Introduction
1.1.
Cohomology in topology and geometry is rarely bounded, even in a single degree with coefficients of rank one. A smooth projective complex curve of genus g g g has first cohomology of dimension 2 g 2g 2 g in singular, de Rham and ℓ \ell ℓ -adic cohomology, and of dimension g g g for the structure sheaf. In 1984, Guralnick conjectured [18 Conjecture 2] that the first cohomology of finite groups behaves very differently:
Conjecture (Guralnick). There is an absolute constant C C C such that
dim k H 1 ( G , V ) ≤ C \dim_k H^1(G,V) \le C dim k H 1 ( G , V ) ≤ C
for every finite group G G G , every field k k k , and every faithful absolutely irreducible finite-dimensional k G kG k G -module V V V .
Guralnick’s conjecture concerns degree one. In higher degrees the analogous statement fails for arbitrary finite groups: Guralnick, Kantor, Kassabov and Lubotzky showed that the dimensions of H n ( G , V ) H^n(G,V) H n ( G , V ) are unbounded for every n ≥ 2 n \ge2 n ≥ 2 as G G G ranges over all finite groups and V V V over faithful absolutely irreducible modules [20 Theorems 12.3 and 12.6] . Their examples use wreath products to build large cohomology, and they pose the following question concerning simple groups:
Question I. [20 Question 12.1] For which n ≥ 1 n \ge1 n ≥ 1 is there an absolute constant C n C_n C n such that
dim k H n ( G , V ) ≤ C n \dim_k H^n(G,V) \le C_n dim k H n ( G , V ) ≤ C n
for every finite simple group G G G , every field k k k , and every faithful absolutely irreducible finite-dimensional k G kG k G -module V V V .
We show that no such constant C n C_n C n exists for any n ≥ 1 n \ge1 n ≥ 1 , answering Question I. The case n = 1 n = 1 n = 1 disproves Guralnick’s conjecture. Before stating our results, we recall the developments that sustained interest in the conjecture for more than forty years.
1.2.
Despite its audacity, the conjecture looked safe for almost two decades. Guralnick and Hoffman even suggested that C = 2 C = 2 C = 2 might suffice [19 Conjecture 2] . Only in 2003 did Scott find modules with dim H 1 = 3 \dim H^1 = 3 dim H 1 = 3 , for P S L 6 \mathrm{PSL}_6 PSL 6 in large defining characteristic [33 ] (see also [8 ] ), and six years later 3 was still the largest known value [13 Introduction] . Meanwhile, a sustained programme of positive results gave bounds increasingly independent of the module. Early bounds depended on dim V \dim V dim V [3 , 19 ] . For finite groups of Lie type in defining characteristic, Cline, Parshall and Scott bounded dim H 1 \dim H^1 dim H 1 in terms of the root system alone, uniformly in the characteristic [13 ] . In the algebraic group setting, Parshall and Scott obtained analogous bounds for Ext 1 \operatorname{Ext}^1 Ext 1 between arbitrary irreducible rational modules [30 Theorem 5.1] . Explicit bounds for finite groups of Lie type were obtained by Guralnick and Tiep in cross characteristic [21 ] and by Parker and Stewart in defining characteristic [29 ] . Moreover, a universal bound for finite simple groups would imply one for all finite groups with faithful absolutely irreducible coefficients [20 Lemma 5.2] .
At the 2012 AIM workshop Cohomology bounds and growth rates , Scott and Sprowl, using a new algorithm for computing individual Kazhdan–Lusztig basis elements [34 ] , reached 5 for P S L 7 \mathrm{PSL}_7 PSL 7 (see [28 §4] ), and then Lübeck, running the Kazhdan–Lusztig algorithm at scale, found
dim H 1 ( P S L N , V ) = 3 , 16 , 469 , 36672 ( N = 6 , 7 , 8 , 9 ) \dim H^1(\mathrm{PSL}_N,V) = 3,\ 16,\ 469,\ 36672 \qquad(N = 6,7,8,9) dim H 1 ( PSL N , V ) = 3 , 16 , 469 , 36672 ( N = 6 , 7 , 8 , 9 )
in sufficiently large defining characteristic, and 3,537,142 in type E 6 E_6 E 6 [28 §4.1 and Theorem 4.7] . In light of these examples Guralnick, Hodge, Parshall, and Scott concluded their AIM workshop saying that Guralnick’s conjecture ‘could no longer be believed’ [17 ] . The theory has advanced since, with Guralnick and Tiep’s bounds in terms of sectional rank [22 , 23 ] and the characteristic-free bounds of [4 , 5 ] ; but the question of a uniform bound has remained open for a further fourteen years. Our first main result settles it, thus proving that Guralnick’s conjecture is false.
Theorem A. For every integer r ≥ 2 r \ge2 r ≥ 2 and every sufficiently large prime p p p , the finite simple group PSL 2 r + 1 ( F p ) \operatorname{PSL}_{2r+1}(\mathbb{F}_p) PSL 2 r + 1 ( F p ) has a faithful absolutely irreducible module V r , p V_{r,p} V r , p over F p \mathbb{F}_p F p such that
dim F p H 1 ( PSL 2 r + 1 ( F p ) , V r , p ) = r . \dim_{\mathbb{F}_p} H^1(\operatorname{PSL}_{2r+1}(\mathbb{F}_p), V_{r,p}) = r. dim F p H 1 ( PSL 2 r + 1 ( F p ) , V r , p ) = r . The module V r , p V_{r,p} V r , p is obtained by first restricting the simple SL 2 r + 1 ( F ‾ p ) \operatorname{SL}_{2r+1}(\overline{\mathbb{F}}_p) SL 2 r + 1 ( F p ) -module labelled by the weight
λ r , p = 2 ϖ r − 1 + ( p − 2 ) ( ϖ r + ϖ r + 1 ) + 2 ϖ r + 2 \lambda_{r,p} = 2\varpi_{r-1} + (p - 2)(\varpi_r + \varpi_{r+1}) + 2\varpi_{r+2} λ r , p = 2 ϖ r − 1 + ( p − 2 ) ( ϖ r + ϖ r + 1 ) + 2 ϖ r + 2 to the finite subgroup SL 2 r + 1 ( F p ) \operatorname{SL}_{2r+1}(\mathbb{F}_p) SL 2 r + 1 ( F p ) , and then descending to the projective quotient.
The computed examples reached dimension in the millions, but they did not lead to a uniform family that could be proven unbounded. Theorem A supplies such a family with an explicit highest weight and an exact dimension formula. Theorem A goes beyond unboundedness: it implies that every nonnegative integer occurs as the dimension of first cohomology of a finite simple group with faithful absolutely irreducible coefficients. This solves an inverse problem in the spirit of constructing varieties with prescribed Hodge numbers [32 Theorem 1] .
1.3.
Our approach to Theorem A is through Kazhdan–Lusztig combinatorics. Let W W W be the affine Weyl group of SL N \operatorname{SL}_N SL N , acting on weights by the dot action x ⋅ λ x \mathbin{\cdot} \lambda x ⋅ λ with translations scaled by p p p , and let w 0 w_0 w 0 be the longest element of its finite Weyl subgroup. We write
h y , w ( v ) = v ℓ ( w ) − ℓ ( y ) P y , w ( v − 2 ) h_{y,w}(v) = v^{\ell(w)-\ell(y)}P_{y,w}(v^{-2}) h y , w ( v ) = v ℓ ( w ) − ℓ ( y ) P y , w ( v − 2 )
for the Kazhdan–Lusztig polynomials in Soergel’s normalization, and μ ( y , w ) \mu(y,w) μ ( y , w ) for the coefficient of v v v in h y , w h_{y,w} h y , w . For p p p large and x ⋅ 0 x \mathbin{\cdot} 0 x ⋅ 0 p p p -restricted,
μ ( w 0 , w 0 x ) = dim H 1 ( SL N ( F p ) , L ( x ⋅ 0 ) ) , \mu(w_0,w_0x) = \dim H^1(\operatorname{SL}_N(\mathbb{F}_p), L(x \mathbin{\cdot} 0)), μ ( w 0 , w 0 x ) = dim H 1 ( SL N ( F p ) , L ( x ⋅ 0 )) ,
by results of Andersen and of Bendel, Nakano and Pillen (see § 2.5 ). As H 1 ≠ 0 H^1 \ne0 H 1 = 0 forces the centre to act trivially, L ( x ⋅ 0 ) L(x \mathbin{\cdot} 0) L ( x ⋅ 0 ) is not faithful for SL N ( F p ) \operatorname{SL}_N(\mathbb{F}_p) SL N ( F p ) ; it descends to a faithful module for the simple group PSL N ( F p ) \operatorname{PSL}_N(\mathbb{F}_p) PSL N ( F p ) with the same H 1 H^1 H 1 , so showing that these coefficients are unbounded would disprove Guralnick’s conjecture. This combinatorial formulation is also what allowed Lübeck to attack the conjecture by computer: his values are coefficients μ ( w 0 , w 0 x ) \mu(w_0,w_0x) μ ( w 0 , w 0 x ) , computed by machine for N ≤ 9 N \le9 N ≤ 9 . The following is the technical heart of the paper.
Theorem B. Let r ≥ 2 r \ge2 r ≥ 2 , put N = 2 r + 1 N = 2r + 1 N = 2 r + 1 , and let W W W be the affine Weyl group of type A ~ 2 r \widetilde{A}_{2r} A 2 r , realized by affine permutations of period N N N . For the affine permutation with window
β r = [ N + 1 , … , N + r − 1 , 1 − r , r + 1 , N + r , 2 − r , … , 0 ] \beta_r = [N+1,\ldots,N+r-1,1-r,r+1,N+r,2-r,\ldots,0] β r = [ N + 1 , … , N + r − 1 , 1 − r , r + 1 , N + r , 2 − r , … , 0 ] one has
μ ( w 0 , w 0 β r ) = r . \mu(w_0,w_0\beta_r) = r. μ ( w 0 , w 0 β r ) = r . More generally, the elements y 0 , … , y r − 1 y_0,\ldots,y_{r-1} y 0 , … , y r − 1 of § 2.3 , with y 0 = β r y_0 = \beta_r y 0 = β r , satisfy
h w 0 , w 0 y s ( v ) = r v s + 1 + O ( v s + 3 ) ( 0 ≤ s ≤ r − 1 ) . h_{w_0,w_0y_s}(v) = r v^{s+1} + O(v^{s+3}) \qquad(0 \le s \le r-1). h w 0 , w 0 y s ( v ) = r v s + 1 + O ( v s + 3 ) ( 0 ≤ s ≤ r − 1 ) .
Remark 1.1. Scott and Xi constructed unbounded μ ( y , w ) \mu(y,w) μ ( y , w ) with both endpoints varying, so that they govern Ext 1 \operatorname{Ext}^1 Ext 1 between two simple modules rather than H 1 H^1 H 1 . They asked whether the coefficients can still be unbounded when the lower endpoint is required to be w 0 w_0 w 0 (as in the formula for first cohomology above) and proposed explicit candidates [35 Section 4] . Theorem B gives an affirmative answer using a different family of upper endpoints, namely the elements w 0 β r w_0\beta_r w 0 β r .
The higher-degree coefficients in Theorem B have consequences for the cohomology of finite simple groups in every degree. Taking s = n − 1 s = n - 1 s = n − 1 in that theorem gives a coefficient of v n v^n v n equal to r r r whenever 1 ≤ n ≤ r 1 \le n \le r 1 ≤ n ≤ r . This is the dimension of the corresponding rational cohomology group in sufficiently large characteristic. The comparison with finite groups in § 2.6 then gives the following result.
Theorem C. Let n ≥ 1 n \ge1 n ≥ 1 and r ≥ max { 2 , n } r \ge\max\{2,n\} r ≥ max { 2 , n } . For every sufficiently large prime p p p and every sufficiently large integer a a a , the finite simple group PSL 2 r + 1 ( F p a ) \operatorname{PSL}_{2r+1}(\mathbb{F}_{p^a}) PSL 2 r + 1 ( F p a ) has a faithful absolutely irreducible module V V V over F p \mathbb{F}_p F p such that
dim F p H n ( PSL 2 r + 1 ( F p a ) , V ) ≥ r . \dim_{\mathbb{F}_p} H^n(\operatorname{PSL}_{2r+1}(\mathbb{F}_{p^a}),V) \ge r. dim F p H n ( PSL 2 r + 1 ( F p a ) , V ) ≥ r .
1.4.
The natural home for our results is the quantum group setting. Let U ζ U_\zeta U ζ be Lusztig’s quantum group of s l N ( C ) \mathfrak{sl}_N(\mathbb{C}) sl N ( C ) at a primitive root of unity ζ \zeta ζ of odd order ℓ > N \ell> N ℓ > N , and let ⋅ ℓ \cdot_\ell ⋅ ℓ denote the dot action of the affine Weyl group with translations scaled by ℓ \ell ℓ . Here Lusztig’s character formula holds [40 ] , and its cohomological interpretation [11 , 30 ] expresses h w 0 , w 0 x h_{w_0,w_0x} h w 0 , w 0 x as the generating function of the dimensions of H j ( U ζ , L ζ ( x ⋅ ℓ 0 ) ) H^j(U_\zeta,L_\zeta(x\mathbin{\cdot_\ell}0)) H j ( U ζ , L ζ ( x ⋅ ℓ 0 )) . The corresponding statement for algebraic groups in characteristic p p p is deduced from the quantum case: Lusztig’s character formula was proved for large p p p by comparison with the quantum group [2 ] . Our results for finite groups (Theorems A and C) therefore pass through the quantum group, and Theorem D below states the quantum result directly.
Parshall and Scott bounded the dimensions of cohomology groups for quantum groups in terms of the degree and the root system [30 Corollary 6.7] , and asked whether the dependence on the root system could be removed [30 Remark 6.6] . Theorem D shows that it cannot, even in a fixed positive degree.
