-1)+E_{1,1} & 1_p & \cdots& 0\\
E_{2,1} & u-p(n-2)+E_{2,2} & \ddots& \vdots\\
\vdots& \ddots& \ddots& 1_p\\
E_{n,1} & \cdots& E_{n,n-1} & u+E_{n,n}
\end{pmatrix} \tag*{(40)} L red ( u ) = u − p ( n − 1 ) + E 1 , 1 E 2 , 1 ⋮ E n , 1 1 p u − p ( n − 2 ) + E 2 , 2 ⋱ ⋯ ⋯ ⋱ ⋱ E n , n − 1 0 ⋮ 1 p u + E n , n (40)
This is a matrix with entries in Q ψ [ u ] Q_\psi[u] Q ψ [ u ] .
Proposition 4.2 (Brundan–Kleshchev [5 ] ). The matrix entries of the polynomial
T ( u ) = rdet ( L r e d ( u ) ) (41) T(u)=\operatorname{rdet}\left(L^{\mathrm{red}}(u)\right) \tag*{(41)} T ( u ) = rdet ( L red ( u ) ) (41) belong to W n ( g l p ) [ u ] W_n(\mathfrak{gl}_p)[u] W n ( gl p ) [ u ] , and their coefficients generate W n ( g l p ) W_n(\mathfrak{gl}_p) W n ( gl p ) .
Example. In case N = 2 p N=2p N = 2 p , n = 2 n=2 n = 2 the Brundan-Kleschev matrix has form:
T ( u ) = ( u + E 2 , 2 ) ( u − p + E 1 , 1 ) − E 2 , 1 . T(u)=(u+E_{2,2})(u-p+E_{1,1})-E_{2,1}. T ( u ) = ( u + E 2 , 2 ) ( u − p + E 1 , 1 ) − E 2 , 1 .
If we set E 2 , 1 = 0 E_{2,1}=0 E 2 , 1 = 0 , than we recover the usual truncated Yangian generating matrix. The generalization to generic N N N is obvious. Consider the projection onto the block-diagonal factor. It gives an algebra homomorphism
η : U ( p n ) ⟶ U ( l n ) , U ( p n ) / m n + U ( p n ) ≃ U ( l n ) . \eta:U(\mathfrak{p}_n)\longrightarrow U(\mathfrak{l}_n),\qquad U(\mathfrak{p}_n)/\mathfrak{m}_n^+U(\mathfrak{p}_n)\simeq U(\mathfrak{l}_n). η : U ( p n ) ⟶ U ( l n ) , U ( p n ) / m n + U ( p n ) ≃ U ( l n ) .
Under the composition η ∘ i \eta\circ i η ∘ i , the image of the generating matrix is
( η ∘ i ) ( T ( u ) ) = ( u + E n ) ( u − p + E n − 1 ) ⋯ ( u − ( n − 1 ) p + E 1 ) . (\eta\circ i)(T(u))=(u+E_n)(u-p+E_{n-1})\cdots(u-(n-1)p+E_1). ( η ∘ i ) ( T ( u )) = ( u + E n ) ( u − p + E n − 1 ) ⋯ ( u − ( n − 1 ) p + E 1 ) .
To remove these scalar shifts, we apply the automorphism E a , a i , j ↦ E a , a i , j + ( n − a ) p δ i , j E_{a,a}^{i,j}\mapsto E_{a,a}^{i,j}+(n-a)p\delta_{i,j} E a , a i , j ↦ E a , a i , j + ( n − a ) p δ i , j . We then use κ \kappa κ to identify the enveloping algebra of the block-diagonal subalgebra with U ( g l p ) ⊗ n U(\mathfrak{gl}_p)^{\otimes n} U ( gl p ) ⊗ n . Composing these maps defines the quantum Miura homomorphism
Θ : W n ( g l p ) ⟶ U ( g l p ) ⊗ n , \Theta:W_n(\mathfrak{gl}_p)\longrightarrow U(\mathfrak{gl}_p)^{\otimes n}, Θ : W n ( gl p ) ⟶ U ( gl p ) ⊗ n ,
which acts on the generating matrix as follows:
Θ ( T ( u ) ) = ( u + E [ n ] ) ( u + E [ n − 1 ] ) ⋯ ( u + E [ 1 ] ) . (42) \Theta(T(u))=(u+E_{[n]})(u+E_{[n-1]})\cdots(u+E_{[1]}). \tag*{(42)} Θ ( T ( u )) = ( u + E [ n ] ) ( u + E [ n − 1 ] ) ⋯ ( u + E [ 1 ] ) . (42)
This is the coproduct-evaluation generating matrix associated with (11) . The quantum Miura homomorphism is summarized by the diagram
W n ( g l p ) → i U ( p n ) Θ ↓ ↓ Y ( n ) ( g l p ) → U ( g l p ) ⊗ n \begin{CD}
\mathcal{W}_n(\mathfrak{gl}_p) @>i>> \mathcal{U}(\mathfrak{p}_n) \\
@V\Theta VV @VVV \\
Y^{(n)}(\mathfrak{gl}_p) @>>> \mathcal{U}(\mathfrak{gl}_p)^{\otimes n}
\end{CD} W n ( gl p ) Θ ↓ ⏐ Y ( n ) ( gl p ) i U ( p n ) ↓ ⏐ U ( gl p ) ⊗ n
Theorem 4.3 (Brundan–Kleshchev [5 ] ). The homomorphism Θ \Theta Θ identifies W n ( g l p ) \mathcal{W}_n(\mathfrak{gl}_p) W n ( gl p ) *with the level-*n truncated Yangian Y ( n ) ( g l p ) Y^{(n)}(\mathfrak{gl}_p) Y ( n ) ( gl p ) generated by the coefficients of (11) .
Reduction of the quasideterminant
Assume that we have constructed a ring A \mathcal{A} A in which the quasideterminant is defined, or in which its rescaled version is defined, and consider any such compatible localization. We now give a general argument showing that the reduction does not depend on the particular choice of localization. Denote by I ψ I_\psi I ψ the left ideal generated by the reduction constraints m − ψ ( m ) m-\psi(m) m − ψ ( m ) , for m ∈ m n m\in\mathfrak{m}_n m ∈ m n . The quotient map restricts to the homomorphism of algebras of invariants A m n → ( A / I ψ ) m n \mathcal{A}^{\mathfrak{m}_n}\to(\mathcal{A}/I_\psi)^{\mathfrak{m}_n} A m n → ( A / I ψ ) m n .
As shown above, the matrix entries of the quasideterminant belong to A m n \mathcal{A}^{\mathfrak{m}_n} A m n . Thus, it follows that the reduction T ( u ) 1 ‾ T(u)\overline{1} T ( u ) 1 remains invariant and satisfies the RTT relation. Let us identify the reduced quasideterminant with the Brundan–Kleshchev generating matrix.
Theorem 4.4. The reduction of the quasideterminant coincides with the Brundan–Kleshchev generating matrix:
( − 1 ) n − 1 T ( u ) 1 ‾ = ( − 1 ) n − 1 ∣ L r e d ( u ) ∣ = T ( u ) . (43) (-1)^{n-1}T(u)\overline{1}=(-1)^{n-1}\left|L^{\mathrm{red}}(u)\right|=\mathbf{T}(u). \tag*{(43)} ( − 1 ) n − 1 T ( u ) 1 = ( − 1 ) n − 1 L red ( u ) = T ( u ) . (43)
To illustrate the proof strategy, consider the case n = 3 n=3 n = 3 . Under reduction, the matrix entries satisfy L 1 , 2 i , j ( u ) 1 ‾ = L 2 , 3 i , j ( u ) 1 ‾ = δ i , j L_{1,2}^{i,j}(u)\overline{1}=L_{2,3}^{i,j}(u)\overline{1}=\delta_{i,j} L 1 , 2 i , j ( u ) 1 = L 2 , 3 i , j ( u ) 1 = δ i , j and L 1 , 3 i , j ( u ) 1 ‾ = 0 L_{1,3}^{i,j}(u)\overline{1}=0 L 1 , 3 i , j ( u ) 1 = 0 . Let
q ( u ) = ( L 1 , 2 ( u ) L 1 , 3 ( u ) L 2 , 2 ( u ) L 2 , 3 ( u ) ) − 1 ( L 1 , 1 ( u ) L 2 , 1 ( u ) ) = ( q 1 ( u ) q 2 ( u ) ) . \mathbf{q}(u)=
\begin{pmatrix}
L_{1,2}(u) & L_{1,3}(u) \\
L_{2,2}(u) & L_{2,3}(u)
\end{pmatrix}^{-1}
\begin{pmatrix}
L_{1,1}(u) \\
L_{2,1}(u)
\end{pmatrix}
=
\begin{pmatrix}
q_1(u) \\
q_2(u)
\end{pmatrix}. q ( u ) = ( L 1 , 2 ( u ) L 2 , 2 ( u ) L 1 , 3 ( u ) L 2 , 3 ( u ) ) − 1 ( L 1 , 1 ( u ) L 2 , 1 ( u ) ) = ( q 1 ( u ) q 2 ( u ) ) .
and we formally set q 0 ( u ) = 0 q_0(u)=0 q 0 ( u ) = 0 . The quasideterminant is then given by
∣ L ( u ) ∣ 3 , 1 = L 3 , 1 ( u ) − ( L 3 , 3 ( u ) L 3 , 2 ( u ) ) ( q 1 ( u ) q 2 ( u ) ) . \left|L(u)\right|_{3,1}=L_{3,1}(u)-\begin{pmatrix}L_{3,3}(u)&L_{3,2}(u)\end{pmatrix}
\begin{pmatrix}q_1(u)\\q_2(u)\end{pmatrix}. ∣ L ( u ) ∣ 3 , 1 = L 3 , 1 ( u ) − ( L 3 , 3 ( u ) L 3 , 2 ( u ) ) ( q 1 ( u ) q 2 ( u ) ) .
We denote the entries of q b ( u ) \mathbf{q}_b(u) q b ( u ) by q b j , k ( u ) q_b^{j,k}(u) q b j , k ( u ) . The following commutator identities hold:
∑ j = 1 p [ L 1 , 2 i , j ( u ) , q 1 j , k ( u ) ] = p δ i , k , ∑ j = 1 p [ L 1 , 3 i , j ( u ) , q 2 j , k ( u ) ] = p δ i , k , \sum_{j=1}^{p}\left[L_{1,2}^{i,j}(u),q_1^{j,k}(u)\right]=p\delta_{i,k},
\qquad
\sum_{j=1}^{p}\left[L_{1,3}^{i,j}(u),q_2^{j,k}(u)\right]=p\delta_{i,k}, j = 1 ∑ p [ L 1 , 2 i , j ( u ) , q 1 j , k ( u ) ] = p δ i , k , j = 1 ∑ p [ L 1 , 3 i , j ( u ) , q 2 j , k ( u ) ] = p δ i , k ,
and
∑ j = 1 p [ L 2 , 3 i , j ( u ) , q 2 j , k ( u ) ] = − p q 1 i , k ( u ) . \sum_{j=1}^{p}\left[L_{2,3}^{i,j}(u),q_2^{j,k}(u)\right]=-p q_1^{i,k}(u). j = 1 ∑ p [ L 2 , 3 i , j ( u ) , q 2 j , k ( u ) ] = − p q 1 i , k ( u ) .
We can combine these into a single formula:
∑ j = 1 p [ L a , b + 1 i , j ( u ) , q b j , k ( u ) ] = − p q a − 1 i , k ( u ) + p δ a , 1 δ i , k . \sum_{j=1}^{p}\left[L_{a,b+1}^{i,j}(u),q_b^{j,k}(u)\right]=-p q_{a-1}^{i,k}(u)+p\delta_{a,1}\delta_{i,k}. j = 1 ∑ p [ L a , b + 1 i , j ( u ) , q b j , k ( u ) ] = − p q a − 1 i , k ( u ) + p δ a , 1 δ i , k .
Next, consider the matrix identity
( L 1 , 2 ( u ) L 1 , 3 ( u ) L 2 , 2 ( u ) L 2 , 3 ( u ) ) ( q 1 ( u ) q 2 ( u ) ) = ( L 1 , 1 ( u ) L 2 , 1 ( u ) ) . \begin{pmatrix} L_{1,2}(u) & L_{1,3}(u) \\ L_{2,2}(u) & L_{2,3}(u) \end{pmatrix}
\begin{pmatrix} q_1(u) \\ q_2(u) \end{pmatrix}
=
\begin{pmatrix} L_{1,1}(u) \\ L_{2,1}(u) \end{pmatrix}. ( L 1 , 2 ( u ) L 2 , 2 ( u ) L 1 , 3 ( u ) L 2 , 3 ( u ) ) ( q 1 ( u ) q 2 ( u ) ) = ( L 1 , 1 ( u ) L 2 , 1 ( u ) ) .
Writing this identity in components and reordering factors as needed, we obtain the following equivalent relations:
2 p δ i , k + ∑ j = 1 p q 1 j , k ( u ) L 1 , 2 i , j ( u ) + q 2 j , k ( u ) L 1 , 3 i , j ( u ) = L 1 , 1 i , k ( u ) , 2p\delta_{i,k}+\sum_{j=1}^{p}q_1^{j,k}(u)L_{1,2}^{i,j}(u)+q_2^{j,k}(u)L_{1,3}^{i,j}(u)=L_{1,1}^{i,k}(u), 2 p δ i , k + j = 1 ∑ p q 1 j , k ( u ) L 1 , 2 i , j ( u ) + q 2 j , k ( u ) L 1 , 3 i , j ( u ) = L 1 , 1 i , k ( u ) ,
− p q 1 i , k + ∑ j = 1 p L 2 , 2 i , j ( u ) q 1 j , k ( u ) + q 2 j , k ( u ) L 2 , 3 i , j ( u ) = L 2 , 1 i , k ( u ) . -pq_1^{i,k}+\sum_{j=1}^{p}L_{2,2}^{i,j}(u)q_1^{j,k}(u)+q_2^{j,k}(u)L_{2,3}^{i,j}(u)=L_{2,1}^{i,k}(u). − p q 1 i , k + j = 1 ∑ p L 2 , 2 i , j ( u ) q 1 j , k ( u ) + q 2 j , k ( u ) L 2 , 3 i , j ( u ) = L 2 , 1 i , k ( u ) .
Applying the quotient map gives:
q 1 i , k ( u ) 1 ‾ = L 1 , 1 i , k ( u ) − 2 p δ i , k , q_1^{i,k}(u)\overline{1}=L_{1,1}^{i,k}(u)-2p\delta_{i,k}, q 1 i , k ( u ) 1 = L 1 , 1 i , k ( u ) − 2 p δ i , k ,
q 2 i , k ( u ) 1 ‾ + ∑ j = 1 p ( L 2 , 2 i , j ( u ) − p δ i , j ) q 1 j , k ( u ) 1 ‾ = L 2 , 1 i , k ( u ) . q_2^{i,k}(u)\overline{1}+\sum_{j=1}^{p}\left(L_{2,2}^{i,j}(u)-p\delta_{i,j}\right)q_1^{j,k}(u)\overline{1}=L_{2,1}^{i,k}(u). q 2 i , k ( u ) 1 + j = 1 ∑ p ( L 2 , 2 i , j ( u ) − p δ i , j ) q 1 j , k ( u ) 1 = L 2 , 1 i , k ( u ) .
Solving for q 1 ( u ) \mathbf{q}_1(u) q 1 ( u ) and q 2 ( u ) \mathbf{q}_2(u) q 2 ( u ) gives:
( q 1 ( u ) 1 ‾ q 2 ( u ) 1 ‾ ) = ( 1 p 0 L 2 , 2 ( u ) − p 1 p ) − 1 ( L 1 , 1 ( u ) L 2 , 1 ( u ) ) . \begin{pmatrix} \mathbf{q}_1(u)\overline{1} \\ \mathbf{q}_2(u)\overline{1} \end{pmatrix}
=
\begin{pmatrix} \mathbf{1}_p & 0 \\ L_{2,2}(u)-p & \mathbf{1}_p \end{pmatrix}^{-1}
\begin{pmatrix} L_{1,1}(u) \\ L_{2,1}(u) \end{pmatrix}. ( q 1 ( u ) 1 q 2 ( u ) 1 ) = ( 1 p L 2 , 2 ( u ) − p 0 1 p ) − 1 ( L 1 , 1 ( u ) L 2 , 1 ( u ) ) .
Substituting this expression into the quasideterminant formula yields
∣ L ( u ) ∣ 3 , 1 1 ‾ = ∣ ( L 1 , 1 ( u ) − 2 p 1 p 0 L 2 , 1 ( u ) L 2 , 2 ( u ) − p 1 p L 3 , 1 ( u ) L 3 , 2 ( u ) L 3 , 3 ( u ) ) ∣ 3 , 1 = rdet ( L 1 , 1 ( u ) − 2 p 1 p 0 L 2 , 1 ( u ) L 2 , 2 ( u ) − p 1 p L 3 , 1 ( u ) L 3 , 2 ( u ) L 3 , 3 ( u ) ) . \left|L(u)\right|_{3,1}\overline{1}
=
\left|
\begin{pmatrix}
L_{1,1}(u)-2p & \mathbf{1}_p & 0 \\
L_{2,1}(u) & L_{2,2}(u)-p & \mathbf{1}_p \\
L_{3,1}(u) & L_{3,2}(u) & L_{3,3}(u)
\end{pmatrix}
\right|_{3,1}
=
\operatorname{rdet}
\begin{pmatrix}
L_{1,1}(u)-2p & \mathbf{1}_p & 0 \\
L_{2,1}(u) & L_{2,2}(u)-p & \mathbf{1}_p \\
L_{3,1}(u) & L_{3,2}(u) & L_{3,3}(u)
\end{pmatrix}. ∣ L ( u ) ∣ 3 , 1 1 = L 1 , 1 ( u ) − 2 p L 2 , 1 ( u ) L 3 , 1 ( u ) 1 p L 2 , 2 ( u ) − p L 3 , 2 ( u ) 0 1 p L 3 , 3 ( u ) 3 , 1 = rdet L 1 , 1 ( u ) − 2 p L 2 , 1 ( u ) L 3 , 1 ( u ) 1 p L 2 , 2 ( u ) − p L 3 , 2 ( u ) 0 1 p L 3 , 3 ( u ) .
The final equality identifies this reduced quasideterminant with the Brundan–Kleshchev generating matrix.
Proof. Set
A ( u ) = ( L 1 , 2 ( u ) ⋯ L 1 , n ( u ) ⋮ ⋱ ⋮ L n − 1 , 2 ( u ) ⋯ L n − 1 , n ( u ) ) , c ( u ) = ( L 1 , 1 ( u ) ⋮ L n − 1 , 1 ( u ) ) . A(u)=
\begin{pmatrix}
\mathbf{L}_{1,2}(u) & \cdots& \mathbf{L}_{1,n}(u) \\
\vdots& \ddots& \vdots\\
\mathbf{L}_{n-1,2}(u) & \cdots& \mathbf{L}_{n-1,n}(u)
\end{pmatrix},
\qquad
\mathbf{c}(u)=
\begin{pmatrix}
L_{1,1}(u) \\
\vdots\\
L_{n-1,1}(u)
\end{pmatrix}. A ( u ) = L 1 , 2 ( u ) ⋮ L n − 1 , 2 ( u ) ⋯ ⋱ ⋯ L 1 , n ( u ) ⋮ L n − 1 , n ( u ) , c ( u ) = L 1 , 1 ( u ) ⋮ L n − 1 , 1 ( u ) .
