The research of S.-j. Oh was supported by the National Research Foundation of Korea (NRF) Grant funded by the Korea government (MSIT) (NRF-2022R1A2C1004045).
Introduction
Cluster algebras and their quantum analogues were introduced as classes of algebras equipped with distinguished generators and mutation relations [8 , 1 ] . They have since appeared naturally in representation theory, where cluster monomials are often expected to correspond to distinguished basis elements. Monoidal categorification provides a categorical form of this relationship: cluster variables are represented by real prime simple modules, cluster monomials by tensor products of pairwise strongly commuting real prime simple modules, and exchange relations by short exact sequences [21 ] .
Quiver Hecke algebras were introduced independently in [46 , 47 , 56 ] to categorify the negative half of a quantum group. Their renormalized R R R -matrix calculus for simple modules gives rise to the numerical invariants Λ \Lambda Λ and d \mathfrak{d} d . These invariants have played a central role in proving monoidal categorification conjectures [30 ] . In particular, Λ ( M , N ) \Lambda(M,N) Λ ( M , N ) records the degree of the renormalized R R R -matrix from M ∘ N M \circ N M ∘ N to N ∘ M N \circ M N ∘ M , where the symmetric invariant
d ( M , N ) = Λ ( M , N ) + Λ ( N , M ) 2 \mathfrak{d}(M,N) = \frac{\Lambda(M,N) + \Lambda(N,M)}{2} d ( M , N ) = 2 Λ ( M , N ) + Λ ( N , M )
measures the interaction between M M M and N N N . Under the usual realness assumptions, it detects strong commutation and controls the structure of their convolution product ([28, 30]). Thus formulas for d \mathfrak{d} d -invariants provide concrete information on the multiplicative structure of these monoidal categories.
Despite the simplicity of their definition, d \mathfrak{d} d -invariants are generally quite difficult to compute, since explicit descriptions of simple modules, and often even their dimensions, are unavailable. Generalized quantum affine Schur–Weyl duality provides a way to compute these invariants in certain cases. Our approach is to compute them through tropical cluster invariants and folding.
Let w w w be a Weyl group element and choose a reduced sequence i ∈ Red ( w ) \mathbf{i} \in\operatorname{Red}(w) i ∈ Red ( w ) . In the category C w \mathcal{C}_w C w of quiver Hecke modules associated with w w w , every simple object is obtained, uniquely up to a grading shift, as the head of an ordered convolution product of the cuspidal modules associated with i \mathbf{i} i [45 , 48 , 57 ] . The d \mathfrak{d} d -invariants between these cuspidal modules govern their commutation relations, while the Λ \Lambda Λ -invariants play a central role in determining the normality of the corresponding sequences. Moreover, determinantial modules provide categorical counterparts of quantum minors [34 ] . The determinantial module associated with an i \mathbf{i} i -box, also called a box module, is the unique simple head of the ordered convolution product of the cuspidal modules indexed by the occurrences of its color in that box. The classes of these modules correspond to cluster variables in box seeds for the quantum unipotent coordinate algebra A q ( n ( w ) ) A_q(\mathfrak{n}(w)) A q ( n ( w )) [39 , 33 ] , whose quantum cluster algebra structure was established in [16 , 17 , 18 ] .
A parallel picture appears in the representation theory of quantum affine algebras U q ′ ( g ( 1 ) ) U'_q(\mathfrak{g}^{(1)}) U q ′ ( g ( 1 ) ) [6 , 31 , 40 ] . Fundamental representations play the role of PBW generators on the quantum affine side, and generalized quantum affine Schur–Weyl duality identifies the corresponding quiver Hecke cuspidal modules with them [27 ] . A ubiquitous combinatorial object in this theory is the inverse C ( q ) − 1 C(q)^{-1} C ( q ) − 1 of the quantum Cartan matrix C ( q ) C(q) C ( q ) : its coefficients occur in the study of q q q -characters, quantum Grothendieck rings, the commutative part of universal R R R -matrices, and denominator formulas for normalized R R R -matrices [10 , 50 , 22 , 14 , 12 ] . In this sense, the inverse quantum Cartan matrix provides a common key to several structures in the representation theory of U q ′ ( g ( 1 ) ) U'_q(\mathfrak{g}^{(1)}) U q ′ ( g ( 1 ) ) ; see [12 Introduction] for this point of view.
An especially important family in the representation theory of quantum affine algebras is formed by the Kirillov–Reshetikhin modules (see [49 , 20 , 5 ] and the references therein). In simply-laced finite type, the Schur–Weyl duality associated with a Dynkin quiver Q Q Q realizes the corresponding quiver Hecke modules as determinantial modules. Their d d d -invariants may therefore be studied through both quiver Hecke and quantum affine representation theory ([40]).
On the cluster-algebraic side, Cao introduced seed-independent tropical and F F F -invariants defined in terms of pointed Laurent expansions [2 ] . The F F F -invariant is the symmetrization of the tropical invariant. When a cluster algebra admits a suitable monoidal categorification, the Laurent maximum property gives categorical interpretations of these invariants:
⟨ Φ ( [ M ] ) , Φ ( [ N ] ) ⟩ t r o p = Λ ( N , M ) , ( Φ ( [ M ] ) ∥ Φ ( [ N ] ) ) F = 2 d ( M , N ) . (1) \langle\Phi([M]),\Phi([N])\rangle_{\mathrm{trop}}=\Lambda(N,M),\qquad\left(\Phi([M])\Vert\Phi([N])\right)_{F}=2\mathfrak{d}(M,N). \tag*{(1)} ⟨ Φ ([ M ]) , Φ ([ N ]) ⟩ trop = Λ ( N , M ) , ( Φ ([ M ]) ∥Φ ([ N ]) ) F = 2 d ( M , N ) . (1)
This provides a direct bridge between tropical cluster data and renormalized R R R -matrices. For a general categorification, however, these interpretations of the tropical and F F F -invariants are not automatic. We obtain such an interpretation (1) by applying the maximum formula for Laurent families in [41 Theorem 2.13] to categorical box seeds (Theorem 3.13). We then use cluster-algebra folding [25 ] to derive the corresponding identities in the folded setting from those for the upstairs box modules.
We briefly fix the folding setting used below. Let g ^ \widehat{\mathfrak{g}} g be a simply-laced Kac–Moody algebra with vertex set I ^ \widehat{I} I , equipped with an admissible action of a finite group Γ \Gamma Γ , and let g \mathfrak{g} g be the symmetrizable Kac–Moody algebra associated with the folded Cartan datum. Thus
( g ^ , I ^ ) simply-laced upstairs data ( g ^ , I ^ ) ↓ / Γ ( g , I = I ^ / Γ ) symmetrizable downstairs data . (2) \begin{gathered}
(\widehat{\mathfrak{g}},\widehat{I})\qquad\text{simply-laced upstairs data}\\
\phantom{(\widehat{\mathfrak{g}},\widehat{I})}\downarrow\!/\Gamma\\
(\mathfrak{g},I=\widehat{I}/\Gamma)\qquad\text{symmetrizable downstairs data}. \tag*{(2)}
\end{gathered} ( g , I ) simply-laced upstairs data ( g , I ) ↓ /Γ ( g , I = I /Γ ) symmetrizable downstairs data . (2)
Set
π : I ^ ⟶ I , I i : = π − 1 ( i ) . \pi:\widehat{I}\longrightarrow I,\qquad I_i:=\pi^{-1}(i). π : I ⟶ I , I i := π − 1 ( i ) .
We refer to the data associated with g ^ \widehat{\mathfrak{g}} g and g \mathfrak{g} g as the upstairs and downstairs data, respectively, and usually denote upstairs objects by hatted symbols. We normalize the symmetrizer
D g = diag ( d i ∣ i ∈ I ) D^{\mathfrak{g}}=\operatorname{diag}(d_i\mid i\in I) D g = diag ( d i ∣ i ∈ I )
so that d i ∣ I i ∣ = m d_i|I_i|=m d i ∣ I i ∣ = m is independent of i i i .
Unless explicitly stated otherwise, no finite-type assumption is imposed.
The framework in (2) is provided by i \mathbf{i} i -boxes and i \mathbf{i} i -box seeds. For a reduced sequence i ∈ Red ( w ) \mathbf{i}\in\operatorname{Red}(w) i ∈ Red ( w ) , each i \mathbf{i} i -box indexes a determinantial module in C w \mathcal{C}_w C w , or equivalently a unipotent quantum minor. Maximal commuting families of i \mathbf{i} i -boxes give cluster seeds, while box moves describe the mutations between them [39 , 33 ] (see Lemma 2.1, Proposition 3.8, and Theorem 3.9 ). Thus i \mathbf{i} i -box seeds provide a common language for determinantial modules, cluster variables, and their Laurent expansions.
This box realization is particularly well suited to folding . The cluster-algebra folding developed in [25 ] organizes the upstairs i \mathbf{i} i -boxes into fibers and folds the associated box seeds and cluster variables to their downstairs counterparts. Using this folding, we derive the corresponding F F F -polynomial identity (Proposition 2.4) and recover the tropical and F F F -invariants between downstairs i \mathbf{i} i -box cluster variables from their simply-laced counterparts. At the cluster-algebraic level, these identities hold in the general symmetrizable setting without a finite Cartan type assumption.
The categorical folding formulas do not require a monoidal functor between the upstairs and downstairs categories.
The same folding philosophy also appears in the theory of t t t -quantized Cartan matrices C ( t ) C(t) C ( t ) . In finite Cartan type, the works [42 , 43 , 24 , 23 ] show that t t t -quantized Cartan matrices and their symmetrized inverses encode important representation-theoretic information and behave naturally under folding. This suggests that, for symmetrizable Cartan data not necessarily of finite type, the symmetrized inverse t t t -quantized Cartan matrix should play a fundamental role in the representation theory of quiver Hecke algebras, analogous to that played by the inverse quantum Cartan matrix in the representation theory of U q ′ ( g ( 1 ) ) U_q'(\mathfrak{g}^{(1)}) U q ′ ( g ( 1 ) ) .
We now state our main results more precisely. Our first result is purely cluster-algebraic and requires neither reducedness nor finite Cartan type. For a positive braid b \mathbf{b} b , let i , j ∈ Seq ( b ) \mathbf{i},\mathbf{j}\in\operatorname{Seq}(\mathbf{b}) i , j ∈ Seq ( b ) , and let C \mathcal{C} C and D \mathcal{D} D be admissible chains associated with i \mathbf{i} i and j \mathbf{j} j , respectively. Let c ∈ F ( C ) \mathfrak{c}\in\mathcal{F}(\mathcal{C}) c ∈ F ( C ) , c ′ ∈ F ( D ) \mathfrak{c}'\in\mathcal{F}(\mathcal{D}) c ′ ∈ F ( D ) .
For any c ^ ′ ∈ Fib D ( c ′ ) \widehat{\mathfrak{c}}'\in\operatorname{Fib}_{\mathcal{D}}(\mathfrak{c}') c ′ ∈ Fib D ( c ′ ) , we prove that
⟨ x c , x c ′ ⟩ t r o p = d i c ⟨ x ^ Fib C ( c ) , x ^ c ^ ′ ⟩ t r o p , ⟨ x c ′ , x c ⟩ t r o p = d i c ⟨ x ^ c ^ ′ , x ^ Fib C ( c ) ⟩ t r o p . \langle x_{\mathfrak{c}},x_{\mathfrak{c}'}\rangle_{\mathrm{trop}}
=d_{i_{\mathfrak{c}}}\langle\widehat{x}_{\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c})},\widehat{x}_{\widehat{\mathfrak{c}}'}\rangle_{\mathrm{trop}},
\qquad
\langle x_{\mathfrak{c}'},x_{\mathfrak{c}}\rangle_{\mathrm{trop}}
=d_{i_{\mathfrak{c}}}\langle\widehat{x}_{\widehat{\mathfrak{c}}'},\widehat{x}_{\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c})}\rangle_{\mathrm{trop}}. ⟨ x c , x c ′ ⟩ trop = d i c ⟨ x Fib C ( c ) , x c ′ ⟩ trop , ⟨ x c ′ , x c ⟩ trop = d i c ⟨ x c ′ , x Fib C ( c ) ⟩ trop .
where x ^ Fib C ( c ) : = ∏ c ^ ∈ Fib C ( c ) x ^ c ^ \widehat{x}_{\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c})}:=\prod_{\widehat{\mathfrak{c}}\in\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c})}\widehat{x}_{\widehat{\mathfrak{c}}} x Fib C ( c ) := ∏ c ∈ Fib C ( c ) x c . Adding these two ordered identities gives
( x c ∥ x c ′ ) F = d i c ( x ^ Fib C ( c ) ∥ x ^ c ^ ′ ) F . (x_{\mathfrak{c}}\mathbin{\|}x_{\mathfrak{c}'})_{F}
=d_{i_{\mathfrak{c}}}(\widehat{x}_{\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c})}\mathbin{\|}\widehat{x}_{\widehat{\mathfrak{c}}'})_{F}. ( x c ∥ x c ′ ) F = d i c ( x Fib C ( c ) ∥ x c ′ ) F .
(Theorem 2.6 ; see also (20) –(22) ). The right-hand sides are independent of the choice of c ^ ′ \widehat{\mathfrak{c}}' c ′ . Thus both ordered tropical pairings, and hence their sum defining the F F F -invariant, are recovered from the lifts.
Let i , j ∈ Red ( w ) \mathbf{i},\mathbf{j}\in\operatorname{Red}(w) i , j ∈ Red ( w ) , and let M i [ c ] M^{\mathbf{i}}[\mathfrak{c}] M i [ c ] and M j [ c ′ ] M^{\mathbf{j}}[\mathfrak{c}'] M j [ c ′ ] be the determinantial modules associated with two downstairs i \mathbf{i} i -boxes. Choose c ^ ′ ∈ Fib D ( c ′ ) \widehat{\mathfrak{c}}'\in\operatorname{Fib}_{\mathcal{D}}(\mathfrak{c}') c ′ ∈ Fib D ( c ′ ) . Our categorical interpretation of the tropical invariant gives folding formulas for both orders of the Λ \Lambda Λ -invariant (Theorem 3.15 ). For these box modules, which may be attached to different reduced words, the formulas are
Λ ( M i [ c ] , M j [ c ′ ] ) = d i c ∑ c ^ ∈ Fib C ( c ) Λ ( M ^ i [ c ^ ] , M ^ j [ c ^ ′ ] ) , \Lambda(M^{\mathbf{i}}[\mathfrak{c}],M^{\mathbf{j}}[\mathfrak{c}'])
=d_{i_{\mathfrak{c}}}\sum_{\widehat{\mathfrak{c}}\in\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c})}\Lambda(\widehat{M}^{\mathbf{i}}[\widehat{\mathfrak{c}}],\widehat{M}^{\mathbf{j}}[\widehat{\mathfrak{c}}']), Λ ( M i [ c ] , M j [ c ′ ]) = d i c c ∈ Fib C ( c ) ∑ Λ ( M i [ c ] , M j [ c ′ ]) ,
Λ ( M j [ c ′ ] , M i [ c ] ) = d i c ∑ c ^ ∈ Fib C ( c ) Λ ( M ^ j [ c ^ ′ ] , M ^ i [ c ^ ] ) . \Lambda(M^{\mathbf{j}}[\mathfrak{c}'],M^{\mathbf{i}}[\mathfrak{c}])
=d_{i_{\mathfrak{c}}}\sum_{\widehat{\mathfrak{c}}\in\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c})}\Lambda(\widehat{M}^{\mathbf{j}}[\widehat{\mathfrak{c}}'],\widehat{M}^{\mathbf{i}}[\widehat{\mathfrak{c}}]). Λ ( M j [ c ′ ] , M i [ c ]) = d i c c ∈ Fib C ( c ) ∑ Λ ( M j [ c ′ ] , M i [ c ]) .
as stated in Corollary 3.17. Thus, symmetrizing the two identities gives the corresponding folding formula for ∂ \mathop{\mathord{\partial}} ∂ (Theorem 3.16 ):
∂ ( M i [ c ] , M j [ c ′ ] ) = d i c ∑ c ^ ∈ Fib C ( c ) ∂ ( M ^ i [ c ^ ] , M ^ j [ c ^ ′ ] ) . \mathop{\mathord{\partial}}(M^{\mathbf{i}}[\mathfrak{c}],M^{\mathbf{j}}[\mathfrak{c}'])
=d_{i_{\mathfrak{c}}}\sum_{\widehat{\mathfrak{c}}\in\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c})}\mathop{\mathord{\partial}}(\widehat{M}^{\mathbf{i}}[\widehat{\mathfrak{c}}],\widehat{M}^{\mathbf{j}}[\widehat{\mathfrak{c}}']). ∂ ( M i [ c ] , M j [ c ′ ]) = d i c c ∈ Fib C ( c ) ∑ ∂ ( M i [ c ] , M j [ c ′ ]) .
All three expressions on the right are independent of the chosen lift c ^ ′ \widehat{\mathfrak{c}}' c ′ . Hence the interaction between two determinantial modules downstairs is completely determined by the interactions among their simply-laced lifts. While the cluster-algebraic formula requires neither reducedness nor finite Cartan type, the categorical formula requires reducedness but remains valid for arbitrary symmetrizable Cartan data (Corollary 3.17 and (39) ).
We next establish, without assuming finite Cartan type, a parallel folding formula for the symmetrized inverse t t t -quantized Cartan matrix. Write
C ~ g ( t ) : = D g C g ( t ) − 1 , \widetilde{C}^{\mathfrak g}(t):=D^{\mathfrak g}C^{\mathfrak g}(t)^{-1}, C g ( t ) := D g C g ( t ) − 1 ,
and define C ~ g ( t ) \widetilde{C}^{\mathfrak g}(t) C g ( t ) similarly. For i , j ∈ I i,j\in I i , j ∈ I , choose ı ∈ I i \imath\in I_i ∈ I i and ȷ ∈ I j \jmath\in I_j ∈ I j . We prove that
C ~ i , j g ( t ) = d i ∑ ı ′ ∈ I i C ~ ı ′ , ȷ g ( t ) = m ∣ Γ ∣ ∑ g ∈ Γ C ~ g ( ı ) , ȷ g ( t ) . \widetilde{C}^{\mathfrak g}_{i,j}(t)=d_i\sum_{\imath'\in I_i}\widetilde{C}^{\mathfrak g}_{\imath',\jmath}(t)=\frac{m}{\lvert\Gamma\rvert}\sum_{g\in\Gamma}\widetilde{C}^{\mathfrak g}_{g(\imath),\jmath}(t). C i , j g ( t ) = d i ′ ∈ I i ∑ C ′ , g ( t ) = ∣ Γ ∣ m g ∈ Γ ∑ C g ( ) , g ( t ) .
The expressions on the right are independent of the chosen lifts (Theorem 4.8 and (43) ). For the standard finite foldings
A 2 n − 1 ⟶ C n , D n + 1 ⟶ B n , E 6 ⟶ F 4 , D 4 ⟶ G 2 , A_{2n-1}\longrightarrow C_n,\qquad D_{n+1}\longrightarrow B_n,\qquad E_6\longrightarrow F_4,\qquad D_4\longrightarrow G_2, A 2 n − 1 ⟶ C n , D n + 1 ⟶ B n , E 6 ⟶ F 4 , D 4 ⟶ G 2 ,
one has m = ∣ Γ ∣ m=\lvert\Gamma\rvert m = ∣ Γ ∣ , and hence the final expression is simply the group sum.
In finite type, [43 Theorem 8.2] identify the coefficients of C ~ ( t ) \widetilde{C}(t) C ( t ) with ∂ \partial ∂ -invariants between Q Q Q -cuspidal modules. Under this interpretation, the specialization of our categorical folding formula to cuspidal modules gives the corresponding coefficient identities in (43) . In particular, the same factor d i d_i d i appears in the categorical folding formula and in the folding formula for the symmetrized inverse t t t -Cartan matrix (Theorem 4.11 ).
This agreement leads to our principal finite-type application. Let g \mathfrak g g be of finite type, and let V ( [ i k ] ) p V\left(\left[\begin{smallmatrix}i\\k\end{smallmatrix}\right]\right)_p V ( [ i k ] ) p and V ( [ j ℓ ] ) s V\left(\left[\begin{smallmatrix}j\\\ell\end{smallmatrix}\right]\right)_s V ( [ j ℓ ] ) s be (Q Q Q -adapted) quiver Hecke Kirillov–Reshetikhin modules ((41) ). Assume that all the vertices defining these two modules lie in the Auslander–Reiten quiver Υ Q \Upsilon_Q Υ Q . Set
K : = max ( k , ℓ ) , L : = min ( k , ℓ ) , u : = ∣ p − s ∣ . K:=\max(k,\ell),\qquad L:=\min(k,\ell),\qquad u:=\lvert p-s\rvert. K := max ( k , ℓ ) , L := min ( k , ℓ ) , u := ∣ p − s ∣ .
Let h h h be the Coxeter number of g \mathfrak g g , and set
C ~ i , j g , [ h ] ( t ) : = ∑ v = 1 h c ~ i , j g ( v ) t v , [ L ] t : = t L − 1 + t L − 3 + ⋯ + t 1 − L . \widetilde{C}^{\mathfrak g,[h]}_{i,j}(t):=\sum_{v=1}^{h}\widetilde{c}^{\mathfrak g}_{i,j}(v)t^v,\qquad[L]_t:=t^{L-1}+t^{L-3}+\cdots+t^{1-L}. C i , j g , [ h ] ( t ) := v = 1 ∑ h c i , j g ( v ) t v , [ L ] t := t L − 1 + t L − 3 + ⋯ + t 1 − L .
Writing [ t u ] f ( t ) [t^u]f(t) [ t u ] f ( t ) for the coefficient of t u t^u t u in f ( t ) f(t) f ( t ) , we have
∂ ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = [ t u ] ( t K [ L ] t C ~ i , j g , [ h ] ( t ) ) . (3) \partial\left(V\left(\left[\begin{smallmatrix}i\\k\end{smallmatrix}\right]\right)_p,V\left(\left[\begin{smallmatrix}j\\\ell\end{smallmatrix}\right]\right)_s\right)
=[t^u]\left(t^K[L]_t\widetilde{C}^{\mathfrak g,[h]}_{i,j}(t)\right). \tag*{(3)} ∂ ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = [ t u ] ( t K [ L ] t C i , j g , [ h ] ( t ) ) . (3)
(Theorems 4.12 and 4.18 ; see (45) and (50) ). Thus the ∂ \partial ∂ -invariants between quiver Hecke KR-modules can be read completely from C ~ ( t ) \widetilde{C}(t) C ( t ) in every finite type. Our results therefore support the expectation that this matrix plays a fundamental role in the representation theory of quiver Hecke algebras.
The closed formulas (3) are particularly transparent for the folding A 2 n − 1 ⟶ C n A_{2n-1}\longrightarrow C_n A 2 n − 1 ⟶ C n , where Γ ≃ Z / 2 Z \Gamma\simeq\mathbb{Z}/2\mathbb{Z} Γ ≃ Z /2 Z acts by i ↔ 2 n − i i\leftrightarrow2n-i i ↔ 2 n − i . For r , L ∈ Z > 0 r,L\in\mathbb{Z}_{>0} r , L ∈ Z > 0 and v ∈ Q v\in\mathbb{Q} v ∈ Q , set
T r , L ( 1 ) ( v ) = δ ( v ∈ Z ) [ min ( v , r , L , r + L − v ) ] + , κ : = ∣ k − ℓ ∣ , u : = ∣ p − s ∣ , \mathsf{T}^{(1)}_{r,L}(v)=\delta(v\in\mathbb{Z})\left[\min(v,r,L,r+L-v)\right]_+,\qquad\kappa:=\lvert k-\ell\rvert,\qquad u:=\lvert p-s\rvert, T r , L ( 1 ) ( v ) = δ ( v ∈ Z ) [ min ( v , r , L , r + L − v ) ] + , κ := ∣ k − ℓ ∣ , u := ∣ p − s ∣ ,
where [ a ] + : = max ( a , 0 ) [a]_+:=\max(a,0) [ a ] + := max ( a , 0 ) and δ ( P ) \delta(P) δ ( P ) denotes the indicator of a statement P P P .
In type A n A_n A n , set ρ i , j ( n ) : = min ( i , j , n + 1 − i , n + 1 − j ) \rho^{(n)}_{i,j}:=\min(i,j,n+1-i,n+1-j) ρ i , j ( n ) := min ( i , j , n + 1 − i , n + 1 − j ) . Then we have
∂ ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = T ρ i , j ( n ) , L ( 1 ) ( u − κ − ∣ i − j ∣ 2 ) . \partial\left(V\left(\left[\begin{smallmatrix}i\\k\end{smallmatrix}\right]\right)_p,V\left(\left[\begin{smallmatrix}j\\\ell\end{smallmatrix}\right]\right)_s\right)
=\mathsf{T}^{(1)}_{\rho^{(n)}_{i,j},L}\left(\frac{u-\kappa-\lvert i-j\rvert}{2}\right). ∂ ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = T ρ i , j ( n ) , L ( 1 ) ( 2 u − κ − ∣ i − j ∣ ) .
(Corollary 4.14; see also (48) ).
The folding A 2 n − 1 ⟶ C n A_{2n-1} \longrightarrow C_n A 2 n − 1 ⟶ C n makes the corresponding type C n C_n C n formula equally explicit. If 1 ≤ i , j < n 1 \le i,j < n 1 ≤ i , j < n , then
∂ ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = T ρ i , j ( 2 n − 1 ) , L ( 1 ) ( u − κ − ∣ i − j ∣ 2 ) + T ρ 2 n − i , j ( 2 n − 1 ) , L ( 1 ) ( u − κ − ( 2 n − i − j ) 2 ) . \partial\left(\mathsf{V}\left([i^k]\right)_p,\mathsf{V}\left([j^\ell]\right)_s\right)=\mathsf{T}^{(1)}_{\rho_{i,j}^{(2n-1)},L}\left(\frac{u-\kappa-\lvert i-j\rvert}{2}\right)+\mathsf{T}^{(1)}_{\rho_{2n-i,j}^{(2n-1)},L}\left(\frac{u-\kappa-(2n-i-j)}{2}\right). ∂ ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = T ρ i , j ( 2 n − 1 ) , L ( 1 ) ( 2 u − κ − ∣ i − j ∣ ) + T ρ 2 n − i , j ( 2 n − 1 ) , L ( 1 ) ( 2 u − κ − ( 2 n − i − j ) ) .
If i = n i=n i = n or j = n j=n j = n , where n n n is the long node, then
∂ ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = 2 T ρ i , j ( 2 n − 1 ) , L ( 1 ) ( u − κ − ∣ i − j ∣ 2 ) . \partial\left(\mathsf{V}\left([i^k]\right)_p,\mathsf{V}\left([j^\ell]\right)_s\right)=2\mathsf{T}^{(1)}_{\rho_{i,j}^{(2n-1)},L}\left(\frac{u-\kappa-\lvert i-j\rvert}{2}\right). ∂ ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = 2 T ρ i , j ( 2 n − 1 ) , L ( 1 ) ( 2 u − κ − ∣ i − j ∣ ) .
The two summands in the first formula arise from the upstairs pairs of colors ( i , j ) (i,j) ( i , j ) and ( 2 n − i , j ) (2n-i,j) ( 2 n − i , j ) , respectively, while the factor 2 in the second is the symmetrizer attached to the fixed vertex n n n . Here ρ i , j ( 2 n − 1 ) = ρ 2 n − i , j ( 2 n − 1 ) = min ( i , j ) \rho_{i,j}^{(2n-1)}=\rho_{2n-i,j}^{(2n-1)}=\min(i,j) ρ i , j ( 2 n − 1 ) = ρ 2 n − i , j ( 2 n − 1 ) = min ( i , j ) for 1 ≤ i , j < n 1\le i,j<n 1 ≤ i , j < n . Thus the closed type C n C_n C n formula visibly retains the folding from type A 2 n − 1 A_{2n-1} A 2 n − 1 by Corollary 4.20.
Analogous closed formulas are obtained in types B B B and D D D (Corollaries 4.22 and 4.16). The uniform coefficient formula also applies to the exceptional types F F F and G G G , as illustrated by explicit examples (Examples 4.24 and 4.25). In particular, we obtain new closed formulas for quiver Hecke KR-modules in types B B B and C C C . Together with the formula in Proposition 3.3 , these ∂ \partial ∂ -formulas allow one to compute both orders of the Λ \Lambda Λ -invariant for KR-modules defined by boxes in the same Q Q Q -adapted reduced sequence (see Remark 4.26 and Example 4.26).
Although (3) is proved here only in finite type, the coefficient formula using the full t t t -adic matrix also holds in three special cases for arbitrary symmetrizable Cartan data. For boxes in a Q Q Q -adapted reduced sequence, the same coefficient expression using the full t t t -adic symmetrized inverse t t t -Cartan matrix gives d i d_i d i both for an i -box and its adjacent singleton i -box and for the exchange pair of boxes in a T T T -system. It vanishes for quiver Hecke KR-modules corresponding to commuting i -boxes. In type B 3 ( 1 ) B_3^{(1)} B 3 ( 1 ) , we illustrate this vanishing with a pair of non-nested commuting i -boxes of different colors. Thus C ~ ( t ) \widetilde{C}(t) C ( t ) detects these categorical interactions beyond finite type; see Proposition 4.30 and Example 4.31.
The folding A 5 ⟶ C 3 A_5\longrightarrow C_3 A 5 ⟶ C 3 illustrates the F F F -invariant, categorical, and inverse t t t -Cartan formulas throughout the paper.
Organization
Organization. The paper is organized as follows. Section 1 recalls Cao’s tropical and F F F -invariants. Section 2 studies the folding of i -box seeds, F F F -polynomials, and tropical invariants. Section 3 establishes categorical interpretations of Cao’s invariants and the Λ \Lambda Λ - and ∂ \partial ∂ -folding formulas. Finally, Section 4 introduces quiver Hecke KR-modules, proves the folding formula for symmetrized inverse t t t -Cartan matrices, derives finite-type ∂ \partial ∂ -formulas, and studies three special cases of the coefficient formula beyond finite type.
Convention
Convention. Throughout this paper, we use the following conventions.
For a statement P P P , we set δ ( P ) \delta(P) δ ( P ) to be 1 or 0 depending on whether P P P is true or not. In particular, δ i , j = δ ( i = j ) \delta_{i,j}=\delta(i=j) δ i , j = δ ( i = j ) is the Kronecker delta.
For integers a ≤ b a\le b a ≤ b , we write [ a , b ] = { a , a + 1 , … , b } [a,b]=\{a,a+1,\ldots,b\} [ a , b ] = { a , a + 1 , … , b } . We understand [ a , b ] = ∅ [a,b]=\varnothing [ a , b ] = ∅ if b < a b<a b < a and set [ a ] : = [ a , a ] [a]:=[a,a] [ a ] := [ a , a ] .
For a power series f ( t ) = ∑ i = 0 ∞ a i t i f(t)=\sum_{i=0}^{\infty}a_i t^i f ( t ) = ∑ i = 0 ∞ a i t i over a commutative ring A A A , we set [ t n ] f ( t ) : = a n for n ∈ Z ≥ 0 [t^n]f(t):=a_n\quad\text{for }n\in\mathbb{Z}_{\ge0} [ t n ] f ( t ) := a n for n ∈ Z ≥ 0 .
Tropical and F-invariants
In this section, we recall Cao’s tropical and F F F -invariants [2 ] for compatibly pointed elements of a Λ \Lambda Λ -cluster algebra. Unless otherwise stated, neither finite type nor symmetry of the underlying Cartan datum is assumed. Their categorical interpretations in terms of the Λ \Lambda Λ - and δ \delta δ -invariants for box seeds will be established in § 4 .
We follow [8 , 1 , 7 ] for compatible pairs and Λ \Lambda Λ -cluster patterns, and [54 , 2 , 32 , 38 ] for dominance orders and pointed elements.
Let J = J e x ⊔ J f r J=J_{\mathrm{ex}}\sqcup J_{\mathrm{fr}} J = J ex ⊔ J fr be a finite index set, where J e x J_{\mathrm{ex}} J ex and J f r J_{\mathrm{fr}} J fr are the exchangeable and frozen indices, respectively.
B = ( b i , j ) i ∈ J , j ∈ J e x B=(b_{i,j})_{i\in J,j\in J_{\mathrm{ex}}} B = ( b i , j ) i ∈ J , j ∈ J ex be an integer matrix whose principal part
B p r i n : = ( b i , j ) i , j ∈ J e x B^{\mathrm{prin}} := (b_{i,j})_{i,j\in J_{\mathrm{ex}}} B prin := ( b i , j ) i , j ∈ J ex is skew-symmetrizable,
L = ( l i , j ) i , j ∈ J L=(l_{i,j})_{i,j\in J} L = ( l i , j ) i , j ∈ J be a skew-symmetric integer matrix.
The pair ( L , B ) (L,B) ( L , B ) is compatible of type S = diag ( s j ∣ j ∈ J e x ) S=\operatorname{diag}(s_j\mid j\in J_{\mathrm{ex}}) S = diag ( s j ∣ j ∈ J ex ) if
B T L = ( S ∣ 0 ) ( s j ∈ 2 Z > 0 ) . (4) B^{T}L=(S\mid0)\qquad(s_j\in2\mathbb{Z}_{>0}). \tag*{(4)} B T L = ( S ∣ 0 ) ( s j ∈ 2 Z > 0 ) . (4)
Here the block decomposition on the right-hand side is taken with respect to J = J e x ⊔ J f r J=J_{\mathrm{ex}}\sqcup J_{\mathrm{fr}} J = J ex ⊔ J fr .
Set F : = C ( X j ∣ j ∈ J ) \mathbb{F}:=\mathbb{C}(X_j\mid j\in J) F := C ( X j ∣ j ∈ J ) , where the X j X_j X j are algebraically independent. A Λ \Lambda Λ -seed of type S S S is a triple
Σ = ( x , L , B ) , x = ( x j ) j ∈ J , \Sigma=(\mathbf{x},L,B),\qquad\mathbf{x}=(x_j)_{j\in J}, Σ = ( x , L , B ) , x = ( x j ) j ∈ J ,
where x \mathbf{x} x is algebraically independent over C \mathbb{C} C , F = C ( x j ∣ j ∈ J ) \mathbb{F}=\mathbb{C}(x_j\mid j\in J) F = C ( x j ∣ j ∈ J ) , and ( L , B ) (L,B) ( L , B ) is compatible of type S S S .
For k ∈ J e x k\in J_{\mathrm{ex}} k ∈ J ex , its mutation
μ k ( Σ ) = ( μ k ( x ) , μ k ( L ) , μ k ( B ) ) \mu_k(\Sigma)=\bigl(\mu_k(\mathbf{x}),\mu_k(L),\mu_k(B)\bigr) μ k ( Σ ) = ( μ k ( x ) , μ k ( L ) , μ k ( B ) )
is defined by the usual seed and compatible-pair mutations. The mutated cluster μ k ( x ) = ( x j ′ ) j ∈ J \mu_k(\mathbf{x})=(x'_j)_{j\in J} μ k ( x ) = ( x j ′ ) j ∈ J is given by
x j ′ = x j ( j ≠ k ) , x k x k ′ = ∏ i ∈ J x i [ b i , k ] + + ∏ i ∈ J x i [ − b i , k ] + . x'_j=x_j\quad(j\ne k),\qquad
x_kx'_k=\prod_{i\in J}x_i^{[b_{i,k}]_+}+\prod_{i\in J}x_i^{[-b_{i,k}]_+}. x j ′ = x j ( j = k ) , x k x k ′ = i ∈ J ∏ x i [ b i , k ] + + i ∈ J ∏ x i [ − b i , k ] + .
In particular, μ k ( L ) \mu_k(L) μ k ( L ) and μ k ( B ) \mu_k(B) μ k ( B ) are again compatible of type S S S . We call x k ′ x'_k x k ′ the cluster variable obtained by mutation in direction k k k .
A Λ \Lambda Λ -cluster pattern of type S S S is a family of Λ \Lambda Λ -seeds
Σ t = ( x t , L t , B t ) ( t ∈ T ) , \Sigma_t=(\mathbf{x}_t,L_t,B_t)\qquad(t\in\mathbb{T}), Σ t = ( x t , L t , B t ) ( t ∈ T ) ,
indexed by the vertices of the J e x J_{\mathrm{ex}} J ex -regular tree T \mathbb{T} T , such that adjacent seeds are related by the mutation labeling the corresponding edge. For each t ∈ T t\in\mathbb{T} t ∈ T , let
L ( Σ t ) : = C [ x j ; t ± 1 ∣ j ∈ J ] ⊂ F \mathcal{L}(\Sigma_t):=\mathbb{C}[x_{j;t}^{\pm1}\mid j\in J]\subset\mathbb{F} L ( Σ t ) := C [ x j ; t ± 1 ∣ j ∈ J ] ⊂ F
be the Laurent polynomial ring associated with Σ t \Sigma_t Σ t . The associated Λ \Lambda Λ -cluster algebra is the C \mathbb{C} C -subalgebra1
A : = C [ x j ; t ∣ j ∈ J , t ∈ T ] ⊂ F . \mathcal{A}:=\mathbb{C}[x_{j;t}\mid j\in J,\ t\in\mathbb{T}]\subset\mathbb{F}. A := C [ x j ; t ∣ j ∈ J , t ∈ T ] ⊂ F .
By the Laurent phenomenon [8 ] , one has
A ⊂ L ( Σ t ) ( t ∈ T ) and hence A ⊂ ⋂ t ∈ T L ( Σ t ) ⊂ F . (5) \mathcal{A} \subset\mathcal{L}(\Sigma_t) \quad(t \in\mathbb{T}) \quad\text{and hence} \quad\mathcal{A} \subset\bigcap_{t\in\mathbb{T}}\mathcal{L}(\Sigma_t) \subset\mathcal{F}. \tag*{(5)} A ⊂ L ( Σ t ) ( t ∈ T ) and hence A ⊂ t ∈ T ⋂ L ( Σ t ) ⊂ F . (5)
An element of the form
x t a : = ∏ j ∈ J x j ; t a j ( a = ( a j ) j ∈ J ∈ Z ≥ 0 ⊕ J ) \mathbf{x}^{\mathbf{a}}_t := \prod_{j\in J} x_{j;t}^{a_j} \qquad\left(\mathbf{a}=(a_j)_{j\in J}\in\mathbb{Z}_{\geq0}^{\oplus J}\right) x t a := j ∈ J ∏ x j ; t a j ( a = ( a j ) j ∈ J ∈ Z ≥ 0 ⊕ J )
is called a cluster monomial .
For t ∈ T t\in\mathbb{T} t ∈ T and j ∈ J e x j\in J_{\mathrm{ex}} j ∈ J ex , set 3
y ˙ j ; t : = ∏ i ∈ J x i ; t b i , j ; t , y ˙ t : = ( y ˙ j ; t ) j ∈ J e x . \dot{y}_{j;t}:=\prod_{i\in J}x_{i;t}^{b_{i,j;t}}, \qquad\dot{\mathbf{y}}_t:=(\dot{y}_{j;t})_{j\in J_{\mathrm{ex}}}. y ˙ j ; t := i ∈ J ∏ x i ; t b i , j ; t , y ˙ t := ( y ˙ j ; t ) j ∈ J ex .
For g , g ′ ∈ Z ⊕ J \mathbf{g},\mathbf{g}'\in\mathbb{Z}^{\oplus J} g , g ′ ∈ Z ⊕ J , we write
g ′ ⪯ t g if g ′ = g + B t v for some v ∈ Z ≥ 0 ⊕ J e x . \mathbf{g}'\preceq_t\mathbf{g}\qquad\text{if }\mathbf{g}'=\mathbf{g}+B_t\mathbf{v}\text{ for some }\mathbf{v}\in\mathbb{Z}_{\geq0}^{\oplus J_{\mathrm{ex}}}. g ′ ⪯ t g if g ′ = g + B t v for some v ∈ Z ≥ 0 ⊕ J ex .
Equivalently,
x t g ′ = x t g y ˙ t v , where x t g : = ∏ i ∈ J x i ; t g i and y ˙ t v : = ∏ j ∈ J e x y ˙ j ; t v j . \mathbf{x}^{\mathbf{g}'}_t=\mathbf{x}^{\mathbf{g}}_t\dot{\mathbf{y}}_t^{\mathbf{v}},\qquad\text{where }\mathbf{x}^{\mathbf{g}}_t:=\prod_{i\in J}x_{i;t}^{g_i}\text{ and }\dot{\mathbf{y}}_t^{\mathbf{v}}:=\prod_{j\in J_{\mathrm{ex}}}\dot{y}_{j;t}^{v_j}. x t g ′ = x t g y ˙ t v , where x t g := i ∈ J ∏ x i ; t g i and y ˙ t v := j ∈ J ex ∏ y ˙ j ; t v j .
The compatibility condition (4) implies that the columns of B t B_t B t are linearly independent, and hence ⪯ t \preceq_t ⪯ t is a partial order.
By the Laurent phenomenon (5) , every u ∈ A u\in\mathcal{A} u ∈ A has a finite expansion
u = ∑ h ∈ Z ⊕ J c h x t h , c h ∈ C . (6) u=\sum_{\mathbf{h}\in\mathbb{Z}^{\oplus J}}c_{\mathbf{h}}\mathbf{x}^{\mathbf{h}}_t,\qquad c_{\mathbf{h}}\in\mathbb{C}. \tag*{(6)} u = h ∈ Z ⊕ J ∑ c h x t h , c h ∈ C . (6)
Following the terminology of pointed and copointed elements, we say that u u u is pointed with respect to t t t if there exists a unique vector deg t ( u ) ∈ Z ⊕ J \deg_t(u)\in\mathbb{Z}^{\oplus J} deg t ( u ) ∈ Z ⊕ J such that
c deg t ( u ) = 1 , c h ≠ 0 ⟹ h ⪯ t deg t ( u ) . c_{\deg_t(u)}=1,\qquad c_{\mathbf{h}}\ne0\Longrightarrow\mathbf{h}\preceq_t\deg_t(u). c d e g t ( u ) = 1 , c h = 0 ⟹ h ⪯ t deg t ( u ) .
Dually, u u u is copointed with respect to t t t if there exists a unique vector \codeg_t(u)\in\mathbb{Z}^{\oplus J} such that
c_{\codeg_t(u)}=1,\qquad c_{\mathbf{h}}\ne0\Longrightarrow\codeg_t(u)\preceq_t\mathbf{h}.
We call deg t ( u ) \deg_t(u) deg t ( u ) and \codeg_t(u) the degree and codegree of u u u with respect to t t t , respectively. An element is bipointed with respect to t t t if it is both pointed and copointed with respect to t t t .
Since the columns of B t B_t B t are linearly independent, every pointed element u u u has a unique expansion
u = x t deg t ( u ) F u ; t ( y ˙ t ) . (7) u=\mathbf{x}_t^{\deg_t(u)}F_{u;t}(\dot{\mathbf{y}}_t). \tag*{(7)} u = x t d e g t ( u ) F u ; t ( y ˙ t ) . (7)
Here
F u ; t ( y ) ∈ C [ y j ∣ j ∈ J e x ] , F u ; t ( 0 ) = 1. F_{u;t}(\mathbf{y})\in\mathbb{C}[y_j\mid j\in J_{\mathrm{ex}}],\qquad F_{u;t}(0)=1. F u ; t ( y ) ∈ C [ y j ∣ j ∈ J ex ] , F u ; t ( 0 ) = 1.
