Introduction
Bender–Knuth involutions and their group
Bender–Knuth involutions t 1 , … , t n − 1 t_1,\ldots,t_{n-1} t 1 , … , t n − 1 act on semistandard Young tableaux with entries in { 1 , … , n } \{1,\ldots,n\} { 1 , … , n } . The involution t i t_i t i rearranges the entries i i i and i + 1 i+1 i + 1 so as to interchange the number of i i i ’s and i + 1 i+1 i + 1 ’s, while preserving the shape and all other entries. They were introduced to give a combinatorial proof that Schur functions are symmetric.
The individual involutions t i t_i t i are simple, but the group they generate is not. Berenstein and Kirillov [6 ] initiated the study of the group B K n \mathrm{BKn} BKn generated by t 1 , … , t n − 1 t_1,\ldots,t_{n-1} t 1 , … , t n − 1 acting on semistandard tableaux of all shapes, now called the Berenstein–Kirillov group. Two distinguished elements are
promotion p n − 1 = t n − 1 t n − 2 ⋯ t 1 p_{n-1}=t_{n-1}t_{n-2}\cdots t_1 p n − 1 = t n − 1 t n − 2 ⋯ t 1 , evacuation q n − 1 = t 1 ( t 2 t 1 ) ( t 3 t 2 t 1 ) ⋯ ( t n − 1 ⋯ t 1 ) q_{n-1}=t_1(t_2t_1)(t_3t_2t_1)\cdots(t_{n-1}\cdots t_1) q n − 1 = t 1 ( t 2 t 1 ) ( t 3 t 2 t 1 ) ⋯ ( t n − 1 ⋯ t 1 ) .
Finding a presentation of B K n \mathrm{BK}_n BK n is still open, and appears to be hard.
Chmutov, Glick and the second author [12 ] showed that the Bender–Knuth involutions satisfy the relations of the cactus group C n C_n C n of Henriques and Kamnitzer [27 ] , giving a surjection C n ↠ B K n C_n \twoheadrightarrow\mathrm{BK}_n C n ↠ BK n . The same conclusion was obtained independently by Halacheva [24 ] using skew Howe duality, and concurrently by Berenstein and Kirillov [7 ] by a group-theoretic argument. Thus the cactus group gives a large family of relations in B K n \mathrm{BK}_n BK n that come from a general structure rather than from computation.
Evacuation words on a rectangle
This paper concerns what happens when one restricts the Bender–Knuth action to rectangular shapes, which in type A A A model Kirillov–Reshetikhin crystals. Let B K R n \mathrm{BKR}_n BKR n be the quotient of B K n \mathrm{BK}_n BK n obtained by imposing all relations that hold on semistandard tableaux of every rectangular shape with entries in { 1 , … , n } \{1,\ldots,n\} { 1 , … , n } .
Call a reduced word i = ( i 1 , … , i N ) \mathbf{i}=(i_1,\ldots,i_N) i = ( i 1 , … , i N ) for w 0 ∈ S n w_0\in\mathfrak{S}_n w 0 ∈ S n an evacuation word if the corresponding product t i 1 ⋯ t i N t_{i_1}\cdots t_{i_N} t i 1 ⋯ t i N acts as evacuation. The standard product
q n − 1 = t 1 ( t 2 t 1 ) ( t 3 t 2 t 1 ) ⋯ ( t n − 1 ⋯ t 1 ) q_{n-1}=t_1(t_2t_1)(t_3t_2t_1)\cdots(t_{n-1}\cdots t_1) q n − 1 = t 1 ( t 2 t 1 ) ( t 3 t 2 t 1 ) ⋯ ( t n − 1 ⋯ t 1 )
is an evacuation word, as is every commutation-equivalent reduced word. On general shapes we know of no others, and conjecture that there are none (Conjecture 1.3, verified for n ≤ 5 n\leq5 n ≤ 5 ). On rectangles, many more words work.
Recall that a reduced word for w 0 w_0 w 0 is adapted to an orientation Ω \Omega Ω of the Dynkin diagram A n − 1 A_{n-1} A n − 1 if i 1 i_1 i 1 is a source of Ω \Omega Ω , i 2 i_2 i 2 is a source of s i 1 Ω s_{i_1}\Omega s i 1 Ω , and so on. Each reduced word is adapted to at most one orientation, and the words adapted to a fixed Ω \Omega Ω form a single commutation class [34 , 5 ] .
Theorem 1.1 (Theorem 5.5). On rectangular tableaux, every reduced word for w 0 w_0 w 0 adapted to an orientation of A n − 1 A_{n-1} A n − 1 is an evacuation word.
Thus the single commutation class coming from the linear orientation is replaced on rectangles by all 2 n − 2 2^{n-2} 2 n − 2 adapted classes. We conjecture that these are all the reduced words representing evacuation on rectangles.
Definition 1.2. A reduced word w w w for w 0 ∈ S n w_0\in\mathfrak{S}_n w 0 ∈ S n is an evacuation word in B K n \mathrm{BK}_n BK n (resp. B K R n \mathrm{BKR}_n BKR n ) if t w = q n − 1 t_w=q_{n-1} t w = q n − 1 in B K n \mathrm{BK}_n BK n (resp. B K R n \mathrm{BKR}_n BKR n ).
Conjecture 1.3. The evacuation words in B K n \mathrm{BK}_n BK n are exactly the reduced words for w 0 w_0 w 0 adapted to the linear orientation.
Here the linear orientation is
1 ⟶ 2 ⟶ ⋯ ⟶ n − 1 1\longrightarrow2\longrightarrow\cdots\longrightarrow n-1 1 ⟶ 2 ⟶ ⋯ ⟶ n − 1
For the opposite linear orientation, the adapted words represent α ( q n − 1 ) \alpha(q_{n-1}) α ( q n − 1 ) , the dual evacuation, on arbitrary tableaux. On rectangles, dual evacuation agrees with evacuation; equivalently, α ( q n − 1 ) = q n − 1 \alpha(q_{n-1})=q_{n-1} α ( q n − 1 ) = q n − 1 . Thus Theorem 1.1 extends this identity from the opposite linear orientation to every orientation of A n − 1 A_{n-1} A n − 1 .
Conjecture 1.4. The evacuation words in B K R n \mathrm{BKR}_n BKR n are exactly the reduced words for w 0 w_0 w 0 adapted to some orientation of A n − 1 A_{n-1} A n − 1 .
We verified both conjectures by computer for n ≤ 5 n\leq5 n ≤ 5 .
Equivalent relations
The exponentially large family of evacuation identities from Theorem 1.1 is equivalent to a very small set of relations. Let α \alpha α be the index-reversing map α ( t i ) = t n − i \alpha(t_i)=t_{n-i} α ( t i ) = t n − i . The Dynkin-reversal relations are
q n − 1 t i q n − 1 = t n − i , i = 1 , … , n − 1. (1) q_{n-1}t_iq_{n-1}=t_{n-i},\qquad i=1,\ldots,n-1. \tag*{(1)} q n − 1 t i q n − 1 = t n − i , i = 1 , … , n − 1. (1)
In other words, conjugation by evacuation acts on the Bender–Knuth generators by the Dynkin diagram automorphism.
Theorem 1.5 (Theorems 3.4 and 5.4 ). The relations ( ∗ ) (*) ( ∗ ) hold in B K R n \mathrm{BKR}_n BKR n . Moreover, modulo the cactus relations, they are equivalent to the whole family of Theorem 1.1 .
Promotion
We also obtain the identity
q n − 1 α ( q n − 1 ) = ( t 1 t 2 ⋯ t n − 1 ) n = p n − n + 1 , q_{n-1}\alpha(q_{n-1})=(t_1t_2\cdots t_{n-1})^n=p_n^{-n+1}, q n − 1 α ( q n − 1 ) = ( t 1 t 2 ⋯ t n − 1 ) n = p n − n + 1 ,
which already holds in the commutation group (Theorem 6.1 ). For the opposite linear orientation
1 ← 2 ← ⋯ ← n − 1 , 1\leftarrow2\leftarrow\cdots\leftarrow n-1, 1 ← 2 ← ⋯ ← n − 1 ,
the corresponding adapted relation on rectangles is
α ( q n − 1 ) = q n − 1 . \alpha(q_{n-1})=q_{n-1}. α ( q n − 1 ) = q n − 1 .
Together with q n − 1 2 = 1 q_{n-1}^2=1 q n − 1 2 = 1 , this gives p n − 1 n = 1 p_{n-1}^n=1 p n − 1 n = 1 on rectangles, which is Haiman’s theorem [23 Thm. 4.4] . We show that for every n ≥ 4 n\ge4 n ≥ 4 , the relation p n − 1 n = 1 p_{n-1}^n=1 p n − 1 n = 1 is strictly weaker than the relations ( ∗ ) (*) ( ∗ ) (Proposition 6.4). Thus Haiman’s relation captures only part of the relations we describe on rectangles.
Adapted words and sorting networks
A reduced word for w 0 w_0 w 0 can be viewed as a sorting network on n n n wires. Choosing four wires and keeping only the crossings between them gives a reduced word for w 0 ∈ S 4 w_0\in S_4 w 0 ∈ S 4 , which we call a 4-subnetwork . Warrington [42 ] introduced this point of view in connection with Sylvester’s four-point problem. Under the Goodman–Pollack correspondence with planar point configurations, the sixteen reduced words for w 0 ∈ S 4 w_0\in S_4 w 0 ∈ S 4 split into twelve corresponding to four points in convex position and four,
X = { 123212 , 321232 , 212321 , 232123 } , X=\{123212,\ 321232,\ 212321,\ 232123\}, X = { 123212 , 321232 , 212321 , 232123 } ,
corresponding to configurations in which one point lies inside the triangle formed by the other three.
Meng [35 ] studied reduced words avoiding these four subnetworks and asked how many X X X -avoiding reduced words for w 0 ∈ S n w_0\in S_n w 0 ∈ S n there are. The counts reported by Warrington begin
2 , 12 , 328 , 54520 , 68641152. 2,12,328,54520,68641152. 2 , 12 , 328 , 54520 , 68641152.
These are exactly the numbers of adapted reduced words that we obtain. In fact, we prove the following.
Theorem 1.6 (Theorem 4.9 ). Every adapted reduced word for w 0 w_0 w 0 is X X X -avoiding.
The rectangular cactus group
Let R C n RC_n R C n , the rectangular cactus group , be the quotient of C n C_n C n by the Dynkin-reversal relations. For n ≥ 3 n\ge3 n ≥ 3 , Theorem 7.2 identifies this quotient with the big cactus group J n J_n J n via the map that forgets the root marked point. By construction,
C n ↠ R C n ↠ B K R n . C_n\twoheadrightarrow RC_n\twoheadrightarrow\mathrm{BKR}_n. C n ↠ R C n ↠ BKR n .
The second map need not be an isomorphism: already for n = 4 n=4 n = 4 , R C 4 ↠ B K R 4 RC_4\twoheadrightarrow\mathrm{BKR}_4 R C 4 ↠ BKR 4 has nontrivial kernel.
Theorem 1.7 (Theorem 3.10 ). The natural map R C 4 ↠ B K R 4 RC_4\twoheadrightarrow\mathrm{BKR}_4 R C 4 ↠ BKR 4 is not an isomorphism.
Conjecture 1.8 (Conjecture 3.11).
B K R n ≅ B K n / ⟨ q n − 1 t i q n − 1 t n − i ⟩ . \mathrm{BKR}_n \cong\mathrm{BK}_n/\langle q_{n-1}t_iq_{n-1}t_{n-i}\rangle. BKR n ≅ BK n / ⟨ q n − 1 t i q n − 1 t n − i ⟩ .
In other words, we conjecture that the Dynkin-reversal relations generate all additional relations that appear when one restricts from arbitrary shapes to rectangles. This avoids the much harder problem of finding a presentation of B K n \mathrm{BK}_n BK n itself.
Crystals and invariant tensors
The cactus group of Henriques and Kamnitzer [27 ] acts on tensor products B 1 ⊗ ⋯ ⊗ B n B_1 \otimes\cdots\otimes B_n B 1 ⊗ ⋯ ⊗ B n of crystals: the generator q [ i , j ] q_{[i,j]} q [ i , j ] reverses the factors i , … , j i,\ldots,j i , … , j using the crystal commutor, built from the Schützenberger involution. We show that the group R C n ≅ J n ′ \mathcal{R}C_n \cong J'_n R C n ≅ J n ′ has a natural meaning in this setting.
An invariant of a crystal is an element on which every e i e_i e i and every f i f_i f i is undefined, equivalently a one-element connected component. We write Inv ( B ) \operatorname{Inv}(B) Inv ( B ) for the set of invariants. Since the cactus action is by crystal isomorphisms, it carries invariants to invariants of the tensor product with the factors permuted.
Theorem 1.9. Let B 1 , … , B n B_1,\ldots,B_n B 1 , … , B n be finite normal crystals for s l m \mathfrak{sl}_m sl m . On Inv ( B 1 ⊗ ⋯ ⊗ B n ) \operatorname{Inv}(B_1 \otimes\cdots\otimes B_n) Inv ( B 1 ⊗ ⋯ ⊗ B n ) , every reduced word for w 0 w_0 w 0 adapted to an orientation of A n − 1 A_{n-1} A n − 1 , read in the elements t i ∈ C n t_i \in C_n t i ∈ C n of Theorem 2.5 , acts by
b 1 ⊗ ⋯ ⊗ b n ⟼ ξ ( b n ) ⊗ ⋯ ⊗ ξ ( b 1 ) , b_1 \otimes\cdots\otimes b_n \longmapsto\xi(b_n) \otimes\cdots\otimes\xi(b_1), b 1 ⊗ ⋯ ⊗ b n ⟼ ξ ( b n ) ⊗ ⋯ ⊗ ξ ( b 1 ) , where ξ \xi ξ is the Schützenberger involution. Equivalently, the cactus action on invariants factors through R C n RC_n R C n .
We give a combinatorial proof in type A and, in § 9.5 , a topological proof valid for finite normal crystals of any complex semisimple Lie algebra; see Corollary 8.19.
Theorem 5.5 is the special case in which all factors are single-row crystals; the standard-tableau case is obtained when all factors are B ( ϖ 1 ) B(\varpi_1) B ( ϖ 1 ) . Thus the tableau relations studied above are the simplest instances of a more general phenomenon for invariant tensors.
The big cactus group and the geometry of invariant tensors
The rectangular cactus group has a topological interpretation. The ordinary cactus group is the equivariant fundamental group [13 , 27 ]
C n = π 1 S n ( M ‾ 0 , n + 1 ( R ) ) , C_n=\pi_1^{\mathfrak{S}_n}\left(\overline{M}_{0,n+1}(\mathbb{R})\right), C n = π 1 S n ( M 0 , n + 1 ( R ) ) ,
where S n \mathfrak{S}_n S n permutes n n n marked points and fixes the remaining point, denoted ∞ \infty ∞ . Henriques’s big cactus group [26 ] is
J n ′ = π 1 S n ( M ‾ 0 , n ( R ) ) . J'_n=\pi_1^{\mathfrak{S}_n}\left(\overline{M}_{0,n}(\mathbb{R})\right). J n ′ = π 1 S n ( M 0 , n ( R ) ) .
Thus the distinction is whether the stable curve has a distinguished root in addition to the n n n permuted markings.
Theorem 1.10 (Theorem 7.2). For n ≥ 3 n \ge3 n ≥ 3 , forgetting ∞ \infty ∞ induces an isomorphism
C n / ⟨ q i t i q n − i : 1 ≤ i < n ⟩ → ∼ J n ′ . C_n/\langle q_i t_i q_{n-i}:1 \le i < n\rangle\xrightarrow{\sim} J'_n. C n / ⟨ q i t i q n − i : 1 ≤ i < n ⟩ ∼ J n ′ . In particular, R C n ≅ J n ′ RC_n \cong J'_n R C n ≅ J n ′ .
Consequently, the Dynkin-reversal relations and the adapted-word relations give two presentations of the big cactus group in the t i t_i t i generators. A different geometric presentation was given by Davis–Januszkiewicz–Scott [13 Section 8.6] ; our description identifies the kernel of the root-forgetting map explicitly. The proof uses the caterpillar paths and the mirabolic generation theorem of Gurenkova, Halacheva and the third author [22 ] .
This interpretation also explains the relations on invariant tensors. By the Gaudin monodromy theorem [25 ] , the cactus action on tensor-product multiplicity sets is realized as monodromy of a covering of M ‾ 0 , n + 1 ( R ) \overline{M}_{0,n+1}(\mathbb{R}) M 0 , n + 1 ( R ) . For the multiplicity of the trivial representation, this covering is pulled back from M ‾ 0 , n ( R ) \overline{M}_{0,n}(\mathbb{R}) M 0 , n ( R ) . On the smooth locus, the descent follows from the projective invariance of the reduced Gaudin algebra [29 Theorem 4.15(2)] ; in § 9.5 , we extend it to the boundary. This gives a topological proof of Theorem 8.1 and extends Theorem 1.9 to finite normal crystals of arbitrary semisimple type.
The proof of Theorem 7.2 uses only § 2 and § 3 , so § 7 can be read directly after them.
Related work
The big cactus group J n ′ = π 1 σ n ( M ‾ 0 , n ( R ) ) J'_n = \pi_1^{\sigma_n}(\overline{M}_{0,n}(\mathbb{R})) J n ′ = π 1 σ n ( M 0 , n ( R )) was introduced by Henriques [26 ] . Our presentation of it in Bender–Knuth generators uses two ingredients from forthcoming work of Gurenkova, Halacheva and the third author [22 ] : a geometric realization of the elements t i t_i t i as paths in the real Deligne–Mumford space, and a generation theorem for the mirabolic cactus group.
The connection between Bender–Knuth involutions and the ordinary cactus group was established independently by Chmutov, Glick and the second author [12 ] , by Halacheva [24 ] , and by Berenstein and Kirillov [7 ] . The closest precursor of Theorem 5.5 is the work of Garver, Patrias and Thomas [19 Thm. 1.10] . They show that reflection at a source of a Dynkin quiver corresponds to a piecewise-linear toggle on the associated minuscule reverse plane partition. Our knitting argument uses the same basic idea in a more elementary form, following source reflections along a full adapted reduced word for w 0 w_0 w 0 .
The local-rule formulation used in § 8 is due to Westbury [44 Def. 3.1] . The general proof for invariant tensors also uses the realization of the cactus action as monodromy of Gaudin eigenlines and the projective invariance of the reduced Gaudin algebra [25 , 29 ] . Further connections, including Bender–Knuth groups in other settings, sorting networks, and related cactus actions, are discussed in § 10 .
AI statement. Claude and ChatGPT helped discover Theorem 8.1 and its combinatorial proof, and suggested, with computational evidence, that the quotient of C n C_n C n by the Dynkin-reversal relations is the big cactus group. AI was also used extensively in editing.
Preliminaries
Tableaux and Bender–Knuth involutions
We identify a partition λ = ( λ 1 ≥ λ 2 ≥ ⋯ ) \lambda= (\lambda_1 \ge\lambda_2 \ge\cdots) λ = ( λ 1 ≥ λ 2 ≥ ⋯ ) with its Young diagram. A semistandard Young tableau (SSYT) of shape λ \lambda λ with entries in [ n ] = { 1 , … , n } [n] = \{1,\ldots,n\} [ n ] = { 1 , … , n } is a filling of λ \lambda λ that is weakly increasing along rows and strictly increasing down columns. We write SSYT ( λ , n ) \operatorname{SSYT}(\lambda,n) SSYT ( λ , n ) for the set of such tableaux, and
wt ( T ) = ( wt 1 ( T ) , … , wt n ( T ) ) \operatorname{wt}(T) = (\operatorname{wt}_1(T), \ldots, \operatorname{wt}_n(T)) wt ( T ) = ( wt 1 ( T ) , … , wt n ( T ))
for the content of T T T , where wt a ( T ) \operatorname{wt}_a(T) wt a ( T ) is the number of entries equal to a a a .
Fix i ∈ { 1 , … , n − 1 } i \in\{1,\ldots,n-1\} i ∈ { 1 , … , n − 1 } . An entry i i i is called free if there is no i + 1 i+1 i + 1 directly below it, and an entry i + 1 i+1 i + 1 is called free if there is no i i i directly above it. In each row, the free i i i ’s and ( i + 1 ) (i+1) ( i + 1 ) ’s form a block
i a ( i + 1 ) b i^a(i+1)^b i a ( i + 1 ) b
for some a , b ≥ 0 a,b \ge0 a , b ≥ 0 . The Bender–Knuth involution t i t_i t i replaces this block by
i b ( i + 1 ) a i^b(i+1)^a i b ( i + 1 ) a
in every row. The resulting tableau is again semistandard. Thus t i t_i t i exchanges the numbers of i i i ’s and ( i + 1 ) (i+1) ( i + 1 ) ’s:
wt ( t i T ) = s i wt ( T ) , \operatorname{wt}(t_iT) = s_i\operatorname{wt}(T), wt ( t i T ) = s i wt ( T ) ,
where s i s_i s i interchanges the i i i -th and ( i + 1 ) (i+1) ( i + 1 ) -st coordinates. In particular, t i 2 = id t_i^2 = \operatorname{id} t i 2 = id .
Example 2.1. Let
T=\begin{array}{|c|c|c|}\hline1&1&2\\\hline2&3&\\\cline{1-2}\end{array}\in\operatorname{SSYT}((3,2),3),\qquad\operatorname{wt}(T)=(2,2,1).
For i = 2 i=2 i = 2 , all three entries 2 , 2 , 3 2,2,3 2 , 2 , 3 are free. The free block in the first row is 2 2 2 , so it becomes 3 3 3 , while the block 23 23 23 in the second row is unchanged. Hence
t_{2}T=\begin{array}{|c|c|c|}\hline1&1&3\\\hline2&3&\\\cline{1-2}\end{array},\qquad\operatorname{wt}(t_{2}T)=(2,1,2).
For i = 1 i=1 i = 1 , the 1 1 1 in the first column and the 2 2 2 below it are not free. The remaining free block in the first row is 12 12 12 , so it is unchanged. Thus t 1 T = T t_{1}T=T t 1 T = T .
Definition 2.2. The Berenstein–Kirillov group B K n \mathrm{BK}_{n} BK n is the group generated by t 1 , … , t n − 1 t_{1},\ldots,t_{n-1} t 1 , … , t n − 1 acting on
⨆ λ SSYT ( λ , n ) , \bigsqcup_{\lambda}\operatorname{SSYT}(\lambda,n), λ ⨆ SSYT ( λ , n ) , where the union is over all partitions λ \lambda λ . The rectangular Berenstein–Kirillov group B K R n \mathrm{BKR}_{n} BKR n is the group generated by the same operators acting only on rectangular shapes:
⨆ k , m ≥ 1 SSYT ( ( m k ) , n ) . \bigsqcup_{k,m\geq1}\operatorname{SSYT}((m^{k}),n). k , m ≥ 1 ⨆ SSYT (( m k ) , n ) . Thus a word in the t i t_i t i is trivial in B K n \mathrm{BK}_{n} BK n if and only if it acts trivially on tableaux of every shape, and trivial in B K R n \mathrm{BKR}_{n} BKR n if and only if it acts trivially on every rectangle. Thus B K R n \mathrm{BKR}_{n} BKR n is a quotient of B K n \mathrm{BK}_{n} BK n , via restriction to rectangular shapes:
B K n ↠ B K R n . \mathrm{BK}_{n}\twoheadrightarrow\mathrm{BKR}_{n}. BK n ↠ BKR n .
We will use repeatedly the following fact, due to Berenstein and Kirillov [6 Corollary 1.1] ; see also [12 Remark 1.9] .
Lemma 2.3. For every n ≥ 3 n\geq3 n ≥ 3 ,
( t 1 t 2 ) 6 = 1 (t_{1}t_{2})^{6}=1 ( t 1 t 2 ) 6 = 1 in B K n \mathrm{BK}_{n} BK n , and hence also in B K R n \mathrm{BKR}_{n} BKR n .
The analogous statement for other pairs of adjacent generators is false.
Evacuation, promotion, and the elements q [ i , j ] q_{[i,j]} q [ i , j ]
Following [6 , 12 ] , define
p i = t i t i − 1 ⋯ t 1 , q i = t 1 ( t 2 t 1 ) ( t 3 t 2 t 1 ) ⋯ ( t i t i − 1 ⋯ t 1 ) = q i − 1 p i , p_i=t_it_{i-1}\cdots t_1,\qquad q_i=t_1(t_2t_1)(t_3t_2t_1)\cdots(t_it_{i-1}\cdots t_1)=q_{i-1}p_i, p i = t i t i − 1 ⋯ t 1 , q i = t 1 ( t 2 t 1 ) ( t 3 t 2 t 1 ) ⋯ ( t i t i − 1 ⋯ t 1 ) = q i − 1 p i ,
and, for 1 ≤ i < j ≤ n 1\leq i<j\leq n 1 ≤ i < j ≤ n ,
q [ i , j ] = q j − 1 q j − i q j − 1 . q_{[i,j]}=q_{j-1}q_{j-i}q_{j-1}. q [ i , j ] = q j − 1 q j − i q j − 1 .
The element p n − 1 p_{n-1} p n − 1 acts as promotion , while q n − 1 q_{n-1} q n − 1 acts as evacuation . We write p : = p n − 1 p:=p_{n-1} p := p n − 1 , and q : = q n − 1 q:=q_{n-1} q := q n − 1 . In particular, q [ 1 , j ] = q j − 1 q_{[1,j]}=q_{j-1} q [ 1 , j ] = q j − 1 . We compose operators from the right: in t j 1 t j 2 ⋯ t j N t_{j_1}t_{j_2}\cdots t_{j_N} t j 1 t j 2 ⋯ t j N , the factor t j N t_{j_N} t j N acts first.
The following classical fact identifies the elements q i q_i q i with evacuation [6 Theorem 2.1] ; see also [12 Proposition 4.2] .
Lemma 2.4. For every partition λ \lambda λ , the element q n − 1 q_{n-1} q n − 1 acts on SSYT ( λ , n ) \operatorname{SSYT}(\lambda,n) SSYT ( λ , n ) as Schützenberger evacuation e λ e_{\lambda} e λ .
We will use the following standard description of evacuation. Let w ( T ) w(T) w ( T ) be the row reading word of T T T , obtained by reading the rows from left to right, starting with the bottom row. If P ( w ) P(w) P ( w ) denotes the RSK row-insertion tableau, then
e λ ( T ) = P ( rc ( w ( T ) ) ) . (2) e_{\lambda}(T)=P(\operatorname{rc}(w(T))). \tag*{(2)} e λ ( T ) = P ( rc ( w ( T ))) . (2)
where rc reverses the word and replaces each letter a a a by n + 1 − a n+1-a n + 1 − a ; see [40 App. A1] and [18 ] . Evacuation is an involution and reverses the content:
e λ 2 = id , wt ( e λ T ) = w 0 ⋅ wt ( T ) , e_{\lambda}^{2}=\operatorname{id},\qquad\operatorname{wt}(e_{\lambda}T)=w_{0}\cdot\operatorname{wt}(T), e λ 2 = id , wt ( e λ T ) = w 0 ⋅ wt ( T ) ,
where w 0 w_{0} w 0 reverses the coordinates; see [18 App. A] and [40 App. A1] .
