The research of W.-S. Jung was supported by the National Research Foundation of Korea (NRF) grants funded by the Korea government (Ministry of Education and MSIT) (RS-2024-00451844 and RS-2023-00273425).
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
Global and canonical bases and their dual bases are fundamental objects in the representation theory of quantum groups. Their positivity and integrality properties are accompanied by a more subtle multiplicative problem: one would like to recognize when products of distinguished basis elements are again distinguished. Cluster algebras and their quantum analogues were introduced in part to provide a combinatorial framework for this problem [7 , 1 ] . In coordinate rings arising from Lie theory and their quantizations, cluster variables and cluster monomials are expected to be closely related to upper global or dual canonical basis elements, while exchange relations make part of their multiplicative structure explicit.
A fundamental model for this interaction is the quantum unipotent coordinate algebra
A q ( n ( w ) ) , \mathcal{A}_q(\mathbf{n}(w)), A q ( n ( w )) ,
where w w w is a Weyl group element. A reduced sequence i \boldsymbol{i} i of w w w determines dual PBW root vectors, and different reduced sequences are related by braid moves. The algebra A q ( n ( w ) ) \mathcal{A}_q(\mathbf{n}(w)) A q ( n ( w )) carries quantum cluster algebra structures [15 , 16 , 17 , 18 ] ; its cluster variables include quantum minors, and its cluster monomials are related to the dual canonical/upper global basis [39 , 26 ] . In symmetric type, modules over quiver Hecke algebras provide a monoidal categorification : Under monoidal categorification, cluster variables correspond to real prime simple modules; the modules attached to a cluster form a maximal commuting family , and mutations are encoded by short exact sequences [26 ] . Thus quantum unipotent coordinate algebras form a setting in which quantum-group-theoretic, cluster-theoretic, and categorical structures meet.
Recent work extends this picture from Weyl group elements to positive braids. Bosonic extensions of quantum groups have braid group actions, and to a symmetrizable Cartan datum g \mathfrak{g} g and a positive braid b ∈ B g + \mathbf{b} \in B_{\mathfrak{g}}^+ b ∈ B g + one associates a positive-braid subalgebra
A q ( b ) \mathcal{A}_q(\mathbf{b}) A q ( b )
of the corresponding bosonic extension [23 , 11 , 24 , 46 , 33 , 30 , 34 ] . When b \mathbf{b} b is represented by a reduced expression of w ∈ W g w \in W_{\mathfrak{g}} w ∈ W g , this construction recovers A q ( n ( w ) ) \mathcal{A}_q(\mathbf{n}(w)) A q ( n ( w )) . Thus positive-braid algebras extend quantum unipotent coordinate algebras to arbitrary positive braids, whose associated root sequences may contain repeated or negative roots. In finite Cartan type, they admit realizations inside quantum (virtual) Grothendieck rings; in simply-laced type A D E ADE A D E , the corresponding realizations are related to quantum Grothendieck rings of finite-dimensional representations of quantum loop algebras. Positive braids of finite type also occur naturally in the geometry of braid varieties and their cluster structures [3 ] .
To make this representation-theoretic connection more explicit, let C g \mathcal{C}_{\mathfrak{g}} C g be the category of finite-dimensional representations of U q ( L g ) U_q(L\mathfrak{g}) U q ( L g ) .1 The q q q -character theory of Frenkel and Reshetikhin [9 , 10 ] realizes its Grothendieck ring as a subring of an ambient Laurent polynomial ring:
K Z ( g ) ↪ Z [ y i , p ± 1 ] . (1) \mathcal{K}_{\mathbb{Z}}(\mathfrak{g}) \hookrightarrow\mathbb{Z}[y_{i,p}^{\pm1}]. \tag*{(1)} K Z ( g ) ↪ Z [ y i , p ± 1 ] . (1)
In simply-laced type, Nakajima’s quiver varieties give geometric constructions of q q q -characters and their quantum deformations [43 , 45 , 55 ] . The corresponding quantum Grothendieck rings K t ( g ) \mathcal{K}_t(\mathfrak{g}) K t ( g ) have canonical or global bases given by the classes of simple modules, while distinguished monoidal subcategories of C g \mathcal{C}_{\mathfrak{g}} C g provide monoidal categorifications of cluster algebras [21 , 23 , 32 , 34 ] . In particular, for the seeds associated with adapted sequences, the initial cluster variables are given by q q q -characters of Kirillov–Reshetikhin modules [23 , 11 , 25 ] .
Even when g \mathfrak{g} g is non-simply-laced, however, these cluster algebras K t ( g ) \mathcal{K}_t(\mathfrak{g}) K t ( g ) are of skew-symmetric type and are naturally related to simply-laced bosonic extensions [23 , 11 , 2 ] . This simply-laced nature of the usual Grothendieck ring led to the introduction of the quantum virtual Grothendieck ring R q ( g ) \mathcal{R}_q(\mathfrak{g}) R q ( g ) as a genuinely non-simply-laced counterpart [35 , 25 ] . Its quantum cluster algebra structure is skew-symmetrizable and reflects the Cartan datum of g \mathfrak{g} g itself [25 ] , as does its braid-group action [24 ] .
The Laurent realizations (1) make the corresponding folding particularly natural. Suppose that the Cartan matrix of g \mathfrak g g is the quotient Cartan matrix associated with an admissible finite-group action on the Dynkin diagram of a simply-laced algebra g ^ \widehat{\mathfrak g} g , and write I i I_i I i for the orbit corresponding to vertex i i i of g \mathfrak g g (see Subsection 1.2). At q = 1 q=1 q = 1 , identifying the variables in each Dynkin-diagram orbit defines
π : C [ y i ^ , p ± 1 ] ⟶ C [ y i , p ± 1 ] , y i ^ , p ⟼ y i , p ∣ I i ∣ . \pi:\mathbb{C}[y_{\widehat{i},p}^{\pm1}] \longrightarrow\mathbb{C}[y_{i,p}^{\pm1}],\qquad y_{\widehat{i},p} \longmapsto y_{i,p}^{\lvert I_i\rvert}. π : C [ y i , p ± 1 ] ⟶ C [ y i , p ± 1 ] , y i , p ⟼ y i , p ∣ I i ∣ .
A nontrivial feature of this elementary identification is that it restricts to a folding homomorphism
π : K ( g ^ ) ⟶ R ( g ) . \pi:\mathcal{K}(\widehat{\mathfrak g}) \longrightarrow\mathcal{R}(\mathfrak g). π : K ( g ) ⟶ R ( g ) .
At the level of heart subrings [22 , 24 ] , this map is somewhat natural in view of the identifications with the unipotent coordinate rings of the simply-laced and folded types:
N ^ g → C [ N ^ g ] → K ( g ^ ) → C [ y i ^ , p ± 1 ] ∼ ↑ ↓ ∼ ↓ π ↓ π N g → C [ N g ] → R ( g ) → C [ y i , p ± 1 ] \begin{CD}
\widehat{N}_{\mathfrak g} @>{}>> \mathbb{C}[\widehat{N}_{\mathfrak g}] @>{}>> \mathcal{K}(\widehat{\mathfrak g}) @>{}>> \mathbb{C}[y_{\widehat{i},p}^{\pm1}] \\
@A{\sim}AA @VV{\sim}V @VV{\pi}V @VV{\pi}V \\
N_{\mathfrak g} @>{}>> \mathbb{C}[N_{\mathfrak g}] @>{}>> \mathcal{R}(\mathfrak g) @>{}>> \mathbb{C}[y_{i,p}^{\pm1}]
\end{CD} N g ∼ ⏐ ↑ N g C [ N g ] ↓ ⏐ ∼ C [ N g ] K ( g ) ↓ ⏐ π R ( g ) C [ y i , p ± 1 ] ↓ ⏐ π C [ y i , p ± 1 ]
What is much less apparent is that this elementary folding is highly compatible with the corresponding cluster structures.
For a non-simply-laced Cartan datum, analogous geometric or monoidal structures are not always available at the same level of generality. This motivates the use of an admissible Dynkin folding
( g ^ , Γ ) ⟶ g . (\widehat{\mathfrak g},\Gamma)\longrightarrow\mathfrak g. ( g , Γ ) ⟶ g .
where g ^ \widehat{\mathfrak g} g is simply-laced and Γ \Gamma Γ acts admissibly on its Dynkin diagram. Our broader goal is to understand how the geometric, representation-theoretic, and categorical structures in simply-laced types are related to those in non-simply-laced types through folding. We construct surjective homomorphisms from cluster algebras associated with g ^ \widehat{\mathfrak g} g to cluster algebras associated with g \mathfrak g g . The direction of this motivation may be summarized as
well-understood simply-laced upstairs structures ↓ folding generally less-understood downstairs structures. (2) \begin{gathered}
\text{well-understood simply-laced upstairs structures}\\
\downarrow\ \text{folding}\\
\text{generally less-understood downstairs structures.} \tag*{(2)}
\end{gathered} well-understood simply-laced upstairs structures ↓ folding generally less-understood downstairs structures. (2)
At the level of Cartan matrices, admissible folding is elementary. If I i I_i I i is the orbit corresponding to i ∈ I i\in I i ∈ I , then (see § 5.2 )
c i j = ∑ a ∈ I i c ^ a , b ( b ∈ I j ) , d i ∣ I i ∣ = m . c_{ij}=\sum_{a\in I_i}\widehat{c}_{a,b}\quad(b\in I_j),\qquad d_i\lvert I_i\rvert=m. c ij = a ∈ I i ∑ c a , b ( b ∈ I j ) , d i ∣ I i ∣ = m .
The corresponding problem for quantum cluster seeds is more delicate: one must fold not only the exchange matrices B B B , but also the compatible skew-symmetric matrices Λ \Lambda Λ , together with mutations and individual cluster variables. A further difficulty is that identifying the variables in each orbit does not in general define a homomorphism between the corresponding quantum tori.
One of our main observations is that the Cartan-matrix quotient rule also governs the root-theoretic construction of the corresponding quantum cluster seeds. For every finite sequence i i i and every symmetrizable Cartan datum g \mathfrak{g} g , we give an explicit root-theoretic formula for the skew-symmetric matrix Λ i \Lambda^{i} Λ i and prove that
( B i , Λ i ) is a compatible pair (B^{i},\Lambda^{i}) \text{ is a compatible pair} ( B i , Λ i ) is a compatible pair
(Proposition 3.1).
For the blockwise lift i ^ \widehat{i} i of i i i , the associated root sequence folds to that of i i i , and the bilinear and skew-symmetric pairings upstairs recover their downstairs counterparts by explicit orbit sums. Schematically,
C ^ ⟶ C ⟹ ( upstairs roots and pairings ) ⟶ ( downstairs roots and pairings ) , \widehat{C}\longrightarrow C \quad\Longrightarrow\quad(\text{upstairs roots and pairings})\longrightarrow(\text{downstairs roots and pairings}), C ⟶ C ⟹ ( upstairs roots and pairings ) ⟶ ( downstairs roots and pairings ) ,
and hence
( B ^ i , Λ ^ i ) → folding ( B i , Λ i ) . (\widehat{B}^{i},\widehat{\Lambda}^{i})\xrightarrow{\text{folding}}(B^{i},\Lambda^{i}). ( B i , Λ i ) folding ( B i , Λ i ) .
By Lemmas 2.4 , 2.6 , and 2.7 , this holds for arbitrary sequences i i i , without assuming reducedness or finite Cartan type.
More importantly, this construction extends uniformly from individual positions to i-boxes. These boxes were introduced in [30 , 27 ] as a combinatorial tool for the monoidal categorification of cluster algebras associated with arbitrary sequences. A maximal commuting family F \mathcal{F} F of i-boxes determines an exchange matrix B F B^{\mathcal{F}} B F . Using the box vectors (28) and the skew-symmetric pairing λ i \lambda^{i} λ i (15) , we define Λ F \Lambda^{\mathcal{F}} Λ F and prove that
( B F , Λ F ) is a compatible pair (B^{\mathcal{F}},\Lambda^{\mathcal{F}}) \text{ is a compatible pair} ( B F , Λ F ) is a compatible pair
(Theorem 3.5 ) for every maximal commuting family F \mathcal{F} F . Thus the compatible-pair construction for ( B i , Λ i ) (B^{i},\Lambda^{i}) ( B i , Λ i ) extends naturally to all i-box seeds.
We then compare this construction under folding. Let C \mathcal{C} C be an admissible chain of i-boxes realizing F \mathcal{F} F (see Subsection 3.3). The canonical lift F ^ \widehat{\mathcal{F}} F of F \mathcal{F} F is organized into fibers over the boxes of F \mathcal{F} F , and the entire compatible pair folds along these fibers:
( B F ^ , Λ F ^ ) → folding ( B F , Λ F ) . (B^{\widehat{\mathcal{F}}},\Lambda^{\widehat{\mathcal{F}}})\xrightarrow{\text{folding}}(B^{\mathcal{F}},\Lambda^{\mathcal{F}}). ( B F , Λ F ) folding ( B F , Λ F ) .
More precisely, for a fixed representative d ^ t ∈ Fib C ( d ) \widehat{d}_{t}\in\operatorname{Fib}^{\mathcal{C}}(d) d t ∈ Fib C ( d ) ,
b c , d F = ∑ c ^ ∈ Fib C ^ ( c ) b c ^ , d ^ t F ^ , Λ c , d F = d i c ∑ c ^ ∈ Fib C ^ ( c ) Λ c ^ , d ^ t F ^ b^{\mathcal{F}}_{c,d}=\sum_{\widehat{c}\in\operatorname{Fib}^{\widehat{\mathcal{C}}}(c)}b^{\widehat{\mathcal{F}}}_{\widehat{c},\widehat{d}_{t}},\qquad
\Lambda^{\mathcal{F}}_{c,d}=d_{i_c}\sum_{\widehat{c}\in\operatorname{Fib}^{\widehat{\mathcal{C}}}(c)}\Lambda^{\widehat{\mathcal{F}}}_{\widehat{c},\widehat{d}_{t}} b c , d F = c ∈ Fib C ( c ) ∑ b c , d t F , Λ c , d F = d i c c ∈ Fib C ( c ) ∑ Λ c , d t F
(Theorem 3.15 ).
Thus the quotient Cartan rule governs not only the root data and the compatible pair associated with the sequence itself, but also the exchange B B B and commutation matrices Λ \Lambda Λ of every i-box seed.
The results above compare the compatible pairs ( B F ^ , Λ F ^ ) (B^{\widehat{\mathcal{F}}},\Lambda^{\widehat{\mathcal{F}}}) ( B F , Λ F ) and ( B F , Λ F ) (B^{\mathcal{F}},\Lambda^{\mathcal{F}}) ( B F , Λ F ) at a fixed i-box seed. Sections 4 and 5 establish the same matrix identities for the seeds obtained after braid moves and box moves. Using the mutation–permutation realizations of these moves, we compare the downstairs seeds before and after each move with the upstairs seeds before and after the corresponding sequence of moves. The Laurent homomorphism sending each upstairs variable in a fiber before the move to the corresponding downstairs variable also sends every upstairs variable in a fiber after the move to the corresponding downstairs cluster variable. For a box move C ⟶ C ′ \mathcal{C}\longrightarrow\mathcal{C}' C ⟶ C ′ , the individual-variable statement takes the uniform form
ϕ C ( x ^ c ^ ′ ) = x c ′ ( c ^ ′ ∈ Fib C ^ ( c ′ ) ) \phi^{\mathcal{C}}(\widehat{x}_{\widehat{c}'})=x_{c'}
\qquad
(\widehat{c}'\in\operatorname{Fib}^{\widehat{\mathcal{C}}}(c')) ϕ C ( x c ′ ) = x c ′ ( c ′ ∈ Fib C ( c ′ ))
(Theorem 5.15 ).
The corresponding g g g -vector statements are Proposition 4.22 and Corollary 5.13 .
The two kinds of moves serve complementary purposes. Braid moves compare the seeds attached to different sequences, whereas box moves compare maximal commuting families for a fixed sequence.
We combine this machinery in § 11 . For a positive braid b \mathbf{b} b , a sequence i ∈ Seq ( b ) \mathbf{i} \in\operatorname{Seq}(\mathbf{b}) i ∈ Seq ( b ) , and a maximal commuting family F \mathcal{F} F of i \mathbf{i} i -boxes, let F ^ \widehat{\mathcal{F}} F be its canonical lift. The fibers over the boxes of F \mathcal{F} F define a surjective Laurent polynomial homomorphism
ϕ F : L ( Σ F ^ ) ⟶ L ( Σ F ) , x ^ c ^ ⟼ x c ( c ^ ∈ Fib C ( c ) ) . \phi^{\mathcal{F}}:\mathcal{L}(\Sigma^{\widehat{\mathcal{F}}}) \longrightarrow\mathcal{L}(\Sigma^{\mathcal{F}}), \qquad\widehat{x}_{\widehat{c}} \longmapsto x_c \quad(\widehat{c} \in\operatorname{Fib}^{\mathsf{C}}(c)). ϕ F : L ( Σ F ) ⟶ L ( Σ F ) , x c ⟼ x c ( c ∈ Fib C ( c )) .
Theorem 6.2 shows that this one map folds all cluster variables and cluster monomials arising from the box seeds connected by the braid and box transformations under consideration. The exponents of a cluster monomial are added along each fiber, and the corresponding g g g -vectors fold by the same summation rule. Explicitly,
ϕ F ( ∏ d ^ ∈ G ^ ( x ^ d ^ ) a d ^ ) = ∏ d ∈ G ( x d ) ∑ d ^ ∈ Fib D ( d ) a d ^ ( x r e f ‘ T h e o r e m 6.2 ‘ ) . \phi^{\mathcal{F}}\left(\prod_{\widehat{d}\in\widehat{\mathcal{G}}}(\widehat{x}_{\widehat{d}})^{a_{\widehat{d}}}\right)=\prod_{d\in\mathcal{G}}(x_d)^{\sum_{\widehat{d}\in\operatorname{Fib}^{\mathsf{D}}(d)}a_{\widehat{d}}}
\quad({xref}`Theorem 6.2`). ϕ F d ∈ G ∏ ( x d ) a d = d ∈ G ∏ ( x d ) ∑ d ∈ Fib D ( d ) a d ( x r e f ‘ T h eor e m 6.2‘ ) .
The Laurent phenomenon, together with the individual-variable folding for i \mathbf{i} i -box seeds, naturally suggests that this fixed Laurent homomorphism should restrict to the corresponding cluster algebras. Our main result confirms this expectation in a stronger form: in the commutative setting, the restriction is a surjective algebra homomorphism.
Main Theorem. Fix an admissible Dynkin folding ( g ^ , Γ ) → g (\widehat{\mathfrak{g}},\Gamma)\to\mathfrak{g} ( g , Γ ) → g . Let b ∈ B g + \mathbf{b}\in B_g^+ b ∈ B g + be an arbitrary positive braid. For every maximal commuting family F \mathcal{F} F of i \mathbf{i} i -boxes and an admissible chain realizing F \mathcal{F} F , the fixed Laurent homomorphism ϕ b F \phi^{\mathcal{F}}_{\mathbf{b}} ϕ b F restricts to a surjective C \mathbb{C} C -algebra homomorphism
ϕ b F : A ( b ^ ) ↠ A ( b ) \phi^{\mathcal{F}}_{\mathbf{b}}:\mathcal{A}(\widehat{\mathbf{b}})\twoheadrightarrow\mathcal{A}(\mathbf{b}) ϕ b F : A ( b ) ↠ A ( b ) (Theorem 6.4, proved in full generality in § 14 ).
We first consider reduced sequences, where the cluster algebras are realized by quantum unipotent coordinate algebras A q ( n ( w ) ) \mathcal{A}_q(\mathfrak{n}(\mathbf{w})) A q ( n ( w )) and have established structures in quantum group theory and related categories.
We then allow arbitrary positive braids while assuming that the Cartan datum is of finite type. In this setting, the cluster algebras are realized by the positive-braid algebras A q ( b ) \mathcal{A}_q(\mathbf{b}) A q ( b ) inside quantum virtual Grothendieck rings and carry interpretations in terms of bosonic extensions and representation theory.
Finally, we remove both reducedness and the finite Cartan-type assumption. For an arbitrary symmetrizable Cartan datum and an arbitrary positive braid, no additional quantum-group realization or categorification is used: the cluster algebra is considered on its own. § 14 shows that the folding machinery developed above is sufficient to prove the theorem in this full generality.
1* Reduced sequences and quantum unipotent coordinate algebras. Let w ∈ W g \mathbf{w}\in W_{\mathfrak{g}} w ∈ W g and i = ( i 1 , … , i r ) ∈ Red ( w ) \mathbf{i}=(i_1,\ldots,i_r)\in\operatorname{Red}(\mathbf{w}) i = ( i 1 , … , i r ) ∈ Red ( w ) . Its blockwise lift i ^ \widehat{\mathbf{i}} i is again reduced and represents an upstairs Weyl group element w ^ \widehat{\mathbf{w}} w (Corollary 2.5). The associated quantum cluster algebras are realized by
A q ( n ( w ^ ) ) and A q ( n ( w ) ) . \mathcal{A}_q(\mathfrak{n}(\widehat{\mathbf{w}})) \qquad\text{and} \qquad\mathcal{A}_q(\mathfrak{n}(\mathbf{w})). A q ( n ( w )) and A q ( n ( w )) .
Thus, in the reduced case, the folding homomorphism can be described in terms of dual PBW generators and dual canonical or upper global basis elements.
The cluster variables associated with singleton i-boxes correspond to the dual PBW generators. Both upstairs and downstairs, these generators are categorified by cuspidal modules over the corresponding quiver Hecke algebras ((Lemma 7.1)). After specialization at q = 1 q=1 q = 1 , the fixed Laurent homomorphism constructed above restricts to a surjective C \mathbb{C} C -algebra homomorphism
ϕ w F : A ( w ^ ) ⟶ A ( w ) . \phi^{\mathcal{F}}_{w}:\mathcal{A}(\widehat{w})\longrightarrow\mathcal{A}(w). ϕ w F : A ( w ) ⟶ A ( w ) .
This restriction is displayed by the commutative diagram
A ( w ^ ) → L ( Σ F ^ ) ϕ w F ↓ ↓ ϕ F A ( w ) → L ( Σ F ) \begin{CD}
\mathcal{A}(\widehat{w}) @>>> \mathcal{L}(\Sigma^{\widehat{\mathcal{F}}}) \\
@V{\phi^{\mathcal{F}}_{w}}VV @VV{\phi^{\mathcal{F}}}V \\
\mathcal{A}(w) @>>> \mathcal{L}(\Sigma^{\mathcal{F}})
\end{CD} A ( w ) ϕ w F ↓ ⏐ A ( w ) L ( Σ F ) ↓ ⏐ ϕ F L ( Σ F )
Under the polynomial presentations by dual PBW generators, the restriction is determined by
A ( w ^ ) = C [ f ^ k , s ∣ 1 ≤ k ≤ r , 1 ≤ s ≤ e i k ] , \mathcal{A}(\widehat{w})=\mathbb{C}[\widehat{f}_{k,s}\mid1\leq k\leq r,\ 1\leq s\leq e_{i_k}], A ( w ) = C [ f k , s ∣ 1 ≤ k ≤ r , 1 ≤ s ≤ e i k ] ,
A ( w ) = C [ f k 1 , … , f k r ] , ϕ w F ( f ^ k , s ) = f k i \mathcal{A}(w)=\mathbb{C}[f^1_k,\ldots,f^r_k],\qquad\phi^{\mathcal{F}}_{w}(\widehat{f}_{k,s})=f^i_k A ( w ) = C [ f k 1 , … , f k r ] , ϕ w F ( f k , s ) = f k i
(Proposition 7.2 and Theorem 7.3). Thus every individual dual PBW generator in an upstairs fiber folds to the corresponding downstairs generator. Moreover, the folding homomorphism sends the cluster monomials associated with canonical lifted i-box seeds to elements of the classical dual canonical basis B c l ( w ) \mathcal{B}^{\mathrm{cl}}(w) B cl ( w ) (Corollary 7.4).
Arbitrary positive braids in finite Cartan type and quantum (virtual) Grothendieck rings
We next assume that g \mathfrak{g} g is of finite type, while the positive braid b ∈ B g + \mathbf{b}\in B_{\mathfrak{g}}^{+} b ∈ B g + is arbitrary. The positive-braid algebras
A q ( b ^ ) and A q ( b ) \mathcal{A}_{q}(\widehat{\mathbf{b}})\qquad\text{and}\qquad\mathcal{A}_{q}(\mathbf{b}) A q ( b ) and A q ( b )
are defined through the braid-group actions on the corresponding bosonic extensions. Put A = Z [ q ± 1 / 2 ] \mathbb{A}=\mathbb{Z}[q^{\pm1/2}] A = Z [ q ± 1/2 ] . Their A \mathbb{A} A -forms are identified with the quantum cluster algebras
A A ( b ^ ) ≃ A q ( b ^ ) , A A ( b ) ≃ A q ( b ) \mathcal{A}_{\mathbb{A}}(\widehat{\mathbf{b}})\simeq\mathcal{A}_{q}(\widehat{\mathbf{b}}),\qquad
\mathcal{A}_{\mathbb{A}}(\mathbf{b})\simeq\mathcal{A}_{q}(\mathbf{b}) A A ( b ) ≃ A q ( b ) , A A ( b ) ≃ A q ( b )
(Corollaries 8.12 and 8.16).
The algebra A A ( b ^ ) \mathcal{A}_{\mathbb{A}}(\widehat{\mathbf{b}}) A A ( b ) admits a monoidal categorification by finite-dimensional representations of the quantum loop algebra of g ^ \widehat{\mathfrak{g}} g , while A A ( b ) \mathcal{A}_{\mathbb{A}}(\mathbf{b}) A A ( b ) is realized in the quantum virtual Grothendieck ring of g \mathfrak{g} g .
After specialization at q = 1 q=1 q = 1 , the ambient Grothendieck-ring folding homomorphism
π : K ( g ^ ) ⟶ K ( g ) \pi:\mathcal{K}(\widehat{\mathfrak{g}})\longrightarrow\mathcal{K}(\mathfrak{g}) π : K ( g ) ⟶ K ( g )
restricts to the positive-braid cluster algebras. More precisely, we obtain the commutative diagram
A ( b ^ ) → K ( g ^ ) → C [ y i , p ± 1 ] π b ↓ ↓ π ↓ π A ( b ) → R ( g ) → C [ y i , p ± 1 ] \begin{CD}
\mathcal{A}(\widehat{\mathbf{b}}) @>>> \mathcal{K}(\widehat{\mathfrak{g}}) @>>> \mathbb{C}[y^{\pm1}_{i,p}] \\
@V{\pi_{\mathbf{b}}}VV @VV{\pi}V @VV{\pi}V \\
\mathcal{A}(\mathbf{b}) @>>> \mathcal{R}(\mathfrak{g}) @>>> \mathbb{C}[y^{\pm1}_{i,p}]
\end{CD} A ( b ) π b ↓ ⏐ A ( b ) K ( g ) ↓ ⏐ π R ( g ) C [ y i , p ± 1 ] ↓ ⏐ π C [ y i , p ± 1 ]
Thus the folding homomorphism for every finite-type positive braid is the restriction of the single global homomorphism π \pi π .
Moreover, for every sequence i ∈ Seq ( b ) \boldsymbol{i} \in\operatorname{Seq}(\mathbf{b}) i ∈ Seq ( b ) and every maximal commuting family F \mathcal{F} F of i \boldsymbol{i} i -boxes, this restriction agrees with ϕ b F \phi_{\mathbf{b}}^{\mathcal{F}} ϕ b F :
π b = ϕ b F \pi_{\mathbf{b}}=\phi_{\mathbf{b}}^{\mathcal{F}} π b = ϕ b F
(Theorem 8.18 ).
Hence the folding homomorphisms arising from all choices of sequences and maximal commuting families are induced by the same ambient map π \pi π .
The same realization shows that cluster monomials associated with canonical lifted i \boldsymbol{i} i -box seeds map to elements of the classical global basis of R ( g ) \mathcal{R}(\mathfrak{g}) R ( g ) (Corollary 8.19 ). The homomorphism π \pi π is also compatible with the substitution isomorphisms on the ambient rational function fields associated with Dynkin quivers: the upstairs substitution formulas specialize under π \pi π to their downstairs counterparts (Corollary 8.21 ). Thus, in finite type, the cluster-theoretic folding is realized by a global homomorphism already present in the quantum virtual Grothendieck-ring framework.
Arbitrary positive braids over symmetrizable Cartan data.
Finally, let both the symmetrizable Cartan datum g \mathfrak{g} g and the positive braid b \mathbf{b} b be arbitrary.
Proposition 9.2 identifies the cluster variables attached to i \boldsymbol{i} i -boxes with the interval variables of Qin. Qin’s standard-monomial basis then gives the polynomial presentation in Proposition 9.3. Write i = ( i 1 , … , i l ) \boldsymbol{i}=(i_1,\ldots,i_l) i = ( i 1 , … , i l ) and e i = ∣ I i ∣ e_i=\lvert I_i\rvert e i = ∣ I i ∣ . 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 , let κ ( k , r ) \kappa(k,r) κ ( k , r ) be the position in i ^ \widehat{\boldsymbol{i}} i of the r r r th letter replacing i k i_k i k . Applying this proposition to i \boldsymbol{i} i and i ^ \widehat{\boldsymbol{i}} i , we obtain
A ( b ) = C [ x [ 1 ] , … , x [ l ] ] , A ( b ^ ) = C [ x [ κ ( k , r ) ] ∣ 1 ≤ k ≤ l , 1 ≤ r ≤ e i k ] . \mathcal{A}(\mathbf{b})=\mathbb{C}[x_{[1]},\ldots,x_{[l]}],\qquad
\mathcal{A}(\widehat{\mathbf{b}})=\mathbb{C}[x_{\left[\kappa(k,r)\right]}\mid1\leq k\leq l,\ 1\leq r\leq e_{i_k}]. A ( b ) = C [ x [ 1 ] , … , x [ l ] ] , A ( b ) = C [ x [ κ ( k , r ) ] ∣ 1 ≤ k ≤ l , 1 ≤ r ≤ e i k ] .
The folding homomorphism sends each upstairs singleton variable to the corresponding downstairs singleton variable, as shown in (9.6):
ϕ b F ( x ^ [ κ ( k , r ) ] ) = x [ k ] . \phi_{\mathbf{b}}^{\mathcal{F}}\left(\widehat{x}_{[\kappa(k,r)]}\right)=x_{[k]}. ϕ b F ( x [ κ ( k , r )] ) = x [ k ] .
Consequently, the fixed Laurent homomorphism restricts to the entire cluster algebra, completing the commutative diagram
A ( b ^ ) → L ( Σ F ^ ) ϕ b F ↓ ↓ ϕ F A ( b ) → L ( Σ F ) . \begin{CD}
\mathcal{A}(\widehat{\mathbf{b}}) @>>> \mathcal{L}(\Sigma^{\widehat{\mathcal{F}}}) \\
@V{\phi_{\mathbf{b}}^{\mathcal{F}}}VV @VV{\phi^{\mathcal{F}}}V \\
\mathcal{A}(\mathbf{b}) @>>> \mathcal{L}(\Sigma^{\mathcal{F}}).
\end{CD} A ( b ) ϕ b F ↓ ⏐ A ( b ) L ( Σ F ) ↓ ⏐ ϕ F L ( Σ F ) .
This proves Theorem 6.4 without reducedness or finite-type assumptions.
As a further application of the preceding framework, let A ~ q ( b ) \widetilde{\mathcal{A}}_q(\mathbf{b}) A q ( b ) denote the localization of A q ( b ) \mathcal{A}_q(\mathbf{b}) A q ( b ) at its frozen variables. In § 15 , we explicitly construct automorphisms τ b , q \tau_{\mathbf{b},q} τ b , q of A ~ q ( b ) \widetilde{\mathcal{A}}_q(\mathbf{b}) A q ( b ) for arbitrary positive braids b \mathbf{b} b over symmetrizable Cartan data, using the quantum cluster variables attached to i \boldsymbol{i} i -boxes and the skew-symmetric pairing on box vectors. These automorphisms recover the quantum twist automorphisms of Kimura–Oya [40 Theorem 6.1] when b \mathbf{b} b is the positive braid lift of a Weyl group element. The automorphism τ b , q \tau_{\mathbf{b},q} τ b , q is a twist automorphism of Donaldson–Thomas type in the sense of Kimura–Qin–Wei [41 ] and permutes the common triangular basis of A ~ q ( b ) \widetilde{\mathcal{A}}_q(\mathbf{b}) A q ( b ) in the sense of [48 Definitions 6.2.1–6.2.2] (Corollary 10.10 ). Let A ~ ( b ) \widetilde{\mathcal{A}}(\mathbf{b}) A ( b ) denote the localization of A ( b ) \mathcal{A}(\mathbf{b}) A ( b ) at its frozen variables, and let τ b \tau_{\mathbf{b}} τ b denote the automorphism of A ~ ( b ) \widetilde{\mathcal{A}}(\mathbf{b}) A ( b ) obtained by specializing τ b , q \tau_{\mathbf{b},q} τ b , q at q 1 / 2 = 1 q^{1/2}=1 q 1/2 = 1 . The folding homomorphism ϕ b F \phi_{\mathbf{b}}^{\mathcal{F}} ϕ b F extends to a surjective homomorphism ϕ b F , l o c : A ~ ( b ~ ) ↠ A ~ ( b ) \phi_b^{\mathcal F,\mathrm{loc}}:\widetilde{\mathcal A}(\widetilde b)\twoheadrightarrow\widetilde{\mathcal A}(b) ϕ b F , loc : A ( b ) ↠ A ( b ) satisfying
ϕ b F , l o c ∘ τ b ~ = τ b ∘ ϕ b F , l o c \phi_b^{\mathcal F,\mathrm{loc}}\circ\tau_{\widetilde b}=\tau_b\circ\phi_b^{\mathcal F,\mathrm{loc}} ϕ b F , loc ∘ τ b = τ b ∘ ϕ b F , loc
(Theorem 10.13 ).
The folding homomorphisms constructed here also suggest a natural direction for further study. It would be interesting to construct their categorical realizations, possibly using module categories over quiver Hecke algebras, so that the induced maps on Grothendieck rings recover these surjective homomorphisms. Such realizations could provide a mechanism for relating the well-established simply-laced structures to their counterparts for the quotient Cartan datum. More broadly, the folding machinery developed here provides a starting point for studying global and canonical bases, and their categorical realizations, under folding.
We conclude with the organization of the paper. Sections 1–3 develop the folding of Cartan and root data and the compatible-pair machinery for i -box seeds. Sections 4 and 5 establish its compatibility with braid moves, box moves, and individual cluster variables. Section 6 combines these results to prove the folding theorem for cluster monomials.
Sections 7 and 8 give, respectively, the quantum-unipotent and finite-type positive-braid interpretations. Section 9 proves the general commutative folding theorem. Section 10 constructs quantum twist automorphisms for arbitrary positive braids, proves that they preserve the common triangular bases, and establishes compatibility with folding after commutative specialization. Finally, Appendix A contains the additional calculations for braid moves with orbit sizes (3, 3) and (4, 2).
Acknowledgements
The authors thank Masaki Kashiwara, Euiyong Park, and Myungho Kim for helpful discussions. They used SageMath [54 ] for computations in the examples.
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 ] .
Cartan data and admissible Dynkin foldings
In this section, we fix the notation for Cartan data and admissible Dynkin foldings that will be used throughout this paper.
Symmetrizable Cartan data
Let I I I be a finite index set. A Cartan datum ( C , P , Π , P ∨ , Π ∨ ) (\mathrm C,\mathrm P,\Pi,\mathrm P^\vee,\Pi^\vee) ( C , P , Π , P ∨ , Π ∨ ) consists of
(a) a symmetrizable Cartan matrix C = ( c i j ) i , j ∈ I \mathrm C=(c_{ij})_{i,j\in I} C = ( c ij ) i , j ∈ I ; i.e., D C \mathrm D\mathrm C DC is symmetric for a diagonal matrix D = diag ( d i ∈ Z ≥ 1 ∣ i ∈ I ) \mathrm D=\operatorname{diag}(d_i\in\mathbb Z_{\ge1}\mid i\in I) D = diag ( d i ∈ Z ≥ 1 ∣ i ∈ I ) ,
(b) a free abelian group P \mathrm P P , called the weight lattice ,
(c) Π = { α i ∣ i ∈ I } \Pi=\{\alpha_i\mid i\in I\} Π = { α i ∣ i ∈ I } , called the set of simple roots ,
(d) Π ∨ = { h i ∣ i ∈ I } ⊂ P ∨ : = Hom ( P , Z ) \Pi^\vee=\{h_i\mid i\in I\}\subset\mathrm P^\vee:=\operatorname{Hom}(\mathrm P,\mathbb Z) Π ∨ = { h i ∣ i ∈ I } ⊂ P ∨ := Hom ( P , Z ) , called the set of simple coroots ,
(e) a Q \mathbb Q Q -valued symmetric bilinear form ( ⋅ , ⋅ ) (\cdot,\cdot) ( ⋅ , ⋅ ) on P \mathrm P P , satisfying the standard property (see [26 §1.1] for instance). Here we take D D D such that
d i = ( α i , α i ) / 2 , ( α i , α j ) = d i c i j ( i , j ∈ I ) , (3) d_i=(\alpha_i,\alpha_i)/2,\qquad(\alpha_i,\alpha_j)=d_i c_{ij}\qquad(i,j\in I), \tag*{(3)} d i = ( α i , α i ) /2 , ( α i , α j ) = d i c ij ( i , j ∈ I ) , (3)
and normalize it by gcd i ∈ I ( d i ) = 1 \gcd_{i\in I}(d_i)=1 g cdi ∈ I ( d i ) = 1 .
For i ∈ I i\in I i ∈ I , we choose ϖ i ∈ P \varpi_i\in P ϖ i ∈ P such that ⟨ h i , ϖ j ⟩ = δ i j \langle h_i,\varpi_j\rangle=\delta_{ij} ⟨ h i , ϖ j ⟩ = δ ij for any j ∈ I j\in I j ∈ I and call it the i i i -th fundamental weight . We denote by P + : = { λ ∈ P ∣ ⟨ h i , λ ⟩ ≥ 0 for all i ∈ I } P^+:=\{\lambda\in P\mid\langle h_i,\lambda\rangle\geq0\text{ for all }i\in I\} P + := { λ ∈ P ∣ ⟨ h i , λ ⟩ ≥ 0 for all i ∈ I } the set of dominant integral weights . The free abelian group Q : = ⨁ i ∈ I Z α i Q:=\bigoplus_{i\in I}\mathbb{Z}\alpha_i Q := ⨁ i ∈ I Z α i is called the root lattice and we set Q + = ∑ i ∈ I Z ≥ 0 α i ⊂ Q Q^+=\sum_{i\in I}\mathbb{Z}_{\geq0}\alpha_i\subset Q Q + = ∑ i ∈ I Z ≥ 0 α i ⊂ Q and Q − = ∑ i ∈ I Z ≤ 0 α i ⊂ Q Q^-=\sum_{i\in I}\mathbb{Z}_{\leq0}\alpha_i\subset Q Q − = ∑ i ∈ I Z ≤ 0 α i ⊂ Q . We denote by Φ \Phi Φ the set of roots and by Φ ± \Phi^\pm Φ ± the set of positive roots (resp. negative roots). We also denote by Φ r e \Phi_{\mathrm{re}} Φ re the set of real roots. We set Φ r e ± : = Φ r e ∩ Φ ± \Phi_{\mathrm{re}}^\pm:=\Phi_{\mathrm{re}}\cap\Phi^\pm Φ re ± := Φ re ∩ Φ ± .
Let g \mathfrak{g} g be the Kac-Moody algebra associated with the Cartan datum ( C , P , Π , P ∨ , Π ∨ ) (C,P,\Pi,P^\vee,\Pi^\vee) ( C , P , Π , P ∨ , Π ∨ ) . We say that g \mathfrak{g} g is simply-laced if c i j < 0 c_{ij}<0 c ij < 0 implies c i j = c j i = − 1 c_{ij}=c_{ji}=-1 c ij = c j i = − 1 for all i ≠ j i\ne j i = j . We denote by W W W the Weyl group of g \mathfrak{g} g . It is generated by the simple reflections s i ∈ Aut ( P ) s_i\in\operatorname{Aut}(P) s i ∈ Aut ( P ) ( i ∈ I ) (i\in I) ( i ∈ I ) defined by s i ( λ ) = λ − ⟨ h i , λ ⟩ α i s_i(\lambda)=\lambda-\langle h_i,\lambda\rangle\alpha_i s i ( λ ) = λ − ⟨ h i , λ ⟩ α i for λ ∈ P \lambda\in P λ ∈ P . In particular,
s i ( α j ) = α j − c i j α i ( i , j ∈ I ) . (4) s_i(\alpha_j)=\alpha_j-c_{ij}\alpha_i\qquad(i,j\in I). \tag*{(4)} s i ( α j ) = α j − c ij α i ( i , j ∈ I ) . (4)
We denote by B g = ⟨ r i ± 1 ⟩ i ∈ I B_{\mathfrak{g}}=\langle r_i^{\pm1}\rangle_{i\in I} B g = ⟨ r i ± 1 ⟩ i ∈ I the braid group associated with g \mathfrak{g} g ; i.e, it is the group generated by { r i ± 1 } i ∈ I \{r_i^{\pm1}\}_{i\in I} { r i ± 1 } i ∈ I subject to the following relations:
(i) r i r i − 1 = r i − 1 r i = 1 and (ii) r i r j ⋯ ⏟ m i , j -times = r j r i ⋯ ⏟ m i , j -times for i ≠ j ∈ I , (5) \text{(i) }\ r_i r_i^{-1}=r_i^{-1}r_i=1\qquad\text{and}\qquad\text{(ii) }\ \underbrace{r_i r_j\cdots}_{m_{i,j}\text{-times}}=\underbrace{r_j r_i\cdots}_{m_{i,j}\text{-times}}\text{ for }i\ne j\in I, \tag*{(5)} (i) r i r i − 1 = r i − 1 r i = 1 and (ii) m i , j -times r i r j ⋯ = m i , j -times r j r i ⋯ for i = j ∈ I , (5)
where
m i , j : = { c i , j c j , i + 2 if c i , j c j , i ≤ 2 , 6 if c i , j c j , i = 3 , ∞ otherwise. m_{i,j}:=
\begin{cases}
c_{i,j}c_{j,i}+2 & \text{if }c_{i,j}c_{j,i}\leq2,\\
6 & \text{if }c_{i,j}c_{j,i}=3,\\
\infty& \text{otherwise.}
\end{cases} m i , j := ⎩ ⎨ ⎧ c i , j c j , i + 2 6 ∞ if c i , j c j , i ≤ 2 , if c i , j c j , i = 3 , otherwise.
We call the relations in (5) (ii) the braid relations of B g B_{\mathfrak{g}} B g . The Weyl group W g W_{\mathfrak{g}} W g is the quotient of B g B_{\mathfrak{g}} B g obtained by imposing r i 2 = 1 r_i^2=1 r i 2 = 1 for all i ∈ I i\in I i ∈ I ; we denote the image of r i r_i r i in W g W_{\mathfrak{g}} W g by s i s_i s i .
We denote by B g + B_{\mathfrak{g}}^+ B g + the positive braid monoid generated by { r i } i ∈ I \{r_i\}_{i\in I} { r i } i ∈ I . There is a natural surjective monoid homomorphism
pr W : B g + ⟶ W g , r i ⟼ s i . \operatorname{pr}_W:B_{\mathfrak{g}}^+\longrightarrow W_{\mathfrak{g}},\qquad r_i\longmapsto s_i. pr W : B g + ⟶ W g , r i ⟼ s i .
For a sequence i = ( i 1 , … , i r ) ∈ I r \mathbf{i}=(i_1,\ldots,i_r)\in I^r i = ( i 1 , … , i r ) ∈ I r , we write
r i : = r i 1 ⋯ r i r ∈ B g + , s i : = s i 1 ⋯ s i r ∈ W g . r_{\mathbf{i}}:=r_{i_1}\cdots r_{i_r}\in B_{\mathfrak{g}}^+,\qquad s_{\mathbf{i}}:=s_{i_1}\cdots s_{i_r}\in W_{\mathfrak{g}}. r i := r i 1 ⋯ r i r ∈ B g + , s i := s i 1 ⋯ s i r ∈ W g .
For w ∈ W w\in W w ∈ W and b ∈ B g + \mathbf{b}\in B_{\mathfrak{g}}^+ b ∈ B g + , set
Red ( w ) : = { i = ( i 1 , … , i r ) ∈ I r ∣ s i = w , r = ℓ ( w ) } and \operatorname{Red}(w):=\{\mathbf{i}=(i_1,\ldots,i_r)\in I^r\mid s_{\mathbf{i}}=w,\ r=\ell(w)\}\qquad\text{and} Red ( w ) := { i = ( i 1 , … , i r ) ∈ I r ∣ s i = w , r = ℓ ( w )} and
Seq ( b ) : = { i = ( i 1 , … , i r ) ∈ I r ∣ r i = b } . \operatorname{Seq}(\mathbf{b}):=\{\mathbf{i}=(i_1,\ldots,i_r)\in I^r\mid r_{\mathbf{i}}=\mathbf{b}\}. Seq ( b ) := { i = ( i 1 , … , i r ) ∈ I r ∣ r i = b } .
Thus Red ( w ) \operatorname{Red}(w) Red ( w ) is the set of reduced sequences of w w w , while Seq ( b ) \operatorname{Seq}(\mathbf{b}) Seq ( b ) is the set of all finite sequences realizing b \mathbf{b} b . Thus, an arbitrary sequence i \mathbf{i} i may always be regarded as an element r i r_{\mathbf{i}} r i of the positive braid monoid B g + B_{\mathfrak{g}}^+ B g + . If i \mathbf{i} i is reduced, then r i r_{\mathbf{i}} r i depends only on s i s_{\mathbf{i}} s i . Thus, in this case, we shall identify s i ∈ W g s_{\mathbf{i}}\in W_{\mathfrak{g}} s i ∈ W g with its positive braid lift r i ∈ B g + r_{\mathbf{i}}\in B_{\mathfrak{g}}^+ r i ∈ B g + whenever no confusion can arise. We usually drop the subscript g \mathfrak{g} g from the notation when there is no danger of confusion.
Admissible Dynkin foldings and quotient Cartan data
Let g \mathfrak{g} g be a symmetrizable Kac–Moody algebra. We denote by △ \triangle △ the Dynkin diagram of g \mathfrak{g} g .
Let Γ \Gamma Γ be a finite group acting on △ \triangle △ by diagram automorphisms, i.e.,
c g ( i ) , g ( j ) = c i j for all g ∈ Γ and i , j ∈ I . c_{g(i),g(j)}=c_{ij}\quad\text{for all }g\in\Gamma\text{ and }i,j\in I. c g ( i ) , g ( j ) = c ij for all g ∈ Γ and i , j ∈ I .
We call the action of Γ \Gamma Γ admissible if
c i , g ( i ) = 0 for all i ∈ I and g ∈ Γ with g ( i ) ≠ i . c_{i,g(i)}=0\quad\text{for all }i\in I\text{ and }g\in\Gamma\text{ with }g(i)\ne i. c i , g ( i ) = 0 for all i ∈ I and g ∈ Γ with g ( i ) = i .
From now on, 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 ^ a , b ) a , b ∈ I ^ \widehat{C}=(\widehat{c}_{a,b})_{a,b\in\widehat{I}} C = ( c a , b ) a , b ∈ I . We normalize the bilinear form (3) of g ^ \widehat{\mathfrak{g}} g so that ( α i , α i ) g ^ = 2 (\alpha_i,\alpha_i)_{\widehat{\mathfrak{g}}}=2 ( α i , α i ) g = 2 for all i ∈ I ^ i\in\widehat{I} i ∈ I .
Assume that a finite group Γ \Gamma Γ acts admissibly on the Dynkin diagram of g ^ \widehat{\mathfrak{g}} g .
We identify the orbit set I ^ / Γ \widehat{I}/\Gamma I /Γ with an index set I I I and denote by
π : I ^ ⟶ I \pi:\widehat{I}\longrightarrow I π : I ⟶ I
the canonical projection. For i ∈ I i\in I i ∈ I , we set
I i : = π − 1 ( i ) and e i : = ∣ I i ∣ . (6) I_i:=\pi^{-1}(i)\quad\text{and}\quad e_i:=|I_i|. \tag*{(6)} I i := π − 1 ( i ) and e i := ∣ I i ∣. (6)
Definition 1.1. The quotient Cartan matrix is the matrix
C = ( c i j ) i , j ∈ I defined by c i j : = ∑ a ∈ I i c ^ a , b , b ∈ I j . (7) C=(c_{ij})_{i,j\in I}\quad\text{defined by}\quad c_{ij}:=\sum_{a\in I_i}\widehat{c}_{a,b},\quad b\in I_j. \tag*{(7)} C = ( c ij ) i , j ∈ I defined by c ij := a ∈ I i ∑ c a , b , b ∈ I j . (7)
Lemma 1.2. The matrix C = ( c i j ) i , j ∈ I C=(c_{ij})_{i,j\in I} C = ( c ij ) i , j ∈ I in (7) is a symmetrizable Cartan matrix. More precisely, there exist positive integers d i d_i d i (i ∈ I i\in I i ∈ I ) with gcd i ∈ I { d i } = 1 \gcd_{i\in I}\{d_i\}=1 g cdi ∈ I { d i } = 1 such that
d i c i j = d j c j i ( i , j ∈ I ) . d_i c_{ij}=d_j c_{ji}\quad(i,j\in I). d i c ij = d j c j i ( i , j ∈ I ) . Moreover, there exists a positive integer m m m such that
d i e i = m ( i ∈ I ) . (8) d_i e_i=m\quad(i\in I). \tag*{(8)} d i e i = m ( i ∈ I ) . (8) Proof. The definition of c i j c_{ij} c ij is independent of the choice of b ∈ I j b\in I_j b ∈ I j , since Γ \Gamma Γ acts transitively on each orbit I j I_j I j and preserves C ^ \widehat{C} C .
By admissibility, for b ∈ I i b\in I_i b ∈ I i , c i i = ∑ a ∈ I i c ^ a , b = c ^ b , b = 2 c_{ii}=\sum_{a\in I_i}\widehat{c}_{a,b}=\widehat{c}_{b,b}=2 c ii = ∑ a ∈ I i c a , b = c b , b = 2 . For i ≠ j i\ne j i = j , we have c i j ∈ Z ≤ 0 c_{ij}\in\mathbb{Z}_{\leq0} c ij ∈ Z ≤ 0 , and clearly c i j = 0 c_{ij}=0 c ij = 0 if and only if c j i = 0 c_{ji}=0 c j i = 0 . Hence C C C is a generalized Cartan matrix.
Since C ^ \widehat{C} C is symmetric, counting the edges between I i I_i I i and I j I_j I j in two ways gives
e j ( − c i j ) = e i ( − c j i ) ⟺ e j c i j = e i c j i . (9) e_j(-c_{ij})=e_i(-c_{ji})\Longleftrightarrow e_jc_{ij}=e_ic_{ji}. \tag*{(9)} e j ( − c ij ) = e i ( − c j i ) ⟺ e j c ij = e i c j i . (9) Thus C C C is symmetrizable.
Since ( L / e i ) i ∈ I (L/e_i)_{i\in I} ( L / e i ) i ∈ I , where L = lcm { e i ∣ i ∈ I } L=\operatorname{lcm}\{e_i\mid i\in I\} L = lcm { e i ∣ i ∈ I } , is an integral symmetrizer, set g = gcd i ∈ I ( L / e i ) g=\gcd_{i\in I}(L/e_i) g = g cdi ∈ I ( L / e i ) , d i = ( L / e i ) / g d_i=(L/e_i)/g d i = ( L / e i ) / g , and m = L / g m=L/g m = L / g . Then gcd i ∈ I d i = 1 \gcd_{i\in I}d_i=1 g cdi ∈ I d i = 1 and d i e i = m d_i e_i=m d i e i = m (i ∈ I i\in I i ∈ I ). □
Remark 1.3. Every symmetrizable Cartan matrix can be obtained as the quotient of a simply-laced Cartan matrix by an admissible action. Indeed, applying [52 Proposition 3.2] to C t r C^{\mathsf{tr}} C tr gives a simple graph with a finite-order diagram automorphism generating an admissible action whose quotient Cartan matrix under (7) is C C C .
In what follows, we mainly consider the pair ( g ^ , g ) (\widehat{\mathfrak{g}},\mathfrak{g}) ( g , g ) arising from an admissible Γ \Gamma Γ -action on g ^ \widehat{\mathfrak{g}} g , where g \mathfrak{g} g denotes the Kac–Moody algebra associated with the quotient Cartan matrix C C C in (7) .
Examples
In this subsection, we collect several examples of pairs ( g ^ , g ) (\widehat{\mathfrak{g}},\mathfrak{g}) ( g , g ) arising from admissible Dynkin foldings, which will be used throughout the paper. We describe them mainly by their Dynkin diagrams, together with the Γ \Gamma Γ -orbits and the corresponding quotient diagrams. Besides the standard finite and affine foldings, we include examples illustrating non-cyclic group actions, fibers of larger cardinality, and indefinite Cartan data.
We first consider the classical folding pair ( A 2 n − 1 , C n ) (A_{2n-1},C_n) ( A 2 n − 1 , C n ) . Let σ \sigma σ be the involutive diagram automorphism of A 2 n − 1 A_{2n-1} A 2 n − 1 given by
σ ( i ) = 2 n − i ( 1 ≤ i ≤ 2 n − 1 ) . \sigma(i)=2n-i \qquad(1\le i\le2n-1). σ ( i ) = 2 n − i ( 1 ≤ i ≤ 2 n − 1 ) .
Then the corresponding admissible folding is
A 2 n − 1 / ⟨ σ ⟩ ⟹ C n (10) A_{2n-1}\mathbin{/ \langle\sigma\rangle}\Longrightarrow C_n \tag*{(10)} A 2 n − 1 / ⟨ σ ⟩ ⟹ C n (10)
Here θ k \theta_k θ k means that ( α k , α k ) = t (\alpha_k,\alpha_k)=t ( α k , α k ) = t and dotted arrows indicate the action of the diagram automorphism σ \sigma σ . In this case, the Γ = ⟨ σ ⟩ \Gamma=\langle\sigma\rangle Γ = ⟨ σ ⟩ -orbits are I i = { i , 2 n − i } I_i=\{i,2n-i\} I i = { i , 2 n − i } ( 1 ≤ i ≤ n − 1 ) (1\le i\le n-1) ( 1 ≤ i ≤ n − 1 ) , and I n = { n } I_n=\{n\} I n = { n } , so that
( e 1 , … , e n ) = ( 2 , … , 2 , 1 ) . (e_1,\ldots,e_n)=(2,\ldots,2,1). ( e 1 , … , e n ) = ( 2 , … , 2 , 1 ) .
The quotient Cartan matrix is of type C n C_n C n , with normalized symmetrizer ( d 1 , … , d n ) = ( 1 , … , 1 , 2 ) (d_1,\ldots,d_n)=(1,\ldots,1,2) ( d 1 , … , d n ) = ( 1 , … , 1 , 2 ) . In particular, d i e i = 2 d_i e_i=2 d i e i = 2 ( 1 ≤ i ≤ n ) (1\le i\le n) ( 1 ≤ i ≤ n ) .
The other standard finite-type admissible foldings are depicted below:
D n + 1 / ⟨ σ ⟩ ⟹ B n D_{n+1}\mathbin{/ \langle\sigma\rangle}\Longrightarrow B_n D n + 1 / ⟨ σ ⟩ ⟹ B n
E 6 / ⟨ σ ⟩ ⟹ F 4 E_6\mathbin{/ \langle\sigma\rangle}\Longrightarrow F_4 E 6 / ⟨ σ ⟩ ⟹ F 4
D 4 / ⟨ σ ⟩ ⟹ G 2 D_4\mathbin{/ \langle\sigma\rangle}\Longrightarrow G_2 D 4 / ⟨ σ ⟩ ⟹ G 2
For later reference, the orbit fibers and their cardinalities in the above three foldings are summarized as follows:
Table 1.
Several affine admissible foldings which will be useful later are depicted below:
Figure 1.
Unlike the preceding three examples, the last folding is induced by a non-cyclic group action. More precisely,
Γ = ⟨ σ 1 , σ 2 ⟩ ≅ C 2 × C 2 , \Gamma= \langle\sigma_1,\sigma_2\rangle\cong C_2 \times C_2, Γ = ⟨ σ 1 , σ 2 ⟩ ≅ C 2 × C 2 ,
and the orbit fibers are
I 0 = { 0 , 1 } , I i = { i + 1 } ( 1 ≤ i ≤ n − 3 ) , I n − 2 = { n − 1 , n } . I_0 = \{0,1\}, \qquad I_i = \{i+1\}\quad(1 \le i \le n-3), \qquad I_{n-2} = \{n-1,n\}. I 0 = { 0 , 1 } , I i = { i + 1 } ( 1 ≤ i ≤ n − 3 ) , I n − 2 = { n − 1 , n } .
Hence
( e 0 , … , e n − 2 ) = ( 2 , 1 , … , 1 , 2 ) , ( d 0 , … , d n − 2 ) = ( 1 , 2 , … , 2 , 1 ) , (e_0,\ldots,e_{n-2})=(2,1,\ldots,1,2), \qquad(d_0,\ldots,d_{n-2})=(1,2,\ldots,2,1), ( e 0 , … , e n − 2 ) = ( 2 , 1 , … , 1 , 2 ) , ( d 0 , … , d n − 2 ) = ( 1 , 2 , … , 2 , 1 ) ,
so that m = 2 m=2 m = 2 in (8) .
Notice that in the folding E 6 ( 1 ) → D 4 ( 3 ) E_6^{(1)} \to D_4^{(3)} E 6 ( 1 ) → D 4 ( 3 ) , the two adjacent quotient vertices 0 and 1 of D 4 ( 3 ) D_4^{(3)} D 4 ( 3 ) have fibers of cardinality three. Some further standard admissible affine foldings are
( A 3 ( 1 ) , A 1 ( 1 ) ) , ( A 2 n − 1 ( 1 ) , C n ( 1 ) ) n ≥ 2 , ( D 2 n ( 1 ) , A 2 n − 1 ( 2 ) ) n ≥ 2 , ( D n + 1 ( 1 ) , B n ( 1 ) ) n ≥ 3 , ( E 6 ( 1 ) , F 4 ( 1 ) ) , (A_3^{(1)},A_1^{(1)}),\quad(A_{2n-1}^{(1)},C_n^{(1)})_{n\ge2},\quad(D_{2n}^{(1)},A_{2n-1}^{(2)})_{n\ge2},\quad(D_{n+1}^{(1)},B_n^{(1)})_{n\ge3},\quad(E_6^{(1)},F_4^{(1)}), ( A 3 ( 1 ) , A 1 ( 1 ) ) , ( A 2 n − 1 ( 1 ) , C n ( 1 ) ) n ≥ 2 , ( D 2 n ( 1 ) , A 2 n − 1 ( 2 ) ) n ≥ 2 , ( D n + 1 ( 1 ) , B n ( 1 ) ) n ≥ 3 , ( E 6 ( 1 ) , F 4 ( 1 ) ) ,
where
in all five cases.
Finally, admissible folding is not restricted to finite or affine Cartan data. The following is a minimal simply-laced indefinite example with a nontrivial admissible action:
Figure 1.9.
Remark 1.4. Neither the simply-laced Cartan datum g ^ \widehat{\mathfrak g} g nor the quotient Cartan datum g \mathfrak g g is determined by the other alone; the quotient depends also on the choice of an admissible group action Γ \Gamma Γ . Conversely, a given Cartan datum g \mathfrak g g may admit different admissible simply-laced realizations. The folding results below apply to any such realization and give the same statements for the quotient datum g \mathfrak g g .
For example, C 2 ( 1 ) C_{2}^{(1)} C 2 ( 1 ) admits admissible simply-laced realizations of types A 7 ( 1 ) A_{7}^{(1)} A 7 ( 1 ) and A 3 ( 1 ) A_{3}^{(1)} A 3 ( 1 ) , with m = 4 m=4 m = 4 and 2 2 2 , respectively.
Figure 2 (PDF p. 14).
Here C 2 × C 2 ≃ ⟨ τ , σ ⟩ C_2 \times C_2 \simeq\langle\tau,\sigma\rangle C 2 × C 2 ≃ ⟨ τ , σ ⟩ , where σ = ( 17 ) ( 26 ) ( 35 ) \sigma=(17)(26)(35) σ = ( 17 ) ( 26 ) ( 35 ) and τ = ( 04 ) ( 13 ) ( 57 ) \tau=(04)(13)(57) τ = ( 04 ) ( 13 ) ( 57 ) . For the left realization, the three fibers over 0 , 1 , 2 0,1,2 0 , 1 , 2 are { 0 , 4 } \{0,4\} { 0 , 4 } , { 1 , 3 , 5 , 7 } \{1,3,5,7\} { 1 , 3 , 5 , 7 } , and { 2 , 6 } \{2,6\} { 2 , 6 } , respectively.
Throughout this paper, in finite and affine types, we use the realizations fixed in the examples above.
Folding of root data
Throughout this section, we retain the notation and conventions of § 5 . We establish the folding properties of root sequences and their associated pairings.
Weyl groups and root lattices under folding
2.1. Weyl groups and root lattices under folding. Let W ^ \widehat{W} W and W W W be the Weyl groups of g ^ \widehat{\mathfrak{g}} g and g \mathfrak{g} g , respectively, and let B ^ + \widehat{B}^{+} B + and B + B^{+} B + be the corresponding positive braid monoids. For each i ∈ I i \in I i ∈ I , set
s ^ i : = ∏ a ∈ I i s a ∈ W ^ and r ^ i : = ∏ a ∈ I i r a ∈ B ^ + . \widehat{s}_i:=\prod_{a\in I_i}s_a\in\widehat{W}\quad\text{and}\quad\widehat{r}_i:=\prod_{a\in I_i}r_a\in\widehat{B}^{+}. s i := a ∈ I i ∏ s a ∈ W and r i := a ∈ I i ∏ r a ∈ B + .
By admissibility, the factors in the above product commute.
We use the same symbol π \pi π for the induced Z \mathbb{Z} Z -linear map
π : Q ^ ⟶ Q , π ( α ^ a ) = α i ( a ∈ I i ) . \pi:\widehat{Q}\longrightarrow Q,\qquad\pi(\widehat{\alpha}_a)=\alpha_i\quad(a\in I_i). π : Q ⟶ Q , π ( α a ) = α i ( a ∈ I i ) .
We also define the Q \mathbb{Q} Q -linear averaging map
ȷ : Q Q ⟶ Q Q ^ , ȷ ( α i ) = 1 e i ∑ a ∈ I i α ^ a . (11) \jmath:\mathbb{Q}Q\longrightarrow\mathbb{Q}\widehat{Q},\qquad\jmath(\alpha_i)=\frac{1}{e_i}\sum_{a\in I_i}\widehat{\alpha}_a. \tag*{(11)} : Q Q ⟶ Q Q , ( α i ) = e i 1 a ∈ I i ∑ α a . (11)
Lemma 2.1. For every i ∈ I i\in I i ∈ I , we have
π ∘ s ^ i = s i ∘ π and ȷ ∘ s i = s ^ i ∘ ȷ . \pi\circ\widehat{s}_i=s_i\circ\pi\quad\text{and}\quad\jmath\circ s_i=\widehat{s}_i\circ\jmath. π ∘ s i = s i ∘ π and ∘ s i = s i ∘ . Proof. Let b ∈ I j b\in I_j b ∈ I j . By admissibility, we have s ^ i ( α ^ b ) = α ^ b − ∑ a ∈ I i c ^ a , b α ^ a \widehat{s}_i(\widehat{\alpha}_b)=\widehat{\alpha}_b-\sum_{a\in I_i}\widehat{c}_{a,b}\widehat{\alpha}_a s i ( α b ) = α b − ∑ a ∈ I i c a , b α a . Hence
π ( s ^ i ( α ^ b ) ) = α j − ( ∑ a ∈ I i c ^ a , b ) α i = α j − c i j α i = s i ( α j ) , \pi\bigl(\widehat{s}_i(\widehat{\alpha}_b)\bigr)=\alpha_j-\left(\sum_{a\in I_i}\widehat{c}_{a,b}\right)\alpha_i=\alpha_j-c_{ij}\alpha_i=s_i(\alpha_j), π ( s i ( α b ) ) = α j − ( a ∈ I i ∑ c a , b ) α i = α j − c ij α i = s i ( α j ) , which proves π ∘ s ^ i = s i ∘ π \pi\circ\widehat{s}_i=s_i\circ\pi π ∘ s i = s i ∘ π .
On the other hand,
s ^ i ( ȷ ( α j ) ) = 1 e j ∑ b ∈ I j ( α ^ b − ∑ a ∈ I i c ^ a , b α ^ a ) = ȷ ( α j ) − c j i e j ∑ a ∈ I i \widehat{s}_i\bigl(\jmath(\alpha_j)\bigr)=\frac{1}{e_j}\sum_{b\in I_j}\left(\widehat{\alpha}_b-\sum_{a\in I_i}\widehat{c}_{a,b}\widehat{\alpha}_a\right)=\jmath(\alpha_j)-\frac{c_{ji}}{e_j}\sum_{a\in I_i} s i ( ( α j ) ) = e j 1 b ∈ I j ∑ ( α b − a ∈ I i ∑ c a , b α a ) = ( α j ) − e j c j i a ∈ I i ∑
= ∗ J ( α j ) − c i j e i ∑ a ∈ I i α ^ a = J ( α j − c i j α i ) = J ( s i ( α j ) ) , \overset{*}{=} J(\alpha_j)-\frac{c_{ij}}{e_i}\sum_{a\in I_i}\widehat{\alpha}_a=J(\alpha_j-c_{ij}\alpha_i)=J(s_i(\alpha_j)), = ∗ J ( α j ) − e i c ij a ∈ I i ∑ α a = J ( α j − c ij α i ) = J ( s i ( α j )) ,
where = ∗ \overset{*}{=} = ∗ holds by e j c i j = e i c j i e_jc_{ij}=e_ic_{ji} e j c ij = e i c j i .
Recall from Lemma 1.2 that
d i e i = m ( i ∈ I ) d_ie_i=m\qquad(i\in I) d i e i = m ( i ∈ I )
for a positive integer m m m independent of i i i .
Lemma 2.2. For α , β ∈ Q \alpha,\beta\in Q α , β ∈ Q , we have
( α , β ) g = m ( J ( α ) , J ( β ) ) g ^ . (\alpha,\beta)_{\mathfrak g}=m\bigl(J(\alpha),J(\beta)\bigr)_{\widehat{\mathfrak g}}. ( α , β ) g = m ( J ( α ) , J ( β ) ) g . Proof. For i , j ∈ I i,j\in I i , j ∈ I , we have
m ( J ( α i ) , J ( α j ) ) g ^ = m e i e j ∑ a ∈ I i ∑ b ∈ I j ( α ^ a , α ^ b ) g ^ = m e i c i j = d i c i j = ( α i , α j ) g . m\bigl(J(\alpha_i),J(\alpha_j)\bigr)_{\widehat{\mathfrak g}}
=\frac{m}{e_ie_j}\sum_{a\in I_i}\sum_{b\in I_j}(\widehat{\alpha}_a,\widehat{\alpha}_b)_{\widehat{\mathfrak g}}
=\frac{m}{e_i}c_{ij}=d_ic_{ij}=(\alpha_i,\alpha_j)_{\mathfrak g}. m ( J ( α i ) , J ( α j ) ) g = e i e j m a ∈ I i ∑ b ∈ I j ∑ ( α a , α b ) g = e i m c ij = d i c ij = ( α i , α j ) g .
Remark 2.3. Throughout this paper, we fix the finite and affine realizations listed below, with the corresponding values of m m m :
m = 2 for ( A 2 n − 1 , C n ) , ( D n + 1 , B n ) , ( E 6 , F 4 ) , ( A 2 n − 1 ( 1 ) , C n ( 1 ) ) , ( A 3 ( 1 ) , A 1 ( 1 ) ) , m=2\quad\text{for }(A_{2n-1},C_n),\ (D_{n+1},B_n),\ (E_6,F_4),\ (A_{2n-1}^{(1)},C_n^{(1)}),\ (A_3^{(1)},A_1^{(1)}), m = 2 for ( A 2 n − 1 , C n ) , ( D n + 1 , B n ) , ( E 6 , F 4 ) , ( A 2 n − 1 ( 1 ) , C n ( 1 ) ) , ( A 3 ( 1 ) , A 1 ( 1 ) ) ,
( D n ( 1 ) , D n − 1 ( 2 ) ) , ( D 2 n ( 1 ) , A 2 n − 1 ( 2 ) ) , ( D n + 1 ( 1 ) , B n ( 1 ) ) , ( E 6 ( 1 ) , F 4 ( 1 ) ) , (D_n^{(1)},D_{n-1}^{(2)}),\ (D_{2n}^{(1)},A_{2n-1}^{(2)}),\ (D_{n+1}^{(1)},B_n^{(1)}),\ (E_6^{(1)},F_4^{(1)}), ( D n ( 1 ) , D n − 1 ( 2 ) ) , ( D 2 n ( 1 ) , A 2 n − 1 ( 2 ) ) , ( D n + 1 ( 1 ) , B n ( 1 ) ) , ( E 6 ( 1 ) , F 4 ( 1 ) ) ,
m = 3 for ( D 4 , G 2 ) , ( D 4 ( 1 ) , G 2 ( 1 ) ) , ( E 6 ( 1 ) , D 4 ( 3 ) ) , m=3\quad\text{for }(D_4,G_2),\ (D_4^{(1)},G_2^{(1)}),\ (E_6^{(1)},D_4^{(3)}), m = 3 for ( D 4 , G 2 ) , ( D 4 ( 1 ) , G 2 ( 1 ) ) , ( E 6 ( 1 ) , D 4 ( 3 ) ) ,
m = 4 for ( D 2 n + 2 ( 1 ) , A 2 n ( 2 ) ) . m=4\quad\text{for }(D_{2n+2}^{(1)},A_{2n}^{(2)}). m = 4 for ( D 2 n + 2 ( 1 ) , A 2 n ( 2 ) ) .
In particular, m m m need not be the order of Γ \Gamma Γ : for ( D n ( 1 ) , D n − 1 ( 2 ) ) (D_n^{(1)},D_{n-1}^{(2)}) ( D n ( 1 ) , D n − 1 ( 2 ) ) , we have Γ ≃ C 2 × C 2 \Gamma\simeq C_2\times C_2 Γ ≃ C 2 × C 2 but m = 2 m=2 m = 2 .
Arbitrary sequences and their lifts
For each i ∈ I g i\in I_{\mathfrak g} i ∈ I g , fix an ordering
I i = ( ı 1 , … , ı e i ) ⊂ I g ^ . I_i=(\imath_1,\ldots,\imath_{e_i})\subset I_{\widehat{\mathfrak g}}. I i = ( 1 , … , e i ) ⊂ I g .
We use ı \imath for an index of g ^ \widehat{\mathfrak g} g when we wish to distinguish it from an index i i i of g \mathfrak g g .
Let i = ( i 1 , … , i l ) ∈ I g l \mathbf{i}=(i_1,\ldots,i_l)\in I_{\mathfrak g}^l i = ( i 1 , … , i l ) ∈ I g l be an arbitrary finite sequence. We do not assume that i \mathbf{i} i is reduced.
For each 1 ≤ k ≤ l 1\leq k\leq l 1 ≤ k ≤ l , write I i k = ( ı k , 1 , … , ı k , e i k ) I_{i_k}=(\imath_{k,1},\ldots,\imath_{k,e_{i_k}}) I i k = ( k , 1 , … , k , e i k ) . We define the blockwise lift of i \mathbf{i} i by
i ^ : = I i 1 ⋆ ⋯ ⋆ I i l = ( ı 1 , 1 , … , ı 1 , e i 1 , … , ı l , 1 , … , ı l , e i l ) ∈ I g ^ N l , (12) \widehat{\mathbf{i}}:=I_{i_1}\star\cdots\star I_{i_l}
=(\imath_{1,1},\ldots,\imath_{1,e_{i_1}},\ldots,\imath_{l,1},\ldots,\imath_{l,e_{i_l}})\in I_{\widehat{\mathfrak g}}^{N_l}, \tag*{(12)} i := I i 1 ⋆ ⋯ ⋆ I i l = ( 1 , 1 , … , 1 , e i 1 , … , l , 1 , … , l , e i l ) ∈ I g N l , (12)
where N s : = ∑ k = 1 s e i k N_s:=\sum_{k=1}^{s}e_{i_k} N s := ∑ k = 1 s e i k for 1 ≤ s ≤ l 1\leq s\leq l 1 ≤ s ≤ l . Here ⋆ \star ⋆ denotes the concatenation operator.
Set N 0 : = 0 N_0:=0 N 0 := 0 and, 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 , define
κ ( k , r ) : = N k − 1 + r . (13) \kappa(k,r):=N_{k-1}+r. \tag*{(13)} κ ( k , r ) := N k − 1 + r . (13)
Thus κ ( k , r ) \kappa(k,r) κ ( k , r ) is the position of ı k , r \imath_{k,r} k , r in i ^ \widehat{\mathbf{i}} i .
We associate with i \mathbf{i} i the root sequence
β k i : = s i 1 ⋯ s i k − 1 ( α i k ) ∈ Φ r e , g ( 1 ≤ k ≤ l ) . \beta_k^{\mathbf{i}}:=s_{i_1}\cdots s_{i_{k-1}}(\alpha_{i_k})\in\Phi_{\mathrm{re},\mathfrak g}\qquad(1\leq k\leq l). β k i := s i 1 ⋯ s i k − 1 ( α i k ) ∈ Φ re , g ( 1 ≤ k ≤ l ) .
For its lift i ^ \widehat{\mathbf{i}} i , we set
β ^ k , r : = β ^ κ ( k , r ) i ^ ∈ Φ r e , g ^ ( 1 ≤ k ≤ l , 1 ≤ r ≤ e i k ) . \widehat{\beta}_{k,r}:=\widehat{\beta}_{\kappa(k,r)}^{\widehat{\mathbf{i}}}\in\Phi_{\mathrm{re},\widehat{\mathfrak g}}\qquad(1\leq k\leq l,\ 1\leq r\leq e_{i_k}). β k , r := β κ ( k , r ) i ∈ Φ re , g ( 1 ≤ k ≤ l , 1 ≤ r ≤ e i k ) .
By admissibility, we have
β ^ k , r = s ^ i 1 ⋯ s ^ i k − 1 s ^ i k , 1 ⋯ s ^ i k , r − 1 ( α ^ i k , r ) = s ^ i 1 ⋯ s ^ i k − 1 ( α ^ i k ) . (14) \widehat{\beta}_{k,r}=\widehat{s}_{i_1}\cdots\widehat{s}_{i_{k-1}}\widehat{s}_{i_k,1}\cdots\widehat{s}_{i_k,r-1}(\widehat{\alpha}_{i_k,r})=\widehat{s}_{i_1}\cdots\widehat{s}_{i_{k-1}}(\widehat{\alpha}_{i_k}). \tag*{(14)} β k , r = s i 1 ⋯ s i k − 1 s i k , 1 ⋯ s i k , r − 1 ( α i k , r ) = s i 1 ⋯ s i k − 1 ( α i k ) . (14)
Lemma 2.4. For any 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 , we have
π ( β ^ k , r ) = β k i and ȷ ( β k i ) = 1 e i k ∑ t = 1 e i k β ^ k , t . \pi(\widehat{\beta}_{k,r})=\beta_k^i\quad\text{and}\quad\jmath(\beta_k^i)=\frac{1}{e_{i_k}}\sum_{t=1}^{e_{i_k}}\widehat{\beta}_{k,t}. π ( β k , r ) = β k i and ( β k i ) = e i k 1 t = 1 ∑ e i k β k , t . Proof. By Lemma 2.1 and (14) , we have
π ( β ^ k , r ) = s i 1 ⋯ s i k − 1 ( π ( α ^ i k , r ) ) = s i 1 ⋯ s i k − 1 ( α i k ) = β k i . \pi(\widehat{\beta}_{k,r})=s_{i_1}\cdots s_{i_{k-1}}\left(\pi(\widehat{\alpha}_{i_k,r})\right)=s_{i_1}\cdots s_{i_{k-1}}(\alpha_{i_k})=\beta_k^i. π ( β k , r ) = s i 1 ⋯ s i k − 1 ( π ( α i k , r ) ) = s i 1 ⋯ s i k − 1 ( α i k ) = β k i . Similarly, we have
ȷ ( β k i ) = s ^ i 1 ⋯ s ^ i k − 1 ( ȷ ( α i k ) ) = 1 e i k ∑ t = 1 e i k s ^ i 1 ⋯ s ^ i k − 1 ( α ^ i k , t ) = 1 e i k ∑ t = 1 e i k β ^ k , t . \jmath(\beta_k^i)=\widehat{s}_{i_1}\cdots\widehat{s}_{i_{k-1}}\left(\jmath(\alpha_{i_k})\right)=\frac{1}{e_{i_k}}\sum_{t=1}^{e_{i_k}}\widehat{s}_{i_1}\cdots\widehat{s}_{i_{k-1}}(\widehat{\alpha}_{i_k,t})=\frac{1}{e_{i_k}}\sum_{t=1}^{e_{i_k}}\widehat{\beta}_{k,t}. ( β k i ) = s i 1 ⋯ s i k − 1 ( ( α i k ) ) = e i k 1 t = 1 ∑ e i k s i 1 ⋯ s i k − 1 ( α i k , t ) = e i k 1 t = 1 ∑ e i k β k , t .
Corollary 2.5. The sequence i i i is reduced if and only if i ^ \widehat{i} i is reduced.
Proof. Recall that i i i is reduced if and only if β k i ∈ Φ r e , g + \beta_k^i\in\Phi_{\mathrm{re},g}^{+} β k i ∈ Φ re , g + (1 ≤ k ≤ l 1\leq k\leq l 1 ≤ k ≤ l ). By Lemma 2.4, we have π ( β ^ k , r ) = β k i \pi(\widehat{\beta}_{k,r})=\beta_k^i π ( β k , r ) = β k i . Since π ( Φ r e , g ^ + ) ⊂ Q + \pi(\Phi_{\mathrm{re},\widehat{g}}^{+})\subset\mathsf{Q}^{+} π ( Φ re , g + ) ⊂ Q + and π ( Φ r e , g ^ − ) ⊂ Q − \pi(\Phi_{\mathrm{re},\widehat{g}}^{-})\subset\mathsf{Q}^{-} π ( Φ re , g − ) ⊂ Q − , we have
β k i ∈ Φ r e , g + if and only if β ^ k , r ∈ Φ r e , g ^ + ( 1 ≤ r ≤ e i k ) . \beta_k^i\in\Phi_{\mathrm{re},g}^{+}\quad\text{if and only if}\quad\widehat{\beta}_{k,r}\in\Phi_{\mathrm{re},\widehat{g}}^{+}\qquad(1\leq r\leq e_{i_k}). β k i ∈ Φ re , g + if and only if β k , r ∈ Φ re , g + ( 1 ≤ r ≤ e i k ) . Thus all roots in the root sequence of i i i are positive if and only if all roots in the root sequence of i ^ \widehat{i} i are positive.
By Corollary 2.5 and [4 Theorem 28] , the assignments s i ↦ s ^ i s_i\mapsto\widehat{s}_i s i ↦ s i and r i ↦ r ^ i r_i\mapsto\widehat{r}_i r i ↦ r i induce injective homomorphisms
ι : W ↪ W ^ and ι : B + ↪ B ^ + . \iota:W\hookrightarrow\widehat{W}\qquad\text{and}\qquad\iota:B^{+}\hookrightarrow\widehat{B}^{+}. ι : W ↪ W and ι : B + ↪ B + .
Moreover,
ι ( W ) = W ^ Γ and ι ( B + ) = ( B ^ + ) Γ . \iota(W)=\widehat{W}^{\Gamma}\qquad\text{and}\qquad\iota(B^{+})=(\widehat{B}^{+})^{\Gamma}. ι ( W ) = W Γ and ι ( B + ) = ( B + ) Γ .
In particular, if i = ( i 1 , … , i l ) i=(i_1,\ldots,i_l) i = ( i 1 , … , i l ) is reduced, then i ^ \widehat{i} i is reduced and
ℓ W ^ ( s ^ i 1 ⋯ s ^ i l ) = N l = ∑ k = 1 l e i k . \ell_{\widehat{W}}(\widehat{s}_{i_1}\cdots\widehat{s}_{i_l})=N_l=\sum_{k=1}^{l}e_{i_k}. ℓ W ( s i 1 ⋯ s i l ) = N l = k = 1 ∑ l e i k .
Folding of root pairings and λ \lambda λ -pairings
The Γ \Gamma Γ -action on I ^ \widehat{I} I preserves each fiber I i k I_{i_k} I i k and permutes its entries. It induces the same permutation on the roots β ^ k , r \widehat{\beta}_{k,r} β k , r ; namely,
g ( i k , r ) = i k , r ′ and g ( β ^ k , r ) = β ^ k , r ′ . g(i_{k,r})=i_{k,r'}\qquad\text{and}\qquad g(\widehat{\beta}_{k,r})=\widehat{\beta}_{k,r'}. g ( i k , r ) = i k , r ′ and g ( β k , r ) = β k , r ′ .
Lemma 2.6. For any 1 ≤ k , p ≤ l 1\leq k,p\leq l 1 ≤ k , p ≤ l and 1 ≤ t ≤ e i p 1\leq t\leq e_{i_p} 1 ≤ t ≤ e i p , we have
( β k i , β p i ) g = m e i k ∑ r = 1 e i k ( β ^ k , r , β ^ p , t ) g ^ . (\beta_k^i,\beta_p^i)_g=\frac{m}{e_{i_k}}\sum_{r=1}^{e_{i_k}}(\widehat{\beta}_{k,r},\widehat{\beta}_{p,t})_{\widehat{g}}. ( β k i , β p i ) g = e i k m r = 1 ∑ e i k ( β k , r , β p , t ) g . In particular, the right-hand side is independent of $t.
Proof. By Lemmas 2.2 and 2.4 ,
( β k i , β p i ) g = m ( J ( β k i ) , J ( β p i ) ) g ^ = m e i k e i p ∑ r = 1 e i k ∑ s = 1 e i p ( β ^ k , r , β ^ p , s ) g ^ . (\beta_k^i,\beta_p^i)_{\mathfrak{g}}=m\bigl(J(\beta_k^i),J(\beta_p^i)\bigr)_{\widehat{\mathfrak{g}}}=\frac{m}{e_{i_k}e_{i_p}}\sum_{r=1}^{e_{i_k}}\sum_{s=1}^{e_{i_p}}(\widehat{\beta}_{k,r},\widehat{\beta}_{p,s})_{\widehat{\mathfrak{g}}}. ( β k i , β p i ) g = m ( J ( β k i ) , J ( β p i ) ) g = e i k e i p m r = 1 ∑ e i k s = 1 ∑ e i p ( β k , r , β p , s ) g .
For 1 ≤ t , t ′ ≤ e i p 1\le t,t'\le e_{i_p} 1 ≤ t , t ′ ≤ e i p , choose g ∈ Γ g\in\Gamma g ∈ Γ such that g ( ν p , t ) = ν p , t ′ g(\nu_{p,t})=\nu_{p,t'} g ( ν p , t ) = ν p , t ′ . Since g g g permutes each fiber and preserves the bilinear form,
∑ r = 1 e i k ( β ^ k , r , β ^ p , t ) g ^ = ∑ r = 1 e i k ( g ( β ^ k , r ) , g ( β ^ p , t ) ) g ^ = ∑ r = 1 e i k ( β ^ k , r , β ^ p , t ′ ) g ^ . \sum_{r=1}^{e_{i_k}}(\widehat{\beta}_{k,r},\widehat{\beta}_{p,t})_{\widehat{\mathfrak{g}}}=\sum_{r=1}^{e_{i_k}}\bigl(g(\widehat{\beta}_{k,r}),g(\widehat{\beta}_{p,t})\bigr)_{\widehat{\mathfrak{g}}}=\sum_{r=1}^{e_{i_k}}(\widehat{\beta}_{k,r},\widehat{\beta}_{p,t'})_{\widehat{\mathfrak{g}}}. r = 1 ∑ e i k ( β k , r , β p , t ) g = r = 1 ∑ e i k ( g ( β k , r ) , g ( β p , t ) ) g = r = 1 ∑ e i k ( β k , r , β p , t ′ ) g .
Hence
( β k i , β p i ) g = m e i k e i p ∑ s = 1 e i p ∑ r = 1 e i k ( β ^ k , r , β ^ p , s ) g ^ = m e i k ∑ r = 1 e i k ( β ^ k , r , β ^ p , t ) g ^ . (\beta_k^i,\beta_p^i)_{\mathfrak{g}}=\frac{m}{e_{i_k}e_{i_p}}\sum_{s=1}^{e_{i_p}}\sum_{r=1}^{e_{i_k}}(\widehat{\beta}_{k,r},\widehat{\beta}_{p,s})_{\widehat{\mathfrak{g}}}=\frac{m}{e_{i_k}}\sum_{r=1}^{e_{i_k}}(\widehat{\beta}_{k,r},\widehat{\beta}_{p,t})_{\widehat{\mathfrak{g}}}. ( β k i , β p i ) g = e i k e i p m s = 1 ∑ e i p r = 1 ∑ e i k ( β k , r , β p , s ) g = e i k m r = 1 ∑ e i k ( β k , r , β p , t ) g .
Let
L i : = ⨁ k = 1 l Z ε k . L^i:=\bigoplus_{k=1}^{l}\mathbb{Z}\varepsilon_k. L i := k = 1 ⨁ l Z ε k .
where ε k \varepsilon_k ε k is labeled by the root β k i \beta_k^i β k i . With the sign opposite to that of [31 (5.4)] , we define the skew-symmetric bilinear form
λ i : L i × L i ⟶ Z \lambda^i:L^i\times L^i\longrightarrow\mathbb{Z} λ i : L i × L i ⟶ Z
by
λ i ( ε k , ε p ) : = ( − 1 ) δ ( k < p ) δ ( k ≠ p ) ( β k i , β p i ) g . (15) \lambda^i(\varepsilon_k,\varepsilon_p):=(-1)^{\delta(k<p)}\delta(k\ne p)(\beta_k^i,\beta_p^i)_{\mathfrak{g}}. \tag*{(15)} λ i ( ε k , ε p ) := ( − 1 ) δ ( k < p ) δ ( k = p ) ( β k i , β p i ) g . (15)
Likewise, let
L ^ i : = ⨁ k = 1 l ⨁ r = 1 e i k Z ε ^ k , r , \widehat{L}^i:=\bigoplus_{k=1}^{l}\bigoplus_{r=1}^{e_{i_k}}\mathbb{Z}\widehat{\varepsilon}_{k,r}, L i := k = 1 ⨁ l r = 1 ⨁ e i k Z ε k , r ,
where ε ^ k , r \widehat{\varepsilon}_{k,r} ε k , r is labeled by β ^ k , r \widehat{\beta}_{k,r} β k , r . For r ≠ s r\ne s r = s , we have
( β ^ k , r , β ^ k , s ) g ^ = ( α ^ ν k , r , α ^ ν k , s ) g ^ = 0 (\widehat{\beta}_{k,r},\widehat{\beta}_{k,s})_{\widehat{\mathfrak{g}}}=(\widehat{\alpha}_{\nu_{k,r}},\widehat{\alpha}_{\nu_{k,s}})_{\widehat{\mathfrak{g}}}=0 ( β k , r , β k , s ) g = ( α ν k , r , α ν k , s ) g = 0
by admissibility. Hence we define
λ ^ i : L ^ i × L ^ i ⟶ Z \widehat{\lambda}^i:\widehat{L}^i\times\widehat{L}^i\longrightarrow\mathbb{Z} λ i : L i × L i ⟶ Z
by
λ ^ i ( ε ^ k , r , ε ^ p , t ) : = ( − 1 ) δ ( k < p ) δ ( ( k , r ) ≠ ( p , t ) ) ( β ^ k , r , β ^ p , t ) g ^ . (16) \widehat{\lambda}^i(\widehat{\varepsilon}_{k,r},\widehat{\varepsilon}_{p,t}):=(-1)^{\delta(k<p)}\delta\bigl((k,r)\ne(p,t)\bigr)(\widehat{\beta}_{k,r},\widehat{\beta}_{p,t})_{\widehat{\mathfrak{g}}}. \tag*{(16)} λ i ( ε k , r , ε p , t ) := ( − 1 ) δ ( k < p ) δ ( ( k , r ) = ( p , t ) ) ( β k , r , β p , t ) g . (16)
Lemma 2.7. For any 1 ≤ k , p ≤ l 1\le k,p\le l 1 ≤ k , p ≤ l and 1 ≤ t ≤ e i p 1\le t\le e_{i_p} 1 ≤ t ≤ e i p , we have
λ i ( ε k , ε p ) = m e i k ∑ r = 1 e i k λ ^ i ( ε ^ k , r , ε ^ p , t ) = d i k ∑ r = 1 e i k λ ^ i ( ε ^ k , r , ε ^ p , t ) . \lambda^i(\varepsilon_k,\varepsilon_p)=\frac{m}{e_{i_k}}\sum_{r=1}^{e_{i_k}}\widehat{\lambda}^i(\widehat{\varepsilon}_{k,r},\widehat{\varepsilon}_{p,t})=d_{i_k}\sum_{r=1}^{e_{i_k}}\widehat{\lambda}^i(\widehat{\varepsilon}_{k,r},\widehat{\varepsilon}_{p,t}). λ i ( ε k , ε p ) = e i k m r = 1 ∑ e i k λ i ( ε k , r , ε p , t ) = d i k r = 1 ∑ e i k λ i ( ε k , r , ε p , t ) . In particular, the right-hand side is independent of t t t .
Proof. Since ( β ^ k , r , β ^ k , t ) g ^ = 0 (\widehat{\beta}_{k,r},\widehat{\beta}_{k,t})_{\widehat{g}}=0 ( β k , r , β k , t ) g = 0 for r ≠ t r\ne t r = t , we have
m e i k ∑ r = 1 e i k λ ^ i ( ε ^ k , r , ε ^ p , t ) = ( − 1 ) δ ( k < p ) δ ( k ≠ p ) m e i k ∑ r = 1 e i k ( β ^ k , r , β ^ p , t ) g ^ \frac{m}{e_{i_k}}\sum_{r=1}^{e_{i_k}}\widehat{\lambda}^{i}(\widehat{\varepsilon}_{k,r},\widehat{\varepsilon}_{p,t})=(-1)^{\delta(k<p)\delta(k\ne p)}\frac{m}{e_{i_k}}\sum_{r=1}^{e_{i_k}}(\widehat{\beta}_{k,r},\widehat{\beta}_{p,t})_{\widehat{g}} e i k m r = 1 ∑ e i k λ i ( ε k , r , ε p , t ) = ( − 1 ) δ ( k < p ) δ ( k = p ) e i k m r = 1 ∑ e i k ( β k , r , β p , t ) g
= ∗ ( − 1 ) δ ( k < p ) δ ( k ≠ p ) ( β k i , β p i ) g = λ ^ i ( ε k , ε p ) , \overset{*}{=}(-1)^{\delta(k<p)\delta(k\ne p)}(\beta^i_k,\beta^i_p)_{g}=\widehat{\lambda}^{i}(\varepsilon_k,\varepsilon_p), = ∗ ( − 1 ) δ ( k < p ) δ ( k = p ) ( β k i , β p i ) g = λ i ( ε k , ε p ) ,
where ∗ * ∗ follows from Lemma 2.6. The remaining equality follows from m / e i k = d i k m/e_{i_k}=d_{i_k} m / e i k = d i k .
Folding of compatible pairs
We study compatible pairs associated with arbitrary finite sequences and i \mathbf{i} i -box families, and establish their folding properties. We slightly generalize the relevant constructions from [13 , 14 , 15 , 12 , 32 , 27 ] to the present setting.
Quantum cluster algebras and pointed elements
We briefly recall the notation for quantum cluster algebras that will be used throughout this paper. We refer to [7 , 1 ] for the general theory and follow the conventions of [32 , 12 ] .
Let J J J be a finite index set and let J e x ⊂ J J_{\mathrm{ex}}\subset J J ex ⊂ J be the set of exchangeable indices. Let B = ( b i j ) i ∈ J , j ∈ J e x B=(b_{ij})_{i\in J,j\in J_{\mathrm{ex}}} B = ( b ij ) i ∈ J , j ∈ J ex be an integer matrix and let Λ = ( λ i j ) i , j ∈ J \Lambda=(\lambda_{ij})_{i,j\in J} Λ = ( λ ij ) i , j ∈ J be a skew-symmetric integer matrix. We say that ( B , Λ ) (B,\Lambda) ( B , Λ ) is a compatible pair if
∑ p ∈ J b p k λ p j = 2 d k δ k , j ( k ∈ J e x , j ∈ J ) (17) \sum_{p\in J}b_{pk}\lambda_{pj}=2d_k\delta_{k,j}\qquad(k\in J_{\mathrm{ex}},\ j\in J) \tag*{(17)} p ∈ J ∑ b p k λ p j = 2 d k δ k , j ( k ∈ J ex , j ∈ J ) (17)
for some positive integers d k d_k d k .
Associated with Λ \Lambda Λ is the quantum torus T ( Λ ) \mathcal{T}(\Lambda) T ( Λ ) over Z [ q ± 1 / 2 ] \mathbb{Z}[q^{\pm1/2}] Z [ q ± 1/2 ] , generated by X i ± 1 X_i^{\pm1} X i ± 1 (i ∈ J i\in J i ∈ J ) with
X i X j = q λ i j X j X i . X_iX_j=q^{\lambda_{ij}}X_jX_i. X i X j = q λ ij X j X i .
Fix a total order on J J J . For a = ( a i ) i ∈ J ∈ Z J \mathbf{a}=(a_i)_{i\in J}\in\mathbb{Z}^{J} a = ( a i ) i ∈ J ∈ Z J , the based quantum-torus monomial X a X^{\mathbf{a}} X a is defined by
X a : = q c ( a ) ∏ i ∈ J X i a i , c ( a ) : = − 1 2 ∑ i , j ∈ J i < j λ i j a i a j , X^{\mathbf{a}}:=q^{c(\mathbf{a})}\prod_{i\in J}X_i^{a_i},\qquad c(\mathbf{a}):=-\frac{1}{2}\sum_{\substack{i,j\in J\\i<j}}\lambda_{ij}a_i a_j, X a := q c ( a ) i ∈ J ∏ X i a i , c ( a ) := − 2 1 i , j ∈ J i < j ∑ λ ij a i a j ,
where the product is taken in increasing order. This definition is independent of the chosen total order. We write
Σ q = ( X , B , Λ ) , X = ( X i ∣ i ∈ J ) , \Sigma_q=(\mathbf{X},B,\Lambda),\qquad\mathbf{X}=(X_i\mid i\in J), Σ q = ( X , B , Λ ) , X = ( X i ∣ i ∈ J ) ,
for the corresponding quantum seed, whose based quantum torus is T ( Λ ) \mathcal{T}(\Lambda) T ( Λ ) , and denote its quantum cluster algebra by A q ( Σ q ) \mathcal{A}_q(\Sigma_q) A q ( Σ q ) . We refer to [31 Section 2.1] for the quantum cluster algebra conventions used here. By specializing q 1 / 2 q^{1/2} q 1/2 to 1 1 1 , a quantum cluster algebra specializes to the corresponding (commutative) cluster algebra (see for instance [11 §A.4] ). We refer to these two settings as the quantum setting and the commutative setting , respectively.