Theorem D. Let r ≥ 2 r \ge2 r ≥ 2 , put N = 2 r + 1 N = 2r + 1 N = 2 r + 1 , and let ℓ > N \ell> N ℓ > N be odd. Let U ζ U_\zeta U ζ be Lusztig’s quantum group of s l N ( C ) \mathfrak{sl}_N(\mathbb{C}) sl N ( C ) at a primitive ℓ \ell ℓ th root of unity ζ \zeta ζ . For 0 ≤ s ≤ r − 1 0 \le s \le r - 1 0 ≤ s ≤ r − 1 , put λ r , s = y s ⋅ ℓ 0 \lambda_{r,s} = y_s\mathbin{\cdot_\ell}0 λ r , s = y s ⋅ ℓ 0 . Then the irreducible modules L ζ ( λ r , s ) L_\zeta(\lambda_{r,s}) L ζ ( λ r , s ) satisfy
dim C H s + 1 ( U ζ , L ζ ( λ r , s ) ) = r . \dim_{\mathbb{C}} H^{s+1}(U_\zeta,L_\zeta(\lambda_{r,s})) = r. dim C H s + 1 ( U ζ , L ζ ( λ r , s )) = r . Moreover, H j ( U ζ , L ζ ( λ r , s ) ) = 0 H^j(U_\zeta,L_\zeta(\lambda_{r,s})) = 0 H j ( U ζ , L ζ ( λ r , s )) = 0 for 0 ≤ j < s + 1 0 \le j < s + 1 0 ≤ j < s + 1 , so the prescribed dimension occurs in the first nonzero cohomological degree.
1.5.
Our group-theoretic constructions require the characteristic to be sufficiently large relative to the rank, and therefore leave open the possibility that Guralnick’s conjecture holds in a given fixed characteristic p > 0 p > 0 p > 0 . For the same reason, the bounds sought in Question I may still hold in every degree, with constants depending on the degree and the characteristic. This leads us to formulate the following fixed-characteristic version of Question I.
Question II. Fix a prime p p p and an integer n ≥ 1 n \ge1 n ≥ 1 . Does there exist a constant C n , p C_{n,p} C n , p such that
dim H n ( G , V ) ≤ C n , p \dim H^n(G,V) \le C_{n,p} dim H n ( G , V ) ≤ C n , p
for every finite simple group G G G and every absolutely irreducible module V V V in characteristic p p p ?
Keeping the characteristic fixed while the rank grows takes us beyond the regime where ordinary Kazhdan–Lusztig polynomials control cohomology. The modular phenomena reflected in the p p p -canonical basis, which is substantially harder to compute than the ordinary Kazhdan–Lusztig basis, suggest that extending our approach to these questions would require considerably deeper control of modular representation theory.
AI statement
This project began while CB, NL, and DP were participating in the 2025 ICERM Semester Program on Categorification and Computation in Algebraic Combinatorics. Our starting point was the observation that, within the parallelepiped of alcoves corresponding to p p p -restricted weights, Lübeck’s examples lie in the regions farthest from the identity. Direct computations using Sage within this region were impractical in higher ranks, and the available values showed no recognizable pattern to guide a proof. At this point DF joined the project, using MAGMA calculations to conduct a more detailed investigation of this region. This led us to ask whether more tractable examples might exist closer to the identity.
We turned to GPT-6 Astra and asked it to explore this region. From the data it provided, we identified the elements β r \beta_r β r as a promising family exhibiting a uniform growth pattern. However, the obstacle remained: although these elements are closer to the identity, and we could compute individual cases in small ranks, the usual Kazhdan–Lusztig algorithm gave no way to directly calculate the entire infinite family.
We therefore asked GPT-6 Astra to find a nearby auxiliary element X 0 X_0 X 0 whose Kazhdan–Lusztig basis element could be computed more directly, and from which β r \beta_r β r could be reached using only simple reflections of the finite Weyl group. The strategy of the proof was ours: we asked for a computable element in a prescribed region, together with a factorization of β r \beta_r β r through simple reflections along which the Kazhdan–Lusztig coefficients remain manageable. Following our instructions, the model produced a proof of Theorem B. All arguments were checked, rewritten, and completed by the authors, who take full responsibility for the results.
Structure of the paper. Our strategy is to estimate the relevant Kazhdan–Lusztig polynomials at X 0 X_0 X 0 and then propagate these estimates along two successive chains leading to β r \beta_r β r , using multiplication formulas in the spherical Hecke module. In Section 2, we introduce the notation and preliminaries, describe these chains, and deduce Theorems A, B, C and D from the coefficient estimates and multiplication formulas proved in the subsequent sections. Section 3 establishes the initial estimates at X 0 X_0 X 0 , and Section 4 develops the tools common to both chains. Finally, Sections 5 and 6 establish the multiplication formulas and coefficient estimates along the first and second chains, respectively.
Acknowledgments. This work began during the Categorification and Computation in Algebraic Combinatorics programme at the Institute for Computational and Experimental Research in Mathematics (ICERM). We thank ICERM for its hospitality and for a productive environment. This work was supported by the National Science Foundation under Grant No. DMS-2424556 while CB, NL and DP were in residence at ICERM. DF acknowledges support from a LACIM postdoctoral fellowship at the Université du Québec à Montréal.
The main theorems
In this section we introduce the affine Weyl group and its Hecke algebra and state the identities and coefficient estimates needed for the Kazhdan–Lusztig calculation. Taking these results as given, we prove Theorem B and use it to deduce the cohomological results in Theorems A, C and D. The proofs of the identities and estimates are deferred to the subsequent sections.
The affine Weyl group and its Hecke algebra
Fix r ≥ 1 r \ge1 r ≥ 1 , put N = 2 r + 1 N = 2r + 1 N = 2 r + 1 , and let W W W be the affine Weyl group of type A ~ 2 r \widetilde{A}_{2r} A 2 r with simple reflections S = { s 0 , … , s N − 1 } S = \{s_0, \ldots, s_{N-1}\} S = { s 0 , … , s N − 1 } and Coxeter graph
We write ℓ \ell ℓ for the Coxeter length and ≤ \le ≤ for the Bruhat order. Let I = { s 1 , … , s N − 1 } I = \{s_1, \ldots, s_{N-1}\} I = { s 1 , … , s N − 1 } and write w I w_I w I for the longest element of the finite parabolic subgroup W I W_I W I of type A 2 r A_{2r} A 2 r (this is called w 0 w_0 w 0 in the Introduction). We use the standard realization of this Coxeter system by affine permutations [7 Section 8.3] . The abstract simple reflections correspond to the periodic adjacent transpositions described below. Under this realization, the elements of W W W are the bijections w : Z → Z w : \mathbb{Z} \to\mathbb{Z} w : Z → Z satisfying
w ( i + N ) = w ( i ) + N ( i ∈ Z ) , ∑ i = 1 N w ( i ) = N ( N + 1 ) 2 . w(i+N)=w(i)+N \quad(i\in\mathbb{Z}), \qquad\sum_{i=1}^{N}w(i)=\frac{N(N+1)}{2}. w ( i + N ) = w ( i ) + N ( i ∈ Z ) , i = 1 ∑ N w ( i ) = 2 N ( N + 1 ) .
Such an affine permutation is determined by its window [ w ( 1 ) , … , w ( N ) ] [w(1),\ldots,w(N)] [ w ( 1 ) , … , w ( N )] . Products are compositions: ( x y ) ( i ) = x ( y ( i ) ) (xy)(i)=x(y(i)) ( x y ) ( i ) = x ( y ( i )) . The simple reflections have windows
s i = [ 1 , … , i − 1 , i + 1 , i , i + 2 , … , N ] ( 1 ≤ i < N ) , s_i=[1,\ldots,i-1,i+1,i,i+2,\ldots,N]\quad(1\le i<N), s i = [ 1 , … , i − 1 , i + 1 , i , i + 2 , … , N ] ( 1 ≤ i < N ) ,
s 0 = [ 0 , 2 , … , N − 1 , N + 1 ] . s_0=[0,2,\ldots,N-1,N+1]. s 0 = [ 0 , 2 , … , N − 1 , N + 1 ] .
Thus right multiplication by s i s_i s i , 1 ≤ i < N 1\le i<N 1 ≤ i < N , exchanges positions i i i and i + 1 i+1 i + 1 , while right multiplication by s 0 s_0 s 0 replaces the first and last entries by w ( N ) − N w(N)-N w ( N ) − N and w ( 1 ) + N w(1)+N w ( 1 ) + N , respectively. Empty intervals in windows are omitted. The subgroup W I W_I W I is the ordinary symmetric group on { 1 , … , N } \{1,\ldots,N\} { 1 , … , N } . We write I W {}^IW I W for the minimal representatives of W I \ W W_I\backslash W W I \ W ; they are characterized by
w − 1 ( 1 ) < ⋯ < w − 1 ( N ) . (1) w^{-1}(1)<\cdots<w^{-1}(N). \tag*{(1)} w − 1 ( 1 ) < ⋯ < w − 1 ( N ) . (1)
More generally, for a proper subset P ⊆ S P\subseteq S P ⊆ S we write W P W_P W P for the finite subgroup it generates and w P w_P w P for its longest element. For Q ⊆ P ⊆ S Q\subseteq P\subseteq S Q ⊆ P ⊆ S , we write Q W P {}^QW_P Q W P for the set of minimal coset representatives of W Q \ W P W_Q\backslash W_P W Q \ W P . Every u ∈ Q W P u\in{}^QW_P u ∈ Q W P factors uniquely as u = a b u=ab u = ab with a ∈ W Q a\in W_Q a ∈ W Q , b ∈ Q W P b\in{}^QW_P b ∈ Q W P and ℓ ( u ) = ℓ ( a ) + ℓ ( b ) \ell(u)=\ell(a)+\ell(b) ℓ ( u ) = ℓ ( a ) + ℓ ( b ) [7 Proposition 2.4.4] .
Let H = H ( W ) \mathcal{H}=\mathcal{H}(W) H = H ( W ) be the Hecke algebra over Z [ v , v − 1 ] \mathbb{Z}[v,v^{-1}] Z [ v , v − 1 ] with standard basis { H w } w ∈ W \{H_w\}_{w\in W} { H w } w ∈ W and relations
H s 2 = 1 + ( v − 1 − v ) H s , H x H y = H x y if ℓ ( x y ) = ℓ ( x ) + ℓ ( y ) . H_s^2=1+(v^{-1}-v)H_s,\qquad H_xH_y=H_{xy}\ \text{if }\ell(xy)=\ell(x)+\ell(y). H s 2 = 1 + ( v − 1 − v ) H s , H x H y = H x y if ℓ ( x y ) = ℓ ( x ) + ℓ ( y ) .
Its bar involution sends v v v to v − 1 v^{-1} v − 1 and H w H_w H w to H w − 1 − 1 H_{w^{-1}}^{-1} H w − 1 − 1 . For every w ∈ W w\in W w ∈ W , we let C w ∈ H C_w\in\mathcal{H} C w ∈ H denote the unique element satisfying
C w ‾ = C w , C w ∈ H w + ∑ y < w v Z [ v ] H y . \overline{C_w}=C_w,\qquad C_w\in H_w+\sum_{y<w}v\mathbb{Z}[v]H_y. C w = C w , C w ∈ H w + y < w ∑ v Z [ v ] H y .
The elements { C w } w ∈ W \{C_w\}_{w\in W} { C w } w ∈ W form the canonical basis of Kazhdan and Lusztig [25 ] . The Kazhdan–Lusztig polynomials h y , w h_{y,w} h y , w in this normalization are the change of basis coefficients defined by
C w = ∑ y ≤ w h y , w H y . C_w=\sum_{y\leq w}h_{y,w}H_y. C w = y ≤ w ∑ h y , w H y .
Thus h w , w = 1 h_{w,w}=1 h w , w = 1 and h y , w ∈ v Z [ v ] h_{y,w}\in v\mathbb{Z}[v] h y , w ∈ v Z [ v ] for y < w y<w y < w ; we put h y , w = 0 h_{y,w}=0 h y , w = 0 when y ≰ w y\not\leq w y ≤ w . It is easy to check that C s = H s + v C_s=H_s+v C s = H s + v . The canonical basis element of w I w_I w I is the symmetrizer [37 Proposition 2.9]
C w I = ∑ u ∈ W I v ℓ ( w I ) − ℓ ( u ) H u . C_{w_I}=\sum_{u\in W_I}v^{\ell(w_I)-\ell(u)}H_u. C w I = u ∈ W I ∑ v ℓ ( w I ) − ℓ ( u ) H u .
Pairing the terms indexed by u u u and u s us u s , and using the quadratic Hecke relation, gives
C w I H s = v − 1 C w I ( s ∈ I ) . (2) C_{w_I}H_s=v^{-1}C_{w_I}\qquad(s\in I). \tag*{(2)} C w I H s = v − 1 C w I ( s ∈ I ) . (2)
The spherical module M \mathcal{M} M is the right ideal M = C w I H \mathcal{M}=C_{w_I}\mathcal{H} M = C w I H . Its standard basis is { M x ∣ x ∈ I W } \{M_x\mid x\in{}^IW\} { M x ∣ x ∈ I W } where
M x = C w I H x . M_x=C_{w_I}H_x. M x = C w I H x .
The elements
C x = C w I x ( x ∈ I W ) C_x=C_{w_Ix}\qquad(x\in{}^IW) C x = C w I x ( x ∈ I W )
lie in M \mathcal{M} M and form its Kazhdan–Lusztig basis [37 Theorem 3.1 and Proposition 3.4] . In particular, both families are Z [ v , v − 1 ] \mathbb{Z}[v,v^{-1}] Z [ v , v − 1 ] -bases of M \mathcal{M} M , and
C x ∈ M x + ∑ y < x v Z [ v ] M y . C_x\in M_x+\sum_{y<x}v\mathbb{Z}[v]M_y. C x ∈ M x + y < x ∑ v Z [ v ] M y .