The cactus group
The cactus group C n C_n C n [27 ] is generated by elements q [ i , j ] q_{[i,j]} q [ i , j ] , 1 ≤ i < j ≤ n 1\le i<j\le n 1 ≤ i < j ≤ n , with relations
(C1) q [ i , j ] 2 = 1 ; (C2) q [ i , j ] q [ k , l ] = q [ k , l ] q [ i , j ] if j < k ; (C3) q [ i , j ] q [ k , l ] q [ i , j ] = q [ i + j − l , i + j − k ] if i ≤ k < l ≤ j . \begin{aligned}
\text{(C1)}\quad&q_{[i,j]}^{2}=1;\\
\text{(C2)}\quad&q_{[i,j]}q_{[k,l]}=q_{[k,l]}q_{[i,j]}\text{ if }j<k;\\
\text{(C3)}\quad&q_{[i,j]}q_{[k,l]}q_{[i,j]}=q_{[i+j-l,i+j-k]}\text{ if }i\le k<l\le j.
\end{aligned} (C1) (C2) (C3) q [ i , j ] 2 = 1 ; q [ i , j ] q [ k , l ] = q [ k , l ] q [ i , j ] if j < k ; q [ i , j ] q [ k , l ] q [ i , j ] = q [ i + j − l , i + j − k ] if i ≤ k < l ≤ j .
We use the same notation q [ i , j ] q_{[i,j]} q [ i , j ] for the cactus generators and for their images in B K n \mathrm{BK}_n BK n ; the ambient group will always be clear from context. The connection with the Berenstein–Kirillov group is given by the following result.
Theorem 2.5 ([12 Thms. 1.4, 1.7, 1.8] ). The Bender–Knuth operators q [ i , j ] q_{[i,j]} q [ i , j ] satisfy the cactus relations, giving a natural quotient
C n ↠ B K n . C_n\twoheadrightarrow\mathrm{BK}_n. C n ↠ BK n .
The cactus group also has a presentation with generators t 1 , … , t n − 1 t_1,\ldots,t_{n-1} t 1 , … , t n − 1 . With
q i = t 1 ( t 2 t 1 ) ⋯ ( t i t i − 1 ⋯ t 1 ) , q [ i , j ] = q j − 1 q j − i q j − 1 , q_i=t_1(t_2t_1)\cdots(t_it_{i-1}\cdots t_1),\qquad q_{[i,j]}=q_{j-1}q_{j-i}q_{j-1}, q i = t 1 ( t 2 t 1 ) ⋯ ( t i t i − 1 ⋯ t 1 ) , q [ i , j ] = q j − 1 q j − i q j − 1 ,
the defining relations are
t i 2 = 1 , t i t j = t j t i ( ∣ i − j ∣ > 1 ) , ( t i q [ j , k ] ) 2 = 1 ( i + 1 < j < k ) . t_i^2=1,\qquad t_it_j=t_jt_i\quad(|i-j|>1),\qquad(t_iq_{[j,k]})^2=1\quad(i+1<j<k). t i 2 = 1 , t i t j = t j t i ( ∣ i − j ∣ > 1 ) , ( t i q [ j , k ] ) 2 = 1 ( i + 1 < j < k ) .
Conversely, t 1 = q [ 1 , 2 ] t_1=q_{[1,2]} t 1 = q [ 1 , 2 ] , t 2 = q [ 1 , 2 ] q [ 1 , 3 ] q [ 1 , 2 ] t_2=q_{[1,2]}q_{[1,3]}q_{[1,2]} t 2 = q [ 1 , 2 ] q [ 1 , 3 ] q [ 1 , 2 ] , and, for i ≥ 3 i\ge3 i ≥ 3 , t i = q [ 1 , i ] q [ 1 , i + 1 ] q [ 1 , i ] q [ 1 , i − 1 ] t_i=q_{[1,i]}q_{[1,i+1]}q_{[1,i]}q_{[1,i-1]} t i = q [ 1 , i ] q [ 1 , i + 1 ] q [ 1 , i ] q [ 1 , i − 1 ] .
We will often work first in the following group, where only the “obvious” relations are imposed.
Definition 2.6. The commutation group is the right-angled Coxeter group
Γ n = ⟨ t 1 , … , t n − 1 ∣ t i 2 = 1 , t i t j = t j t i for ∣ i − j ∣ > 1 ⟩ . \Gamma_n=\langle t_1,\ldots,t_{n-1}\mid t_i^2=1,\ t_it_j=t_jt_i\text{ for }|i-j|>1\rangle. Γ n = ⟨ t 1 , … , t n − 1 ∣ t i 2 = 1 , t i t j = t j t i for ∣ i − j ∣ > 1 ⟩ .
Any identity in Γ n \Gamma_n Γ n therefore holds in C n C_n C n , B K n \mathrm{BK}_n BK n , and B K R n \mathrm{BKR}_n BKR n .
The index-reversal map α : t i ↦ t n − i \alpha:t_i\mapsto t_{n-i} α : t i ↦ t n − i preserves the defining relations of Γ n \Gamma_n Γ n and hence is an automorphism of Γ n \Gamma_n Γ n . It does not, in general, descend to the cactus group: already for n = 4 n=4 n = 4 , the image under α \alpha α of the relation ( t 1 q [ 3 , 4 ] ) 2 = 1 (t_1q_{[3,4]})^2=1 ( t 1 q [ 3 , 4 ] ) 2 = 1 fails in C 4 C_4 C 4 . This is why we state the results below at the level of words in Γ n \Gamma_n Γ n .
For a word w = ( j 1 , … , j N ) w=(j_1,\ldots,j_N) w = ( j 1 , … , j N ) , write t w = t j 1 ⋯ t j N t_w=t_{j_1}\cdots t_{j_N} t w = t j 1 ⋯ t j N , rev ( w ) = ( j N , … , j 1 ) \operatorname{rev}(w)=(j_N,\ldots,j_1) rev ( w ) = ( j N , … , j 1 ) . Then t rev ( w ) = t w − 1 t_{\operatorname{rev}(w)}=t_w^{-1} t rev ( w ) = t w − 1 .
Dynkin-reversal relations
Throughout this section, λ = ( m k ) \lambda=(m^k) λ = ( m k ) is a k × m k\times m k × m rectangle.
Evacuation on a rectangle
Definition 3.1. For T ∈ SSYT ( ( m k ) , n ) T \in\operatorname{SSYT}((m^k),n) T ∈ SSYT (( m k ) , n ) , let θ ( T ) \theta(T) θ ( T ) be obtained by rotating T T T by 180 ∘ 180^\circ 18 0 ∘ and replacing each entry a a a by n + 1 − a n+1-a n + 1 − a . Thus θ ( T ) r , c = n + 1 − T k + 1 − r , m + 1 − c \theta(T)_{r,c}=n+1-T_{k+1-r,m+1-c} θ ( T ) r , c = n + 1 − T k + 1 − r , m + 1 − c .
Both rotation and complementation reverse order, so their composition preserves weak increase along rows and strict increase down columns. Hence
θ ( T ) ∈ SSYT ( ( m k ) , n ) , θ 2 = id . \theta(T) \in\operatorname{SSYT}((m^k),n), \qquad\theta^2=\operatorname{id}. θ ( T ) ∈ SSYT (( m k ) , n ) , θ 2 = id .
Lemma 3.2. For rectangular λ \lambda λ , we have
θ = e λ = q n − 1 . \theta=e_\lambda=q_{n-1}. θ = e λ = q n − 1 .
This is standard. Berenstein and Zelevinsky [8 ] describe evacuation on an arbitrary shape as rotation, complementation, and rectification. For a rectangle, rotation again gives a straight shape, so no rectification is needed. The same statement is explicitly made for row-strict tableaux in [36 Lem. 3.1] .
Lemma 3.3. On rectangular shapes,
θ t i θ = t n − i , 1 ≤ i ≤ n − 1. \theta t_i\theta=t_{n-i}, \qquad1\leq i\leq n-1. θ t i θ = t n − i , 1 ≤ i ≤ n − 1. Proof. Write θ = κ ∘ ρ \theta=\kappa\circ\rho θ = κ ∘ ρ , where ρ \rho ρ is rotation by 180 ∘ 180^\circ 18 0 ∘ and κ ( a ) = n + 1 − a \kappa(a)=n+1-a κ ( a ) = n + 1 − a . Thus { i , i + 1 } \{i,i+1\} { i , i + 1 } is sent to { n − i , n − i + 1 } \{n-i,n-i+1\} { n − i , n − i + 1 } , with the two values interchanged.
Rotation also interchanges “above” and “below”. Hence the non-free pairs for t i t_i t i are carried exactly to the non-free pairs for t n − i t_{n-i} t n − i , and therefore free entries are carried to free entries.
If the free block in a row of T T T is i a ( i + 1 ) b i^a(i+1)^b i a ( i + 1 ) b , then the corresponding free block in θ ( T ) \theta(T) θ ( T ) is ( n − i ) b ( n − i + 1 ) a (n-i)^b(n-i+1)^a ( n − i ) b ( n − i + 1 ) a . Applying t n − i t_{n-i} t n − i gives ( n − i ) a ( n − i + 1 ) b (n-i)^a(n-i+1)^b ( n − i ) a ( n − i + 1 ) b , which is the image under θ \theta θ of the block i b ( i + 1 ) a i^b(i+1)^a i b ( i + 1 ) a . Thus θ t i θ = t n − i \theta t_i\theta=t_{n-i} θ t i θ = t n − i . □
Theorem 3.4. In BKR n \operatorname{BKR}_n BKR n ,
q n − 1 t i q n − 1 = t n − i , 1 ≤ i ≤ n − 1. q_{n-1}t_iq_{n-1}=t_{n-i}, \qquad1\leq i\leq n-1. q n − 1 t i q n − 1 = t n − i , 1 ≤ i ≤ n − 1. Proof. This follows immediately from Lemmas 3.2 3.2 3.2 and 3.3 3.3 3.3 . □
Example 3.5. Let n = 4 n=4 n = 4 , λ = ( 2 , 2 ) \lambda=(2,2) λ = ( 2 , 2 ) , and
T = ( 1 1 2 4 ) . T=
\begin{pmatrix}
1 & 1\\
2 & 4
\end{pmatrix}. T = ( 1 2 1 4 ) . Then
T → q 3 ( 1 3 4 4 ) → t 1 ( 2 3 4 4 ) → q 3 ( 1 1 2 3 ) = t 3 T , T \xrightarrow{q^3}
\begin{pmatrix}
1 & 3\\
4 & 4
\end{pmatrix}
\xrightarrow{t_1}
\begin{pmatrix}
2 & 3\\
4 & 4
\end{pmatrix}
\xrightarrow{q^3}
\begin{pmatrix}
1 & 1\\
2 & 3
\end{pmatrix}
=t_3T, T q 3 ( 1 4 3 4 ) t 1 ( 2 4 3 4 ) q 3 ( 1 2 1 3 ) = t 3 T , illustrating q 3 t 1 q 3 = t 3 q_3t_1q_3=t_3 q 3 t 1 q 3 = t 3 .
Equivalent forms
Define
q [ i , j ] 0 = α ( q [ n + 1 − j , n + 1 − i ] ) , q^0_{[i,j]}=\alpha\left(q_{[n+1-j,n+1-i]}\right), q [ i , j ] 0 = α ( q [ n + 1 − j , n + 1 − i ] ) ,
where α \alpha α acts on words by t a ↦ t n − a t_a\mapsto t_{n-a} t a ↦ t n − a .
Example 3.6. For n = 4 n=4 n = 4 ,
q [ 3 , 4 ] 0 = α ( q [ 1 , 2 ] ) = α ( t 1 ) = t 3 , q^0_{[3,4]}=\alpha(q_{[1,2]})=\alpha(t_1)=t_3, q [ 3 , 4 ] 0 = α ( q [ 1 , 2 ] ) = α ( t 1 ) = t 3 , q [ 2 , 4 ] 0 = α ( q [ 1 , 3 ] ) = α ( t 1 t 2 t 1 ) = t 3 t 2 t 3 , q^0_{[2,4]}=\alpha(q_{[1,3]})=\alpha(t_1t_2t_1)=t_3t_2t_3, q [ 2 , 4 ] 0 = α ( q [ 1 , 3 ] ) = α ( t 1 t 2 t 1 ) = t 3 t 2 t 3 , q [ 1 , 4 ] 0 = α ( q [ 1 , 4 ] ) = α ( t 1 t 2 t 1 t 3 t 2 t 1 ) = t 3 t 2 t 3 t 1 t 2 t 3 . q^0_{[1,4]}=\alpha(q_{[1,4]})=\alpha(t_1t_2t_1t_3t_2t_1)=t_3t_2t_3t_1t_2t_3. q [ 1 , 4 ] 0 = α ( q [ 1 , 4 ] ) = α ( t 1 t 2 t 1 t 3 t 2 t 1 ) = t 3 t 2 t 3 t 1 t 2 t 3 .
Proposition 3.7. In any quotient of C n C_n C n , the following are equivalent:
(a) q [ i , j ] = q [ i , j ] 0 q_{[i,j]} = q^0_{[i,j]} q [ i , j ] = q [ i , j ] 0 for all $1 \le i < j \le n$;
(b) q [ j , n ] = q [ j , n ] 0 q_{[j,n]} = q^0_{[j,n]} q [ j , n ] = q [ j , n ] 0 for all 1 ≤ j < n 1 \le j < n 1 ≤ j < n ;
(c) q t i q = t n − i q t_i q = t_{n-i} q t i q = t n − i for all 1 ≤ i ≤ n − 1 1 \le i \le n-1 1 ≤ i ≤ n − 1 .
Proof. The implication (a)⇒ \Rightarrow ⇒ (b) is immediate.
For (b)⇒ \Rightarrow ⇒ (c), we have q [ j , n ] = q q n − j q q_{[j,n]} = q q_{n-j} q q [ j , n ] = q q n − j q and q [ j , n ] 0 = α ( q n − j ) q^0_{[j,n]} = \alpha(q_{n-j}) q [ j , n ] 0 = α ( q n − j ) , so (b) is equivalent to
q q k q = α ( q k ) , 1 ≤ k ≤ n − 1. (3) q q_k q = \alpha(q_k), \qquad1 \le k \le n-1. \tag*{(3)} q q k q = α ( q k ) , 1 ≤ k ≤ n − 1. (3)
Since q k = q k − 1 p k q_k=q_{k-1}p_k q k = q k − 1 p k , comparing (3) for k k k and k − 1 k-1 k − 1 gives q p k q = α ( p k ) q p_k q = \alpha(p_k) q p k q = α ( p k ) . As p k = t k ⋯ t 1 p_k=t_k\cdots t_1 p k = t k ⋯ t 1 , this becomes
q t k ⋯ t 1 q = t n − k ⋯ t n − 1 . q t_k\cdots t_1 q=t_{n-k}\cdots t_{n-1}. q t k ⋯ t 1 q = t n − k ⋯ t n − 1 .
Comparing with the same identity for k − 1 k-1 k − 1 gives q t k q = t n − k q t_k q=t_{n-k} q t k q = t n − k .
For (c)⇒ \Rightarrow ⇒ (a), conjugation by q q q acts on every word by α \alpha α . Hence
q [ i , j ] 0 = α ( q [ n + 1 − j , n + 1 − i ] ) = q q [ n + 1 − j , n + 1 − i ] q = q [ i , j ] , q^0_{[i,j]}=\alpha\left(q_{[n+1-j,n+1-i]}\right)=q q_{[n+1-j,n+1-i]} q=q_{[i,j]}, q [ i , j ] 0 = α ( q [ n + 1 − j , n + 1 − i ] ) = q q [ n + 1 − j , n + 1 − i ] q = q [ i , j ] ,
where the last equality is the cactus relation (C3), using q = q [ 1 , n ] q=q_{[1,n]} q = q [ 1 , n ] .
Corollary 3.8. In B K R n \mathrm{BKR}_n BKR n ,
q [ i , j ] = q [ i , j ] 0 q_{[i,j]}=q^0_{[i,j]} q [ i , j ] = q [ i , j ] 0 for all $1\le i<j\le n.
Definition 3.9. The rectangular cactus group is the quotient
R C n = C n / ⟨ ⟨ q [ 1 , n ] t i q [ 1 , n ] t n − i , 1 ≤ i ≤ n − 1 ⟩ ⟩ . \mathcal{RC}_n=C_n\mathbin{\Big/}\Big\langle\!\Big\langle q_{[1,n]}t_iq_{[1,n]}t_{n-i},\ 1\le i\le n-1\Big\rangle\!\Big\rangle. R C n = C n / ⟨ ⟨ q [ 1 , n ] t i q [ 1 , n ] t n − i , 1 ≤ i ≤ n − 1 ⟩ ⟩ . Equivalently, R C n \mathcal{RC}_n R C n is the quotient of C n C_n C n by the relations q [ i , j ] = q [ i , j ] 0 q_{[i,j]}=q^0_{[i,j]} q [ i , j ] = q [ i , j ] 0 .
By Corollary 3.8, the natural map to the rectangular Berenstein–Kirillov group factors as
C n ↠ R C n ↠ B K R n . C_n\twoheadrightarrow\mathcal{RC}_n\twoheadrightarrow\mathrm{BKR}_n. C n ↠ R C n ↠ BKR n .
The groups R C n \mathcal{RC}_n R C n and B K R n \mathrm{BKR}_n BKR n
By construction, there is a natural surjection
R C n ↠ B K R n . \mathcal{RC}_n\twoheadrightarrow\mathrm{BKR}_n. R C n ↠ BKR n .
One might hope that it is an isomorphism. Already for n = 4 n=4 n = 4 , this is false.
Theorem 3.10. The map R C 4 ↠ B K R 4 \mathcal{RC}_4\twoheadrightarrow\mathrm{BKR}_4 R C 4 ↠ BKR 4 is not an isomorphism.
Proof. Let D ∞ = ⟨ x , y ∣ x 2 = y 2 = 1 ⟩ D_\infty=\langle x,y\mid x^2=y^2=1\rangle D ∞ = ⟨ x , y ∣ x 2 = y 2 = 1 ⟩ be the infinite dihedral group. Define
t 1 , t 3 ⟼ x , t 2 ⟼ y . t_1,t_3\longmapsto x,\qquad t_2\longmapsto y. t 1 , t 3 ⟼ x , t 2 ⟼ y .
We claim that this defines a homomorphism R C 4 → D ∞ \mathcal{RC}_4\to D_\infty R C 4 → D ∞ .
The involution and commutation relations are immediate. Also q 3 = t 1 t 2 t 1 t 3 t 2 t 1 q_3=t_1t_2t_1t_3t_2t_1 q 3 = t 1 t 2 t 1 t 3 t 2 t 1 maps to x y x x y x = 1 xyxxyx=1 x y xx y x = 1 . Therefore the Dynkin-reversal relations q 3 t i q 3 = t 4 − i q_3t_iq_3=t_{4-i} q 3 t i q 3 = t 4 − i are satisfied, since t 1 t_1 t 1 and t 3 t_3 t 3 have the same image. The only remaining cactus relation is ( t 1 q [ 3 , 4 ] ) 2 = 1 (t_1q_{[3,4]})^2=1 ( t 1 q [ 3 , 4 ] ) 2 = 1 . Since q [ 3 , 4 ] = q 3 q 1 q 3 q_{[3,4]}=q_3q_1q_3 q [ 3 , 4 ] = q 3 q 1 q 3 maps to x x x , this relation is also satisfied.
Thus t 1 t 2 t_1t_2 t 1 t 2 maps to x y xy x y , which has infinite order in D ∞ D_\infty D ∞ , so t 1 t 2 t_1t_2 t 1 t 2 has infinite order in R C 4 \mathcal{RC}_4 R C 4 . On the other hand, ( t 1 t 2 ) 6 = 1 (t_1t_2)^6=1 ( t 1 t 2 ) 6 = 1 in B K R 4 \mathrm{BKR}_4 BKR 4 by Lemma 2.3. Therefore R C 4 ↠ B K R 4 \mathcal{RC}_4\twoheadrightarrow\mathrm{BKR}_4 R C 4 ↠ BKR 4 has nontrivial kernel.
Note that this failure comes from relations that already hold in B K n \mathrm{BK}_n BK n but not in the cactus group. This suggests asking instead whether the Dynkin-reversal relations account for all relations introduced by restricting from arbitrary shapes to rectangles.
Conjecture 3.11. The rectangular Berenstein–Kirillov group is obtained from B K n \mathrm{BK}_n BK n by imposing the Dynkin-reversal relations:
B K R n ≅ B K n / ⟨ ⟨ q n − 1 t i q n − 1 t n − i − 1 : 1 ≤ i ≤ n − 1 ⟩ ⟩ . \mathrm{BKR}_n \cong\mathrm{BK}_n \Big/ \Big\langle\!\Big\langle q_{n-1}t_iq_{n-1}t_{n-i}^{-1}: 1 \le i \le n-1 \Big\rangle\!\Big\rangle. BKR n ≅ BK n / ⟨ ⟨ q n − 1 t i q n − 1 t n − i − 1 : 1 ≤ i ≤ n − 1 ⟩ ⟩ .
In Section 7 we identify R C n RC_n R C n with Henriques’s big cactus group J n ′ J'_n J n ′ (Theorem 7.2). Conjecture 3.11 then says that B K R n \mathrm{BKR}_n BKR n is obtained from B K n \mathrm{BK}_n BK n by imposing the same additional relations that turn C n C_n C n into J n ′ J'_n J n ′ : restricting from arbitrary tableaux to rectangles should introduce exactly the relations introduced by forgetting the root.
Knitting words
The strip, Dynkin paths, and flips
Let
S = { ( i , j ) ∈ Z 2 : 1 ≤ j ≤ n − 1 , i + j even } . S = \{(i,j) \in\mathbb{Z}^2 : 1 \le j \le n-1,\ i+j\ \text{even}\}. S = {( i , j ) ∈ Z 2 : 1 ≤ j ≤ n − 1 , i + j even } .
A Dynkin path is a sequence
D = ( ( d 1 , 1 ) , … , ( d n − 1 , n − 1 ) ) D = ((d_1,1),\ldots,(d_{n-1},n-1)) D = (( d 1 , 1 ) , … , ( d n − 1 , n − 1 ))
with ∣ d j + 1 − d j ∣ = 1 |d_{j+1}-d_j|=1 ∣ d j + 1 − d j ∣ = 1 for all j j j . We usually identify D D D with its height vector ( d 1 , … , d n − 1 ) (d_1,\ldots,d_{n-1}) ( d 1 , … , d n − 1 ) .
A Dynkin path determines an orientation Ω D \Omega_D Ω D of the Dynkin diagram A n − 1 A_{n-1} A n − 1 : orient
j ⟶ j + 1 if d j + 1 = d j + 1 , j \longrightarrow j+1 \quad\text{if } d_{j+1}=d_j+1, j ⟶ j + 1 if d j + 1 = d j + 1 ,
and j + 1 → j j+1 \to j j + 1 → j otherwise. Every orientation arises in this way, uniquely up to horizontal translation of the path.
Figure 1. The strip S S S for n = 5 n=5 n = 5 and the Dynkin path D = ( 3 , 2 , 3 , 4 ) D=(3,2,3,4) D = ( 3 , 2 , 3 , 4 ) (blue). The corresponding orientation of A 4 A_4 A 4 is Ω D : 1 ← 2 → 3 → 4 \Omega_D: 1 \leftarrow2 \to3 \to4 Ω D : 1 ← 2 → 3 → 4 . Thus 2 is a source and 1, 4 are sinks.
A vertex j j j is a source of D D D if d j ′ = d j + 1 d_{j'}=d_j+1 d j ′ = d j + 1 for every neighbour j ′ ∈ { j − 1 , j + 1 } ∩ [ 1 , n − 1 ] j' \in\{j-1,j+1\}\cap[1,n-1] j ′ ∈ { j − 1 , j + 1 } ∩ [ 1 , n − 1 ] . Equivalently, j j j is a source of Ω D \Omega_D Ω D . The flip at a source j j j replaces d j d_j d j by d j + 2 d_j+2 d j + 2 and leaves all other heights unchanged. On orientations, this is reflection at the source j j j .
In coordinates, the flip is the local move
( i , j ) , ( i − 1 , j + 1 ) , ( i , j + 2 ) ⟼ ( i , j ) , ( i + 1 , j + 1 ) , ( i , j + 2 ) , (i,j),(i-1,j+1),(i,j+2)\longmapsto(i,j),(i+1,j+1),(i,j+2), ( i , j ) , ( i − 1 , j + 1 ) , ( i , j + 2 ) ⟼ ( i , j ) , ( i + 1 , j + 1 ) , ( i , j + 2 ) ,
with the obvious boundary versions at j = 1 j=1 j = 1 and j = n − 1 j=n-1 j = n − 1 .
A knitting word from D D D to D ′ D' D ′ is a sequence w = ( j 1 , … , j N ) w=(j_1,\ldots,j_N) w = ( j 1 , … , j N ) such that successive flips at the sources j 1 , … , j N j_1,\ldots,j_N j 1 , … , j N carry D D D to D ′ D' D ′ . We write W ( D , D ′ ) W(D,D') W ( D , D ′ ) for the set of all knitting words from D D D to D ′ D' D ′ .
Figure 2. A knitting word w = ( 2 , 3 , 1 ) ∈ W ( D , D ′ ′ ′ ) w=(2,3,1)\in W(D,D^{\prime\prime\prime}) w = ( 2 , 3 , 1 ) ∈ W ( D , D ′′′ ) for n = 5 n=5 n = 5 . In each panel the source to be flipped is circled, and the corresponding orientation of A 4 A_4 A 4 is shown on the right. Each flip raises one height by 2 and reverses the arrows incident to that vertex.
Lemma 4.1. Every Dynkin path has a source. Moreover, any Dynkin path can be transformed by flips into a path with any prescribed orientation. In particular, every orientation can be reached from the linear path
D l i n = ( 1 , 2 , … , n − 1 ) . D_{\mathrm{lin}}=(1,2,\ldots,n-1). D lin = ( 1 , 2 , … , n − 1 ) .
Proof. A vertex of minimal height is a source. To prove the second statement, encode the orientation by ε j = d j + 1 − d j ∈ { ± 1 } \varepsilon_j=d_{j+1}-d_j\in\{\pm1\} ε j = d j + 1 − d j ∈ { ± 1 } . A flip at an interior source moves a − 1 -1 − 1 one step to the right, while flips at the two ends create a − 1 -1 − 1 on the left or delete one on the right. Thus any sign vector can be transformed into ( + 1 , … , + 1 ) (+1,\ldots,+1) ( + 1 , … , + 1 ) by repeatedly moving its rightmost − 1 -1 − 1 to the right and deleting it at the end. Conversely, starting from ( + 1 , … , + 1 ) (+1,\ldots,+1) ( + 1 , … , + 1 ) , we can obtain any prescribed sign vector by creating − 1 -1 − 1 ’s at the left end and moving them to their desired positions, placing the rightmost ones first. Thus every orientation can be reached from every other one. □
Knitting words define elements of Γ n \Gamma_n Γ n
We now show that a knitting word from D D D to D ′ D^\prime D ′ determines an element of Γ n \Gamma_n Γ n independent of the choice of flips. Assume n ≥ 3 n\ge3 n ≥ 3 .