Write G a , b ( u ) \mathbf{G}_{a,b}(u) G a , b ( u ) for blocks of A ( u ) − 1 \mathbf{A}(u)^{-1} A ( u ) − 1 .
Instead of reducing the whole inverse matrix, we directly reduce the column:
q ( u ) = A ( u ) − 1 c ( u ) , q ( u ) = ( q 1 ( u ) ⋮ q n − 1 ( u ) ) . (44) \mathbf{q}(u)=\mathbf{A}(u)^{-1}\mathbf{c}(u),\qquad
\mathbf{q}(u)=
\begin{pmatrix}
\mathbf{q}_{1}(u)\\
\vdots\\
\mathbf{q}_{n-1}(u)
\end{pmatrix}. \tag*{(44)} q ( u ) = A ( u ) − 1 c ( u ) , q ( u ) = q 1 ( u ) ⋮ q n − 1 ( u ) . (44)
Thus
T ( u ) 1 ‾ = ( L n , 1 ( u ) − ∑ b = 1 n − 1 L n , b + 1 ( u ) q b ( u ) ) 1 ‾ . (45) T(u)\overline{1}=\left(\mathbf{L}_{n,1}(u)-\sum_{b=1}^{n-1}\mathbf{L}_{n,b+1}(u)\mathbf{q}_{b}(u)\right)\overline{1}. \tag*{(45)} T ( u ) 1 = ( L n , 1 ( u ) − b = 1 ∑ n − 1 L n , b + 1 ( u ) q b ( u ) ) 1 . (45)
Let us establish the reduction step by step.
∑ j = 1 p [ L a , b + 1 i , j ( u ) , q b j , k ( u ) ] = − p q a − 1 i , k ( u ) + p δ a , 1 δ i , k . (46) \sum_{j=1}^{p}\left[\mathbf{L}_{a,b+1}^{i,j}(u),\mathbf{q}_{b}^{j,k}(u)\right]
=-p\mathbf{q}_{a-1}^{i,k}(u)+p\delta_{a,1}\delta_{i,k}. \tag*{(46)} j = 1 ∑ p [ L a , b + 1 i , j ( u ) , q b j , k ( u ) ] = − p q a − 1 i , k ( u ) + p δ a , 1 δ i , k . (46)
To verify this identity, write
q b j , k ( u ) = ∑ d = 1 n − 1 ∑ t = 1 p G b , d j , t ( u ) L d , 1 t , k ( u ) . \mathbf{q}_{b}^{j,k}(u)=\sum_{d=1}^{n-1}\sum_{t=1}^{p}\mathbf{G}_{b,d}^{j,t}(u)\mathbf{L}_{d,1}^{t,k}(u). q b j , k ( u ) = d = 1 ∑ n − 1 t = 1 ∑ p G b , d j , t ( u ) L d , 1 t , k ( u ) .
In components, the inverse-matrix commutator formula reads
[ L a , b + 1 i , j ( u ) , G r , d s , t ( u ) ] = δ a , d δ t , i G r , b + 1 s , j ( u ) − δ r , b δ s , j G a − 1 , d i , t ( u ) , \left[\mathbf{L}_{a,b+1}^{i,j}(u),\mathbf{G}_{r,d}^{s,t}(u)\right]
=\delta_{a,d}\delta_{t,i}\mathbf{G}_{r,b+1}^{s,j}(u)-\delta_{r,b}\delta_{s,j}\mathbf{G}_{a-1,d}^{i,t}(u), [ L a , b + 1 i , j ( u ) , G r , d s , t ( u ) ] = δ a , d δ t , i G r , b + 1 s , j ( u ) − δ r , b δ s , j G a − 1 , d i , t ( u ) ,
with the convention that blocks of G \mathbf{G} G outside the range 1 , … , n − 1 1,\ldots,n-1 1 , … , n − 1 are zero. Its contribution to (46) , after summing over j , d , t j,d,t j , d , t , is
tr ( G b , b + 1 ( u ) ) L a , 1 i , k ( u ) − p q a − 1 i , k ( u ) , \operatorname{tr}\left(\mathbf{G}_{b,b+1}(u)\right)\mathbf{L}_{a,1}^{i,k}(u)-p\mathbf{q}_{a-1}^{i,k}(u), tr ( G b , b + 1 ( u ) ) L a , 1 i , k ( u ) − p q a − 1 i , k ( u ) ,
where the first term is absent when b = n − 1 b=n-1 b = n − 1 . On the other hand, (12) gives
[ L a , b + 1 i , j ( u ) , L d , 1 t , k ( u ) ] = δ a , 1 δ i , k L d , b + 1 t , j ( u ) − δ d , b + 1 δ t , j L a , 1 i , k ( u ) . \left[\mathbf{L}_{a,b+1}^{i,j}(u),\mathbf{L}_{d,1}^{t,k}(u)\right]
=\delta_{a,1}\delta_{i,k}\mathbf{L}_{d,b+1}^{t,j}(u)-\delta_{d,b+1}\delta_{t,j}\mathbf{L}_{a,1}^{i,k}(u). [ L a , b + 1 i , j ( u ) , L d , 1 t , k ( u ) ] = δ a , 1 δ i , k L d , b + 1 t , j ( u ) − δ d , b + 1 δ t , j L a , 1 i , k ( u ) .
Using the inverse relation ∑ d G b , d i , k ( u ) L d , b + 1 k , j ( u ) = δ i , j \sum_{d}\mathbf{G}_{b,d}^{i,k}(u)\mathbf{L}_{d,b+1}^{k,j}(u)=\delta^{i,j} ∑ d G b , d i , k ( u ) L d , b + 1 k , j ( u ) = δ i , j , the contribution from this commutator is
p δ a , 1 δ i , k − tr ( G b , b + 1 ( u ) ) L a , 1 i , k ( u ) . p\delta_{a,1}\delta_{i,k}-\operatorname{tr}\left(\mathbf{G}_{b,b+1}(u)\right)\mathbf{L}_{a,1}^{i,k}(u). p δ a , 1 δ i , k − tr ( G b , b + 1 ( u ) ) L a , 1 i , k ( u ) .
The trace terms cancel, proving (46) .
∑ b = 1 n − 1 ∑ j = 1 p L a , b + 1 i , j ( u ) q b j , k ( u ) = L a , 1 i , k ( u ) . (47) \sum_{b=1}^{n-1}\sum_{j=1}^{p}\mathbf{L}_{a,b+1}^{i,j}(u)\mathbf{q}_{b}^{j,k}(u)=\mathbf{L}_{a,1}^{i,k}(u). \tag*{(47)} b = 1 ∑ n − 1 j = 1 ∑ p L a , b + 1 i , j ( u ) q b j , k ( u ) = L a , 1 i , k ( u ) . (47)
For b ≥ a b \ge a b ≥ a , move L a , b + 1 i , j L^{i,j}_{a,b+1} L a , b + 1 i , j to the right before applying the result to 1 ‾ \overline{1} 1 . The reduction relations give
L a , b + 1 i , j ( u ) 1 ‾ = δ a , b δ i , j 1 ‾ , b ≥ a , L^{i,j}_{a,b+1}(u)\overline{1}=\delta_{a,b}\delta_{i,j}\overline{1}, \qquad b\ge a, L a , b + 1 i , j ( u ) 1 = δ a , b δ i , j 1 , b ≥ a ,
while (46) gives the correction term. We now split the sum and commute the generators:
∑ b = 1 n − 1 ∑ j = 1 p L a , b + 1 i , j ( u ) q b j , k ( u ) 1 ‾ = ∑ j = 1 p [ ∑ b = a n − 1 L a , b + 1 i , j ( u ) q b j , k ( u ) + ∑ b = 1 a − 2 L a , b + 1 i , j ( u ) q b j , k ( u ) + L a , a i , j ( u ) q a − 1 j , k ( u ) ] 1 ‾ = ∑ j = 1 p ∑ b = a n − 1 δ a , b δ i , j q b j , k ( u ) 1 ‾ + ∑ j = 1 p ( ∑ b = 1 a − 2 L a , b + 1 i , j ( u ) q b j , k ( u ) + L a , a i , j ( u ) q a − 1 j , k ( u ) ) 1 ‾ − ∑ b = a n − 1 p ( q a − 1 i , k − δ a , 1 δ i , k ) 1 ‾ . \begin{aligned}
\sum_{b=1}^{n-1}\sum_{j=1}^{p}L^{i,j}_{a,b+1}(u)q^{j,k}_{b}(u)\overline{1}=\sum_{j=1}^{p}\left[\sum_{b=a}^{n-1}L^{i,j}_{a,b+1}(u)q^{j,k}_{b}(u)+\sum_{b=1}^{a-2}L^{i,j}_{a,b+1}(u)q^{j,k}_{b}(u)+L^{i,j}_{a,a}(u)q^{j,k}_{a-1}(u)\right]\overline{1} \\
&=\sum_{j=1}^{p}\sum_{b=a}^{n-1}\delta_{a,b}\delta^{i,j}q^{j,k}_{b}(u)\overline{1}+\sum_{j=1}^{p}\left(\sum_{b=1}^{a-2}L^{i,j}_{a,b+1}(u)q^{j,k}_{b}(u)+L^{i,j}_{a,a}(u)q^{j,k}_{a-1}(u)\right)\overline{1}-\sum_{b=a}^{n-1}p(q^{i,k}_{a-1}-\delta_{a,1}\delta^{i,k})\overline{1}.
\end{aligned} b = 1 ∑ n − 1 j = 1 ∑ p L a , b + 1 i , j ( u ) q b j , k ( u ) 1 = j = 1 ∑ p [ b = a ∑ n − 1 L a , b + 1 i , j ( u ) q b j , k ( u ) + b = 1 ∑ a − 2 L a , b + 1 i , j ( u ) q b j , k ( u ) + L a , a i , j ( u ) q a − 1 j , k ( u ) ] 1 = j = 1 ∑ p b = a ∑ n − 1 δ a , b δ i , j q b j , k ( u ) 1 + j = 1 ∑ p ( b = 1 ∑ a − 2 L a , b + 1 i , j ( u ) q b j , k ( u ) + L a , a i , j ( u ) q a − 1 j , k ( u ) ) 1 − b = a ∑ n − 1 p ( q a − 1 i , k − δ a , 1 δ i , k ) 1 .
Next, we sum over j j j and b b b in the first and third terms. Thus the defining relation (47) becomes
q a i , k 1 ‾ + ∑ j = 1 p ( ∑ b = 1 a − 2 L a , b + 1 i , j ( u ) q b j , k ( u ) + ( L a , a i , j ( u ) − p ( n − a ) δ i , j ) q a − 1 j , k ( u ) ) 1 ‾ + p ( n − 1 ) δ a , 1 δ i , k 1 ‾ = L a , 1 i , k 1 ‾ . q^{i,k}_{a}\overline{1}+\sum_{j=1}^{p}\left(\sum_{b=1}^{a-2}L^{i,j}_{a,b+1}(u)q^{j,k}_{b}(u)+\left(L^{i,j}_{a,a}(u)-p(n-a)\delta^{i,j}\right)q^{j,k}_{a-1}(u)\right)\overline{1}+p(n-1)\delta_{a,1}\delta^{i,k}\overline{1}=L^{i,k}_{a,1}\overline{1}. q a i , k 1 + j = 1 ∑ p ( b = 1 ∑ a − 2 L a , b + 1 i , j ( u ) q b j , k ( u ) + ( L a , a i , j ( u ) − p ( n − a ) δ i , j ) q a − 1 j , k ( u ) ) 1 + p ( n − 1 ) δ a , 1 δ i , k 1 = L a , 1 i , k 1 .
We can rewrite this as a matrix recurrence:
q a ( u ) 1 ‾ = ( L a , 1 ( u ) − ∑ b = 1 a − 2 L a , b + 1 ( u ) q b ( u ) − ( L a , a ( u ) − p ( n − a ) 1 p ) q a − 1 ( u ) − p ( n − 1 ) δ a , 1 1 p ) 1 ‾ . (48) \mathbf{q}_{a}(u)\overline{1}=\left(\mathbf{L}_{a,1}(u)-\sum_{b=1}^{a-2}\mathbf{L}_{a,b+1}(u)\mathbf{q}_{b}(u)-\left(\mathbf{L}_{a,a}(u)-p(n-a)\mathbf{1}_{p}\right)\mathbf{q}_{a-1}(u)-p(n-1)\delta_{a,1}\mathbf{1}_{p}\right)\overline{1}. \tag*{(48)} q a ( u ) 1 = ( L a , 1 ( u ) − b = 1 ∑ a − 2 L a , b + 1 ( u ) q b ( u ) − ( L a , a ( u ) − p ( n − a ) 1 p ) q a − 1 ( u ) − p ( n − 1 ) δ a , 1 1 p ) 1 . (48)
Here the sum and the term containing q a − 1 \mathbf{q}_{a-1} q a − 1 are omitted when a = 1 a=1 a = 1 .
( 1 p 0 ⋯ 0 L 2 , 2 ( u ) − ( n − 2 ) p 1 p 1 p ⋱ ⋮ ⋮ ⋱ ⋱ 0 L n − 1 , 2 ( u ) ⋯ L n − 1 , n − 1 ( u ) − p 1 p ) ( q 1 ( u ) 1 ‾ ⋮ ⋮ q n − 1 ( u ) 1 ‾ ) = ( L 1 , 1 ( u ) − ( n − 1 ) p L 2 , 1 ( u ) ⋮ L n − 1 , 1 ( u ) ) , \begin{pmatrix}
\mathbf{1}_{p} & 0 & \cdots& 0\\
\mathbf{L}_{2,2}(u)-(n-2)p\mathbf{1}_{p} & \mathbf{1}_{p} & \ddots& \vdots\\
\vdots& \ddots& \ddots& 0\\
\mathbf{L}_{n-1,2}(u) & \cdots& \mathbf{L}_{n-1,n-1}(u)-p & \mathbf{1}_{p}
\end{pmatrix}
\begin{pmatrix}
\mathbf{q}_{1}(u)\overline{1}\\
\vdots\\
\vdots\\
\mathbf{q}_{n-1}(u)\overline{1}
\end{pmatrix}
=
\begin{pmatrix}
\mathbf{L}_{1,1}(u)-(n-1)p\\
\mathbf{L}_{2,1}(u)\\
\vdots\\
\mathbf{L}_{n-1,1}(u)
\end{pmatrix}, 1 p L 2 , 2 ( u ) − ( n − 2 ) p 1 p ⋮ L n − 1 , 2 ( u ) 0 1 p ⋱ ⋯ ⋯ ⋱ ⋱ L n − 1 , n − 1 ( u ) − p 0 ⋮ 0 1 p q 1 ( u ) 1 ⋮ ⋮ q n − 1 ( u ) 1 = L 1 , 1 ( u ) − ( n − 1 ) p L 2 , 1 ( u ) ⋮ L n − 1 , 1 ( u ) ,
which uniquely determines the class
( q 1 ( u ) 1 ‾ ⋮ ⋮ q n − 1 ( u ) 1 ‾ ) = ( 1 p 0 ⋯ 0 L 2 , 2 ( u ) − ( n − 2 ) p 1 p ⋱ ⋮ ⋮ ⋱ ⋱ 0 L n − 1 , 2 ( u ) ⋯ L n − 1 , n − 1 ( u ) − p 1 p ) − 1 ( L 1 , 1 ( u ) − ( n − 1 ) p L 2 , 1 ( u ) ⋮ L n − 1 , 1 ( u ) ) . \begin{pmatrix}
\mathbf{q}_{1}(u)\overline{1}\\
\vdots\\
\vdots\\
\mathbf{q}_{n-1}(u)\overline{1}
\end{pmatrix}
=
\begin{pmatrix}
\mathbf{1}_{p} & 0 & \cdots& 0\\
\mathbf{L}_{2,2}(u)-(n-2)p & \mathbf{1}_{p} & \ddots& \vdots\\
\vdots& \ddots& \ddots& 0\\
\mathbf{L}_{n-1,2}(u) & \cdots& \mathbf{L}_{n-1,n-1}(u)-p & \mathbf{1}_{p}
\end{pmatrix}^{-1}
\begin{pmatrix}
\mathbf{L}_{1,1}(u)-(n-1)p\\
\mathbf{L}_{2,1}(u)\\
\vdots\\
\mathbf{L}_{n-1,1}(u)
\end{pmatrix}. q 1 ( u ) 1 ⋮ ⋮ q n − 1 ( u ) 1 = 1 p L 2 , 2 ( u ) − ( n − 2 ) p ⋮ L n − 1 , 2 ( u ) 0 1 p ⋱ ⋯ ⋯ ⋱ ⋱ L n − 1 , n − 1 ( u ) − p 0 ⋮ 0 1 p − 1 L 1 , 1 ( u ) − ( n − 1 ) p L 2 , 1 ( u ) ⋮ L n − 1 , 1 ( u ) .
Substituting this expression into (4.10) shows that the reduced quasideterminant is equal to the quasideterminant of L r e d ( u ) L^{\mathrm{red}}(u) L red ( u ) :
T ( u ) ⋅ 1 ˉ = ∣ L r e d ( u ) ∣ n , 1 = ( − 1 ) n − 1 rdet ( L r e d ( u ) ) , T(u)\cdot\bar{1}=\left|L^{\mathrm{red}}(u)\right|_{n,1}=(-1)^{n-1}\operatorname{rdet}\left(L^{\mathrm{red}}(u)\right), T ( u ) ⋅ 1 ˉ = L red ( u ) n , 1 = ( − 1 ) n − 1 rdet ( L red ( u ) ) ,
where the last equality follows from (22) .
Corollary 4.5.
The quotient map restricts to an algebra homomorphism on the invariant subalgebra:
A m n → ( A / I ψ ) m n . \mathcal{A}^{\mathfrak{m}^{n}}\to(\mathcal{A}/I_{\psi})^{\mathfrak{m}^{n}}. A m n → ( A / I ψ ) m n . The Brundan–Kleshchev matrix T ( u ) T(u) T ( u ) satisfies the RTT relation with the Yang matrix R ( u ) R(u) R ( u ) of size p 2 × p 2 p^{2}\times p^{2} p 2 × p 2 :
R ( u − v ) T 1 ( u ) T 2 ( v ) = T 2 ( v ) T 1 ( u ) R ( u − v ) . R(u-v)T_{1}(u)T_{2}(v)=T_{2}(v)T_{1}(u)R(u-v). R ( u − v ) T 1 ( u ) T 2 ( v ) = T 2 ( v ) T 1 ( u ) R ( u − v ) .
Proof.