We call F u ; t ( y ) F_{u;t}(\mathbf{y}) F u ; t ( y ) the F F F -polynomial of u u u with respect to t t t .
For j ∈ J e x j\in J_{\mathrm{ex}} j ∈ J ex , let f j ; u ; t f_{j;u;t} f j ; u ; t be the maximal exponent of y j y_j y j among the monomials occurring in F u ; t ( y ) F_{u;t}(\mathbf{y}) F u ; t ( y ) , and set
f u ; t : = ( f j ; u ; t ) j ∈ J e x ∈ Z ≥ 0 J e x . \mathbf{f}_{u;t}:=(f_{j;u;t})_{j\in J_{\mathrm{ex}}}\in\mathbb{Z}_{\geq0}^{J_{\mathrm{ex}}}. f u ; t := ( f j ; u ; t ) j ∈ J ex ∈ Z ≥ 0 J ex .
We call f u ; t f_{u;t} f u ; t the f f f -vector of u u u with respect to t t t . Then u u u is bipointed with respect to t t t if and only if y f u ; t y^{f_{u;t}} y f u ; t occurs in F u ; t ( y ) F_{u;t}(\mathbf{y}) F u ; t ( y ) with coefficient 1. In this case,
codeg t ( u ) = deg t ( u ) + B t f u ; t . \operatorname{codeg}_{t}(u)=\operatorname{deg}_{t}(u)+B_{t}f_{u;t}. codeg t ( u ) = deg t ( u ) + B t f u ; t .
Following Qin’s notion of compatibly pointed elements [55 ] , as recalled in [2 Definition 3.6] , we say that u ∈ A u\in\mathcal{A} u ∈ A is compatibly pointed if it is pointed with respect to every t ∈ T t\in\mathbb{T} t ∈ T and its degree family forms a tropical point of the cluster pattern. In this case, the family
[ g u ] : = ( deg t ( u ) ) t ∈ T [\mathbf{g}_{u}]:=\left(\operatorname{deg}_{t}(u)\right)_{t\in\mathbb{T}} [ g u ] := ( deg t ( u ) ) t ∈ T
is called the tropical point of u u u . A compatibly pointed element is called bipointed if it is bipointed with respect to every t ∈ T t\in\mathbb{T} t ∈ T .
Proposition 1.1 ([9 , 19 ] (see also [2 Proposition 3.7] ). Every cluster monomial in A \mathcal{A} A is compatibly bipointed.
A compatibly pointed element u ∈ A u\in\mathcal{A} u ∈ A is called good if it is universally positive ; that is, its Laurent expansion with respect to every seed belongs to
Z ≥ 0 [ x j ; t ± 1 ∣ j ∈ J ] . \mathbb{Z}_{\geq0}[x_{j;t}^{\pm1}\mid j\in J]. Z ≥ 0 [ x j ; t ± 1 ∣ j ∈ J ] .
In particular, every cluster monomial is good (see [2 Proposition 3.7] ). For a nonzero polynomial F ( y ) = ∑ v c v y v F(\mathbf{y})=\sum_{\mathbf{v}}c_{\mathbf{v}}\mathbf{y}^{\mathbf{v}} F ( y ) = ∑ v c v y v and r ∈ Z ⊕ J e x \mathbf{r}\in\mathbb{Z}^{\oplus J_{\mathrm{ex}}} r ∈ Z ⊕ J ex , set
F [ r ] : = max { v T r ∣ c v ≠ 0 } . F[\mathbf{r}]:=\max\{\mathbf{v}^{T}\mathbf{r}\mid c_{\mathbf{v}}\ne0\}. F [ r ] := max { v T r ∣ c v = 0 } .
If F ( 0 ) = 1 F(\mathbf{0})=1 F ( 0 ) = 1 , then F [ r ] ≥ 0 F[\mathbf{r}]\geq0 F [ r ] ≥ 0 .
Let u , u ′ ∈ A u,u'\in\mathcal{A} u , u ′ ∈ A be good. For t ∈ T t\in\mathbb{T} t ∈ T , set
g u ; t : = deg t ( u ) , S g u ; t : = ( s j g j ; u ; t ) j ∈ J e x . \mathbf{g}_{u;t}:=\operatorname{deg}_{t}(u),\qquad S\mathbf{g}_{u;t}:=(s_jg_{j;u;t})_{j\in J_{\mathrm{ex}}}. g u ; t := deg t ( u ) , S g u ; t := ( s j g j ; u ; t ) j ∈ J ex .
We regard A \mathcal{A} A as a subalgebra of the corresponding upper cluster algebra, in which Cao’s tropical and F F F -invariants are defined.
Definition 1.2. [2 Definition 4.6] For t ∈ T t\in\mathbb{T} t ∈ T , we define
⟨ u , u ′ ⟩ t r o p , t : = g u ; t T L t g u ′ ; t + F u ; t [ S g u ′ ; t ] \langle u,u'\rangle_{\mathrm{trop},t}:=\mathbf{g}_{u;t}^{T}L_{t}\mathbf{g}_{u';t}+F_{u;t}[S\mathbf{g}_{u';t}] ⟨ u , u ′ ⟩ trop , t := g u ; t T L t g u ′ ; t + F u ; t [ S g u ′ ; t ] and
( u ∥ u ′ ) F , t : = ⟨ u , u ′ ⟩ t r o p , t + ⟨ u ′ , u ⟩ t r o p , t . (u\Vert u')_{F,t}:=\langle u,u'\rangle_{\mathrm{trop},t}+\langle u',u\rangle_{\mathrm{trop},t}. ( u ∥ u ′ ) F , t := ⟨ u , u ′ ⟩ trop , t + ⟨ u ′ , u ⟩ trop , t .
Theorem 1.3 ([2 Theorem 4.10] ). The quantities in Definition 1.2 are independent of t t t . Moreover,
( u ∥ u ′ ) F = F u ; t [ S g u ′ ; t ] + F u ′ ; t [ S g u ; t ] ∈ 2 Z ≥ 0 . (u\Vert u')_{F}=F_{u;t}[S\mathbf{g}_{u';t}]+F_{u';t}[S\mathbf{g}_{u;t}]\in2\mathbb{Z}_{\geq0}. ( u ∥ u ′ ) F = F u ; t [ S g u ′ ; t ] + F u ′ ; t [ S g u ; t ] ∈ 2 Z ≥ 0 .
The evenness follows from s j ∈ 2 Z > 0 s_j\in2\mathbb{Z}_{>0} s j ∈ 2 Z > 0 . Following [3 ] , we call
F u ; t [ S g u ′ ; t ] (8) F_{u;t}[S\mathbf{g}_{u';t}] \tag*{(8)} F u ; t [ S g u ′ ; t ] (8)
the partial F F F -invariant of ( u , u ′ ) (u,u') ( u , u ′ ) with respect to t t t . Unlike ( u ∥ u ′ ) F (u\Vert u')_{F} ( u ∥ u ′ ) F , it may depend on t t t in general. If k ∈ J e x k\in J_{\mathrm{ex}} k ∈ J ex , then
g x k ; t ; t = e k and F x k ; t ; t = 1. \mathbf{g}_{x_{k;t};t}=\mathbf{e}_{k}\qquad\text{and}\qquad F_{x_{k;t};t}=1. g x k ; t ; t = e k and F x k ; t ; t = 1.
Hence, for every good u ∈ A u\in\mathcal{A} u ∈ A ,
( x k ; t ∥ u ) F = F u ; t [ S e k ] = s k f k ; u ; t . (9) (x_{k;t}\Vert u)_{F}=F_{u;t}[S\mathbf{e}_{k}]=s_kf_{k;u;t}. \tag*{(9)} ( x k ; t ∥ u ) F = F u ; t [ S e k ] = s k f k ; u ; t . (9)
Folding of cluster algebras and tropical invariants
In this section, we establish folding formulas for Cao’s tropical and F F F -invariants in cluster algebras associated with positive braids. Starting from an admissible folding of Cartan data, we compare the compatible pairs and cluster variables attached to corresponding i \mathbf{i} i -box seeds. We then show that F F F -polynomials fold under the substitution induced by the box fibers and derive folding formulas for both orders of the tropical invariant between i \mathbf{i} i -box cluster variables. Their symmetrization gives the corresponding formula for the F F F -invariant. Neither the Cartan datum nor its simply-laced cover is assumed to be of finite type.
Admissible foldings and lifted sequences
We follow the folding conventions of [25 Sections 2.1–2.3] , to which we refer for further details.
Let g ^ \widehat{\mathfrak{g}} g be a simply-laced Kac–Moody algebra with index set I ^ \widehat{I} I and Cartan matrix
C ^ = ( c ^ ı , ȷ ) ı , ȷ ∈ I ^ . \widehat{C}=(\widehat{c}_{\imath,\jmath})_{\imath,\jmath\in\widehat{I}}. C = ( c , ) , ∈ I .
Let ( α ^ ı ) ı ∈ I ^ (\widehat{\alpha}_{\imath})_{\imath\in\widehat{I}} ( α ) ∈ I be its simple roots, and set
Q ^ : = ⨁ ı ∈ I ^ Z α ^ ı , Q ^ + : = ∑ ı ∈ I ^ Z ≥ 0 α ^ ı , Q ^ − : = − Q ^ + . \widehat{Q}:=\bigoplus_{\imath\in\widehat{I}}\mathbb{Z}\widehat{\alpha}_{\imath},\qquad
\widehat{Q}_{+}:=\sum_{\imath\in\widehat{I}}\mathbb{Z}_{\geq0}\widehat{\alpha}_{\imath},\qquad
\widehat{Q}_{-}:=-\widehat{Q}_{+}. Q := ∈ I ⨁ Z α , Q + := ∈ I ∑ Z ≥ 0 α , Q − := − Q + .
Let W g ^ W_{\widehat{\mathfrak{g}}} W g and B g ^ + B_{\widehat{\mathfrak{g}}}^{+} B g + be the Weyl group and the positive braid monoid of g ^ \widehat{\mathfrak{g}} g , with standard generators s ı s_{\imath} s and r ı r_{\imath} r (ı ∈ I ^ \imath\in\widehat{I} ∈ I ), respectively.
Let Δ ^ \widehat{\Delta} Δ be the root system of g ^ \widehat{\mathfrak{g}} g , and set Δ ^ + : = Δ ^ ∩ Q ^ + \widehat{\Delta}_{+}:=\widehat{\Delta}\cap\widehat{Q}_{+} Δ + := Δ ∩ Q + . Let
Δ ^ r e : = { w ^ ( α ^ ı ) ∣ w ^ ∈ W g ^ , ı ∈ I ^ } \widehat{\Delta}_{\mathrm{re}}:=\{\widehat{w}(\widehat{\alpha}_{\imath})\mid\widehat{w}\in W_{\widehat{\mathfrak{g}}},\ \imath\in\widehat{I}\} Δ re := { w ( α ) ∣ w ∈ W g , ∈ I }
be the set of real roots, and set
Δ ^ r e , + : = Δ ^ r e ∩ Δ ^ + , Δ ^ r e , − : = − Δ ^ r e , + . \widehat{\Delta}_{\mathrm{re},+}:=\widehat{\Delta}_{\mathrm{re}}\cap\widehat{\Delta}_{+},\qquad
\widehat{\Delta}_{\mathrm{re},-}:=-\widehat{\Delta}_{\mathrm{re},+}. Δ re , + := Δ re ∩ Δ + , Δ re , − := − Δ re , + .
Let Γ \Gamma Γ be a finite group acting on I ^ \widehat{I} I by diagram automorphisms. We assume that the action is admissible , that is,
c ^ ı , g ( ı ) = 0 ( ı ∈ I ^ , g ∈ Γ , g ( ı ) ≠ ı ) . \widehat{c}_{\imath,g(\imath)}=0\qquad(\imath\in\widehat{I},\ g\in\Gamma,\ g(\imath)\ne\imath). c , g ( ) = 0 ( ∈ I , g ∈ Γ , g ( ) = ) .
Set
I : = I ^ / Γ , π : I ^ ⟶ I , I i : = π − 1 ( i ) , e i : = ∣ I i ∣ . I:=\widehat{I}/\Gamma,\qquad\pi:\widehat{I}\longrightarrow I,\qquad I_i:=\pi^{-1}(i),\qquad e_i:=|I_i|. I := I /Γ , π : I ⟶ I , I i := π − 1 ( i ) , e i := ∣ I i ∣.
Thus the vertices belonging to each I i I_i I i are pairwise nonadjacent. To distinguish the two index sets, we generally use ı , ȷ ∈ I ^ \imath,\jmath\in\widehat{I} , ∈ I for the simply-laced datum and i , j ∈ I i,j\in I i , j ∈ I for the quotient datum.
The quotient Cartan matrix C = ( c i , j ) i , j ∈ I C=(c_{i,j})_{i,j\in I} C = ( c i , j ) i , j ∈ I is defined by
c i , j : = ∑ ı ∈ I i c ^ ı , ȷ , ȷ ∈ I j . c_{i,j}:=\sum_{\imath\in I_i}\widehat{c}_{\imath,\jmath},\qquad\jmath\in I_j. c i , j := ∈ I i ∑ c , , ∈ I j .
The right-hand side is independent of the choice of ȷ ∈ I j \jmath\in I_j ∈ I j . Let
D = diag ( d i ∣ i ∈ I ) D=\operatorname{diag}(d_i\mid i\in I) D = diag ( d i ∣ i ∈ I )
be the normalized symmetrizer of C C C . Then there exists m ∈ Z > 0 m\in\mathbb{Z}_{>0} m ∈ Z > 0 such that
d i e i = m ( i ∈ I ) . (10) d_i e_i=m\qquad(i\in I). \tag*{(10)} d i e i = m ( i ∈ I ) . (10)
We denote by g \mathfrak{g} g the symmetrizable Kac–Moody algebra associated with C C C . Let ( α i ) i ∈ I (\alpha_i)_{i\in I} ( α i ) i ∈ I be its simple roots, and set
Q : = ⨁ i ∈ I Z α i , Q + : = ∑ i ∈ I Z ≥ 0 α i , Q − : = − Q + . Q := \bigoplus_{i\in I} \mathbb{Z}\alpha_i,\qquad Q_+ := \sum_{i\in I} \mathbb{Z}_{\geq0}\alpha_i,\qquad Q_- := -Q_+. Q := i ∈ I ⨁ Z α i , Q + := i ∈ I ∑ Z ≥ 0 α i , Q − := − Q + .
We use the same symbol π \pi π for the induced map of root lattices
π : Q ^ ⟶ Q , α ^ ı ⟼ α π ( ı ) . \pi:\widehat{Q}\longrightarrow Q,\qquad\widehat{\alpha}_{\imath}\longmapsto\alpha_{\pi(\imath)}. π : Q ⟶ Q , α ⟼ α π ( ) .
When g ^ \widehat{\mathfrak{g}} g is of finite type, the nontrivial admissible foldings used below are summarized as follows:
Table 2.2.
Here Γ \Gamma Γ acts by the standard diagram automorphisms.
We equip Q Q Q with the symmetric bilinear form
( α i , α j ) g : = d i c i , j ( i , j ∈ I ) . (\alpha_i,\alpha_j)_{\mathfrak{g}} := d_i c_{i,j}\qquad(i,j\in I). ( α i , α j ) g := d i c i , j ( i , j ∈ I ) .
Let W g W_{\mathfrak{g}} W g and B g + B_{\mathfrak{g}}^+ B g + be the Weyl group and the positive braid monoid of g \mathfrak{g} g , with standard generators s i s_i s i and r i r_i r i , respectively. For i ∈ I i\in I i ∈ I , set
s ^ i : = ∏ ı ∈ I i s ı ∈ W g ^ , r ^ i : = ∏ ı ∈ I i r ı ∈ B g ^ + . \widehat{s}_i := \prod_{\imath\in I_i}s_{\imath}\in W_{\widehat{\mathfrak{g}}},\qquad
\widehat{r}_i := \prod_{\imath\in I_i}r_{\imath}\in B_{\widehat{\mathfrak{g}}}^+. s i := ∈ I i ∏ s ∈ W g , r i := ∈ I i ∏ r ∈ B g + .
The products are independent of the ordering of I i I_i I i by admissibility. The assignments
s i ⟼ s ^ i , r i ⟼ r ^ i s_i\longmapsto\widehat{s}_i,\qquad r_i\longmapsto\widehat{r}_i s i ⟼ s i , r i ⟼ r i
induce embeddings
W g ↪ W g ^ , B g + ↪ B g ^ + . W_{\mathfrak{g}}\hookrightarrow W_{\widehat{\mathfrak{g}}},\qquad
B_{\mathfrak{g}}^+\hookrightarrow B_{\widehat{\mathfrak{g}}}^+. W g ↪ W g , B g + ↪ B g + .
See [4 Theorem 28] and [25 Corollary 2.5] .
Now let i = ( i 1 , … , i l ) ∈ I l \boldsymbol{i}=(i_1,\ldots,i_l)\in I^l i = ( i 1 , … , i l ) ∈ I l be an arbitrary finite sequence. Fix an ordering
I i = ( ı i , 1 , … , ı i , e i ) ( i ∈ I ) , I_i=(\imath_{i,1},\ldots,\imath_{i,e_i})\qquad(i\in I), I i = ( i , 1 , … , i , e i ) ( i ∈ I ) ,
and define the blockwise lift of i \boldsymbol{i} i by
i ^ : = I i 1 ⋆ ⋯ ⋆ I i l = ( ı i 1 , 1 , … , ı i 1 , e i 1 , … , ı i l , 1 , … , ı i l , e i l ) . (11) \widehat{\boldsymbol{i}}:=I_{i_1}\star\cdots\star I_{i_l}
=(\imath_{i_1,1},\ldots,\imath_{i_1,e_{i_1}},\ldots,\imath_{i_l,1},\ldots,\imath_{i_l,e_{i_l}}). \tag*{(11)} i := I i 1 ⋆ ⋯ ⋆ I i l = ( i 1 , 1 , … , i 1 , e i 1 , … , i l , 1 , … , i l , e i l ) . (11)
Here ⋆ \star ⋆ denotes concatenation of the ordered blocks. Set
N 0 : = 0 , N k : = ∑ a = 1 k e i a , κ ( k , r ) : = N k − 1 + r N_0:=0,\qquad N_k:=\sum_{a=1}^{k}e_{i_a},\qquad\kappa(k,r):=N_{k-1}+r N 0 := 0 , N k := a = 1 ∑ k e i a , κ ( k , r ) := N k − 1 + r
for 1 ≤ k ≤ l 1\leq k\leq l 1 ≤ k ≤ l and 1 ≤ r ≤ e i k 1\leq r\leq e_{i_k} 1 ≤ r ≤ e i k . The integer κ ( k , r ) \kappa(k,r) κ ( k , r ) is the position in i ^ \widehat{\boldsymbol{i}} i of the r r r -th entry of the k k k -th block I i k I_{i_k} I i k .
Writing r i : = r i 1 ⋯ r i l r_{\boldsymbol{i}}:=r_{i_1}\cdots r_{i_l} r i := r i 1 ⋯ r i l , the lifted sequence represents r ^ i : = r ^ i 1 ⋯ r ^ i l \widehat{r}_{\boldsymbol{i}}:=\widehat{r}_{i_1}\cdots\widehat{r}_{i_l} r i := r i 1 ⋯ r i l . Consequently, if b = r i ∈ B g + b=r_{\boldsymbol{i}}\in B_{\mathfrak{g}}^+ b = r i ∈ B g + , we denote its canonical lift by b ^ : = r ^ i ∈ B g ^ + \widehat{b}:=\widehat{r}_{\boldsymbol{i}}\in B_{\widehat{\mathfrak{g}}}^+ b := r i ∈ B g + . Moreover, i \boldsymbol{i} i is reduced if and only if i ^ \widehat{\boldsymbol{i}} i is reduced.
For b ∈ B g + b\in B_{\mathfrak{g}}^+ b ∈ B g + , we write Seq ( b ) \operatorname{Seq}(b) Seq ( b ) for the set of finite sequences representing b b b .
For later use, define the root sequence of i \mathbf{i} i by
β k i : = s i 1 ⋯ s i k − 1 ( α i k ) ( 1 ≤ k ≤ l ) . (12) \beta_k^{\mathbf{i}}:=s_{i_1}\cdots s_{i_{k-1}}(\alpha_{i_k}) \qquad(1\leq k\leq l). \tag*{(12)} β k i := s i 1 ⋯ s i k − 1 ( α i k ) ( 1 ≤ k ≤ l ) . (12)
In general, the real roots β k i \beta_k^{\mathbf{i}} β k i need not be positive or pairwise distinct. We also write
β ^ k , r : = β ^ κ ( k , r ) i ( 1 ≤ r ≤ e i k ) . \widehat{\beta}_{k,r}:=\widehat{\beta}^{\mathbf{i}}_{\kappa(k,r)} \qquad(1\leq r\leq e_{i_k}). β k , r := β κ ( k , r ) i ( 1 ≤ r ≤ e i k ) .
The folding map on root lattices satisfies
π ( β ^ k , r ) = β k i \pi(\widehat{\beta}_{k,r})=\beta_k^{\mathbf{i}} π ( β k , r ) = β k i
([25 Lemma 2.4] ). In particular, the block
{ β ^ k , r ∣ 1 ≤ r ≤ e i k } \{\widehat{\beta}_{k,r}\mid1\leq r\leq e_{i_k}\} { β k , r ∣ 1 ≤ r ≤ e i k }
is the fiber of the root β k i \beta_k^{\mathbf{i}} β k i . These blockwise fibers will induce the fibers of i \mathbf{i} i -boxes and of the associated cluster variables in the following subsections.
i-box seeds and folding of compatible pairs
Keep the notation of Subsection 2.1. We refer to [38 , 33 ] for further details on i \mathbf{i} i -boxes and their associated seeds.
For 1 ≤ p ≤ l 1\leq p\leq l 1 ≤ p ≤ l , set
p + : = min ( { q > p ∣ i q = i p } ∪ { + ∞ } ) and p − : = max ( { q < p ∣ i q = i p } ∪ { − ∞ } ) . p^+:=\min\bigl(\{q>p\mid i_q=i_p\}\cup\{+\infty\}\bigr)
\qquad\text{and}\qquad
p^-:=\max\bigl(\{q<p\mid i_q=i_p\}\cup\{-\infty\}\bigr). p + := min ( { q > p ∣ i q = i p } ∪ { + ∞ } ) and p − := max ( { q < p ∣ i q = i p } ∪ { − ∞ } ) .
More generally, for j ∈ I j\in I j ∈ I , set
p ( j ) + : = min ( { q ≥ p ∣ i q = j } ∪ { + ∞ } ) and p ( j ) − : = max ( { q ≤ p ∣ i q = j } ∪ { − ∞ } ) . p(j)^+:=\min\bigl(\{q\geq p\mid i_q=j\}\cup\{+\infty\}\bigr)
\qquad\text{and}\qquad
p(j)^-:=\max\bigl(\{q\leq p\mid i_q=j\}\cup\{-\infty\}\bigr). p ( j ) + := min ( { q ≥ p ∣ i q = j } ∪ { + ∞ } ) and p ( j ) − := max ( { q ≤ p ∣ i q = j } ∪ { − ∞ } ) .
A i \mathbf{i} i -box associated with i \mathbf{i} i is an interval
c = [ x , y ] ⊂ [ 1 , l ] such that i x = i y . \mathbf{c}=[x,y]\subset[1,l]\qquad\text{such that}\qquad i_x=i_y. c = [ x , y ] ⊂ [ 1 , l ] such that i x = i y .
We write
i c : = i x = i y i_{\mathbf{c}}:=i_x=i_y i c := i x = i y
for its color , and abbreviate a singleton i \mathbf{i} i -box [ x , x ] [x,x] [ x , x ] to [ x ] [x] [ x ] .
Two i \mathbf{i} i -boxes c = [ x , y ] \mathbf{c}=[x,y] c = [ x , y ] and c ′ = [ x ′ , y ′ ] \mathbf{c}'=[x',y'] c ′ = [ x ′ , y ′ ] associated with i \mathbf{i} i are said to commute if
x − < x ′ ≤ y ′ < y + or ( x ′ ) − < x ≤ y < ( y ′ ) + . (13) x^-<x'\leq y'<y^+
\qquad\text{or}\qquad
(x')^-<x\leq y<(y')^+. \tag*{(13)} x − < x ′ ≤ y ′ < y + or ( x ′ ) − < x ≤ y < ( y ′ ) + . (13)
A collection of pairwise commuting i \mathbf{i} i -boxes associated with i \mathbf{i} i is called a commuting family . Such a family in [ 1 , l ] [1,l] [ 1 , l ] is called maximal if it is maximal with respect to inclusion. Throughout this section, nonempty admissible chains have final envelope [ 1 , l ] [1,l] [ 1 , l ] , called their extent . For the empty sequence, the chain and its underlying family are empty, with extent ∅ \varnothing ∅ .
For an interval [ x , y ] [x,y] [ x , y ] , set
[ x , y : = [ x , y ( i x ) − ] and { x , y ] : = [ x ( i y ) + , y ] [x,y:=[x,y(i_x)^-]
\qquad\text{and}\qquad
\{x,y]:=[x(i_y)^+,y] [ x , y := [ x , y ( i x ) − ] and { x , y ] := [ x ( i y ) + , y ]
whenever these intervals are nonempty. An admissible chain of i \mathbf{i} i -boxes associated with i \mathbf{i} i of extent [ 1 , l ] [1,l] [ 1 , l ] is a sequence
C = ( c k = [ x k , y k ] ) 1 ≤ k ≤ l \mathbf{C}=(\mathbf{c}_k=[x_k,y_k])_{1\leq k\leq l} C = ( c k = [ x k , y k ] ) 1 ≤ k ≤ l
such that
c ~ k : = ⋃ 1 ≤ s ≤ k c s = [ x ~ k , y ~ k ] \widetilde{\mathbf{c}}_k:=\bigcup_{1\leq s\leq k}\mathbf{c}_s=[\widetilde{x}_k,\widetilde{y}_k] c k := ⋃ 1 ≤ s ≤ k c s = [ x k , y k ] is an interval of cardinality k k k , called the k k k -th envelope , and
c k = [ x ~ k , y ~ k \mathbf{c}_k=[\widetilde{x}_k,\widetilde{y}_k c k = [ x k , y k or c k = { x ~ k , y ~ k ] \mathbf{c}_k=\{\widetilde{x}_k,\widetilde{y}_k] c k = { x k , y k ] .
Its underlying family is
F ( C ) : = { c k ∣ 1 ≤ k ≤ l } . \mathcal{F}(\mathcal{C}) := \{c_k \mid1 \leq k \leq l\}. F ( C ) := { c k ∣ 1 ≤ k ≤ l } .
It is a maximal commuting family of i \mathbf{i} i -boxes associated with i \mathbf{i} i , and every maximal commuting family of i \mathbf{i} i -boxes arises in this way. Note that a maximal commuting family of i \mathbf{i} i -boxes need not determine a unique admissible chain.
For l ≥ 1 l \geq1 l ≥ 1 , we describe box moves on admissible chains as follows. Since the first envelope c ~ 1 \widetilde{c}_1 c 1 is a singleton, we have
c ~ 1 = c 1 = [ x ] . \widetilde{c}_1=c_1=[x]. c 1 = c 1 = [ x ] .
For 1 ≤ k < l 1 \leq k<l 1 ≤ k < l , let T k ∈ { L , R } T_k\in\{\mathcal{L},\mathcal{R}\} T k ∈ { L , R } record whether c ~ k + 1 \widetilde{c}_{k+1} c k + 1 is obtained from c ~ k \widetilde{c}_k c k by adjoining its left or right neighboring point, respectively. Thus the chain C \mathcal{C} C has the associated pair
( x , T ) , T = ( T 1 , … , T l − 1 ) ∈ { L , R } l − 1 . (x,\mathcal{T}),\qquad\mathcal{T}=(T_1,\ldots,T_{l-1})\in\{\mathcal{L},\mathcal{R}\}^{l-1}. ( x , T ) , T = ( T 1 , … , T l − 1 ) ∈ { L , R } l − 1 .
Conversely, its extent [ 1 , l ] [1,l] [ 1 , l ] together with T \mathcal{T} T determines x x x , all the envelopes, and hence C \mathcal{C} C .
For 1 ≤ k < l 1\leq k<l 1 ≤ k < l , the member c k c_k c k is called movable if
k = 1 or T k − 1 ≠ T k . k=1\quad\text{or}\quad T_{k-1}\ne T_k. k = 1 or T k − 1 = T k .
For a movable c k c_k c k , the k k k -th box move B k ( C ) \mathcal{B}_k(\mathcal{C}) B k ( C ) is the admissible chain with associated pair ( x ′ , T ′ ) (x',\mathcal{T}') ( x ′ , T ′ ) defined as follows. If k = 1 k=1 k = 1 , then
( x ′ , T 1 ′ ) = { ( x + 1 , L ) , T 1 = R , ( x − 1 , R ) , T 1 = L , T s ′ = T s ( s ≠ 1 ) . (x',T'_1)=
\begin{cases}
(x+1,\mathcal{L}), & T_1=\mathcal{R},\\
(x-1,\mathcal{R}), & T_1=\mathcal{L},
\end{cases}
\qquad T'_s=T_s\quad(s\ne1). ( x ′ , T 1 ′ ) = { ( x + 1 , L ) , ( x − 1 , R ) , T 1 = R , T 1 = L , T s ′ = T s ( s = 1 ) .
If k > 1 k>1 k > 1 , then x ′ = x x'=x x ′ = x ,
T k − 1 ′ = T k , T k ′ = T k − 1 , T s ′ = T s ( s ∉ { k − 1 , k } ) . T'_{k-1}=T_k,\qquad T'_k=T_{k-1},\qquad T'_s=T_s\quad(s\notin\{k-1,k\}). T k − 1 ′ = T k , T k ′ = T k − 1 , T s ′ = T s ( s ∈ / { k − 1 , k }) .
Equivalently, the move shifts the envelope c ~ k \widetilde{c}_k c k by one position inside c ~ k + 1 \widetilde{c}_{k+1} c k + 1 .
Lemma 2.1 ([39 , 33 ] ). Let c k c_k c k be movable in C \mathcal{C} C , and set
C ′ : = B k ( C ) = ( c s ′ ) 1 ≤ s ≤ l . \mathcal{C}':=\mathcal{B}_k(\mathcal{C})=(c'_s)_{1\leq s\leq l}. C ′ := B k ( C ) = ( c s ′ ) 1 ≤ s ≤ l .
Then exactly one of the following holds.
(i) Suppose that c ~ k + 1 \widetilde{c}_{k+1} c k + 1 is not a i \mathbf{i} i -box. Then the box move interchanges the two consecutive members c k c_k c k and c k + 1 c_{k+1} c k + 1 ; namely,
c k ′ = c k + 1 , c k + 1 ′ = c k , c s ′ = c s ( s ∉ { k , k + 1 } ) . c'_k=c_{k+1},\qquad c'_{k+1}=c_k,\qquad c'_s=c_s\quad(s\notin\{k,k+1\}). c k ′ = c k + 1 , c k + 1 ′ = c k , c s ′ = c s ( s ∈ / { k , k + 1 }) .
Consequently,
F ( C ′ ) = F ( C ) . \mathcal{F}(\mathcal{C}')=\mathcal{F}(\mathcal{C}). F ( C ′ ) = F ( C ) .
This is called a permutation-type box move.
(ii) Suppose that
c ~ k + 1 = [ u , v ] \widetilde{c}_{k+1}=[u,v] c k + 1 = [ u , v ]
is an i \mathbf{i} i -box. Then the box move replaces exactly one member of the underlying family. More precisely, there exist unique boxes
c ∈ F ( C ) and c ⋆ ∈ F ( C ′ ) c\in\mathcal{F}(\mathcal{C})\quad\text{and}\quad c^\star\in\mathcal{F}(\mathcal{C}') c ∈ F ( C ) and c ⋆ ∈ F ( C ′ )
such that
{ c , c ⋆ } = { [ u + , v ] , [ u , v − ] } and F ( C ′ ) = ( F ( C ) ∖ { c } ) ⊔ { c ⋆ } . \{c,c^\star\}=\{[u^+,v],[u,v^-]\}\quad\text{and}\quad
\mathcal{F}(\mathcal{C}')=(\mathcal{F}(\mathcal{C})\setminus\{c\})\sqcup\{c^\star\}. { c , c ⋆ } = {[ u + , v ] , [ u , v − ]} and F ( C ′ ) = ( F ( C ) ∖ { c }) ⊔ { c ⋆ } .
This is called a mutation-type box move.
Any two admissible chains of extent [ 1 , l ] [1,l] [ 1 , l ] are connected by a finite sequence of box moves ([39 Lemma 5.10] and [33 Proposition 2.6] ).
We next recall the compatible pair attached to F ( i ) \mathcal{F}(\mathbf{i}) F ( i ) . Let
E i : = ⨁ p = 1 l Z ϵ p . \mathcal{E}^{\mathbf{i}}:=\bigoplus_{p=1}^{l}\mathbb{Z}\epsilon_p. E i := p = 1 ⨁ l Z ϵ p .
and define a skew-symmetric pairing on E i \mathcal{E}^{\mathbf{i}} E i by
λ i ( ϵ p , ϵ q ) : = { − ( β p i , β q i ) g , p < q , 0 , p = q , ( β p i , β q i ) g , p > q . \lambda^{\mathbf{i}}(\epsilon_p,\epsilon_q):=
\begin{cases}
-(\beta_p^{\mathbf{i}},\beta_q^{\mathbf{i}})_g, & p<q,\\
0, & p=q,\\
(\beta_p^{\mathbf{i}},\beta_q^{\mathbf{i}})_g, & p>q.
\end{cases} λ i ( ϵ p , ϵ q ) := ⎩ ⎨ ⎧ − ( β p i , β q i ) g , 0 , ( β p i , β q i ) g , p < q , p = q , p > q .
This is the negative of the pairing in [38 (5.4)] . When i \mathbf{i} i is reduced, the matrix L F L^{\mathcal{F}} L F defined below is the negative of the Λ \Lambda Λ -invariant matrix of the determinantial modules indexed by F \mathcal{F} F ; see Proposition 3.11 below.
For an i \mathbf{i} i -box c = [ x , y ] \mathbf{c}=[x,y] c = [ x , y ] associated with i \mathbf{i} i , set
E c : = ∑ x ≤ p ≤ y i p = i c ϵ p . \mathcal{E}_{\mathbf{c}}:=\sum_{\substack{x\leq p\leq y\\ i_p=i_{\mathbf{c}}}}\epsilon_p. E c := x ≤ p ≤ y i p = i c ∑ ϵ p .
We call E c \mathcal{E}_{\mathbf{c}} E c the box vector associated with c \mathbf{c} c . If F \mathcal{F} F is a maximal commuting family, define
L F : = ( l c , c ′ F ) c , c ′ ∈ F , l c , c ′ F : = λ i ( E c , E c ′ ) . L^{\mathcal{F}}:=\left(l^{\mathcal{F}}_{\mathbf{c},\mathbf{c}'}\right)_{\mathbf{c},\mathbf{c}'\in\mathcal{F}},\qquad
l^{\mathcal{F}}_{\mathbf{c},\mathbf{c}'}:=\lambda^{\mathbf{i}}(\mathcal{E}_{\mathbf{c}},\mathcal{E}_{\mathbf{c}'}). L F := ( l c , c ′ F ) c , c ′ ∈ F , l c , c ′ F := λ i ( E c , E c ′ ) .
An i \mathbf{i} i -box c = [ x , y ] ∈ F \mathbf{c}=[x,y]\in\mathcal{F} c = [ x , y ] ∈ F is called frozen if
x − = − ∞ and y + = + ∞ . x^{-}=-\infty\qquad\text{and}\qquad y^{+}=+\infty. x − = − ∞ and y + = + ∞.
Let F e x \mathcal{F}_{\mathrm{ex}} F ex be the set of the remaining i \mathbf{i} i -boxes. Following [25 Section 3.3] , we use the exchange matrix
B F : = ( b c , c ′ F ) c ∈ F , c ′ ∈ F e x : = − B K K ( F ) , B^{\mathcal{F}}:=\left(b^{\mathcal{F}}_{\mathbf{c},\mathbf{c}'}\right)_{\mathbf{c}\in\mathcal{F},\,\mathbf{c}'\in\mathcal{F}_{\mathrm{ex}}}:=-B_{\mathrm{KK}}(\mathcal{F}), B F := ( b c , c ′ F ) c ∈ F , c ′ ∈ F ex := − B KK ( F ) ,
where B K K ( F ) B_{\mathrm{KK}}(\mathcal{F}) B KK ( F ) is the restriction to the exchangeable columns of the matrix in [33 (3.1)] . By [25 Theorem 3.5] , the pair ( L F , B F ) (L^{\mathcal{F}},B^{\mathcal{F}}) ( L F , B F ) is compatible and satisfies
( B F ) T L F = ( 2 D F ∣ 0 ) , D F : = diag ( d i c ∣ c ∈ F e x ) . (14) (B^{\mathcal{F}})^{T}L^{\mathcal{F}}=(2D^{\mathcal{F}}\mid0),\qquad D^{\mathcal{F}}:=\operatorname{diag}(d_{i_{\mathbf{c}}}\mid\mathbf{c}\in\mathcal{F}_{\mathrm{ex}}). \tag*{(14)} ( B F ) T L F = ( 2 D F ∣ 0 ) , D F := diag ( d i c ∣ c ∈ F ex ) . (14)
We denote the obtained Λ \Lambda Λ -seed by
Σ i ( F ) : = ( x F , L F , B F ) , x F = ( x c ∣ c ∈ F ) . \Sigma^{\mathbf{i}}(\mathcal{F}):=(\mathbf{x}^{\mathcal{F}},L^{\mathcal{F}},B^{\mathcal{F}}),\qquad
\mathbf{x}^{\mathcal{F}}=(x_{\mathbf{c}}\mid\mathbf{c}\in\mathcal{F}). Σ i ( F ) := ( x F , L F , B F ) , x F = ( x c ∣ c ∈ F ) .
We now compare this seed with the corresponding lifted seed. Choose an admissible chain C \mathcal{C} C realizing F \mathcal{F} F . Let C ^ \widehat{\mathcal{C}} C be its canonical lift [25 Section 3.4] , and set
F ^ : = F ( C ^ ) . \widehat{\mathcal{F}}:=\mathcal{F}(\widehat{\mathcal{C}}). F := F ( C ) .
Thus F ^ \widehat{\mathcal{F}} F is a maximal commuting family of i \mathbf{i} i -boxes associated with i ^ \widehat{\mathbf{i}} i . By [33 Section 2.2] and [25 Section 3.4] , we have a surjection
π b o x C : F ^ ⟶ F . \pi^{\mathcal{C}}_{\mathrm{box}}:\widehat{\mathcal{F}}\longrightarrow\mathcal{F}. π box C : F ⟶ F .
For c ∈ J \mathfrak{c} \in\mathcal{J} c ∈ J , define its box fiber by
Fib C ( c ) : = ( π b o x C ) − 1 ( c ) . \operatorname{Fib}_{\mathcal{C}}(\mathfrak{c}) := \left(\pi_{\mathrm{box}}^{\mathcal{C}}\right)^{-1}(\mathfrak{c}). Fib C ( c ) := ( π box C ) − 1 ( c ) .
Then [25 Section 3.4] implies
∣ Fib C ( c ) ∣ = e i c . (15) \left|\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c})\right| = e_{i_{\mathfrak{c}}}. \tag*{(15)} ∣ Fib C ( c ) ∣ = e i c . (15)
Let Σ ^ i ( F ^ ) : = ( x ^ F ^ , L ^ F ^ , B ^ F ^ ) \widehat{\Sigma}^{i}(\widehat{\mathcal{F}}):=(\widehat{\mathbf{x}}^{\widehat{\mathcal{F}}},\widehat{L}^{\widehat{\mathcal{F}}},\widehat{B}^{\widehat{\mathcal{F}}}) Σ i ( F ) := ( x F , L F , B F ) be the Λ \Lambda Λ -seed associated with the lifted family. The exchange and compatibility matrices fold simultaneously as follows.
Theorem 2.2 ([25 Theorem 3.15] ). Let c , c ′ ∈ J \mathfrak{c},\mathfrak{c}' \in\mathcal{J} c , c ′ ∈ J , and choose c ^ ∈ Fib C ( c ′ ) \widehat{\mathfrak{c}} \in\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c}') c ∈ Fib C ( c ′ ) . Then
b c , c ′ F = ∑ c ^ ∈ Fib C ( c ) b ^ c ^ , c ^ ′ F ^ ( c ′ ∈ J e x ) , l c , c ′ F = d i c ∑ c ^ ∈ Fib C ( c ) l ^ c ^ , c ^ ′ F ^ . b^{\mathcal{F}}_{\mathfrak{c},\mathfrak{c}'}=\sum_{\widehat{\mathfrak{c}}\in\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c})}\widehat{b}^{\widehat{\mathcal{F}}}_{\widehat{\mathfrak{c}},\widehat{\mathfrak{c}}'}\quad(\mathfrak{c}'\in\mathcal{J}_{\mathrm{ex}}),\qquad
l^{\mathcal{F}}_{\mathfrak{c},\mathfrak{c}'}=d_{i_{\mathfrak{c}}}\sum_{\widehat{\mathfrak{c}}\in\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c})}\widehat{l}^{\widehat{\mathcal{F}}}_{\widehat{\mathfrak{c}},\widehat{\mathfrak{c}}'}. b c , c ′ F = c ∈ Fib C ( c ) ∑ b c , c ′ F ( c ′ ∈ J ex ) , l c , c ′ F = d i c c ∈ Fib C ( c ) ∑ l c , c ′ F . Both right-hand sides are independent of the choice of c ^ \widehat{\mathfrak{c}} c . In particular,
( B ^ F ^ ) T L ^ F ^ = ( 2 Id ∣ 0 ) and ( B F ) T L F = ( 2 D F ∣ 0 ) . \left(\widehat{B}^{\widehat{\mathcal{F}}}\right)^{T}\widehat{L}^{\widehat{\mathcal{F}}}=(2\operatorname{Id}\mid0)
\qquad\text{and}\qquad
\left(B^{\mathcal{F}}\right)^{T}L^{\mathcal{F}}=(2D^{\mathcal{F}}\mid0). ( B F ) T L F = ( 2 Id ∣ 0 ) and ( B F ) T L F = ( 2 D F ∣ 0 ) .
The construction and Theorem Theorem 2.2 require neither reducedness of i \mathfrak{i} i nor finite type of the Cartan datum. Moreover, all maximal commuting families of i \mathfrak{i} i -boxes associated with i \mathfrak{i} i are connected by box moves. A permutation-type box move only relabels the corresponding seed, whereas a mutation-type box move acts by simultaneous mutation of B F B^{\mathcal{F}} B F and L F L^{\mathcal{F}} L F . Consequently, for any two such families F \mathcal{F} F and G \mathcal{G} G ,
Σ i ( F ) and Σ i ( G ) \Sigma^{\mathfrak{i}}(\mathcal{F})\qquad\text{and}\qquad\Sigma^{\mathfrak{i}}(\mathcal{G}) Σ i ( F ) and Σ i ( G )
are mutation-equivalent up to seed relabeling. In particular, the Λ \Lambda Λ -cluster algebra
A i [ 1 , l ] : = A ( Σ i ( F ) ) , \mathcal{A}^{\mathfrak{i}}[1,l]:=\mathcal{A}\left(\Sigma^{\mathfrak{i}}(\mathcal{F})\right), A i [ 1 , l ] := A ( Σ i ( F ) ) ,
where F \mathcal{F} F is any maximal commuting family of i \mathfrak{i} i -boxes associated with i \mathfrak{i} i and having extent [ 1 , l ] [1,l] [ 1 , l ] , is independent of the choice of F \mathcal{F} F .