For an element X X X of a free module with a distinguished basis, we write [ Y ] X [Y]X [ Y ] X for the coefficient of a basis element Y Y Y in X X X . We define
ε ( F ) = [ M e ] F ( F ∈ M ) , h x = ε ( C x ) ( x ∈ I W ) , \varepsilon(F)=[M_e]F\qquad(F\in\mathcal{M}),\qquad h_x=\varepsilon(C_x)\qquad(x\in{}^IW), ε ( F ) = [ M e ] F ( F ∈ M ) , h x = ε ( C x ) ( x ∈ I W ) ,
and
μ ( y , w ) = [ v ] h y , w ( y , w ∈ W ) , μ I ( c , x ) = μ ( w I c , w I x ) ( c , x ∈ I W ) . \mu(y,w)=[v]h_{y,w}\qquad(y,w\in W),\qquad\mu_I(c,x)=\mu(w_Ic,w_Ix)\qquad(c,x\in{}^IW). μ ( y , w ) = [ v ] h y , w ( y , w ∈ W ) , μ I ( c , x ) = μ ( w I c , w I x ) ( c , x ∈ I W ) .
Lemma 2.1. For c , x ∈ I W c,x\in{}^IW c , x ∈ I W , we have
[ M c ] C x = h w I c , w I x . [M_c]C_x=h_{w_Ic,w_Ix}. [ M c ] C x = h w I c , w I x . In particular, h x = h w I , w I x h_x=h_{w_I,w_Ix} h x = h w I , w I x and μ I ( e , x ) = [ v ] h x \mu_I(e,x)=[v]h_x μ I ( e , x ) = [ v ] h x . Moreover, h e = 1 h_e=1 h e = 1 , while h x ∈ v Z [ v ] h_x\in v\mathbb{Z}[v] h x ∈ v Z [ v ] for x ≠ e x\ne e x = e , and all exponents in h x h_x h x have parity ℓ ( x ) \ell(x) ℓ ( x ) .
Proof. This is standard; see [37 Theorem 3.1 & Proposition 3.4] for the spherical module identification and [25 ] for the parity of polynomial exponents. □ \square □
The chains considered in Section 2.2 below use only right multiplication by generators in I I I . The following rule will turn their Hecke algebra multiplications into scalar recurrences. Put η = v + v − 1 \eta=v+v^{-1} η = v + v − 1 .
Lemma 2.2. For s ∈ I s\in I s ∈ I and F ∈ M F\in\mathcal{M} F ∈ M ,
ε ( F H s ) = v − 1 ε ( F ) , ε ( F C s ) = ( v + v − 1 ) ε ( F ) . (3) \varepsilon(FH_s)=v^{-1}\varepsilon(F),\qquad\varepsilon(FC_s)=(v+v^{-1})\varepsilon(F). \tag*{(3)} ε ( F H s ) = v − 1 ε ( F ) , ε ( F C s ) = ( v + v − 1 ) ε ( F ) . (3)
Proof. Right multiplication by H s H_s H s preserves the span of the standard basis elements indexed by each ( I , I ) (I,I) ( I , I ) double coset. If c ∈ I W c\in{}^IW c ∈ I W and c ≠ e c\ne e c = e , then W I c W I ≠ W I W_IcW_I\ne W_I W I c W I = W I , so M c H s M_cH_s M c H s cannot contribute to M e M_e M e . The remaining basis element satisfies M e H s = v − 1 M e M_eH_s=v^{-1}M_e M e H s = v − 1 M e by (2) . This proves the first identity. The second follows from C s = H s + v C_s=H_s+v C s = H s + v . □ \square □
By Lemma 2.1, Theorem B amounts to computing [ v ] h β r [v]h_{\beta_r} [ v ] h β r . We do this along the two chains defined below.
The first chain
To compute [ v ] h β r [v]h_{\beta_r} [ v ] h β r , we introduce an auxiliary element X 0 X_0 X 0 and connect it to β r \beta_r β r by two chains of finite simple reflections. We first obtain coefficient estimates along the first chain, then use the second chain to determine the desired coefficient.
Definition 2.3. We define X 0 X_0 X 0 and, for completeness, recall the definition of β r \beta_r β r from the introduction:
X 0 = [ 1 − r , r + 1 , N + 1 , … , N + r , 2 − r , … , 0 ] , X_0=[1-r,r+1,N+1,\ldots,N+r,2-r,\ldots,0], X 0 = [ 1 − r , r + 1 , N + 1 , … , N + r , 2 − r , … , 0 ] , β r = [ N + 1 , … , N + r − 1 , 1 − r , r + 1 , N + r , 2 − r , … , 0 ] . (4) \beta_r=[N+1,\ldots,N+r-1,1-r,r+1,N+r,2-r,\ldots,0]. \tag*{(4)} β r = [ N + 1 , … , N + r − 1 , 1 − r , r + 1 , N + r , 2 − r , … , 0 ] . (4)
For r = 5 r=5 r = 5 , Figure 1 and 2 show the affine string diagrams of β r \beta_r β r and X 0 X_0 X 0 , respectively.
Figure 1.
Lemma 2.4. Both elements in Definition 2.3 lie in I W I_W I W , and
β r = X 0 ( s 2 ⋯ s r ) ( s 1 ⋯ s r − 1 ) . \beta_r=X_0(s_2\cdots s_r)(s_1\cdots s_{r-1}). β r = X 0 ( s 2 ⋯ s r ) ( s 1 ⋯ s r − 1 ) . Every prefix of β r \beta_r β r lies in I W I_W I W and each simple reflection increases the length. Consequently, ℓ ( β r ) = ℓ ( X 0 ) + 2 r − 2 \ell(\beta_r)=\ell(X_0)+2r-2 ℓ ( β r ) = ℓ ( X 0 ) + 2 r − 2 .
Proof. Each window in (2.4) contains precisely the values
1 − r , … , 0 , r + 1 , N + 1 , … , N + r . 1-r,\ldots,0,\qquad r+1,\qquad N+1,\ldots,N+r. 1 − r , … , 0 , r + 1 , N + 1 , … , N + r .
The residues 1 , … , r 1,\ldots,r 1 , … , r have been shifted by N N N , those in r + 2 , … , N r+2,\ldots,N r + 2 , … , N by − N -N − N , and r + 1 r+1 r + 1 is unchanged. Hence the residues are distinct and the window sum is N ( N + 1 ) / 2 N(N+1)/2 N ( N + 1 ) /2 . It follows that both X 0 X_0 X 0 and β r \beta_r β r are elements of W W W .
The upper entries N + 1 , … , N + r N+1,\ldots,N+r N + 1 , … , N + r occur in increasing positions, as do the lower entries 1 − r , … , 0 1-r,\ldots,0 1 − r , … , 0 . The inverse images of the residues 1 , … , r 1,\ldots,r 1 , … , r therefore increase in [ 1 − N , 0 ] [1-N,0] [ 1 − N , 0 ] , the inverse image of r + 1 r+1 r + 1 lies in [ 1 , N ] [1,N] [ 1 , N ] , and those of r + 2 , … , N r+2,\ldots,N r + 2 , … , N increase in [ N + 1 , 2 N ] [N+1,2N] [ N + 1 , 2 N ] . Thus both inverse windows are increasing. By (2.1) we conclude that X 0 X_0 X 0 and β r \beta_r β r belong to I W I_W I W .
Right multiplication by s 2 ⋯ s r s_2\cdots s_r s 2 ⋯ s r moves r + 1 r+1 r + 1 past N + 1 , … , N + r − 1 N+1,\ldots,N+r-1 N + 1 , … , N + r − 1 ; multiplication by s 1 ⋯ s r − 1 s_1\cdots s_{r-1} s 1 ⋯ s r − 1 then moves 1 − r 1-r 1 − r past the same entries. This gives the window of β r \beta_r β r . Each swap exchanges an increasing adjacent numerical pair, so it increases Coxeter length. It also preserves the position order of the upper and lower entries, so every intermediate element remains in I W I_W I W . For r = 1 r=1 r = 1 , both products are empty and X 0 = β 1 = s 0 X_0=\beta_1=s_0 X 0 = β 1 = s 0 . □
Figure 1 (PDF p. 6). The affine string diagram of β 5 \beta_5 β 5 in type A ~ 10 \widetilde{A}_{10} A 10 . Solid black strands show its window and dashed gray strands indicate periodic translates.
Definition 2.5. For 0 ≤ k ≤ r − 1 0\leq k\leq r-1 0 ≤ k ≤ r − 1 , put x k = X 0 ( s 2 ⋯ s k + 1 ) x_k=X_0(s_2\cdots s_{k+1}) x k = X 0 ( s 2 ⋯ s k + 1 ) . Equivalently, x k x_k x k can be expressed in terms of its window as follows
x k = [ 1 − r , N + 1 , … , N + k , r + 1 , N + k + 1 , … , N + r , 2 − r , … , 0 ] . x_k=[1-r,N+1,\ldots,N+k,r+1,N+k+1,\ldots,N+r,2-r,\ldots,0]. x k = [ 1 − r , N + 1 , … , N + k , r + 1 , N + k + 1 , … , N + r , 2 − r , … , 0 ] . Empty products are the identity. Hence x 0 = X 0 x_0=X_0 x 0 = X 0 and x k + 1 = x k s k + 2 x_{k+1}=x_ks_{k+2} x k + 1 = x k s k + 2 for 0 ≤ k ≤ r − 2 0\leq k\leq r-2 0 ≤ k ≤ r − 2 . The factorizations are length additive by Lemma 2.4.
Our aim is to estimate h x r − 1 h_{x_{r-1}} h x r − 1 . The starting point is h X 0 h_{X_0} h X 0 , whose initial coefficients are given by the following proposition. Its proof is deferred to Subsection 3.5. For the estimates below, we write f = O ( v d ) f = O(v^d) f = O ( v d ) when f ∈ v d Z [ [ v ] ] f \in v^d\mathbb{Z}[[v]] f ∈ v d Z [[ v ]] .
Proposition 2.6. Let p ( d ) p(d) p ( d ) be the number of partitions of d d d , with p ( 0 ) = 1 p(0) = 1 p ( 0 ) = 1 . For r ≥ 2 r \ge2 r ≥ 2 ,
h X 0 = ∑ d = 0 r − 1 p ( d ) v 2 d + 1 + O ( v 2 r + 1 ) . (5) h_{X_0} = \sum_{d=0}^{r-1} p(d)v^{2d+1} + O(v^{2r+1}). \tag*{(5)} h X 0 = d = 0 ∑ r − 1 p ( d ) v 2 d + 1 + O ( v 2 r + 1 ) . (5)
Proof. See subsection 3.5. □
The coefficient estimate along this chain will be proved together with the following multiplication formulas in Section 5. Their correction terms are indexed by the following family of affine permutations.
Definition 2.7. For 1 ≤ j ≤ r 1 \le j \le r 1 ≤ j ≤ r we let
γ j = [ j + 1 , N + 1 , … , N + j , j + 2 , … , N − j , 1 − j , … , 0 ] . \gamma_j = [j+1,N+1,\ldots,N+j,j+2,\ldots,N-j,1-j,\ldots,0]. γ j = [ j + 1 , N + 1 , … , N + j , j + 2 , … , N − j , 1 − j , … , 0 ] .
As in the proof of Lemma 2.4, the inverse windows of the γ j \gamma_j γ j are increasing, so γ j ∈ I W \gamma_j \in I^W γ j ∈ I W . An example is depicted in Figure 3 .
Figure 3. The affine string diagram for γ 3 \gamma_3 γ 3 , with r = 5 r=5 r = 5 (so N = 11 N=11 N = 11 ).
Proposition 2.8. For r ≥ 2 r \ge2 r ≥ 2 and 0 ≤ k ≤ r − 2 0 \le k \le r-2 0 ≤ k ≤ r − 2 ,
C x k C s k + 2 = { C x k + 1 + C e , k = 0 , C x k + 1 + C x k − 1 + C γ k , k > 0. C_{x_k}C_{s_{k+2}} =
\begin{cases}
C_{x_{k+1}} + C_e, & k=0,\\
C_{x_{k+1}} + C_{x_{k-1}} + C_{\gamma_k}, & k>0.
\end{cases} C x k C s k + 2 = { C x k + 1 + C e , C x k + 1 + C x k − 1 + C γ k , k = 0 , k > 0.
Proof. This proposition and Lemma 2.9 are proved simultaneously in Section 5.
For a , b ≥ 0 a,b \ge0 a , b ≥ 0 , let d a , b d_{a,b} d a , b be the number of partitions of a + b a+b a + b with at least a a a parts, and put
A a , b = ∑ i = 0 a d i , b , A − 1 , b = 0. (6) A_{a,b} = \sum_{i=0}^{a} d_{i,b}, \qquad A_{-1,b}=0. \tag*{(6)} A a , b = i = 0 ∑ a d i , b , A − 1 , b = 0. (6)
These numbers describe the finite approximation to each coefficient h x a h_{x_a} h x a in the following lemma.
Lemma 2.9. For r ≥ 1 r \ge1 r ≥ 1 and 0 ≤ a ≤ r − 1 0 \le a \le r-1 0 ≤ a ≤ r − 1 ,
h x a = ∑ b = 0 r − 1 − a A a , b v a + 1 + 2 b + O ( v 2 r + 1 − a ) . (7) h_{x_a} = \sum_{b=0}^{r-1-a} A_{a,b}v^{a+1+2b} + O(v^{2r+1-a}). \tag*{(7)} h x a = b = 0 ∑ r − 1 − a A a , b v a + 1 + 2 b + O ( v 2 r + 1 − a ) . (7) In particular,
h x a = ( a + 1 ) v a + 1 + O ( v a + 3 ) , h x r − 1 = r v r + O ( v r + 2 ) . h_{x_a} = (a+1)v^{a+1} + O(v^{a+3}), \qquad h_{x_{r-1}} = rv^r + O(v^{r+2}). h x a = ( a + 1 ) v a + 1 + O ( v a + 3 ) , h x r − 1 = r v r + O ( v r + 2 ) .