By definition, multiplication in ( A / I ψ ) m n (\mathcal{A}/I_{\psi})^{\mathfrak{m}^{n}} ( A / I ψ ) m n is given by
x ˉ ⋅ y ˉ : = x y 1 ˉ , \bar{x}\cdot\bar{y}:=xy\bar{1}, x ˉ ⋅ y ˉ := x y 1 ˉ , where x , y ∈ A x,y\in\mathcal{A} x , y ∈ A represent the classes x ˉ \bar{x} x ˉ and y ˉ \bar{y} y ˉ and satisfy
[ x , m ] 1 ˉ = [ y , m ] 1 ˉ = 0. [x,m]\bar{1}=[y,m]\bar{1}=0. [ x , m ] 1 ˉ = [ y , m ] 1 ˉ = 0. These conditions ensure that the product is independent of the choice of representatives. Now let a , b ∈ A m n a,b\in\mathcal{A}^{\mathfrak{m}^{n}} a , b ∈ A m n . Their images a ˉ \bar{a} a ˉ and b ˉ \bar{b} b ˉ lie in ( A / I ψ ) m n (\mathcal{A}/I_{\psi})^{\mathfrak{m}^{n}} ( A / I ψ ) m n . By the definition of multiplication in this algebra, the quotient map preserves products.
By the first point, the quotient map restricts to an algebra homomorphism on the invariant subalgebra. Hence the reduced quasideterminant still satisfies the RTT relation. By Theorem above 4.4, it agrees with the Brundan–Kleshchev generating matrix up to the scalar factor ( − 1 ) n − 1 (-1)^{n-1} ( − 1 ) n − 1 , which does not affect the RTT relation. Therefore, T ( u ) T(u) T ( u ) satisfies the stated relation.
Differential realization and reduction
Differential operators on the lower unipotent group
Let N − ⊂ G L N N_{-}\subset\mathrm{GL}_{N} N − ⊂ GL N be the lower unipotent subgroup, and let n − \mathfrak{n}_{-} n − be its Lie algebra. In this section we use a single pair of indices for coordinates and right-shift operators. Write the coordinate matrix of N − N_{-} N − as
z = 1 N + ∑ 1 ≤ j < i ≤ N z i , j e i , j , C [ N − ] = C [ z i , j ] 1 ≤ j < i ≤ N . z=1_{N}+\sum_{1\leq j<i\leq N}z_{i,j}e_{i,j},\qquad\mathbb{C}[N_{-}]=\mathbb{C}[z_{i,j}]_{1\leq j<i\leq N}. z = 1 N + 1 ≤ j < i ≤ N ∑ z i , j e i , j , C [ N − ] = C [ z i , j ] 1 ≤ j < i ≤ N .
Definition 5.1. The algebra D ( N − ) \mathcal{D}(N_-) D ( N − ) of polynomial differential operators is generated by these coordinates and their partial derivatives, with relations
[ ∂ z i , j , z k , l ] = δ i , k δ j , l , [\partial_{z_{i,j}},z_{k,l}]=\delta_{i,k}\delta_{j,l}, [ ∂ z i , j , z k , l ] = δ i , k δ j , l , [ ∂ z i , j , ∂ z k , l ] = [ z i , j , z k , l ] = 0. [\partial_{z_{i,j}},\partial_{z_{k,l}}]=[z_{i,j},z_{k,l}]=0. [ ∂ z i , j , ∂ z k , l ] = [ z i , j , z k , l ] = 0.
We now recall the standard differential realization of g l N \mathfrak{gl}_N gl N . Let the representation space be a class of functions on N − N_- N − , or on an open subset of N − N_- N − , that is preserved by polynomial multiplication and differentiation. For x ∈ g l N x\in\mathfrak{gl}_N x ∈ gl N and small ε \varepsilon ε , the Gauss decomposition has the form
e − ε x z = z ′ ( ε ) b ( ε ) , z ′ ( ε ) ∈ N − , b ( ε ) ∈ B , e^{-\varepsilon x}z=z'(\varepsilon)b(\varepsilon),\qquad z'(\varepsilon)\in N_-,\qquad b(\varepsilon)\in B, e − ε x z = z ′ ( ε ) b ( ε ) , z ′ ( ε ) ∈ N − , b ( ε ) ∈ B ,
where B ⊂ G L N B\subset GL_N B ⊂ G L N is the upper triangular subgroup. Fix the diagonal matrices
λ = diag ( λ 1 , … , λ N ) ; ρ = diag ( N − 1 , N − 2 , … , 1 , 0 ) . \lambda=\operatorname{diag}(\lambda_1,\ldots,\lambda_N)\ ;\qquad\rho=\operatorname{diag}(N-1,N-2,\ldots,1,0). λ = diag ( λ 1 , … , λ N ) ; ρ = diag ( N − 1 , N − 2 , … , 1 , 0 ) .
For b ∈ B b\in B b ∈ B , define the twist
χ λ ( b ) = exp ( − tr ( ( λ + ρ ) log b ) ) . \chi_\lambda(b)=\exp\left(-\operatorname{tr}((\lambda+\rho)\log b)\right). χ λ ( b ) = exp ( − tr (( λ + ρ ) log b ) ) .
For generic λ \lambda λ , this expression depends on a choice of branch of the logarithm, but its infinitesimal action does not. We denote the realized operator corresponding to x x x by E x E_x E x . Define the left action by
E x Φ ( z ) = d d ε ( χ λ ( b ( ε ) ) Φ ( z ′ ( ε ) ) ) ∣ ε = 0 . (49) E_x\Phi(z)=\left.\frac{d}{d\varepsilon}\left(\chi_\lambda(b(\varepsilon))\Phi(z'(\varepsilon))\right)\right|_{\varepsilon=0}. \tag*{(49)} E x Φ ( z ) = d ε d ( χ λ ( b ( ε )) Φ ( z ′ ( ε )) ) ε = 0 . (49)
If x ∈ n − x\in\mathfrak{n}_- x ∈ n − , then b ( ε ) = 1 N b(\varepsilon)=1_N b ( ε ) = 1 N ; hence the corresponding operators E x E_x E x are independent of λ \lambda λ .
E x Φ ( z ) = d d ε Φ ( z − ε x z ) ∣ ε = 0 = − ∑ i > j ( x z ) i , j ∂ z i , j Φ ( z ) . (50) E_x\Phi(z)=\left.\frac{d}{d\varepsilon}\Phi(z-\varepsilon xz)\right|_{\varepsilon=0}
=-\sum_{i>j}(xz)_{i,j}\partial_{z_{i,j}}\Phi(z). \tag*{(50)} E x Φ ( z ) = d ε d Φ ( z − ε x z ) ε = 0 = − i > j ∑ ( x z ) i , j ∂ z i , j Φ ( z ) . (50)
The standard basis of g l N \mathfrak{gl}_N gl N consists of the matrix units e i , j e_{i,j} e i , j . It follows from definition (49) that the map is linear:
x = ∑ i , j x i , j e i , j ⟼ E x = ∑ i , j x i , j E e i , j . (51) x=\sum_{i,j}x_{i,j}e_{i,j}\longmapsto E_x=\sum_{i,j}x_{i,j}E_{e_{i,j}}. \tag*{(51)} x = i , j ∑ x i , j e i , j ⟼ E x = i , j ∑ x i , j E e i , j . (51)
For x = e i , j x=e_{i,j} x = e i , j we denote the corresponding operator by E i , j = E e i , j E_{i,j}=E_{e_{i,j}} E i , j = E e i , j . The generators naturally split into two classes. Those with i > j i>j i > j correspond to elements e i , j ∈ n − e_{i,j}\in\mathfrak{n}_- e i , j ∈ n − and do not depend on λ \lambda λ :
E i , j = − ∑ p > s ( e i , j z ) p , s ∂ z p , s = − ∂ z i , j − ∑ m = 1 j − 1 z j , m ∂ z i , m . (52) E_{i,j}=-\sum_{p>s}(e_{i,j}z)_{p,s}\partial_{z_{p,s}}=-\partial_{z_{i,j}}-\sum_{m=1}^{j-1}z_{j,m}\partial_{z_{i,m}}. \tag*{(52)} E i , j = − p > s ∑ ( e i , j z ) p , s ∂ z p , s = − ∂ z i , j − m = 1 ∑ j − 1 z j , m ∂ z i , m . (52)
The generators E i , j E_{i,j} E i , j with i ≤ j i\leq j i ≤ j , by contrast, depend on λ \lambda λ . Direct expressions for these generators analogous to (52) would be cumbersome. However, a simple formula expresses all of them in terms of the right-shift generators.
For x ∈ n − x \in\mathfrak{n}_{-} x ∈ n − , define the right-shift operator by
D x Φ ( z ) = d d ε Φ ( z e ε x ) ∣ ε = 0 = ∑ i > j ( z x ) i , j ∂ z i , j Φ ( z ) . (53) D_x\Phi(z)=\left.\frac{d}{d\varepsilon}\Phi\left(ze^{\varepsilon x}\right)\right|_{\varepsilon=0}=\sum_{i>j}(zx)_{i,j}\partial_{z_{i,j}}\Phi(z). \tag*{(53)} D x Φ ( z ) = d ε d Φ ( z e ε x ) ε = 0 = i > j ∑ ( z x ) i , j ∂ z i , j Φ ( z ) . (53)
Both (5.1) and (5.5) define polynomial differential operators. For x = e i , j x=e_{i,j} x = e i , j with i > j i>j i > j , we denote the corresponding right-shift operator by D i , j D_{i,j} D i , j . Any right-shift operator is a linear combination of these basis operators, just as in (5.3). The individual right-shift generators have the following explicit form:
D i , j = ∑ p > s ( z e i , j ) p , s ∂ z p , s = ∂ z i , j + ∑ k = i + 1 N z k , i ∂ z k , j , i > j . (54) D_{i,j}=\sum_{p>s}(ze_{i,j})_{p,s}\partial_{z_{p,s}}=\partial_{z_{i,j}}+\sum_{k=i+1}^{N}z_{k,i}\partial_{z_{k,j}},\qquad i>j. \tag*{(54)} D i , j = p > s ∑ ( z e i , j ) p , s ∂ z p , s = ∂ z i , j + k = i + 1 ∑ N z k , i ∂ z k , j , i > j . (54)
Corollary 5.2. The generators E i , j E_{i,j} E i , j obey the commutation relations (2.1), and the left action (5.1) extends to an algebra homomorphism
ρ λ : U ( g l N ) ⟶ D ( N − ) . \rho_{\lambda}:U(\mathfrak{gl}_{N})\longrightarrow D(N_{-}). ρ λ : U ( gl N ) ⟶ D ( N − ) .
Example. Consider the familiar case N = 2 N=2 N = 2 . Then N − N_{-} N − has one coordinate x = z 2 , 1 x=z_{2,1} x = z 2 , 1 . Take the matrix unit e 1 , 2 e_{1,2} e 1 , 2 and λ + ρ = diag ( λ 1 + 1 , λ 2 ) \lambda+\rho=\operatorname{diag}(\lambda_{1}+1,\lambda_{2}) λ + ρ = diag ( λ 1 + 1 , λ 2 ) . The Gauss decomposition is
( 1 − ε 0 1 ) ( 1 0 x 1 ) = ( 1 0 x 1 − ε x 1 ) ( 1 − ε x − ε 0 1 1 − ε x ) . \begin{pmatrix}1&-\varepsilon\\0&1\end{pmatrix}
\begin{pmatrix}1&0\\x&1\end{pmatrix}
=
\begin{pmatrix}1&0\\\frac{x}{1-\varepsilon x}&1\end{pmatrix}
\begin{pmatrix}1-\varepsilon x&-\varepsilon\\0&\frac{1}{1-\varepsilon x}\end{pmatrix}. ( 1 0 − ε 1 ) ( 1 x 0 1 ) = ( 1 1 − ε x x 0 1 ) ( 1 − ε x 0 − ε 1 − ε x 1 ) .
Thus
χ λ ( b ( ε ) ) = exp ( ( λ 2 − λ 1 − 1 ) log ( 1 − ε x ) ) , \chi_{\lambda}(b(\varepsilon))=\exp\left((\lambda_{2}-\lambda_{1}-1)\log(1-\varepsilon x)\right), χ λ ( b ( ε )) = exp ( ( λ 2 − λ 1 − 1 ) log ( 1 − ε x ) ) ,
and (5.1) gives
E 1 , 2 Φ ( x ) = d d ε [ Φ ( x 1 − ε x ) e ( λ 2 − λ 1 − 1 ) log ( 1 − ε x ) ] ∣ ε = 0 = ( x 2 ∂ x + ( λ 1 − λ 2 + 1 ) x ) Φ ( x ) . E_{1,2}\Phi(x)=\left.\frac{d}{d\varepsilon}\left[\Phi\left(\frac{x}{1-\varepsilon x}\right)e^{(\lambda_{2}-\lambda_{1}-1)\log(1-\varepsilon x)}\right]\right|_{\varepsilon=0}=(x^{2}\partial_{x}+(\lambda_{1}-\lambda_{2}+1)x)\Phi(x). E 1 , 2 Φ ( x ) = d ε d [ Φ ( 1 − ε x x ) e ( λ 2 − λ 1 − 1 ) l o g ( 1 − ε x ) ] ε = 0 = ( x 2 ∂ x + ( λ 1 − λ 2 + 1 ) x ) Φ ( x ) .
Similarly,
E 2 , 1 = − ∂ x , E 1 , 1 = x ∂ x + λ 1 + 1 , E 2 , 2 = − x ∂ x + λ 2 . E_{2,1}=-\partial_{x},\qquad E_{1,1}=x\partial_{x}+\lambda_{1}+1,\qquad E_{2,2}=-x\partial_{x}+\lambda_{2}. E 2 , 1 = − ∂ x , E 1 , 1 = x ∂ x + λ 1 + 1 , E 2 , 2 = − x ∂ x + λ 2 .
The right-shift generator in this case is D 2 , 1 = ∂ x D_{2,1}=\partial_{x} D 2 , 1 = ∂ x .
Corollary 5.3. For all x , y ∈ n − x,y\in\mathfrak{n}_{-} x , y ∈ n − , the left- and right-shift operators commute:
[ E x , D y ] = 0. (55) [E_x,D_y]=0. \tag*{(55)} [ E x , D y ] = 0. (55)
Proof. It is easy to check that
E x D y Φ ( z ) = D y E x Φ ( z ) = d 2 d ε 1 d ε 2 Φ ( e − ε 1 x z e ε 2 y ) ∣ ε 1 = ε 2 = 0 . E_xD_y\Phi(z)=D_yE_x\Phi(z)=\left.\frac{d^2}{d\varepsilon_1d\varepsilon_2}\Phi\left(e^{-\varepsilon_1x}ze^{\varepsilon_2y}\right)\right|_{\varepsilon_1=\varepsilon_2=0}. E x D y Φ ( z ) = D y E x Φ ( z ) = d ε 1 d ε 2 d 2 Φ ( e − ε 1 x z e ε 2 y ) ε 1 = ε 2 = 0 .
We now give a compact formula for the realized universal matrix E E E . Assemble the right-shift operators into the strictly upper-triangular matrix
D = ∑ 1 ≤ j < i ≤ N e j , i ⊗ D i , j . (56) D = \sum_{1 \le j < i \le N} e_{j,i} \otimes D_{i,j}. \tag*{(56)} D = 1 ≤ j < i ≤ N ∑ e j , i ⊗ D i , j . (56)
Proposition 5.4 (Derkachov–Manashov [13 , 14 ] ). The matrix of left-action generators satisfies the factorized formula
E = z ( λ − D ) z − 1 . (57) E = z(\lambda- D)z^{-1}. \tag*{(57)} E = z ( λ − D ) z − 1 . (57)
Equivalently, the image of the L L L -operator is
L ( u ) = z ( u + λ − D ) z − 1 . (58) L(u) = z(u + \lambda- D)z^{-1}. \tag*{(58)} L ( u ) = z ( u + λ − D ) z − 1 . (58)
Remark 5.5. We would like to emphasize that this formula can be interpreted as a quantization of the Springer map [3 ] .
Remark 5.6. It is instructive to present everything in the visual matrix form
E = ( 1 0 0 ⋯ 0 z 2 , 1 1 0 ⋯ 0 z 3 , 1 z 3 , 2 1 ⋯ 0 ⋮ ⋮ ⋮ ⋱ ⋮ z N , 1 z N , 2 z N , 3 ⋯ 1 ) ( λ 1 − D 2 , 1 − D 3 , 1 ⋯ − D N , 1 0 λ 2 − D 3 , 2 ⋯ − D N , 2 0 0 λ 3 ⋯ − D N , 3 ⋮ ⋮ ⋮ ⋱ ⋮ 0 0 0 ⋯ λ N ) ( 1 0 0 ⋯ 0 z 2 , 1 1 0 ⋯ 0 z 3 , 1 z 3 , 2 1 ⋯ 0 ⋮ ⋮ ⋮ ⋱ ⋮ z N , 1 z N , 2 z N , 3 ⋯ 1 ) − 1 . E =
\begin{pmatrix}
1 & 0 & 0 & \cdots& 0 \\
z_{2,1} & 1 & 0 & \cdots& 0 \\
z_{3,1} & z_{3,2} & 1 & \cdots& 0 \\
\vdots& \vdots& \vdots& \ddots& \vdots\\
z_{N,1} & z_{N,2} & z_{N,3} & \cdots& 1
\end{pmatrix}
\begin{pmatrix}
\lambda_1 & -D_{2,1} & -D_{3,1} & \cdots& -D_{N,1} \\
0 & \lambda_2 & -D_{3,2} & \cdots& -D_{N,2} \\
0 & 0 & \lambda_3 & \cdots& -D_{N,3} \\
\vdots& \vdots& \vdots& \ddots& \vdots\\
0 & 0 & 0 & \cdots& \lambda_N
\end{pmatrix}
\begin{pmatrix}
1 & 0 & 0 & \cdots& 0 \\
z_{2,1} & 1 & 0 & \cdots& 0 \\
z_{3,1} & z_{3,2} & 1 & \cdots& 0 \\
\vdots& \vdots& \vdots& \ddots& \vdots\\
z_{N,1} & z_{N,2} & z_{N,3} & \cdots& 1
\end{pmatrix}^{-1}. E = 1 z 2 , 1 z 3 , 1 ⋮ z N , 1 0 1 z 3 , 2 ⋮ z N , 2 0 0 1 ⋮ z N , 3 ⋯ ⋯ ⋯ ⋱ ⋯ 0 0 0 ⋮ 1 λ 1 0 0 ⋮ 0 − D 2 , 1 λ 2 0 ⋮ 0 − D 3 , 1 − D 3 , 2 λ 3 ⋮ 0 ⋯ ⋯ ⋯ ⋱ ⋯ − D N , 1 − D N , 2 − D N , 3 ⋮ λ N 1 z 2 , 1 z 3 , 1 ⋮ z N , 1 0 1 z 3 , 2 ⋮ z N , 2 0 0 1 ⋮ z N , 3 ⋯ ⋯ ⋯ ⋱ ⋯ 0 0 0 ⋮ 1 − 1 .