Folding of cluster variables and F F F -polynomials
Keep the notation of Subsection Subsection 2.2. The box fibers induce a surjective homomorphism between the corresponding seed Laurent rings:
ϕ C : L ( Σ ^ i ( F ^ ) ) ⟶ L ( Σ i ( F ) ) , x ^ c ^ ⟼ x c ( c ^ ∈ Fib C ( c ) ) . (16) \phi_{\mathcal{C}}:\mathcal{L}\left(\widehat{\Sigma}^{\mathfrak{i}}(\widehat{\mathcal{F}})\right)\longrightarrow\mathcal{L}\left(\Sigma^{\mathfrak{i}}(\mathcal{F})\right),
\qquad
\widehat{x}_{\widehat{\mathfrak{c}}}\longmapsto x_{\mathfrak{c}}
\quad(\widehat{\mathfrak{c}}\in\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c})). \tag*{(16)} ϕ C : L ( Σ i ( F ) ) ⟶ L ( Σ i ( F ) ) , x c ⟼ x c ( c ∈ Fib C ( c )) . (16)
They also define the folding map on exponent lattices
Fold C : Z F ^ ⟶ Z F , ( Fold C ( g ^ ) ) c : = ∑ c ^ ∈ Fib C ( c ) g ^ c ^ . \operatorname{Fold}_{\mathcal{C}}:\mathbb{Z}^{\widehat{\mathcal{F}}}\longrightarrow\mathbb{Z}^{\mathcal{F}},
\qquad
\left(\operatorname{Fold}_{\mathcal{C}}(\widehat{\mathfrak{g}})\right)_{\mathfrak{c}}
:=\sum_{\widehat{\mathfrak{c}}\in\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c})}\widehat{\mathfrak{g}}_{\widehat{\mathfrak{c}}}. Fold C : Z F ⟶ Z F , ( Fold C ( g ) ) c := c ∈ Fib C ( c ) ∑ g c .
We use the same notation for its restriction to the exchangeable coordinates. For e ∈ F e x \mathfrak{e}\in\mathcal{F}_{\mathrm{ex}} e ∈ F ex and e ^ ∈ Fib C ( e ) \widehat{\mathfrak{e}}\in\operatorname{Fib}_{\mathcal{C}}(\mathfrak{e}) e ∈ Fib C ( e ) , define the corresponding Laurent monomials by
y ˙ e : = ∏ c ∈ F x c b c , e F and y ˙ ^ e ^ : = ∏ c ^ ∈ F ^ x ^ c ^ b ^ c ^ , e ^ F ^ . \dot{y}_{\mathfrak{e}}:=\prod_{\mathfrak{c}\in\mathcal{F}}x_{\mathfrak{c}}^{b^{\mathcal{F}}_{\mathfrak{c},\mathfrak{e}}}
\qquad\text{and}\qquad
\widehat{\dot{y}}_{\widehat{\mathfrak{e}}}:=\prod_{\widehat{\mathfrak{c}}\in\widehat{\mathcal{F}}}\widehat{x}_{\widehat{\mathfrak{c}}}^{\widehat{b}^{\widehat{\mathcal{F}}}_{\widehat{\mathfrak{c}},\widehat{\mathfrak{e}}}}. y ˙ e := c ∈ F ∏ x c b c , e F and y ˙ e := c ∈ F ∏ x c b c , e F .
Set
y ˙ : = ( y ˙ e ) e ∈ F e x and y ˙ ^ : = ( y ˙ ^ e ^ ) e ^ ∈ F ^ e x . \dot{\mathbf{y}}:=(\dot{y}_{\mathfrak{e}})_{\mathfrak{e}\in\mathcal{F}_{\mathrm{ex}}}
\qquad\text{and}\qquad
\widehat{\dot{\mathbf{y}}}:=(\widehat{\dot{y}}_{\widehat{\mathfrak{e}}})_{\widehat{\mathfrak{e}}\in\widehat{\mathcal{F}}_{\mathrm{ex}}}. y ˙ := ( y ˙ e ) e ∈ F ex and y ˙ := ( y ˙ e ) e ∈ F ex .
The folding of exchange matrices in Theorem Theorem 2.2 gives
ϕ C ( y ˙ ^ e ^ ) = y ˙ e . (17) \phi_{\mathcal{C}}\left(\widehat{\dot{y}}_{\widehat{\mathfrak{e}}}\right)=\dot{y}_{\mathfrak{e}}. \tag*{(17)} ϕ C ( y ˙ e ) = y ˙ e . (17)
Let
y = ( y e ) e ∈ F e x and y ^ = ( y ^ e ^ ) e ^ ∈ F ^ e x \mathbf{y}=(y_e)_{e\in F_{\mathrm{ex}}}\quad\text{and}\quad\widehat{\mathbf{y}}=(\widehat{y}_{\widehat{e}})_{\widehat{e}\in\widehat{F}_{\mathrm{ex}}} y = ( y e ) e ∈ F ex and y = ( y e ) e ∈ F ex
be families of indeterminates. Define the algebra homomorphism
ϕ C , y : C [ y ^ e ^ ∣ e ^ ∈ F ^ e x ] ⟶ C [ y e ∣ e ∈ F e x ] by ϕ C , y ( y ^ e ^ ) = y e ( e ^ ∈ Fib C ( e ) ) . \phi_{\mathcal{C},\mathbf{y}}:\mathbb{C}[\widehat{y}_{\widehat{e}}\mid\widehat{e}\in\widehat{F}_{\mathrm{ex}}]\longrightarrow\mathbb{C}[y_e\mid e\in F_{\mathrm{ex}}]\quad\text{by}\quad\phi_{\mathcal{C},\mathbf{y}}(\widehat{y}_{\widehat{e}})=y_e\quad(\widehat{e}\in\operatorname{Fib}_{\mathcal{C}}(e)). ϕ C , y : C [ y e ∣ e ∈ F ex ] ⟶ C [ y e ∣ e ∈ F ex ] by ϕ C , y ( y e ) = y e ( e ∈ Fib C ( e )) .
The folding relation defined by ϕ C \phi_{\mathcal{C}} ϕ C is preserved at the canonically lifted initial and final seeds for braid moves and box moves. Consequently, it folds all corresponding cluster variables in box seeds.
Theorem 2.3 ([25 Theorem 6.2 and Corollary 6.3] ). Let i , j ∈ Seq ( b ) i,j\in\operatorname{Seq}(\mathbf{b}) i , j ∈ Seq ( b ) and
(i) let D \mathcal{D} D be an admissible chain of i i i -boxes associated with j \mathbf{j} j , with underlying maximal commuting family G : = F ( D ) \mathcal{G}:=\mathcal{F}(\mathcal{D}) G := F ( D ) ,
(ii) let D ^ \widehat{\mathcal{D}} D be the canonical lift of D \mathcal{D} D associated with j ^ \widehat{\mathbf{j}} j , and set G ^ = F ( D ^ ) \widehat{\mathcal{G}}=\mathcal{F}(\widehat{\mathcal{D}}) G = F ( D ) .
For e ∈ G \mathbf{e}\in\mathcal{G} e ∈ G and e ^ ∈ Fib D ( e ) \widehat{\mathbf{e}}\in\operatorname{Fib}_{\mathcal{D}}(\mathbf{e}) e ∈ Fib D ( e ) , we have
ϕ C ( x ^ e ^ ) = x e G and deg Σ i ( F ) ( x e G ) = Fold C ( deg Σ ^ i ^ ( F ^ ) ( x ^ e ^ ) ) . \phi_{\mathcal{C}}(\widehat{x}_{\widehat{\mathbf{e}}})=x_{\mathbf{e}}^{\mathcal{G}}\quad\text{and}\quad\deg_{\Sigma^i(\mathcal{F})}(x_{\mathbf{e}}^{\mathcal{G}})=\operatorname{Fold}_{\mathcal{C}}\left(\deg_{\widehat{\Sigma}^{\widehat{i}}(\widehat{\mathcal{F}})}(\widehat{x}_{\widehat{\mathbf{e}}})\right). ϕ C ( x e ) = x e G and deg Σ i ( F ) ( x e G ) = Fold C ( deg Σ i ( F ) ( x e ) ) . The same statements hold for the corresponding cluster monomials.
We now record the consequence for F F F -polynomials. Let
z ^ = x ^ g ^ F ^ z ^ ( y ^ ) \widehat{z}=\widehat{\mathbf{x}}^{\widehat{\mathbf{g}}}\widehat{F}_{\widehat{z}}(\widehat{\mathbf{y}}) z = x g F z ( y )
be a cluster monomial associated with Σ ^ i ^ ( G ^ ) \widehat{\Sigma}^{\widehat{i}}(\widehat{\mathcal{G}}) Σ i ( G ) , and set
z : = ϕ C ( z ^ ) . z:=\phi_{\mathcal{C}}(\widehat{z}). z := ϕ C ( z ) .
By Theorem 2.3 , we have
g z = Fold C ( g ^ ) . \mathbf{g}_z=\operatorname{Fold}_{\mathcal{C}}(\widehat{\mathbf{g}}). g z = Fold C ( g ) .
where both degrees are taken with respect to the fixed seeds Σ i ( F ) \Sigma^i(\mathcal{F}) Σ i ( F ) and Σ ^ i ^ ( F ^ ) \widehat{\Sigma}^{\widehat{i}}(\widehat{\mathcal{F}}) Σ i ( F ) , respectively. Applying ϕ C \phi_{\mathcal{C}} ϕ C to the pointed expression of z ^ \widehat{z} z and using (17) , we obtain
z = x Fold C ( g ^ ) ϕ C , y ( F ^ z ^ ( y ^ ) ) ∣ y = y ˙ . z=\mathbf{x}^{\operatorname{Fold}_{\mathcal{C}}(\widehat{\mathbf{g}})}\left.\phi_{\mathcal{C},\mathbf{y}}\left(\widehat{F}_{\widehat{z}}(\widehat{\mathbf{y}})\right)\right|_{\mathbf{y}=\dot{\mathbf{y}}}. z = x Fold C ( g ) ϕ C , y ( F z ( y ) ) y = y ˙ .
Comparing this with the downstairs pointed expression
z = x g z F z ( y ˙ ) , z=\mathbf{x}^{\mathbf{g}_z}F_z(\dot{\mathbf{y}}), z = x g z F z ( y ˙ ) ,
the uniqueness of the pointed expression gives the following formal polynomial identity.
Proposition 2.4. With the notation above, we have
F z ( y ) = ϕ C , y ( F ^ z ^ ( y ^ ) ) . F_z(\mathbf{y})=\phi_{\mathcal{C},\mathbf{y}}\left(\widehat{F}_{\widehat{z}}(\widehat{\mathbf{y}})\right). F z ( y ) = ϕ C , y ( F z ( y ) ) .
Global folding and tropical invariants for positive braids
Let b ∈ B g + \mathbf{b} \in\mathrm{B}_{\mathfrak{g}}^{+} b ∈ B g + , and choose i = ( i 1 , … , i l ) ∈ Seq ( b ) i=(i_1,\ldots,i_l)\in\operatorname{Seq}(\mathbf{b}) i = ( i 1 , … , i l ) ∈ Seq ( b ) .
Let F \mathcal{F} F be a maximal commuting family of i \mathbf{i} i -boxes associated with i \mathbf{i} i , and choose an admissible chain C \mathcal{C} C realizing F \mathcal{F} F . We define
A ( b ) : = A ( Σ i ( F ) ) . \mathcal{A}(\mathbf{b}):=\mathcal{A}\left(\Sigma^{\mathbf{i}}(\mathcal{F})\right). A ( b ) := A ( Σ i ( F ) ) .
Its mutation class depends only on b \mathbf{b} b , and not on the choices of i \mathbf{i} i , C \mathcal{C} C , or F \mathcal{F} F . Similarly, the blockwise lift i ^ \widehat{\mathbf{i}} i in (11) represents b ^ ∈ B g ^ + \widehat{\mathbf{b}}\in\mathrm{B}_{\widehat{\mathfrak{g}}}^{+} b ∈ B g + , and we set
A ( b ^ ) : = A ( Σ ^ i ^ ( F ^ ) ) . \mathcal{A}(\widehat{\mathbf{b}}):=\mathcal{A}\left(\widehat{\Sigma}^{\widehat{\mathbf{i}}}(\widehat{\mathcal{F}})\right). A ( b ) := A ( Σ i ( F ) ) .
Keep the Laurent homomorphism ϕ C \phi_{\mathcal{C}} ϕ C defined in (16) . Although this map is defined using the initial box fibers, Theorem 2.3 shows that it simultaneously folds all corresponding box seeds in the mutation classes. More precisely, let j = ( j 1 , … , j l ) ∈ Seq ( b ) \mathbf{j}=(j_1,\ldots,j_l)\in\operatorname{Seq}(\mathbf{b}) j = ( j 1 , … , j l ) ∈ Seq ( b ) , let G \mathcal{G} G be a maximal commuting family of j \mathbf{j} j -boxes associated with j \mathbf{j} j , and choose an admissible chain D \mathcal{D} D realizing G \mathcal{G} G . Let j ^ \widehat{\mathbf{j}} j be the blockwise lift of j \mathbf{j} j , let D ^ \widehat{\mathcal{D}} D be the canonical lift of D \mathcal{D} D , and set G ^ = F ( D ^ ) \widehat{\mathcal{G}}=\mathcal{F}(\widehat{\mathcal{D}}) G = F ( D ) . Then, for every e ∈ G e\in\mathcal{G} e ∈ G and e ^ ∈ Fib D ( e ) \widehat{e}\in\operatorname{Fib}_{\mathcal{D}}(e) e ∈ Fib D ( e ) ,
ϕ C ( x ^ e ^ G ^ ) = x e G . (18) \phi_{\mathcal{C}}\left(\widehat{x}_{\widehat{e}}^{\widehat{\mathcal{G}}}\right)=x_e^{\mathcal{G}}. \tag*{(18)} ϕ C ( x e G ) = x e G . (18)
Combining Theorem 2.3 with Theorem 2.2 , we obtain the simultaneous folding of the entire lifted box seed:
Σ ^ j ^ ( G ^ ) ⟼ Σ j ( G ) . (19) \widehat{\Sigma}^{\widehat{\mathbf{j}}}(\widehat{\mathcal{G}})\longmapsto\Sigma^{\mathbf{j}}(\mathcal{G}). \tag*{(19)} Σ j ( G ) ⟼ Σ j ( G ) . (19)
Here (19) combines the variable folding (18) with the folding of compatible pairs.
In particular, for M ^ = ∏ e ^ ∈ G ^ ( x ^ e ^ G ^ ) a e ^ \widehat{M}=\prod_{\widehat{e}\in\widehat{\mathcal{G}}}\left(\widehat{x}_{\widehat{e}}^{\widehat{\mathcal{G}}}\right)^{a_{\widehat{e}}} M = ∏ e ∈ G ( x e G ) a e , we have
ϕ C ( M ^ ) = ∏ e ∈ G ( x e G ) ∑ e ^ ∈ Fib D ( e ) a e ^ . \phi_{\mathcal{C}}(\widehat{M})=\prod_{e\in\mathcal{G}}\left(x_e^{\mathcal{G}}\right)^{\sum_{\widehat{e}\in\operatorname{Fib}_{\mathcal{D}}(e)}a_{\widehat{e}}}. ϕ C ( M ) = e ∈ G ∏ ( x e G ) ∑ e ∈ Fib D ( e ) a e .
Thus every cluster monomial associated with a lifted box seed is sent to the corresponding downstairs cluster monomial.
The fixed Laurent homomorphism restricts to the cluster algebras as follows.
Theorem 2.5 ([25 Theorems 6.2 and 6.4] ). For every b ∈ B g + \mathbf{b}\in\mathrm{B}_{\mathfrak{g}}^{+} b ∈ B g + , the homomorphism ϕ C \phi_{\mathcal{C}} ϕ C restricts to a surjective C \mathbb{C} C -algebra homomorphism
ϕ C b : A ( b ^ ) ↠ A ( b ) . \phi_{\mathcal{C}}^{\mathbf{b}}:\mathcal{A}(\widehat{\mathbf{b}})\twoheadrightarrow\mathcal{A}(\mathbf{b}). ϕ C b : A ( b ) ↠ A ( b ) . For every j ∈ Seq ( b ) \mathbf{j}\in\operatorname{Seq}(\mathbf{b}) j ∈ Seq ( b ) and every maximal commuting family G \mathcal{G} G of j \mathbf{j} j -boxes associated with j \mathbf{j} j , it maps the cluster variables and cluster monomials of Σ ^ j ^ ( G ^ ) \widehat{\Sigma}^{\widehat{\mathbf{j}}}(\widehat{\mathcal{G}}) Σ j ( G ) to those of Σ j ( G ) \Sigma^{\mathbf{j}}(\mathcal{G}) Σ j ( G ) according to their box fibers.
Theorem 2.5 requires neither reducedness of the sequences nor finite type of the Cartan data. Every symmetrizable Cartan matrix admits an admissible simply-laced realization [25 Remark 1.3] . Combining Theorem 2.5 with Proposition 2.4, we compute the F F F -polynomial and the partial F F F -invariant of every cluster variable arising from a box seed by the corresponding fiber evaluation upstairs.
Theorem 2.6. Let i , j ∈ Seq ( b ) i,j \in\operatorname{Seq}(\mathbf{b}) i , j ∈ Seq ( b ) , and let C \mathcal{C} C and D \mathcal{D} D be admissible chains associated with i i i and j j j , respectively. For c ∈ F ( C ) c \in\mathcal{F}(\mathcal{C}) c ∈ F ( C ) , c ′ ∈ F ( D ) c' \in\mathcal{F}(\mathcal{D}) c ′ ∈ F ( D ) , and every c ^ ∈ Fib D ( c ′ ) \widehat{c} \in\operatorname{Fib}_{\mathcal{D}}(c') c ∈ Fib D ( c ′ ) , set
x ^ Fib C ( c ) : = ∏ c ^ ∈ Fib C ( c ) x ^ c ^ . \widehat{x}_{\operatorname{Fib}_{\mathcal{C}}(c)}:=\prod_{\widehat{c}\in\operatorname{Fib}_{\mathcal{C}}(c)}\widehat{x}_{\widehat{c}}. x Fib C ( c ) := c ∈ Fib C ( c ) ∏ x c . Then the two ordered tropical pairings fold separately:
⟨ x c , x c ′ ⟩ trop = d i c ⟨ x ^ Fib C ( c ) , x ^ c ^ ⟩ trop , (20) \langle x_c,x_{c'}\rangle_{\operatorname{trop}}=d_{i_c}\langle\widehat{x}_{\operatorname{Fib}_{\mathcal{C}}(c)},\widehat{x}_{\widehat{c}}\rangle_{\operatorname{trop}}, \tag*{(20)} ⟨ x c , x c ′ ⟩ trop = d i c ⟨ x Fib C ( c ) , x c ⟩ trop , (20) ⟨ x c ′ , x c ⟩ trop = d i c ⟨ x ^ c ^ , x ^ Fib C ( c ) ⟩ trop . (21) \langle x_{c'},x_c\rangle_{\operatorname{trop}}=d_{i_c}\langle\widehat{x}_{\widehat{c}},\widehat{x}_{\operatorname{Fib}_{\mathcal{C}}(c)}\rangle_{\operatorname{trop}}. \tag*{(21)} ⟨ x c ′ , x c ⟩ trop = d i c ⟨ x c , x Fib C ( c ) ⟩ trop . (21) Both right-hand sides are independent of the choice of c ^ ∈ Fib D ( c ′ ) \widehat{c} \in\operatorname{Fib}_{\mathcal{D}}(c') c ∈ Fib D ( c ′ ) . Adding the two identities gives
( x c ∣ x c ′ ) F = d i c ( x ^ Fib C ( c ) ∣ x ^ c ^ ) F . (22) (x_c\mid x_{c'})_F=d_{i_c}(\widehat{x}_{\operatorname{Fib}_{\mathcal{C}}(c)}\mid\widehat{x}_{\widehat{c}})_F. \tag*{(22)} ( x c ∣ x c ′ ) F = d i c ( x Fib C ( c ) ∣ x c ) F . (22)
Proof. We compute the tropical pairings in the box seeds associated with C \mathcal{C} C and its lift. By Theorem 2.5,
ϕ C b ( x ^ c ^ ) = x c ′ ( c ^ ∈ Fib D ( c ′ ) ) . \phi_{\mathcal{C}}^{\mathbf{b}}(\widehat{x}_{\widehat{c}})=x_{c'}\qquad(\widehat{c}\in\operatorname{Fib}_{\mathcal{D}}(c')). ϕ C b ( x c ) = x c ′ ( c ∈ Fib D ( c ′ )) .
Write g \mathbf{g} g and g ^ \widehat{\mathbf{g}} g for the degrees of x c ′ x_{c'} x c ′ and x ^ c ^ \widehat{x}_{\widehat{c}} x c in these seeds, and set
ε ^ c : = ∑ c ^ ∈ Fib C ( c ) ε ^ c ^ . \widehat{\varepsilon}_c:=\sum_{\widehat{c}\in\operatorname{Fib}_{\mathcal{C}}(c)}\widehat{\varepsilon}_{\widehat{c}}. ε c := c ∈ Fib C ( c ) ∑ ε c .
The folding of cluster variables gives g = Fold C ( g ^ ) \mathbf{g}=\operatorname{Fold}_{\mathcal{C}}(\widehat{\mathbf{g}}) g = Fold C ( g ) . Together with Theorem 2.2 , this yields
ε c T L g = d i c ε ^ c T L ^ g ^ . \varepsilon_c^T L\mathbf{g}=d_{i_c}\widehat{\varepsilon}_c^T\widehat{L}\widehat{\mathbf{g}}. ε c T L g = d i c ε c T L g .
Since x c x_c x c and x ^ Fib C ( c ) \widehat{x}_{\operatorname{Fib}_{\mathcal{C}}(c)} x Fib C ( c ) are initial cluster monomials, their degrees are ε c \varepsilon_c ε c and ε ^ c \widehat{\varepsilon}_c ε c , respectively, and their F F F -polynomials are 1. Hence
⟨ x c , x c ′ ⟩ trop = ε c T L g and ⟨ x ^ Fib C ( c ) , x ^ c ^ ⟩ trop = ε ^ c T L ^ g ^ . \langle x_c,x_{c'}\rangle_{\operatorname{trop}}=\varepsilon_c^T L\mathbf{g}\quad\text{and}\quad\langle\widehat{x}_{\operatorname{Fib}_{\mathcal{C}}(c)},\widehat{x}_{\widehat{c}}\rangle_{\operatorname{trop}}=\widehat{\varepsilon}_c^T\widehat{L}\widehat{\mathbf{g}}. ⟨ x c , x c ′ ⟩ trop = ε c T L g and ⟨ x Fib C ( c ) , x c ⟩ trop = ε c T L g .
This proves (20) .
For the reverse order, let S S S and S ^ \widehat{S} S denote the maps that multiply the exchangeable coordinates by the compatibility types 2 D F 2D^{\mathcal{F}} 2 D F and 2 Id 2\operatorname{Id} 2 Id , respectively. By definition and skew-symmetry,
⟨ x c ′ , x c ⟩ trop = − ε c T L g + F x c ′ [ S ε c ] and ⟨ x ^ c ^ , x ^ Fib C ( c ) ⟩ trop = − ε ^ c T L ^ g ^ + F ^ x ^ c ^ [ S ^ ε ^ c ] . \langle x_{c'},x_c\rangle_{\operatorname{trop}}=-\varepsilon_c^T L\mathbf{g}+F_{x_{c'}}[S\varepsilon_c]\quad\text{and}\quad\langle\widehat{x}_{\widehat{c}},\widehat{x}_{\operatorname{Fib}_{\mathcal{C}}(c)}\rangle_{\operatorname{trop}}=-\widehat{\varepsilon}_c^T\widehat{L}\widehat{\mathbf{g}}+\widehat{F}_{\widehat{x}_{\widehat{c}}}[\widehat{S}\widehat{\varepsilon}_c]. ⟨ x c ′ , x c ⟩ trop = − ε c T L g + F x c ′ [ S ε c ] and ⟨ x c , x Fib C ( c ) ⟩ trop = − ε c T L g + F x c [ S ε c ] .
It remains to compare the partial F F F -invariants.
If c c c is exchangeable, then
S ε c = 2 d i c ε c , S ^ ε ^ c = 2 ε ^ c . S\varepsilon_c=2d_{i_c}\varepsilon_c,\qquad\widehat{S}\widehat{\varepsilon}_c=2\widehat{\varepsilon}_c. S ε c = 2 d i c ε c , S ε c = 2 ε c .
The coefficients of F ^ x ^ c ^ \widehat{F}_{\widehat{x}_{\widehat{c}}} F x c are nonnegative by universal positivity, so identifying variables along box fibers cannot cancel monomials. By Proposition 2.4, we can take the positive factor d i c d_{i_c} d i c outside the maximum to obtain
F x c ′ [ 2 d i c ε c ] = d i c F ^ x ^ c ^ [ 2 ε ^ c ] . F_{x_{c'}}[2d_{i_c}\varepsilon_c]=d_{i_c}\widehat{F}_{\widehat{x}_{\widehat{c}}}[2\widehat{\varepsilon}_c]. F x c ′ [ 2 d i c ε c ] = d i c F x c [ 2 ε c ] .
If c c c is frozen, every box in Fib C ( c ) \operatorname{Fib}_{\mathcal{C}}(c) Fib C ( c ) is frozen by Theorem 2.2 . Thus
S ε c = 0 , S ^ ε ^ c = 0 , S\varepsilon_c=0,\qquad\widehat{S}\widehat{\varepsilon}_c=0, S ε c = 0 , S ε c = 0 ,
so both partial F F F -invariants vanish. This proves (21) .
The left-hand sides of the two identities are independent of the choice of c ^ \widehat{c} c , so their right-hand sides are as well. Adding the identities gives (22) . □
Example 2.7. Consider the folding A 5 → C 3 A_5 \to C_3 A 5 → C 3 with
I 1 = { 1 , 5 } , I 2 = { 2 , 4 } , I 3 = { 3 } , D C 3 = diag ( 1 , 1 , 2 ) . I_1 = \{1,5\}, \qquad I_2 = \{2,4\}, \qquad I_3 = \{3\}, \qquad D^{C_3} = \operatorname{diag}(1,1,2). I 1 = { 1 , 5 } , I 2 = { 2 , 4 } , I 3 = { 3 } , D C 3 = diag ( 1 , 1 , 2 ) . For the height function ξ = ( 1 , 2 , 3 ) \xi= (1,2,3) ξ = ( 1 , 2 , 3 ) , with the convention of Subsection 4.1, the reduced Q Q Q -adapted sequence and its blockwise lift are
i = ( 1 , 2 , 1 , 3 , 2 , 1 , 3 , 2 , 3 ) ∈ Seq ( b ) , b = r i , i ^ = ( 1 , 5 , 2 , 4 , 1 , 5 , 3 , 2 , 4 , 1 , 5 , 3 , 2 , 4 , 3 ) . \mathbf{i} = (1,2,1,3,2,1,3,2,3) \in\operatorname{Seq}(\mathbf{b}), \qquad\mathbf{b} = \mathbf{r}_{\mathbf{i}}, \qquad\widehat{\mathbf{i}} = (1,5,2,4,1,5,3,2,4,1,5,3,2,4,3). i = ( 1 , 2 , 1 , 3 , 2 , 1 , 3 , 2 , 3 ) ∈ Seq ( b ) , b = r i , i = ( 1 , 5 , 2 , 4 , 1 , 5 , 3 , 2 , 4 , 1 , 5 , 3 , 2 , 4 , 3 ) . Let C \mathcal{C} C be the all-R \mathcal{R} R chain and let D \mathcal{D} D be the chain with associated sequence R R R R L L L L \mathcal{R}\mathcal{R}\mathcal{R}\mathcal{R}\mathcal{L}\mathcal{L}\mathcal{L}\mathcal{L} RRRR LLLL . Their underlying families are
F ( C ) = ( [ 1 ] , [ 2 ] , [ 1 , 3 ] , [ 4 ] , [ 2 , 5 ] , [ 1 , 6 ] , [ 4 , 7 ] , [ 2 , 8 ] , [ 4 , 9 ] ) , \mathcal{F}(\mathcal{C}) = ([1],[2],[1,3],[4],[2,5],[1,6],[4,7],[2,8],[4,9]), F ( C ) = ([ 1 ] , [ 2 ] , [ 1 , 3 ] , [ 4 ] , [ 2 , 5 ] , [ 1 , 6 ] , [ 4 , 7 ] , [ 2 , 8 ] , [ 4 , 9 ]) , F ( D ) = ( [ 5 ] , [ 6 ] , [ 7 ] , [ 5 , 8 ] , [ 7 , 9 ] , [ 4 , 9 ] , [ 3 , 6 ] , [ 2 , 8 ] , [ 1 , 6 ] ) . \mathcal{F}(\mathcal{D}) = ([5],[6],[7],[5,8],[7,9],[4,9],[3,6],[2,8],[1,6]). F ( D ) = ([ 5 ] , [ 6 ] , [ 7 ] , [ 5 , 8 ] , [ 7 , 9 ] , [ 4 , 9 ] , [ 3 , 6 ] , [ 2 , 8 ] , [ 1 , 6 ]) . Set c = [ 2 , 5 ] \mathsf{c} = [2,5] c = [ 2 , 5 ] and c ′ = [ 7 , 9 ] \mathsf{c}' = [7,9] c ′ = [ 7 , 9 ] . Then i c = 2 i_{\mathsf{c}} = 2 i c = 2 and i c ′ = 3 i_{\mathsf{c}'} = 3 i c ′ = 3 , and both c \mathsf{c} c and c ′ \mathsf{c}' c ′ are non-singleton boxes. The canonical lifted chains give
Fib C ( [ 2 , 5 ] ) = { [ 3 , 8 ] , [ 4 , 9 ] } , Fib D ( [ 7 , 9 ] ) = { [ 12 , 15 ] } . \operatorname{Fib}_{\mathcal{C}}([2,5]) = \{[3,8],[4,9]\}, \qquad\operatorname{Fib}_{\mathcal{D}}([7,9]) = \{[12,15]\}. Fib C ([ 2 , 5 ]) = {[ 3 , 8 ] , [ 4 , 9 ]} , Fib D ([ 7 , 9 ]) = {[ 12 , 15 ]} . In particular,
ϕ C D ( x ^ [ 12 , 15 ] ) = x [ 7 , 9 ] , x ^ Fib C ( [ 2 , 5 ] ) = x ^ [ 3 , 8 ] x ^ [ 4 , 9 ] . \phi^{\mathcal{D}}_{\mathcal{C}}(\widehat{x}_{[12,15]}) = x_{[7,9]}, \qquad\widehat{x}_{\operatorname{Fib}_{\mathcal{C}}([2,5])} = \widehat{x}_{[3,8]}\widehat{x}_{[4,9]}. ϕ C D ( x [ 12 , 15 ] ) = x [ 7 , 9 ] , x Fib C ([ 2 , 5 ]) = x [ 3 , 8 ] x [ 4 , 9 ] . We record a direct cluster calculation. With respect to the all-R \mathcal{R} R seed, whose exchangeable boxes are ordered as
( [ 1 ] , [ 2 ] , [ 1 , 3 ] , [ 4 ] , [ 2 , 5 ] , [ 4 , 7 ] ) , ([1],[2],[1,3],[4],[2,5],[4,7]), ([ 1 ] , [ 2 ] , [ 1 , 3 ] , [ 4 ] , [ 2 , 5 ] , [ 4 , 7 ]) , the box-move mutation from C \mathcal{C} C to D \mathcal{D} D gives
f x [ 7 , 9 ] ; C = ( 2 , 2 , 2 , 1 , 2 , 1 ) and hence f [ 2 , 5 ] ; x [ 7 , 9 ] = 2. \mathbf{f}_{x_{[7,9]};\mathcal{C}} = (2,2,2,1,2,1) \qquad\text{and hence} \qquad f_{[2,5];x_{[7,9]}} = 2. f x [ 7 , 9 ] ; C = ( 2 , 2 , 2 , 1 , 2 , 1 ) and hence f [ 2 , 5 ] ; x [ 7 , 9 ] = 2. The lifted mutation calculation gives
F ^ x ^ [ 12 , 15 ] [ e [ 3 , 8 ] + e [ 4 , 9 ] ] = 2. \widehat{F}_{\widehat{x}_{[12,15]}}\left[e_{[3,8]} + e_{[4,9]}\right] = 2. F x [ 12 , 15 ] [ e [ 3 , 8 ] + e [ 4 , 9 ] ] = 2. The compatibility types have diagonal entries 2 d i c 2d_{i_{\mathsf{c}}} 2 d i c at exchangeable boxes c \mathsf{c} c downstairs and 2 2 2 upstairs. Since d 2 = 1 d_2 = 1 d 2 = 1 , we obtain
( x [ 2 , 5 ] ∥ x [ 7 , 9 ] ) F = 2 d 2 f [ 2 , 5 ] ; x [ 7 , 9 ] = 4 , ( x ^ [ 3 , 8 ] x ^ [ 4 , 9 ] ∥ x ^ [ 12 , 15 ] ) F = F ^ x ^ [ 12 , 15 ] [ 2 ( e [ 3 , 8 ] + e [ 4 , 9 ] ) ] = 4. \left(x_{[2,5]} \Vert x_{[7,9]}\right)_{\mathcal{F}} = 2d_2 f_{[2,5];x_{[7,9]}} = 4, \qquad\left(\widehat{x}_{[3,8]}\widehat{x}_{[4,9]} \Vert\widehat{x}_{[12,15]}\right)_{\mathcal{F}} = \widehat{F}_{\widehat{x}_{[12,15]}}\left[2\left(e_{[3,8]} + e_{[4,9]}\right)\right] = 4. ( x [ 2 , 5 ] ∥ x [ 7 , 9 ] ) F = 2 d 2 f [ 2 , 5 ] ; x [ 7 , 9 ] = 4 , ( x [ 3 , 8 ] x [ 4 , 9 ] ∥ x [ 12 , 15 ] ) F = F x [ 12 , 15 ] [ 2 ( e [ 3 , 8 ] + e [ 4 , 9 ] ) ] = 4. Thus
( x [ 2 , 5 ] ∥ x [ 7 , 9 ] ) F = d 2 ( x ^ [ 3 , 8 ] x ^ [ 4 , 9 ] ∥ x ^ [ 12 , 15 ] ) F . \left(x_{[2,5]} \Vert x_{[7,9]}\right)_{\mathcal{F}} = d_2\left(\widehat{x}_{[3,8]}\widehat{x}_{[4,9]} \Vert\widehat{x}_{[12,15]}\right)_{\mathcal{F}}. ( x [ 2 , 5 ] ∥ x [ 7 , 9 ] ) F = d 2 ( x [ 3 , 8 ] x [ 4 , 9 ] ∥ x [ 12 , 15 ] ) F .
Categorical folding of Λ \Lambda Λ - and d d d -invariants
For a reduced sequence, the determinantial modules attached to i \mathbf{i} i -boxes realize the box seeds of § 3 . We use these realizations to convert the two ordered tropical folding identities into formulas for the Λ \Lambda Λ -invariants involving sums over box fibers. Symmetrizing the two formulas gives the folding formula for the d d d -invariant.
Quiver Hecke categories and affreal modules
We refer to [30 ] for the basic definitions and standard results on quiver Hecke algebras and their finite-dimensional graded module categories. Keep the Cartan and root data of § 3 . Let P P P be a weight lattice containing the fundamental weights ( ϖ i ) i ∈ I (\varpi_i)_{i\in I} ( ϖ i ) i ∈ I , and set
P + : = ∑ i ∈ I Z ≥ 0 ϖ i . P_+ := \sum_{i\in I} \mathbb{Z}_{\geq0}\varpi_i. P + := i ∈ I ∑ Z ≥ 0 ϖ i .
Fix a quiver Hecke algebra [46 , 47 , 56 ]
R = ⨁ β ∈ Q + R ( β ) R = \bigoplus_{\beta\in Q_+} R(\beta) R = β ∈ Q + ⨁ R ( β )
over a field k \Bbbk k , associated with the Cartan datum of g \mathfrak{g} g . We denote by R R R -gmod the category of finite-dimensional graded R R R -modules and by ∘ \circ ∘ its convolution product.
We generally suppress grading shifts from the notation and understand isomorphisms of graded modules up to a grading shift unless stated otherwise. At the level of Grothendieck rings, we use the ungraded specialization obtained by setting the grading parameter equal to 1 1 1 . The grading is retained in the degrees of renormalized R R R -matrices and hence in the invariants Λ ( − , − ) \Lambda(-,-) Λ ( − , − ) and d ( − , − ) \mathfrak{d}(-,-) d ( − , − ) .
For simple R R R -modules M M M and N N N , we say that they commute if
M ∘ N ≃ N ∘ M M\circ N \simeq N\circ M M ∘ N ≃ N ∘ M
up to a grading shift, and that they strongly commute if, in addition, M ∘ N M\circ N M ∘ N is simple. A simple R R R -module M M M is called real if M ∘ M M\circ M M ∘ M is simple.
A simple R R R -module M M M is called affreal if it is real and admits an affinization ([44 , 36 ] ).
For simple R R R -modules M M M and N N N , suppose that
Hom R ( M ∘ N , N ∘ M ) = k r M , N \operatorname{Hom}_{R}(M\circ N,N\circ M)=\Bbbk r_{M,N} Hom R ( M ∘ N , N ∘ M ) = k r M , N
for a nonzero homogeneous homomorphism r M , N r_{M,N} r M , N . The map r M , N r_{M,N} r M , N is unique up to a nonzero scalar and is called the renormalized R R R -matrix . We call the ordered pair ( M , N ) (M,N) ( M , N ) Λ \Lambda Λ -definable and set
Λ ( M , N ) : = deg ( r M , N ) . \Lambda(M,N):=\deg(r_{M,N}). Λ ( M , N ) := deg ( r M , N ) .
If both ordered pairs ( M , N ) (M,N) ( M , N ) and ( N , M ) (N,M) ( N , M ) are Λ \Lambda Λ -definable, we set
d ( M , N ) : = 1 2 ( Λ ( M , N ) + Λ ( N , M ) ) . \mathfrak{d}(M,N):=\frac{1}{2}\bigl(\Lambda(M,N)+\Lambda(N,M)\bigr). d ( M , N ) := 2 1 ( Λ ( M , N ) + Λ ( N , M ) ) .
In particular, if either M M M or N N N is affreal, then both ( M , N ) (M,N) ( M , N ) and ( N , M ) (N,M) ( N , M ) are Λ \Lambda Λ -definable, and hence d ( M , N ) \mathfrak{d}(M,N) d ( M , N ) is defined.
A simple module S S S is called a convolution factor of a simple module M M M if there exists a simple module X X X such that
M ≃ S ∘ X M\simeq S\circ X M ≃ S ∘ X
up to a grading shift ([41 Section 2.3] ).
We use the standard R R R -matrix calculus for affreal simple modules, including composition and additivity under convolution ([36 Lemma 3.16] and [41 Proposition 2.6] ).
We next recall normal sequences in the almost affreal form of [41 Section 2.3] .
Definition 3.1. [32 Definition 2.5] Let M 1 , … , M r M_1,\ldots,M_r M 1 , … , M r be simple R R R -modules, all but at most one of which are affreal. Let
r M 1 , … , M r : = ( r M r − 1 , M r ) ∘ ⋯ ∘ ( r M 2 , M r ∘ ⋯ ∘ r M 2 , M 3 ) ∘ ( r M 1 , M r ∘ ⋯ ∘ r M 1 , M 2 ) : M 1 ∘ ⋯ ∘ M r ⟶ M r ∘ ⋯ ∘ M 1 . \mathbf{r}_{M_1,\ldots,M_r}:=(\mathbf{r}_{M_{r-1},M_r})\circ\cdots\circ(\mathbf{r}_{M_2,M_r}\circ\cdots\circ\mathbf{r}_{M_2,M_3})\circ(\mathbf{r}_{M_1,M_r}\circ\cdots\circ\mathbf{r}_{M_1,M_2}):M_1\circ\cdots\circ M_r\longrightarrow M_r\circ\cdots\circ M_1. r M 1 , … , M r := ( r M r − 1 , M r ) ∘ ⋯ ∘ ( r M 2 , M r ∘ ⋯ ∘ r M 2 , M 3 ) ∘ ( r M 1 , M r ∘ ⋯ ∘ r M 1 , M 2 ) : M 1 ∘ ⋯ ∘ M r ⟶ M r ∘ ⋯ ∘ M 1 . The ordered sequence ( M 1 , … , M r ) (M_1,\ldots,M_r) ( M 1 , … , M r ) is called normal if
r M 1 , … , M r ≠ 0. \mathbf{r}_{M_1,\ldots,M_r}\ne0. r M 1 , … , M r = 0.
For the almost affreal form of the following lemma, see [37 Lemma 2.8 and Section 2.2] .
Lemma 3.2 ([32 Lemmas 2.6–2.8] ). Let ( M 1 , … , M r ) (M_1,\ldots,M_r) ( M 1 , … , M r ) be a normal sequence as in Definition 3.1 . Then
Im ( r M 1 , … , M r ) ≃ hd ( M 1 ∘ ⋯ ∘ M r ) ≃ soc ( M r ∘ ⋯ ∘ M 1 ) , \operatorname{Im}(\mathbf{r}_{M_1,\ldots,M_r})\simeq\operatorname{hd}(M_1\circ\cdots\circ M_r)\simeq\operatorname{soc}(M_r\circ\cdots\circ M_1), Im ( r M 1 , … , M r ) ≃ hd ( M 1 ∘ ⋯ ∘ M r ) ≃ soc ( M r ∘ ⋯ ∘ M 1 ) , and these modules are simple. Moreover, every consecutive ordered subsequence of ( M 1 , … , M r ) (M_1,\ldots,M_r) ( M 1 , … , M r ) is normal, and
Λ ( M 1 , hd ( M 2 ∘ ⋯ ∘ M r ) ) = ∑ a = 2 r Λ ( M 1 , M a ) , \Lambda(M_1,\operatorname{hd}(M_2\circ\cdots\circ M_r))=\sum_{a=2}^{r}\Lambda(M_1,M_a), Λ ( M 1 , hd ( M 2 ∘ ⋯ ∘ M r )) = a = 2 ∑ r Λ ( M 1 , M a ) , Λ ( hd ( M 1 ∘ ⋯ ∘ M r − 1 ) , M r ) = ∑ a = 1 r − 1 Λ ( M a , M r ) . \Lambda(\operatorname{hd}(M_1\circ\cdots\circ M_{r-1}),M_r)=\sum_{a=1}^{r-1}\Lambda(M_a,M_r). Λ ( hd ( M 1 ∘ ⋯ ∘ M r − 1 ) , M r ) = a = 1 ∑ r − 1 Λ ( M a , M r ) .