Proof. For r = 1 r=1 r = 1 , one has x 0 = s 0 x_0=s_0 x 0 = s 0 and h s 0 = v h_{s_0}=v h s 0 = v , so the assertion holds. For r ≥ 2 r\ge2 r ≥ 2 , the proof of this lemma is intertwined with that of Proposition 2.8: the multiplication identities and the approximation (7) are proved simultaneously in § 5 . □ \square □
The second chain
The first chain gives h x r − 1 = r v r + O ( v r + 2 ) h_{x_{r-1}}=rv^r+O(v^{r+2}) h x r − 1 = r v r + O ( v r + 2 ) . We now complete the path from x r − 1 x_{r-1} x r − 1 to β r \beta_r β r . The indices below are chosen so that each successive multiplication decreases the index by one. We will use the estimate for h x r − 1 h_{x_{r-1}} h x r − 1 to determine the coefficient of v v v at the endpoint.
Definition 2.10. For 0 ≤ s ≤ r − 1 0\le s\le r-1 0 ≤ s ≤ r − 1 , put
y s = x r − 1 ( s 1 ⋯ s r − 1 − s ) . (8) y_s=x_{r-1}(s_1\cdots s_{r-1-s}). \tag*{(8)} y s = x r − 1 ( s 1 ⋯ s r − 1 − s ) . (8) Its window is
y s = [ N + 1 , … , N + r − 1 − s , 1 − r , N + r − s , … , N + r − 1 , r + 1 , N + r , 2 − r , … , 0 ] . (9) y_s=[N+1,\ldots,N+r-1-s,1-r,N+r-s,\ldots,N+r-1,r+1,N+r,2-r,\ldots,0]. \tag*{(9)} y s = [ N + 1 , … , N + r − 1 − s , 1 − r , N + r − s , … , N + r − 1 , r + 1 , N + r , 2 − r , … , 0 ] . (9) Thus y r − 1 = x r − 1 y_{r-1}=x_{r-1} y r − 1 = x r − 1 , y 0 = β r y_0=\beta_r y 0 = β r , and
y t = y t + 1 s r − 1 − t ( 0 ≤ t ≤ r − 2 ) . y_t=y_{t+1}s_{r-1-t}\qquad(0\le t\le r-2). y t = y t + 1 s r − 1 − t ( 0 ≤ t ≤ r − 2 ) . The factorizations are length additive by Lemma 2.4 .
Our starting value is h x r − 1 h_{x_{r-1}} h x r − 1 , and our goal is to estimate h β r h_{\beta_r} h β r . The multiplication formulas again separate the cover from the remaining corrections. Their identity coefficients will be high enough that the leading term can be transported along the chain.
Proposition 2.11. For r ≥ 2 r\ge2 r ≥ 2 and every 0 ≤ t ≤ r − 2 0\le t\le r-2 0 ≤ t ≤ r − 2 , there exists an element E t ∈ M ⊆ H E_t\in\mathcal{M}\subseteq\mathcal{H} E t ∈ M ⊆ H such that
C y t + 1 C s r − 1 − t = C y t + C y t + 2 + E t , (10) C_{y_{t+1}}C_{s_{r-1-t}}=C_{y_t}+C_{y_{t+2}}+E_t, \tag*{(10)} C y t + 1 C s r − 1 − t = C y t + C y t + 2 + E t , (10) where the term C y t + 2 C_{y_{t+2}} C y t + 2 is omitted when t = r − 2 t=r-2 t = r − 2 , and
ε ( E t ) = O ( v t + 3 ) . \varepsilon(E_t)=O(v^{t+3}). ε ( E t ) = O ( v t + 3 ) . The proof is given in § 6 .
We first obtain a coarse estimate from the first-chain endpoint and positivity. It already places the cover contribution beyond the degree needed at each step.
Lemma 2.12. Let r ≥ 2 r\ge2 r ≥ 2 . For 0 ≤ s ≤ r − 1 0\le s\le r-1 0 ≤ s ≤ r − 1 ,
h y s = O ( v s + 1 ) . (11) h_{y_s}=O(v^{s+1}). \tag*{(11)} h y s = O ( v s + 1 ) . (11) Moreover,
h y t = η h y t + 1 + O ( v t + 3 ) ( 0 ≤ t ≤ r − 2 ) . (12) h_{y_t}=\eta h_{y_{t+1}}+O(v^{t+3})\qquad(0\le t\le r-2). \tag*{(12)} h y t = η h y t + 1 + O ( v t + 3 ) ( 0 ≤ t ≤ r − 2 ) . (12)
Proof. The known path from x r − 1 x_{r-1} x r − 1 to y s y_s y s has r − 1 − s r-1-s r − 1 − s finite length-increasing steps. By Corollary 4.5 , each step lowers the smallest possible exponent by at most one. Since h x r − 1 = r v r + O ( v r + 2 ) h_{x_{r-1}}=rv^r+O(v^{r+2}) h x r − 1 = r v r + O ( v r + 2 ) by Lemma 2.9, successive applications give
h y s = O ( v r − ( r − 1 − s ) ) = O ( v s + 1 ) . (13) h_{y_s}=O\left(v^{r-(r-1-s)}\right)=O(v^{s+1}). \tag*{(13)} h y s = O ( v r − ( r − 1 − s ) ) = O ( v s + 1 ) . (13)
This proves (11) independently of the second-chain multiplication formulas. Now apply ε \varepsilon ε to Proposition 2.11. The scalar rule (3) gives η h y t + 1 \eta h_{y_{t+1}} η h y t + 1 on the left. When the cover is present, (11) gives h y t + 2 = O ( v t + 3 ) h_{y_{t+2}}=O(v^{t+3}) h y t + 2 = O ( v t + 3 ) ; the remainder satisfies the same bound by the proposition. Absorbing both contributions proves (12) .
Proof of Theorem B
We are now in position to prove Theorem B.
Proof of Theorem B. Let r ≥ 2 r \ge2 r ≥ 2 and write e t = h y t − η h y t + 1 e_t=h_{y_t}-\eta h_{y_{t+1}} e t = h y t − η h y t + 1 , so that e t = O ( v t + 3 ) e_t=O(v^{t+3}) e t = O ( v t + 3 ) by (12) . Iteration gives
h y s = η r − 1 − s h x r − 1 + ∑ t = s r − 2 η t − s e t = η r − 1 − s h x r − 1 + O ( v s + 3 ) . (14) h_{y_s}=\eta^{r-1-s}h_{x_{r-1}}+\sum_{t=s}^{r-2}\eta^{t-s}e_t=\eta^{r-1-s}h_{x_{r-1}}+O(v^{s+3}). \tag*{(14)} h y s = η r − 1 − s h x r − 1 + t = s ∑ r − 2 η t − s e t = η r − 1 − s h x r − 1 + O ( v s + 3 ) . (14)
The sum is empty when s = r − 1 s=r-1 s = r − 1 . Each summand has order at least ( t + 3 ) − ( t − s ) = s + 3 (t+3)-(t-s)=s+3 ( t + 3 ) − ( t − s ) = s + 3 . By Lemma 2.9, h x r − 1 = r v r + O ( v r + 2 ) h_{x_{r-1}}=rv^r+O(v^{r+2}) h x r − 1 = r v r + O ( v r + 2 ) . Since η r − 1 − s = v − ( r − 1 − s ) ( 1 + v r − 1 − s ) \eta^{r-1-s}=v^{-(r-1-s)}(1+v^{r-1-s}) η r − 1 − s = v − ( r − 1 − s ) ( 1 + v r − 1 − s ) , substitution in (2.13) gives
h w I , w I y s = h y s = r v s + 1 + O ( v s + 3 ) , h_{w_I,w_Iy_s}=h_{y_s}=rv^{s+1}+O(v^{s+3}), h w I , w I y s = h y s = r v s + 1 + O ( v s + 3 ) ,
where the first equality is Lemma 2.1. For s = 0 s=0 s = 0 we have y 0 = β r y_0=\beta_r y 0 = β r , so taking the coefficient of v v v yields μ ( w I , w I β r ) = r \mu(w_I,w_I\beta_r)=r μ ( w I , w I β r ) = r .
Example 2.13. Here r = 2 r=2 r = 2 , N = 5 N=5 N = 5 , and the path is
[ − 1 , 3 , 6 , 7 , 0 ] → s 2 [ − 1 , 6 , 3 , 7 , 0 ] → s 1 [ 6 , − 1 , 3 , 7 , 0 ] . [-1,3,6,7,0]\xrightarrow{s_2}[-1,6,3,7,0]\xrightarrow{s_1}[6,-1,3,7,0]. [ − 1 , 3 , 6 , 7 , 0 ] s 2 [ − 1 , 6 , 3 , 7 , 0 ] s 1 [ 6 , − 1 , 3 , 7 , 0 ] .
The initial coefficient begins h x 0 = v + v 3 + O ( v 5 ) h_{x_0}=v+v^3+O(v^5) h x 0 = v + v 3 + O ( v 5 ) , so h x 1 = η h x 0 − 1 = 2 v 2 + O ( v 4 ) h_{x_1}=\eta h_{x_0}-1=2v^2+O(v^4) h x 1 = η h x 0 − 1 = 2 v 2 + O ( v 4 ) . Along the second chain,
h β 2 = η h x 1 + O ( v 3 ) = 2 v + O ( v 3 ) . h_{\beta_2}=\eta h_{x_1}+O(v^3)=2v+O(v^3). h β 2 = η h x 1 + O ( v 3 ) = 2 v + O ( v 3 ) .
Thus μ ( w I , w I β 2 ) = 2 \mu(w_I,w_I\beta_2)=2 μ ( w I , w I β 2 ) = 2 .
Example 2.14. Let r = 5 r=5 r = 5 . Table 1 lists the multiplication formulas along the chains x 0 = X 0 , … , x 4 x_0=X_0,\ldots,x_4 x 0 = X 0 , … , x 4 and y 4 = x 4 , … , y 0 = β 5 y_4=x_4,\ldots,y_0=\beta_5 y 4 = x 4 , … , y 0 = β 5 ; all multiplicities are one. The column Cover is the previous element of the chain (Lemma 4.15), and the column Extra is the only other term that occurs: C e C_e C e at the first step, then C γ k C_{\gamma_k} C γ k (Theorem 2.7) on the first chain and C ζ t C_{\zeta_t} C ζ t , for certain elements ζ t \zeta_t ζ t found by computer, on the second.
Applying ε \varepsilon ε to a row, the identity coefficient of the new element is ( v + v − 1 ) h x (v+v^{-1})h_x ( v + v − 1 ) h x minus those of the two correction terms (Lemma 2.2). On the first chain the terms of degree k k k cancel, since h x k = ( k + 1 ) v k + 1 + ⋯ h_{x_k}=(k+1)v^{k+1}+\cdots h x k = ( k + 1 ) v k + 1 + ⋯ and h γ k = v k + ⋯ h_{\gamma_k}=v^k+\cdots h γ k = v k + ⋯ (Theorem 3.25), and h x k + 1 h_{x_{k+1}} h x k + 1 begins with ( k + 2 ) v k + 2 (k+2)v^{k+2} ( k + 2 ) v k + 2 . On the second chain both correction terms are of order at least t + 3 t+3 t + 3 , so the leading coefficient 5 5 5 moves down one degree at each step, giving [ v ] h β 5 = 5 [v]h_{\beta_5}=5 [ v ] h β 5 = 5 . The proof does not need the second-chain formulas exactly, only that the correction terms are of this order (Proposition 2.11).
Table 1. The eight multiplication formulas for r = 5 r=5 r = 5 (Theorem 2.14).
First cohomology of finite groups
To prove Theorem A, we interpret the coefficient computed in Theorem B as the dimension of a first cohomology group. The comparison below applies when λ = x ⋅ 0 \lambda=x\cdot0 λ = x ⋅ 0 is p p p -restricted. We will check these conditions for x = β r x=\beta_r x = β r .
Let p p p be a prime number and let G = S L N G=\mathrm{SL}_N G = SL N over F p \mathbb{F}_p F p and let Φ \Phi Φ be its root system. Write α i , ϖ i \alpha_i,\varpi_i α i , ϖ i for the simple roots and fundamental weights, θ \theta θ for the highest root, and ρ = ∑ i ϖ i \rho=\sum_i\varpi_i ρ = ∑ i ϖ i . We identify W W W with p Z Φ ⋊ W I p\mathbb{Z}\Phi\rtimes W_I p Z Φ ⋊ W I , where Z Φ \mathbb{Z}\Phi Z Φ is the root lattice: s i s_i s i is the finite simple reflection for i ≥ 1 i\geq1 i ≥ 1 , and s 0 s_0 s 0 is the affine reflection in ⟨ λ , θ ∨ ⟩ = p \langle\lambda,\theta^\vee\rangle=p ⟨ λ , θ ∨ ⟩ = p . For β ∈ p Z Φ \beta\in p\mathbb{Z}\Phi β ∈ p Z Φ , let t β t_\beta t β denote the translation by β \beta β . We use the dot action
w ⋅ λ = w ( λ + ρ ) − ρ . w\cdot\lambda=w(\lambda+\rho)-\rho. w ⋅ λ = w ( λ + ρ ) − ρ .
For a dominant weight λ \lambda λ , let L ( λ ) L(\lambda) L ( λ ) be the simple rational G G G -module of highest weight λ \lambda λ . We say a weight λ \lambda λ is p p p -restricted if 0 ≤ ⟨ λ , α i ∨ ⟩ < p 0\leq\langle\lambda,\alpha_i^\vee\rangle<p 0 ≤ ⟨ λ , α i ∨ ⟩ < p for every simple root α i \alpha_i α i . Here H 1 ( G , − ) H^1(G,-) H 1 ( G , − ) denotes rational cohomology; for the finite groups below, we use ordinary group cohomology.
The first comparison is Andersen’s interpretation of the spherical KL coefficients; the precise degree-one statement is included in [28 Theorem 4.5] , following [1 ] . The second is the finite-group comparison [6 Theorem 7.5] , in the special case recorded in [28 Theorem 4.6] .