In the commutative setting, a seed is a pair Σ = ( x , B ) \Sigma=(\mathbf{x},B) Σ = ( x , B ) , where x = ( x i ∣ i ∈ J ) \mathbf{x}=(x_i\mid i\in J) x = ( x i ∣ i ∈ J ) is called the cluster of Σ \Sigma Σ . Throughout this paper, all commutative cluster algebras and classical Grothendieck rings are considered over C \mathbb{C} C .2 Whenever we move from the quantum setting to the commutative cluster-algebra setting by specializing at q 1 / 2 = 1 q^{1/2}=1 q 1/2 = 1 , we implicitly extend scalars from Z \mathbb{Z} Z to C \mathbb{C} C . We denote by
L ( Σ ) : = C [ x i ± 1 ∣ i ∈ J ] \mathcal{L}(\Sigma):=\mathbb{C}[x_i^{\pm1}\mid i\in J] L ( Σ ) := C [ x i ± 1 ∣ i ∈ J ]
the Laurent polynomial algebra associated with Σ \Sigma Σ .
For k ∈ J e x k\in J_{\mathrm{ex}} k ∈ J ex , the mutation
μ k ( B , Λ ) = ( B ′ , Λ ′ ) \mu_k(B,\Lambda)=(B',\Lambda') μ k ( B , Λ ) = ( B ′ , Λ ′ )
is defined by
b i j ′ = { − b i j , i = k or j = k , b i j + [ b i k ] + [ b k j ] + − [ − b i k ] + [ − b k j ] + , otherwise . b'_{ij}=
\begin{cases}
-b_{ij}, & i=k\text{ or }j=k,\\
b_{ij}+[b_{ik}]_+[b_{kj}]_+-[-b_{ik}]_+[-b_{kj}]_+, & \text{otherwise}.
\end{cases} b ij ′ = { − b ij , b ij + [ b ik ] + [ b k j ] + − [ − b ik ] + [ − b k j ] + , i = k or j = k , otherwise .
where [ a ] + : = max ( a , 0 ) [a]_+:=\max(a,0) [ a ] + := max ( a , 0 ) , and
λ i j ′ = λ i j ( i , j ≠ k ) , λ k k ′ = 0 , \lambda'_{ij}=\lambda_{ij}\quad(i,j\ne k),\qquad\lambda'_{kk}=0, λ ij ′ = λ ij ( i , j = k ) , λ k k ′ = 0 ,
λ i k ′ = − λ i k + ∑ p ∈ J [ − b p k ] + λ i p , λ k i ′ = − λ i k ′ ( i ≠ k ) . \lambda'_{ik}=-\lambda_{ik}+\sum_{p\in J}[-b_{pk}]_+\lambda_{ip},\qquad\lambda'_{ki}=-\lambda'_{ik}\quad(i\ne k). λ ik ′ = − λ ik + p ∈ J ∑ [ − b p k ] + λ i p , λ k i ′ = − λ ik ′ ( i = k ) .
Then ( B ′ , Λ ′ ) (B',\Lambda') ( B ′ , Λ ′ ) is again a compatible pair with the same positive integers d k d_k d k .
At the commutative level, for a cluster x = ( x i ∣ i ∈ J ) \mathbf{x}=(x_i\mid i\in J) x = ( x i ∣ i ∈ J ) , set
M k ± ( x ) : = ∏ i ∈ J x i [ ± b i k ] + . M_k^\pm(\mathbf{x}):=\prod_{i\in J}x_i^{[\pm b_{ik}]_+}. M k ± ( x ) := i ∈ J ∏ x i [ ± b ik ] + .
The mutation at k ∈ J e x k\in J_{\mathrm{ex}} k ∈ J ex is given by
x i ′ = x i ( i ≠ k ) and x k x k ′ = M k + ( x ) + M k − ( x ) . x'_i=x_i\quad(i\ne k)\qquad\text{and}\qquad x_kx'_k=M_k^+(\mathbf{x})+M_k^-(\mathbf{x}). x i ′ = x i ( i = k ) and x k x k ′ = M k + ( x ) + M k − ( x ) .
The cluster algebra A ( Σ ) \mathcal{A}(\Sigma) A ( Σ ) is the C \mathbb{C} C -subalgebra of L ( Σ ) \mathcal{L}(\Sigma) L ( Σ ) generated by all cluster variables obtained from Σ \Sigma Σ by iterated mutations. By the Laurent phenomenon [7 ] and its quantum analogue [1 ] , for every seed Σ ′ \Sigma' Σ ′ mutation-equivalent to Σ \Sigma Σ and every quantum seed Σ q ′ = ( X ′ , B ′ , Λ ′ ) \Sigma'_q=(\mathbf{X}',B',\Lambda') Σ q ′ = ( X ′ , B ′ , Λ ′ ) mutation-equivalent to Σ q \Sigma_q Σ q , we have
A ( Σ ) ⊆ L ( Σ ′ ) , A q ( Σ q ) ⊆ T ( Λ ′ ) . \mathcal{A}(\Sigma)\subseteq\mathcal{L}(\Sigma'),\qquad\mathcal{A}_q(\Sigma_q)\subseteq\mathcal{T}(\Lambda'). A ( Σ ) ⊆ L ( Σ ′ ) , A q ( Σ q ) ⊆ T ( Λ ′ ) .
We recall pointed elements and g g g -vectors ([8, 47]). An element z ∈ T ( Λ ) z\in\mathcal{T}(\Lambda) z ∈ T ( Λ ) is said to be pointed at g ∈ Z J \mathbf{g}\in\mathbb{Z}^J g ∈ Z J if it has an expansion of the form
z = q a X g + ∑ n ∈ N J e x ∖ { 0 } p n X g + B n . (18) z=q^aX^{\mathbf{g}}+\sum_{\mathbf{n}\in\mathbb{N}^{J_{\mathrm{ex}}}\setminus\{\mathbf{0}\}}p_{\mathbf{n}}X^{\mathbf{g}+B\mathbf{n}}. \tag*{(18)} z = q a X g + n ∈ N J ex ∖ { 0 } ∑ p n X g + B n . (18)
The vector g \mathbf{g} g is called the degree , or the g g g -vector, of z z z with respect to Σ q \Sigma_q Σ q , and we write
deg Σ q ( z ) = g . \deg_{\Sigma_q}(z)=\mathbf{g}. deg Σ q ( z ) = g .
The same terminology will be used in the commutative setting by specializing q 1 / 2 q^{1/2} q 1/2 to 1 1 1 . Cluster monomials are pointed and hence have well-defined g g g -vectors; see [19 §5] .
For later use, we recall the mutation rule for g g g -vectors. Let Σ q ′ = μ k ( Σ q ) \Sigma'_q=\mu_k(\Sigma_q) Σ q ′ = μ k ( Σ q ) , and let z z z be an element compatibly pointed at Σ q \Sigma_q Σ q and Σ q ′ \Sigma'_q Σ q ′ . Write
deg Σ q ( z ) = g = ( g j ) j ∈ J , deg Σ q ′ ( z ) = g ′ = ( g j ′ ) j ∈ J . \deg_{\Sigma_q}(z)=\mathbf{g}=(g_j)_{j\in J},\qquad\deg_{\Sigma'_q}(z)=\mathbf{g}'=(g'_j)_{j\in J}. deg Σ q ( z ) = g = ( g j ) j ∈ J , deg Σ q ′ ( z ) = g ′ = ( g j ′ ) j ∈ J .
Then
g j = { − g k ′ , j = k , g j ′ + [ b j k ′ ] + g k ′ , j ≠ k , g k ′ ≥ 0 , g j ′ + [ − b j k ′ ] + g k ′ , j ≠ k , g k ′ ≤ 0. (19) g_j =
\begin{cases}
-g'_k, & j=k,\\
g'_j+[b'_{jk}]_+g'_k, & j\ne k,\ g'_k\ge0,\\
g'_j+[-b'_{jk}]_+g'_k, & j\ne k,\ g'_k\le0.
\end{cases} \tag*{(19)} g j = ⎩ ⎨ ⎧ − g k ′ , g j ′ + [ b j k ′ ] + g k ′ , g j ′ + [ − b j k ′ ] + g k ′ , j = k , j = k , g k ′ ≥ 0 , j = k , g k ′ ≤ 0. (19)
where B ′ = ( b i j ′ ) B'=(b'_{ij}) B ′ = ( b ij ′ ) is the exchange matrix of the mutated seed (see [8 , 19 , 12 ] for more details).
Compatible pairs for arbitrary sequences
Let i = ( i 1 , … , i l ) ∈ I l \mathbf{i}=(i_1,\ldots,i_l)\in I^l i = ( i 1 , … , i l ) ∈ I l be an arbitrary finite sequence. For 1 ≤ k ≤ l 1\le k\le l 1 ≤ k ≤ l and j ∈ I j\in I j ∈ I , set
k + : = min ( { p ∈ [ 1 , l ] ∣ k < p , i p = i k } ∪ { ∞ } ) , k − : = max ( { p ∈ [ 1 , l ] ∣ p < k , i p = i k } ∪ { − ∞ } ) , k ( j ) + : = min ( { p ∈ [ 1 , l ] ∣ k ≤ p , i p = j } ∪ { ∞ } ) , k ( j ) − : = max ( { p ∈ [ 1 , l ] ∣ p ≤ k , i p = j } ∪ { − ∞ } ) . (20) \begin{aligned}
k_+&:=\min\bigl(\{p\in[1,l]\mid k<p,\ i_p=i_k\}\cup\{\infty\}\bigr),\\
k_-&:=\max\bigl(\{p\in[1,l]\mid p<k,\ i_p=i_k\}\cup\{-\infty\}\bigr),\\
k(j)^+&:=\min\bigl(\{p\in[1,l]\mid k\le p,\ i_p=j\}\cup\{\infty\}\bigr),\\
k(j)^-&:=\max\bigl(\{p\in[1,l]\mid p\le k,\ i_p=j\}\cup\{-\infty\}\bigr). \tag*{(20)}
\end{aligned} k + k − k ( j ) + k ( j ) − := min ( { p ∈ [ 1 , l ] ∣ k < p , i p = i k } ∪ { ∞ } ) , := max ( { p ∈ [ 1 , l ] ∣ p < k , i p = i k } ∪ { − ∞ } ) , := min ( { p ∈ [ 1 , l ] ∣ k ≤ p , i p = j } ∪ { ∞ } ) , := max ( { p ∈ [ 1 , l ] ∣ p ≤ k , i p = j } ∪ { − ∞ } ) . (20)
Set
J i : = [ 1 , l ] and J e x i : = { k ∈ [ 1 , l ] ∣ k + < ∞ } . J^{\mathbf{i}}:=[1,l]\qquad\text{and}\qquad J^{\mathbf{i}}_{\mathrm{ex}}:=\{k\in[1,l]\mid k_+<\infty\}. J i := [ 1 , l ] and J ex i := { k ∈ [ 1 , l ] ∣ k + < ∞ } .
Following the Geiss–Leclerc–Schröer construction [14 ] , we define the exchange matrix
B i = ( b p , k i ) p ∈ J i , k ∈ J e x i by b p , k i : = { 1 , k = p + , − 1 , p = k + , c i p , i k , p < k < p + < k + , − c i p , i k , k < p < k + < p + , 0 , otherwise . (21) B^{\mathbf{i}}=(b^{\mathbf{i}}_{p,k})_{p\in J^{\mathbf{i}},\,k\in J^{\mathbf{i}}_{\mathrm{ex}}}
\qquad\text{by}\qquad
b^{\mathbf{i}}_{p,k}:=
\begin{cases}
1, & k=p_+,\\
-1, & p=k_+,\\
c_{i_p,i_k}, & p<k<p_+<k_+,\\
-c_{i_p,i_k}, & k<p<k_+<p_+,\\
0, & \text{otherwise}.
\end{cases} \tag*{(21)} B i = ( b p , k i ) p ∈ J i , k ∈ J ex i by b p , k i := ⎩ ⎨ ⎧ 1 , − 1 , c i p , i k , − c i p , i k , 0 , k = p + , p = k + , p < k < p + < k + , k < p < k + < p + , otherwise . (21)
For 1 ≤ k ≤ l 1\le k\le l 1 ≤ k ≤ l , set
k min : = min { p ∈ [ 1 , k ] ∣ i p = i k } . k_{\min}:=\min\{p\in[1,k]\mid i_p=i_k\}. k m i n := min { p ∈ [ 1 , k ] ∣ i p = i k } .
Thus [ k min , k ] [k_{\min},k] [ k m i n , k ] is the i \mathbf{i} i -box determined by the position k k k ; see Subsection 3.3 below.
Extending the construction for reduced sequences in [32 , 31 ] , we define the skew-symmetric matrix
Λ i = ( Λ p , k i ) p , k ∈ J i by Λ p , k i : = ∑ p min ≤ a ≤ p i a = i p ∑ k min ≤ b ≤ k i b = i k λ i ( ε a , ε b ) . (22) \Lambda^{\mathbf{i}}=(\Lambda^{\mathbf{i}}_{p,k})_{p,k\in J^{\mathbf{i}}}
\qquad\text{by}\qquad
\Lambda^{\mathbf{i}}_{p,k}:=
\sum_{\substack{p_{\min}\le a\le p\\ i_a=i_p}}
\sum_{\substack{k_{\min}\le b\le k\\ i_b=i_k}}
\lambda^{\mathbf{i}}(\varepsilon_a,\varepsilon_b). \tag*{(22)} Λ i = ( Λ p , k i ) p , k ∈ J i by Λ p , k i := p m i n ≤ a ≤ p i a = i p ∑ k m i n ≤ b ≤ k i b = i k ∑ λ i ( ε a , ε b ) . (22)
We call ( B i , Λ i ) (B^{\mathbf{i}},\Lambda^{\mathbf{i}}) ( B i , Λ i ) the GLS pair associated with i \mathbf{i} i . We denote by Σ i \Sigma^{\mathbf{i}} Σ i the commutative seed associated with the GLS pair ( B i , Λ i ) (B^{\mathbf{i}},\Lambda^{\mathbf{i}}) ( B i , Λ i ) , retaining Λ i \Lambda^{\mathbf{i}} Λ i as part of the seed data, and write
x i = ( x 1 , … , x l ) \mathbf{x}^{\mathbf{i}}=(x_1,\ldots,x_l) x i = ( x 1 , … , x l )
for its cluster. We call Σ i \Sigma^{\mathbf{i}} Σ i the GLS seed associated with i \mathbf{i} i .
Proposition 3.1. For any finite sequence i = ( i 1 , … , i l ) \mathbf{i}=(i_1,\ldots,i_l) i = ( i 1 , … , i l ) , the GLS pair ( B i , Λ i ) (B^{\mathbf{i}},\Lambda^{\mathbf{i}}) ( B i , Λ i ) is compatible. More precisely,
∑ a ∈ J i b a , k i Λ a , p i = 2 d i k δ k , p ( k ∈ J e x i , p ∈ J i ) . \sum_{a\in J^{\mathbf{i}}} b^{\mathbf{i}}_{a,k}\Lambda^{\mathbf{i}}_{a,p}
=2d_{i_k}\delta_{k,p}
\qquad(k\in J^{\mathbf{i}}_{\mathrm{ex}},\ p\in J^{\mathbf{i}}). a ∈ J i ∑ b a , k i Λ a , p i = 2 d i k δ k , p ( k ∈ J ex i , p ∈ J i ) .
Proof. For a ∈ J i a \in J^{i} a ∈ J i , set
E a : = ∑ a min ≤ r ≤ a i r = i a ε r \mathcal{E}_{a}:=\sum_{\substack{a_{\min}\leq r\leq a\\ i_{r}=i_{a}}}\varepsilon_{r} E a := a m i n ≤ r ≤ a i r = i a ∑ ε r
so that Λ a , p i = λ i ( E a , E p ) \Lambda^{i}_{a,p}=\lambda^{i}(\mathcal{E}_{a},\mathcal{E}_{p}) Λ a , p i = λ i ( E a , E p ) .
We use the convention E − ∞ = 0 \mathcal{E}_{-\infty}=0 E − ∞ = 0 .
Fix k ∈ J e x i k\in J^{i}_{\mathrm{ex}} k ∈ J ex i and put i : = i k = i k + i:=i_{k}=i_{k_{+}} i := i k = i k + . In the k k k -th column of B i B^{i} B i , the positions of color i i i contribute
E k − − E k + = − ε k − ε k + . \mathcal{E}_{k_{-}}-\mathcal{E}_{k_{+}}=-\varepsilon_{k}-\varepsilon_{k_{+}}. E k − − E k + = − ε k − ε k + .
For j ≠ i j\ne i j = i , let a 1 < ⋯ < a s a_{1}<\cdots<a_{s} a 1 < ⋯ < a s be the positions of color j j j lying strictly between k k k and k + k_{+} k + . By the conditions (ii) in (21) , the contribution of the j j j -colored positions to ∑ a ∈ J i b a , k i E a \sum_{a\in J^{i}}b^{i}_{a,k}\mathcal{E}_{a} ∑ a ∈ J i b a , k i E a is
− c j , i ( ε a 1 + ⋯ + ε a s ) . -c_{j,i}(\varepsilon_{a_{1}}+\cdots+\varepsilon_{a_{s}}). − c j , i ( ε a 1 + ⋯ + ε a s ) .
Summing over all colors j ≠ i j\ne i j = i , we obtain
− ξ k : = ∑ a ∈ J i b a , k i E a = − ( ε k + ε k + + ∑ k < a < k + c i a , i ε a ) = − ∑ k ≤ a ≤ k + u a ε a , (23) -\xi_{k}:=\sum_{a\in J^{i}}b^{i}_{a,k}\mathcal{E}_{a}
=-\left(\varepsilon_{k}+\varepsilon_{k_{+}}+\sum_{k<a<k_{+}}c_{i_{a},i}\varepsilon_{a}\right)
=-\sum_{k\leq a\leq k_{+}}u_{a}\varepsilon_{a}, \tag*{(23)} − ξ k := a ∈ J i ∑ b a , k i E a = − ε k + ε k + + k < a < k + ∑ c i a , i ε a = − k ≤ a ≤ k + ∑ u a ε a , (23)
where u k = u k + = 1 u_{k}=u_{k_{+}}=1 u k = u k + = 1 and u a = c i a , i u_{a}=c_{i_{a},i} u a = c i a , i (k < a < k + k<a<k_{+} k < a < k + ).
Let w a = s i 1 ⋯ s i a w_{a}=s_{i_{1}}\cdots s_{i_{a}} w a = s i 1 ⋯ s i a , with w 0 = 1 w_{0}=1 w 0 = 1 . Since w a ( α i ) − w a − 1 ( α i ) = − c i a , i β a i w_{a}(\alpha_{i})-w_{a-1}(\alpha_{i})=-c_{i_{a},i}\beta^{i}_{a} w a ( α i ) − w a − 1 ( α i ) = − c i a , i β a i , summing over k ≤ a < k + k\leq a<k_{+} k ≤ a < k + gives
β k + i − β k i = − ∑ a = k k + − 1 c i a , i β a i = − 2 β k i − ∑ k < a < k + c i a , i β a i . \beta^{i}_{k_{+}}-\beta^{i}_{k}
=-\sum_{a=k}^{k_{+}-1}c_{i_{a},i}\beta^{i}_{a}
=-2\beta^{i}_{k}-\sum_{k<a<k_{+}}c_{i_{a},i}\beta^{i}_{a}. β k + i − β k i = − a = k ∑ k + − 1 c i a , i β a i = − 2 β k i − k < a < k + ∑ c i a , i β a i .
which is equivalent to
β k i + β k + i + ∑ k < a < k + c i a , i β a i = 0. (24) \beta^{i}_{k}+\beta^{i}_{k_{+}}+\sum_{k<a<k_{+}}c_{i_{a},i}\beta^{i}_{a}=0. \tag*{(24)} β k i + β k + i + k < a < k + ∑ c i a , i β a i = 0. (24)
In terms of the coefficients u a u_{a} u a , this is
∑ k ≤ a ≤ k + u a β a i = 0. (25) \sum_{k\leq a\leq k_{+}}u_{a}\beta^{i}_{a}=0. \tag*{(25)} k ≤ a ≤ k + ∑ u a β a i = 0. (25)
We now compute λ i ( ξ k , ε q ) \lambda^{i}(\xi_{k},\varepsilon_{q}) λ i ( ξ k , ε q ) . Recall that
λ i ( ε a , ε q ) = δ ( a ≠ q ) ( − 1 ) δ ( a < q ) ( β a i , β q i ) . \lambda^{i}(\varepsilon_{a},\varepsilon_{q})
=\delta(a\ne q)(-1)^{\delta(a<q)}(\beta^{i}_{a},\beta^{i}_{q}). λ i ( ε a , ε q ) = δ ( a = q ) ( − 1 ) δ ( a < q ) ( β a i , β q i ) .
Thus (25) implies that
λ i ( ξ k , ε q ) = 0 for q ∉ [ k , k + ] . \lambda^{i}(\xi_{k},\varepsilon_{q})=0
\qquad\text{for }q\notin[k,k_{+}]. λ i ( ξ k , ε q ) = 0 for q ∈ / [ k , k + ] .
Next, suppose that k < q < k + k<q<k_{+} k < q < k + and j : = i q j:=i_{q} j := i q . Summing w a ( α i ) − w a − 1 ( α i ) = − c i a , i β a i w_{a}(\alpha_{i})-w_{a-1}(\alpha_{i})=-c_{i_{a},i}\beta^{i}_{a} w a ( α i ) − w a − 1 ( α i ) = − c i a , i β a i over k ≤ a < q k\leq a<q k ≤ a < q , we obtain
w q − 1 ( α i ) − w k − 1 ( α i ) = − 2 β k i − ∑ k < a < q c i a , i β a i . w_{q-1}(\alpha_{i})-w_{k-1}(\alpha_{i})
=-2\beta^{i}_{k}-\sum_{k<a<q}c_{i_{a},i}\beta^{i}_{a}. w q − 1 ( α i ) − w k − 1 ( α i ) = − 2 β k i − k < a < q ∑ c i a , i β a i .
Since w k − 1 ( α i ) = β k i w_{k-1}(\alpha_{i})=\beta^{i}_{k} w k − 1 ( α i ) = β k i , it follows that
∑ k ≤ a < q u a β a i = β k i + ∑ k < a < q c i a , i β a i = − w q − 1 ( α i ) . (26) \sum_{k\leq a<q}u_{a}\beta^{i}_{a}
=\beta^{i}_{k}+\sum_{k<a<q}c_{i_{a},i}\beta^{i}_{a}
=-w_{q-1}(\alpha_{i}). \tag*{(26)} k ≤ a < q ∑ u a β a i = β k i + k < a < q ∑ c i a , i β a i = − w q − 1 ( α i ) . (26)
On the other hand, since u q = c j , i u_q=c_{j,i} u q = c j , i , (25) and (26) give
∑ q ≤ a ≤ k + u a β a i = − ∑ k ≤ a < q u a β a i − u q β q i = w q − 1 ( α i ) − c j , i β q i . \sum_{q\leq a\leq k_+}u_a\beta_a^i=-\sum_{k\leq a<q}u_a\beta_a^i-u_q\beta_q^i=w_{q-1}(\alpha_i)-c_{j,i}\beta_q^i. q ≤ a ≤ k + ∑ u a β a i = − k ≤ a < q ∑ u a β a i − u q β q i = w q − 1 ( α i ) − c j , i β q i .
Therefore, by the definition of λ i \lambda^i λ i ,
λ i ( ξ k , ϵ q ) = − ( ∑ k ≤ a < q u a β a i , β q i ) + ( ∑ q < a ≤ k + u a β a i , β q i ) = ( 2 w q − 1 ( α i ) − c j , i β q i , β q i ) . \lambda^i(\xi_k,\epsilon_q)=-\left(\sum_{k\leq a<q}u_a\beta_a^i,\beta_q^i\right)+\left(\sum_{q<a\leq k_+}u_a\beta_a^i,\beta_q^i\right)=\left(2w_{q-1}(\alpha_i)-c_{j,i}\beta_q^i,\beta_q^i\right). λ i ( ξ k , ϵ q ) = − k ≤ a < q ∑ u a β a i , β q i + q < a ≤ k + ∑ u a β a i , β q i = ( 2 w q − 1 ( α i ) − c j , i β q i , β q i ) .
Since β q i = w q − 1 ( α j ) \beta_q^i=w_{q-1}(\alpha_j) β q i = w q − 1 ( α j ) and the bilinear form is W-invariant, we have
( w q − 1 ( α i ) , β q i ) = ( w q − 1 ( α i ) , w q − 1 ( α j ) ) = ( α i , α j ) and ( β q i , β q i ) = ( α j , α j ) . \left(w_{q-1}(\alpha_i),\beta_q^i\right)=\left(w_{q-1}(\alpha_i),w_{q-1}(\alpha_j)\right)=(\alpha_i,\alpha_j)\quad\text{and}\quad\left(\beta_q^i,\beta_q^i\right)=(\alpha_j,\alpha_j). ( w q − 1 ( α i ) , β q i ) = ( w q − 1 ( α i ) , w q − 1 ( α j ) ) = ( α i , α j ) and ( β q i , β q i ) = ( α j , α j ) .
Consequently,
λ i ( ξ k , ϵ q ) = 2 ( α i , α j ) − c j , i ( α j , α j ) = 2 d i c i j − 2 d j c j i = 0 , \lambda^i(\xi_k,\epsilon_q)=2(\alpha_i,\alpha_j)-c_{j,i}(\alpha_j,\alpha_j)=2d_ic_{ij}-2d_jc_{ji}=0, λ i ( ξ k , ϵ q ) = 2 ( α i , α j ) − c j , i ( α j , α j ) = 2 d i c ij − 2 d j c j i = 0 ,
where the last equality follows from d i c i j = d j c j i d_ic_{ij}=d_jc_{ji} d i c ij = d j c j i .
It remains to consider q = k q=k q = k and q = k + q=k_+ q = k + . For q = k q=k q = k , all terms of ξ k \xi_k ξ k other than ϵ k \epsilon_k ϵ k lie to the right of k k k , and hence
λ i ( ξ k , ϵ k ) = ( ∑ k < a ≤ k + u a β a i , β k i ) = ( − β k i , β k i ) = − ( α i , α i ) = − 2 d i . \lambda^i(\xi_k,\epsilon_k)=\left(\sum_{k<a\leq k_+}u_a\beta_a^i,\beta_k^i\right)=(-\beta_k^i,\beta_k^i)=-(\alpha_i,\alpha_i)=-2d_i. λ i ( ξ k , ϵ k ) = k < a ≤ k + ∑ u a β a i , β k i = ( − β k i , β k i ) = − ( α i , α i ) = − 2 d i .
Here we used (25) in the second equality.
For q = k + q=k_+ q = k + , all terms other than ϵ k + \epsilon_{k_+} ϵ k + lie to the left of k + k_+ k + , so
λ i ( ξ k , ϵ k + ) = − ( ∑ k ≤ a < k + u a β a i , β k + i ) = ( β k + i , β k + i ) = ( α i , α i ) = 2 d i . \lambda^i(\xi_k,\epsilon_{k_+})=-\left(\sum_{k\leq a<k_+}u_a\beta_a^i,\beta_{k_+}^i\right)=(\beta_{k_+}^i,\beta_{k_+}^i)=(\alpha_i,\alpha_i)=2d_i. λ i ( ξ k , ϵ k + ) = − k ≤ a < k + ∑ u a β a i , β k + i = ( β k + i , β k + i ) = ( α i , α i ) = 2 d i .
Thus, for every q ∈ J i q\in J^i q ∈ J i ,
λ i ( ξ k , ϵ q ) = { − 2 d i , q = k , 2 d i , q = k + , 0 , otherwise . (27) \lambda^i(\xi_k,\epsilon_q)=
\begin{cases}
-2d_i, & q=k,\\
2d_i, & q=k_+,\\
0, & \text{otherwise}.
\end{cases} \tag*{(27)} λ i ( ξ k , ϵ q ) = ⎩ ⎨ ⎧ − 2 d i , 2 d i , 0 , q = k , q = k + , otherwise . (27)
We now return to the compatibility relation. By (23) and (3.7), we have
∑ a ∈ J i b a , k i Λ a , p i = ∑ a ∈ J i b a , k i λ i ( ϵ a , ϵ p ) = λ i ( ∑ a ∈ J i b a , k i ϵ a , ϵ p ) = − λ i ( ξ k , ϵ p ) . \sum_{a\in J^i}b_{a,k}^i\Lambda_{a,p}^i=\sum_{a\in J^i}b_{a,k}^i\lambda^i(\epsilon_a,\epsilon_p)=\lambda^i\left(\sum_{a\in J^i}b_{a,k}^i\epsilon_a,\epsilon_p\right)=-\lambda^i(\xi_k,\epsilon_p). a ∈ J i ∑ b a , k i Λ a , p i = a ∈ J i ∑ b a , k i λ i ( ϵ a , ϵ p ) = λ i ( a ∈ J i ∑ b a , k i ϵ a , ϵ p ) = − λ i ( ξ k , ϵ p ) .
Since ϵ p \epsilon_p ϵ p is the sum of the basis vectors corresponding to the occurrences of the color i p i_p i p up to p p p , and k + k_+ k + is the next occurrence of the color i k i_k i k after k k k , (27) gives
λ i ( ξ k , ϵ p ) = − 2 d i δ k , p . \lambda^i(\xi_k,\epsilon_p)=-2d_i\delta_{k,p}. λ i ( ξ k , ϵ p ) = − 2 d i δ k , p .
Therefore,
∑ a ∈ J i b a , k i Λ a , p i = − λ i ( ξ k , ϵ p ) = 2 d i δ k , p = 2 d i k δ k , p . \sum_{a\in J^i}b_{a,k}^i\Lambda_{a,p}^i=-\lambda^i(\xi_k,\epsilon_p)=2d_i\delta_{k,p}=2d_{i_k}\delta_{k,p}. a ∈ J i ∑ b a , k i Λ a , p i = − λ i ( ξ k , ϵ p ) = 2 d i δ k , p = 2 d i k δ k , p .
Remark 3.2. For reduced sequences, compatible pairs of this type originate from the Geiss–Leclerc–Schröer construction [13 , 14 , 15 ] for the symmetric setting. Their counterparts in the symmetrizable setting appear in the quantum cluster structures of Goodearl–Yakimov [17 , 18 ] . The formulation used here for arbitrary sequences in finite type is essentially that of [12 Proposition 1.2, Corollary 1.3] . The proof above uses neither reducedness nor finiteness of the Weyl group. Thus, the same argument applies to the present symmetrizable Kac–Moody setting.
i-box families and compatible pairs associated with them
We recall the i-box combinatorics of [27 ] ; see also [32 ] . Let i = ( i 1 , … , i l ) ∈ I l \mathbf{i}=(i_1,\ldots,i_l)\in I^l i = ( i 1 , … , i l ) ∈ I l be an arbitrary finite sequence. An i \mathbf{i} i -box is an interval
c = [ a , b ] ⊂ [ 1 , l ] such that i a = i b . \mathbf{c}=[a,b]\subset[1,l]\text{ such that }i_a=i_b. c = [ a , b ] ⊂ [ 1 , l ] such that i a = i b .
We write i c : = i a = i b i_{\mathbf{c}}:=i_a=i_b i c := i a = i b for its color. When a = b = k a=b=k a = b = k , we simply write [ k ] [k] [ k ] for the singleton i \mathbf{i} i -box. Associated with an i \mathbf{i} i -box c = [ a , b ] \mathbf{c}=[a,b] c = [ a , b ] , we define
E c : = ∑ a ≤ r ≤ b i r = i c ε r (28) E_{\mathbf{c}}:=\sum_{\substack{a\leq r\leq b\\ i_r=i_{\mathbf{c}}}} \varepsilon_r \tag*{(28)} E c := a ≤ r ≤ b i r = i c ∑ ε r (28)
We call E c E_{\mathbf{c}} E c the box vector associated with c \mathbf{c} c . We set E ∅ : = 0 E_{\varnothing}:=0 E ∅ := 0 . In particular, for the i \mathbf{i} i -box c k = [ k min , k ] \mathbf{c}_k=[k_{\min},k] c k = [ k m i n , k ] , we have E c k = E k E_{\mathbf{c}_k}=E_k E c k = E k by (3.7).
Two i \mathbf{i} i -boxes c = [ a , b ] \mathbf{c}=[a,b] c = [ a , b ] and d = [ c , d ] \mathbf{d}=[c,d] d = [ c , d ] are said to commute if
a − < c ≤ d < b + or c − < a ≤ b < d + . a_-<c\leq d<b_+\quad\text{or}\quad c_-<a\leq b<d_+. a − < c ≤ d < b + or c − < a ≤ b < d + .
A collection of pairwise commuting i \mathbf{i} i -boxes is called a commuting family . A commuting family of i \mathbf{i} i -boxes in an interval [ a , b ] [a,b] [ a , b ] is called maximal if it is maximal among the commuting families of i \mathbf{i} i -boxes in [ a , b ] [a,b] [ a , b ] .
For an interval [ a , b ] ⊂ [ 1 , l ] [a,b]\subset[1,l] [ a , b ] ⊂ [ 1 , l ] , set
[ a , b } : = [ a , b ( i a ) − ] and { a , b ] : = [ a ( i b ) + , b ] . [a,b\}:=[a,b(i_a)^-]\quad\text{and}\quad\{a,b]:=[a(i_b)^+,b]. [ a , b } := [ a , b ( i a ) − ] and { a , b ] := [ a ( i b ) + , b ] .
A sequence
C = ( c 1 , … , c r ) , c k = [ x k , y k ] , \mathcal{C}=(\mathbf{c}_1,\ldots,\mathbf{c}_r),\qquad\mathbf{c}_k=[x_k,y_k], C = ( c 1 , … , c r ) , c k = [ x k , y k ] ,
of i \mathbf{i} i -boxes is called an admissible chain if, for each 1 ≤ k ≤ r 1\leq k\leq r 1 ≤ k ≤ r , the union
c ~ k : = ⋃ 1 ≤ s ≤ k [ x s , y s ] \widetilde{\mathbf{c}}_k:=\bigcup_{1\leq s\leq k}[x_s,y_s] c k := 1 ≤ s ≤ k ⋃ [ x s , y s ]
is an interval of cardinality k k k , and
c k = [ x ~ k , y ~ k } or c k = { x ~ k , y ~ k ] , \mathbf{c}_k=[\widetilde{x}_k,\widetilde{y}_k\}\quad\text{or}\quad\mathbf{c}_k=\{\widetilde{x}_k,\widetilde{y}_k], c k = [ x k , y k } or c k = { x k , y k ] ,
where c ~ k = [ x ~ k , y ~ k ] \widetilde{\mathbf{c}}_k=[\widetilde{x}_k,\widetilde{y}_k] c k = [ x k , y k ] . We call c ~ k \widetilde{\mathbf{c}}_k c k the envelope of c k \mathbf{c}_k c k , and c ~ r \widetilde{\mathbf{c}}_r c r the extent of C \mathcal{C} C .
The underlying family
F ( C ) : = { c 1 , … , c r } of C \mathcal{F}(\mathcal{C}):=\{\mathbf{c}_1,\ldots,\mathbf{c}_r\}\text{ of }\mathcal{C} F ( C ) := { c 1 , … , c r } of C
is a maximal commuting family of i \mathbf{i} i -boxes in the extent of C \mathcal{C} C . Conversely, every maximal commuting family of i \mathbf{i} i -boxes in an interval [ a , b ] [a,b] [ a , b ] arises in this way from an admissible chain of extent [ a , b ] [a,b] [ a , b ] ([32 Proposition 5.3] and [27 Corollary 2.10] ). In particular, every such maximal commuting family has cardinality b − a + 1 b-a+1 b − a + 1 .
Let F \mathcal{F} F be a maximal commuting family of i-boxes in [ a , b ] [a,b] [ a , b ] , and choose an admissible chain C = ( c 1 , … , c r ) \mathcal{C}=(c_1,\ldots,c_r) C = ( c 1 , … , c r ) of extent [ a , b ] [a,b] [ a , b ] such that F = F ( C ) \mathcal{F}=\mathcal{F}(\mathcal{C}) F = F ( C ) . Note that such an admissible chain C \mathcal{C} C need not be unique. Since ∣ C ~ k ∣ = ∣ C ~ k − 1 ∣ + 1 |\widetilde{\mathcal{C}}_k|=|\widetilde{\mathcal{C}}_{k-1}|+1 ∣ C k ∣ = ∣ C k − 1 ∣ + 1 , there exists a unique
z k ∈ C ~ k ∖ C ~ k − 1 and z k ∈ { x k , y k } , z_k\in\widetilde{\mathcal{C}}_k\setminus\widetilde{\mathcal{C}}_{k-1}\quad\text{and}\quad z_k\in\{x_k,y_k\}, z k ∈ C k ∖ C k − 1 and z k ∈ { x k , y k } ,
where c k = [ x k , y k ] c_k=[x_k,y_k] c k = [ x k , y k ] and C ~ 0 = ∅ \widetilde{\mathcal{C}}_0=\varnothing C 0 = ∅ .
By [27 Proposition 2.12] , the endpoint z k z_k z k depends only on F \mathcal{F} F and c k c_k c k , and not on the choice of an admissible chain realizing F \mathcal{F} F . We call z k z_k z k the effective end of c k c_k c k in F \mathcal{F} F , and write
eff F ( c k ) : = z k . \operatorname{eff}_{\mathcal{F}}(c_k):=z_k. eff F ( c k ) := z k .
Thus
eff F : F ⟶ [ a , b ] is a bijection. (29) \operatorname{eff}_{\mathcal{F}}:\mathcal{F}\longrightarrow[a,b]\text{ is a bijection.} \tag*{(29)} eff F : F ⟶ [ a , b ] is a bijection. (29)
We now associate an exchange matrix with a maximal commuting family F \mathcal{F} F of i-boxes. We first define the full exchange matrix
B f u l l F = ( b c , d F ) c , d ∈ F , B_{\mathrm{full}}^{\mathcal{F}}=(b_{c,\mathfrak d}^{\mathcal{F}})_{c,\mathfrak d\in\mathcal{F}}, B full F = ( b c , d F ) c , d ∈ F ,
with skew-symmetrizer d c : = d i c ( c ∈ F ) d_c:=d_{i_c}\quad(c\in\mathcal{F}) d c := d i c ( c ∈ F ) .
Let c = [ x , y ] c=[x,y] c = [ x , y ] and d = [ x ′ , y ′ ] \mathfrak d=[x',y'] d = [ x ′ , y ′ ] be distinct i-boxes in F \mathcal{F} F . The negative entries of B f u l l F B_{\mathrm{full}}^{\mathcal{F}} B full F are defined as follows:
b c , d F : = { − 1 , (i-1) x = x ′ and y ′ = y − , or (i-2) y = y ′ and x ′ = x − , c i c , i d , (ii) c i c , i d < 0 and one of (a)–(d) holds: (ii-a) [ x , y + ] ∈ F , eff F ( c ) = x , x − ′ < x < x ′ , y ′ < y + < y + ′ , (ii-b) [ x , y + ] ∈ F , eff F ( d ) = y ′ , x − ′ < x , y < y ′ < y + < y + ′ , (ii-c) [ x − ′ , y ′ ] ∈ F , eff F ( d ) = y ′ , x − < x − ′ < x , y < y ′ < y + , (ii-d) [ x − ′ , y ′ ] ∈ F , eff F ( c ) = x , x − < x − ′ < x < x ′ , y ′ < y + . (30) b_{c,\mathfrak d}^{\mathcal{F}}:=
\begin{cases}
-1, & \text{(i-1) }x=x'\text{ and }y'=y_-,\text{ or (i-2) }y=y'\text{ and }x'=x_-,\\
c_{i_c,i_{\mathfrak d}}, & \text{(ii) }c_{i_c,i_{\mathfrak d}}<0\text{ and one of (a)--(d) holds:}\\
&\text{(ii-a) }[x,y_+]\in\mathcal{F},\ \operatorname{eff}_{\mathcal{F}}(c)=x,\ x'_-<x<x',\ y'<y_+<y'_+,\\
&\text{(ii-b) }[x,y_+]\in\mathcal{F},\ \operatorname{eff}_{\mathcal{F}}(\mathfrak d)=y',\ x'_-<x,\ y<y'<y_+<y'_+,\\
&\text{(ii-c) }[x'_-,y']\in\mathcal{F},\ \operatorname{eff}_{\mathcal{F}}(\mathfrak d)=y',\ x_-<x'_-<x,\ y<y'<y_+,\\
&\text{(ii-d) }[x'_-,y']\in\mathcal{F},\ \operatorname{eff}_{\mathcal{F}}(c)=x,\ x_-<x'_-<x<x',\ y'<y_+.
\end{cases} \tag*{(30)} b c , d F := ⎩ ⎨ ⎧ − 1 , c i c , i d , (i-1) x = x ′ and y ′ = y − , or (i-2) y = y ′ and x ′ = x − , (ii) c i c , i d < 0 and one of (a)–(d) holds: (ii-a) [ x , y + ] ∈ F , eff F ( c ) = x , x − ′ < x < x ′ , y ′ < y + < y + ′ , (ii-b) [ x , y + ] ∈ F , eff F ( d ) = y ′ , x − ′ < x , y < y ′ < y + < y + ′ , (ii-c) [ x − ′ , y ′ ] ∈ F , eff F ( d ) = y ′ , x − < x − ′ < x , y < y ′ < y + , (ii-d) [ x − ′ , y ′ ] ∈ F , eff F ( c ) = x , x − < x − ′ < x < x ′ , y ′ < y + . (30)
We call the conditions (i-1) and (i-2) the horizontal relations , and the conditions (ii-a)--(ii-d) the vertical relations . The remaining entries are uniquely determined by
d c b c , d F = − d d b d , c F , b c , c F = 0. d_c b_{c,\mathfrak d}^{\mathcal{F}}=-d_{\mathfrak d}b_{\mathfrak d,c}^{\mathcal{F}},\qquad b_{c,c}^{\mathcal{F}}=0. d c b c , d F = − d d b d , c F , b c , c F = 0.
Our convention is opposite to that of [27 Section 3.2] ; namely, if B K K ( F ) B_{\mathrm{KK}}(\mathcal{F}) B KK ( F ) denotes the full exchange matrix defined in [27 Section 3.2] ,
B f u l l F = − B K K ( F ) . B_{\mathrm{full}}^{\mathcal{F}}=-B_{\mathrm{KK}}(\mathcal{F}). B full F = − B KK ( F ) .
An i-box c = [ x , y ] ∈ F 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 + = ∞ ,
and exchangeable otherwise. Set
F e x : = { c ∈ F ∣ c is exchangeable } . \mathcal{F}_{\mathrm{ex}}:=\{c\in\mathcal{F}\mid c\text{ is exchangeable}\}. F ex := { c ∈ F ∣ c is exchangeable } .
The exchange matrix associated with F \mathcal{F} F is the restriction
B F : = B f u l l F ∣ F × F e x = ( b c , d F ) c ∈ F , d ∈ F e x . B^{\mathcal{F}}:=B_{\mathrm{full}}^{\mathcal{F}}\big|_{\mathcal{F}\times\mathcal{F}_{\mathrm{ex}}}
=(b_{c,\mathfrak d}^{\mathcal{F}})_{c\in\mathcal{F},\mathfrak d\in\mathcal{F}_{\mathrm{ex}}}. B F := B full F F × F ex = ( b c , d F ) c ∈ F , d ∈ F ex .
For a maximal commuting family F \mathcal{F} F , we define the skew-symmetric matrix
Λ F = ( Λ c , d F ) c , d ∈ F by Λ c , d F : = λ i ( E c , E d ) . (31) \Lambda^{\mathcal{F}} = \left(\Lambda^{\mathcal{F}}_{c,d}\right)_{c,d\in\mathcal{F}} \quad\text{by}\quad\Lambda^{\mathcal{F}}_{c,d} := \lambda^{i}(\mathcal{E}_{c},\mathcal{E}_{d}). \tag*{(31)} Λ F = ( Λ c , d F ) c , d ∈ F by Λ c , d F := λ i ( E c , E d ) . (31)
For 1 ≤ k ≤ l 1 \le k \le l 1 ≤ k ≤ l , set
c k : = [ k min , k ] and F i : = { c k ∣ 1 ≤ k ≤ l } . c_k := [k_{\min},k]\quad\text{and}\quad\mathcal{F}^{i} := \{c_k \mid1 \le k \le l\}. c k := [ k m i n , k ] and F i := { c k ∣ 1 ≤ k ≤ l } .
Then F i \mathcal{F}^{i} F i is a maximal commuting family of i i i -boxes, and
B F i = B i and Λ F i = Λ i . B^{\mathcal{F}^{i}} = B^{i}\quad\text{and}\quad\Lambda^{\mathcal{F}^{i}} = \Lambda^{i}. B F i = B i and Λ F i = Λ i .
Indeed, the equality for the Λ \Lambda Λ -matrices follows immediately from (3.7) and (31) . For the exchange matrices, the horizontal relations in (i) of (30) give precisely the entries in (i) of (21) , while the vertical relations in (ii) of (30) reduce exactly to the conditions in (ii) of (21) .
We recall the { L , R } \{\mathcal{L},\mathcal{R}\} { L , R } -sequence associated with an admissible chain. Let C = ( c 1 , … , c r ) \mathcal{C}=(c_1,\ldots,c_r) C = ( c 1 , … , c r ) be an admissible chain of extent [ a , b ] [a,b] [ a , b ] , and write c k = [ x k , y k ] c_k=[x_k,y_k] c k = [ x k , y k ] and c ~ k = [ x ~ k , y ~ k ] \widetilde{c}_k=[\widetilde{x}_k,\widetilde{y}_k] c k = [ x k , y k ] . Note that C \mathcal{C} C is uniquely determined by its sequence of envelopes ( c ~ 1 , … , c ~ r ) (\widetilde{c}_1,\ldots,\widetilde{c}_r) ( c 1 , … , c r ) . Indeed, for 1 < k ≤ r 1<k\le r 1 < k ≤ r , we have
c k = [ x k , y k ] = { [ x ~ k , y ~ k − 1 ] , x ~ k = x ~ k − 1 − 1 , [ x ~ k − 1 , y ~ k ] , y ~ k = y ~ k − 1 + 1. (32) c_k=[x_k,y_k]=
\begin{cases}
[\widetilde{x}_k,\widetilde{y}_{k-1}],&\widetilde{x}_k=\widetilde{x}_{k-1}-1,\\
[\widetilde{x}_{k-1},\widetilde{y}_k],&\widetilde{y}_k=\widetilde{y}_{k-1}+1.
\end{cases} \tag*{(32)} c k = [ x k , y k ] = { [ x k , y k − 1 ] , [ x k − 1 , y k ] , x k = x k − 1 − 1 , y k = y k − 1 + 1. (32)
In the first case, we set T k − 1 : = L T_{k-1}:=\mathcal{L} T k − 1 := L , while in the second case we set T k − 1 : = R T_{k-1}:=\mathcal{R} T k − 1 := R . Thus, to C \mathcal{C} C we associate an { L , R } \{\mathcal{L},\mathcal{R}\} { L , R } -sequence
T ( C ) = ( T 1 , … , T r − 1 ) ∈ { L , R } r − 1 . \mathcal{T}(\mathcal{C})=(T_1,\ldots,T_{r-1})\in\{\mathcal{L},\mathcal{R}\}^{r-1}. T ( C ) = ( T 1 , … , T r − 1 ) ∈ { L , R } r − 1 .
Conversely, an { L , R } \{\mathcal{L},\mathcal{R}\} { L , R } -sequence
T = ( T 1 , … , T r − 1 ) ∈ { L , R } r − 1 \mathcal{T}=(T_1,\ldots,T_{r-1})\in\{\mathcal{L},\mathcal{R}\}^{r-1} T = ( T 1 , … , T r − 1 ) ∈ { L , R } r − 1
together with the extent [ a , b ] [a,b] [ a , b ] uniquely determines an admissible chain. More precisely, its initial envelope is the singleton
c ~ 1 = [ x 1 ] , x 1 = a + # { k ∣ T k = L } , \widetilde{c}_1=[x_1],\qquad x_1=a+\#\{k\mid T_k=\mathcal{L}\}, c 1 = [ x 1 ] , x 1 = a + # { k ∣ T k = L } ,
and the subsequent envelopes are determined recursively by
c ~ k + 1 = { [ x ~ k − 1 , y ~ k ] , T k = L , [ x ~ k , y ~ k + 1 ] , T k = R , ( 1 ≤ k < r ) . \widetilde{c}_{k+1}=
\begin{cases}
[\widetilde{x}_k-1,\widetilde{y}_k],&T_k=\mathcal{L},\\
[\widetilde{x}_k,\widetilde{y}_k+1],&T_k=\mathcal{R},
\end{cases}
\qquad(1\le k<r). c k + 1 = { [ x k − 1 , y k ] , [ x k , y k + 1 ] , T k = L , T k = R , ( 1 ≤ k < r ) .
The corresponding i i i -boxes are then recovered from (32) . Hence admissible chains of extent [ a , b ] [a,b] [ a , b ] are in bijection with { L , R } \{\mathcal{L},\mathcal{R}\} { L , R } -sequences of length r − 1 = b − a r-1=b-a r − 1 = b − a .
For 1 ≤ k < r 1\le k<r 1 ≤ k < r , we say that c k c_k c k is movable if k = 1 k=1 k = 1 or T k − 1 ≠ T k T_{k-1}\ne T_k T k − 1 = T k (k ≥ 2 k\ge2 k ≥ 2 ). For a movable c k c_k c k in C \mathcal{C} C , the box move of C \mathcal{C} C at k k k is the admissible chain B k ( C ) \mathcal{B}_k(\mathcal{C}) B k ( C ) determined by x 1 ′ x'_1 x 1 ′ and T ′ = ( T 1 ′ , … , T r − 1 ′ ) \mathcal{T}'=(T'_1,\ldots,T'_{r-1}) T ′ = ( T 1 ′ , … , T r − 1 ′ ) as follows:
(i) { x 1 ′ = x 1 ± 1 , if k = 1 and T 1 = R (resp. L ) , x 1 ′ = x 1 , if k > 1 , \text{(i)}\qquad
\begin{cases}
x'_1=x_1\pm1,&\text{if }k=1\text{ and }T_1=\mathcal{R}\text{ (resp. }\mathcal{L}\text{)},\\
x'_1=x_1,&\text{if }k>1,
\end{cases} (i) { x 1 ′ = x 1 ± 1 , x 1 ′ = x 1 , if k = 1 and T 1 = R (resp. L ) , if k > 1 ,
(ii) T s ′ = T s for s ∉ { k − 1 , k } , T s ′ ≠ T s for s ∈ { k − 1 , k } . \text{(ii)}\qquad T'_s=T_s\quad\text{for }s\notin\{k-1,k\},\qquad T'_s\ne T_s\quad\text{for }s\in\{k-1,k\}. (ii) T s ′ = T s for s ∈ / { k − 1 , k } , T s ′ = T s for s ∈ { k − 1 , k } .
Thus B k ( C ) \mathcal{B}_k(\mathcal{C}) B k ( C ) is obtained from C \mathcal{C} C by shifting c ~ k \widetilde{c}_k c k by one step to the right or to the left inside c ~ k + 1 \widetilde{c}_{k+1} c k + 1 .
Proposition 3.3 ([32 Proposition 5.6, Proposition 5.7] ). Let C = ( c k ) 1 ≤ k ≤ l \mathcal{C}=(c_k)_{1\le k\le l} C = ( c k ) 1 ≤ k ≤ l be an admissible chain of i i i -boxes and let k 0 k_0 k 0 be the index of a movable i i i -box (1 ≤ k 0 < l 1\le k_0<l 1 ≤ k 0 < l ). Set B k 0 ( C ) = ( c k ′ ) 1 ≤ k ≤ l \mathcal{B}_{k_0}(\mathcal{C})=(c'_k)_{1\le k\le l} B k 0 ( C ) = ( c k ′ ) 1 ≤ k ≤ l . (i) Assume that i ~ x ~ k 0 + 1 ≠ i ~ y ~ k 0 + 1 \tilde{i}_{\tilde{x}_{k_0+1}} \ne\tilde{i}_{\tilde{y}_{k_0+1}} i ~ x ~ k 0 + 1 = i ~ y ~ k 0 + 1 , i.e., c ~ k 0 + 1 \tilde{c}_{k_0+1} c ~ k 0 + 1 is not an i i i -box. Then we have
c k ′ = c s k 0 ( k ) for 1 ≤ k ≤ l , c'_k = c_{s_{k_0}(k)} \qquad\text{for } 1 \le k \le l, c k ′ = c s k 0 ( k ) for 1 ≤ k ≤ l ,
where s k 0 ∈ S l s_{k_0} \in\mathcal{S}_l s k 0 ∈ S l is the transposition of k 0 k_0 k 0 and k 0 + 1 k_0+1 k 0 + 1 .
(ii) Assume that i ~ x ~ k 0 + 1 = i ~ y ~ k 0 + 1 \tilde{i}_{\tilde{x}_{k_0+1}} = \tilde{i}_{\tilde{y}_{k_0+1}} i ~ x ~ k 0 + 1 = i ~ y ~ k 0 + 1 , i.e., c ~ k 0 + 1 \tilde{c}_{k_0+1} c ~ k 0 + 1 is an i i i -box. Set c k 0 + 1 = [ x , y ] c_{k_0+1} = [x,y] c k 0 + 1 = [ x , y ] with x = x ~ k 0 + 1 x = \tilde{x}_{k_0+1} x = x ~ k 0 + 1 and y = y ~ k 0 + 1 y = \tilde{y}_{k_0+1} y = y ~ k 0 + 1 .
If k 0 > 1 k_0 > 1 k 0 > 1 , then (a) c k 0 = [ x + , y ] c_{k_0} = [x_+,y] c k 0 = [ x + , y ] and c k 0 ′ = [ x , y − ] c'_{k_0} = [x,y_-] c k 0 ′ = [ x , y − ] if T k 0 − 1 = R T_{k_0-1} = \mathcal{R} T k 0 − 1 = R , (b) c k 0 = [ x , y − ] c_{k_0} = [x,y_-] c k 0 = [ x , y − ] and c k 0 ′ = [ x + , y ] c'_{k_0} = [x_+,y] c k 0 ′ = [ x + , y ] if T k 0 − 1 = L T_{k_0-1} = \mathcal{L} T k 0 − 1 = L .
If k 0 = 1 k_0 = 1 k 0 = 1 , then y = x + 1 y = x+1 y = x + 1 and
T 1 = L : c 1 = [ y ] , c 1 ′ = [ x ] , T_1 = \mathcal{L}: \quad c_1 = [y], \quad c'_1 = [x], T 1 = L : c 1 = [ y ] , c 1 ′ = [ x ] ,
T 1 = R : c 1 = [ x ] , c 1 ′ = [ y ] . T_1 = \mathcal{R}: \quad c_1 = [x], \quad c'_1 = [y]. T 1 = R : c 1 = [ x ] , c 1 ′ = [ y ] .
A box move as in (i) is called a permutation-type box move , while a box move as in (ii) is called a mutation-type box move .
By abuse of notation, we also write B k F \mathcal{B}_k\mathcal{F} B k F for the maximal commuting family obtained from F \mathcal{F} F by the box move B k \mathcal{B}_k B k .
Proposition 3.4. Let F ′ = B k F \mathcal{F}' = \mathcal{B}_k\mathcal{F} F ′ = B k F .
(a) If B k \mathcal{B}_k B k is of permutation type, then
B F ′ = P B F ( P e x ) T and Λ F ′ = P Λ F P T , B^{\mathcal{F}'} = P B^{\mathcal{F}} (P_{\mathrm{ex}})^T \qquad\text{and} \qquad\Lambda^{\mathcal{F}'} = P \Lambda^{\mathcal{F}} P^T, B F ′ = P B F ( P ex ) T and Λ F ′ = P Λ F P T , where P P P is the induced permutation matrix and P e x P_{\mathrm{ex}} P ex is its restriction to the exchangeable indices.
(b) If B k \mathcal{B}_k B k is of mutation type at c ∈ F c \in\mathcal{F} c ∈ F , then
B F ′ = μ c ( B F ) and Λ F ′ = μ c ( Λ F ) . B^{\mathcal{F}'} = \mu_c(B^{\mathcal{F}}) \qquad\text{and} \qquad\Lambda^{\mathcal{F}'} = \mu_c(\Lambda^{\mathcal{F}}). B F ′ = μ c ( B F ) and Λ F ′ = μ c ( Λ F ) .
Proof. The permutation case follows immediately from the definitions. Assume that B k \mathcal{B}_k B k is of mutation type at c ∈ F c \in\mathcal{F} c ∈ F , and let c ′ c' c ′ be the new i i i -box. The equality
B F ′ = μ c ( B F ) B^{\mathcal{F}'} = \mu_c(B^{\mathcal{F}}) B F ′ = μ c ( B F )
is the local box-move formula of [27 ] , with our opposite sign convention. We explain below how the relevant entries of the c c c -column follow from (30) . It remains to prove that
Λ F ′ = μ c ( Λ F ) . \Lambda^{\mathcal{F}'} = \mu_c(\Lambda^{\mathcal{F}}). Λ F ′ = μ c ( Λ F ) .
We consider the two orientations separately.
(i) Suppose that
c = [ x + , y ] and c ′ = [ x , y − ] , i : = i x = i y . c = [x_+,y] \qquad\text{and} \qquad c' = [x,y_-], \qquad i := i_x = i_y. c = [ x + , y ] and c ′ = [ x , y − ] , i := i x = i y .
For k > 1 k > 1 k > 1 , one has T k − 1 = R T_{k-1} = \mathcal{R} T k − 1 = R ; for k = 1 k = 1 k = 1 , one has c = [ y ] c = [y] c = [ y ] . Thus
eff F ( c ) = y . \operatorname{eff}_{\mathcal{F}}(c) = y. eff F ( c ) = y .
a j = [ x + ( j ) , y − ( j ) ] a_j = [x^{(j)}_+,y^{(j)}_-] a j = [ x + ( j ) , y − ( j ) ]
whenever this i i i -box is nonempty. By Proposition 3.3 (ii), F \mathcal{F} F contains the i i i -box [ x , y ] [x,y] [ x , y ] . For the ordered pair ( a j , c ) (\mathbf{a}_j,\mathbf{c}) ( a j , c ) , condition (ii-c) of (30) implies that
b a j , c F = c j , i ≤ 0. b^{\mathcal{F}}_{\mathbf{a}_j,\mathbf{c}}=c_{j,i}\leq0. b a j , c F = c j , i ≤ 0.
Thus, apart from the boxes a j \mathbf{a}_j a j above, all boxes contribute zero to the negative part of the c \mathbf{c} c -column. Consequently,
[ − b a j , c F ] + = − c j , i ( j ≠ i ) . (33) \left[-b^{\mathcal{F}}_{\mathbf{a}_j,\mathbf{c}}\right]_+=-c_{j,i}\qquad(j\neq i). \tag*{(33)} [ − b a j , c F ] + = − c j , i ( j = i ) . (33)
where empty boxes a j \mathbf{a}_j a j are omitted.
We next compare the corresponding box vectors. By definition,
E c = ∑ x + ≤ p ≤ y i p = i ε p and E c ′ = ∑ x ≤ p ≤ y − i p = i ε p . \mathcal{E}_{\mathbf{c}}=\sum_{\substack{x+\leq p\leq y\\ i_p=i}}\varepsilon_p
\quad\text{and}\quad
\mathcal{E}_{\mathbf{c}'}=\sum_{\substack{x\leq p\leq y-\\ i_p=i}}\varepsilon_p. E c = x + ≤ p ≤ y i p = i ∑ ε p and E c ′ = x ≤ p ≤ y − i p = i ∑ ε p .
Thus their sum contains ε x \varepsilon_x ε x and ε y \varepsilon_y ε y once and every i i i -colored position strictly between x x x and y y y twice. Since c i , i = 2 c_{i,i}=2 c i , i = 2 , we have
E c ′ + E c = ε x + ε y + ∑ x < p < y i p = i c i p , i ε p . \mathcal{E}_{\mathbf{c}'}+\mathcal{E}_{\mathbf{c}}
=\varepsilon_x+\varepsilon_y+
\sum_{\substack{x<p<y\\ i_p=i}}c_{i_p,i}\varepsilon_p. E c ′ + E c = ε x + ε y + x < p < y i p = i ∑ c i p , i ε p .
For j ≠ i j\neq i j = i , the positions of color j j j in a j \mathbf{a}_j a j are precisely those lying strictly between x x x and y y y . Hence
c j , i E a j = ∑ x < p < y i p = j c i p , i ε p . c_{j,i}\mathcal{E}_{\mathbf{a}_j}
=\sum_{\substack{x<p<y\\ i_p=j}}c_{i_p,i}\varepsilon_p. c j , i E a j = x < p < y i p = j ∑ c i p , i ε p .
Summing over all j ≠ i j\neq i j = i gives
E c ′ + E c − ∑ j ≠ i ( − c j , i ) E a j = ε x + ε y + ∑ x < p < y c i p , i ε p . (34) \mathcal{E}_{\mathbf{c}'}+\mathcal{E}_{\mathbf{c}}
-\sum_{j\neq i}(-c_{j,i})\mathcal{E}_{\mathbf{a}_j}
=\varepsilon_x+\varepsilon_y+
\sum_{x<p<y}c_{i_p,i}\varepsilon_p. \tag*{(34)} E c ′ + E c − j = i ∑ ( − c j , i ) E a j = ε x + ε y + x < p < y ∑ c i p , i ε p . (34)
Since i x = i y = i i_x=i_y=i i x = i y = i , summing w p ( α i ) − w p − 1 ( α i ) = − c i p , i β p i w_p(\alpha_i)-w_{p-1}(\alpha_i)=-c_{i_p,i}\beta_p^i w p ( α i ) − w p − 1 ( α i ) = − c i p , i β p i over x ≤ p < y x\leq p<y x ≤ p < y gives
β x i + β y i + ∑ x < p < y c i p , i β p i = 0. \beta_x^i+\beta_y^i+\sum_{x<p<y}c_{i_p,i}\beta_p^i=0. β x i + β y i + x < p < y ∑ c i p , i β p i = 0.
Summing (27) over the positions p p p with x ≤ p < y x\leq p<y x ≤ p < y and i p = i i_p=i i p = i gives
λ i ( ε x + ε y + ∑ x < p < y c i p , i ε p , ε q ) = 2 d i ( δ q , y − δ q , x ) . (35) \lambda^i\left(\varepsilon_x+\varepsilon_y+
\sum_{x<p<y}c_{i_p,i}\varepsilon_p,\varepsilon_q\right)
=2d_i(\delta_{q,y}-\delta_{q,x}). \tag*{(35)} λ i ( ε x + ε y + x < p < y ∑ c i p , i ε p , ε q ) = 2 d i ( δ q , y − δ q , x ) . (35)
Now let d ∈ F ∩ F ′ \mathbf{d}\in\mathcal{F}\cap\mathcal{F}' d ∈ F ∩ F ′ . By (31) , (33) , and (34) , we have
Λ c ′ , d F ′ + Λ c , d F − ∑ a ∈ F [ − b a , c F ] + Λ a , d F = λ i ( ε x + ε y + ∑ x < p < y c i p , i ε p , ε d ) . \Lambda^{\mathcal{F}'}_{\mathbf{c}',\mathbf{d}}+
\Lambda^{\mathcal{F}}_{\mathbf{c},\mathbf{d}}
-\sum_{\mathbf{a}\in\mathcal{F}}
\left[-b^{\mathcal{F}}_{\mathbf{a},\mathbf{c}}\right]_+
\Lambda^{\mathcal{F}}_{\mathbf{a},\mathbf{d}}
=
\lambda^i\left(\varepsilon_x+\varepsilon_y+
\sum_{x<p<y}c_{i_p,i}\varepsilon_p,\varepsilon_d\right). Λ c ′ , d F ′ + Λ c , d F − a ∈ F ∑ [ − b a , c F ] + Λ a , d F = λ i ( ε x + ε y + x < p < y ∑ c i p , i ε p , ε d ) .
Using (35) and the definition of E d \mathcal{E}_{\mathbf{d}} E d , the right-hand side is
2 d i ∑ q ∈ E d i q = i d ( δ q , y − δ q , x ) . 2d_i\sum_{\substack{\mathbf{q}\in\mathcal{E}_{\mathbf{d}}\\ i_q=i_d}}
(\delta_{q,y}-\delta_{q,x}). 2 d i q ∈ E d i q = i d ∑ ( δ q , y − δ q , x ) .
If i 0 ≠ i i_0 \ne i i 0 = i , both delta terms vanish. Suppose that i 0 = i i_0=i i 0 = i . Since d ∈ F ∩ F ′ \mathfrak{d} \in\mathcal{F}\cap\mathcal{F}' d ∈ F ∩ F ′ , the box d \mathfrak{d} d commutes with both c = [ x + , y ] \mathfrak{c}=[x_+,y] c = [ x + , y ] and c ′ = [ x , y − ] \mathfrak{c}'=[x,y_-] c ′ = [ x , y − ] . The commuting condition gives
x ∈ d ⟺ y ∈ d . x\in\mathfrak{d}\quad\Longleftrightarrow\quad y\in\mathfrak{d}. x ∈ d ⟺ y ∈ d .
Thus the two delta contributions cancel, and therefore
Λ c ′ , d F ′ = − Λ c , d F + ∑ a ∈ F [ − b a , c F ] + Λ a , d F . \Lambda^{\mathcal{F}'}_{\mathfrak{c}',\mathfrak{d}}=-\Lambda^{\mathcal{F}}_{\mathfrak{c},\mathfrak{d}}+\sum_{\mathfrak{a}\in\mathcal{F}}[-b^{\mathcal{F}}_{\mathfrak{a},\mathfrak{c}}]_+\Lambda^{\mathcal{F}}_{\mathfrak{a},\mathfrak{d}}. Λ c ′ , d F ′ = − Λ c , d F + a ∈ F ∑ [ − b a , c F ] + Λ a , d F .
(ii) Suppose that
c = [ x , y − ] and c ′ = [ x + , y ] . \mathfrak{c}=[x,y_-]\quad\text{and}\quad\mathfrak{c}'=[x_+,y]. c = [ x , y − ] and c ′ = [ x + , y ] .
For k > 1 k>1 k > 1 , one has T k − 1 = L T_{k-1}=\mathcal{L} T k − 1 = L ; for k = 1 k=1 k = 1 , one has c = [ x ] \mathfrak{c}=[x] c = [ x ] . Thus
eff F ( c ) = x . \operatorname{eff}_{\mathcal{F}}(\mathfrak{c})=x. eff F ( c ) = x .
By Proposition 3.3, the box [ x , y ] [x,y] [ x , y ] belongs to F \mathcal{F} F . It has the same left endpoint as c = [ x , y − ] \mathfrak{c}=[x,y_-] c = [ x , y − ] . Hence condition (i-1) of (30) gives
b [ x , y ] , c F = − 1. b^{\mathcal{F}}_{[x,y],\mathfrak{c}}=-1. b [ x , y ] , c F = − 1.
If [ x + , y − ] [x_+,y_-] [ x + , y − ] is nonempty, it also belongs to F \mathcal{F} F . This box has the same right endpoint as c \mathfrak{c} c . Hence condition (i-2) gives
b [ x + , y − ] , c F = − 1. b^{\mathcal{F}}_{[x_+,y_-],\mathfrak{c}}=-1. b [ x + , y − ] , c F = − 1.
These are negative entries in the c \mathfrak{c} c -column. Therefore,
[ − b [ x , y ] , c F ] + = 1 and [ − b [ x + , y − ] , c F ] + = δ ( [ x + , y − ] ≠ ∅ ) . (36) [-b^{\mathcal{F}}_{[x,y],\mathfrak{c}}]_+=1\quad\text{and}\quad[-b^{\mathcal{F}}_{[x_+,y_-],\mathfrak{c}}]_+=\delta([x_+,y_-]\ne\varnothing). \tag*{(36)} [ − b [ x , y ] , c F ] + = 1 and [ − b [ x + , y − ] , c F ] + = δ ([ x + , y − ] = ∅ ) . (36)
Note that the effective-end conditions in the vertical cases (ii-a)–(ii-d) are not satisfied for any additional negative entry in this orientation. Thus the only nonzero coefficients [ − b a , c F ] + [-b^{\mathcal{F}}_{\mathfrak{a},\mathfrak{c}}]_+ [ − b a , c F ] + are given in (36) .
The corresponding box vectors satisfy
E c ′ + E c = E [ x + , y − ] + E [ x , y ] . (37) E_{\mathfrak{c}'}+E_{\mathfrak{c}}=E_{[x_+,y_-]}+E_{[x,y]}. \tag*{(37)} E c ′ + E c = E [ x + , y − ] + E [ x , y ] . (37)
Indeed, both sides count ε x \varepsilon_x ε x and ε y \varepsilon_y ε y once and every i i i -colored position strictly between them twice.
Let d ∈ F ∩ F ′ \mathfrak{d}\in\mathcal{F}\cap\mathcal{F}' d ∈ F ∩ F ′ . By (31) , (36) , and (37) , we obtain
Λ c ′ , d F ′ + Λ c , d F = Λ [ x + , y − ] , d F + Λ [ x , y ] , d F = ∑ a ∈ F [ − b a , c F ] + Λ a , d F . \Lambda^{\mathcal{F}'}_{\mathfrak{c}',\mathfrak{d}}+\Lambda^{\mathcal{F}}_{\mathfrak{c},\mathfrak{d}}=\Lambda^{\mathcal{F}}_{[x_+,y_-],\mathfrak{d}}+\Lambda^{\mathcal{F}}_{[x,y],\mathfrak{d}}=\sum_{\mathfrak{a}\in\mathcal{F}}[-b^{\mathcal{F}}_{\mathfrak{a},\mathfrak{c}}]_+\Lambda^{\mathcal{F}}_{\mathfrak{a},\mathfrak{d}}. Λ c ′ , d F ′ + Λ c , d F = Λ [ x + , y − ] , d F + Λ [ x , y ] , d F = a ∈ F ∑ [ − b a , c F ] + Λ a , d F .
Hence
Λ c ′ , d F ′ = − Λ c , d F + ∑ a ∈ F [ − b a , c F ] + Λ a , d F . \Lambda^{\mathcal{F}'}_{\mathfrak{c}',\mathfrak{d}}=-\Lambda^{\mathcal{F}}_{\mathfrak{c},\mathfrak{d}}+\sum_{\mathfrak{a}\in\mathcal{F}}[-b^{\mathcal{F}}_{\mathfrak{a},\mathfrak{c}}]_+\Lambda^{\mathcal{F}}_{\mathfrak{a},\mathfrak{d}}. Λ c ′ , d F ′ = − Λ c , d F + a ∈ F ∑ [ − b a , c F ] + Λ a , d F .
Thus, in both orientations,
Λ F ′ = μ c ( Λ F ) . \Lambda^{\mathcal{F}'}=\mu_{\mathfrak{c}}(\Lambda^{\mathcal{F}}). Λ F ′ = μ c ( Λ F ) .
Theorem 3.5. 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, and let F \mathcal{F} F be a maximal commuting family of i i i -boxes in [ 1 , l ] [1,l] [ 1 , l ] . Then ( B F , Λ F ) (B^{\mathcal{F}},\Lambda^{\mathcal{F}}) ( B F , Λ F ) is a compatible pair. More precisely,
∑ c ∈ F b c , d F Λ c , u F = 2 d i d δ d , u ( d ∈ F e x , u ∈ F ) . \sum_{\mathfrak{c}\in\mathcal{F}} b^{\mathcal{F}}_{\mathfrak{c},\mathfrak{d}}\Lambda^{\mathcal{F}}_{\mathfrak{c},\mathfrak{u}}=2d_{i_{\mathfrak{d}}}\delta_{\mathfrak{d},\mathfrak{u}}\qquad(\mathfrak{d}\in\mathcal{F}_{\mathrm{ex}},\ \mathfrak{u}\in\mathcal{F}). c ∈ F ∑ b c , d F Λ c , u F = 2 d i d δ d , u ( d ∈ F ex , u ∈ F ) .
Proof. Let F i \mathcal{F}^{i} F i be the maximal commuting family corresponding to the constant R \mathcal{R} R -sequence. Since
B F i = B i and Λ F i = Λ i , B^{\mathcal{F}^{i}}=B^{i}\quad\text{and}\quad\Lambda^{\mathcal{F}^{i}}=\Lambda^{i}, B F i = B i and Λ F i = Λ i ,
Proposition 3.1 gives
∑ c ∈ F i b c , d F i Λ c , u F i = 2 d i d δ d , u ( d ∈ F e x i , u ∈ F i ) . \sum_{c\in\mathcal{F}^{i}}b^{\mathcal{F}^{i}}_{c,d}\Lambda^{\mathcal{F}^{i}}_{c,u}=2d_{i_d}\delta_{d,u}\qquad(d\in\mathcal{F}^{i}_{\mathrm{ex}},\ u\in\mathcal{F}^{i}). c ∈ F i ∑ b c , d F i Λ c , u F i = 2 d i d δ d , u ( d ∈ F ex i , u ∈ F i ) .
Every maximal commuting family F \mathcal{F} F is connected to F i \mathcal{F}^{i} F i by a sequence of box moves. By Proposition 3.4, at each step the pair ( B , Λ ) (B,\Lambda) ( B , Λ ) is transformed by either a simultaneous permutation or a simultaneous matrix mutation. Since compatibility is preserved under both operations, we obtain
∑ c ∈ F b c , d F Λ c , u F = 2 d i d δ d , u ( d ∈ F e x , u ∈ F ) . \sum_{c\in\mathcal{F}}b^{\mathcal{F}}_{c,d}\Lambda^{\mathcal{F}}_{c,u}=2d_{i_d}\delta_{d,u}\qquad(d\in\mathcal{F}_{\mathrm{ex}},\ u\in\mathcal{F}). c ∈ F ∑ b c , d F Λ c , u F = 2 d i d δ d , u ( d ∈ F ex , u ∈ F ) .
For reduced sequences, the corresponding compatibility appears in the quiver Hecke setting of [27 ] ; Theorem 3.5 applies to an arbitrary finite sequence.
For a maximal commuting family G \mathcal{G} G of i \mathbf{i} i -boxes, let Σ G = ( x G , B G ) \Sigma^{\mathcal{G}}=(\mathbf{x}^{\mathcal{G}},B^{\mathcal{G}}) Σ G = ( x G , B G ) be the associated commutative seed, where x G = ( x c G ∣ c ∈ G ) \mathbf{x}^{\mathcal{G}}=(x^{\mathcal{G}}_{c}\mid c\in\mathcal{G}) x G = ( x c G ∣ c ∈ G ) .
Proposition 3.6. Let i = ( i 1 , … , i l ) ∈ I l \mathbf{i}=(i_{1},\ldots,i_{l})\in I^{l} i = ( i 1 , … , i l ) ∈ I l be an arbitrary finite sequence, let G \mathcal{G} G be a maximal commuting family of i \mathbf{i} i -boxes in [ 1 , l ] [1,l] [ 1 , l ] , and let G ′ = B k G \mathcal{G}'=\mathcal{B}_{k}\mathcal{G} G ′ = B k G be obtained by a mutation-type box move replacing [ x + , y ] [x_{+},y] [ x + , y ] with [ x , y − ] [x,y_{-}] [ x , y − ] . Set i : = i x = i y i:=i_{x}=i_{y} i := i x = i y and x ∅ G : = 1 x^{\mathcal{G}}_{\varnothing}:=1 x ∅ G := 1 . The corresponding mutation from Σ G \Sigma^{\mathcal{G}} Σ G to Σ G ′ \Sigma^{\mathcal{G}'} Σ G ′ has exchange matrix B G ′ B^{\mathcal{G}'} B G ′ and exchange relation
x [ x + , y ] G x [ x , y − ] G ′ = x [ x , y ] G x [ x + , y − ] G + ∏ j ∈ I ∖ { i } , c j , i < 0 x ( j ) + ≤ y ( j ) − ( x [ x ( j ) + , y ( j ) − ] G ) − c j , i . (38) x^{\mathcal{G}}_{[x_{+},y]}x^{\mathcal{G}'}_{[x,y_{-}]}=x^{\mathcal{G}}_{[x,y]}x^{\mathcal{G}}_{[x_{+},y_{-}]}+\prod_{\substack{j\in I\setminus\{i\},\ c_{j,i}<0\\ x(j)_{+}\leq y(j)_{-}}}\left(x^{\mathcal{G}}_{[x(j)_{+},y(j)_{-}]}\right)^{-c_{j,i}}. \tag*{(38)} x [ x + , y ] G x [ x , y − ] G ′ = x [ x , y ] G x [ x + , y − ] G + j ∈ I ∖ { i } , c j , i < 0 x ( j ) + ≤ y ( j ) − ∏ ( x [ x ( j ) + , y ( j ) − ] G ) − c j , i . (38)
Proof. By Proposition 3.3(ii) and (32) , the box [ x , y ] [x,y] [ x , y ] belongs to G ∩ G ′ \mathcal{G}\cap\mathcal{G}' G ∩ G ′ . For k > 1 k>1 k > 1 , the boxes preceding the mutated box form a maximal commuting family in [ x + 1 , y − 1 ] [x+1,y-1] [ x + 1 , y − 1 ] . Every nonempty box [ x + , y − ] [x_{+},y_{-}] [ x + , y − ] or [ x ( j ) + , y ( j ) − ] [x(j)_{+},y(j)_{-}] [ x ( j ) + , y ( j ) − ] commutes with this family and hence belongs to G ∩ G ′ \mathcal{G}\cap\mathcal{G}' G ∩ G ′ by maximality.
By (33) , the negative entries of the mutation column are c j , i c_{j,i} c j , i at [ x ( j ) + , y ( j ) − ] [x(j)_{+},y(j)_{-}] [ x ( j ) + , y ( j ) − ] for the indicated colors j j j . By (36) and Proposition 3.4, the positive entries are 1 1 1 at [ x , y ] [x,y] [ x , y ] and, when nonempty, [ x + , y − ] [x_{+},y_{-}] [ x + , y − ] . All other entries vanish, so the cluster mutation formula gives (38) . For k = 1 k=1 k = 1 , all interior boxes are empty, and the same argument applies with the empty-box convention.
Remark 3.7. For finite symmetric type, the exchange relation in the proposition is the specialization at q 1 / 2 = 1 q^{1/2}=1 q 1/2 = 1 of the categorical T T T -system in [32 , 34 , 27 ] . The proposition gives the same classical exchange relation for arbitrary finite sequences and symmetrizable Cartan data through the exchange matrices of i \mathbf{i} i -box seeds.
Folding of compatible pairs
We compare the compatible pairs associated with an arbitrary finite sequence i \mathbf{i} i and its lift i ^ \widehat{\mathbf{i}} i . We first establish the folding formulas for the exchange and Λ \Lambda Λ -matrices, and then extend them to maximal commuting i \mathbf{i} i -box families. Throughout this subsection, we refer to the data associated with i \mathbf{i} i and i ^ \widehat{\mathbf{i}} i as the downstairs and upstairs data, respectively.
Retain the notation of Subsections 2.2–2.3, and identify the position κ ( k , r ) \kappa(k,r) κ ( k , r ) of i ^ \hat{i} i ^ with ( k , r ) (k,r) ( k , r ) .
Proposition 3.8. For 1 ≤ k , p ≤ l 1 \leq k,p \leq l 1 ≤ k , p ≤ l and 1 ≤ t ≤ e i p 1 \leq t \leq e_{i_p} 1 ≤ t ≤ e i p , we have
b k , p i = ∑ r = 1 e i k b ^ ( k , r ) , ( p , t ) i ( p ∈ J e x i ) , and Λ k , p i = d i k ∑ r = 1 e i k Λ ^ ( k , r ) , ( p , t ) i . b^i_{k,p}=\sum_{r=1}^{e_{i_k}}\widehat{b}^i_{(k,r),(p,t)}\quad(p\in J^i_{\mathrm{ex}}),\qquad\text{and}\qquad\Lambda^i_{k,p}=d_{i_k}\sum_{r=1}^{e_{i_k}}\widehat{\Lambda}^i_{(k,r),(p,t)}. b k , p i = r = 1 ∑ e i k b ( k , r ) , ( p , t ) i ( p ∈ J ex i ) , and Λ k , p i = d i k r = 1 ∑ e i k Λ ( k , r ) , ( p , t ) i . In particular, both right-hand sides are independent of the choice of t t t .
Proof. Fix t t t . We first prove the assertion for the exchange matrices. If k = p + k=p_+ k = p + , then
b ^ ( k , r ) , ( p , t ) i = { − 1 , r = t , 0 , r ≠ t , \widehat{b}^i_{(k,r),(p,t)}=
\begin{cases}
-1,&r=t,\\
0,&r\ne t,
\end{cases} b ( k , r ) , ( p , t ) i = { − 1 , 0 , r = t , r = t , and hence
∑ r = 1 e i k b ^ ( k , r ) , ( p , t ) i = − 1 = b k , p i . \sum_{r=1}^{e_{i_k}}\widehat{b}^i_{(k,r),(p,t)}=-1=b^i_{k,p}. r = 1 ∑ e i k b ( k , r ) , ( p , t ) i = − 1 = b k , p i . Similarly, if p = k + p=k_+ p = k + , the fiber sum is 1 = b k , p i 1=b^i_{k,p} 1 = b k , p i .
For the vertical relations, the blockwise lift preserves the strict interlacing conditions, and
∑ r = 1 e i k c ^ i k , r ; i p , t = c i k , i p . \sum_{r=1}^{e_{i_k}}\widehat{c}_{i_k,r;i_p,t}=c_{i_k,i_p}. r = 1 ∑ e i k c i k , r ; i p , t = c i k , i p . Hence the corresponding fiber sum is respectively c i k , i p c_{i_k,i_p} c i k , i p or − c i k , i p -c_{i_k,i_p} − c i k , i p , as in (21) . In all remaining cases both sides vanish.
For the Λ \Lambda Λ -matrices, set
E ^ k , r : = ∑ 1 ≤ a ≤ k i a = i k ε ^ a , r . \widehat{E}_{k,r}:=\sum_{\substack{1\leq a\leq k\\ i_a=i_k}}\widehat{\varepsilon}_{a,r}. E k , r := 1 ≤ a ≤ k i a = i k ∑ ε a , r . By (3.7) and Lemma 2.7 ,
Λ k , p i = λ i ( E k , E p ) = d i k ∑ r = 1 e i k λ ^ ( E ^ k , r , E ^ p , t ) = d i k ∑ r = 1 e i k Λ ^ ( k , r ) , ( p , t ) i . \Lambda^i_{k,p}=\lambda^i(E_k,E_p)=d_{i_k}\sum_{r=1}^{e_{i_k}}\widehat{\lambda}(\widehat{E}_{k,r},\widehat{E}_{p,t})=d_{i_k}\sum_{r=1}^{e_{i_k}}\widehat{\Lambda}^i_{(k,r),(p,t)}. Λ k , p i = λ i ( E k , E p ) = d i k r = 1 ∑ e i k λ ( E k , r , E p , t ) = d i k r = 1 ∑ e i k Λ ( k , r ) , ( p , t ) i .
Remark 3.9. The exchange-matrix part of Proposition 3.8 is reminiscent of the classical folding of exchange matrices; see, for example, [6 §4.4] , where global foldability is also considered. To our knowledge, the simultaneous folding of the exchange and compatible Λ \Lambda Λ -matrices for GLS seeds associated with sequences, as above, has not appeared before.
Lemma 3.10. For 1 ≤ k ≤ l 1 \leq k \leq l 1 ≤ k ≤ l , p ∈ J e x i p\in J^i_{\mathrm{ex}} p ∈ J ex i , and 1 ≤ t ≤ e i p 1 \leq t \leq e_{i_p} 1 ≤ t ≤ e i p , the integers
b ^ κ ( k , r ) , κ ( p , t ) i ( 1 ≤ r ≤ e i k ) \widehat{b}^i_{\kappa(k,r),\kappa(p,t)}\qquad(1\leq r\leq e_{i_k}) b κ ( k , r ) , κ ( p , t ) i ( 1 ≤ r ≤ e i k ) have the same weak sign. Consequently,
∑ r = 1 e i k [ b ^ κ ( k , r ) , κ ( p , t ) i ] + = [ b k , p i ] + , and ∑ r = 1 e i k [ − b ^ κ ( k , r ) , κ ( p , t ) i ] + = [ − b k , p i ] + . \sum_{r=1}^{e_{i_k}}\left[\widehat{b}^i_{\kappa(k,r),\kappa(p,t)}\right]_+=[b^i_{k,p}]_+,\qquad\text{and}\qquad\sum_{r=1}^{e_{i_k}}\left[-\widehat{b}^i_{\kappa(k,r),\kappa(p,t)}\right]_+=[-b^i_{k,p}]_+. r = 1 ∑ e i k [ b κ ( k , r ) , κ ( p , t ) i ] + = [ b k , p i ] + , and r = 1 ∑ e i k [ − b κ ( k , r ) , κ ( p , t ) i ] + = [ − b k , p i ] + . Proof. The weak-sign assertion follows directly from the case-by-case description in the proof of Proposition 3.8. Together with
b k , p i = ∑ r = 1 e i k b ^ κ ( k , r ) , κ ( p , t ) i , b^i_{k,p}=\sum_{r=1}^{e_{i_k}}\widehat{b}^i_{\kappa(k,r),\kappa(p,t)}, b k , p i = r = 1 ∑ e i k b κ ( k , r ) , κ ( p , t ) i , it gives the two identities involving positive parts.
For the exchange matrices, the weak-sign property in Lemma 3.10 corresponds to the sign condition in Γ \Gamma Γ -admissibility; see [6 Definition 4.4.1 and Lemma 4.4.4] .
We next describe the lifting of maximal commuting families. Assume first that l ≥ 2 l \ge2 l ≥ 2 . Let C = ( c 1 , … , c l ) \mathcal{C}=(c_1,\ldots,c_l) C = ( c 1 , … , c l ) be an admissible chain of extent [ 1 , l ] [1,l] [ 1 , l ] , with F = F ( C ) \mathcal{F}=\mathcal{F}(\mathcal{C}) F = F ( C ) and associated { L , R } \{L,R\} { L , R } -sequence T = ( T 1 , … , T l − 1 ) \mathcal{T}=(T_1,\ldots,T_{l-1}) T = ( T 1 , … , T l − 1 ) . For 1 ≤ k ≤ l 1\le k\le l 1 ≤ k ≤ l , let p k p_k p k be the unique position such that { p k } = c ~ k ∖ c ~ k − 1 \{p_k\}=\widetilde{c}_k\setminus\widetilde{c}_{k-1} { p k } = c k ∖ c k − 1 (c ~ 0 = ∅ \widetilde{c}_0=\varnothing c 0 = ∅ ). Thus p k = eff F ( c k ) p_k=\operatorname{eff}_{\mathcal{F}}(c_k) p k = eff F ( c k ) . We call
p ( C ) : = ( p 1 , … , p l ) the effective-end sequence of C . \mathbf{p}(\mathcal{C}):=(p_1,\ldots,p_l)\quad\text{the effective-end sequence of }\mathcal{C}. p ( C ) := ( p 1 , … , p l ) the effective-end sequence of C .
For T ∈ { L , R } T\in\{L,R\} T ∈ { L , R } , let T ‾ \overline{T} T denote the opposite symbol; i.e., L ‾ = R \overline{L}=R L = R and R ‾ = L \overline{R}=L R = L . We define the canonical lift T ^ \widehat{\mathcal{T}} T of T \mathcal{T} T by
T ^ : = ( T ‾ 1 e i p 1 − 1 , T 1 e i p 2 , T 2 e i p 3 , … , T l − 1 e i p l ) . (39) \widehat{\mathcal{T}}:=\left(\overline{T}_1^{e_{i_{p_1}}-1},T_1^{e_{i_{p_2}}},T_2^{e_{i_{p_3}}},\ldots,T_{l-1}^{e_{i_{p_l}}}\right). \tag*{(39)} T := ( T 1 e i p 1 − 1 , T 1 e i p 2 , T 2 e i p 3 , … , T l − 1 e i p l ) . (39)
Thus the initial effective end p 1 p_1 p 1 is replaced by an internal block of length e i p 1 − 1 e_{i_{p_1}}-1 e i p 1 − 1 with the opposite orientation, while, for 2 ≤ k ≤ l 2\le k\le l 2 ≤ k ≤ l , the step adjoining p k p_k p k is replaced by a constant block of length e i p k e_{i_{p_k}} e i p k with the same orientation.
Since p ( C ) \mathbf{p}(\mathcal{C}) p ( C ) is a permutation of [ 1 , l ] [1,l] [ 1 , l ] , the sequence T ^ \widehat{\mathcal{T}} T has length
∑ k = 1 l e i p k − 1 = N l − 1. \sum_{k=1}^{l}e_{i_{p_k}}-1=N_l-1. k = 1 ∑ l e i p k − 1 = N l − 1.
Let C ^ : = C ( i , T ^ ) \widehat{\mathcal{C}}:=\mathcal{C}(\mathbf{i},\widehat{\mathcal{T}}) C := C ( i , T ) be the corresponding admissible chain of i \mathbf{i} i -boxes, and set F ^ : = F ( C ^ ) \widehat{\mathcal{F}}:=\mathcal{F}(\widehat{\mathcal{C}}) F := F ( C ) . For l = 1 l=1 l = 1 , set
T ^ : = ( R e i 1 − 1 ) . \widehat{\mathcal{T}}:=(R^{e_{i_1}-1}). T := ( R e i 1 − 1 ) .
The canonical lift of C = ( [ 1 ] ) \mathcal{C}=([1]) C = ([ 1 ]) is then
C ^ = ( [ κ ( 1 , 1 ) ] , … , [ κ ( 1 , e i 1 ) ] ) , F ^ : = F ( C ^ ) . \widehat{\mathcal{C}}=([\kappa(1,1)],\ldots,[\kappa(1,e_{i_1})]),\qquad\widehat{\mathcal{F}}:=\mathcal{F}(\widehat{\mathcal{C}}). C = ([ κ ( 1 , 1 )] , … , [ κ ( 1 , e i 1 )]) , F := F ( C ) .
For the empty sequence, the chain, the family, and their lifts are empty.
Example 3.11. Consider the folding A 3 → C 2 A_3\to C_2 A 3 → C 2 with I 1 = ( 1 , 3 ) I_1=(1,3) I 1 = ( 1 , 3 ) , I 2 = ( 2 ) I_2=(2) I 2 = ( 2 ) , and let
i = ( 1 , 2 , 1 , 2 ) , i ^ = ( 1 , 3 , 2 , 1 , 3 , 2 ) . \mathbf{i}=(1,2,1,2),\qquad\widehat{\mathbf{i}}=(1,3,2,1,3,2). i = ( 1 , 2 , 1 , 2 ) , i = ( 1 , 3 , 2 , 1 , 3 , 2 ) . First take T = ( R , R , R ) \mathcal{T}=(R,R,R) T = ( R , R , R ) . Then
p ( C ) = ( 1 , 2 , 3 , 4 ) . \mathbf{p}(\mathcal{C})=(1,2,3,4). p ( C ) = ( 1 , 2 , 3 , 4 ) . Since p 1 = 1 p_1=1 p 1 = 1 and e i p 1 = e 1 = 2 e_{i_{p_1}}=e_1=2 e i p 1 = e 1 = 2 , the block replacing p 1 p_1 p 1 contributes the internal symbol R ‾ = L \overline{R}=L R = L . The remaining steps are lifted by (39) , and hence
T ^ = ( L , R , R , R , R ) . \widehat{\mathcal{T}}=(L,R,R,R,R). T = ( L , R , R , R , R ) . Notice that ( R , R , R , R , R ) (R,R,R,R,R) ( R , R , R , R , R ) determines the same lifted family, with the first two singleton boxes in the opposite order, but the rule in (39) selects T ^ \widehat{\mathcal{T}} T above.
Now take T = ( L , R , R ) \mathcal{T}=(L,R,R) T = ( L , R , R ) . Then
p ( C ) = ( 2 , 1 , 3 , 4 ) . \mathbf{p}(\mathcal{C})=(2,1,3,4). p ( C ) = ( 2 , 1 , 3 , 4 ) . Here p 1 = 2 p_1=2 p 1 = 2 and e i p 1 = e 2 = 1 e_{i_{p_1}}=e_2=1 e i p 1 = e 2 = 1 , so the first block in (39) is empty. Since
( p 2 , p 3 , p 4 ) = ( 1 , 3 , 4 ) , (p_2,p_3,p_4)=(1,3,4), ( p 2 , p 3 , p 4 ) = ( 1 , 3 , 4 ) ,
(39) gives
I ^ = ( L , L , R , R , R ) . \widehat{\mathcal{I}}=(\mathcal{L},\mathcal{L},\mathcal{R},\mathcal{R},\mathcal{R}). I = ( L , L , R , R , R ) .
By construction, each effective end p p p of C \mathcal{C} C is replaced by the positions
{ κ ( p , r ) ∣ 1 ≤ r ≤ e i p } \{\kappa(p,r)\mid1\leq r\leq e_{i_p}\} { κ ( p , r ) ∣ 1 ≤ r ≤ e i p }
in the effective-end sequence of C ^ \widehat{\mathcal{C}} C . We therefore define the effective-end folding map
π e f f : { effective ends of F ^ } ⟶ [ 1 , l ] by π e f f ( κ ( p , r ) ) : = p . \pi_{\mathrm{eff}}:\{\text{effective ends of }\widehat{\mathcal{F}}\}\longrightarrow[1,l]\quad\text{by}\quad\pi_{\mathrm{eff}}(\kappa(p,r)):=p. π eff : { effective ends of F } ⟶ [ 1 , l ] by π eff ( κ ( p , r )) := p .
Thus
∣ π e f f − 1 ( p ) ∣ = e i p ( 1 ≤ p ≤ l ) . (40) \left|\pi_{\mathrm{eff}}^{-1}(p)\right|=e_{i_p}\qquad(1\leq p\leq l). \tag*{(40)} π eff − 1 ( p ) = e i p ( 1 ≤ p ≤ l ) . (40)
Since the effective-end maps
eff F : F ⟶ [ 1 , l ] and eff F ^ : F ^ ⟶ [ 1 , N l ] \operatorname{eff}_{\mathcal{F}}:\mathcal{F}\longrightarrow[1,l]\qquad\text{and}\qquad\operatorname{eff}_{\widehat{\mathcal{F}}}:\widehat{\mathcal{F}}\longrightarrow[1,N_l] eff F : F ⟶ [ 1 , l ] and eff F : F ⟶ [ 1 , N l ]
are bijections, we define
π b o x C : F ^ ⟶ F by π b o x C : = eff F − 1 ∘ π e f f ∘ eff F ^ . (41) \pi_{\mathrm{box}}^{\mathcal{C}}:\widehat{\mathcal{F}}\longrightarrow\mathcal{F}\quad\text{by}\quad\pi_{\mathrm{box}}^{\mathcal{C}}:=\operatorname{eff}_{\mathcal{F}}^{-1}\circ\pi_{\mathrm{eff}}\circ\operatorname{eff}_{\widehat{\mathcal{F}}}. \tag*{(41)} π box C : F ⟶ F by π box C := eff F − 1 ∘ π eff ∘ eff F . (41)
Example 3.12. Consider the folding A 5 ⟶ C 3 A_5\longrightarrow C_3 A 5 ⟶ C 3 with
I 1 = ( 1 , 5 ) , I 2 = ( 2 , 4 ) , I 3 = ( 3 ) . I_1=(1,5),\qquad I_2=(2,4),\qquad I_3=(3). I 1 = ( 1 , 5 ) , I 2 = ( 2 , 4 ) , I 3 = ( 3 ) .