For M ∈ R ( β ) M\in R(\beta) M ∈ R ( β ) -gmod, its weight is wt ( M ) = − β \operatorname{wt}(M)=-\beta wt ( M ) = − β . Set
W ( M ) : = { γ ∈ Q + ∩ ( β − Q + ) ∣ e ( γ , β − γ ) M ≠ 0 } . W(M):=\{\gamma\in Q_+\cap(\beta-Q_+)\mid e(\gamma,\beta-\gamma)M\ne0\}. W ( M ) := { γ ∈ Q + ∩ ( β − Q + ) ∣ e ( γ , β − γ ) M = 0 } .
Here e ( γ , β − γ ) ∈ R ( β ) e(\gamma,\beta-\gamma)\in R(\beta) e ( γ , β − γ ) ∈ R ( β ) is the standard quiver Hecke idempotent associated with the decomposition β = γ + ( β − γ ) \beta=\gamma+(\beta-\gamma) β = γ + ( β − γ ) (see [34 Section 2.3] for instances). For w ∈ W g w\in W_{\mathfrak{g}} w ∈ W g , let C w \mathcal{C}_w C w be the full subcategory of R R R -gmod consisting of the modules M M M such that
W ( M ) ⊂ Q + ∩ w Q − . W(M)\subset Q_+\cap wQ_-. W ( M ) ⊂ Q + ∩ w Q − .
The category C w \mathcal{C}_w C w is stable under convolution products, extensions, subquotients, and grading shifts. The relation between the Grothendieck ring of C w \mathcal{C}_w C w and the cluster algebra associated with w w w will be recalled in the next subsection. We set
K ( C w ) : = C ⊗ Z K 0 ( C w ) . K(\mathcal{C}_w):=\mathbb{C}\otimes_{\mathbb{Z}}K_0(\mathcal{C}_w). K ( C w ) := C ⊗ Z K 0 ( C w ) .
Determinantial modules and categorical realizations
We recall the determinantial modules attached to i-boxes and their realization of box seeds in C w \mathcal{C}_w C w .
For λ ∈ P + \lambda\in P_+ λ ∈ P + and μ , ζ ∈ W g λ \mu,\zeta\in W_{\mathfrak{g}}\lambda μ , ζ ∈ W g λ with μ ⪯ ζ \mu\preceq\zeta μ ⪯ ζ , let
denote the determinantial R ( ζ − μ ) R(\zeta-\mu) R ( ζ − μ ) -module associated with ( μ , ζ ) (\mu,\zeta) ( μ , ζ ) . It is a self-dual real simple module by [34 Propositions 4.1 and 4.2] and admits an affinization by [36 Theorem 3.26] . Hence M ( μ , ζ ) M(\mu,\zeta) M ( μ , ζ ) is affreal.
Let w ∈ W g w\in W_{\mathfrak{g}} w ∈ W g , and set
Red ( w ) : = { ( i 1 , … , i l ) ∈ I l ∣ w = s i 1 ⋯ s i l , l = ℓ ( w ) } . \operatorname{Red}(w):=\{(i_1,\ldots,i_l)\in I^l\mid w=s_{i_1}\cdots s_{i_l},\ l=\ell(w)\}. Red ( w ) := {( i 1 , … , i l ) ∈ I l ∣ w = s i 1 ⋯ s i l , l = ℓ ( w )} .
Fix i = ( i 1 , … , i l ) ∈ Red ( w ) \mathbf{i}=(i_1,\ldots,i_l)\in\operatorname{Red}(w) i = ( i 1 , … , i l ) ∈ Red ( w ) and define
w ≤ k i : = s i 1 ⋯ s i k , w < k i : = s i 1 ⋯ s i k − 1 ( 1 ≤ k ≤ l ) . w^i_{\leq k}:=s_{i_1}\cdots s_{i_k},\qquad w^i_{<k}:=s_{i_1}\cdots s_{i_{k-1}}\qquad(1\leq k\leq l). w ≤ k i := s i 1 ⋯ s i k , w < k i := s i 1 ⋯ s i k − 1 ( 1 ≤ k ≤ l ) .
For 1 ≤ k ≤ l 1\leq k\leq l 1 ≤ k ≤ l , define the cuspidal module associated with the position k k k of i \mathbf{i} i by
S k i : = M ( w ≤ k i ϖ i k , w < k i ϖ i k ) . (23) S^i_k:=M(w^i_{\leq k}\varpi_{i_k},w^i_{<k}\varpi_{i_k}). \tag*{(23)} S k i := M ( w ≤ k i ϖ i k , w < k i ϖ i k ) . (23)
Let c = [ x , y ] \mathbf{c}=[x,y] c = [ x , y ] be an i \mathbf{i} i -box associated with i \mathbf{i} i . Thus
1 ≤ x ≤ y ≤ l and i x = i y = i c . 1\leq x\leq y\leq l\qquad\text{and}\qquad i_x=i_y=i_c. 1 ≤ x ≤ y ≤ l and i x = i y = i c .
We define the determinantial module associated with c \mathbf{c} c by
M i [ c ] : = M i [ x , y ] : = M ( w ≤ y i ϖ i y , w < x i ϖ i x ) . (24) M^i[\mathbf{c}]:=M^i[x,y]:=M(w^i_{\leq y}\varpi_{i_y},w^i_{<x}\varpi_{i_x}). \tag*{(24)} M i [ c ] := M i [ x , y ] := M ( w ≤ y i ϖ i y , w < x i ϖ i x ) . (24)
In particular, M i [ k ] = S k i M^i[k]=S^i_k M i [ k ] = S k i .
Let c = [ x , y ] \mathbf{c}=[x,y] c = [ x , y ] be an i \mathbf{i} i -box associated with i \mathbf{i} i , and write
x = x 0 < x 1 < ⋯ < x r − 1 = y x=x_0<x_1<\cdots<x_{r-1}=y x = x 0 < x 1 < ⋯ < x r − 1 = y
for the positions of color i c i_c i c in [ x , y ] [x,y] [ x , y ] . Then [38 Proposition 4.3] and [33 Section 5.1.1] give
( S x r − 1 i , … , S x 1 i , S x 0 i ) is a normal sequence and M i [ c ] ≃ hd ( S x r − 1 i ∘ ⋯ ∘ S x 1 i ∘ S x 0 i ) . (25) \left(S^i_{x_{r-1}},\ldots,S^i_{x_1},S^i_{x_0}\right)\text{ is a normal sequence}\qquad\text{and}\qquad M^i[\mathbf{c}]\simeq\operatorname{hd}\left(S^i_{x_{r-1}}\circ\cdots\circ S^i_{x_1}\circ S^i_{x_0}\right). \tag*{(25)} ( S x r − 1 i , … , S x 1 i , S x 0 i ) is a normal sequence and M i [ c ] ≃ hd ( S x r − 1 i ∘ ⋯ ∘ S x 1 i ∘ S x 0 i ) . (25)
The following formula for one order of the Λ \Lambda Λ -invariant is the form of [38 Proposition 5.7, (5.5)–(5.6)] in our sign convention.
Proposition 3.3. Let [ x , y ] [x,y] [ x , y ] and [ x ′ , y ′ ] [x',y'] [ x ′ , y ′ ] be two i \mathbf{i} i -boxes for the same reduced sequence i \mathbf{i} i , and put
M = M i [ x , y ] , N = M i [ x ′ , y ′ ] . M=M^i[x,y],\qquad N=M^i[x',y']. M = M i [ x , y ] , N = M i [ x ′ , y ′ ] . If x > ( x ′ ) − x>(x')^- x > ( x ′ ) − or y + > y ′ y^+>y' y + > y ′ , then
Λ ( M , N ) = − λ i ( E [ x , y ] , E [ x ′ , y ′ ] ) . (26) \Lambda(M,N)=-\lambda^i(\mathcal{E}_{[x,y]},\mathcal{E}_{[x',y']}) . \tag*{(26)} Λ ( M , N ) = − λ i ( E [ x , y ] , E [ x ′ , y ′ ] ) . (26) Consequently,
Λ ( N , M ) = 2 d ( M , N ) + λ i ( E [ x , y ] , E [ x ′ , y ′ ] ) . \Lambda(N,M)=2\mathfrak{d}(M,N)+\lambda^i(\mathcal{E}_{[x,y]},\mathcal{E}_{[x',y']}). Λ ( N , M ) = 2 d ( M , N ) + λ i ( E [ x , y ] , E [ x ′ , y ′ ] ) . After interchanging the boxes, if x ′ > x − x'>x^- x ′ > x − or ( y ′ ) + > y (y')^+>y ( y ′ ) + > y , the same proposition computes Λ ( N , M ) \Lambda(N,M) Λ ( N , M ) directly.
Remark 3.4. Let M = M i [ x , y ] M=M^i[x,y] M = M i [ x , y ] and N = M i [ x ′ , y ′ ] N=M^i[x',y'] N = M i [ x ′ , y ′ ] be two box modules associated with the same reduced sequence i \mathbf{i} i . At least one of the two orders satisfies the hypotheses of Proposition 3.3. Indeed, if x ≥ x ′ x\geq x' x ≥ x ′ , then x > ( x ′ ) − x>(x')^- x > ( x ′ ) − , while if x ′ > x x'>x x ′ > x , then x ′ > x − x'>x^- x ′ > x − . Thus the pairing λ i \lambda^i λ i determines one of Λ ( M , N ) \Lambda(M,N) Λ ( M , N ) and Λ ( N , M ) \Lambda(N,M) Λ ( N , M ) , and the identity
Λ ( M , N ) + Λ ( N , M ) = 2 d ( M , N ) \Lambda(M,N)+\Lambda(N,M)=2\mathfrak{d}(M,N) Λ ( M , N ) + Λ ( N , M ) = 2 d ( M , N )
determines the other once d ( M , N ) \mathfrak{d}(M,N) d ( M , N ) is known.
If the hypotheses hold in both orders, skew-symmetry of λ i \lambda^i λ i gives d ( M , N ) = 0 \mathfrak{d}(M,N)=0 d ( M , N ) = 0 .
Let K K K be a finite index set and let
M = { M k ∣ k ∈ K } \mathcal{M}=\{M_k\mid k\in K\} M = { M k ∣ k ∈ K }
be a family of affreal simple modules. We call M \mathcal{M} M a commuting family if
M k ∘ M l ≃ M l ∘ M k ( k , l ∈ K ) . M_k\circ M_l\simeq M_l\circ M_k\qquad(k,l\in K). M k ∘ M l ≃ M l ∘ M k ( k , l ∈ K ) .
up to a grading shift. For a = ( a k ) k ∈ K ∈ Z ≥ 0 ⊕ K \mathbf{a}=(a_k)_{k\in K}\in\mathbb{Z}_{\geq0}^{\oplus K} a = ( a k ) k ∈ K ∈ Z ≥ 0 ⊕ K , the corresponding convolution product is simple and independent of the ordering of its factors, up to a grading shift. We denote it by
M ( a ) : = ∘ k ∈ K M k ∘ a k . M(\mathbf{a}):=\mathop{\circ}_{k\in K}M_k^{\circ a_k}. M ( a ) := ∘ k ∈ K M k ∘ a k .
Thus
M ( a ) ∘ M ( b ) ≅ M ( a + b ) ( a , b ∈ Z ≥ 0 ⊕ K ) M(\mathbf{a})\circ M(\mathbf{b})\cong M(\mathbf{a}+\mathbf{b}) \qquad(\mathbf{a},\mathbf{b}\in\mathbb{Z}_{\geq0}^{\oplus K}) M ( a ) ∘ M ( b ) ≅ M ( a + b ) ( a , b ∈ Z ≥ 0 ⊕ K )
up to a grading shift.
A simple module X X X is called an M-monomial if X ≅ M ( a ) X\cong M(\mathbf{a}) X ≅ M ( a ) for some a ∈ Z ≥ 0 ⊕ K \mathbf{a}\in\mathbb{Z}_{\geq0}^{\oplus K} a ∈ Z ≥ 0 ⊕ K . The commuting family M is called independent if M ( a ) ≅ M ( b ) M(\mathbf{a})\cong M(\mathbf{b}) M ( a ) ≅ M ( b ) implies a = b \mathbf{a}=\mathbf{b} a = b .
Let C \mathcal{C} C be an admissible chain of i-boxes associated with i \boldsymbol{i} i , and set
F : = F ( C ) . \mathcal{F}:=\mathcal{F}(\mathcal{C}). F := F ( C ) .
For c ∈ F \mathbf{c}\in\mathcal{F} c ∈ F , write
M c i : = M i [ c ] , M F i : = { M c i ∣ c ∈ F } . M_{\mathbf{c}}^i:=M^i[\mathbf{c}],\qquad M_{\mathcal{F}}^i:=\{M_{\mathbf{c}}^i\mid\mathbf{c}\in\mathcal{F}\}. M c i := M i [ c ] , M F i := { M c i ∣ c ∈ F } .
Lemma 3.5 ([39 ] , Theorem 4.21(b); [33 ] , Proposition 5.1). Let c \mathbf{c} c and c ′ \mathbf{c}' c ′ be commuting i \boldsymbol{i} i -boxes associated with i \boldsymbol{i} i . Then
M i [ c ] and M i [ c ′ ] strongly commute. M^i[\mathbf{c}] \text{ and } M^i[\mathbf{c}'] \text{ strongly commute.} M i [ c ] and M i [ c ′ ] strongly commute.
Proposition 3.6. The family M F i M_{\mathcal{F}}^i M F i is an independent commuting family of affreal simple objects in C w \mathcal{C}_w C w . More precisely, for any c , c ′ ∈ F \mathbf{c},\mathbf{c}'\in\mathcal{F} c , c ′ ∈ F , the modules M c i M_{\mathbf{c}}^i M c i and M c ′ i M_{\mathbf{c}'}^i M c ′ i strongly commute. Consequently, for every a = ( a c ) c ∈ F ∈ Z ≥ 0 ⊕ F \mathbf{a}=(a_{\mathbf{c}})_{\mathbf{c}\in\mathcal{F}}\in\mathbb{Z}_{\geq0}^{\oplus\mathcal{F}} a = ( a c ) c ∈ F ∈ Z ≥ 0 ⊕ F , the convolution product
M F i ( a ) : = ∘ c ∈ F ( M c i ) ∘ a c M_{\mathcal{F}}^i(\mathbf{a}):=\mathop{\circ}_{\mathbf{c}\in\mathcal{F}}(M_{\mathbf{c}}^i)^{\circ a_{\mathbf{c}}} M F i ( a ) := ∘ c ∈ F ( M c i ) ∘ a c is simple and independent of the ordering of its factors, up to a grading shift.
Proof. The affreal property and Lemma 3.5 give the assertions on convolution products. The independence follows from the algebraic independence of the corresponding cluster variables under the Grothendieck-ring realization in Theorem 3.9 . □
We next recall the cluster algebra structure on the Grothendieck ring of C w \mathcal{C}_w C w . Let A q ( n ( w ) ) A_q(\mathfrak{n}(w)) A q ( n ( w )) denote the quantum unipotent coordinate algebra associated with w w w , and write A q ( n ( w ) ) Z [ q ± 1 ] A_q(\mathfrak{n}(w))_{\mathbb{Z}[q^{\pm1}]} A q ( n ( w ) ) Z [ q ± 1 ] for its Z [ q ± 1 ] \mathbb{Z}[q^{\pm1}] Z [ q ± 1 ] -form. Let K q ( C w ) K_q(\mathcal{C}_w) K q ( C w ) denote the graded Grothendieck ring of C w \mathcal{C}_w C w , where a grading shift acts by a power of q q q . The quiver Hecke categorification induces an isomorphism of Z [ q ± 1 ] \mathbb{Z}[q^{\pm1}] Z [ q ± 1 ] -algebras
Ω w : K q ( C w ) → ∼ A q ( n ( w ) ) Z [ q ± 1 ] (27) \Omega_w:K_q(\mathcal{C}_w)\xrightarrow{\sim}A_q(\mathfrak{n}(w))_{\mathbb{Z}[q^{\pm1}]} \tag*{(27)} Ω w : K q ( C w ) ∼ A q ( n ( w ) ) Z [ q ± 1 ] (27)
(see [34 ] , Theorem 2.20(ii-a)). Under this isomorphism, the class of a determinantial module M ( μ , ζ ) M(\mu,\zeta) M ( μ , ζ ) is sent to the corresponding unipotent quantum minor D ( μ , ζ ) D(\mu,\zeta) D ( μ , ζ ) (see [34 ] , Proposition 4.1 for more detail):
Ω w ( [ M ( μ , ζ ) ] ) = D ( μ , ζ ) . \Omega_w([M(\mu,\zeta)])=D(\mu,\zeta). Ω w ([ M ( μ , ζ )]) = D ( μ , ζ ) .
The algebra A q ( n ( w ) ) A_q(\mathfrak{n}(w)) A q ( n ( w )) admits a quantum cluster algebra structure associated with every reduced sequence i ∈ Red ( w ) \boldsymbol{i}\in\operatorname{Red}(w) i ∈ Red ( w ) . In the symmetric setting, this structure originates from the work of Geiß–Leclerc–Schröer, while its symmetrizable counterpart is due to Goodearl–Yakimov [15 , 16 , 17 , 18 ] .
Let
R R i : = { { 1 , k } ∣ 1 ≤ k ≤ l } and F R i : = { { 1 , k } ∣ 1 ≤ k ≤ l } . \mathcal{R}_{R}^{i}:=\{ \{1,k\}\mid1\leq k\leq l\}
\qquad\text{and}\qquad
\mathcal{F}_{R}^{i}:=\{\{1,k\}\mid1\leq k\leq l\}. R R i := {{ 1 , k } ∣ 1 ≤ k ≤ l } and F R i := {{ 1 , k } ∣ 1 ≤ k ≤ l } .
be the constant R \mathcal{R} R -chain and its underlying maximal commuting family. For 1 ≤ k ≤ l 1 \le k \le l 1 ≤ k ≤ l , one has
M i [ { 1 , k } ] ≃ M ( w ≤ k i ϖ i k , ϖ i k ) . M^{\mathbf{i}}[\{1,k\}] \simeq M\left(w_{\leq k}^{\mathbf{i}}\varpi_{i_k},\varpi_{i_k}\right). M i [{ 1 , k }] ≃ M ( w ≤ k i ϖ i k , ϖ i k ) .
Consequently,
Ω w ( M i [ { 1 , k } ] ) = D ( w ≤ k i ϖ i k , ϖ i k ) . \Omega_w\left(M^{\mathbf{i}}[\{1,k\}]\right)=D\left(w_{\leq k}^{\mathbf{i}}\varpi_{i_k},\varpi_{i_k}\right). Ω w ( M i [{ 1 , k }] ) = D ( w ≤ k i ϖ i k , ϖ i k ) .
These unipotent quantum minors form the initial cluster associated with i \mathbf{i} i . We transport this quantum cluster algebra structure along Ω w \Omega_w Ω w . Specializing the grading parameter to 1 1 1 , we obtain a C \mathbb{C} C -algebra isomorphism
Φ w : K ( C w ) ⟶ ∼ A ( w ) , \Phi_w:K(\mathcal{C}_w)\overset{\sim}{\longrightarrow}\mathcal{A}(w), Φ w : K ( C w ) ⟶ ∼ A ( w ) ,
where A ( w ) \mathcal{A}(w) A ( w ) is the corresponding classical cluster algebra. We use the opposite exchange matrices in the specialized classical seed pattern, in accordance with our convention for B F B^{\mathcal{F}} B F . Under this isomorphism,
Φ w ( [ M c i ] ) = x c ( c ∈ F R i ) . (28) \Phi_w\left([M^{\mathbf{i}}_c]\right)=x_c \qquad(c\in\mathcal{F}^{\mathbf{i}}_{\mathcal{R}}). \tag*{(28)} Φ w ( [ M c i ] ) = x c ( c ∈ F R i ) . (28)
Thus the family
M R i : = { M c i ∣ c ∈ F R i } M^{\mathbf{i}}_{\mathcal{R}}:=\{M^{\mathbf{i}}_c\mid c\in\mathcal{F}^{\mathbf{i}}_{\mathcal{R}}\} M R i := { M c i ∣ c ∈ F R i }
realizes the initial box seed Σ i ( F R i ) \Sigma^{\mathbf{i}}(\mathcal{F}^{\mathbf{i}}_{\mathcal{R}}) Σ i ( F R i ) .
We next show that mutation-type box moves are realized by categorical mutations. Let C = ( c k ) 1 ≤ k ≤ l \mathscr{C}=(c_k)_{1\leq k\leq l} C = ( c k ) 1 ≤ k ≤ l be an admissible chain of i \mathbf{i} i -boxes associated with i \mathbf{i} i , and set F : = F ( C ) \mathcal{F}:=\mathcal{F}(\mathscr{C}) F := F ( C ) and M F i = { M c i ∣ c ∈ F } M^{\mathbf{i}}_{\mathcal{F}}=\{M^{\mathbf{i}}_c\mid c\in\mathcal{F}\} M F i = { M c i ∣ c ∈ F } . The box-move terminology and the exchanged pair { [ u + , v ] , [ u , v − ] } \{[u^+,v],[u,v^-]\} {[ u + , v ] , [ u , v − ]} are as in Lemma 2.1.
Lemma 3.7 ([33 Proposition 5.2] ). Let [ u , v ] [u,v] [ u , v ] be an i \mathbf{i} i -box of color i i i with u < v u<v u < v . Up to grading shifts, there is a short exact sequence
0 ⟶ ◯ j ∈ I ∖ { i } c j , i < 0 M i [ u ( j ) + , v ( j ) − ] ∘ ( − c j , i ) ⟶ M i [ u + , v ] ∘ M i [ u , v − ] ⟶ M i [ u + , v − ] ∘ M i [ u , v ] ⟶ 0. (29) 0\longrightarrow\mathop{\bigcirc}_{\substack{j\in I\setminus\{i\}\\ c_{j,i}<0}} M^{\mathbf{i}}[u(j)^+,v(j)^-]^{\circ(-c_{j,i})}\longrightarrow M^{\mathbf{i}}[u^+,v]\circ M^{\mathbf{i}}[u,v^-]\longrightarrow M^{\mathbf{i}}[u^+,v^-]\circ M^{\mathbf{i}}[u,v]\longrightarrow0. \tag*{(29)} 0 ⟶ ◯ j ∈ I ∖ { i } c j , i < 0 M i [ u ( j ) + , v ( j ) − ] ∘ ( − c j , i ) ⟶ M i [ u + , v ] ∘ M i [ u , v − ] ⟶ M i [ u + , v − ] ∘ M i [ u , v ] ⟶ 0. (29) An undefined or empty interval is omitted, and an empty convolution product is understood to be the monoidal unit. We call (29) the T T T -system.
Proposition 3.8. Let C ′ : = B k ( C ) \mathscr{C}':=\mathcal{B}_k(\mathscr{C}) C ′ := B k ( C ) be a mutation-type box move replacing
c ∈ F ( C ) with c ⋆ ∈ F ( C ′ ) . c\in\mathcal{F}(\mathscr{C})\quad\text{with}\quad c^\star\in\mathcal{F}(\mathscr{C}'). c ∈ F ( C ) with c ⋆ ∈ F ( C ′ ) .
Assume that
( ( [ M c i ] ) c ∈ F ( C ) , B F ( C ) ) \left(\left([M^{\mathbf{i}}_c]\right)_{c\in\mathcal{F}(\mathscr{C})},B^{\mathcal{F}(\mathscr{C})}\right) ( ( [ M c i ] ) c ∈ F ( C ) , B F ( C ) )
is a seed of K ( C w ) K(\mathcal{C}_w) K ( C w ) . Then, up to the natural relabeling,
μ c ( ( ( [ M c i ] ) c ∈ F ( C ) , B F ( C ) ) ) = ( ( [ M c i ] ) c ∈ F ( C ′ ) , B F ( C ′ ) ) . \mu_c\left(\left(\left([M^{\mathbf{i}}_c]\right)_{c\in\mathcal{F}(\mathscr{C})},B^{\mathcal{F}(\mathscr{C})}\right)\right)
=
\left(\left([M^{\mathbf{i}}_c]\right)_{c\in\mathcal{F}(\mathscr{C}')},B^{\mathcal{F}(\mathscr{C}')}\right). μ c ( ( ( [ M c i ] ) c ∈ F ( C ) , B F ( C ) ) ) = ( ( [ M c i ] ) c ∈ F ( C ′ ) , B F ( C ′ ) ) .
Proof. Put F = F ( C ) \mathcal{F}=\mathcal{F}(\mathscr{C}) F = F ( C ) and F ′ = F ( C ′ ) \mathcal{F}'=\mathcal{F}(\mathscr{C}') F ′ = F ( C ′ ) . By Lemma 2.1, there is an i \mathbf{i} i -box [ u , v ] [u,v] [ u , v ] of color i i i such that
{ c , c ⋆ } = { [ u + , v ] , [ u , v − ] } , c k + 1 = [ u , v ] . \{c,c^\star\}=\{[u^+,v],[u,v^-]\},\qquad c_{k+1}=[u,v]. { c , c ⋆ } = {[ u + , v ] , [ u , v − ]} , c k + 1 = [ u , v ] .
Both box modules are affreal, and each strongly commutes with the modules indexed by F ∩ F ′ \mathcal{F}\cap\mathcal{F}' F ∩ F ′ . The first k − 1 k-1 k − 1 boxes are unchanged by the move and have envelope [ u + 1 , v − 1 ] [u+1,v-1] [ u + 1 , v − 1 ] . For each color j ≠ i j\ne i j = i occurring in this interval, set
r j = max { r < k ∣ i c r = j } . r_j=\max\{r<k\mid i_{c_r}=j\}. r j = max { r < k ∣ i c r = j } .
By the definition of an admissible chain,
c r j = [ u ( j ) + , v ( j ) − ] . c_{r_j}=[u(j)^+,v(j)^-]. c r j = [ u ( j ) + , v ( j ) − ] .
If [ u + , v − ] [u^+,v^-] [ u + , v − ] is nonempty, it is the last box of color i i i among the first k − 1 k-1 k − 1 boxes. Thus all nonempty boxes appearing in the two end terms of (29) belong to F ∩ F ′ \mathcal F\cap\mathcal F' F ∩ F ′ . By [33 (3.1)] , these end terms correspond to the two exchange monomials. After specializing the grading parameter to 1 1 1 , we obtain
[ M c i ] [ M c ∗ i ] = ∏ e ∈ F [ M e i ] [ b e , c F ] + + ∏ e ∈ F [ M e i ] [ − b e , c F ] + . [M_{\mathfrak c}^{i}][M_{\mathfrak c_*}^{i}]=\prod_{\mathfrak e\in\mathcal F}[M_{\mathfrak e}^{i}]^{[b_{\mathfrak e,\mathfrak c}^{\mathcal F}]_+}+\prod_{\mathfrak e\in\mathcal F}[M_{\mathfrak e}^{i}]^{[-b_{\mathfrak e,\mathfrak c}^{\mathcal F}]_+}. [ M c i ] [ M c ∗ i ] = e ∈ F ∏ [ M e i ] [ b e , c F ] + + e ∈ F ∏ [ M e i ] [ − b e , c F ] + .
Hence the new cluster variable is [ M c ∗ i ] [M_{\mathfrak c_*}^{i}] [ M c ∗ i ] . The box move also gives B F ′ = μ c ( B F ) B^{\mathcal F'}=\mu_{\mathfrak c}(B^{\mathcal F}) B F ′ = μ c ( B F ) , proving the asserted seed identity. □ \square □
Theorem 3.9 ([33 Theorem 5.20] ). Let C \mathcal C C be an admissible chain of i \mathbf i i -boxes associated with i i i , and set F : = F ( C ) \mathcal F:=\mathcal F(\mathcal C) F := F ( C ) . Then
( [ M c i ] c ∈ F , B F ) \left([M_{\mathfrak c}^{i}]_{\mathfrak c\in\mathcal F},B^{\mathcal F}\right) ( [ M c i ] c ∈ F , B F ) is a seed of K ( C w ) K(\mathcal C_w) K ( C w ) , identified under Φ w \Phi_w Φ w with the box seed Σ i ( F ) \Sigma^i(\mathcal F) Σ i ( F ) . Moreover,
Φ w ( [ M c i ] ) = x c ( c ∈ F ) , L F = − ( Λ ( M c i , M c ′ i ) ) c , c ′ ∈ F . \Phi_w([M_{\mathfrak c}^{i}])=x_{\mathfrak c}\quad(\mathfrak c\in\mathcal F),\qquad L^{\mathcal F}=-\left(\Lambda(M_{\mathfrak c}^{i},M_{\mathfrak c'}^{i})\right)_{\mathfrak c,\mathfrak c'\in\mathcal F}. Φ w ([ M c i ]) = x c ( c ∈ F ) , L F = − ( Λ ( M c i , M c ′ i ) ) c , c ′ ∈ F . Proof. The seed identification holds for the constant R \mathcal R R -chain by (28) . Every admissible chain associated with i i i is connected to C R i \mathcal C_{\mathcal R}^{i} C R i by box moves. Permutation-type moves only relabel the seed, and mutation-type moves are treated by Proposition 3.8. Induction along these moves proves the seed identification. The identity for L F L^{\mathcal F} L F follows from [38 Propositions 5.7 and 5.9] and our sign convention for λ i \lambda^i λ i .
Corollary 3.10. For every i \mathbf i i -box c \mathfrak c c associated with i i i , the class [ M c i ] [M_{\mathfrak c}^{i}] [ M c i ] is a cluster variable of K ( C w ) K(\mathcal C_w) K ( C w ) . More precisely, there exists an admissible chain C \mathcal C C such that
c ∈ F ( C ) and Φ w ( [ M c i ] ) = x c \mathfrak c\in\mathcal F(\mathcal C)\qquad\text{and}\qquad\Phi_w([M_{\mathfrak c}^{i}])=x_{\mathfrak c} c ∈ F ( C ) and Φ w ([ M c i ]) = x c in the corresponding box seed.
Proof. Every i \mathbf i i -box associated with i i i is contained in the underlying maximal commuting family of an admissible chain. The assertion follows from Theorem 3.9. □ \square □
Quasi-Laurent families and the categorical F F F -invariant
Following [38 Definition 3.2] , an independent commuting family M \mathbf M M of affreal simple objects of a full monoidal subcategory C ⊂ R -gmod \mathcal C\subset R\text{-gmod} C ⊂ R -gmod is called a quasi-Laurent family in C \mathcal C C if, for every simple object X ∈ C X\in\mathcal C X ∈ C commuting with all M k M_k M k , there exist a , b ∈ Z ≥ 0 K \mathbf a,\mathbf b\in\mathbb Z_{\geq0}^{K} a , b ∈ Z ≥ 0 K such that
X ∘ M ( a ) ≃ M ( b ) X\circ\mathbf M(\mathbf a)\simeq\mathbf M(\mathbf b) X ∘ M ( a ) ≃ M ( b )
up to a grading shift. The family M \mathbf M M is called a Laurent family in C \mathcal C C if every simple object of C \mathcal C C commuting with all M k M_k M k is an M \mathbf M M -monomial. Every Laurent family is, in particular, a quasi-Laurent family.
Proposition 3.11 ([38 Proposition 3.6 and Corollary 3.4] ). Let C \mathcal{C} C be an admissible chain of i i i -boxes associated with i i i , and set F : = F ( C ) \mathcal{F}:=\mathcal{F}(\mathcal{C}) F := F ( C ) and M F i = { M c i ∣ c ∈ F } \mathsf{M}^{i}_{\mathcal{F}}=\{\mathsf{M}^{i}_{c}\mid c\in\mathcal{F}\} M F i = { M c i ∣ c ∈ F } . Then M F i \mathsf{M}^{i}_{\mathcal{F}} M F i is a quasi-Laurent family in C w \mathcal{C}_{w} C w , and
Φ w ( [ M c i ] ) = x c ( c ∈ F ) , and L F = − Λ M F i . \Phi_{w}([\mathsf{M}^{i}_{c}])=x_{c}\quad(c\in\mathcal{F}),\qquad\text{and}\qquad L^{\mathcal{F}}=-\Lambda_{\mathsf{M}^{i}_{\mathcal{F}}}. Φ w ([ M c i ]) = x c ( c ∈ F ) , and L F = − Λ M F i . In particular, every simple object X ∈ C w X\in\mathcal{C}_{w} X ∈ C w has a positive Laurent expansion
Φ w ( [ X ] ) = ∑ ν = 1 r c ν ( x F ) v ν , c ν ∈ Z > 0 . (30) \Phi_{w}([X])=\sum_{\nu=1}^{r}c_{\nu}(\mathbf{x}^{\mathcal{F}})^{\mathbf{v}_{\nu}},\qquad c_{\nu}\in\mathbb{Z}_{>0}. \tag*{(30)} Φ w ([ X ]) = ν = 1 ∑ r c ν ( x F ) v ν , c ν ∈ Z > 0 . (30)
Proof. The identification Φ w ( [ M c i ] ) = x c \Phi_{w}([\mathsf{M}^{i}_{c}])=x_{c} Φ w ([ M c i ]) = x c follows from Theorem 3.9 , and L F = − Λ M F i L^{\mathcal{F}}=-\Lambda_{\mathsf{M}^{i}_{\mathcal{F}}} L F = − Λ M F i follows from [38 Propositions 5.7 and 5.9] . The quasi-Laurent property and the positive Laurent expansion then follow from [38 Proposition 3.6 and Corollary 3.4] , using the quantum seed associated with the box family in [33 Section 5.1.4 and Theorem 5.20] . □
Let C \mathcal{C} C be a full monoidal subcategory of R R R -gmod stable under subquotients, extensions, and grading shifts. Let K q ( C ) K_{q}(\mathcal{C}) K q ( C ) denote its graded Grothendieck ring. For a quasi-Laurent family M = { M k ∣ k ∈ K } \mathsf{M}=\{\mathsf{M}_{k}\mid k\in K\} M = { M k ∣ k ∈ K } in C \mathcal{C} C , let T M \mathcal{T}_{\mathsf{M}} T M be the quantum torus over Z [ q ± 1 / 2 ] \mathbb{Z}[q^{\pm1/2}] Z [ q ± 1/2 ] generated by the classes [ M k ] [\mathsf{M}_{k}] [ M k ] and their inverses. We denote by
\phi_{\mathsf{M}}:\mathbb{Z}[q^{\pm1/2}]_{\mathbb{Z}[q^{\pm1}]}K_{q}(\mathcal{C})\lhook\joinrel\longrightarrow\mathcal{T}_{\mathsf{M}}
the injective algebra homomorphism given by [41 Lemma 2.12] . For a ∈ Z ⊕ K \mathbf{a}\in\mathbb{Z}^{\oplus K} a ∈ Z ⊕ K , the notation [ M ( a ) ] [\mathsf{M}(\mathbf{a})] [ M ( a )] denotes the corresponding Laurent monomial in T M ; w h e n \mathcal{T}_{\mathsf{M}}; when T M ; w h e n \mathbf{a}\in\mathbb{Z}_{\geq0}^{\oplus K}$, it is represented by the class of the corresponding convolution module.
Theorem 3.12 ([41 Theorem 2.13] ). With C \mathcal{C} C and M \mathsf{M} M as above, let X , L ∈ C X,L\in\mathcal{C} X , L ∈ C be simple modules such that L L L is affreal, and suppose that the Laurent expansion of [ X ] [X] [ X ] with respect to M \mathsf{M} M is
ϕ M ( [ X ] ) = ∑ ν q c ν [ M ( a ν ) ] , c ν ∈ Z , a ν ∈ Z ⊕ K . \phi_{\mathsf{M}}([X])=\sum_{\nu}q^{c_{\nu}}[\mathsf{M}(\mathbf{a}_{\nu})],\qquad c_{\nu}\in\mathbb{Z},\quad\mathbf{a}_{\nu}\in\mathbb{Z}^{\oplus K}. ϕ M ([ X ]) = ν ∑ q c ν [ M ( a ν )] , c ν ∈ Z , a ν ∈ Z ⊕ K . Then we have
Λ ( L , X ) = max ν Λ ( L , M ( a ν ) ) and Λ ( X , L ) = max ν Λ ( M ( a ν ) , L ) . \Lambda(L,X)=\max_{\nu}\Lambda(L,\mathsf{M}(\mathbf{a}_{\nu}))\qquad\text{and}\qquad\Lambda(X,L)=\max_{\nu}\Lambda(\mathsf{M}(\mathbf{a}_{\nu}),L). Λ ( L , X ) = ν max Λ ( L , M ( a ν )) and Λ ( X , L ) = ν max Λ ( M ( a ν ) , L ) .
Theorem 3.13. Let X X X be an affreal simple object of C w \mathcal{C}_{w} C w such that z X : = Φ w ( [ X ] ) z_{X}:=\Phi_{w}([X]) z X := Φ w ([ X ]) is good and bipointed. For a ∈ Z ≥ 0 ⊕ F \mathbf{a}\in\mathbb{Z}_{\geq0}^{\oplus\mathcal{F}} a ∈ Z ≥ 0 ⊕ F , set
N : = M F i ( a ) . N:=\mathsf{M}^{i}_{\mathcal{F}}(\mathbf{a}). N := M F i ( a ) . Then
Λ ( N , X ) = ⟨ z X , Φ w ( [ N ] ) ⟩ t r o p and Λ ( X , N ) = ⟨ Φ w ( [ N ] ) , z X ⟩ t r o p , \Lambda(N,X)=\langle z_{X},\Phi_{w}([N])\rangle_{\mathrm{trop}}\qquad\text{and}\qquad\Lambda(X,N)=\langle\Phi_{w}([N]),z_{X}\rangle_{\mathrm{trop}}, Λ ( N , X ) = ⟨ z X , Φ w ([ N ]) ⟩ trop and Λ ( X , N ) = ⟨ Φ w ([ N ]) , z X ⟩ trop , and consequently
2 d ( X , N ) = ( z X ∣ Φ w ( [ N ] ) ) F ∈ 2 Z ≥ 0 . 2d(X,N)=(z_{X}\mid\Phi_{w}([N]))_{\mathcal{F}}\in2\mathbb{Z}_{\geq0}. 2 d ( X , N ) = ( z X ∣ Φ w ([ N ]) ) F ∈ 2 Z ≥ 0 . Proof. Choose a graded positive Laurent expansion supplied by [38 Corollary 3.4] . After specializing q = 1 q=1 q = 1 and collecting equal Laurent monomials, it gives (30) . Since z X z_{X} z X is good and bipointed, its exponents have the form
v ν = g X ; F + B F w ν , w ν ∈ Z ≥ 0 F e x , \mathbf{v}_{\nu}=\mathbf{g}_{X;\mathcal{F}}+B^{\mathcal{F}}\mathbf{w}_{\nu},\qquad\mathbf{w}_{\nu}\in\mathbb{Z}_{\geq0}^{\mathcal{F}_{\mathrm{ex}}}, v ν = g X ; F + B F w ν , w ν ∈ Z ≥ 0 F ex ,
where both w ν = 0 \mathbf{w}_{\nu}=0 w ν = 0 and w ν = f X ; F \mathbf{w}_{\nu}=\mathbf{f}_{X;\mathcal{F}} w ν = f X ; F occur, and
0 ≤ ( w ν ) c ≤ f c ; X ; F ( c ∈ F e x ) . 0\leq(\mathbf{w}_{\nu})_{c}\leq f_{c;X;\mathcal{F}}\qquad(c\in\mathcal{F}_{\mathrm{ex}}). 0 ≤ ( w ν ) c ≤ f c ; X ; F ( c ∈ F ex ) .
For each c ∈ F c\in\mathcal{F} c ∈ F with a c > 0 a_c>0 a c > 0 , the affreal simple module M c i M_c^i M c i is a convolution factor of N N N . Apply Theorem 3.12 with L = M c i L=M_c^i L = M c i , and then sum with multiplicity a c a_c a c . Since N = M F i ( a ) N=M_{\mathcal{F}}^i(\mathbf{a}) N = M F i ( a ) is simple, [41 Proposition 2.6] gives
Λ ( N , X ) = ∑ c ∈ F a c Λ ( M c i , X ) and Λ ( X , N ) = ∑ c ∈ F a c Λ ( X , M c i ) . (31) \Lambda(N,X)=\sum_{c\in\mathcal{F}}a_c\Lambda(M_c^i,X)\quad\text{and}\quad\Lambda(X,N)=\sum_{c\in\mathcal{F}}a_c\Lambda(X,M_c^i). \tag*{(31)} Λ ( N , X ) = c ∈ F ∑ a c Λ ( M c i , X ) and Λ ( X , N ) = c ∈ F ∑ a c Λ ( X , M c i ) . (31)
Set a e x : = ( a c ) c ∈ F e x \mathbf{a}_{\mathrm{ex}}:=(a_c)_{c\in\mathcal{F}_{\mathrm{ex}}} a ex := ( a c ) c ∈ F ex . The compatibility relation
( B F ) T L F = ( 2 D F ∣ 0 ) (B^{\mathcal{F}})^{\mathsf{T}}L^{\mathcal{F}}=(2D^{\mathcal{F}}\mid0) ( B F ) T L F = ( 2 D F ∣ 0 )
gives
v ν T L F a = g X ; F T L F a + w ν T ( 2 D F ) a e x . \mathbf{v}_{\nu}^{\mathsf{T}}L^{\mathcal{F}}\mathbf{a}
=\mathbf{g}_{X;\mathcal{F}}^{\mathsf{T}}L^{\mathcal{F}}\mathbf{a}
+\mathbf{w}_{\nu}^{\mathsf{T}}(2D^{\mathcal{F}})\mathbf{a}_{\mathrm{ex}}. v ν T L F a = g X ; F T L F a + w ν T ( 2 D F ) a ex .
Thus the term w ν = f X ; F \mathbf{w}_{\nu}=\mathbf{f}_{X;\mathcal{F}} w ν = f X ; F is a common maximizer for all these convolution factors and their multiplicities. Theorem 3.12 and (31) give
Λ ( N , X ) = g X ; F T L F a + F X ; F [ 2 D F a e x ] = ⟨ z X , Φ w ( [ N ] ) ⟩ t r o p . (32) \Lambda(N,X)=\mathbf{g}_{X;\mathcal{F}}^{\mathsf{T}}L^{\mathcal{F}}\mathbf{a}
+F_{X;\mathcal{F}}[2D^{\mathcal{F}}\mathbf{a}_{\mathrm{ex}}]
=\langle\mathbf{z}_X,\Phi_{\mathbf{w}}([N])\rangle_{\mathrm{trop}}. \tag*{(32)} Λ ( N , X ) = g X ; F T L F a + F X ; F [ 2 D F a ex ] = ⟨ z X , Φ w ([ N ]) ⟩ trop . (32)
On the other hand, skew-symmetry and compatibility give
L F B F = − ( 2 D F 0 ) , L^{\mathcal{F}}B^{\mathcal{F}}
=-\begin{pmatrix}2D^{\mathcal{F}}\\0\end{pmatrix}, L F B F = − ( 2 D F 0 ) ,
and hence
a T L F v ν = a T L F g X ; F − a e x T ( 2 D F ) w ν . \mathbf{a}^{\mathsf{T}}L^{\mathcal{F}}\mathbf{v}_{\nu}
=\mathbf{a}^{\mathsf{T}}L^{\mathcal{F}}\mathbf{g}_{X;\mathcal{F}}
-\mathbf{a}_{\mathrm{ex}}^{\mathsf{T}}(2D^{\mathcal{F}})\mathbf{w}_{\nu}. a T L F v ν = a T L F g X ; F − a ex T ( 2 D F ) w ν .