Proof. Let g l N = n − ⊕ b \mathfrak{gl}_N = \mathfrak{n}_{-} \oplus\mathfrak{b} gl N = n − ⊕ b , where b \mathfrak{b} b is the upper triangular subalgebra. For x ∈ g l N x \in\mathfrak{gl}_N x ∈ gl N , write
x = x − + x + , x − ∈ n − , x + ∈ b , x = x_{-} + x_{+}, \qquad x_{-} \in\mathfrak{n}_{-}, \qquad x_{+} \in\mathfrak{b}, x = x − + x + , x − ∈ n − , x + ∈ b ,
We first consider an element e ε a = 1 N + ε a + O ( ε 2 ) e^{\varepsilon a} = 1_N + \varepsilon a + O(\varepsilon^2) e ε a = 1 N + ε a + O ( ε 2 ) in a small neighborhood of the identity. Put x = z − 1 a z x = z^{-1}az x = z − 1 a z . To first order in ε \varepsilon ε , the Gauss decomposition used in (49) takes the form
z ′ ( ε ) = z ( 1 N − ε x − ) + O ( ε 2 ) , z'(\varepsilon) = z(1_N - \varepsilon x_{-}) + O(\varepsilon^2), z ′ ( ε ) = z ( 1 N − ε x − ) + O ( ε 2 ) ,
b ( ε ) = 1 N − ε x + + O ( ε 2 ) , b(\varepsilon) = 1_N - \varepsilon x_{+} + O(\varepsilon^2), b ( ε ) = 1 N − ε x + + O ( ε 2 ) ,
which could be checked directly:
e − ε a z = ( 1 N − ε a ) z + O ( ε 2 ) = z ( 1 − ε z − 1 a z ) + O ( ε 2 ) e^{-\varepsilon a}z = (1_N - \varepsilon a)z + O(\varepsilon^2) = z(1 - \varepsilon z^{-1}az) + O(\varepsilon^2) e − ε a z = ( 1 N − ε a ) z + O ( ε 2 ) = z ( 1 − ε z − 1 a z ) + O ( ε 2 )
z ′ ( ε ) b ( ε ) = z ( 1 N − ε x − ) ( 1 N − ε x + ) + O ( ε 2 ) = z ( 1 − ε ( x − + x + ) ) + O ( ε 2 ) z'(\varepsilon)b(\varepsilon) = z(1_N - \varepsilon x_{-})(1_N - \varepsilon x_{+}) + O(\varepsilon^2) = z(1 - \varepsilon(x_{-} + x_{+})) + O(\varepsilon^2) z ′ ( ε ) b ( ε ) = z ( 1 N − ε x − ) ( 1 N − ε x + ) + O ( ε 2 ) = z ( 1 − ε ( x − + x + )) + O ( ε 2 )
Therefore the infinitesimal action can be written as
E a = ∑ k = 1 N ( λ k + ρ k ) ( x + ) k , k − D x − , x = z − 1 a z . (59) E_a = \sum_{k=1}^{N}(\lambda_k + \rho_k)(x_+)_{k,k} - D_{x_-}, \qquad x = z^{-1}az. \tag*{(59)} E a = k = 1 ∑ N ( λ k + ρ k ) ( x + ) k , k − D x − , x = z − 1 a z . (59)
Here λ + ρ = diag ( λ 1 + ρ 1 , … , λ N + ρ N ) \lambda+\rho=\operatorname{diag}(\lambda_1+\rho_1,\ldots,\lambda_N+\rho_N) λ + ρ = diag ( λ 1 + ρ 1 , … , λ N + ρ N ) . For a = e i , j a=e_{i,j} a = e i , j we have explicit expressions
( x − ) k , l = z j , l ( z − 1 ) k , i ; k > l ; ( x + ) k , k = z j , k ( z − 1 ) k , i (60) (x_-)_{k,l}=z_{j,l}\left(z^{-1}\right)_{k,i}\ ;\quad k>l\ ;\quad(x_+)_{k,k}=z_{j,k}\left(z^{-1}\right)_{k,i} \tag*{(60)} ( x − ) k , l = z j , l ( z − 1 ) k , i ; k > l ; ( x + ) k , k = z j , k ( z − 1 ) k , i (60)
so that
E i , j = ∑ k = 1 N z j , k ( λ k + ρ k ) ( z − 1 ) k , i − ∑ k > l z j , l ( z − 1 ) k , i D k , l = E_{i,j}=\sum_{k=1}^{N}z_{j,k}(\lambda_k+\rho_k)\left(z^{-1}\right)_{k,i}-\sum_{k>l}z_{j,l}\left(z^{-1}\right)_{k,i}D_{k,l}= E i , j = k = 1 ∑ N z j , k ( λ k + ρ k ) ( z − 1 ) k , i − k > l ∑ z j , l ( z − 1 ) k , i D k , l =
( z ( λ + ρ ) z − 1 ) j , i − ∑ k > l z j , l D k , l ( z − 1 ) k , i + ∑ k > l z j , l [ D k , l , ( z − 1 ) k , i ] \left(z(\lambda+\rho)z^{-1}\right)_{j,i}-\sum_{k>l}z_{j,l}D_{k,l}\left(z^{-1}\right)_{k,i}+\sum_{k>l}z_{j,l}\left[D_{k,l},\left(z^{-1}\right)_{k,i}\right] ( z ( λ + ρ ) z − 1 ) j , i − k > l ∑ z j , l D k , l ( z − 1 ) k , i + k > l ∑ z j , l [ D k , l , ( z − 1 ) k , i ]
It remains to calculate the commutator. From the definition of the right action, we have
D x z = z D x + x z → z − 1 D x = ( D x + x ) z − 1 → [ D x , z − 1 ] = − x z − 1 D_xz=zD_x+xz\ \to\ z^{-1}D_x=(D_x+x)z^{-1}\ \to\ [D_x,z^{-1}]=-xz^{-1} D x z = z D x + x z → z − 1 D x = ( D x + x ) z − 1 → [ D x , z − 1 ] = − x z − 1
or in explicit index form
[ D k , l , ( z − 1 ) j , i ] = − ( e k , l z − 1 ) j , i = − δ k , j ( z − 1 ) l , i \left[D_{k,l},\left(z^{-1}\right)_{j,i}\right]=-\left(e_{k,l}z^{-1}\right)_{j,i}=-\delta_{k,j}\left(z^{-1}\right)_{l,i} [ D k , l , ( z − 1 ) j , i ] = − ( e k , l z − 1 ) j , i = − δ k , j ( z − 1 ) l , i
and consequently
∑ k > l z j , l [ D k , l , ( z − 1 ) k , i ] = − ∑ l ( N − l ) z j , l ( z − 1 ) l , i = − ( z ρ z − 1 ) j , i \sum_{k>l}z_{j,l}\left[D_{k,l},\left(z^{-1}\right)_{k,i}\right]=-\sum_l(N-l)z_{j,l}\left(z^{-1}\right)_{l,i}=-\left(z\rho z^{-1}\right)_{j,i} k > l ∑ z j , l [ D k , l , ( z − 1 ) k , i ] = − l ∑ ( N − l ) z j , l ( z − 1 ) l , i = − ( z ρ z − 1 ) j , i
Collecting everything together we obtain
E i , j = ( z λ z − 1 ) j , i − ∑ k > l z j , l D k , l ( z − 1 ) k , i (61) E_{i,j}=\left(z\lambda z^{-1}\right)_{j,i}-\sum_{k>l}z_{j,l}D_{k,l}\left(z^{-1}\right)_{k,i} \tag*{(61)} E i , j = ( z λ z − 1 ) j , i − k > l ∑ z j , l D k , l ( z − 1 ) k , i (61)
This formula is exactly the formula (57) in index notations.
Block notation for differential matrices
We use n × n n\times n n × n block matrices with blocks of size p × p p\times p p × p . Decompose the diagonal parameter matrix into blocks as
λ = diag ( λ 1 , … , λ n ) , λ a = diag ( λ ( a − 1 ) p + 1 , … , λ a p ) . \lambda=\operatorname{diag}(\lambda_1,\ldots,\lambda_n),\qquad\lambda_a=\operatorname{diag}(\lambda_{(a-1)p+1},\ldots,\lambda_{ap}). λ = diag ( λ 1 , … , λ n ) , λ a = diag ( λ ( a − 1 ) p + 1 , … , λ a p ) .
In the same block notation, the upper-triangular right-shift matrix is
D = ( D 1 D 1 , 2 D 1 , 3 ⋯ D 1 , n 0 D 2 D 2 , 3 ⋯ D 2 , n 0 0 D 3 ⋯ D 3 , n ⋮ ⋮ ⋮ ⋱ ⋮ 0 0 0 ⋯ D n ) D=\begin{pmatrix}
\mathbf{D}_1 & \mathbf{D}_{1,2} & \mathbf{D}_{1,3} & \cdots& \mathbf{D}_{1,n}\\
0 & \mathbf{D}_2 & \mathbf{D}_{2,3} & \cdots& \mathbf{D}_{2,n}\\
0 & 0 & \mathbf{D}_3 & \cdots& \mathbf{D}_{3,n}\\
\vdots& \vdots& \vdots& \ddots& \vdots\\
0 & 0 & 0 & \cdots& \mathbf{D}_n
\end{pmatrix} D = D 1 0 0 ⋮ 0 D 1 , 2 D 2 0 ⋮ 0 D 1 , 3 D 2 , 3 D 3 ⋮ 0 ⋯ ⋯ ⋯ ⋱ ⋯ D 1 , n D 2 , n D 3 , n ⋮ D n
where the off-diagonal blocks and diagonal blocks are given, respectively, by
D a , b = ∑ i , j = 1 p e j , i ⊗ D ( b − 1 ) p + i , ( a − 1 ) p + j , a < b , \mathbf{D}_{a,b}=\sum_{i,j=1}^{p}e_{j,i}\otimes D_{(b-1)p+i,(a-1)p+j},\qquad a<b, D a , b = i , j = 1 ∑ p e j , i ⊗ D ( b − 1 ) p + i , ( a − 1 ) p + j , a < b ,
D a = ∑ 1 ≤ j < i ≤ p e j , i ⊗ D ( a − 1 ) p + i , ( a − 1 ) p + j , 1 ≤ a ≤ n . \mathbf{D}_{a}=\sum_{1\leq j<i\leq p}e_{j,i}\otimes D_{(a-1)p+i,(a-1)p+j},\qquad1\leq a\leq n. D a = 1 ≤ j < i ≤ p ∑ e j , i ⊗ D ( a − 1 ) p + i , ( a − 1 ) p + j , 1 ≤ a ≤ n .
The coordinate matrix z z z has the block form
z = ( z 1 0 0 ⋯ 0 z 2 , 1 z 2 0 ⋯ 0 z 3 , 1 z 3 , 2 z 3 ⋯ 0 ⋮ ⋮ ⋮ ⋱ ⋮ z n , 1 z n , 2 z n , 3 ⋯ z n ) . z=\begin{pmatrix}
z_{1} & 0 & 0 & \cdots& 0\\
z_{2,1} & z_{2} & 0 & \cdots& 0\\
z_{3,1} & z_{3,2} & z_{3} & \cdots& 0\\
\vdots& \vdots& \vdots& \ddots& \vdots\\
z_{n,1} & z_{n,2} & z_{n,3} & \cdots& z_{n}
\end{pmatrix}. z = z 1 z 2 , 1 z 3 , 1 ⋮ z n , 1 0 z 2 z 3 , 2 ⋮ z n , 2 0 0 z 3 ⋮ z n , 3 ⋯ ⋯ ⋯ ⋱ ⋯ 0 0 0 ⋮ z n .
For a > b a>b a > b , the entries of the off-diagonal coordinate blocks are
( z a , b ) i , j = z ( a − 1 ) p + i , ( b − 1 ) p + j , 1 ≤ i , j ≤ p . (z_{a,b})_{i,j}=z_{(a-1)p+i,(b-1)p+j},\qquad1\leq i,j\leq p. ( z a , b ) i , j = z ( a − 1 ) p + i , ( b − 1 ) p + j , 1 ≤ i , j ≤ p .
Each diagonal block z a z_a z a is lower unitriangular, with entries
( z a ) i , j = z ( a − 1 ) p + i , ( a − 1 ) p + j . (z_a)_{i,j}=z_{(a-1)p+i,(a-1)p+j}. ( z a ) i , j = z ( a − 1 ) p + i , ( a − 1 ) p + j .
We also record useful block-factorization formulas in terms of the usual derivatives ∂ z i , j \partial_{z_{i,j}} ∂ z i , j . It is useful to introduce block matrices of derivatives
d a , b i , j = ∂ z j + p ( b − 1 ) , i + p ( a − 1 ) ; d a i , j = ( d a , a ) + i , j = { ∂ z j + p ( a − 1 ) , i + p ( a − 1 ) i < j , 0 i = j , 0 i > j . (62) d_{a,b}^{i,j}=\partial_{z_{j+p(b-1),i+p(a-1)}}\ ;\qquad d_a^{i,j}=(d_{a,a})_{+}^{i,j}=
\begin{cases}
\partial_{z_{j+p(a-1),i+p(a-1)}} & i<j,\\
0 & i=j,\\
0 & i>j.
\end{cases} \tag*{(62)} d a , b i , j = ∂ z j + p ( b − 1 ) , i + p ( a − 1 ) ; d a i , j = ( d a , a ) + i , j = ⎩ ⎨ ⎧ ∂ z j + p ( a − 1 ) , i + p ( a − 1 ) 0 0 i < j , i = j , i > j . (62)
Proposition 5.7. For a < b a<b a < b , the off-diagonal blocks D a , b \mathbf{D}_{a,b} D a , b of D D D and E a , b \mathbf{E}_{a,b} E a , b of the generator matrix (see (14) ) are expressed in terms of d a , b d_{a,b} d a , b and z a z_a z a as follows. The diagonal blocks of D D D are given by:
D a , b = d a , b z b + ∑ c = b + 1 n d a , c z c , b ; E a , b = − z a d a , b − ∑ c = 1 a − 1 z a , c d c , b (63) \mathbf{D}_{a,b}=d_{a,b}z_b+\sum_{c=b+1}^{n}d_{a,c}z_{c,b}\ ;\qquad\mathbf{E}_{a,b}=-z_a d_{a,b}-\sum_{c=1}^{a-1}z_{a,c}d_{c,b} \tag*{(63)} D a , b = d a , b z b + c = b + 1 ∑ n d a , c z c , b ; E a , b = − z a d a , b − c = 1 ∑ a − 1 z a , c d c , b (63) D a = ( d a z a ) + + ∑ c = a + 1 n ( d a , c z c , a ) + (64) \mathbf{D}_{a}=(d_a z_a)_{+}+\sum_{c=a+1}^{n}(d_{a,c}z_{c,a})_{+} \tag*{(64)} D a = ( d a z a ) + + c = a + 1 ∑ n ( d a , c z c , a ) + (64)
The proof is a direct check.
Remark 5.8. Note the following matrix relation
( ( D 1 D 1 , 2 D 1 , 3 ⋯ D 1 , n 0 D 2 D 2 , 3 ⋯ D 2 , n 0 0 D 3 ⋯ D 3 , n ⋮ ⋮ ⋮ ⋱ ⋮ 0 0 0 ⋯ D n ) ( z 1 0 0 ⋯ 0 z 2 , 1 z 2 0 ⋯ 0 z 3 , 1 z 3 , 2 z 3 ⋯ 0 ⋮ ⋮ ⋮ ⋱ ⋮ z n , 1 z n , 2 z n , 3 ⋯ z n ) − 1 ) = ( 0 d 1 , 2 d 1 , 3 ⋯ d 1 , n 0 0 d 2 , 3 ⋯ d 2 , n 0 0 0 ⋯ d 3 , n ⋮ ⋮ ⋮ ⋱ ⋮ 0 0 0 ⋯ 0 ) + \left(\begin{pmatrix}
\mathbf{D}_1 & \mathbf{D}_{1,2} & \mathbf{D}_{1,3} & \cdots& \mathbf{D}_{1,n}\\
0 & \mathbf{D}_2 & \mathbf{D}_{2,3} & \cdots& \mathbf{D}_{2,n}\\
0 & 0 & \mathbf{D}_3 & \cdots& \mathbf{D}_{3,n}\\
\vdots& \vdots& \vdots& \ddots& \vdots\\
0 & 0 & 0 & \cdots& \mathbf{D}_n
\end{pmatrix}
\begin{pmatrix}
z_1 & 0 & 0 & \cdots& 0\\
z_{2,1} & z_2 & 0 & \cdots& 0\\
z_{3,1} & z_{3,2} & z_3 & \cdots& 0\\
\vdots& \vdots& \vdots& \ddots& \vdots\\
z_{n,1} & z_{n,2} & z_{n,3} & \cdots& z_n
\end{pmatrix}^{-1}\right)
=
\begin{pmatrix}
0 & \mathbf{d}_{1,2} & \mathbf{d}_{1,3} & \cdots& \mathbf{d}_{1,n}\\
0 & 0 & \mathbf{d}_{2,3} & \cdots& \mathbf{d}_{2,n}\\
0 & 0 & 0 & \cdots& \mathbf{d}_{3,n}\\
\vdots& \vdots& \vdots& \ddots& \vdots\\
0 & 0 & 0 & \cdots& 0
\end{pmatrix}
+ D 1 0 0 ⋮ 0 D 1 , 2 D 2 0 ⋮ 0 D 1 , 3 D 2 , 3 D 3 ⋮ 0 ⋯ ⋯ ⋯ ⋱ ⋯ D 1 , n D 2 , n D 3 , n ⋮ D n z 1 z 2 , 1 z 3 , 1 ⋮ z n , 1 0 z 2 z 3 , 2 ⋮ z n , 2 0 0 z 3 ⋮ z n , 3 ⋯ ⋯ ⋯ ⋱ ⋯ 0 0 0 ⋮ z n − 1 = 0 0 0 ⋮ 0 d 1 , 2 0 0 ⋮ 0 d 1 , 3 d 2 , 3 0 ⋮ 0 ⋯ ⋯ ⋯ ⋱ ⋯ d 1 , n d 2 , n d 3 , n ⋮ 0 +
follows from the formulas in (63) –(64) . It expresses the strictly upper block-triangular part of the generator matrix E E E in the form
E + = − ( z 1 0 0 ⋯ 0 z 2 , 1 z 2 0 ⋯ 0 z 3 , 1 z 3 , 2 z 3 ⋯ 0 ⋮ ⋮ ⋮ ⋱ ⋮ z n , 1 z n , 2 z n , 3 ⋯ z n ) ( 0 d 1 , 2 d 1 , 3 ⋯ d 1 , n 0 0 d 2 , 3 ⋯ d 2 , n 0 0 0 ⋯ d 3 , n ⋮ ⋮ ⋮ ⋱ ⋮ 0 0 0 ⋯ 0 ) + (65) E_+=-
\begin{pmatrix}
z_1 & 0 & 0 & \cdots& 0\\
z_{2,1} & z_2 & 0 & \cdots& 0\\
z_{3,1} & z_{3,2} & z_3 & \cdots& 0\\
\vdots& \vdots& \vdots& \ddots& \vdots\\
z_{n,1} & z_{n,2} & z_{n,3} & \cdots& z_n
\end{pmatrix}
\begin{pmatrix}
0 & \mathbf{d}_{1,2} & \mathbf{d}_{1,3} & \cdots& \mathbf{d}_{1,n}\\
0 & 0 & \mathbf{d}_{2,3} & \cdots& \mathbf{d}_{2,n}\\
0 & 0 & 0 & \cdots& \mathbf{d}_{3,n}\\
\vdots& \vdots& \vdots& \ddots& \vdots\\
0 & 0 & 0 & \cdots& 0
\end{pmatrix}
+ \tag*{(65)} E + = − z 1 z 2 , 1 z 3 , 1 ⋮ z n , 1 0 z 2 z 3 , 2 ⋮ z n , 2 0 0 z 3 ⋮ z n , 3 ⋯ ⋯ ⋯ ⋱ ⋯ 0 0 0 ⋮ z n 0 0 0 ⋮ 0 d 1 , 2 0 0 ⋮ 0 d 1 , 3 d 2 , 3 0 ⋮ 0 ⋯ ⋯ ⋯ ⋱ ⋯ d 1 , n d 2 , n d 3 , n ⋮ 0 + (65)
which is equivalent to the formula for E a , b \mathbf{E}_{a,b} E a , b for a < b a<b a < b in (63) .