Let i = ( 1 , 2 , 3 , 2 , 1 , 3 , 2 , 3 ) \boldsymbol{i}=(1,2,3,2,1,3,2,3) i = ( 1 , 2 , 3 , 2 , 1 , 3 , 2 , 3 ) , so that
i ^ = ( 1 , 5 , 2 , 4 , 3 , 2 , 4 , 1 , 5 , 3 , 2 , 4 , 3 ) . \widehat{\boldsymbol{i}}=(1,5,2,4,3,2,4,1,5,3,2,4,3). i = ( 1 , 5 , 2 , 4 , 3 , 2 , 4 , 1 , 5 , 3 , 2 , 4 , 3 ) .
Consider the maximal commuting families
F = { [ 1 , 5 ] , [ 2 , 4 ] , [ 3 ] , [ 4 ] , [ 5 ] , [ 3 , 6 ] , [ 2 , 7 ] , [ 3 , 8 ] } \mathcal{F}=\{[1,5],[2,4],[3],[4],[5],[3,6],[2,7],[3,8]\} F = {[ 1 , 5 ] , [ 2 , 4 ] , [ 3 ] , [ 4 ] , [ 5 ] , [ 3 , 6 ] , [ 2 , 7 ] , [ 3 , 8 ]}
and
F ^ = { [ 1 , 8 ] , [ 2 , 9 ] , [ 3 , 6 ] , [ 4 , 7 ] , [ 5 ] , [ 6 ] , [ 7 ] , [ 8 ] , [ 9 ] , [ 5 , 10 ] , [ 3 , 11 ] , [ 4 , 12 ] , [ 5 , 13 ] } , \widehat{\mathcal{F}}=\{[1,8],[2,9],[3,6],[4,7],[5],[6],[7],[8],[9],[5,10],[3,11],[4,12],[5,13]\}, F = {[ 1 , 8 ] , [ 2 , 9 ] , [ 3 , 6 ] , [ 4 , 7 ] , [ 5 ] , [ 6 ] , [ 7 ] , [ 8 ] , [ 9 ] , [ 5 , 10 ] , [ 3 , 11 ] , [ 4 , 12 ] , [ 5 , 13 ]} ,
ordered by their effective ends. Thus
and
The effective-end folding is
{ 1 , 2 } { 3 , 4 } { 5 } { 6 , 7 } { 8 , 9 } { 10 } { 11 , 12 } { 13 } ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ 1 2 3 4 5 6 7 8. \begin{array}{cccccccc}
\{1,2\} & \{3,4\} & \{5\} & \{6,7\} & \{8,9\} & \{10\} & \{11,12\} & \{13\}\\
\downarrow& \downarrow& \downarrow& \downarrow& \downarrow& \downarrow& \downarrow& \downarrow\\
1 & 2 & 3 & 4 & 5 & 6 & 7 & 8.
\end{array} { 1 , 2 } ↓ 1 { 3 , 4 } ↓ 2 { 5 } ↓ 3 { 6 , 7 } ↓ 4 { 8 , 9 } ↓ 5 { 10 } ↓ 6 { 11 , 12 } ↓ 7 { 13 } ↓ 8.
Hence the box folding map
π b o x C = eff F − 1 ∘ π e f f ∘ eff F ^ \pi_{\mathrm{box}}^{\mathcal{C}}=\operatorname{eff}_{\mathcal{F}}^{-1}\circ\pi_{\mathrm{eff}}\circ\operatorname{eff}_{\widehat{\mathcal{F}}} π box C = eff F − 1 ∘ π eff ∘ eff F
is given by
{ [ 1 , 8 ] , [ 2 , 9 ] } ⟼ [ 1 , 5 ] , { [ 3 , 6 ] , [ 4 , 7 ] } ⟼ [ 2 , 4 ] , \{\,[1,8],[2,9]\,\}\longmapsto[1,5],\qquad\{\,[3,6],[4,7]\,\}\longmapsto[2,4], { [ 1 , 8 ] , [ 2 , 9 ] } ⟼ [ 1 , 5 ] , { [ 3 , 6 ] , [ 4 , 7 ] } ⟼ [ 2 , 4 ] ,
{ [ 5 ] } ⟼ [ 3 ] , { [ 6 ] , [ 7 ] } ⟼ [ 4 ] , \{\,[5]\,\}\longmapsto[3],\qquad\{\,[6],[7]\,\}\longmapsto[4], { [ 5 ] } ⟼ [ 3 ] , { [ 6 ] , [ 7 ] } ⟼ [ 4 ] ,
{ [ 8 ] , [ 9 ] } ⟼ [ 5 ] , { [ 5 , 10 ] } ⟼ [ 3 , 6 ] , \{\,[8],[9]\,\}\longmapsto[5],\qquad\{\,[5,10]\,\}\longmapsto[3,6], { [ 8 ] , [ 9 ] } ⟼ [ 5 ] , { [ 5 , 10 ] } ⟼ [ 3 , 6 ] ,
{ [ 3 , 11 ] , [ 4 , 12 ] } ⟼ [ 2 , 7 ] , { [ 5 , 13 ] } ⟼ [ 3 , 8 ] . \{\,[3,11],[4,12]\,\}\longmapsto[2,7],\qquad\{\,[5,13]\,\}\longmapsto[3,8]. { [ 3 , 11 ] , [ 4 , 12 ] } ⟼ [ 2 , 7 ] , { [ 5 , 13 ] } ⟼ [ 3 , 8 ] .
For c ∈ F \mathfrak c\in\mathcal F c ∈ F , we define the box fiber over c \mathfrak c c by
Fib c ( c ) : = ( π b o x F ^ ) − 1 ( c ) ⊂ F ^ . \operatorname{Fib}^{\mathfrak c}(\mathfrak c):=(\pi_{\mathrm{box}}^{\widehat{\mathcal F}})^{-1}(\mathfrak c)\subset\widehat{\mathcal F}. Fib c ( c ) := ( π box F ) − 1 ( c ) ⊂ F .
We say that the i \mathbf i i -boxes in Fib c ( c ) \operatorname{Fib}^{\mathfrak c}(\mathfrak c) Fib c ( c ) fold to c \mathfrak c c . By (40) ,
∣ Fib c ( c ) ∣ = e i c . \left|\operatorname{Fib}^{\mathfrak c}(\mathfrak c)\right|=e_{\mathfrak i_{\mathfrak c}}. ∣ Fib c ( c ) ∣ = e i c .
Lemma 3.13. Let c = [ a , b ] ∈ F \mathfrak c=[a,b]\in\mathcal F c = [ a , b ] ∈ F , and write Fib c ( c ) = ( c ^ 1 , … , c ^ e i c ) \operatorname{Fib}^{\mathfrak c}(\mathfrak c)=(\widehat{\mathfrak c}_1,\ldots,\widehat{\mathfrak c}_{e_{\mathfrak i_{\mathfrak c}}}) Fib c ( c ) = ( c 1 , … , c e i c ) in increasing order of their effective ends. Then
E c ^ r = ∑ a ≤ p ≤ b i p = i c ε ^ p , r ( 1 ≤ r ≤ e i c ) . E_{\widehat{\mathfrak c}_r}=\sum_{\substack{a\le p\le b\\ i_p=i_{\mathfrak c}}}\widehat{\varepsilon}_{p,r}\qquad(1\le r\le e_{\mathfrak i_{\mathfrak c}}). E c r = a ≤ p ≤ b i p = i c ∑ ε p , r ( 1 ≤ r ≤ e i c ) . In particular,
c ^ r = [ κ ( a , r ) , κ ( b , r ) ] ( 1 ≤ r ≤ e i c ) . \widehat{\mathfrak c}_r=[\kappa(a,r),\kappa(b,r)]\qquad(1\le r\le e_{\mathfrak i_{\mathfrak c}}). c r = [ κ ( a , r ) , κ ( b , r )] ( 1 ≤ r ≤ e i c ) .
Proof. The lemma follows directly from the fiberwise construction of F ^ \widehat{\mathcal F} F . By (28) , the endpoints of a box are the positions of the first and last nonzero coordinates of its box vector. □
Lemma 3.14. The Γ \Gamma Γ -action on the effective ends of F ^ \widehat{\mathcal F} F induces an action on F ^ \widehat{\mathcal F} F by
g ⋅ c ^ : = eff F ^ − 1 ( g ⋅ eff F ^ ( c ^ ) ) ( g ∈ Γ ) . g\cdot\widehat{\mathfrak c}:=\operatorname{eff}_{\widehat{\mathcal F}}^{-1}\bigl(g\cdot\operatorname{eff}_{\widehat{\mathcal F}}(\widehat{\mathfrak c})\bigr)\qquad(g\in\Gamma). g ⋅ c := eff F − 1 ( g ⋅ eff F ( c ) ) ( g ∈ Γ ) . For every c ∈ F \mathfrak c\in\mathcal F c ∈ F and c ^ ∈ Fib c ( c ) \widehat{\mathfrak c}\in\operatorname{Fib}^{\mathfrak c}(\mathfrak c) c ∈ Fib c ( c ) , we have
Γ ⋅ c ^ = Fib c ( c ) . \Gamma\cdot\widehat{\mathfrak c}=\operatorname{Fib}^{\mathfrak c}(\mathfrak c). Γ ⋅ c = Fib c ( c ) . Moreover, for c ^ , d ^ ∈ F ^ \widehat{\mathfrak c},\widehat{\mathfrak d}\in\widehat{\mathcal F} c , d ∈ F and g ∈ Γ g\in\Gamma g ∈ Γ ,
b g ⋅ c ^ , g ⋅ d ^ F ^ = b c ^ , d ^ F ^ and Λ g ⋅ c ^ , g ⋅ d ^ F ^ = Λ c ^ , d ^ F ^ . b^{\widehat{\mathcal F}}_{g\cdot\widehat{\mathfrak c},g\cdot\widehat{\mathfrak d}}=b^{\widehat{\mathcal F}}_{\widehat{\mathfrak c},\widehat{\mathfrak d}}
\qquad\text{and}\qquad
\Lambda^{\widehat{\mathcal F}}_{g\cdot\widehat{\mathfrak c},g\cdot\widehat{\mathfrak d}}=\Lambda^{\widehat{\mathcal F}}_{\widehat{\mathfrak c},\widehat{\mathfrak d}}. b g ⋅ c , g ⋅ d F = b c , d F and Λ g ⋅ c , g ⋅ d F = Λ c , d F .
Proof. By construction,
eff F ^ ( Fib c ( c ) ) = π e f f − 1 ( eff F ( c ) ) , \operatorname{eff}_{\widehat{\mathcal F}}\bigl(\operatorname{Fib}^{\mathfrak c}(\mathfrak c)\bigr)=\pi_{\mathrm{eff}}^{-1}\bigl(\operatorname{eff}_{\mathcal F}(\mathfrak c)\bigr), eff F ( Fib c ( c ) ) = π eff − 1 ( eff F ( c ) ) ,
and the latter is a single Γ \Gamma Γ -orbit. Hence
Γ ⋅ c ^ = Fib c ( c ) . \Gamma\cdot\widehat{\mathfrak c}=\operatorname{Fib}^{\mathfrak c}(\mathfrak c). Γ ⋅ c = Fib c ( c ) .
The remaining assertions follow from the Γ \Gamma Γ -invariance of the upstairs Cartan data and of λ ^ i \widehat{\lambda}^{\mathfrak i} λ i . □
Theorem 3.15. Let c , d ∈ F \mathfrak c,\mathfrak d\in\mathcal F c , d ∈ F , and write Fib c ( c ) = ( c ^ 1 , … , c ^ e i c ) \operatorname{Fib}^{\mathfrak c}(\mathfrak c)=(\widehat{\mathfrak c}_1,\ldots,\widehat{\mathfrak c}_{e_{\mathfrak i_{\mathfrak c}}}) Fib c ( c ) = ( c 1 , … , c e i c ) . For any d ^ t ∈ Fib d ( d ) \widehat{\mathfrak d}_t\in\operatorname{Fib}^{\mathfrak d}(\mathfrak d) d t ∈ Fib d ( d ) , we have
b c , d F = ∑ c ^ ∈ Fib c ( c ) b c ^ , d ^ t F ^ , and Λ c , d F = d i c ∑ c ^ ∈ Fib c ( c ) Λ c ^ , d ^ t F ^ . b^{\mathcal F}_{\mathfrak c,\mathfrak d}=\sum_{\widehat{\mathfrak c}\in\operatorname{Fib}^{\mathfrak c}(\mathfrak c)}b^{\widehat{\mathcal F}}_{\widehat{\mathfrak c},\widehat{\mathfrak d}_t},
\qquad\text{and}\qquad
\Lambda^{\mathcal F}_{\mathfrak c,\mathfrak d}=d_{\mathfrak i_{\mathfrak c}}\sum_{\widehat{\mathfrak c}\in\operatorname{Fib}^{\mathfrak c}(\mathfrak c)}\Lambda^{\widehat{\mathcal F}}_{\widehat{\mathfrak c},\widehat{\mathfrak d}_t}. b c , d F = c ∈ Fib c ( c ) ∑ b c , d t F , and Λ c , d F = d i c c ∈ Fib c ( c ) ∑ Λ c , d t F . In particular, both right-hand sides are independent of the choice of d ^ t \widehat{\mathfrak d}_t d t .
Proof. We first prove the assertion for B B B . The construction of E ^ \widehat{\mathcal{E}} E preserves the effective-end and endpoint conditions in (30) .
Suppose first that b c ^ , d F < 0 b^{\mathcal{F}}_{\hat{c},d}<0 b c ^ , d F < 0 . If this entry comes from a horizontal relation, then, for the fixed d ^ t ∈ Fib E ( d ) \hat{d}_t\in\operatorname{Fib}^{\mathcal{E}}(d) d ^ t ∈ Fib E ( d ) , exactly one c ^ ∈ Fib E ( c ) \hat{c}\in\operatorname{Fib}^{\mathcal{E}}(c) c ^ ∈ Fib E ( c ) satisfies the corresponding upstairs horizontal relation. Its contribution is − 1 -1 − 1 , while all other members of the fiber contribute 0 0 0 .
If b c ^ , d F b^{\mathcal{F}}_{\hat{c},d} b c ^ , d F comes from a vertical relation, then the same effective-end and endpoint conditions hold fiberwise upstairs. Hence
∑ c ^ ∈ Fib E ( c ) b c ^ , d ^ t F ^ = ∑ c ^ ∈ Fib E ( c ) c ^ c ^ , d ^ t = c i c , i d = b c ^ , d F . \sum_{\hat{c}\in\operatorname{Fib}^{\mathcal{E}}(c)} b^{\widehat{\mathcal{F}}}_{\hat{c},\hat{d}_t}
=
\sum_{\hat{c}\in\operatorname{Fib}^{\mathcal{E}}(c)} \widehat{c}_{\hat{c},\hat{d}_t}
=
c_{i_c,i_d}
=
b^{\mathcal{F}}_{\hat{c},d}. c ^ ∈ Fib E ( c ) ∑ b c ^ , d ^ t F = c ^ ∈ Fib E ( c ) ∑ c c ^ , d ^ t = c i c , i d = b c ^ , d F .
where the second equality is the definition of the quotient Cartan matrix. If b c ^ , d F = 0 b^{\mathcal{F}}_{\hat{c},d}=0 b c ^ , d F = 0 , none of the corresponding upstairs horizontal or vertical relations occurs, and every summand is zero.
The case b c ^ , d F > 0 b^{\mathcal{F}}_{\hat{c},d}>0 b c ^ , d F > 0 follows from the negative case by skew-symmetrizability, together with d i e i = m d_i e_i=m d i e i = m and the Γ \Gamma Γ -invariance of the upstairs exchange matrix. Therefore,
b c ^ , d F = ∑ c ^ ∈ Fib E ( c ) b c ^ , d ^ t F ^ . b^{\mathcal{F}}_{\hat{c},d}
=
\sum_{\hat{c}\in\operatorname{Fib}^{\mathcal{E}}(c)}
b^{\widehat{\mathcal{F}}}_{\hat{c},\hat{d}_t}. b c ^ , d F = c ^ ∈ Fib E ( c ) ∑ b c ^ , d ^ t F .
Write Fib E ( c ) = ( c ^ 1 , … , c ^ e i c ) \operatorname{Fib}^{\mathcal{E}}(c)=(\hat{c}_1,\ldots,\hat{c}_{e_{i_c}}) Fib E ( c ) = ( c ^ 1 , … , c ^ e i c ) in the fixed fiber order. By Lemma 2.7 and Lemma 3.13 ,
Λ c ^ , d F = λ i ( E c , E d ) = d i c ∑ r = 1 e i c λ ^ i ( E c ^ r , E d ^ t ) = d i c ∑ r = 1 e i c Λ ^ c ^ r , d ^ t F . \Lambda^{\mathcal{F}}_{\hat{c},d}
=
\lambda^i(\mathcal{E}_{c},\mathcal{E}_{d})
=
d_{i_c}\sum_{r=1}^{e_{i_c}}\hat{\lambda}^i(\mathcal{E}_{\hat{c}_r},\mathcal{E}_{\hat{d}_t})
=
d_{i_c}\sum_{r=1}^{e_{i_c}}\widehat{\Lambda}^{\mathcal{F}}_{\hat{c}_r,\hat{d}_t}. Λ c ^ , d F = λ i ( E c , E d ) = d i c r = 1 ∑ e i c λ ^ i ( E c ^ r , E d ^ t ) = d i c r = 1 ∑ e i c Λ c ^ r , d ^ t F .
The independence of d ^ t \hat{d}_t d ^ t follows from the Γ \Gamma Γ -invariance above.
Folding of g g g -vectors
We introduce the folding map on exponent lattices that will be used later for the comparison of g g g -vectors.
Define
Fold i : Z J i ⟶ Z J i by ( Fold i ( g ^ ) ) k : = ∑ r = 1 e i k g ^ ( k , r ) ( k ∈ J i ) . (42) \operatorname{Fold}^i:\mathbb{Z}^{J^i}\longrightarrow\mathbb{Z}^{J^i}
\quad\text{by}\quad
\left(\operatorname{Fold}^i(\hat{g})\right)_k:=\sum_{r=1}^{e_{i_k}}\hat{g}_{(k,r)}
\qquad(k\in J^i). \tag*{(42)} Fold i : Z J i ⟶ Z J i by ( Fold i ( g ^ ) ) k := r = 1 ∑ e i k g ^ ( k , r ) ( k ∈ J i ) . (42)
We denote by
Fold e x i : Z J e x i ⟶ Z J e x i \operatorname{Fold}^i_{\mathrm{ex}}:\mathbb{Z}^{J^i_{\mathrm{ex}}}\longrightarrow\mathbb{Z}^{J^i_{\mathrm{ex}}} Fold ex i : Z J ex i ⟶ Z J ex i
its restriction to the exchangeable coordinates.
Proposition 3.16. We have
Fold i ∘ B ^ i = B i ∘ Fold e x i . \operatorname{Fold}^i\circ\widehat{B}^i=B^i\circ\operatorname{Fold}^i_{\mathrm{ex}}. Fold i ∘ B i = B i ∘ Fold ex i . Proof. For the standard basis vector ε ( p , t ) \varepsilon_{(p,t)} ε ( p , t ) , Proposition 3.8 gives
( Fold i ( B ^ i ε ( p , t ) ) ) k = ∑ r = 1 e i k b ^ ( k , r ) , ( p , t ) i = b k , p i . \left(\operatorname{Fold}^i(\widehat{B}^i\varepsilon_{(p,t)})\right)_k
=
\sum_{r=1}^{e_{i_k}}\widehat{b}^i_{(k,r),(p,t)}
=
b^i_{k,p}. ( Fold i ( B i ε ( p , t ) ) ) k = r = 1 ∑ e i k b ( k , r ) , ( p , t ) i = b k , p i . On the other hand,
B i Fold e x i ( ε ( p , t ) ) = B i ε p . B^i\operatorname{Fold}^i_{\mathrm{ex}}(\varepsilon_{(p,t)})
=
B^i\varepsilon_p. B i Fold ex i ( ε ( p , t ) ) = B i ε p . Hence the assertion follows.
Braid moves and folding
In this section, we lift finite braid moves to sequences of 2- and 3-moves and realize these sequences by mutations and permutations. We prove the folding identities for the exchange and commutation matrices, cluster variables, and g g g -vectors before and after each finite braid move and its lifted sequence.
A motivating example: the 4-move A 3 → C 2 A_3 \to C_2 A 3 → C 2
We begin with the 4-move
i = 1212 ⟶ i ′ = 2121 \mathbf{i}=1212 \longrightarrow\mathbf{i}'=2121 i = 1212 ⟶ i ′ = 2121
of type C 2 C_2 C 2 .
With our fixed ordering of the fibers,
i ^ = 132132 , i ′ ^ = 213213. \widehat{\mathbf{i}}=132132,\qquad\widehat{\mathbf{i}'}=213213. i = 132132 , i ′ = 213213.
The positions in i ^ \widehat{\mathbf{i}} i corresponding to the successive positions of i \mathbf{i} i are
( 1 , 2 ) , ( 3 ) , ( 4 , 5 ) , ( 6 ) , (1,2),\ (3),\ (4,5),\ (6), ( 1 , 2 ) , ( 3 ) , ( 4 , 5 ) , ( 6 ) ,
while the positions in i ′ ^ \widehat{\mathbf{i}'} i ′ corresponding to those in i ′ \mathbf{i}' i ′ are
( 1 ) , ( 2 , 3 ) , ( 4 ) , ( 5 , 6 ) . (1),\ (2,3),\ (4),\ (5,6). ( 1 ) , ( 2 , 3 ) , ( 4 ) , ( 5 , 6 ) .
Let x = ( x 1 , x 2 , x 3 , x 4 ) \mathbf{x}=(x_1,x_2,x_3,x_4) x = ( x 1 , x 2 , x 3 , x 4 ) and y = ( y 1 , … , y 6 ) \mathbf{y}=(y_1,\ldots,y_6) y = ( y 1 , … , y 6 ) be the initial downstairs and upstairs clusters, respectively. Define
ϕ : Z [ y 1 ± 1 , … , y 6 ± 1 ] ⟶ Z [ x 1 ± 1 , … , x 4 ± 1 ] \phi:\mathbb{Z}[y_1^{\pm1},\ldots,y_6^{\pm1}]\longrightarrow\mathbb{Z}[x_1^{\pm1},\ldots,x_4^{\pm1}] ϕ : Z [ y 1 ± 1 , … , y 6 ± 1 ] ⟶ Z [ x 1 ± 1 , … , x 4 ± 1 ]
by
ϕ ( y 1 ) = ϕ ( y 2 ) = x 1 , ϕ ( y 3 ) = x 2 , ϕ ( y 4 ) = ϕ ( y 5 ) = x 3 , ϕ ( y 6 ) = x 4 . \phi(y_1)=\phi(y_2)=x_1,\qquad\phi(y_3)=x_2,\qquad\phi(y_4)=\phi(y_5)=x_3,\qquad\phi(y_6)=x_4. ϕ ( y 1 ) = ϕ ( y 2 ) = x 1 , ϕ ( y 3 ) = x 2 , ϕ ( y 4 ) = ϕ ( y 5 ) = x 3 , ϕ ( y 6 ) = x 4 .
The downstairs calculation. The initial compatible pair is
B ( 0 ) = ( 0 − 2 1 0 − 1 2 0 − 1 ) , Λ ( 0 ) = ( 0 − 2 0 0 2 0 0 0 0 0 0 0 0 0 0 0 ) . B^{(0)}=
\begin{pmatrix}
0 & -2\\
1 & 0\\
-1 & 2\\
0 & -1
\end{pmatrix},
\qquad
\Lambda^{(0)}=
\begin{pmatrix}
0 & -2 & 0 & 0\\
2 & 0 & 0 & 0\\
0 & 0 & 0 & 0\\
0 & 0 & 0 & 0
\end{pmatrix}. B ( 0 ) = 0 1 − 1 0 − 2 0 2 − 1 , Λ ( 0 ) = 0 2 0 0 − 2 0 0 0 0 0 0 0 0 0 0 0 .
The exchangeable positions are 1 , 2 1,2 1 , 2 . Let P p P_p P p denote the permutation matrix exchanging positions p p p and p + 1 p+1 p + 1 . The 4-move is realized by
μ 1 , μ 2 , μ 1 , P 3 , P 1 \mu_1,\ \mu_2,\ \mu_1,\ P_3,\ P_1 μ 1 , μ 2 , μ 1 , P 3 , P 1
from left to right (see Subsection 4.3 below).
Starting with x ( 0 ) = ( x 1 , x 2 , x 3 , x 4 ) \mathbf{x}^{(0)}=(x_1,x_2,x_3,x_4) x ( 0 ) = ( x 1 , x 2 , x 3 , x 4 ) , the first mutation gives
x 1 a = x 2 + x 3 , a = x 2 + x 3 x 1 x_1a=x_2+x_3,\qquad a=\frac{x_2+x_3}{x_1} x 1 a = x 2 + x 3 , a = x 1 x 2 + x 3
and
B ( 1 ) = ( 0 2 − 1 0 1 0 0 − 1 ) . B^{(1)}=
\begin{pmatrix}
0 & 2\\
-1 & 0\\
1 & 0\\
0 & -1
\end{pmatrix}. B ( 1 ) = 0 − 1 1 0 2 0 0 − 1 .
The second mutation gives
x 2 b = a 2 + x 4 , b = x 1 2 x 4 + ( x 2 + x 3 ) 2 x 1 2 x 2 x_2b=a^2+x_4,\qquad b=\frac{x_1^2x_4+(x_2+x_3)^2}{x_1^2x_2} x 2 b = a 2 + x 4 , b = x 1 2 x 2 x 1 2 x 4 + ( x 2 + x 3 ) 2
with
B ( 2 ) = ( 0 − 2 1 0 1 0 − 1 1 ) . B^{(2)}=
\begin{pmatrix}
0 & -2\\
1 & 0\\
1 & 0\\
-1 & 1
\end{pmatrix}. B ( 2 ) = 0 1 1 − 1 − 2 0 0 1 .
Finally,
a c = b x 3 + x 4 , c = x 1 2 x 4 + x 2 x 3 + x 3 2 x 1 x 2 and B ( 3 ) = ( 0 2 − 1 0 − 1 0 1 − 1 ) . ac=bx_{3}+x_{4},\qquad c=\frac{x_{1}^{2}x_{4}+x_{2}x_{3}+x_{3}^{2}}{x_{1}x_{2}}\qquad\text{and}\qquad B^{(3)}=\begin{pmatrix}0&2\\-1&0\\-1&0\\1&-1\end{pmatrix}. a c = b x 3 + x 4 , c = x 1 x 2 x 1 2 x 4 + x 2 x 3 + x 3 2 and B ( 3 ) = 0 − 1 − 1 1 2 0 0 − 1 .
Applying P 3 P_{3} P 3 and then P 1 P_{1} P 1 , we obtain
x ′ = ( b , c , x 4 , x 3 ) , \mathbf{x}'=(b,c,x_{4},x_{3}), x ′ = ( b , c , x 4 , x 3 ) ,
that is,
x 1 ′ = x 1 2 x 4 + ( x 2 + x 3 ) 2 x 1 2 x 2 , x 2 ′ = x 1 2 x 4 + x 2 x 3 + x 3 2 x 1 x 2 , x 3 ′ = x 4 , x 4 ′ = x 3 . x'_{1}=\frac{x_{1}^{2}x_{4}+(x_{2}+x_{3})^{2}}{x_{1}^{2}x_{2}},\qquad x'_{2}=\frac{x_{1}^{2}x_{4}+x_{2}x_{3}+x_{3}^{2}}{x_{1}x_{2}},\qquad x'_{3}=x_{4},\qquad x'_{4}=x_{3}. x 1 ′ = x 1 2 x 2 x 1 2 x 4 + ( x 2 + x 3 ) 2 , x 2 ′ = x 1 x 2 x 1 2 x 4 + x 2 x 3 + x 3 2 , x 3 ′ = x 4 , x 4 ′ = x 3 .
The final compatible pair is
B i ′ = ( 0 − 1 2 0 − 1 1 0 − 1 ) , Λ i ′ = ( 0 − 2 ∣ 0 0 2 0 ∣ 0 0 0 0 ∣ 0 0 0 0 ∣ 0 0 ) . B^{i'}=\begin{pmatrix}0&-1\\2&0\\-1&1\\0&-1\end{pmatrix},\qquad\Lambda^{i'}=\begin{pmatrix}0&-2&|&0&0\\2&0&|&0&0\\\hline0&0&|&0&0\\0&0&|&0&0\end{pmatrix}. B i ′ = 0 2 − 1 0 − 1 0 1 − 1 , Λ i ′ = 0 2 0 0 − 2 0 0 0 ∣ ∣ ∣ ∣ 0 0 0 0 0 0 0 0 .
The upstairs calculation. The initial compatible pair associated with i ^ = 132132 \hat{i}=132132 i ^ = 132132 is
B ^ i ^ , ( 0 ) = ( 0 0 − 1 0 0 − 1 1 1 0 − 1 0 1 0 − 1 1 0 0 − 1 ) , Λ ^ i ^ , ( 0 ) = ( 0 0 ∣ − 1 1 − 1 ∣ 0 0 0 ∣ − 1 − 1 1 ∣ 0 1 1 ∣ 0 0 0 ∣ 0 − 1 1 ∣ 0 0 0 ∣ 0 1 − 1 ∣ 0 0 0 ∣ 0 0 0 ∣ 0 0 0 ∣ 0 ) . \widehat{B}^{\hat{i},(0)}=\begin{pmatrix}0&0&-1\\0&0&-1\\\hline1&1&0\\-1&0&1\\0&-1&1\\\hline0&0&-1\end{pmatrix},\qquad\widehat{\Lambda}^{\hat{i},(0)}=\begin{pmatrix}0&0&|&-1&1&-1&|&0\\0&0&|&-1&-1&1&|&0\\\hline1&1&|&0&0&0&|&0\\-1&1&|&0&0&0&|&0\\1&-1&|&0&0&0&|&0\\\hline0&0&|&0&0&0&|&0\end{pmatrix}. B i ^ , ( 0 ) = 0 0 1 − 1 0 0 0 0 1 0 − 1 0 − 1 − 1 0 1 1 − 1 , Λ i ^ , ( 0 ) = 0 0 1 − 1 1 0 0 0 1 1 − 1 0 ∣ ∣ ∣ ∣ ∣ ∣ − 1 − 1 0 0 0 0 1 − 1 0 0 0 0 − 1 1 0 0 0 0 ∣ ∣ ∣ ∣ ∣ ∣ 0 0 0 0 0 0 .
The exchangeable positions are 1, 2, 3. The following sequence of 2- and 3-moves connects 132132 to 213213:
132132 ‾ → 312132 ‾ → 321232 → 321323 ‾ → 323123 ‾ → 232123 → 231213 ‾ → 213213 , \underline{132132}\to\underline{312132}\to321232\to\underline{321323}\to\underline{323123}\to232123\to\underline{231213}\to213213, 132132 → 312132 → 321232 → 321323 → 323123 → 232123 → 231213 → 213213 ,
with corresponding transformations
P 1 , μ 2 P 3 , μ 4 P 5 , P 3 , μ 1 P 2 , μ 3 P 4 , P 2 P_{1},\quad\mu_{2}P_{3},\quad\mu_{4}P_{5},\quad P_{3},\quad\mu_{1}P_{2},\quad\mu_{3}P_{4},\quad P_{2} P 1 , μ 2 P 3 , μ 4 P 5 , P 3 , μ 1 P 2 , μ 3 P 4 , P 2
from left to right (see Subsection 4.3 below). Starting with y ( 0 ) = ( y 1 , … , y 6 ) \mathbf{y}^{(0)}=(y_{1},\ldots,y_{6}) y ( 0 ) = ( y 1 , … , y 6 ) , the first move gives
y ( 1 ) = ( y 2 , y 1 , y 3 , y 4 , y 5 , y 6 ) . \mathbf{y}^{(1)}=(y_{2},y_{1},y_{3},y_{4},y_{5},y_{6}). y ( 1 ) = ( y 2 , y 1 , y 3 , y 4 , y 5 , y 6 ) .
Next,
y 1 w = y 3 + y 4 , w = y 3 + y 4 y 1 , y_{1}w=y_{3}+y_{4},\qquad w=\frac{y_{3}+y_{4}}{y_{1}}, y 1 w = y 3 + y 4 , w = y 1 y 3 + y 4 ,
and after P 3 P_{3} P 3 ,
y ( 3 ) = ( y 2 , u , y 4 , y 3 , y 5 , y 6 ) , B ^ i , ( 3 ) = ( 0 0 − 1 0 0 1 0 1 0 1 − 1 0 − 1 0 1 0 0 − 1 ) . \mathbf{y}^{(3)}=(y_{2},u,y_{4},y_{3},y_{5},y_{6}),\qquad\widehat{B}^{i,(3)}=
\begin{pmatrix}
0 & 0 & -1\\
0 & 0 & 1\\
0 & 1 & 0\\
1 & -1 & 0\\
-1 & 0 & 1\\
0 & 0 & -1
\end{pmatrix}. y ( 3 ) = ( y 2 , u , y 4 , y 3 , y 5 , y 6 ) , B i , ( 3 ) = 0 0 0 1 − 1 0 0 0 1 − 1 0 0 − 1 1 0 0 1 − 1 .
The next mutation gives
y 3 v = y 5 + y 2 y 6 , v = y 1 y 2 y 6 + y 3 y 5 + y 4 y 5 y 1 y 3 . y_{3}v=y_{5}+y_{2}y_{6},\qquad v=\frac{y_{1}y_{2}y_{6}+y_{3}y_{5}+y_{4}y_{5}}{y_{1}y_{3}}. y 3 v = y 5 + y 2 y 6 , v = y 1 y 3 y 1 y 2 y 6 + y 3 y 5 + y 4 y 5 .
and after P 5 P_{5} P 5 and then P 3 P_{3} P 3 ,
y ( 6 ) = ( y 2 , u , v , y 4 , y 6 , y 5 ) , B ^ i , ( 6 ) = ( 0 − 1 1 1 0 − 1 − 1 1 0 0 1 0 0 − 1 1 0 0 − 1 ) . \mathbf{y}^{(6)}=(y_{2},u,v,y_{4},y_{6},y_{5}),\qquad\widehat{B}^{i,(6)}=
\begin{pmatrix}
0 & -1 & 1\\
1 & 0 & -1\\
-1 & 1 & 0\\
0 & 1 & 0\\
0 & -1 & 1\\
0 & 0 & -1
\end{pmatrix}. y ( 6 ) = ( y 2 , u , v , y 4 , y 6 , y 5 ) , B i , ( 6 ) = 0 1 − 1 0 0 0 − 1 0 1 1 − 1 0 1 − 1 0 0 1 − 1 .
Next,
y 2 w = u + v , w = y 1 y 2 y 6 + y 3 2 + y 3 y 4 + y 3 y 5 + y 4 y 5 y 1 y 2 y 3 , y_{2}w=u+v,\qquad w=\frac{y_{1}y_{2}y_{6}+y_{3}^{2}+y_{3}y_{4}+y_{3}y_{5}+y_{4}y_{5}}{y_{1}y_{2}y_{3}}, y 2 w = u + v , w = y 1 y 2 y 3 y 1 y 2 y 6 + y 3 2 + y 3 y 4 + y 3 y 5 + y 4 y 5 ,
and after P 2 P_{2} P 2 ,
y ( 8 ) = ( w , v , u , y 4 , y 6 , y 5 ) , B ^ i , ( 8 ) = ( 0 − 1 1 1 0 0 − 1 0 0 0 0 1 0 1 − 1 0 − 1 0 ) . \mathbf{y}^{(8)}=(w,v,u,y_{4},y_{6},y_{5}),\qquad\widehat{B}^{i,(8)}=
\begin{pmatrix}
0 & -1 & 1\\
1 & 0 & 0\\
-1 & 0 & 0\\
0 & 0 & 1\\
0 & 1 & -1\\
0 & -1 & 0
\end{pmatrix}. y ( 8 ) = ( w , v , u , y 4 , y 6 , y 5 ) , B i , ( 8 ) = 0 1 − 1 0 0 0 − 1 0 0 0 1 − 1 1 0 0 1 − 1 0 .
Finally,
u z = w y 4 + y 6 , z = y 1 y 2 y 6 + y 3 y 4 + y 4 y 5 y 2 y 3 . uz=wy_{4}+y_{6},\qquad z=\frac{y_{1}y_{2}y_{6}+y_{3}y_{4}+y_{4}y_{5}}{y_{2}y_{3}}. u z = w y 4 + y 6 , z = y 2 y 3 y 1 y 2 y 6 + y 3 y 4 + y 4 y 5 .
After P 4 P_{4} P 4 and P 2 P_{2} P 2 , we obtain
y ′ = ( w , z , y 9 , y 6 , y 4 , y 5 ) , B ^ i = ( 0 − 1 − 1 1 0 0 1 0 0 − 1 1 1 0 − 1 0 0 0 − 1 ) and Λ ^ i = ( 0 − 1 − 1 0 0 0 1 0 0 0 1 − 1 1 0 0 0 − 1 1 0 0 0 0 0 0 0 − 1 1 0 0 0 0 1 − 1 0 0 0 ) . \mathbf{y}'=(w,z,y_{9},y_{6},y_{4},y_{5}),\qquad\widehat{B}^{i}=
\begin{pmatrix}
0 & -1 & -1\\
1 & 0 & 0\\
1 & 0 & 0\\
-1 & 1 & 1\\
0 & -1 & 0\\
0 & 0 & -1
\end{pmatrix}
\quad\text{and}\quad
\widehat{\Lambda}^{i}=
\begin{pmatrix}
0 & -1 & -1 & 0 & 0 & 0\\
1 & 0 & 0 & 0 & 1 & -1\\
1 & 0 & 0 & 0 & -1 & 1\\
0 & 0 & 0 & 0 & 0 & 0\\
0 & -1 & 1 & 0 & 0 & 0\\
0 & 1 & -1 & 0 & 0 & 0
\end{pmatrix}. y ′ = ( w , z , y 9 , y 6 , y 4 , y 5 ) , B i = 0 1 1 − 1 0 0 − 1 0 0 1 − 1 0 − 1 0 0 1 0 − 1 and Λ i = 0 1 1 0 0 0 − 1 0 0 0 − 1 1 − 1 0 0 0 1 − 1 0 0 0 0 0 0 0 1 − 1 0 0 0 0 − 1 1 0 0 0 .
Thus
y 1 ′ = y 1 y 2 y 6 + y 3 2 + y 3 y 4 + y 3 y 5 + y 4 y 5 y 1 y 2 y 3 , y'_1=\frac{y_1y_2y_6+y_3^2+y_3y_4+y_3y_5+y_4y_5}{y_1y_2y_3}, y 1 ′ = y 1 y 2 y 3 y 1 y 2 y 6 + y 3 2 + y 3 y 4 + y 3 y 5 + y 4 y 5 ,
y 2 ′ = y 1 y 2 y 6 + y 3 y 4 + y 4 y 5 y 2 y 3 , y 3 ′ = y 1 y 2 y 6 + y 3 y 5 + y 4 y 5 y 1 y 3 , y'_2=\frac{y_1y_2y_6+y_3y_4+y_4y_5}{y_2y_3},\qquad
y'_3=\frac{y_1y_2y_6+y_3y_5+y_4y_5}{y_1y_3}, y 2 ′ = y 2 y 3 y 1 y 2 y 6 + y 3 y 4 + y 4 y 5 , y 3 ′ = y 1 y 3 y 1 y 2 y 6 + y 3 y 5 + y 4 y 5 ,
y 4 ′ = y 6 , y 5 ′ = y 4 , y 6 ′ = y 5 . y'_4=y_6,\qquad y'_5=y_4,\qquad y'_6=y_5. y 4 ′ = y 6 , y 5 ′ = y 4 , y 6 ′ = y 5 .
Comparison under folding. We now apply the initial folding homomorphism ϕ \phi ϕ to the final upstairs cluster y ′ \mathbf{y}' y ′ . The positions in I ^ \hat{I} I ^ corresponding to the successive positions of i i i are
( 1 ) , ( 2 , 3 ) , ( 4 ) , ( 5 , 6 ) . (1),\qquad(2,3),\qquad(4),\qquad(5,6). ( 1 ) , ( 2 , 3 ) , ( 4 ) , ( 5 , 6 ) .
Proposition 4.1. With the notation above,
ϕ ( y 1 ′ ) = x 1 ′ , ϕ ( y 2 ′ ) = ϕ ( y 3 ′ ) = x 2 ′ , ϕ ( y 4 ′ ) = x 3 ′ , ϕ ( y 5 ′ ) = ϕ ( y 6 ′ ) = x 4 ′ . \phi(y'_1)=x'_1,\qquad\phi(y'_2)=\phi(y'_3)=x'_2,\qquad\phi(y'_4)=x'_3,\qquad\phi(y'_5)=\phi(y'_6)=x'_4. ϕ ( y 1 ′ ) = x 1 ′ , ϕ ( y 2 ′ ) = ϕ ( y 3 ′ ) = x 2 ′ , ϕ ( y 4 ′ ) = x 3 ′ , ϕ ( y 5 ′ ) = ϕ ( y 6 ′ ) = x 4 ′ . Thus ϕ \phi ϕ sends every variable in y ′ \mathbf{y}' y ′ to the corresponding variable in x ′ \mathbf{x}' x ′ .
Proof. Using
ϕ ( y 1 ) = ϕ ( y 2 ) = x 1 , ϕ ( y 3 ) = x 2 , ϕ ( y 4 ) = ϕ ( y 5 ) = x 3 , ϕ ( y 6 ) = x 4 , \phi(y_1)=\phi(y_2)=x_1,\qquad\phi(y_3)=x_2,\qquad\phi(y_4)=\phi(y_5)=x_3,\qquad\phi(y_6)=x_4, ϕ ( y 1 ) = ϕ ( y 2 ) = x 1 , ϕ ( y 3 ) = x 2 , ϕ ( y 4 ) = ϕ ( y 5 ) = x 3 , ϕ ( y 6 ) = x 4 ,
we obtain
ϕ ( y 1 ′ ) = x 1 2 x 4 + x 2 2 + x 2 x 3 + x 2 x 3 + x 3 2 x 1 2 x 2 = x 1 2 x 4 + ( x 2 + x 3 ) 2 x 1 2 x 2 = x 1 ′ . \phi(y'_1)=\frac{x_1^2x_4+x_2^2+x_2x_3+x_2x_3+x_3^2}{x_1^2x_2}
=\frac{x_1^2x_4+(x_2+x_3)^2}{x_1^2x_2}=x'_1. ϕ ( y 1 ′ ) = x 1 2 x 2 x 1 2 x 4 + x 2 2 + x 2 x 3 + x 2 x 3 + x 3 2 = x 1 2 x 2 x 1 2 x 4 + ( x 2 + x 3 ) 2 = x 1 ′ .
Notice that the coefficient 2 in x 1 ′ x'_1 x 1 ′ comes from the two distinct upstairs monomials y 3 y 4 y_3y_4 y 3 y 4 and y 3 y 5 y_3y_5 y 3 y 5 , both of which fold to x 2 x 3 x_2x_3 x 2 x 3 . Similarly,
ϕ ( y 2 ′ ) = x 1 2 x 4 + x 2 x 3 + x 3 2 x 1 x 2 = ϕ ( y 3 ′ ) = x 2 ′ . \phi(y'_2)=\frac{x_1^2x_4+x_2x_3+x_3^2}{x_1x_2}=\phi(y'_3)=x'_2. ϕ ( y 2 ′ ) = x 1 x 2 x 1 2 x 4 + x 2 x 3 + x 3 2 = ϕ ( y 3 ′ ) = x 2 ′ .
For the remaining variables,
ϕ ( y 4 ′ ) = ϕ ( y 6 ) = x 4 = x 3 ′ , ϕ ( y 5 ′ ) = ϕ ( y 4 ) = x 3 = x 4 ′ , ϕ ( y 6 ′ ) = ϕ ( y 5 ) = x 3 = x 4 ′ . \phi(y'_4)=\phi(y_6)=x_4=x'_3,\qquad\phi(y'_5)=\phi(y_4)=x_3=x'_4,\qquad\phi(y'_6)=\phi(y_5)=x_3=x'_4. ϕ ( y 4 ′ ) = ϕ ( y 6 ) = x 4 = x 3 ′ , ϕ ( y 5 ′ ) = ϕ ( y 4 ) = x 3 = x 4 ′ , ϕ ( y 6 ′ ) = ϕ ( y 5 ) = x 3 = x 4 ′ .
Hence the assertion follows. □
Finite braid moves and their unfoldings
We describe how a finite braid move lifts to the upstairs sequence. Let i , j ∈ I g i,j\in I_g i , j ∈ I g be distinct vertices such that
c i , j c j , i < 4 , c_{i,j}c_{j,i}<4, c i , j c j , i < 4 ,
and write
I i = ( ı i , 1 , … , ı i , e i ) , I j = ( ı j , 1 , … , ı j , e j ) I_i=(\imath_{i,1},\ldots,\imath_{i,e_i}),\qquad I_j=(\imath_{j,1},\ldots,\imath_{j,e_j}) I i = ( i , 1 , … , i , e i ) , I j = ( j , 1 , … , j , e j )
for the corresponding Γ \Gamma Γ -orbits in I ^ g ^ \hat{I}_{\hat{g}} I ^ g ^ .
Since the Γ \Gamma Γ -action is admissible, there are no edges between two vertices in the same orbit. Since g ^ \hat{g} g ^ is simply-laced, the Cartan folding formula (7) gives
− c i , j = # { ı ∈ I i ∣ ı ∼ ȷ } , − c j , i = # { ȷ ∈ I j ∣ ȷ ∼ ı } , -c_{i,j}=\#\{\imath\in I_i\mid\imath\sim\jmath\},\qquad
-c_{j,i}=\#\{\jmath\in I_j\mid\jmath\sim\imath\}, − c i , j = # { ∈ I i ∣ ∼ } , − c j , i = # { ∈ I j ∣ ∼ } ,
for ı ∈ I i \imath\in I_i ∈ I i and ȷ ∈ I j \jmath\in I_j ∈ I j . Hence
c i , j c j , i ∈ { 0 , 1 , 2 , 3 } , m i , j ∈ { 2 , 3 , 4 , 6 } . c_{i,j}c_{j,i}\in\{0,1,2,3\},\qquad m_{i,j}\in\{2,3,4,6\}. c i , j c j , i ∈ { 0 , 1 , 2 , 3 } , m i , j ∈ { 2 , 3 , 4 , 6 } .
Lemma 4.2. Suppose that 0 < c i j c j i < 4 0<c_{ij}c_{ji}<4 0 < c ij c j i < 4 . Then every connected component of the subdiagram on I i ⊔ I j I_i\sqcup I_j I i ⊔ I j consists of one vertex on one side joined to respectively one, two, or three vertices on the other side, according as
c i j c j i = 1 , 2 , or 3. c_{ij}c_{ji}=1,\ 2,\ \text{or }3. c ij c j i = 1 , 2 , or 3. Moreover, all connected components are isomorphic.
Proof. Every vertex of I j I_j I j has − c i j -c_{ij} − c ij neighbors in I i I_i I i , and every vertex of I i I_i I i has − c j i -c_{ji} − c j i neighbors in I j I_j I j . Since
c i j c j i ∈ { 1 , 2 , 3 } , c_{ij}c_{ji}\in\{1,2,3\}, c ij c j i ∈ { 1 , 2 , 3 } , one of these two numbers is 1. Hence each connected component has one vertex on one side and respectively one, two, or three vertices on the other side. The last assertion follows from the transitivity of the Γ \Gamma Γ -action on I i I_i I i . □
The subdiagram on I i ⊔ I j I_i\sqcup I_j I i ⊔ I j is therefore as follows:
(a) If m i , j = 2 m_{i,j}=2 m i , j = 2 , there are no edges between I i I_i I i and I j I_j I j .
(b) If m i , j = 3 m_{i,j}=3 m i , j = 3 , every vertex in either orbit has exactly one neighbor in the other orbit. Hence e i = e j e_i=e_j e i = e j , and the edges give a bijection between I i I_i I i and I j I_j I j .
(c) If m i , j = 4 m_{i,j}=4 m i , j = 4 , up to interchanging i i i and j j j , we have ( c i , j , c j , i ) = ( − 1 , − 2 ) (c_{i,j},c_{j,i})=(-1,-2) ( c i , j , c j , i ) = ( − 1 , − 2 ) . Thus every vertex of I i I_i I i has two neighbors in I j I_j I j , while every vertex of I j I_j I j has one neighbor in I i I_i I i .
(d) If m i , j = 6 m_{i,j}=6 m i , j = 6 , up to interchanging i i i and j j j , we have ( c i , j , c j , i ) = ( − 1 , − 3 ) (c_{i,j},c_{j,i})=(-1,-3) ( c i , j , c j , i ) = ( − 1 , − 3 ) . Thus every vertex of I i I_i I i has three neighbors in I j I_j I j , while every vertex of I j I_j I j has one neighbor in I i I_i I i .
We shall use the following terminology repeatedly. Let i = ( i 1 , … , i l ) i=(i_1,\ldots,i_l) i = ( i 1 , … , i l ) and let k k k be a position.
(i) A 2-move at k k k replaces ( i k , i k + 1 ) = ( i , j ) (i_k,i_{k+1})=(i,j) ( i k , i k + 1 ) = ( i , j ) by ( j , i ) (j,i) ( j , i ) , when c i , j = 0 c_{i,j}=0 c i , j = 0 .
(ii) A 3-move at k k k replaces ( i k , i k + 1 , i k + 2 ) = ( i , j , i ) (i_k,i_{k+1},i_{k+2})=(i,j,i) ( i k , i k + 1 , i k + 2 ) = ( i , j , i ) by ( j , i , j ) (j,i,j) ( j , i , j ) , when c i , j c j , i = 1 c_{i,j}c_{j,i}=1 c i , j c j , i = 1 .
(iii) A 4-move at k k k replaces ( i , j , i , j ) (i,j,i,j) ( i , j , i , j ) by ( j , i , j , i ) (j,i,j,i) ( j , i , j , i ) , when c i , j c j , i = 2 c_{i,j}c_{j,i}=2 c i , j c j , i = 2 .
(iv) A 6-move at k k k replaces ( i , j , i , j , i , j ) (i,j,i,j,i,j) ( i , j , i , j , i , j ) by ( j , i , j , i , j , i ) (j,i,j,i,j,i) ( j , i , j , i , j , i ) , when c i , j c j , i = 3 c_{i,j}c_{j,i}=3 c i , j c j , i = 3 .
We denote an m m m -move starting at position k k k by ( m ; k ) (m;k) ( m ; k ) .
Example 4.3. Consider the folding
Figure 3.
by the order-three diagram automorphism. With the numbering above and our fixed ordering of the fibers, we have
I 0 = ( 1 , 0 , 4 ) , I 1 = ( 3 , 2 , 5 ) , I 2 = ( 6 ) . I_0=(1,0,4),\qquad I_1=(3,2,5),\qquad I_2=(6). I 0 = ( 1 , 0 , 4 ) , I 1 = ( 3 , 2 , 5 ) , I 2 = ( 6 ) .
Thus e 0 = e 1 = 3 e_0=e_1=3 e 0 = e 1 = 3 and e 2 = 1 e_2=1 e 2 = 1 .
The edges between I 0 I_0 I 0 and I 1 I_1 I 1 are
1 ∼ 3 , 0 ∼ 2 , 4 ∼ 5 , while 3 ∼ 6 , 2 ∼ 6 , 5 ∼ 6. 1\sim3,\quad0\sim2,\quad4\sim5,\quad\text{while}\quad3\sim6,\quad2\sim6,\quad5\sim6. 1 ∼ 3 , 0 ∼ 2 , 4 ∼ 5 , while 3 ∼ 6 , 2 ∼ 6 , 5 ∼ 6.
There are no other edges between distinct fibers. Hence, by (7) ,
C = ( 2 − 1 0 − 1 2 − 3 0 − 1 2 ) . C=\begin{pmatrix}2 & -1 & 0 \\ -1 & 2 & -3 \\ 0 & -1 & 2\end{pmatrix}. C = 2 − 1 0 − 1 2 − 1 0 − 3 2 .
Indeed, c 0 , 1 = c 1 , 0 = − 1 c_{0,1}=c_{1,0}=-1 c 0 , 1 = c 1 , 0 = − 1 , c 1 , 2 = − 3 c_{1,2}=-3 c 1 , 2 = − 3 , and c 2 , 1 = − 1 c_{2,1}=-1 c 2 , 1 = − 1 .
The pair 0 , 1 0,1 0 , 1 therefore gives a 3-move, with e 0 = e 1 = 3 e_0=e_1=3 e 0 = e 1 = 3 . The subdiagram on I 0 ⊔ I 1 I_0\sqcup I_1 I 0 ⊔ I 1 consists of the three disjoint edges
1 ∼ 3 , 0 ∼ 2 , 4 ∼ 5. 1\sim3,\qquad0\sim2,\qquad4\sim5. 1 ∼ 3 , 0 ∼ 2 , 4 ∼ 5.
In particular, the two sides of the braid relation lift to
010 ⟼ ( 1 , 0 , 4 ) ( 3 , 2 , 5 ) ( 1 , 0 , 4 ) , and 101 ⟼ ( 3 , 2 , 5 ) ( 1 , 0 , 4 ) ( 3 , 2 , 5 ) . 010\longmapsto(1,0,4)(3,2,5)(1,0,4),\qquad\text{and}\qquad101\longmapsto(3,2,5)(1,0,4)(3,2,5). 010 ⟼ ( 1 , 0 , 4 ) ( 3 , 2 , 5 ) ( 1 , 0 , 4 ) , and 101 ⟼ ( 3 , 2 , 5 ) ( 1 , 0 , 4 ) ( 3 , 2 , 5 ) .
On the other hand, the pair 1 , 2 1,2 1 , 2 gives a 6-move. Here
( c 2 , 1 , c 1 , 2 ) = ( − 1 , − 3 ) , (c_{2,1},c_{1,2})=(-1,-3), ( c 2 , 1 , c 1 , 2 ) = ( − 1 , − 3 ) ,
and the unique vertex 6 ∈ I 2 6\in I_2 6 ∈ I 2 is adjacent to all three vertices 3 , 2 , 5 ∈ I 1 3,2,5\in I_1 3 , 2 , 5 ∈ I 1 . Thus this single affine folding exhibits both the 3-move with e 0 = e 1 = 3 e_0=e_1=3 e 0 = e 1 = 3 and the 6-move with ( e 1 , e 2 ) = ( 3 , 1 ) (e_1,e_2)=(3,1) ( e 1 , e 2 ) = ( 3 , 1 ) .
Let i = ( i 1 , … , i l ) \boldsymbol{i}=(i_1,\ldots,i_l) i = ( i 1 , … , i l ) , and suppose that a finite braid move of length m = m i , j m=m_{i,j} m = m i , j starts at position k k k . We call the subword ( i k , … , i k + m − 1 ) (i_k,\ldots,i_{k+m-1}) ( i k , … , i k + m − 1 ) the braid block of this move and its blockwise lift in i ^ \widehat{\boldsymbol{i}} i the lifted braid block . The subword I i k I_{i_k} I i k starts at position κ ( k , 1 ) \kappa(k,1) κ ( k , 1 ) in i ^ \widehat{\boldsymbol{i}} i .
Proposition 4.4. Let i ′ \boldsymbol{i}' i ′ be obtained from i \boldsymbol{i} i by a finite braid move at position k k k . Then i ^ ′ \widehat{\boldsymbol{i}}' i ′ is obtained from i ^ \widehat{\boldsymbol{i}} i by a finite sequence of 2- and 3-moves depending only on the subdiagram on I i ⊔ I j I_i\sqcup I_j I i ⊔ I j described in (4.1). The reverse braid move is obtained by reversing the same sequence.
Proof. Since the lift is blockwise, it is enough to consider the subword involving I i I_i I i and I j I_j I j . In what follows, → ( 2 ) \xrightarrow{(2)} ( 2 ) and → ( 3 ) \xrightarrow{(3)} ( 3 ) denote sequences of 2-moves and 3-moves, respectively, while → ( 2 , 3 ) \xrightarrow{(2,3)} ( 2 , 3 ) denotes a sequence involving both.
(i) m i , j = 2 m_{i,j}=2 m i , j = 2 . Write
I i = ( a 1 , … , a p ) , I j = ( b 1 , … , b q ) . I_i=(a_1,\ldots,a_p),\qquad I_j=(b_1,\ldots,b_q). I i = ( a 1 , … , a p ) , I j = ( b 1 , … , b q ) .
Since a r ≁ b s a_r\nsim b_s a r ≁ b s for all r , s r,s r , s ,
( a 1 ⋯ a p ) ( b 1 ⋯ b q ) → ( 2 ) ( b 1 ⋯ b q ) ( a 1 ⋯ a p ) . (a_1\cdots a_p)(b_1\cdots b_q)\xrightarrow{(2)}(b_1\cdots b_q)(a_1\cdots a_p). ( a 1 ⋯ a p ) ( b 1 ⋯ b q ) ( 2 ) ( b 1 ⋯ b q ) ( a 1 ⋯ a p ) .
(ii) m i , j = 3 m_{i,j}=3 m i , j = 3 . Put r = e i = e j r=e_i=e_j r = e i = e j . Using 2-moves inside the fibers, we write I i = ( a 1 , … , a r ) I_i=(a_1,\ldots,a_r) I i = ( a 1 , … , a r ) and I j = ( b 1 , … , b r ) I_j=(b_1,\ldots,b_r) I j = ( b 1 , … , b r ) so that
a s ∼ b s , a s ≁ b t ( s ≠ t ) . a_s\sim b_s,\qquad a_s\nsim b_t\quad(s\ne t). a s ∼ b s , a s ≁ b t ( s = t ) .
Then
( a 1 ⋯ a r ) ( b 1 ⋯ b r ) ( a 1 ⋯ a r ) → ( 2 ) ( a 1 b 1 a 1 ) ⋯ ( a r b r a r ) → ( 3 ) ( b 1 a 1 b 1 ) ⋯ ( b r a r b r ) → ( 2 ) ( b 1 ⋯ b r ) ( a 1 ⋯ a r ) ( b 1 ⋯ b r ) . \begin{aligned}
&(a_1\cdots a_r)(b_1\cdots b_r)(a_1\cdots a_r)
\xrightarrow{(2)}
(a_1b_1a_1)\cdots(a_rb_ra_r)\\
&\xrightarrow{(3)}
(b_1a_1b_1)\cdots(b_ra_rb_r)
\xrightarrow{(2)}
(b_1\cdots b_r)(a_1\cdots a_r)(b_1\cdots b_r).
\end{aligned} ( a 1 ⋯ a r ) ( b 1 ⋯ b r ) ( a 1 ⋯ a r ) ( 2 ) ( a 1 b 1 a 1 ) ⋯ ( a r b r a r ) ( 3 ) ( b 1 a 1 b 1 ) ⋯ ( b r a r b r ) ( 2 ) ( b 1 ⋯ b r ) ( a 1 ⋯ a r ) ( b 1 ⋯ b r ) .
(iii) m i , j = 4 m_{i,j}=4 m i , j = 4 . Up to interchanging i i i and j j j ,
( c i , j , c j , i ) = ( − 1 , − 2 ) . (c_{i,j},c_{j,i})=(-1,-2). ( c i , j , c j , i ) = ( − 1 , − 2 ) .
Writing r = e i r=e_i r = e i , we have e j = 2 r e_j=2r e j = 2 r . Using 2-moves inside the fibers, we write I i = ( c 1 , … , c r ) I_i=(c_1,\ldots,c_r) I i = ( c 1 , … , c r ) , and I j = ( a 1 , b 1 , … , a r , b r ) I_j=(a_1,b_1,\ldots,a_r,b_r) I j = ( a 1 , b 1 , … , a r , b r ) , where
a s ∼ c s ∼ b s , a s ≁ b s , a_s\sim c_s\sim b_s,\qquad a_s\not\sim b_s, a s ∼ c s ∼ b s , a s ∼ b s ,
and vertices with distinct subscripts are mutually nonadjacent. Hence
I i I j I i I j → ( 2 ) ∏ s = 1 r c s a s b s c s a s b s . I_iI_jI_iI_j\xrightarrow{(2)}\prod_{s=1}^{r}c_sa_sb_sc_sa_sb_s. I i I j I i I j ( 2 ) s = 1 ∏ r c s a s b s c s a s b s .
For each s s s , suppressing the subscript, we have
c a b c a b → ( 2 ) c b a c a b → ( 3 ) c b c a c b → ( 3 ) b c b a c b → ( 2 ) b c a b c b → ( 3 ) b c a c b c → ( 3 ) b a c a b c → ( 2 ) a b c a b c . cabcab\xrightarrow{(2)}cbacab\xrightarrow{(3)}cbcacb\xrightarrow{(3)}bcbacb\xrightarrow{(2)}bcabcb\xrightarrow{(3)}bcacbc\xrightarrow{(3)}bacabc\xrightarrow{(2)}abcabc. c ab c ab ( 2 ) c ba c ab ( 3 ) c b c a c b ( 3 ) b c ba c b ( 2 ) b c ab c b ( 3 ) b c a c b c ( 3 ) ba c ab c ( 2 ) ab c ab c .
Therefore
I i I j I i I j → ( 2 , 3 ) ∏ s = 1 r a s b s c s a s b s c s → ( 2 ) I j I i I j I i . I_iI_jI_iI_j\xrightarrow{(2,3)}\prod_{s=1}^{r}a_sb_sc_sa_sb_sc_s\xrightarrow{(2)}I_jI_iI_jI_i. I i I j I i I j ( 2 , 3 ) s = 1 ∏ r a s b s c s a s b s c s ( 2 ) I j I i I j I i .
(iv) m i , j = 6 m_{i,j}=6 m i , j = 6 . Up to interchanging i i i and j j j ,
( c i , j , c j , i ) = ( − 1 , − 3 ) . (c_{i,j},c_{j,i})=(-1,-3). ( c i , j , c j , i ) = ( − 1 , − 3 ) .
Writing r = e i r=e_i r = e i , we have e j = 3 r e_j=3r e j = 3 r . Using 2-moves inside the fibers, we write I i = ( d 1 , … , d r ) I_i=(d_1,\ldots,d_r) I i = ( d 1 , … , d r ) , and I j = ( a 1 , b 1 , c 1 , … , a r , b r , c r ) I_j=(a_1,b_1,c_1,\ldots,a_r,b_r,c_r) I j = ( a 1 , b 1 , c 1 , … , a r , b r , c r ) , where
a s ∼ d s , b s ∼ d s , c s ∼ d s , a_s\sim d_s,\qquad b_s\sim d_s,\qquad c_s\sim d_s, a s ∼ d s , b s ∼ d s , c s ∼ d s ,
the vertices a s , b s , c s a_s,b_s,c_s a s , b s , c s are pairwise nonadjacent, and vertices with distinct subscripts are mutually nonadjacent. Thus
I i I j I i I j I i I j → ( 2 ) ∏ s = 1 r d s a s b s c s d s a s b s c s d s a s b s c s . I_iI_jI_iI_jI_iI_j\xrightarrow{(2)}\prod_{s=1}^{r}d_sa_sb_sc_sd_sa_sb_sc_sd_sa_sb_sc_s. I i I j I i I j I i I j ( 2 ) s = 1 ∏ r d s a s b s c s d s a s b s c s d s a s b s c s .
For one component, suppressing the subscript, the fixed sequence
( 2 ; 5 ) , ( 2 ; 6 ) , ( 2 ; 5 ) , ( 3 ; 3 ) , ( 3 ; 7 ) , ( 3 ; 5 ) , ( 2 ; 4 ) , ( 2 ; 7 ) , (2;5),(2;6),(2;5),(3;3),(3;7),(3;5),(2;4),(2;7), ( 2 ; 5 ) , ( 2 ; 6 ) , ( 2 ; 5 ) , ( 3 ; 3 ) , ( 3 ; 7 ) , ( 3 ; 5 ) , ( 2 ; 4 ) , ( 2 ; 7 ) ,
( 3 ; 2 ) , ( 3 ; 4 ) , ( 2 ; 6 ) , ( 3 ; 8 ) , ( 3 ; 10 ) , ( 2 ; 9 ) , (3;2),(3;4),(2;6),(3;8),(3;10),(2;9), ( 3 ; 2 ) , ( 3 ; 4 ) , ( 2 ; 6 ) , ( 3 ; 8 ) , ( 3 ; 10 ) , ( 2 ; 9 ) ,
( 3 ; 7 ) , ( 3 ; 5 ) , ( 2 ; 4 ) , ( 2 ; 3 ) , ( 3 ; 9 ) , ( 2 ; 7 ) , (3;7),(3;5),(2;4),(2;3),(3;9),(2;7), ( 3 ; 7 ) , ( 3 ; 5 ) , ( 2 ; 4 ) , ( 2 ; 3 ) , ( 3 ; 9 ) , ( 2 ; 7 ) ,
( 3 ; 5 ) , ( 3 ; 1 ) , ( 3 ; 3 ) , ( 2 ; 8 ) , ( 2 ; 5 ) , ( 3 ; 6 ) , ( 3 ; 8 ) , ( 3 ; 4 ) , ( 2 ; 6 ) , ( 2 ; 7 ) , ( 2 ; 6 ) (3;5),(3;1),(3;3),(2;8),(2;5),(3;6),(3;8),(3;4),(2;6),(2;7),(2;6) ( 3 ; 5 ) , ( 3 ; 1 ) , ( 3 ; 3 ) , ( 2 ; 8 ) , ( 2 ; 5 ) , ( 3 ; 6 ) , ( 3 ; 8 ) , ( 3 ; 4 ) , ( 2 ; 6 ) , ( 2 ; 7 ) , ( 2 ; 6 )
gives
a b c d a b c d a b c d → ( 2 , 3 ) d a b c d a b c d a b c . abcdabcdabcd\xrightarrow{(2,3)}dabcdabcdabc. ab c d ab c d ab c d ( 2 , 3 ) d ab c d ab c d ab c .
Hence, by reversing this sequence,
d a b c d a b c d a b c → ( 2 , 3 ) a b c d a b c d a b c d . dabcdabcdabc\xrightarrow{(2,3)}abcdabcdabcd. d ab c d ab c d ab c ( 2 , 3 ) ab c d ab c d ab c d .
Applying this to each component gives
I i I j I i I j I i I j → ( 2 , 3 ) ∏ s = 1 r a s b s c s d s a s b s c s d s a s b s c s → ( 2 ) I j I i I j I i I j I i . I_iI_jI_iI_jI_iI_j\xrightarrow{(2,3)}\prod_{s=1}^{r}a_sb_sc_sd_sa_sb_sc_sd_sa_sb_sc_s\xrightarrow{(2)}I_jI_iI_jI_iI_jI_i. I i I j I i I j I i I j ( 2 , 3 ) s = 1 ∏ r a s b s c s d s a s b s c s d s a s b s c s ( 2 ) I j I i I j I i I j I i .
Thus, in every case, the two lifted words are connected by a finite sequence of 2- and 3-moves. Since both moves are involutive, the reverse braid move is obtained by reversing the same sequence. □
Mutation–permutation realizations and folding
We recall the mutation–permutation realizations of finite braid moves. The 2- and 3-moves were treated in [12 ] , and the 4- and 6-moves in [42 ] (see also [53 ] ).
As in Subsection 4.1, mutations and permutations below are applied from left to right.
Proposition 4.5. [12 , 42 ] Let i ′ i' i ′ be obtained from i i i by an ( m ; k ) (m;k) ( m ; k ) -move. Then the compatible pair ( B i ′ , Λ i ′ ) (B^{i'},\Lambda^{i'}) ( B i ′ , Λ i ′ ) is obtained from ( B i , Λ i ) (B^i,\Lambda^i) ( B i , Λ i ) by the following mutation–permutation sequence:
For an ( m ; k ) (m;k) ( m ; k ) -move, denote by T \mathrm{T} T the corresponding mutation–permutation sequence in Proposition 4.5 . For the sequence of 2- and 3-moves
i ^ = j ^ ( 0 ) ⟶ j ^ ( 1 ) ⟶ ⋯ ⟶ j ^ ( s ) = i ^ , \widehat{i}=\widehat{j}^{(0)}\longrightarrow\widehat{j}^{(1)}\longrightarrow\cdots\longrightarrow\widehat{j}^{(s)}=\widehat{i}, i = j ( 0 ) ⟶ j ( 1 ) ⟶ ⋯ ⟶ j ( s ) = i ,
let T ^ a \widehat{\mathrm{T}}_a T a be the mutation–permutation sequence corresponding to
j ^ ( a − 1 ) ⟶ j ^ ( a ) ( 1 ≤ a ≤ s ) . \widehat{j}^{(a-1)}\longrightarrow\widehat{j}^{(a)}\qquad(1\leq a\leq s). j ( a − 1 ) ⟶ j ( a ) ( 1 ≤ a ≤ s ) .
We call the sequence from i i i to i ^ \widehat{i} i a lifted path of the ( m ; k ) (m;k) ( m ; k ) -move and ( T ^ 1 , … , T ^ s ) (\widehat{\mathrm{T}}_1,\ldots,\widehat{\mathrm{T}}_s) ( T 1 , … , T s ) its mutation–permutation realization.
Proposition 4.6. Let i ′ i' i ′ be obtained from i i i by an ( m ; k ) (m;k) ( m ; k ) -move, and let
i ^ = j ^ ( 0 ) ⟶ j ^ ( 1 ) ⟶ ⋯ ⟶ j ^ ( s ) = i ^ \widehat{i}=\widehat{j}^{(0)}\longrightarrow\widehat{j}^{(1)}\longrightarrow\cdots\longrightarrow\widehat{j}^{(s)}=\widehat{i} i = j ( 0 ) ⟶ j ( 1 ) ⟶ ⋯ ⟶ j ( s ) = i be the sequence of 2- and 3-moves in Proposition 4.4 . Then
( B i ^ , Λ i ^ ) → T ^ 1 , … , T ^ s ( B i ′ ^ , Λ i ′ ^ ) F o l d ↓ ↓ F o l d ( B i , Λ i ) → T ( B i ′ , Λ i ′ ) \begin{CD}
(B^{\widehat{i}},\Lambda^{\widehat{i}}) @>{\widehat{\mathrm{T}}_1,\ldots,\widehat{\mathrm{T}}_s}>> (B^{\widehat{i'}},\Lambda^{\widehat{i'}})\\
@V{\mathrm{Fold}}VV @VV{\mathrm{Fold}}V\\
(B^i,\Lambda^i) @>{\mathrm{T}}>> (B^{i'},\Lambda^{i'})
\end{CD} ( B i , Λ i ) Fold ↓ ⏐ ( B i , Λ i ) T 1 , … , T s T ( B i ′ , Λ i ′ ) ↓ ⏐ Fold ( B i ′ , Λ i ′ ) where the vertical arrows are given by Proposition 3.8 .
Proof. By Proposition 4.5 ,
( B i , Λ i ) → ( m ; k ) -move ( B i ′ , Λ i ′ ) . (B^i,\Lambda^i)\xrightarrow{\ (m;k)\text{-move}\ }(B^{i'},\Lambda^{i'}). ( B i , Λ i ) ( m ; k ) -move ( B i ′ , Λ i ′ ) .
Applying the same proposition successively to
j ^ ( a − 1 ) ⟶ j ^ ( a ) ( 1 ≤ a ≤ s ) \widehat{j}^{(a-1)}\longrightarrow\widehat{j}^{(a)}\qquad(1\leq a\leq s) j ( a − 1 ) ⟶ j ( a ) ( 1 ≤ a ≤ s )
gives
( B i ^ , Λ i ^ ) → T ^ 1 , … , T ^ s ( B i ′ ^ , Λ i ′ ^ ) . (B^{\widehat{i}},\Lambda^{\widehat{i}})\xrightarrow{\ \widehat{\mathrm{T}}_1,\ldots,\widehat{\mathrm{T}}_s\ }(B^{\widehat{i'}},\Lambda^{\widehat{i'}}). ( B i , Λ i ) T 1 , … , T s ( B i ′ , Λ i ′ ) .
The two vertical arrows are given by Proposition 3.8 , applied to ( i , i ) (i,i) ( i , i ) and ( i ′ ^ , i ′ ) (\widehat{i'},i') ( i ′ , i ′ ) , respectively. □
Folding of cluster variables and g-vectors
We work in the commutative setting throughout this subsection. Let Σ i \Sigma^{i} Σ i and Σ ^ i \widehat{\Sigma}^{i} Σ i be the initial seeds associated with i i i and i ^ \widehat{i} i , respectively. Define the algebra homomorphism between the seed Laurent polynomial rings
ϕ i : L ( Σ ^ i ) ⟶ L ( Σ i ) , x ^ ( k , r ) ⟼ x k ( 1 ≤ k ≤ l , 1 ≤ r ≤ e i k ) . \phi^{i}:\mathcal{L}(\widehat{\Sigma}^{i})\longrightarrow\mathcal{L}(\Sigma^{i}),\qquad\widehat{x}_{(k,r)}\longmapsto x_k\quad(1\leq k\leq l,\ 1\leq r\leq e_{i_k}). ϕ i : L ( Σ i ) ⟶ L ( Σ i ) , x ( k , r ) ⟼ x k ( 1 ≤ k ≤ l , 1 ≤ r ≤ e i k ) .
We show that ϕ i \phi^{i} ϕ i identifies the cluster variables of Σ ^ i \widehat{\Sigma}^{i} Σ i and Σ i \Sigma^{i} Σ i obtained by the mutation–permutation sequences of Proposition 4.6.
Remark 4.7. For m = 3 , 4 , 6 m=3,4,6 m = 3 , 4 , 6 , the connected components of the subdiagram on I i ⊔ I j I_i\sqcup I_j I i ⊔ I j are classified in Lemma 4.2. The case m = 2 m=2 m = 2 involves only permutations. The additional affine calculations are recorded in Appendix A, and Proposition A.6 reduces (43) to the calculation for each connected component. The indefinite example in (1.9) introduces no additional case of a finite braid move.
By the Laurent phenomenon, the resulting cluster variables belong to L ( Σ ^ i ) \mathcal{L}(\widehat{\Sigma}^{i}) L ( Σ i ) and L ( Σ i ) \mathcal{L}(\Sigma^{i}) L ( Σ i ) , respectively. We shall prove
( x ^ κ ( p , r ) ) p , r → T ^ 1 , … , T ^ s ( x ^ κ ′ ( p , r ) ′ ) p , r ϕ i ↓ ↓ ϕ i ( x p ) p → T ( x p ′ ) p \begin{CD}
\left(\widehat{x}_{\kappa(p,r)}\right)_{p,r} @>{\widehat{T}_1,\ldots,\widehat{T}_s}>> \left(\widehat{x}'_{\kappa'(p,r)}\right)_{p,r}\\
@V{\phi^{i}}VV @VV{\phi^{i}}V\\
\left(x_p\right)_p @>{T}>> \left(x'_p\right)_p
\end{CD} ( x κ ( p , r ) ) p , r ϕ i ↓ ⏐ ( x p ) p T 1 , … , T s T ( x κ ′ ( p , r ) ′ ) p , r ↓ ⏐ ϕ i ( x p ′ ) p
Notice that the right vertical arrow is the same Laurent homomorphism ϕ i \phi^{i} ϕ i , rather than the Laurent homomorphism ϕ i ′ \phi^{i'} ϕ i ′ defined from Σ ^ i ′ \widehat{\Sigma}^{i'} Σ i ′ . Equivalently, we shall prove
ϕ i ( x ^ κ ′ ( p , r ) ) = x p ′ ( 1 ≤ p ≤ l , 1 ≤ r ≤ e i p ′ ) . (43) \phi^{i}\left(\widehat{x}_{\kappa'(p,r)}\right)=x'_p\quad(1\leq p\leq l,\ 1\leq r\leq e_{i'_p}). \tag*{(43)} ϕ i ( x κ ′ ( p , r ) ) = x p ′ ( 1 ≤ p ≤ l , 1 ≤ r ≤ e i p ′ ) . (43)
where κ ′ \kappa' κ ′ is defined for i ′ i' i ′ by (13) .
Proposition 4.8. Suppose that i ′ i' i ′ is obtained from i i i by a 2-move. Then (43) holds.
The 3-move. We first consider the case e i = e j = 2 e_i=e_j=2 e i = e j = 2 . This case occurs in the folding A 2 n − 1 → C n A_{2n-1}\to C_n A 2 n − 1 → C n away from the multiple edge. Write I i = ( a , A ) I_i=(a,A) I i = ( a , A ) and I j = ( b , B ) I_j=(b,B) I j = ( b , B ) , where a ∼ b a\sim b a ∼ b , A ∼ B A\sim B A ∼ B , a ≁ B a\not\sim B a ∼ B and A ≁ b A\not\sim b A ∼ b . Thus the braid move
i j i ⟶ j i j iji\longrightarrow jij ij i ⟶ j ij
lifts to
a A b B a A ⟶ b B a A b B . aAbBaA\longrightarrow bBaAbB. a A b B a A ⟶ b B a A b B .
By Proposition 4.4, the lifted 3-move decomposes into e i = e j e_i=e_j e i = e j simply-laced 3-moves, together with 2-moves rearranging the fibers. Thus, in the present case e i = e j = 2 e_i=e_j=2 e i = e j = 2 , exactly two 3-moves occur. For the calculation below, we choose the following sequence of 2- and 3-moves:
a A b B a A ⟶ a b A B a A ⟶ a b A a B A ⟶ a b a A B A ⟶ b a b A B A ⟶ b a b B A B ⟶ b a B b A B ⟶ b B a b A B ⟶ b B a A b B . (44) \begin{aligned}
aAbBaA&\longrightarrow abABaA\longrightarrow abAaBA\longrightarrow abaABA\\
&\longrightarrow babABA\longrightarrow babBAB\longrightarrow baBbAB\longrightarrow bBabAB\longrightarrow bBaAbB. \tag*{(44)}
\end{aligned} a A b B a A ⟶ ab A B a A ⟶ ab A a B A ⟶ aba A B A ⟶ bab A B A ⟶ bab B A B ⟶ ba B b A B ⟶ b B ab A B ⟶ b B a A b B . (44)
Here the underlined subwords indicate the successive braid moves. More precisely, the successive moves are
( 2 ; κ ( k , 1 ) + 1 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 3 ; κ ( k , 1 ) ) , (2;\kappa(k,1)+1),\quad(2;\kappa(k,1)+3),\quad(2;\kappa(k,1)+2),\quad(3;\kappa(k,1)), ( 2 ; κ ( k , 1 ) + 1 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 3 ; κ ( k , 1 )) ,
( 3 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 1 ) , ( 2 ; κ ( k , 1 ) + 3 ) . (3;\kappa(k,1)+3),\quad(2;\kappa(k,1)+2),\quad(2;\kappa(k,1)+1),\quad(2;\kappa(k,1)+3). ( 3 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 1 ) , ( 2 ; κ ( k , 1 ) + 3 ) .
The two 3-moves occur at positions κ ( k , 1 ) \kappa(k,1) κ ( k , 1 ) and κ ( k , 1 ) + 3 \kappa(k,1)+3 κ ( k , 1 ) + 3 ; all the remaining moves are 2-moves. We compare the exchange relations associated with these mutations. For the three positions of the downstairs braid block, write
x 1 : = x k , x 2 : = x k + 1 , x 3 : = x k + 2 . x_1:=x_k,\qquad x_2:=x_{k+1},\qquad x_3:=x_{k+2}. x 1 := x k , x 2 := x k + 1 , x 3 := x k + 2 .
By Proposition 4.5, the downstairs 3-move is realized by ( μ k , P k + 1 ) (\mu_k,P_{k+1}) ( μ k , P k + 1 ) in order. Let D + D^+ D + and D − D^- D − denote the contributions of the rows indexed by r ∉ { k , k + 1 , k + 2 } r\notin\{k,k+1,k+2\} r ∈ / { k , k + 1 , k + 2 } to the positive and negative exchange monomials at the mutation μ k \mu_k μ k . Then
x 1 x 1 ′ = D + x 2 + D − x 3 , x_1x'_1=D^+x_2+D^-x_3, x 1 x 1 ′ = D + x 2 + D − x 3 ,
and hence, after P k + 1 P_{k+1} P k + 1 ,
x 1 ′ = D + x 2 + D − x 3 x 1 , x 2 ′ = x 3 , x 3 ′ = x 2 . x'_1=\frac{D^+x_2+D^-x_3}{x_1},\qquad x'_2=x_3,\qquad x'_3=x_2. x 1 ′ = x 1 D + x 2 + D − x 3 , x 2 ′ = x 3 , x 3 ′ = x 2 .
For the lifted braid block a A b B a A aAbBaA a A b B a A , write
x ^ 1 , x ^ 2 , x ^ 3 , x ^ 4 , x ^ 5 , x ^ 6 \widehat{x}_1,\widehat{x}_2,\widehat{x}_3,\widehat{x}_4,\widehat{x}_5,\widehat{x}_6 x 1 , x 2 , x 3 , x 4 , x 5 , x 6
for the corresponding cluster variables, in this order. Thus
ϕ i ( x ^ 1 ) = ϕ i ( x ^ 2 ) = x 1 , ϕ i ( x ^ 3 ) = ϕ i ( x ^ 4 ) = x 2 , ϕ i ( x ^ 5 ) = ϕ i ( x ^ 6 ) = x 3 . \phi^i(\widehat{x}_1)=\phi^i(\widehat{x}_2)=x_1,\qquad\phi^i(\widehat{x}_3)=\phi^i(\widehat{x}_4)=x_2,\qquad\phi^i(\widehat{x}_5)=\phi^i(\widehat{x}_6)=x_3. ϕ i ( x 1 ) = ϕ i ( x 2 ) = x 1 , ϕ i ( x 3 ) = ϕ i ( x 4 ) = x 2 , ϕ i ( x 5 ) = ϕ i ( x 6 ) = x 3 .
For a = 1 , 2 a=1,2 a = 1 , 2 , let U a + U_a^+ U a + and U a − U_a^- U a − denote the contributions of the rows with indices outside [ κ ( k , 1 ) , κ ( k + 2 , 2 ) ] [\kappa(k,1),\kappa(k+2,2)] [ κ ( k , 1 ) , κ ( k + 2 , 2 )] to the positive and negative exchange monomials at the a a a th upstairs mutation, computed after all preceding mutations and permutations in (44) . Thus U 1 ± U_1^\pm U 1 ± are computed after the three preceding 2-moves, whereas U 2 ± U_2^\pm U 2 ± are computed after all preceding moves, including the first 3-move.
Lemma 4.9. With the notation above, we have
ϕ i ( U 1 ± ) = ϕ i ( U 2 ± ) = D ± . \phi^i(U_1^\pm)=\phi^i(U_2^\pm)=D^\pm. ϕ i ( U 1 ± ) = ϕ i ( U 2 ± ) = D ± .
Proof. Set
k a : = κ ( k , a ) ( a = 1 , 2 ) . k_a:=\kappa(k,a)\qquad(a=1,2). k a := κ ( k , a ) ( a = 1 , 2 ) .
Let B ^ ( a ) \widehat{B}^{(a)} B ( a ) denote the exchange matrix immediately before the a a a -th upstairs mutation, and let k ^ a \widehat{k}_a k a be the corresponding mutation position.
By (44) , the two mutation vertices are the positions of k 1 k_1 k 1 and k 2 k_2 k 2 after the preceding permutations, respectively. Since k 1 k_1 k 1 and k 2 k_2 k 2 belong to the same fiber,
b ^ k ^ 1 , k ^ 2 i = 0. \widehat{b}^{\,i}_{\widehat{k}_1,\widehat{k}_2}=0. b k 1 , k 2 i = 0.
Therefore the first mutation does not change the k 2 k_2 k 2 -column:
( μ k ^ 1 B ^ i ) r , k ^ 2 = b ^ r , k ^ 2 i + [ b ^ r , k ^ 1 i ] + [ b ^ k ^ 1 , k ^ 2 i ] + − [ − b ^ r , k ^ 1 i ] + [ − b ^ k ^ 1 , k ^ 2 i ] + = b ^ r , k ^ 2 i . (\mu_{\widehat{k}_1}\widehat{B}^{\,i})_{r,\widehat{k}_2}
=\widehat{b}^{\,i}_{r,\widehat{k}_2}
+\left[\widehat{b}^{\,i}_{r,\widehat{k}_1}\right]_+
\left[\widehat{b}^{\,i}_{\widehat{k}_1,\widehat{k}_2}\right]_+
-\left[-\widehat{b}^{\,i}_{r,\widehat{k}_1}\right]_+
\left[-\widehat{b}^{\,i}_{\widehat{k}_1,\widehat{k}_2}\right]_+
=\widehat{b}^{\,i}_{r,\widehat{k}_2}. ( μ k 1 B i ) r , k 2 = b r , k 2 i + [ b r , k 1 i ] + [ b k 1 , k 2 i ] + − [ − b r , k 1 i ] + [ − b k 1 , k 2 i ] + = b r , k 2 i .
All moves other than the two 3-moves in (44) are 2-moves and only permute rows and columns. Hence, up to these permutations, the mutation column of B ^ ( a ) \widehat{B}^{(a)} B ( a ) is the initial k a k_a k a -column of B ^ i \widehat{B}^{i} B i , for a = 1 , 2 a=1,2 a = 1 , 2 .
Now fix r ∉ { k , k + 1 , k + 2 } r \notin\{k,k+1,k+2\} r ∈ / { k , k + 1 , k + 2 } , and put r s : = κ ( r , s ) r_s:=\kappa(r,s) r s := κ ( r , s ) for ( 1 ≤ s ≤ e i r ) (1\le s\le e_{ir}) ( 1 ≤ s ≤ e i r ) . By Proposition 3.8 and Lemma 3.10 ,
∑ s = 1 e i r [ ± b ^ r s , k a i ] + = [ ± b r , k i ] + ( a = 1 , 2 ) . \sum_{s=1}^{e_{ir}}\left[\pm\widehat{b}^{i}_{r_s,k_a}\right]_{+}=\left[\pm b^{i}_{r,k}\right]_{+}\qquad(a=1,2). s = 1 ∑ e i r [ ± b r s , k a i ] + = [ ± b r , k i ] + ( a = 1 , 2 ) .
Following the corresponding rows and columns through (44) , we obtain
∑ s = 1 e i r [ ± b ^ r ^ s , k ^ a ( a ) ] + = [ ± b r , k i ] + ( a = 1 , 2 ) , \sum_{s=1}^{e_{ir}}\left[\pm\widehat{b}^{(a)}_{\widehat{r}_s,\widehat{k}_a}\right]_{+}=\left[\pm b^{i}_{r,k}\right]_{+}\qquad(a=1,2), s = 1 ∑ e i r [ ± b r s , k a ( a ) ] + = [ ± b r , k i ] + ( a = 1 , 2 ) ,
where r ^ s \widehat{r}_s r s is the position of r s r_s r s immediately before the a a a -th mutation. Since no position outside [ κ ( k , 1 ) , κ ( k + 2 , 2 ) ] [\kappa(k,1),\kappa(k+2,2)] [ κ ( k , 1 ) , κ ( k + 2 , 2 )] is mutated, the corresponding cluster variables are only permuted, and hence
ϕ i ( x ^ r ^ s ) = x r . \phi^{i}(\widehat{x}_{\widehat{r}_s})=x_r. ϕ i ( x r s ) = x r .
It follows that the contribution of the rows indexed by r ^ s \widehat{r}_s r s ( 1 ≤ s ≤ e i r ) (1\le s\le e_{ir}) ( 1 ≤ s ≤ e i r ) to U a ± U_a^{\pm} U a ± satisfies
ϕ i ( ∏ s = 1 e i r x ^ r ^ s [ ± b ^ r ^ s , k ^ a ( a ) ] + ) = x r [ ± b r , k i ] + . \phi^{i}\left(\prod_{s=1}^{e_{ir}}\widehat{x}_{\widehat{r}_s}^{[\pm\widehat{b}^{(a)}_{\widehat{r}_s,\widehat{k}_a}]_{+}}\right)=x_r^{[\pm b^{i}_{r,k}]_{+}}. ϕ i ( s = 1 ∏ e i r x r s [ ± b r s , k a ( a ) ] + ) = x r [ ± b r , k i ] + .
Multiplying over all r ∉ { k , k + 1 , k + 2 } r\notin\{k,k+1,k+2\} r ∈ / { k , k + 1 , k + 2 } gives
ϕ i ( U a ± ) = D ± ( a = 1 , 2 ) . \phi^{i}(U_a^{\pm})=D^{\pm}\qquad(a=1,2). ϕ i ( U a ± ) = D ± ( a = 1 , 2 ) .
Proposition 4.10. Suppose that i ′ i' i ′ is obtained from i i i by a 3-move
i j i ⟶ j i j with e i = e j = 2. iji\longrightarrow jij\qquad\text{with }e_i=e_j=2. ij i ⟶ j ij with e i = e j = 2. Then (43) holds.
Proof. With the local notation introduced above, the final downstairs variables are
x 1 ′ = D + x 2 + D − x 3 x 1 , x 2 ′ = x 3 , x 3 ′ = x 2 . x'_1=\frac{D^{+}x_2+D^{-}x_3}{x_1},\qquad x'_2=x_3,\qquad x'_3=x_2. x 1 ′ = x 1 D + x 2 + D − x 3 , x 2 ′ = x 3 , x 3 ′ = x 2 .
We now follow the two mutations in (44) . For the first 3-move, the exchange relation is
x ^ 1 x ^ 1 ′ = U 1 + x ^ 3 + U 1 − x ^ 5 . \widehat{x}_1\widehat{x}'_1=U_1^{+}\widehat{x}_3+U_1^{-}\widehat{x}_5. x 1 x 1 ′ = U 1 + x 3 + U 1 − x 5 .
whereas for the second 3-move it is
x ^ 2 x ^ 2 ′ = U 2 + x ^ 4 + U 2 − x ^ 6 . \widehat{x}_2\widehat{x}'_2=U_2^{+}\widehat{x}_4+U_2^{-}\widehat{x}_6. x 2 x 2 ′ = U 2 + x 4 + U 2 − x 6 .
All other steps in (44) are 2-moves and hence only permute the cluster variables. Therefore
x ^ 3 ′ = x ^ 5 , x ^ 4 ′ = x ^ 6 , x ^ 5 ′ = x ^ 3 , x ^ 6 ′ = x ^ 4 . \widehat{x}'_3=\widehat{x}_5,\qquad\widehat{x}'_4=\widehat{x}_6,\qquad\widehat{x}'_5=\widehat{x}_3,\qquad\widehat{x}'_6=\widehat{x}_4. x 3 ′ = x 5 , x 4 ′ = x 6 , x 5 ′ = x 3 , x 6 ′ = x 4 .
Thus
x ^ 1 ′ = U 1 + x ^ 3 + U 1 − x ^ 5 x ^ 1 , x ^ 2 ′ = U 2 + x ^ 4 + U 2 − x ^ 6 x ^ 2 . \widehat{x}'_1=\frac{U_1^{+}\widehat{x}_3+U_1^{-}\widehat{x}_5}{\widehat{x}_1},\qquad\widehat{x}'_2=\frac{U_2^{+}\widehat{x}_4+U_2^{-}\widehat{x}_6}{\widehat{x}_2}. x 1 ′ = x 1 U 1 + x 3 + U 1 − x 5 , x 2 ′ = x 2 U 2 + x 4 + U 2 − x 6 .
By Lemma 4.9 ,
ϕ i ( U 1 ± ) = ϕ i ( U 2 ± ) = D ± . \phi^{i}(U_1^{\pm})=\phi^{i}(U_2^{\pm})=D^{\pm}. ϕ i ( U 1 ± ) = ϕ i ( U 2 ± ) = D ± .
Using
ϕ i ( x ^ 1 ) = ϕ i ( x ^ 2 ) = x 1 , ϕ i ( x ^ 3 ) = ϕ i ( x ^ 4 ) = x 2 , ϕ i ( x ^ 5 ) = ϕ i ( x ^ 6 ) = x 3 , \phi^{i}(\widehat{x}_{1})=\phi^{i}(\widehat{x}_{2})=x_{1},\qquad\phi^{i}(\widehat{x}_{3})=\phi^{i}(\widehat{x}_{4})=x_{2},\qquad\phi^{i}(\widehat{x}_{5})=\phi^{i}(\widehat{x}_{6})=x_{3}, ϕ i ( x 1 ) = ϕ i ( x 2 ) = x 1 , ϕ i ( x 3 ) = ϕ i ( x 4 ) = x 2 , ϕ i ( x 5 ) = ϕ i ( x 6 ) = x 3 ,
we obtain
ϕ i ( x ^ 1 ′ ) = D + x 2 + D − x 3 x 1 = x 1 ′ , ϕ i ( x ^ 2 ′ ) = D + x 2 + D − x 3 x 1 = x 1 ′ . \phi^{i}(\widehat{x}'_{1})=\frac{D^{+}x_{2}+D^{-}x_{3}}{x_{1}}=x'_{1},\qquad\phi^{i}(\widehat{x}'_{2})=\frac{D^{+}x_{2}+D^{-}x_{3}}{x_{1}}=x'_{1}. ϕ i ( x 1 ′ ) = x 1 D + x 2 + D − x 3 = x 1 ′ , ϕ i ( x 2 ′ ) = x 1 D + x 2 + D − x 3 = x 1 ′ .
For the remaining variables,
ϕ i ( x ^ 3 ′ ) = ϕ i ( x ^ 4 ′ ) = x 3 = x 2 ′ , and ϕ i ( x ^ 5 ′ ) = ϕ i ( x ^ 6 ′ ) = x 2 = x 3 ′ . \phi^{i}(\widehat{x}'_{3})=\phi^{i}(\widehat{x}'_{4})=x_{3}=x'_{2},\qquad\text{and}\qquad\phi^{i}(\widehat{x}'_{5})=\phi^{i}(\widehat{x}'_{6})=x_{2}=x'_{3}. ϕ i ( x 3 ′ ) = ϕ i ( x 4 ′ ) = x 3 = x 2 ′ , and ϕ i ( x 5 ′ ) = ϕ i ( x 6 ′ ) = x 2 = x 3 ′ .
Therefore (43) holds. □
We next consider the case e i = e j = 1 e_i=e_j=1 e i = e j = 1 . Write I i = ( a ) I_i=(a) I i = ( a ) and I j = ( b ) I_j=(b) I j = ( b ) , with a ∼ b a\sim b a ∼ b . Then
i j i ⟶ j i j lifts simply to a b a ⟶ b a b . iji\longrightarrow jij\quad\text{lifts simply to}\quad aba\longrightarrow bab. ij i ⟶ j ij lifts simply to aba ⟶ bab .
Thus the lifted path consists of a single 3-move. This case occurs, for example, in the folding D n + 1 ⟶ B n D_{n+1}\longrightarrow B_n D n + 1 ⟶ B n .
Proposition 4.11. Suppose that i ′ i' i ′ is obtained from i i i by a 3-move with e i = e j = 1 e_i=e_j=1 e i = e j = 1 . Then (43) holds.
Proof. This follows immediately from Proposition 3.8 and Lemma 3.10. □
The additional affine case e i = e j = 3 e_i=e_j=3 e i = e j = 3 is recorded in Appendix A, where the lifted path and the identities for the final cluster variables yielding (43) are given.
• The 4-move. We first consider the case e i = 1 e_i=1 e i = 1 and e j = 2 e_j=2 e j = 2 . This case occurs in the folding A 2 n − 1 ⟶ C n A_{2n-1}\longrightarrow C_n A 2 n − 1 ⟶ C n at the multiple edge; in this case, i = n i=n i = n and j = n − 1 j=n-1 j = n − 1 . Write I i = ( c ) I_i=(c) I i = ( c ) and I j = ( a , b ) I_j=(a,b) I j = ( a , b ) , where a ∼ c ∼ b a\sim c\sim b a ∼ c ∼ b and a ≁ b a\nsim b a ≁ b . Thus the braid move
i j i j ⟶ j i j i lifts to c a b c a b ⟶ a b c a b c . ijij\longrightarrow jiji\quad\text{lifts to}\quad cabcab\longrightarrow abcabc. ij ij ⟶ j ij i lifts to c ab c ab ⟶ ab c ab c .