Here w ν = 0 \mathbf{w}_{\nu}=0 w ν = 0 is a common maximizer, and therefore
Λ ( X , N ) = a T L F g X ; F = ⟨ Φ w ( [ N ] ) , z X ⟩ t r o p . (33) \Lambda(X,N)=\mathbf{a}^{\mathsf{T}}L^{\mathcal{F}}\mathbf{g}_{X;\mathcal{F}}
=\langle\Phi_{\mathbf{w}}([N]),\mathbf{z}_X\rangle_{\mathrm{trop}}. \tag*{(33)} Λ ( X , N ) = a T L F g X ; F = ⟨ Φ w ([ N ]) , z X ⟩ trop . (33)
The last assertion follows by adding the two identities and applying Theorem 1.3. □
By Proposition 1.1 and universal positivity, the good and bipointed hypothesis is automatic when z X \mathbf{z}_X z X is a cluster variable or, more generally, a cluster monomial.
Corollary 3.14 (cf. [2 Proposition 4.13] ). Let c ∈ F e x c\in\mathcal{F}_{\mathrm{ex}} c ∈ F ex , and let X X X be an affreal simple object of C w \mathcal{C}_{\mathbf{w}} C w such that Φ w ( [ X ] ) \Phi_{\mathbf{w}}([X]) Φ w ([ X ]) is good and bipointed. Then
d ( M c i , X ) = d i c f c ; Φ w ( [ X ] ) ; F . \mathfrak{d}(M_c^i,X)=d_{i_c}f_{c;\Phi_{\mathbf{w}}([X]);\mathcal{F}}. d ( M c i , X ) = d i c f c ; Φ w ([ X ]) ; F .
Folding formulas for Lambda- and d-invariants
We now combine the categorical interpretation of the F F F -invariant with the folding of F F F -polynomials.
Let C \mathcal{C} C be an admissible chain of i \mathbf{i} i -boxes associated with i \mathbf{i} i , and let C ^ \widehat{\mathcal{C}} C be its canonical lift. Set
F : = F ( C ) and F ^ : = F ( C ^ ) . \mathcal{F}:=\mathcal{F}(\mathcal{C})\qquad\text{and}\qquad
\widehat{\mathcal{F}}:=\mathcal{F}(\widehat{\mathcal{C}}). F := F ( C ) and F := F ( C ) .
For c ∈ F c\in\mathcal{F} c ∈ F , recall that Fib C ( c ) ⊂ F ^ \operatorname{Fib}_{\mathcal{C}}(c)\subset\widehat{\mathcal{F}} Fib C ( c ) ⊂ F denotes its box fiber.
Set w ^ : = s ^ i 1 ⋯ s ^ i l ∈ W g ^ \widehat{w}:=\widehat{s}_{i_1}\cdots\widehat{s}_{i_l}\in W_{\widehat{\mathfrak{g}}} w := s i 1 ⋯ s i l ∈ W g . Let C w ^ \mathcal{C}_{\widehat{w}} C w be the quiver Hecke category associated with the lifted reduced sequence i ^ \widehat{\mathbf{i}} i . For c ^ ∈ F ^ \widehat{c}\in\widehat{\mathcal{F}} c ∈ F , write
M ^ c ^ : = M i ^ [ c ^ ] . \widehat{M}_{\widehat{c}}:=M^{\widehat{\mathbf{i}}}[\widehat{c}]. M c := M i [ c ] .
Let Φ ^ w ^ : K ( C w ^ ) → ∼ A ( w ^ ) \widehat{\Phi}_{\widehat{w}}:K(\mathcal{C}_{\widehat{w}})\xrightarrow{\sim}A(\widehat{w}) Φ w : K ( C w ) ∼ A ( w ) denote the upstairs counterpart of Φ w \Phi_w Φ w . The lifted box family categorically realizes Σ ^ i ^ ( F ^ ) \widehat{\Sigma}_{\widehat{\mathbf{i}}}(\widehat{\mathcal{F}}) Σ i ( F ) , and hence
Φ ^ w ^ ( [ M ^ c ^ ] ) = x ^ c ^ . \widehat{\Phi}_{\widehat{w}}([\widehat{M}_{\widehat{c}}])=\widehat{x}_{\widehat{c}}. Φ w ([ M c ]) = x c .
For c ∈ F c \in\mathcal{F} c ∈ F , define the convolution product
M ^ Fib C ^ ( c ) : = ∘ c ^ ∈ Fib C ^ ( c ) M ^ c ^ . \widehat{M}_{\operatorname{Fib}_{\widehat{\mathcal{C}}}(c)} := \mathop{\circ}_{\widehat{c}\in\operatorname{Fib}_{\widehat{\mathcal{C}}}(c)} \widehat{M}_{\widehat{c}}. M Fib C ( c ) := ∘ c ∈ Fib C ( c ) M c .
The modules occurring in this convolution belong to a common lifted box seed and hence pairwise strongly commute. Thus
Φ ^ w ^ ( [ M ^ Fib C ^ ( c ) ] ) = x ^ Fib C ^ ( c ) : = ∏ c ^ ∈ Fib C ^ ( c ) x ^ c ^ . \widehat{\Phi}_{\widehat{w}}\left([\widehat{M}_{\operatorname{Fib}_{\widehat{\mathcal{C}}}(c)}]\right)=\widehat{x}_{\operatorname{Fib}_{\widehat{\mathcal{C}}}(c)}:=\prod_{\widehat{c}\in\operatorname{Fib}_{\widehat{\mathcal{C}}}(c)}\widehat{x}_{\widehat{c}}. Φ w ( [ M Fib C ( c ) ] ) = x Fib C ( c ) := c ∈ Fib C ( c ) ∏ x c .
Theorem 3.15. Keep the preceding notation. Let c ∈ F c \in\mathcal{F} c ∈ F , and let N ∈ C w N \in\mathcal{C}_{w} N ∈ C w be simple such that z N = Φ w ( [ N ] ) z_{N}=\Phi_{w}([N]) z N = Φ w ([ N ]) is a cluster variable. Suppose that z N z_{N} z N and z ^ N \widehat{z}_{N} z N are corresponding cluster variables in the box seeds of (19) , with ϕ C ( z ^ N ) = z N \phi_{\mathcal{C}}(\widehat{z}_{N})=z_{N} ϕ C ( z N ) = z N . Let N ^ \widehat{N} N be the simple object of C ^ w ^ \widehat{\mathcal{C}}_{\widehat{w}} C w with Φ ^ w ^ ( [ N ^ ] ) = z ^ N \widehat{\Phi}_{\widehat{w}}([\widehat{N}])=\widehat{z}_{N} Φ w ([ N ]) = z N . Then
Λ ( M c i , N ) = d i c ∑ c ^ ∈ Fib C ^ ( c ) Λ ( M ^ c ^ , N ^ ) , (34) \Lambda(M^{i}_{c},N)=d_{i_{c}}\sum_{\widehat{c}\in\operatorname{Fib}_{\widehat{\mathcal{C}}}(c)}\Lambda(\widehat{M}_{\widehat{c}},\widehat{N}), \tag*{(34)} Λ ( M c i , N ) = d i c c ∈ Fib C ( c ) ∑ Λ ( M c , N ) , (34) Λ ( N , M c i ) = d i c ∑ c ^ ∈ Fib C ^ ( c ) Λ ( N ^ , M ^ c ^ ) . (35) \Lambda(N,M^{i}_{c})=d_{i_{c}}\sum_{\widehat{c}\in\operatorname{Fib}_{\widehat{\mathcal{C}}}(c)}\Lambda(\widehat{N},\widehat{M}_{\widehat{c}}). \tag*{(35)} Λ ( N , M c i ) = d i c c ∈ Fib C ( c ) ∑ Λ ( N , M c ) . (35)
Proof. The cluster variables in the statement are good and bipointed, and their categorical realizations are affreal. We may therefore use Theorem 3.13 in both box seeds. With the order convention in that theorem, we have
Λ ( M c i , N ) = ⟨ z N , x c ⟩ t r o p and Λ ( N , M c i ) = ⟨ x c , z N ⟩ t r o p , \Lambda(M^{i}_{c},N)=\langle z_{N},x_{c}\rangle_{\mathrm{trop}}\quad\text{and}\quad\Lambda(N,M^{i}_{c})=\langle x_{c},z_{N}\rangle_{\mathrm{trop}}, Λ ( M c i , N ) = ⟨ z N , x c ⟩ trop and Λ ( N , M c i ) = ⟨ x c , z N ⟩ trop ,
and the analogous identities upstairs. Applying respectively (21) and (20) from Theorem 2.6 therefore gives
Λ ( M c i , N ) = d i c Λ ( M ^ Fib C ^ ( c ) , N ^ ) and Λ ( N , M c i ) = d i c Λ ( N ^ , M ^ Fib C ^ ( c ) ) . \Lambda(M^{i}_{c},N)=d_{i_{c}}\Lambda(\widehat{M}_{\operatorname{Fib}_{\widehat{\mathcal{C}}}(c)},\widehat{N})\quad\text{and}\quad\Lambda(N,M^{i}_{c})=d_{i_{c}}\Lambda(\widehat{N},\widehat{M}_{\operatorname{Fib}_{\widehat{\mathcal{C}}}(c)}). Λ ( M c i , N ) = d i c Λ ( M Fib C ( c ) , N ) and Λ ( N , M c i ) = d i c Λ ( N , M Fib C ( c ) ) .
Finally, applying [41 ] Proposition 2.6 to M ^ Fib C ^ ( c ) = ∘ c ^ ∈ Fib C ^ ( c ) M ^ c ^ \widehat{M}_{\operatorname{Fib}_{\widehat{\mathcal{C}}}(c)}=\mathop{\circ}_{\widehat{c}\in\operatorname{Fib}_{\widehat{\mathcal{C}}}(c)}\widehat{M}_{\widehat{c}} M Fib C ( c ) = ∘ c ∈ Fib C ( c ) M c in both orders gives (3.12). □
Theorem 3.16. Let c ∈ F c \in\mathcal{F} c ∈ F , and let N N N be a simple object of C w \mathcal{C}_{w} C w such that
z N : = Φ w ( [ N ] ) z_{N}:=\Phi_{w}([N]) z N := Φ w ([ N ]) is a cluster variable. Suppose that z N z_{N} z N and z ^ N \widehat{z}_{N} z N are corresponding cluster variables in the box seeds of (19) and satisfy
ϕ C ( z ^ N ) = z N , \phi_{\mathcal{C}}(\widehat{z}_{N})=z_{N}, ϕ C ( z N ) = z N , and let N ^ \widehat{N} N be the corresponding simple object of C ^ w ^ \widehat{\mathcal{C}}_{\widehat{w}} C w , determined by
Φ ^ w ^ ( [ N ^ ] ) = z ^ N . \widehat{\Phi}_{\widehat{w}}([\widehat{N}])=\widehat{z}_{N}. Φ w ([ N ]) = z N . Then we have
∂ ( M c i , N ) = d i c ∑ c ^ ∈ Fib C ^ ( c ) ∂ ( M ^ c ^ , N ^ ) = d i c ∂ ( M ^ Fib C ^ ( c ) , N ^ ) . (36) \partial(M^{i}_{c},N)=d_{i_{c}}\sum_{\widehat{c}\in\operatorname{Fib}_{\widehat{\mathcal{C}}}(c)}\partial(\widehat{M}_{\widehat{c}},\widehat{N})=d_{i_{c}}\partial(\widehat{M}_{\operatorname{Fib}_{\widehat{\mathcal{C}}}(c)},\widehat{N}). \tag*{(36)} ∂ ( M c i , N ) = d i c c ∈ Fib C ( c ) ∑ ∂ ( M c , N ) = d i c ∂ ( M Fib C ( c ) , N ) . (36)
Proof. Add equations (34) and (35) , and divide by 2. The additivity already used in their proof identifies the symmetrized upstairs sum with ∂ ( M ^ Fib C ^ ( c ) , N ^ ) \partial(\widehat{M}_{\operatorname{Fib}_{\widehat{\mathcal{C}}}(c)},\widehat{N}) ∂ ( M Fib C ( c ) , N ) .
Corollary 3.17. Let i , j ∈ Red ( w ) i,j \in\operatorname{Red}(w) i , j ∈ Red ( w ) , and let c c c and c ′ c' c ′ be i i i -boxes associated with i i i and j j j , respectively. Choose admissible chains C \mathcal{C} C and D \mathcal{D} D whose underlying families contain c c c and c ′ c' c ′ , respectively, and choose c ^ ′ ∈ Fib D ( c ′ ) \widehat{c}' \in\operatorname{Fib}_{\mathcal{D}}(c') c ′ ∈ Fib D ( c ′ ) . Then we have formulas for both orders of the Λ \Lambda Λ -invariant:
Λ ( M i [ c ] , M j [ c ′ ] ) = d i c ∑ c ^ ∈ Fib C ( c ) Λ ( M i [ c ^ ] , M j [ c ^ ′ ] ) . (37) \Lambda(M^i[c],M^j[c'])=d_{i_c}\sum_{\widehat{c}\in\operatorname{Fib}_{\mathcal{C}}(c)}\Lambda(M^i[\widehat{c}],M^j[\widehat{c}']). \tag*{(37)} Λ ( M i [ c ] , M j [ c ′ ]) = d i c c ∈ Fib C ( c ) ∑ Λ ( M i [ c ] , M j [ c ′ ]) . (37) Λ ( M j [ c ′ ] , M i [ c ] ) = d i c ∑ c ^ ∈ Fib C ( c ) Λ ( M j [ c ^ ′ ] , M i [ c ^ ] ) . (38) \Lambda(M^j[c'],M^i[c])=d_{i_c}\sum_{\widehat{c}\in\operatorname{Fib}_{\mathcal{C}}(c)}\Lambda(M^j[\widehat{c}'],M^i[\widehat{c}]). \tag*{(38)} Λ ( M j [ c ′ ] , M i [ c ]) = d i c c ∈ Fib C ( c ) ∑ Λ ( M j [ c ′ ] , M i [ c ]) . (38) Their symmetrization gives
d ( M i [ c ] , M j [ c ′ ] ) = d i c ∑ c ^ ∈ Fib C ( c ) d ( M i [ c ^ ] , M j [ c ^ ′ ] ) . (39) \mathsf{d}(M^i[c],M^j[c'])=d_{i_c}\sum_{\widehat{c}\in\operatorname{Fib}_{\mathcal{C}}(c)}\mathsf{d}(M^i[\widehat{c}],M^j[\widehat{c}']). \tag*{(39)} d ( M i [ c ] , M j [ c ′ ]) = d i c c ∈ Fib C ( c ) ∑ d ( M i [ c ] , M j [ c ′ ]) . (39) The right-hand side is independent of the choice of the lifted box c ^ ′ \widehat{c}' c ′ .
Proof. By Corollary 3.10, z c ′ : = Φ w ( [ M j [ c ′ ] ] ) z_{c'}:=\Phi_w([M^j[c']]) z c ′ := Φ w ([ M j [ c ′ ]]) is a cluster variable. For every c ^ ′ ∈ Fib D ( c ′ ) \widehat{c}'\in\operatorname{Fib}_{\mathcal{D}}(c') c ′ ∈ Fib D ( c ′ ) , set
z ^ c ^ ′ : = Φ w ^ ( [ M j [ c ′ ] ] ) . \widehat{z}_{\widehat{c}'}:=\Phi_{\widehat{w}}([M^j[c']]). z c ′ := Φ w ([ M j [ c ′ ]]) . Theorem 2.5 gives
ϕ C b ( z ^ c ^ ′ ) = z c ′ ( c ^ ′ ∈ Fib D ( c ′ ) ) . \phi_{\mathcal{C}}^{\mathrm{b}}(\widehat{z}_{\widehat{c}'})=z_{c'}\qquad(\widehat{c}'\in\operatorname{Fib}_{\mathcal{D}}(c')). ϕ C b ( z c ′ ) = z c ′ ( c ′ ∈ Fib D ( c ′ )) . Hence Theorem 3.15 , applied with N = M j [ c ′ ] N=M^j[c'] N = M j [ c ′ ] and N ^ = M i [ c ^ ] \widehat{N}=M^i[\widehat{c}] N = M i [ c ] , gives (37) and (38) . Adding these identities and dividing by 2 gives (39) . Since the left-hand side is independent of c ^ ′ \widehat{c}' c ′ , the right-hand side has the same value for every c ^ ′ ∈ Fib D ( c ′ ) \widehat{c}'\in\operatorname{Fib}_{\mathcal{D}}(c') c ′ ∈ Fib D ( c ′ ) . □ \square □
In the notation of Example 2.7, the categorical counterpart of the two box variables is
x [ 2 , 5 ] ⟷ M i [ 2 , 5 ] , x [ 7 , 9 ] ⟷ M i [ 7 , 9 ] , x_{[2,5]}\longleftrightarrow M^i[2,5],\qquad x_{[7,9]}\longleftrightarrow M^i[7,9], x [ 2 , 5 ] ⟷ M i [ 2 , 5 ] , x [ 7 , 9 ] ⟷ M i [ 7 , 9 ] ,
and the relevant lifted variables correspond to M ı ^ [ 3 , 8 ] M^{\widehat{\imath}}[3,8] M [ 3 , 8 ] , M ı ^ [ 4 , 9 ] M^{\widehat{\imath}}[4,9] M [ 4 , 9 ] , and M ı ^ [ 12 , 15 ] M^{\widehat{\imath}}[12,15] M [ 12 , 15 ] . The A 5 A_5 A 5 involution exchanges x ^ [ 3 , 8 ] \widehat{x}_{[3,8]} x [ 3 , 8 ] and x ^ [ 4 , 9 ] \widehat{x}_{[4,9]} x [ 4 , 9 ] , fixes x ^ [ 12 , 15 ] \widehat{x}_{[12,15]} x [ 12 , 15 ] , and preserves the lifted seed data by [25 Lemma 3.14] . The categorical interpretation of the F F F -invariant therefore gives equal d d d -values for the two upstairs pairs. Since the F F F -invariant for x [ 2 , 5 ] x_{[2,5]} x [ 2 , 5 ] and x [ 7 , 9 ] x_{[7,9]} x [ 7 , 9 ] is 4, Theorem 3.13 gives
d ( M i [ 2 , 5 ] , M i [ 7 , 9 ] ) = d 2 ∑ e ^ ∈ { [ 3 , 8 ] , [ 4 , 9 ] } d ( M ı ^ [ e ^ ] , M ı ^ [ 12 , 15 ] ) = 1 ⋅ ( 1 + 1 ) = 2. \mathsf{d}(M^i[2,5],M^i[7,9])=d_2\sum_{\widehat{e}\in\{[3,8],[4,9]\}}\mathsf{d}(M^{\widehat{\imath}}[\widehat{e}],M^{\widehat{\imath}}[12,15])=1\cdot(1+1)=2. d ( M i [ 2 , 5 ] , M i [ 7 , 9 ]) = d 2 e ∈ {[ 3 , 8 ] , [ 4 , 9 ]} ∑ d ( M [ e ] , M [ 12 , 15 ]) = 1 ⋅ ( 1 + 1 ) = 2.
Unlike the cuspidal modules in (23) , both determinantial modules in this example arise from boxes containing two occurrences of their respective colors.
KR-modules over quiver Hecke algebras and symmetrized inverse t-Cartan matrices
In this final section, we apply the folding formulas established above to Kirillov–Reshetikhin modules over quiver Hecke algebras . We first introduce these modules as determinantial modules attached to i i i -boxes arising from adapted reduced sequences. We then relate their d d d -invariants to the symmetrized inverse t t t -Cartan matrix. In finite type, this relation, together with folding, yields explicit formulas in all finite types and closed formulas in the classical finite types. Together with the formula in Proposition 3.3 , these d d d -formulas also permit the computation of both orders of the Λ \Lambda Λ -invariant for boxes in the same Q Q Q -adapted reduced sequence.
We begin by describing adapted reduced sequences in terms of height functions and repetition quivers. Although the usual Auslander–Reiten interpretation is restricted to finite type, the construction below makes sense for every symmetrizable Cartan datum whose Dynkin diagram admits a height function.
Height functions and Q Q Q -adapted sequences
We use coordinates on repetition quivers to define the KR-module parameters and show that these parameters are preserved under folding.
Keep the symmetrizable Cartan datum
C = ( c i , j ) i , j ∈ I , D = diag ( d i ∣ i ∈ I ) , C=(c_{i,j})_{i,j\in I}, \qquad D=\operatorname{diag}(d_i\mid i\in I), C = ( c i , j ) i , j ∈ I , D = diag ( d i ∣ i ∈ I ) ,
and let △ \triangle △ be the Dynkin diagram of g \mathfrak{g} g with vertex set I I I . Let { α i ∣ i ∈ I } \{\alpha_i\mid i\in I\} { α i ∣ i ∈ I } be the set of simple roots, normalized by ( α i , α i ) g = 2 d i (\alpha_i,\alpha_i)_{\mathfrak{g}}=2d_i ( α i , α i ) g = 2 d i .
For i , j ∈ I i,j\in I i , j ∈ I , we write i ∼ j i\sim j i ∼ j if c i , j c j , i > 0 c_{i,j}c_{j,i}>0 c i , j c j , i > 0 . Thus i ∼ j i\sim j i ∼ j records adjacency in △ \triangle △ .
A Dynkin quiver is a pair Q = ( △ , ξ ) Q=(\triangle,\xi) Q = ( △ , ξ ) , where a height function
ξ : I ⟶ Z \xi:I\longrightarrow\mathbb{Z} ξ : I ⟶ Z
satisfies ∣ ξ i − ξ j ∣ = 1 |\xi_i-\xi_j|=1 ∣ ξ i − ξ j ∣ = 1 whenever i ∼ j i\sim j i ∼ j .
The height function determines an orientation of the underlying graph of △ \triangle △ by
i ⟶ j ⟺ ξ i = ξ j + 1. i\longrightarrow j \Longleftrightarrow\xi_i=\xi_j+1. i ⟶ j ⟺ ξ i = ξ j + 1.
This quiver orientation should not be confused with the arrows in the usual Dynkin diagram that indicate relative root lengths.
Remark 4.1. Not every Dynkin diagram admits a height function. Indeed, such a function exists if and only if the underlying graph of △ \triangle △ is bipartite. Thus every finite-type Dynkin diagram admits a height function, whereas the affine Dynkin diagram of type A 2 n ( 1 ) A_{2n}^{(1)} A 2 n ( 1 ) , whose underlying graph is an odd cycle, does not.
Throughout this section, we assume that △ \triangle △ is connected and admits a height function ξ \xi ξ , and fix the corresponding Dynkin quiver Q = ( △ , ξ ) Q=(\triangle,\xi) Q = ( △ , ξ ) .
A vertex i ∈ I i\in I i ∈ I is called a sink of Q Q Q if
ξ i = ξ j − 1 for every j ∼ i . \xi_i=\xi_j-1 \qquad\text{for every }j\sim i. ξ i = ξ j − 1 for every j ∼ i .
If i i i is a sink, let s i Q s_iQ s i Q be the Dynkin quiver obtained by reversing all arrows incident with i i i . Its height function is given by
( s i ξ ) j : = ξ j + 2 δ i , j ( j ∈ I ) . (s_i\xi)_j:=\xi_j+2\delta_{i,j}\qquad(j\in I). ( s i ξ ) j := ξ j + 2 δ i , j ( j ∈ I ) .
A sequence i = ( i 1 , … , i l ) \mathbf{i}=(i_1,\ldots,i_l) i = ( i 1 , … , i l ) is called Q Q Q -adapted if
i k is a sink of s i k − 1 ⋯ s i 1 Q ( 1 ≤ k ≤ l ) . i_k\text{ is a sink of }s_{i_{k-1}}\cdots s_{i_1}Q \qquad(1\leq k\leq l). i k is a sink of s i k − 1 ⋯ s i 1 Q ( 1 ≤ k ≤ l ) .
If moreover i ∈ Red ( w ) \mathbf{i}\in\operatorname{Red}(w) i ∈ Red ( w ) , we call it a Q Q Q -adapted reduced sequence of w w w .
Fix a parity function ϵ : I ⟶ { 0 , 1 } \epsilon:I\longrightarrow\{0,1\} ϵ : I ⟶ { 0 , 1 } such that
ϵ i ≠ ϵ j whenever i ∼ j . \epsilon_i\ne\epsilon_j \qquad\text{whenever }i\sim j. ϵ i = ϵ j whenever i ∼ j .
After adding a constant to ξ \xi ξ if necessary, we assume that ξ i ≡ ϵ i ( m o d 2 ) \xi_i\equiv\epsilon_i\pmod{2} ξ i ≡ ϵ i ( mod 2 ) ( i ∈ I ) (i\in I) ( i ∈ I ) .
The repetition quiver associated with △ \triangle △ is the colored quiver △ ~ \widetilde{\triangle} △ with vertex set
△ ~ 0 : = { ( i , p ) ∈ I × Z ∣ p ≡ ϵ i ( m o d 2 ) } (40) \widetilde{\triangle}_0:=\{(i,p)\in I\times\mathbb{Z}\mid p\equiv\epsilon_i\pmod{2}\} \tag*{(40)} △ 0 := {( i , p ) ∈ I × Z ∣ p ≡ ϵ i ( mod 2 )} (40)
and arrow set
Δ ~ 1 : = { ( i , p ) ⟶ ( j , p − 1 ) ∣ i ∼ j } . \widetilde{\Delta}_1 := \{(i,p) \longrightarrow(j,p-1) \mid i \sim j\}. Δ 1 := {( i , p ) ⟶ ( j , p − 1 ) ∣ i ∼ j } .
Here every arrow is drawn as a single arrow. The lacing data of Δ \Delta Δ are retained by the colors of the vertices (see Example 4.2 below).
The height function ξ \xi ξ determines the full subquiver Δ ~ ≥ ξ \widetilde{\Delta}_{\geq\xi} Δ ≥ ξ of Δ ~ \widetilde{\Delta} Δ with vertex set
( Δ ~ ≥ ξ ) 0 : = { ( i , p ) ∈ Δ ~ 0 ∣ p ≥ ξ i } . \left(\widetilde{\Delta}_{\geq\xi}\right)_0 := \{(i,p) \in\widetilde{\Delta}_0 \mid p \geq\xi_i\}. ( Δ ≥ ξ ) 0 := {( i , p ) ∈ Δ 0 ∣ p ≥ ξ i } .
Thus the vertices of color i i i are
( i , ξ i ) , ( i , ξ i + 2 ) , ( i , ξ i + 4 ) , … . (i,\xi_i),\ (i,\xi_i+2),\ (i,\xi_i+4),\ldots. ( i , ξ i ) , ( i , ξ i + 2 ) , ( i , ξ i + 4 ) , … .
The boundary vertices { ( i , ξ i ) ∣ i ∈ I } \{(i,\xi_i) \mid i \in I\} {( i , ξ i ) ∣ i ∈ I } form a copy of the Dynkin quiver Q Q Q .
The same one-sided indexing is used by Fujita and Murakami in their definition of Φ ξ \Phi_\xi Φ ξ for bipartite generalized Cartan matrices [13 Definition 2.13] .
Example 4.2. We give three examples of the one-sided repetition quiver Δ ~ ≥ ξ \widetilde{\Delta}_{\geq\xi} Δ ≥ ξ .
(i) Let Q = ( A 3 , ξ ) Q=(A_3,\xi) Q = ( A 3 , ξ ) with ( ξ 1 , ξ 2 , ξ 3 ) = ( 1 , 2 , 3 ) (\xi_1,\xi_2,\xi_3)=(1,2,3) ( ξ 1 , ξ 2 , ξ 3 ) = ( 1 , 2 , 3 ) . Thus the boundary orientation is 3 → 2 → 1 3 \to2 \to1 3 → 2 → 1 . The one-sided repetition quiver is
Figure 1.
(ii) Let Q = ( C 3 , ξ ) Q=(C_3,\xi) Q = ( C 3 , ξ ) with ( ξ 1 , ξ 2 , ξ 3 ) = ( 1 , 2 , 3 ) (\xi_1,\xi_2,\xi_3)=(1,2,3) ( ξ 1 , ξ 2 , ξ 3 ) = ( 1 , 2 , 3 ) . Again, the boundary orientation is 3 → 2 → 1 3 \to2 \to1 3 → 2 → 1 . The one-sided repetition quiver is
Figure 2.
Here the two vertex styles distinguish the two root lengths. As in the simply-laced case, only one arrow is drawn between adjacent colored vertices.
(iii) Let Q = ( A 3 ( 1 ) , ξ ) Q=(A_3^{(1)},\xi) Q = ( A 3 ( 1 ) , ξ ) with ( ξ 0 , ξ 1 , ξ 2 , ξ 3 ) = ( 1 , 2 , 1 , 2 ) (\xi_0,\xi_1,\xi_2,\xi_3)=(1,2,1,2) ( ξ 0 , ξ 1 , ξ 2 , ξ 3 ) = ( 1 , 2 , 1 , 2 ) . Thus the boundary orientation is
1 ⟶ 0 ⟵ 3 , 1 ⟶ 2 ⟵ 3. 1 \longrightarrow0 \longleftarrow3,\qquad1 \longrightarrow2 \longleftarrow3. 1 ⟶ 0 ⟵ 3 , 1 ⟶ 2 ⟵ 3. The one-sided repetition quiver is represented by the following fundamental strip:
Figure 3.
Here the first and last rows are two copies of the color-0 boundary and are identified. Thus the displayed fundamental strip is naturally regarded as a quiver drawn on a cylinder.
Let Γ \Gamma Γ be a finite full subquiver of △ ~ ≥ ξ \widetilde{\triangle}_{\geq\xi} △ ≥ ξ . A sequence
R = ( v 1 , … , v r ) \mathrm{R}=(v_1,\ldots,v_r) R = ( v 1 , … , v r )
containing every vertex of Γ \Gamma Γ exactly once is called a compatible reading of Γ \Gamma Γ if
v a ⟶ v b ⟹ b < a . v_a \longrightarrow v_b \quad\Longrightarrow\quad b<a. v a ⟶ v b ⟹ b < a .
Thus the target of every arrow is read before its source.
Write v a = ( i a , p a ) v_a=(i_a,p_a) v a = ( i a , p a ) for 1 ≤ a ≤ r 1\leq a\leq r 1 ≤ a ≤ r . For 1 ≤ k ≤ r 1\leq k\leq r 1 ≤ k ≤ r , set
m i ( k ) : = ∣ { 1 ≤ a < k ∣ i a = i } ∣ , i ( R ) : = ( i 1 , … , i r ) . m_i(k):=\left|\{1\leq a<k\mid i_a=i\}\right|,\qquad\mathbf{i}(\mathrm{R}):=(i_1,\ldots,i_r). m i ( k ) := ∣ { 1 ≤ a < k ∣ i a = i } ∣ , i ( R ) := ( i 1 , … , i r ) .
After v 1 , … , v k − 1 v_1,\ldots,v_{k-1} v 1 , … , v k − 1 have been read, set
∂ k : = { ( i , ξ i + 2 m i ( k ) ) ∣ i ∈ I } \partial_k:=\{(i,\xi_i+2m_i(k))\mid i\in I\} ∂ k := {( i , ξ i + 2 m i ( k )) ∣ i ∈ I }
and for a vertex
( i , ξ i + 2 m i ( k ) ) ∈ ∂ k (i,\xi_i+2m_i(k))\in\partial_k ( i , ξ i + 2 m i ( k )) ∈ ∂ k
consider the condition
ξ i + 2 m i ( k ) = ξ j + 2 m j ( k ) − 1 for every j ∼ i . \xi_i+2m_i(k)=\xi_j+2m_j(k)-1\qquad\text{for every }j\sim i. ξ i + 2 m i ( k ) = ξ j + 2 m j ( k ) − 1 for every j ∼ i .
We consider compatible readings R \mathrm{R} R such that, for every 1 ≤ k ≤ r 1\leq k\leq r 1 ≤ k ≤ r ,
( i k , p k ) = ( i k , ξ i k + 2 m i k ( k ) ) ∈ ∂ k (i_k,p_k)=(i_k,\xi_{i_k}+2m_{i_k}(k))\in\partial_k ( i k , p k ) = ( i k , ξ i k + 2 m i k ( k )) ∈ ∂ k
and the equality for adjacent colors holds with i = i k i=i_k i = i k .
Lemma 4.3. Let R = ( ( i 1 , p 1 ) , … , ( i r , p r ) ) \mathrm{R}=((i_1,p_1),\ldots,(i_r,p_r)) R = (( i 1 , p 1 ) , … , ( i r , p r )) be a compatible reading of a finite full subquiver of △ ~ ≥ ξ \widetilde{\triangle}_{\geq\xi} △ ≥ ξ , satisfying the two conditions above. Then
i ( R ) : = ( i 1 , … , i r ) is Q -adapted . \mathbf{i}(\mathrm{R}):=(i_1,\ldots,i_r)\text{ is }Q\text{-adapted}. i ( R ) := ( i 1 , … , i r ) is Q -adapted .
Proof. Reflection at a sink increases its height by 2 2 2 and leaves the other heights unchanged. Thus, just before the k k k -th vertex is read, the height function is
ξ ( k − 1 ) = s i k − 1 ⋯ s i 1 ξ , ξ i ( k − 1 ) = ξ i + 2 m i ( k ) ( i ∈ I ) . \xi^{(k-1)}=s_{i_{k-1}}\cdots s_{i_1}\xi,\qquad\xi_i^{(k-1)}=\xi_i+2m_i(k)\quad(i\in I). ξ ( k − 1 ) = s i k − 1 ⋯ s i 1 ξ , ξ i ( k − 1 ) = ξ i + 2 m i ( k ) ( i ∈ I ) .
Thus i k i_k i k is a sink of s i k − 1 ⋯ s i 1 Q s_{i_{k-1}}\cdots s_{i_1}Q s i k − 1 ⋯ s i 1 Q , so i ( R ) \mathbf{i}(\mathrm{R}) i ( R ) is Q Q Q -adapted. □ \square □
Let
R = ( ( i 1 , p 1 ) , … , ( i l , p l ) ) \mathrm{R}=((i_1,p_1),\ldots,(i_l,p_l)) R = (( i 1 , p 1 ) , … , ( i l , p l ))
be a finite compatible reading as in Lemma 4.3, and set
i : = i ( R ) = ( i 1 , … , i l ) . \mathbf{i}:=\mathbf{i}(\mathrm{R})=(i_1,\ldots,i_l). i := i ( R ) = ( i 1 , … , i l ) .
We denote by Υ Q i \Upsilon_Q^{\mathbf{i}} Υ Q i the full subquiver of △ ~ ≥ ξ \widetilde{\triangle}_{\geq\xi} △ ≥ ξ whose vertex set is
( Υ Q i ) 0 : = { ( i k , p k ) ∣ 1 ≤ k ≤ l } . (\Upsilon_Q^{\mathbf{i}})_0:=\{(i_k,p_k)\mid1\leq k\leq l\}. ( Υ Q i ) 0 := {( i k , p k ) ∣ 1 ≤ k ≤ l } .
We call Υ Q i \Upsilon_Q^{\mathbf{i}} Υ Q i the Q Q Q -adapted quiver associated with i \mathbf{i} i .
The position of the k k k -th vertex is determined by Q Q Q and i \mathbf{i} i . Since Q Q Q is fixed, we write
p k i : = ξ i k + 2 ∣ { 1 ≤ a < k ∣ i a = i k } ∣ ( 1 ≤ k ≤ l ) . p_k^{\mathbf{i}}:=\xi_{i_k}+2\left|\{1\leq a<k\mid i_a=i_k\}\right|\qquad(1\leq k\leq l). p k i := ξ i k + 2 ∣ { 1 ≤ a < k ∣ i a = i k } ∣ ( 1 ≤ k ≤ l ) .
Thus the compatible reading associated with i \mathbf{i} i is
( ( i 1 , p 1 i ) , … , ( i l , p l i ) ) . ((i_1,p_1^i),\ldots,(i_l,p_l^i)). (( i 1 , p 1 i ) , … , ( i l , p l i )) .
We shall also use Dynkin quivers compatible with the folding datum of § 3 . Let
Q ^ = ( Δ ^ , ξ ^ ) \widehat{Q}=(\widehat{\Delta},\widehat{\xi}) Q = ( Δ , ξ )
be a Dynkin quiver for the simply-laced Cartan datum of g ^ \widehat{\mathfrak{g}} g . We call Q ^ \widehat{Q} Q Γ \Gamma Γ -foldable if
ξ ^ g ( i ) = ξ ^ i ( g ∈ Γ , i ∈ I ^ ) . \widehat{\xi}_{g(i)}=\widehat{\xi}_i \qquad(g\in\Gamma,\ i\in\widehat{I}). ξ g ( i ) = ξ i ( g ∈ Γ , i ∈ I ) .
In this case, ξ ^ \widehat{\xi} ξ induces a height function on the quotient Dynkin diagram by
ξ i : = ξ ^ i ( i ∈ I i ) . \xi_i:=\widehat{\xi}_i \qquad(i\in I_i). ξ i := ξ i ( i ∈ I i ) .
We denote the quotient Dynkin quiver by
Q : = Q ^ / Γ = ( Δ , ξ ) . Q:=\widehat{Q}/\Gamma=(\Delta,\xi). Q := Q /Γ = ( Δ , ξ ) .
Lemma 4.4. Let Q ^ \widehat{Q} Q be a Γ \Gamma Γ -foldable simply-laced Dynkin quiver and set Q : = Q ^ / Γ Q:=\widehat{Q}/\Gamma Q := Q /Γ . Then ( i , p ) ↦ ( π ( i ) , p ) (i,p)\mapsto(\pi(i),p) ( i , p ) ↦ ( π ( i ) , p ) defines the natural folding of repetition quivers
Δ ^ ≥ ξ ^ ⟶ Δ ≥ ξ . \widehat{\Delta}_{\geq\widehat{\xi}}\longrightarrow\Delta_{\geq\xi}. Δ ≥ ξ ⟶ Δ ≥ ξ . Moreover, if i = ( i 1 , … , i l ) \mathbf{i}=(i_1,\ldots,i_l) i = ( i 1 , … , i l ) is Q Q Q -adapted, then its blockwise lift i ^ \widehat{\mathbf{i}} i is Q ^ \widehat{Q} Q -adapted ([25 Lemma 8.1] , stated there in the opposite source-adapted convention). If i \mathbf{i} i and i ^ \widehat{\mathbf{i}} i are reduced, the preceding folding restricts to
Υ Q ^ i ^ ⟶ Υ Q i . \Upsilon_{\widehat{Q}}^{\widehat{\mathbf{i}}}\longrightarrow\Upsilon_Q^{\mathbf{i}}. Υ Q i ⟶ Υ Q i . Under the canonical lift of an admissible chain, the obtained repetition-quiver vertex fibers induce the box fibers used in § 3 .
Lemma 4.5. Assume that i \mathbf{i} i is a Q Q Q -adapted reduced sequence. Let C \mathcal{C} C be an admissible chain associated with i \mathbf{i} i , and let C ^ \widehat{\mathcal{C}} C be its canonical lift.
(i) For c = [ x , y ] ∈ F ( C ) \mathfrak{c}=[x,y]\in F(\mathcal{C}) c = [ x , y ] ∈ F ( C ) , set k c : = ∣ { a ∈ [ x , y ] ∣ i a = i c } ∣ k_{\mathfrak{c}}:=\lvert\{a\in[x,y]\mid i_a=i_{\mathfrak{c}}\}\rvert k c := ∣{ a ∈ [ x , y ] ∣ i a = i c }∣ and p c : = p x i + p y i 2 p_{\mathfrak{c}}:=\frac{p_x^i+p_y^i}{2} p c := 2 p x i + p y i .
(ii) For c ^ = [ x ^ , y ^ ] ∈ Fib C ( c ) \widehat{\mathfrak{c}}=[\widehat{x},\widehat{y}]\in\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c}) c = [ x , y ] ∈ Fib C ( c ) , let i c ^ ∈ I ^ i_{\widehat{\mathfrak{c}}}\in\widehat{I} i c ∈ I be its color, so that π ( i c ^ ) = i c \pi(i_{\widehat{\mathfrak{c}}})=i_{\mathfrak{c}} π ( i c ) = i c , and set
k c ^ : = ∣ { a ^ ∈ [ x ^ , y ^ ] ∣ i ^ a ^ = i c ^ } ∣ and p c ^ : = p ^ x ^ i ^ + p ^ y ^ i ^ 2 . k_{\widehat{\mathfrak{c}}}:=\lvert\{\widehat{a}\in[\widehat{x},\widehat{y}]\mid\widehat{i}_{\widehat{a}}=i_{\widehat{\mathfrak{c}}}\}\rvert
\qquad\text{and}\qquad
p_{\widehat{\mathfrak{c}}}:=\frac{\widehat{p}_{\widehat{x}}^{\widehat{i}}+\widehat{p}_{\widehat{y}}^{\widehat{i}}}{2}. k c := ∣{ a ∈ [ x , y ] ∣ i a = i c }∣ and p c := 2 p x i + p y i . Then we have
π ( i c ^ ) = i c , k c ^ = k c , p c ^ = p c for every c ^ ∈ Fib C ( c ) . \pi(i_{\widehat{\mathfrak{c}}})=i_{\mathfrak{c}},\qquad
k_{\widehat{\mathfrak{c}}}=k_{\mathfrak{c}},\qquad
p_{\widehat{\mathfrak{c}}}=p_{\mathfrak{c}}
\qquad\text{for every }\widehat{\mathfrak{c}}\in\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c}). π ( i c ) = i c , k c = k c , p c = p c for every c ∈ Fib C ( c ) . Proof. Let ρ : [ 1 , N l ] ⟶ [ 1 , l ] \rho:[1,N_l]\longrightarrow[1,l] ρ : [ 1 , N l ] ⟶ [ 1 , l ] be defined by ρ ( κ ( a , r ) ) = a \rho(\kappa(a,r))=a ρ ( κ ( a , r )) = a . By the construction of the canonical lift [25 Lemma 3.13] , ρ \rho ρ induces an order-preserving bijection
{ a ^ ∈ [ x ^ , y ^ ] ∣ i ^ a ^ = i c ^ } ⟶ ∼ { a ∈ [ x , y ] ∣ i a = i c } . \{\widehat{a}\in[\widehat{x},\widehat{y}]\mid\widehat{i}_{\widehat{a}}=i_{\widehat{\mathfrak{c}}}\}
\overset{\sim}{\longrightarrow}
\{a\in[x,y]\mid i_a=i_{\mathfrak{c}}\}. { a ∈ [ x , y ] ∣ i a = i c } ⟶ ∼ { a ∈ [ x , y ] ∣ i a = i c } . Hence k c ^ = k c k_{\widehat{\mathfrak{c}}}=k_{\mathfrak{c}} k c = k c , ρ ( x ^ ) = x \rho(\widehat{x})=x ρ ( x ) = x , and ρ ( y ^ ) = y \rho(\widehat{y})=y ρ ( y ) = y . Since the repetition-quiver folding preserves the second coordinate,
p c ^ = p ^ x ^ i ^ + p ^ y ^ i ^ 2 = p x i + p y i 2 = p c . p_{\widehat{\mathfrak{c}}}
=\frac{\widehat{p}_{\widehat{x}}^{\widehat{i}}+\widehat{p}_{\widehat{y}}^{\widehat{i}}}{2}
=\frac{p_x^i+p_y^i}{2}
=p_{\mathfrak{c}}. p c = 2 p x i + p y i = 2 p x i + p y i = p c . The equality π ( i c ^ ) = i c \pi(i_{\widehat{\mathfrak{c}}})=i_{\mathfrak{c}} π ( i c ) = i c holds by construction. □ \square □
Suppose now that i = ( i 1 , … , i l ) ∈ Red ( w ) \mathbf{i}=(i_1,\ldots,i_l)\in\operatorname{Red}(w) i = ( i 1 , … , i l ) ∈ Red ( w ) is Q Q Q -adapted, where w ∈ W g w\in W_{\mathfrak{g}} w ∈ W g . Recall from (12) that
β k i = s i 1 ⋯ s i k − 1 ( α i k ) . \beta_k^{\mathbf{i}}=s_{i_1}\cdots s_{i_{k-1}}(\alpha_{i_k}). β k i = s i 1 ⋯ s i k − 1 ( α i k ) .