Proposition 2.15. Assume Lusztig’s character formula for type A N − 1 A_{N-1} A N − 1 in characteristic p p p , and suppose p ≥ 5 ( N − 1 ) p\geq5(N-1) p ≥ 5 ( N − 1 ) . If x ∈ I W x\in{}^IW x ∈ I W and λ = x ⋅ 0 \lambda=x\cdot0 λ = x ⋅ 0 is p p p -restricted, then
[ v ] h x = dim H 1 ( G , L ( λ ) ) = dim H 1 ( S L N ( F p ) , L ( λ ) ) . (15) [v]h_x=\dim H^1(G,L(\lambda))=\dim H^1(\mathrm{SL}_N(\mathbb{F}_p),L(\lambda)). \tag*{(15)} [ v ] h x = dim H 1 ( G , L ( λ )) = dim H 1 ( SL N ( F p ) , L ( λ )) . (15) For fixed N N N , these assumptions on p p p hold for every sufficiently large prime, by [2 , 16 ] ; see also [28 Theorem 3.1] .
We now check that the element β r \beta_r β r gives a weight to which this comparison applies.
Lemma 2.16. Let r ≥ 2 r\geq2 r ≥ 2 , N = 2 r + 1 N=2r+1 N = 2 r + 1 , and p > 2 p>2 p > 2 . With ν = ϖ r + ϖ r + 1 \nu=\varpi_r+\varpi_{r+1} ν = ϖ r + ϖ r + 1 , the window of β r \beta_r β r gives
β r = t p ν ( s r s r + 1 s r ) , λ r , p : = β r ⋅ 0 = p ν − 2 ( α r + α r + 1 ) . (16) \beta_r=t_{p\nu}(s_rs_{r+1}s_r),\qquad\lambda_{r,p}:=\beta_r\cdot0=p\nu-2(\alpha_r+\alpha_{r+1}). \tag*{(16)} β r = t p ν ( s r s r + 1 s r ) , λ r , p := β r ⋅ 0 = p ν − 2 ( α r + α r + 1 ) . (16) Equivalently,
λ r , p = 2 ϖ r − 1 + ( p − 2 ) ( ϖ r + ϖ r + 1 ) + 2 ϖ r + 2 . (17) \lambda_{r,p}=2\varpi_{r-1}+(p-2)(\varpi_r+\varpi_{r+1})+2\varpi_{r+2}. \tag*{(17)} λ r , p = 2 ϖ r − 1 + ( p − 2 ) ( ϖ r + ϖ r + 1 ) + 2 ϖ r + 2 . (17) This is a nonzero p p p -restricted weight in Z Φ \mathbb{Z}\Phi Z Φ .
Proof. Use the usual sum-zero coordinates for type A N − 1 A_{N-1} A N − 1 . If w ( j ) = a j + m j N w(j)=a_j+m_jN w ( j ) = a j + m j N , with 1 ≤ a j ≤ N 1\leq a_j\leq N 1 ≤ a j ≤ N , the affine action on a vector μ \mu μ is
( w μ ) a j = μ j + p m j . (w\mu)_{a_j}=\mu_j+pm_j. ( w μ ) a j = μ j + p m j .
The residue permutation of β r \beta_r β r is the transposition of r r r and r + 2 r+2 r + 2 , namely s r s r + 1 s r s_rs_{r+1}s_r s r s r + 1 s r . Its shifts, indexed by value residue, are
ν = ( 1 , … , 1 ⏟ r , 0 , − 1 , … , − 1 ⏟ r ) = ϖ r + ϖ r + 1 . \nu=(\underbrace{1,\ldots,1}_{r},0,\underbrace{-1,\ldots,-1}_{r})=\varpi_r+\varpi_{r+1}. ν = ( r 1 , … , 1 , 0 , r − 1 , … , − 1 ) = ϖ r + ϖ r + 1 .
This proves the factorization in (16) . The finite transposition sends ρ \rho ρ to ρ − 2 ( α r + α r + 1 ) \rho-2(\alpha_r+\alpha_{r+1}) ρ − 2 ( α r + α r + 1 ) , giving its weight formula. Expanding the simple roots in fundamental weights yields (17) . Its nonzero coefficients are 2 , p − 2 , p − 2 , 2 2,p-2,p-2,2 2 , p − 2 , p − 2 , 2 , all between 0 0 0 and p − 1 p-1 p − 1 . Finally ν \nu ν is an integral sum-zero vector, so ν ∈ Z Φ \nu\in\mathbb{Z}\Phi ν ∈ Z Φ ; the same is true of λ r , p \lambda_{r,p} λ r , p . □ \square □
Proof of Theorem A. Choose p p p satisfying Proposition 2.15 . Lemma 2.16 verifies its weight hypotheses for x = β r x=\beta_r x = β r . Equations (15) and Theorem B give
dim H 1 ( S L N ( F p ) , L ( λ r , p ) ) = [ v ] h β r = r . \dim H^1(\mathrm{SL}_N(\mathbb{F}_p),L(\lambda_{r,p}))=[v]h_{\beta_r}=r. dim H 1 ( SL N ( F p ) , L ( λ r , p )) = [ v ] h β r = r .
By Steinberg’s restriction theorem [39 ] , the restriction of L ( λ r , p ) L(\lambda_{r,p}) L ( λ r , p ) to S L N ( F p ) \mathrm{SL}_N(\mathbb{F}_p) SL N ( F p ) is nontrivial and absolutely irreducible, since λ r , p \lambda_{r,p} λ r , p is nonzero and p p p -restricted. Its central character is trivial because λ r , p ∈ Z Φ \lambda_{r,p}\in\mathbb{Z}\Phi λ r , p ∈ Z Φ . It therefore descends to a nontrivial absolutely irreducible module V ‾ \overline{V} V for P S L N ( F p ) \mathrm{PSL}_N(\mathbb{F}_p) PSL N ( F p ) . Let Z Z Z be the center of S L N ( F p ) \mathrm{SL}_N(\mathbb{F}_p) SL N ( F p ) . Since Z Z Z acts trivially and has order prime to p p p , its positive-degree cohomology with these coefficients vanishes. Inflation therefore gives isomorphisms in every degree [36 Proposition 1.8.11] . In particular, it identifies the first cohomology groups, so dim H 1 ( P S L N ( F p ) , V ‾ ) = r \dim H^1(\mathrm{PSL}_N(\mathbb{F}_p),\overline{V})=r dim H 1 ( PSL N ( F p ) , V ) = r . Finally P S L N ( F p ) \mathrm{PSL}_N(\mathbb{F}_p) PSL N ( F p ) is simple for N ≥ 5 N\geq5 N ≥ 5 ; the kernel of its nontrivial representation on V ‾ \overline{V} V is therefore trivial, proving faithfulness.
Since G G G is split over F p \mathbb{F}_p F p , L ( λ r , p ) L(\lambda_{r,p}) L ( λ r , p ) admits an F p \mathbb{F}_p F p -form, so V ‾ = V r , p ⊗ F p F ‾ p \overline{V}=V_{r,p}\otimes_{\mathbb{F}_p}\overline{\mathbb{F}}_p V = V r , p ⊗ F p F p for an F p P S L N ( F p ) \mathbb{F}_p\mathrm{PSL}_N(\mathbb{F}_p) F p PSL N ( F p ) -module V r , p V_{r,p} V r , p , which is therefore faithful and absolutely irreducible. As the cohomology of finite groups commutes with extension of scalars, dim F p H 1 ( PSL N ( F p ) , V r , p ) = r \dim_{\mathbb{F}_p} H^1(\operatorname{PSL}_N(\mathbb{F}_p),V_{r,p})=r dim F p H 1 ( PSL N ( F p ) , V r , p ) = r . This proves (1.1) for the module V r , p V_{r,p} V r , p of Theorem A. □
Higher cohomology of finite simple groups
Recall the notation and affine dot action of Subsection 2.5. Let k = F ‾ p k=\overline{\mathbb{F}}_p k = F p and let G = SL N G=\operatorname{SL}_N G = SL N be the algebraic group over k k k , with its standard F p \mathbb{F}_p F p -structure. We write H j ( G , − ) H^j(G,-) H j ( G , − ) for rational cohomology and H j ( G ( F q ) , − ) H^j(G(\mathbb{F}_q),-) H j ( G ( F q ) , − ) for ordinary cohomology of the finite group G ( F q ) G(\mathbb{F}_q) G ( F q ) , where q q q is a power of p p p . All dimensions below are taken over k k k .
Much of Subsection 2.5 generalises to all cohomological degrees. Indeed, Andersen’s theorem, in the formulation of [28 Theorem 4.5] , gives
h x ( v ) = ∑ j ≥ 0 dim k H j ( G , L ( x ⋅ 0 ) ) v j = ∑ j ≥ 0 dim k Ext G j ( k , L ( x ⋅ 0 ) ) v j (18) h_x(v)=\sum_{j\geq0}\dim_k H^j(G,L(x\mathbin{\cdot}0))v^j=\sum_{j\geq0}\dim_k\operatorname{Ext}^j_G(k,L(x\mathbin{\cdot}0))v^j \tag*{(18)} h x ( v ) = j ≥ 0 ∑ dim k H j ( G , L ( x ⋅ 0 )) v j = j ≥ 0 ∑ dim k Ext G j ( k , L ( x ⋅ 0 )) v j (18)
for x ∈ I W x\in{}^IW x ∈ I W such that x ⋅ 0 x\mathbin{\cdot}0 x ⋅ 0 is p p p -restricted, provided p ≥ N p\geq N p ≥ N and Lusztig’s character formula holds in characteristic p p p . The Ext groups are computed in the category of rational G G G -modules. The polynomial denoted m 1 , x m_{1,x} m 1 , x in that reference is our h x h_x h x .
To prove Theorem C, we will restrict the action to a sufficiently large finite subgroup G ( F q ) G(\mathbb{F}_q) G ( F q ) . Restriction need not preserve the cohomology dimension, but eventual injectivity ensures that it cannot decrease. We first verify the weight hypotheses for every element of the second chain, rather than only its endpoint β r \beta_r β r .
Lemma 2.17. Let r ≥ 2 r\geq2 r ≥ 2 , N = 2 r + 1 N=2r+1 N = 2 r + 1 , 0 ≤ s ≤ r − 1 0\leq s\leq r-1 0 ≤ s ≤ r − 1 , and p > N p>N p > N . In the sum-zero coordinates of type A N − 1 A_{N-1} A N − 1 , put λ r , s = y s ⋅ 0 \lambda_{r,s}=y_s\mathbin{\cdot}0 λ r , s = y s ⋅ 0 . Then
λ r , s = ( p , … , p ⏟ r − 1 − s , p − 1 , … , p − 1 ⏟ s , p − 2 , 0 , − p + s + 2 , − p , … , − p ⏟ r − 1 ) . \lambda_{r,s}=\left(\underbrace{p,\ldots,p}_{r-1-s},\underbrace{p-1,\ldots,p-1}_{s},p-2,0,-p+s+2,\underbrace{-p,\ldots,-p}_{r-1}\right). λ r , s = r − 1 − s p , … , p , s p − 1 , … , p − 1 , p − 2 , 0 , − p + s + 2 , r − 1 − p , … , − p . In particular, λ r , s \lambda_{r,s} λ r , s is a nonzero p p p -restricted weight in the root lattice. For s = 0 s=0 s = 0 this recovers the weight λ r , p \lambda_{r,p} λ r , p of Lemma 2.16.
Proof. Set b = r − 1 − s b=r-1-s b = r − 1 − s and
ν = ( 1 , … , 1 ⏟ r , 0 , − 1 , … , − 1 ⏟ r ) . \nu=(\underbrace{1,\ldots,1}_{r},0,\underbrace{-1,\ldots,-1}_{r}). ν = ( r 1 , … , 1 , 0 , r − 1 , … , − 1 ) .
The window in (2.9) has translation part p ν p\nu p ν , so write y s = t p ν σ s y_s=t_{p\nu}\sigma_s y s = t p ν σ s with σ s ∈ S N \sigma_s\in S_N σ s ∈ S N . Reading the positions of its residues gives
( σ s − 1 ( 1 ) , … , σ s − 1 ( N ) ) = ( 1 , … , b , b + 2 , … , r , r + 2 , r + 1 , b + 1 , r + 3 , … , N ) , \left(\sigma_s^{-1}(1),\ldots,\sigma_s^{-1}(N)\right)=(1,\ldots,b,b+2,\ldots,r,r+2,r+1,b+1,r+3,\ldots,N), ( σ s − 1 ( 1 ) , … , σ s − 1 ( N ) ) = ( 1 , … , b , b + 2 , … , r , r + 2 , r + 1 , b + 1 , r + 3 , … , N ) ,
where empty intervals are omitted. Since ρ i = r + 1 − i \rho_i=r+1-i ρ i = r + 1 − i , we have
( y s ⋅ 0 ) i = p ν i + i − σ s − 1 ( i ) . (y_s\mathbin{\cdot}0)_i=p\nu_i+i-\sigma_s^{-1}(i). ( y s ⋅ 0 ) i = p ν i + i − σ s − 1 ( i ) .
This proves the displayed coordinate formula. Its coordinates sum to zero and are integers, so it lies in the root lattice. Its consecutive coordinate differences are nonnegative and belong to
{ 0 , 1 , 2 , s + 2 , p − 2 , p − s − 2 } . \{0,1,2,s+2,p-2,p-s-2\}. { 0 , 1 , 2 , s + 2 , p − 2 , p − s − 2 } .
As s ≤ r − 1 s\leq r-1 s ≤ r − 1 and p > 2 r + 1 p>2r+1 p > 2 r + 1 , all these differences are less than p p p . They are the simple-coroot pairings, proving dominance and p p p -restrictedness. The coordinate p − 2 p-2 p − 2 is positive, so the weight is nonzero. □
We also use eventual injectivity of restriction.