Quasideterminant with differential entries
Proposition 5.9. The corner quasideterminant of L ( u ) L(u) L ( u ) can be written entirely in terms of the diagonal coordinate blocks, the diagonal parameter blocks, and the right-shift blocks:
∣ L ( u ) ∣ n , 1 = ∣ ( ( u + λ 1 − D 1 ) z 1 − 1 − D 1 , 2 − D 1 , 3 ⋯ − D 1 , n 0 u + λ 2 − D 2 − D 2 , 3 ⋯ − D 2 , n 0 0 u + λ 3 − D 3 ⋯ − D 3 , n ⋮ ⋮ ⋮ ⋱ ⋮ 0 0 0 ⋯ z n ( u + λ n − D n ) ) ∣ n , 1 . (66) \lvert L(u)\rvert_{n,1}=
\left\lvert
\begin{pmatrix}
(u+\lambda_1-\mathbf{D}_1)z_1^{-1} & -\mathbf{D}_{1,2} & -\mathbf{D}_{1,3} & \cdots& -\mathbf{D}_{1,n}\\
0 & u+\lambda_2-\mathbf{D}_2 & -\mathbf{D}_{2,3} & \cdots& -\mathbf{D}_{2,n}\\
0 & 0 & u+\lambda_3-\mathbf{D}_3 & \cdots& -\mathbf{D}_{3,n}\\
\vdots& \vdots& \vdots& \ddots& \vdots\\
0 & 0 & 0 & \cdots& z_n(u+\lambda_n-\mathbf{D}_n)
\end{pmatrix}
\right\rvert_{n,1}. \tag*{(66)} ∣ L ( u ) ∣ n , 1 = ( u + λ 1 − D 1 ) z 1 − 1 0 0 ⋮ 0 − D 1 , 2 u + λ 2 − D 2 0 ⋮ 0 − D 1 , 3 − D 2 , 3 u + λ 3 − D 3 ⋮ 0 ⋯ ⋯ ⋯ ⋱ ⋯ − D 1 , n − D 2 , n − D 3 , n ⋮ z n ( u + λ n − D n ) n , 1 . (66)
Proof. The block entries of z z z give the following explicit factorization:
z = ( 1 p 0 0 ⋯ 0 z 2 , 1 z 1 − 1 1 p 0 ⋯ 0 z 3 , 1 z 1 − 1 z 3 , 2 z 2 − 1 1 p ⋯ 0 ⋮ ⋮ ⋮ ⋱ ⋮ z n , 1 z 1 − 1 z n , 2 z 2 − 1 z n , 3 z 3 − 1 ⋯ 1 p ) ⋅ ( z 1 0 0 ⋯ 0 0 z 2 0 ⋯ 0 0 0 z 3 ⋯ 0 ⋮ ⋮ ⋮ ⋱ ⋮ 0 0 0 ⋯ z n ) . z=
\begin{pmatrix}
1_p & 0 & 0 & \cdots& 0\\
z_{2,1}z_1^{-1} & 1_p & 0 & \cdots& 0\\
z_{3,1}z_1^{-1} & z_{3,2}z_2^{-1} & 1_p & \cdots& 0\\
\vdots& \vdots& \vdots& \ddots& \vdots\\
z_{n,1}z_1^{-1} & z_{n,2}z_2^{-1} & z_{n,3}z_3^{-1} & \cdots& 1_p
\end{pmatrix}
\cdot
\begin{pmatrix}
z_1 & 0 & 0 & \cdots& 0\\
0 & z_2 & 0 & \cdots& 0\\
0 & 0 & z_3 & \cdots& 0\\
\vdots& \vdots& \vdots& \ddots& \vdots\\
0 & 0 & 0 & \cdots& z_n
\end{pmatrix}. z = 1 p z 2 , 1 z 1 − 1 z 3 , 1 z 1 − 1 ⋮ z n , 1 z 1 − 1 0 1 p z 3 , 2 z 2 − 1 ⋮ z n , 2 z 2 − 1 0 0 1 p ⋮ z n , 3 z 3 − 1 ⋯ ⋯ ⋯ ⋱ ⋯ 0 0 0 ⋮ 1 p ⋅ z 1 0 0 ⋮ 0 0 z 2 0 ⋮ 0 0 0 z 3 ⋮ 0 ⋯ ⋯ ⋯ ⋱ ⋯ 0 0 0 ⋮ z n .
Substituting this factorization into L ( u ) = z ( u + λ − D ) z − 1 L(u)=z(u+\lambda-D)z^{-1} L ( u ) = z ( u + λ − D ) z − 1 , the first lower block-unitriangular factor and its inverse flank the conjugate of u + λ − D u+\lambda-D u + λ − D by the displayed diagonal block matrix. The first factor and its inverse have the forms allowed in (19) . Applying that invariance therefore removes these two factors, leaving the corner quasideterminant of the block-conjugated upper-triangular matrix displayed below:
∣ ( z 1 0 0 ⋯ 0 0 z 2 0 ⋯ 0 0 0 z 3 ⋯ 0 ⋮ ⋮ ⋮ ⋱ ⋮ 0 0 0 ⋯ z n ) ( u + λ − ( D 1 D 1 , 2 D 1 , 3 ⋯ D 1 , n 0 D 2 D 2 , 3 ⋯ D 2 , n 0 0 D 3 ⋯ D 3 , n ⋮ ⋮ ⋮ ⋱ ⋮ 0 0 0 ⋯ D n ) ) ( z 1 0 0 ⋯ 0 0 z 2 0 ⋯ 0 0 0 z 3 ⋯ 0 ⋮ ⋮ ⋮ ⋱ ⋮ 0 0 0 ⋯ z n ) − 1 ∣ n , 1 \left\lvert
\begin{pmatrix}
\mathbf z_1 & 0 & 0 & \cdots& 0 \\
0 & \mathbf z_2 & 0 & \cdots& 0 \\
0 & 0 & \mathbf z_3 & \cdots& 0 \\
\vdots& \vdots& \vdots& \ddots& \vdots\\
0 & 0 & 0 & \cdots& \mathbf z_n
\end{pmatrix}
\left(
u+\lambda-
\begin{pmatrix}
\mathbf D_1 & \mathbf D_{1,2} & \mathbf D_{1,3} & \cdots& \mathbf D_{1,n} \\
0 & \mathbf D_2 & \mathbf D_{2,3} & \cdots& \mathbf D_{2,n} \\
0 & 0 & \mathbf D_3 & \cdots& \mathbf D_{3,n} \\
\vdots& \vdots& \vdots& \ddots& \vdots\\
0 & 0 & 0 & \cdots& \mathbf D_n
\end{pmatrix}
\right)
\begin{pmatrix}
\mathbf z_1 & 0 & 0 & \cdots& 0 \\
0 & \mathbf z_2 & 0 & \cdots& 0 \\
0 & 0 & \mathbf z_3 & \cdots& 0 \\
\vdots& \vdots& \vdots& \ddots& \vdots\\
0 & 0 & 0 & \cdots& \mathbf z_n
\end{pmatrix}^{-1}
\right\rvert_{n,1} z 1 0 0 ⋮ 0 0 z 2 0 ⋮ 0 0 0 z 3 ⋮ 0 ⋯ ⋯ ⋯ ⋱ ⋯ 0 0 0 ⋮ z n u + λ − D 1 0 0 ⋮ 0 D 1 , 2 D 2 0 ⋮ 0 D 1 , 3 D 2 , 3 D 3 ⋮ 0 ⋯ ⋯ ⋯ ⋱ ⋯ D 1 , n D 2 , n D 3 , n ⋮ D n z 1 0 0 ⋮ 0 0 z 2 0 ⋮ 0 0 0 z 3 ⋮ 0 ⋯ ⋯ ⋯ ⋱ ⋯ 0 0 0 ⋮ z n − 1 n , 1
For the second application of (19) , use the following explicit left and right multipliers, respectively:
( z 1 0 0 ⋯ 0 0 z 2 0 ⋯ 0 0 0 z 3 ⋯ 0 ⋮ ⋮ ⋮ ⋱ ⋮ 0 0 0 ⋯ 1 p ) , ( 1 p 0 0 ⋯ 0 0 z 2 − 1 0 ⋯ 0 0 0 z 3 − 1 ⋯ 0 ⋮ ⋮ ⋮ ⋱ ⋮ 0 0 0 ⋯ z n − 1 ) . \begin{pmatrix}
\mathbf z_1 & 0 & 0 & \cdots& 0 \\
0 & \mathbf z_2 & 0 & \cdots& 0 \\
0 & 0 & \mathbf z_3 & \cdots& 0 \\
\vdots& \vdots& \vdots& \ddots& \vdots\\
0 & 0 & 0 & \cdots& 1_p
\end{pmatrix},
\qquad
\begin{pmatrix}
1_p & 0 & 0 & \cdots& 0 \\
0 & \mathbf z_2^{-1} & 0 & \cdots& 0 \\
0 & 0 & \mathbf z_3^{-1} & \cdots& 0 \\
\vdots& \vdots& \vdots& \ddots& \vdots\\
0 & 0 & 0 & \cdots& \mathbf z_n^{-1}
\end{pmatrix}. z 1 0 0 ⋮ 0 0 z 2 0 ⋮ 0 0 0 z 3 ⋮ 0 ⋯ ⋯ ⋯ ⋱ ⋯ 0 0 0 ⋮ 1 p , 1 p 0 0 ⋮ 0 0 z 2 − 1 0 ⋮ 0 0 0 z 3 − 1 ⋮ 0 ⋯ ⋯ ⋯ ⋱ ⋯ 0 0 0 ⋮ z n − 1 .
Their required principal blocks are invertible, and they have the forms prescribed in (19) . Factoring out the left matrix from the left, and right matrix from the right, we arrive to the formula (66) .
Reduction of the algebra of differential operators
Let us proceed with algebraic reduction of the differential realization. Let M n = exp ( m n ) M_n=\exp(\mathfrak m_n) M n = exp ( m n ) be the block-lower unipotent subgroup, let N d N_d N d be the block-diagonal subgroup of N − N_- N − , and let N − ′ ⊂ G L p N'_-\subset GL_p N − ′ ⊂ G L p be the lower unipotent subgroup. Every z ∈ N − z\in N_- z ∈ N − has a unique factorization
z = m z d , m ∈ M n , z d = diag ( z 1 , … , z n ) , z a ∈ N − ′ . (67) z=mz_d,\qquad m\in M_n,\qquad z_d=\operatorname{diag}(\mathbf z_1,\ldots,\mathbf z_n),\qquad\mathbf z_a\in N'_-. \tag*{(67)} z = m z d , m ∈ M n , z d = diag ( z 1 , … , z n ) , z a ∈ N − ′ . (67)
Indeed, the diagonal blocks of z z z determine z d z_d z d , after which
m = z z d − 1 . (68) m=zz_d^{-1}. \tag*{(68)} m = z z d − 1 . (68)
Thus multiplication gives an algebraic isomorphism M n × N d ≃ N − M_n\times N_d\simeq N_- M n × N d ≃ N − , with N d ≃ ( N − ′ ) n N_d\simeq(N'_-)^n N d ≃ ( N − ′ ) n . Use the character ψ \psi ψ from (37) and set
I : = D ( N − ) ⟨ x − ψ ( x ) ∣ x ∈ m n ⟩ , D Q = D ( N − ) / I . I:=\mathcal D(N_-)\langle x-\psi(x)\mid x\in\mathfrak m_n\rangle,\qquad\mathcal DQ=\mathcal D(N_-)/I. I := D ( N − ) ⟨ x − ψ ( x ) ∣ x ∈ m n ⟩ , D Q = D ( N − ) / I .
We define the reduced algebra by
D − = ( D Q ) m n . (69) \mathcal D_-=(\mathcal DQ)^{\mathfrak m_n}. \tag*{(69)} D − = ( D Q ) m n . (69)
i.e. the elements of the quotient D Q \mathcal DQ D Q that are invariant under the action of m n \mathfrak m_n m n .
Proposition 5.10. There is a natural algebra isomorphism
ξ : D − ⟶ D ( N − ′ ) ⊗ n . \xi:\mathcal D_-\longrightarrow\mathcal D(N'_-)^{\otimes n}. ξ : D − ⟶ D ( N − ′ ) ⊗ n .
The reduced algebra is generated by the entries of the diagonal coordinate blocks z a \mathbf z_a z a and by the classes of the right-shift operators D ( a − 1 ) p + i , ( a − 1 ) p + j D_{(a-1)p+i,(a-1)p+j} D ( a − 1 ) p + i , ( a − 1 ) p + j , i > j i>j i > j .
Proof. • First, we verify that the diagonal-block right-shift operators D ( a − 1 ) p + i , ( a − 1 ) p + j D_{(a-1)p+i,(a-1)p+j} D ( a − 1 ) p + i , ( a − 1 ) p + j and the diagonal coordinates z ( a − 1 ) p + i , ( a − 1 ) p + j z_{(a-1)p+i,(a-1)p+j} z ( a − 1 ) p + i , ( a − 1 ) p + j are invariant under m n \mathfrak{m}_n m n . By Corollary 5.3, the right-shift operators commute with the infinitesimal left action of m n \mathfrak{m}_n m n . The diagonal factor z d z_{\mathrm{d}} z d is invariant as well: left multiplication by M n M_n M n changes the factor m m m but leaves the diagonal blocks of z z z unchanged. In the coordinates ( m , z d ) (m,z_{\mathrm{d}}) ( m , z d ) , the action of a ∈ M n a \in M_n a ∈ M n is therefore
z ↦ a z ⇒ m ↦ a m , z d ↦ z d . z \mapsto az \Rightarrow m \mapsto am,\qquad z_{\mathrm{d}} \mapsto z_{\mathrm{d}}. z ↦ a z ⇒ m ↦ am , z d ↦ z d .
• To describe the quotient, we order the generators so that the diagonal-block right shifts precede the infinitesimal action generators from m n \mathfrak{m}_n m n . In this order, every element has a unique decomposition as a sum of terms of the form
P ( z ) D ( a 1 − 1 ) p + i 1 , ( a 1 − 1 ) p + j 1 ⋯ D ( a k − 1 ) p + i k , ( a k − 1 ) p + j k ⋅ E a k + 1 , b k + 1 i k + 1 , j k + 1 ⋯ E a k + s , b k + s i k + s , j k + s , P(z)D_{(a_1-1)p+i_1,(a_1-1)p+j_1}\cdots D_{(a_k-1)p+i_k,(a_k-1)p+j_k}\cdot E_{a_{k+1},b_{k+1}}^{i_{k+1},j_{k+1}}\cdots E_{a_{k+s},b_{k+s}}^{i_{k+s},j_{k+s}}, P ( z ) D ( a 1 − 1 ) p + i 1 , ( a 1 − 1 ) p + j 1 ⋯ D ( a k − 1 ) p + i k , ( a k − 1 ) p + j k ⋅ E a k + 1 , b k + 1 i k + 1 , j k + 1 ⋯ E a k + s , b k + s i k + s , j k + s ,
where P ( z ) ∈ C [ N − ] P(z) \in\mathbb{C}[N_-] P ( z ) ∈ C [ N − ] is a polynomial function and the E a r , b r i r , j r E_{a_r,b_r}^{i_r,j_r} E a r , b r i r , j r are the images of generators of m n \mathfrak{m}_n m n . Replacing each such generator by its shifted form E a r , b r i r , j r − ψ ( E a r , b r i r , j r ) E_{a_r,b_r}^{i_r,j_r}-\psi(E_{a_r,b_r}^{i_r,j_r}) E a r , b r i r , j r − ψ ( E a r , b r i r , j r ) separates terms with no shifted action generator from terms containing at least one. The latter lie in I I I , while uniqueness of the ordered decomposition makes the resulting sum direct:
D ( N − ) = C [ N − ] ⋅ D ⊕ I . \mathcal{D}(N_-)=\mathbb{C}[N_-]\cdot\mathcal{D}\oplus I. D ( N − ) = C [ N − ] ⋅ D ⊕ I .
where D \mathcal{D} D is generated by the diagonal-block right-shift operators D ( a − 1 ) p + i , ( a − 1 ) p + j D_{(a-1)p+i,(a-1)p+j} D ( a − 1 ) p + i , ( a − 1 ) p + j and the diagonal-block coordinates z ( a − 1 ) p + i , ( a − 1 ) p + j z_{(a-1)p+i,(a-1)p+j} z ( a − 1 ) p + i , ( a − 1 ) p + j . Consequently,
D Q ≅ C [ N − ] ⋅ D . \mathcal{D}Q\cong\mathbb{C}[N_-]\cdot\mathcal{D}. D Q ≅ C [ N − ] ⋅ D .
This is an isomorphism of vector spaces. In the factorization z = m z d z=mz_{\mathrm{d}} z = m z d , the left action of M n M_n M n acts on m m m and fixes z d z_{\mathrm{d}} z d ; hence invariant coefficient functions depend only on z d z_{\mathrm{d}} z d . Therefore, using
C [ N − ] m n ≅ C [ N d ] \mathbb{C}[N_-]^{\mathfrak{m}_n}\cong\mathbb{C}[N_{\mathrm{d}}] C [ N − ] m n ≅ C [ N d ]
we identify the reduction D − \mathcal{D}_- D − with the subalgebra D \mathcal{D} D of D ( N − ) \mathcal{D}(N_-) D ( N − ) generated by the block-diagonal right shifts D ( a − 1 ) p + i , ( b − 1 ) p + j D_{(a-1)p+i,(b-1)p+j} D ( a − 1 ) p + i , ( b − 1 ) p + j and the block-diagonal coordinates z ( a − 1 ) p + i , ( b − 1 ) p + j z_{(a-1)p+i,(b-1)p+j} z ( a − 1 ) p + i , ( b − 1 ) p + j .
• Finally, we identify D \mathcal{D} D with D ( N ′ ) ⊗ n \mathcal{D}(N')^{\otimes n} D ( N ′ ) ⊗ n . We note that the previous point can be understood as follows: for any element of the reduced algebra D − \mathcal{D}_- D − , we can find a unique representative in the algebra D \mathcal{D} D . In particular, the elements E a + 1 , a i , j E_{a+1,a}^{i,j} E a + 1 , a i , j have representative δ i , j \delta_{i,j} δ i , j in D \mathcal{D} D .