Since i \mathbf{i} i is reduced, one has
β k i ∈ Δ + ( 1 ≤ k ≤ l ) \beta_k^{\mathbf{i}}\in\Delta_+\qquad(1\leq k\leq l) β k i ∈ Δ + ( 1 ≤ k ≤ l )
and
{ β k i ∣ 1 ≤ k ≤ l } = Δ + ∩ w ( Δ − ) . \{\beta_k^{\mathbf{i}}\mid1\leq k\leq l\}=\Delta_+\cap w(\Delta_-). { β k i ∣ 1 ≤ k ≤ l } = Δ + ∩ w ( Δ − ) .
We label the vertices of Γ Q i \Gamma_Q^{\mathbf{i}} Γ Q i by ( i k , p k ) ⟷ β k i (i_k,p_k)\longleftrightarrow\beta_k^{\mathbf{i}} ( i k , p k ) ⟷ β k i .
Different compatible readings of the same quiver Γ Q i \Gamma_Q^{\mathbf{i}} Γ Q i are related by commutation moves, that is, by interchanging consecutive vertices with distinct, nonadjacent colors in Δ \Delta Δ . Since the corresponding simple reflections commute, the positive-root label of each vertex is independent of the chosen compatible reading.
When g \mathfrak{g} g is of finite type and w = w 0 w=w_0 w = w 0 , we write
Γ Q : = Γ Q i . \Gamma_Q:=\Gamma_Q^{\mathbf{i}}. Γ Q := Γ Q i .
Then Γ Q \Gamma_Q Γ Q is the usual Auslander–Reiten quiver associated with Q Q Q , its vertex set is labeled by Δ + \Delta_+ Δ + , and
∣ ( Γ Q ) 0 ∣ = ∣ Δ + ∣ . |(\Gamma_Q)_0|=|\Delta_+|. ∣ ( Γ Q ) 0 ∣ = ∣ Δ + ∣.
For a general symmetrizable Cartan datum, we use the finite Q Q Q -adapted quiver Γ Q i \Gamma_Q^{\mathbf{i}} Γ Q i associated with a chosen Q Q Q -adapted sequence, imposing reducedness only when a positive-root labeling is required.
KR-modules over quiver Hecke algebras
Let i = ( i 1 , … , i l ) ∈ Red ( w ) \mathbf{i}=(i_1,\ldots,i_l)\in\operatorname{Red}(w) i = ( i 1 , … , i l ) ∈ Red ( w ) be a Q Q Q -adapted reduced sequence, and let
( i k , p k ) ⟷ β k i ( 1 ≤ k ≤ l ) (i_k,p_k)\longleftrightarrow\beta_k^{\mathbf{i}}\qquad(1\leq k\leq l) ( i k , p k ) ⟷ β k i ( 1 ≤ k ≤ l )
be the corresponding labeling of Γ Q i \Gamma_Q^{\mathbf{i}} Γ Q i . Recall from (23) that the cuspidal module associated with the k k k -th position is S k i S_k^{\mathbf{i}} S k i . For a vertex ( i k , p k ) ∈ Γ Q i (i_k,p_k)\in\Gamma_Q^{\mathbf{i}} ( i k , p k ) ∈ Γ Q i , we write
S Q ( i k , p k ) : = S k i . S_Q(i_k,p_k):=S_k^{\mathbf{i}}. S Q ( i k , p k ) := S k i .
Let r ∈ Z ≥ 0 r\in\mathbb{Z}_{\geq0} r ∈ Z ≥ 0 , and suppose that the vertices
( i , p − 2 r ) , ( i , p − 2 r + 2 ) , … , ( i , p − 2 ) , ( i , p ) (i,p-2r),\ (i,p-2r+2),\ldots,(i,p-2),\ (i,p) ( i , p − 2 r ) , ( i , p − 2 r + 2 ) , … , ( i , p − 2 ) , ( i , p )
occur in Γ Q i \Gamma_Q^{\mathbf{i}} Γ Q i . Then
( S Q ( i , p ) , S Q ( i , p − 2 ) , … , S Q ( i , p − 2 r ) ) \left(S_Q(i,p),S_Q(i,p-2),\ldots,S_Q(i,p-2r)\right) ( S Q ( i , p ) , S Q ( i , p − 2 ) , … , S Q ( i , p − 2 r ) )
is a normal sequence ([38 Proposition 4.3] ). In particular, the convolution product S Q ( i , p ) ∘ S Q ( i , p − 2 ) ∘ ⋯ ∘ S Q ( i , p − 2 r ) S_Q(i,p)\circ S_Q(i,p-2)\circ\cdots\circ S_Q(i,p-2r) S Q ( i , p ) ∘ S Q ( i , p − 2 ) ∘ ⋯ ∘ S Q ( i , p − 2 r ) has a simple head.
We define the corresponding KR-module over the quiver Hecke algebra by
V ( [ i r + 1 ] ) p − r : = hd ( S Q ( i , p ) ∘ S Q ( i , p − 2 ) ∘ ⋯ ∘ S Q ( i , p − 2 r ) ) . (41) \mathbf{V}\left(\left[i^{r+1}\right]\right)_{p-r}:=\operatorname{hd}\left(S_Q(i,p)\circ S_Q(i,p-2)\circ\cdots\circ S_Q(i,p-2r)\right). \tag*{(41)} V ( [ i r + 1 ] ) p − r := hd ( S Q ( i , p ) ∘ S Q ( i , p − 2 ) ∘ ⋯ ∘ S Q ( i , p − 2 r ) ) . (41)
Here p − r p-r p − r is the center of the parameters
p − 2 r , p − 2 r + 2 , … , p − 2 , p . p-2r,\ p-2r+2,\ldots,p-2,\ p. p − 2 r , p − 2 r + 2 , … , p − 2 , p .
Equivalently, for k ≥ 1 k\geq1 k ≥ 1 , the centered notation V ( [ i k ] ) a \mathbf{V}\left(\left[i^k\right]\right)_a V ( [ i k ] ) a is characterized by
V ( [ i k ] ) a = hd ( S Q ( i , a + k − 1 ) ∘ S Q ( i , a + k − 3 ) ∘ ⋯ ∘ S Q ( i , a − k + 1 ) ) , \mathbf{V}\left(\left[i^k\right]\right)_a=\operatorname{hd}\left(S_Q(i,a+k-1)\circ S_Q(i,a+k-3)\circ\cdots\circ S_Q(i,a-k+1)\right), V ( [ i k ] ) a = hd ( S Q ( i , a + k − 1 ) ∘ S Q ( i , a + k − 3 ) ∘ ⋯ ∘ S Q ( i , a − k + 1 ) ) ,
whenever all the corresponding vertices occur in Γ Q i \Gamma_Q^{\mathbf{i}} Γ Q i .
More precisely, suppose that ( i , p − 2 r ) (i,p-2r) ( i , p − 2 r ) and ( i , p ) (i,p) ( i , p ) are the x x x -th and y y y -th vertices, respectively, in the compatible reading defining i i i . Then
i x = i y = i i_x=i_y=i i x = i y = i
and the positions of color i i i in [ x , y ] [x,y] [ x , y ] correspond exactly to
( i , p − 2 r ) , ( i , p − 2 r + 2 ) , … , ( i , p ) . (i,p-2r),\ (i,p-2r+2),\ldots,(i,p). ( i , p − 2 r ) , ( i , p − 2 r + 2 ) , … , ( i , p ) .
Consequently,
V ( [ i r + 1 ] n − r ) ≃ M i [ x , y ] = M ( w ≤ y ϖ i i , w < x ϖ i i ) . V\left(\left[\begin{smallmatrix}i^{r+1}\end{smallmatrix}\right]_{n-r}\right)\simeq M^i[x,y]=M\left(w^i_{\leq y\varpi_i},w^i_{<x\varpi_i}\right). V ( [ i r + 1 ] n − r ) ≃ M i [ x , y ] = M ( w ≤ y ϖ i i , w < x ϖ i i ) .
Thus the KR-modules defined above are precisely the determinantial modules attached to the i i i -boxes associated with the Q Q Q -adapted reduced sequence i \boldsymbol{i} i .
Symmetrized inverse t t t -Cartan matrices
Let
A = ( a i , j ) i , j ∈ I , a i , j : = 2 δ i , j − c i , j . A=(a_{i,j})_{i,j\in I},\qquad a_{i,j}:=2\delta_{i,j}-c_{i,j}. A = ( a i , j ) i , j ∈ I , a i , j := 2 δ i , j − c i , j .
Thus
a i , i = 0 and a i , j = − c i , j ∈ Z ≥ 0 ( i ≠ j ) . a_{i,i}=0\qquad\text{and}\qquad a_{i,j}=-c_{i,j}\in\mathbb{Z}_{\geq0}\quad(i\ne j). a i , i = 0 and a i , j = − c i , j ∈ Z ≥ 0 ( i = j ) .
The t t t -quantized Cartan matrix is
C ( t ) : = ( t + t − 1 ) Id − A = ( c i , j ( t ) ) i , j ∈ I ⟺ c i , j ( t ) = { t + t − 1 , i = j , c i , j , i ≠ j . C(t):=(t+t^{-1})\operatorname{Id}-A=(c_{i,j}(t))_{i,j\in I}\Longleftrightarrow c_{i,j}(t)=
\begin{cases}
t+t^{-1},&i=j,\\
c_{i,j},&i\ne j.
\end{cases} C ( t ) := ( t + t − 1 ) Id − A = ( c i , j ( t ) ) i , j ∈ I ⟺ c i , j ( t ) = { t + t − 1 , c i , j , i = j , i = j .
Since
C ( t ) = t − 1 ( Id − t A + t 2 Id ) , C(t)=t^{-1}(\operatorname{Id}-tA+t^2\operatorname{Id}), C ( t ) = t − 1 ( Id − t A + t 2 Id ) ,
the inverse t t t -quantized Cartan matrix has the t t t -adic expansion
C ( t ) − 1 = t ( Id − t A + t 2 Id ) − 1 ∈ t Mat I ( Z [ [ t ] ] ) . C(t)^{-1}=t(\operatorname{Id}-tA+t^2\operatorname{Id})^{-1}\in t\operatorname{Mat}_{I}(\mathbb{Z}[[t]]). C ( t ) − 1 = t ( Id − t A + t 2 Id ) − 1 ∈ t Mat I ( Z [[ t ]]) .
We refer to
C ~ ( t ) : = D C ( t ) − 1 = ( C ~ i , j ( t ) ) i , j ∈ I \widetilde{C}(t):=DC(t)^{-1}=(\widetilde{C}_{i,j}(t))_{i,j\in I} C ( t ) := D C ( t ) − 1 = ( C i , j ( t ) ) i , j ∈ I
as the symmetrized inverse t t t -quantized Cartan matrix , or simply the symmetrized inverse t t t -Cartan matrix . Note that C ~ ( t ) \widetilde{C}(t) C ( t ) is symmetric.
For i , j ∈ I i,j\in I i , j ∈ I , write
C ~ i , j ( t ) : = ∑ u ∈ Z c ~ i , j ( u ) t u . (42) \widetilde{C}_{i,j}(t):=\sum_{u\in\mathbb{Z}}\widetilde{c}_{i,j}(u)t^u. \tag*{(42)} C i , j ( t ) := u ∈ Z ∑ c i , j ( u ) t u . (42)
Hence c ~ i , j ( u ) = c ~ j , i ( u ) \widetilde{c}_{i,j}(u)=\widetilde{c}_{j,i}(u) c i , j ( u ) = c j , i ( u ) for i , j ∈ I i,j\in I i , j ∈ I and u ∈ Z u\in\mathbb{Z} u ∈ Z .
For i , j ∈ I i,j\in I i , j ∈ I , let d ( i , j ) d(i,j) d ( i , j ) denote the distance between i i i and j j j in △ \triangle △ .
Proposition 4.6 ([11 , 14 , 42 , 12 , 13 ] ). For i , j ∈ I i,j\in I i , j ∈ I and u ∈ Z u\in\mathbb{Z} u ∈ Z , the coefficients c ~ i , j ( u ) \widetilde{c}_{i,j}(u) c i , j ( u ) in (4.2) satisfy the following properties.
(i) We have c ~ i , j ( u ) = 0 \widetilde{c}_{i,j}(u)=0 c i , j ( u ) = 0 for u ≤ d ( i , j ) u\leq d(i,j) u ≤ d ( i , j ) and, in particular, c ~ i , j ( 1 ) = d i δ i , j \widetilde{c}_{i,j}(1)=d_i\delta_{i,j} c i , j ( 1 ) = d i δ i , j .
(ii) If c ~ i , j ( u ) ≠ 0 \widetilde{c}_{i,j}(u)\ne0 c i , j ( u ) = 0 then u + ϵ i + ϵ j ≡ 1 ( m o d 2 ) u+\epsilon_i+\epsilon_j\equiv1\pmod{2} u + ϵ i + ϵ j ≡ 1 ( mod 2 ) .
(iii) If C C C is indecomposable and of infinite type, then c ~ i , j ( u ) ∈ Z ≥ 0 \widetilde{c}_{i,j}(u)\in\mathbb{Z}_{\geq0} c i , j ( u ) ∈ Z ≥ 0 for u ∈ Z u\in\mathbb{Z} u ∈ Z , and hence C ~ i , j ( t ) ∈ t Z ≥ 0 [ [ t ] ] \widetilde{C}_{i,j}(t)\in t\mathbb{Z}_{\geq0}[[t]] C i , j ( t ) ∈ t Z ≥ 0 [[ t ]] .
Proof. The finite-type case of (i), as well as (ii) and (iii), is covered by the cited references. For completeness, we give a direct proof of (i), valid in arbitrary type. Write
( I − t A + t 2 I ) − 1 = ∑ r ≥ 0 Q r t r . (I-tA+t^{2}I)^{-1}=\sum_{r\geq0}Q_{r}t^{r}. ( I − t A + t 2 I ) − 1 = r ≥ 0 ∑ Q r t r .
Comparing coefficients gives
Q 0 = I , Q 1 = A , Q r = A Q r − 1 − Q r − 2 ( r ≥ 2 ) . Q_{0}=I,\qquad Q_{1}=A,\qquad Q_{r}=AQ_{r-1}-Q_{r-2}\quad(r\geq2). Q 0 = I , Q 1 = A , Q r = A Q r − 1 − Q r − 2 ( r ≥ 2 ) .
It follows inductively that each Q r Q_{r} Q r is a polynomial in A A A of degree at most r r r . Since
( A k ) i , j = 0 whenever k < d ( i , j ) , (A^{k})_{i,j}=0\qquad\text{whenever }k<d(i,j), ( A k ) i , j = 0 whenever k < d ( i , j ) ,
we have ( Q r ) i , j = 0 (Q_{r})_{i,j}=0 ( Q r ) i , j = 0 for r < d ( i , j ) r<d(i,j) r < d ( i , j ) . Now
C ~ ( t ) = D t ∑ r ≥ 0 Q r t r . \widetilde{C}(t)=Dt\sum_{r\geq0}Q_{r}t^{r}. C ( t ) = D t r ≥ 0 ∑ Q r t r .
Thus there are no terms of nonpositive degree, and for 1 ≤ u ≤ d ( i , j ) 1\leq u\leq d(i,j) 1 ≤ u ≤ d ( i , j ) the coefficient of t u t^{u} t u vanishes because u − 1 < d ( i , j ) u-1<d(i,j) u − 1 < d ( i , j ) . Hence
c ~ i , j ( u ) = 0 for u ≤ d ( i , j ) . \widetilde{c}_{i,j}(u)=0\qquad\text{for }u\leq d(i,j). c i , j ( u ) = 0 for u ≤ d ( i , j ) .
Moreover, C ~ ( t ) = t D + O ( t 2 ) \widetilde{C}(t)=tD+O(t^{2}) C ( t ) = t D + O ( t 2 ) , and therefore
c ~ i , j ( 1 ) = d i δ i , j . \widetilde{c}_{i,j}(1)=d_{i}\delta_{i,j}. c i , j ( 1 ) = d i δ i , j .
We next record the compatibility of C ~ ( t ) \widetilde{C}(t) C ( t ) with folding. Keep the folding datum of § 3 . Let
C ~ g ^ ( t ) = ( C ~ ı , ȷ g ^ ( t ) ) ı , ȷ ∈ I ^ and C ~ g ( t ) = ( C ~ i , j g ( t ) ) i , j ∈ I \widetilde{C}^{\widehat{\mathfrak{g}}}(t)=\left(\widetilde{C}^{\widehat{\mathfrak{g}}}_{\imath,\jmath}(t)\right)_{\imath,\jmath\in\widehat{I}}
\qquad\text{and}\qquad
\widetilde{C}^{\mathfrak{g}}(t)=\left(\widetilde{C}^{\mathfrak{g}}_{i,j}(t)\right)_{i,j\in I} C g ( t ) = ( C , g ( t ) ) , ∈ I and C g ( t ) = ( C i , j g ( t ) ) i , j ∈ I
be the corresponding symmetrized inverse t t t -Cartan matrices. Since the upstairs datum is simply-laced, we have
C ~ g ^ ( t ) = C g ^ ( t ) − 1 . \widetilde{C}^{\widehat{\mathfrak{g}}}(t)=C^{\widehat{\mathfrak{g}}}(t)^{-1}. C g ( t ) = C g ( t ) − 1 .
Let R : Q ( t ) I ⟶ Q ( t ) I ^ R:\mathbb{Q}(t)^{I}\longrightarrow\mathbb{Q}(t)^{\widehat{I}} R : Q ( t ) I ⟶ Q ( t ) I be the lifting map onto the fiber-constant subspace, defined by
( R v ) ı : = v π ( ı ) . (Rv)_{\imath}:=v_{\pi(\imath)}. ( R v ) := v π ( ) .
With respect to the standard bases, its matrix is given by
R ı , j = δ π ( ı ) , j ( ı ∈ I ^ , j ∈ I ) . R_{\imath,j}=\delta_{\pi(\imath),j}\qquad(\imath\in\widehat{I},\ j\in I). R , j = δ π ( ) , j ( ∈ I , j ∈ I ) .
The quotient definition c j , i = ∑ ȷ ∈ I j c ^ ȷ , ı c_{j,i}=\sum_{\jmath\in I_{j}}\widehat{c}_{\jmath,\imath} c j , i = ∑ ∈ I j c , for ı ∈ I i \imath\in I_{i} ∈ I i , together with the symmetry of C ^ \widehat{C} C , gives (4.3)
C ~ g ^ ( t ) R = R C ~ g ( t ) T . \widetilde{C}^{\widehat{\mathfrak{g}}}(t)R=R\widetilde{C}^{\mathfrak{g}}(t)^{\mathsf{T}}. C g ( t ) R = R C g ( t ) T .
Example 4.7. Consider the standard folding A 5 ⟶ C 3 A_{5}\longrightarrow C_{3} A 5 ⟶ C 3 induced by the involution 1 ⟷ 5 1\longleftrightarrow5 1 ⟷ 5 , 2 ⟷ 4 2\longleftrightarrow4 2 ⟷ 4 , 3 ⟷ 3 3\longleftrightarrow3 3 ⟷ 3 . The vertex fibers and the downstairs symmetrizer are
I 1 = { 1 , 5 } , I 2 = { 2 , 4 } , I 3 = { 3 } , D C 3 = diag ( 1 , 1 , 2 ) . I_{1}=\{1,5\},\qquad I_{2}=\{2,4\},\qquad I_{3}=\{3\},\qquad D^{C_{3}}=\operatorname{diag}(1,1,2). I 1 = { 1 , 5 } , I 2 = { 2 , 4 } , I 3 = { 3 } , D C 3 = diag ( 1 , 1 , 2 ) .
Set a : = t + t − 1 a:=t+t^{-1} a := t + t − 1 . Then C A 5 ( t ) C^{A_{5}}(t) C A 5 ( t ) , C C 3 ( t ) C^{C_{3}}(t) C C 3 ( t ) and the lifting map R R R are given as follows:
C A 5 ( t ) = ( a − 1 0 0 0 − 1 a − 1 0 0 0 − 1 a − 1 0 0 0 − 1 a − 1 0 0 0 − 1 a ) , C C 3 ( t ) = ( a − 1 0 − 1 a − 2 0 − 1 a ) , R = ( 1 0 0 0 1 0 0 0 1 0 1 0 1 0 0 ) . C^{A_{5}}(t)=
\begin{pmatrix}
a&-1&0&0&0\\
-1&a&-1&0&0\\
0&-1&a&-1&0\\
0&0&-1&a&-1\\
0&0&0&-1&a
\end{pmatrix},
\qquad
C^{C_{3}}(t)=
\begin{pmatrix}
a&-1&0\\
-1&a&-2\\
0&-1&a
\end{pmatrix},
\qquad
R=
\begin{pmatrix}
1&0&0\\
0&1&0\\
0&0&1\\
0&1&0\\
1&0&0
\end{pmatrix}. C A 5 ( t ) = a − 1 0 0 0 − 1 a − 1 0 0 0 − 1 a − 1 0 0 0 − 1 a − 1 0 0 0 − 1 a , C C 3 ( t ) = a − 1 0 − 1 a − 1 0 − 2 a , R = 1 0 0 0 1 0 1 0 1 0 0 0 1 0 0 .
satisfying
C A 5 ( t ) R = ( a − 1 0 − 1 a − 1 0 − 2 a − 1 a − 1 a − 1 0 ) = R C C 3 ( t ) T , C^{A_5}(t)R =
\begin{pmatrix}
a & -1 & 0 \\
-1 & a & -1 \\
0 & -2 & a \\
-1 & a & -1 \\
a & -1 & 0
\end{pmatrix}
= R C^{C_3}(t)^{T}, C A 5 ( t ) R = a − 1 0 − 1 a − 1 a − 2 a − 1 0 − 1 a − 1 0 = R C C 3 ( t ) T ,
which illustrates (4.3).
Theorem 4.8. Let i , j ∈ I i,j \in I i , j ∈ I , and choose ι ∈ I i \iota\in I_i ι ∈ I i and ȷ ∈ I j \jmath\in I_j ∈ I j . Then
C ~ i , j g ( t ) = d i ∑ ι ′ ∈ I i C ~ ι ′ , ȷ g ( t ) = m ∣ Γ ∣ ∑ g ∈ Γ C ~ g ( ι ) , ȷ g ( t ) . (43) \widetilde{C}^{\mathfrak{g}}_{i,j}(t) = d_i \sum_{\iota' \in I_i} \widetilde{C}^{\mathfrak{g}}_{\iota',\jmath}(t) = \frac{m}{|\Gamma|} \sum_{g \in\Gamma} \widetilde{C}^{\mathfrak{g}}_{g(\iota),\jmath}(t). \tag*{(43)} C i , j g ( t ) = d i ι ′ ∈ I i ∑ C ι ′ , g ( t ) = ∣Γ∣ m g ∈ Γ ∑ C g ( ι ) , g ( t ) . (43) These expressions are independent of the chosen lifts. By symmetry, one also has
C ~ i , j g ( t ) = d j ∑ ȷ ′ ∈ I j C ~ ι , ȷ ′ g ( t ) = m ∣ Γ ∣ ∑ g ∈ Γ C ~ ι , g ( ȷ ) g ( t ) . (44) \widetilde{C}^{\mathfrak{g}}_{i,j}(t) = d_j \sum_{\jmath' \in I_j} \widetilde{C}^{\mathfrak{g}}_{\iota,\jmath'}(t) = \frac{m}{|\Gamma|} \sum_{g \in\Gamma} \widetilde{C}^{\mathfrak{g}}_{\iota,g(\jmath)}(t). \tag*{(44)} C i , j g ( t ) = d j ′ ∈ I j ∑ C ι , ′ g ( t ) = ∣Γ∣ m g ∈ Γ ∑ C ι , g ( ) g ( t ) . (44)
Proof. Transposing (4.3) and using the symmetry of C g ( t ) C^{\mathfrak{g}}(t) C g ( t ) gives R T C ~ g ( t ) = C g ( t ) R T R^{T}\widetilde{C}^{\mathfrak{g}}(t) = C^{\mathfrak{g}}(t)R^{T} R T C g ( t ) = C g ( t ) R T . Since both Cartan matrices are invertible as t t t -adic formal matrices,
R T C ~ g ( t ) − 1 = C g ( t ) − 1 R T . R^{T}\widetilde{C}^{\mathfrak{g}}(t)^{-1} = C^{\mathfrak{g}}(t)^{-1}R^{T}. R T C g ( t ) − 1 = C g ( t ) − 1 R T .
Using C ~ g ( t ) = C g ( t ) − 1 \widetilde{C}^{\mathfrak{g}}(t) = C^{\mathfrak{g}}(t)^{-1} C g ( t ) = C g ( t ) − 1 and comparing the ( i , ȷ ) (i,\jmath) ( i , ) -entries for ȷ ∈ I j \jmath\in I_j ∈ I j , we obtain
∑ ι ′ ∈ I i C ~ ι ′ , ȷ g ( t ) = ( C g ( t ) − 1 ) i , j . \sum_{\iota' \in I_i} \widetilde{C}^{\mathfrak{g}}_{\iota',\jmath}(t) = \left(C^{\mathfrak{g}}(t)^{-1}\right)_{i,j}. ι ′ ∈ I i ∑ C ι ′ , g ( t ) = ( C g ( t ) − 1 ) i , j .
Multiplication by d i d_i d i gives the first fiber formula in (43) , independently of the chosen lift ȷ \jmath . For any ι ∈ I i \iota\in I_i ι ∈ I i , orbit–stabilizer gives
∑ g ∈ Γ C ~ g ( ι ) , ȷ g ( t ) = ∣ Γ ∣ e i ∑ ι ′ ∈ I i C ~ ι ′ , ȷ g ( t ) . \sum_{g \in\Gamma} \widetilde{C}^{\mathfrak{g}}_{g(\iota),\jmath}(t) = \frac{|\Gamma|}{e_i} \sum_{\iota' \in I_i} \widetilde{C}^{\mathfrak{g}}_{\iota',\jmath}(t). g ∈ Γ ∑ C g ( ι ) , g ( t ) = e i ∣Γ∣ ι ′ ∈ I i ∑ C ι ′ , g ( t ) .
Together with d i e i = m d_i e_i = m d i e i = m , this gives the group-sum factor m / ∣ Γ ∣ m/|\Gamma| m /∣Γ∣ , independently of ι \iota ι . The second fiber formula (44) follows by symmetry of C ~ g ( t ) \widetilde{C}^{\mathfrak{g}}(t) C g ( t ) and C ~ g ( t ) \widetilde{C}^{\mathfrak{g}}(t) C g ( t ) .
The folding identity also holds coefficientwise:
c i , j g ( u ) = d i ∑ ι ′ ∈ I i c ι ′ , ȷ g ( u ) = m ∣ Γ ∣ ∑ g ∈ Γ c g ( ι ) , ȷ g ( u ) ( u ≥ 1 ) . c^{\mathfrak{g}}_{i,j}(u) = d_i \sum_{\iota' \in I_i} c^{\mathfrak{g}}_{\iota',\jmath}(u) = \frac{m}{|\Gamma|} \sum_{g \in\Gamma} c^{\mathfrak{g}}_{g(\iota),\jmath}(u) \qquad(u \ge1). c i , j g ( u ) = d i ι ′ ∈ I i ∑ c ι ′ , g ( u ) = ∣Γ∣ m g ∈ Γ ∑ c g ( ι ) , g ( u ) ( u ≥ 1 ) .
Remark 4.9. Theorem 4.8 applies to admissible foldings of arbitrary Cartan type, including indefinite type.
Example 4.10. Consider the standard folding D 4 ( 1 ) ⟶ B 3 ( 1 ) D_4^{(1)} \longrightarrow B_3^{(1)} D 4 ( 1 ) ⟶ B 3 ( 1 ) . We label the vertices of D 4 ( 1 ) D_4^{(1)} D 4 ( 1 ) so that 2 is the central vertex and 0, 1, 3, 4 are its four adjacent vertices. Let Γ \Gamma Γ be generated by the diagram automorphism interchanging 3 and 4. The vertex fibers are I 0 = { 0 } I_0 = \{0\} I 0 = { 0 } , I 1 = { 1 } I_1 = \{1\} I 1 = { 1 } , I 2 = { 2 } I_2 = \{2\} I 2 = { 2 } , I 3 = { 3 , 4 } I_3 = \{3,4\} I 3 = { 3 , 4 } , and the symmetrizer of B 3 ( 1 ) B_3^{(1)} B 3 ( 1 ) is D = diag ( 2 , 2 , 2 , 1 ) D = \operatorname{diag}(2,2,2,1) D = diag ( 2 , 2 , 2 , 1 ) .
The first terms of the relevant entries upstairs are
C ~ 0 , 0 D 4 ( 1 ) ( t ) = t + 2 t 5 + t 7 + O ( t 9 ) , C ~ 0 , 1 D 4 ( 1 ) ( t ) = t 3 + t 5 + 2 t 7 + O ( t 9 ) . \widetilde{C}^{D_4^{(1)}}_{0,0}(t) = t + 2t^5 + t^7 + O(t^9), \qquad\widetilde{C}^{D_4^{(1)}}_{0,1}(t) = t^3 + t^5 + 2t^7 + O(t^9). C 0 , 0 D 4 ( 1 ) ( t ) = t + 2 t 5 + t 7 + O ( t 9 ) , C 0 , 1 D 4 ( 1 ) ( t ) = t 3 + t 5 + 2 t 7 + O ( t 9 ) .
The corresponding entries downstairs are
C ~ 0 , 0 B 3 ( 1 ) ( t ) = 2 t + 4 t 5 + 2 t 7 + O ( t 9 ) , C ~ 0 , 1 B 3 ( 1 ) ( t ) = 2 t 3 + 2 t 5 + 4 t 7 + O ( t 9 ) . \widetilde{C}^{B_3^{(1)}}_{0,0}(t)=2t+4t^5+2t^7+O(t^9),\qquad\widetilde{C}^{B_3^{(1)}}_{0,1}(t)=2t^3+2t^5+4t^7+O(t^9). C 0 , 0 B 3 ( 1 ) ( t ) = 2 t + 4 t 5 + 2 t 7 + O ( t 9 ) , C 0 , 1 B 3 ( 1 ) ( t ) = 2 t 3 + 2 t 5 + 4 t 7 + O ( t 9 ) .
Thus Theorem 4.8 gives
C ~ 0 , 0 B 3 ( 1 ) ( t ) = d 0 ∑ r ∈ I 0 C ~ r , 0 D 4 ( 1 ) ( t ) = 2 C ~ 0 , 0 D 4 ( 1 ) ( t ) , C ~ 0 , 1 B 3 ( 1 ) ( t ) = d 0 ∑ r ∈ I 0 C ~ r , 1 D 4 ( 1 ) ( t ) = 2 C ~ 0 , 1 D 4 ( 1 ) ( t ) . \widetilde{C}^{B_3^{(1)}}_{0,0}(t)=d_0\sum_{r\in I_0}\widetilde{C}^{D_4^{(1)}}_{r,0}(t)=2\widetilde{C}^{D_4^{(1)}}_{0,0}(t),\qquad\widetilde{C}^{B_3^{(1)}}_{0,1}(t)=d_0\sum_{r\in I_0}\widetilde{C}^{D_4^{(1)}}_{r,1}(t)=2\widetilde{C}^{D_4^{(1)}}_{0,1}(t). C 0 , 0 B 3 ( 1 ) ( t ) = d 0 r ∈ I 0 ∑ C r , 0 D 4 ( 1 ) ( t ) = 2 C 0 , 0 D 4 ( 1 ) ( t ) , C 0 , 1 B 3 ( 1 ) ( t ) = d 0 r ∈ I 0 ∑ C r , 1 D 4 ( 1 ) ( t ) = 2 C 0 , 1 D 4 ( 1 ) ( t ) .
The nontrivial vertex fiber gives a sum of two upstairs entries. Indeed,
C ~ 3 , 0 D 4 ( 1 ) ( t ) = t 3 + t 5 + 2 t 7 + O ( t 9 ) , C ~ 4 , 0 D 4 ( 1 ) ( t ) = t 3 + t 5 + 2 t 7 + O ( t 9 ) , \widetilde{C}^{D_4^{(1)}}_{3,0}(t)=t^3+t^5+2t^7+O(t^9),\qquad\widetilde{C}^{D_4^{(1)}}_{4,0}(t)=t^3+t^5+2t^7+O(t^9), C 3 , 0 D 4 ( 1 ) ( t ) = t 3 + t 5 + 2 t 7 + O ( t 9 ) , C 4 , 0 D 4 ( 1 ) ( t ) = t 3 + t 5 + 2 t 7 + O ( t 9 ) ,
and hence
C ~ 3 , 0 B 3 ( 1 ) ( t ) = ∑ r ∈ I 3 C ~ r , 0 D 4 ( 1 ) ( t ) = C ~ 3 , 0 D 4 ( 1 ) ( t ) + C ~ 4 , 0 D 4 ( 1 ) ( t ) = 2 t 3 + 2 t 5 + 4 t 7 + O ( t 9 ) . \widetilde{C}^{B_3^{(1)}}_{3,0}(t)=\sum_{r\in I_3}\widetilde{C}^{D_4^{(1)}}_{r,0}(t)=\widetilde{C}^{D_4^{(1)}}_{3,0}(t)+\widetilde{C}^{D_4^{(1)}}_{4,0}(t)=2t^3+2t^5+4t^7+O(t^9). C 3 , 0 B 3 ( 1 ) ( t ) = r ∈ I 3 ∑ C r , 0 D 4 ( 1 ) ( t ) = C 3 , 0 D 4 ( 1 ) ( t ) + C 4 , 0 D 4 ( 1 ) ( t ) = 2 t 3 + 2 t 5 + 4 t 7 + O ( t 9 ) .
This illustrates both the factor d 0 = 2 d_0=2 d 0 = 2 on a singleton fiber and the sum over a nontrivial orbit.
We next turn to the explicit formulas for the d \mathfrak{d} d -invariants between Q Q Q -adapted KR-modules of finite types and their compatibility with folding.
Explicit finite-type formulas and folding
Assume in this subsection that C \mathcal{C} C is of finite type. Let Q = ( △ , ξ ) Q=(\triangle,\xi) Q = ( △ , ξ ) be a Dynkin quiver, and let i = ( i 1 , … , i N ) ∈ Red ( w 0 ) \mathbf{i}=(i_1,\ldots,i_N)\in\operatorname{Red}(w_0) i = ( i 1 , … , i N ) ∈ Red ( w 0 ) (N = ∣ Δ + ∣ N=|\Delta_+| N = ∣ Δ + ∣ ) be a Q Q Q -adapted reduced sequence. Then the associated quiver
Υ Q = Υ i Q \Upsilon_Q=\Upsilon_{\mathbf{i}}^Q Υ Q = Υ i Q
is the usual Auslander–Reiten quiver of Q Q Q . Its vertices are labeled by the positive roots through
( i k , p k ) ⟷ β k i , (i_k,p_k)\longleftrightarrow\beta_k^{\mathbf{i}}, ( i k , p k ) ⟷ β k i ,
where β k i \beta_k^{\mathbf{i}} β k i is given by (12) , and the corresponding cuspidal module is
S Q ( i k , p k ) = S k i . S_Q(i_k,p_k)=S_{k}^{\mathbf{i}}. S Q ( i k , p k ) = S k i .
The d \mathfrak{d} d -values between these cuspidal modules are completely determined by the symmetrized inverse t t t -Cartan matrix.
Theorem 4.11 ([43 Theorem 8.2] ). For any ( i , p ) , ( j , q ) ∈ ( Υ Q ) 0 (i,p),(j,q)\in(\Upsilon_Q)_0 ( i , p ) , ( j , q ) ∈ ( Υ Q ) 0 , one has
d ( S Q ( i , p ) , S Q ( j , q ) ) = c ~ i , j ( ∣ p − q ∣ − 1 ) . \mathfrak{d}\bigl(S_Q(i,p),S_Q(j,q)\bigr)=\widetilde{c}_{i,j}(|p-q|-1). d ( S Q ( i , p ) , S Q ( j , q ) ) = c i , j ( ∣ p − q ∣ − 1 ) .
The formula depends only on the colors i , j i,j i , j and the relative position ∣ p − q ∣ |p-q| ∣ p − q ∣ , and hence applies directly to the sink-adapted convention used here.
Although our main objects are modules over quiver Hecke algebras, in simply-laced finite type g \mathfrak{g} g they are related to modules over quantum affine algebras U q ′ ( g ( 1 ) ) U'_q(\mathfrak{g}^{(1)}) U q ′ ( g ( 1 ) ) through generalized quantum affine Schur–Weyl duality. For a Dynkin quiver Q Q Q , the construction in [26 , 27 , 29 ] gives an exact monoidal functor
F Q : R - g m o d ⟶ C Q , \mathcal{F}_Q:R\text{-}\mathrm{gmod}\longrightarrow\mathcal{C}_Q, F Q : R - gmod ⟶ C Q ,
where C Q \mathcal{C}_Q C Q is the heart subcategory associated with Q Q Q inside the category of finite-dimensional integrable U q ′ ( g ( 1 ) ) U'_q(\mathfrak{g}^{(1)}) U q ′ ( g ( 1 ) ) -modules. It sends the cuspidal module attached to a vertex ( i , p ) ∈ ( Υ Q ) 0 (i,p)\in(\Upsilon_Q)_0 ( i , p ) ∈ ( Υ Q ) 0 to the corresponding fundamental module:
F Q ( S Q ( i , p ) ) ≃ V ( ϖ i ) ( − q ) p . \mathcal{F}_Q\bigl(S_Q(i,p)\bigr)\simeq V(\varpi_i)_{(-q)^p}. F Q ( S Q ( i , p ) ) ≃ V ( ϖ i ) ( − q ) p .
For quantum affine modules, Λ \Lambda Λ and d \mathfrak{d} d denote the integer invariants defined from the renormalizing coefficient of the universal R R R -matrix in [40 Definition 2.10] . Generalized quantum affine Schur–Weyl duality preserves these invariants:
Λ ( M , N ) = Λ ( F ~ Q ( M ) , F ~ Q ( N ) ) , d ( M , N ) = d ( F ~ Q ( M ) , F ~ Q ( N ) ) \Lambda(M,N)=\Lambda(\widetilde{\mathcal{F}}_Q(M),\widetilde{\mathcal{F}}_Q(N)),\qquad\mathfrak{d}(M,N)=\mathfrak{d}(\widetilde{\mathcal{F}}_Q(M),\widetilde{\mathcal{F}}_Q(N)) Λ ( M , N ) = Λ ( F Q ( M ) , F Q ( N )) , d ( M , N ) = d ( F Q ( M ) , F Q ( N ))
for simple modules M , N ∈ R - g m o d M,N\in R\text{-}\mathrm{gmod} M , N ∈ R - gmod ([40 Theorem 4.12] ). In particular, the quiver Hecke KR-module associated with the consecutive vertices
( i , p ) , ( i , p − 2 ) , … , ( i , p − 2 r ) (i,p),\ (i,p-2),\ldots,(i,p-2r) ( i , p ) , ( i , p − 2 ) , … , ( i , p − 2 r )
is sent to the KR-module over U q ′ ( g ( 1 ) ) U_q'(\mathfrak{g}^{(1)}) U q ′ ( g ( 1 ) )
V ( [ i r + 1 ] ) p − r \mathrm{V}\left(\left[\begin{smallmatrix}i\\r+1\end{smallmatrix}\right]\right)_{p-r} V ( [ i r + 1 ] ) p − r
(see [53 ] for the centered notation). Hence, whenever all the vertices defining the corresponding KR-modules lie in Υ Q \Upsilon_Q Υ Q , the quiver Hecke and quantum affine d \mathfrak{d} d -invariants agree.
The quantum affine formulation also allows arbitrary centered parameters, without requiring all the vertices defining the corresponding KR-modules to lie in Υ Q \Upsilon_Q Υ Q . The denominator formulas for KR-modules and their relation with the symmetrized inverse t t t -Cartan matrix are established in [14 , 53 , 3 ] . We first state this formula for arbitrary centered quantum affine KR-modules of type A D E ADE A D E . Since F ~ Q \widetilde{\mathcal{F}}_Q F Q preserves d \mathfrak{d} d , the same formula holds for quiver Hecke KR-modules whenever all their defining vertices lie in Υ Q \Upsilon_Q Υ Q .
Let g \mathfrak{g} g be of finite type A A A , D D D , or E E E , and let h h h be its Coxeter number. For i ∈ I i\in I i ∈ I , k ∈ Z > 0 k\in\mathbb{Z}_{>0} k ∈ Z > 0 , and p ∈ Z p\in\mathbb{Z} p ∈ Z , write V ( [ i k ] ) p \mathrm{V}\left(\left[\begin{smallmatrix}i\\k\end{smallmatrix}\right]\right)_p V ( [ i k ] ) p for the quantum affine KR-module over U q ′ ( g ( 1 ) ) U_q'(\mathfrak{g}^{(1)}) U q ′ ( g ( 1 ) ) with the centered spectral parameters
V ( [ i k ] ) p : = hd ( V ( ϖ i ) ( − q ) p + k − 1 ⊗ V ( ϖ i ) ( − q ) p + k − 3 ⊗ ⋯ ⊗ V ( ϖ i ) ( − q ) p − k + 1 ) . \mathrm{V}\left(\left[\begin{smallmatrix}i\\k\end{smallmatrix}\right]\right)_p:=\operatorname{hd}\left(\mathrm{V}(\varpi_i)_{(-q)^{p+k-1}}\otimes\mathrm{V}(\varpi_i)_{(-q)^{p+k-3}}\otimes\cdots\otimes\mathrm{V}(\varpi_i)_{(-q)^{p-k+1}}\right). V ( [ i k ] ) p := hd ( V ( ϖ i ) ( − q ) p + k − 1 ⊗ V ( ϖ i ) ( − q ) p + k − 3 ⊗ ⋯ ⊗ V ( ϖ i ) ( − q ) p − k + 1 ) .
For finite-dimensional simple modules U U U and V V V over U q ′ ( g ( 1 ) ) U_q'(\mathfrak{g}^{(1)}) U q ′ ( g ( 1 ) ) , let V z V_z V z be the spectral shift of V V V by the indeterminate z z z . Following [3 (5.3)] , let o ( U , V ) o(U,V) o ( U , V ) denote the pole order at z = 1 z=1 z = 1 of the normalized R R R -matrix from U ⊗ V z U\otimes V_z U ⊗ V z to V z ⊗ U V_z\otimes U V z ⊗ U , and set
d ( U , V ) : = o ( U , V ) + o ( V , U ) . \mathfrak{d}(U,V):=o(U,V)+o(V,U). d ( U , V ) := o ( U , V ) + o ( V , U ) .