Say that D ′ D^\prime D ′ lies weakly to the right of D D D , and write D ≤ D ′ D\le D^\prime D ≤ D ′ , if d j ′ ≥ d j d^\prime_j\ge d_j d j ′ ≥ d j for all j j j . Let
R ( D , D ′ ) = { ( i , j ) ∈ S : d j ≤ i < d j ′ } . R(D,D^\prime)=\{(i,j)\in S:d_j\le i<d^\prime_j\}. R ( D , D ′ ) = {( i , j ) ∈ S : d j ≤ i < d j ′ } .
Order R ( D , D ′ ) R(D,D^\prime) R ( D , D ′ ) by
( i − 1 , j ± 1 ) ⋖ ( i , j ) (i-1,j\pm1)\lessdot(i,j) ( i − 1 , j ± 1 ) ⋖ ( i , j )
whenever both nodes lie in the region.
Lemma 4.2. Let D ′ D^\prime D ′ lie weakly to the right of D D D .
(a) Knitting sequences from D D D to D ′ D^\prime D ′ are in bijection with the linear extensions of R ( D , D ′ ) R(D,D^\prime) R ( D , D ′ ) , with the knitting word given by the levels of the nodes.
(b) Every knitting word contains the letter j j j exactly ( d j ′ − d j ) / 2 (d^\prime_j-d_j)/2 ( d j ′ − d j ) /2 times. In particular, ∣ w ∣ = ∣ R ( D , D ′ ) ∣ |w|=|R(D,D^\prime)| ∣ w ∣ = ∣ R ( D , D ′ ) ∣ .
(c) Any two knitting words are related by commutations of adjacent letters j , j ′ j,j^\prime j , j ′ with ∣ j − j ′ ∣ > 1 |j-j^\prime|>1 ∣ j − j ′ ∣ > 1 . Hence they define the same element
Q ( D , D ′ ) ∈ Γ n . Q(D,D^\prime)\in\Gamma_n. Q ( D , D ′ ) ∈ Γ n .
and therefore the same elements of C n C_n C n , B K n \mathrm{BK}_n BK n , and B K R n \mathrm{BKR}_n BKR n .
Figure 3. The region R ( D , D ′ ) R(D,D') R ( D , D ′ ) for D = ( 3 , 2 , 3 , 4 ) D=(3,2,3,4) D = ( 3 , 2 , 3 , 4 ) and D ′ = ( 5 , 6 , 5 , 4 ) D'=(5,6,5,4) D ′ = ( 5 , 6 , 5 , 4 ) . The two linear extensions give the knitting words ( 2 , 1 , 3 , 2 ) (2,1,3,2) ( 2 , 1 , 3 , 2 ) and ( 2 , 3 , 1 , 2 ) (2,3,1,2) ( 2 , 3 , 1 , 2 ) , differing by the commutation of 1 and 3.
Proof. A flip at level j j j moves the path across the next node of R ( D , D ′ ) R(D,D') R ( D , D ′ ) on that level. It is legal exactly when the predecessors of that node have already been crossed. Thus knitting sequences are precisely the linear extensions of R ( D , D ′ ) R(D,D') R ( D , D ′ ) , proving (a). Part (b) is immediate, since each flip at level j j j raises d j d_j d j by 2.
For (c), any two linear extensions differ by swaps of adjacent incomparable elements. In R ( D , D ′ ) R(D,D') R ( D , D ′ ) , nodes on the same or adjacent levels are comparable, so incomparable nodes have levels j , j ′ j,j' j , j ′ with ∣ j − j ′ ∣ > 1 \lvert j-j'\rvert>1 ∣ j − j ′ ∣ > 1 . The corresponding generators commute in Γ n \Gamma_n Γ n . □
Later, we will use the composition rule
Q ( D , D ′ ′ ) = Q ( D , D ′ ) Q ( D ′ , D ′ ′ ) Q(D,D'')=Q(D,D')Q(D',D'') Q ( D , D ′′ ) = Q ( D , D ′ ) Q ( D ′ , D ′′ )
( D ≤ D ′ ≤ D ′ ′ ) (D\le D'\le D'') ( D ≤ D ′ ≤ D ′′ ) , which follows by concatenating knitting sequences.
Knitting w 0 w_0 w 0 -words
Definition 4.3. For a Dynkin path D D D , let D ˉ \bar{D} D ˉ be its image under ( i , j ) ↦ ( i + n , n − j ) (i,j)\mapsto(i+n,n-j) ( i , j ) ↦ ( i + n , n − j ) , so that d ˉ j = d n − j + n \bar{d}_j=d_{n-j}+n d ˉ j = d n − j + n . A knitting w 0 w_0 w 0 -word for D D D is a knitting word from D D D to D ˉ \bar{D} D ˉ . We write
K ( D ) = W ( D , D ˉ ) , Q D = Q ( D , D ˉ ) ∈ Γ n . \mathcal{K}(D)=W(D,\bar{D}),\qquad Q_D=Q(D,\bar{D})\in\Gamma_n. K ( D ) = W ( D , D ˉ ) , Q D = Q ( D , D ˉ ) ∈ Γ n .
The path D ˉ \bar{D} D ˉ has orientation α ( Ω D ) \alpha(\Omega_D) α ( Ω D ) and lies weakly to the right of D D D , since
d ˉ j − d j = n + d n − j − d j > 0. \bar{d}_j-d_j=n+d_{n-j}-d_j>0. d ˉ j − d j = n + d n − j − d j > 0.
Horizontal translation of D D D translates the whole picture, so K ( D ) \mathcal{K}(D) K ( D ) and Q D Q_D Q D depend only on the orientation Ω D \Omega_D Ω D .
Lemma 4.4. Every knitting w 0 w_0 w 0 -word has length ( n 2 ) \binom{n}{2} ( 2 n ) .
Proof. By Lemma 4.2 (b),
∣ R ( D , D ˉ ) ∣ = 1 2 ∑ j = 1 n − 1 ( d ˉ j − d j ) = 1 2 ∑ j = 1 n − 1 ( n + d n − j − d j ) = ( n 2 ) . \lvert R(D,\bar{D})\rvert=\frac{1}{2}\sum_{j=1}^{n-1}(\bar{d}_j-d_j)=\frac{1}{2}\sum_{j=1}^{n-1}(n+d_{n-j}-d_j)=\binom{n}{2}. ∣ R ( D , D ˉ )∣ = 2 1 j = 1 ∑ n − 1 ( d ˉ j − d j ) = 2 1 j = 1 ∑ n − 1 ( n + d n − j − d j ) = ( 2 n ) .
A word w = ( j 1 , … , j N ) w=(j_1,\ldots,j_N) w = ( j 1 , … , j N ) is adapted to an orientation Ω \Omega Ω if j 1 j_1 j 1 is a source of Ω \Omega Ω and, successively, j r j_r j r is a source after reflections at j 1 , … , j r − 1 j_1,\ldots,j_{r-1} j 1 , … , j r − 1 . This is the usual notion of an adapted reduced word; see, for example, [34 §4.7] , [5 ] , and [2 §VII] .
Lemma 4.5. The knitting w 0 w_0 w 0 -words for D D D are exactly the reduced words for w 0 w_0 w 0 adapted to Ω D \Omega_D Ω D . Moreover, a reduced word for w 0 w_0 w 0 is adapted to at most one orientation.
Proof. Every knitting word is adapted. We first show that it is a reduced word for w 0 w_0 w 0 . For the linear path D l i n = ( 1 , 2 , … , n − 1 ) D_{\mathrm{lin}}=(1,2,\ldots,n-1) D lin = ( 1 , 2 , … , n − 1 ) , the standard word
( 1 ) ( 2 , 1 ) ⋯ ( n − 1 , … , 1 ) (1)(2,1)\cdots(n-1,\ldots,1) ( 1 ) ( 2 , 1 ) ⋯ ( n − 1 , … , 1 )
is a knitting word. Indeed, after the segments through ( k , … , 1 ) (k,\ldots,1) ( k , … , 1 ) have been performed, the heights are
2 k + 1 , 2 k , … , k + 2 , k + 1 , k + 2 , … , n − 1 2k+1,2k,\ldots,k+2,k+1,k+2,\ldots,n-1 2 k + 1 , 2 k , … , k + 2 , k + 1 , k + 2 , … , n − 1
so k + 1 k+1 k + 1 is the next source. At the end the j j j -th height is 2 n − j 2n-j 2 n − j , which is exactly the j j j -th height of D ‾ l i n \overline{D}_{\mathrm{lin}} D lin . The word is also the standard reduced word for w 0 w_0 w 0 .
Suppose now that D ′ D' D ′ is obtained from D D D by flipping a source j j j . Choose ( j , x ) ∈ K ( D ) (j,x)\in\mathcal{K}(D) ( j , x ) ∈ K ( D ) . Then ( x , n − j ) ∈ K ( D ′ ) (x,n-j)\in\mathcal{K}(D') ( x , n − j ) ∈ K ( D ′ ) , since the flip at n − j n-j n − j takes D ‾ \overline{D} D to D ‾ ′ \overline{D}' D ′ ; see Lemma 5.1 (1). If s j s x = w 0 s_js_x=w_0 s j s x = w 0 , then
s x s n − j = s j w 0 s n − j = w 0 . s_xs_{n-j}=s_jw_0s_{n-j}=w_0. s x s n − j = s j w 0 s n − j = w 0 .
By Lemma 4.1 , every orientation can be reached from the linear one by source flips. It follows that some knitting word for every D D D represents w 0 w_0 w 0 . By Lemma 4.2 (c), they all do. Their length ( n 2 ) = ℓ ( w 0 ) \binom{n}{2}=\ell(w_0) ( 2 n ) = ℓ ( w 0 ) by Lemma 4.4, so they are reduced.
Conversely, let w w w and w ′ w' w ′ be reduced words for the same u ∈ S n u\in\mathfrak{S}_n u ∈ S n , both adapted to the same orientation. We claim that they differ only by commutations of nonadjacent letters.
Let j j j be the first letter of w w w . Since j j j lies in the support of u u u , it also occurs in w ′ w' w ′ . Before its first occurrence in w ′ w' w ′ , no neighbour of j j j can occur: such a neighbour cannot become a source before the edge joining it to j j j is reversed, and that requires reflecting at j j j first. Hence every preceding letter commutes with j j j , so we may move j j j to the front of w ′ w' w ′ . Deleting the initial j j j from both words and reflecting the orientation at j j j , the claim follows by induction.
Now let w w w be a reduced word for w 0 w_0 w 0 adapted to Ω D \Omega_D Ω D , and choose v ∈ K ( D ) v\in\mathcal{K}(D) v ∈ K ( D ) . By the claim, w w w and v v v differ only by distant commutations. These commutations preserve legality and the endpoint of the flip sequence, so w ∈ K ( D ) w\in\mathcal{K}(D) w ∈ K ( D ) .
Finally, an adapted reduced word for w 0 w_0 w 0 determines the orientation. Every letter 1 , … , n − 1 1,\ldots,n-1 1 , … , n − 1 occurs in such a word, and for each edge j − ( j + 1 ) j-(j+1) j − ( j + 1 ) , whichever of j j j and j + 1 j+1 j + 1 occurs first determines its direction, since no other reflection changes that edge. □
Remark 4.6. Lemma 4.5 is standard in the representation theory of quivers. In this language, knitting a slice D D D of the repetition quiver Z A n − 1 \mathbb{Z}A_{n-1} Z A n − 1 to its translate D ‾ \overline{D} D sweeps out the Auslander–Reiten quiver of rep ( Ω D ) \operatorname{rep}(\Omega_D) rep ( Ω D ) . Reading the levels of its vertices in any linear extension gives a reduced word for w 0 w_0 w 0 adapted to Ω D \Omega_D Ω D . Conversely, every adapted reduced word arises this way, and the adapted words for a fixed orientation form a single commutation class. See, for example, Lusztig [34 §4.7] , Bédard [5 ] , and [2 §VII] . We included the elementary proof above to keep the discussion accessible.
Much of the literature uses sinks rather than sources. We use sources throughout, as do Garver, Patrias and Thomas [19 Thm. 1.10] , because a flip of a Dynkin path moves it to the right. The two conventions are exchanged by reversing all arrows, i.e. by replacing Ω \Omega Ω with Ω o p \Omega^{\mathrm{op}} Ω op .
Corollary 4.7. For the linear path D l i n = ( 1 , 2 , … , n − 1 ) D_{\mathrm{lin}}=(1,2,\ldots,n-1) D lin = ( 1 , 2 , … , n − 1 ) , Q D l i n = q n − 1 = q Q_{D_{\mathrm{lin}}}=q_{n-1}=q Q D lin = q n − 1 = q in Γ n \Gamma_n Γ n . Thus every knitting w 0 w_0 w 0 -word for D l i n D_{\mathrm{lin}} D lin represents evacuation in B K n \mathrm{BK}_n BK n .
Proof. The standard word
( 1 ) ( 2 , 1 ) ⋯ ( n − 1 , … , 1 ) (1)(2,1)\cdots(n-1,\ldots,1) ( 1 ) ( 2 , 1 ) ⋯ ( n − 1 , … , 1 )
is adapted to the linear orientation, so by Lemma 4.5 it is a knitting w 0 w_0 w 0 -word for D l i n D_{\mathrm{lin}} D lin . Since this is the defining word for q n − 1 q_{n-1} q n − 1 , we get Q D l i n = q n − 1 = q Q_{D_{\mathrm{lin}}}=q_{n-1}=q Q D lin = q n − 1 = q . The last statement follows because q q q acts as evacuation in B K n n \mathrm{BKn}_n BKn n .
Example 4.8. Let n = 4 n=4 n = 4 and let D = ( 1 , 2 , 3 ) D=(1,2,3) D = ( 1 , 2 , 3 ) be the linear path, so that Ω D : 1 → 2 → 3 \Omega_D:1\to2\to3 Ω D : 1 → 2 → 3 . Then D ‾ = ( 7 , 6 , 5 ) \overline{D}=(7,6,5) D = ( 7 , 6 , 5 ) , and R ( D , D ‾ ) R(D,\overline{D}) R ( D , D ) consists of the six nodes
( 1 , 1 ) , ( 3 , 1 ) , ( 5 , 1 ) ; ( 2 , 2 ) , ( 4 , 2 ) ; ( 3 , 3 ) . (1,1),(3,1),(5,1);\qquad(2,2),(4,2);\qquad(3,3). ( 1 , 1 ) , ( 3 , 1 ) , ( 5 , 1 ) ; ( 2 , 2 ) , ( 4 , 2 ) ; ( 3 , 3 ) . They are shown in Figure 4 , together with the cover relations. The only incomparable pair is { ( 3 , 1 ) , ( 3 , 3 ) } \{(3,1),(3,3)\} {( 3 , 1 ) , ( 3 , 3 )} , so there are exactly two knitting w 0 w_0 w 0 -words:
( 1 , 2 , 1 , 3 , 2 , 1 ) and ( 1 , 2 , 3 , 1 , 2 , 1 ) . (1,2,1,3,2,1)\qquad\text{and}\qquad(1,2,3,1,2,1). ( 1 , 2 , 1 , 3 , 2 , 1 ) and ( 1 , 2 , 3 , 1 , 2 , 1 ) . These are the two reduced words for w 0 ∈ S 4 w_0\in\mathcal{S}_4 w 0 ∈ S 4 adapted to 1 → 2 → 3 1\to2\to3 1 → 2 → 3 , and they differ by the commutation of 1 and 3. The first is the standard word for q 3 q_3 q 3 .
Figure 4. The region R ( D , D ‾ ) R(D,\overline{D}) R ( D , D ) for the linear path D = ( 1 , 2 , 3 ) D=(1,2,3) D = ( 1 , 2 , 3 ) , with cover relations shown as arrows. The incomparable nodes ( 3 , 1 ) (3,1) ( 3 , 1 ) and ( 3 , 3 ) (3,3) ( 3 , 3 ) may be flipped in either order, giving the two knitting w 0 w_0 w 0 -words.
Adapted words and sorting networks
A reduced word for w 0 ∈ S n w_0\in\mathcal{S}_n w 0 ∈ S n can be viewed as a sorting network on n n n wires. Following Warrington [42 ] , choose a subset of m m m wires and record the crossings among them. The resulting network is again a reduced word for w 0 ∈ S m w_0\in\mathcal{S}_m w 0 ∈ S m , called an m m m -subnetwork . If X X X is a set of reduced words in S m \mathcal{S}_m S m , we say that a reduced word is X X X -avoiding if none of its m m m -subnetworks belongs to X X X .
Warrington used this notion to formulate a combinatorial analogue of Sylvester’s four-point problem. Goodman and Pollack [20 ] associate to a configuration of n n n points in general position an allowable sequence by rotating a directed line and recording the order of the orthogonal projections of the points onto it. Each time two points exchange order we record the corresponding adjacent transposition; over a half-turn every pair exchanges order exactly once, giving a reduced word for w 0 ∈ S n w_0\in\mathcal{S}_n w 0 ∈ S n . For n = 4 n=4 n = 4 , the sixteen reduced words split into twelve of convex type and four of reentrant type [42 (1) and Fig. 1] ; the latter are
X = { 123212 , 321232 , 212321 , 232123 } . X=\{123212,\ 321232,\ 212321,\ 232123\}. X = { 123212 , 321232 , 212321 , 232123 } .
Sylvester’s problem asks for the probability that four random points are reentrant. Warrington’s combinatorial analogue asks for the probability that a uniformly random reduced word for w 0 ∈ S n w_0\in\mathcal{S}_n w 0 ∈ S n , together with a uniformly random 4-subset of its wires, induces a subnetwork in X X X . He conjectured that this probability is 1 / 4 1/4 1/4 for every n ≥ 4 n\ge4 n ≥ 4 , and Angel and Holroyd [1 ] proved the conjecture. For n ≥ 5 n\ge5 n ≥ 5 , not every reduced word comes from a point configuration under the Goodman–Pollack construction. Thus “convex” and “reentrant” refer here to the corresponding combinatorial types of reduced words, not necessarily to actual point configurations.
Meng [35 §6.3] studied X X X -avoiding reduced words for w 0 ∈ S n w_0 \in\mathfrak{S}_n w 0 ∈ S n and asked for their number. The following theorem connects this question with adapted words.
Theorem 4.9. Every knitting w 0 w_0 w 0 -word is X X X -avoiding.
Proof. Write w = ( i 1 , … , i N ) ∈ K ( D ) w=(i_1,\ldots,i_N)\in\mathcal{K}(D) w = ( i 1 , … , i N ) ∈ K ( D ) . By Lemma 4.2(b), the letter j j j occurs
m j = d ˉ j − d j 2 = n + d n − j − d j 2 m_j=\frac{\bar{d}_j-d_j}{2}=\frac{n+d_{n-j}-d_j}{2} m j = 2 d ˉ j − d j = 2 n + d n − j − d j
times. Hence
∑ r = 1 N ( i r − 1 ) ( n − i r − 1 ) = n 2 ∑ j = 1 n − 1 ( j − 1 ) ( n − j − 1 ) + 1 2 ∑ j = 1 n − 1 ( j − 1 ) ( n − j − 1 ) ( d n − j − d j ) . \sum_{r=1}^{N}(i_r-1)(n-i_r-1)=\frac{n}{2}\sum_{j=1}^{n-1}(j-1)(n-j-1)+\frac{1}{2}\sum_{j=1}^{n-1}(j-1)(n-j-1)(d_{n-j}-d_j). r = 1 ∑ N ( i r − 1 ) ( n − i r − 1 ) = 2 n j = 1 ∑ n − 1 ( j − 1 ) ( n − j − 1 ) + 2 1 j = 1 ∑ n − 1 ( j − 1 ) ( n − j − 1 ) ( d n − j − d j ) .
The second sum is zero by the symmetry j ↔ n − j j\leftrightarrow n-j j ↔ n − j . One can check that
∑ j = 1 n − 1 ( j − 1 ) ( n − j − 1 ) = ( n − 1 ) ( n − 2 ) ( n − 3 ) 6 , \sum_{j=1}^{n-1}(j-1)(n-j-1)=\frac{(n-1)(n-2)(n-3)}{6}, j = 1 ∑ n − 1 ( j − 1 ) ( n − j − 1 ) = 6 ( n − 1 ) ( n − 2 ) ( n − 3 ) ,
so
∑ r = 1 N ( i r − 1 ) ( n − i r − 1 ) = n ( n − 1 ) ( n − 2 ) ( n − 3 ) 12 = 2 ( n 4 ) . \sum_{r=1}^{N}(i_r-1)(n-i_r-1)=\frac{n(n-1)(n-2)(n-3)}{12}=2\binom{n}{4}. r = 1 ∑ N ( i r − 1 ) ( n − i r − 1 ) = 12 n ( n − 1 ) ( n − 2 ) ( n − 3 ) = 2 ( 4 n ) .
By [35 Prop. 5.3] , the number of X X X -subnetworks of w w w is
∑ r = 1 N ( i r − 1 ) ( n − i r − 1 ) − 2 ( n 4 ) , \sum_{r=1}^{N}(i_r-1)(n-i_r-1)-2\binom{n}{4}, r = 1 ∑ N ( i r − 1 ) ( n − i r − 1 ) − 2 ( 4 n ) ,
and is therefore zero.
:::
Conjecture 4.10. A reduced word for w 0 ∈ S n w_0\in\mathfrak{S}_n w 0 ∈ S n is X X X -avoiding if and only if it is adapted to an orientation of A n − 1 A_{n-1} A n − 1 . Equivalently, a reduced word for w 0 w_0 w 0 is adapted to an orientation of A n − 1 A_{n-1} A n − 1 if and only if
∑ r ( i r − 1 ) ( n − i r − 1 ) = 2 ( n 4 ) . \sum_r(i_r-1)(n-i_r-1)=2\binom{n}{4}. r ∑ ( i r − 1 ) ( n − i r − 1 ) = 2 ( 4 n ) .
By Meng’s formula, the displayed equality is exactly the condition of being X X X -avoiding, so the content of the conjecture is that this numerical condition forces adaptedness. We verified the conjecture by computer for n = 4 , 5 , 6 n=4,5,6 n = 4 , 5 , 6 : in these cases the X X X -avoiding reduced words are exactly the knitting w 0 w_0 w 0 -words. For n = 7 n=7 n = 7 the number of adapted reduced words is 68641152, the number of X X X -avoiding reduced words reported by Warrington; by Theorem 4.9 the two sets coincide, so the conjecture holds for n ≤ 7 n\leq7 n ≤ 7 . We have no conceptual explanation for why adapted reduced words should be governed by a convexity condition on four-point configurations.
Adapted relations
We now show that, modulo the cactus relations, the adapted relations are equivalent to the Dynkin-reversal relations of § 3 . The key is the following simple observation describing how the knitting element Q D Q_D Q D changes under a flip.
Lemma 5.1 (Flip Lemma). Let j j j be a source of a Dynkin path D D D , and let D ′ D' D ′ be the flip of D D D at j j j . Then
n − j n-j n − j is a source of D ˉ \bar{D} D ˉ , and D ˉ ′ \bar{D}' D ˉ ′ is the flip of D ˉ \bar{D} D ˉ at n − j n-j n − j ;
in Γ n \Gamma_n Γ n ,
Q D ′ = t j Q D t n − j . Q_{D'}=t_jQ_Dt_{n-j}. Q D ′ = t j Q D t n − j . :::`
Proof. Part (1) is immediate from
d ˉ k = d n − k + n . \bar{d}_k=d_{n-k}+n. d ˉ k = d n − k + n .
a source at j j j becomes a source at n − j n-j n − j , and flipping there replaces d ˉ n − j = d j + n \bar{d}_{n-j}=d_j+n d ˉ n − j = d j + n by d j + n + 2 d_j+n+2 d j + n + 2 , which is exactly D ˉ ′ \bar{D}' D ˉ ′ .
For (2), note that D ′ ≤ D ˉ D'\leq\bar{D} D ′ ≤ D ˉ , since the flip raises one height by 2 and d ˉ j − d j ≥ 2 \bar{d}_j-d_j\geq2 d ˉ j − d j ≥ 2 . Since j j j is a source, we can write a knitting word for D D D as ( j , x ) (j,x) ( j , x ) with x ∈ W ( D ′ , D ˉ ) x\in W(D',\bar{D}) x ∈ W ( D ′ , D ˉ ) . Thus Q D = t j t x Q_D=t_jt_x Q D = t j t x , so t x = t j Q D t_x=t_jQ_D t x = t j Q D . By (1), ( n − j ) ∈ W ( D ˉ , D ˉ ′ ) (n-j)\in W(\bar{D},\bar{D}') ( n − j ) ∈ W ( D ˉ , D ˉ ′ ) , and therefore
Q D ′ = t x t n − j = t j Q D t n − j . Q_{D'}=t_xt_{n-j}=t_jQ_Dt_{n-j}. Q D ′ = t x t n − j = t j Q D t n − j .
Proposition 5.2. Let u ∈ W ( D l i n , D ) u\in W(D_{\mathrm{lin}},D) u ∈ W ( D lin , D ) . Then Q D = t rev ( u ) q t α ( u ) Q_D=t_{\operatorname{rev}(u)}q\,t_{\alpha(u)} Q D = t rev ( u ) q t α ( u ) in Γ n \Gamma_n Γ n .
Proof. This follows by induction on the length of u u u from the Flip Lemma, with Corollary 4.7 as the base case.
Definition 5.3. The adapted relations are the relations Q D = q Q_D=q Q D = q for every Dynkin path D D D . Equivalently, they are the relations t w = q t_w=q t w = q for every reduced word w w w for w 0 w_0 w 0 adapted to an orientation of A n − 1 A_{n-1} A n − 1 .
Theorem 5.4. In C n C_n C n , the Dynkin-reversal relations are equivalent to the adapted relations . Thus,
R C n = C n / ⟨ ⟨ Q D = q for all Dynkin paths D ⟩ ⟩ . \mathcal{R}C_n=C_n\mathbin{/}\left\langle\!\left\langle Q_D=q\text{ for all Dynkin paths }D\right\rangle\!\right\rangle. R C n = C n / ⟨ ⟨ Q D = q for all Dynkin paths D ⟩ ⟩ . Proof. Assume first the Dynkin-reversal relations q t i q = t n − i q\,t_i\,q=t_{n-i} q t i q = t n − i , equivalently t j q = q t n − j t_jq=qt_{n-j} t j q = q t n − j . If D ′ D' D ′ is obtained from D D D by flipping a source j j j and Q D = q Q_D=q Q D = q , then the Flip Lemma gives
Q D ′ = t j q t n − j = q . Q_{D'}=t_jqt_{n-j}=q. Q D ′ = t j q t n − j = q . Since Q D l i n = q Q_{D_{\mathrm{lin}}}=q Q D lin = q and every orientation can be reached from the linear one by flips, we get Q D = q Q_D=q Q D = q for every D D D .