The latter algebra D \mathcal{D} D can be identified with differential algebra in diagonal variables D ( N d ) \mathcal{D}(N_{\mathrm{d}}) D ( N d ) by considering the action of the diagonal-block right shifts on functions of the diagonal-block coordinates. Their expressions on N − N_- N − may also contain derivatives in off-diagonal coordinates, but these derivatives vanish on such functions.
Thus, their restrictions give the corresponding right shifts on N d N_{\mathrm{d}} N d . Conversely, any differential operator on N d N_{\mathrm{d}} N d extends to a differential operator on N − N_- N − by expressing it in the basis of right shifts and extending those shifts via formula (64) . The polynomial prefactor of an operator of the form P ( z d ) D i 1 , j 1 ⋯ D i n , j n P(z_d)D_{i_1,j_1}\cdots D_{i_n,j_n} P ( z d ) D i 1 , j 1 ⋯ D i n , j n is unchanged under this extension. Thus, we define mutually inverse homomorphisms.
Thus, we have constructed a map from D − \mathcal{D}_{-} D − to D ( N d ) \mathcal{D}(N_d) D ( N d ) in two steps — firstly by mapping to D \mathcal{D} D via finding a good representative, and then setting to zero all the off-diagonal variables in the operators D i , j D_{i,j} D i , j .
Since N d ≃ ( N − ′ ) ⊗ n N_d \simeq(N^{\prime}_{-})^{\otimes n} N d ≃ ( N − ′ ) ⊗ n , this identification extends naturally to the differential algebras:
ξ : D − ⟶ D ( N − ′ ) ⊗ n \xi:\mathcal{D}_{-}\longrightarrow\mathcal{D}(N^{\prime}_{-})^{\otimes n} ξ : D − ⟶ D ( N − ′ ) ⊗ n
z a ⟼ z [ a ] , D ( a − 1 ) p + i , ( a − 1 ) p + j ⟼ D i , j , [ a ] = ∂ z i , j , [ a ] + ∑ k > i p z k , i , [ a ] ∂ z k , j , [ a ] . z_a\longmapsto z_{[a]},\qquad D_{(a-1)p+i,(a-1)p+j}\longmapsto D_{i,j,[a]}=\partial_{z_{i,j,[a]}}+\sum_{k>i}^{p}z_{k,i,[a]}\partial_{z_{k,j,[a]}}. z a ⟼ z [ a ] , D ( a − 1 ) p + i , ( a − 1 ) p + j ⟼ D i , j , [ a ] = ∂ z i , j , [ a ] + k > i ∑ p z k , i , [ a ] ∂ z k , j , [ a ] .
where the square-bracketed subscripts label the tensor factors of the D ( N − ′ ) ⊗ n \mathcal{D}(N^{\prime}_{-})^{\otimes n} D ( N − ′ ) ⊗ n .
The reduction of ρ λ \rho_\lambda ρ λ therefore defines the homomorphism
ρ λ r e d : W n ( g l p ) ⟶ D − . \rho^{\mathrm{red}}_{\lambda}:W_n(\mathfrak{gl}_p)\longrightarrow\mathcal{D}_{-}. ρ λ red : W n ( gl p ) ⟶ D − .
Composing with κ \kappa κ identifies its target with D ( N − ′ ) ⊗ n \mathcal{D}(N^{\prime}_{-})^{\otimes n} D ( N − ′ ) ⊗ n . However, the choice of the nontrivial character ψ \psi ψ imposes nontrivial constraints on the off-diagonal right-shift generators. We express the reduction constraints x − ψ ( x ) x-\psi(x) x − ψ ( x ) in terms of the generators D i , j D_{i,j} D i , j and z i , j z_{i,j} z i , j .
Proposition 5.12. For a < b a<b a < b , the generators D ( b − 1 ) p + i , ( a − 1 ) p + j D_{(b-1)p+i,(a-1)p+j} D ( b − 1 ) p + i , ( a − 1 ) p + j satisfy
D ( b − 1 ) p + i , ( a − 1 ) p + j 1 = − δ b , a + 1 ( z a − 1 z b ) j , i . (72) D_{(b-1)p+i,(a-1)p+j}\mathbf{1}=-\delta_{b,a+1}(z_a^{-1}z_b)_{j,i}. \tag*{(72)} D ( b − 1 ) p + i , ( a − 1 ) p + j 1 = − δ b , a + 1 ( z a − 1 z b ) j , i . (72) In the block notation, this is equivalent to
D a , b 1 = − δ b , a + 1 z a − 1 z b , a < b . (73) \mathbf{D}_{a,b}\mathbf{1}=-\delta_{b,a+1}z_a^{-1}z_b,\qquad a<b. \tag*{(73)} D a , b 1 = − δ b , a + 1 z a − 1 z b , a < b . (73)
Proof. As before, denote by ( ) + (\ )_{+} ( ) + the projection onto the strictly upper block-triangular part. The formulas (64) , (65) imply the relation
( D z − 1 ) + = − ( z − 1 E + ) + , (Dz^{-1})_{+}=-(z^{-1}E_{+})_{+}, ( D z − 1 ) + = − ( z − 1 E + ) + ,
where E + = ( E ) + E_{+}=(E)_{+} E + = ( E ) + is the strictly upper block-triangular part of the generator matrix. We now apply this relation to 1 ‾ \overline{1} 1 and obtain
( D z − 1 ) + 1 ‾ = − ( z − 1 ( 0 1 p 0 ⋯ 0 0 0 1 p ⋱ ⋮ ⋮ ⋱ ⋱ ⋱ 0 0 ⋯ ⋯ 0 1 p 0 ⋯ ⋯ 0 0 ) ) + 1 ‾ , (Dz^{-1})_{+}\overline{1}=-\left(z^{-1}
\begin{pmatrix}
0 & \mathbf{1}_{p} & 0 & \cdots& 0\\
0 & 0 & \mathbf{1}_{p} & \ddots& \vdots\\
\vdots& \ddots& \ddots& \ddots& 0\\
0 & \cdots& \cdots& 0 & \mathbf{1}_{p}\\
0 & \cdots& \cdots& 0 & 0
\end{pmatrix}
\right)_{+}\overline{1}, ( D z − 1 ) + 1 = − z − 1 0 0 ⋮ 0 0 1 p 0 ⋱ ⋯ ⋯ 0 1 p ⋱ ⋯ ⋯ ⋯ ⋱ ⋱ 0 0 0 ⋮ 0 1 p 0 + 1 ,
and solve the resulting relation iteratively to obtain (73) .
However, all the abstract relations and identifications (such as the map ξ \xi ξ ) can be understood directly in terms of manipulations with the functional spaces — similarly to the construction of the map ρ λ \rho_{\lambda} ρ λ . We explain this point of view on the reduction of the differential algebra in Appendix § 9 .
Reduction of the differential quasideterminant
Assume that we have chosen a localization D N \mathcal{D}\mathcal{N} DN of the differential-operator algebra in which the quasideterminant with differential-operator entries is defined, and that the homomorphism ρ λ \rho_{\lambda} ρ λ extends to a homomorphism of the localized algebras, denoted by ρ ^ λ \widehat{\rho}_{\lambda} ρ λ . The image of the quasideterminant under this extension is the quasideterminant of the matrix of generators in (57) ; this matrix is L ( u ) = z ( u + λ − D ) z − 1 L(u)=z(u+\lambda-D)z^{-1} L ( u ) = z ( u + λ − D ) z − 1 .
Both algebras have left ideals generated by m − ψ ( m ) m-\psi(m) m − ψ ( m ) , for m ∈ m n m\in\mathfrak{m}_{n} m ∈ m n . The quasideterminant with differential entries belongs to D N m n \mathcal{D}\mathcal{N}^{\mathfrak{m}_{n}} D N m n . We denote the corresponding left ideal in the localized differential algebra by D I \mathcal{D}I D I , so that the quotient is D N / D I \mathcal{D}\mathcal{N}/\mathcal{D}I DN / D I and the reduced differential algebra is its invariant subalgebra ( D N / D I ) m n (\mathcal{D}\mathcal{N}/\mathcal{D}I)^{\mathfrak{m}_{n}} ( DN / D I ) m n . The homomorphism ρ ^ λ \widehat{\rho}_{\lambda} ρ λ therefore descends to a homomorphism of the invariant reductions:
ρ ^ λ : A ⟶ D N , \widehat{\rho}_{\lambda}:\mathcal{A}\longrightarrow\mathcal{D}\mathcal{N}, ρ λ : A ⟶ DN ,
ρ ^ λ r e d : ( A / I ψ ) m n ⟶ ( D N / D I ) m n , \widehat{\rho}_{\lambda}^{\mathrm{red}}:(\mathcal{A}/I_{\psi})^{\mathfrak{m}_{n}}\longrightarrow(\mathcal{D}\mathcal{N}/\mathcal{D}I)^{\mathfrak{m}_{n}}, ρ λ red : ( A / I ψ ) m n ⟶ ( DN / D I ) m n ,
where D I \mathcal{D}I D I is the left ideal implementing the reduction constraints in the localized differential algebra.
Remark 5.13. The appropriate construction of the rings A \mathcal{A} A and D N \mathcal{D}\mathcal{N} DN is by Rees localization. One can introduce the Kazhdan filtration on U ( g l N ) U(\mathfrak{gl}_{N}) U ( gl N ) :
deg E a , b i , j = 1 + b − a (74) \deg E^{i,j}_{a,b}=1+b-a \tag*{(74)} deg E a , b i , j = 1 + b − a (74)
and the corresponding filtration on D ( N − ) \mathcal{D}(\mathbb{N}_{-}) D ( N − ) :
deg z ( a − 1 ) p + i , ( b − 1 ) p + j = a − b , deg ∂ z ( a − 1 ) p + i , ( b − 1 ) p + j = 1 + b − a . (75) \deg z_{(a-1)p+i,(b-1)p+j}=a-b,\qquad
\deg\partial_{z_{(a-1)p+i,(b-1)p+j}}=1+b-a. \tag*{(75)} deg z ( a − 1 ) p + i , ( b − 1 ) p + j = a − b , deg ∂ z ( a − 1 ) p + i , ( b − 1 ) p + j = 1 + b − a . (75)
The homomorphism ρ λ \rho_\lambda ρ λ extends to a homomorphism between the filtered algebras. The ideals I ψ I_\psi I ψ and I I I are filtered, so the quotient modules inherit induced filtrations. Thus, one can consider the Rees rings and the Rees modules of these quotients (see [10 ] ). The homomorphism ρ λ \rho_\lambda ρ λ therefore extends to the Rees rings and Rees modules. Rees localization is constructed by successively adding formal series that become invertible after taking the quotient. In this way, the map ρ λ \rho_\lambda ρ λ extends to the localized rings ρ ~ λ \widetilde{\rho}_\lambda ρ λ . It follows from the results of [10 ] that all the used quasideterminants are well-defined in the Rees localization – since all the matrices in the quasideterminants become invertible after reduction. Hence, all the formulas are justified by considering this concrete localization.
The compatibility conditions show that the quasideterminant with differential entries agrees with the Brundan–Kleshchev generating matrix after the generators E a , b i , j E^{i,j}_{a,b} E a , b i , j are replaced by their differential realizations. On the other hand, it agrees with the reduction of ∣ z ( u + λ − D ) z − 1 ∣ n , 1 |z(u+\lambda-D)z^{-1}|_{n,1} ∣ z ( u + λ − D ) z − 1 ∣ n , 1 . The latter can be computed directly in terms of the invariant coordinates z i \mathbf{z}_i z i and the operators D i , j D_{i,j} D i , j . This computation is independent of the choice of localization: changing the localization does not alter the identities used below or the resulting expression in the reduced differential algebra. Accordingly, no particular construction of D N \mathcal{D}N D N or of the extended homomorphism is needed for the computation below.
Theorem 5.14. The Brundan–Kleshchev matrix, evaluated via the reduced differential realization ρ λ r e d \rho^{\mathrm{red}}_\lambda ρ λ red induced by ρ λ \rho_\lambda ρ λ in (57) , satisfies the factorization
( ξ ∘ ρ λ r e d ) ( T ( u ) ) = ( u + z [ n ] ( λ n − D [ n ] ) z [ n ] − 1 ) ⋯ ( u + z [ 1 ] ( λ 1 − D [ 1 ] ) z [ 1 ] − 1 ) ⋯ (76) (\xi\circ\rho^{\mathrm{red}}_\lambda)(\mathrm{T}(u))=(u+z_{[n]}(\lambda_n-D_{[n]})z_{[n]}^{-1})\cdots(u+z_{[1]}(\lambda_1-D_{[1]})z_{[1]}^{-1})\cdots \tag*{(76)} ( ξ ∘ ρ λ red ) ( T ( u )) = ( u + z [ n ] ( λ n − D [ n ] ) z [ n ] − 1 ) ⋯ ( u + z [ 1 ] ( λ 1 − D [ 1 ] ) z [ 1 ] − 1 ) ⋯ (76)
Proof. Recall that the quasideterminant of the Lax matrix in the differential realization is
∣ L ( u ) ∣ n , 1 = ∣ ( ( u + λ 1 − D 1 ) z 1 − 1 − D 1 , 2 − D 1 , 3 ⋯ − D 1 , n 0 u + λ 2 − D 2 − D 2 , 3 ⋯ − D 2 , n 0 0 u + λ 3 − D 3 ⋯ − D 3 , n ⋮ ⋮ ⋱ ⋱ ⋮ 0 0 ⋯ 0 z n ( u + λ n − D n ) ) ∣ n , 1 . |L(u)|_{n,1}=
\left|
\begin{pmatrix}
(u+\lambda_1-D_1)z_1^{-1} & -D_{1,2} & -D_{1,3} & \cdots& -D_{1,n}\\
0 & u+\lambda_2-D_2 & -D_{2,3} & \cdots& -D_{2,n}\\
0 & 0 & u+\lambda_3-D_3 & \cdots& -D_{3,n}\\
\vdots& \vdots& \ddots& \ddots& \vdots\\
0 & 0 & \cdots& 0 & z_n(u+\lambda_n-D_n)
\end{pmatrix}
\right|_{n,1}. ∣ L ( u ) ∣ n , 1 = ( u + λ 1 − D 1 ) z 1 − 1 0 0 ⋮ 0 − D 1 , 2 u + λ 2 − D 2 0 ⋮ 0 − D 1 , 3 − D 2 , 3 u + λ 3 − D 3 ⋱ ⋯ ⋯ ⋯ ⋯ ⋱ 0 − D 1 , n − D 2 , n − D 3 , n ⋮ z n ( u + λ n − D n ) n , 1 .
To establish the explicit form after reduction, we note that the entries of the matrix in this quasideterminant commute with m n \mathfrak{m}_n m n . Indeed, the right-shift operators D i , j D_{i,j} D i , j commute with the infinitesimal left action of m n \mathfrak{m}_n m n , while the diagonal coordinates are invariant under this action, by Proposition 5.10.
Hence, we can reduce the entries of the matrix directly using Proposition 5.12. The result is
∣ L ( u ) ∣ n , 1 ⋅ 1 ‾ = ∣ ( ( u + λ 1 − D 1 ) z 1 − 1 z 1 − 1 z 2 0 ⋯ 0 0 u + λ 2 − D 2 z 2 − 1 z 3 ⋱ ⋮ 0 0 u + λ 3 − D 3 ⋱ 0 ⋮ ⋮ ⋱ ⋱ z n − 1 − 1 z n 0 ⋯ 0 0 z n ( u + λ n − D n ) ) ∣ n , 1 . |L(u)|_{n,1}\cdot\overline{1}=
\left|
\begin{pmatrix}
(u+\lambda_1-D_1)z_1^{-1} & z_1^{-1}z_2 & 0 & \cdots& 0\\
0 & u+\lambda_2-D_2 & z_2^{-1}z_3 & \ddots& \vdots\\
0 & 0 & u+\lambda_3-D_3 & \ddots& 0\\
\vdots& \vdots& \ddots& \ddots& z_{n-1}^{-1}z_n\\
0 & \cdots& 0 & 0 & z_n(u+\lambda_n-D_n)
\end{pmatrix}
\right|_{n,1}. ∣ L ( u ) ∣ n , 1 ⋅ 1 = ( u + λ 1 − D 1 ) z 1 − 1 0 0 ⋮ 0 z 1 − 1 z 2 u + λ 2 − D 2 0 ⋮ ⋯ 0 z 2 − 1 z 3 u + λ 3 − D 3 ⋱ 0 ⋯ ⋱ ⋱ ⋱ 0 0 ⋮ 0 z n − 1 − 1 z n z n ( u + λ n − D n ) n , 1 .
The last quasideterminant can be computed explicitly using Proposition 3.6 . Thus, in the reduced algebra,
∣ L ( u ) ∣ n , 1 ⋅ 1 ‾ = ( − 1 ) n − 1 z n ( u + λ n − D n ) z n − 1 ⋯ z 1 ( u + λ 1 − D 1 ) z 1 − 1 ⋅ 1 ‾ . (77) |L(u)|_{n,1}\cdot\overline{1}=(-1)^{n-1}z_n(u+\lambda_n-D_n)z_n^{-1}\cdots z_1(u+\lambda_1-D_1)z_1^{-1}\cdot\overline{1}. \tag*{(77)} ∣ L ( u ) ∣ n , 1 ⋅ 1 = ( − 1 ) n − 1 z n ( u + λ n − D n ) z n − 1 ⋯ z 1 ( u + λ 1 − D 1 ) z 1 − 1 ⋅ 1 . (77)
On the other hand, by (43) , the Brundan–Kleshchev matrix is ( − 1 ) n − 1 (-1)^{n-1} ( − 1 ) n − 1 times the reduced quasideterminant. Thus the sign in the last display cancels under this normalization.
Next, we use the identification ξ \xi ξ between the reduced algebra D − \mathcal{D}_{-} D − and D ( N − ′ ) ⊗ n \mathcal{D}(N'_{-})^{\otimes n} D ( N − ′ ) ⊗ n to obtain (76) . In the original coordinates on N − N_{-} N − , the operators D a \mathrm{D}_{a} D a may involve derivatives with respect to off-diagonal coordinates. Under the isomorphism ξ : D − ⟶ ∼ D ( N − ′ ) ⊗ n \xi:\mathcal{D}_{-}\overset{\sim}{\longrightarrow}\mathcal{D}(N'_{-})^{\otimes n} ξ : D − ⟶ ∼ D ( N − ′ ) ⊗ n from Proposition 5.10, each factor is expressed in terms of differential operators acting in the individual tensor factors.