By Proposition 4.4, the lifted 4-move is obtained by four simply-laced 3-moves together with three 2-moves. For the calculation below, we choose the following sequence of 2- and 3-moves:
c a b c a b ⟶ c b a c a b ⟶ c b c a c b ⟶ b c b a c b ⟶ b c a b c b ⟶ b c a c b c ⟶ b a c a b c ⟶ a b c a b c . (45) cabcab\longrightarrow cbacab\longrightarrow cbcacb\longrightarrow bcbacb\longrightarrow bcabcb\longrightarrow bcacbc\longrightarrow bacabc\longrightarrow abcabc. \tag*{(45)} c ab c ab ⟶ c ba c ab ⟶ c b c a c b ⟶ b c ba c b ⟶ b c ab c b ⟶ b c a c b c ⟶ ba c ab c ⟶ ab c ab c . (45)
Here the underlined subwords indicate the successive braid moves. More precisely, the successive moves are
( 2 ; κ ( k , 1 ) + 1 ) , ( 3 ; κ ( k , 1 ) + 2 ) , ( 3 ; κ ( k , 1 ) ) , ( 2 ; κ ( k , 1 ) + 2 ) , (2;\kappa(k,1)+1),\qquad(3;\kappa(k,1)+2),\qquad(3;\kappa(k,1)),\qquad(2;\kappa(k,1)+2), ( 2 ; κ ( k , 1 ) + 1 ) , ( 3 ; κ ( k , 1 ) + 2 ) , ( 3 ; κ ( k , 1 )) , ( 2 ; κ ( k , 1 ) + 2 ) ,
( 3 ; κ ( k , 1 ) + 3 ) , ( 3 ; κ ( k , 1 ) + 1 ) , ( 2 ; κ ( k , 1 ) ) . (3;\kappa(k,1)+3),\qquad(3;\kappa(k,1)+1),\qquad(2;\kappa(k,1)). ( 3 ; κ ( k , 1 ) + 3 ) , ( 3 ; κ ( k , 1 ) + 1 ) , ( 2 ; κ ( k , 1 )) .
Thus the four 3-moves occur at positions
κ ( k , 1 ) + 2 , κ ( k , 1 ) , κ ( k , 1 ) + 3 , κ ( k , 1 ) + 1 , \kappa(k,1)+2,\qquad\kappa(k,1),\qquad\kappa(k,1)+3,\qquad\kappa(k,1)+1, κ ( k , 1 ) + 2 , κ ( k , 1 ) , κ ( k , 1 ) + 3 , κ ( k , 1 ) + 1 ,
and the remaining three moves are 2-moves.
Lemma 4.12. For r ∉ { k , k + 1 , k + 2 , k + 3 } r\notin\{k,k+1,k+2,k+3\} r ∈ / { k , k + 1 , k + 2 , k + 3 } , the entries in columns k , k + 1 k,k+1 k , k + 1 satisfy
( b r , k i , b r , k + 1 i ) = { ( − 2 , 1 ) , r + = k + 1 , ( 1 , 0 ) , r + = k , ( 0 , 0 ) , otherwise . (46) \left(b^{i}_{r,k},b^{i}_{r,k+1}\right)=
\begin{cases}
(-2,1), & r_{+}=k+1,\\
(1,0), & r_{+}=k,\\
(0,0), & \text{otherwise}.
\end{cases} \tag*{(46)} ( b r , k i , b r , k + 1 i ) = ⎩ ⎨ ⎧ ( − 2 , 1 ) , ( 1 , 0 ) , ( 0 , 0 ) , r + = k + 1 , r + = k , otherwise . (46)
In particular, for the mutation sequence
( μ k , μ k + 1 , μ k ) , (\mu_k,\mu_{k+1},\mu_k), ( μ k , μ k + 1 , μ k ) ,
the relevant initial data are
ρ d o w n ( r ) : = ( b r , k i , b r , k + 1 i ) ∈ { ( − 2 , 1 ) , ( 1 , 0 ) , ( 0 , 0 ) } . \rho_{\mathrm{down}}(r):=\left(b^i_{r,k},b^i_{r,k+1}\right)\in\{(-2,1),(1,0),(0,0)\}. ρ down ( r ) := ( b r , k i , b r , k + 1 i ) ∈ {( − 2 , 1 ) , ( 1 , 0 ) , ( 0 , 0 )} .
We call the tuple of entries of a row in a specified ordered list of columns its row pattern. The three patterns above define the classes P \mathcal{P} P , Q \mathcal{Q} Q , Z \mathcal{Z} Z of row indices, respectively. The entries of the rows of B i B^i B i indexed by κ ( r , s ) \kappa(r,s) κ ( r , s ) (1 ≤ s ≤ e i r 1\le s\le e_{i_r} 1 ≤ s ≤ e i r ) in columns κ ( k , 1 ) , κ ( k + 1 , 1 ) , κ ( k + 1 , 2 ) \kappa(k,1),\kappa(k+1,1),\kappa(k+1,2) κ ( k , 1 ) , κ ( k + 1 , 1 ) , κ ( k + 1 , 2 ) , in this order, are
P Q Z ρ d o w n ( − 2 , 1 ) ( 1 , 0 ) ( 0 , 0 ) ρ u p ( − 1 , 1 , 0 ) , ( − 1 , 0 , 1 ) ( 1 , 0 , 0 ) ( 0 , 0 , 0 ) (47) \begin{array}{c|ccc}
& \mathcal{P} & \mathcal{Q} & \mathcal{Z}\\
\rho_{\mathrm{down}} & (-2,1) & (1,0) & (0,0)\\
\rho_{\mathrm{up}} & (-1,1,0),\ (-1,0,1) & (1,0,0) & (0,0,0) \tag*{(47)}
\end{array} ρ down ρ up P ( − 2 , 1 ) ( − 1 , 1 , 0 ) , ( − 1 , 0 , 1 ) Q ( 1 , 0 ) ( 1 , 0 , 0 ) Z ( 0 , 0 ) ( 0 , 0 , 0 ) (47)
Here P \mathcal{P} P occurs only when i r = j i_r=j i r = j and hence e j = 2 e_j=2 e j = 2 , whereas Q \mathcal{Q} Q occurs only when i r = i i_r=i i r = i and hence e i = 1 e_i=1 e i = 1 . The class Z \mathcal{Z} Z consists of the remaining cases with ρ d o w n ( r ) = ( 0 , 0 ) \rho_{\mathrm{down}}(r)=(0,0) ρ down ( r ) = ( 0 , 0 ) ; every row indexed by κ ( r , s ) \kappa(r,s) κ ( r , s ) (1 ≤ s ≤ e i r 1\le s\le e_{i_r} 1 ≤ s ≤ e i r ) has entries ( 0 , 0 , 0 ) (0,0,0) ( 0 , 0 , 0 ) in the three columns specified above.
Proof. By the defining formula for the GLS exchange matrix, a nonzero entry in columns k , k + 1 k,k+1 k , k + 1 can occur only from (i) (21) or from (ii) (21) . Since
k + = k + 2 , ( k + 1 ) + = k + 3 , k_+=k+2,\qquad(k+1)_+=k+3, k + = k + 2 , ( k + 1 ) + = k + 3 ,
a direct case-by-case check, according as r < k r<k r < k or r > k + 3 r>k+3 r > k + 3 , gives (46) .
For the mutation sequence ( μ k , μ k + 1 , μ k ) (\mu_k,\mu_{k+1},\mu_k) ( μ k , μ k + 1 , μ k ) , the initial entries in columns k , k + 1 k,k+1 k , k + 1 are
( − 2 , 1 ) , ( 1 , 0 ) , ( 0 , 0 ) , (-2,1),\qquad(1,0),\qquad(0,0), ( − 2 , 1 ) , ( 1 , 0 ) , ( 0 , 0 ) ,
which give the classes P \mathcal{P} P , Q \mathcal{Q} Q , Z \mathcal{Z} Z , respectively.
For the subword c a b c a b cabcab c ab c ab of ı ⃗ \vec{\imath} , the simply-laced GLS formula gives, for a row of class P \mathcal{P} P , the two triples in columns κ ( k , 1 ) , κ ( k + 1 , 1 ) , κ ( k + 1 , 2 ) \kappa(k,1),\kappa(k+1,1),\kappa(k+1,2) κ ( k , 1 ) , κ ( k + 1 , 1 ) , κ ( k + 1 , 2 )
( − 1 , 1 , 0 ) , ( − 1 , 0 , 1 ) , (-1,1,0),\qquad(-1,0,1), ( − 1 , 1 , 0 ) , ( − 1 , 0 , 1 ) ,
corresponding to the rows indexed by κ ( r , 1 ) \kappa(r,1) κ ( r , 1 ) and κ ( r , 2 ) \kappa(r,2) κ ( r , 2 ) . For r ∈ Q r\in\mathcal{Q} r ∈ Q , the row indexed by κ ( r , 1 ) \kappa(r,1) κ ( r , 1 ) has the triple
For r ∈ Z r\in\mathcal{Z} r ∈ Z , every row indexed by κ ( r , s ) \kappa(r,s) κ ( r , s ) (1 ≤ s ≤ e i r 1\le s\le e_{i_r} 1 ≤ s ≤ e i r ) has the triple
Example 4.13
For the folding A 5 ⟶ C 3 A_5\longrightarrow C_3 A 5 ⟶ C 3 , consider the reduced sequence
ı ⃗ = ( 2 , 1 , 3 , 2 , 3 , 2 ) \vec{\imath}=(2,1,3,2,3,2) = ( 2 , 1 , 3 , 2 , 3 , 2 )
of type C 3 C_3 C 3 , whose 4-braid block ( 3 , 2 , 3 , 2 ) (3,2,3,2) ( 3 , 2 , 3 , 2 ) starts at k = 3 k=3 k = 3 . Its exchange matrix is
B i = ( 0 − 2 1 1 0 0 1 0 − 1 − 1 2 0 0 − 1 1 0 0 − 1 ) . B^i=
\begin{pmatrix}
0 & -2 & 1\\
1 & 0 & 0\\
1 & 0 & -1\\
-1 & 2 & 0\\
0 & -1 & 1\\
0 & 0 & -1
\end{pmatrix}. B i = 0 1 1 − 1 0 0 − 2 0 0 2 − 1 0 1 0 − 1 0 1 − 1 .
The columns of B i B^{i} B i are indexed by 1 , 3 , 4 1,3,4 1 , 3 , 4 , in this order. Thus, for the outside row r = 1 r=1 r = 1 ,
ρ d o w n ( 1 ) = ( − 2 , 1 ) , \rho_{\mathrm{down}}(1)=(-2,1), ρ down ( 1 ) = ( − 2 , 1 ) ,
so that r = 1 r=1 r = 1 belongs to the class P \mathcal{P} P .
Its lift is
ı ^ = ( 2 , 4 , 1 , 5 , 3 , 2 , 4 , 3 , 2 , 4 ) , \hat{\imath}=(2,4,1,5,3,2,4,3,2,4), ^ = ( 2 , 4 , 1 , 5 , 3 , 2 , 4 , 3 , 2 , 4 ) ,
with exchange matrix
B ı ^ = ( 0 0 − 1 1 0 0 0 − 1 0 1 1 0 0 0 0 0 1 0 0 0 1 1 0 − 1 − 1 − 1 0 1 0 0 0 − 1 1 0 0 0 0 − 1 1 1 0 0 0 − 1 0 0 0 0 0 − 1 ) . B^{\hat{\imath}}=
\begin{pmatrix}
0 & 0 & -1 & 1 & 0 \\
0 & 0 & -1 & 0 & 1 \\
1 & 0 & 0 & 0 & 0 \\
0 & 1 & 0 & 0 & 0 \\
1 & 1 & 0 & -1 & -1 \\
-1 & 0 & 1 & 0 & 0 \\
0 & -1 & 1 & 0 & 0 \\
0 & 0 & -1 & 1 & 1 \\
0 & 0 & 0 & -1 & 0 \\
0 & 0 & 0 & 0 & -1
\end{pmatrix}. B ^ = 0 0 1 0 1 − 1 0 0 0 0 0 0 0 1 1 0 − 1 0 0 0 − 1 − 1 0 0 0 1 1 − 1 0 0 1 0 0 0 − 1 0 0 1 − 1 0 0 1 0 0 − 1 0 0 1 0 − 1 .
The columns of B ^ ı \hat B^{\imath} B ^ are indexed by 1 , 2 , 5 , 6 , 7 1,2,5,6,7 1 , 2 , 5 , 6 , 7 , in this order. The rows corresponding to r = 1 r=1 r = 1 are indexed by 1 , 2 1,2 1 , 2 , and we consider columns 5 , 6 , 7 5,6,7 5 , 6 , 7 . The entries in these columns are
( − 1 , 1 , 0 ) , ( − 1 , 0 , 1 ) , (-1,1,0),\qquad(-1,0,1), ( − 1 , 1 , 0 ) , ( − 1 , 0 , 1 ) ,
as in (47) . Moreover,
b ^ 1 , 5 ı + b ^ 2 , 5 ı = − 2 , b ^ 1 , 6 ı + b ^ 2 , 6 ı = b ^ 1 , 7 ı + b ^ 2 , 7 ı = 1 , \hat b^{\imath}_{1,5}+\hat b^{\imath}_{2,5}=-2,\qquad
\hat b^{\imath}_{1,6}+\hat b^{\imath}_{2,6}=\hat b^{\imath}_{1,7}+\hat b^{\imath}_{2,7}=1, b ^ 1 , 5 + b ^ 2 , 5 = − 2 , b ^ 1 , 6 + b ^ 2 , 6 = b ^ 1 , 7 + b ^ 2 , 7 = 1 ,
which recovers
ρ d o w n ( 1 ) = ( − 2 , 1 ) \rho_{\mathrm{down}}(1)=(-2,1) ρ down ( 1 ) = ( − 2 , 1 )
under folding.
We introduce notation for the outside variables in the 4-move. Set
O : = { r ∉ { k , k + 1 , k + 2 , k + 3 } } , \mathcal{O}:=\{r\notin\{k,k+1,k+2,k+3\}\}, O := { r ∈ / { k , k + 1 , k + 2 , k + 3 }} ,
and
O P : = { r ∈ O ∣ ρ d o w n ( r ) = ( − 2 , 1 ) } , O Q : = { r ∈ O ∣ ρ d o w n ( r ) = ( 1 , 0 ) } . \mathcal{O}_{P}:=\{r\in\mathcal{O}\mid\rho_{\mathrm{down}}(r)=(-2,1)\},\qquad
\mathcal{O}_{Q}:=\{r\in\mathcal{O}\mid\rho_{\mathrm{down}}(r)=(1,0)\}. O P := { r ∈ O ∣ ρ down ( r ) = ( − 2 , 1 )} , O Q := { r ∈ O ∣ ρ down ( r ) = ( 1 , 0 )} .
Set
P : = ∏ r ∈ O P x r , Q : = ∏ r ∈ O Q x r , P:=\prod_{r\in\mathcal{O}_{P}}x_r,\qquad
Q:=\prod_{r\in\mathcal{O}_{Q}}x_r, P := r ∈ O P ∏ x r , Q := r ∈ O Q ∏ x r ,
where an empty product is understood to be 1.
By Lemma 4.12 , for r ∈ O P r\in\mathcal{O}_{P} r ∈ O P , let r ^ A , r ^ B \hat r_A,\hat r_B r ^ A , r ^ B be the indices κ ( r , 1 ) , κ ( r , 2 ) \kappa(r,1),\kappa(r,2) κ ( r , 1 ) , κ ( r , 2 ) , ordered so that the corresponding triples in (47) are
( − 1 , 1 , 0 ) , ( − 1 , 0 , 1 ) , (-1,1,0),\qquad(-1,0,1), ( − 1 , 1 , 0 ) , ( − 1 , 0 , 1 ) ,
respectively. Set
P ^ A : = ∏ r ∈ O P x ^ r ^ A , P ^ B : = ∏ r ∈ O P x ^ r ^ B . \widehat P_A:=\prod_{r\in\mathcal{O}_{P}}\hat x_{\hat r_A},\qquad
\widehat P_B:=\prod_{r\in\mathcal{O}_{P}}\hat x_{\hat r_B}. P A := r ∈ O P ∏ x ^ r ^ A , P B := r ∈ O P ∏ x ^ r ^ B .
Similarly, writing r ^ = κ ( r , 1 ) \widehat{r}=\kappa(r,1) r = κ ( r , 1 ) for r ∈ O Q r\in\mathcal{O}_Q r ∈ O Q , set
Q ^ : = ∏ r ∈ O Q x ^ r ^ . \widehat{Q}:=\prod_{r\in\mathcal{O}_Q}\widehat{x}_{\widehat{r}}. Q := r ∈ O Q ∏ x r .
Then
ϕ i ( P ^ A ) = ϕ i ( P ^ B ) = P , ϕ i ( Q ^ ) = Q . \phi^i(\widehat{P}_A)=\phi^i(\widehat{P}_B)=P,\qquad\phi^i(\widehat{Q})=Q. ϕ i ( P A ) = ϕ i ( P B ) = P , ϕ i ( Q ) = Q .
Lemma 4.14. In the setting of Lemma 4.12 , let D s ± D_s^\pm D s ± (s = 1 , 2 , 3 s=1,2,3 s = 1 , 2 , 3 ) be the factors from rows indexed by r ∈ O r\in\mathcal{O} r ∈ O in the successive downstairs exchange relations. Let U a ± U_a^\pm U a ± (a = 1 , … , 4 a=1,\ldots,4 a = 1 , … , 4 ) be those for the successive upstairs mutations, determined immediately before each mutation after all preceding mutations and permutations in the mutation–permutation sequence realizing (45) . Then
( D 1 + , D 1 − ) = ( Q , P 2 ) , ( D 2 + , D 2 − ) = ( 1 , P ) , ( D 3 + , D 3 − ) = ( 1 , Q ) , (D_1^+,D_1^-)=(Q,P^2),\qquad(D_2^+,D_2^-)=(1,P),\qquad(D_3^+,D_3^-)=(1,Q), ( D 1 + , D 1 − ) = ( Q , P 2 ) , ( D 2 + , D 2 − ) = ( 1 , P ) , ( D 3 + , D 3 − ) = ( 1 , Q ) , whereas
( U 1 + , U 1 − ) = ( P ^ A , 1 ) , ( U 2 + , U 2 − ) = ( Q ^ , P ^ B ) , (U_1^+,U_1^-)=(\widehat{P}_A,1),\qquad(U_2^+,U_2^-)=(\widehat{Q},\widehat{P}_B), ( U 1 + , U 1 − ) = ( P A , 1 ) , ( U 2 + , U 2 − ) = ( Q , P B ) , ( U 3 + , U 3 − ) = ( 1 , 1 ) , ( U 4 + , U 4 − ) = ( Q ^ , P ^ A ) . (U_3^+,U_3^-)=(1,1),\qquad(U_4^+,U_4^-)=(\widehat{Q},\widehat{P}_A). ( U 3 + , U 3 − ) = ( 1 , 1 ) , ( U 4 + , U 4 − ) = ( Q , P A ) .
Proof. For r ∈ O r\in\mathcal{O} r ∈ O , set
( α , β ) : = ( b r , k i , b r , k + 1 i ) , B ( 0 ) = B i , B ( 1 ) = μ k B ( 0 ) , B ( 2 ) = μ k + 1 B ( 1 ) , (\alpha,\beta):=(b^i_{r,k},b^i_{r,k+1}),\qquad B^{(0)}=B^i,\qquad B^{(1)}=\mu_kB^{(0)},\qquad B^{(2)}=\mu_{k+1}B^{(1)}, ( α , β ) := ( b r , k i , b r , k + 1 i ) , B ( 0 ) = B i , B ( 1 ) = μ k B ( 0 ) , B ( 2 ) = μ k + 1 B ( 1 ) ,
and
η 1 = b r , k ( 0 ) , η 2 = b r , k + 1 ( 1 ) , η 3 = b r , k ( 2 ) . \eta_1=b^{(0)}_{r,k},\qquad\eta_2=b^{(1)}_{r,k+1},\qquad\eta_3=b^{(2)}_{r,k}. η 1 = b r , k ( 0 ) , η 2 = b r , k + 1 ( 1 ) , η 3 = b r , k ( 2 ) .
Thus η a \eta_a η a is the entry in row r r r of the exchange column used at the a a a -th mutation. Since the final permutations occur only after the three mutations, they play no role here. By (21) ,
b k , k + 1 ( 0 ) = − 1 , b k + 1 , k ( 0 ) = 2 , b^{(0)}_{k,k+1}=-1,\qquad b^{(0)}_{k+1,k}=2, b k , k + 1 ( 0 ) = − 1 , b k + 1 , k ( 0 ) = 2 ,
and hence η 1 = α \eta_1=\alpha η 1 = α .
For the second mutation happening at k + 1 k+1 k + 1 ,
η 2 = b r , k + 1 ( 1 ) = β + [ α ] + [ b k , k + 1 ( 0 ) ] + + [ − α ] + [ − b k , k + 1 ( 0 ) ] + = β − [ − α ] + . \eta_2=b^{(1)}_{r,k+1}=\beta+[\alpha]_+[b^{(0)}_{k,k+1}]_++[-\alpha]_+[-b^{(0)}_{k,k+1}]_+=\beta-[-\alpha]_+. η 2 = b r , k + 1 ( 1 ) = β + [ α ] + [ b k , k + 1 ( 0 ) ] + + [ − α ] + [ − b k , k + 1 ( 0 ) ] + = β − [ − α ] + .
Note that we have b r , k ( 1 ) = − α b^{(1)}_{r,k}=-\alpha b r , k ( 1 ) = − α and b k + 1 , k ( 1 ) = − b k + 1 , k ( 0 ) = − 2 b^{(1)}_{k+1,k}=-b^{(0)}_{k+1,k}=-2 b k + 1 , k ( 1 ) = − b k + 1 , k ( 0 ) = − 2 . Therefore, for the third mutation happening at k k k ,
η 3 = b r , k ( 2 ) = − α + [ η 2 ] + [ b k + 1 , k ( 1 ) ] + + [ − η 2 ] + [ − b k + 1 , k ( 1 ) ] + = − α − 2 [ − η 2 ] + . \eta_3=b^{(2)}_{r,k}=-\alpha+[\eta_2]_+[b^{(1)}_{k+1,k}]_+ +[-\eta_2]_+[-b^{(1)}_{k+1,k}]_+=-\alpha-2[-\eta_2]_+. η 3 = b r , k ( 2 ) = − α + [ η 2 ] + [ b k + 1 , k ( 1 ) ] + + [ − η 2 ] + [ − b k + 1 , k ( 1 ) ] + = − α − 2 [ − η 2 ] + .
Thus Lemma 4.12 gives
η 1 η 2 η 3 P − 2 − 1 0 Q 1 0 − 1 Z 0 0 0 (48) \begin{array}{c|ccc}
& \eta_1 & \eta_2 & \eta_3\\
\mathcal{P} & -2 & -1 & 0\\
Q & 1 & 0 & -1\\
\mathcal{Z} & 0 & 0 & 0 \tag*{(48)}
\end{array} P Q Z η 1 − 2 1 0 η 2 − 1 0 0 η 3 0 − 1 0 (48)
and hence, reading the three mutation columns,
( D 1 + , D 1 − ) = ( Q , P 2 ) , ( D 2 + , D 2 − ) = ( 1 , P ) , ( D 3 + , D 3 − ) = ( 1 , Q ) . (D_1^+,D_1^-)=(Q,P^2),\qquad(D_2^+,D_2^-)=(1,P),\qquad(D_3^+,D_3^-)=(1,Q). ( D 1 + , D 1 − ) = ( Q , P 2 ) , ( D 2 + , D 2 − ) = ( 1 , P ) , ( D 3 + , D 3 − ) = ( 1 , Q ) .
We now consider the mutation–permutation sequence realizing (45) . Number the six positions of the subword c a b c a b cabcab c ab c ab by 1 , … , 6 1,\ldots,6 1 , … , 6 from left to right. With respect to these local positions, the four mutations occur at
( m 1 , m 2 , m 3 , m 4 ) = ( 3 , 1 , 4 , 2 ) . (m_1,m_2,m_3,m_4)=(3,1,4,2). ( m 1 , m 2 , m 3 , m 4 ) = ( 3 , 1 , 4 , 2 ) .
For r ∈ O r\in\mathcal{O} r ∈ O and 1 ≤ s ≤ e i r 1\le s\le e_{ir} 1 ≤ s ≤ e i r , put r s = κ ( r , s ) r_s=\kappa(r,s) r s = κ ( r , s ) and let
( α , β , γ ) (\alpha,\beta,\gamma) ( α , β , γ )
denote its entries in local columns 1 , 2 , 3 1,2,3 1 , 2 , 3 of the initial exchange matrix, as in (47) . Let B ^ [ a ] \widehat{B}^{[a]} B [ a ] be the exchange matrix immediately before the a a a -th mutation, after all preceding mutations and permutations, and let r s [ a ] r_s^{[a]} r s [ a ] be the current position of r s r_s r s . Set
η ^ a ( r s ) : = B ^ r s [ a ] , m a [ a ] . \widehat{\eta}_a(r_s):=\widehat{B}^{[a]}_{r_s^{[a]},m_a}. η a ( r s ) := B r s [ a ] , m a [ a ] .
The words immediately before the four mutations in (45) are
W 1 = ( c , b , a ‾ , c , a , b ) , W 2 = ( c ‾ , b , c , a , c , b ) , W 3 = ( b , c , a , b ‾ , c , b ) , W 4 = ( b , c ‾ , a , c , b , c ) . \begin{aligned}
W_1&=(c,b,\underline{a},c,a,b),& W_2&=(\underline{c},b,c,a,c,b),\\
W_3&=(b,c,a,\underline{b},c,b),& W_4&=(b,\underline{c},a,c,b,c).
\end{aligned} W 1 W 3 = ( c , b , a , c , a , b ) , = ( b , c , a , b , c , b ) , W 2 W 4 = ( c , b , c , a , c , b ) , = ( b , c , a , c , b , c ) .
Thus the mutation colors are ( a , c , b , c ) (a,c,b,c) ( a , c , b , c ) , respectively.
For the word W 1 W_1 W 1 , (21) gives
B ^ 3 , 1 [ 1 ] = 1 , B ^ 3 , 2 [ 1 ] = 0. \widehat{B}^{[1]}_{3,1}=1,\qquad\widehat{B}^{[1]}_{3,2}=0. B 3 , 1 [ 1 ] = 1 , B 3 , 2 [ 1 ] = 0.
Since the initial permutation gives
( α , β , γ ) ⟼ ( α , γ , β ) , (\alpha,\beta,\gamma)\longmapsto(\alpha,\gamma,\beta), ( α , β , γ ) ⟼ ( α , γ , β ) ,
and the mutation happens at the third position,
η ^ 1 ( r s ) = β . \widehat{\eta}_1(r_s)=\beta. η 1 ( r s ) = β .
Mutation at m 1 = 3 m_1=3 m 1 = 3 gives
B ^ r s , 1 ′ = α + [ β ] + [ 1 ] + − [ − β ] + [ − 1 ] + = α + [ β ] + ( ∵ B ^ 3 , 1 [ 1 ] = 1 ) , (49) \widehat{B}'_{r_s,1}=\alpha+[\beta]_+[1]_+-[-\beta]_+[-1]_+=\alpha+[\beta]_+
\qquad(\because\widehat{B}^{[1]}_{3,1}=1), \tag*{(49)} B r s , 1 ′ = α + [ β ] + [ 1 ] + − [ − β ] + [ − 1 ] + = α + [ β ] + ( ∵ B 3 , 1 [ 1 ] = 1 ) , (49)
B ^ r s , 2 ′ = γ + [ β ] + [ 0 ] + − [ − β ] + [ 0 ] + = γ ( ∵ B ^ 3 , 2 [ 1 ] = 0 ) , (50) \widehat{B}'_{r_s,2}=\gamma+[\beta]_+[0]_+-[-\beta]_+[0]_+=\gamma
\qquad(\because\widehat{B}^{[1]}_{3,2}=0), \tag*{(50)} B r s , 2 ′ = γ + [ β ] + [ 0 ] + − [ − β ] + [ 0 ] + = γ ( ∵ B 3 , 2 [ 1 ] = 0 ) , (50)
B ^ r s , 3 ′ = − β . \widehat{B}'_{r_s,3}=-\beta. B r s , 3 ′ = − β .
The following permutation does not affect positions 1 , 2 , 3 1,2,3 1 , 2 , 3 . Since the second mutation happens at the first position,
η ^ 2 ( r s ) = α + [ β ] + = α + [ η ^ 1 ( r s ) ] + . \widehat{\eta}_2(r_s)=\alpha+[\beta]_+=\alpha+[\widehat{\eta}_1(r_s)]_+. η 2 ( r s ) = α + [ β ] + = α + [ η 1 ( r s ) ] + .
At W 2 W_2 W 2 , (21) gives
B ^ 1 , 2 [ 2 ] = − 1 , B ^ 1 , 3 [ 2 ] = 1. \widehat{B}^{[2]}_{1,2}=-1,\qquad\widehat{B}^{[2]}_{1,3}=1. B 1 , 2 [ 2 ] = − 1 , B 1 , 3 [ 2 ] = 1.
The entry γ \gamma γ is still in position 2 2 2 by (49) , so after the second mutation it becomes
γ + [ η ^ 2 ( r s ) ] + [ − 1 ] + − [ − η ^ 2 ( r s ) ] + [ 1 ] + = γ − [ − η ^ 2 ( r s ) ] + . \gamma+[\widehat{\eta}_2(r_s)]_+[-1]_+-[-\widehat{\eta}_2(r_s)]_+[1]_+
=\gamma-[-\widehat{\eta}_2(r_s)]_+. γ + [ η 2 ( r s ) ] + [ − 1 ] + − [ − η 2 ( r s ) ] + [ 1 ] + = γ − [ − η 2 ( r s ) ] + .
After the following two permutations, the position of this entry changes as
2 ⟼ 3 ⟼ 4. 2\longmapsto3\longmapsto4. 2 ⟼ 3 ⟼ 4.
Since the third mutation happens at the fourth position,
η ^ 3 ( r s ) = γ − [ − η ^ 2 ( r s ) ] + . \widehat{\eta}_3(r_s)=\gamma-\left[-\widehat{\eta}_2(r_s)\right]_+. η 3 ( r s ) = γ − [ − η 2 ( r s ) ] + .
At W 2 W_2 W 2 , the entry originating from β \beta β has already changed to − β -\beta − β by (50) , and B ^ 1 , 3 [ 2 ] = 1 \widehat{B}^{[2]}_{1,3}=1 B 1 , 3 [ 2 ] = 1 . Thus, after the second mutation, it becomes
− β ⟼ − β + [ η ^ 2 ( r s ) ] + [ 1 ] + − [ − η ^ 2 ( r s ) ] + [ − 1 ] + = − β + [ η ^ 2 ( r s ) ] + ( ∵ B ^ 1 , 3 [ 2 ] = 1 ) . -\beta\longmapsto-\beta+\left[\widehat{\eta}_2(r_s)\right]_+[1]_+-\left[-\widehat{\eta}_2(r_s)\right]_+[-1]_+=-\beta+\left[\widehat{\eta}_2(r_s)\right]_+\quad\left(\because\widehat{B}^{[2]}_{1,3}=1\right). − β ⟼ − β + [ η 2 ( r s ) ] + [ 1 ] + − [ − η 2 ( r s ) ] + [ − 1 ] + = − β + [ η 2 ( r s ) ] + ( ∵ B 1 , 3 [ 2 ] = 1 ) .
The subsequent permutation moves this entry from position 3 to position 2, while the next one leaves position 2 fixed.
At W 3 W_3 W 3 , (21) gives
B ^ 4 , 2 [ 3 ] = 1. \widehat{B}^{[3]}_{4,2}=1. B 4 , 2 [ 3 ] = 1.
Thus the third mutation gives
− β + [ η ^ 2 ( r s ) ] + ⟼ − β + [ η ^ 2 ( r s ) ] + + [ η ^ 3 ( r s ) ] + [ 1 ] + − [ − η ^ 3 ( r s ) ] + [ − 1 ] + = − β + [ η ^ 2 ( r s ) ] + + [ η ^ 3 ( r s ) ] + . \begin{aligned}
-\beta+\left[\widehat{\eta}_2(r_s)\right]_+\longmapsto-\beta+\left[\widehat{\eta}_2(r_s)\right]_++\left[\widehat{\eta}_3(r_s)\right]_+[1]_+-\left[-\widehat{\eta}_3(r_s)\right]_+[-1]_+ \\
&=-\beta+\left[\widehat{\eta}_2(r_s)\right]_++\left[\widehat{\eta}_3(r_s)\right]_+.
\end{aligned} − β + [ η 2 ( r s ) ] + ⟼ − β + [ η 2 ( r s ) ] + + [ η 3 ( r s ) ] + [ 1 ] + − [ − η 3 ( r s ) ] + [ − 1 ] + = − β + [ η 2 ( r s ) ] + + [ η 3 ( r s ) ] + .
The following permutation leaves position 2 fixed. Since the last mutation is at m 4 = 2 m_4=2 m 4 = 2 ,
η ^ 4 ( r s ) = − β + [ η ^ 2 ( r s ) ] + + [ η ^ 3 ( r s ) ] + . \widehat{\eta}_4(r_s)=-\beta+\left[\widehat{\eta}_2(r_s)\right]_++\left[\widehat{\eta}_3(r_s)\right]_+. η 4 ( r s ) = − β + [ η 2 ( r s ) ] + + [ η 3 ( r s ) ] + .
Consequently,
η ^ 1 ( r s ) = β , η ^ 2 ( r s ) = α + [ η ^ 1 ( r s ) ] + , \widehat{\eta}_1(r_s)=\beta,\qquad\widehat{\eta}_2(r_s)=\alpha+\left[\widehat{\eta}_1(r_s)\right]_+, η 1 ( r s ) = β , η 2 ( r s ) = α + [ η 1 ( r s ) ] + ,
η ^ 3 ( r s ) = γ − [ − η ^ 2 ( r s ) ] + , η ^ 4 ( r s ) = − β + [ η ^ 2 ( r s ) ] + + [ η ^ 3 ( r s ) ] + . \widehat{\eta}_3(r_s)=\gamma-\left[-\widehat{\eta}_2(r_s)\right]_+,\qquad\widehat{\eta}_4(r_s)=-\beta+\left[\widehat{\eta}_2(r_s)\right]_++\left[\widehat{\eta}_3(r_s)\right]_+. η 3 ( r s ) = γ − [ − η 2 ( r s ) ] + , η 4 ( r s ) = − β + [ η 2 ( r s ) ] + + [ η 3 ( r s ) ] + .
For r ∈ P r\in\mathcal{P} r ∈ P , the rows indexed by κ ( r , 1 ) \kappa(r,1) κ ( r , 1 ) and κ ( r , 2 ) \kappa(r,2) κ ( r , 2 ) have initial triples
( − 1 , 1 , 0 ) , ( − 1 , 0 , 1 ) , (-1,1,0),\qquad(-1,0,1), ( − 1 , 1 , 0 ) , ( − 1 , 0 , 1 ) ,
and hence
( η ^ 1 , η ^ 2 , η ^ 3 , η ^ 4 ) = ( 1 , 0 , 0 , − 1 ) , ( 0 , − 1 , 0 , 0 ) . (\widehat{\eta}_1,\widehat{\eta}_2,\widehat{\eta}_3,\widehat{\eta}_4)=(1,0,0,-1),\qquad(0,-1,0,0). ( η 1 , η 2 , η 3 , η 4 ) = ( 1 , 0 , 0 , − 1 ) , ( 0 , − 1 , 0 , 0 ) .
For Q \mathcal{Q} Q ,
( α , β , γ ) = ( 1 , 0 , 0 ) , ( η ^ 1 , η ^ 2 , η ^ 3 , η ^ 4 ) = ( 0 , 1 , 0 , 1 ) . (\alpha,\beta,\gamma)=(1,0,0),\qquad(\widehat{\eta}_1,\widehat{\eta}_2,\widehat{\eta}_3,\widehat{\eta}_4)=(0,1,0,1). ( α , β , γ ) = ( 1 , 0 , 0 ) , ( η 1 , η 2 , η 3 , η 4 ) = ( 0 , 1 , 0 , 1 ) .
For every row indexed by κ ( r , s ) \kappa(r,s) κ ( r , s ) with r ∈ Z r\in\mathbb{Z} r ∈ Z and 1 ≤ s ≤ e i r 1\leq s\leq e_{ir} 1 ≤ s ≤ e i r ,
( α , β , γ ) = ( 0 , 0 , 0 ) , η ^ a = 0 ( 1 ≤ a ≤ 4 ) . (\alpha,\beta,\gamma)=(0,0,0),\qquad\widehat{\eta}_a=0\quad(1\leq a\leq4). ( α , β , γ ) = ( 0 , 0 , 0 ) , η a = 0 ( 1 ≤ a ≤ 4 ) .
Therefore
Table 2.
and hence
( U 1 + , U 1 − ) = ( P ^ A , 1 ) , ( U 2 + , U 2 − ) = ( Q ^ , P ^ B ) , (U_1^+,U_1^-)=(\widehat{P}_A,1),\qquad(U_2^+,U_2^-)=(\widehat{Q},\widehat{P}_B), ( U 1 + , U 1 − ) = ( P A , 1 ) , ( U 2 + , U 2 − ) = ( Q , P B ) ,
( U 3 + , U 3 − ) = ( 1 , 1 ) , ( U 4 + , U 4 − ) = ( Q ^ , P ^ A ) . (U_3^+,U_3^-)=(1,1),\qquad(U_4^+,U_4^-)=(\widehat{Q},\widehat{P}_A). ( U 3 + , U 3 − ) = ( 1 , 1 ) , ( U 4 + , U 4 − ) = ( Q , P A ) .
Hence the assertion follows. ∎
Proposition 4.15. Suppose that i ′ i' i ′ is obtained from i i i by a 4-move
i j i j ⟶ j i j i with e i = 1 and e j = 2. ijij \longrightarrow jiji \quad\text{with } e_i=1 \text{ and } e_j=2. ij ij ⟶ j ij i with e i = 1 and e j = 2. Then (43) holds.
Proof. Write
x 1 = x k , x 2 = x k + 1 , x 3 = x k + 2 , x 4 = x k + 3 . x_1=x_k,\qquad x_2=x_{k+1},\qquad x_3=x_{k+2},\qquad x_4=x_{k+3}. x 1 = x k , x 2 = x k + 1 , x 3 = x k + 2 , x 4 = x k + 3 .
By Lemma 4.14 ,
( D 1 + , D 1 − ) = ( Q , P 2 ) , ( D 2 + , D 2 − ) = ( 1 , P ) , ( D 3 + , D 3 − ) = ( 1 , Q ) . (D_1^+,D_1^-)=(Q,P^2),\qquad(D_2^+,D_2^-)=(1,P),\qquad(D_3^+,D_3^-)=(1,Q). ( D 1 + , D 1 − ) = ( Q , P 2 ) , ( D 2 + , D 2 − ) = ( 1 , P ) , ( D 3 + , D 3 − ) = ( 1 , Q ) .
Hence the three downstairs exchange relations are
x 1 x 1 ( 1 ) = Q x 2 2 + P 2 x 3 , x 2 x 2 ( 1 ) = x 1 ( 1 ) + P x 4 , x 1 ( 1 ) x 1 ( 2 ) = ( x 2 ( 1 ) ) 2 x 3 + Q x 2 4 . x_1x_1^{(1)}=Qx_2^2+P^2x_3,\qquad x_2x_2^{(1)}=x_1^{(1)}+Px_4,\qquad x_1^{(1)}x_1^{(2)}=(x_2^{(1)})^2x_3+Qx_2^4. x 1 x 1 ( 1 ) = Q x 2 2 + P 2 x 3 , x 2 x 2 ( 1 ) = x 1 ( 1 ) + P x 4 , x 1 ( 1 ) x 1 ( 2 ) = ( x 2 ( 1 ) ) 2 x 3 + Q x 2 4 .
Set
T : = Q x 2 2 + P 2 x 3 , W : = T + P x 1 x 4 . T:=Qx_2^2+P^2x_3,\qquad W:=T+Px_1x_4. T := Q x 2 2 + P 2 x 3 , W := T + P x 1 x 4 .
Then
x 1 ( 1 ) = T x 1 , x 2 ( 1 ) = W x 1 x 2 , x 1 ( 2 ) = x 3 W 2 + Q x 1 2 x 2 2 x 4 2 x 1 x 2 2 T . x_1^{(1)}=\frac{T}{x_1},\qquad x_2^{(1)}=\frac{W}{x_1x_2},\qquad x_1^{(2)}=\frac{x_3W^2+Qx_1^2x_2^2x_4^2}{x_1x_2^2T}. x 1 ( 1 ) = x 1 T , x 2 ( 1 ) = x 1 x 2 W , x 1 ( 2 ) = x 1 x 2 2 T x 3 W 2 + Q x 1 2 x 2 2 x 4 2 .
After the final permutations,
x k ′ = W x 1 x 2 , x k + 1 ′ = x 3 W 2 + Q x 1 2 x 2 2 x 4 2 x 1 x 2 2 T , x k + 2 ′ = x 4 , x k + 3 ′ = x 3 . x'_k=\frac{W}{x_1x_2},\qquad x'_{k+1}=\frac{x_3W^2+Qx_1^2x_2^2x_4^2}{x_1x_2^2T},\qquad x'_{k+2}=x_4,\qquad x'_{k+3}=x_3. x k ′ = x 1 x 2 W , x k + 1 ′ = x 1 x 2 2 T x 3 W 2 + Q x 1 2 x 2 2 x 4 2 , x k + 2 ′ = x 4 , x k + 3 ′ = x 3 .
For the lifted braid block ( c , a , b , c , a , b ) (c,a,b,c,a,b) ( c , a , b , c , a , b ) , write ( x ^ 1 , … , x ^ 6 ) (\widehat{x}_1,\ldots,\widehat{x}_6) ( x 1 , … , x 6 ) for the corresponding cluster variables. Then
ϕ i ( x ^ 1 ) = x 1 , ϕ i ( x ^ 2 ) = ϕ i ( x ^ 3 ) = x 2 , ϕ i ( x ^ 4 ) = x 3 , ϕ i ( x ^ 5 ) = ϕ i ( x ^ 6 ) = x 4 . \phi^i(\widehat{x}_1)=x_1,\qquad\phi^i(\widehat{x}_2)=\phi^i(\widehat{x}_3)=x_2,\qquad\phi^i(\widehat{x}_4)=x_3,\qquad\phi^i(\widehat{x}_5)=\phi^i(\widehat{x}_6)=x_4. ϕ i ( x 1 ) = x 1 , ϕ i ( x 2 ) = ϕ i ( x 3 ) = x 2 , ϕ i ( x 4 ) = x 3 , ϕ i ( x 5 ) = ϕ i ( x 6 ) = x 4 .
The lifted path in (45) ,
( ( 2 ; κ ( k , 1 ) + 1 ) , ( 3 ; κ ( k , 1 ) + 2 ) , ( 3 ; κ ( k , 1 ) ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 3 ; κ ( k , 1 ) + 3 ) , ( 3 ; κ ( k , 1 ) + 1 ) , ( 2 ; κ ( k , 1 ) ) ) , ((2;\kappa(k,1)+1),(3;\kappa(k,1)+2),(3;\kappa(k,1)),(2;\kappa(k,1)+2),
(3;\kappa(k,1)+3),(3;\kappa(k,1)+1),(2;\kappa(k,1))), (( 2 ; κ ( k , 1 ) + 1 ) , ( 3 ; κ ( k , 1 ) + 2 ) , ( 3 ; κ ( k , 1 )) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 3 ; κ ( k , 1 ) + 3 ) , ( 3 ; κ ( k , 1 ) + 1 ) , ( 2 ; κ ( k , 1 ))) ,
has a mutation–permutation realization with four mutations at the current local positions 3 , 1 , 4 , 2 3,1,4,2 3 , 1 , 4 , 2 . Using Lemma 4.14 , set
R : = U 1 + x ^ 4 + U 1 − x ^ 1 x ^ 5 , C : = U 2 + x ^ 2 x ^ 3 + U 2 − R , E : = U 3 + x ^ 4 C + U 3 − x ^ 1 x ^ 6 R . R:=U_1^+\widehat{x}_4+U_1^-\widehat{x}_1\widehat{x}_5,\qquad C:=U_2^+\widehat{x}_2\widehat{x}_3+U_2^-R,\qquad E:=U_3^+\widehat{x}_4C+U_3^-\widehat{x}_1\widehat{x}_6R. R := U 1 + x 4 + U 1 − x 1 x 5 , C := U 2 + x 2 x 3 + U 2 − R , E := U 3 + x 4 C + U 3 − x 1 x 6 R .
Eliminating the intermediate mutated variables in this realization gives
x ^ 1 ′ = U 4 − E + U 4 + x ^ 1 x ^ 2 x ^ 3 x ^ 5 x ^ 1 x ^ 3 R , x ^ 2 ′ = C x ^ 1 x ^ 2 , x ^ 3 ′ = E x ^ 1 x ^ 2 x ^ 3 , \widehat{x}'_1=\frac{U_4^-E+U_4^+\widehat{x}_1\widehat{x}_2\widehat{x}_3\widehat{x}_5}{\widehat{x}_1\widehat{x}_3R},\qquad\widehat{x}'_2=\frac{C}{\widehat{x}_1\widehat{x}_2},\qquad\widehat{x}'_3=\frac{E}{\widehat{x}_1\widehat{x}_2\widehat{x}_3}, x 1 ′ = x 1 x 3 R U 4 − E + U 4 + x 1 x 2 x 3 x 5 , x 2 ′ = x 1 x 2 C , x 3 ′ = x 1 x 2 x 3 E ,
and
( x ^ 4 ′ , x ^ 5 ′ , x ^ 6 ′ ) = ( x ^ 5 , x ^ 6 , x ^ 4 ) . (\widehat{x}'_4,\widehat{x}'_5,\widehat{x}'_6)=(\widehat{x}_5,\widehat{x}_6,\widehat{x}_4). ( x 4 ′ , x 5 ′ , x 6 ′ ) = ( x 5 , x 6 , x 4 ) .
Again by Lemma 4.14 and
ϕ i ( P ^ A ) = ϕ i ( P ^ B ) = P , ϕ i ( Q ^ ) = Q , \phi^i(\widehat{P}_A)=\phi^i(\widehat{P}_B)=P,\qquad\phi^i(\widehat{Q})=Q, ϕ i ( P A ) = ϕ i ( P B ) = P , ϕ i ( Q ) = Q ,
we have
ϕ i ( U 1 + , U 1 − ) = ( P , 1 ) , ϕ i ( U 2 + , U 2 − ) = ( Q , P ) , ϕ i ( U 3 + , U 3 − ) = ( 1 , 1 ) , ϕ i ( U 4 + , U 4 − ) = ( Q , P ) . \phi^i(U_1^+,U_1^-)=(P,1),\qquad\phi^i(U_2^+,U_2^-)=(Q,P),\qquad\phi^i(U_3^+,U_3^-)=(1,1),\qquad\phi^i(U_4^+,U_4^-)=(Q,P). ϕ i ( U 1 + , U 1 − ) = ( P , 1 ) , ϕ i ( U 2 + , U 2 − ) = ( Q , P ) , ϕ i ( U 3 + , U 3 − ) = ( 1 , 1 ) , ϕ i ( U 4 + , U 4 − ) = ( Q , P ) .
Consequently
ϕ i ( R ) = P x 3 + x 1 x 4 , ϕ i ( C ) = Q x 2 2 + P ϕ i ( R ) = W , ϕ i ( E ) = x 3 W + x 1 x 4 ϕ i ( R ) . \phi^{i}(R) = Px_{3} + x_{1}x_{4}, \qquad\phi^{i}(C) = Qx_{2}^{2} + P\phi^{i}(R) = W, \qquad\phi^{i}(E) = x_{3}W + x_{1}x_{4}\phi^{i}(R). ϕ i ( R ) = P x 3 + x 1 x 4 , ϕ i ( C ) = Q x 2 2 + P ϕ i ( R ) = W , ϕ i ( E ) = x 3 W + x 1 x 4 ϕ i ( R ) .
Moreover,
P ϕ i ( E ) + Q x 1 x 2 2 x 4 = W ϕ i ( R ) , T ϕ i ( E ) = x 3 W 2 + Q x 1 2 x 2 2 x 4 2 , P\phi^{i}(E) + Qx_{1}x_{2}^{2}x_{4} = W\phi^{i}(R), \qquad T\phi^{i}(E) = x_{3}W^{2} + Qx_{1}^{2}x_{2}^{2}x_{4}^{2}, P ϕ i ( E ) + Q x 1 x 2 2 x 4 = W ϕ i ( R ) , T ϕ i ( E ) = x 3 W 2 + Q x 1 2 x 2 2 x 4 2 ,
where the first follows from W = Q x 2 2 + P ϕ i ( R ) W = Qx_{2}^{2} + P\phi^{i}(R) W = Q x 2 2 + P ϕ i ( R ) and the second from T = Q x 2 2 + P 2 x 3 T = Qx_{2}^{2} + P^{2}x_{3} T = Q x 2 2 + P 2 x 3 and W = T + P x 1 x 4 W = T + Px_{1}x_{4} W = T + P x 1 x 4 .
Therefore
ϕ i ( x ~ 1 ) = P ϕ i ( E ) + Q x 1 x 2 2 x 4 x 1 x 2 ϕ i ( R ) = W x 1 x 2 = x k ′ = ϕ i ( x ~ 2 ) . \phi^{i}(\widetilde{x}_{1}) = \frac{P\phi^{i}(E) + Qx_{1}x_{2}^{2}x_{4}}{x_{1}x_{2}\phi^{i}(R)} = \frac{W}{x_{1}x_{2}} = x_{k}^{\prime} = \phi^{i}(\widetilde{x}_{2}). ϕ i ( x 1 ) = x 1 x 2 ϕ i ( R ) P ϕ i ( E ) + Q x 1 x 2 2 x 4 = x 1 x 2 W = x k ′ = ϕ i ( x 2 ) .
and
ϕ i ( x ~ 3 ) = ϕ i ( E ) x 1 x 2 2 = x 3 W 2 + Q x 1 2 x 2 2 x 4 2 x 1 x 2 2 T = x k + 1 ′ . \phi^{i}(\widetilde{x}_{3}) = \frac{\phi^{i}(E)}{x_{1}x_{2}^{2}} = \frac{x_{3}W^{2} + Qx_{1}^{2}x_{2}^{2}x_{4}^{2}}{x_{1}x_{2}^{2}T} = x_{k+1}^{\prime}. ϕ i ( x 3 ) = x 1 x 2 2 ϕ i ( E ) = x 1 x 2 2 T x 3 W 2 + Q x 1 2 x 2 2 x 4 2 = x k + 1 ′ .
Finally,
ϕ i ( x ~ 4 ) = ϕ i ( x ~ 5 ) = x 4 = x k + 2 ′ , ϕ i ( x ~ 6 ) = x 3 = x k + 3 ′ . \phi^{i}(\widetilde{x}_{4}) = \phi^{i}(\widetilde{x}_{5}) = x_{4} = x_{k+2}^{\prime}, \qquad\phi^{i}(\widetilde{x}_{6}) = x_{3} = x_{k+3}^{\prime}. ϕ i ( x 4 ) = ϕ i ( x 5 ) = x 4 = x k + 2 ′ , ϕ i ( x 6 ) = x 3 = x k + 3 ′ .
Thus the final variables at positions 1 , … , 6 1,\ldots,6 1 , … , 6 satisfy
ϕ i ( x ~ 1 ) = ϕ i ( x ~ 2 ) = x k ′ , ϕ i ( x ~ 3 ) = x k + 1 ′ , ϕ i ( x ~ 4 ) = ϕ i ( x ~ 5 ) = x k + 2 ′ , ϕ i ( x ~ 6 ) = x k + 3 ′ . \phi^{i}(\widetilde{x}_{1}) = \phi^{i}(\widetilde{x}_{2}) = x_{k}^{\prime}, \qquad\phi^{i}(\widetilde{x}_{3}) = x_{k+1}^{\prime}, \qquad\phi^{i}(\widetilde{x}_{4}) = \phi^{i}(\widetilde{x}_{5}) = x_{k+2}^{\prime}, \qquad\phi^{i}(\widetilde{x}_{6}) = x_{k+3}^{\prime}. ϕ i ( x 1 ) = ϕ i ( x 2 ) = x k ′ , ϕ i ( x 3 ) = x k + 1 ′ , ϕ i ( x 4 ) = ϕ i ( x 5 ) = x k + 2 ′ , ϕ i ( x 6 ) = x k + 3 ′ .
Hence (43) holds. □
We next consider the case e i = 2 e_i = 2 e i = 2 and e j = 1 e_j = 1 e j = 1 . This case occurs in the folding D n + 1 ⟶ B n D_{n+1} \longrightarrow B_n D n + 1 ⟶ B n at the multiple edge; in this case, i = n i = n i = n and j = n − 1 j = n - 1 j = n − 1 . Write I i = ( a , b ) I_i = (a,b) I i = ( a , b ) and I j = ( c ) I_j = (c) I j = ( c ) , where a ∼ c ∼ b a \sim c \sim b a ∼ c ∼ b and a ≁ b a \not\sim b a ∼ b . Thus the braid move
i j i j ⟶ j i j i ijij \longrightarrow jiji ij ij ⟶ j ij i
lifts to
a b c a b c ⟶ c a b c a b . abcabc \longrightarrow cabcab. ab c ab c ⟶ c ab c ab .
For the calculation below, we choose the following sequence of 2- and 3-moves:
a b c a b c ⟶ b a c a b c ⟶ b c a c b c ⟶ b c a b c b ⟶ b c b a c b ⟶ c b c a c b ⟶ c b a c a b ⟶ c a b c a b . (51) \begin{aligned}
abcabc &\longrightarrow bacabc \longrightarrow bcacbc \longrightarrow bcabcb\\
&\longrightarrow bcbacb \longrightarrow cbcacb \longrightarrow cbacab \longrightarrow cabcab. \tag*{(51)}
\end{aligned} ab c ab c ⟶ ba c ab c ⟶ b c a c b c ⟶ b c ab c b ⟶ b c ba c b ⟶ c b c a c b ⟶ c ba c ab ⟶ c ab c ab . (51)
More precisely, the successive moves are
( 2 ; κ ( k , 1 ) ) , ( 3 ; κ ( k , 1 ) + 1 ) , ( 3 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , (2;\kappa(k,1)), \qquad(3;\kappa(k,1)+1), \qquad(3;\kappa(k,1)+3), \qquad(2;\kappa(k,1)+2), ( 2 ; κ ( k , 1 )) , ( 3 ; κ ( k , 1 ) + 1 ) , ( 3 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) ,
( 3 ; κ ( k , 1 ) ) , ( 3 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 1 ) . (3;\kappa(k,1)), \qquad(3;\kappa(k,1)+2), \qquad(2;\kappa(k,1)+1). ( 3 ; κ ( k , 1 )) , ( 3 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 1 ) .
Thus the four 3-moves occur at positions
κ ( k , 1 ) + 1 , κ ( k , 1 ) + 3 , κ ( k , 1 ) , κ ( k , 1 ) + 2. \kappa(k,1)+1, \qquad\kappa(k,1)+3, \qquad\kappa(k,1), \qquad\kappa(k,1)+2. κ ( k , 1 ) + 1 , κ ( k , 1 ) + 3 , κ ( k , 1 ) , κ ( k , 1 ) + 2.
The following calculation is entirely parallel to the case e i = 1 e_i = 1 e i = 1 and e j = 2 e_j = 2 e j = 2 above, so we record only the data and omit the proof.
Lemma 4.16. For r ∉ { k , k + 1 , k + 2 , k + 3 } r \notin\{k,k+1,k+2,k+3\} r ∈ / { k , k + 1 , k + 2 , k + 3 } , the entries in columns k , k + 1 k,k+1 k , k + 1 satisfy
( b r , k i , b r , k + 1 i ) = { ( − 1 , 1 ) , r + = k + 1 , ( 1 , 0 ) , r + = k , ( 0 , 0 ) , otherwise . (52) \left(b^{i}_{r,k},b^{i}_{r,k+1}\right) =
\begin{cases}
(-1,1), & r_{+}=k+1,\\
(1,0), & r_{+}=k,\\
(0,0), & \text{otherwise}.
\end{cases} \tag*{(52)} ( b r , k i , b r , k + 1 i ) = ⎩ ⎨ ⎧ ( − 1 , 1 ) , ( 1 , 0 ) , ( 0 , 0 ) , r + = k + 1 , r + = k , otherwise . (52) In particular,
ρ d o w n ( r ) : = ( b r , k i , b r , k + 1 i ) ∈ { ( − 1 , 1 ) , ( 1 , 0 ) , ( 0 , 0 ) } , \rho_{\mathrm{down}}(r) := \left(b^{i}_{r,k},b^{i}_{r,k+1}\right) \in\{(-1,1),(1,0),(0,0)\}, ρ down ( r ) := ( b r , k i , b r , k + 1 i ) ∈ {( − 1 , 1 ) , ( 1 , 0 ) , ( 0 , 0 )} ,
defining the classes N , P , Z \mathcal{N},\mathcal{P},\mathcal{Z} N , P , Z , respectively. The entries of the rows of B ^ i \widehat{B}^{i} B i indexed by κ ( r , s ) \kappa(r,s) κ ( r , s ) (1 ≤ s ≤ e i r 1 \le s \le e_{i_r} 1 ≤ s ≤ e i r ) in columns κ ( k , 1 ) , κ ( k , 2 ) , κ ( k + 1 , 1 ) \kappa(k,1),\kappa(k,2),\kappa(k+1,1) κ ( k , 1 ) , κ ( k , 2 ) , κ ( k + 1 , 1 ) , in this order, are
N P Z ρ d o w n ( − 1 , 1 ) ( 1 , 0 ) ( 0 , 0 ) ρ u p ( − 1 , − 1 , 1 ) ( 1 , 0 , 0 ) , ( 0 , 1 , 0 ) ( 0 , 0 , 0 ) (53) \begin{array}{c|ccc}
& \mathcal{N} & \mathcal{P} & \mathcal{Z} \\
\rho_{\mathrm{down}} & (-1,1) & (1,0) & (0,0) \\
\rho_{\mathrm{up}} & (-1,-1,1) & (1,0,0),\ (0,1,0) & (0,0,0) \tag*{(53)}
\end{array} ρ down ρ up N ( − 1 , 1 ) ( − 1 , − 1 , 1 ) P ( 1 , 0 ) ( 1 , 0 , 0 ) , ( 0 , 1 , 0 ) Z ( 0 , 0 ) ( 0 , 0 , 0 ) (53)
Here N \mathcal{N} N occurs only when i r = j i_r=j i r = j , and hence e j = 1 e_j=1 e j = 1 , whereas P \mathcal{P} P occurs only when i r = i i_r=i i r = i , and hence e i = 2 e_i=2 e i = 2 .
We introduce notation for the outside variables in the 4-move. Set
O : = { r ∉ { k , k + 1 , k + 2 , k + 3 } } , \mathcal{O}:=\{r\notin\{k,k+1,k+2,k+3\}\}, O := { r ∈ / { k , k + 1 , k + 2 , k + 3 }} ,
and
O N : = { r ∈ O ∣ ρ d o w n ( r ) = ( − 1 , 1 ) } , O P : = { r ∈ O ∣ ρ d o w n ( r ) = ( 1 , 0 ) } . \mathcal{O}_{\mathcal{N}}:=\{r\in\mathcal{O}\mid\rho_{\mathrm{down}}(r)=(-1,1)\},\qquad
\mathcal{O}_{\mathcal{P}}:=\{r\in\mathcal{O}\mid\rho_{\mathrm{down}}(r)=(1,0)\}. O N := { r ∈ O ∣ ρ down ( r ) = ( − 1 , 1 )} , O P := { r ∈ O ∣ ρ down ( r ) = ( 1 , 0 )} .
Set
N : = ∏ r ∈ O N x r , P : = ∏ r ∈ O P x r , N:=\prod_{r\in\mathcal{O}_{\mathcal{N}}}x_r,\qquad
P:=\prod_{r\in\mathcal{O}_{\mathcal{P}}}x_r, N := r ∈ O N ∏ x r , P := r ∈ O P ∏ x r ,
where an empty product is understood to be 1.
By Lemma 4.16 , for r ∈ O N r\in\mathcal{O}_{\mathcal{N}} r ∈ O N , put r ^ = κ ( r , 1 ) \widehat{r}=\kappa(r,1) r = κ ( r , 1 ) and set
N ^ : = ∏ r ∈ O N x r ^ . \widehat{N}:=\prod_{r\in\mathcal{O}_{\mathcal{N}}}x_{\widehat{r}}. N := r ∈ O N ∏ x r .
For r ∈ O P r\in\mathcal{O}_{\mathcal{P}} r ∈ O P , let r ^ A , r ^ B \widehat{r}_A,\widehat{r}_B r A , r B be the indices κ ( r , 1 ) , κ ( r , 2 ) \kappa(r,1),\kappa(r,2) κ ( r , 1 ) , κ ( r , 2 ) , ordered so that the corresponding triples in (53) are
( 1 , 0 , 0 ) , ( 0 , 1 , 0 ) , (1,0,0),\qquad(0,1,0), ( 1 , 0 , 0 ) , ( 0 , 1 , 0 ) ,
respectively, and set
P ^ A : = ∏ r ∈ O P x r ^ A , P ^ B : = ∏ r ∈ O P x r ^ B . \widehat{P}_A:=\prod_{r\in\mathcal{O}_{\mathcal{P}}}x_{\widehat{r}_A},\qquad
\widehat{P}_B:=\prod_{r\in\mathcal{O}_{\mathcal{P}}}x_{\widehat{r}_B}. P A := r ∈ O P ∏ x r A , P B := r ∈ O P ∏ x r B .
Then
ϕ i ( N ^ ) = N , ϕ i ( P ^ A ) = ϕ i ( P ^ B ) = P . \phi^i(\widehat{N})=N,\qquad\phi^i(\widehat{P}_A)=\phi^i(\widehat{P}_B)=P. ϕ i ( N ) = N , ϕ i ( P A ) = ϕ i ( P B ) = P .
The verification of the next coefficient identities is again parallel to that of Lemma 4.14 , so we omit the calculation.
Lemma 4.17. Let D s ± D_s^{\pm} D s ± (s = 1 , 2 , 3 s=1,2,3 s = 1 , 2 , 3 ) be the factors from rows indexed by r ∈ O r\in\mathcal{O} r ∈ O in the successive downstairs exchange relations. Let U a ± U_a^{\pm} U a ± (a = 1 , … , 4 a=1,\ldots,4 a = 1 , … , 4 ) be those for the successive upstairs mutations, determined immediately before each mutation after all preceding mutations and permutations in the mutation–permutation sequence realizing (51) . Then
( D 1 + , D 1 − ) = ( P , N ) , ( D 2 + , D 2 − ) = ( 1 , N ) , ( D 3 + , D 3 − ) = ( 1 , P ) , (D_1^+,D_1^-)=(P,N),\qquad(D_2^+,D_2^-)=(1,N),\qquad(D_3^+,D_3^-)=(1,P), ( D 1 + , D 1 − ) = ( P , N ) , ( D 2 + , D 2 − ) = ( 1 , N ) , ( D 3 + , D 3 − ) = ( 1 , P ) , whereas
( U 1 + , U 1 − ) = ( P ^ A , N ^ ) , ( U 2 + , U 2 − ) = ( 1 , 1 ) , (U_1^+,U_1^-)=(\widehat{P}_A,\widehat{N}),\qquad(U_2^+,U_2^-)=(1,1), ( U 1 + , U 1 − ) = ( P A , N ) , ( U 2 + , U 2 − ) = ( 1 , 1 ) , ( U 3 + , U 3 − ) = ( P ^ B , N ^ ) , ( U 4 + , U 4 − ) = ( 1 , P ^ A ) . (U_3^+,U_3^-)=(\widehat{P}_B,\widehat{N}),\qquad(U_4^+,U_4^-)=(1,\widehat{P}_A). ( U 3 + , U 3 − ) = ( P B , N ) , ( U 4 + , U 4 − ) = ( 1 , P A ) .
Proposition 4.18. Suppose that i ′ i' i ′ is obtained from i i i by a 4-move i j i j ⟶ j i j i ijij \longrightarrow jiji ij ij ⟶ j ij i with e i = 2 e_i=2 e i = 2 and e j = 1 e_j=1 e j = 1 . Then (43) holds.
Proof. Write x 1 = x k x_1=x_k x 1 = x k , x 2 = x k + 1 x_2=x_{k+1} x 2 = x k + 1 , x 3 = x k + 2 x_3=x_{k+2} x 3 = x k + 2 and x 4 = x k + 3 x_4=x_{k+3} x 4 = x k + 3 . By Lemma 4.17 ,
( D 1 + , D 1 − ) = ( P , N ) , ( D 2 + , D 2 − ) = ( 1 , N ) , ( D 3 + , D 3 − ) = ( 1 , P ) . (D_1^+,D_1^-)=(P,N),\qquad(D_2^+,D_2^-)=(1,N),\qquad(D_3^+,D_3^-)=(1,P). ( D 1 + , D 1 − ) = ( P , N ) , ( D 2 + , D 2 − ) = ( 1 , N ) , ( D 3 + , D 3 − ) = ( 1 , P ) .
Set
T : = P x 2 + N x 3 , W : = T 2 + N x 1 2 x 4 . T:=Px_2+Nx_3,\qquad W:=T^2+Nx_1^2x_4. T := P x 2 + N x 3 , W := T 2 + N x 1 2 x 4 .
Composing the three downstairs mutations and the final permutations gives
x k ′ = W x 1 2 x 2 , x k + 1 ′ = x 3 W + P x 1 2 x 2 x 4 x 1 x 2 T , x k + 2 ′ = x 4 , x k + 3 ′ = x 3 . x'_k=\frac{W}{x_1^2x_2},\qquad x'_{k+1}=\frac{x_3W+Px_1^2x_2x_4}{x_1x_2T},\qquad x'_{k+2}=x_4,\qquad x'_{k+3}=x_3. x k ′ = x 1 2 x 2 W , x k + 1 ′ = x 1 x 2 T x 3 W + P x 1 2 x 2 x 4 , x k + 2 ′ = x 4 , x k + 3 ′ = x 3 .
For the lifted braid block ( a , b , c , a , b , c ) (a,b,c,a,b,c) ( a , b , c , a , b , c ) , write ( x ^ 1 , … , x ^ 6 ) (\widehat{x}_1,\ldots,\widehat{x}_6) ( x 1 , … , x 6 ) for the corresponding cluster variables. Then
ϕ i ( x ^ 1 ) = ϕ i ( x ^ 2 ) = x 1 , ϕ i ( x ^ 3 ) = x 2 , ϕ i ( x ^ 4 ) = ϕ i ( x ^ 5 ) = x 3 , ϕ i ( x ^ 6 ) = x 4 . \phi^i(\widehat{x}_1)=\phi^i(\widehat{x}_2)=x_1,\qquad\phi^i(\widehat{x}_3)=x_2,\qquad\phi^i(\widehat{x}_4)=\phi^i(\widehat{x}_5)=x_3,\qquad\phi^i(\widehat{x}_6)=x_4. ϕ i ( x 1 ) = ϕ i ( x 2 ) = x 1 , ϕ i ( x 3 ) = x 2 , ϕ i ( x 4 ) = ϕ i ( x 5 ) = x 3 , ϕ i ( x 6 ) = x 4 .
The mutation–permutation realization of the lifted path in (51) has four mutations at the current local positions 2 , 4 , 1 , 3 2,4,1,3 2 , 4 , 1 , 3 . Set
R : = U 1 + x ^ 3 + U 1 − x ^ 4 , C : = U 2 + x ^ 5 R + U 2 − x ^ 1 x ^ 2 x ^ 6 , E : = U 3 − C + U 3 + x ^ 3 R . R:=U_1^+\widehat{x}_3+U_1^-\widehat{x}_4,\qquad C:=U_2^+\widehat{x}_5R+U_2^-\widehat{x}_1\widehat{x}_2\widehat{x}_6,\qquad E:=U_3^-C+U_3^+\widehat{x}_3R. R := U 1 + x 3 + U 1 − x 4 , C := U 2 + x 5 R + U 2 − x 1 x 2 x 6 , E := U 3 − C + U 3 + x 3 R .
Eliminating the intermediate mutated variables in this realization gives
x ^ 1 ′ = E x ^ 1 x ^ 2 x ^ 3 , x ^ 2 ′ = U 4 + x ^ 4 E + U 4 − x ^ 1 x ^ 2 x ^ 3 x ^ 6 x ^ 2 x ^ 3 R , x ^ 3 ′ = C x ^ 1 x ^ 3 , \widehat{x}'_1=\frac{E}{\widehat{x}_1\widehat{x}_2\widehat{x}_3},\qquad\widehat{x}'_2=\frac{U_4^+\widehat{x}_4E+U_4^-\widehat{x}_1\widehat{x}_2\widehat{x}_3\widehat{x}_6}{\widehat{x}_2\widehat{x}_3R},\qquad\widehat{x}'_3=\frac{C}{\widehat{x}_1\widehat{x}_3}, x 1 ′ = x 1 x 2 x 3 E , x 2 ′ = x 2 x 3 R U 4 + x 4 E + U 4 − x 1 x 2 x 3 x 6 , x 3 ′ = x 1 x 3 C ,
and
( x ^ 4 ′ , x ^ 5 ′ , x ^ 6 ′ ) = ( x ^ 6 , x ^ 4 , x ^ 5 ) . (\widehat{x}'_4,\widehat{x}'_5,\widehat{x}'_6)=(\widehat{x}_6,\widehat{x}_4,\widehat{x}_5). ( x 4 ′ , x 5 ′ , x 6 ′ ) = ( x 6 , x 4 , x 5 ) .
By Lemma 4.17 and
ϕ i ( N ^ ) = N , ϕ i ( P ^ A ) = ϕ i ( P ^ B ) = P , \phi^i(\widehat{N})=N,\qquad\phi^i(\widehat{P}_A)=\phi^i(\widehat{P}_B)=P, ϕ i ( N ) = N , ϕ i ( P A ) = ϕ i ( P B ) = P ,
we have
ϕ i ( U 1 + , U 1 − ) = ( P , N ) , ϕ i ( U 2 + , U 2 − ) = ( 1 , 1 ) , \phi^i(U_1^+,U_1^-)=(P,N),\qquad\phi^i(U_2^+,U_2^-)=(1,1), ϕ i ( U 1 + , U 1 − ) = ( P , N ) , ϕ i ( U 2 + , U 2 − ) = ( 1 , 1 ) ,
ϕ i ( U 3 + , U 3 − ) = ( P , N ) , ϕ i ( U 4 + , U 4 − ) = ( 1 , P ) . \phi^i(U_3^+,U_3^-)=(P,N),\qquad\phi^i(U_4^+,U_4^-)=(1,P). ϕ i ( U 3 + , U 3 − ) = ( P , N ) , ϕ i ( U 4 + , U 4 − ) = ( 1 , P ) .
Hence
ϕ i ( R ) = T , ϕ i ( C ) = x 3 T + x 1 2 x 4 , \phi^i(R)=T,\qquad\phi^i(C)=x_3T+x_1^2x_4, ϕ i ( R ) = T , ϕ i ( C ) = x 3 T + x 1 2 x 4 ,
and
ϕ i ( E ) = N ϕ i ( C ) + P x 2 T = T 2 + N x 1 2 x 4 = W . \phi^i(E)=N\phi^i(C)+Px_2T=T^2+Nx_1^2x_4=W. ϕ i ( E ) = N ϕ i ( C ) + P x 2 T = T 2 + N x 1 2 x 4 = W .
Moreover,
T ϕ i ( C ) = x 3 W + P x 1 2 x 2 x 4 . T\phi^i(C)=x_3W+Px_1^2x_2x_4. T ϕ i ( C ) = x 3 W + P x 1 2 x 2 x 4 .
Indeed, this follows directly from
ϕ i ( C ) = x 3 T + x 1 2 x 4 , T = P x 2 + N x 3 , W = T 2 + N x 1 2 x 4 . \phi^i(C)=x_3T+x_1^2x_4,\qquad T=Px_2+Nx_3,\qquad W=T^2+Nx_1^2x_4. ϕ i ( C ) = x 3 T + x 1 2 x 4 , T = P x 2 + N x 3 , W = T 2 + N x 1 2 x 4 .
Therefore
ϕ i ( x ^ 1 ′ ) = W x 1 2 x 2 = x k ′ , while ϕ i ( x ^ 2 ′ ) = ϕ i ( x ^ 3 ′ ) = x 3 W + P x 1 2 x 2 x 4 x 1 x 2 T = x k + 1 ′ . \phi^i(\widehat{x}'_1)=\frac{W}{x_1^2x_2}=x'_k,\qquad\text{while}\qquad\phi^i(\widehat{x}'_2)=\phi^i(\widehat{x}'_3)=\frac{x_3W+Px_1^2x_2x_4}{x_1x_2T}=x'_{k+1}. ϕ i ( x 1 ′ ) = x 1 2 x 2 W = x k ′ , while ϕ i ( x 2 ′ ) = ϕ i ( x 3 ′ ) = x 1 x 2 T x 3 W + P x 1 2 x 2 x 4 = x k + 1 ′ .
Finally,
ϕ i ( x ^ 4 ) = x 4 = x k + 2 ′ , ϕ i ( x ^ 5 ) = ϕ i ( x ^ 6 ) = x 3 = x k + 3 ′ . \phi^{i}(\widehat{x}_{4}) = x_{4} = x'_{k+2}, \qquad\phi^{i}(\widehat{x}_{5}) = \phi^{i}(\widehat{x}_{6}) = x_{3} = x'_{k+3}. ϕ i ( x 4 ) = x 4 = x k + 2 ′ , ϕ i ( x 5 ) = ϕ i ( x 6 ) = x 3 = x k + 3 ′ .
Thus the final variables at the positions
( 1 ) , ( 2 , 3 ) , ( 4 ) , ( 5 , 6 ) (1), \qquad(2,3), \qquad(4), \qquad(5,6) ( 1 ) , ( 2 , 3 ) , ( 4 ) , ( 5 , 6 )
of the subword ( c , a , b , c , a , b ) (c,a,b,c,a,b) ( c , a , b , c , a , b ) of i ^ \widehat{i} i satisfy (43) .
The remaining affine 4-move configuration
{ e i , e j } = { 2 , 4 } , \{e_i,e_j\} = \{2,4\}, { e i , e j } = { 2 , 4 } ,
arising from the folding
D 2 n + 2 ( 1 ) ⟶ A 2 n ( 2 ) , D^{(1)}_{2n+2} \longrightarrow A^{(2)}_{2n}, D 2 n + 2 ( 1 ) ⟶ A 2 n ( 2 ) ,
is recorded separately in Appendix A. There we give the corresponding lifted path and the identities for the final cluster variables yielding (43) .
D 4 ⟶ G 2 . D_4 \longrightarrow G_2. D 4 ⟶ G 2 .
With our fixed ordering, write
I i = ( a , b , c ) , I j = ( d ) , I_i = (a,b,c), \qquad I_j = (d), I i = ( a , b , c ) , I j = ( d ) ,
where a , b , c a,b,c a , b , c are the outer vertices of D 4 D_4 D 4 and d d d is the central vertex. Thus
e i = 3 , e j = 1. e_i = 3, \qquad e_j = 1. e i = 3 , e j = 1.
For finite and affine admissible foldings, the subdiagram on I i ⊔ I j I_i \sqcup I_j I i ⊔ I j associated with a 6-move is of type D 4 D_4 D 4 , with orbit sizes ( 3 , 1 ) (3,1) ( 3 , 1 ) . Thus the braid move
i j i j i j ⟶ j i j i j i ijijij \longrightarrow jijiji ij ij ij ⟶ j ij ij i
lifts to
( a , b , c , d , a , b , c , d , a , b , c , d ) ⟶ ( d , a , b , c , d , a , b , c , d , a , b , c ) . (a,b,c,d,a,b,c,d,a,b,c,d) \longrightarrow(d,a,b,c,d,a,b,c,d,a,b,c). ( a , b , c , d , a , b , c , d , a , b , c , d ) ⟶ ( d , a , b , c , d , a , b , c , d , a , b , c ) .
For the calculation below, we choose the following sequence of 2- and 3-moves:
( ( 2 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 2 ; κ ( k , 1 ) + 4 ) , ( 3 ; κ ( k , 1 ) + 2 ) , ( 3 ; κ ( k , 1 ) + 6 ) , ( 3 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 3 ; κ ( k , 1 ) + 1 ) , ( 3 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 3 ; κ ( k , 1 ) + 7 ) , ( 3 ; κ ( k , 1 ) + 9 ) , ( 2 ; κ ( k , 1 ) + 8 ) , ( 3 ; κ ( k , 1 ) + 6 ) , ( 3 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 3 ; κ ( k , 1 ) + 8 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 3 ; κ ( k , 1 ) + 4 ) , ( 3 ; κ ( k , 1 ) ) , ( 3 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 7 ) , ( 2 ; κ ( k , 1 ) + 4 ) , ( 3 ; κ ( k , 1 ) + 5 ) , ( 3 ; κ ( k , 1 ) + 7 ) , ( 3 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 2 ; κ ( k , 1 ) + 5 ) ) . (54) \begin{aligned}
(& (2;\kappa(k,1)+4),(2;\kappa(k,1)+5),(2;\kappa(k,1)+4),(3;\kappa(k,1)+2),\\
& (3;\kappa(k,1)+6),(3;\kappa(k,1)+4),(2;\kappa(k,1)+3),(2;\kappa(k,1)+6),\\
& (3;\kappa(k,1)+1),(3;\kappa(k,1)+3),(2;\kappa(k,1)+5),(3;\kappa(k,1)+7),\\
& (3;\kappa(k,1)+9),(2;\kappa(k,1)+8),(3;\kappa(k,1)+6),(3;\kappa(k,1)+4),\\
& (2;\kappa(k,1)+3),(2;\kappa(k,1)+2),(3;\kappa(k,1)+8),(2;\kappa(k,1)+6),\\
& (3;\kappa(k,1)+4),(3;\kappa(k,1)),(3;\kappa(k,1)+2),(2;\kappa(k,1)+7),\\
& (2;\kappa(k,1)+4),(3;\kappa(k,1)+5),(3;\kappa(k,1)+7),(3;\kappa(k,1)+3),\\
& (2;\kappa(k,1)+5),(2;\kappa(k,1)+6),(2;\kappa(k,1)+5)). \tag*{(54)}
\end{aligned} ( ( 2 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 2 ; κ ( k , 1 ) + 4 ) , ( 3 ; κ ( k , 1 ) + 2 ) , ( 3 ; κ ( k , 1 ) + 6 ) , ( 3 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 3 ; κ ( k , 1 ) + 1 ) , ( 3 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 3 ; κ ( k , 1 ) + 7 ) , ( 3 ; κ ( k , 1 ) + 9 ) , ( 2 ; κ ( k , 1 ) + 8 ) , ( 3 ; κ ( k , 1 ) + 6 ) , ( 3 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 3 ; κ ( k , 1 ) + 8 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 3 ; κ ( k , 1 ) + 4 ) , ( 3 ; κ ( k , 1 )) , ( 3 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 7 ) , ( 2 ; κ ( k , 1 ) + 4 ) , ( 3 ; κ ( k , 1 ) + 5 ) , ( 3 ; κ ( k , 1 ) + 7 ) , ( 3 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 2 ; κ ( k , 1 ) + 5 )) . (54)
The sixteen 3-moves occur at the current positions
( κ ( k , 1 ) + 2 , κ ( k , 1 ) + 6 , κ ( k , 1 ) + 4 , κ ( k , 1 ) + 1 , κ ( k , 1 ) + 3 , κ ( k , 1 ) + 7 , κ ( k , 1 ) + 9 , κ ( k , 1 ) + 6 , κ ( k , 1 ) + 4 , κ ( k , 1 ) + 8 , κ ( k , 1 ) + 4 , κ ( k , 1 ) , κ ( k , 1 ) + 2 , κ ( k , 1 ) + 5 , κ ( k , 1 ) + 7 , κ ( k , 1 ) + 3 ) . \begin{aligned}
(&\kappa(k,1)+2,\kappa(k,1)+6,\kappa(k,1)+4,\kappa(k,1)+1,\kappa(k,1)+3,\\
&\kappa(k,1)+7,\kappa(k,1)+9,\kappa(k,1)+6,\kappa(k,1)+4,\kappa(k,1)+8,\\
&\kappa(k,1)+4,\kappa(k,1),\kappa(k,1)+2,\kappa(k,1)+5,\kappa(k,1)+7,\kappa(k,1)+3).
\end{aligned} ( κ ( k , 1 ) + 2 , κ ( k , 1 ) + 6 , κ ( k , 1 ) + 4 , κ ( k , 1 ) + 1 , κ ( k , 1 ) + 3 , κ ( k , 1 ) + 7 , κ ( k , 1 ) + 9 , κ ( k , 1 ) + 6 , κ ( k , 1 ) + 4 , κ ( k , 1 ) + 8 , κ ( k , 1 ) + 4 , κ ( k , 1 ) , κ ( k , 1 ) + 2 , κ ( k , 1 ) + 5 , κ ( k , 1 ) + 7 , κ ( k , 1 ) + 3 ) .
Downstairs, the 6-move is realized by ten mutations at the current positions ( k , k + 1 , k + 2 , k , k + 3 , k + 1 , k , k + 2 , k + 3 , k ) (k, k + 1, k + 2, k, k + 3, k + 1, k, k + 2, k + 3, k) ( k , k + 1 , k + 2 , k , k + 3 , k + 1 , k , k + 2 , k + 3 , k ) , followed by the prescribed permutations at positions ( k + 4 , k + 2 , k ) (k + 4, k + 2, k) ( k + 4 , k + 2 , k ) .
The following outside-row calculation is obtained by the same analysis as in the 4-move cases above, so we record the data without proof.
Lemma 4.19. For r ∉ { k , … , k + 5 } r \notin\{k,\ldots,k+5\} r ∈ / { k , … , k + 5 } , the entries in columns k , … , k + 3 k,\ldots,k+3 k , … , k + 3 satisfy
( b r , k + t i ) 0 ≤ t ≤ 3 = { ( − 1 , 1 , 0 , 0 ) , r + = k + 1 , ( 1 , 0 , 0 , 0 ) , r + = k , ( 0 , 0 , 0 , 0 ) , otherwise . (55) \left(b^{i}_{r,k+t}\right)_{0\leq t\leq3} =
\begin{cases}
(-1,1,0,0), & r_{+}=k+1,\\
(1,0,0,0), & r_{+}=k,\\
(0,0,0,0), & \text{otherwise}.
\end{cases} \tag*{(55)} ( b r , k + t i ) 0 ≤ t ≤ 3 = ⎩ ⎨ ⎧ ( − 1 , 1 , 0 , 0 ) , ( 1 , 0 , 0 , 0 ) , ( 0 , 0 , 0 , 0 ) , r + = k + 1 , r + = k , otherwise . (55) The first two row patterns in (55) define the classes N \mathcal{N} N and P \mathcal{P} P , respectively. The entries of the rows of B i B^{i} B i indexed by κ ( r , s ) \kappa(r,s) κ ( r , s ) (1 ≤ s ≤ e i r 1\leq s\leq e_{i_r} 1 ≤ s ≤ e i r ) in columns κ ( k , 1 ) , … , κ ( k , 1 ) + 7 \kappa(k,1),\ldots,\kappa(k,1)+7 κ ( k , 1 ) , … , κ ( k , 1 ) + 7 are
N : ( − 1 , − 1 , − 1 , 1 , 0 , 0 , 0 , 0 ) , P A : ( 1 , 0 , 0 , 0 , 0 , 0 , 0 , 0 ) , P B : ( 0 , 1 , 0 , 0 , 0 , 0 , 0 , 0 ) , P C : ( 0 , 0 , 1 , 0 , 0 , 0 , 0 , 0 ) . (56) \begin{aligned}
\mathcal{N}:&\ (-1,-1,-1,1,0,0,0,0),\\
\mathcal{P}_{A}:&\ (1,0,0,0,0,0,0,0),\\
\mathcal{P}_{B}:&\ (0,1,0,0,0,0,0,0),\\
\mathcal{P}_{C}:&\ (0,0,1,0,0,0,0,0). \tag*{(56)}
\end{aligned} N : P A : P B : P C : ( − 1 , − 1 , − 1 , 1 , 0 , 0 , 0 , 0 ) , ( 1 , 0 , 0 , 0 , 0 , 0 , 0 , 0 ) , ( 0 , 1 , 0 , 0 , 0 , 0 , 0 , 0 ) , ( 0 , 0 , 1 , 0 , 0 , 0 , 0 , 0 ) . (56) Here N \mathcal{N} N occurs only when i r = j i_r=j i r = j , and hence e i r = e j = 1 e_{i_r}=e_j=1 e i r = e j = 1 , whereas P \mathcal{P} P occurs only when i r = i i_r=i i r = i , and hence e i r = e i = 3 e_{i_r}=e_i=3 e i r = e i = 3 . The zero row pattern in (55) defines the class Z \mathcal{Z} Z . Rows r ∈ Z r\in\mathcal{Z} r ∈ Z and rows indexed by κ ( r , s ) \kappa(r,s) κ ( r , s ) (1 ≤ s ≤ e i r 1\leq s\leq e_{i_r} 1 ≤ s ≤ e i r ) have zero entries in the exchange column at every mutation.
Set O : = { r ∉ { k , … , k + 5 } } \mathcal{O}:=\{r\notin\{k,\ldots,k+5\}\} O := { r ∈ / { k , … , k + 5 }} , O N : = { r ∈ O ∣ r is of class N } \mathcal{O}_{\mathcal{N}}:=\{r\in\mathcal{O}\mid r\text{ is of class }\mathcal{N}\} O N := { r ∈ O ∣ r is of class N } , O P : = { r ∈ O ∣ r is of class P } \mathcal{O}_{\mathcal{P}}:=\{r\in\mathcal{O}\mid r\text{ is of class }\mathcal{P}\} O P := { r ∈ O ∣ r is of class P } , and
N : = ∏ r ∈ O N x r , P : = ∏ r ∈ O P x r , N:=\prod_{r\in\mathcal{O}_{\mathcal{N}}}x_r,\qquad
P:=\prod_{r\in\mathcal{O}_{\mathcal{P}}}x_r, N := r ∈ O N ∏ x r , P := r ∈ O P ∏ x r ,
where an empty product is understood to be 1.
By Lemma 4.19 , for r ∈ O N r\in\mathcal{O}_{\mathcal{N}} r ∈ O N , put r ^ = κ ( r , 1 ) \widehat{r}=\kappa(r,1) r = κ ( r , 1 ) and set
N ^ : = ∏ r ∈ O N x r ^ . \widehat{N}:=\prod_{r\in\mathcal{O}_{\mathcal{N}}}x_{\widehat{r}}. N := r ∈ O N ∏ x r .
For r ∈ O P r\in\mathcal{O}_{\mathcal{P}} r ∈ O P , let r ^ A , r ^ B , r ^ C \widehat{r}_{A},\widehat{r}_{B},\widehat{r}_{C} r A , r B , r C be the indices κ ( r , 1 ) , κ ( r , 2 ) , κ ( r , 3 ) \kappa(r,1),\kappa(r,2),\kappa(r,3) κ ( r , 1 ) , κ ( r , 2 ) , κ ( r , 3 ) , ordered as in (56) , and set
P ^ A : = ∏ r ∈ O P x r ^ A , P ^ B : = ∏ r ∈ O P x r ^ B , P ^ C : = ∏ r ∈ O P x r ^ C . \widehat{P}_{A}:=\prod_{r\in\mathcal{O}_{\mathcal{P}}}x_{\widehat{r}_{A}},\qquad
\widehat{P}_{B}:=\prod_{r\in\mathcal{O}_{\mathcal{P}}}x_{\widehat{r}_{B}},\qquad
\widehat{P}_{C}:=\prod_{r\in\mathcal{O}_{\mathcal{P}}}x_{\widehat{r}_{C}}. P A := r ∈ O P ∏ x r A , P B := r ∈ O P ∏ x r B , P C := r ∈ O P ∏ x r C .
Then
ϕ i ( N ^ ) = N , ϕ i ( P ^ A ) = ϕ i ( P ^ B ) = ϕ i ( P ^ C ) = P . \phi^{i}(\widehat{N})=N,\qquad
\phi^{i}(\widehat{P}_{A})=\phi^{i}(\widehat{P}_{B})=\phi^{i}(\widehat{P}_{C})=P. ϕ i ( N ) = N , ϕ i ( P A ) = ϕ i ( P B ) = ϕ i ( P C ) = P .
The verification of the following coefficient identities is a longer instance of the same exchange-column calculation used for the 4-move above, and can be checked directly from (54) . We therefore omit the proof.
Lemma 4.20. Let D s ± D_s^{\pm} D s ± (1 ≤ s ≤ 10 1 \le s \le10 1 ≤ s ≤ 10 ) be the factors from rows indexed by r ∈ O r \in O r ∈ O in the successive downstairs exchange relations. Let U a ± U_a^{\pm} U a ± (1 ≤ a ≤ 16 1 \le a \le16 1 ≤ a ≤ 16 ) be those for the successive upstairs mutations, determined immediately before each mutation after all preceding mutations and permutations in the mutation–permutation sequence realizing (54) . Then
( D 1 + , D 1 − ) = ( P , N ) , ( D 2 + , D 2 − ) = ( 1 , N 2 ) , ( D 3 + , D 3 − ) = ( P , 1 ) , ( D 4 + , D 4 − ) = ( 1 , N ) , ( D 5 + , D 5 − ) = ( 1 , 1 ) , ( D 6 + , D 6 − ) = ( 1 , N ) , ( D 7 + , D 7 − ) = ( 1 , 1 ) , ( D 8 + , D 8 − ) = ( 1 , P ) , ( D 9 + , D 9 − ) = ( 1 , 1 ) , ( D 10 + , D 10 − ) = ( 1 , P ) . \begin{aligned}
(D_1^+,D_1^-) &= (P,N), & (D_2^+,D_2^-) &= (1,N^2),\\
(D_3^+,D_3^-) &= (P,1), & (D_4^+,D_4^-) &= (1,N),\\
(D_5^+,D_5^-) &= (1,1), & (D_6^+,D_6^-) &= (1,N),\\
(D_7^+,D_7^-) &= (1,1), & (D_8^+,D_8^-) &= (1,P),\\
(D_9^+,D_9^-) &= (1,1), & (D_{10}^+,D_{10}^-) &= (1,P).
\end{aligned} ( D 1 + , D 1 − ) ( D 3 + , D 3 − ) ( D 5 + , D 5 − ) ( D 7 + , D 7 − ) ( D 9 + , D 9 − ) = ( P , N ) , = ( P , 1 ) , = ( 1 , 1 ) , = ( 1 , 1 ) , = ( 1 , 1 ) , ( D 2 + , D 2 − ) ( D 4 + , D 4 − ) ( D 6 + , D 6 − ) ( D 8 + , D 8 − ) ( D 10 + , D 10 − ) = ( 1 , N 2 ) , = ( 1 , N ) , = ( 1 , N ) , = ( 1 , P ) , = ( 1 , P ) . The upstairs pairs are
( U 1 + , U 1 − ) = ( P ^ C , N ^ ) , ( U 2 + , U 2 − ) = ( 1 , 1 ) , ( U 3 + , U 3 − ) = ( 1 , 1 ) , ( U 4 + , U 4 − ) = ( P ^ B , N ^ ) , ( U 5 + , U 5 − ) = ( 1 , P ^ C ) , ( U 6 + , U 6 − ) = ( 1 , 1 ) , ( U 7 + , U 7 − ) = ( 1 , 1 ) , ( U 8 + , U 8 − ) = ( 1 , 1 ) , ( U 9 + , U 9 − ) = ( 1 , 1 ) , ( U 10 + , U 10 − ) = ( 1 , 1 ) , ( U 11 + , U 11 − ) = ( P ^ C , 1 ) , ( U 12 + , U 12 − ) = ( P ^ A , N ^ ) , ( U 13 + , U 13 − ) = ( 1 , P ^ B ) , ( U 14 + , U 14 − ) = ( 1 , 1 ) , ( U 15 + , U 15 − ) = ( 1 , 1 ) , ( U 16 + , U 16 − ) = ( 1 , P ^ C ) . \begin{aligned}
(U_1^+,U_1^-) &= (\widehat{P}_C,\widehat{N}), & (U_2^+,U_2^-) &= (1,1), & (U_3^+,U_3^-) &= (1,1), & (U_4^+,U_4^-) &= (\widehat{P}_B,\widehat{N}),\\
(U_5^+,U_5^-) &= (1,\widehat{P}_C), & (U_6^+,U_6^-) &= (1,1), & (U_7^+,U_7^-) &= (1,1), & (U_8^+,U_8^-) &= (1,1),\\
(U_9^+,U_9^-) &= (1,1), & (U_{10}^+,U_{10}^-) &= (1,1), & (U_{11}^+,U_{11}^-) &= (\widehat{P}_C,1), & (U_{12}^+,U_{12}^-) &= (\widehat{P}_A,\widehat{N}),\\
(U_{13}^+,U_{13}^-) &= (1,\widehat{P}_B), & (U_{14}^+,U_{14}^-) &= (1,1), & (U_{15}^+,U_{15}^-) &= (1,1), & (U_{16}^+,U_{16}^-) &= (1,\widehat{P}_C).
\end{aligned} ( U 1 + , U 1 − ) ( U 5 + , U 5 − ) ( U 9 + , U 9 − ) ( U 13 + , U 13 − ) = ( P C , N ) , = ( 1 , P C ) , = ( 1 , 1 ) , = ( 1 , P B ) , ( U 2 + , U 2 − ) ( U 6 + , U 6 − ) ( U 10 + , U 10 − ) ( U 14 + , U 14 − ) = ( 1 , 1 ) , = ( 1 , 1 ) , = ( 1 , 1 ) , = ( 1 , 1 ) , ( U 3 + , U 3 − ) ( U 7 + , U 7 − ) ( U 11 + , U 11 − ) ( U 15 + , U 15 − ) = ( 1 , 1 ) , = ( 1 , 1 ) , = ( P C , 1 ) , = ( 1 , 1 ) , ( U 4 + , U 4 − ) ( U 8 + , U 8 − ) ( U 12 + , U 12 − ) ( U 16 + , U 16 − ) = ( P B , N ) , = ( 1 , 1 ) , = ( P A , N ) , = ( 1 , P C ) . For later use, we describe the composite transformations recursively. Write ( x 1 , … , x 6 ) (x_1,\ldots,x_6) ( x 1 , … , x 6 ) for the initial cluster variables at positions k , … , k + 5 k,\ldots,k+5 k , … , k + 5 of i i i , and ( x ^ 1 , … , x ^ 12 ) (\widehat{x}_1,\ldots,\widehat{x}_{12}) ( x 1 , … , x 12 ) for those at the corresponding positions of i ^ \widehat{i} i , in order. At the s s s -th downstairs mutation, let p s p_s p s be the current mutation position, B q , p s ( s ) B^{(s)}_{q,p_s} B q , p s ( s ) the actual exchange column restricted to q ∈ { k , … , k + 5 } q \in\{k,\ldots,k+5\} q ∈ { k , … , k + 5 } , and V ( s − 1 ) = ( V q ( s − 1 ) ) \mathbf{V}^{(s-1)}=(V_q^{(s-1)}) V ( s − 1 ) = ( V q ( s − 1 ) ) the current cluster restricted to these six positions. Set
E s : = D s + ∏ q ( V q ( s − 1 ) ) [ B q , p s ( s ) ] + + D s − ∏ q ( V q ( s − 1 ) ) [ − B q , p s ( s ) ] + , Z s : = E s V p s ( s − 1 ) . E_s := D_s^+ \prod_q \left(V_q^{(s-1)}\right)^{[B_{q,p_s}^{(s)}]_+} + D_s^- \prod_q \left(V_q^{(s-1)}\right)^{[-B_{q,p_s}^{(s)}]_+}, \qquad Z_s := \frac{E_s}{V_{p_s}^{(s-1)}}. E s := D s + q ∏ ( V q ( s − 1 ) ) [ B q , p s ( s ) ] + + D s − q ∏ ( V q ( s − 1 ) ) [ − B q , p s ( s ) ] + , Z s := V p s ( s − 1 ) E s . Thus V p s ( s ) = Z s V_{p_s}^{(s)}=Z_s V p s ( s ) = Z s , with the other coordinates unchanged. Recursively substituting the earlier Z Z Z 's into the later exchange numerators expresses the composite transformation solely in the initial variables and the factors D s ± D_s^{\pm} D s ± .