In the other direction, assume Q D = q Q_D=q Q D = q for every Dynkin path D D D . Choose D D D with j j j as a source, and let D ′ D' D ′ be its flip at j j j . Then Q D = Q D ′ = q Q_D=Q_{D'}=q Q D = Q D ′ = q , so the Flip Lemma gives q = t j q t n − j q=t_jqt_{n-j} q = t j q t n − j , or equivalently q t n − j q = t j qt_{n-j}q=t_j q t n − j q = t j . As j j j varies, these are exactly the Dynkin-reversal relations.
Theorem 5.5. Every reduced word for w 0 w_0 w 0 adapted to an orientation of A n − 1 A_{n-1} A n − 1 represents evacuation in B K R n \mathrm{BKR}_n BKR n . Equivalently, Q D = q Q_D=q Q D = q in B K R n \mathrm{BKR}_n BKR n for every Dynkin path D D D .
Proof. The Dynkin-reversal relations hold in B K R n \mathrm{BKR}_n BKR n by Theorem 3.4 , so the adapted relations hold by Theorem 5.4 .
Promotion
We now relate the adapted relations to promotion. The key is the following identity, which already holds in the commutation group.
Recall p = p n − 1 = t n − 1 t n − 2 ⋯ t 1 p=p_{n-1}=t_{n-1}t_{n-2}\cdots t_1 p = p n − 1 = t n − 1 t n − 2 ⋯ t 1 , so that p − 1 = t 1 t 2 ⋯ t n − 1 p^{-1}=t_1t_2\cdots t_{n-1} p − 1 = t 1 t 2 ⋯ t n − 1 .
Theorem 6.1. In Γ n \Gamma_n Γ n , and hence in C n C_n C n , B K n \mathrm{BK}_n BK n , and B K R n \mathrm{BKR}_n BKR n ,
q n − 1 α ( q n − 1 ) = ( t 1 t 2 ⋯ t n − 1 ) n = p − n . q_{n-1}\alpha(q_{n-1})=(t_1t_2\cdots t_{n-1})^n=p^{-n}. q n − 1 α ( q n − 1 ) = ( t 1 t 2 ⋯ t n − 1 ) n = p − n . Proof. We compute
Q ( D l i n , D l i n + ( 2 n , 0 ) ) Q(D_{\mathrm{lin}},D_{\mathrm{lin}}+(2n,0)) Q ( D lin , D lin + ( 2 n , 0 )) in two ways.
First, ( 1 , 2 , … , n − 1 ) (1,2,\ldots,n-1) ( 1 , 2 , … , n − 1 ) knits D l i n D_{\mathrm{lin}} D lin to D l i n + ( 2 , 0 ) D_{\mathrm{lin}}+(2,0) D lin + ( 2 , 0 ) : after flipping 1 , 2 , … , j 1,2,\ldots,j 1 , 2 , … , j , the next level j + 1 j+1 j + 1 is a source, and after all n − 1 n-1 n − 1 flips every height has increased by 2 2 2 . Repeating this n n n times gives
Q ( D l i n , D l i n + ( 2 n , 0 ) ) = ( t 1 t 2 ⋯ t n − 1 ) n . Q(D_{\mathrm{lin}},D_{\mathrm{lin}}+(2n,0))=(t_1t_2\cdots t_{n-1})^n. Q ( D lin , D lin + ( 2 n , 0 )) = ( t 1 t 2 ⋯ t n − 1 ) n .
On the other hand,
D l i n + ( 2 n , 0 ) = D ‾ ‾ l i n . D_{\mathrm{lin}}+(2n,0)=\overline{\overline{D}}_{\mathrm{lin}}. D lin + ( 2 n , 0 ) = D lin .
The map ( i , j ) ↦ ( i + n , n − j ) (i,j)\mapsto(i+n,n-j) ( i , j ) ↦ ( i + n , n − j ) sends knitting words from D D D to D ′ D' D ′ to knitting words from D ‾ \overline{D} D to D ‾ ′ \overline{D}' D ′ , replacing every letter j j j by n − j n-j n − j . Therefore Q ( D ‾ , D ‾ ′ ) = α ( Q ( D , D ′ ) ) Q(\overline{D},\overline{D}')=\alpha(Q(D,D')) Q ( D , D ′ ) = α ( Q ( D , D ′ )) . Using the composition rule,
Q ( D l i n , D ‾ ‾ l i n ) = Q ( D l i n , D ‾ l i n ) Q ( D ‾ l i n , D ‾ ‾ l i n ) = q α ( q ) . Q(D_{\mathrm{lin}},\overline{\overline{D}}_{\mathrm{lin}})=Q(D_{\mathrm{lin}},\overline{D}_{\mathrm{lin}})Q(\overline{D}_{\mathrm{lin}},\overline{\overline{D}}_{\mathrm{lin}})=q\alpha(q). Q ( D lin , D lin ) = Q ( D lin , D lin ) Q ( D lin , D lin ) = q α ( q ) .
Comparing the two expressions gives the result. □
Corollary 6.2 (Haiman’s promotion relation). The adapted relations imply
In particular, promotion has order dividing n n n on every rectangular shape. For standard rectangular tableaux this is Haiman’s theorem [23 Thm. 4.4] ; the semistandard statement follows by standardisation, as observed for example in [4 Prop. 3.2] . The order is exactly n n n in B K R n \mathrm{BKR}_n BKR n .
Proof. For the anti-linear path D ‾ l i n \overline{D}_{\mathrm{lin}} D lin , we have Q D ‾ l i n = α ( q ) Q_{\overline{D}_{\mathrm{lin}}}=\alpha(q) Q D lin = α ( q ) . Thus its adapted relation is α ( q ) = q \alpha(q)=q α ( q ) = q . By Theorem 6.1 ,
p − n = q α ( q ) = q 2 = 1. p^{-n}=q\alpha(q)=q^2=1. p − n = q α ( q ) = q 2 = 1.
Therefore p n = 1 p^n=1 p n = 1 .
The order is exactly n n n in B K R n \mathrm{BKR}_n BKR n , since already on the 1 × 1 1\times1 1 × 1 rectangle, p p p acts as an n n n -cycle. □
Remark 6.3. For standard rectangular tableaux, Haiman’s proof uses dual equivalence [23 Thm. 4.4] . Several other approaches are known. Bandlow, Schilling and Thiéry [4 Prop. 3.2] use crystal operations and jeu de taquin ; Rhoades [38 ] proves the result representation-theoretically as part of his cyclic-sieving theorem for rectangular tableaux. Einstein and Propp [15 ] interpret semistandard promotion through piecewise-linear rowmotion on a product of two chains and deduce its periodicity from the birational rowmotion results of Grinberg and Roby [21 ] . The first author [17 ] realizes promotion as the tropicalization of a cyclic shift on the Grassmannian, and Westbury [45 ] uses the cactus-group action on seminormal bases for Hecke algebras.
One might ask whether the other way around, Haiman’s relation p n = 1 p^n=1 p n = 1 implies the Dynkin-reversal relations. It does not.
Proposition 6.4. For every n ≥ 4 n\geq4 n ≥ 4 , the promotion relation p n = 1 p^n=1 p n = 1 is strictly weaker than the Dynkin-reversal relations. Equivalently, the natural map
C n / ⟨ p n ⟩ ↠ R C n C_n/\langle p^n\rangle\twoheadrightarrow RC_n C n / ⟨ p n ⟩ ↠ R C n is not an isomorphism.
Proof. Any choice of signs χ ( t i ) ∈ { ± 1 } \chi(t_i)\in\{\pm1\} χ ( t i ) ∈ { ± 1 } defines a homomorphism χ : C n → { ± 1 } \chi:C_n\to\{\pm1\} χ : C n → { ± 1 } : the signs commute and square to 1 1 1 , so all the relations of Theorem 2.5 are satisfied.
The Dynkin-reversal relations would force χ ( t i ) = χ ( t n − i ) \chi(t_i)=\chi(t_{n-i}) χ ( t i ) = χ ( t n − i ) . We construct a character satisfying p n = 1 p^n=1 p n = 1 but not this relation.
If n n n is even, set χ ( t 1 ) = − 1 \chi(t_1)=-1 χ ( t 1 ) = − 1 and χ ( t i ) = 1 \chi(t_i)=1 χ ( t i ) = 1 for i > 1 i>1 i > 1 . Then χ ( p n ) = 1 \chi(p^n)=1 χ ( p n ) = 1 , but χ ( t 1 ) ≠ χ ( t n − 1 ) \chi(t_1)\ne\chi(t_{n-1}) χ ( t 1 ) = χ ( t n − 1 ) .
If n ≥ 5 n\geq5 n ≥ 5 is odd, set χ ( t 1 ) = χ ( t 2 ) = − 1 \chi(t_1)=\chi(t_2)=-1 χ ( t 1 ) = χ ( t 2 ) = − 1 and χ ( t i ) = 1 \chi(t_i)=1 χ ( t i ) = 1 otherwise. Then χ ( p ) = 1 \chi(p)=1 χ ( p ) = 1 , so again χ ( p n ) = 1 \chi(p^n)=1 χ ( p n ) = 1 , while χ ( t 1 ) ≠ χ ( t n − 1 ) \chi(t_1)\ne\chi(t_{n-1}) χ ( t 1 ) = χ ( t n − 1 ) .
The case n = 3 n=3 n = 3 is an exception: the promotion relation p 3 = 1 p^3=1 p 3 = 1 is equivalent to the Dynkin-reversal relation: both are equivalent to t 1 t 2 t 1 = t 2 t 1 t 2 t_1t_2t_1=t_2t_1t_2 t 1 t 2 t 1 = t 2 t 1 t 2 . □
The big cactus group
We identify the quotient of Definition 3.9 with the big cactus group. The proof uses the geometry of real moduli spaces and a generation theorem for a parabolic subgroup of the cactus group.
Real Deligne–Mumford spaces and equivariant fundamental groups
For r ≥ 3 r \ge3 r ≥ 3 , let M ‾ 0 , r ( R ) \overline{M}_{0,r}(\mathbb{R}) M 0 , r ( R ) be the real Deligne–Mumford space of stable genus-zero curves with r r r labeled real marked points. A point of M ‾ 0 , r ( R ) \overline{M}_{0,r}(\mathbb{R}) M 0 , r ( R ) is represented by a connected nodal curve whose irreducible components are copies of P 1 ( R ) \mathbb{P}^{1}(\mathbb{R}) P 1 ( R ) and such that every component contains at least three special points, meaning marked points or nodes. Two such marked curves represent the same point if they are related by projective automorphisms on their components.
The boundary stratification is described by the dual tree of the stable curve: vertices correspond to irreducible components, internal edges to nodes, and leaves to marked points. The codimension of a stratum is the number of internal edges of its dual tree. In particular, the zero-dimensional strata are represented by trivalent trees; we will use special zero-dimensional points whose dual trees are caterpillars.
There is one feature of the real locus that will be used repeatedly below. On every irreducible component, the order of the special points is intrinsic only up to reversal. Indeed, after choosing one special point as ∞ \infty ∞ , the remaining automorphisms have the form
x ⟼ a x + b , a ∈ R × x \longmapsto ax+b,\qquad a \in\mathbb{R}^{\times} x ⟼ a x + b , a ∈ R ×
and a < 0 a<0 a < 0 reverses the order. Equivalently, after normalizing the translation, the residual ambiguity includes the sign change x ↦ − x x \mapsto-x x ↦ − x . Thus a real boundary stratum may be encoded by a planar dual tree, but at each internal vertex the cyclic order of the incident half-edges is taken modulo reversal, independently at different vertices.
Figure 5. A boundary point of M ‾ 0 , n ( R ) \overline{M}_{0,n}(\mathbb{R}) M 0 , n ( R ) and its dual tree. Each vertex of the tree represents an irreducible component, each internal edge represents a node, and the labeled leaves represent marked points.
The connected components of the open stratum M 0 , r ( R ) M_{0,r}(\mathbb{R}) M 0 , r ( R ) are therefore indexed by cyclic orders of the r r r labels, modulo reversal. If one marking is distinguished and denoted by ∞ \infty ∞ , a chamber is equivalently described by a linear order of the remaining labels, again modulo reversal. The closure of every chamber is an associahedron, hence is contractible; see [13 ] and the references therein.
For a finite group G G G acting on a connected space X X X , we use the equivariant fundamental group π 1 G ( X , x ) \pi_{1}^{G}(X,x) π 1 G ( X , x ) . An element is represented by a pair ( g , γ ) (g,\gamma) ( g , γ ) , where g ∈ G g \in G g ∈ G and γ \gamma γ is a path from x x x to g x gx g x , modulo homotopy relative to the endpoints. There is an exact sequence
1 ⟶ π 1 ( X , x ) ⟶ π 1 G ( X , x ) ⟶ G ⟶ 1. (4) 1 \longrightarrow\pi_{1}(X,x) \longrightarrow\pi_{1}^{G}(X,x) \longrightarrow G \longrightarrow1. \tag*{(4)} 1 ⟶ π 1 ( X , x ) ⟶ π 1 G ( X , x ) ⟶ G ⟶ 1. (4)
Figure 6. On an irreducible real component, the special points are considered modulo PGL 2 ( R ) \operatorname{PGL}_2(\mathbb{R}) PGL 2 ( R ) . Consequently, their cyclic order is intrinsic only up to reversal.
In particular, if g g g fixes x x x , then ( g , const x ) (g,\operatorname{const}_x) ( g , const x ) is an element of π 1 S n ( X , x ) \pi_1^{\mathfrak{S}_n}(X,x) π 1 S n ( X , x ) ; its underlying path is constant although its permutation part is nontrivial.
We use the topological realization of the cactus group [13 , 27 ] :
C n = π 1 S n ( M ‾ 0 , n + 1 ( R ) ) , C_n=\pi_1^{\mathfrak{S}_n}\left(\overline{M}_{0,n+1}(\mathbb{R})\right), C n = π 1 S n ( M 0 , n + 1 ( R ) ) ,
where the marked points are 1 , … , n , ∞ 1,\ldots,n,\infty 1 , … , n , ∞ and S n \mathfrak{S}_n S n fixes ∞ \infty ∞ . Under this identification the permutation of q [ a , b ] q_{[a,b]} q [ a , b ] is the reversal w [ a , b ] w_{[a,b]} w [ a , b ] of the interval [ a , b ] [a,b] [ a , b ] . Henriques’s big cactus group is
J n ′ = π 1 S n ( M ‾ 0 , n ( R ) ) , \mathfrak{J}'_n=\pi_1^{\mathfrak{S}_n}\left(\overline{M}_{0,n}(\mathbb{R})\right), J n ′ = π 1 S n ( M 0 , n ( R ) ) ,
where all n n n markings are permuted [26 ] .
We will also use the mirabolic cactus group studied by Gurenkova, Halacheva and the third author [22 ] :
M C m = π 1 S m ( M ‾ 0 , m + 2 ( R ) ) ⊂ C m + 1 . \mathrm{MC}_m=\pi_1^{\mathfrak{S}_m}\left(\overline{M}_{0,m+2}(\mathbb{R})\right)\subset C_{m+1}. MC m = π 1 S m ( M 0 , m + 2 ( R ) ) ⊂ C m + 1 .
Here the markings are 0 , 1 , … , m , ∞ 0,1,\ldots,m,\infty 0 , 1 , … , m , ∞ ; S m \mathfrak{S}_m S m fixes 0 0 0 and ∞ \infty ∞ and permutes 1 , … , m 1,\ldots,m 1 , … , m . Thus M C m \mathrm{MC}_m MC m is the full preimage of S m ⊂ S m + 1 \mathfrak{S}_m\subset\mathfrak{S}_{m+1} S m ⊂ S m + 1 under the permutation map. In particular, the subgroup fixing the first marking in C n C_n C n is a relabeled copy of M C n − 1 \mathrm{MC}_{n-1} MC n − 1 .
Caterpillar paths and forgetting the root.
Fix n ≥ 3 n\ge3 n ≥ 3 and write
X n = M ‾ 0 , n + 1 ( R ) , Y n = M ‾ 0 , n ( R ) . X_n=\overline{M}_{0,n+1}(\mathbb{R}),\qquad Y_n=\overline{M}_{0,n}(\mathbb{R}). X n = M 0 , n + 1 ( R ) , Y n = M 0 , n ( R ) .
We use the caterpillar point x ∈ X n x\in X_n x ∈ X n whose two end components carry 1 , 2 1,2 1 , 2 and n , ∞ n,\infty n , ∞ , respectively, and whose remaining markings occur in increasing order along the spine. Let D ⊂ M 0 , n + 1 ( R ) D\subset M_{0,n+1}(\mathbb{R}) D ⊂ M 0 , n + 1 ( R ) be the chamber with cyclic order 1 , 2 , … , n , ∞ 1,2,\ldots,n,\infty 1 , 2 , … , n , ∞ . Since D ‾ \overline{D} D is contractible, there is a preferred homotopy class of paths γ ⊂ D \gamma\subset D γ ⊂ D from x x x to a generic point of D D D . This identifies the equivariant fundamental group based at the caterpillar orbit with the usual cactus group C n C_n C n .
The standard topological representative of the cactus generator q [ a , b ] q_{[a,b]} q [ a , b ] may be seen directly on the open chambers. Starting from a configuration in D D D , move the consecutive block a , a + 1 , … , b a,a+1,\ldots,b a , a + 1 , … , b onto a separate component. At the resulting boundary point the order on that new real component is defined only up to reversal. Leaving the boundary in the other adjacent chamber therefore reverses the block. The path ends at the configuration obtained by w [ a , b ] w_{[a,b]} w [ a , b ] . Conjugating this path by γ \gamma γ gives a representative based at x x x .
There is a second, boundary-based description which is especially convenient for us. Let
t 1 , … , t n − 1 t_1,\ldots,t_{n-1} t 1 , … , t n − 1
be the elementary caterpillar paths of [22 ] . For 2 ≤ i < n 2\le i<n 2 ≤ i < n , the path t i t_i t i is the shortest edge of the 1 1 1 -skeleton joining x x x to the adjacent caterpillar obtained by exchanging the positions of i i i and i + 1 i+1 i + 1 . Locally, only the four branches
L = { 1 , … , i − 1 } , i , i + 1 , R = { i + 2 , … , n , ∞ } L=\{1,\ldots,i-1\},\qquad i,\qquad i+1,\qquad R=\{i+2,\ldots,n,\infty\} L = { 1 , … , i − 1 } , i , i + 1 , R = { i + 2 , … , n , ∞ }
move. The corresponding one-dimensional stratum is a copy of M ‾ 0 , 4 ( R ) \overline{M}_{0,4}(\mathbb{R}) M 0 , 4 ( R ) : t i t_i t i is the arc joining the two displayed splittings and avoiding the third zero-dimensional point. For i = 1 i=1 i = 1 , the first two markings lie on the same end component, so the underlying path is constant and the equivariant permutation is ( 1 , 2 ) (1,2) ( 1 , 2 ) .
The comparison theorem of [22 ] identifies these geometric paths with the elements t i t_i t i of Theorem 2.5 . Moreover, if
p i = t i ⋯ t 1 , q i = p 1 ⋯ p i , p_i=t_i\cdots t_1,\qquad q_i=p_1\cdots p_i, p i = t i ⋯ t 1 , q i = p 1 ⋯ p i ,
then the path
q [ a , b ] = q b − 1 q b − a q b − 1 q_{[a,b]}=q_{b-1}q_{b-a}q_{b-1} q [ a , b ] = q b − 1 q b − a q b − 1
is homotopic to the standard interval-reversal path of Figure 7 . We will use one further result from the same paper. If π : C n → S n \pi:C_n\to\mathfrak{S}_n π : C n → S n is the permutation map, then
Figure 7. A topological representative of the generator q [ a , b ] q_{[a,b]} q [ a , b ] . First, the consecutive marked points a , … , b a,\ldots,b a , … , b bubble off onto a new component. Since the real projective line is considered modulo PGL 2 ( R ) \operatorname{PGL}_{2}(\mathbb{R}) PGL 2 ( R ) , the order on that component is defined only up to reversal. Reversing the bubbled component and smoothing the node produces the order obtained from the original one by reversing the interval [ a , b ] [a,b] [ a , b ] .
π − 1 ( Stab S n ( 1 ) ) = ⟨ q [ a , b ] ( 2 ≤ a < b ≤ n ) , t 2 , … , t n − 1 ⟩ (5) \pi^{-1}\bigl(\operatorname{Stab}_{\mathfrak{S}_n}(1)\bigr)=\left\langle q_{[a,b]}\ (2\leq a<b\leq n),\ t_2,\ldots,t_{n-1}\right\rangle \tag*{(5)} π − 1 ( Stab S n ( 1 ) ) = ⟨ q [ a , b ] ( 2 ≤ a < b ≤ n ) , t 2 , … , t n − 1 ⟩ (5)
This is the mirabolic generation theorem, with the labels shifted by one.
Now let
f : X n ⟶ Y n f:X_n\longrightarrow Y_n f : X n ⟶ Y n
forget ∞ \infty ∞ and stabilize, and put y = f ( x ) y=f(x) y = f ( x ) . The map is S n \mathfrak{S}_n S n -equivariant, so it induces
f ∗ : C n ⟶ J n ′ . (6) f_*:C_n\longrightarrow J'_n. \tag*{(6)} f ∗ : C n ⟶ J n ′ . (6)
For n ≥ 4 n\geq4 n ≥ 4 , forgetting ∞ \infty ∞ contracts the last component of the rooted caterpillar. Thus the two end components of y y y carry the pairs 1 , 2 1,2 1 , 2 and n − 1 , n n-1,n n − 1 , n . The resulting unrooted caterpillar has a reflection symmetry which, together with the relabeling a ↦ n + 1 − a a\mapsto n+1-a a ↦ n + 1 − a , fixes the point y y y .
Figure 8. A topological representative of the elementary path t i t_i t i . Here L L L and R R R denote the unchanged left and right parts of the caterpillar curve. The path t i t_i t i is realized by moving the marked point i i i to the lower side of its component, smoothing the node between the two adjacent components, passing the two marked points through the smoothed component, and then splitting the component again. The resulting caterpillar has the same shape, with the order of i i i and i + 1 i+1 i + 1 exchanged.
Figure 9. Forgetting the root marked point ∞ \infty ∞ . Top: if a terminal component carries ∞ \infty ∞ and one additional marked point p p p , then after forgetting ∞ \infty ∞ that component becomes unstable and is contracted. Bottom: for the rooted caterpillar x ∈ X n x \in X_n x ∈ X n , the last component carries n n n and ∞ \infty ∞ ; forgetting ∞ \infty ∞ contracts this component and produces the ordinary caterpillar y = f ( x ) ∈ Y n y=f(x) \in Y_n y = f ( x ) ∈ Y n .
Recall q = q [ 1 , n ] = q n − 1 q=q_{[1,n]}=q_{n-1} q = q [ 1 , n ] = q n − 1 , and set
t i R = q [ i + 1 , n ] q [ i , n ] q [ i + 1 , n ] q [ i + 2 , n ] ( 1 ≤ i < n ) , (7) t_i^{R}=q_{[i+1,n]}q_{[i,n]}q_{[i+1,n]}q_{[i+2,n]}\qquad(1\le i<n), \tag*{(7)} t i R = q [ i + 1 , n ] q [ i , n ] q [ i + 1 , n ] q [ i + 2 , n ] ( 1 ≤ i < n ) , (7)
with a singleton or empty interval interpreted as the identity. The cactus relations and Theorem 2.5 give
q t i q = t n − i R . (8) qt_iq=t_{n-i}^{R}. \tag*{(8)} q t i q = t n − i R . (8)
Lemma 7.1. For 1 ≤ i < n 1 \le i < n 1 ≤ i < n ,
f ∗ ( q t i q ) = f ∗ ( t n − i ) ; equivalently, f ∗ ( t i ) = f ∗ ( t i R ) . f_*(q t_i q) = f_*(t_{n-i}); \qquad\text{equivalently,} \qquad f_*(t_i) = f_*(t_i^{\mathrm{R}}). f ∗ ( q t i q ) = f ∗ ( t n − i ) ; equivalently, f ∗ ( t i ) = f ∗ ( t i R ) . Proof. The crucial observation is that the unrooted caterpillar y = f ( x ) y=f(x) y = f ( x ) is fixed by the longest permutation w 0 ( a ) = n + 1 − a w_0(a)=n+1-a w 0 ( a ) = n + 1 − a . Indeed, after forgetting ∞ \infty ∞ , the last component of x x x becomes unstable and is contracted. The resulting caterpillar is symmetric under reversal of the chain: reflecting the chain and reversing the order of special points on each component gives an isomorphism of the corresponding marked curves, since the order on each real component is defined only up to reversal. Thus
Let r = ( w 0 , const y ) ∈ J n ′ r=(w_0,\operatorname{const}_y)\in J'_n r = ( w 0 , const y ) ∈ J n ′ be the constant-path lift of w 0 w_0 w 0 .
Since w 0 w_0 w 0 reverses the cyclic order 1 , … , n , ∞ 1,\ldots,n,\infty 1 , … , n , ∞ , it maps the chamber D D D to itself, and the full cactus reversal q = q [ 1 , n ] q=q_{[1,n]} q = q [ 1 , n ] is represented by a path from x x x to w 0 x w_0x w 0 x contained in D ‾ \overline{D} D . After forgetting ∞ \infty ∞ , this path becomes a loop at y y y contained in the closure of the corresponding chamber of M ‾ 0 , n ( R ) \overline{M}_{0,n}(\mathbb{R}) M 0 , n ( R ) . Since this chamber closure is contractible, the loop is null-homotopic. Hence
f ∗ ( q ) = r . f_*(q)=r. f ∗ ( q ) = r . For 2 ≤ i ≤ n − 2 2\le i\le n-2 2 ≤ i ≤ n − 2 , the projected path f ( t i ) f(t_i) f ( t i ) varies only the four-pointed component with branches
L = { 1 , … , i − 1 } , i , i + 1 , R = { i + 2 , … , n } . L=\{1,\ldots,i-1\},\qquad i,\qquad i+1,\qquad R=\{i+2,\ldots,n\}. L = { 1 , … , i − 1 } , i , i + 1 , R = { i + 2 , … , n } . It joins the splittings ( L , i ) ∣ ( i + 1 , R ) (L,i)\mid(i+1,R) ( L , i ) ∣ ( i + 1 , R ) and ( L , i + 1 ) ∣ ( i , R ) (L,i+1)\mid(i,R) ( L , i + 1 ) ∣ ( i , R ) along the arc of M ‾ 0 , 4 ( R ) \overline{M}_{0,4}(\mathbb{R}) M 0 , 4 ( R ) avoiding ( L , R ) ∣ ( i , i + 1 ) (L,R)\mid(i,i+1) ( L , R ) ∣ ( i , i + 1 ) . Reflection exchanges the two outer branches and, after relabeling by w 0 w_0 w 0 , carries this arc to the corresponding arc for n − i n-i n − i . Therefore
r f ∗ ( t i ) r = f ∗ ( t n − i ) . r f_*(t_i) r=f_*(t_{n-i}). r f ∗ ( t i ) r = f ∗ ( t n − i ) . At the two ends, f ∗ ( t 1 ) f_*(t_1) f ∗ ( t 1 ) is the constant-path transposition of the first cherry, while f ( t n − 1 ) f(t_{n-1}) f ( t n − 1 ) becomes constant after the component carrying n , ∞ n,\infty n , ∞ is contracted; its equivariant permutation is the transposition of the last cherry. Reflection exchanges these two cases. For n = 3 n=3 n = 3 the same statement is simply the corresponding identity in S 3 \mathfrak{S}_3 S 3 . Combining this with f ∗ ( q ) = r f_*(q)=r f ∗ ( q ) = r proves
f ∗ ( q t i q ) = f ∗ ( t n − i ) f_*(q t_i q)=f_*(t_{n-i}) f ∗ ( q t i q ) = f ∗ ( t n − i ) for every 1 ≤ i < n 1\le i<n 1 ≤ i < n . The equivalent form follows from (8) .