Together, the Miura factorization (42) and Theorem 5.14 are summarized by the commutative diagram
W n ( g l p ) → Θ U ( g l p ) ⊗ n ρ λ r e d ↓ ↓ ρ λ 1 ⊗ ⋯ ⊗ ρ λ n D − → ξ D ( N − ′ ) ⊗ n (78) \begin{CD}
W_n(\mathfrak{gl}_p) @>{\Theta}>> U(\mathfrak{gl}_p)^{\otimes n} \\
@V{\rho_{\lambda}^{\mathrm{red}}}VV @VV{\rho_{\lambda_1}\otimes\cdots\otimes\rho_{\lambda_n}}V \\
\mathcal{D}_{-} @>{\xi}>> \mathcal{D}(N'_{-})^{\otimes n} \tag*{(78)}
\end{CD} W n ( gl p ) ρ λ red ↓ ⏐ D − Θ ξ U ( gl p ) ⊗ n ↓ ⏐ ρ λ 1 ⊗ ⋯ ⊗ ρ λ n D ( N − ′ ) ⊗ n (78)
Each factor in (76) is the differential realization of the corresponding factor u + E i i u+E_{ii} u + E ii in the Miura product. Consequently,
ξ ∘ ρ λ r e d = ( ρ λ 1 ⊗ ⋯ ⊗ ρ λ n ) ∘ Θ . (79) \xi\circ\rho_{\lambda}^{\mathrm{red}}=(\rho_{\lambda_1}\otimes\cdots\otimes\rho_{\lambda_n})\circ\Theta. \tag*{(79)} ξ ∘ ρ λ red = ( ρ λ 1 ⊗ ⋯ ⊗ ρ λ n ) ∘ Θ. (79)
Thus the direct Hamiltonian reduction agrees with first applying the quantum Miura homomorphism and then applying the tensor-product differential realization.
Conclusion
In this work, we studied the Hamiltonian reduction of the big-cell differential realization in the rectangular case. We showed that this realization is naturally compatible with Hamiltonian reduction, yielding an intrinsic factorization of the reduced differential realization of the rectangular W W W -algebra.
We also want to emphasize the usefulness of quasideterminants for describing invariants in the W W W -algebra and the reduced differential algebra. To illustrate the geometric picture, recall the classical analogue of our construction: the differential realization is replaced by the Springer map from the cotangent bundle T ∗ N − T^{*}N_{-} T ∗ N − to the coadjoint orbit O λ \mathcal{O}_{\lambda} O λ [2 , 8 ] . Hamiltonian reduction of this map gives Darboux coordinates on a symplectic leaf of the Slodowy slice, which is quantized by the finite W W W -algebra [24 , 32 ] . In the rectangular case, this picture extends to a description of the symplectic leaves of the Yangian [34 ] .
The main steps of our construction have classical geometric counterparts. In the classical setting, the matrix entries commute, so the localization and factorization procedures for the quasideterminant can be carried out directly. The quasideterminant then reduces to the row determinant of the Lax operator with constrained entries.
The natural properties of the invariant suggested by the geometric construction also hold for the quasideterminant in the noncommutative, quantum setting considered here. Thus, quasideterminants provide a natural language for extending the classical picture to the quantum setting. They also make the underlying algebraic structure transparent: the RTT relations can be checked directly in this framework.
We would like to note that our analysis is restricted to the rectangular reduction and to the differential realization considered above, which is associated with a generic coadjoint orbit. More general
Lax operators associated with special orbits have also been studied [22 , 23 ] . However, when the differential realization is specialized to a concrete space, our construction applies to a large class of Yangian representations and thus appears quite general.
The natural question is what obstacles arise in extending this picture. We outline several possible directions:
How can the factorization mechanism be extended to non-rectangular reductions?
What obstructs generalization to more general type-A Lax operators? We suspect that difficulties arise only when the Lax operator’s Jordan-block structure interacts nontrivially with the reduction.
How can intrinsic factorization be extended to types B, C, and D and their associated W W W -algebras?
The first and third questions have been considered in related quasideterminant constructions. For non-rectangular reductions in type A, De Sole, Kac, and Valeri constructed a generating matrix using a generalized quasideterminant [10 ] . This matrix does not by itself provide a full set of generators for the W W W -algebra; its relation to a full generating set was subsequently established in [12 ] . For classical simple Lie algebras, generalized-quasideterminant generating matrices with coefficients in the finite W W W -algebra were constructed in [11 ] . It remains useful to understand how such generating matrices encode a full set of generators. Our differential-realization approach, grounded in geometric reasoning, may offer a complementary perspective and reveal a simpler structure for the generators in the differential realization.
Finally, we emphasize that the present result provides a starting point for studying the reduction of intertwiners and factorized R R R -matrices, which we plan to investigate in future work.
Acknowledgments
We thank BIMSA for hospitality. A big part of the work was done during our visit to BIMSA.
We thank N. Belousov, S. Khoroshkin, N. Reshetikhin, A. Litvinov, and V. Rubtsov for insightful discussions. We would like to express our special thanks to Vitaly Tarasov. Thanks to our joint discussions, all participants were able to appreciate the power and beauty of quasideterminants.
The work of D.G. performed in Skoltech has been supported by the Russian Science Foundation under the grant 26-11-00342.
Commutation relations
Details of calculations in index form
Throughout this subsection, we use the summation convention: repeated block indices are summed over 1 through n n n , and repeated internal indices over 1 through p p p , whether written as subscripts or superscripts. We start from relation
( u − v ) L a , b k i ( u ) L c , d s j ( v ) + L c , b s j ( u ) L a , d k i ( v ) = ( u − v ) L c , d s j ( v ) L a , b k i ( u ) + L c , b s j ( v ) L a , d k i ( u ) , (80) (u-v)L^{ki}_{a,b}(u)L^{sj}_{c,d}(v)+L^{sj}_{c,b}(u)L^{ki}_{a,d}(v)=(u-v)L^{sj}_{c,d}(v)L^{ki}_{a,b}(u)+L^{sj}_{c,b}(v)L^{ki}_{a,d}(u), \tag*{(80)} ( u − v ) L a , b k i ( u ) L c , d s j ( v ) + L c , b s j ( u ) L a , d k i ( v ) = ( u − v ) L c , d s j ( v ) L a , b k i ( u ) + L c , b s j ( v ) L a , d k i ( u ) , (80)
where
L a , b i , j ( u ) = u δ a , b δ i , j + E b , a j , i . (81) L^{i,j}_{a,b}(u)=u\delta_{a,b}\delta^{i,j}+E^{j,i}_{b,a}. \tag*{(81)} L a , b i , j ( u ) = u δ a , b δ i , j + E b , a j , i . (81)
Let us introduce the inverse matrix
L a , b i , j ( u ) M b , c j , k ( u ) = δ a , c δ i , k ; M a , b i , j ( u ) L b , c j , k ( u ) = δ a , c δ i , k . (82) L^{i,j}_{a,b}(u)M^{j,k}_{b,c}(u)=\delta_{a,c}\delta^{i,k}\ ;\ M^{i,j}_{a,b}(u)L^{j,k}_{b,c}(u)=\delta_{a,c}\delta^{i,k}. \tag*{(82)} L a , b i , j ( u ) M b , c j , k ( u ) = δ a , c δ i , k ; M a , b i , j ( u ) L b , c j , k ( u ) = δ a , c δ i , k . (82)
Let us transform relation (A.1) in a following way. On the first step we multiply on the matrix M ( v ) M(v) M ( v ) from the right
( u − v ) L a , b k , i ( u ) L c , d s , j ( v ) M d , f j , m ( v ) + L c , b s , i ( u ) L a , d k , j ( v ) M d , f j , m ( v ) = ( u − v ) L c , d s , j ( v ) L a , b k , i ( u ) M d , f j , m ( v ) + L c , b s , i ( v ) L a , d k , j ( u ) M d , f j , m ( v ) , \begin{aligned}
(u-v)L^{k,i}_{a,b}(u)L^{s,j}_{c,d}(v)M^{j,m}_{d,f}(v)+L^{s,i}_{c,b}(u)L^{k,j}_{a,d}(v)M^{j,m}_{d,f}(v)\\
=(u-v)L^{s,j}_{c,d}(v)L^{k,i}_{a,b}(u)M^{j,m}_{d,f}(v)+L^{s,i}_{c,b}(v)L^{k,j}_{a,d}(u)M^{j,m}_{d,f}(v),
\end{aligned} ( u − v ) L a , b k , i ( u ) L c , d s , j ( v ) M d , f j , m ( v ) + L c , b s , i ( u ) L a , d k , j ( v ) M d , f j , m ( v ) = ( u − v ) L c , d s , j ( v ) L a , b k , i ( u ) M d , f j , m ( v ) + L c , b s , i ( v ) L a , d k , j ( u ) M d , f j , m ( v ) ,
which gives
( u − v ) L a , b k , i ( u ) δ s , m δ c , f + L c , b s , i ( u ) δ k , m δ a , f = ( u − v ) L c , d s , j ( v ) L a , b k , i ( u ) M d , f j , m ( v ) + L c , b s , i ( v ) L a , d k , j ( u ) M d , f j , m ( v ) , \begin{aligned}
(u-v)L^{k,i}_{a,b}(u)\delta^{s,m}\delta_{c,f}+L^{s,i}_{c,b}(u)\delta^{k,m}\delta_{a,f}
=(u-v)L^{s,j}_{c,d}(v)L^{k,i}_{a,b}(u)M^{j,m}_{d,f}(v)\\
+L^{s,i}_{c,b}(v)L^{k,j}_{a,d}(u)M^{j,m}_{d,f}(v),
\end{aligned} ( u − v ) L a , b k , i ( u ) δ s , m δ c , f + L c , b s , i ( u ) δ k , m δ a , f = ( u − v ) L c , d s , j ( v ) L a , b k , i ( u ) M d , f j , m ( v ) + L c , b s , i ( v ) L a , d k , j ( u ) M d , f j , m ( v ) ,
and on the second step multiply on M ( v ) M(v) M ( v ) from the left
( u − v ) M e , c q , s ( v ) L a , b k , i ( u ) δ s , m δ c , f + M e , c q , s ( v ) L c , b s , i ( u ) δ k , m δ a , f = ( u − v ) M e , c q , s ( v ) L c , d s , j ( v ) L a , b k , i ( u ) M d , f j , m ( v ) + M e , c q , s ( v ) L c , b s , i ( v ) L a , d k , j ( u ) M d , f j , m ( v ) , \begin{aligned}
(u-v)M^{q,s}_{e,c}(v)L^{k,i}_{a,b}(u)\delta^{s,m}\delta_{c,f}+M^{q,s}_{e,c}(v)L^{s,i}_{c,b}(u)\delta^{k,m}\delta_{a,f}\\
=(u-v)M^{q,s}_{e,c}(v)L^{s,j}_{c,d}(v)L^{k,i}_{a,b}(u)M^{j,m}_{d,f}(v)+M^{q,s}_{e,c}(v)L^{s,i}_{c,b}(v)L^{k,j}_{a,d}(u)M^{j,m}_{d,f}(v),
\end{aligned} ( u − v ) M e , c q , s ( v ) L a , b k , i ( u ) δ s , m δ c , f + M e , c q , s ( v ) L c , b s , i ( u ) δ k , m δ a , f = ( u − v ) M e , c q , s ( v ) L c , d s , j ( v ) L a , b k , i ( u ) M d , f j , m ( v ) + M e , c q , s ( v ) L c , b s , i ( v ) L a , d k , j ( u ) M d , f j , m ( v ) ,
which gives
( u − v ) M e , c q , s ( v ) L a , b k , i ( u ) δ s , m δ c , f + M e , c q , s ( v ) L c , b s , i ( u ) δ k , m δ a , f = ( u − v ) δ q , j δ e , d L a , b k , i ( u ) M d , f j , m ( v ) + δ q , i δ e , b L a , d k , j ( u ) M d , f j , m ( v ) , \begin{aligned}
(u-v)M^{q,s}_{e,c}(v)L^{k,i}_{a,b}(u)\delta^{s,m}\delta_{c,f}+M^{q,s}_{e,c}(v)L^{s,i}_{c,b}(u)\delta^{k,m}\delta_{a,f}\\
=(u-v)\delta^{q,j}\delta_{e,d}L^{k,i}_{a,b}(u)M^{j,m}_{d,f}(v)+\delta^{q,i}\delta_{e,b}L^{k,j}_{a,d}(u)M^{j,m}_{d,f}(v),
\end{aligned} ( u − v ) M e , c q , s ( v ) L a , b k , i ( u ) δ s , m δ c , f + M e , c q , s ( v ) L c , b s , i ( u ) δ k , m δ a , f = ( u − v ) δ q , j δ e , d L a , b k , i ( u ) M d , f j , m ( v ) + δ q , i δ e , b L a , d k , j ( u ) M d , f j , m ( v ) ,
and finally
( u − v ) M e , f q , m ( v ) L a , b k , i ( u ) + M e , c q , s ( v ) L c , b s , i ( u ) δ k , m δ a , f = ( u − v ) L a , b k , i ( u ) M e , f q , m ( v ) + δ q , i δ e , b L a , d k , j ( u ) M d , f j , m ( v ) . (83) \begin{aligned}
(u-v)M^{q,m}_{e,f}(v)L^{k,i}_{a,b}(u)+M^{q,s}_{e,c}(v)L^{s,i}_{c,b}(u)\delta^{k,m}\delta_{a,f}\\
=(u-v)L^{k,i}_{a,b}(u)M^{q,m}_{e,f}(v)+\delta^{q,i}\delta_{e,b}L^{k,j}_{a,d}(u)M^{j,m}_{d,f}(v). \tag*{(83)}
\end{aligned} ( u − v ) M e , f q , m ( v ) L a , b k , i ( u ) + M e , c q , s ( v ) L c , b s , i ( u ) δ k , m δ a , f = ( u − v ) L a , b k , i ( u ) M e , f q , m ( v ) + δ q , i δ e , b L a , d k , j ( u ) M d , f j , m ( v ) . (83)
To prove the Yangian commutation relations (35) for the quasideterminant’s matrix elements, we continue transforming the main relation (A.1) and multiply (A.4) by M ( u ) M(u) M ( u ) from the right.
( u − v ) M e , f q , m ( v ) L a , b k , i ( u ) M b , h i , j ( u ) + M e , c q , s ( v ) L c , b s , i ( u ) M b , h i , j ( u ) δ k , m δ a , f = ( u − v ) L a , b k , i ( u ) M e , f q , m ( v ) M b , h i , j ( u ) + δ q , i δ e , b L a , β k , α ( u ) M β , f α , m ( v ) M b , h i , j ( u ) , \begin{aligned}
(u-v)M^{q,m}_{e,f}(v)L^{k,i}_{a,b}(u)M^{i,j}_{b,h}(u)+M^{q,s}_{e,c}(v)L^{s,i}_{c,b}(u)M^{i,j}_{b,h}(u)\delta^{k,m}\delta_{a,f}\\
=(u-v)L^{k,i}_{a,b}(u)M^{q,m}_{e,f}(v)M^{i,j}_{b,h}(u)+\delta^{q,i}\delta_{e,b}L^{k,\alpha}_{a,\beta}(u)M^{\alpha,m}_{\beta,f}(v)M^{i,j}_{b,h}(u),
\end{aligned} ( u − v ) M e , f q , m ( v ) L a , b k , i ( u ) M b , h i , j ( u ) + M e , c q , s ( v ) L c , b s , i ( u ) M b , h i , j ( u ) δ k , m δ a , f = ( u − v ) L a , b k , i ( u ) M e , f q , m ( v ) M b , h i , j ( u ) + δ q , i δ e , b L a , β k , α ( u ) M β , f α , m ( v ) M b , h i , j ( u ) ,
so that
( u − v ) M e , f q , m ( v ) δ k , j δ a , h + M e , c q , s ( v ) δ s , j δ c , h δ k , m δ a , f = ( u − v ) L a , b k , i ( u ) M e , f q , m ( v ) M b , h i , j ( u ) + δ q , i δ e , b L a , β k , α ( u ) M β , f α , m ( v ) M b , h i , j ( u ) . \begin{aligned}
(u-v)M^{q,m}_{e,f}(v)\delta^{k,j}\delta_{a,h}+M^{q,s}_{e,c}(v)\delta^{s,j}\delta_{c,h}\delta^{k,m}\delta_{a,f}\\
=(u-v)L^{k,i}_{a,b}(u)M^{q,m}_{e,f}(v)M^{i,j}_{b,h}(u)+\delta^{q,i}\delta_{e,b}L^{k,\alpha}_{a,\beta}(u)M^{\alpha,m}_{\beta,f}(v)M^{i,j}_{b,h}(u).
\end{aligned} ( u − v ) M e , f q , m ( v ) δ k , j δ a , h + M e , c q , s ( v ) δ s , j δ c , h δ k , m δ a , f = ( u − v ) L a , b k , i ( u ) M e , f q , m ( v ) M b , h i , j ( u ) + δ q , i δ e , b L a , β k , α ( u ) M β , f α , m ( v ) M b , h i , j ( u ) .