Lemma 2.18. Let M M M be a finite-dimensional rational G G G -module over k k k admitting an F p \mathbb{F}_p F p -form. Then, for each fixed degree j j j and all sufficiently large a ∈ Z > 0 a\in\mathbb{Z}_{>0} a ∈ Z > 0 , restriction induces an injection
H j ( G , M ) ⟶ H j ( G ( F p a ) , M ) . H^j(G,M)\longrightarrow H^j(G(\mathbb{F}_{p^a}),M). H j ( G , M ) ⟶ H j ( G ( F p a ) , M ) .
Here the source is rational cohomology of the algebraic group G G G and the target is ordinary group cohomology of the finite group G ( F p a ) G(\mathbb{F}_{p^a}) G ( F p a ) . The assertion over F p \mathbb{F}_p F p is stated explicitly in [31 Theorem 13.2] , as a consequence of [10 Theorem 6.6] and [12 Theorem 2.1] . The stated version over k k k follows by extension of scalars.
Proof of Theorem C. Fix n ≥ 1 n \ge1 n ≥ 1 and r ≥ max { 2 , n } r \ge\max\{2,n\} r ≥ max { 2 , n } , and set s = n − 1 s=n-1 s = n − 1 and N = 2 r + 1 N=2r+1 N = 2 r + 1 . Choose p > N p>N p > N large enough so that (2.17) applies in type A N − 1 A_{N-1} A N − 1 , as in Subsection 2.5. Put L = L ( λ r , s ) L=L(\lambda_{r,s}) L = L ( λ r , s ) .
By Lemma 2.17 , the weight λ r , s \lambda_{r,s} λ r , s is nonzero and p p p -restricted. Thus (2.17) and Theorem B give
dim k H n ( G , L ) = [ v n ] h y n − 1 ( v ) = r . \dim_k H^n(G,L)=[v^n]h_{y_{n-1}}(v)=r. dim k H n ( G , L ) = [ v n ] h y n − 1 ( v ) = r .
The simple highest-weight module L L L admits an F p \mathbb{F}_p F p -form, so Lemma 2.18 applies. Keeping p p p and L L L fixed, we obtain, for every sufficiently large a ∈ Z > 0 a\in\mathbb{Z}_{>0} a ∈ Z > 0 ,
dim k H n ( G ( F q ) , L ) ≥ r , with q = p a . \dim_k H^n(G(\mathbb{F}_q),L)\ge r,\qquad\text{with }q=p^a. dim k H n ( G ( F q ) , L ) ≥ r , with q = p a .
The required threshold may depend on p p p , r r r and n n n .
We now repeat the argument in the proof of Theorem A, with q q q in place of p p p . Since λ r , s \lambda_{r,s} λ r , s is nonzero and p p p -restricted, it is also q q q -restricted. By Steinberg’s restriction theorem, L ( λ r , s ) L(\lambda_{r,s}) L ( λ r , s ) restricts to a nontrivial irreducible k k k -module for S L N ( F q ) \mathrm{SL}_N(\mathbb{F}_q) SL N ( F q ) . Moreover, since λ r , s \lambda_{r,s} λ r , s lies in the root lattice, the center acts trivially, so this module descends to a nontrivial irreducible k k k -module V V V for P S L N ( F q ) \mathrm{PSL}_N(\mathbb{F}_q) PSL N ( F q ) . This module is faithful because P S L N ( F q ) \mathrm{PSL}_N(\mathbb{F}_q) PSL N ( F q ) is simple (N = 2 r + 1 ≥ 5 N=2r+1\ge5 N = 2 r + 1 ≥ 5 ), and absolutely irreducible because k k k is algebraically closed.
As explained in that proof, inflation across the center, whose order is prime to p p p , is an isomorphism in every degree. Hence
H n ( P S L N ( F q ) , V ) ≅ H n ( S L N ( F q ) , L ) , H^n(\mathrm{PSL}_N(\mathbb{F}_q),V)\cong H^n(\mathrm{SL}_N(\mathbb{F}_q),L), H n ( PSL N ( F q ) , V ) ≅ H n ( SL N ( F q ) , L ) ,
and therefore
dim k H n ( P S L N ( F q ) , V ) ≥ r . \dim_k H^n(\mathrm{PSL}_N(\mathbb{F}_q),V)\ge r. dim k H n ( PSL N ( F q ) , V ) ≥ r .
Finally, with n n n fixed, let r r r tend to infinity. This excludes an absolute bound in degree n n n . □
Remark 2.19. The prime p p p is allowed to depend on r r r . Thus Theorem C excludes a bound depending only on the cohomological degree, but does not exclude a bound depending on both the degree and the characteristic.
Corollary 2.20. For every n ≥ 1 n\ge1 n ≥ 1 and r ≥ max { 2 , n } r\ge\max\{2,n\} r ≥ max { 2 , n } , and every sufficiently large prime p p p , the algebraic group G = S L 2 r + 1 G=\mathrm{SL}_{2r+1} G = SL 2 r + 1 over k = F p k=\mathbb{F}_p k = F p has a simple rational module L L L such that
dim k H n ( G , L ) = r . \dim_k H^n(G,L)=r. dim k H n ( G , L ) = r .
In particular, these dimensions are unbounded in every fixed positive degree.
Proof. This is the rational cohomology calculation in the proof of Theorem C. □
Quantum groups
In the quantum case, the polynomials computed above determine the cohomology directly. Let N ≥ 2 N\ge2 N ≥ 2 , let ℓ > N \ell>N ℓ > N be odd, and let ζ ∈ C \zeta\in\mathbb{C} ζ ∈ C be a primitive ℓ \ell ℓ th root of unity. Let U ζ U_\zeta U ζ be Lusztig’s quantum enveloping algebra (with divided powers) of s l N ( C ) \mathfrak{sl}_N(\mathbb{C}) sl N ( C ) . We work in the category of finite-dimensional integrable type 1 modules. We define cohomology in this category by
H n ( U ζ , M ) = Ext U ζ n ( C , M ) . H^n(U_\zeta,M)=\operatorname{Ext}_{U_\zeta}^n(\mathbb{C},M). H n ( U ζ , M ) = Ext U ζ n ( C , M ) .
For a dominant weight λ \lambda λ , we write L ζ ( λ ) L_\zeta(\lambda) L ζ ( λ ) for the irreducible module of highest weight λ \lambda λ . We use the same affine action as in Subsection 2.5, with ℓ \ell ℓ in place of p p p : the generators s i s_i s i for 1 ≤ i < N 1\le i<N 1 ≤ i < N act as the finite simple reflections, and
s 0 ( u ) = u − ( ⟨ u , θ ∨ ⟩ − ℓ ) θ . s_0(u)=u-\left(\langle u,\theta^\vee\rangle-\ell\right)\theta. s 0 ( u ) = u − ( ⟨ u , θ ∨ ⟩ − ℓ ) θ .
These extend to an action of W W W by affine transformations, and we write w ⋅ ℓ λ = w ( λ + ρ ) − ρ w\mathbin{\cdot_\ell}\lambda=w(\lambda+\rho)-\rho w ⋅ ℓ λ = w ( λ + ρ ) − ρ for the corresponding dot action. For x ∈ I W x\in{}^IW x ∈ I W , the cohomological Kazhdan–Lusztig formula [26 (2.3.1)] gives, in our notation,
∑ n ≥ 0 dim C H n ( U ζ , L ζ ( x ⋅ ℓ 0 ) ) v n = h w I , w I x ( v ) = h x ( v ) . (19) \sum_{n\ge0}\dim_{\mathbb{C}}H^n(U_\zeta,L_\zeta(x\mathbin{\cdot_\ell}0))v^n=h_{w_I,w_Ix}(v)=h_x(v). \tag*{(19)} n ≥ 0 ∑ dim C H n ( U ζ , L ζ ( x ⋅ ℓ 0 )) v n = h w I , w I x ( v ) = h x ( v ) . (19)
where the second equality follows from Lemma 2.1 .
We explain how to obtain this identity from Ko’s formula. In type A A A , every positive integer ℓ \ell ℓ is KL-good [26 §2.2] , and ℓ > N \ell>N ℓ > N ensures that − 2 ρ -2\rho − 2 ρ lies in the interior of the antidominant fundamental alcove. Ko’s dot action is on our ℓ \ell ℓ ; the difference lies in the choice of Coxeter generators. We therefore take λ = − 2 ρ \lambda=-2\rho λ = − 2 ρ , y = w I y=w_I y = w I , and w = x w I w=xw_I w = x w I in [26 (2.3.1)] . These choices give
w I ⋅ ℓ ( − 2 ρ ) = 0 , ( x w I ) ⋅ ℓ ( − 2 ρ ) = x ⋅ ℓ 0. w_I\mathbin{\cdot_\ell}(-2\rho)=0,\qquad(xw_I)\mathbin{\cdot_\ell}(-2\rho)=x\mathbin{\cdot_\ell}0. w I ⋅ ℓ ( − 2 ρ ) = 0 , ( x w I ) ⋅ ℓ ( − 2 ρ ) = x ⋅ ℓ 0.
Both weights are dominant because x ∈ I W x \in{}^{I}W x ∈ I W . Since the standard module of highest weight 0 0 0 is C \mathbb{C} C , the left-hand side of Ko’s formula is the cohomology series in (19) .
For the right-hand side, Ko uses the Coxeter generators associated with the antidominant fundamental alcove. Conjugation by w I w_I w I identifies this Coxeter system with ours and sends ( w I , x w I ) (w_I,xw_I) ( w I , x w I ) to ( w I , w I x ) (w_I,w_Ix) ( w I , w I x ) . Thus, in our normalization, Ko’s right-hand side is h w I , w I x ( v ) h_{w_I,w_Ix}(v) h w I , w I x ( v ) , proving (19) .
Proof of Theorem D. Fix r ≥ 2 r \ge2 r ≥ 2 and put N = 2 r + 1 N=2r+1 N = 2 r + 1 . For 0 ≤ s ≤ r − 1 0 \le s \le r-1 0 ≤ s ≤ r − 1 , put
λ r , s = y s ⋅ ℓ 0. \lambda_{r,s}=y_s\cdot\ell0. λ r , s = y s ⋅ ℓ 0.
The calculation in Lemma 2.17 applies with ℓ \ell ℓ in place of p p p , since it uses only the inequality p > N p>N p > N , not primality. In particular, λ r , s \lambda_{r,s} λ r , s is a nonzero dominant weight. By Lemma 2.4, we have y s ∈ I W y_s \in{}^{I}W y s ∈ I W , and applying (19) and Theorem B gives
∑ n ≥ 0 dim C H n ( U ζ , L ζ ( λ r , s ) ) v n = h y s ( v ) = r v s + 1 + O ( v s + 3 ) . \sum_{n\ge0}\dim_{\mathbb{C}} H^n(U_{\zeta},L_{\zeta}(\lambda_{r,s}))v^n=h_{y_s}(v)=rv^{s+1}+O(v^{s+3}). n ≥ 0 ∑ dim C H n ( U ζ , L ζ ( λ r , s )) v n = h y s ( v ) = r v s + 1 + O ( v s + 3 ) .
Comparing coefficients we obtain that
dim C H s + 1 ( U ζ , L ζ ( λ r , s ) ) = r , \dim_{\mathbb{C}}H^{s+1}(U_{\zeta},L_{\zeta}(\lambda_{r,s}))=r, dim C H s + 1 ( U ζ , L ζ ( λ r , s )) = r ,
and that the cohomology vanishes in every degree 0 ≤ n < s + 1 0 \le n < s+1 0 ≤ n < s + 1 . □
The parabolic construction of the initial coefficient
In this section we write the canonical basis element of X 0 X_0 X 0 as a combination of normalized parabolic products, read off its identity coefficient, and prove one further coefficient estimate. The final subsection computes the polynomials h γ j h_{\gamma_j} h γ j . These results are the inputs to Section 5, where the multiplication formulas of Theorem 2.8 and the coefficient estimates of Theorem 2.9 for the first chain x 0 = X 0 , … , x r − 1 x_0=X_0,\ldots,x_{r-1} x 0 = X 0 , … , x r − 1 are proved together.
For large r r r the element C X 0 C_{X_0} C X 0 is too long to calculate using the Kazhdan–Lusztig algorithm, and in any case we need it for all r r r at once. This leads us to build it as a combination of products of symmetrizers C w P C_{w_P} C w P of finite parabolic subgroups. Such products are bar-invariant by construction. We show (Proposition 3.5) that
C X 0 = B S r − B S r − 1 + ( v + v − 1 ) ∑ s = 2 r − 1 C B S s . C_{X_0}=\mathcal{B}S_r-\mathcal{B}S_{r-1}+(v+v^{-1})\sum_{s=2}^{r-1}C\mathcal{B}S_s. C X 0 = B S r − B S r − 1 + ( v + v − 1 ) s = 2 ∑ r − 1 C B S s .
where B S j \mathcal{B}S_j B S j and C B S s C\mathcal{B}S_s C B S s are the products of Definition 3.2. By Lemma 3.1 it is enough to see that the right-hand side equals M X 0 M_{X_0} M X 0 plus terms in v Z [ v ] M c v\mathbb{Z}[v]M_c v Z [ v ] M c ; so we need the lowest power of v v v in each standard coefficient. These coefficients are, up to explicit powers of v v v , constant on orbits of words in three letters (Subsections 3.2 and 3.3), and one computation per orbit suffices (Proposition 3.21). It turns out that B S r − B S r − 1 \mathcal{B}S_r-\mathcal{B}S_{r-1} B S r − B S r − 1 has the right coefficient at X 0 X_0 X 0 but a spurious − v − 1 -v^{-1} − v − 1 at certain orbit maxima, and the terms ( v + v − 1 ) C B S s (v+v^{-1})C\mathcal{B}S_s ( v + v − 1 ) C B S s cancel exactly this.
Having the formula, we apply ε \varepsilon ε . All factors after the first lie in W I W_I W I , so each acts by a scalar (23) , and the result is a sum of Gaussian binomial coefficients. The partition numbers appear in this way in Subsection 3.5.