The defining relations.
Theorem 7.2. For n ≥ 3 n\ge3 n ≥ 3 , forgetting ∞ \infty ∞ induces an isomorphism
R C n → ∼ J n ′ . \mathcal{R}C_n \xrightarrow{\sim} J'_n. R C n ∼ J n ′ . Equivalently, the map (6) is surjective and
ker f ∗ = ⟨ q t i q t n − i : 1 ≤ i < n ⟩ . (9) \ker f_*=\left\langle q t_i q t_{n-i}:1\le i<n\right\rangle. \tag*{(9)} ker f ∗ = ⟨ q t i q t n − i : 1 ≤ i < n ⟩ . (9) Proof. There is a section σ n : Y n → X n \sigma_n:Y_n\to X_n σ n : Y n → X n of f f f : attach at the marking n n n a three-pointed component carrying n , ∞ n,\infty n , ∞ , and the node. It is equivariant for S n − 1 = Stab S n ( n ) \mathfrak{S}_{n-1}=\operatorname{Stab}_{\mathfrak{S}_n}(n) S n − 1 = Stab S n ( n ) , and σ n ( y ) = x \sigma_n(y)=x σ n ( y ) = x . Set
H = ⟨ q [ a , b ] : 1 ≤ a < b ≤ n − 1 ⟩ ⊂ C n , K = π 1 S n − 1 ( Y n , y ) ⊂ J n ′ . H=\left\langle q_{[a,b]}:1\le a<b\le n-1\right\rangle\subset C_n,\qquad K=\pi_1^{\mathfrak{S}_{n-1}}(Y_n,y)\subset J'_n. H = ⟨ q [ a , b ] : 1 ≤ a < b ≤ n − 1 ⟩ ⊂ C n , K = π 1 S n − 1 ( Y n , y ) ⊂ J n ′ . Regarding n n n as the root identifies K K K with C n − 1 C_{n-1} C n − 1 . The section identifies its image with the standard subgroup H H H . Since f σ n = id f\sigma_n=\operatorname{id} f σ n = id , we have an isomorphism
f ∗ ∣ H : H → ∼ K . (10) f_*|_H:H\xrightarrow{\sim}K. \tag*{(10)} f ∗ ∣ H : H ∼ K . (10) In particular, the image of f ∗ f_* f ∗ contains the pure subgroup J n ′ J'_n J n ′ and projects onto S n \mathfrak{S}_n S n , so f ∗ f_* f ∗ is surjective. @page 22/38 starts_mid=0 ends_mid=1
Lemma 7.1. For 1 ≤ i < n 1 \le i < n 1 ≤ i < n ,
f ∗ ( q t i q ) = f ∗ ( t n − i ) ; equivalently, f ∗ ( t i ) = f ∗ ( t i R ) . f_*(q t_i q) = f_*(t_{n-i}); \qquad\text{equivalently,} \qquad f_*(t_i) = f_*(t_i^{\mathrm{R}}). f ∗ ( q t i q ) = f ∗ ( t n − i ) ; equivalently, f ∗ ( t i ) = f ∗ ( t i R ) . Proof. The crucial observation is that the unrooted caterpillar y = f ( x ) y=f(x) y = f ( x ) is fixed by the longest permutation w 0 ( a ) = n + 1 − a w_0(a)=n+1-a w 0 ( a ) = n + 1 − a . Indeed, after forgetting ∞ \infty ∞ , the last component of x x x becomes unstable and is contracted. The resulting caterpillar is symmetric under reversal of the chain: reflecting the chain and reversing the order of special points on each component gives an isomorphism of the corresponding marked curves, since the order on each real component is defined only up to reversal. Thus
Let r = ( w 0 , const y ) ∈ J n ′ r=(w_0,\operatorname{const}_y)\in J'_n r = ( w 0 , const y ) ∈ J n ′ be the constant-path lift of w 0 w_0 w 0 .
Since w 0 w_0 w 0 reverses the cyclic order 1 , … , n , ∞ 1,\ldots,n,\infty 1 , … , n , ∞ , it maps the chamber D D D to itself, and the full cactus reversal q = q [ 1 , n ] q=q_{[1,n]} q = q [ 1 , n ] is represented by a path from x x x to w 0 x w_0x w 0 x contained in D ‾ \overline{D} D . After forgetting ∞ \infty ∞ , this path becomes a loop at y y y contained in the closure of the corresponding chamber of M ‾ 0 , n ( R ) \overline{M}_{0,n}(\mathbb{R}) M 0 , n ( R ) . Since this chamber closure is contractible, the loop is null-homotopic. Hence
f ∗ ( q ) = r . f_*(q)=r. f ∗ ( q ) = r . For 2 ≤ i ≤ n − 2 2\le i\le n-2 2 ≤ i ≤ n − 2 , the projected path f ( t i ) f(t_i) f ( t i ) varies only the four-pointed component with branches
L = { 1 , … , i − 1 } , i , i + 1 , R = { i + 2 , … , n } . L=\{1,\ldots,i-1\},\qquad i,\qquad i+1,\qquad R=\{i+2,\ldots,n\}. L = { 1 , … , i − 1 } , i , i + 1 , R = { i + 2 , … , n } . It joins the splittings ( L , i ) ∣ ( i + 1 , R ) (L,i)\mid(i+1,R) ( L , i ) ∣ ( i + 1 , R ) and ( L , i + 1 ) ∣ ( i , R ) (L,i+1)\mid(i,R) ( L , i + 1 ) ∣ ( i , R ) along the arc of M ‾ 0 , 4 ( R ) \overline{M}_{0,4}(\mathbb{R}) M 0 , 4 ( R ) avoiding ( L , R ) ∣ ( i , i + 1 ) (L,R)\mid(i,i+1) ( L , R ) ∣ ( i , i + 1 ) . Reflection exchanges the two outer branches and, after relabeling by w 0 w_0 w 0 , carries this arc to the corresponding arc for n − i n-i n − i . Therefore
r f ∗ ( t i ) r = f ∗ ( t n − i ) . r f_*(t_i) r=f_*(t_{n-i}). r f ∗ ( t i ) r = f ∗ ( t n − i ) . At the two ends, f ∗ ( t 1 ) f_*(t_1) f ∗ ( t 1 ) is the constant-path transposition of the first cherry, while f ( t n − 1 ) f(t_{n-1}) f ( t n − 1 ) becomes constant after the component carrying n , ∞ n,\infty n , ∞ is contracted; its equivariant permutation is the transposition of the last cherry. Reflection exchanges these two cases. For n = 3 n=3 n = 3 the same statement is simply the corresponding identity in S 3 \mathfrak{S}_3 S 3 . Combining this with f ∗ ( q ) = r f_*(q)=r f ∗ ( q ) = r proves
f ∗ ( q t i q ) = f ∗ ( t n − i ) f_*(q t_i q)=f_*(t_{n-i}) f ∗ ( q t i q ) = f ∗ ( t n − i ) for every 1 ≤ i < n 1\le i<n 1 ≤ i < n . The equivalent form follows from (8) .
The defining relations. Theorem 7.2. For n ≥ 3 n\ge3 n ≥ 3 , forgetting ∞ \infty ∞ induces an isomorphism
R C n → ∼ J n ′ . \mathcal{R}C_n \xrightarrow{\sim} J'_n. R C n ∼ J n ′ . Equivalently, the map (6) is surjective and
ker f ∗ = ⟨ q t i q t n − i : 1 ≤ i < n ⟩ . (11) \ker f_*=\left\langle q t_i q t_{n-i}:1\le i<n\right\rangle. \tag*{(11)} ker f ∗ = ⟨ q t i q t n − i : 1 ≤ i < n ⟩ . (11) Proof. There is a section σ n : Y n → X n \sigma_n:Y_n\to X_n σ n : Y n → X n of f f f : attach at the marking n n n a three-pointed component carrying n , ∞ n,\infty n , ∞ , and the node. It is equivariant for S n − 1 = Stab S n ( n ) \mathfrak{S}_{n-1}=\operatorname{Stab}_{\mathfrak{S}_n}(n) S n − 1 = Stab S n ( n ) , and σ n ( y ) = x \sigma_n(y)=x σ n ( y ) = x . Set
H = ⟨ q [ a , b ] : 1 ≤ a < b ≤ n − 1 ⟩ ⊂ C n , K = π 1 S n − 1 ( Y n , y ) ⊂ J n ′ . H=\left\langle q_{[a,b]}:1\le a<b\le n-1\right\rangle\subset C_n,\qquad K=\pi_1^{\mathfrak{S}_{n-1}}(Y_n,y)\subset J'_n. H = ⟨ q [ a , b ] : 1 ≤ a < b ≤ n − 1 ⟩ ⊂ C n , K = π 1 S n − 1 ( Y n , y ) ⊂ J n ′ . Regarding n n n as the root identifies K K K with C n − 1 C_{n-1} C n − 1 . The section identifies its image with the standard subgroup H H H . Since f σ n = id f\sigma_n=\operatorname{id} f σ n = id , we have an isomorphism
f ∗ ∣ H : H → ∼ K . (12) f_*|_H:H\xrightarrow{\sim}K. \tag*{(12)} f ∗ ∣ H : H ∼ K . (12) In particular, the image of f ∗ f_* f ∗ contains the pure subgroup J n ′ J'_n J n ′ and projects onto S n \mathfrak{S}_n S n , so f ∗ f_* f ∗ is surjective.
By Lemma 7.1, write f ∗ = f ˉ η f_*=\bar{f}\eta f ∗ = f ˉ η , where η : C n → R C n \eta:C_n\to\mathcal{R}C_n η : C n → R C n is the quotient map. Conjugating (5) by q q q gives
P : = π − 1 ( S n − 1 ) = ⟨ H , t 1 R , … , t n − 2 R ⟩ . P:=\pi^{-1}(\mathfrak{S}_{n-1})=\langle H,t_1^{R},\ldots,t_{n-2}^{R}\rangle. P := π − 1 ( S n − 1 ) = ⟨ H , t 1 R , … , t n − 2 R ⟩ .
In R C n \mathcal{R}C_n R C n we have η ( t i R ) = η ( t i ) \eta(t_i^{R})=\eta(t_i) η ( t i R ) = η ( t i ) . Moreover, t i ∈ H t_i\in H t i ∈ H for i ≤ n − 2 i\le n-2 i ≤ n − 2 by Theorem 2.5. Hence η ( P ) = η ( H ) \eta(P)=\eta(H) η ( P ) = η ( H ) .
The permutation map descends to R C n \mathcal{R}C_n R C n , since all the additional relators have trivial permutation. If g ∈ ker f ˉ g\in\ker\bar{f} g ∈ ker f ˉ , any lift of g g g to C n C_n C n has trivial permutation and therefore belongs to P P P . Thus g = η ( h ) g=\eta(h) g = η ( h ) for some h ∈ H h\in H h ∈ H . Now f ∗ ( h ) = f ˉ ( g ) = 1 f_*(h)=\bar{f}(g)=1 f ∗ ( h ) = f ˉ ( g ) = 1 , and (9) implies h = 1 h=1 h = 1 . Therefore f ˉ \bar{f} f ˉ is injective. □
Corollary 7.3. The big cactus group has a presentation on t 1 , … , t n − 1 t_1,\ldots,t_{n-1} t 1 , … , t n − 1 with relations
t i 2 = 1 , t i t j = t j t i ( ∣ i − j ∣ > 1 ) , t_i^2=1,\qquad t_i t_j=t_j t_i\quad(|i-j|>1), t i 2 = 1 , t i t j = t j t i ( ∣ i − j ∣ > 1 ) , ( t i q [ j , k ] ) 2 = 1 ( i + 1 < j < k ) , (t_iq_{[j,k]})^2=1\quad(i+1<j<k), ( t i q [ j , k ] ) 2 = 1 ( i + 1 < j < k ) , q t i q = t n − i ( 1 ≤ i < n ) , qt_iq=t_{n-i}\quad(1\le i<n), q t i q = t n − i ( 1 ≤ i < n ) , where q i q_i q i and q [ j , k ] q_{[j,k]} q [ j , k ] are the words of Theorem 2.5. A second presentation is obtained by replacing the last family by t w = q t_w=q t w = q for every reduced word w w w for w 0 ∈ S n w_0\in\mathfrak{S}_n w 0 ∈ S n adapted to an orientation of A n − 1 A_{n-1} A n − 1 .
Proof. Combine Theorem 7.2 with Theorems 2.5 and 5.4. □
Remark 7.4. Davis–Januszkiewicz–Scott give a geometric presentation of the big cactus group in [13 Theorem 8.6.2 and Remark 8.6.5] , in terms of polygonal subdivisions and dihedral symmetries (in their notation, the group is Aut ( X ) \operatorname{Aut}(X) Aut ( X ) ). The presentations above instead use the elements t i t_i t i and describe precisely the relations introduced by forgetting a marking.
Invariant tensors
We now prove Theorem 1.9. The main step is the Dynkin-reversal relation on invariants.
Theorem 8.1. On Inv ( B 1 ⊗ ⋯ ⊗ B n ) \operatorname{Inv}(B_1\otimes\cdots\otimes B_n) Inv ( B 1 ⊗ ⋯ ⊗ B n ) ,
q [ 1 , n ] t i q [ 1 , n ] = t n − i , 1 ≤ i ≤ n − 1. q_{[1,n]}t_iq_{[1,n]}=t_{n-i},\qquad1\le i\le n-1. q [ 1 , n ] t i q [ 1 , n ] = t n − i , 1 ≤ i ≤ n − 1.
Theorem 1.9 follows from Theorem 5.4 and Proposition 8.8: the Dynkin-reversal relations give Q D = q [ 1 , n ] Q_D=q_{[1,n]} Q D = q [ 1 , n ] for every Dynkin path D D D , while q [ 1 , n ] q_{[1,n]} q [ 1 , n ] acts on invariants by
b 1 ⊗ ⋯ ⊗ b n ⟼ ξ ( b n ) ⊗ ⋯ ⊗ ξ ( b 1 ) . b_1\otimes\cdots\otimes b_n\longmapsto\xi(b_n)\otimes\cdots\otimes\xi(b_1). b 1 ⊗ ⋯ ⊗ b n ⟼ ξ ( b n ) ⊗ ⋯ ⊗ ξ ( b 1 ) .
For the combinatorial proof of Theorem 8.1 in type A A A , we refine the factors into single boxes and factor the two local rules into the same sequence of pairwise box interchanges. Proposition 8.13 then gives the result one crossing at a time. Section 8.7 gives a second proof, valid in every semisimple type.
Crystals, the Schützenberger involution, and the commutor
Fix s l m \mathfrak{sl}_m sl m . A (finite, normal) crystal is a finite set B B B equipped with a weight map wt : B → Z m \operatorname{wt}:B\to\mathbb{Z}^m wt : B → Z m and partial operators
e i , f i : B → B , 1 ≤ i < m , e_i,f_i:B\to B,\qquad1\le i<m, e i , f i : B → B , 1 ≤ i < m ,
such that f i ( b ) = b ′ f_i(b)=b' f i ( b ) = b ′ if and only if e i ( b ′ ) = b e_i(b')=b e i ( b ′ ) = b , and
wt ( f i b ) = wt ( b ) − α i . \operatorname{wt}(f_i b)=\operatorname{wt}(b)-\alpha_i. wt ( f i b ) = wt ( b ) − α i .
We will only consider normal crystals, so every connected component is the crystal B ( λ ) B(\lambda) B ( λ ) of an irreducible representation. The standard model is B ( λ ) = SSYT ( λ , m ) B(\lambda)=\operatorname{SSYT}(\lambda,m) B ( λ ) = SSYT ( λ , m ) , and in particular B ( ϖ 1 ) = { 1 , … , m } B(\varpi_1)=\{1,\ldots,m\} B ( ϖ 1 ) = { 1 , … , m } . We write
ε i ( b ) = max { k : e i k b is defined } , φ i ( b ) = max { k : f i k b is defined } . \varepsilon_i(b)=\max\{k:e_i^k b\text{ is defined}\},\qquad\varphi_i(b)=\max\{k:f_i^k b\text{ is defined}\}. ε i ( b ) = max { k : e i k b is defined } , φ i ( b ) = max { k : f i k b is defined } .
By a morphism of crystals we mean a strict morphism : a map g : B → B ′ g:B\to B' g : B → B ′ satisfying
wt ( g ( b ) ) = wt ( b ) , g ( e i b ) = e i g ( b ) , g ( f i b ) = f i g ( b ) \operatorname{wt}(g(b))=\operatorname{wt}(b),\qquad g(e_i b)=e_i g(b),\qquad g(f_i b)=f_i g(b) wt ( g ( b )) = wt ( b ) , g ( e i b ) = e i g ( b ) , g ( f i b ) = f i g ( b )
for every b ∈ B b\in B b ∈ B and every simple root index i i i . Here we adjoin a symbol 0 0 0 to each crystal, interpret every undefined crystal operator as having value 0 0 0 , and extend g g g by g ( 0 ) = 0 g(0)=0 g ( 0 ) = 0 . Thus the identities above require preservation of undefinedness as well. The image of a morphism is a union of connected components of B ′ B' B ′ ; in particular, an injective morphism identifies B B B with such a union. Thus, if g : B → B ′ g:B\to B' g : B → B ′ is an injective morphism, its image is a union of connected components of B ′ B' B ′ .
An element is highest weight if every e i e_i e i is undefined, and lowest weight if every f i f_i f i is undefined. We call it invariant if it is both, or equivalently if it forms a one-element connected component.
Definition 8.2 (tensor product). The tensor product B ⊗ C B\otimes C B ⊗ C is the set B × C B\times C B × C , with
wt ( b ⊗ c ) = wt ( b ) + wt ( c ) , \operatorname{wt}(b\otimes c)=\operatorname{wt}(b)+\operatorname{wt}(c), wt ( b ⊗ c ) = wt ( b ) + wt ( c ) , and
f i ( b ⊗ c ) = { f i ( b ) ⊗ c , ε i ( b ) ≥ φ i ( c ) , b ⊗ f i ( c ) , otherwise . f_i(b\otimes c)=
\begin{cases}
f_i(b)\otimes c, & \varepsilon_i(b)\ge\varphi_i(c),\\
b\otimes f_i(c), & \text{otherwise}.
\end{cases} f i ( b ⊗ c ) = { f i ( b ) ⊗ c , b ⊗ f i ( c ) , ε i ( b ) ≥ φ i ( c ) , otherwise . while
e i ( b ⊗ c ) = { e i ( b ) ⊗ c , ε i ( b ) > φ i ( c ) , b ⊗ e i ( c ) , otherwise . e_i(b\otimes c)=
\begin{cases}
e_i(b)\otimes c, & \varepsilon_i(b)>\varphi_i(c),\\
b\otimes e_i(c), & \text{otherwise}.
\end{cases} e i ( b ⊗ c ) = { e i ( b ) ⊗ c , b ⊗ e i ( c ) , ε i ( b ) > φ i ( c ) , otherwise . This convention is associative, and in a highest weight element of B ⊗ C B\otimes C B ⊗ C the second factor is highest weight.
The ordinary flip b ⊗ c ⟼ c ⊗ b b\otimes c\longmapsto c\otimes b b ⊗ c ⟼ c ⊗ b is not a crystal morphism B ⊗ C → C ⊗ B B\otimes C\to C\otimes B B ⊗ C → C ⊗ B , although the two crystals are isomorphic. For an irreducible crystal B ( λ ) B(\lambda) B ( λ ) , the Schützenberger involution ξ \xi ξ is the unique bijection satisfying
ξ ( b λ ) = b λ l o w , ξ ( e i b ) = f m − i ξ ( b ) , \xi(b_\lambda)=b_\lambda^{\mathrm{low}},\qquad\xi(e_i b)=f_{m-i}\xi(b), ξ ( b λ ) = b λ low , ξ ( e i b ) = f m − i ξ ( b ) ,
where b λ b_\lambda b λ and b λ l o w b_\lambda^{\mathrm{low}} b λ low are the highest and lowest weight elements. It is an involution, and on SSYT ( λ , m ) \operatorname{SSYT}(\lambda,m) SSYT ( λ , m ) it is the usual Schützenberger evacuation [27 , 32 ] ; by Lemma 2.4 , this is the Bender–Knuth word q m − 1 q_{m-1} q m − 1 . For a general finite crystal, we define ξ \xi ξ componentwise.
The Schützenberger involution is natural with respect to crystal isomorphisms:
g ∘ ξ B = ξ B ′ ∘ g g\circ\xi_B=\xi_{B'}\circ g g ∘ ξ B = ξ B ′ ∘ g
for every isomorphism g : B → B ′ g:B\to B' g : B → B ′ . Indeed, g ξ B g − 1 g\xi_Bg^{-1} g ξ B g − 1 sends the highest weight element of each component to its lowest weight element and intertwines e i e_i e i with f m − i f_{m-i} f m − i , so it must equal ξ B ′ \xi_{B'} ξ B ′ . In particular, ξ \xi ξ fixes every invariant, since on a one-element component the highest and lowest weight elements are the same.
Definition 8.3 (Henriques–Kamnitzer [27 ] ). The crystal commutor
σ B , C : B ⊗ C ⟶ C ⊗ B \sigma_{B,C}:B\otimes C\longrightarrow C\otimes B σ B , C : B ⊗ C ⟶ C ⊗ B is defined by
σ B , C ( b ⊗ c ) = ξ C ⊗ B ( ξ C ( c ) ⊗ ξ B ( b ) ) . \sigma_{B,C}(b\otimes c)=\xi_{C\otimes B}\bigl(\xi_C(c)\otimes\xi_B(b)\bigr). σ B , C ( b ⊗ c ) = ξ C ⊗ B ( ξ C ( c ) ⊗ ξ B ( b ) ) .
Example 8.4. For s l 2 \mathfrak{sl}_2 sl 2 write B ( 1 ) = { 1 , 2 } B(1)=\{1,2\} B ( 1 ) = { 1 , 2 } and B ( 2 ) = { 11 , 12 , 22 } B(2)=\{11,12,22\} B ( 2 ) = { 11 , 12 , 22 } , with ξ ( 1 ) = 2 \xi(1)=2 ξ ( 1 ) = 2 , ξ ( 11 ) = 22 \xi(11)=22 ξ ( 11 ) = 22 and ξ ( 12 ) = 12 \xi(12)=12 ξ ( 12 ) = 12 . The crystal B ( 1 ) ⊗ B ( 2 ) B(1)\otimes B(2) B ( 1 ) ⊗ B ( 2 ) has components
{ 1 ⊗ 11 , 1 ⊗ 12 , 1 ⊗ 22 , 2 ⊗ 22 } ≅ B ( 3 ) , { 2 ⊗ 11 , 2 ⊗ 12 } ≅ B ( 1 ) , \{1\otimes11,\ 1\otimes12,\ 1\otimes22,\ 2\otimes22\}\cong B(3),\qquad\{2\otimes11,\ 2\otimes12\}\cong B(1), { 1 ⊗ 11 , 1 ⊗ 12 , 1 ⊗ 22 , 2 ⊗ 22 } ≅ B ( 3 ) , { 2 ⊗ 11 , 2 ⊗ 12 } ≅ B ( 1 ) ,
and B ( 2 ) ⊗ B ( 1 ) B(2)\otimes B(1) B ( 2 ) ⊗ B ( 1 ) has components { 11 ⊗ 1 , 11 ⊗ 2 , 12 ⊗ 2 , 22 ⊗ 2 } ≅ B ( 3 ) \{11\otimes1, 11\otimes2, 12\otimes2, 22\otimes2\}\cong B(3) { 11 ⊗ 1 , 11 ⊗ 2 , 12 ⊗ 2 , 22 ⊗ 2 } ≅ B ( 3 ) and { 12 ⊗ 1 , 22 ⊗ 1 } ≅ B ( 1 ) \{12\otimes1, 22\otimes1\}\cong B(1) { 12 ⊗ 1 , 22 ⊗ 1 } ≅ B ( 1 ) . Unwinding Definition 8.3, for instance σ ( 1 ⊗ 22 ) = ξ B ( 2 ) ⊗ B ( 1 ) ( ξ ( 22 ) ⊗ ξ ( 1 ) ) = ξ ( 11 ⊗ 2 ) = 12 ⊗ 2 \sigma(1\otimes22)=\xi_{B(2)\otimes B(1)}(\xi(22)\otimes\xi(1))=\xi(11\otimes2)=12\otimes2 σ ( 1 ⊗ 22 ) = ξ B ( 2 ) ⊗ B ( 1 ) ( ξ ( 22 ) ⊗ ξ ( 1 )) = ξ ( 11 ⊗ 2 ) = 12 ⊗ 2 , one finds
Compare the flip: it sends 1 ⊗ 22 1\otimes22 1 ⊗ 22 to 22 ⊗ 1 22\otimes1 22 ⊗ 1 , which lies in the wrong component. Here B ( 1 ) ⊗ B ( 2 ) B(1)\otimes B(2) B ( 1 ) ⊗ B ( 2 ) is multiplicity-free, so σ \sigma σ is the only isomorphism B ( 1 ) ⊗ B ( 2 ) → B ( 2 ) ⊗ B ( 1 ) B(1)\otimes B(2)\to B(2)\otimes B(1) B ( 1 ) ⊗ B ( 2 ) → B ( 2 ) ⊗ B ( 1 ) ; in general there are several, and the content of Definition 8.3 is the choice of one.
Thus the ordinary flip is corrected by applying the Schützenberger involution to the two factors, and then once more to the resulting tensor product. The two applications of ξ \xi ξ reverse the crystal operators twice, so the resulting map is a crystal isomorphism. We will use the following properties, proved in [27 ] .