By [35 Proposition 3.16] , it is expressed in terms of the orders of zeros of the normalized R R R -matrix denominators.
Finally, set
C ~ i , j [ h ] ( t ) : = ∑ v = 1 h c ~ i , j ( v ) t v , [ m ] t : = t m − 1 + t m − 3 + ⋯ + t 1 − m . \widetilde{C}^{[h]}_{i,j}(t):=\sum_{v=1}^{h}\widetilde{c}_{i,j}(v)t^v,\qquad[m]_t:=t^{m-1}+t^{m-3}+\cdots+t^{1-m}. C i , j [ h ] ( t ) := v = 1 ∑ h c i , j ( v ) t v , [ m ] t := t m − 1 + t m − 3 + ⋯ + t 1 − m .
Theorem 4.12 ([53 ] and [3 Theorem 6.27] ). Assume that g \mathfrak{g} g is of simply-laced finite type A A A , D D D , or E E E , and let h h h be its Coxeter number. Let i , j ∈ I i,j\in I i , j ∈ I , k , ℓ ∈ Z > 0 k,\ell\in\mathbb{Z}_{>0} k , ℓ ∈ Z > 0 , and p , s ∈ Z p,s\in\mathbb{Z} p , s ∈ Z . Set
K : = max ( k , ℓ ) , L : = min ( k , ℓ ) , u : = ∣ p − s ∣ . K:=\max(k,\ell),\qquad L:=\min(k,\ell),\qquad u:=|p-s|. K := max ( k , ℓ ) , L := min ( k , ℓ ) , u := ∣ p − s ∣.
Then we have
d ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = [ t u ] ( t K [ L ] t C ~ i , j [ h ] ( t ) ) . (45) \mathfrak{d}\left(\mathrm{V}\left(\left[\begin{smallmatrix}i\\k\end{smallmatrix}\right]\right)_p,\mathrm{V}\left(\left[\begin{smallmatrix}j\\\ell\end{smallmatrix}\right]\right)_s\right)
=
\left[t^u\right]\left(t^K[L]_t\widetilde{C}^{[h]}_{i,j}(t)\right). \tag*{(45)} d ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = [ t u ] ( t K [ L ] t C i , j [ h ] ( t ) ) . (45)
For v ∈ Z v \in\mathbb{Z} v ∈ Z , set
c ~ i , j [ h ] ( v ) : = { c ~ i , j ( v ) , 1 ≤ v ≤ h , 0 , otherwise . \widetilde{c}^{[h]}_{i,j}(v):=
\begin{cases}
\widetilde{c}_{i,j}(v), & 1 \le v \le h,\\
0, & \text{otherwise}.
\end{cases} c i , j [ h ] ( v ) := { c i , j ( v ) , 0 , 1 ≤ v ≤ h , otherwise .
Corollary 4.13. With the notation of Theorem 4.12, one has
d ( V ( [ i k ] p ) , V ( [ j ℓ ] s ) ) = ∑ a = 0 L − 1 c ~ i , j [ h ] ( u − ∣ k − ℓ ∣ − 2 a − 1 ) . (46) \mathsf{d}\left(\mathsf{V}\left(\left[\begin{smallmatrix}i^k\end{smallmatrix}\right]_p\right),\mathsf{V}\left(\left[\begin{smallmatrix}j^\ell\end{smallmatrix}\right]_s\right)\right)
=
\sum_{a=0}^{L-1}\widetilde{c}^{[h]}_{i,j}(u-|k-\ell|-2a-1). \tag*{(46)} d ( V ( [ i k ] p ) , V ( [ j ℓ ] s ) ) = a = 0 ∑ L − 1 c i , j [ h ] ( u − ∣ k − ℓ ∣ − 2 a − 1 ) . (46)
Proof. Since t K [ L ] t = ∑ r = 0 L − 1 t K + L − 1 − 2 r t^K[L]_t=\sum_{r=0}^{L-1}t^{K+L-1-2r} t K [ L ] t = ∑ r = 0 L − 1 t K + L − 1 − 2 r , the coefficient of t u t^u t u in the right-hand side of (45) is
∑ r = 0 L − 1 c ~ i , j [ h ] ( u − K − L + 1 + 2 r ) . \sum_{r=0}^{L-1}\widetilde{c}^{[h]}_{i,j}(u-K-L+1+2r). r = 0 ∑ L − 1 c i , j [ h ] ( u − K − L + 1 + 2 r ) .
Replacing r r r by L − 1 − a L-1-a L − 1 − a and using K − L = ∣ k − ℓ ∣ K-L=|k-\ell| K − L = ∣ k − ℓ ∣ gives the desired formula. □
We now make Corollary 4.13 explicit in type A A A . Assume that g \mathfrak{g} g is of type A n A_n A n , and set
ρ i , j ( n ) : = min ( i , j , n + 1 − i , n + 1 − j ) . \rho^{(n)}_{i,j}:=\min(i,j,n+1-i,n+1-j). ρ i , j ( n ) := min ( i , j , n + 1 − i , n + 1 − j ) .
The coefficients of C ~ i , j [ h ] ( t ) \widetilde{C}^{[h]}_{i,j}(t) C i , j [ h ] ( t ) are
c ~ i , j [ h ] ( v ) = { 1 , v = ∣ i − j ∣ + 2 r − 1 for some 1 ≤ r ≤ ρ i , j ( n ) , 0 , otherwise . (47) \widetilde{c}^{[h]}_{i,j}(v)=
\begin{cases}
1, & v=|i-j|+2r-1\text{ for some }1\le r\le\rho^{(n)}_{i,j},\\
0, & \text{otherwise}.
\end{cases} \tag*{(47)} c i , j [ h ] ( v ) = { 1 , 0 , v = ∣ i − j ∣ + 2 r − 1 for some 1 ≤ r ≤ ρ i , j ( n ) , otherwise . (47)
(see [42 Theorem 4.12(1)] for more details).
Corollary 4.14. Let g \mathfrak{g} g be of type A n A_n A n . For i , j ∈ I i,j\in I i , j ∈ I , k , ℓ ∈ Z > 0 k,\ell\in\mathbb{Z}_{>0} k , ℓ ∈ Z > 0 , and p , s ∈ Z p,s\in\mathbb{Z} p , s ∈ Z , set
L : = min ( k , ℓ ) , u : = ∣ p − s ∣ , ν : = u − ∣ i − j ∣ − ∣ k − ℓ ∣ 2 . L:=\min(k,\ell),\qquad u:=|p-s|,\qquad\nu:=\frac{u-|i-j|-|k-\ell|}{2}. L := min ( k , ℓ ) , u := ∣ p − s ∣ , ν := 2 u − ∣ i − j ∣ − ∣ k − ℓ ∣ . Then we have
d ( V ( [ i k ] p ) , V ( [ j ℓ ] s ) ) = { [ min ( ν , ρ i , j ( n ) , L , ρ i , j ( n ) + L − ν ) ] + , ν ∈ Z , 0 , ν ∉ Z . (48) \mathsf{d}\left(\mathsf{V}\left(\left[\begin{smallmatrix}i^k\end{smallmatrix}\right]_p\right),\mathsf{V}\left(\left[\begin{smallmatrix}j^\ell\end{smallmatrix}\right]_s\right)\right)
=
\begin{cases}
\left[\min(\nu,\rho^{(n)}_{i,j},L,\rho^{(n)}_{i,j}+L-\nu)\right]_+, & \nu\in\mathbb{Z},\\
0, & \nu\notin\mathbb{Z}.
\end{cases} \tag*{(48)} d ( V ( [ i k ] p ) , V ( [ j ℓ ] s ) ) = ⎩ ⎨ ⎧ [ min ( ν , ρ i , j ( n ) , L , ρ i , j ( n ) + L − ν ) ] + , 0 , ν ∈ Z , ν ∈ / Z . (48) Proof. By (47) and Corollary 4.13, the desired d \mathsf{d} d -value is the number of pairs
( r , a ) ∈ Z ≥ 1 × Z ≥ 0 (r,a)\in\mathbb{Z}_{\ge1}\times\mathbb{Z}_{\ge0} ( r , a ) ∈ Z ≥ 1 × Z ≥ 0 satisfying
1 ≤ r ≤ ρ i , j ( n ) , 0 ≤ a ≤ L − 1 , r + a = ν . 1\le r\le\rho^{(n)}_{i,j},\qquad0\le a\le L-1,\qquad r+a=\nu. 1 ≤ r ≤ ρ i , j ( n ) , 0 ≤ a ≤ L − 1 , r + a = ν . It vanishes unless ν ∈ Z \nu\in\mathbb{Z} ν ∈ Z . If ν ∈ Z \nu\in\mathbb{Z} ν ∈ Z , the possible values of r r r form the interval
max ( 1 , ν − L + 1 ) ≤ r ≤ min ( ρ i , j ( n ) , ν ) . \max(1,\nu-L+1)\le r\le\min(\rho^{(n)}_{i,j},\nu). max ( 1 , ν − L + 1 ) ≤ r ≤ min ( ρ i , j ( n ) , ν ) . Its cardinality is
[ min ( ρ i , j ( n ) , ν ) − max ( 1 , ν − L + 1 ) + 1 ] + , \left[\min(\rho^{(n)}_{i,j},\nu)-\max(1,\nu-L+1)+1\right]_+, [ min ( ρ i , j ( n ) , ν ) − max ( 1 , ν − L + 1 ) + 1 ] + , which is equal to the right-hand side of (48) . □
Example 4.15. Let g \mathfrak{g} g be of type A 5 A_5 A 5 , and consider V ( [ 2 2 ] ) p V([2^2])_p V ([ 2 2 ] ) p and V ( [ 3 4 ] ) s V([3^4])_s V ([ 3 4 ] ) s . Here we have ρ 2 , 3 2 = 2 \rho_{2,3}^2=2 ρ 2 , 3 2 = 2 and L = 2 L=2 L = 2 . Consequently,
∂ ( V ( [ 2 2 ] ) p , V ( [ 3 4 ] ) s ) = { 1 , ∣ p − s ∣ = 5 , or 9 , 2 , ∣ p − s ∣ = 7 , 0 , otherwise . \partial\left(V\left([2^2]\right)_p,V\left([3^4]\right)_s\right)=
\begin{cases}
1, & |p-s|=5,\text{ or }9,\\
2, & |p-s|=7,\\
0, & \text{otherwise}.
\end{cases} ∂ ( V ( [ 2 2 ] ) p , V ( [ 3 4 ] ) s ) = ⎩ ⎨ ⎧ 1 , 2 , 0 , ∣ p − s ∣ = 5 , or 9 , ∣ p − s ∣ = 7 , otherwise .
We next give the corresponding closed formulas in type D D D . We use the standard numbering in which n − 1 n-1 n − 1 and n n n are the two spin nodes of D n D_n D n .
For r , m ∈ Z > 0 r,m\in\mathbb{Z}_{>0} r , m ∈ Z > 0 , v ∈ Q v\in\mathbb{Q} v ∈ Q , and d ∈ { 1 , 2 } d\in\{1,2\} d ∈ { 1 , 2 } , define
T r , m ( d ) ( v ) : = ∣ { ( a , b ) ∈ Z ≥ 1 × Z ≥ 0 ∣ 1 ≤ a ≤ r , 0 ≤ b ≤ m − 1 , d a + b = v } ∣ . (49) T_{r,m}^{(d)}(v):=\left|\left\{(a,b)\in\mathbb{Z}_{\geq1}\times\mathbb{Z}_{\geq0}\mid1\leq a\leq r,\ 0\leq b\leq m-1,\ da+b=v\right\}\right|. \tag*{(49)} T r , m ( d ) ( v ) := ∣ { ( a , b ) ∈ Z ≥ 1 × Z ≥ 0 ∣ 1 ≤ a ≤ r , 0 ≤ b ≤ m − 1 , d a + b = v } ∣ . (49)
Equivalently,
T r , m ( 1 ) ( v ) : = { [ min ( v , r , m , r + m − v ) ] + , v ∈ Z , 0 , v ∉ Z . T_{r,m}^{(1)}(v):=
\begin{cases}
\left[\min(v,r,m,r+m-v)\right]_+, & v\in\mathbb{Z},\\
0, & v\notin\mathbb{Z}.
\end{cases} T r , m ( 1 ) ( v ) := { [ min ( v , r , m , r + m − v ) ] + , 0 , v ∈ Z , v ∈ / Z .
and
T r , m ( 2 ) ( v ) : = { [ ⌊ min ( r , v 2 ) ⌋ − max ( 1 , ⌈ v − m + 1 2 ⌉ ) + 1 ] + , v ∈ Z , 0 , v ∉ Z . T_{r,m}^{(2)}(v):=
\begin{cases}
\left[\left\lfloor\min\left(r,\frac{v}{2}\right)\right\rfloor-\max\left(1,\left\lceil\frac{v-m+1}{2}\right\rceil\right)+1\right]_+, & v\in\mathbb{Z},\\
0, & v\notin\mathbb{Z}.
\end{cases} T r , m ( 2 ) ( v ) := { [ ⌊ min ( r , 2 v ) ⌋ − max ( 1 , ⌈ 2 v − m + 1 ⌉ ) + 1 ] + , 0 , v ∈ Z , v ∈ / Z .
Corollary 4.16. Let g \mathfrak{g} g be of type D n D_n D n . For i , j ∈ I i,j\in I i , j ∈ I , k , ℓ ∈ Z > 0 k,\ell\in\mathbb{Z}_{>0} k , ℓ ∈ Z > 0 , and p , s ∈ Z p,s\in\mathbb{Z} p , s ∈ Z , set
L : = min ( k , ℓ ) , u : = ∣ p − s ∣ , κ : = ∣ k − ℓ ∣ . L:=\min(k,\ell),\qquad u:=|p-s|,\qquad\kappa:=|k-\ell|. L := min ( k , ℓ ) , u := ∣ p − s ∣ , κ := ∣ k − ℓ ∣. Then the following formulas hold.
(i) Suppose that 1 ≤ i , j ≤ n − 2 1\leq i,j\leq n-2 1 ≤ i , j ≤ n − 2 . Set
r : = min ( i , j ) , ν 1 : = u − κ − ∣ i − j ∣ 2 , ν 2 : = u − κ − ( 2 n − i − j − 2 ) 2 . r:=\min(i,j),\qquad\nu_1:=\frac{u-\kappa-|i-j|}{2},\qquad\nu_2:=\frac{u-\kappa-(2n-i-j-2)}{2}. r := min ( i , j ) , ν 1 := 2 u − κ − ∣ i − j ∣ , ν 2 := 2 u − κ − ( 2 n − i − j − 2 ) . Then we have
∂ ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = T r , L ( 1 ) ( ν 1 ) + T r , L ( 1 ) ( ν 2 ) . \partial\left(V\left([i^k]\right)_p,V\left([j^\ell]\right)_s\right)=T_{r,L}^{(1)}(\nu_1)+T_{r,L}^{(1)}(\nu_2). ∂ ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = T r , L ( 1 ) ( ν 1 ) + T r , L ( 1 ) ( ν 2 ) . (ii) Suppose that exactly one of i , j i,j i , j is a spin node. Let
c : = min ( i , j , n − 2 ) , c:=\min(i,j,n-2), c := min ( i , j , n − 2 ) , so that c c c is the non-spin node, and set
ν : = u − κ − ( n − c − 1 ) 2 . \nu:=\frac{u-\kappa-(n-c-1)}{2}. ν := 2 u − κ − ( n − c − 1 ) . Then we have
∂ ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = T c , L ( 1 ) ( ν ) . \partial\left(V\left([i^k]\right)_p,V\left([j^\ell]\right)_s\right)=T_{c,L}^{(1)}(\nu). ∂ ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = T c , L ( 1 ) ( ν ) . (iii) Suppose that i = j ∈ { n − 1 , n } i=j\in\{n-1,n\} i = j ∈ { n − 1 , n } . Set
r s p : = ⌊ n 2 ⌋ and ν : = u − κ + 2 2 . r_{\mathrm{sp}}:=\left\lfloor\frac{n}{2}\right\rfloor\qquad\text{and}\qquad\nu:=\frac{u-\kappa+2}{2}. r sp := ⌊ 2 n ⌋ and ν := 2 u − κ + 2 . Then we have
∂ ( V ( [ i k ] ) p , V ( [ ℓ ℓ ] ) s ) = T r s p , L ( 2 ) ( ν ) . \partial\left(V\left([i^k]\right)_p,V\left([\ell^\ell]\right)_s\right)=T_{r_{\mathrm{sp}},L}^{(2)}(\nu). ∂ ( V ( [ i k ] ) p , V ( [ ℓ ℓ ] ) s ) = T r sp , L ( 2 ) ( ν ) . :::`
(iv) Suppose that { i , j } = { n − 1 , n } \{i,j\}=\{n-1,n\} { i , j } = { n − 1 , n } . Set
r o p : = ⌊ n − 1 2 ⌋ and ν : = u − κ 2 . r_{\mathrm{op}}:=\left\lfloor\frac{n-1}{2}\right\rfloor\quad\text{and}\quad\nu:=\frac{u-\kappa}{2}. r op := ⌊ 2 n − 1 ⌋ and ν := 2 u − κ .
Then we have
d ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = Γ r o p , L ( 2 ) ( ν ) . \mathfrak{d}\bigl(\mathsf{V}([i^{k}])_{p},\mathsf{V}([j^{\ell}])_{s}\bigr)=\Gamma^{(2)}_{r_{\mathrm{op}},L}(\nu). d ( V ([ i k ] ) p , V ([ j ℓ ] ) s ) = Γ r op , L ( 2 ) ( ν ) .
Proof. By [42 Theorem 4.12(2)] , the coefficients of C ~ i , j [ h ] ( t ) \widetilde{C}^{[h]}_{i,j}(t) C i , j [ h ] ( t ) in type D n D_n D n are given as follows. Set
i ˉ : = min ( i , n − 1 ) and j ˉ : = min ( j , n − 1 ) . \bar{i}:=\min(i,n-1)\quad\text{and}\quad\bar{j}:=\min(j,n-1). i ˉ := min ( i , n − 1 ) and j ˉ := min ( j , n − 1 ) .
Then
∑ v C ~ i , j [ h ] ( v ) t v = { ∑ a = 1 min ( i ˉ , j ˉ ) ( t ∣ i ˉ − j ˉ ∣ + 2 a − 1 + δ ( max ( i ˉ , j ˉ ) < n − 1 ) t 2 n − i ˉ − j ˉ + 2 a − 3 ) , min ( i ˉ , j ˉ ) < n − 1 , ∑ a = 1 ⌊ ( n − 1 + δ i , j ) / 2 ⌋ t 4 a − 1 − 2 δ i , j , min ( i ˉ , j ˉ ) ≥ n − 1. \sum_v\widetilde{C}^{[h]}_{i,j}(v)t^v=
\begin{cases}
\sum_{a=1}^{\min(\bar{i},\bar{j})}\left(t^{|\bar{i}-\bar{j}|+2a-1}+\delta(\max(\bar{i},\bar{j})<n-1)t^{2n-\bar{i}-\bar{j}+2a-3}\right),&\min(\bar{i},\bar{j})<n-1,\\[1.2em]
\sum_{a=1}^{\lfloor(n-1+\delta_{i,j})/2\rfloor}t^{4a-1-2\delta_{i,j}},&\min(\bar{i},\bar{j})\ge n-1.
\end{cases} v ∑ C i , j [ h ] ( v ) t v = ⎩ ⎨ ⎧ ∑ a = 1 m i n ( i ˉ , j ˉ ) ( t ∣ i ˉ − j ˉ ∣ + 2 a − 1 + δ ( max ( i ˉ , j ˉ ) < n − 1 ) t 2 n − i ˉ − j ˉ + 2 a − 3 ) , ∑ a = 1 ⌊( n − 1 + δ i , j ) /2 ⌋ t 4 a − 1 − 2 δ i , j , min ( i ˉ , j ˉ ) < n − 1 , min ( i ˉ , j ˉ ) ≥ n − 1.
Substituting these coefficients into (46) and using (49) gives the four formulas. □
Remark 4.17. Theorem 4.12 also gives explicit finite formulas in types E 6 E_6 E 6 , E 7 E_7 E 7 , and E 8 E_8 E 8 . We do not list all exceptional cases, since they are obtained directly from the finitely many coefficients
C ~ i , j [ h ] ( v ) ( i , j ∈ I , 1 ≤ v ≤ h ) . \widetilde{C}^{[h]}_{i,j}(v)\qquad(i,j\in I,\ 1\le v\le h). C i , j [ h ] ( v ) ( i , j ∈ I , 1 ≤ v ≤ h ) .
The higher root multiplicities in the exceptional denominator formulas are understood according to the formulas in [51 , 52 ] . For example, for the normalized R R R -matrix denominator d 4 , 4 ( z ) d_{4,4}(z) d 4 , 4 ( z ) between two color-4 fundamental quantum affine modules of type E 6 E_6 E 6 , with the numbering used in [52 Proposition 2.1] , one has
d 4 , 4 ( z ) = ( z − q 2 ) ( z − q 4 ) 2 ( z − q 6 ) 3 ( z − q 8 ) 3 ( z − q 10 ) 2 ( z − q 12 ) . d_{4,4}(z)=(z-q^2)(z-q^4)^2(z-q^6)^3(z-q^8)^3(z-q^{10})^2(z-q^{12}). d 4 , 4 ( z ) = ( z − q 2 ) ( z − q 4 ) 2 ( z − q 6 ) 3 ( z − q 8 ) 3 ( z − q 10 ) 2 ( z − q 12 ) .
In this example, the multiplicity of each root q a q^a q a in d 4 , 4 ( z ) d_{4,4}(z) d 4 , 4 ( z ) is c ~ 4 , 4 ( a − 1 ) \widetilde{c}_{4,4}(a-1) c 4 , 4 ( a − 1 ) . Since the Coxeter number of E 6 E_6 E 6 is 12, the corresponding truncation of C ~ 4 , 4 ( t ) \widetilde{C}_{4,4}(t) C 4 , 4 ( t ) through degree 12 is
C ~ 4 , 4 [ 12 ] ( t ) = t + 2 t 3 + 3 t 5 + 3 t 7 + 2 t 9 + t 11 . \widetilde{C}^{[12]}_{4,4}(t)=t+2t^3+3t^5+3t^7+2t^9+t^{11}. C 4 , 4 [ 12 ] ( t ) = t + 2 t 3 + 3 t 5 + 3 t 7 + 2 t 9 + t 11 .
Since c ~ 4 , 4 [ 12 ] ( 5 ) = 3 \widetilde{c}^{[12]}_{4,4}(5)=3 c 4 , 4 [ 12 ] ( 5 ) = 3 , (45) gives
d ( V ( [ 4 ] ) p , V ( [ 4 ] ) s ) = 3 if ∣ p − s ∣ = 6. \mathfrak{d}\bigl(\mathsf{V}([4])_{p},\mathsf{V}([4])_{s}\bigr)=3\qquad\text{if}\qquad|p-s|=6. d ( V ([ 4 ] ) p , V ([ 4 ] ) s ) = 3 if ∣ p − s ∣ = 6.
In particular, the multiplicity 3 of the root q 6 q^6 q 6 agrees with the symmetrized inverse t t t -Cartan coefficient prescribed by Theorem 4.12.
For arbitrary i , j ∈ I i,j\in I i , j ∈ I and k , ℓ ∈ Z > 0 k,\ell\in\mathbb{Z}_{>0} k , ℓ ∈ Z > 0 , the exceptional d \mathfrak{d} d -values are computed uniformly by (45) , with K = max ( k , ℓ ) K=\max(k,\ell) K = max ( k , ℓ ) , L = min ( k , ℓ ) L=\min(k,\ell) L = min ( k , ℓ ) , and u = ∣ p − s ∣ u=|p-s| u = ∣ p − s ∣ .
We now return to KR-modules over quiver Hecke algebras of non-simply-laced finite type. Keep the finite-type folding datum
( g ^ , Γ ) ⟶ g (\widehat{\mathfrak{g}},\Gamma)\longrightarrow\mathfrak{g} ( g , Γ ) ⟶ g
of § 3 , where g ^ \widehat{\mathfrak{g}} g is of simply-laced finite type and g \mathfrak{g} g is of finite type B B B , C C C , F F F , or G G G ((2.2)).
Let Q ^ \widehat{Q} Q be a Γ \Gamma Γ -foldable simply-laced Dynkin quiver and set
Q : = Q ^ / Γ . Q := \widehat{Q}/\Gamma. Q := Q /Γ.
Keep the notation of Lemma 4.4 . In particular, the resulting vertex fibers induce the chain-dependent box fibers used in Theorem 3.16 and Corollary 3.17 .
The Coxeter numbers of g ^ \widehat{\mathfrak{g}} g and g \mathfrak{g} g have the same value h h h . We set
C ~ i , j g , [ h ] ( t ) : = ∑ v = 1 h c ~ i , j g ( v ) t v . \widetilde{C}^{\mathfrak{g},[h]}_{i,j}(t) := \sum_{v=1}^{h} \widetilde{c}^{\mathfrak{g}}_{i,j}(v)t^v. C i , j g , [ h ] ( t ) := v = 1 ∑ h c i , j g ( v ) t v .
For ı , ȷ ∈ I ^ \imath,\jmath\in\widehat{I} , ∈ I , we likewise set
C ~ ı , ȷ g ^ , [ h ] ( t ) : = ∑ v = 1 h c ~ ı , ȷ g ^ ( v ) t v . \widetilde{C}^{\widehat{\mathfrak{g}},[h]}_{\imath,\jmath}(t) := \sum_{v=1}^{h} \widetilde{c}^{\widehat{\mathfrak{g}}}_{\imath,\jmath}(v)t^v. C , g , [ h ] ( t ) := v = 1 ∑ h c , g ( v ) t v .
For these standard finite-type foldings one has m = ∣ Γ ∣ m=|\Gamma| m = ∣Γ∣ , so the group-sum form of Theorem 4.8 has no additional scalar factor.
Theorem 4.18. Assume that g \mathfrak{g} g is of finite type, and let h h h be its Coxeter number. Let i , j ∈ I i,j\in I i , j ∈ I , k , ℓ ∈ Z > 0 k,\ell\in\mathbb{Z}_{>0} k , ℓ ∈ Z > 0 , and p , s ∈ Z p,s\in\mathbb{Z} p , s ∈ Z . Suppose that all vertices
( i , p + k − 1 − 2 a ) ( 0 ≤ a < k ) , ( j , s + ℓ − 1 − 2 b ) ( 0 ≤ b < ℓ ) (i,p+k-1-2a)\quad(0\leq a<k),\qquad(j,s+\ell-1-2b)\quad(0\leq b<\ell) ( i , p + k − 1 − 2 a ) ( 0 ≤ a < k ) , ( j , s + ℓ − 1 − 2 b ) ( 0 ≤ b < ℓ ) belong to Υ Q \Upsilon_Q Υ Q . Set
K : = max ( k , ℓ ) , L : = min ( k , ℓ ) , u : = ∣ p − s ∣ . K:=\max(k,\ell),\qquad L:=\min(k,\ell),\qquad u:=|p-s|. K := max ( k , ℓ ) , L := min ( k , ℓ ) , u := ∣ p − s ∣. Then
∂ ( V ( i [ k ] ) p , V ( j [ ℓ ] ) s ) = [ t u ] ( t K [ L ] t C ~ i , j g , [ h ] ( t ) ) = ∑ a = 0 L − 1 c ~ i , j g , [ h ] ( u − ∣ k − ℓ ∣ − 2 a − 1 ) . (50) \partial\bigl(V({}^{[k]}_{i})_p,V({}^{[\ell]}_{j})_s\bigr)
=[t^u]\left(t^K[L]_t\widetilde{C}^{\mathfrak{g},[h]}_{i,j}(t)\right)
=\sum_{a=0}^{L-1}\widetilde{c}^{\mathfrak{g},[h]}_{i,j}(u-|k-\ell|-2a-1). \tag*{(50)} ∂ ( V ( i [ k ] ) p , V ( j [ ℓ ] ) s ) = [ t u ] ( t K [ L ] t C i , j g , [ h ] ( t ) ) = a = 0 ∑ L − 1 c i , j g , [ h ] ( u − ∣ k − ℓ ∣ − 2 a − 1 ) . (50) where
c ~ i , j g , [ h ] ( v ) : = { c i , j g ( v ) , 1 ≤ v ≤ h , 0 , otherwise . \widetilde{c}^{\mathfrak{g},[h]}_{i,j}(v):=
\begin{cases}
c^{\mathfrak{g}}_{i,j}(v), & 1\leq v\leq h,\\
0, & \text{otherwise}.
\end{cases} c i , j g , [ h ] ( v ) := { c i , j g ( v ) , 0 , 1 ≤ v ≤ h , otherwise .
Proof. In simply-laced type, the assertion follows from Theorem 4.12 and generalized quantum affine Schur–Weyl duality. For the remaining types, use the standard finite-type folding g ^ → g \widehat{\mathfrak{g}}\to\mathfrak{g} g → g , for which the Coxeter numbers agree.
Let c , c ′ \mathfrak{c},\mathfrak{c}' c , c ′ be the boxes corresponding to V ( i [ k ] ) p , V ( j [ ℓ ] ) s V({}^{[k]}_{i})_p,V({}^{[\ell]}_{j})_s V ( i [ k ] ) p , V ( j [ ℓ ] ) s , respectively. Choose admissible chains C , D \mathcal{C},\mathcal{D} C , D containing these boxes, and choose c ^ ′ ∈ Fib D ( c ′ ) \widehat{\mathfrak{c}}'\in\operatorname{Fib}_{\mathcal{D}}(\mathfrak{c}') c ′ ∈ Fib D ( c ′ ) of color j ∈ I j j\in I_j j ∈ I j . By Corollary 3.17 ,
∂ ( M i [ c ] , M i [ c ′ ] ) = d i ∑ c ^ ∈ Fib C ( c ) ∂ ( M i ^ [ c ^ ] , M i ^ [ c ^ ′ ] ) . \partial(M^i[\mathfrak{c}],M^i[\mathfrak{c}'])
=d_i\sum_{\widehat{\mathfrak{c}}\in\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c})}
\partial(M^{\widehat{i}}[\widehat{\mathfrak{c}}],M^{\widehat{i}}[\widehat{\mathfrak{c}}']). ∂ ( M i [ c ] , M i [ c ′ ]) = d i c ∈ Fib C ( c ) ∑ ∂ ( M i [ c ] , M i [ c ′ ]) .
By Lemma 4.5 , the modules M i ^ [ c ^ ] M^{\widehat{i}}[\widehat{\mathfrak{c}}] M i [ c ] with c ^ ∈ Fib C ( c ) \widehat{\mathfrak{c}}\in\operatorname{Fib}_{\mathcal{C}}(\mathfrak{c}) c ∈ Fib C ( c ) have parameters k , p k,p k , p , and M i ^ [ c ^ ′ ] M^{\widehat{i}}[\widehat{\mathfrak{c}}'] M i [ c ′ ] has parameters ℓ , s \ell,s ℓ , s . All their defining vertices lie in Υ Q ^ \Upsilon_{\widehat{Q}} Υ Q . Their colors run through I i I_i I i and the fixed color j j j , respectively. Applying the simply-laced formula and Theorem 4.8 , we obtain
∂ ( V ( i [ k ] ) p , V ( j [ ℓ ] ) s ) = [ t u ] ( t K [ L ] t d i ∑ ı ∈ I i C ~ ı , j g ^ , [ h ] ( t ) ) = [ t u ] ( t K [ L ] t C ~ i , j g , [ h ] ( t ) ) . \partial\bigl(V({}^{[k]}_{i})_p,V({}^{[\ell]}_{j})_s\bigr)
=[t^u]\left(t^K[L]_t d_i\sum_{\imath\in I_i}\widetilde{C}^{\widehat{\mathfrak{g}},[h]}_{\imath,j}(t)\right)
=[t^u]\left(t^K[L]_t\widetilde{C}^{\mathfrak{g},[h]}_{i,j}(t)\right). ∂ ( V ( i [ k ] ) p , V ( j [ ℓ ] ) s ) = [ t u ] ( t K [ L ] t d i ∈ I i ∑ C , j g , [ h ] ( t ) ) = [ t u ] ( t K [ L ] t C i , j g , [ h ] ( t ) ) .
Finally,
t K [ L ] t = ∑ a = 0 L − 1 t ∣ k − ℓ ∣ + 2 a + 1 , t^{K}[L]_{t}=\sum_{a=0}^{L-1}t^{|k-\ell|+2a+1}, t K [ L ] t = a = 0 ∑ L − 1 t ∣ k − ℓ ∣ + 2 a + 1 ,
which gives the coefficient sum.
Example 4.19. Return to the folding A 5 → C 3 A_{5}\to C_{3} A 5 → C 3 of Examples 2.7 and 4.7. For the reduced sequence i \boldsymbol{i} i in the first example, the modules
M = V ( [ 2 2 ] ) 3 = M i [ 2 , 5 ] , N = V ( [ 3 2 ] ) 6 = M i [ 7 , 9 ] M=V\left([2^{2}]\right)_{3}=M^{\boldsymbol{i}}[2,5],\qquad N=V\left([3^{2}]\right)_{6}=M^{\boldsymbol{i}}[7,9] M = V ( [ 2 2 ] ) 3 = M i [ 2 , 5 ] , N = V ( [ 3 2 ] ) 6 = M i [ 7 , 9 ] have defining vertices ( 2 , 2 ) , ( 2 , 4 ) (2,2),(2,4) ( 2 , 2 ) , ( 2 , 4 ) and ( 3 , 5 ) , ( 3 , 7 ) (3,5),(3,7) ( 3 , 5 ) , ( 3 , 7 ) in the same Υ Q \Upsilon_{Q} Υ Q ; see also Example 4.2 (ii). Here K = L = 2 K=L=2 K = L = 2 , u = 3 u=3 u = 3 , I 2 = { 2 , 4 } I_{2}=\{2,4\} I 2 = { 2 , 4 } , I 3 = { 3 } I_{3}=\{3\} I 3 = { 3 } , and d 2 = 1 d_{2}=1 d 2 = 1 . Theorem 4.8 gives
c ~ 2 , 3 C 3 ( 2 ) = c ~ 2 , 3 A 5 ( 2 ) + c ~ 4 , 3 A 5 ( 2 ) = 2. \widetilde{c}^{C_{3}}_{2,3}(2)=\widetilde{c}^{A_{5}}_{2,3}(2)+\widetilde{c}^{A_{5}}_{4,3}(2)=2. c 2 , 3 C 3 ( 2 ) = c 2 , 3 A 5 ( 2 ) + c 4 , 3 A 5 ( 2 ) = 2. Since h = 6 h=6 h = 6 , we have
c ~ i , j [ 6 ] ( v ) = c ~ i , j ( v ) ( v = 0 , 2 ) , \widetilde{c}^{[6]}_{i,j}(v)=\widetilde{c}_{i,j}(v)\qquad(v=0,2), c i , j [ 6 ] ( v ) = c i , j ( v ) ( v = 0 , 2 ) , where both coefficients at v = 0 v=0 v = 0 vanish. Thus Theorem 4.18 gives
d ( M , N ) = c ~ 2 , 3 C 3 , [ 6 ] ( 2 ) + c ~ 2 , 3 C 3 , [ 6 ] ( 0 ) = 2 , \mathfrak{d}(M,N)=\widetilde{c}^{C_{3},[6]}_{2,3}(2)+\widetilde{c}^{C_{3},[6]}_{2,3}(0)=2, d ( M , N ) = c 2 , 3 C 3 , [ 6 ] ( 2 ) + c 2 , 3 C 3 , [ 6 ] ( 0 ) = 2 , in agreement with the categorical calculation following Corollary 3.17.
Recall the counting functions T r , L ( d ) ( ν ) T^{(d)}_{r,L}(\nu) T r , L ( d ) ( ν ) from (4.10).
Corollary 4.20. Let g \mathfrak{g} g be of type C n C_{n} C n , with n n n the long node. Keep the notation
L : = min ( k , ℓ ) , u : = ∣ p − s ∣ , κ : = ∣ k − ℓ ∣ . L:=\min(k,\ell),\qquad u:=|p-s|,\qquad\kappa:=|k-\ell|. L := min ( k , ℓ ) , u := ∣ p − s ∣ , κ := ∣ k − ℓ ∣. Assume that all the vertices defining the two quiver Hecke KR-modules in question lie in Υ Q \Upsilon_{Q} Υ Q .
(i) If 1 ≤ i , j < n 1\le i,j<n 1 ≤ i , j < n , set
ν 1 : = u − κ − ∣ i − j ∣ 2 and ν 2 : = u − κ − ( 2 n − i − j ) 2 . \nu_{1}:=\frac{u-\kappa-|i-j|}{2}\qquad\text{and}\qquad\nu_{2}:=\frac{u-\kappa-(2n-i-j)}{2}. ν 1 := 2 u − κ − ∣ i − j ∣ and ν 2 := 2 u − κ − ( 2 n − i − j ) . Then we have
d ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = T ρ i , j ( 2 n − 1 ) , L ( 1 ) ( ν 1 ) + T ρ 2 n − i , j ( 2 n − 1 ) , L ( 1 ) ( ν 2 ) . \mathfrak{d}\left(V\left(\left[i^{k}\right]\right)_{p},V\left(\left[j^{\ell}\right]\right)_{s}\right)
=T^{(1)}_{\rho_{i,j}^{(2n-1)},L}(\nu_{1})+
T^{(1)}_{\rho_{2n-i,j}^{(2n-1)},L}(\nu_{2}). d ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = T ρ i , j ( 2 n − 1 ) , L ( 1 ) ( ν 1 ) + T ρ 2 n − i , j ( 2 n − 1 ) , L ( 1 ) ( ν 2 ) . (ii) If i = n i=n i = n or j = n j=n j = n , set
ν : = u − κ − ∣ i − j ∣ 2 . \nu:=\frac{u-\kappa-|i-j|}{2}. ν := 2 u − κ − ∣ i − j ∣ . Then we have
d ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = 2 T ρ i , j ( 2 n − 1 ) , L ( 1 ) ( ν ) . \mathfrak{d}\left(V\left(\left[i^{k}\right]\right)_{p},V\left(\left[j^{\ell}\right]\right)_{s}\right)
=2T^{(1)}_{\rho_{i,j}^{(2n-1)},L}(\nu). d ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = 2 T ρ i , j ( 2 n − 1 ) , L ( 1 ) ( ν ) . Proof. Consider the standard folding
A 2 n − 1 ⟶ C n , I i = { i , 2 n − i } ( 1 ≤ i < n ) , I n = { n } . A_{2n-1}\longrightarrow C_{n},\qquad I_{i}=\{i,2n-i\}\quad(1\le i<n),\qquad I_{n}=\{n\}. A 2 n − 1 ⟶ C n , I i = { i , 2 n − i } ( 1 ≤ i < n ) , I n = { n } . Let M M M and N N N denote the two downstairs KR-modules in the statement. For ı ∈ I i \imath\in I_{i} ∈ I i and ȷ ∈ I j \jmath\in I_{j} ∈ I j , denote their corresponding lifted KR-modules by M ^ ı \widehat{M}_{\imath} M and N ^ ȷ \widehat{N}_{\jmath} N , respectively. By Lemma 4.5, M ^ ı ≅ V ( [ i k ] ) p \widehat{M}_{\imath}\cong V\left(\left[i^{k}\right]\right)_{p} M ≅ V ( [ i k ] ) p and N ^ ȷ ≅ V ( [ j ℓ ] ) s \widehat{N}_{\jmath}\cong V\left(\left[j^{\ell}\right]\right)_{s} N ≅ V ( [ j ℓ ] ) s .
Suppose first that i , j < n i,j<n i , j < n . Identify the upstairs vertices with the standard numeric labels of A 2 n − 1 A_{2n-1} A 2 n − 1 , and choose the lifts ı = i ∈ I i \imath=i\in I_i = i ∈ I i and ȷ = j ∈ I j \jmath=j\in I_j = j ∈ I j . Thus the other lift of i i i is 2 n − i 2n-i 2 n − i . Since d i = 1 d_i=1 d i = 1 , Corollary 3.17 gives
d ( M , N ) = d ( M ^ ı , N ^ ȷ ) + d ( M ^ 2 n − i , N ^ ȷ ) . \mathfrak{d}(M,N)=\mathfrak{d}(\widehat{M}_{\imath},\widehat{N}_{\jmath})+\mathfrak{d}(\widehat{M}_{2n-i},\widehat{N}_{\jmath}). d ( M , N ) = d ( M , N ) + d ( M 2 n − i , N ) .
For the two pairs of upstairs colors, we have
∣ ı − ȷ ∣ = ∣ i − j ∣ and ∣ ( 2 n − i ) − ȷ ∣ = 2 n − i − j , |\imath-\jmath|=|i-j|\quad\text{and}\quad|(2n-i)-\jmath|=2n-i-j, ∣ − ∣ = ∣ i − j ∣ and ∣ ( 2 n − i ) − ∣ = 2 n − i − j ,
and
ρ i , j ( 2 n − 1 ) = ρ 2 n − i , j ( 2 n − 1 ) = min ( i , j ) . \rho_{i,j}^{(2n-1)}=\rho_{2n-i,j}^{(2n-1)}=\min(i,j). ρ i , j ( 2 n − 1 ) = ρ 2 n − i , j ( 2 n − 1 ) = min ( i , j ) .
Applying Corollary 4.14 to the two summands yields
d ( M , N ) = T ρ i , j ( 2 n − 1 ) , L ( 1 ) ( u − ∣ i − j ∣ − ∣ k − ℓ ∣ 2 ) + T ρ 2 n − i , j ( 2 n − 1 ) , L ( 1 ) ( u − ( 2 n − i − j ) − ∣ k − ℓ ∣ 2 ) , \mathfrak{d}(M,N)=\mathrm{T}_{\rho_{i,j}^{(2n-1)},L}^{(1)}\left(\frac{u-|i-j|-|k-\ell|}{2}\right)+\mathrm{T}_{\rho_{2n-i,j}^{(2n-1)},L}^{(1)}\left(\frac{u-(2n-i-j)-|k-\ell|}{2}\right), d ( M , N ) = T ρ i , j ( 2 n − 1 ) , L ( 1 ) ( 2 u − ∣ i − j ∣ − ∣ k − ℓ ∣ ) + T ρ 2 n − i , j ( 2 n − 1 ) , L ( 1 ) ( 2 u − ( 2 n − i − j ) − ∣ k − ℓ ∣ ) ,
which is the first formula.
Suppose now that i = n i=n i = n . Since I n = { n } I_n=\{n\} I n = { n } and d n = 2 d_n=2 d n = 2 , Corollary 3.17 gives
d ( M , N ) = 2 d ( M ^ n , N ^ ȷ ) ( ȷ ∈ I j ) . \mathfrak{d}(M,N)=2\mathfrak{d}(\widehat{M}_n,\widehat{N}_{\jmath})\qquad(\jmath\in I_j). d ( M , N ) = 2 d ( M n , N ) ( ∈ I j ) .
The type A 2 n − 1 A_{2n-1} A 2 n − 1 closed formula therefore gives
d ( M , N ) = 2 T ρ i , j ( 2 n − 1 ) , L ( 1 ) ( u − ∣ i − j ∣ − ∣ k − ℓ ∣ 2 ) . \mathfrak{d}(M,N)=2\mathrm{T}_{\rho_{i,j}^{(2n-1)},L}^{(1)}\left(\frac{u-|i-j|-|k-\ell|}{2}\right). d ( M , N ) = 2 T ρ i , j ( 2 n − 1 ) , L ( 1 ) ( 2 u − ∣ i − j ∣ − ∣ k − ℓ ∣ ) .