After the ten mutations and final permutations, the six downstairs cluster variables are
( Z 6 , Z 10 , Z 9 , Z 8 , x 6 , x 5 ) . (Z_6,Z_{10},Z_9,Z_8,x_6,x_5). ( Z 6 , Z 10 , Z 9 , Z 8 , x 6 , x 5 ) . Define Z ^ a \widehat{Z}_a Z a (1 ≤ a ≤ 16 1 \le a \le16 1 ≤ a ≤ 16 ) analogously upstairs, using the actual exchange columns and U a ± U_a^{\pm} U a ± after all preceding mutations and permutations. After the final permutations, the twelve upstairs cluster variables are
( Z ^ 12 , Z ^ 9 , Z ^ 13 , Z ^ 16 , Z ^ 14 , Z ^ 15 , Z ^ 10 , Z ^ 7 , x ^ 12 , x ^ 9 , x ^ 10 , x ^ 11 ) . (\widehat{Z}_{12},\widehat{Z}_9,\widehat{Z}_{13},\widehat{Z}_{16},\widehat{Z}_{14},\widehat{Z}_{15},\widehat{Z}_{10},\widehat{Z}_7,\widehat{x}_{12},\widehat{x}_9,\widehat{x}_{10},\widehat{x}_{11}). ( Z 12 , Z 9 , Z 13 , Z 16 , Z 14 , Z 15 , Z 10 , Z 7 , x 12 , x 9 , x 10 , x 11 ) . Proposition 4.21. Suppose that i ^ \widehat{i} i is obtained from i i i by a 6-move
i j i j i j ⟶ j i j i j i ijijij \longrightarrow jijiji ij ij ij ⟶ j ij ij i with e i = 3 e_i=3 e i = 3 and e j = 1 e_j=1 e j = 1 . Then (43) holds.
Proof. The lifted braid block
( a , b , c , d , a , b , c , d , a , b , c , d ) (a,b,c,d,a,b,c,d,a,b,c,d) ( a , b , c , d , a , b , c , d , a , b , c , d )
has fibers
( 1 , 2 , 3 ) , ( 4 ) , ( 5 , 6 , 7 ) , ( 8 ) , ( 9 , 10 , 11 ) , ( 12 ) . (1,2,3),\quad(4),\quad(5,6,7),\quad(8),\quad(9,10,11),\quad(12). ( 1 , 2 , 3 ) , ( 4 ) , ( 5 , 6 , 7 ) , ( 8 ) , ( 9 , 10 , 11 ) , ( 12 ) .
Hence
ϕ i ( x ^ 1 ) = ϕ i ( x ^ 2 ) = ϕ i ( x ^ 3 ) = x 1 , ϕ i ( x ^ 4 ) = x 2 , ϕ i ( x ^ 5 ) = ϕ i ( x ^ 6 ) = ϕ i ( x ^ 7 ) = x 3 , ϕ i ( x ^ 8 ) = x 4 , \phi^{i}(\widehat{x}_{1})=\phi^{i}(\widehat{x}_{2})=\phi^{i}(\widehat{x}_{3})=x_{1},\qquad\phi^{i}(\widehat{x}_{4})=x_{2},\qquad\phi^{i}(\widehat{x}_{5})=\phi^{i}(\widehat{x}_{6})=\phi^{i}(\widehat{x}_{7})=x_{3},\qquad\phi^{i}(\widehat{x}_{8})=x_{4}, ϕ i ( x 1 ) = ϕ i ( x 2 ) = ϕ i ( x 3 ) = x 1 , ϕ i ( x 4 ) = x 2 , ϕ i ( x 5 ) = ϕ i ( x 6 ) = ϕ i ( x 7 ) = x 3 , ϕ i ( x 8 ) = x 4 ,
ϕ i ( x ^ 9 ) = ϕ i ( x ^ 10 ) = ϕ i ( x ^ 11 ) = x 5 , ϕ i ( x ^ 12 ) = x 6 , \phi^{i}(\widehat{x}_{9})=\phi^{i}(\widehat{x}_{10})=\phi^{i}(\widehat{x}_{11})=x_{5},\qquad\phi^{i}(\widehat{x}_{12})=x_{6}, ϕ i ( x 9 ) = ϕ i ( x 10 ) = ϕ i ( x 11 ) = x 5 , ϕ i ( x 12 ) = x 6 ,
together with
ϕ i ( N ^ ) = N , ϕ i ( P ^ A ) = ϕ i ( P ^ B ) = ϕ i ( P ^ C ) = P . \phi^{i}(\widehat{N})=N,\qquad\phi^{i}(\widehat{P}_{A})=\phi^{i}(\widehat{P}_{B})=\phi^{i}(\widehat{P}_{C})=P. ϕ i ( N ) = N , ϕ i ( P A ) = ϕ i ( P B ) = ϕ i ( P C ) = P .
Substitute the expressions for D s ± D_s^{\pm} D s ± and U a ± U_a^{\pm} U a ± in Lemma 4.20 into the recursive transformations above. After all mutations and permutations, the six and twelve cluster variables are
( Z 6 , Z 10 , Z 9 , Z 8 , x 6 , x 5 ) (Z_{6},Z_{10},Z_{9},Z_{8},x_{6},x_{5}) ( Z 6 , Z 10 , Z 9 , Z 8 , x 6 , x 5 )
downstairs and
( Z ^ 12 , Z ^ 9 , Z ^ 13 , Z ^ 16 , Z ^ 14 , Z ^ 15 , Z ^ 10 , Z ^ 7 , x ^ 12 , x ^ 9 , x ^ 10 , x ^ 11 ) (\widehat{Z}_{12},\widehat{Z}_{9},\widehat{Z}_{13},\widehat{Z}_{16},\widehat{Z}_{14},\widehat{Z}_{15},\widehat{Z}_{10},\widehat{Z}_{7},\widehat{x}_{12},\widehat{x}_{9},\widehat{x}_{10},\widehat{x}_{11}) ( Z 12 , Z 9 , Z 13 , Z 16 , Z 14 , Z 15 , Z 10 , Z 7 , x 12 , x 9 , x 10 , x 11 )
upstairs. All intermediate variables are recursively substituted, so these expressions involve only the initial variables and the factors D s ± D_s^{\pm} D s ± and U a ± U_a^{\pm} U a ± . Applying the fixed Laurent homomorphism ϕ i \phi^{i} ϕ i and simplifying the Laurent expressions gives
ϕ i ( x ^ 1 ′ ) = x 1 ′ , ϕ i ( x ^ 2 ′ ) = ϕ i ( x ^ 3 ′ ) = ϕ i ( x ^ 4 ′ ) = x 2 ′ , ϕ i ( x ^ 5 ′ ) = x 3 ′ , \phi^{i}(\widehat{x}'_{1})=x'_{1},\qquad\phi^{i}(\widehat{x}'_{2})=\phi^{i}(\widehat{x}'_{3})=\phi^{i}(\widehat{x}'_{4})=x'_{2},\qquad\phi^{i}(\widehat{x}'_{5})=x'_{3}, ϕ i ( x 1 ′ ) = x 1 ′ , ϕ i ( x 2 ′ ) = ϕ i ( x 3 ′ ) = ϕ i ( x 4 ′ ) = x 2 ′ , ϕ i ( x 5 ′ ) = x 3 ′ ,
ϕ i ( x ^ 6 ′ ) = ϕ i ( x ^ 7 ′ ) = ϕ i ( x ^ 8 ′ ) = x 4 ′ , ϕ i ( x ^ 9 ′ ) = x 5 ′ , ϕ i ( x ^ 10 ′ ) = ϕ i ( x ^ 11 ′ ) = ϕ i ( x ^ 12 ′ ) = x 6 ′ . \phi^{i}(\widehat{x}'_{6})=\phi^{i}(\widehat{x}'_{7})=\phi^{i}(\widehat{x}'_{8})=x'_{4},\qquad\phi^{i}(\widehat{x}'_{9})=x'_{5},\qquad\phi^{i}(\widehat{x}'_{10})=\phi^{i}(\widehat{x}'_{11})=\phi^{i}(\widehat{x}'_{12})=x'_{6}. ϕ i ( x 6 ′ ) = ϕ i ( x 7 ′ ) = ϕ i ( x 8 ′ ) = x 4 ′ , ϕ i ( x 9 ′ ) = x 5 ′ , ϕ i ( x 10 ′ ) = ϕ i ( x 11 ′ ) = ϕ i ( x 12 ′ ) = x 6 ′ .
The corresponding subword of i ′ i' i ′ is
( d , a , b , c , d , a , b , c , d , a , b , c ) , (d,a,b,c,d,a,b,c,d,a,b,c), ( d , a , b , c , d , a , b , c , d , a , b , c ) ,
where positions k , … , k + 5 k,\ldots,k+5 k , … , k + 5 of i ′ i' i ′ correspond to the groups
( 1 ) , ( 2 , 3 , 4 ) , ( 5 ) , ( 6 , 7 , 8 ) , ( 9 ) , ( 10 , 11 , 12 ) . (1),\qquad(2,3,4),\qquad(5),\qquad(6,7,8),\qquad(9),\qquad(10,11,12). ( 1 ) , ( 2 , 3 , 4 ) , ( 5 ) , ( 6 , 7 , 8 ) , ( 9 ) , ( 10 , 11 , 12 ) .
Hence (43) holds. □
Folding of g g g -vectors. For a pointed Laurent polynomial z z z , write
deg i ( z ) ∈ Z l , deg i ^ ( z ) ∈ Z l ^ ( l ^ : = ℓ ( i ^ ) ) \deg_{i}(z)\in\mathbb{Z}^{l},\qquad\deg_{\widehat{i}}(z)\in\mathbb{Z}^{\widehat{l}}\qquad(\widehat{l}:=\ell(\widehat{i})) deg i ( z ) ∈ Z l , deg i ( z ) ∈ Z l ( l := ℓ ( i ))
for its g g g -vector with respect to the initial clusters x i x^{i} x i and x ^ i \widehat{x}^{i} x i , respectively. Recall the fiber-sum map
Fold i : Z l ^ ⟶ Z l , ( Fold i ( a ^ ) ) p = ∑ r = 1 e i p a ^ κ ′ ( p , r ) , \operatorname{Fold}^{i}:\mathbb{Z}^{\widehat{l}}\longrightarrow\mathbb{Z}^{l},\qquad\left(\operatorname{Fold}^{i}(\widehat{a})\right)_{p}=\sum_{r=1}^{e_{i_{p}}}\widehat{a}_{\kappa'(p,r)}, Fold i : Z l ⟶ Z l , ( Fold i ( a ) ) p = r = 1 ∑ e i p a κ ′ ( p , r ) ,
and its restriction Fold e x i \operatorname{Fold}^{i}_{\mathrm{ex}} Fold ex i to the exchangeable coordinates. By Proposition 3.8 ,
Fold i ∘ B ^ i = B i ∘ Fold e x i . (57) \operatorname{Fold}^{i}\circ\widehat{B}^{i}=B^{i}\circ\operatorname{Fold}^{i}_{\mathrm{ex}}. \tag*{(57)} Fold i ∘ B i = B i ∘ Fold ex i . (57)
Proposition 4.22. Let i ′ i' i ′ be obtained from i i i by a braid move. Then
deg i ( x p ′ ) = Fold i ( deg i ^ ( x ^ κ ′ ( p , r ) ′ ) ) ( 1 ≤ p ≤ l , 1 ≤ r ≤ e i p ′ ) . \deg_{i}(x'_{p})=\operatorname{Fold}^{i}\left(\deg_{\widehat{i}}(\widehat{x}'_{\kappa'(p,r)})\right)\qquad(1\le p\le l,\ 1\le r\le e_{i'_{p}}). deg i ( x p ′ ) = Fold i ( deg i ( x κ ′ ( p , r ) ′ ) ) ( 1 ≤ p ≤ l , 1 ≤ r ≤ e i p ′ ) .
Proof. Set
g ^ : = deg i ( x ^ κ ′ ( p , r ) ) . \widehat{\mathbf{g}}:=\deg_i\left(\widehat{x}_{\kappa'(p,r)}\right). g := deg i ( x κ ′ ( p , r ) ) .
Put l ^ e x : = ∣ J ^ e x i ∣ \widehat{l}_{\mathrm{ex}}:=\lvert\widehat{J}^{i}_{\mathrm{ex}}\rvert l ex := ∣ J ex i ∣ . By pointedness,
x ^ κ ′ ( p , r ) = x ^ g ^ ( 1 + ∑ n ^ ∈ N l ^ e x n ^ ≠ 0 a n ^ x ^ B i n ^ ) . \widehat{x}_{\kappa'(p,r)}=\widehat{x}^{\widehat{\mathbf{g}}}\left(1+\sum_{\substack{\widehat{\mathbf{n}}\in\mathbb{N}^{\widehat{l}_{\mathrm{ex}}}\\\widehat{\mathbf{n}}\ne0}}a_{\widehat{\mathbf{n}}}\widehat{x}^{B^{i}\widehat{\mathbf{n}}}\right). x κ ′ ( p , r ) = x g 1 + n ∈ N l ex n = 0 ∑ a n x B i n .
Applying ϕ i \phi^{i} ϕ i and using (57) , we obtain
ϕ i ( x ^ κ ′ ( p , r ) ) = x Fold i ( g ^ ) ( 1 + ∑ n ^ ≠ 0 a n ^ x B i Fold e x i ( n ^ ) ) . \phi^{i}\left(\widehat{x}_{\kappa'(p,r)}\right)=x^{\operatorname{Fold}^{i}(\widehat{\mathbf{g}})}\left(1+\sum_{\widehat{\mathbf{n}}\ne0}a_{\widehat{\mathbf{n}}}x^{B^{i}\operatorname{Fold}^{i}_{\mathrm{ex}}(\widehat{\mathbf{n}})}\right). ϕ i ( x κ ′ ( p , r ) ) = x Fold i ( g ) 1 + n = 0 ∑ a n x B i Fold ex i ( n ) .
Since n ^ ∈ N l ^ e x \widehat{\mathbf{n}}\in\mathbb{N}^{\widehat{l}_{\mathrm{ex}}} n ∈ N l ex , we have
n ^ ≠ 0 if and only if Fold e x i ( n ^ ) ≠ 0. \widehat{\mathbf{n}}\ne0\quad\text{if and only if}\quad\operatorname{Fold}^{i}_{\mathrm{ex}}(\widehat{\mathbf{n}})\ne0. n = 0 if and only if Fold ex i ( n ) = 0.
Since ( B i , Λ i ) (B^{i},\Lambda^{i}) ( B i , Λ i ) is a compatible pair, B i B^{i} B i has full column rank, and hence
B i Fold e x i ( n ^ ) ≠ 0 ( n ^ ≠ 0 ) . B^{i}\operatorname{Fold}^{i}_{\mathrm{ex}}(\widehat{\mathbf{n}})\ne0\qquad(\widehat{\mathbf{n}}\ne0). B i Fold ex i ( n ) = 0 ( n = 0 ) .
Therefore, the right-hand side is pointed of degree Fold i ( g ^ ) \operatorname{Fold}^{i}(\widehat{\mathbf{g}}) Fold i ( g ) . By (43) ,
ϕ i ( x ^ κ ′ ( p , r ) ) = x p ′ . \phi^{i}\left(\widehat{x}_{\kappa'(p,r)}\right)=x'_{p}. ϕ i ( x κ ′ ( p , r ) ) = x p ′ .
Therefore
deg i ( x p ′ ) = Fold i ( deg i ( x ^ κ ′ ( p , r ) ) ) . \deg_i(x'_{p})=\operatorname{Fold}^{i}\left(\deg_i\left(\widehat{x}_{\kappa'(p,r)}\right)\right). deg i ( x p ′ ) = Fold i ( deg i ( x κ ′ ( p , r ) ) ) .
i-box moves and folding
In this section, we study the behavior of the canonical lifting of i-box families under box moves. Starting from a box move downstairs, we construct a sequence of box moves upstairs, called its lifted path , and realize this sequence by mutations and permutations. We then prove the corresponding identities for the exchange and commutation matrices, cluster variables, and g g g -vectors at the seeds before and after the downstairs move and its lifted path.
Box moves and their lifts
Retain the notation and conventions of Subsection 3.3. Let C = ( c 1 , … , c l ) \mathcal{C}=(c_{1},\ldots,c_{l}) C = ( c 1 , … , c l ) be an admissible chain of i-boxes, and write p ( C ) = ( p 1 , … , p l ) p(\mathcal{C})=(p_{1},\ldots,p_{l}) p ( C ) = ( p 1 , … , p l ) for its effective-end sequence. We write B k \mathcal{B}_{k} B k and B u p \mathcal{B}^{\mathrm{up}} B up for box moves downstairs and upstairs, respectively. By definition, we have the following.
Lemma 5.1. Let C ′ = B k ( C ) \mathcal{C}'=\mathcal{B}_{k}(\mathcal{C}) C ′ = B k ( C ) where c k c_{k} c k is movable. Then
p ( C ′ ) = ( p 1 , … , p k − 1 , p k + 1 , p k , p k + 2 , … , p l ) . p(\mathcal{C}')=(p_{1},\ldots,p_{k-1},p_{k+1},p_{k},p_{k+2},\ldots,p_{l}). p ( C ′ ) = ( p 1 , … , p k − 1 , p k + 1 , p k , p k + 2 , … , p l ) .
Moreover,
(a) B k \mathcal{B}_{k} B k is of mutation type if and only if i p k = i p k + 1 i_{p_{k}}=i_{p_{k+1}} i p k = i p k + 1 .
(b) B k \mathcal{B}_{k} B k is of permutation type if and only if i p k ≠ i p k + 1 i_{p_{k}}\ne i_{p_{k+1}} i p k = i p k + 1 .
Corollary 5.2. If B k \mathcal{B}_{k} B k is of mutation type, then we have e i p k = e i p k + 1 e_{i_{p_{k}}}=e_{i_{p_{k+1}}} e i p k = e i p k + 1 .
Put
p : = p k , q : = p k + 1 , r : = e i p , s : = e i q , p := p_k,\qquad q := p_{k+1},\qquad r := e_{ip},\qquad s := e_{iq}, p := p k , q := p k + 1 , r := e i p , s := e i q ,
and set
p ( a ) : = κ ( p , a ) ( 1 ≤ a ≤ r ) , q ( b ) : = κ ( q , b ) ( 1 ≤ b ≤ s ) . p^{(a)} := \kappa(p,a)\quad(1 \le a \le r),\qquad q^{(b)} := \kappa(q,b)\quad(1 \le b \le s). p ( a ) := κ ( p , a ) ( 1 ≤ a ≤ r ) , q ( b ) := κ ( q , b ) ( 1 ≤ b ≤ s ) .
Write
p : = ( p ( 1 ) , … , p ( r ) ) , q : = ( q ( 1 ) , … , q ( s ) ) , \mathbf{p} := (p^{(1)},\ldots,p^{(r)}),\qquad\mathbf{q} := (q^{(1)},\ldots,q^{(s)}), p := ( p ( 1 ) , … , p ( r ) ) , q := ( q ( 1 ) , … , q ( s ) ) ,
and
p r e v : = ( p ( r ) , … , p ( 1 ) ) , q r e v : = ( q ( s ) , … , q ( 1 ) ) . \mathbf{p}^{\mathrm{rev}} := (p^{(r)},\ldots,p^{(1)}),\qquad\mathbf{q}^{\mathrm{rev}} := (q^{(s)},\ldots,q^{(1)}). p rev := ( p ( r ) , … , p ( 1 ) ) , q rev := ( q ( s ) , … , q ( 1 ) ) .
By (39) , the entries p ( 1 ) , … , p ( r ) p^{(1)},\ldots,p^{(r)} p ( 1 ) , … , p ( r ) occur in p ( C ^ ) \mathbf{p}(\widehat{\mathcal{C}}) p ( C ) in this order if the block for p p p in T ^ \widehat{\mathcal{T}} T consists of R \mathcal{R} R ’s, and in the reverse order if it consists of L \mathcal{L} L ’s. The same holds for q ( 1 ) , … , q ( s ) q^{(1)},\ldots,q^{(s)} q ( 1 ) , … , q ( s ) . Since c k c_k c k is movable, the two adjacent fibers occur with opposite orientations. Thus, up to reversing the orientation, the relevant part of the canonical lifted effective-end sequence is
( p , q r e v ) . (58) (\mathbf{p},\mathbf{q}^{\mathrm{rev}}). \tag*{(58)} ( p , q rev ) . (58)
Equivalently, in the opposite orientation it is
( p r e v , q ) . (\mathbf{p}^{\mathrm{rev}},\mathbf{q}). ( p rev , q ) .
Let
h : = ∑ a = 1 k − 1 e i p a h := \sum_{a=1}^{k-1} e_{ip_a} h := a = 1 ∑ k − 1 e i p a
be the number of lifted effective ends preceding the entries of p \mathbf{p} p and q \mathbf{q} q . We first consider the case where B k \mathcal{B}_k B k is of permutation type. In what follows, a sequence of box moves
( B a 1 u p , … , B a m u p ) (\mathcal{B}^{\mathrm{up}}_{a_1},\ldots,\mathcal{B}^{\mathrm{up}}_{a_m}) ( B a 1 up , … , B a m up )
is applied from left to right, in the order displayed.
Proposition 5.3. Suppose that B k \mathcal{B}_k B k is of permutation type. Then its lift is
B ^ k : = ( B h + r u p , … , B h + 1 u p ) ( B h + r + 1 u p , … , B h + 2 u p ) ⋯ ( B h + r + s − 1 u p , … , B h + s u p ) . \widehat{\mathcal{B}}_k := (\mathcal{B}^{\mathrm{up}}_{h+r},\ldots,\mathcal{B}^{\mathrm{up}}_{h+1})(\mathcal{B}^{\mathrm{up}}_{h+r+1},\ldots,\mathcal{B}^{\mathrm{up}}_{h+2})\cdots(\mathcal{B}^{\mathrm{up}}_{h+r+s-1},\ldots,\mathcal{B}^{\mathrm{up}}_{h+s}). B k := ( B h + r up , … , B h + 1 up ) ( B h + r + 1 up , … , B h + 2 up ) ⋯ ( B h + r + s − 1 up , … , B h + s up ) . These are r s rs r s permutation-type box moves, and they give
( p , q r e v ) ⟼ ( q r e v , p ) . (\mathbf{p},\mathbf{q}^{\mathrm{rev}})\longmapsto(\mathbf{q}^{\mathrm{rev}},\mathbf{p}). ( p , q rev ) ⟼ ( q rev , p ) . The reverse transformation is obtained by reversing the order of the box moves.
Proof. By Lemma 5.1, B k \mathcal{B}_k B k interchanges p = p k p=p_k p = p k and q = p k + 1 q=p_{k+1} q = p k + 1 . In the chosen orientation, their lifted effective-end blocks occur as
p , q r e v . \mathbf{p},\qquad\mathbf{q}^{\mathrm{rev}}. p , q rev .
For k ≥ 2 k\ge2 k ≥ 2 , movability of c k c_k c k implies that the two corresponding L / R \mathcal{L}/\mathcal{R} L / R -blocks have opposite orientations. Hence each prescribed box move can be performed. The b b b -th parenthesized block of moves shifts q ( s − b + 1 ) q^{(s-b+1)} q ( s − b + 1 ) successively across
p ( r ) , … , p ( 1 ) . p^{(r)},\ldots,p^{(1)}. p ( r ) , … , p ( 1 ) .
Thus the successive blocks of moves shift
q ( s ) , q ( s − 1 ) , … , q ( 1 ) q^{(s)},q^{(s-1)},\ldots,q^{(1)} q ( s ) , q ( s − 1 ) , … , q ( 1 )
across the whole p-block, and hence
( p , q r e v ) ⟼ ( q r e v , p ) . (\mathbf{p},\mathbf{q}^{\mathrm{rev}}) \longmapsto(\mathbf{q}^{\mathrm{rev}},\mathbf{p}). ( p , q rev ) ⟼ ( q rev , p ) .
For k = 1 k=1 k = 1 , the definition of the canonical lift in (39) gives the same sequence of box moves, and the chain is precisely the canonical lift of C ′ \mathcal{C}' C ′ .
Finally, since B k \mathcal{B}_k B k is of permutation type,
by Lemma 5.1. Hence every crossing between a p p p - and a q q q -effective end is of permutation type. There are r s rs r s such crossings.
We next consider the case where B k \mathcal{B}_k B k is of mutation type. By Lemma 5.1 and Corollary 5.2, we have
i p = i q , r = s . i_p=i_q,\qquad r=s. i p = i q , r = s .
We keep the notation r r r for this common value.
Set
δ h : = { 1 , h = 0 , 0 , h > 0. \delta_h:=\begin{cases}
1, & h=0,\\
0, & h>0.
\end{cases} δ h := { 1 , 0 , h = 0 , h > 0.
Lemma 5.4. Let r ≥ 1 r\ge1 r ≥ 1 , and let h ≥ 0 h\ge0 h ≥ 0 be the number of lifted effective ends preceding the two consecutive r r r -blocks. If the relevant { L , R } \{\mathcal{L},\mathcal{R}\} { L , R } -word is
R r − δ h L r , \mathcal{R}^{r-\delta_h}\mathcal{L}^{r}, R r − δ h L r , then
( B h + r u p , … , B h + 1 u p ) ( B h + r + 1 u p , … , B h + 2 u p ) ⋯ ( B h + 2 r − 1 u p , … , B h + r u p ) (\mathcal{B}^{\mathrm{up}}_{h+r},\ldots,\mathcal{B}^{\mathrm{up}}_{h+1})(\mathcal{B}^{\mathrm{up}}_{h+r+1},\ldots,\mathcal{B}^{\mathrm{up}}_{h+2})\cdots(\mathcal{B}^{\mathrm{up}}_{h+2r-1},\ldots,\mathcal{B}^{\mathrm{up}}_{h+r}) ( B h + r up , … , B h + 1 up ) ( B h + r + 1 up , … , B h + 2 up ) ⋯ ( B h + 2 r − 1 up , … , B h + r up ) consists of r 2 r^2 r 2 box moves and transforms it into
L r − δ h R r . \mathcal{L}^{r-\delta_h}\mathcal{R}^{r}. L r − δ h R r . The reverse transformation is obtained by reversing the order of the box moves.
Proof. For 0 ≤ t ≤ r 0\le t\le r 0 ≤ t ≤ r , set
W t : = ( q ( r ) , … , q ( r − t + 1 ) , p ( 1 ) , … , p ( r ) , q ( r − t ) , … , q ( 1 ) ) , W_t:=\left(q^{(r)},\ldots,q^{(r-t+1)},p^{(1)},\ldots,p^{(r)},q^{(r-t)},\ldots,q^{(1)}\right), W t := ( q ( r ) , … , q ( r − t + 1 ) , p ( 1 ) , … , p ( r ) , q ( r − t ) , … , q ( 1 ) ) , with empty blocks omitted. The t t t -th parenthesized block of moves shifts q ( r − t + 1 ) q^{(r-t+1)} q ( r − t + 1 ) across p ( r ) , … , p ( 1 ) p^{(r)},\ldots,p^{(1)} p ( r ) , … , p ( 1 ) and hence W t − 1 ⟼ W t ( 1 ≤ t ≤ r ) W_{t-1}\longmapsto W_t\quad(1\le t\le r) W t − 1 ⟼ W t ( 1 ≤ t ≤ r ) .
Immediately before a prescribed box move at position a > 1 a>1 a > 1 , the direction symbols at positions a − 1 a-1 a − 1 and a a a are R \mathcal{R} R and L \mathcal{L} L , respectively. The box at position a a a is therefore movable. A move at position 1 occurs only when h = 0 h=0 h = 0 , and the first box is movable by definition. Thus all r 2 r^2 r 2 moves can be performed and
( p , q r e v ) ⟼ ( q r e v , p ) . (\mathbf{p},\mathbf{q}^{\mathrm{rev}})\longmapsto(\mathbf{q}^{\mathrm{rev}},\mathbf{p}). ( p , q rev ) ⟼ ( q rev , p ) . Accordingly,
R r − δ h L r ⟼ L r − δ h R r . \mathcal{R}^{r-\delta_h}\mathcal{L}^{r}\longmapsto\mathcal{L}^{r-\delta_h}\mathcal{R}^{r}. R r − δ h L r ⟼ L r − δ h R r . The reverse transformation follows since every box move is involutive. □
Proposition 5.5. Suppose that B k \mathcal{B}_k B k is of mutation type. Then its lift is
B ^ k : = ( B h + r + 1 u p , … , B h + 1 u p ) ( B h + r + 1 u p , … , B h + 2 r u p ) ⋯ ( B h + 2 r − 1 u p , … , B h + r u p ) . \widehat{\mathcal{B}}_k := (\mathcal{B}^{\mathrm{up}}_{h+r+1},\ldots,\mathcal{B}^{\mathrm{up}}_{h+1})(\mathcal{B}^{\mathrm{up}}_{h+r+1},\ldots,\mathcal{B}^{\mathrm{up}}_{h+2r})\cdots(\mathcal{B}^{\mathrm{up}}_{h+2r-1},\ldots,\mathcal{B}^{\mathrm{up}}_{h+r}). B k := ( B h + r + 1 up , … , B h + 1 up ) ( B h + r + 1 up , … , B h + 2 r up ) ⋯ ( B h + 2 r − 1 up , … , B h + r up ) . It consists of r 2 r^2 r 2 box moves. Writing M M M for a mutation-type move and P P P for a permutation-type move, their sequence of types is
M P r − 1 ⋅ P M P r − 2 ⋯ P r − 2 M P ⋅ P r − 1 M . MP^{r-1}\cdot PMP^{r-2}\cdots P^{r-2}MP\cdot P^{r-1}M. M P r − 1 ⋅ P M P r − 2 ⋯ P r − 2 M P ⋅ P r − 1 M . In particular, B ^ k \widehat{\mathcal{B}}_k B k contains exactly r r r mutation-type box moves and r ( r − 1 ) r(r-1) r ( r − 1 ) permutation-type box moves.
Proof. By Lemma 5.4 , all r 2 r^2 r 2 prescribed moves can be performed and
( p , q r e v ) ⟼ ( q r e v , p ) . (\mathbf{p},\mathbf{q}^{\mathrm{rev}})\longmapsto(\mathbf{q}^{\mathrm{rev}},\mathbf{p}). ( p , q rev ) ⟼ ( q rev , p ) . By the fixed fiber ordering, a crossing of p ( a ) p^{(a)} p ( a ) and q ( b ) q^{(b)} q ( b ) is of mutation type exactly when a = b a=b a = b , and of permutation type otherwise. Hence
# M = r , # P = r ( r − 1 ) . \#M=r,\qquad\#P=r(r-1). # M = r , # P = r ( r − 1 ) . In the ( t + 1 ) (t+1) ( t + 1 ) -st block of moves, for 0 ≤ t < r 0\le t<r 0 ≤ t < r , the element q ( r − t ) q^{(r-t)} q ( r − t ) crosses
p ( r ) , … , p ( 1 ) , p^{(r)},\ldots,p^{(1)}, p ( r ) , … , p ( 1 ) , and the corresponding sequence of types is
P t M P r − 1 − t . P^tMP^{r-1-t}. P t M P r − 1 − t . Therefore the full sequence of types is
M P r − 1 ⋅ P M P r − 2 ⋯ P r − 2 M P ⋅ P r − 1 M = ( M P r ) r − 1 M . MP^{r-1}\cdot PMP^{r-2}\cdots P^{r-2}MP\cdot P^{r-1}M=(MP^r)^{r-1}M. M P r − 1 ⋅ P M P r − 2 ⋯ P r − 2 M P ⋅ P r − 1 M = ( M P r ) r − 1 M . The opposite orientation is obtained by reversing the moves. □
Seed transformations under lifted box moves
Let
C ′ : = B k ( C ) , C u p : = B ^ k ( C ^ ) and F ′ : = F ( C ′ ) , F u p : = F ( C u p ) . \mathcal{C}' := \mathcal{B}_k(\mathcal{C}),\qquad\mathcal{C}^{\mathrm{up}} := \widehat{\mathcal{B}}_k(\widehat{\mathcal{C}}) \qquad\text{and}\qquad\mathcal{F}' := \mathcal{F}(\mathcal{C}'),\qquad\mathcal{F}^{\mathrm{up}} := \mathcal{F}(\mathcal{C}^{\mathrm{up}}). C ′ := B k ( C ) , C up := B k ( C ) and F ′ := F ( C ′ ) , F up := F ( C up ) .
Let C ^ ′ \widehat{\mathcal{C}}' C ′ denote the canonical lift of C ′ \mathcal{C}' C ′ and set
F ^ ′ : = F ( C ^ ′ ) . \widehat{\mathcal{F}}' := \mathcal{F}(\widehat{\mathcal{C}}'). F ′ := F ( C ′ ) .
Thus C u p \mathcal{C}^{\mathrm{up}} C up is obtained by performing the lifted box moves of Propositions 5.3 and 5.5, whereas C ^ ′ \widehat{\mathcal{C}}' C ′ is the canonical lift constructed directly from C ′ \mathcal{C}' C ′ . We first compare these two admissible chains.
Proposition 5.6. Let p ( C ′ ) = ( p 1 ′ , … , p l ′ ) \mathbf{p}(\mathcal{C}')=(p'_1,\ldots,p'_l) p ( C ′ ) = ( p 1 ′ , … , p l ′ ) . Then the effective ends of C u p \mathcal{C}^{\mathrm{up}} C up decompose into consecutive complete fibers over
p 1 ′ , … , p l ′ . p'_1,\ldots,p'_l. p 1 ′ , … , p l ′ . More precisely, the block over p a ′ p'_a p a ′ consists of
{ κ ( p a ′ , 1 ) , … , κ ( p a ′ , e p a ′ ) } ( 1 ≤ a ≤ l ) . \{\kappa(p'_a,1),\ldots,\kappa(p'_a,e_{p'_a})\}\qquad(1\le a\le l). { κ ( p a ′ , 1 ) , … , κ ( p a ′ , e p a ′ )} ( 1 ≤ a ≤ l ) . Moreover, we have
F u p = F ^ ′ . \mathcal{F}^{\mathrm{up}}=\widehat{\mathcal{F}}'. F up = F ′ .
Proof. Let p ( C ) = ( p 1 , … , p l ) \mathbf{p}(\mathcal{C})=(p_1,\ldots,p_l) p ( C ) = ( p 1 , … , p l ) . By Lemma 5.1,
p ( C ′ ) = ( p 1 , … , p k − 1 , p k + 1 , p k , p k + 2 , … , p l ) . \mathbf{p}(\mathcal{C}')=(p_1,\ldots,p_{k-1},p_{k+1},p_k,p_{k+2},\ldots,p_l). p ( C ′ ) = ( p 1 , … , p k − 1 , p k + 1 , p k , p k + 2 , … , p l ) .
By Propositions 5.3 and 5.5, B ^ k \widehat{\mathcal{B}}_k B k interchanges exactly the two complete effective-end fibers over p k p_k p k and p k + 1 p_{k+1} p k + 1 and leaves all other fibers fixed. Hence the effective ends of C u p \mathcal{C}^{\mathrm{up}} C up decompose into consecutive complete fibers over
p 1 ′ , … , p l ′ . p'_1,\ldots,p'_l. p 1 ′ , … , p l ′ .
It remains to compare C u p \mathcal{C}^{\mathrm{up}} C up with C ^ ′ \widehat{\mathcal{C}}' C ′ . They agree for k = 1 k=1 k = 1 , for k ≥ 3 k\geq3 k ≥ 3 , and for k = 2 k=2 k = 2 with e i p 1 = 1 e_{i p_1}=1 e i p 1 = 1 . If k = 2 k=2 k = 2 and e i p 1 > 1 e_{i p_1}>1 e i p 1 > 1 , the change of the first { L , R } \{L,R\} { L , R } -symbol reverses only the order of the singleton boxes in the initial lifted fiber; both orders determine the same maximal commuting family. Therefore
F u p = F ^ ′ . \mathcal{F}^{\mathrm{up}}=\widehat{\mathcal{F}}'. F up = F ′ .
By Proposition 5.6,
F u p = F ( C u p ) = F ( C ^ ′ ) = F ^ ′ . \mathcal{F}^{\mathrm{up}}=\mathcal{F}(\mathcal{C}^{\mathrm{up}})=\mathcal{F}(\widehat{\mathcal{C}}')=\widehat{\mathcal{F}}'. F up = F ( C up ) = F ( C ′ ) = F ′ .
We next consider the mutation–permutation transformations of compatible pairs corresponding to the lifted box moves. As above, positions and box labels refer to the family obtained after the preceding box moves.
Proposition 5.7. Let
C ′ = B k ( C ) . \mathcal{C}'=\mathcal{B}_k(\mathcal{C}). C ′ = B k ( C ) . Then the mutation–permutation realization of B k \mathcal{B}_k B k transforms
( B F , Λ F ) ⟼ ( B F ′ , Λ F ′ ) , (B^{\mathcal{F}},\Lambda^{\mathcal{F}})\longmapsto(B^{\mathcal{F}'},\Lambda^{\mathcal{F}'}), ( B F , Λ F ) ⟼ ( B F ′ , Λ F ′ ) , while the mutation–permutation realizations of the successive upstairs box moves in B ^ k \widehat{\mathcal{B}}_k B k transform
( B F ^ , Λ F ^ ) ⟼ ( B F ′ ^ , Λ F ′ ^ ) . (B^{\widehat{\mathcal{F}}},\Lambda^{\widehat{\mathcal{F}}})\longmapsto(B^{\widehat{\mathcal{F}'}},\Lambda^{\widehat{\mathcal{F}'}}). ( B F , Λ F ) ⟼ ( B F ′ , Λ F ′ ) . If B k \mathcal{B}_k B k is of permutation type, the mutation–permutation sequence induced by B ^ k \widehat{\mathcal{B}}_k B k contains no mutation. If B k \mathcal{B}_k B k is of mutation type, this sequence contains exactly r r r mutations and r ( r − 1 ) r(r-1) r ( r − 1 ) permutation-type transformations.
Proof. By Proposition 3.4, a permutation-type box move acts on the associated compatible pair by the corresponding permutation, whereas a mutation-type box move acts by the corresponding mutation. Applying this successively to the box moves in B ^ k \widehat{\mathcal{B}}_k B k , with respect to the current family, transforms
( B F ^ , Λ F ^ ) ⟼ ( B F u p , Λ F u p ) . (B^{\widehat{\mathcal{F}}},\Lambda^{\widehat{\mathcal{F}}})\longmapsto(B^{\mathcal{F}^{\mathrm{up}}},\Lambda^{\mathcal{F}^{\mathrm{up}}}). ( B F , Λ F ) ⟼ ( B F up , Λ F up ) .
By Proposition 5.6,
F u p = F ^ ′ , \mathcal{F}^{\mathrm{up}}=\widehat{\mathcal{F}}', F up = F ′ ,
and hence
( B F u p , Λ F u p ) = ( B F ′ ^ , Λ F ′ ^ ) . (B^{\mathcal{F}^{\mathrm{up}}},\Lambda^{\mathcal{F}^{\mathrm{up}}})=(B^{\widehat{\mathcal{F}'}},\Lambda^{\widehat{\mathcal{F}'}}). ( B F up , Λ F up ) = ( B F ′ , Λ F ′ ) .
The assertion for B k \mathcal{B}_k B k follows directly from Proposition 3.4. Finally, the assertions on the numbers of mutations and permutation-type transformations follow from Propositions 5.3 and 5.5.
We next compare the compatible pairs before and after B k \mathcal{B}_k B k and B ^ k \widehat{\mathcal{B}}_k B k .
Proposition 5.8. With the notation above, the following diagram commutes:
( B F ^ , Λ ^ F ^ ) → B ^ k ( B F ^ ′ , Λ ^ F ^ ′ ) Fold ↓ ↓ Fold ( B F , Λ F ) → B k ( B F ′ , Λ F ′ ) \begin{CD}
(\mathcal{B}^{\widehat{\mathcal{F}}},\widehat{\Lambda}^{\widehat{\mathcal{F}}}) @>{\widehat{\mathcal{B}}_k}>> (\mathcal{B}^{\widehat{\mathcal{F}}'},\widehat{\Lambda}^{\widehat{\mathcal{F}}'}) \\
@V{\operatorname{Fold}}VV @VV{\operatorname{Fold}}V \\
(\mathcal{B}^{\mathcal{F}},\Lambda^{\mathcal{F}}) @>{\mathcal{B}_k}>> (\mathcal{B}^{\mathcal{F}'},\Lambda^{\mathcal{F}'})
\end{CD} ( B F , Λ F ) Fold ↓ ⏐ ( B F , Λ F ) B k B k ( B F ′ , Λ F ′ ) ↓ ⏐ Fold ( B F ′ , Λ F ′ ) where the horizontal arrows are the mutation–permutation transformations of Proposition 5.7 , and the vertical arrows are given by Theorem 3.15 .
Proof. The horizontal arrows are given by Proposition 5.7 , while the two vertical arrows are given by Theorem 3.15 , applied to ( F , F ^ ) (\mathcal{F},\widehat{\mathcal{F}}) ( F , F ) and ( F ′ , F ^ ′ ) (\mathcal{F}',\widehat{\mathcal{F}}') ( F ′ , F ′ ) , respectively. □
Folding of i-box cluster variables
We work in the commutative setting throughout this subsection. Let F = F ( C ) \mathcal{F}=\mathcal{F}(\mathcal{C}) F = F ( C ) and F ^ = F ( C ^ ) \widehat{\mathcal{F}}=\mathcal{F}(\widehat{\mathcal{C}}) F = F ( C ) be the downstairs and upstairs commuting families considered above. For each c ∈ F c\in\mathcal{F} c ∈ F , write
Fib C ( c ) : = { c ^ 1 , … , c ^ e c } . \operatorname{Fib}^{\mathcal{C}}(c):=\{\widehat{c}_1,\ldots,\widehat{c}_{e_c}\}. Fib C ( c ) := { c 1 , … , c e c } .
Let
x F = ( x c ∣ c ∈ F ) , x F ^ = ( x ^ c ^ ∣ c ^ ∈ F ^ ) \mathbf{x}^{\mathcal{F}}=(x_c\mid c\in\mathcal{F}),\qquad\mathbf{x}^{\widehat{\mathcal{F}}}=(\widehat{x}_{\widehat{c}}\mid\widehat{c}\in\widehat{\mathcal{F}}) x F = ( x c ∣ c ∈ F ) , x F = ( x c ∣ c ∈ F )
be the corresponding clusters. Define the Laurent homomorphism
ϕ C : L ( Σ ^ F ^ ) ⟶ L ( Σ F ) , x ^ c ^ ⟼ x c ( c ^ ∈ Fib C ( c ) ) . \phi^{\mathcal{C}}:\mathcal{L}(\widehat{\Sigma}^{\widehat{\mathcal{F}}})\longrightarrow\mathcal{L}(\Sigma^{\mathcal{F}}),\qquad\widehat{x}_{\widehat{c}}\longmapsto x_c\quad(\widehat{c}\in\operatorname{Fib}^{\mathcal{C}}(c)). ϕ C : L ( Σ F ) ⟶ L ( Σ F ) , x c ⟼ x c ( c ∈ Fib C ( c )) .
As in the braid-move case, we shall compare the final cluster variables using only this fixed Laurent homomorphism.
To compare the exchange monomials, we first strengthen the fiber-sum identity of Theorem 3.15 to its positive and negative parts.
Lemma 5.9. Let c , d ∈ F c,d\in\mathcal{F} c , d ∈ F , with d d d exchangeable, and fix d ^ ∈ Fib C ( d ) \widehat{d}\in\operatorname{Fib}^{\mathcal{C}}(d) d ∈ Fib C ( d ) . Then the integers
b c ^ , d ^ F ^ ( c ^ ∈ Fib C ( c ) ) b^{\widehat{\mathcal{F}}}_{\widehat{c},\widehat{d}}\quad(\widehat{c}\in\operatorname{Fib}^{\mathcal{C}}(c)) b c , d F ( c ∈ Fib C ( c )) have the same weak sign.
Consequently, we have
∑ c ^ ∈ Fib C ( c ) [ b c ^ , d ^ F ^ ] + = [ b c , d F ] + , and ∑ c ^ ∈ Fib C ( c ) [ − b c ^ , d ^ F ^ ] + = [ − b c , d F ] + . \sum_{\widehat{c}\in\operatorname{Fib}^{\mathcal{C}}(c)}
\left[b^{\widehat{\mathcal{F}}}_{\widehat{c},\widehat{d}}\right]_{+}
=
\left[b^{\mathcal{F}}_{c,d}\right]_{+},
\qquad\text{and}\qquad
\sum_{\widehat{c}\in\operatorname{Fib}^{\mathcal{C}}(c)}
\left[-b^{\widehat{\mathcal{F}}}_{\widehat{c},\widehat{d}}\right]_{+}
=
\left[-b^{\mathcal{F}}_{c,d}\right]_{+}. c ∈ Fib C ( c ) ∑ [ b c , d F ] + = [ b c , d F ] + , and c ∈ Fib C ( c ) ∑ [ − b c , d F ] + = [ − b c , d F ] + .
Proof. We use the local description of the i -box exchange matrix in (30) together with the fiberwise construction of F ^ \widehat{\mathcal{F}} F .
(a) Assume that b c , d F < 0 b^{\mathcal{F}}_{c,d}<0 b c , d F < 0 . By (30) , this negative entry arises either from one of the horizontal relations in (i) or from one of the vertical relations in (ii).
Suppose first that it arises from (i) of (30) . Fix d ^ ∈ Fib C ( d ) \widehat{d}\in\operatorname{Fib}^{\mathcal{C}}(d) d ∈ Fib C ( d ) . By the fiberwise construction, for c ^ ∈ Fib C ( c ) \widehat{c}\in\operatorname{Fib}^{\mathcal{C}}(c) c ∈ Fib C ( c ) , the pair ( c ^ , d ^ ) (\widehat{c},\widehat{d}) ( c , d ) either satisfies the corresponding horizontal relation in (i) of (30) , in which case b c ^ , d ^ F ^ = − 1 b^{\widehat{\mathcal{F}}}_{\widehat{c},\widehat{d}}=-1 b c , d F = − 1 , or it does not, in which case b c ^ , d ^ F ^ = 0 b^{\widehat{\mathcal{F}}}_{\widehat{c},\widehat{d}}=0 b c , d F = 0 . The opposite horizontal relation cannot occur within the same fiber. Hence
b c ^ , d ^ F ^ ≤ 0 ( c ^ ∈ Fib C ( c ) ) . (59) b^{\widehat{\mathcal{F}}}_{\widehat{c},\widehat{d}}\leq0\qquad(\widehat{c}\in\operatorname{Fib}^{\mathcal{C}}(c)). \tag*{(59)} b c , d F ≤ 0 ( c ∈ Fib C ( c )) . (59)
Suppose next that b c , d F b^{F}_{c,d} b c , d F arises from one of the vertical relations in (ii) of (30) . The endpoint inequalities, the effective-end conditions, and the required boxes in (ii) of (30) lift fiberwise. Thus, for fixed d ^ \widehat d d , every c ^ ∈ Fib c ( c ) \widehat c\in\operatorname{Fib}^{c}(c) c ∈ Fib c ( c ) either satisfies the corresponding lifted vertical relation, in which case its exchange-matrix entry is the corresponding negative upstairs Cartan entry, or contributes zero. The opposite vertical relation would require the reversed system of inequalities in (ii) of (30) , and hence cannot occur within the same fiber. Therefore
b c ^ , d ^ F ^ ≤ 0 ( c ^ ∈ Fib c ( c ) ) . b^{\widehat F}_{\widehat c,\widehat d}\leq0\qquad(\widehat c\in\operatorname{Fib}^{c}(c)). b c , d F ≤ 0 ( c ∈ Fib c ( c )) .
Thus
b c , d F < 0 implies b c ^ , d ^ F ^ ≤ 0 ( c ^ ∈ Fib c ( c ) ) . b^{F}_{c,d}<0\quad\text{implies}\quad b^{\widehat F}_{\widehat c,\widehat d}\leq0\qquad(\widehat c\in\operatorname{Fib}^{c}(c)). b c , d F < 0 implies b c , d F ≤ 0 ( c ∈ Fib c ( c )) .
(b) Assume that b c , d F > 0 b^{F}_{c,d}>0 b c , d F > 0 . By skew-symmetrizability, b d , c F < 0 b^{F}_{d,c}<0 b d , c F < 0 . For any c ^ ∈ Fib c ( c ) \widehat c\in\operatorname{Fib}^{c}(c) c ∈ Fib c ( c ) , applying (a) to the ordered pair ( d , c ) (d,c) ( d , c ) with the fixed representative c ^ \widehat c c gives b d ^ , c ^ F ^ ≤ 0 b^{\widehat F}_{\widehat d,\widehat c}\leq0 b d , c F ≤ 0 for every d ^ ∈ Fib c ( d ) \widehat d\in\operatorname{Fib}^{c}(d) d ∈ Fib c ( d ) . Since B f u l l F ^ B^{\widehat F}_{\mathrm{full}} B full F is skew-symmetrizable with positive symmetrizer, it follows that
b c ^ , d ^ F ^ ≥ 0 ( c ^ ∈ Fib c ( c ) ) . b^{\widehat F}_{\widehat c,\widehat d}\geq0\qquad(\widehat c\in\operatorname{Fib}^{c}(c)). b c , d F ≥ 0 ( c ∈ Fib c ( c )) .
(c) Finally, assume that b c , d F = 0 b^{F}_{c,d}=0 b c , d F = 0 . Then neither a horizontal relation in (i) nor a vertical relation in (ii) of (30) occurs for the pair ( c , d ) (c,d) ( c , d ) . By the fiberwise construction, no pair
( c ^ , d ^ ) , c ^ ∈ Fib c ( c ) , (\widehat c,\widehat d),\qquad\widehat c\in\operatorname{Fib}^{c}(c), ( c , d ) , c ∈ Fib c ( c ) ,
satisfies a corresponding upstairs relation in (i) or (ii) of (30) . Hence
b c ^ , d ^ F ^ = 0 ( c ^ ∈ Fib c ( c ) ) . b^{\widehat F}_{\widehat c,\widehat d}=0\qquad(\widehat c\in\operatorname{Fib}^{c}(c)). b c , d F = 0 ( c ∈ Fib c ( c )) .
Thus the entries
b c ^ , d ^ F ^ ( c ^ ∈ Fib c ( c ) ) b^{\widehat F}_{\widehat c,\widehat d}\qquad(\widehat c\in\operatorname{Fib}^{c}(c)) b c , d F ( c ∈ Fib c ( c ))
have the same weak sign. Together with Theorem 3.15 , this gives
b c , d F = ∑ c ^ ∈ Fib c ( c ) b c ^ , d ^ F ^ , b^{F}_{c,d}=\sum_{\widehat c\in\operatorname{Fib}^{c}(c)}b^{\widehat F}_{\widehat c,\widehat d}, b c , d F = c ∈ Fib c ( c ) ∑ b c , d F ,
this gives
[ b c , d F ] + = ∑ c ^ ∈ Fib c ( c ) [ b c ^ , d ^ F ^ ] + , [ − b c , d F ] + = ∑ c ^ ∈ Fib c ( c ) [ − b c ^ , d ^ F ^ ] + . [b^{F}_{c,d}]_{+}=\sum_{\widehat c\in\operatorname{Fib}^{c}(c)}[b^{\widehat F}_{\widehat c,\widehat d}]_{+},
\qquad
[-b^{F}_{c,d}]_{+}=\sum_{\widehat c\in\operatorname{Fib}^{c}(c)}[-b^{\widehat F}_{\widehat c,\widehat d}]_{+}. [ b c , d F ] + = c ∈ Fib c ( c ) ∑ [ b c , d F ] + , [ − b c , d F ] + = c ∈ Fib c ( c ) ∑ [ − b c , d F ] + .
For the mutation-type case, the next lemma shows that the exchange column at each lifted mutation agrees, up to the preceding permutations, with the corresponding column of the initial seed Σ F ^ \Sigma^{\widehat F} Σ F .
Lemma 5.10. Suppose that B k \mathcal{B}_{k} B k is of mutation type, and let c ^ 1 , … , c ^ r \widehat c_{1},\ldots,\widehat c_{r} c 1 , … , c r be the mutation vertices of B ^ k \widehat{\mathcal{B}}_{k} B k , in their order of occurrence. Immediately before the mutation at c ^ a \widehat c_{a} c a , we have
b c ^ a , c ^ b = 0 ( a < b ≤ r ) . b_{\widehat c_{a},\widehat c_{b}}=0\qquad(a<b\leq r). b c a , c b = 0 ( a < b ≤ r ) . Consequently, mutation at c ^ a \widehat c_{a} c a leaves the c ^ b \widehat c_{b} c b -column unchanged for every b > a b>a b > a . The permutation-type box moves only permute these columns.
Proof. By the proof of Proposition 5.5, the r r r mutation-type moves occur when p ( a ) p^{(a)} p ( a ) and q ( a ) q^{(a)} q ( a ) cross, one for each 1 ≤ a ≤ r 1 \leq a \leq r 1 ≤ a ≤ r . The permutation-type box moves only permute their positions. Hence
{ ı ^ c 1 , … , ı ^ c r } = I i p . \{\widehat{\imath}_{c_1},\ldots,\widehat{\imath}_{c_r}\}=I_{i_p}. { c 1 , … , c r } = I i p .
By admissibility, these colors are pairwise nonadjacent. Therefore, neither a horizontal relation in (i) nor a vertical relation in (ii) of (3.15) can occur between two distinct mutation vertices. Thus
b ^ c a ^ , c b ^ = 0 ( a ≠ b ) \widehat{b}_{\widehat{c_a},\widehat{c_b}}=0 \qquad(a\ne b) b c a , c b = 0 ( a = b )
at the corresponding stage of B ^ k \widehat{\mathcal{B}}_k B k . Now fix a < b a<b a < b . Under mutation at c a ^ \widehat{c_a} c a , the c b ^ \widehat{c_b} c b -column changes according to
b d ^ , c b ^ ′ = b d ^ , c b ^ + [ b d ^ , c a ^ ] + + [ b c a ^ , c b ^ ] + − [ − b d ^ , c a ^ ] + [ − b c a ^ , c b ^ ] + . b'_{\widehat{d},\widehat{c_b}}
=b_{\widehat{d},\widehat{c_b}}
+[b_{\widehat{d},\widehat{c_a}}]_+
+[b_{\widehat{c_a},\widehat{c_b}}]_+
-[-b_{\widehat{d},\widehat{c_a}}]_+
[-b_{\widehat{c_a},\widehat{c_b}}]_+. b d , c b ′ = b d , c b + [ b d , c a ] + + [ b c a , c b ] + − [ − b d , c a ] + [ − b c a , c b ] + .
Since b ^ c a ^ , c b ^ = 0 \widehat{b}_{\widehat{c_a},\widehat{c_b}}=0 b c a , c b = 0 , we obtain
b d ^ , c b ^ ′ = b d ^ , c b ^ b'_{\widehat{d},\widehat{c_b}}=b_{\widehat{d},\widehat{c_b}} b d , c b ′ = b d , c b
for every row d ^ \widehat{d} d . Thus the mutation at c a ^ \widehat{c_a} c a leaves the c b ^ \widehat{c_b} c b -column unchanged. Applying this successively along B ^ k \widehat{\mathcal{B}}_k B k , and observing that the permutation-type box moves only permute box labels and seed positions, proves the assertion. □
We first consider the permutation-type case.
Proposition 5.11. Suppose that B k \mathcal{B}_k B k is of permutation type. Then
ϕ C ( x ^ c ^ ′ ) = x c ′ ( c ′ ∈ F ′ , c ^ ′ ∈ Fib C ′ ( c ′ ) ) . \phi^{\mathcal{C}}(\widehat{x}_{\widehat{c}'})=x_{c'}
\qquad(c'\in\mathcal{F}',\ \widehat{c}'\in\operatorname{Fib}^{\mathcal{C}'}(c')). ϕ C ( x c ′ ) = x c ′ ( c ′ ∈ F ′ , c ′ ∈ Fib C ′ ( c ′ )) . Proof. By Proposition 5.3 , both B k \mathcal{B}_k B k and B ^ k \widehat{\mathcal{B}}_k B k act on the corresponding clusters only by permutations. Proposition 5.6 identifies the permuted blocks with the box fibers for C ′ \mathcal{C}' C ′ . Hence the definition of ϕ C \phi^{\mathcal{C}} ϕ C gives the asserted identities. □
We next consider the mutation-type case.
Proposition 5.12. Suppose that B k \mathcal{B}_k B k is of mutation type. Then
ϕ C ( x ^ d ^ ) = x d ′ ( d ∈ F ′ , d ^ ∈ Fib C ′ ( d ) ) . \phi^{\mathcal{C}}(\widehat{x}_{\widehat{d}})=x'_d
\qquad(d\in\mathcal{F}',\ \widehat{d}\in\operatorname{Fib}^{\mathcal{C}'}(d)). ϕ C ( x d ) = x d ′ ( d ∈ F ′ , d ∈ Fib C ′ ( d )) . Proof. Let c ∈ F c\in\mathcal{F} c ∈ F be the i \mathbf{i} i -box at which the downstairs mutation occurs. By Proposition 5.5, B ^ k \widehat{\mathcal{B}}_k B k contains exactly r r r mutation-type box moves, separated by permutation-type box moves. Let
c ^ 1 , … , c ^ r \widehat{c}_1,\ldots,\widehat{c}_r c 1 , … , c r be the corresponding mutation vertices, in their order of occurrence. We first consider the cluster variables not mutated by B k \mathcal{B}_k B k . The variables in their corresponding fibers are not mutated along B ^ k \widehat{\mathcal{B}}_k B k ; the permutation-type box moves only permute their labels and positions. Hence the definition of ϕ C \phi^{\mathcal{C}} ϕ C gives
ϕ C ( x ^ d ^ ) = x d ′ \phi^{\mathcal{C}}(\widehat{x}_{\widehat{d}})=x'_d ϕ C ( x d ) = x d ′ for every unchanged box d ∈ F d\in\mathcal{F} d ∈ F and each d ^ ∈ Fib C ( d ) \widehat{d}\in\operatorname{Fib}^{\mathcal{C}}(d) d ∈ Fib C ( d ) . It remains to consider the affected fiber. Write the exchange relation at c c c as
x c x c ′ = M + + M − , x_cx_{c'}=M^+ + M^-, x c x c ′ = M + + M − ,
where
M + : = ∏ d ∈ F x d [ b d , c F ] + , M − : = ∏ d ∈ F x d [ − b d , c F ] + . M^{+}:=\prod_{\mathfrak d\in\mathcal F}x_{\mathfrak d}^{[b^{\mathcal F}_{\mathfrak d,\mathfrak c}]_{+}},\qquad M^{-}:=\prod_{\mathfrak d\in\mathcal F}x_{\mathfrak d}^{[-b^{\mathcal F}_{\mathfrak d,\mathfrak c}]_{+}}. M + := d ∈ F ∏ x d [ b d , c F ] + , M − := d ∈ F ∏ x d [ − b d , c F ] + .
and c ′ ∈ F ′ \mathfrak c' \in\mathcal F' c ′ ∈ F ′ is the new box replacing c \mathfrak c c .
Fix 1 ≤ a ≤ r 1\leq a\leq r 1 ≤ a ≤ r and consider the mutation at c ^ a \widehat{\mathfrak c}_a c a . By Lemma 5.10 , after identifying the vertices through the preceding permutation-type box moves, the exchange column used at c ^ a \widehat{\mathfrak c}_a c a is obtained from its corresponding initial exchange column only by relabeling. In particular, none of the preceding mutations changes this column.
Moreover, every previously mutated vertex c ^ b \widehat{\mathfrak c}_b c b satisfies
b c ^ b , c ^ a = 0. b_{\widehat{\mathfrak c}_b,\widehat{\mathfrak c}_a}=0. b c b , c a = 0.
Hence the cluster variables created by the preceding mutations do not occur in the exchange relation at c ^ a \widehat{\mathfrak c}_a c a . Thus, after identifying vertices through the preceding permutations, the exchange relation at c ^ a \widehat{\mathfrak c}_a c a has the form
x ^ c ^ a x ^ c ^ a ′ = M ^ a + + M ^ a − . \widehat{x}_{\widehat{\mathfrak c}_a}\widehat{x}'_{\widehat{\mathfrak c}_a}=\widehat{M}^{+}_a+\widehat{M}^{-}_a. x c a x c a ′ = M a + + M a − .
where
M ^ a + : = ∏ d ^ ∈ F ^ x ^ d ^ [ b d ^ , c ^ a F ^ ] + , M ^ a − : = ∏ d ^ ∈ F ^ x ^ d ^ [ − b d ^ , c ^ a F ^ ] + . \widehat{M}^{+}_a:=\prod_{\widehat{\mathfrak d}\in\widehat{\mathcal F}}\widehat{x}_{\widehat{\mathfrak d}}^{[b^{\widehat{\mathcal F}}_{\widehat{\mathfrak d},\widehat{\mathfrak c}_a}]_{+}},\qquad
\widehat{M}^{-}_a:=\prod_{\widehat{\mathfrak d}\in\widehat{\mathcal F}}\widehat{x}_{\widehat{\mathfrak d}}^{[-b^{\widehat{\mathcal F}}_{\widehat{\mathfrak d},\widehat{\mathfrak c}_a}]_{+}}. M a + := d ∈ F ∏ x d [ b d , c a F ] + , M a − := d ∈ F ∏ x d [ − b d , c a F ] + .
We now apply the fixed Laurent homomorphism ϕ E \phi^{\mathcal E} ϕ E . For each d ∈ F \mathfrak d\in\mathcal F d ∈ F , we have
ϕ E ( x ^ d ^ ) = x d ( d ^ ∈ Fib E ( d ) ) . \phi^{\mathcal E}(\widehat{x}_{\widehat{\mathfrak d}})=x_{\mathfrak d}\qquad(\widehat{\mathfrak d}\in\operatorname{Fib}^{\mathcal E}(\mathfrak d)). ϕ E ( x d ) = x d ( d ∈ Fib E ( d )) .
Hence
ϕ E ( M ^ a + ) = ∏ d ∈ F x d ∑ d ^ ∈ Fib E ( d ) [ b d ^ , c ^ a F ^ ] + . \phi^{\mathcal E}(\widehat{M}^{+}_a)=\prod_{\mathfrak d\in\mathcal F}x_{\mathfrak d}^{\sum_{\widehat{\mathfrak d}\in\operatorname{Fib}^{\mathcal E}(\mathfrak d)}[b^{\widehat{\mathcal F}}_{\widehat{\mathfrak d},\widehat{\mathfrak c}_a}]_{+}}. ϕ E ( M a + ) = d ∈ F ∏ x d ∑ d ∈ Fib E ( d ) [ b d , c a F ] + .
By Lemma 5.9 ,
[ b d , c F ] + = ∑ d ^ ∈ Fib E ( d ) [ b d ^ , c ^ a F ^ ] + . [b^{\mathcal F}_{\mathfrak d,\mathfrak c}]_{+}=\sum_{\widehat{\mathfrak d}\in\operatorname{Fib}^{\mathcal E}(\mathfrak d)}[b^{\widehat{\mathcal F}}_{\widehat{\mathfrak d},\widehat{\mathfrak c}_a}]_{+}. [ b d , c F ] + = d ∈ Fib E ( d ) ∑ [ b d , c a F ] + .
Therefore
ϕ E ( M ^ a + ) = ∏ d ∈ F x d [ b d , c F ] + = M + . \phi^{\mathcal E}(\widehat{M}^{+}_a)=\prod_{\mathfrak d\in\mathcal F}x_{\mathfrak d}^{[b^{\mathcal F}_{\mathfrak d,\mathfrak c}]_{+}}=M^{+}. ϕ E ( M a + ) = d ∈ F ∏ x d [ b d , c F ] + = M + .
Similarly,
[ − b d , c F ] + = ∑ d ^ ∈ Fib E ( d ) [ − b d ^ , c ^ a F ^ ] + , [-b^{\mathcal F}_{\mathfrak d,\mathfrak c}]_{+}=\sum_{\widehat{\mathfrak d}\in\operatorname{Fib}^{\mathcal E}(\mathfrak d)}[-b^{\widehat{\mathcal F}}_{\widehat{\mathfrak d},\widehat{\mathfrak c}_a}]_{+}, [ − b d , c F ] + = d ∈ Fib E ( d ) ∑ [ − b d , c a F ] + ,
and hence ϕ E ( M ^ a − ) = M − \phi^{\mathcal E}(\widehat{M}^{-}_a)=M^{-} ϕ E ( M a − ) = M − .
Since c ^ a \widehat{\mathfrak c}_a c a belongs to the fiber over c \mathfrak c c , we have
ϕ E ( x ^ c ^ a ) = x c . \phi^{\mathcal E}(\widehat{x}_{\widehat{\mathfrak c}_a})=x_{\mathfrak c}. ϕ E ( x c a ) = x c .
Applying ϕ E \phi^{\mathcal E} ϕ E to the exchange relation at c ^ a \widehat{\mathfrak c}_a c a gives
x c ϕ E ( x ^ c ^ a ′ ) = M + + M − . x_{\mathfrak c}\phi^{\mathcal E}(\widehat{x}'_{\widehat{\mathfrak c}_a})=M^{+}+M^{-}. x c ϕ E ( x c a ′ ) = M + + M − .
Comparing this with x c x c ′ = M + + M − x_{\mathfrak c}x_{\mathfrak c'}=M^{+}+M^{-} x c x c ′ = M + + M − , we obtain
ϕ E ( x ^ c ^ a ′ ) = x c ′ . \phi^{\mathcal E}(\widehat{x}'_{\widehat{\mathfrak c}_a})=x_{\mathfrak c'}. ϕ E ( x c a ′ ) = x c ′ .
Since a a a was arbitrary, this holds for all r r r mutation vertices. After the remaining permutation-type box moves, the variables
x ^ c 1 ′ , … , x ^ c r ′ \widehat{x}_{c'_1},\ldots,\widehat{x}_{c'_r} x c 1 ′ , … , x c r ′
are precisely those labelled by the boxes in
Fib E ′ ( c ′ ) . \operatorname{Fib}^{\mathcal{E}'}(c'). Fib E ′ ( c ′ ) .
Together with the unchanged-variable case above, this proves the assertion. □
Corollary 5.13. Define
Fold E : Z F ^ ⟶ Z F , ( Fold E ( g ^ ) ) c = ∑ c ^ ∈ Fib E ( c ) g ^ c ^ . \operatorname{Fold}^{\mathcal{E}}:\mathbb{Z}^{\widehat{F}}\longrightarrow\mathbb{Z}^{F},\qquad
\bigl(\operatorname{Fold}^{\mathcal{E}}(\widehat{g})\bigr)_c
=
\sum_{\widehat{c}\in\operatorname{Fib}^{\mathcal{E}}(c)}
\widehat{g}_{\widehat{c}}. Fold E : Z F ⟶ Z F , ( Fold E ( g ) ) c = c ∈ Fib E ( c ) ∑ g c . Then, for every c ′ ∈ F ′ c'\in F' c ′ ∈ F ′ and c ^ ′ ∈ Fib E ′ ( c ′ ) \widehat{c}'\in\operatorname{Fib}^{\mathcal{E}'}(c') c ′ ∈ Fib E ′ ( c ′ ) ,
deg Σ F ( x c ′ ′ ) = Fold E ( deg Σ ^ F ^ ( x ^ c ^ ′ ) ) . \deg_{\Sigma_F}(x'_{c'})
=
\operatorname{Fold}^{\mathcal{E}}\bigl(\deg_{\widehat{\Sigma}_{\widehat{F}}}(\widehat{x}_{\widehat{c}'})\bigr). deg Σ F ( x c ′ ′ ) = Fold E ( deg Σ F ( x c ′ ) ) . Proof. The proof is the same as that of Proposition 4.22, using Theorem 3.15 and Propositions 5.11 and 5.12 . □
Example 5.14. Consider the folding
A 5 ⟶ C 3 , 1 ^ = ( 1 , 5 ) , 2 ^ = ( 2 , 4 ) , 3 ^ = ( 3 ) . A_5\longrightarrow C_3,\qquad
\widehat{1}=(1,5),\quad\widehat{2}=(2,4),\quad\widehat{3}=(3). A 5 ⟶ C 3 , 1 = ( 1 , 5 ) , 2 = ( 2 , 4 ) , 3 = ( 3 ) .
Take
i = ( 1 , 2 , 1 , 3 , 2 , 3 , 1 ) , T = ( L , L , R , L , L , L ) . \mathbf{i}=(1,2,1,3,2,3,1),\qquad
\mathcal{T}=(\mathcal{L},\mathcal{L},\mathcal{R},\mathcal{L},\mathcal{L},\mathcal{L}). i = ( 1 , 2 , 1 , 3 , 2 , 3 , 1 ) , T = ( L , L , R , L , L , L ) .
The corresponding admissible chain and effective-end sequence are
C = ( [ 6 ] , [ 5 ] , [ 4 , 6 ] , [ 7 ] , [ 3 , 7 ] , [ 2 , 5 ] , [ 1 , 7 ] ) , p ( C ) = ( 6 , 5 , 4 , 7 , 3 , 2 , 1 ) . \mathcal{C}=([6],[5],[4,6],[7],[3,7],[2,5],[1,7]),\qquad
\mathbf{p}(\mathcal{C})=(6,5,4,7,3,2,1). C = ([ 6 ] , [ 5 ] , [ 4 , 6 ] , [ 7 ] , [ 3 , 7 ] , [ 2 , 5 ] , [ 1 , 7 ]) , p ( C ) = ( 6 , 5 , 4 , 7 , 3 , 2 , 1 ) .
Thus, in effective-end order,
F = ( [ 1 , 7 ] , [ 2 , 5 ] , [ 3 , 7 ] , [ 4 , 6 ] , [ 5 ] , [ 6 ] , [ 7 ] ) . \mathcal{F}=([1,7],[2,5],[3,7],[4,6],[5],[6],[7]). F = ([ 1 , 7 ] , [ 2 , 5 ] , [ 3 , 7 ] , [ 4 , 6 ] , [ 5 ] , [ 6 ] , [ 7 ]) .
Consider the box move B 4 \mathcal{B}_4 B 4 . Since
p 4 = 7 , p 5 = 3 , i p 4 = i p 5 = 1 , p_4=7,\qquad p_5=3,\qquad i_{p_4}=i_{p_5}=1, p 4 = 7 , p 5 = 3 , i p 4 = i p 5 = 1 ,
it is of mutation type and replaces
c = [ 7 ] with c ′ = [ 3 ] . c=[7]\quad\text{with}\quad c'=[3]. c = [ 7 ] with c ′ = [ 3 ] .
The corresponding exchange relation is
x [ 7 ] x [ 3 ] ′ = x [ 3 , 7 ] + x [ 5 ] . x_{[7]}x'_{[3]}=x_{[3,7]}+x_{[5]}. x [ 7 ] x [ 3 ] ′ = x [ 3 , 7 ] + x [ 5 ] .
The fixed lift of i \mathbf{i} i is
i ^ = ( 1 , 5 , 2 , 4 , 1 , 5 , 3 , 2 , 4 , 3 , 1 , 5 ) , \widehat{\mathbf{i}}=(1,5,2,4,1,5,3,2,4,3,1,5), i = ( 1 , 5 , 2 , 4 , 1 , 5 , 3 , 2 , 4 , 3 , 1 , 5 ) ,
and the canonical lifted effective-end sequence is
( 10 , 9 , 8 , 7 , 11 , 12 , 6 , 5 , 4 , 3 , 2 , 1 ) . (10,9,8,7,11,12,6,5,4,3,2,1). ( 10 , 9 , 8 , 7 , 11 , 12 , 6 , 5 , 4 , 3 , 2 , 1 ) .
The relevant part of the lifted family F ^ \widehat{\mathcal{F}} F before the move is
Fib E ( [ 3 , 7 ] ) = ( [ 5 , 11 ] , [ 6 , 12 ] ) , Fib E ( [ 5 ] ) = ( [ 8 ] , [ 9 ] ) , Fib E ( [ 7 ] ) = ( [ 11 ] , [ 12 ] ) . \operatorname{Fib}^{\mathcal{E}}([3,7])=([5,11],[6,12]),\qquad
\operatorname{Fib}^{\mathcal{E}}([5])=([8],[9]),\qquad
\operatorname{Fib}^{\mathcal{E}}([7])=([11],[12]). Fib E ([ 3 , 7 ]) = ([ 5 , 11 ] , [ 6 , 12 ]) , Fib E ([ 5 ]) = ([ 8 ] , [ 9 ]) , Fib E ([ 7 ]) = ([ 11 ] , [ 12 ]) .
Hence
ϕ E ( x ^ [ 5 , 11 ] ) = ϕ E ( x ^ [ 6 , 12 ] ) = x [ 3 , 7 ] , \phi^{\mathcal{E}}(\widehat{x}_{[5,11]})=\phi^{\mathcal{E}}(\widehat{x}_{[6,12]})=x_{[3,7]}, ϕ E ( x [ 5 , 11 ] ) = ϕ E ( x [ 6 , 12 ] ) = x [ 3 , 7 ] ,
ϕ E ( x ^ [ 8 ] ) = ϕ E ( x ^ [ 9 ] ) = x [ 5 ] , \phi^{\mathcal{E}}(\widehat{x}_{[8]})=\phi^{\mathcal{E}}(\widehat{x}_{[9]})=x_{[5]}, ϕ E ( x [ 8 ] ) = ϕ E ( x [ 9 ] ) = x [ 5 ] ,
ϕ E ( x ^ [ 11 ] ) = ϕ E ( x ^ [ 12 ] ) = x [ 7 ] . \phi^{\mathcal{E}}(\widehat{x}_{[11]})=\phi^{\mathcal{E}}(\widehat{x}_{[12]})=x_{[7]}. ϕ E ( x [ 11 ] ) = ϕ E ( x [ 12 ] ) = x [ 7 ] .
Here r = 2 r=2 r = 2 and h = 4 h=4 h = 4 , so Proposition 5.5 gives
B ^ 4 = ( B 6 u p , B 5 u p , B 7 u p , B 6 u p ) , \widehat{\mathbb{B}}_4=(\mathbb{B}^{\mathrm{up}}_6,\mathbb{B}^{\mathrm{up}}_5,\mathbb{B}^{\mathrm{up}}_7,\mathbb{B}^{\mathrm{up}}_6), B 4 = ( B 6 up , B 5 up , B 7 up , B 6 up ) ,
with the sequence of types
The two mutation-type moves are
[ 12 ] ⟼ [ 6 ] , [ 11 ] ⟼ [ 5 ] . [12]\longmapsto[6],\qquad[11]\longmapsto[5]. [ 12 ] ⟼ [ 6 ] , [ 11 ] ⟼ [ 5 ] .
Their exchange relations are
x ^ [ 12 ] x ^ [ 6 ] = x ^ [ 6 , 12 ] + x ^ [ 9 ] and x ^ [ 11 ] x ^ [ 5 ] = x ^ [ 5 , 11 ] + x ^ [ 8 ] . \widehat{x}_{[12]}\widehat{x}_{[6]}=\widehat{x}_{[6,12]}+\widehat{x}_{[9]}
\qquad\text{and}\qquad
\widehat{x}_{[11]}\widehat{x}_{[5]}=\widehat{x}_{[5,11]}+\widehat{x}_{[8]}. x [ 12 ] x [ 6 ] = x [ 6 , 12 ] + x [ 9 ] and x [ 11 ] x [ 5 ] = x [ 5 , 11 ] + x [ 8 ] .
Applying ϕ E \phi^{\mathcal{E}} ϕ E gives in both cases
x [ 7 ] ϕ E ( x ^ ) = x [ 3 , 7 ] + x [ 5 ] = x [ 7 ] x [ 3 ] ′ . x_{[7]}\phi^{\mathcal{E}}(\widehat{x})=x_{[3,7]}+x_{[5]}=x_{[7]}x'_{[3]}. x [ 7 ] ϕ E ( x ) = x [ 3 , 7 ] + x [ 5 ] = x [ 7 ] x [ 3 ] ′ .
Therefore
ϕ E ( x ^ [ 5 ] ′ ) = ϕ E ( x ^ [ 6 ] ′ ) = x [ 3 ] ′ . \phi^{\mathcal{E}}(\widehat{x}'_{[5]})=\phi^{\mathcal{E}}(\widehat{x}'_{[6]})=x'_{[3]}. ϕ E ( x [ 5 ] ′ ) = ϕ E ( x [ 6 ] ′ ) = x [ 3 ] ′ .
After applying B ^ 4 \widehat{\mathbb{B}}_4 B 4 ,
Fib E ′ ( [ 3 ] ) = ( [ 5 ] , [ 6 ] ) , \operatorname{Fib}^{\mathcal{E}'}([3])=([5],[6]), Fib E ′ ([ 3 ]) = ([ 5 ] , [ 6 ]) ,
so the two mutated variables are labelled by the boxes in Fib E ′ ( [ 3 ] ) \operatorname{Fib}^{\mathcal{E}'}([3]) Fib E ′ ([ 3 ]) . This example also makes visible the two ingredients in the mutation-type argument: the permutation-type moves only permute the relevant exchange columns, while the positive and negative exchange monomials fold separately.
Folding for i-box moves
We conclude this section by collecting the preceding results for a single box move. Let c k c_k c k be movable, put C ′ = B k ( C ) \mathcal{C}'=\mathbb{B}_k(\mathcal{C}) C ′ = B k ( C ) , and let B ^ k \widehat{\mathbb{B}}_k B k be the sequence of box moves constructed in Propositions 5.3 and 5.5. Let
F = F ( C ) , F ′ = F ( C ′ ) and F ^ = F ( C ^ ) , F ^ ′ = F ( C ^ ′ ) \mathcal{F}=\mathcal{F}(\mathcal{C}),\qquad
\mathcal{F}'=\mathcal{F}(\mathcal{C}')
\qquad\text{and}\qquad
\widehat{\mathcal{F}}=\mathcal{F}(\widehat{\mathcal{C}}),\qquad
\widehat{\mathcal{F}}'=\mathcal{F}(\widehat{\mathcal{C}}') F = F ( C ) , F ′ = F ( C ′ ) and F = F ( C ) , F ′ = F ( C ′ )
be the corresponding families and their canonical lifts. Let
Σ F = ( x F , B F , Λ F ) , Σ F ^ = ( x ^ F , B ^ F , Λ ^ F ) \Sigma^{\mathcal{F}}=(\boldsymbol{x}^{\mathcal{F}},B^{\mathcal{F}},\Lambda^{\mathcal{F}}),\qquad
\Sigma^{\widehat{\mathcal{F}}}=(\widehat{\boldsymbol{x}}^{\mathcal{F}},\widehat{B}^{\mathcal{F}},\widehat{\Lambda}^{\mathcal{F}}) Σ F = ( x F , B F , Λ F ) , Σ F = ( x F , B F , Λ F )
be the seeds before the move, and let
Σ F ′ = ( x F ′ , B F ′ , Λ F ′ ) , Σ F ^ ′ = ( x ^ F ′ , B ^ F ′ , Λ ^ F ′ ) \Sigma^{\mathcal{F}'}=(\boldsymbol{x}^{\mathcal{F}'},B^{\mathcal{F}'},\Lambda^{\mathcal{F}'}),\qquad
\Sigma^{\widehat{\mathcal{F}}'}=(\widehat{\boldsymbol{x}}^{\mathcal{F}'},\widehat{B}^{\mathcal{F}'},\widehat{\Lambda}^{\mathcal{F}'}) Σ F ′ = ( x F ′ , B F ′ , Λ F ′ ) , Σ F ′ = ( x F ′ , B F ′ , Λ F ′ )
be the seeds after the move.
Recall the fixed Laurent homomorphism
ϕ E : L ( Σ F ^ ) ⟶ L ( Σ F ) , x ^ c ^ ⟼ x c ( c ^ ∈ Fib C ( c ) ) . \phi^{\mathcal{E}}:\mathcal{L}(\Sigma^{\widehat{\mathcal{F}}})\longrightarrow\mathcal{L}(\Sigma^{\mathcal{F}}),\qquad
\widehat{x}_{\widehat{c}}\longmapsto x_c\quad(\widehat{c}\in\operatorname{Fib}^{\mathcal{C}}(c)). ϕ E : L ( Σ F ) ⟶ L ( Σ F ) , x c ⟼ x c ( c ∈ Fib C ( c )) .
Theorem 5.15. The sequence B ^ k \widehat{\mathcal{B}}_k B k transforms Σ ^ E \widehat{\Sigma}^{\mathcal{E}} Σ E into Σ ^ E ^ \widehat{\Sigma}^{\widehat{\mathcal{E}}} Σ E . For c ′ , d ′ ∈ F ′ c',d' \in\mathcal{F}' c ′ , d ′ ∈ F ′ and any d ^ ′ ∈ Fib E ′ ( d ′ ) \widehat{d}' \in\operatorname{Fib}^{\mathcal{E}'}(d') d ′ ∈ Fib E ′ ( d ′ ) , we have
b c ′ , d ′ F ′ = ∑ c ^ ′ ∈ Fib E ′ ( c ′ ) b c ^ ′ , d ^ ′ F ^ , Λ c ′ , d ′ F ′ = d i c ′ ∑ c ^ ′ ∈ Fib E ′ ( c ′ ) Λ c ^ ′ , d ^ ′ F ^ . b_{c',d'}^{\mathcal{F}'}=\sum_{\widehat{c}'\in\operatorname{Fib}^{\mathcal{E}'}(c')} b_{\widehat{c}',\widehat{d}'}^{\widehat{\mathcal{F}}},\qquad\Lambda_{c',d'}^{\mathcal{F}'}=d_{i_{c'}}\sum_{\widehat{c}'\in\operatorname{Fib}^{\mathcal{E}'}(c')} \Lambda_{\widehat{c}',\widehat{d}'}^{\widehat{\mathcal{F}}}. b c ′ , d ′ F ′ = c ′ ∈ Fib E ′ ( c ′ ) ∑ b c ′ , d ′ F , Λ c ′ , d ′ F ′ = d i c ′ c ′ ∈ Fib E ′ ( c ′ ) ∑ Λ c ′ , d ′ F . Moreover,
ϕ E ( x ^ c ^ ′ ) = x c ′ ( c ′ ∈ F ′ , c ^ ′ ∈ Fib E ′ ( c ′ ) ) . \phi^{\mathcal{E}}(\widehat{x}_{\widehat{c}'})=x_{c'}\qquad(c'\in\mathcal{F}',\ \widehat{c}'\in\operatorname{Fib}^{\mathcal{E}'}(c')). ϕ E ( x c ′ ) = x c ′ ( c ′ ∈ F ′ , c ′ ∈ Fib E ′ ( c ′ )) . Proof. By Proposition 5.6, the maximal commuting family obtained after B ^ k \widehat{\mathcal{B}}_k B k is F ′ \mathcal{F}' F ′ . The two matrix identities follow from Proposition 5.8 .
For cluster variables, B k \mathcal{B}_k B k is either of permutation or mutation type. The desired identity follows respectively from Proposition 5.11 or Proposition 5.12 .
Global folding of cluster monomials
We now collect the results of § 8 and § 10 and consider the cluster algebras associated with positive braid monoid elements.
Let b ∈ B g + \mathrm{b}\in\mathrm{B}_{\mathfrak{g}}^{+} b ∈ B g + . Thus
i = ( i 1 , … , i l ) ∈ Seq ( b ) \mathbf{i}=(i_1,\ldots,i_l)\in\operatorname{Seq}(\mathrm{b}) i = ( i 1 , … , i l ) ∈ Seq ( b )
if b = r i 1 ⋯ r i l \mathrm{b}=r_{i_1}\cdots r_{i_l} b = r i 1 ⋯ r i l . Any two elements of Seq ( b ) \operatorname{Seq}(\mathrm{b}) Seq ( b ) are connected by a sequence of braid moves.
For i ∈ Seq ( b ) \mathbf{i}\in\operatorname{Seq}(\mathrm{b}) i ∈ Seq ( b ) , choose a maximal commuting family F \mathcal{F} F of i \mathbf{i} i -boxes associated with i \mathbf{i} i , and let Σ F \Sigma^{\mathcal{F}} Σ F be the corresponding box seed. Let
Σ q F = ( X F , B F , Λ F ) \Sigma_q^{\mathcal{F}}=(X^{\mathcal{F}},B^{\mathcal{F}},\Lambda^{\mathcal{F}}) Σ q F = ( X F , B F , Λ F )
be the corresponding quantum seed, where X F = ( X c ∣ c ∈ F ) X^{\mathcal{F}}=(X_c\mid c\in\mathcal{F}) X F = ( X c ∣ c ∈ F ) . By the braid-move and box-move results above, the mutation class of Σ q F \Sigma_q^{\mathcal{F}} Σ q F depends only on b \mathrm{b} b , and not on the choices of i \mathbf{i} i and F \mathcal{F} F .
In view of the preceding results, the following definition makes sense for an arbitrary symmetrizable Cartan datum and an arbitrary positive braid.
Definition 6.1. For b ∈ B g + \mathrm{b}\in\mathrm{B}_{\mathfrak{g}}^{+} b ∈ B g + , we define
A q ( b ) : = A q ( Σ q F ) \mathcal{A}_q(\mathrm{b}):=\mathcal{A}_q(\Sigma_q^{\mathcal{F}}) A q ( b ) := A q ( Σ q F )
to be the quantum cluster algebra associated with b \mathrm{b} b . Specializing at q = 1 q=1 q = 1 , we define
A ( b ) : = A ( Σ F ) ⊆ L ( Σ F ) , Σ F = ( X F , B F ) . \mathcal{A}(\mathrm{b}):=\mathcal{A}(\Sigma^{\mathcal{F}})\subseteq\mathcal{L}(\Sigma^{\mathcal{F}}),\qquad\Sigma^{\mathcal{F}}=(X^{\mathcal{F}},B^{\mathcal{F}}). A ( b ) := A ( Σ F ) ⊆ L ( Σ F ) , Σ F = ( X F , B F ) .
At the commutative level, § 14 identifies A ( b ) \mathcal{A}(\mathrm{b}) A ( b ) with the cluster algebra of Qin’s all-positive signed-word seed associated with i \mathbf{i} i ; see [50 Section 3.1] .
Fix a maximal commuting family F \mathcal{F} F of i \mathbf{i} i -boxes, and choose an admissible chain C \mathcal{C} C realizing F \mathcal{F} F . Let i ^ \widehat{\mathbf{i}} i be the blockwise lift of i \mathbf{i} i , let C ^ \widehat{\mathcal{C}} C be the canonical lift of C \mathcal{C} C , and set F ^ = F ( C ^ ) \widehat{\mathcal{F}}=\mathcal{F}(\widehat{\mathcal{C}}) F = F ( C ) . Then i ^ ∈ Seq ( b ^ ) \widehat{\mathbf{i}}\in\operatorname{Seq}(\widehat{\mathrm{b}}) i ∈ Seq ( b ) for the corresponding upstairs positive braid monoid element b ^ ∈ B g ^ + \widehat{\mathrm{b}}\in\mathrm{B}_{\widehat{\mathfrak{g}}}^{+} b ∈ B g + , and F ^ \widehat{\mathcal{F}} F determines the lifted box seed
Σ ^ F ^ = ( X F ^ , B F ^ ) \widehat{\Sigma}^{\widehat{\mathcal{F}}}=(X^{\widehat{\mathcal{F}}},B^{\widehat{\mathcal{F}}}) Σ F = ( X F , B F )
of A ( b ^ ) \mathcal{A}(\widehat{\mathrm{b}}) A ( b ) .
We fix the surjective Laurent polynomial homomorphism
ϕ F : L ( Σ ^ F ^ ) ↠ L ( Σ F ) , x ^ c ^ ⟼ x c ( c ^ ∈ Fib E ^ ( c ) ) . \phi^{\mathcal{F}}:\mathcal{L}(\widehat{\Sigma}^{\widehat{\mathcal{F}}})\twoheadrightarrow\mathcal{L}(\Sigma^{\mathcal{F}}),\qquad\widehat{x}_{\widehat{c}}\longmapsto x_c\quad(\widehat{c}\in\operatorname{Fib}^{\widehat{\mathcal{E}}}(c)). ϕ F : L ( Σ F ) ↠ L ( Σ F ) , x c ⟼ x c ( c ∈ Fib E ( c )) .
Our main result in this section is that this single Laurent homomorphism simultaneously folds all cluster monomials arising from box seeds associated with arbitrary sequences of b.
Theorem 6.2. Let j ∈ Seq ( b ) \mathbf{j} \in\operatorname{Seq}(\mathbf{b}) j ∈ Seq ( b ) , and let G \mathcal{G} G be any maximal commuting family of i i i -boxes associated with j \mathbf{j} j . 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
ϕ F ( x ^ d ^ ) = x d G ( d ∈ G , d ^ ∈ Fib b ( D ) ) . \phi^{\mathcal{F}}\left(\widehat{x}_{\widehat{d}}\right)=x_{d}^{\mathcal{G}}\qquad\left(d \in\mathcal{G},\ \widehat{d} \in\operatorname{Fib}^{\mathbf{b}}(\mathcal{D})\right). ϕ F ( x d ) = x d G ( d ∈ G , d ∈ Fib b ( D ) ) . In particular, ϕ F \phi^{\mathcal{F}} ϕ F sends every cluster monomial associated with Σ G ^ \Sigma^{\widehat{\mathcal{G}}} Σ G to a cluster monomial associated with Σ G \Sigma^{\mathcal{G}} Σ G .
Proof. We first show that if ϕ F \phi^{\mathcal{F}} ϕ F sends each upstairs cluster variable before a braid move or box move to the corresponding downstairs variable, then the same assertion holds after the move.
Let u ^ \widehat{\mathbf{u}} u be the upstairs cluster before the chosen lifted sequence, and suppose that ϕ F \phi^{\mathcal{F}} ϕ F sends each of its variables to the corresponding downstairs variable. Set v : = ϕ F ( u ^ ) \mathbf{v}:=\phi^{\mathcal{F}}(\widehat{\mathbf{u}}) v := ϕ F ( u ) , where ϕ F \phi^{\mathcal{F}} ϕ F is applied componentwise. Thus v \mathbf{v} v lists the downstairs cluster variables, with repetitions along the fibers. Let z z z be a downstairs cluster variable after the move, and let z ^ \widehat{z} z be a variable in the corresponding upstairs fiber. By the Laurent phenomenon, there are a polynomial P P P and a monomial M M M with nonnegative exponents such that
M ( u ^ ) z ^ = P ( u ^ ) . M(\widehat{\mathbf{u}})\widehat{z}=P(\widehat{\mathbf{u}}). M ( u ) z = P ( u ) .
Every entry of u ^ \widehat{\mathbf{u}} u and the variable z ^ \widehat{z} z belong to L ( Σ F ^ ) \mathcal{L}(\Sigma^{\widehat{\mathcal{F}}}) L ( Σ F ) by the Laurent phenomenon. Applying ϕ F \phi^{\mathcal{F}} ϕ F to this identity therefore gives
M ( v ) ϕ F ( z ^ ) = P ( v ) . M(\mathbf{v})\phi^{\mathcal{F}}(\widehat{z})=P(\mathbf{v}). M ( v ) ϕ F ( z ) = P ( v ) .
By the braid-move results of § 8 and Theorem 5.15 for box moves, the Laurent homomorphism determined by u ^ ↦ v \widehat{\mathbf{u}}\mapsto\mathbf{v} u ↦ v sends z ^ \widehat{z} z to z z z . Hence
P ( v ) = M ( v ) z . P(\mathbf{v})=M(\mathbf{v})z. P ( v ) = M ( v ) z .
Comparing the last two identities and cancelling the nonzero monomial M ( v ) M(\mathbf{v}) M ( v ) gives ϕ F ( z ^ ) = z \phi^{\mathcal{F}}(\widehat{z})=z ϕ F ( z ) = z .
To apply the braid-move results, choose a sequence of box moves from C \mathcal{C} C to the chain with constant R \mathcal{R} R -sequence, whose family is F i \mathcal{F}^{i} F i . By (39) and the convention for sequences of length 1, its canonical lift has family F ^ i \widehat{\mathcal{F}}^{i} F i : the order of the singleton boxes in the first fiber does not affect the family. By Theorem 5.15 and the preceding argument, ϕ F \phi^{\mathcal{F}} ϕ F identifies the cluster variables of Σ F ^ i \Sigma^{\widehat{\mathcal{F}}^{i}} Σ F i with those of Σ F i \Sigma^{\mathcal{F}^{i}} Σ F i .
Choose a sequence of braid moves
i = i ( 0 ) ⟶ i ( 1 ) ⟶ ⋯ ⟶ i ( N ) = j . i=i^{(0)}\longrightarrow i^{(1)}\longrightarrow\cdots\longrightarrow i^{(N)}=j. i = i ( 0 ) ⟶ i ( 1 ) ⟶ ⋯ ⟶ i ( N ) = j .
By the folding results of § 8 , together with Proposition A.6 , the identity ϕ F ( z ^ ) = z \phi^{\mathcal{F}}(\widehat{z})=z ϕ F ( z ) = z is preserved under each braid move. Induction along the chosen braid-move sequence shows that ϕ F \phi^{\mathcal{F}} ϕ F identifies the cluster variables of Σ F ^ j \Sigma^{\widehat{\mathcal{F}}^{j}} Σ F j with those of Σ F j \Sigma^{\mathcal{F}^{j}} Σ F j .
For the fixed sequence j j j , connect F j \mathcal{F}^{j} F j to G \mathcal{G} G by box moves. By Theorem 5.15 , the identity ϕ F ( z ^ ) = z \phi^{\mathcal{F}}(\widehat{z})=z ϕ F ( z ) = z is preserved under each box move. Hence
ϕ F ( x ^ d ^ ) = x d G ( d ∈ G , d ^ ∈ Fib b ( D ) ) . (60) \phi^{\mathcal{F}}\left(\widehat{x}_{\widehat{d}}\right)=x_{d}^{\mathcal{G}}\qquad\left(d \in\mathcal{G},\ \widehat{d} \in\operatorname{Fib}^{\mathbf{b}}(\mathcal{D})\right). \tag*{(60)} ϕ F ( x d ) = x d G ( d ∈ G , d ∈ Fib b ( D ) ) . (60)
Since ϕ F \phi^{\mathcal{F}} ϕ F is an algebra homomorphism,
ϕ F ( ∏ d ^ ∈ G ^ ( x d ^ G ^ ) a d ^ ) = ∏ d ∈ G ( x d G ) ∑ d ^ ∈ Fib D ( d ) a d ^ , \phi^{\mathcal{F}}\left(\prod_{\widehat{\mathsf{d}}\in\widehat{\mathcal{G}}}(x_{\widehat{\mathsf{d}}}^{\widehat{\mathcal{G}}})^{a_{\widehat{\mathsf{d}}}}\right)=\prod_{\mathsf{d}\in\mathcal{G}}(x_{\mathsf{d}}^{\mathcal{G}})^{\sum_{\widehat{\mathsf{d}}\in\operatorname{Fib}^{\mathcal{D}}(\mathsf{d})}a_{\widehat{\mathsf{d}}}}, ϕ F d ∈ G ∏ ( x d G ) a d = d ∈ G ∏ ( x d G ) ∑ d ∈ Fib D ( d ) a d ,
which is a cluster monomial associated with Σ G \Sigma^{\mathcal{G}} Σ G . □
Corollary 6.3. With the notation of Theorem 6.2 , we have
deg Σ F ( x d G ) = Fold E ( deg Σ F ^ ( x d ^ G ^ ) ) \deg_{\Sigma^{\mathcal{F}}}(x_{\mathsf{d}}^{\mathcal{G}})=\operatorname{Fold}^{\mathcal{E}}\left(\deg_{\Sigma^{\widehat{\mathcal{F}}}}(x_{\widehat{\mathsf{d}}}^{\widehat{\mathcal{G}}})\right) deg Σ F ( x d G ) = Fold E ( deg Σ F ( x d G ) ) for every d ∈ G \mathsf{d}\in\mathcal{G} d ∈ G and d ^ ∈ Fib D ( d ) \widehat{\mathsf{d}}\in\operatorname{Fib}^{\mathcal{D}}(\mathsf{d}) d ∈ Fib D ( d ) . The same identity holds for the corresponding cluster monomials.