or
( u − v ) M e , f q , m ( v ) δ k , j δ a , h + M e , h q , j ( v ) δ k , m δ a , f = ( u − v ) L a , b k , i ( u ) M e , f q , m ( v ) M b , h i , j ( u ) + L a , d k , t ( u ) M d , f t , m ( v ) M e , h q , j ( u ) (u-v)M^{q,m}_{e,f}(v)\delta^{k,j}\delta_{a,h}+M^{q,j}_{e,h}(v)\delta^{k,m}\delta_{a,f}
= (u-v)L^{k,i}_{a,b}(u)M^{q,m}_{e,f}(v)M^{i,j}_{b,h}(u)+L^{k,t}_{a,d}(u)M^{t,m}_{d,f}(v)M^{q,j}_{e,h}(u) ( u − v ) M e , f q , m ( v ) δ k , j δ a , h + M e , h q , j ( v ) δ k , m δ a , f = ( u − v ) L a , b k , i ( u ) M e , f q , m ( v ) M b , h i , j ( u ) + L a , d k , t ( u ) M d , f t , m ( v ) M e , h q , j ( u )
and then multiply on the matrix M ( u ) M(u) M ( u ) from the left
( u − v ) M c , a s , k ( u ) M e , f q , m ( v ) δ k , j δ a , h + M c , a s , k ( u ) M e , h q , j ( v ) δ k , m δ a , f = ( u − v ) M c , a s , k ( u ) L a , b k , i ( u ) M e , f q , m ( v ) M b , h i , j ( u ) + M c , a s , k ( u ) L a , d k , t ( u ) M d , f t , m ( v ) M e , h q , j ( u ) (u-v)M^{s,k}_{c,a}(u)M^{q,m}_{e,f}(v)\delta^{k,j}\delta_{a,h}+M^{s,k}_{c,a}(u)M^{q,j}_{e,h}(v)\delta^{k,m}\delta_{a,f}
= (u-v)M^{s,k}_{c,a}(u)L^{k,i}_{a,b}(u)M^{q,m}_{e,f}(v)M^{i,j}_{b,h}(u)+M^{s,k}_{c,a}(u)L^{k,t}_{a,d}(u)M^{t,m}_{d,f}(v)M^{q,j}_{e,h}(u) ( u − v ) M c , a s , k ( u ) M e , f q , m ( v ) δ k , j δ a , h + M c , a s , k ( u ) M e , h q , j ( v ) δ k , m δ a , f = ( u − v ) M c , a s , k ( u ) L a , b k , i ( u ) M e , f q , m ( v ) M b , h i , j ( u ) + M c , a s , k ( u ) L a , d k , t ( u ) M d , f t , m ( v ) M e , h q , j ( u )
which gives
( u − v ) M c , h s , j ( u ) M e , f q , m ( v ) + M c , f s , m ( u ) M e , h q , j ( v ) = ( u − v ) M e , f q , m ( v ) M c , h s , j ( u ) + M c , f s , m ( v ) M e , h q , j ( u ) (u-v)M^{s,j}_{c,h}(u)M^{q,m}_{e,f}(v)+M^{s,m}_{c,f}(u)M^{q,j}_{e,h}(v)
= (u-v)M^{q,m}_{e,f}(v)M^{s,j}_{c,h}(u)+M^{s,m}_{c,f}(v)M^{q,j}_{e,h}(u) ( u − v ) M c , h s , j ( u ) M e , f q , m ( v ) + M c , f s , m ( u ) M e , h q , j ( v ) = ( u − v ) M e , f q , m ( v ) M c , h s , j ( u ) + M c , f s , m ( v ) M e , h q , j ( u )
or
( u − v ) M c , h s , j ( u ) M e , f q , m ( v ) + M c , f s , m ( u ) M e , h q , j ( v ) = ( u − v ) M e , f q , m ( v ) M c , h s , j ( u ) + M c , f s , m ( v ) M e , h q , j ( u ) (84) (u-v)M^{s,j}_{c,h}(u)M^{q,m}_{e,f}(v)+M^{s,m}_{c,f}(u)M^{q,j}_{e,h}(v)=(u-v)M^{q,m}_{e,f}(v)M^{s,j}_{c,h}(u)+M^{s,m}_{c,f}(v)M^{q,j}_{e,h}(u) \tag*{(84)} ( u − v ) M c , h s , j ( u ) M e , f q , m ( v ) + M c , f s , m ( u ) M e , h q , j ( v ) = ( u − v ) M e , f q , m ( v ) M c , h s , j ( u ) + M c , f s , m ( v ) M e , h q , j ( u ) (84)
We obtain the same Yangian relations as for L L L -matrices and the particular case c = e = 1 c=e=1 c = e = 1 , h = f = n h=f=n h = f = n gives Yangian relations for the quasideterminants
( u − v ) M 1 , n s , j ( u ) M 1 , n q , m ( v ) + M 1 , n s , m ( u ) M 1 , n q , j ( v ) = ( u − v ) M 1 , n q , m ( v ) M 1 , n s , j ( u ) + M 1 , n s , m ( v ) M 1 , n q , j ( u ) . (u-v)M^{s,j}_{1,n}(u)M^{q,m}_{1,n}(v)+M^{s,m}_{1,n}(u)M^{q,j}_{1,n}(v)=(u-v)M^{q,m}_{1,n}(v)M^{s,j}_{1,n}(u)+M^{s,m}_{1,n}(v)M^{q,j}_{1,n}(u). ( u − v ) M 1 , n s , j ( u ) M 1 , n q , m ( v ) + M 1 , n s , m ( u ) M 1 , n q , j ( v ) = ( u − v ) M 1 , n q , m ( v ) M 1 , n s , j ( u ) + M 1 , n s , m ( v ) M 1 , n q , j ( u ) .
Functional realization of the reduction
Let us briefly comment on the functional side of the reduction described above. We postpone the detailed study to our second work; however, here we clarify the meaning of the reduction in a concrete realization on some functional space.
Similarly to the construction of the homomorphism ρ λ \rho_{\lambda} ρ λ , we start from the test functions on N − N_{-} N − with coordinates ( m , z d ) (m,z_d) ( m , z d ) . For the purpose of reduction, we have to fix the behaviour of the test function with respect to the left action of m n \mathfrak{m}_{n} m n (i.e. to consider the subspace of Whittaker vectors as a representation of our W W W -algebra or reduced algebra D − D_{-} D − ).
It is straightforward to check that the relation
x ⋅ Φ ( z ) = ψ ( x ) Φ ( z ) , ∀ x ∈ m n , x\cdot\Phi(z)=\psi(x)\Phi(z),\quad\forall x\in\mathfrak{m}_{n}, x ⋅ Φ ( z ) = ψ ( x ) Φ ( z ) , ∀ x ∈ m n ,
implies
Φ ( m , z d ) = e − ψ ( log m ) Ψ ( z d ) , \Phi(m,z_d)=e^{-\psi(\log m)}\Psi(z_d), Φ ( m , z d ) = e − ψ ( l o g m ) Ψ ( z d ) ,
where Ψ \Psi Ψ can be an arbitrary function of z d z_d z d . For the usual choice of the character ψ ( E a , b i , j ) = δ b , a − 1 δ i , j \psi(E^{i,j}_{a,b})=\delta_{b,a-1}\delta_{i,j} ψ ( E a , b i , j ) = δ b , a − 1 δ i , j , the prefactor is e − ψ ( log m ) = e − ∑ k = 1 n − 1 tr ( m k + 1 , k ) e^{-\psi(\log m)}=e^{-\sum_{k=1}^{n-1}\operatorname{tr}(m_{k+1,k})} e − ψ ( l o g m ) = e − ∑ k = 1 n − 1 tr ( m k + 1 , k ) , where m k + 1 , k = z k + 1 , k z k − 1 \mathbf{m}_{k+1,k}=\mathbf{z}_{k+1,k}\mathbf{z}_{k}^{-1} m k + 1 , k = z k + 1 , k z k − 1 are blocks of m m m .
We now interpret the abstract map ξ \xi ξ as an intertwining map between the subspace of functions Φ ( m , z d ) \Phi(m,z_d) Φ ( m , z d ) with prefactor and the subspace of Ψ ( z d ) \Psi(z_d) Ψ ( z d ) , via conjugation by exponentials and setting the off-diagonal coordinate parts in differential operators to zero. Let us derive the explicit form of the reduction of the generators D a , b \mathbf{D}_{a,b} D a , b . To be explicit, we consider the n = 2 n=2 n = 2 case.
The subspace consists of the following functions:
Φ ( z ) = e − ψ ( z 2 , 1 z 1 − 1 ) Ψ ( z 1 , z 2 ) . \Phi(z)=e^{-\psi(z_{2,1}z_1^{-1})}\Psi(z_1,z_2). Φ ( z ) = e − ψ ( z 2 , 1 z 1 − 1 ) Ψ ( z 1 , z 2 ) .
For the right shifts D a D_a D a corresponding to D 1 \mathbf{D}_1 D 1 we have
( z 1 0 z 2 , 1 z 2 ) ( 1 + ε a 0 0 1 ) = ( 1 0 z 2 , 1 z 1 − 1 1 ) ( z 1 ( 1 + ε a ) 0 0 z 2 ) , \begin{pmatrix} z_1 & 0 \\ z_{2,1} & z_2 \end{pmatrix}
\begin{pmatrix} 1+\varepsilon a & 0 \\ 0 & 1 \end{pmatrix}
=
\begin{pmatrix} 1 & 0 \\ z_{2,1}z_1^{-1} & 1 \end{pmatrix}
\begin{pmatrix} z_1(1+\varepsilon a) & 0 \\ 0 & z_2 \end{pmatrix}, ( z 1 z 2 , 1 0 z 2 ) ( 1 + ε a 0 0 1 ) = ( 1 z 2 , 1 z 1 − 1 0 1 ) ( z 1 ( 1 + ε a ) 0 0 z 2 ) ,
so that
D a Φ ( z ) = e − ψ ( z 2 , 1 z 1 − 1 ) d d ε Ψ ( z 1 ( 1 + ε a ) , z 2 ) ∣ ε = 0 . D_a\Phi(z)=e^{-\psi(z_{2,1}z_1^{-1})}\left.\frac{d}{d\varepsilon}\Psi\bigl(z_1(1+\varepsilon a),z_2\bigr)\right|_{\varepsilon=0}. D a Φ ( z ) = e − ψ ( z 2 , 1 z 1 − 1 ) d ε d Ψ ( z 1 ( 1 + ε a ) , z 2 ) ε = 0 .
In matrix form we have the intertwining relation
D 1 [ e − ψ ( z 2 , 1 z 1 − 1 ) Ψ ( z d ) ] = e − ψ ( z 2 , 1 z 1 − 1 ) [ ( d 1 z 1 ) + Ψ ( z d ) ] , \mathbf{D}_1\left[e^{-\psi(z_{2,1}z_1^{-1})}\Psi(z_d)\right]
=e^{-\psi(z_{2,1}z_1^{-1})}\left[(d_1z_1)_+\Psi(z_d)\right], D 1 [ e − ψ ( z 2 , 1 z 1 − 1 ) Ψ ( z d ) ] = e − ψ ( z 2 , 1 z 1 − 1 ) [ ( d 1 z 1 ) + Ψ ( z d ) ] ,
and a similar relation holds for D 2 \mathbf{D}_2 D 2 . It is instructive to compare with the initial picture without reduction. In this case one works with functions Φ ( z 2 , 1 , z 1 , z 2 ) \Phi(z_{2,1},z_1,z_2) Φ ( z 2 , 1 , z 1 , z 2 ) and
( z 1 0 z 2 , 1 z 2 ) ( 1 + ε a 0 0 1 ) = ( z 1 ( 1 + ε a ) 0 z 2 , 1 ( 1 + ε a ) z 2 ) (85) \begin{pmatrix} z_1 & 0 \\ z_{2,1} & z_2 \end{pmatrix}
\begin{pmatrix} 1+\varepsilon a & 0 \\ 0 & 1 \end{pmatrix}
=
\begin{pmatrix} z_1(1+\varepsilon a) & 0 \\ z_{2,1}(1+\varepsilon a) & z_2 \tag*{(85)} \end{pmatrix} ( z 1 z 2 , 1 0 z 2 ) ( 1 + ε a 0 0 1 ) = ( z 1 ( 1 + ε a ) z 2 , 1 ( 1 + ε a ) 0 z 2 ) (85)
so that D 1 \mathbf{D}_1 D 1 contains contributions from all the variables:
D 1 = ( d 1 z 1 ) + + ( d 1 , 2 z 2 , 1 ) + . \mathbf{D}_1=(d_1z_1)_+ +(d_{1,2}z_{2,1})_+. D 1 = ( d 1 z 1 ) + + ( d 1 , 2 z 2 , 1 ) + .
On a formal level D 1 \mathbf{D}_1 D 1 commutes with e − ψ ( z 2 , 1 z 1 − 1 ) e^{-\psi(z_{2,1}z_1^{-1})} e − ψ ( z 2 , 1 z 1 − 1 ) so that we have
D 1 [ e − ψ ( z 2 , 1 z 1 − 1 ) Ψ ( z d ) ] = e − ψ ( z 2 , 1 z 1 − 1 ) [ ( d 1 z 1 ) + + ( d 1 , 2 z 2 , 1 ) + ] Ψ ( z d ) = e − ψ ( z 2 , 1 z 1 − 1 ) [ ( d 1 z 1 ) + ] Ψ ( z d ) . \mathbf{D}_1\left[e^{-\psi(z_{2,1}z_1^{-1})}\Psi(z_d)\right]
=e^{-\psi(z_{2,1}z_1^{-1})}\left[(d_1z_1)_+ +(d_{1,2}z_{2,1})_+\right]\Psi(z_d)
=e^{-\psi(z_{2,1}z_1^{-1})}\left[(d_1z_1)_+\right]\Psi(z_d). D 1 [ e − ψ ( z 2 , 1 z 1 − 1 ) Ψ ( z d ) ] = e − ψ ( z 2 , 1 z 1 − 1 ) [ ( d 1 z 1 ) + + ( d 1 , 2 z 2 , 1 ) + ] Ψ ( z d ) = e − ψ ( z 2 , 1 z 1 − 1 ) [ ( d 1 z 1 ) + ] Ψ ( z d ) .
If we additionally relabel the variables z i → z [ i ] z_i\to z_{[i]} z i → z [ i ] , then we find that the formal map ξ \xi ξ implements exactly the passage from the subspace of functions e − ψ ( log m ) Ψ ( z [ 1 ] , z [ 2 ] ) e^{-\psi(\log m)}\Psi(z_{[1]},z_{[2]}) e − ψ ( l o g m ) Ψ ( z [ 1 ] , z [ 2 ] ) to the subspace Ψ ( z [ 1 ] , z [ 2 ] ) \Psi(z_{[1]},z_{[2]}) Ψ ( z [ 1 ] , z [ 2 ] ) .
The reduction of the operators D a , b \mathbf{D}_{a,b} D a , b depends on the concrete choice of the character e − ψ ( log m ) e^{-\psi(\log m)} e − ψ ( l o g m ) .
For the left shifts E a E_a E a corresponding to E 1 , 2 \mathbf{E}_{1,2} E 1 , 2 , we have
e − ε a z = ( 1 0 − ε a 1 ) ( z 1 0 z 2 , 1 z 2 ) = ( 1 0 z 2 , 1 z 1 − 1 − ε a 1 ) ( z 1 0 0 z 2 ) ; a = ( 1 0 a 1 ) ∈ exp ( m 2 ) , e^{-\varepsilon a}z=
\begin{pmatrix} 1 & 0 \\ -\varepsilon a & 1 \end{pmatrix}
\begin{pmatrix} z_1 & 0 \\ z_{2,1} & z_2 \end{pmatrix}
=
\begin{pmatrix} 1 & 0 \\ z_{2,1}z_1^{-1}-\varepsilon a & 1 \end{pmatrix}
\begin{pmatrix} z_1 & 0 \\ 0 & z_2 \end{pmatrix};
\qquad
a=\begin{pmatrix} 1 & 0 \\ a & 1 \end{pmatrix}\in\exp(\mathfrak{m}_2), e − ε a z = ( 1 − ε a 0 1 ) ( z 1 z 2 , 1 0 z 2 ) = ( 1 z 2 , 1 z 1 − 1 − ε a 0 1 ) ( z 1 0 0 z 2 ) ; a = ( 1 a 0 1 ) ∈ exp ( m 2 ) ,
and therefore
E a Φ ( z ) = d d ε e ψ ( z 2 , 1 z 1 − 1 − ε a ) Ψ ( z 1 , z 2 ) ∣ ε = 0 E_a\Phi(\mathbf{z})=\left.\frac{d}{d\varepsilon}e^{\psi(\mathbf{z}_{2,1}\mathbf{z}_1^{-1}-\varepsilon a)}\Psi(\mathbf{z}_1,\mathbf{z}_2)\right|_{\varepsilon=0} E a Φ ( z ) = d ε d e ψ ( z 2 , 1 z 1 − 1 − ε a ) Ψ ( z 1 , z 2 ) ε = 0
We consider the reduction associated with the character ψ ( E 2 , 1 i , j ) = δ i , j \psi(E_{2,1}^{i,j})=\delta_{i,j} ψ ( E 2 , 1 i , j ) = δ i , j , and this defines the prefactor e − tr ( z 2 , 1 z 1 − 1 ) e^{-\operatorname{tr}(\mathbf{z}_{2,1}\mathbf{z}_1^{-1})} e − tr ( z 2 , 1 z 1 − 1 ) . For the right shifts D a D_a D a corresponding to D 1 , 2 D_{1,2} D 1 , 2 we have
( z 1 0 z 2 , 1 z 2 ) ( 1 0 ε a 1 ) = ( 1 0 z 2 , 1 z 1 − 1 + ε z 2 a z 1 − 1 1 ) ( z 1 0 0 z 2 ) \begin{pmatrix}
\mathbf{z}_1 & 0\\
\mathbf{z}_{2,1} & \mathbf{z}_2
\end{pmatrix}
\begin{pmatrix}
1 & 0\\
\varepsilon a & 1
\end{pmatrix}
=
\begin{pmatrix}
1 & 0\\
\mathbf{z}_{2,1}\mathbf{z}_1^{-1}+\varepsilon\mathbf{z}_2a\mathbf{z}_1^{-1} & 1
\end{pmatrix}
\begin{pmatrix}
\mathbf{z}_1 & 0\\
0 & \mathbf{z}_2
\end{pmatrix} ( z 1 z 2 , 1 0 z 2 ) ( 1 ε a 0 1 ) = ( 1 z 2 , 1 z 1 − 1 + ε z 2 a z 1 − 1 0 1 ) ( z 1 0 0 z 2 )
so that
D a [ e − tr ( z 2 , 1 z 1 − 1 ) Ψ ( z 1 , z 2 ) ] = d d ε e − ε tr ( z 2 a z 1 − 1 ) ∣ ε = 0 e − tr ( z 2 , 1 z 1 − 1 ) Ψ ( z 1 , z 2 ) = e − tr ( z 2 , 1 z 1 − 1 ) [ − tr ( a z 1 − 1 z 2 ) Ψ ( z 1 , z 2 ) ] . \begin{aligned}
D_a\left[e^{-\operatorname{tr}(\mathbf{z}_{2,1}\mathbf{z}_1^{-1})}\Psi(\mathbf{z}_1,\mathbf{z}_2)\right]
=
\left.\frac{d}{d\varepsilon}e^{-\varepsilon\operatorname{tr}(\mathbf{z}_2a\mathbf{z}_1^{-1})}\right|_{\varepsilon=0}
e^{-\operatorname{tr}(\mathbf{z}_{2,1}\mathbf{z}_1^{-1})}\Psi(\mathbf{z}_1,\mathbf{z}_2) \\
&=
e^{-\operatorname{tr}(\mathbf{z}_{2,1}\mathbf{z}_1^{-1})}
\left[-\operatorname{tr}(a\mathbf{z}_1^{-1}\mathbf{z}_2)\Psi(\mathbf{z}_1,\mathbf{z}_2)\right].
\end{aligned} D a [ e − tr ( z 2 , 1 z 1 − 1 ) Ψ ( z 1 , z 2 ) ] = d ε d e − ε tr ( z 2 a z 1 − 1 ) ε = 0 e − tr ( z 2 , 1 z 1 − 1 ) Ψ ( z 1 , z 2 ) = e − tr ( z 2 , 1 z 1 − 1 ) [ − tr ( a z 1 − 1 z 2 ) Ψ ( z 1 , z 2 ) ] .
This formula gives an intertwining relation
D 1 , 2 [ e − tr ( z 2 , 1 z 1 − 1 ) Ψ ( z 1 , z 2 ) ] = e − tr ( z 2 , 1 z 1 − 1 ) [ − z 1 − 1 z 2 Ψ ( z 1 , z 2 ) ] . D_{1,2}\left[e^{-\operatorname{tr}(\mathbf{z}_{2,1}\mathbf{z}_1^{-1})}\Psi(\mathbf{z}_1,\mathbf{z}_2)\right]
=
e^{-\operatorname{tr}(\mathbf{z}_{2,1}\mathbf{z}_1^{-1})}
\left[-\mathbf{z}_1^{-1}\mathbf{z}_2\Psi(\mathbf{z}_1,\mathbf{z}_2)\right]. D 1 , 2 [ e − tr ( z 2 , 1 z 1 − 1 ) Ψ ( z 1 , z 2 ) ] = e − tr ( z 2 , 1 z 1 − 1 ) [ − z 1 − 1 z 2 Ψ ( z 1 , z 2 ) ] .