Proposition 8.5. (a) The commutor is involutive:
σ C , B ∘ σ B , C = id . \sigma_{C,B}\circ\sigma_{B,C}=\operatorname{id}. σ C , B ∘ σ B , C = id . (b) It is natural with respect to isomorphisms and inclusions of unions of connected components. Thus, if g : B → B ′ g:B\to B' g : B → B ′ and h : C → C ′ h:C\to C' h : C → C ′ are maps of either kind, then
σ B ′ , C ′ ∘ ( g ⊗ h ) = ( h ⊗ g ) ∘ σ B , C . \sigma_{B',C'}\circ(g\otimes h)=(h\otimes g)\circ\sigma_{B,C}. σ B ′ , C ′ ∘ ( g ⊗ h ) = ( h ⊗ g ) ∘ σ B , C . (c) It satisfies the cactus axiom:
σ B , D ⊗ C ∘ ( 1 B ⊗ σ C , D ) = σ C ⊗ B , D ∘ ( σ B , C ⊗ 1 D ) . \sigma_{B,D\otimes C}\circ(1_B\otimes\sigma_{C,D})=\sigma_{C\otimes B,D}\circ(\sigma_{B,C}\otimes1_D). σ B , D ⊗ C ∘ ( 1 B ⊗ σ C , D ) = σ C ⊗ B , D ∘ ( σ B , C ⊗ 1 D ) . (d) In general it does not satisfy the hexagon axiom
σ B ⊗ C , D = ( σ B , D ⊗ 1 C ) ∘ ( 1 B ⊗ σ C , D ) . \sigma_{B\otimes C,D}=(\sigma_{B,D}\otimes1_C)\circ(1_B\otimes\sigma_{C,D}). σ B ⊗ C , D = ( σ B , D ⊗ 1 C ) ∘ ( 1 B ⊗ σ C , D ) .
Henriques and Kamnitzer state (b) for isomorphisms. The extension to inclusions of unions of connected components follows immediately from the definition, since the Schützenberger involution is defined componentwise.
A monoidal category with a natural involutive commutor satisfying the cactus axiom is called a coboundary category . Thus crystals form a coboundary category, but not a symmetric one.
The cactus axiom says, in particular, that there is a canonical way to reverse any number of tensor factors. For three factors, reversing C , D C,D C , D and then moving B B B to the end gives the same map as reversing B , C B,C B , C and then moving D D D to the front. Iterating this construction gives a canonical full reversal
ρ : X 1 ⊗ ⋯ ⊗ X k ⟶ X k ⊗ ⋯ ⊗ X 1 . \rho:X_1\otimes\cdots\otimes X_k\longrightarrow X_k\otimes\cdots\otimes X_1. ρ : X 1 ⊗ ⋯ ⊗ X k ⟶ X k ⊗ ⋯ ⊗ X 1 .
If
X = X 1 ⊗ ⋯ ⊗ X j , Y = X j + 1 ⊗ ⋯ ⊗ X k , X=X_1\otimes\cdots\otimes X_j,\qquad Y=X_{j+1}\otimes\cdots\otimes X_k, X = X 1 ⊗ ⋯ ⊗ X j , Y = X j + 1 ⊗ ⋯ ⊗ X k ,
then
ρ X ⊗ Y = ( ρ Y ⊗ ρ X ) ∘ σ X , Y . (13) \rho_{X\otimes Y}=(\rho_Y\otimes\rho_X)\circ\sigma_{X,Y}. \tag*{(13)} ρ X ⊗ Y = ( ρ Y ⊗ ρ X ) ∘ σ X , Y . (13)
By the cactus axiom this is independent of the choice of splitting. Moreover, ρ \rho ρ is an involution, and for two factors ρ B ⊗ C = σ B , C \rho_{B\otimes C}=\sigma_{B,C} ρ B ⊗ C = σ B , C .
The cactus group action
The cactus group C n C_n C n acts on tensor products of n n n crystals. The generator q [ i , j ] q_{[i,j]} q [ i , j ] acts by the full reversal ρ \rho ρ on the factors in positions i , … , j i,\ldots,j i , … , j , and as the identity on the remaining factors. Thus it maps B 1 ⊗ ⋯ ⊗ B n B_1\otimes\cdots\otimes B_n B 1 ⊗ ⋯ ⊗ B n to the tensor product in which the factors B i , … , B j B_i,\ldots,B_j B i , … , B j appear in reverse order. More precisely, this is an action on the groupoid of all orderings of the factors; when all B i B_i B i are equal, it is an ordinary action on a single crystal.
The cactus relations (C1)–(C3) follow from the properties of the commutor [27 ] . Relation (C1) follows because ρ \rho ρ is an involution, and (C2) because reversals of disjoint intervals act on disjoint factors. For (C3), reversing [ i , j ] [i,j] [ i , j ] carries the interval [ k , l ] [k,l] [ k , l ] to [ i + j − l , i + j − k ] [i+j-l,i+j-k] [ i + j − l , i + j − k ] , and the independence of the splitting in (13) shows that conjugating the reversal of [ k , l ] [k,l] [ k , l ] gives the reversal of this new interval. In particular, the elements t i ∈ C n t_i\in C_n t i ∈ C n from Theorem 2.5 act on tensor products, with
t i 2 = 1 , t i t j = t j t i for ∣ i − j ∣ > 1. t_i^2=1,\qquad t_i t_j=t_j t_i\quad\text{for }|i-j|>1. t i 2 = 1 , t i t j = t j t i for ∣ i − j ∣ > 1.
We will also need to express the commutor itself in terms of interval reversals. Suppose that X X X occupies positions [ p , r ′ ] [p,r'] [ p , r ′ ] and Y Y Y occupies positions [ r ′ + 1 , r ] [r'+1,r] [ r ′ + 1 , r ] . Then
σ X , Y = q [ p , p + r − r ′ − 1 ] q [ p + r − r ′ , r ] q [ p , r ] = q [ p , r ] q [ p , r ′ ] q [ r ′ + 1 , r ] . (14) \sigma_{X,Y}=q_{[p,p+r-r'-1]}q_{[p+r-r',r]}q_{[p,r]}=q_{[p,r]}q_{[p,r']}q_{[r'+1,r]}. \tag*{(14)} σ X , Y = q [ p , p + r − r ′ − 1 ] q [ p + r − r ′ , r ] q [ p , r ] = q [ p , r ] q [ p , r ′ ] q [ r ′ + 1 , r ] . (14)
The first expression reverses the whole block and then reverses X X X and Y Y Y separately in their new positions; the second reverses the two blocks first and then reverses the whole interval. A one-element interval is understood to act trivially.
Lemma 8.6 (naturality of the action). Suppose that each g i : B i → B i ′ g_i:B_i\to B'_i g i : B i → B i ′ is an isomorphism or an inclusion of a union of connected components, and let w ∈ C n w\in C_n w ∈ C n permute the positions by π \pi π . Then
w ∘ ( g 1 ⊗ ⋯ ⊗ g n ) = ( g π − 1 ( 1 ) ⊗ ⋯ ⊗ g π − 1 ( n ) ) ∘ w . w\circ(g_1\otimes\cdots\otimes g_n)=(g_{\pi^{-1}(1)}\otimes\cdots\otimes g_{\pi^{-1}(n)})\circ w. w ∘ ( g 1 ⊗ ⋯ ⊗ g n ) = ( g π − 1 ( 1 ) ⊗ ⋯ ⊗ g π − 1 ( n ) ) ∘ w .
Proof. Each interval reversal is obtained from commutors and identity maps using (13) , and the commutor is natural. □
Example 8.7. For n = 4 n=4 n = 4 and i = 2 i=2 i = 2 , the four factors of t 2 = q [ 1 , 2 ] q [ 1 , 3 ] q [ 1 , 2 ] q [ 1 , 1 ] t_2=q_{[1,2]}q_{[1,3]}q_{[1,2]}q_{[1,1]} t 2 = q [ 1 , 2 ] q [ 1 , 3 ] q [ 1 , 2 ] q [ 1 , 1 ] act, from right to left, as
B 1 B 2 B 3 B 4 → q [ 1 , 1 ] B 1 B 2 B 3 B 4 → q [ 1 , 2 ] B 2 B 1 B 3 B 4 → q [ 1 , 3 ] B 3 B 1 B 2 B 4 → q [ 1 , 2 ] B 1 B 3 B 2 B 4 . B_1B_2B_3B_4\xrightarrow{q_{[1,1]}}B_1B_2B_3B_4\xrightarrow{q_{[1,2]}}B_2B_1B_3B_4\xrightarrow{q_{[1,3]}}B_3B_1B_2B_4\xrightarrow{q_{[1,2]}}B_1B_3B_2B_4. B 1 B 2 B 3 B 4 q [ 1 , 1 ] B 1 B 2 B 3 B 4 q [ 1 , 2 ] B 2 B 1 B 3 B 4 q [ 1 , 3 ] B 3 B 1 B 2 B 4 q [ 1 , 2 ] B 1 B 3 B 2 B 4 . Thus B 2 B_2 B 2 first moves past B 1 B_1 B 1 to the front, and then past B 1 ⊗ B 3 B_1\otimes B_3 B 1 ⊗ B 3 to the back.
Similarly, the right-anchored element t 2 ′ = q [ 3 , 4 ] q [ 2 , 4 ] q [ 3 , 4 ] q [ 4 , 4 ] t'_2=q_{[3,4]}q_{[2,4]}q_{[3,4]}q_{[4,4]} t 2 ′ = q [ 3 , 4 ] q [ 2 , 4 ] q [ 3 , 4 ] q [ 4 , 4 ] acts by
B 1 B 2 B 3 B 4 → q [ 4 , 4 ] B 1 B 2 B 3 B 4 → q [ 3 , 4 ] B 1 B 2 B 4 B 3 → q [ 2 , 4 ] B 1 B 3 B 4 B 2 → q [ 3 , 4 ] B 1 B 3 B 2 B 4 . B_1B_2B_3B_4\xrightarrow{q_{[4,4]}}B_1B_2B_3B_4\xrightarrow{q_{[3,4]}}B_1B_2B_4B_3\xrightarrow{q_{[2,4]}}B_1B_3B_4B_2\xrightarrow{q_{[3,4]}}B_1B_3B_2B_4. B 1 B 2 B 3 B 4 q [ 4 , 4 ] B 1 B 2 B 3 B 4 q [ 3 , 4 ] B 1 B 2 B 4 B 3 q [ 2 , 4 ] B 1 B 3 B 4 B 2 q [ 3 , 4 ] B 1 B 3 B 2 B 4 . Here B 3 B_3 B 3 first moves past B 4 B_4 B 4 to the back, and then past B 2 ⊗ B 4 B_2\otimes B_4 B 2 ⊗ B 4 to the front. Both maps therefore interchange B 2 B_2 B 2 and B 3 B_3 B 3 , ending at B 1 B 3 B 2 B 4 B_1B_3B_2B_4 B 1 B 3 B 2 B 4 . Theorem 8.1 says that on invariants these two maps are equal.
Proposition 8.8 (reversal on invariants). On Inv ( B 1 ⊗ ⋯ ⊗ B n ) \operatorname{Inv}(B_1\otimes\cdots\otimes B_n) Inv ( B 1 ⊗ ⋯ ⊗ B n ) , the full reversal q [ 1 , n ] q_{[1,n]} q [ 1 , n ] acts by
b 1 ⊗ ⋯ ⊗ b n ⟼ ξ ( b n ) ⊗ ⋯ ⊗ ξ ( b 1 ) . b_1\otimes\cdots\otimes b_n\longmapsto\xi(b_n)\otimes\cdots\otimes\xi(b_1). b 1 ⊗ ⋯ ⊗ b n ⟼ ξ ( b n ) ⊗ ⋯ ⊗ ξ ( b 1 ) . Proof. Write
Θ ( b 1 ⊗ ⋯ ⊗ b n ) = ξ ( b n ) ⊗ ⋯ ⊗ ξ ( b 1 ) . \Theta(b_1\otimes\cdots\otimes b_n)=\xi(b_n)\otimes\cdots\otimes\xi(b_1). Θ ( b 1 ⊗ ⋯ ⊗ b n ) = ξ ( b n ) ⊗ ⋯ ⊗ ξ ( b 1 ) . Iterating the recursive definition of the full reversal, using naturality of ξ \xi ξ and the fact that Θ \Theta Θ intertwines e i e_i e i with f m − i f_{m-i} f m − i , and hence conjugates the Schützenberger involution to the Schützenberger involution on the reversed tensor product, gives
q [ 1 , n ] = ξ B n ⊗ ⋯ ⊗ B 1 ∘ Θ . q_{[1,n]}=\xi_{B_n\otimes\cdots\otimes B_1}\circ\Theta. q [ 1 , n ] = ξ B n ⊗ ⋯ ⊗ B 1 ∘ Θ. @page 26/38 starts_mid=0 ends_mid=1
The cactus group action {#sec:8.2} The cactus group C n C_n C n acts on tensor products of n n n crystals. The generator q [ i , j ] q_{[i,j]} q [ i , j ] acts by the full reversal ρ \rho ρ on the factors in positions i , … , j i,\ldots,j i , … , j , and as the identity on the remaining factors. Thus it maps B 1 ⊗ ⋯ ⊗ B n B_1\otimes\cdots\otimes B_n B 1 ⊗ ⋯ ⊗ B n to the tensor product in which the factors B i , … , B j B_i,\ldots,B_j B i , … , B j appear in reverse order. More precisely, this is an action on the groupoid of all orderings of the factors; when all B i B_i B i are equal, it is an ordinary action on a single crystal.
The cactus relations (C1)–(C3) follow from the properties of the commutor [27 ] . Relation (C1) follows because ρ \rho ρ is an involution, and (C2) because reversals of disjoint intervals act on disjoint factors. For (C3), reversing [ i , j ] [i,j] [ i , j ] carries the interval [ k , l ] [k,l] [ k , l ] to [ i + j − l , i + j − k ] [i+j-l,i+j-k] [ i + j − l , i + j − k ] , and the independence of the splitting in (13) shows that conjugating the reversal of [ k , l ] [k,l] [ k , l ] gives the reversal of this new interval. In particular, the elements t i ∈ C n t_i\in C_n t i ∈ C n from Theorem 2.5 act on tensor products, with
t i 2 = 1 , t i t j = t j t i for ∣ i − j ∣ > 1. t_i^2=1,\qquad t_i t_j=t_j t_i\quad\text{for }|i-j|>1. t i 2 = 1 , t i t j = t j t i for ∣ i − j ∣ > 1. We will also need to express the commutor itself in terms of interval reversals. Suppose that X X X occupies positions [ p , r ′ ] [p,r'] [ p , r ′ ] and Y Y Y occupies positions [ r ′ + 1 , r ] [r'+1,r] [ r ′ + 1 , r ] . Then
σ X , Y = q [ p , p + r − r ′ − 1 ] q [ p + r − r ′ , r ] q [ p , r ] = q [ p , r ] q [ p , r ′ ] q [ r ′ + 1 , r ] . (15) \sigma_{X,Y}=q_{[p,p+r-r'-1]}q_{[p+r-r',r]}q_{[p,r]}=q_{[p,r]}q_{[p,r']}q_{[r'+1,r]}. \tag*{(15)} σ X , Y = q [ p , p + r − r ′ − 1 ] q [ p + r − r ′ , r ] q [ p , r ] = q [ p , r ] q [ p , r ′ ] q [ r ′ + 1 , r ] . (15) The first expression reverses the whole block and then reverses X X X and Y Y Y separately in their new positions; the second reverses the two blocks first and then reverses the whole interval. A one-element interval is understood to act trivially.
Lemma 8.6 (naturality of the action). Suppose that each g i : B i → B i ′ g_i:B_i\to B'_i g i : B i → B i ′ is an isomorphism or an inclusion of a union of connected components, and let w ∈ C n w\in C_n w ∈ C n permute the positions by π \pi π . Then
w ∘ ( g 1 ⊗ ⋯ ⊗ g n ) = ( g π − 1 ( 1 ) ⊗ ⋯ ⊗ g π − 1 ( n ) ) ∘ w . w\circ(g_1\otimes\cdots\otimes g_n)=(g_{\pi^{-1}(1)}\otimes\cdots\otimes g_{\pi^{-1}(n)})\circ w. w ∘ ( g 1 ⊗ ⋯ ⊗ g n ) = ( g π − 1 ( 1 ) ⊗ ⋯ ⊗ g π − 1 ( n ) ) ∘ w . Proof. Each interval reversal is obtained from commutors and identity maps using (13) , and the commutor is natural. □
Example 8.7. For n = 4 n=4 n = 4 and i = 2 i=2 i = 2 , the four factors of t 2 = q [ 1 , 2 ] q [ 1 , 3 ] q [ 1 , 2 ] q [ 1 , 1 ] t_2=q_{[1,2]}q_{[1,3]}q_{[1,2]}q_{[1,1]} t 2 = q [ 1 , 2 ] q [ 1 , 3 ] q [ 1 , 2 ] q [ 1 , 1 ] act, from right to left, as
B 1 B 2 B 3 B 4 → q [ 1 , 1 ] B 1 B 2 B 3 B 4 → q [ 1 , 2 ] B 2 B 1 B 3 B 4 → q [ 1 , 3 ] B 3 B 1 B 2 B 4 → q [ 1 , 2 ] B 1 B 3 B 2 B 4 . B_1B_2B_3B_4\xrightarrow{q_{[1,1]}}B_1B_2B_3B_4\xrightarrow{q_{[1,2]}}B_2B_1B_3B_4\xrightarrow{q_{[1,3]}}B_3B_1B_2B_4\xrightarrow{q_{[1,2]}}B_1B_3B_2B_4. B 1 B 2 B 3 B 4 q [ 1 , 1 ] B 1 B 2 B 3 B 4 q [ 1 , 2 ] B 2 B 1 B 3 B 4 q [ 1 , 3 ] B 3 B 1 B 2 B 4 q [ 1 , 2 ] B 1 B 3 B 2 B 4 . Thus B 2 B_2 B 2 first moves past B 1 B_1 B 1 to the front, and then past B 1 ⊗ B 3 B_1\otimes B_3 B 1 ⊗ B 3 to the back.
Similarly, the right-anchored element t 2 ′ = q [ 3 , 4 ] q [ 2 , 4 ] q [ 3 , 4 ] q [ 4 , 4 ] t'_2=q_{[3,4]}q_{[2,4]}q_{[3,4]}q_{[4,4]} t 2 ′ = q [ 3 , 4 ] q [ 2 , 4 ] q [ 3 , 4 ] q [ 4 , 4 ] acts by
B 1 B 2 B 3 B 4 → q [ 4 , 4 ] B 1 B 2 B 3 B 4 → q [ 3 , 4 ] B 1 B 2 B 4 B 3 → q [ 2 , 4 ] B 1 B 3 B 4 B 2 → q [ 3 , 4 ] B 1 B 3 B 2 B 4 . B_1B_2B_3B_4\xrightarrow{q_{[4,4]}}B_1B_2B_3B_4\xrightarrow{q_{[3,4]}}B_1B_2B_4B_3\xrightarrow{q_{[2,4]}}B_1B_3B_4B_2\xrightarrow{q_{[3,4]}}B_1B_3B_2B_4. B 1 B 2 B 3 B 4 q [ 4 , 4 ] B 1 B 2 B 3 B 4 q [ 3 , 4 ] B 1 B 2 B 4 B 3 q [ 2 , 4 ] B 1 B 3 B 4 B 2 q [ 3 , 4 ] B 1 B 3 B 2 B 4 . Here B 3 B_3 B 3 first moves past B 4 B_4 B 4 to the back, and then past B 2 ⊗ B 4 B_2\otimes B_4 B 2 ⊗ B 4 to the front. Both maps therefore interchange B 2 B_2 B 2 and B 3 B_3 B 3 , ending at $B_1B_3B_2B_4. Theorem 8.1 says that on invariants these two maps are equal.
Reversal on invariants {#sec:8.8} Proposition 8.8 (reversal on invariants). On Inv ( B 1 ⊗ ⋯ ⊗ B n ) \operatorname{Inv}(B_1\otimes\cdots\otimes B_n) Inv ( B 1 ⊗ ⋯ ⊗ B n ) , the full reversal q [ 1 , n ] q_{[1,n]} q [ 1 , n ] acts by
b 1 ⊗ ⋯ ⊗ b n ⟼ ξ ( b n ) ⊗ ⋯ ⊗ ξ ( b 1 ) . b_1\otimes\cdots\otimes b_n\longmapsto\xi(b_n)\otimes\cdots\otimes\xi(b_1). b 1 ⊗ ⋯ ⊗ b n ⟼ ξ ( b n ) ⊗ ⋯ ⊗ ξ ( b 1 ) . Proof. Write
Θ ( b 1 ⊗ ⋯ ⊗ b n ) = ξ ( b n ) ⊗ ⋯ ⊗ ξ ( b 1 ) . \Theta(b_1\otimes\cdots\otimes b_n)=\xi(b_n)\otimes\cdots\otimes\xi(b_1). Θ ( b 1 ⊗ ⋯ ⊗ b n ) = ξ ( b n ) ⊗ ⋯ ⊗ ξ ( b 1 ) . Iterating the recursive definition of the full reversal, using naturality of ξ \xi ξ and the fact that Θ \Theta Θ intertwines e i e_i e i with f m − i f_{m-i} f m − i , and hence conjugates the Schützenberger involution to the Schützenberger involution on the reversed tensor product, gives
q [ 1 , n ] = ξ B n ⊗ ⋯ ⊗ B 1 ∘ Θ . (16) q_{[1,n]}=\xi_{B_n\otimes\cdots\otimes B_1}\circ\Theta. \tag*{(16)} q [ 1 , n ] = ξ B n ⊗ ⋯ ⊗ B 1 ∘ Θ. (16)
Now let x ∈ Inv ( B 1 ⊗ ⋯ ⊗ B n ) x \in\operatorname{Inv}(B_1 \otimes\cdots\otimes B_n) x ∈ Inv ( B 1 ⊗ ⋯ ⊗ B n ) . Since q [ 1 , n ] q_{[1,n]} q [ 1 , n ] is a crystal isomorphism, q [ 1 , n ] x q_{[1,n]}x q [ 1 , n ] x is again invariant, now in B n ⊗ ⋯ ⊗ B 1 B_n \otimes\cdots\otimes B_1 B n ⊗ ⋯ ⊗ B 1 . The Schützenberger involution fixes every invariant. Therefore q [ 1 , n ] x = ξ ( Θ ( x ) ) q_{[1,n]}x = \xi(\Theta(x)) q [ 1 , n ] x = ξ ( Θ ( x )) implies
Θ ( x ) = ξ ( q [ 1 , n ] x ) = q [ 1 , n ] x , \Theta(x) = \xi(q_{[1,n]}x) = q_{[1,n]}x, Θ ( x ) = ξ ( q [ 1 , n ] x ) = q [ 1 , n ] x ,
where in the first equality we use that ξ \xi ξ is an involution, and in the second that q [ 1 , n ] x q_{[1,n]}x q [ 1 , n ] x is invariant. Therefore on invariants the full reversal is exactly
b 1 ⊗ ⋯ ⊗ b n ⟼ ξ ( b n ) ⊗ ⋯ ⊗ ξ ( b 1 ) . b_1 \otimes\cdots\otimes b_n \longmapsto\xi(b_n) \otimes\cdots\otimes\xi(b_1). b 1 ⊗ ⋯ ⊗ b n ⟼ ξ ( b n ) ⊗ ⋯ ⊗ ξ ( b 1 ) .
This is the general form of Lemma 3.2 : on invariants, the full cactus reversal is simply reversal of the factors together with the Schützenberger involution on each factor.
The Bender–Knuth elements as local rules
Fix 1 ≤ i < n 1 \le i < n 1 ≤ i < n , and write
A = B 1 ⊗ ⋯ ⊗ B i − 1 , B = B i , C = B i + 1 , D = B i + 2 ⊗ ⋯ ⊗ B n , A = B_1 \otimes\cdots\otimes B_{i-1}, \qquad B = B_i, \qquad C = B_{i+1}, \qquad D = B_{i+2} \otimes\cdots\otimes B_n, A = B 1 ⊗ ⋯ ⊗ B i − 1 , B = B i , C = B i + 1 , D = B i + 2 ⊗ ⋯ ⊗ B n ,
where an empty tensor product is the unit. We use the convention that the reversal of a one-element or empty interval is the identity:
q [ p , p ] = q [ p , p − 1 ] = 1. q_{[p,p]} = q_{[p,p-1]} = 1. q [ p , p ] = q [ p , p − 1 ] = 1.
Then, uniformly for all 1 ≤ i < n 1 \le i < n 1 ≤ i < n ,
t i = q [ 1 , i ] q [ 1 , i + 1 ] q [ 1 , i ] q [ 1 , i − 1 ] . t_i = q_{[1,i]}q_{[1,i+1]}q_{[1,i]}q_{[1,i-1]}. t i = q [ 1 , i ] q [ 1 , i + 1 ] q [ 1 , i ] q [ 1 , i − 1 ] .
Define the right-anchored analogue
t i ′ : = q [ i + 1 , n ] q [ i , n ] q [ i + 1 , n ] q [ i + 2 , n ] . t'_i := q_{[i+1,n]}q_{[i,n]}q_{[i+1,n]}q_{[i+2,n]}. t i ′ := q [ i + 1 , n ] q [ i , n ] q [ i + 1 , n ] q [ i + 2 , n ] .
The point is that t i t_i t i and t i ′ t'_i t i ′ give two ways to interchange the adjacent factors B B B and C C C : t i t_i t i is anchored on the prefix A A A , while t i ′ t'_i t i ′ is anchored on the suffix D D D .
Lemma 8.9 (local-rule form). As maps
A ⊗ B ⊗ C ⊗ D ⟶ A ⊗ C ⊗ B ⊗ D , A \otimes B \otimes C \otimes D \longrightarrow A \otimes C \otimes B \otimes D, A ⊗ B ⊗ C ⊗ D ⟶ A ⊗ C ⊗ B ⊗ D , we have
t i = [ σ B , A ⊗ C ∘ ( σ A , B ⊗ 1 C ) ] ⊗ 1 D , (17) t_i = \left[\sigma_{B,A\otimes C} \circ(\sigma_{A,B} \otimes1_C)\right] \otimes1_D, \tag*{(17)} t i = [ σ B , A ⊗ C ∘ ( σ A , B ⊗ 1 C ) ] ⊗ 1 D , (17) t i ′ = 1 A ⊗ [ σ B ⊗ D , C ∘ ( 1 B ⊗ σ C , D ) ] . (18) t'_i = 1_A \otimes\left[\sigma_{B\otimes D,C} \circ(1_B \otimes\sigma_{C,D})\right]. \tag*{(18)} t i ′ = 1 A ⊗ [ σ B ⊗ D , C ∘ ( 1 B ⊗ σ C , D ) ] . (18) Moreover, q [ 1 , n ] t i q [ 1 , n ] = t n − i ′ q_{[1,n]}t_iq_{[1,n]} = t'_{n-i} q [ 1 , n ] t i q [ 1 , n ] = t n − i ′ . Thus Theorem 8.1 is equivalent to t i = t i ′ t_i = t'_i t i = t i ′ , 1 ≤ i < n 1 \le i < n 1 ≤ i < n .