Finally, suppose that j = n j=n j = n and i < n i<n i < n . Then
d ( M , N ) = d ( M ^ i , N ^ n ) + d ( M ^ 2 n − i , N ^ n ) . \mathfrak{d}(M,N)=\mathfrak{d}(\widehat{M}_i,\widehat{N}_n)+\mathfrak{d}(\widehat{M}_{2n-i},\widehat{N}_n). d ( M , N ) = d ( M i , N n ) + d ( M 2 n − i , N n ) .
The two summands are equal, since
∣ i − n ∣ = ∣ ( 2 n − i ) − n ∣ and ρ i , n ( 2 n − 1 ) = ρ 2 n − i , n ( 2 n − 1 ) = i . |i-n|=|(2n-i)-n|\quad\text{and}\quad\rho_{i,n}^{(2n-1)}=\rho_{2n-i,n}^{(2n-1)}=i. ∣ i − n ∣ = ∣ ( 2 n − i ) − n ∣ and ρ i , n ( 2 n − 1 ) = ρ 2 n − i , n ( 2 n − 1 ) = i .
Consequently,
d ( M , N ) = 2 T ρ i , j ( 2 n − 1 ) , L ( 1 ) ( u − ∣ i − j ∣ − ∣ k − ℓ ∣ 2 ) = 2 T ρ i , j ( 2 n − 1 ) , L ( 1 ) ( ν ) . \mathfrak{d}(M,N)=2\mathrm{T}_{\rho_{i,j}^{(2n-1)},L}^{(1)}\left(\frac{u-|i-j|-|k-\ell|}{2}\right)=2\mathrm{T}_{\rho_{i,j}^{(2n-1)},L}^{(1)}(\nu). d ( M , N ) = 2 T ρ i , j ( 2 n − 1 ) , L ( 1 ) ( 2 u − ∣ i − j ∣ − ∣ k − ℓ ∣ ) = 2 T ρ i , j ( 2 n − 1 ) , L ( 1 ) ( ν ) .
This proves the result. □
Example 2.7 is the case n = 3 n=3 n = 3 , i = 2 i=2 i = 2 , j = 3 j=3 j = 3 , k = ℓ = 2 k=\ell=2 k = ℓ = 2 , and u = 3 u=3 u = 3 of the second formula:
2 T 2 , 2 ( 1 ) ( 1 ) = 2. 2\mathrm{T}_{2,2}^{(1)}(1)=2. 2 T 2 , 2 ( 1 ) ( 1 ) = 2.
Example 4.21. Let g \mathfrak{g} g be of type C 5 C_5 C 5 , with node 5 long, and suppose that all the vertices defining the modules below lie in Υ Q \Upsilon_Q Υ Q . Consider
V ( [ 5 2 ] ) p and V ( [ 5 3 ] ) s V([5^2])_p\quad\text{and}\quad V([5^3])_s V ([ 5 2 ] ) p and V ([ 5 3 ] ) s with ∣ p − s ∣ = 5 |p-s|=5 ∣ p − s ∣ = 5 .
Here L = 2 L=2 L = 2 , κ = 1 \kappa=1 κ = 1 and ν = 5 − 1 2 = 2 \nu=\frac{5-1}{2}=2 ν = 2 5 − 1 = 2 .
Hence
d ( V ( [ 5 2 ] ) p , V ( [ 5 3 ] ) s ) = 2 T 5 , 2 ( 1 ) ( 2 ) = 4. \mathfrak{d}\left(V([5^2])_p,V([5^3])_s\right)=2\mathrm{T}_{5,2}^{(1)}(2)=4. d ( V ([ 5 2 ] ) p , V ([ 5 3 ] ) s ) = 2 T 5 , 2 ( 1 ) ( 2 ) = 4. This value may also be read directly from the folding
A 9 ⟶ C 5 . A_9\longrightarrow C_5. A 9 ⟶ C 5 .
The middle vertex 5 5 5 of A 9 A_9 A 9 is fixed by the diagram involution and lies above the long vertex 5 5 5 of C 5 C_5 C 5 . Thus its fiber has cardinality 1 1 1 , while the corresponding symmetrizer is
The type A 9 A_9 A 9 formula gives
d ( V ( [ 5 2 ] ) p , V ( [ 5 3 ] ) s ) = T 5 , 2 ( 1 ) ( 2 ) = 2 , \mathfrak{d}\bigl(\mathsf{V}([5^2])_p,\mathsf{V}([5^3])_s\bigr)=T_{5,2}^{(1)}(2)=2, d ( V ([ 5 2 ] ) p , V ([ 5 3 ] ) s ) = T 5 , 2 ( 1 ) ( 2 ) = 2 ,
and categorical folding therefore gives the downstairs type C 5 C_5 C 5 value
d ( V ( [ 5 2 ] ) p , V ( [ 5 3 ] ) s ) = 2 ⋅ 2 = 4. \mathfrak{d}\bigl(\mathsf{V}([5^2])_p,\mathsf{V}([5^3])_s\bigr)=2\cdot2=4. d ( V ([ 5 2 ] ) p , V ([ 5 3 ] ) s ) = 2 ⋅ 2 = 4.
The type B n B_n B n formula is obtained in the same way from the standard folding D n + 1 → B n D_{n+1}\to B_n D n + 1 → B n , using Corollary 3.17 and the type D D D closed formula. We therefore omit its proof.
Corollary 4.22. Let g \mathfrak{g} g be of type B n B_n B n , with n n n the short node. Keep the notation
L : = min ( k , ℓ ) , u : = ∣ p − s ∣ , κ : = ∣ k − ℓ ∣ . L:=\min(k,\ell),\qquad u:=|p-s|,\qquad\kappa:=|k-\ell|. L := min ( k , ℓ ) , u := ∣ p − s ∣ , κ := ∣ k − ℓ ∣. Assume that all the vertices defining the two quiver Hecke KR-modules in question lie in Υ Q \Upsilon_Q Υ Q .
(i) If 1 ≤ i , j < n 1\leq i,j<n 1 ≤ i , j < n , set
r : = min ( i , j ) , ν 1 : = u − κ − ∣ i − j ∣ 2 , ν 2 : = u − κ − ( 2 n − i − j ) 2 . r:=\min(i,j),\qquad\nu_1:=\frac{u-\kappa-|i-j|}{2},\qquad\nu_2:=\frac{u-\kappa-(2n-i-j)}{2}. r := min ( i , j ) , ν 1 := 2 u − κ − ∣ i − j ∣ , ν 2 := 2 u − κ − ( 2 n − i − j ) . Then we have
d ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = 2 T r , L ( 1 ) ( ν 1 ) + 2 T r , L ( 1 ) ( ν 2 ) . \mathfrak{d}\bigl(\mathsf{V}([i^k])_p,\mathsf{V}([j^\ell])_s\bigr)=2T_{r,L}^{(1)}(\nu_1)+2T_{r,L}^{(1)}(\nu_2). d ( V ([ i k ] ) p , V ([ j ℓ ] ) s ) = 2 T r , L ( 1 ) ( ν 1 ) + 2 T r , L ( 1 ) ( ν 2 ) . (ii) If exactly one of i , j i,j i , j is n n n , let
c : = min ( i , j ) and ν : = u − κ − ( n − c ) 2 . c:=\min(i,j)\qquad\text{and}\qquad\nu:=\frac{u-\kappa-(n-c)}{2}. c := min ( i , j ) and ν := 2 u − κ − ( n − c ) . Then we have
d ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = 2 T c , L ( 1 ) ( ν ) . \mathfrak{d}\bigl(\mathsf{V}([i^k])_p,\mathsf{V}([j^\ell])_s\bigr)=2T_{c,L}^{(1)}(\nu). d ( V ([ i k ] ) p , V ([ j ℓ ] ) s ) = 2 T c , L ( 1 ) ( ν ) . (iii) Finally, if i = j = n i=j=n i = j = n , set
r 1 : = ⌊ n + 1 2 ⌋ , r 2 : = ⌊ n 2 ⌋ , ν 1 : = u − κ + 2 2 , ν 2 : = u − κ 2 . r_1:=\left\lfloor\frac{n+1}{2}\right\rfloor,\qquad r_2:=\left\lfloor\frac{n}{2}\right\rfloor,\qquad\nu_1:=\frac{u-\kappa+2}{2},\qquad\nu_2:=\frac{u-\kappa}{2}. r 1 := ⌊ 2 n + 1 ⌋ , r 2 := ⌊ 2 n ⌋ , ν 1 := 2 u − κ + 2 , ν 2 := 2 u − κ . Then
d ( V ( [ n k ] ) p , V ( [ n ℓ ] ) s ) = T r 1 , L ( 2 ) ( ν 1 ) + T r 2 , L ( 2 ) ( ν 2 ) . \mathfrak{d}\bigl(\mathsf{V}([n^k])_p,\mathsf{V}([n^\ell])_s\bigr)=T_{r_1,L}^{(2)}(\nu_1)+T_{r_2,L}^{(2)}(\nu_2). d ( V ([ n k ] ) p , V ([ n ℓ ] ) s ) = T r 1 , L ( 2 ) ( ν 1 ) + T r 2 , L ( 2 ) ( ν 2 ) .
Example 4.23. Let g \mathfrak{g} g be of type B 5 B_5 B 5 , with node 5 5 5 short, and suppose that all the vertices defining the modules below lie in Υ Q \Upsilon_Q Υ Q . Consider
V ( [ 2 2 ] ) p and V ( [ 3 3 ] ) s with ∣ p − s ∣ = 6. \mathsf{V}([2^2])_p\quad\text{and}\quad\mathsf{V}([3^3])_s\qquad\text{with}\qquad|p-s|=6. V ([ 2 2 ] ) p and V ([ 3 3 ] ) s with ∣ p − s ∣ = 6. In this case,
L = 2 , κ = 1 , ν 1 = 6 − 1 − ∣ 2 − 3 ∣ 2 = 2 , ν 2 = 6 − 1 − ( 10 − 2 − 3 ) 2 = 0. L=2,\qquad\kappa=1,\qquad\nu_1=\frac{6-1-|2-3|}{2}=2,\qquad\nu_2=\frac{6-1-(10-2-3)}{2}=0. L = 2 , κ = 1 , ν 1 = 2 6 − 1 − ∣2 − 3∣ = 2 , ν 2 = 2 6 − 1 − ( 10 − 2 − 3 ) = 0. Therefore
d ( V ( [ 2 2 ] ) p , V ( [ 3 3 ] ) s ) = 2 T 2 , 2 ( 1 ) ( 2 ) + 2 T 2 , 2 ( 1 ) ( 0 ) = 4. \mathfrak{d}\bigl(\mathsf{V}([2^2])_p,\mathsf{V}([3^3])_s\bigr)=2T_{2,2}^{(1)}(2)+2T_{2,2}^{(1)}(0)=4. d ( V ([ 2 2 ] ) p , V ([ 3 3 ] ) s ) = 2 T 2 , 2 ( 1 ) ( 2 ) + 2 T 2 , 2 ( 1 ) ( 0 ) = 4.
This value is obtained from the folding
D 6 ⟶ B 5 . D_6 \longrightarrow B_5. D 6 ⟶ B 5 .
The vertices 2 2 2 and 3 3 3 are fixed by the involution exchanging the two spin vertices of D 6 D_6 D 6 , and their corresponding symmetrizer is 2 2 2 . The type D 6 D_6 D 6 formula gives
∂ ( V ( 2 2 ) p , V ( 3 3 ) s ) = 2. \partial\left(\mathrm{V}\left(\boxed{2^2}\right)_p,\mathrm{V}\left(\boxed{3^3}\right)_s\right)=2. ∂ ( V ( 2 2 ) p , V ( 3 3 ) s ) = 2.
Consequently, the categorical fiber formula yields the downstairs type B 5 B_5 B 5 value
∂ ( V ( 2 2 ) p , V ( 3 3 ) s ) = 2 ⋅ 2 = 4. \partial\left(\mathrm{V}\left(\boxed{2^2}\right)_p,\mathrm{V}\left(\boxed{3^3}\right)_s\right)=2\cdot2=4. ∂ ( V ( 2 2 ) p , V ( 3 3 ) s ) = 2 ⋅ 2 = 4.
Theorem 4.18 shows that the categorical folding of ∂ \partial ∂ -invariants and the folding of the symmetrized inverse t t t -Cartan matrix coincide. Both formulas contain the factor d i d_i d i .
Example 4.24. Consider type F 4 F_4 F 4 with the convention
C = ( 2 − 1 0 0 − 1 2 − 1 0 0 − 2 2 − 1 0 0 − 1 2 ) , D = diag ( 2 , 2 , 1 , 1 ) . C=\begin{pmatrix}2&-1&0&0\\-1&2&-1&0\\0&-2&2&-1\\0&0&-1&2\end{pmatrix},\qquad D=\operatorname{diag}(2,2,1,1). C = 2 − 1 0 0 − 1 2 − 2 0 0 − 1 2 − 1 0 0 − 1 2 , D = diag ( 2 , 2 , 1 , 1 ) .
Truncating through degree 12 12 12 , we obtain
C ~ 2 , 3 [ 12 ] ( t ) = 2 t 2 + 4 t 4 + 4 t 6 + 4 t 8 + 2 t 10 . \widetilde{C}^{[12]}_{2,3}(t)=2t^2+4t^4+4t^6+4t^8+2t^{10}. C 2 , 3 [ 12 ] ( t ) = 2 t 2 + 4 t 4 + 4 t 6 + 4 t 8 + 2 t 10 .
Suppose that
V ( 2 2 ) p and V ( 3 3 ) s \mathrm{V}\left(\boxed{2^2}\right)_p\quad\text{and}\quad\mathrm{V}\left(\boxed{3^3}\right)_s V ( 2 2 ) p and V ( 3 3 ) s
are defined by i i i -boxes whose defining vertices all lie in Υ Q \Upsilon_Q Υ Q and that
∣ p − s ∣ = 6. \lvert p-s\rvert=6. ∣ p − s ∣ = 6.
Here k = 2 k=2 k = 2 and ℓ = 3 \ell=3 ℓ = 3 , so L = 2 L=2 L = 2 and ∣ k − ℓ ∣ = 1 \lvert k-\ell\rvert=1 ∣ k − ℓ ∣ = 1 . Theorem 4.18 gives
∂ ( V ( 2 2 ) p , V ( 3 3 ) s ) = C ~ 2 , 3 [ 12 ] ( 4 ) + C ~ 2 , 3 [ 12 ] ( 2 ) = 4 + 2 = 6. \partial\left(\mathrm{V}\left(\boxed{2^2}\right)_p,\mathrm{V}\left(\boxed{3^3}\right)_s\right)=\widetilde{C}^{[12]}_{2,3}(4)+\widetilde{C}^{[12]}_{2,3}(2)=4+2=6. ∂ ( V ( 2 2 ) p , V ( 3 3 ) s ) = C 2 , 3 [ 12 ] ( 4 ) + C 2 , 3 [ 12 ] ( 2 ) = 4 + 2 = 6.
Under the folding E 6 ⟶ F 4 E_6\longrightarrow F_4 E 6 ⟶ F 4 , the fibers are I 1 = { 2 } I_1=\{2\} I 1 = { 2 } , I 2 = { 4 } I_2=\{4\} I 2 = { 4 } , I 3 = { 3 , 5 } I_3=\{3,5\} I 3 = { 3 , 5 } , I 4 = { 1 , 6 } I_4=\{1,6\} I 4 = { 1 , 6 } . Thus this equality is obtained from the singleton vertex fiber over 2 2 2 , with folding factor d 2 = 2 d_2=2 d 2 = 2 , by summing the corresponding type E 6 E_6 E 6 KR-module ∂ \partial ∂ -values. Thus it is the higher KR analogue of the cuspidal folding formula.
Example 4.25. Consider type G 2 G_2 G 2 with the convention
C = ( 2 − 3 − 1 2 ) , D = diag ( 1 , 3 ) . C=\begin{pmatrix}2&-3\\-1&2\end{pmatrix},\qquad D=\operatorname{diag}(1,3). C = ( 2 − 1 − 3 2 ) , D = diag ( 1 , 3 ) .
The truncation through degree 6 6 6 of the off-diagonal entry of the symmetrized inverse t t t -Cartan matrix is
C ~ 1 , 2 [ 6 ] ( t ) = 3 t 2 + 3 t 4 . \widetilde{C}^{[6]}_{1,2}(t)=3t^2+3t^4. C 1 , 2 [ 6 ] ( t ) = 3 t 2 + 3 t 4 .
Suppose that
V ( 1 2 ) p and V ( 2 2 ) s \mathrm{V}\left(\boxed{1^2}\right)_p\quad\text{and}\quad\mathrm{V}\left(\boxed{2^2}\right)_s V ( 1 2 ) p and V ( 2 2 ) s
are defined by i i i -boxes whose defining vertices all lie in Υ Q \Upsilon_Q Υ Q and that ∣ p − s ∣ = 3 \lvert p-s\rvert=3 ∣ p − s ∣ = 3 .
Since L = 2 L=2 L = 2 and ∣ 2 − 2 ∣ = 0 \lvert2-2\rvert=0 ∣ 2 − 2 ∣ = 0 , Theorem 4.18 yields
∂ ( V ( 1 2 ) p , V ( 2 2 ) s ) = C ~ 1 , 2 [ 6 ] ( 2 ) + C ~ 1 , 2 [ 6 ] ( 0 ) = 3. \partial\left(\mathrm{V}\left(\boxed{1^2}\right)_p,\mathrm{V}\left(\boxed{2^2}\right)_s\right)=\widetilde{C}^{[6]}_{1,2}(2)+\widetilde{C}^{[6]}_{1,2}(0)=3. ∂ ( V ( 1 2 ) p , V ( 2 2 ) s ) = C 1 , 2 [ 6 ] ( 2 ) + C 1 , 2 [ 6 ] ( 0 ) = 3.
This value is obtained from the folding
D 4 ⟶ G 2 D_{4} \longrightarrow G_{2} D 4 ⟶ G 2
by summing the corresponding type D 4 D_{4} D 4 KR-module d \mathfrak{d} d -values over the three outer vertices above color 1, while keeping the central vertex above color 2 fixed. Here d 1 = 1 d_{1}=1 d 1 = 1 .
Remark 4.26. For two quiver Hecke KR-modules M , N M,N M , N defined by boxes in the same Q Q Q -adapted reduced sequence, Proposition 3.3 computes at least one of Λ ( M , N ) \Lambda(M,N) Λ ( M , N ) and Λ ( N , M ) \Lambda(N,M) Λ ( N , M ) , as explained in Remark 3.4. (50) then determines the other through
Λ ( M , N ) + Λ ( N , M ) = 2 d ( M , N ) . \Lambda(M,N)+\Lambda(N,M)=2\mathfrak{d}(M,N). Λ ( M , N ) + Λ ( N , M ) = 2 d ( M , N ) .
Example 4.27. In type C 5 C_{5} C 5 , take ξ i = i − 1 \xi_i=i-1 ξ i = i − 1 and the Q Q Q -adapted reduced sequence i = ( 1 , 2 , 3 , 4 , 5 ) 5 \mathbf{i}=(1,2,3,4,5)^{5} i = ( 1 , 2 , 3 , 4 , 5 ) 5 . We use the standard realization of C 5 C_{5} C 5 in R 5 \mathbb{R}^{5} R 5 with orthonormal basis e 1 , … , e 5 \mathbf{e}_{1},\ldots,\mathbf{e}_{5} e 1 , … , e 5 . The positions 4, 9, 14 have color 4, with coordinates 3, 5, 7 in the repetition quiver and root labels
β 4 i = e 1 − e 5 , β 9 i = e 1 + e 2 , β 14 i = e 2 + e 3 . \beta^{\mathbf{i}}_{4}=\mathbf{e}_{1}-\mathbf{e}_{5},\qquad
\beta^{\mathbf{i}}_{9}=\mathbf{e}_{1}+\mathbf{e}_{2},\qquad
\beta^{\mathbf{i}}_{14}=\mathbf{e}_{2}+\mathbf{e}_{3}. β 4 i = e 1 − e 5 , β 9 i = e 1 + e 2 , β 14 i = e 2 + e 3 .
Set
M = V ( [ 4 2 ] 6 ) = M i [ 9 , 14 ] , N = V ( [ 4 2 ] 4 ) = M i [ 4 , 9 ] . M=V\left(\begin{bmatrix}4^{2}\end{bmatrix}_{6}\right)=M^{\mathbf{i}}[9,14],\qquad
N=V\left(\begin{bmatrix}4^{2}\end{bmatrix}_{4}\right)=M^{\mathbf{i}}[4,9]. M = V ( [ 4 2 ] 6 ) = M i [ 9 , 14 ] , N = V ( [ 4 2 ] 4 ) = M i [ 4 , 9 ] .
Since 9 > 4 − = − ∞ 9>4^{-}=-\infty 9 > 4 − = − ∞ , Proposition 3.3 applies to ( M , N ) (M,N) ( M , N ) . The two boxes share position 9, and λ i ( ε 9 , ε 9 ) = 0 \lambda^{\mathbf{i}}(\varepsilon_{9},\varepsilon_{9})=0 λ i ( ε 9 , ε 9 ) = 0 . Hence
λ i ( ε [ 9 , 14 ] , ε [ 4 , 9 ] ) = ( β 9 i , β 4 i ) g + ( β 14 i , β 4 i ) g + ( β 14 i , β 9 i ) g = 1 + 0 + 1 = 2. \lambda^{\mathbf{i}}(\varepsilon_{[9,14]},\varepsilon_{[4,9]})
=(\beta^{\mathbf{i}}_{9},\beta^{\mathbf{i}}_{4})_{\mathfrak{g}}
+(\beta^{\mathbf{i}}_{14},\beta^{\mathbf{i}}_{4})_{\mathfrak{g}}
+(\beta^{\mathbf{i}}_{14},\beta^{\mathbf{i}}_{9})_{\mathfrak{g}}
=1+0+1=2. λ i ( ε [ 9 , 14 ] , ε [ 4 , 9 ] ) = ( β 9 i , β 4 i ) g + ( β 14 i , β 4 i ) g + ( β 14 i , β 9 i ) g = 1 + 0 + 1 = 2.
Thus Λ ( M , N ) = − 2 \Lambda(M,N)=-2 Λ ( M , N ) = − 2 . Since k = ℓ = 2 k=\ell=2 k = ℓ = 2 and u = 2 u=2 u = 2 , Theorem 4.18 gives
d ( M , N ) = c 4 , 4 [ 10 ] ( 1 ) + c 4 , 4 [ 10 ] ( − 1 ) = d 4 = 1. \mathfrak{d}(M,N)=\mathfrak{c}^{[10]}_{4,4}(1)+\mathfrak{c}^{[10]}_{4,4}(-1)=d_{4}=1. d ( M , N ) = c 4 , 4 [ 10 ] ( 1 ) + c 4 , 4 [ 10 ] ( − 1 ) = d 4 = 1.
Therefore,
Λ ( N , M ) = 2 ⋅ 1 − ( − 2 ) = 4. \Lambda(N,M)=2\cdot1-(-2)=4. Λ ( N , M ) = 2 ⋅ 1 − ( − 2 ) = 4.
Beyond finite type
We conclude by returning to arbitrary symmetrizable Cartan data. Although the uniform formula of Theorem 4.18 is proved here only in finite type, its coefficient expression continues to detect three elementary local categorical interactions without any finite-type assumption: the interaction across the boundary of a box, the exchange interaction in a T T T -system, and the vanishing associated with commuting boxes.
Let i = ( i 1 , … , i l ) \mathbf{i}=(i_{1},\ldots,i_{l}) i = ( i 1 , … , i l ) be a reduced sequence, and let c = [ x , y ] \mathfrak{c}=[x,y] c = [ x , y ] be an i \mathbf{i} i -box of color i i i associated with i \mathbf{i} i . By [33 Lemma 5.8] , we have
d ( M i [ x , y ] , S y + i ) = d i if y + ≤ l and d ( S x − i , M i [ x , y ] ) = d i if x − ≥ 1. (51) \mathfrak{d}\left(M^{\mathbf{i}}[x,y],S^{i}_{y^{+}}\right)=d_{i}\ \text{if }y^{+}\leq l
\quad\text{and}\quad
\mathfrak{d}\left(S^{i}_{x^{-}},M^{\mathbf{i}}[x,y]\right)=d_{i}\ \text{if }x^{-}\geq1. \tag*{(51)} d ( M i [ x , y ] , S y + i ) = d i if y + ≤ l and d ( S x − i , M i [ x , y ] ) = d i if x − ≥ 1. (51)
We next record a coordinate estimate for commuting i \mathbf{i} i -boxes.
Lemma 4.28. Let i \mathbf{i} i be a finite Q Q Q -adapted sequence, and let c = [ x , y ] \mathfrak{c}=[x,y] c = [ x , y ] and c ′ = [ x ′ , y ′ ] \mathfrak{c}'=[x',y'] c ′ = [ x ′ , y ′ ] be commuting i \mathbf{i} i -boxes of colors i i i and j j j , respectively. Let k k k and ℓ \ell ℓ be the numbers of occurrences of colors i i i and j j j in [ x , y ] [x,y] [ x , y ] and [ x ′ , y ′ ] [x',y'] [ x ′ , y ′ ] , respectively, and set p = ( p x i + p y i ) / 2 p=(p^{i}_{x}+p^{i}_{y})/2 p = ( p x i + p y i ) /2 and s = ( p x ′ j + p y ′ j ) / 2 s=(p^{j}_{x'}+p^{j}_{y'})/2 s = ( p x ′ j + p y ′ j ) /2 . Then
∣ p − s ∣ ≤ ∣ k − ℓ ∣ + d ( i , j ) . |p-s|\leq|k-\ell|+d(i,j). ∣ p − s ∣ ≤ ∣ k − ℓ ∣ + d ( i , j ) .
When i \mathbf{i} i is reduced, the corresponding KR-modules are
V ( [ i k ] ) p and V ( [ j ℓ ] ) s . V\left(\begin{bmatrix}i^{k}\end{bmatrix}\right)_{p}
\qquad\text{and}\qquad
V\left(\begin{bmatrix}j^{\ell}\end{bmatrix}\right)_{s}. V ( [ i k ] ) p and V ( [ j ℓ ] ) s .
Proof. Write P a = p a i P_a=p_a^i P a = p a i . If a < b a<b a < b , let η \eta η be the height function immediately before position b b b is read. Since position a a a has already been read, we have
P a + 2 ≤ η i a ≤ η i b + d ( i a , i b ) = P b + d ( i a , i b ) . (52) P_a+2\leq\eta_{i_a}\leq\eta_{i_b}+d(i_a,i_b)=P_b+d(i_a,i_b). \tag*{(52)} P a + 2 ≤ η i a ≤ η i b + d ( i a , i b ) = P b + d ( i a , i b ) . (52)
Here the second inequality follows from ∣ η i − η j ∣ = 1 |\eta_i-\eta_j|=1 ∣ η i − η j ∣ = 1 for adjacent vertices i , j i,j i , j .
Suppose first that
x − < x ′ ≤ y ′ < y + . x^-<x'\leq y'<y^+. x − < x ′ ≤ y ′ < y + .
We have
P x = p − k + 1 , P y = p + k − 1 and P x ′ = s − ℓ + 1 , P y ′ = s + ℓ − 1. P_x=p-k+1,\qquad P_y=p+k-1\qquad\text{and}\qquad P_{x'}=s-\ell+1,\qquad P_{y'}=s+\ell-1. P x = p − k + 1 , P y = p + k − 1 and P x ′ = s − ℓ + 1 , P y ′ = s + ℓ − 1.
If x − x^- x − exists, then P x − = p − k − 1 P_{x^-}=p-k-1 P x − = p − k − 1 , and (52) gives
p − s ≤ k − ℓ + d ( i , j ) . p-s\leq k-\ell+d(i,j). p − s ≤ k − ℓ + d ( i , j ) .
If x − = − ∞ x^-=-\infty x − = − ∞ , then P x = ξ i P_x=\xi_i P x = ξ i and P x ′ ≥ ξ j P_{x'}\geq\xi_j P x ′ ≥ ξ j , so the same inequality follows by applying the preceding shortest-path argument to the initial height function.
Similarly, if y + y^+ y + exists, then P y + = p + k + 1 P_{y^+}=p+k+1 P y + = p + k + 1 , and (52) gives
s − p ≤ k − ℓ + d ( i , j ) . s-p\leq k-\ell+d(i,j). s − p ≤ k − ℓ + d ( i , j ) .
If y + = + ∞ y^+=+\infty y + = + ∞ , then, for the height function obtained after the finite sequence has been read, the heights at i i i and j j j are P y + 2 P_y+2 P y + 2 and at least P y ′ + 2 P_{y'}+2 P y ′ + 2 , respectively. The same shortest-path argument therefore gives the inequality.
Consequently,
∣ p − s ∣ ≤ k − ℓ + d ( i , j ) . |p-s|\leq k-\ell+d(i,j). ∣ p − s ∣ ≤ k − ℓ + d ( i , j ) .
For the other commutation condition
( x ′ ) − < x ≤ y < ( y ′ ) + , (x')^-<x\leq y<(y')^+, ( x ′ ) − < x ≤ y < ( y ′ ) + ,
interchanging the two boxes gives
∣ p − s ∣ ≤ ℓ − k + d ( i , j ) . |p-s|\leq\ell-k+d(i,j). ∣ p − s ∣ ≤ ℓ − k + d ( i , j ) .
We also recall the following consequence of the determinantial module T T T -system.
Proposition 4.29 ([33 Lemma 5.18] ). Let i = ( i 1 , … , i l ) \boldsymbol{i}=(i_1,\ldots,i_l) i = ( i 1 , … , i l ) be a reduced sequence, and let [ x , y ] [x,y] [ x , y ] be an i \boldsymbol{i} i -box of color i i i with x < y x<y x < y . Then the two determinantial modules associated with the T T T -system exchange pair [ x + , y ] [x^+,y] [ x + , y ] and [ x , y − ] [x,y^-] [ x , y − ] satisfy
d ( M i [ x + , y ] , M i [ x , y − ] ) = d i . \mathfrak{d}\left(M^i[x^+,y],M^i[x,y^-]\right)=d_i. d ( M i [ x + , y ] , M i [ x , y − ] ) = d i .
Proposition 4.30. Let the Cartan datum be symmetrizable, and let the indicated i \boldsymbol{i} i -boxes occur in a Q Q Q -adapted reduced sequence i \boldsymbol{i} i .
(i) If an i \boldsymbol{i} i -box of color i i i corresponds to V ( i k ℓ ) p V\left(\begin{smallmatrix}i^k\\ \ell\end{smallmatrix}\right)_p V ( i k ℓ ) p and the singleton box at the next occurrence of color i i i corresponds to V ( i ℓ ) p + k + 1 V\left(\begin{smallmatrix}i\\ \ell\end{smallmatrix}\right)_{p+k+1} V ( i ℓ ) p + k + 1 , then
d ( V ( i k ℓ ) p , V ( i ℓ ) p + k + 1 ) = [ t k + 1 ] ( t k [ 1 ] t C ~ i , i ( t ) ) = d i . \mathfrak{d}\left(V\left(\begin{smallmatrix}i^k\\ \ell\end{smallmatrix}\right)_p,V\left(\begin{smallmatrix}i\\ \ell\end{smallmatrix}\right)_{p+k+1}\right)
=[t^{k+1}]\left(t^k[1]_t\widetilde{C}_{i,i}(t)\right)=d_i. d ( V ( i k ℓ ) p , V ( i ℓ ) p + k + 1 ) = [ t k + 1 ] ( t k [ 1 ] t C i , i ( t ) ) = d i . (ii) If V ( i k ℓ ) p V\left(\begin{smallmatrix}i^k\\ \ell\end{smallmatrix}\right)_p V ( i k ℓ ) p and V ( i k ℓ ) s V\left(\begin{smallmatrix}i^k\\ \ell\end{smallmatrix}\right)_s V ( i k ℓ ) s correspond to a T T T -system exchange pair of i \boldsymbol{i} i -boxes, then ∣ p − s ∣ = 2 |p-s|=2 ∣ p − s ∣ = 2 and
d ( V ( i k ℓ ) p , V ( i k ℓ ) s ) = [ t 2 ] ( t k [ k ] t C ~ i , i ( t ) ) = d i . \mathfrak{d}\left(V\left(\begin{smallmatrix}i^k\\ \ell\end{smallmatrix}\right)_p,V\left(\begin{smallmatrix}i^k\\ \ell\end{smallmatrix}\right)_s\right)
=[t^2]\left(t^k[k]_t\widetilde{C}_{i,i}(t)\right)=d_i. d ( V ( i k ℓ ) p , V ( i k ℓ ) s ) = [ t 2 ] ( t k [ k ] t C i , i ( t ) ) = d i .
(iii) If two commuting i i i -boxes of colors i i i and j j j correspond, respectively, to V ( [ i k ] ) p V\left(\left[\begin{smallmatrix}i^k\end{smallmatrix}\right]\right)_p V ( [ i k ] ) p and V ( [ j ℓ ] ) s V\left(\left[\begin{smallmatrix}j^\ell\end{smallmatrix}\right]\right)_s V ( [ j ℓ ] ) s , then
∂ ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = [ t ∣ p − s ∣ ] ( t K [ L ] t C ~ i , j ( t ) ) = 0 , \partial\left(V\left(\left[\begin{smallmatrix}i^k\end{smallmatrix}\right]\right)_p,V\left(\left[\begin{smallmatrix}j^\ell\end{smallmatrix}\right]\right)_s\right)=[t^{|p-s|}]\left(t^K[L]_t\widetilde{C}_{i,j}(t)\right)=0, ∂ ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = [ t ∣ p − s ∣ ] ( t K [ L ] t C i , j ( t ) ) = 0 ,
where
K = max ( k , ℓ ) , L = min ( k , ℓ ) . K=\max(k,\ell),\qquad L=\min(k,\ell). K = max ( k , ℓ ) , L = min ( k , ℓ ) .
Proof. For (i), the categorical equality is given in (51) . Since [ 1 ] t = 1 [1]_t=1 [ 1 ] t = 1 , we have
[ t k + 1 ] ( t k [ 1 ] t C ~ i , i ( t ) ) = [ t 1 ] C ~ i , i ( t ) = c ~ i , i ( 1 ) = d i . [t^{k+1}]\left(t^k[1]_t\widetilde{C}_{i,i}(t)\right)=[t^1]\widetilde{C}_{i,i}(t)=\widetilde{c}_{i,i}(1)=d_i. [ t k + 1 ] ( t k [ 1 ] t C i , i ( t ) ) = [ t 1 ] C i , i ( t ) = c i , i ( 1 ) = d i .
For (ii), the categorical equality follows from Proposition 4.29 . Indeed, the two subboxes [ x + , y ] [x^+,y] [ x + , y ] and [ x , y − ] [x,y^-] [ x , y − ] each contain k k k occurrences of color i i i , and their centered coordinates differ by 2 2 2 . On the coefficient side,
t k [ k ] t = t + t 3 + ⋯ + t 2 k − 1 . t^k[k]_t=t+t^3+\cdots+t^{2k-1}. t k [ k ] t = t + t 3 + ⋯ + t 2 k − 1 .
Since C ~ i , i ( t ) ∈ t Z [ [ t ] ] \widetilde{C}_{i,i}(t)\in t\mathbb{Z}[[t]] C i , i ( t ) ∈ t Z [[ t ]] and [ t 1 ] C ~ i , i ( t ) = c ~ i , i ( 1 ) = d i [t^1]\widetilde{C}_{i,i}(t)=\widetilde{c}_{i,i}(1)=d_i [ t 1 ] C i , i ( t ) = c i , i ( 1 ) = d i , it follows that
[ t 2 ] ( t k [ k ] t C ~ i , i ( t ) ) = [ t 1 ] C ~ i , i ( t ) = c ~ i , i ( 1 ) = d i . [t^2]\left(t^k[k]_t\widetilde{C}_{i,i}(t)\right)=[t^1]\widetilde{C}_{i,i}(t)=\widetilde{c}_{i,i}(1)=d_i. [ t 2 ] ( t k [ k ] t C i , i ( t ) ) = [ t 1 ] C i , i ( t ) = c i , i ( 1 ) = d i .
For (iii), Lemma 3.5 implies that the corresponding quiver Hecke KR-modules strongly commute. Hence
∂ ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = 0. \partial\left(V\left(\left[\begin{smallmatrix}i^k\end{smallmatrix}\right]\right)_p,V\left(\left[\begin{smallmatrix}j^\ell\end{smallmatrix}\right]\right)_s\right)=0. ∂ ( V ( [ i k ] ) p , V ( [ j ℓ ] ) s ) = 0.
Set
u = ∣ p − s ∣ , K = max ( k , ℓ ) , L = min ( k , ℓ ) . u=|p-s|,\qquad K=\max(k,\ell),\qquad L=\min(k,\ell). u = ∣ p − s ∣ , K = max ( k , ℓ ) , L = min ( k , ℓ ) .
By Lemma 4.28, we have u ≤ ∣ k − ℓ ∣ + d ( i , j ) u\leq|k-\ell|+d(i,j) u ≤ ∣ k − ℓ ∣ + d ( i , j ) .
Since t K [ L ] t = ∑ a = 0 L − 1 t K − L + 1 + 2 a t^K[L]_t=\sum_{a=0}^{L-1}t^{K-L+1+2a} t K [ L ] t = ∑ a = 0 L − 1 t K − L + 1 + 2 a , we obtain
[ t u ] ( t K [ L ] t C ~ i , j ( t ) ) = ∑ a = 0 L − 1 c ~ i , j ( u − ∣ k − ℓ ∣ − 2 a − 1 ) . [t^u]\left(t^K[L]_t\widetilde{C}_{i,j}(t)\right)=\sum_{a=0}^{L-1}\widetilde{c}_{i,j}\left(u-|k-\ell|-2a-1\right). [ t u ] ( t K [ L ] t C i , j ( t ) ) = a = 0 ∑ L − 1 c i , j ( u − ∣ k − ℓ ∣ − 2 a − 1 ) .
For every 0 ≤ a ≤ L − 1 0\leq a\leq L-1 0 ≤ a ≤ L − 1 , we have
u − ∣ k − ℓ ∣ − 2 a − 1 ≤ d ( i , j ) − 2 a − 1 < d ( i , j ) . u-|k-\ell|-2a-1\leq d(i,j)-2a-1<d(i,j). u − ∣ k − ℓ ∣ − 2 a − 1 ≤ d ( i , j ) − 2 a − 1 < d ( i , j ) .
Therefore, Proposition 4.6 (i) shows that every summand on the right-hand side vanishes:
c ~ i , j ( u − ∣ k − ℓ ∣ − 2 a − 1 ) = 0. \widetilde{c}_{i,j}\left(u-|k-\ell|-2a-1\right)=0. c i , j ( u − ∣ k − ℓ ∣ − 2 a − 1 ) = 0.
Consequently,
[ t u ] ( t K [ L ] t C ~ i , j ( t ) ) = 0. [t^u]\left(t^K[L]_t\widetilde{C}_{i,j}(t)\right)=0. [ t u ] ( t K [ L ] t C i , j ( t ) ) = 0.
Thus the ∂ \partial ∂ -invariant in each of these three cases is given by the coefficient formula with the full t t t -adic entry C ~ i , j ( t ) \widetilde{C}_{i,j}(t) C i , j ( t ) , without a finite-type assumption.
Example 4.31. Consider type B 3 ( 1 ) B_3^{(1)} B 3 ( 1 ) with D = diag ( 2 , 2 , 2 , 1 ) D=\operatorname{diag}(2,2,2,1) D = diag ( 2 , 2 , 2 , 1 ) . Choose the height function ( ξ 0 , ξ 1 , ξ 2 , ξ 3 ) = ( 1 , 1 , 2 , 1 ) (\xi_0,\xi_1,\xi_2,\xi_3)=(1,1,2,1) ( ξ 0 , ξ 1 , ξ 2 , ξ 3 ) = ( 1 , 1 , 2 , 1 ) . The sequence
i = ( 0 , 1 , 3 , 2 , 0 , 1 , 3 , 2 , 0 , 1 , 3 , 2 , 0 ) \mathbf{i}=(0,1,3,2,0,1,3,2,0,1,3,2,0) i = ( 0 , 1 , 3 , 2 , 0 , 1 , 3 , 2 , 0 , 1 , 3 , 2 , 0 )
is Q Q Q -adapted and reduced. Consider the boxes [ 1 , 5 ] [1,5] [ 1 , 5 ] and [ 4 , 8 ] [4,8] [ 4 , 8 ] of colors 0 and 2, respectively. Their corresponding quiver Hecke KR-modules are
M i [ 1 , 5 ] ≃ V ( [ 0 2 ] ) 2 and M i [ 4 , 8 ] ≃ V ( [ 2 2 ] ) 3 . M^{i}[1,5]\simeq V([0^{2}])_{2}\quad\text{and}\quad M^{i}[4,8]\simeq V([2^{2}])_{3}. M i [ 1 , 5 ] ≃ V ([ 0 2 ] ) 2 and M i [ 4 , 8 ] ≃ V ([ 2 2 ] ) 3 .
The boxes are not nested, since 1 < 4 ≤ 5 < 8 1<4\leq5<8 1 < 4 ≤ 5 < 8 . Moreover, 1 − < 4 ≤ 8 < 5 + = 9 1^{-}<4\leq8<5^{+}=9 1 − < 4 ≤ 8 < 5 + = 9 , so they commute by the definition of commuting i-boxes. Thus Lemma 3.5 gives
∂ ( V ( [ 0 2 ] ) 2 , V ( [ 2 2 ] ) 3 ) = 0. \partial\left(V([0^{2}])_{2},V([2^{2}])_{3}\right)=0. ∂ ( V ([ 0 2 ] ) 2 , V ([ 2 2 ] ) 3 ) = 0.
The same value is obtained directly from the full t t t -adic symmetrized inverse t t t -Cartan matrix. Indeed, K = L = 2 K=L=2 K = L = 2 and ∣ 2 − 3 ∣ = 1 |2-3|=1 ∣2 − 3∣ = 1 , and therefore
[ t 1 ] ( t 2 [ 2 ] t C ~ 0 , 2 B ( 1 ) ( t ) ) = C ~ 0 , 2 B ( 1 ) ( − 2 ) + C ~ 0 , 2 B ( 1 ) ( 0 ) = 0. [t^{1}]\left(t^{2}[2]_{t}\widetilde{C}^{B^{(1)}}_{0,2}(t)\right)=\widetilde{C}^{B^{(1)}}_{0,2}(-2)+\widetilde{C}^{B^{(1)}}_{0,2}(0)=0. [ t 1 ] ( t 2 [ 2 ] t C 0 , 2 B ( 1 ) ( t ) ) = C 0 , 2 B ( 1 ) ( − 2 ) + C 0 , 2 B ( 1 ) ( 0 ) = 0.
Here the last equality follows from Proposition 4.6 , since both coefficient indices are at most d ( 0 , 2 ) = 1 d(0,2)=1 d ( 0 , 2 ) = 1 . This illustrates Proposition 4.30 in affine type, using two non-nested boxes of different colors.
References
(S.-j. Oh) Department of Mathematics, Sungkyunkwan University, Suwon, South Korea
Email address , S.-j. Oh: sejin092@gmail.com