Proof. The proof is the same as that of Proposition 4.22, using Theorem 6.2 . □
The cluster algebras A ( b ^ ) \mathcal{A}(\widehat{\mathbf{b}}) A ( b ) and A ( b ) \mathcal{A}(\mathbf{b}) A ( b ) are contained in the Laurent polynomial algebras L ( Σ F ^ ) \mathcal{L}(\Sigma^{\widehat{\mathcal{F}}}) L ( Σ F ) and L ( Σ F ) \mathcal{L}(\Sigma^{\mathcal{F}}) L ( Σ F ) , respectively. Theorem 6.2 shows that the Laurent polynomial homomorphism ϕ F \phi^{\mathcal{F}} ϕ F is compatible with all cluster monomials arising from box seeds. Our main commutative folding theorem strengthens this compatibility to the cluster algebras themselves.
Theorem 6.4. The Laurent polynomial homomorphism ϕ F \phi^{\mathcal{F}} ϕ F restricts to a surjective C \mathbb{C} C -algebra homomorphism
ϕ b F : A ( b ^ ) ↠ A ( b ) . \phi_{\mathbf{b}}^{\mathcal{F}}:\mathcal{A}(\widehat{\mathbf{b}})\twoheadrightarrow\mathcal{A}(\mathbf{b}). ϕ b F : A ( b ) ↠ A ( b ) .
We first study two settings in which this folding homomorphism admits quantum-group-theoretic realizations and the simply-laced upstairs algebras admit monoidal categorifications. In § 12 , we consider positive braids represented by reduced sequences and use quantum unipotent coordinate algebras. In § 13 , we treat arbitrary positive braids in finite type through positive-braid algebras and quantum virtual Grothendieck rings. We then prove Theorem 6.4 in full generality in § 14 using the standard-monomial basis [50 ] .
Reduced sequences and quantum unipotent coordinate algebras
Let w ∈ W w\in W w ∈ W and choose a reduced sequence i ∈ Red ( w ) \mathbf{i}\in\operatorname{Red}(w) i ∈ Red ( w ) . In this setting, the quantum cluster algebra associated with i \mathbf{i} i will be identified with the quantum unipotent coordinate algebra A q ( n ( w ) ) \mathcal{A}_{q}(\mathfrak{n}(w)) A q ( n ( w )) . The blockwise lift i ^ \widehat{\mathbf{i}} i is again reduced and determines the corresponding upstairs algebra A q ( n ( w ^ ) ) \mathcal{A}_{q}(\mathfrak{n}(\widehat{w})) A q ( n ( w )) . Both upstairs and downstairs, the dual PBW generators are categorified by cuspidal modules over the corresponding quiver Hecke algebras. Moreover, since the upstairs Cartan datum is simply-laced, the upstairs cluster algebra admits a monoidal categorification by quiver Hecke modules.
In this section, we show that, after specialization at q = 1 q=1 q = 1 , the homomorphism ϕ F \phi^{\mathcal{F}} ϕ F of § 11 restricts to a surjective algebra homomorphism
A ( w ^ ) → ϕ w F A ( w ) . \mathcal{A}(\widehat{w})\xrightarrow{\ \phi_{\mathbf{w}}^{\mathcal{F}}\ }\mathcal{A}(w). A ( w ) ϕ w F A ( w ) .
The proof uses these realizations and the polynomial presentations by dual PBW generators.
Quantum unipotent coordinate algebras
We briefly recall only the notation and properties of quantum unipotent coordinate algebras needed below. For the definitions, notation, and basic properties, we refer to [39 ] (see also [26 , 11 ] ).
Recall that g \mathfrak{g} g is the symmetrizable Kac–Moody algebra and let U q ( g ) U_q(\mathfrak{g}) U q ( g ) be its quantized enveloping algebra over Q ( q 1 / 2 ) \mathbb{Q}(q^{1/2}) Q ( q 1/2 ) . We denote by U q + ( g ) U_q^+(\mathfrak{g}) U q + ( g ) the subalgebra generated by e i e_i e i (i ∈ I i \in I i ∈ I ), and by U A + ( g ) U_{\mathbb{A}}^+(\mathfrak{g}) U A + ( g ) its A \mathbb{A} A -form generated by the divided powers e i ( n ) e_i^{(n)} e i ( n ) , where A = Z [ q ± 1 / 2 ] \mathbb{A} = \mathbb{Z}[q^{\pm1/2}] A = Z [ q ± 1/2 ] .
Set
A q ( n ) : = ⨁ β ∈ Q − Hom Q ( q 1 / 2 ) ( U q + ( g ) − β , Q ( q 1 / 2 ) ) . A_q(\mathfrak{n}) := \bigoplus_{\beta\in Q^-} \operatorname{Hom}_{\mathbb{Q}(q^{1/2})}(U_q^+(\mathfrak{g})_{-\beta},\mathbb{Q}(q^{1/2})). A q ( n ) := β ∈ Q − ⨁ Hom Q ( q 1/2 ) ( U q + ( g ) − β , Q ( q 1/2 )) .
The multiplication dual to the coproduct on U q + ( g ) U_q^+(\mathfrak{g}) U q + ( g ) endows A q ( n ) A_q(\mathfrak{n}) A q ( n ) with an algebra structure. We denote by A A ( n ) A_{\mathbb{A}}(\mathfrak{n}) A A ( n ) the corresponding integral form consisting of those linear forms taking U A + ( g ) U_{\mathbb{A}}^+(\mathfrak{g}) U A + ( g ) into A \mathbb{A} A .
For w ∈ W w \in W w ∈ 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 . Equivalently, it is the subalgebra of A q ( n ) A_q(\mathfrak{n}) A q ( n ) corresponding to Φ + ∩ w Φ − \Phi^+ \cap w\Phi^- Φ + ∩ w Φ − .
For i = ( i 1 , … , i r ) ∈ Red ( w ) \boldsymbol{i}=(i_1,\ldots,i_r)\in\operatorname{Red}(w) i = ( i 1 , … , i r ) ∈ Red ( w ) with w ∈ W w\in W w ∈ W , set
w k i = s i 1 ⋯ s i k , w < k i = s i 1 ⋯ s i k − 1 , β k i = w < k i ( α i k ) . w_k^{\boldsymbol{i}}=s_{i_1}\cdots s_{i_k},\qquad
w_{<k}^{\boldsymbol{i}}=s_{i_1}\cdots s_{i_{k-1}},\qquad
\beta_k^{\boldsymbol{i}}=w_{<k}^{\boldsymbol{i}}(\alpha_{i_k}). w k i = s i 1 ⋯ s i k , w < k i = s i 1 ⋯ s i k − 1 , β k i = w < k i ( α i k ) .
Then we have
Φ + ∩ w Φ − = { β 1 i , … , β r i } . \Phi^+\cap w\Phi^-=\{\beta_1^{\boldsymbol{i}},\ldots,\beta_r^{\boldsymbol{i}}\}. Φ + ∩ w Φ − = { β 1 i , … , β r i } .
For λ ∈ P + \lambda\in P^+ λ ∈ P + and μ , ζ ∈ W λ \mu,\zeta\in W\lambda μ , ζ ∈ W λ , we denote by
D ( μ , ζ ) ∈ A A ( n ) D(\mu,\zeta)\in A_{\mathbb{A}}(\mathfrak{n}) D ( μ , ζ ) ∈ A A ( n )
the corresponding unipotent quantum minor. Recall that D ( μ , ζ ) D(\mu,\zeta) D ( μ , ζ ) is either zero or an element of the upper global basis (or dual canonical basis) B \mathbf{B} B of A q ( n ) A_q(\mathfrak{n}) A q ( n ) ; see [26 ] , Section 9.1.
For i = ( i 1 , … , i r ) ∈ Red ( w ) \boldsymbol{i}=(i_1,\ldots,i_r)\in\operatorname{Red}(w) i = ( i 1 , … , i r ) ∈ Red ( w ) with w ∈ W w\in W w ∈ W and 1 ≤ k ≤ r 1\leq k\leq r 1 ≤ k ≤ r , let
F up ( β k i ) ∈ A q ( n ( w ) ) F^{\operatorname{up}}(\beta_k^{\boldsymbol{i}})\in A_q(\mathfrak{n}(w)) F up ( β k i ) ∈ A q ( n ( w ))
denote the dual PBW root vector corresponding to β k i \beta_k^{\boldsymbol{i}} β k i . With a proper q q q -normalization, [27 ] , Section 5.1.1 gives the exact identity
D ( w k i ϖ i k , w < k i ϖ i k ) = F up ( β k i ) ( 1 ≤ k ≤ r ) . D(w_k^{\boldsymbol{i}}\varpi_{i_k},w_{<k}^{\boldsymbol{i}}\varpi_{i_k})=F^{\operatorname{up}}(\beta_k^{\boldsymbol{i}})\qquad(1\leq k\leq r). D ( w k i ϖ i k , w < k i ϖ i k ) = F up ( β k i ) ( 1 ≤ k ≤ r ) .
Thus, F up ( β k i ) F^{\operatorname{up}}(\beta_k^{\boldsymbol{i}}) F up ( β k i ) belongs to A A ( n ( w ) ) A_{\mathbb{A}}(\mathfrak{n}(w)) A A ( n ( w )) , and A A ( n ( w ) ) A_{\mathbb{A}}(\mathfrak{n}(w)) A A ( n ( w )) is generated by these elements as an A \mathbb{A} A -algebra. The generator indexed by k k k corresponds to the positive root β k i \beta_k^{\boldsymbol{i}} β k i .
We shall mainly work with the specialization at q = 1 q=1 q = 1 . Regarding C \mathbb{C} C as an A \mathbb{A} A -algebra via q 1 / 2 ↦ 1 q^{1/2}\mapsto1 q 1/2 ↦ 1 , we set
A ( n ( w ) ) : = C ⊗ A A A ( n ( w ) ) . A(\mathfrak{n}(w)):=\mathbb{C}\otimes_{\mathbb{A}}A_{\mathbb{A}}(\mathfrak{n}(w)). A ( n ( w )) := C ⊗ A A A ( n ( w )) .
For x ∈ A A ( n ( w ) ) x\in A_{\mathbb{A}}(\mathfrak{n}(w)) x ∈ A A ( n ( w )) , we write
x ∣ q = 1 : = 1 ⊗ x ∈ A ( n ( w ) ) x|_{q=1}:=1\otimes x\in A(\mathfrak{n}(w)) x ∣ q = 1 := 1 ⊗ x ∈ A ( n ( w ))
for its specialization at q 1 / 2 = 1 q^{1/2}=1 q 1/2 = 1 .
By [39 Theorem 4.39] , we have an algebra isomorphism
A ( n ( w ) ) ≅ C [ N ( w ) ] , A(\mathfrak{n}(w))\cong\mathbb{C}[N(w)], A ( n ( w )) ≅ C [ N ( w )] ,
where N ( w ) N(w) N ( w ) is the unipotent algebraic group with Lie algebra
n ( w ) : = ⨁ α ∈ Φ + ∩ w Φ − g − α . \mathfrak{n}(w):=\bigoplus_{\alpha\in\Phi^+\cap w\Phi^-}\mathfrak{g}_{-\alpha}. n ( w ) := α ∈ Φ + ∩ w Φ − ⨁ g − α .
and C [ N ( w ) ] \mathbb{C}[N(w)] C [ N ( w )] denotes its coordinate algebra.
It is known that A A ( n ( w ) ) \mathcal{A}_{\mathbb{A}}(\mathfrak{n}(w)) A A ( n ( w )) admits a quantum cluster algebra structure associated with each i = ( i 1 , … , i r ) ∈ Red ( w ) \boldsymbol{i}=(i_1,\ldots,i_r)\in\operatorname{Red}(w) i = ( i 1 , … , i r ) ∈ Red ( w ) ; see [15 , 17 , 18 , 26 , 48 ] . The corresponding quantum seed has the compatible pair ( B i , Λ i ) (B^{\boldsymbol{i}},\Lambda^{\boldsymbol{i}}) ( B i , Λ i ) and cluster variables, up to the standard q q q -normalization,
D ( w k ϖ i k , ϖ i k ) ( 1 ≤ k ≤ r ) . D(w_k\varpi_{i_k},\varpi_{i_k}) \qquad(1\leq k\leq r). D ( w k ϖ i k , ϖ i k ) ( 1 ≤ k ≤ r ) .
By specializing the above quantum cluster algebra structure at q = 1 q=1 q = 1 , we obtain a cluster algebra structure on A ( n ( w ) ) \mathcal{A}(\mathfrak{n}(w)) A ( n ( w )) whose initial seed is Σ i \Sigma^{\boldsymbol{i}} Σ i . Hence
A ( w ) : = A ( Σ i ) ≃ A ( n ( w ) ) ≃ C [ N ( w ) ] . \mathcal{A}(w):=\mathcal{A}(\Sigma^{\boldsymbol{i}})\simeq\mathcal{A}(\mathfrak{n}(w))\simeq\mathbb{C}[N(w)]. A ( w ) := A ( Σ i ) ≃ A ( n ( w )) ≃ C [ N ( w )] .
Henceforth, we identify A ( w ) \mathcal{A}(w) A ( w ) with A ( n ( w ) ) \mathcal{A}(\mathfrak{n}(w)) A ( n ( w )) via the above isomorphism.
We recall the dual PBW basis associated with i ∈ Red ( w ) \boldsymbol{i}\in\operatorname{Red}(w) i ∈ Red ( w ) . The suitably normalized ordered monomials in the dual PBW root vectors F up ( β k i ) F^{\operatorname{up}}(\beta_k^{\boldsymbol{i}}) F up ( β k i ) form the dual PBW basis of A q ( n ( w ) ) \mathcal{A}_q(\mathfrak{n}(w)) A q ( n ( w )) . In particular,
A q ( n ( w ) ) = Q ( q 1 / 2 ) ⟨ F up ( β 1 i ) , … , F up ( β r i ) ⟩ . \mathcal{A}_q(\mathfrak{n}(w))=\mathbb{Q}(q^{1/2})\langle F^{\operatorname{up}}(\beta_1^{\boldsymbol{i}}),\ldots,F^{\operatorname{up}}(\beta_r^{\boldsymbol{i}})\rangle. A q ( n ( w )) = Q ( q 1/2 ) ⟨ F up ( β 1 i ) , … , F up ( β r i )⟩ .
Moreover, each F up ( β k i ) F^{\operatorname{up}}(\beta_k^{\boldsymbol{i}}) F up ( β k i ) belongs to [39 Proposition 4.23]
B ( w ) : = B ∩ A A ( n ( w ) ) . \mathbf{B}(w):=\mathbf{B}\cap\mathcal{A}_{\mathbb{A}}(\mathfrak{n}(w)). B ( w ) := B ∩ A A ( n ( w )) .
We call
B c l ( w ) : = { b ∣ q = 1 ∣ b ∈ B ( w ) } \mathbf{B}^{\mathrm{cl}}(w):=\{b|_{q=1}\mid b\in\mathbf{B}(w)\} B cl ( w ) := { b ∣ q = 1 ∣ b ∈ B ( w )}
the classical dual canonical basis of A ( n ( w ) ) \mathcal{A}(\mathfrak{n}(w)) A ( n ( w )) .
For 1 ≤ k ≤ r 1\leq k\leq r 1 ≤ k ≤ r , set
f k i : = F up ( β k i ) ∣ q = 1 ∈ A ( n ( w ) ) . f_k^{\boldsymbol{i}}:=F^{\operatorname{up}}(\beta_k^{\boldsymbol{i}})|_{q=1}\in\mathcal{A}(\mathfrak{n}(w)). f k i := F up ( β k i ) ∣ q = 1 ∈ A ( n ( w )) .
Under the identification A ( n ( w ) ) ≃ C [ N ( w ) ] \mathcal{A}(\mathfrak{n}(w))\simeq\mathbb{C}[N(w)] A ( n ( w )) ≃ C [ N ( w )] , the elements f 1 i , … , f r i f_1^{\boldsymbol{i}},\ldots,f_r^{\boldsymbol{i}} f 1 i , … , f r i are the coordinate functions dual to the root vectors corresponding to β 1 i , … , β r i \beta_1^{\boldsymbol{i}},\ldots,\beta_r^{\boldsymbol{i}} β 1 i , … , β r i . By the PBW theorem, their monomials form a C \mathbb{C} C -basis of C [ N ( w ) ] \mathbb{C}[N(w)] C [ N ( w )] . Hence
A ( w ) ≃ A ( n ( w ) ) = C [ f 1 i , … , f r i ] . \mathcal{A}(w)\simeq\mathcal{A}(\mathfrak{n}(w))=\mathbb{C}[f_1^{\boldsymbol{i}},\ldots,f_r^{\boldsymbol{i}}]. A ( w ) ≃ A ( n ( w )) = C [ f 1 i , … , f r i ] .
In particular, A ( w ) \mathcal{A}(w) A ( w ) is a polynomial algebra in r = ℓ ( w ) r=\ell(w) r = ℓ ( w ) variables.
Quantum cluster monomials of A A ( n ( w ) ) \mathcal{A}_{\mathbb{A}}(\mathfrak{n}(w)) A A ( n ( w )) belong, up to q 1 2 Z q^{\frac{1}{2}\mathbb{Z}} q 2 1 Z , to B ( w ) \mathbf{B}(w) B ( w ) ; this follows from [26 ] in the symmetric case and from [48 ] in the symmetrizable case. Consequently, every cluster monomial of A ( w ) \mathcal{A}(w) A ( w ) belongs to B c l ( w ) \mathbf{B}^{\mathrm{cl}}(w) B cl ( w ) via the isomorphism.
Categorification and dual PBW vectors
We briefly recall only the notation and results on quiver Hecke algebras that are needed below. Throughout this subsection, we follow the notation and conventions of [26 , 28 ] , where quiver Hecke algebras, their module categories, and determinantial modules are developed in detail.
Let R ( β ) R(\beta) R ( β ) (β ∈ Q + \beta\in Q^+ β ∈ Q + ) be a quiver Hecke algebra ([37 , 38 , 51 ] ) associated with a Cartan datum. We denote by R ( β ) R(\beta) R ( β ) -gmod the category of finite-dimensional Z \mathbb{Z} Z -graded R ( β ) R(\beta) R ( β ) -modules, and set
R -gmod : = ⨁ β ∈ Q + R ( β ) -gmod . R\text{-gmod}:=\bigoplus_{\beta\in Q^+}R(\beta)\text{-gmod}. R -gmod := β ∈ Q + ⨁ R ( β ) -gmod .
The convolution product makes R R R -gmod into a monoidal category. There is an A \mathbb{A} A -algebra isomorphism [37 , 38 , 51 ]
Ω : A ⊗ Z [ q ± 1 ] K Z [ q ± 1 ] ( R -gmod ) ⟶ ∼ A A ( n ) . \Omega:\mathbb{A}\otimes_{\mathbb{Z}[q^{\pm1}]}K_{\mathbb{Z}[q^{\pm1}]}(R\text{-gmod})\overset{\sim}{\longrightarrow}\mathcal{A}_{\mathbb{A}}(\mathfrak{n}). Ω : A ⊗ Z [ q ± 1 ] K Z [ q ± 1 ] ( R -gmod ) ⟶ ∼ A A ( n ) .
For λ ∈ P + \lambda\in P^{+} λ ∈ P + and μ , ζ ∈ W λ \mu,\zeta\in W\lambda μ , ζ ∈ W λ such that D ( μ , ζ ) ≠ 0 D(\mu,\zeta) \ne0 D ( μ , ζ ) = 0 , we denote by M ( μ , ζ ) M(\mu,\zeta) M ( μ , ζ ) the corresponding determinantial module [28 ] . It is the unique self-dual simple module satisfying
Ω ( [ M ( μ , ζ ) ] ) = D ( μ , ζ ) . \Omega([M(\mu,\zeta)])=D(\mu,\zeta). Ω ([ M ( μ , ζ )]) = D ( μ , ζ ) .
Let w ∈ W w \in W w ∈ W and i = ( i 1 , … , i r ) ∈ Red ( w ) \boldsymbol{i}=(i_1,\ldots,i_r)\in\operatorname{Red}(w) i = ( i 1 , … , i r ) ∈ Red ( w ) . For 1 ≤ k ≤ r 1\leq k\leq r 1 ≤ k ≤ r , set
S k i : = M ( w k i ϖ i k , w < k i ϖ i k ) . S_k^{\boldsymbol{i}}:=M(w_k^{\boldsymbol{i}}\varpi_{i_k},w_{<k}^{\boldsymbol{i}}\varpi_{i_k}). S k i := M ( w k i ϖ i k , w < k i ϖ i k ) .
Then
− wt ( S k i ) = β k i and Ω ( [ S k i ] ) = F up ( β k i ) . -\operatorname{wt}(S_k^{\boldsymbol{i}})=\beta_k^{\boldsymbol{i}}\quad\text{and}\quad\Omega([S_k^{\boldsymbol{i}}])=F^{\operatorname{up}}(\beta_k^{\boldsymbol{i}}). − wt ( S k i ) = β k i and Ω ([ S k i ]) = F up ( β k i ) .
The module S k i S_k^{\boldsymbol{i}} S k i is the k k k -th cuspidal module associated with i \boldsymbol{i} i .
The modules S k i S_k^{\boldsymbol{i}} S k i are the categorical counterparts of the dual PBW root generators associated with i \boldsymbol{i} i ; see [27 Section 5.1] .
Recall that, for an i \boldsymbol{i} i -box [ x , y ] [x,y] [ x , y ] , the corresponding determinantial module is
M i [ x , y ] = M ( w y i ϖ i y , w < x i ϖ i x ) ; M^{\boldsymbol{i}}[x,y]=M(w_y^{\boldsymbol{i}}\varpi_{i_y},w_{<x}^{\boldsymbol{i}}\varpi_{i_x}); M i [ x , y ] = M ( w y i ϖ i y , w < x i ϖ i x ) ;
(see [26 ] and [27 Section 5.1] ). Therefore, for a singleton box [ k ] [k] [ k ] , we have M i [ k ] = S k i M^{\boldsymbol{i}}[k]=S_k^{\boldsymbol{i}} M i [ k ] = S k i , and the corresponding cluster variable is precisely f k i f_k^{\boldsymbol{i}} f k i .
We now apply the preceding discussion to the folding setting. Recall that i ^ \widehat{\boldsymbol{i}} i is the blockwise lift of i \boldsymbol{i} i . Since i \boldsymbol{i} i is reduced, so is i ^ \widehat{\boldsymbol{i}} i by Corollary 2.5 . For 1 ≤ k ≤ r 1\leq k\leq r 1 ≤ k ≤ r and 1 ≤ s ≤ e i k 1\leq s\leq e_{i_k} 1 ≤ s ≤ e i k , set
f ^ k , s : = F up ( β ^ k , s ) ∣ q = 1 ∈ A ( w ^ ) , \widehat{f}_{k,s}:=F^{\operatorname{up}}(\widehat{\beta}_{k,s})\big|_{q=1}\in\mathcal{A}(\widehat{w}), f k , s := F up ( β k , s ) q = 1 ∈ A ( w ) ,
where i ^ ∈ Red ( w ^ ) \widehat{\boldsymbol{i}}\in\operatorname{Red}(\widehat{w}) i ∈ Red ( w ) for w ^ ∈ W ^ \widehat{w}\in\widehat{W} w ∈ W .
Lemma 7.1. For every 1 ≤ k ≤ r 1\leq k\leq r 1 ≤ k ≤ r , the cluster variable associated with the singleton i \boldsymbol{i} i -box [ k ] [k] [ k ] is f k i f_k^{\boldsymbol{i}} f k i . Moreover, the fiber of [ k ] [k] [ k ] consists of the singleton i \boldsymbol{i} i -boxes associated with i ^ \widehat{\boldsymbol{i}} i
[ κ ( k , 1 ) ] , … , [ κ ( k , e i k ) ] , [\kappa(k,1)],\ldots,[\kappa(k,e_{i_k})], [ κ ( k , 1 )] , … , [ κ ( k , e i k )] , whose corresponding cluster variables are
f ^ k , 1 , … , f ^ k , e i k . \widehat{f}_{k,1},\ldots,\widehat{f}_{k,e_{i_k}}. f k , 1 , … , f k , e i k . Furthermore,
A ( w ) = C [ f 1 i , … , f r i ] and A ( w ^ ) = C [ f ^ k , s ∣ 1 ≤ k ≤ r , 1 ≤ s ≤ e i k ] . \mathcal{A}(w)=\mathbb{C}[f_1^{\boldsymbol{i}},\ldots,f_r^{\boldsymbol{i}}]\quad\text{and}\quad\mathcal{A}(\widehat{w})=\mathbb{C}[\widehat{f}_{k,s}\mid1\leq k\leq r,\ 1\leq s\leq e_{i_k}]. A ( w ) = C [ f 1 i , … , f r i ] and A ( w ) = C [ f k , s ∣ 1 ≤ k ≤ r , 1 ≤ s ≤ e i k ] .
Proof. By [28 ] (see also [27 Section 5.1] ), the module associated with an i \boldsymbol{i} i -box [ x , y ] [x,y] [ x , y ] is
M i [ x , y ] = M ( w y i ϖ i y , w < x i ϖ i x ) M^{\boldsymbol{i}}[x,y]=M(w_y^{\boldsymbol{i}}\varpi_{i_y},w_{<x}^{\boldsymbol{i}}\varpi_{i_x}) M i [ x , y ] = M ( w y i ϖ i y , w < x i ϖ i x )
and hence M i [ k ] = M ( w k i ϖ i k , w < k i ϖ i k ) = S k i M^{\boldsymbol{i}}[k]=M(w_k^{\boldsymbol{i}}\varpi_{i_k},w_{<k}^{\boldsymbol{i}}\varpi_{i_k})=S_k^{\boldsymbol{i}} M i [ k ] = M ( w k i ϖ i k , w < k i ϖ i k ) = S k i .
By the preceding discussion, the corresponding cluster variable is f k i f_k^{\boldsymbol{i}} f k i .
Let F \mathcal{F} F be a maximal commuting family containing [ k ] [k] [ k ] , choose an admissible chain C \mathcal{C} C realizing F \mathcal{F} F , and let F ^ \widehat{\mathcal{F}} F be its canonical lift. By Lemma 3.13 ,
Fib C ^ ( [ k ] ) = { [ κ ( k , 1 ) ] , … , [ κ ( k , e i k ) ] } . \operatorname{Fib}^{\widehat{\mathcal{C}}}([k])=\{[\kappa(k,1)],\ldots,[\kappa(k,e_{i_k})]\}. Fib C ([ k ]) = {[ κ ( k , 1 )] , … , [ κ ( k , e i k )]} .
Since i ^ \widehat{\boldsymbol{i}} i is reduced, applying the first assertion to i ^ \widehat{\boldsymbol{i}} i shows that the corresponding upstairs cluster variables are
f ^ k , 1 , … , f ^ k , e i k . \widehat{f}_{k,1},\ldots,\widehat{f}_{k,e_{i_k}}. f k , 1 , … , f k , e i k .
Finally, the PBW generation recalled above gives
A ( w ) = C [ f 1 i , … , f r i ] . \mathcal{A}(w)=\mathbb{C}[f^{i}_{1},\ldots,f^{i}_{r}]. A ( w ) = C [ f 1 i , … , f r i ] .
Applying the same argument to i ^ ∈ Red ( w ^ ) \hat{\boldsymbol{i}}\in\operatorname{Red}(\hat{w}) i ^ ∈ Red ( w ^ ) , we obtain
A ( w ^ ) = C [ f ^ k , s ∣ 1 ≤ k ≤ r , 1 ≤ s ≤ e i k ] , \mathcal{A}(\hat{w})=\mathbb{C}[\hat{f}_{k,s}\mid1\leq k\leq r,\ 1\leq s\leq e_{i_k}], A ( w ^ ) = C [ f ^ k , s ∣ 1 ≤ k ≤ r , 1 ≤ s ≤ e i k ] ,
which completes the proof. □ \square □
The folding homomorphism via PBW generators
We now realize the folding homomorphism in terms of dual PBW generators. Fix a maximal commuting family F \mathcal{F} F of i \boldsymbol{i} i -boxes, choose an admissible chain C \mathcal{C} C realizing F \mathcal{F} F , and let F ^ \hat{\mathcal{F}} F ^ be its canonical lift. Recall from § 11 the fixed surjective Laurent polynomial homomorphism
ϕ F : L ( Σ F ^ ) ↠ L ( Σ F ) , x ^ c ^ ⟼ x c ( c ^ ∈ Fib C ^ ( c ) ) . \phi^{\mathcal{F}}:\mathcal{L}(\Sigma^{\hat{\mathcal{F}}})\twoheadrightarrow\mathcal{L}(\Sigma^{\mathcal{F}}),\qquad\hat{x}_{\hat{c}}\longmapsto x_c\quad(\hat{c}\in\operatorname{Fib}^{\hat{\mathcal{C}}}(c)). ϕ F : L ( Σ F ^ ) ↠ L ( Σ F ) , x ^ c ^ ⟼ x c ( c ^ ∈ Fib C ^ ( c )) .
Since
A ( w ^ ) ⊂ L ( Σ F ^ ) and A ( w ) ⊂ L ( Σ F ) , \mathcal{A}(\hat{w})\subset\mathcal{L}(\Sigma^{\hat{\mathcal{F}}})\qquad\text{and}\qquad\mathcal{A}(w)\subset\mathcal{L}(\Sigma^{\mathcal{F}}), A ( w ^ ) ⊂ L ( Σ F ^ ) and A ( w ) ⊂ L ( Σ F ) ,
it remains to show that ϕ F \phi^{\mathcal{F}} ϕ F restricts to a surjective homomorphism from A ( w ^ ) \mathcal{A}(\hat{w}) A ( w ^ ) onto A ( w ) \mathcal{A}(w) A ( w ) .
Proposition 7.2. For every 1 ≤ k ≤ r 1\leq k\leq r 1 ≤ k ≤ r and 1 ≤ s ≤ e i k 1\leq s\leq e_{i_k} 1 ≤ s ≤ e i k , we have
ϕ F ( f ^ k , s ) = f k i . \phi^{\mathcal{F}}(\hat{f}_{k,s})=f^i_k. ϕ F ( f ^ k , s ) = f k i .
Proof. Fix 1 ≤ k ≤ r 1\leq k\leq r 1 ≤ k ≤ r , and choose a maximal commuting family G \mathcal{G} G of i \boldsymbol{i} i -boxes containing the singleton [ k ] [k] [ k ] . By Lemma 7.1 , the cluster variable associated with [ k ] [k] [ k ] is f k i f^i_k f k i , while its lifted fiber consists of
[ κ ( k , 1 ) ] , … , [ κ ( k , e i k ) ] , [\kappa(k,1)],\ldots,[\kappa(k,e_{i_k})], [ κ ( k , 1 )] , … , [ κ ( k , e i k )] ,
with corresponding cluster variables
f ^ k , 1 , … , f ^ k , e i k . \hat{f}_{k,1},\ldots,\hat{f}_{k,e_{i_k}}. f ^ k , 1 , … , f ^ k , e i k .
Since i \boldsymbol{i} i remains fixed, F \mathcal{F} F and G \mathcal{G} G are connected by i \boldsymbol{i} i -box moves. Hence repeated application of Theorem 5.15 gives
ϕ F ( f ^ k , s ) = f k i ( 1 ≤ s ≤ e i k ) , \phi^{\mathcal{F}}(\hat{f}_{k,s})=f^i_k\qquad(1\leq s\leq e_{i_k}), ϕ F ( f ^ k , s ) = f k i ( 1 ≤ s ≤ e i k ) ,
as desired. □ \square □
Theorem 7.3. Let w ∈ W w\in W w ∈ W and i ∈ Red ( w ) \boldsymbol{i}\in\operatorname{Red}(w) i ∈ Red ( w ) . Then
ϕ w F : = ϕ F ∣ A ( w ^ ) : A ( w ^ ) ⟶ A ( w ) \phi^{\mathcal{F}}_w:=\left.\phi^{\mathcal{F}}\right|_{\mathcal{A}(\hat{w})}:\mathcal{A}(\hat{w})\longrightarrow\mathcal{A}(w) ϕ w F := ϕ F A ( w ^ ) : A ( w ^ ) ⟶ A ( w ) is a well-defined surjective algebra homomorphism. Under the polynomial presentations by dual PBW generators, it is determined by
ϕ w F ( f ^ k , s ) = f k i ( 1 ≤ k ≤ r , 1 ≤ s ≤ e i k ) . \phi^{\mathcal{F}}_w(\hat{f}_{k,s})=f^i_k\qquad(1\leq k\leq r,\ 1\leq s\leq e_{i_k}). ϕ w F ( f ^ k , s ) = f k i ( 1 ≤ k ≤ r , 1 ≤ s ≤ e i k ) .
Proof. By Lemma 7.1 , we have
A ( w ^ ) = C [ f ^ k , s ∣ 1 ≤ k ≤ r , 1 ≤ s ≤ e i k ] and A ( w ) = C [ f 1 i , … , f r i ] . \mathcal{A}(\hat{w})=\mathbb{C}[\hat{f}_{k,s}\mid1\leq k\leq r,\ 1\leq s\leq e_{i_k}]
\qquad\text{and}\qquad
\mathcal{A}(w)=\mathbb{C}[f^i_1,\ldots,f^i_r]. A ( w ^ ) = C [ f ^ k , s ∣ 1 ≤ k ≤ r , 1 ≤ s ≤ e i k ] and A ( w ) = C [ f 1 i , … , f r i ] .
By Proposition 7.2 ,
ϕ F ( f ^ k , s ) = f k i ( 1 ≤ k ≤ r , 1 ≤ s ≤ e i k ) . \phi^{\mathcal{F}}(\hat{f}_{k,s})=f^i_k\qquad(1\leq k\leq r,\ 1\leq s\leq e_{i_k}). ϕ F ( f ^ k , s ) = f k i ( 1 ≤ k ≤ r , 1 ≤ s ≤ e i k ) .
It follows that
ϕ w F ( A ( w ^ ) ) ⊆ A ( w ) , \phi_w^F\bigl(\mathcal{A}(\widehat{w})\bigr) \subseteq\mathcal{A}(w), ϕ w F ( A ( w ) ) ⊆ A ( w ) ,
so that the restriction ϕ w F \phi_w^F ϕ w F is well-defined. Moreover, for every 1 ≤ k ≤ r 1 \le k \le r 1 ≤ k ≤ r , we have
f k i = ϕ w F ( f ^ k , 1 ) , f_k^i=\phi_w^F(\widehat{f}_{k,1}), f k i = ϕ w F ( f k , 1 ) ,
and hence all algebra generators of A ( w ) \mathcal{A}(w) A ( w ) belong to the image of ϕ w F \phi_w^F ϕ w F . Therefore
ϕ w F ( A ( w ^ ) ) = A ( w ) , \phi_w^F\bigl(\mathcal{A}(\widehat{w})\bigr)=\mathcal{A}(w), ϕ w F ( A ( w ) ) = A ( w ) ,
which proves the surjectivity. □
Corollary 7.4. Let i ∈ Red ( w ) i \in\operatorname{Red}(w) i ∈ Red ( w ) , let G \mathcal{G} G be any maximal commuting family of i i i -boxes, and let G ^ \widehat{\mathcal{G}} G be its canonical lift. Then
ϕ w F ( ∏ c ^ ∈ G ^ ( x ^ c ^ ) a c ^ ) = ∏ c ∈ G ( x c ) ∑ c ^ ∈ Fib ( c ) a c ^ ∈ B c l ( w ) \phi_w^F\left(\prod_{\widehat{c}\in\widehat{\mathcal{G}}}(\widehat{x}_{\widehat{c}})^{a_{\widehat{c}}}\right)
=\prod_{c\in\mathcal{G}}(x_c)^{\sum_{\widehat{c}\in\operatorname{Fib}(c)}a_{\widehat{c}}}
\in\mathbf{B}^{\mathrm{cl}}(w) ϕ w F c ∈ G ∏ ( x c ) a c = c ∈ G ∏ ( x c ) ∑ c ∈ Fib ( c ) a c ∈ B cl ( w ) for every ( a c ^ ) c ^ ∈ G ^ ∈ Z ≥ 0 G ^ (a_{\widehat{c}})_{\widehat{c}\in\widehat{\mathcal{G}}}\in\mathbb{Z}_{\geq0}^{\widehat{\mathcal{G}}} ( a c ) c ∈ G ∈ Z ≥ 0 G . In particular, ϕ w F \phi_w^F ϕ w F sends every cluster monomial associated with Σ ^ \widehat{\Sigma} Σ to an element of $\mathbf{B}^{\mathrm{cl}}(w).
Proof. This follows from Theorem 6.2 and the fact that cluster monomials of A ( w ) \mathcal{A}(w) A ( w ) belong to B c l ( w ) \mathbf{B}^{\mathrm{cl}}(w) B cl ( w ) . □
Remark 7.5. The preceding folding result also admits a quantum interpretation at the level of canonical bases. Recall that quantum cluster monomials belong, up to the standard q q q -normalization, to B \mathbf{B} B . In the symmetric case this follows from the monoidal categorification by quiver Hecke algebras; see [26 ] , while the symmetrizable case follows from [48 ] .
Let z ^ \widehat{z} z be a quantum cluster monomial associated with a canonical lifted box seed, and let z z z be the corresponding downstairs quantum cluster monomial. Then
z ^ ∈ q Z / 2 B ^ , z ∈ q Z / 2 B . \widehat{z}\in q^{\mathbb{Z}/2}\widehat{\mathbf{B}},\qquad z\in q^{\mathbb{Z}/2}\mathbf{B}. z ∈ q Z /2 B , z ∈ q Z /2 B . Moreover, their g g g -vectors satisfy
deg i ( z ) = Fold i ( deg i ^ ( z ^ ) ) , \deg_i(z)=\operatorname{Fold}^i\bigl(\deg_{\widehat{i}}(\widehat{z})\bigr), deg i ( z ) = Fold i ( deg i ( z ) ) , by the g g g -vector folding established above. Thus Theorem 6.2 gives a folding correspondence between the classical dual canonical basis elements z ^ ∣ q = 1 \widehat{z}|_{q=1} z ∣ q = 1 and z ∣ q = 1 z|_{q=1} z ∣ q = 1 .
We emphasize that this interpretation does not arise from an algebra homomorphism between the corresponding quantum tori. In general, the fiberwise identification of quantum torus generators does not preserve their q q q -commutation relations. Rather, the quantum correspondence is detected through the folding of g g g -vectors. After specialization at q = 1 q=1 q = 1 , this obstruction disappears, and the correspondence is realized by the Laurent algebra homomorphism ϕ F \phi^F ϕ F .
Positive braids of finite Cartan type and quantum virtual Grothendieck rings
Throughout this section, we assume that g \mathfrak{g} g is of finite type, but the sequence representing b ∈ B g + \mathbf{b} \in B_{\mathfrak{g}}^{+} b ∈ B g + need not be reduced. In this setting, we identify the associated quantum cluster algebra A q ( b ) \mathcal{A}_{q}(\mathbf{b}) A q ( b ) with the positive-braid subalgebra A ( b ) \mathscr{A}(\mathbf{b}) A ( b ) defined through the braid group action on the bosonic extension, also known as the quantum virtual Grothendieck ring. For the simply-laced upstairs algebra A ( b ^ ) \mathscr{A}(\widehat{\mathbf{b}}) A ( b ) , we use its monoidal categorification by the category C g ^ ( b ^ ) \mathcal{C}_{\widehat{\mathfrak{g}}}(\widehat{\mathbf{b}}) C g ( b ) .
Using these realizations, we show that, after specialization at q = 1 q=1 q = 1 , the ambient Grothendieck-ring homomorphism π \pi π restricts to a surjective algebra homomorphism
π b : A ( b ^ ) ↠ A ( b ) \pi_{\mathbf{b}}:\mathcal{A}(\widehat{\mathbf{b}})\twoheadrightarrow\mathcal{A}(\mathbf{b}) π b : A ( b ) ↠ A ( b )
and agrees with ϕ b F \phi_{\mathbf{b}}^{\mathcal{F}} ϕ b F .
Quantum virtual Grothendieck rings
We briefly recall only the notation and properties needed below. Throughout this section, we follow the notation and conventions of [11 , 25 , 24 , 50 ] , where the quantum virtual Grothendieck rings associated with g \mathfrak{g} g and its simply-laced unfolding g ^ \widehat{\mathfrak{g}} g are developed in detail.
Fix t ∈ C × t \in\mathbb{C}^{\times} t ∈ C × which is not a root of unity. Let U t ( L g ) U_{t}(L\mathfrak{g}) U t ( L g ) be the quantum loop algebra associated with g \mathfrak{g} g , and let C g \mathcal{C}_{\mathfrak{g}} C g be the category of finite-dimensional type 1 U t ( L g ) U_{t}(L\mathfrak{g}) U t ( L g ) -modules. It is a rigid monoidal category, and its Grothendieck ring K ( C g ) K(\mathcal{C}_{\mathfrak{g}}) K ( C g ) is commutative.
The simple objects of C g \mathcal{C}_{\mathfrak{g}} C g are parametrized by dominant monomials in the variables y i , a y_{i,a} y i , a (i ∈ I i \in I i ∈ I , a ∈ C × a \in\mathbb{C}^{\times} a ∈ C × ). For a dominant monomial m m m , we denote by L ( m ) L(m) L ( m ) the corresponding simple module. The t t t -character homomorphism of Frenkel–Reshetikhin [5 , 9 , 10 ] 3 gives an injective algebra homomorphism
χ t : K ( C g ) ↪ Z [ y i , a ± 1 ∣ i ∈ I , a ∈ C × ] . \chi_{t}:K(\mathcal{C}_{\mathfrak{g}})\hookrightarrow\mathbb{Z}[y_{i,a}^{\pm1}\mid i\in I,\ a\in\mathbb{C}^{\times}]. χ t : K ( C g ) ↪ Z [ y i , a ± 1 ∣ i ∈ I , a ∈ C × ] .
We first consider the simply-laced Lie algebra g ^ \widehat{\mathfrak{g}} g . Choose a Γ \Gamma Γ -invariant parity function
ϵ : I ^ ⟶ { 0 , 1 } , ϵ ı ≡ ϵ ȷ + 1 ( m o d 2 ) if ı ∼ ȷ , \epsilon:\widehat{I}\longrightarrow\{0,1\},\qquad\epsilon_{\imath}\equiv\epsilon_{\jmath}+1\pmod{2}\quad\text{if }\imath\sim\jmath, ϵ : I ⟶ { 0 , 1 } , ϵ ≡ ϵ + 1 ( mod 2 ) if ∼ ,
and set
Δ ^ 0 g ^ : = { ( ı , p ) ∈ I ^ × Z ∣ p ≡ ϵ ı ( m o d 2 ) } . \widehat{\Delta}_{0}^{\widehat{\mathfrak{g}}}:=\{(\imath,p)\in\widehat{I}\times\mathbb{Z}\mid p\equiv\epsilon_{\imath}\pmod{2}\}. Δ 0 g := {( , p ) ∈ I × Z ∣ p ≡ ϵ ( mod 2 )} .
For ( ı , p ) ∈ Δ ^ 0 g ^ (\imath,p)\in\widehat{\Delta}_{0}^{\widehat{\mathfrak{g}}} ( , p ) ∈ Δ 0 g , we write
y ^ ı , p : = y ı , t p . \widehat{y}_{\imath,p}:=y_{\imath,t^{p}}. y , p := y , t p .
Following Hernandez–Leclerc, let C g ^ 0 \mathcal{C}_{\widehat{\mathfrak{g}}}^{0} C g 0 be the Serre subcategory of C g ^ \mathcal{C}_{\widehat{\mathfrak{g}}} C g whose simple objects have dominant monomials in the variables y ^ ı , p \widehat{y}_{\imath,p} y , p (( ı , p ) ∈ Δ ^ 0 g ^ (\imath,p)\in\widehat{\Delta}_{0}^{\widehat{\mathfrak{g}}} ( , p ) ∈ Δ 0 g ).
The category C g ^ 0 \mathcal{C}_{\widehat{\mathfrak{g}}}^{0} C g 0 is a rigid monoidal subcategory of C g ^ \mathcal{C}_{\widehat{\mathfrak{g}}} C g and is often referred to as the Hernandez–Leclerc category, or the skeleton category. Indeed, every simple object of C g ^ \mathcal{C}_{\widehat{\mathfrak{g}}} C g can be expressed as a commuting tensor product of spectral-parameter shifts of simple objects in C g ^ 0 \mathcal{C}_{\widehat{\mathfrak{g}}}^{0} C g 0 (see [21 , 23 , 11 ] ).
The t t t -character homomorphism restricts to an injective algebra homomorphism
\chi_t: K(\mathcal{C}^{0}_{\widehat{\mathfrak g}}) \lhook\joinrel\longrightarrow\mathcal{Y}_{\mathbb{Z}}(\widehat{\mathfrak g}) := \mathbb{Z}[{\widehat{y}}_{i,p}^{\pm1} \mid(i,p) \in\widehat{\Delta}^{\mathfrak g}_{0}].
Via the t t t -character homomorphism, we identify K ( C g ^ 0 ) K(\mathcal{C}^{0}_{\widehat{\mathfrak g}}) K ( C g 0 ) with its image in Y Z ( g ^ ) \mathcal{Y}_{\mathbb{Z}}(\widehat{\mathfrak g}) Y Z ( g ) and write
K Z ( g ^ ) : = χ t ( K ( C g ^ 0 ) ) ⊂ Y Z ( g ^ ) . \mathcal{K}_{\mathbb{Z}}(\widehat{\mathfrak g}) := \chi_t(K(\mathcal{C}^{0}_{\widehat{\mathfrak g}})) \subset\mathcal{Y}_{\mathbb{Z}}(\widehat{\mathfrak g}). K Z ( g ) := χ t ( K ( C g 0 )) ⊂ Y Z ( g ) .
We use the complexified classical rings
Y ( g ^ ) : = C ⊗ Z Y Z ( g ^ ) ≃ C [ y ^ i , p ± 1 ∣ ( i , p ) ∈ Δ ^ 0 g ] , K ( g ^ ) : = C ⊗ Z K Z ( g ^ ) ⊂ Y ( g ^ ) . \mathcal{Y}(\widehat{\mathfrak g}) := \mathbb{C} \otimes_{\mathbb{Z}} \mathcal{Y}_{\mathbb{Z}}(\widehat{\mathfrak g}) \simeq\mathbb{C}[{\widehat{y}}_{i,p}^{\pm1} \mid(i,p) \in\widehat{\Delta}^{\mathfrak g}_{0}], \qquad\mathcal{K}(\widehat{\mathfrak g}) := \mathbb{C} \otimes_{\mathbb{Z}} \mathcal{K}_{\mathbb{Z}}(\widehat{\mathfrak g}) \subset\mathcal{Y}(\widehat{\mathfrak g}). Y ( g ) := C ⊗ Z Y Z ( g ) ≃ C [ y i , p ± 1 ∣ ( i , p ) ∈ Δ 0 g ] , K ( g ) := C ⊗ Z K Z ( g ) ⊂ Y ( g ) .
Thus K ( g ^ ) \mathcal{K}(\widehat{\mathfrak g}) K ( g ) is the complexified Grothendieck ring of C g ^ 0 \mathcal{C}^{0}_{\widehat{\mathfrak g}} C g 0 realized as a ring of Laurent polynomials.
We recall the virtual counterpart associated with the possibly non-simply-laced Lie algebra g \mathfrak g g . The Γ \Gamma Γ -invariant parity function ϵ \epsilon ϵ descends to I I I by setting
ϵ i : = ϵ ι ( ι ∈ I i ) . \epsilon_i := \epsilon_{\iota} \qquad(\iota\in I_i). ϵ i := ϵ ι ( ι ∈ I i ) .
Accordingly, we set
Δ ~ 0 g : = { ( i , p ) ∈ I × Z ∣ p ≡ ϵ i ( m o d 2 ) } , Y Z ( g ) : = Z [ y i , p ± 1 ∣ ( i , p ) ∈ Δ ~ 0 g ] . \widetilde{\Delta}^{\mathfrak g}_{0} := \{(i,p) \in I \times\mathbb{Z} \mid p \equiv\epsilon_i \pmod{2}\}, \qquad\mathcal{Y}_{\mathbb{Z}}(\mathfrak g) := \mathbb{Z}[y_{i,p}^{\pm1} \mid(i,p) \in\widetilde{\Delta}^{\mathfrak g}_{0}]. Δ 0 g := {( i , p ) ∈ I × Z ∣ p ≡ ϵ i ( mod 2 )} , Y Z ( g ) := Z [ y i , p ± 1 ∣ ( i , p ) ∈ Δ 0 g ] .
We use its complexification
Y ( g ) : = C ⊗ Z Y Z ( g ) ≃ C [ y i , p ± 1 ∣ ( i , p ) ∈ Δ ~ 0 g ] . \mathcal{Y}(\mathfrak g) := \mathbb{C} \otimes_{\mathbb{Z}} \mathcal{Y}_{\mathbb{Z}}(\mathfrak g) \simeq\mathbb{C}[y_{i,p}^{\pm1} \mid(i,p) \in\widetilde{\Delta}^{\mathfrak g}_{0}]. Y ( g ) := C ⊗ Z Y Z ( g ) ≃ C [ y i , p ± 1 ∣ ( i , p ) ∈ Δ 0 g ] .
The quotient map I ^ ↠ I \widehat{I} \twoheadrightarrow I I ↠ I induces a surjective algebra homomorphism
π Z : Y Z ( g ^ ) ↠ Y Z ( g ) , y ^ ι , p ⟼ y i , p ( ι ∈ I i ) . \pi_{\mathbb{Z}}: \mathcal{Y}_{\mathbb{Z}}(\widehat{\mathfrak g}) \twoheadrightarrow\mathcal{Y}_{\mathbb{Z}}(\mathfrak g), \qquad\widehat{y}_{\iota,p} \longmapsto y_{i,p} \qquad(\iota\in I_i). π Z : Y Z ( g ) ↠ Y Z ( g ) , y ι , p ⟼ y i , p ( ι ∈ I i ) .
By extension of scalars, it induces a surjective C \mathbb{C} C -algebra homomorphism
π : Y ( g ^ ) ↠ Y ( g ) , y ^ ι , p ⟼ y i , p ( ι ∈ I i ) . (61) \pi: \mathcal{Y}(\widehat{\mathfrak g}) \twoheadrightarrow\mathcal{Y}(\mathfrak g), \qquad\widehat{y}_{\iota,p} \longmapsto y_{i,p} \qquad(\iota\in I_i). \tag*{(61)} π : Y ( g ) ↠ Y ( g ) , y ι , p ⟼ y i , p ( ι ∈ I i ) . (61)
Following [25 ] , we define the virtual Grothendieck ring of g \mathfrak g g and its complexification by
K Z ( g ) : = π Z ( K Z ( g ^ ) ) ⊂ Y Z ( g ) , K ( g ) : = C ⊗ Z K Z ( g ) = π ( K ( g ^ ) ) ⊂ Y ( g ) . \mathcal{K}_{\mathbb{Z}}(\mathfrak g) := \pi_{\mathbb{Z}}(\mathcal{K}_{\mathbb{Z}}(\widehat{\mathfrak g})) \subset\mathcal{Y}_{\mathbb{Z}}(\mathfrak g), \qquad\mathcal{K}(\mathfrak g) := \mathbb{C} \otimes_{\mathbb{Z}} \mathcal{K}_{\mathbb{Z}}(\mathfrak g) = \pi(\mathcal{K}(\widehat{\mathfrak g})) \subset\mathcal{Y}(\mathfrak g). K Z ( g ) := π Z ( K Z ( g )) ⊂ Y Z ( g ) , K ( g ) := C ⊗ Z K Z ( g ) = π ( K ( g )) ⊂ Y ( g ) .
Thus π \pi π restricts to a surjective algebra homomorphism
π : K ( g ^ ) ↠ K ( g ) . \pi: \mathcal{K}(\widehat{\mathfrak g}) \twoheadrightarrow\mathcal{K}(\mathfrak g). π : K ( g ) ↠ K ( g ) .
See [25 ] for the construction and basic properties of K ( g ) \mathcal{K}(\mathfrak g) K ( g ) .
We recall the fundamental and Kirillov–Reshetikhin polynomials. For ( i , p ) , ( i , s ) ∈ Δ ~ 0 g (i,p),(i,s) \in\widetilde{\Delta}^{\mathfrak g}_{0} ( i , p ) , ( i , s ) ∈ Δ 0 g with p ≤ s p \leq s p ≤ s , set
m ( i ) [ p , s ] : = y i , p y i , p + 2 ⋯ y i , s . m^{(i)}[p,s] := y_{i,p}y_{i,p+2} \cdots y_{i,s}. m ( i ) [ p , s ] := y i , p y i , p + 2 ⋯ y i , s .
We set m ( i ) [ p , s ] : = 1 m^{(i)}[p,s] := 1 m ( i ) [ p , s ] := 1 if p > s p > s p > s . We call m ( i ) [ p , s ] m^{(i)}[p,s] m ( i ) [ p , s ] a Kirillov–Reshetikhin (K R KR K R ) monomial. Associated with it is a distinguished element
F ( m ( i ) [ p , s ] ) ∈ K ( g ) , F(m^{(i)}[p,s]) \in\mathcal{K}(\mathfrak g), F ( m ( i ) [ p , s ]) ∈ K ( g ) ,
which we call the corresponding K R KR K R -polynomial. In particular,
F ( y i , p ) = F ( m ( i ) [ p ] ) F(y_{i,p}) = F(m^{(i)}[p]) F ( y i , p ) = F ( m ( i ) [ p ])
is called a fundamental polynomial.
When g \mathfrak{g} g is simply laced, these elements are the t t t -characters of the corresponding Kirillov–Reshetikhin modules. In particular, F ( y i , p ) F(y_{i,p}) F ( y i , p ) is the t t t -character of the fundamental module L ( y i , p ) L(y_{i,p}) L ( y i , p ) . Thus, for ı ∈ I ^ \imath\in\widehat{I} ∈ I , the Laurent polynomial
F ( y ^ ı , p ) ∈ R ( g ^ ) F(\widehat{y}_{\imath,p}) \in\mathcal{R}(\widehat{\mathfrak{g}}) F ( y , p ) ∈ R ( g )
is the t t t -character of the fundamental U t ( L g ^ ) U_t(L\widehat{\mathfrak{g}}) U t ( L g ) -module L ( y ^ ı , p ) L(\widehat{y}_{\imath,p}) L ( y , p ) .
The Frenkel–Mukhin algorithm provides an algorithmic construction of these Laurent polynomials. In particular, the KR-polynomials above are well defined and can be obtained from their simply-laced counterparts by folding.
More precisely, for ı ∈ I i \imath\in I_i ∈ I i and ( ı , p ) , ( ı , s ) ∈ △ ^ 0 g (\imath,p),( \imath,s) \in\widehat{\triangle}^{\mathfrak{g}}_0 ( , p ) , ( , s ) ∈ △ 0 g , set
m ^ ( i ) [ p , s ] : = y ^ ı , p y ^ ı , p + 2 ⋯ y ^ ı , s . \widehat{m}^{(i)}[p,s]:=\widehat{y}_{\imath,p}\widehat{y}_{\imath,p+2}\cdots\widehat{y}_{\imath,s}. m ( i ) [ p , s ] := y , p y , p + 2 ⋯ y , s .
Then
π ( F ( m ^ ( i ) [ p , s ] ) ) = F ( m ( i ) [ p , s ] ) . \pi\left(F\left(\widehat{m}^{(i)}[p,s]\right)\right)=F\left(m^{(i)}[p,s]\right). π ( F ( m ( i ) [ p , s ] ) ) = F ( m ( i ) [ p , s ] ) .
In particular,
π ( F ( y ^ ı , p ) ) = F ( y i , p ) \pi\left(F\left(\widehat{y}_{\imath,p}\right)\right)=F\left(y_{i,p}\right) π ( F ( y , p ) ) = F ( y i , p )
(see [25 Corollary 3.10 and Remark 5.40] for instance).
We next consider the quantum deformation. Recall that D = diag ( d i ∣ i ∈ I ) \mathrm{D}=\operatorname{diag}(d_i\mid i\in I) D = diag ( d i ∣ i ∈ I ) is the symmetrizer of the Cartan matrix C = ( c i j ) i , j ∈ I \mathrm{C}=(c_{ij})_{i,j\in I} C = ( c ij ) i , j ∈ I . For an indeterminate t t t , set
c i j ( t ) : = δ i j ( t + t − 1 ) + δ i ≠ j c i j , C ( t ) : = ( c i j ( t ) ) i , j ∈ I . c_{ij}(t):=\delta_{ij}(t+t^{-1})+\delta_{i\ne j}c_{ij},\qquad\mathrm{C}(t):=(c_{ij}(t))_{i,j\in I}. c ij ( t ) := δ ij ( t + t − 1 ) + δ i = j c ij , C ( t ) := ( c ij ( t ) ) i , j ∈ I .
We call C ( t ) \mathrm{C}(t) C ( t ) the t t t -quantized Cartan matrix of g \mathfrak{g} g . Set
B ( t ) : = C ( t ) D − 1 . \mathrm{B}(t):=\mathrm{C}(t)\mathrm{D}^{-1}. B ( t ) := C ( t ) D − 1 .
Then B ( t ) \mathrm{B}(t) B ( t ) is symmetric. Regarding B ( t ) \mathrm{B}(t) B ( t ) as a matrix over Q ( t ) \mathbb{Q}(t) Q ( t ) , let
B ~ ( t ) = ( B ~ i j ( t ) ) i , j ∈ I : = B ( t ) − 1 . \widetilde{\mathrm{B}}(t)=\left(\widetilde{\mathrm{B}}_{ij}(t)\right)_{i,j\in I}:=\mathrm{B}(t)^{-1}. B ( t ) = ( B ij ( t ) ) i , j ∈ I := B ( t ) − 1 .
For i , j ∈ I i,j\in I i , j ∈ I , we write
B ~ i j ( t ) = ∑ u ∈ Z b ~ i j ( u ) t u ∈ Q ( ( t ) ) \widetilde{\mathrm{B}}_{ij}(t)=\sum_{u\in\mathbb{Z}}\widetilde{b}_{ij}(u)t^u\in\mathbb{Q}((t)) B ij ( t ) = u ∈ Z ∑ b ij ( u ) t u ∈ Q (( t ))
for its Laurent expansion at t = 0 t=0 t = 0 .
For ( i , p ) , ( j , s ) ∈ △ ^ 0 g (i,p),(j,s)\in\widehat{\triangle}^{\mathfrak{g}}_0 ( i , p ) , ( j , s ) ∈ △ 0 g , set
N ( i , p ; j , s ) : = b ~ i j ( p − s − 1 ) − b ~ i j ( s − p − 1 ) − b ~ i j ( p − s + 1 ) + b ~ i j ( s − p + 1 ) . \mathcal{N}(i,p;j,s):=\widetilde{b}_{ij}(p-s-1)-\widetilde{b}_{ij}(s-p-1)-\widetilde{b}_{ij}(p-s+1)+\widetilde{b}_{ij}(s-p+1). N ( i , p ; j , s ) := b ij ( p − s − 1 ) − b ij ( s − p − 1 ) − b ij ( p − s + 1 ) + b ij ( s − p + 1 ) .
Since B ~ ( t ) \widetilde{\mathrm{B}}(t) B ( t ) is symmetric, we have N ( i , p ; j , s ) = − N ( j , s ; i , p ) \mathcal{N}(i,p;j,s)=-\mathcal{N}(j,s;i,p) N ( i , p ; j , s ) = − N ( j , s ; i , p ) .
Let Y q ( g ) \mathcal{Y}_q(\mathfrak{g}) Y q ( g ) be the A \mathbb{A} A -algebra generated by
Y i , p ± 1 ( ( i , p ) ∈ △ ^ 0 g ) Y_{i,p}^{\pm1}\qquad((i,p)\in\widehat{\triangle}^{\mathfrak{g}}_0) Y i , p ± 1 (( i , p ) ∈ △ 0 g )
subject to the relations
Y i , p Y i , p − 1 = Y i , p − 1 Y i , p = 1 and Y i , p Y j , s = q N ( i , p ; j , s ) Y j , s Y i , p . Y_{i,p}Y_{i,p}^{-1}=Y_{i,p}^{-1}Y_{i,p}=1\qquad\text{and}\qquad Y_{i,p}Y_{j,s}=q^{\mathcal{N}(i,p;j,s)}Y_{j,s}Y_{i,p}. Y i , p Y i , p − 1 = Y i , p − 1 Y i , p = 1 and Y i , p Y j , s = q N ( i , p ; j , s ) Y j , s Y i , p .
We call Y q ( g ) \mathcal{Y}_q(\mathfrak{g}) Y q ( g ) the quantum torus associated with C ( t ) \mathrm{C}(t) C ( t ) .
Specializing q 1 / 2 q^{1/2} q 1/2 to 1 1 1 , we recover the commutative Laurent polynomial rings over Z \mathbb{Z} Z introduced above. More precisely, there are surjective algebra homomorphisms
ev q = 1 : Y q ( g ) ↠ Y Z ( g ) , Y i , p ⟼ y i , p . \operatorname{ev}_{q=1}:\mathcal{Y}_q(\mathfrak{g})\twoheadrightarrow\mathcal{Y}_{\mathbb{Z}}(\mathfrak{g}),\qquad Y_{i,p}\longmapsto y_{i,p}. ev q = 1 : Y q ( g ) ↠ Y Z ( g ) , Y i , p ⟼ y i , p .
Passing to the commutative cluster-algebra setting means composing this specialization with extension of scalars from Z \mathbb{Z} Z to C \mathbb{C} C .
We recall the quantum Grothendieck rings inside these quantum tori. When g ^ \widehat{\mathfrak{g}} g is of simply-laced finite type, the quantum Grothendieck ring
K q ( g ^ ) ⊆ Y q ( g ^ ) \mathcal{K}_q(\widehat{\mathfrak{g}}) \subseteq\mathcal{Y}_q(\widehat{\mathfrak{g}}) K q ( g ) ⊆ Y q ( g )
is the quantum deformation of the integral form K Z ( g ^ ) \mathcal{K}_{\mathbb{Z}}(\widehat{\mathfrak{g}}) K Z ( g ) constructed geometrically by Nakajima and Varagnolo–Vasserot; see [43 , 45 , 55 ] . An algebraic approach using deformed screening operators was developed by Hernandez [20 ] .
For every dominant monomial M M M , there is a distinguished element
F q ( M ) ∈ K q ( g ^ ) F_q(M) \in\mathcal{K}_q(\widehat{\mathfrak{g}}) F q ( M ) ∈ K q ( g )
having M M M as its unique dominant monomial. The elements
{ F q ( M ) ∣ M is dominant } \{F_q(M) \mid M\text{ is dominant}\} { F q ( M ) ∣ M is dominant }
form an A \mathcal{A} A -basis of K q ( g ^ ) \mathcal{K}_q(\widehat{\mathfrak{g}}) K q ( g ) . Moreover, specialization at q = 1 q=1 q = 1 recovers the corresponding elements of the ordinary integral Grothendieck ring:
ev q = 1 ( F q ( M ) ) = F ( ev q = 1 ( M ) ) . \operatorname{ev}_{q=1}\left(F_q(M)\right)=F\left(\operatorname{ev}_{q=1}(M)\right). ev q = 1 ( F q ( M ) ) = F ( ev q = 1 ( M ) ) .
For the possibly non-simply-laced Lie algebra g \mathfrak{g} g , we consider the quantum virtual Grothendieck ring
K q ( g ) ⊆ Y q ( g ) , \mathcal{K}_q(\mathfrak{g}) \subseteq\mathcal{Y}_q(\mathfrak{g}), K q ( g ) ⊆ Y q ( g ) ,
introduced by Kashiwara–Oh [35 ] and further developed in [25 , 24 ] . It admits an A \mathcal{A} A -basis { F q ( M ) ∣ M is dominant } \{F_q(M) \mid M\text{ is dominant}\} { F q ( M ) ∣ M is dominant } , whose specialization at q = 1 q=1 q = 1 gives the F F F -basis of K Z ( g ) \mathcal{K}_{\mathbb{Z}}(\mathfrak{g}) K Z ( g ) (see [25 Corollary 5.28] ):
ev q = 1 ( F q ( M ) ) = F ( ev q = 1 ( M ) ) . \operatorname{ev}_{q=1}\left(F_q(M)\right)=F\left(\operatorname{ev}_{q=1}(M)\right). ev q = 1 ( F q ( M ) ) = F ( ev q = 1 ( M ) ) .
For a quantum KR-monomial
M ( i ) [ p , s ] = Y i , p Y i , p + 2 ⋯ Y i , s , M^{(i)}[p,s]=Y_{i,p}Y_{i,p+2}\cdots Y_{i,s}, M ( i ) [ p , s ] = Y i , p Y i , p + 2 ⋯ Y i , s ,
where the product is understood with the standard q q q -normalization, we call
F q ( M ( i ) [ p , s ] ) ∈ K q ( g ) F_q\left(M^{(i)}[p,s]\right)\in\mathcal{K}_q(\mathfrak{g}) F q ( M ( i ) [ p , s ] ) ∈ K q ( g )
the corresponding quantum KR-polynomial . In particular, F q ( Y i , p ) F_q(Y_{i,p}) F q ( Y i , p ) is called a quantum fundamental polynomial . By the specialization property above,
ev q = 1 ( F q ( M ( i ) [ p , s ] ) ) = F ( m ( i ) [ p , s ] ) . \operatorname{ev}_{q=1}\left(F_q\left(M^{(i)}[p,s]\right)\right)=F\left(m^{(i)}[p,s]\right). ev q = 1 ( F q ( M ( i ) [ p , s ] ) ) = F ( m ( i ) [ p , s ] ) .
The same notation and terminology will be used for g ^ \widehat{\mathfrak{g}} g .
The quantum Grothendieck ring K q ( g ^ ) \mathcal{K}_q(\widehat{\mathfrak{g}}) K q ( g ) also admits the basis
{ L q ( M ) ∣ M is dominant } , \{L_q(M)\mid M\text{ is dominant}\}, { L q ( M ) ∣ M is dominant } ,
constructed by Nakajima and referred to as the canonical basis . Its specialization at q = 1 q=1 q = 1 recovers the t t t -characters of the corresponding simple modules:
ev q = 1 ( L q ( M ) ) = χ t ( L ( m ) ) , m = ev q = 1 ( M ) . \operatorname{ev}_{q=1}\left(L_q(M)\right)=\chi_t\left(L(m)\right),\qquad m=\operatorname{ev}_{q=1}(M). ev q = 1 ( L q ( M ) ) = χ t ( L ( m ) ) , m = ev q = 1 ( M ) .
Moreover, if M M M is a KR-monomial, then
F q ( M ) = L q ( M ) . F_q(M)=L_q(M). F q ( M ) = L q ( M ) .
These properties follow from Nakajima’s results [43 ] .
For a dominant monomial M M M , let E q ( M ) E_q(M) E q ( M ) be the product of quantum fundamental polynomials with the multiplicities prescribed by M M M , ordered by increasing spectral label and normalized to have coefficient 1 1 1 at M M M ; see [25 (5.21)] . These elements form the standard basis of K q ( g ) \mathcal{K}_q(\mathfrak{g}) K q ( g ) .
For the quantum virtual Grothendieck ring K q ( g ) \mathcal{K}_q(\mathfrak{g}) K q ( g ) , the corresponding L q L_q L q -basis { L q ( M ) ∣ M is dominant } \{L_q(M)\mid M\text{ is dominant}\} { L q ( M ) ∣ M is dominant } was constructed in [25 ] . It is characterized by the bar-invariance and unitriangularity properties with respect to the standard basis { E q ( M ) ∣ M is dominant } \{E_q(M)\mid M\text{ is dominant}\} { E q ( M ) ∣ M is dominant } . A fundamental result concerning quantum KR-polynomials is
F q ( M ( i ) [ p , s ] ) = L q ( M ( i ) [ p , s ] ) F_q\left(M^{(i)}[p,s]\right)=L_q\left(M^{(i)}[p,s]\right) F q ( M ( i ) [ p , s ] ) = L q ( M ( i ) [ p , s ] )
for every KR-monomial M ( i ) [ p , s ] M^{(i)}[p,s] M ( i ) [ p , s ] . In the simply-laced case, this is due to Nakajima, while the general finite-type case was proved by Qin; see [44 ] and [50 Theorem 1.1] .
The rings K ( g ) \mathcal{K}(\mathfrak{g}) K ( g ) and K q ( g ) \mathcal{K}_q(\mathfrak{g}) K q ( g ) are generated by their fundamental polynomials. More precisely,
K ( g ) = C [ F ( y i , p ) ∣ ( i , p ) ∈ △ ~ 0 g ] and K q ( g ) = A ⟨ F q ( Y i , p ) ∣ ( i , p ) ∈ △ ~ 0 g ⟩ . \mathcal{K}(\mathfrak{g})=\mathbb{C}\left[F(y_{i,p})\mid(i,p)\in\widetilde{\triangle}^{\mathfrak{g}}_0\right]\qquad\text{and}\qquad\mathcal{K}_q(\mathfrak{g})=\mathbb{A}\left\langle F_q(Y_{i,p})\mid(i,p)\in\widetilde{\triangle}^{\mathfrak{g}}_0\right\rangle. K ( g ) = C [ F ( y i , p ) ∣ ( i , p ) ∈ △ 0 g ] and K q ( g ) = A ⟨ F q ( Y i , p ) ∣ ( i , p ) ∈ △ 0 g ⟩ .
For i ∈ I i\in I i ∈ I , let h h h denote the Coxeter number of g \mathfrak{g} g , and let i ∗ ∈ I i^*\in I i ∗ ∈ I be determined by w 0 ( α i ) = − α i ∗ w_0(\alpha_i)=-\alpha_{i^*} w 0 ( α i ) = − α i ∗ , where w 0 w_0 w 0 is the longest element of the Weyl group. We denote by D q D_q D q the duality automorphism, given on the ambient variables by
D q ( Y i , p ) = Y i ∗ , p + h . D_q(Y_{i,p})=Y_{i^*,p+h}. D q ( Y i , p ) = Y i ∗ , p + h .
We denote by D D D its specialization at q 1 / 2 = 1 q^{1/2}=1 q 1/2 = 1 , so that
D ( y i , p ) = y i ∗ , p + h . D(y_{i,p})=y_{i^*,p+h}. D ( y i , p ) = y i ∗ , p + h .
We use the same notation for their natural extensions to the ambient quantum torus and its skew field of fractions.
A Dynkin quiver of g \mathfrak{g} g is a pair Q = ( △ , ξ ) Q=(\triangle,\xi) Q = ( △ , ξ ) , where △ \triangle △ is the Dynkin diagram of g \mathfrak{g} g endowed with an orientation and
ξ : I ⟶ Z \xi:I\longrightarrow\mathbb{Z} ξ : I ⟶ Z
is a height function satisfying
ξ i = ξ j + 1 whenever i → j . \xi_i=\xi_j+1\qquad\text{whenever }i\to j. ξ i = ξ j + 1 whenever i → j .
We choose ξ \xi ξ compatibly with the parity function fixed above, so that
ξ i ≡ ϵ i ( m o d 2 ) ( i ∈ I ) . \xi_i\equiv\epsilon_i\pmod{2}\qquad(i\in I). ξ i ≡ ϵ i ( mod 2 ) ( i ∈ I ) .
Set
△ ~ ≤ ξ g : = { ( i , p ) ∈ △ ~ 0 g ∣ p ≤ ξ i } . \widetilde{\triangle}^{\mathfrak{g}}_{\leq\xi}:=\{(i,p)\in\widetilde{\triangle}^{\mathfrak{g}}_0\mid p\leq\xi_i\}. △ ≤ ξ g := {( i , p ) ∈ △ 0 g ∣ p ≤ ξ i } .
For a Dynkin quiver Q = ( △ , ξ ) Q=(\triangle,\xi) Q = ( △ , ξ ) , set
K q ( g ) ≤ ξ : = A ⟨ F q ( Y i , p ) ∣ ( i , p ) ∈ △ ~ ≤ ξ g ⟩ . \mathcal{K}_q(\mathfrak{g})_{\leq\xi}:=\mathbb{A}\left\langle F_q(Y_{i,p})\mid(i,p)\in\widetilde{\triangle}^{\mathfrak{g}}_{\leq\xi}\right\rangle. K q ( g ) ≤ ξ := A ⟨ F q ( Y i , p ) ∣ ( i , p ) ∈ △ ≤ ξ g ⟩ .
It is known that
K q ( g ) ≤ ξ = ⨁ M A F q ( M ) , \mathcal{K}_q(\mathfrak{g})_{\leq\xi}=\bigoplus_M\mathbb{A}\,F_q(M), K q ( g ) ≤ ξ = M ⨁ A F q ( M ) ,
where the sum runs over all dominant monomials M M M involving only the variables Y i , p Y_{i,p} Y i , p with ( i , p ) ∈ △ ~ ≤ ξ g (i,p)\in\widetilde{\triangle}^{\mathfrak{g}}_{\leq\xi} ( i , p ) ∈ △ ≤ ξ g ; see [23 , 2 , 11 , 25 ] .
For a Dynkin quiver Q = ( △ , ξ ) Q=(\triangle,\xi) Q = ( △ , ξ ) , a vertex i ∈ I i\in I i ∈ I is called a source if
ξ i > ξ j for every j ∈ I adjacent to i . \xi_i>\xi_j \qquad\text{for every }j\in I\text{ adjacent to }i. ξ i > ξ j for every j ∈ I adjacent to i .
If i i i is a source, let s i Q = ( △ , s i ξ ) s_iQ=(\triangle,s_i\xi) s i Q = ( △ , s i ξ ) , where
( s i ξ ) j = ξ j − 2 δ i j . (s_i\xi)_j=\xi_j-2\delta_{ij}. ( s i ξ ) j = ξ j − 2 δ ij .
We call s i Q s_iQ s i Q the reflection of Q Q Q at i i i .
Let Q = ( △ , ξ ) Q=(\triangle,\xi) Q = ( △ , ξ ) be a Dynkin quiver. A sequence i = ( i 1 , i 2 , … ) \mathbf{i}=(i_1,i_2,\ldots) i = ( i 1 , i 2 , … ) of finite or infinite length is said to be adapted to Q Q Q if i 1 i_1 i 1 is a source of Q Q Q and, inductively, i k i_k i k is a source of the Dynkin quiver obtained by successively reflecting at i 1 , … , i k − 1 i_1,\ldots,i_{k-1} i 1 , … , i k − 1 .
We shall also use the lifted Dynkin quiver associated with Q Q Q . Define
Q ^ = ( △ ^ , ξ ^ ) , ξ ^ ı ^ : = ξ i ( ı ^ ∈ I i ) , \widehat{Q}=(\widehat{\triangle},\widehat{\xi}),\qquad\widehat{\xi}_{\hat{\imath}}:=\xi_i\quad(\hat{\imath}\in I_i), Q = ( △ , ξ ) , ξ ^ := ξ i ( ^ ∈ I i ) ,
where each edge of △ ^ \widehat{\triangle} △ lying above an oriented edge i → j i\to j i → j of △ \triangle △ is oriented from I i I_i I i to I j I_j I j . Then ξ ^ \widehat{\xi} ξ is a Γ \Gamma Γ -invariant height function on △ ^ \widehat{\triangle} △ . We call Q ^ \widehat{Q} Q the lifted Dynkin quiver associated with Q Q Q . Its quotient by Γ \Gamma Γ is Q Q Q .
Lemma 8.1. Let Q = ( △ , ξ ) Q=(\triangle,\xi) Q = ( △ , ξ ) be a Dynkin quiver and let Q ^ = ( △ ^ , ξ ^ ) \widehat{Q}=(\widehat{\triangle},\widehat{\xi}) Q = ( △ , ξ ) be its lifted Dynkin quiver . If i = ( i 1 , i 2 , … ) \mathbf{i}=(i_1,i_2,\ldots) i = ( i 1 , i 2 , … ) is a sequence of finite or infinite length adapted to Q Q Q , then its blockwise lift
i ^ = ( ı 1 , 1 , … , ı 1 , e i 1 , ı 2 , 1 , … , ı 2 , e i 2 , … ) , I i k = ( ı k , 1 , … , ı k , e i k ) \widehat{\mathbf{i}}=(\imath_{1,1},\ldots,\imath_{1,e_{i_1}},\imath_{2,1},\ldots,\imath_{2,e_{i_2}},\ldots),\qquad I_{i_k}=(\imath_{k,1},\ldots,\imath_{k,e_{i_k}}) i = ( 1 , 1 , … , 1 , e i 1 , 2 , 1 , … , 2 , e i 2 , … ) , I i k = ( k , 1 , … , k , e i k ) is adapted to Q ^ \widehat{Q} Q .
Proof. If i i i is a source of Q Q Q , then every vertex of I i I_i I i is a source of Q ^ \widehat{Q} Q . Moreover, the vertices in I i I_i I i are pairwise nonadjacent, so the corresponding source reflections commute. After reflecting successively at all vertices of I i I_i I i , the height function is
ξ ^ ℓ = ξ ^ ℓ − 2 δ ℓ ∈ I i \widehat{\xi}_{\ell}=\widehat{\xi}_{\ell}-2\delta_{\ell\in I_i} ξ ℓ = ξ ℓ − 2 δ ℓ ∈ I i
which is precisely the lift of the height function obtained from ξ \xi ξ by reflecting Q Q Q at i i i . The assertion now follows inductively. □
Every infinite adapted sequence considered below is assumed to contain each color infinitely often.
For each position k k k of i \mathbf{i} i , set
p k : = ξ i k − 2 # { 1 ≤ a < k ∣ i a = i k } . p_k:=\xi_{i_k}-2\#\{1\leq a<k\mid i_a=i_k\}. p k := ξ i k − 2# { 1 ≤ a < k ∣ i a = i k } .
If i \mathbf{i} i is infinite, then
{ ( i k , p k ) ∣ k ≥ 1 } = △ ^ ≤ ξ . \{(i_k,p_k)\mid k\geq1\}=\widehat{\triangle}_{\leq\xi}. {( i k , p k ) ∣ k ≥ 1 } = △ ≤ ξ .
Theorem 8.2 ([23 , 2 , 11 , 25 , 32 ] ). Let Q = ( △ , ξ ) Q=(\triangle,\xi) Q = ( △ , ξ ) be a Dynkin quiver and let i = ( i 1 , i 2 , … ) \mathbf{i}=(i_1,i_2,\ldots) i = ( i 1 , i 2 , … ) be an infinite sequence adapted to Q Q Q . Then R q ( g ) ≤ ξ \mathcal{R}_q(\mathfrak{g})_{\leq\xi} R q ( g ) ≤ ξ has a quantum cluster algebra structure whose initial quantum seed is
( { F q ( M ( i k ) [ p k , ξ i k ] ) ∣ k ≥ 1 } , B i , Λ i ) . \left(\{F_q(M^{(i_k)}[p_k,\xi_{i_k}])\mid k\geq1\},B^{\mathbf{i}},\Lambda^{\mathbf{i}}\right). ( { F q ( M ( i k ) [ p k , ξ i k ]) ∣ k ≥ 1 } , B i , Λ i ) .
Theorem 8.3 ([32 , 48 , 50 ] ). Every quantum cluster monomial of R q ( g ) ≤ ξ \mathcal{R}_q(\mathfrak{g})_{\leq\xi} R q ( g ) ≤ ξ is an element of the L q L_q L q -basis of R q ( g ) \mathcal{R}_q(\mathfrak{g}) R q ( g ) .
By the Kazhdan–Lusztig characterization of the L q \mathcal{L}_q L q -basis, its transition matrix from the standard basis { E q ( M ) } \{E_q(M)\} { E q ( M )} is unitriangular over Z [ q ± 1 ] \mathbb{Z}[q^{\pm1}] Z [ q ± 1 ] . Specializing at q = 1 q=1 q = 1 therefore gives a unitriangular transition matrix over Z \mathbb{Z} Z . Hence
L : = { ev q = 1 ( L q ( M ) ) ∣ M is dominant } \mathcal{L}:=\{\operatorname{ev}_{q=1}(L_q(M))\mid M\text{ is dominant}\} L := { ev q = 1 ( L q ( M )) ∣ M is dominant }
forms a Z \mathbb{Z} Z -basis of K Z ( g ) \mathcal{K}_{\mathbb{Z}}(\mathfrak{g}) K Z ( g ) and, after complexification, a C \mathbb{C} C -basis of K ( g ) \mathcal{K}(\mathfrak{g}) K ( g ) , which we call the classical global basis .4 By Remark 8.4 , it is the specialization at q = 1 q=1 q = 1 of the normalized global basis G ~ \widetilde{\mathbf{G}} G . Set
K ( g ) ≤ ξ ~ : = C [ F ( y i , p ) ∣ ( i , p ) ∈ △ ~ ≤ ξ ~ ] . \mathcal{K}(\mathfrak{g})_{\leq\widetilde{\xi}}:=\mathbb{C}[F(y_{i,p})\mid(i,p)\in\widetilde{\triangle}_{\leq\widetilde{\xi}}]. K ( g ) ≤ ξ := C [ F ( y i , p ) ∣ ( i , p ) ∈ △ ≤ ξ ] .
Specializing Theorem 8.2 at q = 1 q=1 q = 1 , we obtain the following.
Corollary 8.5. Let Q = ( △ , ξ ) Q=(\triangle,\xi) Q = ( △ , ξ ) be a Dynkin quiver and let i = ( i 1 , i 2 , … ) \mathbf{i}=(i_1,i_2,\ldots) i = ( i 1 , i 2 , … ) be an infinite sequence adapted to Q Q Q . Then K ( g ) ≤ ξ ~ \mathcal{K}(\mathfrak{g})_{\leq\widetilde{\xi}} K ( g ) ≤ ξ has a cluster algebra structure whose initial seed is
( { F ( m ( i k ) [ p k , ξ i k ] ) ∣ k ≥ 1 } , B i ) . \left(\{F(m^{(i_k)}[p_k,\xi_{i_k}])\mid k\geq1\},B^{\mathbf{i}}\right). ( { F ( m ( i k ) [ p k , ξ i k ]) ∣ k ≥ 1 } , B i ) . Moreover, every cluster monomial of K ( g ) ≤ ξ ~ \mathcal{K}(\mathfrak{g})_{\leq\widetilde{\xi}} K ( g ) ≤ ξ belongs to the L \mathcal{L} L -basis L \mathcal{L} L of K ( g ) \mathcal{K}(\mathfrak{g}) K ( g ) .
Proof. This follows by specializing Theorem 8.2 and Theorem 8.3 at q = 1 q=1 q = 1 .
Cluster structures on Grothendieck subrings
For i ∈ I i\in I i ∈ I , set q i : = q d i q_i:=q^{d_i} q i := q d i and
[ n ] q i : = q i n − q i − n q i − q i − 1 , [ n ] q i ! : = ∏ k = 1 n [ k ] q i , [ n k ] q i : = [ n ] q i ! [ k ] q i ! [ n − k ] q i ! . [n]_{q_i}:=\frac{q_i^n-q_i^{-n}}{q_i-q_i^{-1}},\qquad[n]_{q_i}!:=\prod_{k=1}^{n}[k]_{q_i},\qquad
\begin{bmatrix}n\\k\end{bmatrix}_{q_i}:=\frac{[n]_{q_i}!}{[k]_{q_i}![n-k]_{q_i}!}. [ n ] q i := q i − q i − 1 q i n − q i − n , [ n ] q i ! := k = 1 ∏ n [ k ] q i , [ n k ] q i := [ k ] q i ! [ n − k ] q i ! [ n ] q i ! .
Definition 8.6. [22 , 11 , 24 ] The bosonic extension A q ( g ) \mathcal{A}_q(\mathfrak{g}) A q ( g ) associated with g \mathfrak{g} g is the Q ( q 1 / 2 ) \mathbb{Q}(q^{1/2}) Q ( q 1/2 ) -algebra generated by { f i , p ∣ ( i ∈ I , p ∈ Z ) } \{f_{i,p}\mid(i\in I,\ p\in\mathbb{Z})\} { f i , p ∣ ( i ∈ I , p ∈ Z )} , subject to the relations
∑ k = 0 1 − c i j ( − 1 ) k [ 1 − c i j k ] q i f i , p k f j , p f i , p 1 − c i j − k = 0 ( i ≠ j ) \sum_{k=0}^{1-c_{ij}}(-1)^k\begin{bmatrix}1-c_{ij}\\k\end{bmatrix}_{q_i}f_{i,p}^{k}f_{j,p}f_{i,p}^{1-c_{ij}-k}=0\qquad(i\ne j) k = 0 ∑ 1 − c ij ( − 1 ) k [ 1 − c ij k ] q i f i , p k f j , p f i , p 1 − c ij − k = 0 ( i = j )
f i , m f j , p = q ( − 1 ) p − m + 1 ( α i , α j ) f j , p f i , m + δ ( j , p ) , ( i , m + 1 ) ( 1 − q i 2 ) ( m < p ) . f_{i,m}f_{j,p}=q^{(-1)^{p-m+1}(\alpha_i,\alpha_j)}f_{j,p}f_{i,m}+\delta_{(j,p),(i,m+1)}(1-q_i^2)\qquad(m<p). f i , m f j , p = q ( − 1 ) p − m + 1 ( α i , α j ) f j , p f i , m + δ ( j , p ) , ( i , m + 1 ) ( 1 − q i 2 ) ( m < p ) .
For − ∞ ≤ a ≤ b ≤ ∞ -\infty\leq a\leq b\leq\infty − ∞ ≤ a ≤ b ≤ ∞ , let A q ( g ) [ a , b ] \mathcal{A}_q(\mathfrak{g})[a,b] A q ( g ) [ a , b ] be the subalgebra generated by f i , p f_{i,p} f i , p with a ≤ p ≤ b . a\leq p\leq b. a ≤ p ≤ b . We set
A q ( g ) ≥ m : = A q ( g ) [ m , ∞ ] , A q ( g ) ≤ m : = A q ( g ) [ − ∞ , m ] , \mathcal{A}_q(\mathfrak{g})_{\geq m}:=\mathcal{A}_q(\mathfrak{g})[m,\infty],\qquad\mathcal{A}_q(\mathfrak{g})_{\leq m}:=\mathcal{A}_q(\mathfrak{g})[-\infty,m], A q ( g ) ≥ m := A q ( g ) [ m , ∞ ] , A q ( g ) ≤ m := A q ( g ) [ − ∞ , m ] ,
and A q ( g ) > m \mathcal{A}_q(\mathfrak{g})_{>m} A q ( g ) > m , A q ( g ) < m \mathcal{A}_q(\mathfrak{g})_{<m} A q ( g ) < m similarly. We also abbreviate A q ( g ) [ m ] : = A q ( g ) [ m , m ] \mathcal{A}_q(\mathfrak{g})[m]:=\mathcal{A}_q(\mathfrak{g})[m,m] A q ( g ) [ m ] := A q ( g ) [ m , m ] .
When there is no danger of confusion, we omit g \mathfrak{g} g from the notation and simply write
A q , A q [ a , b ] , A q [ m ] , A q , ≥ m , A q , ≤ m , \mathcal{A}_q,\qquad\mathcal{A}_q[a,b],\qquad\mathcal{A}_q[m],\qquad\mathcal{A}_{q,\geq m},\qquad\mathcal{A}_{q,\leq m}, A q , A q [ a , b ] , A q [ m ] , A q , ≥ m , A q , ≤ m ,
and similarly for the other subalgebras.
Recall that there are anti-involutions ω \omega ω and ( ⋅ ) ‾ \overline{(\mathord\cdot)} ( ⋅ ) of A q ( g ) \mathcal{A}_q(\mathfrak{g}) A q ( g ) defined by
ω ( q 1 / 2 ) = q 1 / 2 , ω ( f i , p ) = f i , − p , and q 1 / 2 ‾ = q − 1 / 2 , f i , p ‾ = f i , p , \omega(q^{1/2})=q^{1/2},\qquad\omega(f_{i,p})=f_{i,-p},\qquad\text{and}\qquad\overline{q^{1/2}}=q^{-1/2},\qquad\overline{f_{i,p}}=f_{i,p}, ω ( q 1/2 ) = q 1/2 , ω ( f i , p ) = f i , − p , and q 1/2 = q − 1/2 , f i , p = f i , p ,
respectively ([30 (2.3)] ). They commute, and hence Ω : = ω ∘ ( ⋅ ) ‾ \Omega:=\omega\circ\overline{(\mathord\cdot)} Ω := ω ∘ ( ⋅ ) is a multiplicative Q \mathbb{Q} Q -algebra automorphism satisfying
Ω ( q 1 / 2 ) = q − 1 / 2 , Ω ( f i , p ) = f i , − p . \Omega(q^{1/2})=q^{-1/2},\qquad\Omega(f_{i,p})=f_{i,-p}. Ω ( q 1/2 ) = q − 1/2 , Ω ( f i , p ) = f i , − p .
Set
κ i : = q i − 1 / 2 ( 1 − q i 2 ) , F i , p : = κ i − 1 f i , p , F i , p ( r ) : = F i , p r [ r ] q i ! . \kappa_i:=q_i^{-1/2}(1-q_i^2),\qquad F_{i,p}:=\kappa_i^{-1}f_{i,p},\qquad F_{i,p}^{(r)}:=\frac{F_{i,p}^r}{[r]_{q_i}!}. κ i := q i − 1/2 ( 1 − q i 2 ) , F i , p := κ i − 1 f i , p , F i , p ( r ) := [ r ] q i ! F i , p r .
Definition 8.7. [29 , 24 , 30 ] For each i ∈ I i\in I i ∈ I , define the braid symmetry
T i : A q ⟶ A q T_i:\mathcal{A}_q\longrightarrow\mathcal{A}_q T i : A q ⟶ A q on the generators by
T i ( f j , p ) = { f i , p + 1 , j = i , ∑ r , s ≥ 0 r + s = − c i j ( − q i ) s F i , p ( r ) f j , p F i , p ( s ) , j ≠ i . T_i(f_{j,p})=
\begin{cases}
f_{i,p+1}, & j=i,\\
\sum_{\substack{r,s\geq0\\r+s=-c_{ij}}}(-q_i)^sF_{i,p}^{(r)}f_{j,p}F_{i,p}^{(s)}, & j\ne i.
\end{cases} T i ( f j , p ) = ⎩ ⎨ ⎧ f i , p + 1 , ∑ r , s ≥ 0 r + s = − c ij ( − q i ) s F i , p ( r ) f j , p F i , p ( s ) , j = i , j = i . The maps T i T_i T i are Q ( q 1 / 2 ) \mathbb{Q}(q^{1/2}) Q ( q 1/2 ) -algebra automorphisms and satisfy the braid relations. Hence they define an action of the braid group B g B_{\mathfrak{g}} B g on A q \mathcal{A}_q A q . We denote the inverse of T i T_i T i by T i ∗ T_i^* T i ∗ , so that
T i ∗ T i = T i T i ∗ = id . T_i^*T_i=T_iT_i^*=\operatorname{id}. T i ∗ T i = T i T i ∗ = id .
For b ∈ B g + \mathbf{b}\in B_{\mathfrak{g}}^+ b ∈ B g + and i = ( i 1 , … , i l ) ∈ Seq ( b ) \mathbf{i}=(i_1,\ldots,i_l)\in\operatorname{Seq}(\mathbf{b}) i = ( i 1 , … , i l ) ∈ Seq ( b ) , set
T b : = T i 1 ⋯ T i l . T_{\mathbf{b}}:=T_{i_1}\cdots T_{i_l}. T b := T i 1 ⋯ T i l .
By the braid relations, T b T_{\mathbf{b}} T b is independent of the choice of i ∈ Seq ( b ) \mathbf{i}\in\operatorname{Seq}(\mathbf{b}) i ∈ Seq ( b ) .
Definition 8.9. [46 , 30 ] For b ∈ B g + \mathbf{b}\in B_{\mathfrak{g}}^+ b ∈ B g + , define
A q ( b ) : = A q , ≥ 0 ∩ T b ( A q , < 0 ) . \mathcal{A}_q(\mathbf{b}):=\mathcal{A}_{q,\geq0}\cap T_{\mathbf{b}}\left(\mathcal{A}_{q,<0}\right). A q ( b ) := A q , ≥ 0 ∩ T b ( A q , < 0 ) . We call A q ( b ) \mathcal{A}_q(\mathbf{b}) A q ( b ) the positive-braid subalgebra associated with b \mathbf{b} b .
Let i = ( i 1 , … , i l ) ∈ Seq ( b ) \mathbf{i}=(i_1,\ldots,i_l)\in\operatorname{Seq}(\mathbf{b}) i = ( i 1 , … , i l ) ∈ Seq ( b ) . For 1 ≤ k ≤ l 1\leq k\leq l 1 ≤ k ≤ l , set
P k i : = T i 1 ⋯ T i k − 1 ( q i k 1 / 2 f i k , 0 ) , P_k^{\mathbf{i}}:=T_{i_1}\cdots T_{i_{k-1}}\left(q_{i_k}^{1/2}f_{i_k,0}\right), P k i := T i 1 ⋯ T i k − 1 ( q i k 1/2 f i k , 0 ) ,
and call it the cuspidal element of i \mathbf{i} i at k k k . For m ∈ Z ≥ 0 m\in\mathbb{Z}_{\geq0} m ∈ Z ≥ 0 , set
P k i , { m } : = q i k m ( m − 1 ) / 2 ( P k i ) m . P_k^{\mathbf{i},\{m\}}:=q_{i_k}^{m(m-1)/2}\left(P_k^{\mathbf{i}}\right)^m. P k i , { m } := q i k m ( m − 1 ) /2 ( P k i ) m .
For u = ( u 1 , … , u l ) ∈ Z ≥ 0 l \mathbf{u}=(u_1,\ldots,u_l)\in\mathbb{Z}_{\geq0}^l u = ( u 1 , … , u l ) ∈ Z ≥ 0 l , define
P i ( u ) : = ∏ → k ∈ [ 1 , l ] P k i , { u k } . P^i(\mathbf{u}):=\overrightarrow{\prod}_{k\in[1,l]}P_k^{i,\{u_k\}}. P i ( u ) := ∏ k ∈ [ 1 , l ] P k i , { u k } .