Proof. We use the uniform expression t i = q [ 1 , i ] q [ 1 , i + 1 ] q [ 1 , i ] q [ 1 , i − 1 ] t_i = q_{[1,i]}q_{[1,i+1]}q_{[1,i]}q_{[1,i-1]} t i = q [ 1 , i ] q [ 1 , i + 1 ] q [ 1 , i ] q [ 1 , i − 1 ] , with the rightmost factor acting first. By (11), σ A , B ⊗ 1 C \sigma_{A,B} \otimes1_C σ A , B ⊗ 1 C interchanges the blocks [ 1 , i − 1 ] [1,i-1] [ 1 , i − 1 ] and [ i , i ] [i,i] [ i , i ] , and thus
σ A , B ⊗ 1 C = q [ 2 , i ] q [ 1 , i ] = q [ 1 , i ] q [ 1 , i − 1 ] . \sigma_{A,B} \otimes1_C = q_{[2,i]}q_{[1,i]} = q_{[1,i]}q_{[1,i-1]}. σ A , B ⊗ 1 C = q [ 2 , i ] q [ 1 , i ] = q [ 1 , i ] q [ 1 , i − 1 ] .
The second equality is (C3) when the intervals are nondegenerate, and follows directly from our convention at the boundary. These are precisely the two rightmost factors in t i t_i t i .
After this interchange the factors are ordered as B , A , C B,A,C B , A , C . The map σ B , A ⊗ C \sigma_{B,A\otimes C} σ B , A ⊗ C moves B B B past the block A ⊗ C A \otimes C A ⊗ C , so again by (11),
σ B , A ⊗ C = q [ 1 , i ] q [ 1 , i + 1 ] . \sigma_{B,A\otimes C} = q_{[1,i]}q_{[1,i+1]}. σ B , A ⊗ C = q [ 1 , i ] q [ 1 , i + 1 ] .
These are the two remaining factors in t i t_i t i , proving (12). The proof of (18) is the same argument from the right.
Finally, conjugation by q [ 1 , n ] q_{[1,n]} q [ 1 , n ] reverses the interval labels: q [ 1 , n ] q [ 1 , j ] q [ 1 , n ] = q [ n + 1 − j , n ] q_{[1,n]}q_{[1,j]}q_{[1,n]} = q_{[n+1-j,n]} q [ 1 , n ] q [ 1 , j ] q [ 1 , n ] = q [ n + 1 − j , n ] . Thus conjugating the four factors in t i t_i t i gives q [ 1 , n ] t i q [ 1 , n ] = t n − i ′ q_{[1,n]}t_iq_{[1,n]} = t'_{n-i} q [ 1 , n ] t i q [ 1 , n ] = t n − i ′ .
Westbury [44 Def. 3.1] interprets the Bender–Knuth generators of the cactus group as local rules in a coboundary category. In his notation,
τ B , C A : = σ B , A ⊗ C ∘ ( σ A , B ⊗ 1 C ) . \tau^{A}_{B,C} := \sigma_{B,A\otimes C} \circ(\sigma_{A,B} \otimes1_C). τ B , C A := σ B , A ⊗ C ∘ ( σ A , B ⊗ 1 C ) .
We will also use the right-anchored version
τ ˉ B , C D : = σ B ⊗ D , C ∘ ( 1 B ⊗ σ C , D ) . \bar{\tau}^{D}_{B,C} := \sigma_{B\otimes D,C} \circ(1_B \otimes\sigma_{C,D}). τ ˉ B , C D := σ B ⊗ D , C ∘ ( 1 B ⊗ σ C , D ) .
Thus t i t_i t i acts by the local rule τ B , C A \tau^{A}_{B,C} τ B , C A , while t i ′ t'_i t i ′ acts by the corresponding right-anchored local rule τ ˉ B , C D \bar{\tau}^{D}_{B,C} τ ˉ B , C D . By the lemma, Theorem 8.1 is equivalent to saying that these two local rules agree on invariants.
Remark 8.10. There is a third natural way to interchange B B B and C C C , namely the adjacent commutor
q [ i , i + 1 ] = 1 A ⊗ σ B , C ⊗ 1 D . q_{[i,i+1]} = 1_A \otimes\sigma_{B,C} \otimes1_D. q [ i , i + 1 ] = 1 A ⊗ σ B , C ⊗ 1 D .
In general this is different from both t i t_i t i and t i ′ t'_i t i ′ . Indeed, the equality t i = q [ i , i + 1 ] t_i = q_{[i,i+1]} t i = q [ i , i + 1 ] would amount to
σ B , A ⊗ C = ( 1 A ⊗ σ B , C ) ∘ ( σ B , A ⊗ 1 C ) , \sigma_{B,A\otimes C} = (1_A \otimes\sigma_{B,C}) \circ(\sigma_{B,A} \otimes1_C), σ B , A ⊗ C = ( 1 A ⊗ σ B , C ) ∘ ( σ B , A ⊗ 1 C ) ,
which is the hexagon axiom, and this fails for crystals. Thus there are in general three different maps interchanging the same two adjacent factors: t i t_i t i , t i ′ t'_i t i ′ , and q [ i , i + 1 ] q_{[i,i+1]} q [ i , i + 1 ] . On tableaux, q [ i , i + 1 ] q_{[i,i+1]} q [ i , i + 1 ] is the crystal reflection operator; it satisfies the braid relations, while the Bender–Knuth involutions do not.
The atomic case
We first prove Theorem 8.1 in the case where every factor is the single-box crystal
B 1 = ⋯ = B N = B ( ϖ 1 ) . B_1 = \cdots= B_N = B(\varpi_1). B 1 = ⋯ = B N = B ( ϖ 1 ) .
This is the only place in the proof where tableaux enter. An element of B ( ϖ 1 ) ⊗ N B(\varpi_1)^{\otimes N} B ( ϖ 1 ) ⊗ N is a word
u = u 1 ⋯ u N , u i ∈ [ m ] . u = u_1 \cdots u_N,\qquad u_i \in[m]. u = u 1 ⋯ u N , u i ∈ [ m ] .
Under RSK, the crystal operators preserve the recording tableau Q ( u ) Q(u) Q ( u ) and act on the insertion tableau P ( u ) P(u) P ( u ) [18 ] . Thus the connected components of the crystal are the fibres of Q ( u ) Q(u) Q ( u ) . If Q ( u ) Q(u) Q ( u ) has shape λ \lambda λ , the corresponding component is isomorphic to B ( λ ) B(\lambda) B ( λ ) . Thus, if u u u is invariant, its component must have λ = ( d m ) \lambda= (d^m) λ = ( d m ) for some d d d , and N = d m N = dm N = d m . We therefore obtain a bijection
Inv ( B ( ϖ 1 ) ⊗ N ) ⟷ SYT ( d m ) . (19) \operatorname{Inv}(B(\varpi_1)^{\otimes N}) \longleftrightarrow\operatorname{SYT}(d^m). \tag*{(19)} Inv ( B ( ϖ 1 ) ⊗ N ) ⟷ SYT ( d m ) . (19)
Thus the positions of the tensor factors become the entries 1 , … , N 1,\ldots,N 1 , … , N of a rectangular standard tableau, exactly as in § 3 –§ 6 .
Lemma 8.11 (the atomic dictionary). Under the bijection (19) , the interval reversal q [ 1 , r ] q_{[1,r]} q [ 1 , r ] acts by evacuation on the subtableau formed by the entries 1 , … , r 1,\ldots,r 1 , … , r , therefore by the Bender–Knuth word q r − 1 q_{r-1} q r − 1 . Thus, the element t i ∈ C N t_i \in C_N t i ∈ C N acts by the Bender–Knuth involution t i t_i t i .
Example 8.12. Let m = 3 m = 3 m = 3 , N = 6 N = 6 N = 6 , and take the invariant word u = 332121 u = 332121 u = 332121 , with
Q ( u ) = ( 1 2 3 5 4 6 ) . Q(u) =
\begin{pmatrix}
1 & 2 \\
3 & 5 \\
4 & 6
\end{pmatrix}. Q ( u ) = 1 3 4 2 5 6 . r = 3 r = 3 r = 3 . On the first three factors,
Θ ( 332 ) = rc ( 332 ) = 211. \Theta(332) = \operatorname{rc}(332) = 211. Θ ( 332 ) = rc ( 332 ) = 211.
The word 211 211 211 is highest weight in its component, which is isomorphic to B ( ( 2 , 1 ) ) B((2,1)) B (( 2 , 1 )) . Applying lowering operators,
211 → f 1 212 → f 2 213 → f 2 313 → f 1 323 , 211 \xrightarrow{f_1} 212 \xrightarrow{f_2} 213 \xrightarrow{f_2} 313 \xrightarrow{f_1} 323, 211 f 1 212 f 2 213 f 2 313 f 1 323 ,
so ξ ( 211 ) = 323 \xi(211)=323 ξ ( 211 ) = 323 . Then
q [ 1 , 3 ] ( u ) = 323121 , Q ( 323121 ) = 1 3 2 5 4 6 . q_{[1,3]}(u)=323121,\qquad Q(323121)=\begin{array}{|c|c|}\hline1&3\\\hline2&5\\\hline4&6\\\hline\end{array}. q [ 1 , 3 ] ( u ) = 323121 , Q ( 323121 ) = 1 2 4 3 5 6 .
Thus the subtableau 1 2 3 \begin{array}{|c|c|}\hline1&2\\\hline3&\\\hline\end{array} 1 3 2 is evacuated to 1 3 2 \begin{array}{|c|c|}\hline1&3\\\hline2&\\\hline\end{array} 1 2 3 , while the entries 4 , 5 , 6 4,5,6 4 , 5 , 6 stay in place. r = 6 r=6 r = 6 . Since u u u is invariant, Proposition 8.8 gives
q [ 1 , 6 ] ( u ) = rc ( u ) = 323211. q_{[1,6]}(u)=\operatorname{rc}(u)=323211. q [ 1 , 6 ] ( u ) = rc ( u ) = 323211.
Its recording tableau is
Q ( 323211 ) = 1 3 2 4 5 6 , Q(323211)=\begin{array}{|c|c|}\hline1&3\\\hline2&4\\\hline5&6\\\hline\end{array}, Q ( 323211 ) = 1 2 5 3 4 6 ,
which is the evacuation of Q ( u ) Q(u) Q ( u ) ; on a rectangle, this is rotation by 180 ∘ 180^\circ 18 0 ∘ together with a ↦ 7 − a a\mapsto7-a a ↦ 7 − a , as in Lemma 3.2 .
Proof. Let v = u 1 ⋯ u r v=u_1\cdots u_r v = u 1 ⋯ u r . On the first r r r factors, q [ 1 , r ] q_{[1,r]} q [ 1 , r ] acts as ξ ∘ Θ \xi\circ\Theta ξ ∘ Θ , where Θ ( v ) = rc ( v ) \Theta(v)=\operatorname{rc}(v) Θ ( v ) = rc ( v ) is reverse-complement, with a ↦ m + 1 − a a\mapsto m+1-a a ↦ m + 1 − a . A standard symmetry of RSK gives
P ( rc ( v ) ) = e ( P ( v ) ) , Q ( rc ( v ) ) = e ( Q ( v ) ) P(\operatorname{rc}(v))=e(P(v)),\qquad Q(\operatorname{rc}(v))=e(Q(v)) P ( rc ( v )) = e ( P ( v )) , Q ( rc ( v )) = e ( Q ( v ))
[18 , 40 App. A1] . The Schützenberger involution ξ \xi ξ preserves the recording tableau and acts by evacuation on the insertion tableau. Then
P ( q [ 1 , r ] v ) = P ( v ) , Q ( q [ 1 , r ] v ) = e ( Q ( v ) ) . P(q_{[1,r]}v)=P(v),\qquad Q(q_{[1,r]}v)=e(Q(v)). P ( q [ 1 , r ] v ) = P ( v ) , Q ( q [ 1 , r ] v ) = e ( Q ( v )) .
Since the insertion tableau after the first r r r letters is unchanged, inserting the remaining letters u r + 1 , … , u N u_{r+1},\ldots,u_N u r + 1 , … , u N adds the same boxes as before. Thus q [ 1 , r ] q_{[1,r]} q [ 1 , r ] evacuates the subtableau of Q ( u ) Q(u) Q ( u ) formed by the entries 1 , … , r 1,\ldots,r 1 , … , r and leaves the remaining entries in place. By Lemma 2.4 , this is q r − 1 q_{r-1} q r − 1 .
Using t i = q [ 1 , i ] q [ 1 , i + 1 ] q [ 1 , i ] q [ 1 , i − 1 ] t_i=q_{[1,i]}q_{[1,i+1]}q_{[1,i]}q_{[1,i-1]} t i = q [ 1 , i ] q [ 1 , i + 1 ] q [ 1 , i ] q [ 1 , i − 1 ] in the cactus group, we conclude that its action on tableaux is q i − 1 q i q i − 1 q i − 2 q_{i-1}q_iq_{i-1}q_{i-2} q i − 1 q i q i − 1 q i − 2 . Since q i = q i − 1 p i q_i=q_{i-1}p_i q i = q i − 1 p i and each q j q_j q j is an involution, this is p i p i − 1 p_ip_{i-1} p i p i − 1 , which is precisely the Bender–Knuth involution t i t_i t i on tableaux. □
Proposition 8.13 (the atomic relation). On Inv ( B ( ϖ 1 ) ⊗ N ) \operatorname{Inv}(B(\varpi_1)^{\otimes N}) Inv ( B ( ϖ 1 ) ⊗ N ) ,
t j = t N − j ′ , 1 ≤ j < N . t_j=t'_{N-j},\qquad1\leq j<N. t j = t N − j ′ , 1 ≤ j < N . Proof. By Lemma 8.9 ,
t N − j ′ = q [ 1 , N ] t j q [ 1 , N ] . t'_{N-j}=q_{[1,N]}t_jq_{[1,N]}. t N − j ′ = q [ 1 , N ] t j q [ 1 , N ] . Under the dictionary of Lemma 8.11 , the right-hand side acts on (14) as q t j q qt_jq q t j q . By Theorem 3.4 , this is t N − j t_{N-j} t N − j . Replacing N − j N-j N − j by j j j gives the result. □
Only the standard-tableau case of Theorem 3.4 is used here. In this case t j t_j t j interchanges the entries j j j and j + 1 j+1 j + 1 unless they lie in the same row or column. Thus Lemma 3.3 reduces to the simple observation that rotation by 180 ∘ 180^\circ 18 0 ∘ , together with a ↦ N + 1 − a a\mapsto N+1-a a ↦ N + 1 − a , preserves this condition.
Factorising a block interchange
The local rules factor when one of the two moving blocks is itself a tensor product. This is essentially Westbury’s factorisation [44 Lemmas 2, 3] .
Lemma 8.14 (factorisation). For any crystals A A A , B B B , C C C , D D D ,
τ B , C ⊗ D A = τ B , D A ⊗ C ∘ ( τ B , C A ⊗ 1 D ) . (20) \tau^{A}_{B,C\otimes D}=\tau^{A\otimes C}_{B,D}\circ(\tau^{A}_{B,C}\otimes1_D). \tag*{(20)} τ B , C ⊗ D A = τ B , D A ⊗ C ∘ ( τ B , C A ⊗ 1 D ) . (20) τ B ⊗ C , D A = ( τ B , D A ⊗ 1 C ) ∘ τ C , D A ⊗ B . (21) \tau^{A}_{B\otimes C,D}=(\tau^{A}_{B,D}\otimes1_C)\circ\tau^{A\otimes B}_{C,D}. \tag*{(21)} τ B ⊗ C , D A = ( τ B , D A ⊗ 1 C ) ∘ τ C , D A ⊗ B . (21) The mirror identities hold for τ ˉ \bar{\tau} τ ˉ , with prefixes replaced by suffixes.
Thus (20) moves B B B past C ⊗ D C\otimes D C ⊗ D in two steps: first past C C C , anchored on A A A , and then past D D D , now anchored on A ⊗ C A\otimes C A ⊗ C . Similarly, (21) moves D D D left past B ⊗ C B\otimes C B ⊗ C by first crossing C C C and then B B B .
Proof. Expanding the right-hand side of (20) gives
σ B , A ⊗ C ⊗ D ∘ ( σ A ⊗ C , B ⊗ 1 D ) ∘ ( σ B , A ⊗ C ⊗ 1 D ) ∘ ( σ A , B ⊗ 1 C ⊗ 1 D ) . \sigma_{B,A\otimes C\otimes D}\circ(\sigma_{A\otimes C,B}\otimes1_D)\circ(\sigma_{B,A\otimes C}\otimes1_D)\circ(\sigma_{A,B}\otimes1_C\otimes1_D). σ B , A ⊗ C ⊗ D ∘ ( σ A ⊗ C , B ⊗ 1 D ) ∘ ( σ B , A ⊗ C ⊗ 1 D ) ∘ ( σ A , B ⊗ 1 C ⊗ 1 D ) .
The two middle commutors cancel by Proposition 8.5 (a), leaving τ B , C ⊗ D A \tau^{A}_{B,C\otimes D} τ B , C ⊗ D A .
For (21) , note that
τ C , B A ∘ τ B , C A = id , \tau^{A}_{C,B}\circ\tau^{A}_{B,C}=\operatorname{id}, τ C , B A ∘ τ B , C A = id ,
since both maps are given by the same cactus generator t i t_i t i , and t i 2 = 1 t_i^2=1 t i 2 = 1 . Apply (20) with ( B , C , D ) (B,C,D) ( B , C , D ) replaced by ( D , B , C ) (D,B,C) ( D , B , C ) and invert the resulting identity.
The identities for τ ˉ \bar{\tau} τ ˉ follow by conjugating with the full reversal. □
Now suppose
B = X 1 ⊗ ⋯ ⊗ X b , C = Y 1 ⊗ ⋯ ⊗ Y c , B=X_1\otimes\cdots\otimes X_b,\qquad C=Y_1\otimes\cdots\otimes Y_c, B = X 1 ⊗ ⋯ ⊗ X b , C = Y 1 ⊗ ⋯ ⊗ Y c ,
where the X p X_p X p and Y r Y_r Y r are atomic factors, and suppose that A A A contains u u u atomic factors. Thus X p X_p X p initially occupies position u + p u+p u + p , while Y r Y_r Y r occupies position u + b + r u+b+r u + b + r .
Apply (21) first, splitting off X b X_b X b from the left block:
τ B , C A = ( τ X 1 ⊗ ⋯ ⊗ X b − 1 , C A ⊗ 1 X b ) ∘ τ X b , C A ⊗ X 1 ⊗ ⋯ ⊗ X b − 1 . \tau^{A}_{B,C}=\left(\tau^{A}_{X_1\otimes\cdots\otimes X_{b-1},C}\otimes1_{X_b}\right)\circ\tau^{A\otimes X_1\otimes\cdots\otimes X_{b-1}}_{X_b,C}. τ B , C A = ( τ X 1 ⊗ ⋯ ⊗ X b − 1 , C A ⊗ 1 X b ) ∘ τ X b , C A ⊗ X 1 ⊗ ⋯ ⊗ X b − 1 .
Thus X b X_b X b crosses the whole block C C C first, then X b − 1 X_{b-1} X b − 1 does the same, and so on. Applying (20) to each of these block crossings, X p X_p X p crosses Y 1 , Y 2 , … , Y c Y_1,Y_2,\ldots,Y_c Y 1 , Y 2 , … , Y c in that order.
Each map in this decomposition has the form τ X p , Y r P \tau^{P}_{X_p,Y_r} τ X p , Y r P , where P P P is the tensor product of the factors currently to the left of X p , Y r X_p,Y_r X p , Y r ; in particular, the anchor changes after each crossing.
By Lemma 8.9 , if X p X_p X p and Y r Y_r Y r occupy adjacent positions j , j + 1 j,j+1 j , j + 1 , the map τ X p , Y r P \tau^{P}_{X_p,Y_r} τ X p , Y r P is precisely the action of the cactus generator t j t_j t j . Applying the mirror identities in the opposite order, first the mirror of (20) to split off X b X_b X b from B B B and then the mirror of (21) to split C C C , gives the same sequence of crossings for τ ˉ B , C D \bar{\tau}^{D}_{B,C} τ ˉ B , C D : X b X_b X b crosses Y 1 , … , Y c Y_1,\ldots,Y_c Y 1 , … , Y c , then X b − 1 X_{b-1} X b − 1 does the same, and so on, now with each t j t_j t j replaced by t j ′ t'_j t j ′ . We therefore obtain the following.
Corollary 8.15. The maps τ B , C A \tau^{A}_{B,C} τ B , C A and τ ˉ B , C D \bar{\tau}^{D}_{B,C} τ ˉ B , C D factor into the same sequence of b c bc b c adjacent box interchanges. In chronological order, their indices are
u + b , … , u + b + c − 1 ; u + b − 1 , … , u + b + c − 2 ; … ; u + 1 , … , u + c . (22) u+b,\ldots,u+b+c-1;\ u+b-1,\ldots,u+b+c-2;\ \ldots;\ u+1,\ldots,u+c. \tag*{(22)} u + b , … , u + b + c − 1 ; u + b − 1 , … , u + b + c − 2 ; … ; u + 1 , … , u + c . (22) For τ B , C A \tau^{A}_{B,C} τ B , C A the corresponding factors are t j t_j t j , while for τ ˉ B , C D \bar{\tau}^{D}_{B,C} τ ˉ B , C D they are t j ′ t'_j t j ′ .
For example, when b = c = 2 b=c=2 b = c = 2 and u = 0 u=0 u = 0 , both factorizations have chronological index word ( 2 , 3 , 1 , 2 ) (2,3,1,2) ( 2 , 3 , 1 , 2 ) .
Propagation from the atomic case.
Lemma 8.16 (atomic propagation). Suppose that t j = t j ′ t_j=t'_j t j = t j ′ on Inv ( B ( ϖ 1 ) ⊗ N ) \operatorname{Inv}(B(\varpi_1)^{\otimes N}) Inv ( B ( ϖ 1 ) ⊗ N ) for every N N N and every j j j . Then t i = t i ′ t_i=t'_i t i = t i ′ on
Inv ( B ( λ 1 ) ⊗ ⋯ ⊗ B ( λ n ) ) \operatorname{Inv}\left(B(\lambda_1)\otimes\cdots\otimes B(\lambda_n)\right) Inv ( B ( λ 1 ) ⊗ ⋯ ⊗ B ( λ n ) ) for all λ 1 , … , λ n \lambda_1,\ldots,\lambda_n λ 1 , … , λ n and all i i i .
Proof. Fix i i i , and write A , B , C , D A,B,C,D A , B , C , D as in § 9.1 , with
B = B ( λ i ) , C = B ( λ i + 1 ) . B=B(\lambda_i), \qquad C=B(\lambda_{i+1}). B = B ( λ i ) , C = B ( λ i + 1 ) .
Refine into boxes. For each j j j , use the row-reading embedding
ι j : B ( λ j ) ↪ B ( ϖ 1 ) ⊗ ∣ λ j ∣ , \iota_j:B(\lambda_j)\hookrightarrow B(\varpi_1)^{\otimes|\lambda_j|}, ι j : B ( λ j ) ↪ B ( ϖ 1 ) ⊗ ∣ λ j ∣ ,
where the rows of a tableau are read from left to right, starting with the bottom row and moving upward. Its image is a connected component of B ( ϖ 1 ) ⊗ ∣ λ j ∣ B(\varpi_1)^{\otimes|\lambda_j|} B ( ϖ 1 ) ⊗ ∣ λ j ∣ . Tensoring these inclusions gives
ι : B ( λ 1 ) ⊗ ⋯ ⊗ B ( λ n ) ↪ B ( ϖ 1 ) ⊗ N , N = ∑ j ∣ λ j ∣ . \iota:B(\lambda_1)\otimes\cdots\otimes B(\lambda_n)\hookrightarrow B(\varpi_1)^{\otimes N}, \qquad N=\sum_j|\lambda_j|. ι : B ( λ 1 ) ⊗ ⋯ ⊗ B ( λ n ) ↪ B ( ϖ 1 ) ⊗ N , N = j ∑ ∣ λ j ∣.
The image is a union of connected components, so invariants map to invariants. Write A ~ , B ~ , C ~ , D ~ \widetilde A,\widetilde B,\widetilde C,\widetilde D A , B , C , D for the tensor products of single-box crystals obtained by refining the factors in A , B , C , D A,B,C,D A , B , C , D . Thus
B ~ = B ( ϖ 1 ) ⊗ b , C ~ = B ( ϖ 1 ) ⊗ c , b = ∣ λ i ∣ , c = ∣ λ i + 1 ∣ . \widetilde B=B(\varpi_1)^{\otimes b}, \qquad\widetilde C=B(\varpi_1)^{\otimes c}, \qquad b=|\lambda_i|,\qquad c=|\lambda_{i+1}|. B = B ( ϖ 1 ) ⊗ b , C = B ( ϖ 1 ) ⊗ c , b = ∣ λ i ∣ , c = ∣ λ i + 1 ∣.
Let
ι s r c : A ⊗ B ⊗ C ⊗ D ↪ A ~ ⊗ B ~ ⊗ C ~ ⊗ D ~ \iota_{\mathrm{src}}:A\otimes B\otimes C\otimes D\hookrightarrow\widetilde A\otimes\widetilde B\otimes\widetilde C\otimes\widetilde D ι src : A ⊗ B ⊗ C ⊗ D ↪ A ⊗ B ⊗ C ⊗ D
be the tensor product of the row-reading embeddings above, and let
ι t g t : A ⊗ C ⊗ B ⊗ D ↪ A ~ ⊗ C ~ ⊗ B ~ ⊗ D ~ \iota_{\mathrm{tgt}}:A\otimes C\otimes B\otimes D\hookrightarrow\widetilde A\otimes\widetilde C\otimes\widetilde B\otimes\widetilde D ι tgt : A ⊗ C ⊗ B ⊗ D ↪ A ⊗ C ⊗ B ⊗ D
be the corresponding embedding after the two middle factors have been interchanged. Set
L : = τ B ~ , C ~ A ~ , R : = τ ‾ B ~ , C ~ D ~ . L:=\tau^{\widetilde A}_{\widetilde B,\widetilde C}, \qquad R:=\overline{\tau}^{\widetilde D}_{\widetilde B,\widetilde C}. L := τ B , C A , R := τ B , C D .
Thus L L L and R R R interchange the two blocks of b b b and c c c single-box factors corresponding to B B B and C C C , anchored respectively on the refined blocks A ~ \widetilde A A and D ~ \widetilde D D . Since the ι j \iota_j ι j are inclusions of connected components, Lemma 8.6 gives
ι t g t ∘ τ B , C A = L ∘ ι s r c , ι t g t ∘ τ ‾ B , C D = R ∘ ι s r c . \iota_{\mathrm{tgt}}\circ\tau^A_{B,C}=L\circ\iota_{\mathrm{src}}, \qquad\iota_{\mathrm{tgt}}\circ\overline{\tau}^D_{B,C}=R\circ\iota_{\mathrm{src}}. ι tgt ∘ τ B , C A = L ∘ ι src , ι tgt ∘ τ B , C D = R ∘ ι src .
Thus the local rules on B B B and C C C are the restrictions of the corresponding block interchanges after the factors have been refined into single boxes.