Products of symmetrizers
We put q = v 2 q=v^2 q = v 2 . For n ≥ 1 n\ge1 n ≥ 1 write
[ n ] q = 1 + q + ⋯ + q n − 1 , [ n ] q ! = ∏ i = 1 n [ i ] q , [ 0 ] q ! = 1 , [ n k ] q = [ n ] q ! [ k ] q ! [ n − k ] q ! . [n]_q=1+q+\cdots+q^{n-1},\qquad[n]_q!=\prod_{i=1}^{n}[i]_q,\qquad[0]_q!=1,\qquad\begin{bmatrix}n\\k\end{bmatrix}_q=\frac{[n]_q!}{[k]_q![n-k]_q!}. [ n ] q = 1 + q + ⋯ + q n − 1 , [ n ] q ! = i = 1 ∏ n [ i ] q , [ 0 ] q ! = 1 , [ n k ] q = [ k ] q ! [ n − k ] q ! [ n ] q ! .
for 0 ≤ k ≤ n 0\le k\le n 0 ≤ k ≤ n be the Gaussian binomial coefficients, which we set to zero outside this range. We write f ~ ∼ v b \widetilde{f}\sim v^b f ∼ v b to mean f ∈ v b ( 1 + v 2 Z [ v 2 ] ) f\in v^b(1+v^2\mathbb{Z}[v^2]) f ∈ v b ( 1 + v 2 Z [ v 2 ]) .
For Q ⊆ I Q\subseteq I Q ⊆ I , we define the position blocks of Q Q Q to be the orbits of W Q W_Q W Q on { 1 , … , N } \{1,\ldots,N\} { 1 , … , N } . For a finite parabolic subset P ⊆ S P\subseteq S P ⊆ S put
C w P = ∑ u ∈ W P v ℓ ( w P ) − ℓ ( u ) H u , π P = ∑ u ∈ W P v ℓ ( w P ) − 2 ℓ ( u ) . C_{w_P}=\sum_{u\in W_P}v^{\ell(w_P)-\ell(u)}H_u,\qquad\pi_P=\sum_{u\in W_P}v^{\ell(w_P)-2\ell(u)}. C w P = u ∈ W P ∑ v ℓ ( w P ) − ℓ ( u ) H u , π P = u ∈ W P ∑ v ℓ ( w P ) − 2 ℓ ( u ) .
As for I I I in (2) , pairing u u u with u s us u s gives
C w P H s = H s C w P = v − 1 C w P ( s ∈ P ) . (20) C_{w_P}H_s=H_sC_{w_P}=v^{-1}C_{w_P}\qquad(s\in P). \tag*{(20)} C w P H s = H s C w P = v − 1 C w P ( s ∈ P ) . (20)
and π P \pi_P π P is bar-invariant (replace u u u by w P u w_Pu w P u ). If T ⊆ P T \subseteq P T ⊆ P , splitting W P W_P W P into cosets W T u W_Tu W T u with u ∈ T W P u \in{}^T W_P u ∈ T W P gives
C w P = C w T ∑ u ∈ T W P v ℓ ( w P ) − ℓ ( w T ) − ℓ ( u ) H u . (21) C_{w_P}=C_{w_T}\sum_{u\in{}^T W_P}v^{\ell(w_P)-\ell(w_T)-\ell(u)}H_u. \tag*{(21)} C w P = C w T u ∈ T W P ∑ v ℓ ( w P ) − ℓ ( w T ) − ℓ ( u ) H u . (21)
For Q ⊆ I Q\subseteq I Q ⊆ I , the Poincaré polynomial of S m S_m S m is [ m ] q ! [m]_q! [ m ] q ! , so
π Q = ∏ B Π ∣ B ∣ , Π m : = v − ( m 2 ) [ m ] q ! , (22) \pi_Q=\prod_B\Pi_{\lvert B\rvert},\qquad\Pi_m:=v^{-\binom{m}{2}}[m]_q!, \tag*{(22)} π Q = B ∏ Π ∣ B ∣ , Π m := v − ( 2 m ) [ m ] q ! , (22)
the product running over the position blocks B B B of Q Q Q . In particular π Q ∼ v − ℓ ( w Q ) \pi_Q\sim v^{-\ell(w_Q)} π Q ∼ v − ℓ ( w Q ) . Lemma 2.2 gives ε ( F H u ) = v − ℓ ( u ) ε ( F ) \varepsilon(FH_u)=v^{-\ell(u)}\varepsilon(F) ε ( F H u ) = v − ℓ ( u ) ε ( F ) for u ∈ W I u\in W_I u ∈ W I , hence
ε ( F C w Q ) = π Q ε ( F ) ( F ∈ M , Q ⊆ I ) . (23) \varepsilon(FC_{w_Q})=\pi_Q\varepsilon(F)\qquad(F\in\mathcal{M},\ Q\subseteq I). \tag*{(23)} ε ( F C w Q ) = π Q ε ( F ) ( F ∈ M , Q ⊆ I ) . (23)
Lemma 3.1. Let F ∈ M F\in\mathcal{M} F ∈ M be bar-invariant, and suppose F − M x ∈ L : = ⨁ c ∈ I W v Z [ v ] M c F-M_x\in\mathcal{L}:=\bigoplus_{c\in I W}v\mathbb{Z}[v]M_c F − M x ∈ L := ⨁ c ∈ I W v Z [ v ] M c for some x ∈ I W x\in I W x ∈ I W . Then F = C x F=C_x F = C x .
Proof. Since C x ∈ M x + L C_x\in M_x+\mathcal{L} C x ∈ M x + L , the element F − C x F-C_x F − C x is bar-invariant and lies in L \mathcal{L} L . By [37 Claim 2.4] it is zero.
Definition 3.2. Let r ≥ 2 r\geq2 r ≥ 2 and N = 2 r + 1 N=2r+1 N = 2 r + 1 . For a sequence ( P 1 , … , P k ) (P_1,\ldots,P_k) ( P 1 , … , P k ) of finite parabolic subsets of S S S , put P 0 = I P_0=I P 0 = I and
B S ( P 1 , … , P k ) = C w I C w P 1 ⋯ C w P k ∏ i = 1 k π P i − 1 ∩ P i . (24) BS(P_1,\ldots,P_k)=\frac{C_{w_I}C_{w_{P_1}}\cdots C_{w_{P_k}}}{\prod_{i=1}^k\pi_{P_{i-1}\cap P_i}}. \tag*{(24)} B S ( P 1 , … , P k ) = ∏ i = 1 k π P i − 1 ∩ P i C w I C w P 1 ⋯ C w P k . (24) Let E = { s 3 , … , s N − 1 } E=\{s_3,\ldots,s_{N-1}\} E = { s 3 , … , s N − 1 } and, for 1 ≤ s < N 1\leq s<N 1 ≤ s < N and 1 ≤ j ≤ r 1\leq j\leq r 1 ≤ j ≤ r ,
P s = S ∖ { s s } , J j = S ∖ { s j , s j + 1 } , K j = I ∖ { s 1 , s j + 1 } . P_s=S\setminus\{s_s\},\qquad J_j=S\setminus\{s_j,s_{j+1}\},\qquad K_j=I\setminus\{s_1,s_{j+1}\}. P s = S ∖ { s s } , J j = S ∖ { s j , s j + 1 } , K j = I ∖ { s 1 , s j + 1 } . We set
B S j = B S ( J j , K j , E ) ( 1 ≤ j ≤ r ) , C B S s = B S ( P s , E ) ( 2 ≤ s ≤ r − 1 ) . BS_j=BS(J_j,K_j,E)\quad(1\leq j\leq r),\qquad CBS_s=BS(P_s,E)\quad(2\leq s\leq r-1). B S j = B S ( J j , K j , E ) ( 1 ≤ j ≤ r ) , C B S s = B S ( P s , E ) ( 2 ≤ s ≤ r − 1 ) .
Lemma 3.3 (Overlaps cancel). For finite parabolic subsets P 1 , … , P k P_1,\ldots,P_k P 1 , … , P k the element B S ( P 1 , … , P k ) BS(P_1,\ldots,P_k) B S ( P 1 , … , P k ) lies in M \mathcal{M} M and is bar-invariant, and
B S ( P 1 , … , P k ) H s = v − 1 B S ( P 1 , … , P k ) ( s ∈ P k ) . BS(P_1,\ldots,P_k)H_s=v^{-1}BS(P_1,\ldots,P_k)\qquad(s\in P_k). B S ( P 1 , … , P k ) H s = v − 1 B S ( P 1 , … , P k ) ( s ∈ P k ) . Proof. Put F i = B S ( P 1 , … , P i ) F_i=BS(P_1,\ldots,P_i) F i = B S ( P 1 , … , P i ) , so F 0 = C w I F_0=C_{w_I} F 0 = C w I , which satisfies the claim for P 0 = I P_0=I P 0 = I by (2) . Suppose F k − 1 ∈ M F_{k-1}\in\mathcal{M} F k − 1 ∈ M and F k − 1 H s = v − 1 F k − 1 F_{k-1}H_s=v^{-1}F_{k-1} F k − 1 H s = v − 1 F k − 1 for s ∈ P k − 1 s\in P_{k-1} s ∈ P k − 1 . Let T = P k − 1 ∩ P k T=P_{k-1}\cap P_k T = P k − 1 ∩ P k . Then F k − 1 C w T = π T F k − 1 F_{k-1}C_{w_T}=\pi_TF_{k-1} F k − 1 C w T = π T F k − 1 , since C w T = ∑ u ∈ W T v ℓ ( w T ) − ℓ ( u ) H u C_{w_T}=\sum_{u\in W_T}v^{\ell(w_T)-\ell(u)}H_u C w T = ∑ u ∈ W T v ℓ ( w T ) − ℓ ( u ) H u and H u H_u H u acts on F k − 1 F_{k-1} F k − 1 by v − ℓ ( u ) v^{-\ell(u)} v − ℓ ( u ) . Using (21) with T ⊆ P k T\subseteq P_k T ⊆ P k ,
F k = F k − 1 C w P k π T = F k − 1 ∑ u ∈ T W P k v ℓ ( w P k ) − ℓ ( w T ) − ℓ ( u ) H u . F_k=\frac{F_{k-1}C_{w_{P_k}}}{\pi_T}=F_{k-1}\sum_{u\in{}^T W_{P_k}}v^{\ell(w_{P_k})-\ell(w_T)-\ell(u)}H_u. F k = π T F k − 1 C w P k = F k − 1 u ∈ T W P k ∑ v ℓ ( w P k ) − ℓ ( w T ) − ℓ ( u ) H u . This lies in F k − 1 H ⊆ M F_{k-1}\mathcal{H}\subseteq\mathcal{M} F k − 1 H ⊆ M , and the eigenrelation for s ∈ P k s\in P_k s ∈ P k follows from (20) . Bar-invariance holds because the numerator in (3.5) is a product of bar-invariant elements and each π P \pi_P π P is bar-invariant.
Proposition 3.5. For r ≥ 2 r\geq2 r ≥ 2 we have
C X 0 = B S r − B S r − 1 + ( v + v − 1 ) ∑ s = 2 r − 1 C B S s . C_{X_0}=BS_r-BS_{r-1}+(v+v^{-1})\sum_{s=2}^{r-1}CBS_s. C X 0 = B S r − B S r − 1 + ( v + v − 1 ) s = 2 ∑ r − 1 C B S s . The proof is given in Subsection 3.4, after the expansions of B S j BS_j B S j and C B S s CBS_s C B S s have been computed.
Words, windows and orbits
Elements of I W {}^{I}W I W whose window lies in [ 1 − N , 2 N ] [1-N,2N] [ 1 − N , 2 N ] are encoded by words in three letters, and the sets on which right multiplication by C w Q C_{wQ} C w Q acts turn out to be sets of rearrangements of a word. We use the standard length formula [7 Section 8.3]
ℓ ( w ) = ∑ 1 ≤ i < j ≤ N ⌊ w ( j ) − w ( i ) N ⌋ . (25) \ell(w)=\sum_{1\leq i<j\leq N}\left\lfloor\frac{w(j)-w(i)}{N}\right\rfloor. \tag*{(25)} ℓ ( w ) = 1 ≤ i < j ≤ N ∑ ⌊ N w ( j ) − w ( i ) ⌋ . (25)
Definition 3.6. Let t ∈ { 0 , … , r } t\in\{0,\ldots,r\} t ∈ { 0 , … , r } . We define an admissible word of type t t t to be a word ω = ω 1 ⋯ ω N \omega=\omega_1\cdots\omega_N ω = ω 1 ⋯ ω N in the letters A, B, C such that A and C appear exactly t t t times and B appears exactly N − 2 t N-2t N − 2 t times. We write d ω , i ( X ) d_{\omega,i}(X) d ω , i ( X ) for the number of times X appears in ω 1 , … , ω i \omega_1,\ldots,\omega_i ω 1 , … , ω i . We define the window c ω = [ c ω ( 1 ) , … , c ω ( N ) ] c_\omega=[c_\omega(1),\ldots,c_\omega(N)] c ω = [ c ω ( 1 ) , … , c ω ( N )] as follows:
c ω ( i ) = { d ω , i ( A ) − t if ω i = A , d ω , i ( B ) + t if ω i = B , d ω , i ( C ) + N if ω i = C . c_\omega(i)=
\begin{cases}
d_{\omega,i}(\mathrm{A})-t & \text{if }\omega_i=\mathrm{A},\\
d_{\omega,i}(\mathrm{B})+t & \text{if }\omega_i=\mathrm{B},\\
d_{\omega,i}(\mathrm{C})+N & \text{if }\omega_i=\mathrm{C}.