The product is taken in decreasing order of k k k . We shall also use the integral form of the bosonic extension. Let
A A ⊆ A q \mathcal{A}_{\mathbb{A}}\subseteq\mathcal{A}_q A A ⊆ A q
be its A \mathbb{A} A -integral form. The integral form A A \mathcal{A}_{\mathbb{A}} A A is preserved by T i ± 1 T_i^{\pm1} T i ± 1 . For b ∈ B g + \mathbf{b}\in\mathcal{B}_{\mathfrak{g}}^+ b ∈ B g + , set
A A ( b ) : = A q ( b ) ∩ A A . \mathcal{A}_{\mathbb{A}}(\mathbf{b}):=\mathcal{A}_q(\mathbf{b})\cap\mathcal{A}_{\mathbb{A}}. A A ( b ) := A q ( b ) ∩ A A .
For finite type g \mathfrak{g} g and each Dynkin quiver Q Q Q of g \mathfrak{g} g , the A \mathbb{A} A -integral form A A ( g ) \mathcal{A}_{\mathbb{A}}(\mathfrak{g}) A A ( g ) is isomorphic to the quantum virtual Grothendieck ring K q ( g ) \mathcal{K}_q(\mathfrak{g}) K q ( g ) introduced above [34 Theorem 6.2] ; see also [22 ] and [24 Theorems 5.2 and 7.2] . In the simply-laced case, this recovers the quantum Grothendieck ring of the Hernandez–Leclerc category.
More precisely, for every Dynkin quiver Q = ( Δ , ξ ) Q=(\Delta,\xi) Q = ( Δ , ξ ) , there is a Z \mathbb{Z} Z -algebra isomorphism
Φ Q : A A → ∼ K q ( g ) , Φ Q ( q 1 / 2 ) = q − 1 / 2 \Phi_Q:\mathcal{A}_{\mathbb{A}}\xrightarrow{\sim}\mathcal{K}_q(\mathfrak{g}),\qquad\Phi_Q(q^{1/2})=q^{-1/2} Φ Q : A A ∼ K q ( g ) , Φ Q ( q 1/2 ) = q − 1/2
(see [34 Theorem 6.2] ). It restricts to a Z \mathbb{Z} Z -isomorphism
Φ Q : ( A A ) ≤ 0 → ∼ K q ( g ) ≤ ξ . \Phi_Q:(\mathcal{A}_{\mathbb{A}})_{\leq0}\xrightarrow{\sim}\mathcal{K}_q(\mathfrak{g})_{\leq\xi}. Φ Q : ( A A ) ≤ 0 ∼ K q ( g ) ≤ ξ .
We set
Ψ Q : = Φ Q ∘ Ω . \Psi_Q:=\Phi_Q\circ\Omega. Ψ Q := Φ Q ∘ Ω.
Since both Φ Q \Phi_Q Φ Q and Ω \Omega Ω invert the quantum parameter q q q , Ψ Q \Psi_Q Ψ Q is an A \mathbb{A} A -algebra isomorphism
Ψ Q : A A → ∼ K q ( g ) , \Psi_Q:\mathcal{A}_{\mathbb{A}}\xrightarrow{\sim}\mathcal{K}_q(\mathfrak{g}), Ψ Q : A A ∼ K q ( g ) ,
whose restriction gives an A \mathbb{A} A -algebra isomorphism
Ψ Q : ( A A ) ≥ 0 → ∼ K q ( g ) ≤ ξ . \Psi_Q:(\mathcal{A}_{\mathbb{A}})_{\geq0}\xrightarrow{\sim}\mathcal{K}_q(\mathfrak{g})_{\leq\xi}. Ψ Q : ( A A ) ≥ 0 ∼ K q ( g ) ≤ ξ .
For later use, we also recall the substitution isomorphisms of [12 , 42 ] . Let F q ( g ) \mathbb{F}_q(\mathfrak{g}) F q ( g ) denote the skew field of fractions of the ambient quantum torus, and set
F ( g ) : = C ( y i , p ∣ ( i , p ) ∈ Δ ~ 0 g ) , \mathbb{F}(\mathfrak{g}):=\mathbb{C}(y_{i,p}\mid(i,p)\in\widetilde{\Delta}_0^g), F ( g ) := C ( y i , p ∣ ( i , p ) ∈ Δ 0 g ) ,
the corresponding field of rational functions after specializing q 1 / 2 = 1 q^{1/2}=1 q 1/2 = 1 .
For two Dynkin quivers Q Q Q and Q ′ Q' Q ′ of g \mathfrak{g} g , set
Ψ Q , Q ′ : = Ψ Q ∘ Ψ Q ′ − 1 = Φ Q ∘ Φ Q ′ − 1 , \Psi_{Q,Q'}:=\Psi_Q\circ\Psi_{Q'}^{-1}=\Phi_Q\circ\Phi_{Q'}^{-1}, Ψ Q , Q ′ := Ψ Q ∘ Ψ Q ′ − 1 = Φ Q ∘ Φ Q ′ − 1 ,
where the second equality follows from the definition of Ψ Q \Psi_Q Ψ Q . Thus Ψ Q , Q ′ \Psi_{Q,Q'} Ψ Q , Q ′ is an A \mathbb{A} A -algebra automorphism of K q ( g ) \mathcal{K}_q(\mathfrak{g}) K q ( g ) . By [12 , 42 ] , this automorphism extends to an A \mathbb{A} A -algebra automorphism
Ψ ~ Q , Q ′ : F q ( g ) → ∼ F q ( g ) , \widetilde{\Psi}_{Q,Q'}:\mathbb{F}_q(\mathfrak{g})\xrightarrow{\sim}\mathbb{F}_q(\mathfrak{g}), Ψ Q , Q ′ : F q ( g ) ∼ F q ( g ) ,
called the substitution isomorphism , which commutes with the duality automorphism D q D_q D q . We denote its specialization at q 1 / 2 = 1 q^{1/2}=1 q 1/2 = 1 by the C \mathbb{C} C -algebra automorphism
Ψ ~ Q , Q ′ c l : F ( g ) → ∼ F ( g ) . \widetilde{\Psi}^{\mathrm{cl}}_{Q,Q'}:\mathbb{F}(\mathfrak{g})\xrightarrow{\sim}\mathbb{F}(\mathfrak{g}). Ψ Q , Q ′ cl : F ( g ) ∼ F ( g ) .
Let B ( ∞ ) \mathcal{B}(\infty) B ( ∞ ) denote the crystal of U q − ( g ) U_q^-(\mathfrak{g}) U q − ( g ) , and let 1 1 1 denote its highest weight element. Recall that Kashiwara–Park [36 ] introduced the extended crystal
B ^ ( ∞ ) : = { ( b m ) m ∈ Z ∈ ∏ m ∈ Z B ( ∞ ) ∣ b m = 1 for all but finitely many m } . \widehat{\mathcal{B}}(\infty):=\left\{(b_m)_{m\in\mathbb{Z}}\in\prod_{m\in\mathbb{Z}}\mathcal{B}(\infty)\mathrel{\Big|}b_m=1\text{ for all but finitely many }m\right\}. B ( ∞ ) := { ( b m ) m ∈ Z ∈ m ∈ Z ∏ B ( ∞ ) b m = 1 for all but finitely many m } .
The bosonic extension A q \mathcal{A}_q A q admits the normalized global basis
G ~ = { G ~ ( b ) ∣ b ∈ B ^ ( ∞ ) } (62) \widetilde{\mathbf{G}}=\left\{\widetilde{G}(b)\mathrel{\Big|}b\in\widehat{\mathcal{B}}(\infty)\right\} \tag*{(62)} G = { G ( b ) b ∈ B ( ∞ ) } (62)
which is compatible with the integral form and is preserved by the braid symmetries T i T_i T i ; see [33 , 30 ] .
Theorem 8.10 ([46 , 30 ] ). Let b ∈ B g + b\in B_{\mathfrak{g}}^+ b ∈ B g + with ℓ ( b ) = l \ell(b)=l ℓ ( b ) = l and let i ∈ Seq ( b ) \boldsymbol{i}\in\operatorname{Seq}(b) i ∈ Seq ( b ) . Then
P i : = { P i ( u ) ∣ u ∈ Z ≥ 0 l } \mathcal{P}^{\boldsymbol{i}}:=\left\{P^{\boldsymbol{i}}(\boldsymbol{u})\mathrel{\Big|}\boldsymbol{u}\in\mathbb{Z}_{\geq0}^{l}\right\} P i := { P i ( u ) u ∈ Z ≥ 0 l } forms an A \mathbb{A} A -basis of A A ( b ) \mathcal{A}_{\mathbb{A}}(b) A A ( b ) . In particular, A q ( b ) \mathcal{A}_q(b) A q ( b ) is generated by the cuspidal elements
P 1 i , … , P l i . P_1^{\boldsymbol{i}},\ldots,P_l^{\boldsymbol{i}}. P 1 i , … , P l i . Moreover, the global basis is compatible with A q ( b ) \mathcal{A}_q(b) A q ( b ) ; namely,
G ~ ( b ) : = G ~ ∩ A q ( b ) \widetilde{\mathbf{G}}(b):=\widetilde{\mathbf{G}}\cap\mathcal{A}_q(b) G ( b ) := G ∩ A q ( b ) forms an A \mathbb{A} A -basis of A A ( b ) \mathcal{A}_{\mathbb{A}}(b) A A ( b ) .
In particular,
A A ( b ) = A ⟨ P 1 i , … , P l i ⟩ , \mathcal{A}_{\mathbb{A}}(b)=\mathbb{A}\langle P_1^{\boldsymbol{i}},\ldots,P_l^{\boldsymbol{i}}\rangle, A A ( b ) = A ⟨ P 1 i , … , P l i ⟩ ,
with PBW basis P i \mathcal{P}^{\boldsymbol{i}} P i .
We recall a consequence of the Garside normal form that will be used to realize arbitrary finite sequences inside adapted ones. Let
π : B g ⟶ W \pi:B_{\mathfrak{g}}\longrightarrow W π : B g ⟶ W
be the canonical projection, and let Δ ∈ B g + \Delta\in B_{\mathfrak{g}}^+ Δ ∈ B g + be the positive braid lift of the longest element w 0 ∈ W w_0\in W w 0 ∈ W , characterized by
π ( Δ ) = w 0 , ℓ ( Δ ) = ℓ ( w 0 ) . \pi(\Delta)=w_0,\qquad\ell(\Delta)=\ell(w_0). π ( Δ ) = w 0 , ℓ ( Δ ) = ℓ ( w 0 ) .
A sequence is called locally reduced if every consecutive subsequence of length at most ℓ ( w 0 ) \ell(w_0) ℓ ( w 0 ) is reduced [34 Definition 2.3(iii)] . By [34 Remark 5.4] , for every finite sequence
i = ( i 1 , … , i l ) , \boldsymbol{i}=(i_1,\ldots,i_l), i = ( i 1 , … , i l ) ,
there exist m ≫ 0 m\gg0 m ≫ 0 and a sequence
i ~ ∈ Seq ( Δ m ) \widetilde{\boldsymbol{i}}\in\operatorname{Seq}(\Delta^m) i ∈ Seq ( Δ m )
such that i ~ [ 1 , l ] = i \widetilde{\boldsymbol{i}}_{[1,l]}=\boldsymbol{i} i [ 1 , l ] = i .
Moreover, i ~ \widetilde{\boldsymbol{i}} i can be obtained from a locally reduced sequence in Seq ( Δ m ) \operatorname{Seq}(\Delta^m) Seq ( Δ m ) by finitely many commutation and braid moves, where the locally reduced sequence may be chosen to be adapted to a suitable Dynkin quiver.
Now let b ∈ B g + b\in B_{\mathfrak{g}}^+ b ∈ B g + and i = ( i 1 , … , i l ) ∈ Seq ( b ) \boldsymbol{i}=(i_1,\ldots,i_l)\in\operatorname{Seq}(b) i = ( i 1 , … , i l ) ∈ Seq ( b ) , and choose i ~ ∈ Seq ( Δ m ) \widetilde{\boldsymbol{i}}\in\operatorname{Seq}(\Delta^m) i ∈ Seq ( Δ m ) as above. Thus i \boldsymbol{i} i occurs as the initial segment of i ~ \widetilde{\boldsymbol{i}} i :
i ~ [ 1 , l ] = i . \widetilde{\boldsymbol{i}}_{[1,l]}=\boldsymbol{i}. i [ 1 , l ] = i .
Choose a locally reduced sequence j ∈ Seq ( Δ m ) \mathbf{j} \in\operatorname{Seq}(\Delta^m) j ∈ Seq ( Δ m ) adapted to a Dynkin quiver Q = ( Δ , ξ ) Q=(\Delta,\xi) Q = ( Δ , ξ ) such that i \mathbf{i} i is obtained from j \mathbf{j} j by finitely many commutation and braid moves. Let Q ^ = ( Δ ^ , ξ ^ ) \widehat{Q}=(\widehat{\Delta},\widehat{\xi}) Q = ( Δ , ξ ) be the lifted Dynkin quiver associated with Q Q Q .
By Lemma 8.1 8.1 8.1 , the blockwise lift j ^ \widehat{\mathbf{j}} j of j \mathbf{j} j is adapted to Q ^ \widehat{Q} Q . Moreover, by the lifting results of the preceding sections, the sequence of commutation and braid moves from j \mathbf{j} j to i \mathbf{i} i lifts to a sequence of moves from j ^ \widehat{\mathbf{j}} j to the blockwise lift i ^ \widehat{\mathbf{i}} i of i \mathbf{i} i .
Since i \mathbf{i} i is the initial segment of i ~ \widetilde{\mathbf{i}} i , the corresponding initial block of i ^ \widehat{\mathbf{i}} i is precisely the blockwise lift i ^ \widehat{\mathbf{i}} i of i \mathbf{i} i . Writing
i ~ = i ⋆ τ , i ~ ^ = i ^ ⋆ τ ^ , \widetilde{\mathbf{i}}=\mathbf{i}\star\tau,\qquad
\widehat{\widetilde{\mathbf{i}}}=\widehat{\mathbf{i}}\star\widehat{\tau}, i = i ⋆ τ , i = i ⋆ τ ,
where τ ^ \widehat{\tau} τ is the blockwise lift of τ \tau τ , we obtain
j ^ adapted to Q ^ ⟶ i ~ ^ = i ^ ⋆ τ ^ ↓ ↓ j adapted to Q ⟶ i ~ = i ⋆ τ . \begin{aligned}
\widehat{\mathbf{j}}\text{ adapted to }\widehat{Q} &\longrightarrow\widehat{\widetilde{\mathbf{i}}}=\widehat{\mathbf{i}}\star\widehat{\tau}\\
\downarrow\quad&\qquad\qquad\downarrow\\
\mathbf{j}\text{ adapted to }Q &\longrightarrow\widetilde{\mathbf{i}}=\mathbf{i}\star\tau.
\end{aligned} j adapted to Q ↓ j adapted to Q ⟶ i = i ⋆ τ ↓ ⟶ i = i ⋆ τ .
where the horizontal arrows are given by the corresponding sequences of commutation and braid moves.
We now apply this construction to the simply-laced upstairs Grothendieck ring K q ( g ^ ) \mathcal{K}_q(\widehat{\mathfrak{g}}) K q ( g ) . Let b ^ ∈ B g ^ + \widehat{b}\in B_{\widehat{\mathfrak{g}}}^{+} b ∈ B g + be the positive braid monoid element represented by the blockwise lift i ^ \widehat{\mathbf{i}} i of i \mathbf{i} i . By the monoidal categorification of [34 ] , the positive-braid subalgebra A A ( b ^ ) \mathcal{A}_{A}(\widehat{b}) A A ( b ) admits a categorical realization by the monoidal subcategory C g ^ ( b ^ ) \mathcal{C}_{\widehat{\mathfrak{g}}}(\widehat{b}) C g ( b ) of the Hernandez–Leclerc category C g ^ ∘ \mathcal{C}_{\widehat{\mathfrak{g}}}^{\circ} C g ∘ . In particular, admissible chains of i \mathbf{i} i -boxes give monoidal seeds whose quantum cluster algebra realizes A A ( b ^ ) \mathcal{A}_{A}(\widehat{b}) A A ( b ) .
For a simple module L L L in C g ^ ∘ \mathcal{C}_{\widehat{\mathfrak{g}}}^{\circ} C g ∘ , let [ L ] q [L]_q [ L ] q denote its ( q , t ) (q,t) ( q , t ) -character, with deformation parameter q q q , and set
L q : = Φ Q ^ − 1 ( [ L ] q ) ∈ A A ( b ^ ) . L_q:=\Phi_{\widehat{Q}}^{-1}([L]_q)\in\mathcal{A}_{A}(\widehat{b}). L q := Φ Q − 1 ([ L ] q ) ∈ A A ( b ) .
The modules and the map Φ Q ^ \Phi_{\widehat{Q}} Φ Q are associated with the same Dynkin quiver Q ^ \widehat{Q} Q . For an admissible chain C ^ \widehat{\mathcal{C}} C , the exchange and commutation matrices in [34 Definition 9.5 and Proposition 5.20] are − B C ^ -B^{\widehat{\mathcal{C}}} − B C and − Λ C ^ -\Lambda^{\widehat{\mathcal{C}}} − Λ C , respectively. Thus [34 Theorem 9.7] , with t 1 / 2 = q − 1 / 2 t^{1/2}=q^{-1/2} t 1/2 = q − 1/2 and the exchange matrix negated, gives the quantum seed below.
Theorem 8.11 ([34 Theorems 9.6 and 9.12] ). Let b ^ ∈ B g ^ + \widehat{b}\in B_{\widehat{\mathfrak{g}}}^{+} b ∈ B g + and i ^ ∈ Seq ( b ^ ) \widehat{\mathbf{i}}\in\operatorname{Seq}(\widehat{b}) i ∈ Seq ( b ) . For each i \mathbf{i} i -box c ^ \widehat{c} c , let M ^ [ c ^ ] \widehat{M}[\widehat{c}] M [ c ] be the corresponding affine determinantial module. For a singleton box [ k ] [k] [ k ] , M ^ [ k ] \widehat{M}[k] M [ k ] is the corresponding affine cuspidal module.
For every admissible chain C ^ = ( c ^ 1 , … , c ^ ℓ ^ ) \widehat{\mathcal{C}}=(\widehat{c}_1,\ldots,\widehat{c}_{\widehat{\ell}}) C = ( c 1 , … , c ℓ ) of i \mathbf{i} i -boxes, the family { M ^ [ c ^ k ] ∣ 1 ≤ k ≤ ℓ ^ } \{\widehat{M}[\widehat{c}_k]\mid1\leq k\leq\widehat{\ell}\} { M [ c k ] ∣ 1 ≤ k ≤ ℓ } forms a monoidal seed of C g ^ ( b ^ ) \mathcal{C}_{\widehat{\mathfrak{g}}}(\widehat{b}) C g ( b ) , and
( { M ^ [ c ^ k ] q } 1 ≤ k ≤ ℓ ^ , B C ^ , Λ C ^ ) (63) \left(\{\widehat{M}[\widehat{c}_k]_q\}_{1\leq k\leq\widehat{\ell}},B^{\widehat{\mathcal{C}}},\Lambda^{\widehat{\mathcal{C}}}\right) \tag*{(63)} ( { M [ c k ] q } 1 ≤ k ≤ ℓ , B C , Λ C ) (63) is a quantum seed of A A ( b ^ ) \mathcal{A}_{A}(\widehat{b}) A A ( b ) .
Corollary 8.12. For every b ^ ∈ B g ^ + \widehat{b}\in B_{\widehat{\mathfrak{g}}}^{+} b ∈ B g + , there is an A \mathbb{A} A -algebra isomorphism
A q ( b ^ ) → ∼ A A ( b ^ ) . \mathcal{A}_q(\widehat{b})\xrightarrow{\sim}\mathcal{A}_{A}(\widehat{b}). A q ( b ) ∼ A A ( b ) .
Under this isomorphism, the cluster variable associated with an i \mathbf{i} i -box c ^ \widehat{c} c is identified with M ^ [ c ^ ] q \widehat{M}[\widehat{c}]_q M [ c ] q .
Proof. Fix i ^ ∈ Seq ( b ) \widehat{\mathbf{i}}\in\operatorname{Seq}(\mathbf{b}) i ∈ Seq ( b ) . By Theorem 8.11 , for every admissible chain C ^ = ( c ^ 1 , … , c ^ l ^ ) \widehat{C}=(\widehat{c}_1,\ldots,\widehat{c}_{\widehat{l}}) C = ( c 1 , … , c l ) of i \mathbf{i} i -boxes,
( { M ^ [ c ^ k ] q } 1 ≤ k ≤ l ^ , B C ^ , Λ C ^ ) \left(\{\widehat{M}[\widehat{c}_k]_q\}_{1\leq k\leq\widehat{l}},B^{\widehat{C}},\Lambda^{\widehat{C}}\right) ( { M [ c k ] q } 1 ≤ k ≤ l , B C , Λ C )
is a quantum seed of A A ( b ^ ) \mathcal{A}_{\mathcal{A}}(\widehat{\mathbf{b}}) A A ( b ) . On the other hand, the i \mathbf{i} i -box seed defining A q ( b ^ ) A_q(\widehat{\mathbf{b}}) A q ( b ) has the same compatible pair ( B C ^ , Λ C ^ ) (B^{\widehat{C}},\Lambda^{\widehat{C}}) ( B C , Λ C ) . Moreover, under box moves, the two families of seeds are related by the same mutation–permutation transformations. Hence the identification of the initial cluster variables
X c ^ ⟼ M ^ [ c ^ ] q ( c ^ ∈ C ^ ) X_{\widehat{c}}\longmapsto\widehat{M}[\widehat{c}]_q\qquad(\widehat{c}\in\widehat{C}) X c ⟼ M [ c ] q ( c ∈ C )
extends throughout the corresponding mutation classes and induces the desired isomorphism
A q ( b ^ ) → ∼ A A ( b ^ ) . A_q(\widehat{\mathbf{b}})\xrightarrow{\sim}\mathcal{A}_{\mathcal{A}}(\widehat{\mathbf{b}}). A q ( b ) ∼ A A ( b ) .
Theorem 8.13 ([34 Theorem 9.12 and its proof] ). Let b ^ ∈ B g + \widehat{\mathbf{b}}\in B_{\mathfrak{g}}^+ b ∈ B g + and write i ^ = ( ı ^ 1 , … , ı ^ l ^ ) ∈ Seq ( b ) \widehat{\mathbf{i}}=(\widehat{\imath}_1,\ldots,\widehat{\imath}_{\widehat{l}})\in\operatorname{Seq}(\mathbf{b}) i = ( 1 , … , l ) ∈ Seq ( b ) . Then, for every 1 ≤ k ≤ l ^ 1\leq k\leq\widehat{l} 1 ≤ k ≤ l ,
M ^ [ k ] q = q − 1 2 q ı ^ k P ^ k . \widehat{M}[k]_q=q^{-\frac{1}{2}q_{\widehat{\imath}_k}}\widehat{P}_k. M [ k ] q = q − 2 1 q k P k .
We compare the quantum seeds associated with i ^ \widehat{\mathbf{i}} i with those obtained from the ambient sequence i ~ = i ^ ⋆ τ ^ \widetilde{\mathbf{i}}=\widehat{\mathbf{i}}\star\widehat{\tau} i = i ⋆ τ .
Extend j ^ \widehat{\mathbf{j}} j to an infinite Q ^ \widehat{Q} Q -adapted sequence and take the KR seed of Theorem 8.2 . Freeze the last occurrence of each color in j ^ \widehat{\mathbf{j}} j and delete all vertices outside j ^ \widehat{\mathbf{j}} j . By (21) , the deleted rows have zero entries in every retained exchangeable column, and (22) restricts to Λ j ^ \Lambda^{\widehat{\mathbf{j}}} Λ j . This gives a finite quantum seed in R q ( g ^ ) ≤ ξ ^ \mathcal{R}_q(\widehat{\mathfrak{g}})_{\leq\widehat{\xi}} R q ( g ) ≤ ξ with compatible pair ( B j ^ , Λ j ^ ) (B^{\widehat{\mathbf{j}}},\Lambda^{\widehat{\mathbf{j}}}) ( B j , Λ j ) . The same construction applies to j ^ \widehat{\mathbf{j}} j and Q ^ \widehat{Q} Q downstairs. The sequence of braid moves from j ^ \widehat{\mathbf{j}} j to i ^ \widehat{\mathbf{i}} i is realized by the corresponding mutation–permutation sequence. Hence the quantum seed associated with i ^ \widehat{\mathbf{i}} i belongs to the same mutation class.
Proposition 8.14 ([34 Proposition 8.10] ). Let i ~ = i ^ ⋆ τ ^ \widetilde{\mathbf{i}}=\widehat{\mathbf{i}}\star\widehat{\tau} i = i ⋆ τ be as above, and let C ^ \widehat{C} C be an admissible chain of i \mathbf{i} i -boxes associated with i ^ \widehat{\mathbf{i}} i . Extend C ^ \widehat{C} C to an admissible chain C ~ \widetilde{C} C of i ~ \widetilde{\mathbf{i}} i -boxes associated with i ~ \widetilde{\mathbf{i}} i . Assume that C ^ \widehat{C} C has extent [ 1 , l ^ ] [1,\widehat{l}] [ 1 , l ] and that C ~ \widetilde{C} C has length equal to that of i ~ \widetilde{\mathbf{i}} i . Write
J ^ = J ^ e x ⊔ J ^ f r \widehat{J}=\widehat{J}_{\mathrm{ex}}\sqcup\widehat{J}_{\mathrm{fr}} J = J ex ⊔ J fr for the vertices corresponding to the i ^ \widehat{\mathbf{i}} i -part, and let J ^ t a i l \widehat{J}_{\mathrm{tail}} J tail denote the remaining vertices. Then
b ~ a b C ~ = 0 ( a ∈ J ^ t a i l , b ∈ J ^ e x ) , B C ~ ∣ J ^ × J ^ e x = B C ^ , Λ C ~ ∣ J ^ × J ^ = Λ C ^ . \widetilde{b}^{\widetilde{C}}_{ab}=0\quad(a\in\widehat{J}_{\mathrm{tail}},\ b\in\widehat{J}_{\mathrm{ex}}),\qquad
B^{\widetilde{C}}\big|_{\widehat{J}\times\widehat{J}_{\mathrm{ex}}}=B^{\widehat{C}},\qquad
\Lambda^{\widetilde{C}}\big|_{\widehat{J}\times\widehat{J}}=\Lambda^{\widehat{C}}. b ab C = 0 ( a ∈ J tail , b ∈ J ex ) , B C J × J ex = B C , Λ C J × J = Λ C . Consequently, declaring the vertices in J ^ f r \widehat{J}_{\mathrm{fr}} J fr frozen and deleting the vertices in J ^ t a i l \widehat{J}_{\mathrm{tail}} J tail from the quantum seed in (63) for C ~ \widetilde{C} C gives
Σ C ~ = ( { M ^ [ c ^ ] q } c ^ ∈ C ^ , B C ~ , Λ C ~ ) , \Sigma^{\widetilde{C}}=\left(\{\widehat{M}[\widehat{c}]_q\}_{\widehat{c}\in\widehat{C}},B^{\widetilde{C}},\Lambda^{\widetilde{C}}\right), Σ C = ( { M [ c ] q } c ∈ C , B C , Λ C ) , the quantum seed of A A ( b ^ ) \mathcal{A}_{\mathcal{A}}(\widehat{\mathbf{b}}) A A ( b ) in (63) .
We return to the downstairs sequence i ∈ Seq ( b ) \mathbf{i} \in\operatorname{Seq}(\mathbf{b}) i ∈ Seq ( b ) . Let C = ( c 1 , … , c l ) \mathcal{C}=(\mathbf{c}_1,\ldots,\mathbf{c}_l) C = ( c 1 , … , c l ) be an admissible chain of i \mathbf{i} i -boxes, and let C ^ \widehat{\mathcal{C}} C be its canonical lift. By Theorem 3.15 , the compatible pair of the upstairs quantum seed Σ C ^ \Sigma^{\widehat{\mathcal{C}}} Σ C folds to ( B C , Λ C ) (B^{\mathcal{C}},\Lambda^{\mathcal{C}}) ( B C , Λ C ) . More precisely, for c , d ∈ C \mathbf{c},\mathbf{d}\in\mathcal{C} c , d ∈ C and d ^ ∈ Fib C ( d ) \widehat{\mathbf{d}}\in\operatorname{Fib}^{\mathcal{C}}(\mathbf{d}) d ∈ Fib C ( d ) ,
b c , d C = ∑ c ^ ∈ Fib C ( c ) b c ^ , d ^ C ^ , Λ c , d C = d i c ∑ c ^ ∈ Fib C ( c ) Λ c ^ , d ^ C ^ . b^{\mathcal{C}}_{\mathbf{c},\mathbf{d}}=\sum_{\widehat{\mathbf{c}}\in\operatorname{Fib}^{\mathcal{C}}(\mathbf{c})}b^{\widehat{\mathcal{C}}}_{\widehat{\mathbf{c}},\widehat{\mathbf{d}}},\qquad\Lambda^{\mathcal{C}}_{\mathbf{c},\mathbf{d}}=d_{i_{\mathbf{c}}}\sum_{\widehat{\mathbf{c}}\in\operatorname{Fib}^{\mathcal{C}}(\mathbf{c})}\Lambda^{\widehat{\mathcal{C}}}_{\widehat{\mathbf{c}},\widehat{\mathbf{d}}}. b c , d C = c ∈ Fib C ( c ) ∑ b c , d C , Λ c , d C = d i c c ∈ Fib C ( c ) ∑ Λ c , d C .
Thus the exchange and quantum commutation data of the upstairs seed fold precisely to the pair ( B C , Λ C ) (B^{\mathcal{C}},\Lambda^{\mathcal{C}}) ( B C , Λ C ) associated with the downstairs admissible chain C \mathcal{C} C .
From now on, we distinguish the positive-braid subalgebra A A ( b ) \mathcal{A}_{\mathsf{A}}(\mathbf{b}) A A ( b ) of the bosonic extension from the quantum cluster algebra associated with b \mathbf{b} b , which will be denoted by A q ( b ) \mathcal{A}_{q}(\mathbf{b}) A q ( b ) .
We consider the cluster variables on the downstairs side. We first identify the cluster variables associated with the singleton boxes. For this reason, we use the following result of Qin. For the adapted sequences in [50 Section 5.1] , the quantization matrix is Λ i \Lambda^{\mathbf{i}} Λ i by [35 (7.1) and (7.5)] and (22) . The equality is preserved by braid moves, by Proposition 4.5 . Restriction to an initial segment preserves the root sequence and box vectors, and hence also this equality. The quantization in [50 Section 6.3] is obtained by these operations, so it agrees with the present quantization.
Theorem 8.15 ([50 Theorem 6.6] ). Let i = ( i 1 , … , i l ) \mathbf{i}=(i_1,\ldots,i_l) i = ( i 1 , … , i l ) be an arbitrary finite sequence. Under the bosonic realization, the cluster variable associated with the singleton i \mathbf{i} i -box [ k ] [k] [ k ] is represented by
T i 1 ⋯ T i k − 1 ( f i k , 0 ) ( 1 ≤ k ≤ l ) . T_{i_1}\cdots T_{i_{k-1}}(f_{i_k,0})\qquad(1\leq k\leq l). T i 1 ⋯ T i k − 1 ( f i k , 0 ) ( 1 ≤ k ≤ l ) .
Here the braid symmetries in Theorem 8.15 agree with the automorphisms T i T_i T i defined above. Comparing the formula in Theorem 8.15 with
P k i = T i 1 ⋯ T i k − 1 ( q i k 1 / 2 f i k , 0 ) , P_k^{\mathbf{i}}=T_{i_1}\cdots T_{i_{k-1}}\left(q_{i_k}^{1/2}f_{i_k,0}\right), P k i = T i 1 ⋯ T i k − 1 ( q i k 1/2 f i k , 0 ) ,
we obtain the correspondence
X [ k ] ⟼ q i k − 1 / 2 P k i ( 1 ≤ k ≤ l ) . X_{[k]}\longmapsto q_{i_k}^{-1/2}P_k^{\mathbf{i}}\qquad(1\leq k\leq l). X [ k ] ⟼ q i k − 1/2 P k i ( 1 ≤ k ≤ l ) .
Corollary 8.16. Let b ∈ B g + \mathbf{b}\in\mathbf{B}_{\mathfrak{g}}^{+} b ∈ B g + and i = ( i 1 , … , i l ) ∈ Seq ( b ) \mathbf{i}=(i_1,\ldots,i_l)\in\operatorname{Seq}(\mathbf{b}) i = ( i 1 , … , i l ) ∈ Seq ( b ) . The assignment
X [ k ] ⟼ q i k − 1 / 2 P k i ( 1 ≤ k ≤ l ) X_{[k]}\longmapsto q_{i_k}^{-1/2}P_k^{\mathbf{i}}\qquad(1\leq k\leq l) X [ k ] ⟼ q i k − 1/2 P k i ( 1 ≤ k ≤ l ) extends to an A \mathbb{A} A -algebra isomorphism
Υ i : A q ( b ) → ∼ A A ( b ) . \Upsilon_{\mathbf{i}}:\mathcal{A}_{q}(\mathbf{b})\xrightarrow{\sim}\mathcal{A}_{\mathsf{A}}(\mathbf{b}). Υ i : A q ( b ) ∼ A A ( b ) . For the empty sequence, Υ ∅ = id A \Upsilon_{\varnothing}=\operatorname{id}_{\mathbb{A}} Υ ∅ = id A .
Proof. The assertion is immediate for b = 1 \mathbf{b}=1 b = 1 . For b ≠ 1 \mathbf{b}\ne1 b = 1 , Theorem 8.15 gives the stated isomorphism after extension of scalars to Q ( q 1 / 2 ) \mathbb{Q}(q^{1/2}) Q ( q 1/2 ) . By [49 Theorem 8.10(1)] , A q ( b ) \mathcal{A}_{q}(\mathbf{b}) A q ( b ) is generated over A \mathbb{A} A by the X [ k ] X_{[k]} X [ k ] . By Theorem 8.10, A A ( b ) \mathcal{A}_{\mathsf{A}}(\mathbf{b}) A A ( b ) is generated over A \mathbb{A} A by the P k i P_k^{\mathbf{i}} P k i . Since q i k − 1 / 2 ∈ A × q_{i_k}^{-1/2}\in\mathbb{A}^{\times} q i k − 1/2 ∈ A × , the isomorphism restricts to the asserted A \mathbb{A} A -algebra isomorphism. □ \square □
We identify A q ( b ) \mathcal{A}_q(\mathbf{b}) A q ( b ) with A A ( b ) \mathcal{A}_A(\mathbf{b}) A A ( b ) via Υ i \Upsilon_{\mathbf{i}} Υ i . Under this identification, the KR realization constructed above is the restriction of Ψ Q \Psi_Q Ψ Q . Indeed, with the generator correspondence in [50 Section 6.1 and footnote 3] , the bosonic map used there sends
f a , m ⟼ Φ Q ( f a , − m ) = Ψ Q ( f a , m ) ( a ∈ I , m ∈ Z ) ; f_{a,m} \longmapsto\Phi_Q(f_{a,-m})=\Psi_Q(f_{a,m}) \qquad(a\in I,\ m\in\mathbb{Z}); f a , m ⟼ Φ Q ( f a , − m ) = Ψ Q ( f a , m ) ( a ∈ I , m ∈ Z ) ;
see [24 (7.2) and Theorem 7.2] . Thus this map agrees with Ψ Q \Psi_Q Ψ Q on every generator.
For the chosen Q Q Q -adapted word j = ( j 1 , … , j N ) \mathbf{j}=(j_1,\ldots,j_N) j = ( j 1 , … , j N ) , put p r = ξ j r − 2 # { a < r ∣ j a = j r } p_r=\xi_{j_r}-2\#\{a<r\mid j_a=j_r\} p r = ξ j r − 2# { a < r ∣ j a = j r } . The initial variables satisfy
( Ψ Q ∘ Υ j ) ( X [ r min , r ] F j ) = F q ( j r ) ( M [ p r , ξ j r ] ( j r ) ) ( 1 ≤ r ≤ N ) . (\Psi_Q\circ\Upsilon_{\mathbf{j}})\left(X^{F_j}_{[r_{\min,r]}}\right)=F_q^{(j_r)}(M^{(j_r)}_{[p_r,\xi_{j_r}]})\qquad(1\leq r\leq N). ( Ψ Q ∘ Υ j ) ( X [ r m i n , r ] F j ) = F q ( j r ) ( M [ p r , ξ j r ] ( j r ) ) ( 1 ≤ r ≤ N ) .
By [50 Lemma 5.10] , this identity holds for repeated Coxeter words when ξ \xi ξ takes at most two consecutive values on each connected component. Commutations and restriction to prefixes give it for the chosen adapted word. For a source i i i of Q Q Q , [25 (8.16)] gives the mutation from the KR seed for Q Q Q to that for s i Q s_iQ s i Q , while [50 (6.3)] gives
Ψ s i Q = Ψ Q ∘ T i . \Psi_{s_iQ}=\Psi_Q\circ T_i. Ψ s i Q = Ψ Q ∘ T i .
These identities preserve the initial-variable correspondence under source reflections. Together with even shifts of the height function on each connected component, they give the same correspondence for arbitrary Q Q Q . After the chosen braid moves from j \mathbf{j} j to i \mathbf{i} i and restriction to the prefix i \mathbf{i} i , Theorem 8.15 gives the image Ψ Q ( q i − 1 / 2 p i ) \Psi_Q(q_i^{-1/2}p_i) Ψ Q ( q i − 1/2 p i ) for each singleton variable X [ k ] X_{[k]} X [ k ] in the KR realization. Since these variables generate A q ( b ) \mathcal{A}_q(\mathbf{b}) A q ( b ) , this realization is Ψ Q ∘ Υ i \Psi_Q\circ\Upsilon_{\mathbf{i}} Ψ Q ∘ Υ i . The same argument applies upstairs.
Remark 8.17. The algebra A q ( b ) \mathcal{A}_q(\mathbf{b}) A q ( b ) is defined combinatorially for an arbitrary symmetrizable Cartan datum in Definition 6.1, while A A ( b ) \mathcal{A}_A(\mathbf{b}) A A ( b ) is the positive-braid algebra introduced in [30 ] . In finite type, these algebras are isomorphic, as recalled in Corollary 8.16. We expect that
A q ( b ) ≃ A A ( b ) \mathcal{A}_q(\mathbf{b})\simeq\mathcal{A}_A(\mathbf{b}) A q ( b ) ≃ A A ( b )
for arbitrary symmetrizable Cartan data.
Surjectivity for arbitrary sequences
By Corollary 8.16, after specialization at q = 1 q=1 q = 1 and extension of scalars to C \mathbb{C} C , we identify
A ^ ( b ^ ) ⊂ K ( g ^ ) and A ( b ) ⊂ K ( g ) (64) \widehat{\mathcal{A}}(\widehat{\mathbf{b}})\subset\mathcal{K}(\widehat{\mathfrak{g}})\qquad\text{and}\qquad\mathcal{A}(\mathbf{b})\subset\mathcal{K}(\mathfrak{g}) \tag*{(64)} A ( b ) ⊂ K ( g ) and A ( b ) ⊂ K ( g ) (64)
with the corresponding specialized positive-braid algebras through Ψ ^ Q ^ \widehat{\Psi}_{\widehat{Q}} Ψ Q and Ψ Q \Psi_Q Ψ Q , respectively. Under these identifications, we shall show that π \pi π restricts to a surjective homomorphism
π b : A ^ ( b ^ ) ↠ A ( b ) . \pi_{\mathbf{b}}:\widehat{\mathcal{A}}(\widehat{\mathbf{b}})\twoheadrightarrow\mathcal{A}(\mathbf{b}). π b : A ( b ) ↠ A ( b ) .
Let b ∈ B g + \mathbf{b}\in B_{\mathfrak{g}}^+ b ∈ B g + and let i = ( i 1 , … , i l ) ∈ Seq ( b ) \mathbf{i}=(i_1,\ldots,i_l)\in\operatorname{Seq}(\mathbf{b}) i = ( i 1 , … , i l ) ∈ Seq ( b ) . Fix a maximal commuting family F \mathcal{F} F of i \mathbf{i} i -boxes, choose an admissible chain C \mathcal{C} C realizing F \mathcal{F} F , and let F ^ \widehat{\mathcal{F}} F be its canonical lift.
Theorem 8.18. Assume that g \mathfrak{g} g is of finite type. The ambient homomorphism π \pi π restricts to a surjective algebra homomorphism
π b : A ^ ( b ^ ) ↠ A ( b ) . \pi_{\mathbf{b}}:\widehat{\mathcal{A}}(\widehat{\mathbf{b}})\twoheadrightarrow\mathcal{A}(\mathbf{b}). π b : A ( b ) ↠ A ( b ) . Moreover, for any i ∈ Seq ( b ) \mathbf{i}\in\operatorname{Seq}(\mathbf{b}) i ∈ Seq ( b ) and any maximal commuting family F \mathcal{F} F of i \mathbf{i} i -boxes, under the Laurent realizations associated with its canonical lift F ^ \widehat{\mathcal{F}} F and F \mathcal{F} F , we have
π b = ϕ b F . \pi_{\mathbf{b}}=\phi_{\mathbf{b}}^{\mathcal{F}}. π b = ϕ b F .
In particular, the folding homomorphisms associated with all choices of sequences and maximal commuting families are induced by the single ambient homomorphism π \pi π .
Proof. We first compare π \pi π with the Laurent folding map on the adapted KR seeds for j ^ \hat{\boldsymbol{j}} j ^ and j \boldsymbol{j} j constructed above. By the folding formula for KR-polynomials, for each position r r r of j \boldsymbol{j} j , π \pi π sends every variable at a corresponding position of j ^ \hat{\boldsymbol{j}} j ^ to the variable at position r r r of j \boldsymbol{j} j . Hence the two maps agree on the Laurent rings of these seeds. Theorem 6.2 then gives, for every 1 ≤ k ≤ l 1 \leq k \leq l 1 ≤ k ≤ l and every c ^ ∈ Fib E ( c k ) \hat{c} \in\operatorname{Fib}^{\mathcal{E}}(c_k) c ^ ∈ Fib E ( c k ) ,
π ( x ^ c ^ ) = ϕ F ( x ^ c ^ ) = x c k . \pi(\hat{x}_{\hat{c}})=\phi^{\mathcal{F}}(\hat{x}_{\hat{c}})=x_{c_k}. π ( x ^ c ^ ) = ϕ F ( x ^ c ^ ) = x c k .
By Corollary 8.16 and specialization at q = 1 q=1 q = 1 , the cluster algebras A ( b ^ ) \mathcal{A}(\hat{\boldsymbol{b}}) A ( b ^ ) and A ( b ) \mathcal{A}(\boldsymbol{b}) A ( b ) are generated by the cluster variables corresponding to singleton boxes. It follows that
π ( A ( b ^ ) ) = A ( b ) . \pi(\mathcal{A}(\hat{\boldsymbol{b}}))=\mathcal{A}(\boldsymbol{b}). π ( A ( b ^ )) = A ( b ) .
Since π \pi π and ϕ F \phi^{\mathcal{F}} ϕ F agree on these generators, their restrictions to A ( b ) \mathcal{A}(\boldsymbol{b}) A ( b ) coincide. This proves the assertion. □
Corollary 8.19. Let i ∈ Seq ( b ) \boldsymbol{i} \in\operatorname{Seq}(\boldsymbol{b}) i ∈ Seq ( b ) , let G \mathcal{G} G be any maximal commuting family of i \boldsymbol{i} i -boxes, and let G ^ \hat{\mathcal{G}} G ^ be its canonical lift. Then
π b ( ∏ c ^ ∈ G ^ ( x ^ c ^ G ) a c ^ ) = ∏ c ∈ G ( x c G ) ∑ c ^ ∈ Fib ( c ) a c ^ ∈ L \pi_{\boldsymbol{b}}\left(\prod_{\hat{c}\in\hat{\mathcal{G}}}\left(\hat{x}_{\hat{c}}^{\mathcal{G}}\right)^{a_{\hat{c}}}\right)
=
\prod_{c\in\mathcal{G}}\left(x_c^{\mathcal{G}}\right)^{\sum_{\hat{c}\in\operatorname{Fib}(c)}a_{\hat{c}}}
\in\mathrm{L} π b c ^ ∈ G ^ ∏ ( x ^ c ^ G ) a c ^ = c ∈ G ∏ ( x c G ) ∑ c ^ ∈ Fib ( c ) a c ^ ∈ L for every ( a c ^ ) c ^ ∈ G ^ ∈ Z ≥ 0 G ^ (a_{\hat{c}})_{\hat{c}\in\hat{\mathcal{G}}}\in\mathbb{Z}_{\geq0}^{\hat{\mathcal{G}}} ( a c ^ ) c ^ ∈ G ^ ∈ Z ≥ 0 G ^ . In particular, π b \pi_{\boldsymbol{b}} π b sends every cluster monomial associated with Σ ^ G \hat{\Sigma}^{\mathcal{G}} Σ ^ G to an element of the classical global basis L \mathrm{L} L .
Proof. This follows from Theorem 6.2 and the fact that cluster monomials of K ( g ) ≤ ξ \mathcal{K}(\mathfrak{g})_{\leq\xi} K ( g ) ≤ ξ belong to the classical global basis L \mathrm{L} L . □
Remark 8.20. Theorem 8.18 is a commutative statement obtained after specializing q 1 / 2 = 1 q^{1/2}=1 q 1/2 = 1 . At the quantum level, we do not claim the existence of a folding homomorphism between the corresponding quantum cluster algebras.
Quantum cluster monomials correspond, up to the standard q q q -normalization, to global basis elements, and the folding of their g g g -vectors gives a correspondence between the associated upstairs and downstairs global basis elements. More precisely, if z ^ \hat{z} z ^ and z z z are corresponding quantum cluster monomials, then
deg i ( z ) = Fold i ( deg i ^ ( z ^ ) ) . \deg_{\boldsymbol{i}}(z)=\operatorname{Fold}_{\boldsymbol{i}}\left(\deg_{\hat{\boldsymbol{i}}}(\hat{z})\right). deg i ( z ) = Fold i ( deg i ^ ( z ^ ) ) .
Thus the quantum correspondence is detected by g g g -vector folding, rather than by an algebra homomorphism between quantum tori.
Folding of substitution isomorphisms
Let
S π : = { f ∈ Y ( g ^ ) ∣ π ( f ) ≠ 0 } . S_{\pi}:=\{f\in\mathcal{Y}(\hat{\mathfrak{g}})\mid\pi(f)\ne0\}. S π := { f ∈ Y ( g ^ ) ∣ π ( f ) = 0 } .
Then π \pi π extends uniquely to
π : S π − 1 Y ( g ^ ) ⟶ F ( g ) , f g ⟼ π ( f ) π ( g ) . \pi:S_{\pi}^{-1}\mathcal{Y}(\hat{\mathfrak{g}})\longrightarrow F(\mathfrak{g}),\qquad
\frac{f}{g}\longmapsto\frac{\pi(f)}{\pi(g)}. π : S π − 1 Y ( g ^ ) ⟶ F ( g ) , g f ⟼ π ( g ) π ( f ) .
We call the elements of S π − 1 Y ( g ^ ) S_{\pi}^{-1}\mathcal{Y}(\hat{\mathfrak{g}}) S π − 1 Y ( g ^ ) π \pi π -specializable.
Corollary 8.21. Let Q , Q ′ Q,Q' Q , Q ′ be Dynkin quivers of g \mathfrak{g} g , and let Q ^ , Q ^ ′ \widehat{Q},\widehat{Q}' Q , Q ′ be their lifted Dynkin quivers. For every ( i , p ) ∈ Δ ~ 0 g (i,p) \in\widetilde{\Delta}^{\mathfrak{g}}_0 ( i , p ) ∈ Δ 0 g and every ι ∈ I i \iota\in I_i ι ∈ I i , the rational function
Ψ ~ Q ^ , Q ^ ′ c l ( y ^ ι , p ) is π -specializable and π ( Ψ ~ Q ^ , Q ^ ′ c l ( y ^ ι , p ) ) = Ψ ~ Q , Q ′ c l ( y i , p ) . \widetilde{\Psi}^{\mathrm{cl}}_{\widehat{Q},\widehat{Q}'}(\widehat{y}_{\iota,p}) \text{ is }\pi\text{-specializable and } \pi\left(\widetilde{\Psi}^{\mathrm{cl}}_{\widehat{Q},\widehat{Q}'}(\widehat{y}_{\iota,p})\right)=\widetilde{\Psi}^{\mathrm{cl}}_{Q,Q'}(y_{i,p}). Ψ Q , Q ′ cl ( y ι , p ) is π -specializable and π ( Ψ Q , Q ′ cl ( y ι , p ) ) = Ψ Q , Q ′ cl ( y i , p ) . Proof. We first assume that ( i , p ) ∈ Δ ~ ≤ ξ ′ g (i,p) \in\widetilde{\Delta}^{\mathfrak{g}}_{\leq\xi'} ( i , p ) ∈ Δ ≤ ξ ′ g . Choose a reduced expression of w 0 w_0 w 0 adapted to Q ′ Q' Q ′ and extend it to an infinite sequence i = ( i 1 , i 2 , … ) \boldsymbol{i}=(i_1,i_2,\ldots) i = ( i 1 , i 2 , … ) by i r + ℓ ( w 0 ) = i r ∗ i_{r+\ell(w_0)}=i_r^* i r + ℓ ( w 0 ) = i r ∗ for r ≥ 1 r\geq1 r ≥ 1 . This sequence is Q ′ Q' Q ′ -adapted; see [42 (5.11)] . Then ( i , p ) = ( i k , p k ) (i,p)=(i_k,p_k) ( i , p ) = ( i k , p k ) for some k k k , where
p k = ξ i k ′ − 2 # { 1 ≤ a < k ∣ i a = i k } . p_k=\xi'_{i_k}-2\#\{1\leq a<k\mid i_a=i_k\}. p k = ξ i k ′ − 2# { 1 ≤ a < k ∣ i a = i k } . By Lemma 8.1, its blockwise lift i ^ \widehat{\boldsymbol{i}} i is Q ^ ′ \widehat{Q}' Q ′ -adapted.
By Theorem 8.2 and specialization at q = 1 q=1 q = 1 , the corresponding initial cluster variables of the Q ′ Q' Q ′ - and Q ^ ′ \widehat{Q}' Q ′ -adapted seeds are the KR polynomials
F ( m ( i ) [ p , ξ i ′ ] ) and F ( m ^ ( i ) [ p , ξ ^ l ′ ] ) , F\left(m^{(i)}[p,\xi'_i]\right) \quad\text{and}\quad F\left(\widehat{m}^{(i)}[p,\widehat{\xi}'_l]\right), F ( m ( i ) [ p , ξ i ′ ] ) and F ( m ( i ) [ p , ξ l ′ ] ) , respectively. On the other hand, under the ambient Laurent realizations of these same adapted seeds, the corresponding initial seed variables are represented by
m ( i ) [ p , ξ i ′ ] and m ^ ( i ) [ p , ξ ^ l ′ ] , m^{(i)}[p,\xi'_i]\quad\text{and}\quad\widehat{m}^{(i)}[p,\widehat{\xi}'_l], m ( i ) [ p , ξ i ′ ] and m ( i ) [ p , ξ l ′ ] , respectively; see [12 ] and [42 Proposition 5.7 and Theorem 5.12] .
Under the monomial realizations, π \pi π identifies the initial variables along their fibers, since
π ( m ^ ( i ) [ p , ξ ^ l ′ ] ) = m ( i ) [ p , ξ i ′ ] ( ι ∈ I i ) . \pi\left(\widehat{m}^{(i)}[p,\widehat{\xi}'_l]\right)=m^{(i)}[p,\xi'_i]\qquad(\iota\in I_i). π ( m ( i ) [ p , ξ l ′ ] ) = m ( i ) [ p , ξ i ′ ] ( ι ∈ I i ) . Choose the analogous infinite Q Q Q -adapted sequence. By [12 Theorem 6.3 and diagram (7.1)] and [42 Theorem 5.21 and diagram (5.19)] , the substitution isomorphism between the two monomial realizations is obtained by repeating a finite braid sequence between their reduced words for w 0 w_0 w 0 . For the variable at position k k k , use the finite transformation between sufficiently long initial segments of length a multiple of ℓ ( w 0 ) \ell(w_0) ℓ ( w 0 ) . The finite words obtained in this way represent the same power of Δ \Delta Δ . Every mutated position has a later occurrence of the same color within the subword to which the braid move applies, so all mutations are at exchangeable vertices. Theorem 6.2 , applied to this finite braid sequence and its lift, together with the initial monomial folding above, gives
π ( Ψ ~ Q ^ , Q ^ ′ c l ( m ^ ( i ) [ p , ξ ^ l ′ ] ) ) = Ψ ~ Q , Q ′ c l ( m ( i ) [ p , ξ i ′ ] ) . \pi\left(\widetilde{\Psi}^{\mathrm{cl}}_{\widehat{Q},\widehat{Q}'}\left(\widehat{m}^{(i)}[p,\widehat{\xi}'_l]\right)\right)=\widetilde{\Psi}^{\mathrm{cl}}_{Q,Q'}\left(m^{(i)}[p,\xi'_i]\right). π ( Ψ Q , Q ′ cl ( m ( i ) [ p , ξ l ′ ] ) ) = Ψ Q , Q ′ cl ( m ( i ) [ p , ξ i ′ ] ) . The same argument with p p p replaced by p + 2 p+2 p + 2 gives
π ( Ψ ~ Q ^ , Q ^ ′ c l ( m ^ ( i ) [ p + 2 , ξ ^ l ′ ] ) ) = Ψ ~ Q , Q ′ c l ( m ( i ) [ p + 2 , ξ i ′ ] ) . \pi\left(\widetilde{\Psi}^{\mathrm{cl}}_{\widehat{Q},\widehat{Q}'}\left(\widehat{m}^{(i)}[p+2,\widehat{\xi}'_l]\right)\right)=\widetilde{\Psi}^{\mathrm{cl}}_{Q,Q'}\left(m^{(i)}[p+2,\xi'_i]\right). π ( Ψ Q , Q ′ cl ( m ( i ) [ p + 2 , ξ l ′ ] ) ) = Ψ Q , Q ′ cl ( m ( i ) [ p + 2 , ξ i ′ ] ) . Using the usual empty-product convention when p = ξ i ′ p=\xi'_i p = ξ i ′ , we have
y i , p = m ( i ) [ p , ξ i ′ ] m ( i ) [ p + 2 , ξ i ′ ] − 1 , y_{i,p}=m^{(i)}[p,\xi'_i]m^{(i)}[p+2,\xi'_i]^{-1}, y i , p = m ( i ) [ p , ξ i ′ ] m ( i ) [ p + 2 , ξ i ′ ] − 1 , and, since ξ ^ l ′ = ξ i ′ \widehat{\xi}'_l=\xi'_i ξ l ′ = ξ i ′ ,
y ^ ι , p = m ^ ( i ) [ p , ξ ^ l ′ ] m ^ ( i ) [ p + 2 , ξ ^ l ′ ] − 1 . \widehat{y}_{\iota,p}=\widehat{m}^{(i)}[p,\widehat{\xi}'_l]\widehat{m}^{(i)}[p+2,\widehat{\xi}'_l]^{-1}. y ι , p = m ( i ) [ p , ξ l ′ ] m ( i ) [ p + 2 , ξ l ′ ] − 1 .
The second folding identity shows that the denominator of Ψ ~ Q ^ , Q ^ ′ c l ( y ^ i , p ) \widetilde{\Psi}^{\mathrm{cl}}_{\widehat{Q},\widehat{Q}'}(\widehat{y}_{i,p}) Ψ Q , Q ′ cl ( y i , p ) has nonzero image under π \pi π . Hence Ψ ~ Q ^ , Q ^ ′ c l ( y ^ i , p ) \widetilde{\Psi}^{\mathrm{cl}}_{\widehat{Q},\widehat{Q}'}(\widehat{y}_{i,p}) Ψ Q , Q ′ cl ( y i , p ) is π \pi π -specializable, and
π ( Ψ ~ Q ^ , Q ^ ′ c l ( y ^ i , p ) ) = Ψ ~ Q , Q ′ c l ( m ( i ) [ p , ξ i ′ ] ) Ψ ~ Q , Q ′ c l ( m ( i ) [ p + 2 , ξ i ′ ] ) = Ψ ~ Q , Q ′ c l ( y i , p ) . \pi\left(\widetilde{\Psi}^{\mathrm{cl}}_{\widehat{Q},\widehat{Q}'}(\widehat{y}_{i,p})\right)=\frac{\widetilde{\Psi}^{\mathrm{cl}}_{Q,Q'}(m^{(i)}[p,\xi'_i])}{\widetilde{\Psi}^{\mathrm{cl}}_{Q,Q'}(m^{(i)}[p+2,\xi'_i])}=\widetilde{\Psi}^{\mathrm{cl}}_{Q,Q'}(y_{i,p}). π ( Ψ Q , Q ′ cl ( y i , p ) ) = Ψ Q , Q ′ cl ( m ( i ) [ p + 2 , ξ i ′ ]) Ψ Q , Q ′ cl ( m ( i ) [ p , ξ i ′ ]) = Ψ Q , Q ′ cl ( y i , p ) .
Finally, let h ^ \widehat{h} h denote the Coxeter number of g ^ \widehat{\mathfrak{g}} g . For the finite-type admissible foldings considered here, we have
h ^ = h , I i ∗ = I i ∗ . \widehat{h}=h,\qquad I_i^*=I_{i_*}. h = h , I i ∗ = I i ∗ .
Hence
π ( D ^ ( y ^ i , p ) ) = D ( π ( y ^ i , p ) ) ( i ∈ I i ) . \pi\left(\widehat{D}(\widehat{y}_{i,p})\right)=D\left(\pi(\widehat{y}_{i,p})\right)\qquad(i\in I_i). π ( D ( y i , p ) ) = D ( π ( y i , p ) ) ( i ∈ I i ) .
Since the classical substitution isomorphisms commute with the duality automorphisms, [11 Theorem 6.3] and [42 Proposition 5.16] imply that the identity proved above for ( i , p ) ∈ Δ ^ ≤ ξ ′ g (i,p)\in\widehat{\Delta}^{\mathfrak{g}}_{\leq\xi'} ( i , p ) ∈ Δ ≤ ξ ′ g extends to every ( i , p ) ∈ Δ ^ 0 g (i,p)\in\widehat{\Delta}^{\mathfrak{g}}_0 ( i , p ) ∈ Δ 0 g .
In particular, since the substitution isomorphisms preserve the classical Grothendieck rings, Corollary 8.21 gives the commutative diagram
R ( g ^ ) → Ψ ~ Q ^ , Q ^ ′ c l R ( g ^ ) π ↓ ↓ π R ( g ) → Ψ ~ Q , Q ′ c l R ( g ) \begin{CD}
\mathcal{R}(\widehat{\mathfrak{g}}) @>{\widetilde{\Psi}^{\mathrm{cl}}_{\widehat{Q},\widehat{Q}'}}>> \mathcal{R}(\widehat{\mathfrak{g}}) \\
@V{\pi}VV @VV{\pi}V \\
\mathcal{R}(\mathfrak{g}) @>{\widetilde{\Psi}^{\mathrm{cl}}_{Q,Q'}}>> \mathcal{R}(\mathfrak{g})
\end{CD} R ( g ) π ↓ ⏐ R ( g ) Ψ Q , Q ′ cl Ψ Q , Q ′ cl R ( g ) ↓ ⏐ π R ( g )
The settings considered in § 12 and § 13 admit quantum-group-theoretic realizations, and their simply-laced upstairs cluster algebras admit monoidal categorifications. We now turn to arbitrary positive braids for general symmetrizable Cartan data, where no such representation-theoretic realization is used.
Folding for positive braids over arbitrary symmetrizable Cartan data
We prove Theorem 6.4 for an arbitrary symmetrizable Cartan datum and an arbitrary positive braid. We first show that the commutative cluster algebra is freely generated by the cluster variables attached to singleton i \mathbf{i} i -boxes.
Fix a nonempty sequence
i = ( i 1 , … , i l ) ∈ Seq ( b ) . \mathbf{i}=(i_1,\ldots,i_l)\in\operatorname{Seq}(\mathfrak{b}). i = ( i 1 , … , i l ) ∈ Seq ( b ) .
We take Σ F i \Sigma^{\mathbf{i}}_F Σ F i as the initial seed of A ( b ) \mathcal{A}(\mathfrak{b}) A ( b ) , where F i = { [ k min , k ] ∣ 1 ≤ k ≤ l } F^{\mathbf{i}}=\{[k_{\min},k]\mid1\leq k\leq l\} F i = {[ k m i n , k ] ∣ 1 ≤ k ≤ l } . By Subsection 3.3, the exchangeable set is J e x i J^{\mathbf{i}}_{\mathrm{ex}} J ex i and the exchange matrix is B i B^{\mathbf{i}} B i , given by (21) .
We recall the definition of interval variables from [49 ] for the commutative cluster algebra A ( b ) \mathcal{A}(\mathfrak{b}) A ( b ) . For p ∈ [ 1 , l ] p\in[1,l] p ∈ [ 1 , l ] , set p [ 0 ] = p p[0]=p p [ 0 ] = p and, for each integer d ≥ 0 d\geq0 d ≥ 0 with p [ d ] p[d] p [ d ] finite, define
p [ d + 1 ] : = ( p [ d ] ) + . p[d+1]:=(p[d])_+. p [ d + 1 ] := ( p [ d ] ) + .
For 1 ≤ k ≤ l 1\leq k\leq l 1 ≤ k ≤ l , set
o + i ( k ) : = # { p ∈ [ 1 , l ] ∣ k < p , i p = i k } . o_+^{\mathbf{i}}(k):=\#\{p\in[1,l]\mid k<p,\ i_p=i_k\}. o + i ( k ) := # { p ∈ [ 1 , l ] ∣ k < p , i p = i k } .
and define the mutation block
μ ( k ) : = { μ k min [ o + i ( k ) − 1 ] ⋯ μ k min [ 0 ] , o + i ( k ) > 0 , id , o + i ( k ) = 0 , (65) \mu^{(k)} := \begin{cases} \mu_{k_{\min}[o^i_+(k)-1]} \cdots\mu_{k_{\min}[0]}, & o^i_+(k)>0, \\ \operatorname{id}, & o^i_+(k)=0, \end{cases} \tag*{(65)} μ ( k ) := { μ k m i n [ o + i ( k ) − 1 ] ⋯ μ k m i n [ 0 ] , id , o + i ( k ) > 0 , o + i ( k ) = 0 , (65)
where the rightmost mutation is performed first. If o + i ( k ) > 0 o^i_+(k)>0 o + i ( k ) > 0 , then k min [ s ] ∈ J e x i k_{\min}[s]\in J^i_{\mathrm{ex}} k m i n [ s ] ∈ J ex i for every integer s s s with 0 ≤ s < o + i ( k ) 0\le s<o^i_+(k) 0 ≤ s < o + i ( k ) ; hence every mutation in μ ( k ) \mu^{(k)} μ ( k ) is performed at an exchangeable vertex.
For a seed Σ \Sigma Σ indexed by J i J^i J i and k ∈ J i k\in J^i k ∈ J i , let x k ( Σ ) x_k(\Sigma) x k ( Σ ) denote its cluster variable at vertex k k k . An empty composition of mutation blocks is understood to be the identity.
Definition 9.1 ([49 Definition 8.3] ). Let [ a , b ] [a,b] [ a , b ] be an i-box, and let d ≥ 0 d\ge0 d ≥ 0 be the unique integer such that b = a [ d ] b=a[d] b = a [ d ] . The interval variable attached to [ a , b ] [a,b] [ a , b ] is defined by
W [ a , b ] : = x a min [ d ] ( μ ( a − 1 ) ⋯ μ ( 1 ) ( Σ F i ) ) . W_{[a,b]}:=x_{a_{\min[d]}}\left(\mu^{(a-1)}\cdots\mu^{(1)}\left(\Sigma^{\mathcal{F}^i}\right)\right). W [ a , b ] := x a m i n [ d ] ( μ ( a − 1 ) ⋯ μ ( 1 ) ( Σ F i ) ) .
For 1 ≤ k ≤ l 1\le k\le l 1 ≤ k ≤ l , set
W k : = W [ k , k ] = W [ k ] . W_k:=W_{[k,k]}=W_{[k]}. W k := W [ k , k ] = W [ k ] .
The variables W 1 , … , W l W_1,\ldots,W_l W 1 , … , W l are called the fundamental variables . We also set W ∅ : = 1 W_{\varnothing}:=1 W ∅ := 1 .
Each block μ ( r ) \mu^{(r)} μ ( r ) mutates only vertices whose labels p p p satisfy i p = i r i_p=i_r i p = i r . For 1 ≤ k ≤ l 1\le k\le l 1 ≤ k ≤ l , one has i r ≠ i k i_r\ne i_k i r = i k whenever r < k min r<k_{\min} r < k m i n , so none of these blocks mutates the vertex k k k . Hence,
W [ k min , k ] = x [ k min , k ] F i ( 1 ≤ k ≤ l ) . (66) W_{[k_{\min,k]}}=x^{\mathcal{F}^i}_{[k_{\min,k]}} \qquad(1\le k\le l). \tag*{(66)} W [ k m i n , k ] = x [ k m i n , k ] F i ( 1 ≤ k ≤ l ) . (66)
The following proposition identifies the cluster variables attached to i-boxes with the corresponding interval variables.
Proposition 9.2. Let G \mathcal{G} G be a maximal commuting family of i-boxes associated with i i i . Then
x [ a , b ] G = W [ a , b ] ( [ a , b ] ∈ G ) . x^{\mathcal{G}}_{[a,b]}=W_{[a,b]} \qquad([a,b]\in\mathcal{G}). x [ a , b ] G = W [ a , b ] ([ a , b ] ∈ G ) .
Consequently, the cluster variable attached to an i-box is independent of the maximal commuting family containing it.
Proof. The exchangeable set and exchange matrix in [50 Section 3.1, equation (3.1)] , with sign + + + attached to every letter of i i i , agree with those of Σ F i \Sigma^{\mathcal{F}^i} Σ F i . The mutation blocks μ ( 1 ) , … , μ ( l ) \mu^{(1)},\ldots,\mu^{(l)} μ ( 1 ) , … , μ ( l ) also agree with those in [50 Section 3.3] , and (66) identifies the variables of the initial seed. Let x < y x<y x < y be positions of the same color i i i . By [49 Proposition 8.6] at q 1 / 2 = 1 q^{1/2}=1 q 1/2 = 1 , the interval variables satisfy
W [ x + , y ] W [ x , y − ] = W [ x , y ] W [ x + , y − ] + ∏ j ∈ I ∖ { i } , c j , i < 0 x ( j ) + ≤ y ( j ) − W [ x ( j ) + , y ( j ) − ] − c j , i . (67) W_{[x_+,y]}W_{[x,y_-]}=W_{[x,y]}W_{[x_+,y_-]}+\prod_{\substack{j\in I\setminus\{i\},\ c_{j,i}<0\\ x(j)_+\le y(j)_-}}W_{[x(j)_+,y(j)_-]}^{-c_{j,i}}. \tag*{(67)} W [ x + , y ] W [ x , y − ] = W [ x , y ] W [ x + , y − ] + j ∈ I ∖ { i } , c j , i < 0 x ( j ) + ≤ y ( j ) − ∏ W [ x ( j ) + , y ( j ) − ] − c j , i . (67)
For each color j ≠ i j\ne i j = i , the interval in Qin’s product is [ x ( j ) + , y ( j ) − ] [x(j)_+,y(j)_-] [ x ( j ) + , y ( j ) − ] , when nonempty, and its exponent is − c j , i -c_{j,i} − c j , i . Thus (67) agrees with (38) ; the conventions W ∅ = x ∅ G = 1 W_{\varnothing}=x^{\mathcal{G}}_{\varnothing}=1 W ∅ = x ∅ G = 1 cover the empty intervals.
Every maximal commuting family is connected to F i \mathcal{F}^i F i by box moves. We prove by induction on the length of a box-move sequence starting at F i \mathcal{F}^i F i that the seed Σ G \Sigma^{\mathcal{G}} Σ G satisfies
x [ a , b ] G = W [ a , b ] ( [ a , b ] ∈ G ) . x^{\mathcal{G}}_{[a,b]}=W_{[a,b]} \qquad([a,b]\in\mathcal{G}). x [ a , b ] G = W [ a , b ] ([ a , b ] ∈ G ) .
The initial case is (66) . Suppose that the assertion holds for G \mathcal{G} G , and let G ′ = B k G \mathcal{G}'=\mathcal{B}_k\mathcal{G} G ′ = B k G be obtained by one further box move. A permutation-type move preserves the variable attached to each box, so the assertion holds for G ′ \mathcal{G}' G ′ . For a mutation-type move, Proposition 3.6 and (67) give the same exchange relation. The induction hypothesis identifies all common variables and the variable being mutated. Since the mutated variable is nonzero in the ambient field, the exchange relation uniquely determines the new variable. Solving the exchange relation for the other factor gives the reverse mutation-type move. All other cluster variables are unchanged by the mutation, so the assertion holds for every box of G ′ \mathcal{G}' G ′ . This completes the induction. Since W [ a , b ] W_{[a,b]} W [ a , b ] is defined from Σ F i \Sigma^{\mathcal{F}^i} Σ F i and [ a , b ] [a,b] [ a , b ] alone, the asserted independence follows.
We henceforth write x [ a , b ] x_{[a,b]} x [ a , b ] for this family-independent variable. In particular,
x [ k ] = W k ( 1 ≤ k ≤ l ) . (68) x_{[k]}=W_k \qquad(1\leq k\leq l). \tag*{(68)} x [ k ] = W k ( 1 ≤ k ≤ l ) . (68)
The singleton box variables give the following polynomial presentation of A ( b ) \mathcal{A}(\mathsf{b}) A ( b ) .
Proposition 9.3. For every finite sequence i = ( i 1 , … , i l ) ∈ Seq ( b ) \mathbf{i}=(i_1,\ldots,i_l)\in\operatorname{Seq}(\mathsf{b}) i = ( i 1 , … , i l ) ∈ Seq ( b ) ,
A ( b ) = C [ x [ 1 ] , … , x [ l ] ] . \mathcal{A}(\mathsf{b})=\mathbb{C}[x_{[1]},\ldots,x_{[l]}]. A ( b ) = C [ x [ 1 ] , … , x [ l ] ] . More precisely, the monomials
x [ 1 ] a 1 ⋯ x [ l ] a l ( ( a 1 , … , a l ) ∈ Z ≥ 0 l ) x_{[1]}^{a_1}\cdots x_{[l]}^{a_l}\qquad((a_1,\ldots,a_l)\in\mathbb{Z}_{\geq0}^{l}) x [ 1 ] a 1 ⋯ x [ l ] a l (( a 1 , … , a l ) ∈ Z ≥ 0 l ) form a C \mathbb{C} C -basis of A ( b ) \mathcal{A}(\mathsf{b}) A ( b ) . In particular, x [ 1 ] , … , x [ l ] x_{[1]},\ldots,x_{[l]} x [ 1 ] , … , x [ l ] are algebraically independent.
Proof. For the empty sequence, the algebra is C \mathbb{C} C and the asserted basis is { 1 } \{1\} { 1 } . Suppose that l > 0 l>0 l > 0 . In the commutative case, [50 Theorem 3.6(1)] states that the monomials W 1 a 1 ⋯ W l a l W_1^{a_1}\cdots W_l^{a_l} W 1 a 1 ⋯ W l a l , with ( a 1 , … , a l ) ∈ Z ≥ 0 l (a_1,\ldots,a_l)\in\mathbb{Z}_{\geq0}^{l} ( a 1 , … , a l ) ∈ Z ≥ 0 l , form a Z \mathbb{Z} Z -basis of the Z \mathbb{Z} Z -algebra generated by the cluster variables obtained from Σ F i \Sigma^{\mathcal{F}^i} Σ F i . After extension of scalars to C \mathbb{C} C , these monomials form a C \mathbb{C} C -basis of A ( b ) \mathcal{A}(\mathsf{b}) A ( b ) . Equation (68) gives the asserted basis in the singleton box variables. This proves both generation and algebraic independence.
We now prove Theorem 6.4 using Proposition 9.3 .
Let i = ( i 1 , … , i l ) \mathbf{i}=(i_1,\ldots,i_l) i = ( i 1 , … , i l ) , F \mathcal{F} F , and C \mathcal{C} C be the sequence, maximal commuting family, and admissible chain used to define φ F \varphi^{\mathcal{F}} φ F in § 11 . Recall from (13) that κ ( k , r ) \kappa(k,r) κ ( k , r ) is the position in i ^ \widehat{\mathbf{i}} i of the r r r th entry of the sequence replacing i k i_k i k .
Proof of Theorem 6.4 . The assertion is immediate for the identity braid, for which both cluster algebras are C \mathbb{C} C . Suppose that b ≠ 1 \mathsf{b}\ne1 b = 1 .
By Proposition 9.3 ,
A ( b ) = C [ x [ k ] ∣ 1 ≤ k ≤ l ] , (69) \mathcal{A}(\mathsf{b})=\mathbb{C}[x_{[k]}\mid1\leq k\leq l], \tag*{(69)} A ( b ) = C [ x [ k ] ∣ 1 ≤ k ≤ l ] , (69)
A ( b ^ ) = C [ x ^ [ κ ( k , r ) ] ∣ 1 ≤ k ≤ l , 1 ≤ r ≤ e i k ] . \mathcal{A}(\widehat{\mathsf{b}})=\mathbb{C}[\widehat{x}_{[\kappa(k,r)]}\mid1\leq k\leq l,\ 1\leq r\leq e_{i_k}]. A ( b ) = C [ x [ κ ( k , r )] ∣ 1 ≤ k ≤ l , 1 ≤ r ≤ e i k ] .
We determine the images of the generators x ^ [ κ ( k , r ) ] \widehat{x}_{[\kappa(k,r)]} x [ κ ( k , r )] under φ F \varphi^{\mathcal{F}} φ F . Fix k ∈ [ 1 , l ] k\in[1,l] k ∈ [ 1 , l ] and choose a maximal commuting family G \mathcal{G} G containing [ k ] [k] [ k ] . Let D \mathcal{D} D be an admissible chain realizing G \mathcal{G} G . By Lemma 3.13 , the r r r th box in Fib D ( [ k ] ) \operatorname{Fib}^{\mathcal{D}}([k]) Fib D ([ k ]) , in the fixed fiber order, has box vector ε ~ k , r \widetilde{\varepsilon}_{k,r} ε k , r for 1 ≤ r ≤ e i k 1\leq r\leq e_{i_k} 1 ≤ r ≤ e i k . By (28) , each endpoint of a box indexes a nonzero coordinate of its box vector. Since the r r r th box has box vector ε ^ k , r \widehat{\varepsilon}_{k,r} ε k , r , both endpoints are κ ( k , r ) \kappa(k,r) κ ( k , r ) , and hence this box is [ κ ( k , r ) ] [\kappa(k,r)] [ κ ( k , r )] . Theorem 6.2 therefore gives
ϕ b F ( x ^ [ κ ( k , r ) ] ) = x [ k ] ( 1 ≤ k ≤ l , 1 ≤ r ≤ e i k ) . (70) \phi_b^F\left(\widehat{x}_{[\kappa(k,r)]}\right)=x_{[k]} \qquad(1\leq k\leq l,\ 1\leq r\leq e_{i_k}). \tag*{(70)} ϕ b F ( x [ κ ( k , r )] ) = x [ k ] ( 1 ≤ k ≤ l , 1 ≤ r ≤ e i k ) . (70)
The restriction ϕ b F \phi_b^F ϕ b F is well defined, since (69) and (70) show that every generator of A ^ ( b ^ ) \widehat{\mathcal{A}}(\widehat{\mathbf{b}}) A ( b ) is mapped into A ( b ) \mathcal{A}(\mathbf{b}) A ( b ) . It is surjective because every generator x [ k ] x_{[k]} x [ k ] of A ( b ) \mathcal{A}(\mathbf{b}) A ( b ) is the image of x ^ [ κ ( k , 1 ) ] \widehat{x}_{[\kappa(k,1)]} x [ κ ( k , 1 )] . □
Twist automorphisms and folding for positive braids
Kimura–Oya constructed quantum twist automorphisms of localized quantum unipotent coordinate algebras and proved that they permute the localized dual canonical bases [40 Theorem 6.1] . We express their twist formula in terms of i i i -box variables and box vectors, and use this expression to extend the construction to arbitrary positive braids. The automorphism is a twist automorphism of Donaldson–Thomas type in the sense of Kimura–Qin–Wei [41 ] . It is independent of the sequence representing the braid and permutes the common triangular basis. We prove that the commutative specializations of these twists commute with the folding homomorphisms.
Quantum twists and box vectors
Fix a positive braid b ∈ B g + \mathbf{b}\in B_{\mathfrak{g}}^+ b ∈ B g + and a sequence i = ( i 1 , … , i l ) ∈ Seq ( b ) \mathbf{i}=(i_1,\ldots,i_l)\in\operatorname{Seq}(\mathbf{b}) i = ( i 1 , … , i l ) ∈ Seq ( b ) .
We use Σ F i \Sigma^{F_{\mathbf{i}}} Σ F i as the initial quantum seed of A q ( b ) \mathcal{A}_q(\mathbf{b}) A q ( b ) . For a maximal commuting family G \mathcal{G} G of i i i -boxes, write X δ G X_{\delta}^{\mathcal{G}} X δ G for the cluster variable labelled by δ ∈ G \delta\in\mathcal{G} δ ∈ G in Σ q G \Sigma_q^{\mathcal{G}} Σ q G .
By Proposition 9.2, the specialization of X [ a , b ] G X_{[a,b]}^{\mathcal{G}} X [ a , b ] G at q 1 / 2 = 1 q^{1/2}=1 q 1/2 = 1 is independent of G \mathcal{G} G . Since specialization is injective on quantum cluster monomials [12 Lemma A.4] , the same holds for X [ a , b ] G X_{[a,b]}^{\mathcal{G}} X [ a , b ] G . We write X [ a , b ] X_{[a,b]} X [ a , b ] for this variable and set X ∅ = 1 X_{\varnothing}=1 X ∅ = 1 .
For 1 ≤ k ≤ l 1\leq k\leq l 1 ≤ k ≤ l , let k max k_{\max} k m a x be the last occurrence of i k i_k i k in i \mathbf{i} i . Set
c k : = [ k min , k ] , f k : = [ k min , k max ] , n k : = { [ k + , k max ] , k < k max , ∅ , k = k max . (71) \mathbf{c}_k:=[k_{\min},k],\qquad
\mathbf{f}_k:=[k_{\min},k_{\max}],\qquad
\mathbf{n}_k:=
\begin{cases}
[k_+,k_{\max}], & k<k_{\max},\\
\varnothing, & k=k_{\max}.
\end{cases} \tag*{(71)} c k := [ k m i n , k ] , f k := [ k m i n , k m a x ] , n k := { [ k + , k m a x ] , ∅ , k < k m a x , k = k m a x . (71)
Thus c k \mathbf{c}_k c k labels the initial variable at k k k , f k \mathbf{f}_k f k is the frozen box of color i k i_k i k , and n k \mathbf{n}_k n k contains the occurrences of this color after k k k . Their box vectors satisfy
E f k = E c k + E n k ( 1 ≤ k ≤ l ) . (72) \mathcal{E}_{\mathbf{f}_k}=\mathcal{E}_{\mathbf{c}_k}+\mathcal{E}_{\mathbf{n}_k}
\qquad(1\leq k\leq l). \tag*{(72)} E f k = E c k + E n k ( 1 ≤ k ≤ l ) . (72)
Let A ~ q ( b ) \widetilde{\mathcal{A}}_q(\mathbf{b}) A q ( b ) be the A \mathcal{A} A -subalgebra of the skew field of fractions of T ( Λ i ) \mathcal{T}(\Lambda^{\mathbf{i}}) T ( Λ i ) generated by A q ( b ) \mathcal{A}_q(\mathbf{b}) A q ( b ) and X f k − 1 X_{\mathbf{f}_k}^{-1} X f k − 1 ( k = k max ) (k=k_{\max}) ( k = k m a x ) .
For the identity braid, set A ~ q ( 1 ) = A \widetilde{\mathcal{A}}_q(1)=\mathcal{A} A q ( 1 ) = A .
Suppose now that i ∈ Red ( w ) \mathbf{i}\in\operatorname{Red}(w) i ∈ Red ( w ) for w ∈ W w\in W w ∈ W . Let A ~ A ( n ( w ) ) \widetilde{\mathcal{A}}_{\mathbb{A}}(n(w)) A A ( n ( w )) be the localization of A A ( n ( w ) ) \mathcal{A}_{\mathbb{A}}(n(w)) A A ( n ( w )) obtained by adjoining the inverses of D ( w ϖ i , ϖ i ) D(w\varpi_i,\varpi_i) D ( w ϖ i , ϖ i ) for the colors occurring in i \mathbf{i} i . Let
η w , q : A ~ A ( n ( w ) ) → ∼ A ~ A ( n ( w ) ) \eta_{w,q}:\widetilde{\mathcal{A}}_{\mathbb{A}}(n(w))\xrightarrow{\sim}\widetilde{\mathcal{A}}_{\mathbb{A}}(n(w)) η w , q : A A ( n ( w )) ∼ A A ( n ( w ))
denote the quantum twist automorphism of Kimura–Oya [40 Theorems 4.13 and 6.1] . For 1 ≤ k ≤ l 1 \le k \le l 1 ≤ k ≤ l , put ω = ϖ i k \omega=\varpi_{i_k} ω = ϖ i k . Then
η w , q ( D ( w i k ω , ω ) ) = q ( ω , ω − w i k ω ) D ( w ω , ω ) − 1 D ( w ω , w i k ω ) . (73) \eta_{w,q}\bigl(D(w_{i_k}^{\omega},\omega)\bigr)=q^{(\omega,\omega-w_{i_k}^{\omega})}D(w\omega,\omega)^{-1}D(w\omega,w_{i_k}^{\omega}). \tag*{(73)} η w , q ( D ( w i k ω , ω ) ) = q ( ω , ω − w i k ω ) D ( w ω , ω ) − 1 D ( w ω , w i k ω ) . (73)
For an i \mathbf{i} i -box d = [ a , b ] \mathbf{d}=[a,b] d = [ a , b ] , set
β d : = w < a i ϖ i a − w b i ϖ i a , D d : = D ( w b i ϖ i a , w < a i ϖ i a ) . \beta_{\mathbf{d}}:=w_{<a}^{i}\varpi_{i_a}-w_{b}^{i}\varpi_{i_a},\qquad D_{\mathbf{d}}:=D(w_{b}^{i}\varpi_{i_a},w_{<a}^{i}\varpi_{i_a}). β d := w < a i ϖ i a − w b i ϖ i a , D d := D ( w b i ϖ i a , w < a i ϖ i a ) .
Lemma 10.1 ([18 Theorem 7.3] (see also [26 §8] )). Let i ∈ Red ( w ) \mathbf{i}\in\operatorname{Red}(w) i ∈ Red ( w ) . The assignment
X d ⟼ q − ( β d , β d ) / 4 D d ( d an i -box ) X_{\mathbf{d}}\longmapsto q^{-(\beta_{\mathbf{d}},\beta_{\mathbf{d}})/4}D_{\mathbf{d}}\qquad(\mathbf{d}\text{ an }\mathbf{i}\text{-box}) X d ⟼ q − ( β d , β d ) /4 D d ( d an i -box ) extends to an A \mathbb{A} A -algebra isomorphism
A q ( w ) ⟶ ∼ A A ( n ( w ) ) . \mathcal{A}_{q}(w)\overset{\sim}{\longrightarrow}\mathcal{A}_{\mathbb{A}}(\mathfrak{n}(w)). A q ( w ) ⟶ ∼ A A ( n ( w )) . This isomorphism extends to an A ~ \widetilde{\mathbb{A}} A -algebra isomorphism
A ~ q ( w ) ⟶ ∼ A ~ A ( n ( w ) ) . \widetilde{\mathcal{A}}_{q}(w)\overset{\sim}{\longrightarrow}\widetilde{\mathcal{A}}_{\mathbb{A}}(\mathfrak{n}(w)). A q ( w ) ⟶ ∼ A A ( n ( w )) . Proof. The first assertion follows from [18 Theorem 7.3] , after applying the normalization isomorphism
A A ′ ( n ( w ) ) ⟶ ∼ A A ( n ( w ) ) \mathcal{A}'_{\mathbb{A}}(\mathfrak{n}(w))\overset{\sim}{\longrightarrow}\mathcal{A}_{\mathbb{A}}(\mathfrak{n}(w)) A A ′ ( n ( w )) ⟶ ∼ A A ( n ( w )) of [26 Remark 8.2.3 and (8.3)] . Under this isomorphism, the quantum cluster variable associated with an i \mathbf{i} i -box d \mathbf{d} d is sent to
q − ( β d , β d ) / 4 D d . q^{-(\beta_{\mathbf{d}},\beta_{\mathbf{d}})/4}D_{\mathbf{d}}. q − ( β d , β d ) /4 D d . The isomorphism extends to the localizations because each frozen variable is sent to a unit in A \mathbb{A} A times the corresponding frozen minor.
We identify A ~ q ( b ) \widetilde{\mathcal{A}}_{q}(\mathbf{b}) A q ( b ) with A ~ A ( n ( w ) ) \widetilde{\mathcal{A}}_{\mathbb{A}}(\mathfrak{n}(w)) A A ( n ( w )) by this isomorphism and write
X d = q − ( β d , β d ) / 4 D d . (74) X_{\mathbf{d}}=q^{-(\beta_{\mathbf{d}},\beta_{\mathbf{d}})/4}D_{\mathbf{d}}. \tag*{(74)} X d = q − ( β d , β d ) /4 D d . (74)
Set β ∅ = 0 \beta_{\varnothing}=0 β ∅ = 0 and D ∅ = 1 D_{\varnothing}=1 D ∅ = 1 .
The following proposition expresses the formula (73) in terms of the quantum cluster variables X d X_{\mathbf{d}} X d and the pairing λ i \lambda^{i} λ i on box vectors.
Proposition 10.2. For 1 ≤ k ≤ l 1\le k\le l 1 ≤ k ≤ l , one has
η w , q ( X c k ) = q 1 2 λ i ( ε f k , ε n k ) X f k − 1 X n k . (75) \eta_{w,q}(X_{\mathbf{c}_k})=q^{\frac{1}{2}\lambda^{i}(\varepsilon_{\mathbf{f}_k},\varepsilon_{\mathbf{n}_k})}X_{\mathbf{f}_k}^{-1}X_{\mathbf{n}_k}. \tag*{(75)} η w , q ( X c k ) = q 2 1 λ i ( ε f k , ε n k ) X f k − 1 X n k . (75) In particular, η w , q ( X f k ) = X f k − 1 \eta_{w,q}(X_{\mathbf{f}_k})=X_{\mathbf{f}_k}^{-1} η w , q ( X f k ) = X f k − 1 .
Proof. Fix k k k and put ω = ϖ i k \omega=\varpi_{i_k} ω = ϖ i k . For j ≠ i k j\ne i_k j = i k , the simple reflection s j s_j s j fixes ω \omega ω . Hence
D c k = D ( w k i ω , ω ) , β c k = ω − w k i ω , D_{\mathbf{c}_k}=D(w_{k}^{i}\omega,\omega),\qquad\beta_{\mathbf{c}_k}=\omega-w_{k}^{i}\omega, D c k = D ( w k i ω , ω ) , β c k = ω − w k i ω , D f k = D ( w ω , ω ) , β f k = ω − w ω , D_{\mathbf{f}_k}=D(w\omega,\omega),\qquad\beta_{\mathbf{f}_k}=\omega-w\omega, D f k = D ( w ω , ω ) , β f k = ω − w ω , D n k = D ( w ω , w k i ω ) , β n k = w k i ω − w ω . D_{\mathbf{n}_k}=D(w\omega,w_{k}^{i}\omega),\qquad\beta_{\mathbf{n}_k}=w_{k}^{i}\omega-w\omega. D n k = D ( w ω , w k i ω ) , β n k = w k i ω − w ω . The last line also holds for k = k max k=k_{\max} k = k m a x , using D ∅ = 1 D_{\varnothing}=1 D ∅ = 1 and β ∅ = 0 \beta_{\varnothing}=0 β ∅ = 0 . In particular,
β f k = β c k + β n k . \beta_{\mathbf{f}_k}=\beta_{\mathbf{c}_k}+\beta_{\mathbf{n}_k}. β f k = β c k + β n k .
Since ( w k i ω , w k i ω ) = ( ω , ω ) (w_k^i\omega,w_k^i\omega)=(\omega,\omega) ( w k i ω , w k i ω ) = ( ω , ω ) , we have
( β c k , β c k ) = 2 ( ω , ω ) − 2 ( ω , w k i ω ) = 2 ( ω , β c k ) . (\beta_{c_k},\beta_{c_k})=2(\omega,\omega)-2(\omega,w_k^i\omega)=2(\omega,\beta_{c_k}). ( β c k , β c k ) = 2 ( ω , ω ) − 2 ( ω , w k i ω ) = 2 ( ω , β c k ) .
Substituting (74) into (73) , we obtain
η w , q ( X c k ) = q ( ω , β c k ) − 1 4 ( β c k , β c k ) − 1 4 ( β f k , β f k ) + 1 4 ( β n k , β n k ) X f k − 1 X n k = q − 1 2 ( β c k , β n k ) X f k − 1 X n k . \eta_{w,q}(X_{c_k})=q^{(\omega,\beta_{c_k})-\frac{1}{4}(\beta_{c_k},\beta_{c_k})-\frac{1}{4}(\beta_{f_k},\beta_{f_k})+\frac{1}{4}(\beta_{n_k},\beta_{n_k})}X_{f_k}^{-1}X_{n_k}=q^{-\frac{1}{2}(\beta_{c_k},\beta_{n_k})}X_{f_k}^{-1}X_{n_k}. η w , q ( X c k ) = q ( ω , β c k ) − 4 1 ( β c k , β c k ) − 4 1 ( β f k , β f k ) + 4 1 ( β n k , β n k ) X f k − 1 X n k = q − 2 1 ( β c k , β n k ) X f k − 1 X n k .
It remains to express this exponent in terms of box vectors. For an i -box d = [ a , b ] \mathfrak{d}=[a,b] d = [ a , b ] , one has
β d = ∑ p = a b ( w i ≤ p i ϖ i a − w p i ϖ i a ) = ∑ a ≤ p ≤ b i p = i a β p i . \beta_{\mathfrak{d}}=\sum_{p=a}^{b}(w_{i_{\leq p}}^i\varpi_{i_a}-w_p^i\varpi_{i_a})=\sum_{\substack{a\leq p\leq b\\i_p=i_a}}\beta_p^i. β d = p = a ∑ b ( w i ≤ p i ϖ i a − w p i ϖ i a ) = a ≤ p ≤ b i p = i a ∑ β p i .
Every position contributing to E c k \mathcal{E}_{c_k} E c k precedes every position contributing to E n k \mathcal{E}_{n_k} E n k . Thus (15) , (72) , and skew-symmetry give
λ i ( E f k , E n k ) = λ i ( E c k , E n k ) = − ( β c k , β n k ) . \lambda^i(\mathcal{E}_{f_k},\mathcal{E}_{n_k})=\lambda^i(\mathcal{E}_{c_k},\mathcal{E}_{n_k})=-(\beta_{c_k},\beta_{n_k}). λ i ( E f k , E n k ) = λ i ( E c k , E n k ) = − ( β c k , β n k ) .
Substituting this into the formula for η w , q ( X c k ) \eta_{w,q}(X_{c_k}) η w , q ( X c k ) proves (75) . When k = k max k=k_{\max} k = k m a x , we have c k = f k c_k=f_k c k = f k and n k = ∅ n_k=\varnothing n k = ∅ . Since X ∅ = 1 X_{\varnothing}=1 X ∅ = 1 and E ∅ = 0 \mathcal{E}_{\varnothing}=0 E ∅ = 0 , the formula reduces to η w , q ( X f k ) = X f k − 1 \eta_{w,q}(X_{f_k})=X_{f_k}^{-1} η w , q ( X f k ) = X f k − 1 .
Quantum twists for arbitrary positive braids
The i -box variables and the pairing on box vectors in (75) are defined for arbitrary sequences. Motivated by this formula, we define a quantum twist automorphism for an arbitrary positive braid.
Fix a nonidentity positive braid b ∈ B g + \mathbf{b}\in B_g^+ b ∈ B g + and a sequence
i = ( i 1 , … , i l ) ∈ Seq ( b ) . \mathbf{i}=(i_1,\ldots,i_l)\in\operatorname{Seq}(\mathbf{b}). i = ( i 1 , … , i l ) ∈ Seq ( b ) .
For a quantum torus R R R , we write F ( R ) \operatorname{F}(R) F ( R ) for its skew field of fractions.
Lemma 10.3. The assignment
X c k ⟼ q 1 2 λ i ( E f k , E n k ) X f k − 1 X n k ( 1 ≤ k ≤ l ) (76) X_{c_k}\longmapsto q^{\frac{1}{2}\lambda^i(\mathcal{E}_{f_k},\mathcal{E}_{n_k})}X_{f_k}^{-1}X_{n_k}\qquad(1\leq k\leq l) \tag*{(76)} X c k ⟼ q 2 1 λ i ( E f k , E n k ) X f k − 1 X n k ( 1 ≤ k ≤ l ) (76) extends uniquely to a Q ( q 1 / 2 ) \mathbb{Q}(q^{1/2}) Q ( q 1/2 ) -algebra automorphism of F ( T ( Λ i ) ) \operatorname{F}(\mathcal{T}(\Lambda^i)) F ( T ( Λ i )) .
Proof. Set
G i : = { [ k , k max ] ∣ 1 ≤ k ≤ l } . \mathcal{G}^i:=\{[k,k_{\max}]\mid1\leq k\leq l\}. G i := {[ k , k m a x ] ∣ 1 ≤ k ≤ l } .
This family is associated with the constant L \mathcal{L} L -sequence. We compare the quantum seeds
Σ q F i = ( X F i , B i , Λ i ) and Σ q G i = ( X G i , B G i , Λ G i ) , \Sigma_q^{\mathcal{F}^i}=(X^{\mathcal{F}^i},B^i,\Lambda^i)
\qquad\text{and}\qquad
\Sigma_q^{\mathcal{G}^i}=(X^{\mathcal{G}^i},B^{\mathcal{G}^i},\Lambda^{\mathcal{G}^i}), Σ q F i = ( X F i , B i , Λ i ) and Σ q G i = ( X G i , B G i , Λ G i ) ,
whose based quantum tori are T ( Λ i ) \mathcal{T}(\Lambda^i) T ( Λ i ) and T ( Λ G i ) \mathcal{T}(\Lambda^{\mathcal{G}^i}) T ( Λ G i ) , respectively. The cluster of the latter seed is
{ X n k ∣ k < k max } ∪ { X f k ∣ k = k max } . \{X_{n_k}\mid k<k_{\max}\}\cup\{X_{f_k}\mid k=k_{\max}\}. { X n k ∣ k < k m a x } ∪ { X f k ∣ k = k m a x } .
Write T k T_k T k for the image prescribed by (76) . Each T k T_k T k is a based monomial in T ( Λ G i ) \mathcal{T}(\Lambda^{\mathcal{G}^i}) T ( Λ G i ) . By (72) ,
E n k − E f k = − E c k . \mathcal{E}_{n_k}-\mathcal{E}_{f_k}=-\mathcal{E}_{c_k}. E n k − E f k = − E c k .
Hence (31) gives
λ i ( E n k − E f k , E n p − E f p ) = λ i ( − E c k , − E c p ) = Λ k , p i . \lambda^i(\mathcal{E}_{n_k}-\mathcal{E}_{f_k},\mathcal{E}_{n_p}-\mathcal{E}_{f_p})
=\lambda^i(-\mathcal{E}_{c_k},-\mathcal{E}_{c_p})
=\Lambda^i_{k,p}. λ i ( E n k − E f k , E n p − E f p ) = λ i ( − E c k , − E c p ) = Λ k , p i .