Common factorization. By Corollary 8.15 , L L L and R R R factor into the same chronological sequence of adjacent positions, given by (22) . At each step the factor for L L L is t j t_j t j , while the corresponding factor for R R R is t j ′ t'_j t j ′ . Starting from an invariant, the two intermediate tensors agree inductively: every local rule is a crystal isomorphism and hence preserves invariants, and the hypothesis gives t j = t j ′ t_j=t'_j t j = t j ′ on invariants at the next crossing. Therefore L = R L=R L = R on Inv ( B ( ϖ 1 ) ⊗ N ) \operatorname{Inv}(B(\varpi_1)^{\otimes N}) Inv ( B ( ϖ 1 ) ⊗ N ) .
Pull back. If
x ∈ Inv ( B ( λ 1 ) ⊗ ⋯ ⊗ B ( λ n ) ) , x\in\operatorname{Inv}\left(B(\lambda_1)\otimes\cdots\otimes B(\lambda_n)\right), x ∈ Inv ( B ( λ 1 ) ⊗ ⋯ ⊗ B ( λ n ) ) ,
then ι s r c ( x ) \iota_{\mathrm{src}}(x) ι src ( x ) is an invariant of B ( ϖ 1 ) ⊗ N B(\varpi_1)^{\otimes N} B ( ϖ 1 ) ⊗ N . By the naturality identities above,
ι t g t ( τ B , C A x ) = L ( ι s r c ( x ) ) , ι t g t ( τ ‾ B , C D x ) = R ( ι s r c ( x ) ) . \iota_{\mathrm{tgt}}\left(\tau^A_{B,C}x\right)=L\left(\iota_{\mathrm{src}}(x)\right), \qquad\iota_{\mathrm{tgt}}\left(\overline{\tau}^D_{B,C}x\right)=R\left(\iota_{\mathrm{src}}(x)\right). ι tgt ( τ B , C A x ) = L ( ι src ( x ) ) , ι tgt ( τ B , C D x ) = R ( ι src ( x ) ) .
We have just proved that L = R L=R L = R on invariants, so L ( ι s r c ( x ) ) = R ( ι s r c ( x ) ) L(\iota_{\mathrm{src}}(x))=R(\iota_{\mathrm{src}}(x)) L ( ι src ( x )) = R ( ι src ( x )) . Combining these equalities gives ι t g t ( τ B , C A x ) = ι t g t ( τ ‾ B , C D x ) \iota_{\mathrm{tgt}}(\tau^{A}_{B,C}x)=\iota_{\mathrm{tgt}}(\overline{\tau}^{D}_{B,C}x) ι tgt ( τ B , C A x ) = ι tgt ( τ B , C D x ) . Since ι t g t \iota_{\mathrm{tgt}} ι tgt is injective,
τ B , C A x = τ ‾ B , C D x . \tau^{A}_{B,C}x=\overline{\tau}^{D}_{B,C}x. τ B , C A x = τ B , C D x .
By Lemma 8.9, this is exactly t i x = t i ′ x t_{i}x=t'_{i}x t i x = t i ′ x .
Proof of Theorem 8.1. A finite normal crystal is a disjoint union of irreducible crystals, and the cactus action respects this decomposition. We may therefore assume that every B j B_j B j is irreducible. Proposition 8.13 gives the hypothesis of Lemma 8.16, so t j = t j ′ t_j=t'_j t j = t j ′ on invariants for every j j j . Lemma 8.9 now gives
q [ 1 , n ] t i q [ 1 , n ] = t n − i ′ = t n − i , q_{[1,n]}t_iq_{[1,n]}=t'_{n-i}=t_{n-i}, q [ 1 , n ] t i q [ 1 , n ] = t n − i ′ = t n − i ,
which is the desired Dynkin-reversal relation.
Proof of Theorem 1.9. By Theorem 8.1, the cactus action on invariants satisfies the Dynkin-reversal relations, and thus factors through R C n RC_n R C n . By Theorem 5.4,
Q D = q [ 1 , n ] Q_D=q_{[1,n]} Q D = q [ 1 , n ]
in R C n RC_n R C n for every Dynkin path D D D . By Lemma 4.5, the words in K ( D ) \mathcal{K}(D) K ( D ) are exactly the reduced words for w 0 w_0 w 0 adapted to the orientation Ω D \Omega_D Ω D , and by Lemma 4.2 every such word represents Q D Q_D Q D . Thus every adapted reduced word acts on invariants as q [ 1 , n ] q_{[1,n]} q [ 1 , n ] . Proposition 8.8 identifies this action with
b 1 ⊗ ⋯ ⊗ b n ⟼ ξ ( b n ) ⊗ ⋯ ⊗ ξ ( b 1 ) , b_1\otimes\cdots\otimes b_n\longmapsto\xi(b_n)\otimes\cdots\otimes\xi(b_1), b 1 ⊗ ⋯ ⊗ b n ⟼ ξ ( b n ) ⊗ ⋯ ⊗ ξ ( b 1 ) ,
as claimed.
A topological proof and extension to other types
Theorem 7.2 gives a second proof of Theorem 8.1, valid in every semisimple type. Let g \mathfrak{g} g be a complex semisimple Lie algebra and λ = ( λ 1 , … , λ n ) \lambda=(\lambda_1,\ldots,\lambda_n) λ = ( λ 1 , … , λ n ) a tuple of dominant integral weights. Set
W λ = ( V ( λ 1 ) ⊗ ⋯ ⊗ V ( λ n ) ) g . W_\lambda=\left(V(\lambda_1)\otimes\cdots\otimes V(\lambda_n)\right)^{\mathfrak{g}}. W λ = ( V ( λ 1 ) ⊗ ⋯ ⊗ V ( λ n ) ) g .
For z ∈ X n = M ‾ 0 , n + 1 ( R ) z\in X_n=\overline{M}_{0,n+1}(\mathbb{R}) z ∈ X n = M 0 , n + 1 ( R ) , the rational Gaudin algebra A z A_z A z acts on W λ W_\lambda W λ with simple spectrum. Its joint eigenlines form an unramified covering E λ 0 → X n \mathcal{E}^{0}_{\lambda}\to X_n E λ 0 → X n . The Gaudin monodromy theorem [25 Theorems 1.4 and 8.5] identifies its monodromy with the cactus action on Inv ( B ( λ 1 ) ⊗ ⋯ ⊗ B ( λ n ) ) \operatorname{Inv}(B(\lambda_1)\otimes\cdots\otimes B(\lambda_n)) Inv ( B ( λ 1 ) ⊗ ⋯ ⊗ B ( λ n )) . The identification is compatible with permuting the marked points and the labeled factors. For unequal factors we consider all orderings, as in § 8.2 .
Lemma 8.17. Forgetting the root f : X n → Y n = M ‾ 0 , n ( R ) f:X_n\to Y_n=\overline{M}_{0,n}(\mathbb{R}) f : X n → Y n = M 0 , n ( R ) gives
E λ 0 ≅ f ∗ F λ \mathcal{E}^{0}_{\lambda}\cong f^{*}\mathcal{F}_{\lambda} E λ 0 ≅ f ∗ F λ for an unramified covering F λ → Y n \mathcal{F}_{\lambda}\to Y_n F λ → Y n . This descent is compatible with simultaneous permutations of the markings and the labeled factors.
Proof. Put U = U ( g ) ⊗ n \mathcal{U}=U(\mathfrak{g})^{\otimes n} U = U ( g ) ⊗ n . Since the diagonal copy Δ ( g ) \Delta(\mathfrak{g}) Δ ( g ) annihilates W λ W_\lambda W λ , the action of U g \mathcal{U}^{\mathfrak{g}} U g on this space factors through the quantum Hamiltonian reduction
( U / U Δ ( g ) ) g , \left(\mathcal{U}/\mathcal{U}\Delta(\mathfrak{g})\right)^{\mathfrak{g}}, ( U / U Δ ( g ) ) g ,
where U Δ ( g ) \mathcal{U}\Delta(\mathfrak{g}) U Δ ( g ) is a left ideal. By [29 Section 2.6 and Theorem 4.15(2)] , the image of the full homogeneous Gaudin algebra in this reduction is invariant under simultaneous projective transformations of its marked points. For semisimple g \mathfrak{g} g , apply this statement to each simple summand. Hence, on a smooth curve, the image in End ( W λ ) \operatorname{End}(W_\lambda) End ( W λ ) , and therefore its joint eigenline decomposition, depends only on the n n n original markings, not on the additional root.
Here is the extension to the boundary. Using the section σ n \sigma_n σ n from the proof of Theorem 7.2, define F λ = σ n ∗ E λ 0 \mathcal{F}_{\lambda}=\sigma_n^*\mathcal{E}_{\lambda}^{0} F λ = σ n ∗ E λ 0 . Let D ( z ) D(z) D ( z ) be the unordered set of joint eigenlines at z z z , viewed in Sym dim W λ P ( W λ ) \operatorname{Sym}^{\dim W_{\lambda}}\mathbb{P}(W_{\lambda}) Sym d i m W λ P ( W λ ) ; the case W λ = 0 W_{\lambda}=0 W λ = 0 is immediate. This set varies continuously and has distinct elements, including at the boundary. For a smooth z z z , move the root towards the marking n n n . Root-independence and continuity give
D ( z ) = D ( σ n ( f ( z ) ) ) . D(z)=D\bigl(\sigma_n(f(z))\bigr). D ( z ) = D ( σ n ( f ( z )) ) .
Both sides are continuous, so density of the smooth real locus gives the equality for every z ∈ X n z\in X_n z ∈ X n . Identifying the same eigenlines proves the asserted pullback isomorphism. Its description is independent of the chosen lift of a point of Y n Y_n Y n , so the permutation maps on the original coverings descend as well. □
Theorem 8.18. For any complex semisimple g \mathfrak g g and finite normal g \mathfrak g g -crystals B 1 , … , B n B_1,\ldots,B_n B 1 , … , B n , n ≥ 3 n\geq3 n ≥ 3 , the cactus action on their invariant tensors factors through
C n → f ∗ J n ′ ≅ R C n . C_n \xrightarrow{f_*} J'_n \cong RC_n. C n f ∗ J n ′ ≅ R C n . In particular, Theorem 8.1 holds in every semisimple type.
Proof. For irreducible factors, apply the Gaudin monodromy theorem to Lemma 8.17. Monodromy of a pullback is obtained by composing with the induced map on fundamental groups. Thus the action factors through f ∗ f_* f ∗ , whose kernel is given by Theorem 7.2. Naturality of the commutor for inclusions of connected components reduces arbitrary finite normal crystals to this case. □
Corollary 8.19. Theorem 1.9 holds for finite normal crystals of any complex semisimple Lie algebra: every adapted reduced word acts on invariant tensors as
b 1 ⊗ ⋯ ⊗ b n ⟼ ξ B n ( b n ) ⊗ ⋯ ⊗ ξ B 1 ( b 1 ) . b_1\otimes\cdots\otimes b_n\longmapsto\xi_{B_n}(b_n)\otimes\cdots\otimes\xi_{B_1}(b_1). b 1 ⊗ ⋯ ⊗ b n ⟼ ξ B n ( b n ) ⊗ ⋯ ⊗ ξ B 1 ( b 1 ) .
Proof. Use Theorems 8.18 and 5.4 . The proof of Proposition 8.8 uses only the commutor formula and the fact that the Schützenberger involution fixes singleton components, so it applies in every semisimple type. □
The diagram A n − 1 A_{n-1} A n − 1 in this corollary indexes tensor positions, not the simple roots of g \mathfrak g g . The invariant condition is essential: on a general multiplicity space Hom g ( V ( μ ) , ⨂ i V ( λ i ) ) \operatorname{Hom}_{\mathfrak g}(V(\mu),\bigotimes_i V(\lambda_i)) Hom g ( V ( μ ) , ⨂ i V ( λ i )) , the root carries V ( μ ) ∗ V(\mu)^* V ( μ ) ∗ rather than the trivial representation, so the preceding root-forgetting argument does not apply. The assertions for n ≤ 2 n\leq2 n ≤ 2 follow directly from the commutor formula.
Rectangular tableaux as invariants
Finally, we explain how the rectangular tableaux of § 3 –§ 6 arise from Theorem 1.9 when the factors are single rows.
Let R a = B ( a ϖ 1 ) R_a=B(a\varpi_1) R a = B ( a ϖ 1 ) be the crystal of a single row of length a a a . Under the row-reading embedding, an element of R a R_a R a is simply a weakly increasing word of length a a a . Thus
r 1 ⊗ ⋯ ⊗ r n ∈ R c 1 ⊗ ⋯ ⊗ R c n r_1\otimes\cdots\otimes r_n\in R_{c_1}\otimes\cdots\otimes R_{c_n} r 1 ⊗ ⋯ ⊗ r n ∈ R c 1 ⊗ ⋯ ⊗ R c n
refines to the word
u = r 1 ⋯ r n ∈ B ( ω ‾ 1 ) ⊗ N , N = c 1 + ⋯ + c n . u=r_1\cdots r_n\in B(\overline{\omega}_1)^{\otimes N},\qquad N=c_1+\cdots+c_n. u = r 1 ⋯ r n ∈ B ( ω 1 ) ⊗ N , N = c 1 + ⋯ + c n .
Set
C j = c 1 + ⋯ + c j , C 0 = 0. C_j=c_1+\cdots+c_j,\qquad C_0=0. C j = c 1 + ⋯ + c j , C 0 = 0.
Under RSK, the connected component of u u u is determined by its recording tableau Q ( u ) Q(u) Q ( u ) . Since r j r_j r j is weakly increasing, the entries
C j − 1 + 1 , … , C j C_{j-1}+1,\ldots,C_j C j − 1 + 1 , … , C j
form a horizontal strip in Q ( u ) Q(u) Q ( u ) , with the entries increasing from left to right. Conversely, this condition is exactly what is needed for the corresponding block of the word to be weakly increasing. Replacing all entries in this interval by j j j , for each j j j , therefore gives a semistandard tableau T ( u ) T(u) T ( u ) of the same shape and content ( c 1 , … , c n ) (c_1,\ldots,c_n) ( c 1 , … , c n ) , and standardisation recovers Q ( u ) Q(u) Q ( u ) . The word u u u is invariant exactly when its RSK component is a one-element s l m \mathfrak{sl}_m sl m -crystal, or equivalently when Q ( u ) Q(u) Q ( u ) has rectangular shape ( k m ) (k^m) ( k m ) . Therefore
Inv ( R c 1 ⊗ ⋯ ⊗ R c n ) ⟷ SSYT ( ( k m ) , n ) content ( c 1 , … , c n ) (23) \operatorname{Inv}\left(R_{c_1}\otimes\cdots\otimes R_{c_n}\right)\longleftrightarrow\operatorname{SSYT}\left((k^m),n\right)_{\operatorname{content}(c_1,\ldots,c_n)} \tag*{(23)} Inv ( R c 1 ⊗ ⋯ ⊗ R c n ) ⟷ SSYT ( ( k m ) , n ) content ( c 1 , … , c n ) (23)
Thus the j j j -th row factor corresponds to the letter j j j in the tableau. The same description identifies the cactus action. First note that the full reversal of a single row factor acts trivially on its row-reading image:
ρ ( r ) = ξ ( rc ( r ) ) = r . \rho(r)=\xi(\operatorname{rc}(r))=r. ρ ( r ) = ξ ( rc ( r )) = r .
Consequently, after refining the first k k k row factors into boxes, the coarse reversal q [ 1 , k ] q_{[1,k]} q [ 1 , k ] is the restriction of the fine reversal q [ 1 , C k ] q_{[1,C_k]} q [ 1 , C k ] on the first C k C_k C k boxes: by (11), the only additional maps are the internal reversals of the row blocks, and these act trivially. By Lemma 8.11, the fine reversal q [ 1 , C k ] q_{[1,C_k]} q [ 1 , C k ] evacuates the subtableau of Q ( u ) Q(u) Q ( u ) formed by the entries 1 , … , C k 1,\ldots,C_k 1 , … , C k . Evacuation commutes with standardisation, by the compatibility of RSK with standardisation and reverse-complement. Hence after destandardisation this becomes evacuation of the letters 1 , … , k 1,\ldots,k 1 , … , k in T ( u ) T(u) T ( u ) . In the notation of the first part of the paper, this is precisely q k − 1 q_{k-1} q k − 1 . Thus, under (23) ,
q [ 1 , k ] ⟼ q k − 1 . q_{[1,k]}\longmapsto q_{k-1}. q [ 1 , k ] ⟼ q k − 1 .
In particular, q [ 1 , n ] q_{[1,n]} q [ 1 , n ] acts as q n − 1 = q q_{n-1}=q q n − 1 = q , which on a rectangular tableau is θ \theta θ by Lemma 3.2. Thus Proposition 8.8 recovers that lemma. Similarly,
t j = q [ 1 , j ] q [ 1 , j + 1 ] q [ 1 , j ] q [ 1 , j − 1 ] t_j=q_{[1,j]}q_{[1,j+1]}q_{[1,j]}q_{[1,j-1]} t j = q [ 1 , j ] q [ 1 , j + 1 ] q [ 1 , j ] q [ 1 , j − 1 ]
acts as
q j − 1 q j q j − 1 q j − 2 = t j , q_{j-1}q_jq_{j-1}q_{j-2}=t_j, q j − 1 q j q j − 1 q j − 2 = t j ,
the Bender–Knuth involution on T ( u ) T(u) T ( u ) . It follows that B K R n \mathrm{BKR}_n BKR n is the image of C n C_n C n on the invariants of tensor products of rows, and Theorem 5.5 is precisely Theorem 1.9 for row factors.
Related work
Bender–Knuth groups
The group B K n \mathrm{BK}_n BK n was introduced by Berenstein and Kirillov [6 ] , who found, among other relations, ( t 1 t 2 ) 6 = 1 (t_1t_2)^6=1 ( t 1 t 2 ) 6 = 1 . Its connection with the cactus group was established independently in [12 ] , [24 ] and [7 ] .
Several related Bender–Knuth and partial-evacuation actions have since been studied. Rodrigues [39 ] defines a shifted Berenstein–Kirillov group and shows that it is a quotient of the cactus group. Azenhas, Tarighat Feller and Torres [3 ] define a symplectic Berenstein–Kirillov group acting on Kashiwara–Nakashima tableaux and exhibit two relations which do not follow from the defining relations of the symplectic cactus group. Liao and the third author [33 ] study the image of the cactus group on standard Young tableaux of a fixed shape. Bender–Knuth moves on linear extensions of posets are studied in [11 ] , where the cactus relations are shown to hold for all minuscule posets.
The rectangular group B K R n \mathrm{BKR}_n BKR n fits naturally into this picture. Conjecture 3.11 says that the additional relations obtained by restricting from arbitrary tableaux to rectangles are exactly the Dynkin-reversal relations.
The big cactus group
The big cactus group J n ′ = π 1 S n ( M ‾ 0 , n ( R ) ) J'_n = \pi_1^{S_n}(\overline{M}_{0,n}(\mathbb{R})) J n ′ = π 1 S n ( M 0 , n ( R )) was introduced by Henriques [26 ] . The space M ‾ 0 , n ( R ) \overline{M}_{0,n}(\mathbb{R}) M 0 , n ( R ) is tiled by associahedra, with one tile for each dihedral order of the markings, and is aspherical [13 ] . Davis, Januszkiewicz and Scott studied the symmetry group Aut ( X ) \operatorname{Aut}(X) Aut ( X ) of the corresponding associahedral tiling of the universal cover, which in our notation identifies with J n ′ J'_n J n ′ , and gave a presentation with generators indexed by subpolygons of an n n n -gon and by its diagonals [13 Thm. 8.6.2 and Rem. 8.6.5] . Theorem 7.2 instead presents J n ′ J'_n J n ′ as the quotient of the ordinary cactus group by the n − 1 n-1 n − 1 Dynkin-reversal relations, and Corollary 7.3 gives a second presentation by adapted words. The proof uses the mirabolic cactus group and the caterpillar realization of the elements t i t_i t i from forthcoming work of Gurenkova, Halacheva and the third author [22 ] .
Quivers and toggles
In the language of Garver, Patrias and Thomas [19 ] , a flip of a Dynkin path at a source corresponds to reflection at a source of the associated quiver. The knitting region R ( D , D ‾ ) R(D,\overline{D}) R ( D , D ) can be identified with the Auslander–Reiten quiver of rep ( Ω D ) \operatorname{rep}(\Omega_D) rep ( Ω D ) ; see Remark 4.6. Garver, Patrias and Thomas show that source reflections are realized by piecewise-linear toggles on minuscule reverse plane partitions. Our knitting argument similarly follows source reflections along a full adapted reduced word for w 0 w_0 w 0 ; on rectangles the resulting action is evacuation.
Dranowski, Elek, Kamnitzer and Morton-Ferguson [14 ] develop a uniform toggle model for minuscule heaps in simply-laced type. Brown, Elek and Halacheva [10 ] relate this toggle action to the cactus action for the crystals B ( n ϖ 1 ) B(n\varpi_1) B ( n ϖ 1 ) in type D D D . This suggests asking for an analogue of Theorem 5.4 in other types.
Sorting networks
Warrington [42 ] introduced the four reentrant subnetworks X X X as a combinatorial analogue of Sylvester’s four-point problem, using the Goodman–Pollack correspondence with planar point configurations [20 ] . He conjectured that a random 4-subnetwork of a random sorting network is reentrant with probability 1/4; this was proved by Angel and Holroyd [1 ] . Meng [35 ] studied X X X -avoiding reduced words and asked for their number.
Theorem 4.9 gives one direction of a surprisingly close connection: every adapted reduced word is X X X -avoiding. Conjecture 4.10 says that the converse also holds. We verified the conjecture directly for n ≤ 6 n \le6 n ≤ 6 , and for n = 7 n = 7 n = 7 by comparison of counts.
If the conjecture holds in general, Meng’s question becomes the enumeration of adapted reduced words, equivalently of linear extensions of the Auslander–Reiten quivers associated to the orientations of A n − 1 A_{n-1} A n − 1 . We have no conceptual explanation for why adaptedness should be governed by a convexity condition on four points.
Promotion
Rectangular promotion periodicity and some of its proofs are discussed in Remark 6.3. On tableaux, the identity of Theorem 6.1 is the Bender–Knuth form of the classical relation between promotion, evacuation and dual evacuation; see Stanley’s survey [41 ] .
Crystals and invariant tensors
The cactus action on tensor products of crystals was constructed by Henriques and Kamnitzer [27 ] . In joint work with Halacheva, Kamnitzer and Weekes, the third author [25 ] proved Etingof’s conjecture identifying this action with the monodromy of Bethe eigenvectors for Gaudin algebras on tensor-product multiplicity spaces. The invariants considered in § 8 correspond to the multiplicity space of the trivial representation.
Halacheva [24 ] used skew Howe duality to identify the internal cactus action on tableaux with the external action on tensor products of column crystals. For row crystals, § 9.6 gives the analogous identification directly by RSK and standardisation. Neither identification is used in the proof of Theorem 8.1 .
Westbury [44 ] introduced local rules in coboundary categories; the factorisation in Lemma 8.14 is essentially his factorisation of these local rules. Pfannerer, Rubey and Westbury [37 ] use local rules and growth diagrams to study promotion and evacuation on highest weight words, relating them to rotation and reversal of chord diagrams and permutations.
There are also several general periodicity results for promotion on invariants. For tensor powers of a single crystal, in arbitrary semisimple type, Westbury [43 ] obtains promotion periodicity as part of a cyclic sieving phenomenon. Fontaine and Kamnitzer [16 ] prove a corresponding result for rotation on invariant vectors in tensor products of minuscule representations.
Theorems 6.1 and 8.18 also give promotion periodicity for arbitrary finite normal crystals B 1 , … , B n B_1,\ldots,B_n B 1 , … , B n : the n n n -fold composite of the promotion maps through the cyclic orderings of the factors is the identity on Inv ( B 1 ⊗ ⋯ ⊗ B n ) \operatorname{Inv}(B_1\otimes\cdots\otimes B_n) Inv ( B 1 ⊗ ⋯ ⊗ B n ) .
Finally, Borodin [9 ] studies the cactus action on s l 2 \mathfrak{sl}_2 sl 2 multiplicity sets through arc diagrams and finds additional relations, including braid relations.
Affine cactus groups and Kirillov–Reshetikhin crystals
For 1 ≤ r < n 1\le r<n 1 ≤ r < n , the tableaux SSYT ( ( s r ) , n ) \operatorname{SSYT}((s^r),n) SSYT (( s r ) , n ) form the underlying finite-type crystal of the Kirillov–Reshetikhin crystal B r , s B^{r,s} B r , s of type A n − 1 ( 1 ) A_{n-1}^{(1)} A n − 1 ( 1 ) . Under the reality and separation hypotheses of their construction, Krylov, Mashanova-Golikova and the third author [31 Theorems 17.4 and 18.2] identify Yangian Bethe eigenlines with tensor products of Kirillov–Reshetikhin crystals and identify the interval monodromies with partial Schützenberger involutions for finite-type subdiagrams of the affine Dynkin diagram. This extends the internal cactus action to the affine setting. A complementary monodromy description for the trigonometric Gaudin model, in terms of crystal commutors and cyclic permutation of tensor factors, is given in [30 Theorem 1.3(3)] ; this is an external action on tensor positions, rather than an internal action on Dynkin colours.
The affine parameter space is the compact real form Z n = M ‾ 0 , n + 2 c o m p ( R ) Z_n=\overline{M}_{0,n+2}^{\mathrm{comp}}(\mathbb{R}) Z n = M 0 , n + 2 comp ( R ) , on which 1 , … , n 1,\ldots,n 1 , … , n are real markings and 0 , ∞ 0,\infty 0 , ∞ are exchanged by the real structure. Its S ~ n \widetilde{S}_n S n -equivariant fundamental group is the extended affine cactus group A C ~ n \widetilde{AC}_n A C n [28 Theorem 11.12] . There is a natural comparison with the spaces used here. Forgetting 0 , ∞ 0,\infty 0 , ∞ defines F : Z n → Y n F:Z_n\to Y_n F : Z n → Y n , while attaching a rigid three-pointed component carrying this conjugate pair at the root defines j : X n → Z n j:X_n\to Z_n j : X n → Z n . Since the attached component is contracted after the pair is forgotten, F ∘ j = f F\circ j=f F ∘ j = f . Thus, with compatible basepoints, we obtain
C n → j ∗ A C ~ n → F ∗ J n ′ , F ∗ j ∗ = f ∗ . C_n \xrightarrow{j_*} \widetilde{AC}_n \xrightarrow{F_*} J'_n,\qquad F_*j_*=f_*. C n j ∗ A C n F ∗ J n ′ , F ∗ j ∗ = f ∗ .
This compares the groups and parameter spaces, but does not imply that the full Kirillov–Reshetikhin monodromy descends along F F F .
Gabriel Frieden
Desautels Faculty of Management, McGill University, Montreal QC, Canada
gabriel.frieden@mcgill.ca
Pavlo Pylyavskyy
School of Mathematics, University of Minnesota, Minneapolis, Minnesota, United States
ppylyavs@umn.edu
P.P. was supported by a Simons Foundation International grant administered by the Simons Foundation, SFI-MPS-TSM-00013120.
Leonid Rybnikov
Department of Mathematics and Statistics, University of Montreal, Montreal QC, Canada
leonid.rybnikov@umontreal.ca
L.R. was supported by Fondation Courtois and by the NSERC Discovery program.