Thus T k T p = q λ k , p i T p T k T_kT_p=q^{\lambda^{i}_{k,p}}T_pT_k T k T p = q λ k , p i T p T k , and the assignment defines an A \mathbb{A} A -algebra homomorphism
Φ : T ( Λ i ) ⟶ T ( Λ G i ) , X c k ⟼ T k . \Phi:\mathcal{T}(\Lambda^i)\longrightarrow\mathcal{T}(\Lambda^{G^i}),\qquad X_{c_k}\longmapsto T_k. Φ : T ( Λ i ) ⟶ T ( Λ G i ) , X c k ⟼ T k .
For each k k k , we have f k max = f k f_{k_{\max}}=f_k f k m a x = f k and T k max = X f k − 1 T_{k_{\max}}=X_{f_k}^{-1} T k m a x = X f k − 1 . Consequently,
X f k = T k max − 1 ( 1 ≤ k ≤ l ) X_{f_k}=T_{k_{\max}}^{-1}\qquad(1\leq k\leq l) X f k = T k m a x − 1 ( 1 ≤ k ≤ l )
and
X n k = q − 1 2 λ i ( ε f k , ε n k ) T k max − 1 T k ( k < k max ) . X_{n_k}=q^{-\frac{1}{2}\lambda^i(\varepsilon_{f_k},\varepsilon_{n_k})}T_{k_{\max}}^{-1}T_k\qquad(k<k_{\max}). X n k = q − 2 1 λ i ( ε f k , ε n k ) T k m a x − 1 T k ( k < k m a x ) .
These identities show that the exponent vectors of T 1 , … , T l T_1,\ldots,T_l T 1 , … , T l form a Z \mathbb{Z} Z -basis of the exponent lattice of T ( Λ G i ) \mathcal{T}(\Lambda^{G^i}) T ( Λ G i ) . Thus Φ \Phi Φ maps the based monomial basis of T ( Λ i ) \mathcal{T}(\Lambda^i) T ( Λ i ) bijectively onto that of T ( Λ G i ) \mathcal{T}(\Lambda^{G^i}) T ( Λ G i ) and is therefore an A \mathbb{A} A -algebra isomorphism.
Since Σ q F i \Sigma^{F^i}_q Σ q F i and Σ q G i \Sigma^{G^i}_q Σ q G i are mutation-equivalent, we have
F ( T ( Λ i ) ) = F ( T ( Λ G i ) ) . \mathbb{F}(\mathcal{T}(\Lambda^i))=\mathbb{F}(\mathcal{T}(\Lambda^{G^i})). F ( T ( Λ i )) = F ( T ( Λ G i )) .
Hence Φ \Phi Φ extends uniquely to a Q ( q 1 / 2 ) \mathbb{Q}(q^{1/2}) Q ( q 1/2 ) -algebra automorphism of this common skew field. □
Definition 10.4. The quantum twist automorphism associated with i i i is the Q ( q 1 / 2 ) \mathbb{Q}(q^{1/2}) Q ( q 1/2 ) -algebra automorphism τ i , q \tau_{i,q} τ i , q of F ( T ( Λ i ) ) \mathbb{F}(\mathcal{T}(\Lambda^i)) F ( T ( Λ i )) determined by (76) , namely,
τ i , q ( X c k ) = q 1 2 λ i ( ε f k , ε n k ) X f k − 1 X n k ( 1 ≤ k ≤ l ) . \tau_{i,q}(X_{c_k})=q^{\frac{1}{2}\lambda^i(\varepsilon_{f_k},\varepsilon_{n_k})}X_{f_k}^{-1}X_{n_k}\qquad(1\leq k\leq l). τ i , q ( X c k ) = q 2 1 λ i ( ε f k , ε n k ) X f k − 1 X n k ( 1 ≤ k ≤ l ) .
We recall the notion of a twist automorphism of Donaldson–Thomas type from [41 ] , restricting to integral powers of the frozen variables.
Let Σ q = ( X , B , Λ ) \Sigma_q=(X,B,\Lambda) Σ q = ( X , B , Λ ) be a quantum seed indexed by J J J , with exchangeable set J e x J_{\mathrm{ex}} J ex , and put J f r = J ∖ J e x J_{\mathrm{fr}}=J\setminus J_{\mathrm{ex}} J fr = J ∖ J ex . If Ξ q \Xi_q Ξ q is a quantum seed indexed by J J J , we denote by X j ( Ξ q ) X_j(\Xi_q) X j ( Ξ q ) its quantum cluster variable at j ∈ J j\in J j ∈ J . Let μ \mu μ be a sequence of mutations and write Σ q ′ = μ ( Σ q ) = ( Y , B ′ , Λ ′ ) \Sigma'_q=\mu(\Sigma_q)=(Y,B',\Lambda') Σ q ′ = μ ( Σ q ) = ( Y , B ′ , Λ ′ ) . Write ( ε j ) j ∈ J (\varepsilon_j)_{j\in J} ( ε j ) j ∈ J and ( ε k ) k ∈ J e x (\varepsilon_k)_{k\in J_{\mathrm{ex}}} ( ε k ) k ∈ J ex for the standard bases of Z J \mathbb{Z}^J Z J and Z J e x \mathbb{Z}^{J_{\mathrm{ex}}} Z J ex , respectively. Let σ \sigma σ be a permutation of J e x J_{\mathrm{ex}} J ex preserving the skew-symmetrizer. Assume, as in [41 Section 2.12] , that there are integers a j , k a_{j,k} a j , k ( j ∈ J f r , k ∈ J e x ) (j\in J_{\mathrm{fr}}, k\in J_{\mathrm{ex}}) ( j ∈ J fr , k ∈ J ex ) such that
deg Σ q ( Y σ ( k ) ) = − ε k + ∑ j ∈ J f r a j , k ε j ( k ∈ J e x ) . (77) \deg_{\Sigma_q}(Y_{\sigma(k)})=-\varepsilon_k+\sum_{j\in J_{\mathrm{fr}}}a_{j,k}\varepsilon_j\qquad(k\in J_{\mathrm{ex}}). \tag*{(77)} deg Σ q ( Y σ ( k ) ) = − ε k + j ∈ J fr ∑ a j , k ε j ( k ∈ J ex ) . (77)
For the fixed seeds Σ q \Sigma_q Σ q and Σ q ′ \Sigma'_q Σ q ′ , condition (77) determines σ \sigma σ uniquely. Indeed, the restrictions of deg Σ q ( Y j ) \deg_{\Sigma_q}(Y_j) deg Σ q ( Y j ) , j ∈ J e x j\in J_{\mathrm{ex}} j ∈ J ex , to the exchangeable coordinates are the distinct vectors − ε k -\varepsilon_k − ε k , k ∈ J e x k\in J_{\mathrm{ex}} k ∈ J ex , each occurring once, so σ ( k ) \sigma(k) σ ( k ) is the unique index for which this restriction equals − ε k -\varepsilon_k − ε k .
Definition 10.5 ([41 Definitions 3.3 and 3.15 and Section 5.1] ). A Q ( q 1 / 2 ) \mathbb{Q}(q^{1/2}) Q ( q 1/2 ) -algebra automorphism ψ \psi ψ of the ambient skew field is a twist automorphism of Donaldson–Thomas type passing through Σ q ′ \Sigma'_q Σ q ′ if there are vectors v j ∈ Z J v_j\in\mathbb{Z}^J v j ∈ Z J ( j ∈ J ) (j\in J) ( j ∈ J ) supported on J f r J_{\mathrm{fr}} J fr such that
ψ ( X k ) = Y ε σ ( k ) + v k ( k ∈ J e x ) , ψ ( X j ) = Y v j ( j ∈ J f r ) , \psi(X_k)=Y^{\varepsilon_{\sigma(k)}+v_k}\qquad(k\in J_{\mathrm{ex}}),\qquad\psi(X_j)=Y^{v_j}\qquad(j\in J_{\mathrm{fr}}), ψ ( X k ) = Y ε σ ( k ) + v k ( k ∈ J ex ) , ψ ( X j ) = Y v j ( j ∈ J fr ) , and
ψ ( X B ε k ) = Y B ′ ε σ ( k ) ( k ∈ J e x ) . \psi(X^{B\varepsilon_k})=Y^{B'\varepsilon_{\sigma(k)}}\qquad(k\in J_{\mathrm{ex}}). ψ ( X B ε k ) = Y B ′ ε σ ( k ) ( k ∈ J ex ) .
A based monomial in a quantum seed is called a localized quantum cluster monomial if its exponents at all exchangeable vertices are nonnegative.
Theorem 10.6. The automorphism τ i , q \tau_{i,q} τ i , q is a twist automorphism of Donaldson–Thomas type and restricts to an A \mathbb{A} A -algebra automorphism
τ i , q : A ~ q ( b ) → ∼ A ~ q ( b ) . \tau_{i,q}:\widetilde{\mathcal{A}}_q(\mathbf{b})\xrightarrow{\sim}\widetilde{\mathcal{A}}_q(\mathbf{b}). τ i , q : A q ( b ) ∼ A q ( b ) . Both τ i , q \tau_{i,q} τ i , q and τ ~ i , q − 1 \widetilde{\tau}_{i,q}^{-1} τ i , q − 1 send every quantum cluster variable to a localized quantum cluster monomial.
Proof. Recall the mutation blocks μ ( k ) \mu^{(k)} μ ( k ) defined in (65) . We use the same notation for the corresponding mutation sequences acting on quantum seeds. Set
Σ q = Σ q F i = ( X , B i , Λ i ) , μ i = μ ( l ) ⋯ μ ( 1 ) , Σ q ′ = μ i ( Σ q ) = ( Y , B ′ , Λ ′ ) , \Sigma_q=\Sigma_q^{F_i}=(\mathbf{X},B^i,\Lambda^i),\qquad\mu_i=\mu^{(l)}\cdots\mu^{(1)},\qquad\Sigma'_q=\mu_i(\Sigma_q)=(\mathbf{Y},B',\Lambda'), Σ q = Σ q F i = ( X , B i , Λ i ) , μ i = μ ( l ) ⋯ μ ( 1 ) , Σ q ′ = μ i ( Σ q ) = ( Y , B ′ , Λ ′ ) ,
and write J f r i = J i ∖ J e x i = { j ∈ J i ∣ j = j max } J^i_{\mathrm{fr}}=J^i\setminus J^i_{\mathrm{ex}}=\{j\in J^i\mid j=j_{\max}\} J fr i = J i ∖ J ex i = { j ∈ J i ∣ j = j m a x } . If J e x i = ∅ J^i_{\mathrm{ex}}=\varnothing J ex i = ∅ , then τ i , q \tau_{i,q} τ i , q inverts every cluster variable, and the assertion is immediate.
We first identify the variables of Σ q ′ \Sigma'_q Σ q ′ . For each color occurring at p 0 < ⋯ < p m − 1 p_0<\cdots<p_{m-1} p 0 < ⋯ < p m − 1 with m ≥ 2 m\ge2 m ≥ 2 , define
σ ( p r ) = p m − 2 − r ( 0 ≤ r ≤ m − 2 ) . (78) \sigma(p_r)=p_{m-2-r}\qquad(0\le r\le m-2). \tag*{(78)} σ ( p r ) = p m − 2 − r ( 0 ≤ r ≤ m − 2 ) . (78)
This permutation preserves colors and hence the skew-symmetrizer.
Fix k = p r ∈ J e x i k=p_r\in J^i_{\mathrm{ex}} k = p r ∈ J ex i . By Definition 9.1, W [ k + , k max ] W_{[k_+,k_{\max}]} W [ k + , k m a x ] occurs at σ ( k ) = p m − 2 − r \sigma(k)=p_{m-2-r} σ ( k ) = p m − 2 − r after applying μ ( k + − 1 ) ⋯ μ ( 1 ) \mu^{(k_+-1)}\cdots\mu^{(1)} μ ( k + − 1 ) ⋯ μ ( 1 ) . Each subsequent mutation block of the same color is μ ( p s ) \mu^{(p_s)} μ ( p s ) with s ≥ r + 1 s\ge r+1 s ≥ r + 1 and mutates only vertices p t p_t p t with
0 ≤ t < m − 1 − s ≤ m − 2 − r . 0\le t<m-1-s\le m-2-r. 0 ≤ t < m − 1 − s ≤ m − 2 − r .
Blocks of other colors do not mutate σ ( k ) \sigma(k) σ ( k ) either. Thus Y σ ( k ) Y_{\sigma(k)} Y σ ( k ) specializes to W [ k + , k max ] W_{[k_+,k_{\max}]} W [ k + , k m a x ] , which is also the specialization of X n k X_{n_k} X n k by Proposition 9.2. Since quantum cluster variables are uniquely determined by their classical specializations, we have Y σ ( k ) = X n k Y_{\sigma(k)}=X_{n_k} Y σ ( k ) = X n k . Together with [49 Proposition 8.7] , this gives
Y σ ( k ) = X n k , deg Σ q ( Y σ ( k ) ) = − ε k + ε k max ( k ∈ J e x i ) . (79) Y_{\sigma(k)}=X_{n_k},\qquad\deg_{\Sigma_q}(Y_{\sigma(k)})=-\varepsilon_k+\varepsilon_{k_{\max}}\qquad(k\in J^i_{\mathrm{ex}}). \tag*{(79)} Y σ ( k ) = X n k , deg Σ q ( Y σ ( k ) ) = − ε k + ε k m a x ( k ∈ J ex i ) . (79)
Frozen variables are unchanged by mutation, so
Y j = X f j , deg Σ q ( Y j ) = ε j ( j ∈ J f r i ) . (80) Y_j=X_{f_j},\qquad\deg_{\Sigma_q}(Y_j)=\varepsilon_j\qquad(j\in J^i_{\mathrm{fr}}). \tag*{(80)} Y j = X f j , deg Σ q ( Y j ) = ε j ( j ∈ J fr i ) . (80)
In particular, (77) holds with a j , k = δ j , k max a_{j,k}=\delta_{j,k_{\max}} a j , k = δ j , k m a x .
We now express τ i , q \tau_{i,q} τ i , q in the variables of Σ q ′ \Sigma'_q Σ q ′ . By (79) , (80) , and (31) , we have
Λ k max , σ ( k ) ′ = λ i ( e f k , e n k ) ( k ∈ J e x i ) . \Lambda'_{k_{\max},\sigma(k)}=\lambda^i(\mathbf{e}_{f_k},\mathbf{e}_{n_k})\qquad(k\in J^i_{\mathrm{ex}}). Λ k m a x , σ ( k ) ′ = λ i ( e f k , e n k ) ( k ∈ J ex i ) .
Hence (76) , with the normalization of based monomials, becomes
τ i , q ( X c k ) = { Y ε σ ( k ) − ε k max , k ∈ J e x i , Y − ε k , k ∈ J f r i . (81) \tau_{i,q}(X_{c_k})=
\begin{cases}
Y^{\varepsilon_{\sigma(k)}-\varepsilon_{k_{\max}}}, & k\in J^i_{\mathrm{ex}},\\
Y^{-\varepsilon_k}, & k\in J^i_{\mathrm{fr}}.
\end{cases} \tag*{(81)} τ i , q ( X c k ) = { Y ε σ ( k ) − ε k m a x , Y − ε k , k ∈ J ex i , k ∈ J fr i . (81)
It remains to verify
τ i , q ( X B ε k ) = Y B ′ ε σ ( k ) ( k ∈ J e x i ) . (82) \tau_{i,q}(X^{B\varepsilon_k})=Y^{B'\varepsilon_{\sigma(k)}}\qquad(k\in J^i_{\mathrm{ex}}). \tag*{(82)} τ i , q ( X B ε k ) = Y B ′ ε σ ( k ) ( k ∈ J ex i ) . (82)
Let G G G be the matrix whose j j j -th column is deg Σ q ( Y j ) \deg_{\Sigma_q}(Y_j) deg Σ q ( Y j ) .5 The degree formulas above show that G G G is unimodular and that
G ( ε σ ( k ) − ε k max ) = − ε k ( k ∈ J e x i ) , G ( − ε j ) = − ε j ( j ∈ J f r i ) . G(\varepsilon_{\sigma(k)}-\varepsilon_{k_{\max}})=-\varepsilon_k\qquad(k\in J^i_{\mathrm{ex}}),\qquad G(-\varepsilon_j)=-\varepsilon_j\qquad(j\in J^i_{\mathrm{fr}}). G ( ε σ ( k ) − ε k m a x ) = − ε k ( k ∈ J ex i ) , G ( − ε j ) = − ε j ( j ∈ J fr i ) .
Consequently,
τ i , q ( X a ) = Y − G − 1 a ( a ∈ Z J i ) . \tau_{i,q}(X^{\mathbf a})=Y^{-G^{-1}\mathbf a}\qquad(\mathbf a\in\mathbb Z^{J^i}). τ i , q ( X a ) = Y − G − 1 a ( a ∈ Z J i ) .
Since the exchangeable parts of the columns of G G G indexed by σ ( k ) \sigma(k) σ ( k ) are − ε k -\varepsilon_k − ε k and σ \sigma σ preserves the skew-symmetrizer, [41 equation (2.4) and Lemma 2.15(1)] give
G B ′ ε σ ( k ) = − B i ε k ( k ∈ J e x i ) . GB'\varepsilon_{\sigma(k)}=-B^i\varepsilon_k\qquad(k\in J^i_{\mathrm{ex}}). G B ′ ε σ ( k ) = − B i ε k ( k ∈ J ex i ) .
Therefore
τ i , q ( X B i ε k ) = Y − G − 1 B i ε k = Y B ′ ε σ ( k ) , \tau_{i,q}\left(X^{B^i\varepsilon_k}\right)=Y^{-G^{-1}B^i\varepsilon_k}=Y^{B'\varepsilon_{\sigma(k)}}, τ i , q ( X B i ε k ) = Y − G − 1 B i ε k = Y B ′ ε σ ( k ) ,
which verifies the remaining condition in Definition 10.5 .
Finally, (81) gives
τ i , q − 1 ( Y σ ( k ) ) = X ε k − ε k max ( k ∈ J e x i ) , τ i , q − 1 ( Y j ) = X − ε j ( j ∈ J f r i ) . (83) \tau_{i,q}^{-1}(Y_{\sigma(k)})=X^{\varepsilon_k-\varepsilon_{k_{\max}}}\qquad(k\in J^i_{\mathrm{ex}}),\qquad\tau_{i,q}^{-1}(Y_j)=X^{-\varepsilon_j}\qquad(j\in J^i_{\mathrm{fr}}). \tag*{(83)} τ i , q − 1 ( Y σ ( k ) ) = X ε k − ε k m a x ( k ∈ J ex i ) , τ i , q − 1 ( Y j ) = X − ε j ( j ∈ J fr i ) . (83)
Applying τ i , q − 1 \tau_{i,q}^{-1} τ i , q − 1 to (82) shows that τ i , q − 1 \tau_{i,q}^{-1} τ i , q − 1 satisfies the same condition with the two seeds interchanged. By [41 Propositions 3.12 and 3.14] , the identities in (83) , with integral frozen exponents, are preserved under mutations at corresponding vertices. Iterating along a mutation sequence therefore shows that both τ i , q \tau_{i,q} τ i , q and τ i , q − 1 \tau_{i,q}^{-1} τ i , q − 1 send every exchangeable quantum cluster variable to a cluster variable multiplied by a frozen Laurent monomial, with the based normalization. These are localized quantum cluster monomials in seeds mutation-equivalent to Σ q \Sigma_q Σ q .
Thus both maps send every quantum cluster variable into A ~ q ( b ) \widetilde{\mathcal A}_q(\mathbf b) A q ( b ) . They also invert the frozen variables and fix A \mathcal A A . Since A ~ q ( b ) \widetilde{\mathcal A}_q(\mathbf b) A q ( b ) is generated over A \mathcal A A by the quantum cluster variables and the inverses of the frozen variables, both maps preserve A ~ q ( b ) \widetilde{\mathcal A}_q(\mathbf b) A q ( b ) . Their restrictions are mutually inverse, proving the assertion. □
Lemma 10.7. Let i ∈ Seq ( b ) i\in\operatorname{Seq}(\mathbf b) i ∈ Seq ( b ) . For every quantum seed Σ q \Sigma_q Σ q mutation-equivalent to Σ q F i \Sigma_q^{F^i} Σ q F i , we have
deg Σ q τ i , q ( X j ( Σ q ) ) = − ε j ( j ∈ J i ) . (84) \deg_{\Sigma_q}\tau_{i,q}(X_j(\Sigma_q))=-\varepsilon_j\qquad(j\in J^i). \tag*{(84)} deg Σ q τ i , q ( X j ( Σ q )) = − ε j ( j ∈ J i ) . (84) Proof. By Theorem 10.6 , the images under τ i , q \tau_{i,q} τ i , q of quantum cluster variables are localized quantum cluster monomials. Quantum cluster monomials are pointed in every seed, as recalled in § 7 . Multiplication by a frozen Laurent monomial shifts the degree in (18) by its exponent vector. Hence localized quantum cluster monomials are also pointed in every seed, and their degrees satisfy (19) .
We prove (84) by induction on the length of a mutation sequence from Σ q F i \Sigma_q^{F^i} Σ q F i . For the initial seed, (79) and (81) give
deg Σ q F i τ i , q ( X c k ) = ( − ε k + ε k max ) − ε k max = − ε k ( k ∈ J e x i ) . \deg_{\Sigma_q^{F^i}}\tau_{i,q}(X_{c_k})=(-\varepsilon_k+\varepsilon_{k_{\max}})-\varepsilon_{k_{\max}}=-\varepsilon_k\qquad(k\in J^i_{\mathrm{ex}}). deg Σ q F i τ i , q ( X c k ) = ( − ε k + ε k m a x ) − ε k m a x = − ε k ( k ∈ J ex i ) . The assertion also holds at frozen vertices, since τ i , q \tau_{i,q} τ i , q inverts the frozen variables.
Suppose that (84) holds for Σ q = ( X , B , Λ ) \Sigma_q=(\mathbf X,B,\Lambda) Σ q = ( X , B , Λ ) . Let k ∈ J e x i k\in J^i_{\mathrm{ex}} k ∈ J ex i and Σ q ′ = μ k ( Σ q ) = ( X ′ , B ′ , Λ ′ ) \Sigma'_q=\mu_k(\Sigma_q)=(\mathbf X',B',\Lambda') Σ q ′ = μ k ( Σ q ) = ( X ′ , B ′ , Λ ′ ) . Set
b ± = ∑ j ∈ J i [ ± b j , k ] + ε j . \mathbf b^\pm=\sum_{j\in J^i}[\pm b_{j,k}]_+\varepsilon_j. b ± = j ∈ J i ∑ [ ± b j , k ] + ε j . The quantum exchange relation has the form
X k ( Σ q ) X k ( Σ q ′ ) = q r + X b + + q r − X b − , r ± ∈ 1 2 Z , X_k(\Sigma_q)X_k(\Sigma'_q)=q^{r_+}X^{\mathbf b^+}+q^{r_-}X^{\mathbf b^-},\qquad r_\pm\in\tfrac{1}{2}\mathbb Z, X k ( Σ q ) X k ( Σ q ′ ) = q r + X b + + q r − X b − , r ± ∈ 2 1 Z ,
where the based monomials on the right are taken in Σ q \Sigma_q Σ q . Applying τ i , q \tau_{i,q} τ i , q gives
τ i , q ( X k ( Σ q ) ) τ i , q ( X k ( Σ q ′ ) ) = q r + τ i , q ( X b + ) + q r − τ i , q ( X b − ) . \tau_{i,q}\left(X_k(\Sigma_q)\right)\tau_{i,q}\left(X_k(\Sigma'_q)\right)=q^{r^+}\tau_{i,q}\left(X^{\mathbf b^+}\right)+q^{r^-}\tau_{i,q}\left(X^{\mathbf b^-}\right). τ i , q ( X k ( Σ q ) ) τ i , q ( X k ( Σ q ′ ) ) = q r + τ i , q ( X b + ) + q r − τ i , q ( X b − ) .
By the induction hypothesis and additivity of degrees under multiplication, the two summands on the right are pointed at − b + -\mathbf b^+ − b + and − b − -\mathbf b^- − b − , respectively. Since
− b − = − b + + B e k , -\mathbf b^-=-\mathbf b^+ + B\mathbf e_k, − b − = − b + + B e k ,
their expansions in (18) take the form
q r + τ i , q ( X b + ) = q s + X − b + + ∑ n ∈ N e x J i ∖ { 0 } c n X − b + + B n , q^{r^+}\tau_{i,q}\left(X^{\mathbf b^+}\right)=q^{s^+}X^{-\mathbf b^+}+\sum_{\mathbf n\in\mathbb N^{J^i}_{\mathrm{ex}}\setminus\{0\}}c_{\mathbf n}X^{-\mathbf b^++B\mathbf n}, q r + τ i , q ( X b + ) = q s + X − b + + n ∈ N ex J i ∖ { 0 } ∑ c n X − b + + B n ,
q r − τ i , q ( X b − ) = ∑ n ∈ N e x J i d n X − b + + B ( e k + n ) . q^{r^-}\tau_{i,q}\left(X^{\mathbf b^-}\right)=\sum_{\mathbf n\in\mathbb N^{J^i}_{\mathrm{ex}}}d_{\mathbf n}X^{-\mathbf b^++B(\mathbf e_k+\mathbf n)}. q r − τ i , q ( X b − ) = n ∈ N ex J i ∑ d n X − b + + B ( e k + n ) .
Compatibility implies that B B B has full column rank. Thus B ( e k + n ) ≠ 0 B(\mathbf e_k+\mathbf n)\ne0 B ( e k + n ) = 0 for every n ∈ N e x J i \mathbf n\in\mathbb N^{J^i}_{\mathrm{ex}} n ∈ N ex J i , so the second summand contains no term with exponent − b + -\mathbf b^+ − b + . The leading term q s + X − b + q^{s^+}X^{-\mathbf b^+} q s + X − b + therefore cannot be cancelled, and the sum is pointed at − b + -\mathbf b^+ − b + .
Both factors on the left are pointed in Σ q \Sigma_q Σ q , and the induction hypothesis gives
deg Σ q τ i , q ( X k ( Σ q ) ) = − e k . \deg_{\Sigma_q}\tau_{i,q}\left(X_k(\Sigma_q)\right)=-\mathbf e_k. deg Σ q τ i , q ( X k ( Σ q ) ) = − e k .
Comparing degrees on both sides, we obtain
− e k + deg Σ q τ i , q ( X k ( Σ q ′ ) ) = − b + . -\mathbf e_k+\deg_{\Sigma_q}\tau_{i,q}\left(X_k(\Sigma'_q)\right)=-\mathbf b^+. − e k + deg Σ q τ i , q ( X k ( Σ q ′ ) ) = − b + .
Write
g = ( g j ) j ∈ J i : = deg Σ q τ i , q ( X k ( Σ q ′ ) ) = e k − b + , \mathbf g=(g_j)_{j\in J_i}:=\deg_{\Sigma_q}\tau_{i,q}\left(X_k(\Sigma'_q)\right)=\mathbf e_k-\mathbf b^+, g = ( g j ) j ∈ J i := deg Σ q τ i , q ( X k ( Σ q ′ ) ) = e k − b + ,
g ′ = ( g j ′ ) j ∈ J i : = deg Σ q ′ τ i , q ( X k ( Σ q ′ ) ) . \mathbf g'=(g'_j)_{j\in J_i}:=\deg_{\Sigma'_q}\tau_{i,q}\left(X_k(\Sigma'_q)\right). g ′ = ( g j ′ ) j ∈ J i := deg Σ q ′ τ i , q ( X k ( Σ q ′ ) ) .
Since b k , k = 0 b_{k,k}=0 b k , k = 0 , the coordinates of g \mathbf g g are
g k = 1 , g j = − [ b j , k ] + ( j ≠ k ) . g_k=1,\qquad g_j=-[b_{j,k}]_+\quad(j\ne k). g k = 1 , g j = − [ b j , k ] + ( j = k ) .
The k k k -th equation in (19) gives
g k ′ = − g k = − 1. g'_k=-g_k=-1. g k ′ = − g k = − 1.
For j ≠ k j\ne k j = k , the case g k ′ ≤ 0 g'_k\le0 g k ′ ≤ 0 of the same formula gives
g j ′ = g j + [ − b j , k ′ ] + g k ′ = g j ′ − [ b j , k ] + , g'_j=g_j+[-b'_{j,k}]_+g'_k=g'_j-[b_{j,k}]_+, g j ′ = g j + [ − b j , k ′ ] + g k ′ = g j ′ − [ b j , k ] + ,
where we used b j , k ′ = − b j , k b'_{j,k}=-b_{j,k} b j , k ′ = − b j , k . Substituting g j = − [ b j , k ] + g_j=-[b_{j,k}]_+ g j = − [ b j , k ] + , we obtain
g j ′ = − [ b j , k ] + + [ b j , k ] + = 0 ( j ≠ k ) . g'_j=-[b_{j,k}]_++[b_{j,k}]_+=0\qquad(j\ne k). g j ′ = − [ b j , k ] + + [ b j , k ] + = 0 ( j = k ) .
Thus g ′ = − e k \mathbf g'=-\mathbf e_k g ′ = − e k , so
deg Σ q ′ τ i , q ( X k ( Σ q ′ ) ) = − e k . \deg_{\Sigma'_q}\tau_{i,q}\left(X_k(\Sigma'_q)\right)=-\mathbf e_k. deg Σ q ′ τ i , q ( X k ( Σ q ′ ) ) = − e k .
For j ≠ k j\ne k j = k , we have X j ( Σ q ′ ) = X j ( Σ q ) X_j(\Sigma'_q)=X_j(\Sigma_q) X j ( Σ q ′ ) = X j ( Σ q ) . Hence the induction hypothesis gives
deg Σ q τ i , q ( X j ( Σ q ′ ) ) = deg Σ q τ i , q ( X j ( Σ q ) ) = − e j . \deg_{\Sigma_q}\tau_{i,q}\left(X_j(\Sigma'_q)\right)=\deg_{\Sigma_q}\tau_{i,q}\left(X_j(\Sigma_q)\right)=-\mathbf e_j. deg Σ q τ i , q ( X j ( Σ q ′ ) ) = deg Σ q τ i , q ( X j ( Σ q ) ) = − e j .
Since the k k k -th coordinate of − e j -\mathbf e_j − e j is zero, (19) gives
deg Σ q ′ τ i , q ( X j ( Σ q ′ ) ) = − e j ( j ≠ k ) . \deg_{\Sigma'_q}\tau_{i,q}\left(X_j(\Sigma'_q)\right)=-\mathbf e_j\qquad(j\ne k). deg Σ q ′ τ i , q ( X j ( Σ q ′ ) ) = − e j ( j = k ) .
Together with the calculation for j = k j=k j = k , this proves (84) for Σ q ′ \Sigma'_q Σ q ′ , completing the induction. □
Proposition 10.8. The automorphism τ i , q \tau_{i,q} τ i , q is independent of the choice of i ∈ Seq ( b ) i \in\operatorname{Seq}(b) i ∈ Seq ( b ) .
Proof. Let i , i ′ ∈ Seq ( b ) i,i' \in\operatorname{Seq}(b) i , i ′ ∈ Seq ( b ) and put Σ q = Σ q F i ′ \Sigma_q=\Sigma_q^{F_{i'}} Σ q = Σ q F i ′ . By Proposition 4.5, Σ q \Sigma_q Σ q is obtained from Σ q F i \Sigma_q^{F_i} Σ q F i by mutations and permutations. Relabelling vertices also relabels the coordinates of degree vectors, so Lemma 10.7 applies and gives
deg Σ q τ i , q ( X j ( Σ q ) ) = − ε j , deg Σ q τ i ′ , q ( X j ( Σ q ) ) = − ε j ( j ∈ J i ′ ) . \begin{aligned}
\deg_{\Sigma_q}\tau_{i,q}\left(X_j(\Sigma_q)\right)&=-\varepsilon_j,\\
\deg_{\Sigma_q}\tau_{i',q}\left(X_j(\Sigma_q)\right)&=-\varepsilon_j
\end{aligned}
\qquad(j\in J^{i'}). deg Σ q τ i , q ( X j ( Σ q ) ) deg Σ q τ i ′ , q ( X j ( Σ q ) ) = − ε j , = − ε j ( j ∈ J i ′ ) . Both images are localized quantum cluster monomials by Theorem 10.6. Multiplying both images by a common frozen monomial and normalizing by the same power of q 1 / 2 q^{1/2} q 1/2 gives quantum cluster monomials of the same degree. By [12 Theorem A.8] , these monomials are equal. Cancelling the common factors, we obtain
τ i , q ( X j ( Σ q ) ) = τ i ′ , q ( X j ( Σ q ) ) ( j ∈ J i ′ ) . \tau_{i,q}\left(X_j(\Sigma_q)\right)=\tau_{i',q}\left(X_j(\Sigma_q)\right)
\qquad(j\in J^{i'}). τ i , q ( X j ( Σ q ) ) = τ i ′ , q ( X j ( Σ q ) ) ( j ∈ J i ′ ) . The two automorphisms agree on all variables of Σ q \Sigma_q Σ q and on Q ( q 1 / 2 ) \mathbb{Q}(q^{1/2}) Q ( q 1/2 ) , hence on their common skew field of fractions. □
We write τ b , q : = τ i , q \tau_{b,q}:=\tau_{i,q} τ b , q := τ i , q and set τ 1 , q = id A ~ A \tau_{1,q}=\operatorname{id}_{\widetilde{\mathcal A}_{A}} τ 1 , q = id A A .
If b b b is the positive braid lift of w ∈ W w\in W w ∈ W , we identify A ~ q ( b ) \widetilde{\mathcal A}_q(b) A q ( b ) with A ~ A ( n ( w ) ) \widetilde{\mathcal A}_A(n(w)) A A ( n ( w )) by Lemma 10.1.
Corollary 10.9. Let w ∈ W w\in W w ∈ W , and let b b b be its positive braid lift. Then
τ b , q = η w , q . \tau_{b,q}=\eta_{w,q}. τ b , q = η w , q . Proof. Choose i ∈ Red ( w ) i\in\operatorname{Red}(w) i ∈ Red ( w ) . Proposition 10.2 and (76) show that τ b , q \tau_{b,q} τ b , q and η w , q \eta_{w,q} η w , q agree on all initial quantum cluster variables. These generate the ambient skew field over Q ( q 1 / 2 ) \mathbb{Q}(q^{1/2}) Q ( q 1/2 ) , so the two automorphisms agree. □
By [49 Theorems 6.13(1) and 6.14] , for b ≠ 1 b\ne1 b = 1 the algebra A ~ q ( b ) \widetilde{\mathcal A}_q(b) A q ( b ) admits a common triangular A \mathcal A A -basis in the sense of [48 Definitions 6.2.1–6.2.2] . This basis is unique by [48 Lemma 6.2.4] . We denote it by L b \mathbf L_b L b and set L 1 = { 1 } \mathbf L_1=\{1\} L 1 = { 1 } .
For i ∈ Red ( w ) i\in\operatorname{Red}(w) i ∈ Red ( w ) , the isomorphism in Lemma 10.1 identifies L b \mathbf L_b L b with the localized dual canonical basis: the products of an element of B ( w ) \mathbf B(w) B ( w ) with inverses of the frozen minors, normalized by powers of q 1 / 2 q^{1/2} q 1/2 so that their leading coefficients in Σ q F i \Sigma_q^{F_i} Σ q F i are 1 1 1 ; see [48 Theorems 9.3.2 and 9.5.1] . The following corollary extends the basis-preservation property of the Kimura–Oya quantum twist η w , q \eta_{w,q} η w , q [40 Theorem 6.1] to arbitrary positive braids.
Corollary 10.10. The quantum twist automorphism τ b , q \tau_{b,q} τ b , q permutes L b \mathbf L_b L b :
τ b , q ( L b ) = L b . \tau_{b,q}(\mathbf L_b)=\mathbf L_b. τ b , q ( L b ) = L b . Proof. The assertion is immediate for b = 1 b=1 b = 1 . For b ≠ 1 b\ne1 b = 1 , it follows from Theorem 10.6, [49 Theorem 6.14] , and [41 Section 6.1 and Theorem 6.2(2)] . □
In finite Cartan type, the preceding corollary also gives preservation of the normalized global basis of Kashiwara–Kim–Oh–Park [33 ] after localization. We first identify this basis with the common triangular basis.
Assume that C C C is of finite type, and fix i = ( i 1 , … , i l ) ∈ Seq ( b ) i=(i_1,\ldots,i_l)\in\operatorname{Seq}(b) i = ( i 1 , … , i l ) ∈ Seq ( b ) . Recall from Section 8 that the cuspidal elements are
P k i = T i 1 ⋯ T i k − 1 ( q i k 1 / 2 f i k , 0 ) ( 1 ≤ k ≤ l ) , P_k^i=T_{i_1}\cdots T_{i_{k-1}}\left(q_{i_k}^{1/2}f_{i_k,0}\right)\qquad(1\le k\le l), P k i = T i 1 ⋯ T i k − 1 ( q i k 1/2 f i k , 0 ) ( 1 ≤ k ≤ l ) ,
where q i = q d i q_i=q^{d_i} q i = q d i , the f i , p f_{i,p} f i , p are the generators of the bosonic extension A q \mathcal{A}_q A q , and the T i T_i T i are its braid symmetries. Recall also the normalized global basis G ~ ( b ) \widetilde{\mathbf{G}}(\mathbf{b}) G ( b ) of A A ( b ) \mathcal{A}_{\mathbb{A}}(\mathbf{b}) A A ( b ) from Theorem 8.10. Corollary 8.16 gives the A \mathbb{A} A -algebra isomorphism
Υ i : A q ( b ) → ∼ A A ( b ) . \Upsilon_i:\mathcal{A}_q(\mathbf{b})\xrightarrow{\sim}\mathcal{A}_{\mathbb{A}}(\mathbf{b}). Υ i : A q ( b ) ∼ A A ( b ) .
Set
P ~ k i : = Υ i ( X [ k ] ) = q i − 1 / 2 P k i ( 1 ≤ k ≤ l ) . \widetilde{P}^{i}_k:=\Upsilon_i(X_{[k]})=q_i^{-1/2}P^i_k \qquad(1\leq k\leq l). P k i := Υ i ( X [ k ] ) = q i − 1/2 P k i ( 1 ≤ k ≤ l ) .
Proposition 10.11. The isomorphism Υ i \Upsilon_i Υ i identifies the two bases:
Υ i ( L b ∩ A q ( b ) ) = G ~ ( b ) . \Upsilon_i(\mathbf{L}_{\mathbf{b}}\cap\mathcal{A}_q(\mathbf{b}))=\widetilde{\mathbf{G}}(\mathbf{b}). Υ i ( L b ∩ A q ( b )) = G ( b ) .
Proof. The assertion is immediate for b = 1 \mathbf{b}=1 b = 1 . Suppose that b ≠ 1 \mathbf{b}\ne1 b = 1 . The variables X [ k ] X_{[k]} X [ k ] are bar-invariant, and their images P ~ k i \widetilde{P}^{i}_k P k i are bar-invariant by [30 (2.4) and Lemma 4.17] . Since these elements generate the two algebras by Corollary 8.16 and Theorem 8.10, the isomorphism Υ i \Upsilon_i Υ i commutes with the bar anti-involutions.
For u ∈ Z ≥ 0 l \mathbf{u}\in\mathbb{Z}_{\geq0}^{l} u ∈ Z ≥ 0 l , let M ( u ) M(\mathbf{u}) M ( u ) and P ~ i ( u ) \widetilde{P}^{i}(\mathbf{u}) P i ( u ) denote the standard monomial of [49 Section 8.3] and the normalized PBW element of [30 (2.4) and (4.7)] , respectively. The definitions, (22) , (15) , and [49 Lemma 8.4] give
Υ i ( M ( u ) ) = P ~ i ( u ) ‾ . \Upsilon_i(M(\mathbf{u}))=\overline{\widetilde{P}^{i}(\mathbf{u})}. Υ i ( M ( u )) = P i ( u ) .
The bar-invariance and PBW triangularity in [30 Definition 4.14 and Lemmas 4.16–4.17] , after applying the bar anti-involution, give the standard monomial characterization in [49 Definition 2.20 and Theorem 8.10(4)] . The uniqueness in that characterization applies because the bi-lexicographic order refines the reverse lexicographic order and normalization preserves the coefficients of terms of the same weight. Hence Υ i − 1 ( G ~ ( b ) ) = L b ∩ A q ( b ) \Upsilon_i^{-1}(\widetilde{\mathbf{G}}(\mathbf{b}))=\mathbf{L}_{\mathbf{b}}\cap\mathcal{A}_q(\mathbf{b}) Υ i − 1 ( G ( b )) = L b ∩ A q ( b ) . □
We extend the basis identification to the frozen localizations. For j ∈ J f r i j\in J_{\mathrm{fr}}^i j ∈ J fr i , put H j = Υ i ( X f j ) H_j=\Upsilon_i(X_{f_j}) H j = Υ i ( X f j ) , where f j = [ j min , j max ] f_j=[j_{\min},j_{\max}] f j = [ j m i n , j m a x ] is the frozen i \mathbf{i} i -box of color i j i_j i j . Let A ~ A ( b ) \widetilde{\mathcal{A}}_{\mathbb{A}}(\mathbf{b}) A A ( b ) be the localization of A A ( b ) \mathcal{A}_{\mathbb{A}}(\mathbf{b}) A A ( b ) at the elements H j H_j H j . Then Υ i \Upsilon_i Υ i extends to an isomorphism
Υ i : A ~ q ( b ) → ∼ A ~ A ( b ) . \Upsilon_i:\widetilde{\mathcal{A}}_q(\mathbf{b})\xrightarrow{\sim}\widetilde{\mathcal{A}}_{\mathbb{A}}(\mathbf{b}). Υ i : A q ( b ) ∼ A A ( b ) .
Let G ~ l o c ( b ) \widetilde{\mathbf{G}}^{\mathrm{loc}}(\mathbf{b}) G loc ( b ) be the set obtained from the products
( ∏ j ∈ J f r i H j − a j ) g , a j ∈ Z ≥ 0 , g ∈ G ~ ( b ) , \left(\prod_{j\in J_{\mathrm{fr}}^i}H_j^{-a_j}\right)g,\qquad a_j\in\mathbb{Z}_{\geq0},\quad g\in\widetilde{\mathbf{G}}(\mathbf{b}), j ∈ J fr i ∏ H j − a j g , a j ∈ Z ≥ 0 , g ∈ G ( b ) ,
by normalizing their leading coefficients in the image of Σ q F i \Sigma_q^{\mathcal{F}^i} Σ q F i to $1. The product is taken in increasing order of j j j . Normalized multiplication by frozen variables and their inverses preserves the common triangular basis [49 Definition 2.20] . Clearing the frozen denominators and applying the proposition gives
Υ i ( L b ) = G ~ l o c ( b ) . \Upsilon_i(\mathbf{L}_{\mathbf{b}})=\widetilde{\mathbf{G}}^{\mathrm{loc}}(\mathbf{b}). Υ i ( L b ) = G loc ( b ) .
In particular, G ~ l o c ( b ) \widetilde{\mathbf{G}}^{\mathrm{loc}}(\mathbf{b}) G loc ( b ) is an A \mathbb{A} A -basis of A ~ A ( b ) \widetilde{\mathcal{A}}_{\mathbb{A}}(\mathbf{b}) A A ( b ) . Write
τ i , q b o s : = Υ i ∘ τ b , q ∘ Υ i − 1 . \tau^{\mathrm{bos}}_{i,q}:=\Upsilon_i\circ\tau_{\mathbf{b},q}\circ\Upsilon_i^{-1}. τ i , q bos := Υ i ∘ τ b , q ∘ Υ i − 1 .
Corollary 10.12. If C C C is of finite type, then
τ ı ~ , q b o s ( G ~ l o c ( b ) ) = G ~ l o c ( b ) . \tau^{\mathrm{bos}}_{\tilde{\imath},q}(\widetilde{\mathbf{G}}^{\mathrm{loc}}(\mathbf{b}))=\widetilde{\mathbf{G}}^{\mathrm{loc}}(\mathbf{b}). τ ~ , q bos ( G loc ( b )) = G loc ( b ) . Proof. This follows from Corollary 10.10 and Proposition 10.11 . □
Compatibility with folding
We prove that the commutative specializations of the quantum twists commute with the folding homomorphisms.
Let S b S_{\mathbf{b}} S b be the multiplicative subset of A ( b ) \mathcal{A}(\mathbf{b}) A ( b ) generated by its frozen variables, and set
A ~ ( b ) : = S b − 1 A ( b ) . \widetilde{\mathcal{A}}(\mathbf{b}) := S_{\mathbf{b}}^{-1}\mathcal{A}(\mathbf{b}). A ( b ) := S b − 1 A ( b ) .
Evaluation at q 1 / 2 = 1 q^{1/2}=1 q 1/2 = 1 induces a surjective homomorphism
C ⊗ A A ~ q ( b ) ⟶ A ~ ( b ) , \mathbb{C}\otimes_{\mathbb{A}}\widetilde{\mathcal{A}}_q(\mathbf{b})\longrightarrow\widetilde{\mathcal{A}}(\mathbf{b}), C ⊗ A A q ( b ) ⟶ A ( b ) ,
since quantum cluster variables and inverses of frozen variables specialize to the corresponding generators of A ~ ( b ) \widetilde{\mathcal{A}}(\mathbf{b}) A ( b ) . The elements of L b \mathbf{L}_{\mathbf{b}} L b are pointed at distinct degrees with leading coefficient 1 1 1 in Z F i \mathbb{Z}\mathbf{F}^{\mathbf{i}} Z F i . Their specializations at q 1 / 2 = 1 q^{1/2}=1 q 1/2 = 1 are therefore linearly independent. Since L b \mathbf{L}_{\mathbf{b}} L b is an A \mathbb{A} A -basis of A ~ q ( b ) \widetilde{\mathcal{A}}_q(\mathbf{b}) A q ( b ) , this homomorphism is also injective and hence an isomorphism. We denote the specialization of τ b , q \tau_{\mathbf{b},q} τ b , q by
τ b : A ~ ( b ) → ∼ A ~ ( b ) . \tau_{\mathbf{b}}:\widetilde{\mathcal{A}}(\mathbf{b})\xrightarrow{\sim}\widetilde{\mathcal{A}}(\mathbf{b}). τ b : A ( b ) ∼ A ( b ) .
For the identity braid, S 1 = { 1 } S_1=\{1\} S 1 = { 1 } , A ~ ( 1 ) = C \widetilde{\mathcal{A}}(1)=\mathbb{C} A ( 1 ) = C , and τ 1 = id C \tau_1=\operatorname{id}_{\mathbb{C}} τ 1 = id C . By (76) , for every i ∈ Seq ( b ) \mathbf{i}\in\operatorname{Seq}(\mathbf{b}) i ∈ Seq ( b ) one has
τ b ( x c k ) = x f k − 1 x n k ( 1 ≤ k ≤ l ) . (85) \tau_{\mathbf{b}}(x_{\mathbf{c}_k})=x_{\mathbf{f}_k}^{-1}x_{\mathbf{n}_k}\qquad(1\leq k\leq l). \tag*{(85)} τ b ( x c k ) = x f k − 1 x n k ( 1 ≤ k ≤ l ) . (85)
where x ∅ = 1 x_{\varnothing}=1 x ∅ = 1 . In particular, τ b ( x f k ) = x f k − 1 \tau_{\mathbf{b}}(x_{\mathbf{f}_k})=x_{\mathbf{f}_k}^{-1} τ b ( x f k ) = x f k − 1 . By Corollary 10.10 , τ b \tau_{\mathbf{b}} τ b permutes the basis L b c l \mathbf{L}_{\mathbf{b}}^{\mathrm{cl}} L b cl obtained by specializing L b \mathbf{L}_{\mathbf{b}} L b at q 1 / 2 = 1 q^{1/2}=1 q 1/2 = 1 .
Fix a maximal commuting family F \mathcal{F} F of i \mathbf{i} i -boxes and an admissible chain C \mathcal{C} C realizing F \mathcal{F} F , and let ϕ b F \phi_{\mathbf{b}}^{\mathcal{F}} ϕ b F be the Laurent homomorphism defined in § 11 . Let i ^ ∈ Seq ( b ^ ) \hat{\mathbf{i}}\in\operatorname{Seq}(\hat{\mathbf{b}}) i ^ ∈ Seq ( b ^ ) be the blockwise lift of i \mathbf{i} i . Recall from (13) that κ ( k , a ) \kappa(k,a) κ ( k , a ) is the position of the a a a th lifted letter over i k i_k i k in i ^ \hat{\mathbf{i}} i ^ . Theorem 6.4 gives the surjective homomorphism
ϕ b F : A ( b ^ ) ⟶ A ( b ) . \phi_{\mathbf{b}}^{\mathcal{F}}:\mathcal{A}(\hat{\mathbf{b}})\longrightarrow\mathcal{A}(\mathbf{b}). ϕ b F : A ( b ^ ) ⟶ A ( b ) .
For 1 ≤ k ≤ l 1\leq k\leq l 1 ≤ k ≤ l and 1 ≤ a ≤ e i k 1\leq a\leq e_{i_k} 1 ≤ a ≤ e i k , set
c ^ k , a : = [ κ ( k min , a ) , κ ( k , a ) ] , f ^ k , a : = [ κ ( k min , a ) , κ ( k max , a ) ] , \hat{\mathbf{c}}_{k,a}:=[\kappa(k_{\min},a),\kappa(k,a)],\qquad
\hat{\mathbf{f}}_{k,a}:=[\kappa(k_{\min},a),\kappa(k_{\max},a)], c ^ k , a := [ κ ( k m i n , a ) , κ ( k , a )] , f ^ k , a := [ κ ( k m i n , a ) , κ ( k m a x , a )] ,
and set
n ^ k , a : = { [ κ ( k + , a ) , κ ( k max , a ) ] , k < k max , ∅ , k = k max . \hat{\mathbf{n}}_{k,a}:=
\begin{cases}
[\kappa(k_{+},a),\kappa(k_{\max},a)], & k<k_{\max},\\
\varnothing, & k=k_{\max}.
\end{cases} n ^ k , a := { [ κ ( k + , a ) , κ ( k m a x , a )] , ∅ , k < k m a x , k = k m a x .
Set x ∅ ^ = 1 x_{\hat{\varnothing}}=1 x ∅ ^ = 1 . Lemma 3.13 identifies the a a a th box in a fiber over [ u , v ] [u,v] [ u , v ] as [ κ ( u , a ) , κ ( v , a ) ] [\kappa(u,a),\kappa(v,a)] [ κ ( u , a ) , κ ( v , a )] . Applying Theorem 6.2 and Proposition 9.2, we obtain
ϕ b F ( x c ^ k , a ) = x c k , ϕ b F ( x f ^ k , a ) = x f k , ϕ b F ( x n ^ k , a ) = x n k . (86) \phi_{\mathbf{b}}^{\mathcal{F}}(x_{\hat{\mathbf{c}}_{k,a}})=x_{\mathbf{c}_k},\qquad
\phi_{\mathbf{b}}^{\mathcal{F}}(x_{\hat{\mathbf{f}}_{k,a}})=x_{\mathbf{f}_k},\qquad
\phi_{\mathbf{b}}^{\mathcal{F}}(x_{\hat{\mathbf{n}}_{k,a}})=x_{\mathbf{n}_k}. \tag*{(86)} ϕ b F ( x c ^ k , a ) = x c k , ϕ b F ( x f ^ k , a ) = x f k , ϕ b F ( x n ^ k , a ) = x n k . (86)
By (86) , ϕ b F ( S b ^ ) = S b \phi_{\mathbf{b}}^{\mathcal{F}}(S_{\hat{\mathbf{b}}})=S_{\mathbf{b}} ϕ b F ( S b ^ ) = S b . Thus ϕ b F \phi_{\mathbf{b}}^{\mathcal{F}} ϕ b F extends to a surjective homomorphism
ϕ b F , l o c : A ~ ( b ^ ) ⟶ A ~ ( b ) . (87) \phi_{\mathbf{b}}^{\mathcal{F},\mathrm{loc}}:\widetilde{\mathcal{A}}(\hat{\mathbf{b}})\longrightarrow\widetilde{\mathcal{A}}(\mathbf{b}). \tag*{(87)} ϕ b F , loc : A ( b ^ ) ⟶ A ( b ) . (87)
Theorem 10.13. The twist automorphisms satisfy
ϕ b F , l o c ∘ τ ~ b = τ b ∘ ϕ b F , l o c . (88) \phi^{\mathcal{F},\mathrm{loc}}_{b} \circ\widetilde{\tau}_{b} = \tau_{b} \circ\phi^{\mathcal{F},\mathrm{loc}}_{b}. \tag*{(88)} ϕ b F , loc ∘ τ b = τ b ∘ ϕ b F , loc . (88)
Proof. For b = 1 b=1 b = 1 , all maps are the identity of C \mathbb{C} C , so the assertion holds. Assume that b ≠ 1 b\ne1 b = 1 . The two homomorphisms agree on x ^ c ^ k , a \widehat{x}_{\widehat{c}_{k,a}} x c k , a for 1 ≤ k ≤ l 1\le k\le l 1 ≤ k ≤ l and 1 ≤ a ≤ e i k 1\le a\le e_{i_k} 1 ≤ a ≤ e i k . The occurrences of the lifted color at κ ( k , a ) \kappa(k,a) κ ( k , a ) are precisely the positions κ ( p , a ) \kappa(p,a) κ ( p , a ) with i p = i k i_p=i_k i p = i k . Hence the three boxes in (85) for this lifted position are c ^ k , a \widehat{c}_{k,a} c k , a , f ^ k , a \widehat{f}_{k,a} f k , a , and n ^ k , a \widehat{n}_{k,a} n k , a . By (85) and (86) ,
ϕ b F , l o c τ ~ b ( x ^ c ^ k , a ) = ϕ b F , l o c ( x ^ f ^ k , a x ^ n ^ k , a − 1 ) = x f ^ k , a x n ^ k , a − 1 = τ b ( x c k ) = τ b ϕ b F , l o c ( x ^ c ^ k , a ) . \phi^{\mathcal{F},\mathrm{loc}}_{b}\widetilde{\tau}_{b}\left(\widehat{x}_{\widehat{c}_{k,a}}\right)
=\phi^{\mathcal{F},\mathrm{loc}}_{b}\left(\widehat{x}_{\widehat{f}_{k,a}}\widehat{x}_{\widehat{n}_{k,a}}^{-1}\right)
=x_{\widehat{f}_{k,a}}x_{\widehat{n}_{k,a}}^{-1}
=\tau_{b}\left(x_{c_k}\right)
=\tau_{b}\phi^{\mathcal{F},\mathrm{loc}}_{b}\left(\widehat{x}_{\widehat{c}_{k,a}}\right). ϕ b F , loc τ b ( x c k , a ) = ϕ b F , loc ( x f k , a x n k , a − 1 ) = x f k , a x n k , a − 1 = τ b ( x c k ) = τ b ϕ b F , loc ( x c k , a ) .
The variables x ^ c ^ k , a \widehat{x}_{\widehat{c}_{k,a}} x c k , a , over all k k k and a a a , form the cluster of Σ ^ F ^ \widehat{\Sigma}^{\widehat{\mathcal{F}}} Σ F , and the two homomorphisms have the same nonzero values on them. By the Laurent phenomenon, every element of A ( b ^ ) \mathcal{A}(\widehat{b}) A ( b ) is a Laurent polynomial in these variables. Multiplying by a monomial in the initial variables clears the denominators. The two homomorphisms agree on the polynomial, and the common image of the monomial is nonzero. Cancelling this image shows that the two homomorphisms agree on A ( b ^ ) \mathcal{A}(\widehat{b}) A ( b ) . □
For w ∈ W w\in W w ∈ W , the specialization of η w , q \eta_{w,q} η w , q at q 1 / 2 = 1 q^{1/2}=1 q 1/2 = 1 is an automorphism of C ⊗ A A ~ A ( n ( w ) ) \mathbb{C}\otimes_A\widetilde{\mathcal{A}}_A(n(w)) C ⊗ A A A ( n ( w )) by [40 Corollary 6.2] . If b b b is the positive braid lift of w w w , we identify C ⊗ A A ~ A ( n ( w ) ) \mathbb{C}\otimes_A\widetilde{\mathcal{A}}_A(n(w)) C ⊗ A A A ( n ( w )) with A ( b ) \mathcal{A}(b) A ( b ) by specializing the isomorphism in Lemma 10.1 .
Corollary 10.14. Let w ∈ W w\in W w ∈ W , set w ~ = ι ( w ) \widetilde{w}=\iota(w) w = ι ( w ) , and let b b b be the positive braid lift of w w w . Then
ϕ b F , l o c ∘ η w ~ , q ∣ q = 1 = η w , q ∣ q = 1 ∘ ϕ b F , l o c . \phi^{\mathcal{F},\mathrm{loc}}_{b}\circ\left.\eta_{\widetilde{w},q}\right|_{q=1}
=\left.\eta_{w,q}\right|_{q=1}\circ\phi^{\mathcal{F},\mathrm{loc}}_{b}. ϕ b F , loc ∘ η w , q ∣ q = 1 = η w , q ∣ q = 1 ∘ ϕ b F , loc .
Proof. Since i ∈ Seq ( b ) i\in\operatorname{Seq}(b) i ∈ Seq ( b ) is reduced, its blockwise lift i ^ \widehat{i} i belongs to Red ( w ~ ) \operatorname{Red}(\widetilde{w}) Red ( w ) by Corollary 2.5 . Specializing Corollary 10.9 for w w w and w ~ \widetilde{w} w gives
τ b = η w , q ∣ q = 1 , τ ~ b = η w ~ , q ∣ q = 1 . \tau_b=\left.\eta_{w,q}\right|_{q=1},
\qquad
\widetilde{\tau}_b=\left.\eta_{\widetilde{w},q}\right|_{q=1}. τ b = η w , q ∣ q = 1 , τ b = η w , q ∣ q = 1 .
The assertion now follows from Theorem 10.13 . □
Appendix A. Additional calculations for affine braid moves
The braid-move calculations in § 8 treat the subdiagrams on the two color orbits arising in finite-type foldings. Here we give the affine cases with orbit sizes ( 3 , 3 ) (3,3) ( 3 , 3 ) and ( 4 , 2 ) (4,2) ( 4 , 2 ) . The exchange columns are computed as in § 8 . We describe the lifted paths and the corresponding mutation–permutation sequences, and prove (43) . Throughout, the homomorphism ϕ i ^ \phi^{\widehat{i}} ϕ i is the one defined on L ( Σ ^ i ^ ) \mathcal{L}(\widehat{\Sigma}^{\widehat{i}}) L ( Σ i ) .
The 3-move with orbit sizes (3,3)
A.1. The 3-move with orbit sizes ( 3 , 3 ) (3,3) ( 3 , 3 ) . Use the Sage numbering of E 6 ( 1 ) E^{(1)}_6 E 6 ( 1 ) , whose three arms are
0 − 2 − 4 , 1 − 3 − 4 , 6 − 5 − 4. 0-2-4,\qquad1-3-4,\qquad6-5-4. 0 − 2 − 4 , 1 − 3 − 4 , 6 − 5 − 4.
The automorphism
σ = ( 0 1 6 ) ( 2 3 5 ) , σ ( 4 ) = 4 , \sigma=(0\ 1\ 6)(2\ 3\ 5),\qquad\sigma(4)=4, σ = ( 0 1 6 ) ( 2 3 5 ) , σ ( 4 ) = 4 ,
is admissible, with ordered orbits
I i = ( a 1 , a 2 , a 3 ) = ( 0 , 1 , 6 ) , I j = ( b 1 , b 2 , b 3 ) = ( 2 , 3 , 5 ) . I_i=(a_1,a_2,a_3)=(0,1,6),\qquad I_j=(b_1,b_2,b_3)=(2,3,5). I i = ( a 1 , a 2 , a 3 ) = ( 0 , 1 , 6 ) , I j = ( b 1 , b 2 , b 3 ) = ( 2 , 3 , 5 ) .
The quotient matrix is
( 2 − 1 0 − 1 2 − 3 0 − 1 2 ) \begin{pmatrix}
2 & -1 & 0 \\
-1 & 2 & -3 \\
0 & -1 & 2
\end{pmatrix} 2 − 1 0 − 1 2 − 1 0 − 3 2
the Cartan matrix of D 4 ( 3 ) D_4^{(3)} D 4 ( 3 ) . In particular, c i , j = c j , i = − 1 c_{i,j}=c_{j,i}=-1 c i , j = c j , i = − 1 and e i = e j = 3 e_i=e_j=3 e i = e j = 3 . The only cross edges between I i I_i I i and I j I_j I j are the three mutually disjoint edges a r − b r a_r-b_r a r − b r (1 ≤ r ≤ 3 1 \le r \le3 1 ≤ r ≤ 3 ).
The blockwise lift of i j i → j i j iji \to jij ij i → j ij is
( a 1 , a 2 , a 3 , b 1 , b 2 , b 3 , a 1 , a 2 , a 3 ) ⟶ ( b 1 , b 2 , b 3 , a 1 , a 2 , a 3 , b 1 , b 2 , b 3 ) . (a_1,a_2,a_3,b_1,b_2,b_3,a_1,a_2,a_3) \longrightarrow(b_1,b_2,b_3,a_1,a_2,a_3,b_1,b_2,b_3). ( a 1 , a 2 , a 3 , b 1 , b 2 , b 3 , a 1 , a 2 , a 3 ) ⟶ ( b 1 , b 2 , b 3 , a 1 , a 2 , a 3 , b 1 , b 2 , b 3 ) .
The following sequence of 2- and 3-moves is applied from left to right at the current positions:
( 2 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 1 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 2 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 3 ; κ ( k , 1 ) ) , ( 3 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 1 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 3 ; κ ( k , 1 ) + 6 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 2 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 2 ; κ ( k , 1 ) + 5 ) . (89) \begin{aligned}
&(2;\kappa(k,1)+2),(2;\kappa(k,1)+1),(2;\kappa(k,1)+3),(2;\kappa(k,1)+5),(2;\kappa(k,1)+4),\\
&(2;\kappa(k,1)+3),(2;\kappa(k,1)+2),(2;\kappa(k,1)+6),(2;\kappa(k,1)+5),\\
&(3;\kappa(k,1)),(3;\kappa(k,1)+3),\\
&(2;\kappa(k,1)+2),(2;\kappa(k,1)+1),(2;\kappa(k,1)+3),(3;\kappa(k,1)+6),\\
&(2;\kappa(k,1)+5),(2;\kappa(k,1)+4),(2;\kappa(k,1)+3),(2;\kappa(k,1)+2),\\
&(2;\kappa(k,1)+6),(2;\kappa(k,1)+5). \tag*{(89)}
\end{aligned} ( 2 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 1 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 2 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 3 ; κ ( k , 1 )) , ( 3 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 1 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 3 ; κ ( k , 1 ) + 6 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 2 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 2 ; κ ( k , 1 ) + 5 ) . (89)
After the first nine commuting moves, the nine-letter subword has the form
a 1 b 1 a 1 ‾ a 2 b 2 a 2 ‾ a 3 b 3 a 3 ‾ . \underline{a_1b_1a_1}\,\underline{a_2b_2a_2}\,\underline{a_3b_3a_3}. a 1 b 1 a 1 a 2 b 2 a 2 a 3 b 3 a 3 .
The three underlined factors are changed to b r a r b r b_ra_rb_r b r a r b r ; the remaining moves give the word b 1 b 2 b 3 a 1 a 2 a 3 b 1 b 2 b 3 b_1b_2b_3a_1a_2a_3b_1b_2b_3 b 1 b 2 b 3 a 1 a 2 a 3 b 1 b 2 b 3 . Thus (A.1) consists of eighteen 2-moves and exactly three simply-laced 3-moves. Its mutation–permutation realization is obtained by replacing ( 2 ; p ) (2;p) ( 2 ; p ) with P p P_p P p and ( 3 ; p ) (3;p) ( 3 ; p ) with ( μ p , P p + 1 ) (\mu_p,P_{p+1}) ( μ p , P p + 1 ) , with every index read in the current seed.
Let D ± D^\pm D ± denote the contributions of the rows indexed by p ∉ { k , k + 1 , k + 2 } p \notin\{k,k+1,k+2\} p ∈ / { k , k + 1 , k + 2 } to the positive and negative exchange monomials at the unique downstairs mutation. For r = 1 , 2 , 3 r=1,2,3 r = 1 , 2 , 3 , let U r ± U_r^\pm U r ± denote the contributions of the rows with indices outside [ κ ( k , 1 ) , κ ( k + 2 , 3 ) ] [\kappa(k,1),\kappa(k+2,3)] [ κ ( k , 1 ) , κ ( k + 2 , 3 )] to the positive and negative exchange monomials, computed immediately before the r r r th upstairs mutation in (A.1).
For r = 1 , 2 , 3 r=1,2,3 r = 1 , 2 , 3 , the r r r th upstairs mutation in (A.1) occurs at k ^ r : = κ ( k , 1 ) + 3 ( r − 1 ) \widehat{k}_r:=\kappa(k,1)+3(r-1) k r := κ ( k , 1 ) + 3 ( r − 1 ) . Let B ^ ( r ) = ( b ^ u , v ( r ) ) \widehat{B}^{(r)}=(\widehat{b}^{(r)}_{u,v}) B ( r ) = ( b u , v ( r ) ) be the exchange matrix immediately before the mutation at k ^ r \widehat{k}_r k r .
Lemma A.1. For r = 1 , 2 , 3 r=1,2,3 r = 1 , 2 , 3 ,
ϕ i ( U r ± ) = D ± . \phi^i(U_r^\pm)=D^\pm. ϕ i ( U r ± ) = D ± . More precisely, if p ∉ { k , k + 1 , k + 2 } p\notin\{k,k+1,k+2\} p ∈ / { k , k + 1 , k + 2 } and δ p = b p , k i \delta_p=b^i_{p,k} δ p = b p , k i , then
∑ s [ ± b ^ κ ( p , s ) , k ^ r ( r ) ] + = [ ± δ p ] + . \sum_s [\pm\widehat{b}^{(r)}_{\kappa(p,s),\widehat{k}_r}]_+=[\pm\delta_p]_+. s ∑ [ ± b κ ( p , s ) , k r ( r ) ] + = [ ± δ p ] + .
Proof. The verification is parallel to the size-(2,2) calculation in Lemma 4.9 . Here the subdiagram on I i ⊔ I j I_i\sqcup I_j I i ⊔ I j has the three mutually disjoint edges a r − b r a_r-b_r a r − b r . Fix an outside position p p p with δ p ≠ 0 \delta_p\ne0 δ p = 0 . In the r r r th mutation column, only the row κ ( p , s ) \kappa(p,s) κ ( p , s ) whose color lies in the component containing a r a_r a r has a nonzero entry. This entry has the same sign as δ p \delta_p δ p . The exchange matrix entries between the three mutation vertices vanish. Hence mutating at one of these vertices does not change the columns indexed by the other two. The 2-moves only relabel the vertices. □ \square □
Proposition A.2. Suppose that i ′ \mathbf{i}' i ′ is obtained from i \mathbf{i} i by a 3-move
i j i ⟶ j i j , e i = e j = 3 , iji \longrightarrow jij,\qquad e_i=e_j=3, ij i ⟶ j ij , e i = e j = 3 , in the affine folding E 6 ( 1 ) ⟶ D 4 ( 3 ) E_6^{(1)} \longrightarrow D_4^{(3)} E 6 ( 1 ) ⟶ D 4 ( 3 ) above. Then (43) holds.
Proof. Write x 1 , x 2 , x 3 x_1,x_2,x_3 x 1 , x 2 , x 3 for the initial cluster variables at positions k k k , k + 1 k+1 k + 1 , k + 2 k+2 k + 2 , respectively. The final downstairs cluster variables are
x 1 ′ = D + x 2 + D − x 3 x 1 , x 2 ′ = x 3 , x 3 ′ = x 2 . x'_1=\frac{D^+x_2+D^-x_3}{x_1},\qquad x'_2=x_3,\qquad x'_3=x_2. x 1 ′ = x 1 D + x 2 + D − x 3 , x 2 ′ = x 3 , x 3 ′ = x 2 . For the cluster variables at positions κ ( k , 1 ) , … , κ ( k + 2 , 3 ) \kappa(k,1),\ldots,\kappa(k+2,3) κ ( k , 1 ) , … , κ ( k + 2 , 3 ) in Σ ^ i \widehat{\Sigma}^{\mathbf{i}} Σ i , write
x ^ 1 , x ^ 2 , x ^ 3 ; x ^ 4 , x ^ 5 , x ^ 6 ; x ^ 7 , x ^ 8 , x ^ 9 \widehat{x}_1,\widehat{x}_2,\widehat{x}_3;\qquad\widehat{x}_4,\widehat{x}_5,\widehat{x}_6;\qquad\widehat{x}_7,\widehat{x}_8,\widehat{x}_9 x 1 , x 2 , x 3 ; x 4 , x 5 , x 6 ; x 7 , x 8 , x 9 with three variables for each successive position of i j i iji ij i . The three mutations create variables z r z_r z r satisfying
x ^ r z r = U r + x ^ 3 + r + U r − x ^ 6 + r ( 1 ≤ r ≤ 3 ) . \widehat{x}_r z_r=U_r^+\widehat{x}_{3+r}+U_r^-\widehat{x}_{6+r}\qquad(1\le r\le3). x r z r = U r + x 3 + r + U r − x 6 + r ( 1 ≤ r ≤ 3 ) . Lemma A.1 gives
ϕ i ( z 1 ) = ϕ i ( z 2 ) = ϕ i ( z 3 ) = x 1 ′ . \phi^i(z_1)=\phi^i(z_2)=\phi^i(z_3)=x'_1. ϕ i ( z 1 ) = ϕ i ( z 2 ) = ϕ i ( z 3 ) = x 1 ′ . After the final permutations, the nine upstairs cluster variables are
( z 1 , z 2 , z 3 , x ^ 7 , x ^ 8 , x ^ 9 , x ^ 4 , x ^ 5 , x ^ 6 ) . (z_1,z_2,z_3,\widehat{x}_7,\widehat{x}_8,\widehat{x}_9,\widehat{x}_4,\widehat{x}_5,\widehat{x}_6). ( z 1 , z 2 , z 3 , x 7 , x 8 , x 9 , x 4 , x 5 , x 6 ) . Consequently, ϕ i \phi^i ϕ i sends the variables corresponding to the successive positions of j i j jij j ij respectively to x 1 ′ , x 2 ′ , x 3 ′ x'_1,x'_2,x'_3 x 1 ′ , x 2 ′ , x 3 ′ , proving (43) . □
The 4-move with orbit sizes (4, 2)
For D 2 n + 2 ( 1 ) D_{2n+2}^{(1)} D 2 n + 2 ( 1 ) with n ≥ 2 n\ge2 n ≥ 2 , put N = 2 n + 2 N=2n+2 N = 2 n + 2 and consider the order-four automorphism whose orbits are
I 0 = ( 0 , 1 , N − 1 , N ) , I q = ( q + 1 , N − 1 − q ) ( 1 ≤ q < n ) , I n = ( n + 1 ) . I_0=(0,1,N-1,N),\qquad I_q=(q+1,N-1-q)\ (1\le q<n),\qquad I_n=(n+1). I 0 = ( 0 , 1 , N − 1 , N ) , I q = ( q + 1 , N − 1 − q ) ( 1 ≤ q < n ) , I n = ( n + 1 ) .
The action is admissible. Direct orbit summation gives the Cartan matrix of A 2 n ( 2 ) A_{2n}^{(2)} A 2 n ( 2 ) ; in particular,
( c 0 , 1 , c 1 , 0 ) = ( − 2 , − 1 ) , ( e 0 , e 1 ) = ( 4 , 2 ) . (c_{0,1},c_{1,0})=(-2,-1),\qquad(e_0,e_1)=(4,2). ( c 0 , 1 , c 1 , 0 ) = ( − 2 , − 1 ) , ( e 0 , e 1 ) = ( 4 , 2 ) .
We take i = 0 i=0 i = 0 , j = 1 j=1 j = 1 and write
I i = ( a 1 , a 2 , a 3 , a 4 ) , I j = ( b 1 , b 2 ) . I_i=(a_1,a_2,a_3,a_4),\qquad I_j=(b_1,b_2). I i = ( a 1 , a 2 , a 3 , a 4 ) , I j = ( b 1 , b 2 ) .
The subdiagram on I i ⊔ I j I_i\sqcup I_j I i ⊔ I j has two connected components, each consisting of one vertex of I j I_j I j adjacent to two vertices of I i I_i I i . The size-(2, 4) case is obtained by reversing the path below. The blockwise lifts of i j i j ijij ij ij and j i j i jiji j ij i are, respectively
( a 1 , a 2 , a 3 , a 4 , b 1 , b 2 , a 1 , a 2 , a 3 , a 4 , b 1 , b 2 ) , (a_1,a_2,a_3,a_4,b_1,b_2,a_1,a_2,a_3,a_4,b_1,b_2), ( a 1 , a 2 , a 3 , a 4 , b 1 , b 2 , a 1 , a 2 , a 3 , a 4 , b 1 , b 2 ) ,
( b 1 , b 2 , a 1 , a 2 , a 3 , a 4 , b 1 , b 2 , a 1 , a 2 , a 3 , a 4 ) . (b_1,b_2,a_1,a_2,a_3,a_4,b_1,b_2,a_1,a_2,a_3,a_4). ( b 1 , b 2 , a 1 , a 2 , a 3 , a 4 , b 1 , b 2 , a 1 , a 2 , a 3 , a 4 ) .
The following sequence of 2- and 3-moves connects the two displayed words:
( 2 ; κ ( k , 1 ) ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 5 ) ( 2 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 2 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 9 ) , ( 2 ; κ ( k , 1 ) + 8 ) , ( 2 ; κ ( k , 1 ) + 7 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 3 ; κ ( k , 1 ) + 7 ) , ( 3 ; κ ( k , 1 ) + 9 ) , ( 2 ; κ ( k , 1 ) + 8 ) , ( 3 ; κ ( k , 1 ) + 6 ) , ( 3 ; κ ( k , 1 ) + 8 ) , ( 3 ; κ ( k , 1 ) + 1 ) , ( 3 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 3 ; κ ( k , 1 ) ) , ( 3 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 2 ; κ ( k , 1 ) + 7 ) , ( 2 ; κ ( k , 1 ) + 8 ) , ( 2 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 2 ; κ ( k , 1 ) + 7 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 2 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 1 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 1 ) . (90) \begin{aligned}
&(2;\kappa(k,1)),\ (2;\kappa(k,1)+2),\ (2;\kappa(k,1)+3),\ (2;\kappa(k,1)+2),\ (2;\kappa(k,1)+5)\\
&(2;\kappa(k,1)+4),\ (2;\kappa(k,1)+3),\ (2;\kappa(k,1)+6),\ (2;\kappa(k,1)+5),\ (2;\kappa(k,1)+4),\\
&(2;\kappa(k,1)+9),\ (2;\kappa(k,1)+8),\ (2;\kappa(k,1)+7),\ (2;\kappa(k,1)+6),\ (2;\kappa(k,1)+5),\\
&(3;\kappa(k,1)+7),\ (3;\kappa(k,1)+9),\ (2;\kappa(k,1)+8),\ (3;\kappa(k,1)+6),\ (3;\kappa(k,1)+8),\\
&(3;\kappa(k,1)+1),\ (3;\kappa(k,1)+3),\ (2;\kappa(k,1)+2),\ (3;\kappa(k,1)),\ (3;\kappa(k,1)+2),\\
&(2;\kappa(k,1)+5),\ (2;\kappa(k,1)+6),\ (2;\kappa(k,1)+7),\ (2;\kappa(k,1)+8),\ (2;\kappa(k,1)+4),\\
&(2;\kappa(k,1)+5),\ (2;\kappa(k,1)+6),\ (2;\kappa(k,1)+7),\ (2;\kappa(k,1)+3),\ (2;\kappa(k,1)+4),\\
&(2;\kappa(k,1)+5),\ (2;\kappa(k,1)+4),\ (2;\kappa(k,1)+1),\ (2;\kappa(k,1)+2),\ (2;\kappa(k,1)+1). \tag*{(90)}
\end{aligned} ( 2 ; κ ( k , 1 )) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 5 ) ( 2 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 2 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 9 ) , ( 2 ; κ ( k , 1 ) + 8 ) , ( 2 ; κ ( k , 1 ) + 7 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 3 ; κ ( k , 1 ) + 7 ) , ( 3 ; κ ( k , 1 ) + 9 ) , ( 2 ; κ ( k , 1 ) + 8 ) , ( 3 ; κ ( k , 1 ) + 6 ) , ( 3 ; κ ( k , 1 ) + 8 ) , ( 3 ; κ ( k , 1 ) + 1 ) , ( 3 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 3 ; κ ( k , 1 )) , ( 3 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 2 ; κ ( k , 1 ) + 7 ) , ( 2 ; κ ( k , 1 ) + 8 ) , ( 2 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 2 ; κ ( k , 1 ) + 6 ) , ( 2 ; κ ( k , 1 ) + 7 ) , ( 2 ; κ ( k , 1 ) + 3 ) , ( 2 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 5 ) , ( 2 ; κ ( k , 1 ) + 4 ) , ( 2 ; κ ( k , 1 ) + 1 ) , ( 2 ; κ ( k , 1 ) + 2 ) , ( 2 ; κ ( k , 1 ) + 1 ) . (90)
It contains thirty-two 2-moves and eight simply-laced 3-moves. Replacing the former by adjacent permutations and each ( 3 ; p ) (3;p) ( 3 ; p ) by ( μ p , P p + 1 ) (\mu_p,P_{p+1}) ( μ p , P p + 1 ) gives eight mutations and forty permutations, again with the labels at the current step.
For a = 1 , 2 , 3 a=1,2,3 a = 1 , 2 , 3 , let D a ± D_a^{\pm} D a ± denote the contributions of the rows indexed by p ∉ { k , k + 1 , k + 2 , k + 3 } p\notin\{k,k+1,k+2,k+3\} p ∈ / { k , k + 1 , k + 2 , k + 3 } to the positive and negative exchange monomials at the a a a th downstairs mutation in the 4-move realization. For q = 1 , … , 8 q=1,\ldots,8 q = 1 , … , 8 , let U q ± U_q^{\pm} U q ± denote the contributions of the rows with indices outside [ κ ( k , 1 ) , κ ( k + 3 , 2 ) ] [\kappa(k,1),\kappa(k+3,2)] [ κ ( k , 1 ) , κ ( k + 3 , 2 )] to the positive and negative exchange monomials, computed immediately before the q q q th upstairs mutation in (A.2).
Lemma A.3. For the 4-move i j i j → j i j i ijij\to jiji ij ij → j ij i at position k k k in i i i , the coefficient factors satisfy
( D 2 + , D 2 − ) = ( 1 , D 1 − ) , ( D 3 + , D 3 − ) = ( 1 , D 1 + ) , (D_2^+,D_2^-)=(1,D_1^-),\qquad(D_3^+,D_3^-)=(1,D_1^+), ( D 2 + , D 2 − ) = ( 1 , D 1 − ) , ( D 3 + , D 3 − ) = ( 1 , D 1 + ) , and
( ϕ i ( U q + ) , ϕ i ( U q − ) ) = { ( D 1 + , D 1 − ) , q = 1 , 3 , 5 , 7 , ( 1 , 1 ) , q = 2 , 6 , ( 1 , D 1 + ) , q = 4 , 8. \left(\phi^i(U_q^+),\phi^i(U_q^-)\right)=
\begin{cases}
(D_1^+,D_1^-),&q=1,3,5,7,\\
(1,1),&q=2,6,\\
(1,D_1^+),&q=4,8.
\end{cases} ( ϕ i ( U q + ) , ϕ i ( U q − ) ) = ⎩ ⎨ ⎧ ( D 1 + , D 1 − ) , ( 1 , 1 ) , ( 1 , D 1 + ) , q = 1 , 3 , 5 , 7 , q = 2 , 6 , q = 4 , 8. Consequently, ϕ i \phi^i ϕ i sends the variables corresponding to each position of j i j i jiji j ij i to the corresponding downstairs cluster variable, as in (43) .
Proof. The verification uses the same calculation of the factors D a ± D_a^{\pm} D a ± and U q ± U_q^{\pm} U q ± from exchange matrix entries as for the ( e i , e j ) = ( 2 , 1 ) (e_i,e_j)=(2,1) ( e i , e j ) = ( 2 , 1 ) 4-move in Lemma 4.17, applied independently to the two connected components described above. Moves between the components are 2-moves and hence only permute the current labels. Substituting these coefficient identities into the calculation in Proposition 4.18 gives the cluster-variable identities for each component. □
Proposition A.4. Suppose that i ′ i' i ′ is obtained from i i i by the affine 4-move
i j i j → j i j i , ( e i , e j ) = ( 4 , 2 ) , ijij\to jiji,\qquad(e_i,e_j)=(4,2), ij ij → j ij i , ( e i , e j ) = ( 4 , 2 ) , above, or by the reverse move. Then (43) holds.
Proof. We first compute the final cluster variables with all factors D a ± D_a^{\pm} D a ± and U q ± U_q^{\pm} U q ± set to 1. Write x 1 , x 2 , x 3 , x 4 x_1,x_2,x_3,x_4 x 1 , x 2 , x 3 , x 4 for the four initial downstairs variables and put
A = x 1 2 x 4 + ( x 2 + x 3 ) 2 , C = x 1 2 x 4 + x 2 x 3 + x 3 2 . A=x_1^2x_4+(x_2+x_3)^2,\qquad C=x_1^2x_4+x_2x_3+x_3^2. A = x 1 2 x 4 + ( x 2 + x 3 ) 2 , C = x 1 2 x 4 + x 2 x 3 + x 3 2 .
The three downstairs mutations and final permutations give
( x 1 ′ , x 2 ′ , x 3 ′ , x 4 ′ ) = ( A x 1 2 x 2 , C x 1 x 2 , x 4 , x 3 ) . (x'_1,x'_2,x'_3,x'_4)=\left(\frac{A}{x_1^2x_2},\frac{C}{x_1x_2},x_4,x_3\right). ( x 1 ′ , x 2 ′ , x 3 ′ , x 4 ′ ) = ( x 1 2 x 2 A , x 1 x 2 C , x 4 , x 3 ) .
Let x ^ 1 , … , x ^ 12 \widehat{x}_1,\ldots,\widehat{x}_{12} x 1 , … , x 12 be the cluster variables at positions κ ( k , 1 ) , … , κ ( k + 3 , 2 ) \kappa(k,1),\ldots,\kappa(k+3,2) κ ( k , 1 ) , … , κ ( k + 3 , 2 ) in Σ ^ i \widehat{\Sigma}^{i} Σ i , in this order. Thus ϕ i \phi^{i} ϕ i sends
x ^ 1 , … , x ^ 4 ↦ x 1 , x ^ 5 , x ^ 6 ↦ x 2 , x ^ 7 , … , x ^ 10 ↦ x 3 , x ^ 11 , x ^ 12 ↦ x 4 . \widehat{x}_1,\ldots,\widehat{x}_4\mapsto x_1,\quad\widehat{x}_5,\widehat{x}_6\mapsto x_2,\quad\widehat{x}_7,\ldots,\widehat{x}_{10}\mapsto x_3,\quad\widehat{x}_{11},\widehat{x}_{12}\mapsto x_4. x 1 , … , x 4 ↦ x 1 , x 5 , x 6 ↦ x 2 , x 7 , … , x 10 ↦ x 3 , x 11 , x 12 ↦ x 4 .
The eight upstairs mutations in (A.2) produce
z 1 = x ^ 1 x ^ 2 x ^ 11 + x ^ 5 2 + x ^ 5 x ^ 7 + x ^ 5 x ^ 8 + x ^ 7 x ^ 8 x ^ 1 x ^ 2 x ^ 5 , z_1=\frac{\widehat{x}_1\widehat{x}_2\widehat{x}_{11}+\widehat{x}_5^2+\widehat{x}_5\widehat{x}_7+\widehat{x}_5\widehat{x}_8+\widehat{x}_7\widehat{x}_8}{\widehat{x}_1\widehat{x}_2\widehat{x}_5}, z 1 = x 1 x 2 x 5 x 1 x 2 x 11 + x 5 2 + x 5 x 7 + x 5 x 8 + x 7 x 8 ,
z 2 = x ^ 3 x ^ 4 x ^ 12 + x ^ 6 2 + x ^ 6 x ^ 9 + x ^ 6 x ^ 10 + x ^ 9 x ^ 10 x ^ 3 x ^ 4 x ^ 6 , z_2=\frac{\widehat{x}_3\widehat{x}_4\widehat{x}_{12}+\widehat{x}_6^2+\widehat{x}_6\widehat{x}_9+\widehat{x}_6\widehat{x}_{10}+\widehat{x}_9\widehat{x}_{10}}{\widehat{x}_3\widehat{x}_4\widehat{x}_6}, z 2 = x 3 x 4 x 6 x 3 x 4 x 12 + x 6 2 + x 6 x 9 + x 6 x 10 + x 9 x 10 ,
w 1 = x ^ 1 x ^ 2 x ^ 11 + x ^ 5 x ^ 7 + x ^ 7 x ^ 8 x ^ 2 x ^ 5 , w 2 = x ^ 1 x ^ 2 x ^ 11 + x ^ 5 x ^ 8 + x ^ 7 x ^ 8 x ^ 1 x ^ 5 , w 3 = x ^ 3 x ^ 4 x ^ 12 + x ^ 6 x ^ 9 + x ^ 9 x ^ 10 x ^ 4 x ^ 6 , w 4 = x ^ 3 x ^ 4 x ^ 12 + x ^ 6 x ^ 10 + x ^ 9 x ^ 10 x ^ 3 x ^ 6 . \begin{aligned}
w_1&=\frac{\widehat{x}_1\widehat{x}_2\widehat{x}_{11}+\widehat{x}_5\widehat{x}_7+\widehat{x}_7\widehat{x}_8}{\widehat{x}_2\widehat{x}_5},&
w_2&=\frac{\widehat{x}_1\widehat{x}_2\widehat{x}_{11}+\widehat{x}_5\widehat{x}_8+\widehat{x}_7\widehat{x}_8}{\widehat{x}_1\widehat{x}_5},\\
w_3&=\frac{\widehat{x}_3\widehat{x}_4\widehat{x}_{12}+\widehat{x}_6\widehat{x}_9+\widehat{x}_9\widehat{x}_{10}}{\widehat{x}_4\widehat{x}_6},&
w_4&=\frac{\widehat{x}_3\widehat{x}_4\widehat{x}_{12}+\widehat{x}_6\widehat{x}_{10}+\widehat{x}_9\widehat{x}_{10}}{\widehat{x}_3\widehat{x}_6}.
\end{aligned} w 1 w 3 = x 2 x 5 x 1 x 2 x 11 + x 5 x 7 + x 7 x 8 , = x 4 x 6 x 3 x 4 x 12 + x 6 x 9 + x 9 x 10 , w 2 w 4 = x 1 x 5 x 1 x 2 x 11 + x 5 x 8 + x 7 x 8 , = x 3 x 6 x 3 x 4 x 12 + x 6 x 10 + x 9 x 10 .
After the final permutations, the twelve upstairs cluster variables are
( z 1 , z 2 , w 1 , w 2 , w 3 , w 4 , x ^ 11 , x ^ 12 , x ^ 7 , x ^ 8 , x ^ 9 , x ^ 10 ) . (z_1,z_2,w_1,w_2,w_3,w_4,\widehat{x}_{11},\widehat{x}_{12},\widehat{x}_7,\widehat{x}_8,\widehat{x}_9,\widehat{x}_{10}). ( z 1 , z 2 , w 1 , w 2 , w 3 , w 4 , x 11 , x 12 , x 7 , x 8 , x 9 , x 10 ) .
Direct substitution gives
ϕ i ( z 1 ) = ϕ i ( z 2 ) = A x 1 2 x 2 = x 1 ′ , \phi^{i}(z_1)=\phi^{i}(z_2)=\frac{A}{x_1^2x_2}=x'_1, ϕ i ( z 1 ) = ϕ i ( z 2 ) = x 1 2 x 2 A = x 1 ′ ,
ϕ i ( w 1 ) = ⋯ = ϕ i ( w 4 ) = C x 1 x 2 = x 2 ′ , \phi^{i}(w_1)=\cdots=\phi^{i}(w_4)=\frac{C}{x_1x_2}=x'_2, ϕ i ( w 1 ) = ⋯ = ϕ i ( w 4 ) = x 1 x 2 C = x 2 ′ ,
ϕ i ( x ^ 11 ) = ϕ i ( x ^ 12 ) = x 4 = x 3 ′ , \phi^{i}(\widehat{x}_{11})=\phi^{i}(\widehat{x}_{12})=x_4=x'_3, ϕ i ( x 11 ) = ϕ i ( x 12 ) = x 4 = x 3 ′ ,
ϕ i ( x ^ 7 ) = ⋯ = ϕ i ( x ^ 10 ) = x 3 = x 4 ′ . \phi^{i}(\widehat{x}_7)=\cdots=\phi^{i}(\widehat{x}_{10})=x_3=x'_4. ϕ i ( x 7 ) = ⋯ = ϕ i ( x 10 ) = x 3 = x 4 ′ .
The four groups correspond to the successive positions of j i j i jiji j ij i and have cardinalities 2 , 4 , 2 , 4 2,4,2,4 2 , 4 , 2 , 4 . For a general ambient sequence, the same calculation is performed with the factors D a ± D_a^{\pm} D a ± and U q ± U_q^{\pm} U q ± retained. By Lemma A.3 , after applying ϕ i \phi^{i} ϕ i the exchange relations agree componentwise with those of the ( e i , e j ) = ( 2 , 1 ) (e_i,e_j)=(2,1) ( e i , e j ) = ( 2 , 1 ) calculation in Proposition 4.18 . Hence the same elimination proves (43) . For the reverse move, we use the sequence (A.2) in reverse order. □
Together with the braid-move calculations in § 8 , these two propositions complete (43) for every finite braid move in the admissible finite and affine foldings considered here.
For a general admissible folding, the subdiagram on the two color orbits involved in the braid move may have more connected components. In the next subsection, we show that the verification of (43) reduces to the case in which the subdiagram on I i ⊔ I j I_i\sqcup I_j I i ⊔ I j is connected. Thus no new calculation is required when the subdiagram on I i ⊔ I j I_i\sqcup I_j I i ⊔ I j has several connected components.
Reduction to connected subdiagrams
Let i , j ∈ I i,j \in I i , j ∈ I be the vertices involved in a downstairs braid move. The subdiagram of Δ ^ \widehat{\Delta} Δ on I i ⊔ I j I_i \sqcup I_j I i ⊔ I j decomposes into isomorphic connected components
Δ ^ 1 ⊔ ⋯ ⊔ Δ ^ r . \widehat{\Delta}_1 \sqcup\cdots\sqcup\widehat{\Delta}_r. Δ 1 ⊔ ⋯ ⊔ Δ r .
For 1 ≤ s ≤ r 1 \le s \le r 1 ≤ s ≤ r , let J ^ s \widehat{J}_s J s be the positions in the lifted braid block whose colors belong to Δ ^ s \widehat{\Delta}_s Δ s . When the vertex labels are permuted, the sets J ^ s \widehat{J}_s J s are relabelled in the same way; mutation leaves these sets unchanged.
Lemma A.5. At every stage of the mutation–permutation sequence realizing the 2- and 3-moves in Proposition 4.4 ,
b a b = 0 ( a ∈ J ^ s , b ∈ J ^ t , s ≠ t ) b_{ab}=0 \qquad(a\in\widehat{J}_s,\ b\in\widehat{J}_t,\ s\ne t) b ab = 0 ( a ∈ J s , b ∈ J t , s = t ) whenever b b b is exchangeable. Consequently, mutations supported on distinct J ^ s \widehat{J}_s J s commute, and mutations supported on J ^ s \widehat{J}_s J s do not change the mutation columns indexed by J ^ t \widehat{J}_t J t for s ≠ t s\ne t s = t .
Proof. For the initial exchange matrix,
b a b = 0 ( a ∈ J ^ s , b ∈ J ^ t , s ≠ t ) . b_{ab}=0 \qquad(a\in\widehat{J}_s,\ b\in\widehat{J}_t,\ s\ne t). b ab = 0 ( a ∈ J s , b ∈ J t , s = t ) .
If k ∈ J ^ s k\in\widehat{J}_s k ∈ J s is mutated and b ∈ J ^ t b\in\widehat{J}_t b ∈ J t , s ≠ t s\ne t s = t , then b k b = 0 b_{kb}=0 b k b = 0 . Hence
b a b ′ = b a b + [ b a k ] + [ b k b ] + − [ − b a k ] + [ − b k b ] + = b a b ( a ≠ k ) b'_{ab}=b_{ab}+[b_{ak}]_+[b_{kb}]_+-[-b_{ak}]_+[-b_{kb}]_+=b_{ab}\qquad(a\ne k) b ab ′ = b ab + [ b ak ] + [ b k b ] + − [ − b ak ] + [ − b k b ] + = b ab ( a = k )
and
b k b ′ = − b k b = 0. b'_{kb}=-b_{kb}=0. b k b ′ = − b k b = 0.
Thus the entries between distinct components remain zero. The permutations in this sequence only permute the labels, and the assertion follows inductively. □
Proposition A.6. If (4.2) holds for each connected component Δ ^ s \widehat{\Delta}_s Δ s of the subdiagram on I i ⊔ I j I_i\sqcup I_j I i ⊔ I j , then it holds for the original braid move.
Proof. In the construction of Proposition 4.4 , 2-moves first group the letters by connected component. For each s s s , let T ^ s \widehat{T}_s T s be the mutation–permutation sequence for Δ ^ s \widehat{\Delta}_s Δ s ; it acts only at positions in J ^ s \widehat{J}_s J s . Further 2-moves then give the letter order of i ^ \widehat{\boldsymbol{i}} i .
By Lemma A.5 , mutations supported on distinct components commute, and the exchange columns occurring along T ^ s \widehat{T}_s T s , including their entries in rows indexed outside ⋃ u = 1 r J ^ u \bigcup_{u=1}^r\widehat{J}_u ⋃ u = 1 r J u , are unchanged by mutations supported on J ^ t \widehat{J}_t J t , t ≠ s t\ne s t = s . Thus the calculations for the connected components may be performed independently.
Assume that (4.2) holds for every connected component. Then, for each s s s ,
ϕ i ( x ~ κ ′ ( p , u ) ) = x p ′ \phi^i\left(\widetilde{x}_{\kappa'(p,u)}\right)=x'_p ϕ i ( x κ ′ ( p , u ) ) = x p ′
whenever the position κ ′ ( p , u ) \kappa'(p,u) κ ′ ( p , u ) belongs to J ^ s \widehat{J}_s J s .
The 2-moves before and after the sequences T ^ s \widehat{T}_s T s only permute the position labels. Hence
ϕ i ( x ~ κ ′ ( p , u ) ) = x p ′ ( 1 ≤ p ≤ l , 1 ≤ u ≤ e i p ′ ) , \phi^i\left(\widetilde{x}_{\kappa'(p,u)}\right)=x'_p \qquad(1\le p\le l,\ 1\le u\le e_{i'_p}), ϕ i ( x κ ′ ( p , u ) ) = x p ′ ( 1 ≤ p ≤ l , 1 ≤ u ≤ e i p ′ ) ,
which proves the assertion. □
Thus the verification of (4.2) reduces to the calculations for the connected subdiagrams treated in § 8 .
References
(W.-S. Jung) Department of Mathematics, University of Seoul, Seoul 02504, South Korea
Email address , W.-S. Jung: jungws@uos.ac.kr
(S.-j. Oh) Department of Mathematics, Sungkyunkwan University, Suwon, South Korea
Email address , S.-j. Oh: sejin092@gmail.com