\end{cases} c ω ( i ) = ⎩ ⎨ ⎧ d ω , i ( A ) − t d ω , i ( B ) + t d ω , i ( C ) + N if ω i = A , if ω i = B , if ω i = C . We will use admissible words to describe permutations like those in Figures 1 and Figure 2 where the entries A, B, C record whether an entry is non-positive, positive but less than or equal to N N N , or strictly greater than N N N respectively. We have that the entries of c ω c_\omega c ω at the letters A, B, C are, in increasing order of position
A t = ( 1 − t , … , 0 ) , B t = ( t + 1 , … , N − t ) , C t = ( N + 1 , … , N + t ) . (26) A_t=(1-t,\ldots,0),\qquad B_t=(t+1,\ldots,N-t),\qquad C_t=(N+1,\ldots,N+t). \tag*{(26)} A t = ( 1 − t , … , 0 ) , B t = ( t + 1 , … , N − t ) , C t = ( N + 1 , … , N + t ) . (26) By construction, c ω ∈ W c_\omega\in W c ω ∈ W (to see this note that the entries of c ω c_\omega c ω have distinct residues modulo N N N and sum to $N(N+1)/2).
Figure 2. The affine string diagram for X 0 X_0 X 0 , with N = 11 N = 11 N = 11 . Compare X 0 ( s 2 s 3 s 4 s 5 ) ( s 1 s 2 s 3 s 4 ) X_0(s_2s_3s_4s_5)(s_1s_2s_3s_4) X 0 ( s 2 s 3 s 4 s 5 ) ( s 1 s 2 s 3 s 4 ) with the diagram in Figure 1 (see also Theorem 2.4). To see that this is an element of I W I^W I W simply note that the strands that connect to the top vertices 1 , … , 11 1,\ldots,11 1 , … , 11 are non-crossing. The labels A, B, C are used in § 3 .
Lemma 3.7. If ω \omega ω is an admissible word of type t t t , then c ω ∈ I W c_\omega\in{}^{I}W c ω ∈ I W . Conversely, every x ∈ I W x\in{}^{I}W x ∈ I W whose window lies in [ 1 − N , 2 N ] [1-N,2N] [ 1 − N , 2 N ] equals c ω c_\omega c ω for a unique admissible word ω \omega ω .
Proof. Let p 1 < ⋯ < p t p_1<\cdots<p_t p 1 < ⋯ < p t , q 1 < ⋯ < q N − 2 t q_1<\cdots<q_{N-2t} q 1 < ⋯ < q N − 2 t and u 1 < ⋯ < u t u_1<\cdots<u_t u 1 < ⋯ < u t be the positions of A, B and C in ω \omega ω . We have that
c ω − 1 ( a ) = u a − N , c ω − 1 ( t + b ) = q b , c ω − 1 ( N − t + a ) = p a + N c_\omega^{-1}(a)=u_a-N,\qquad c_\omega^{-1}(t+b)=q_b,\qquad c_\omega^{-1}(N-t+a)=p_a+N c ω − 1 ( a ) = u a − N , c ω − 1 ( t + b ) = q b , c ω − 1 ( N − t + a ) = p a + N for 1 ≤ a ≤ t 1\leq a\leq t 1 ≤ a ≤ t , 1 ≤ b ≤ N − 2 t 1\leq b\leq N-2t 1 ≤ b ≤ N − 2 t . The groups are increasing and lie in [ 1 − N , 0 ] [1-N,0] [ 1 − N , 0 ] , [ 1 , N ] [1,N] [ 1 , N ] and [ N + 1 , 2 N ] [N+1,2N] [ N + 1 , 2 N ] respectively; therefore c ω − 1 ( 1 ) < ⋯ < c ω − 1 ( N ) c_\omega^{-1}(1)<\cdots<c_\omega^{-1}(N) c ω − 1 ( 1 ) < ⋯ < c ω − 1 ( N ) . Hence c ω ∈ I W c_\omega\in{}^{I}W c ω ∈ I W .
We now suppose that x ∈ I W x\in{}^{I}W x ∈ I W has window in [ 1 − N , 2 N ] [1-N,2N] [ 1 − N , 2 N ] . We let u u u (respectively t t t ) be the total number of entries of x x x that are strictly greater than N N N (respectively non-positive). By assumption x − 1 x^{-1} x − 1 is increasing on 1 , … , N 1,\ldots,N 1 , … , N ; therefore these entries are N + 1 , … , N + u N+1,\ldots,N+u N + 1 , … , N + u and 1 − t , … , 0 1-t,\ldots,0 1 − t , … , 0 . The remaining N − u − t N-u-t N − u − t entries are given by u + 1 , … , N − t u+1,\ldots,N-t u + 1 , … , N − t in increasing order. The entries sum to N ( N + 1 ) / 2 + N ( u − t ) N(N+1)/2+N(u-t) N ( N + 1 ) /2 + N ( u − t ) and x ∈ W x\in W x ∈ W by assumption, and so u = t u=t u = t . Replacing the three lists by C, A, B gives the unique word.
Lemma 3.8. Let ω \omega ω be admissible and s i ∈ I s_i\in I s i ∈ I with ω i = ω i + 1 \omega_i=\omega_{i+1} ω i = ω i + 1 . Then c ω s i = s ′ c ω c_\omega s_i=s'c_\omega c ω s i = s ′ c ω for some s ′ ∈ I s'\in I s ′ ∈ I , ℓ ( c ω s i ) = ℓ ( c ω ) + 1 \ell(c_\omega s_i)=\ell(c_\omega)+1 ℓ ( c ω s i ) = ℓ ( c ω ) + 1 , and M c ω H s i = v − 1 M c ω M_{c_\omega}H_{s_i}=v^{-1}M_{c_\omega} M c ω H s i = v − 1 M c ω .
Proof. Equal adjacent letters give c ω ( i + 1 ) = c ω ( i ) + 1 c_\omega(i+1)=c_\omega(i)+1 c ω ( i + 1 ) = c ω ( i ) + 1 , and both values lie in one interval [ k N + 1 , ( k + 1 ) N ] [kN+1,(k+1)N] [ k N + 1 , ( k + 1 ) N ] . Hence c ω s i c ω − 1 c_\omega s_i c_\omega^{-1} c ω s i c ω − 1 , which exchanges the values c ω ( i ) c_\omega(i) c ω ( i ) and c ω ( i ) + 1 c_\omega(i)+1 c ω ( i ) + 1 (and their translates by N Z N\mathbb{Z} N Z ), is a simple reflection s ′ ∈ I s'\in I s ′ ∈ I , and c ω ( i ) < c ω ( i + 1 ) c_\omega(i)<c_\omega(i+1) c ω ( i ) < c ω ( i + 1 ) gives ℓ ( c ω s i ) = ℓ ( c ω ) + 1 = ℓ ( s ′ c ω ) \ell(c_\omega s_i)=\ell(c_\omega)+1=\ell(s'c_\omega) ℓ ( c ω s i ) = ℓ ( c ω ) + 1 = ℓ ( s ′ c ω ) . So M c ω H s i = C w I H s ′ c ω = C w I H s ′ H c ω = v − 1 M c ω M_{c_\omega}H_{s_i}=C_{w_I}H_{s'c_\omega}=C_{w_I}H_{s'}H_{c_\omega}=v^{-1}M_{c_\omega} M c ω H s i = C w I H s ′ c ω = C w I H s ′ H c ω = v − 1 M c ω by (2.2).
Lemma 3.9. Let ω \omega ω be an admissible word of type t t t , and let ι ( ω ) \iota(\omega) ι ( ω ) be its inversion number for the letter order C < < < B < < < A,
ι ( ω ) = # { ( i , j ) : 1 ≤ i < j ≤ N , ω i > ω j } . \iota(\omega)=\#\{(i,j):1\leq i<j\leq N,\ \omega_i>\omega_j\}. ι ( ω ) = # {( i , j ) : 1 ≤ i < j ≤ N , ω i > ω j } . Then ℓ ( c ω ) = 2 t ( N − t ) − ι ( ω ) \ell(c_\omega)=2t(N-t)-\iota(\omega) ℓ ( c ω ) = 2 t ( N − t ) − ι ( ω ) .
Proof. Let ω 0 = A t B N − 2 t C t \omega_0=\mathrm{A}^t\mathrm{B}^{N-2t}\mathrm{C}^t ω 0 = A t B N − 2 t C t . Its window is increasing, and in (25) only the pairs (an entry of A t A_t A t , an entry of C t C_t C t ) contribute, each by one. So ℓ ( c ω 0 ) = t 2 \ell(c_{\omega_0})=t^2 ℓ ( c ω 0 ) = t 2 , while ι ( ω 0 ) = t 2 + 2 t ( N − 2 t ) \iota(\omega_0)=t^2+2t(N-2t) ι ( ω 0 ) = t 2 + 2 t ( N − 2 t ) . Suppose σ ′ \sigma' σ ′ is obtained from σ \sigma σ by one of the interchanges AB → \to → BA, AC → \to → CA or BC → \to → CB in positions i , i + 1 i,i+1 i , i + 1 . The two letters keep their counts, so c σ ′ = c σ s i c_{\sigma'}=c_\sigma s_i c σ ′ = c σ s i , and c σ ( i ) < c σ ( i + 1 ) c_\sigma(i)<c_\sigma(i+1) c σ ( i ) < c σ ( i + 1 ) because A, B, C entries are ≤ 0 \leq0 ≤ 0 , in [ 1 , N ] [1,N] [ 1 , N ] and > N >N > N respectively. Hence ℓ ( c σ ′ ) = ℓ ( c σ ) + 1 \ell(c_{\sigma'})=\ell(c_\sigma)+1 ℓ ( c σ ′ ) = ℓ ( c σ ) + 1 , whereas ι ( σ ′ ) = ι ( σ ) − 1 \iota(\sigma')=\iota(\sigma)-1 ι ( σ ′ ) = ι ( σ ) − 1 . Every admissible word of type t t t is reached from ω 0 \omega_0 ω 0 by such interchanges, so ℓ ( c ω ) + ι ( ω ) = t 2 + ι ( ω 0 ) = 2 t ( N − t ) \ell(c_\omega)+\iota(\omega)=t^2+\iota(\omega_0)=2t(N-t) ℓ ( c ω ) + ι ( ω ) = t 2 + ι ( ω 0 ) = 2 t ( N − t ) .
Definition 3.10. For x ∈ I W x \in{}^{I}W x ∈ I W and Q ⊆ I Q \subseteq I Q ⊆ I put O Q ( x ) = I W ∩ W I x W Q \mathcal{O}_{Q}(x) = {}^{I}W \cap W_{I}xW_{Q} O Q ( x ) = I W ∩ W I x W Q . It consists of the minimal representatives of the cosets W I x q W_{I}xq W I x q with q ∈ W Q q \in W_{Q} q ∈ W Q ; it is finite, depends only on W I x W Q W_{I}xW_{Q} W I x W Q , and for fixed Q Q Q these sets partition I W {}^{I}W I W .
Lemma 3.11. Let Q ⊆ I Q \subseteq I Q ⊆ I and ω , ω ′ \omega,\omega' ω , ω ′ be admissible words of type t t t . We write ω ′ ∼ ω \omega' \sim\omega ω ′ ∼ ω if ω , ω ′ \omega,\omega' ω , ω ′ differ only by permutation of the letters within the position blocks of Q Q Q . We have that W I c ω W Q = W I c ω ′ W Q W_{I}c_{\omega}W_{Q}=W_{I}c_{\omega'}W_{Q} W I c ω W Q = W I c ω ′ W Q if and only if ω ′ ∼ ω \omega' \sim\omega ω ′ ∼ ω ; hence O Q ( c ω ) = { c ω ′ ∣ ω ′ ∼ ω } \mathcal{O}_{Q}(c_{\omega})=\{c_{\omega'}\mid\omega'\sim\omega\} O Q ( c ω ) = { c ω ′ ∣ ω ′ ∼ ω } . Let ω Q \omega_{Q} ω Q and ω Q \omega^{Q} ω Q be obtained from ω \omega ω by sorting within each position block of Q Q Q in the orders A < < < B < < < C and C < < < B < < < A respectively. Then c ω Q c_{\omega_{Q}} c ω Q and c ω Q c_{\omega^{Q}} c ω Q are the unique minimum and maximum elements of O Q ( c ω ) \mathcal{O}_{Q}(c_{\omega}) O Q ( c ω ) in the Bruhat order.
Proof. Say that position i i i of x ∈ W x \in W x ∈ W carries the letter A, B or C if x ( i ) ≤ 0 x(i) \leq0 x ( i ) ≤ 0 , x ( i ) ∈ [ 1 , N ] x(i) \in[1,N] x ( i ) ∈ [ 1 , N ] or x ( i ) > N x(i) > N x ( i ) > N . Left multiplication by W I W_{I} W I moves each value inside its interval [ k N + 1 , ( k + 1 ) N ] [kN+1,(k+1)N] [ k N + 1 , ( k + 1 ) N ] , so it preserves the letters carried; right multiplication by τ ∈ W Q \tau\in W_{Q} τ ∈ W Q rearranges them within position blocks of Q Q Q . Since c ω ′ c_{\omega'} c ω ′ carries the word ω ′ \omega' ω ′ , equality of double cosets forces ω ′ \omega' ω ′ to be a rearrangement of ω \omega ω .
Conversely let ω ′ \omega' ω ′ be obtained from ω \omega ω by a permutation τ ∈ W Q \tau\in W_{Q} τ ∈ W Q of positions. Then c ω τ c_{\omega}\tau c ω τ and c ω ′ c_{\omega'} c ω ′ carry the same word, and the entries carrying each letter form the same set A t A_t A t , B t B_t B t or C t C_t C t , inside one interval [ k N + 1 , ( k + 1 ) N ] [kN+1,(k+1)N] [ k N + 1 , ( k + 1 ) N ] . Matching them residue by residue defines σ ∈ W I \sigma\in W_{I} σ ∈ W I with σ c ω τ = c ω ′ \sigma c_{\omega}\tau=c_{\omega'} σ c ω τ = c ω ′ . As c ω ′ ∈ I W c_{\omega'} \in{}^{I}W c ω ′ ∈ I W (Lemma 3.7), it is the minimal representative of W I c ω τ W_{I}c_{\omega}\tau W I c ω τ . Every element of W I c ω W Q ∩ I W W_{I}c_{\omega}W_{Q}\cap{}^{I}W W I c ω W Q ∩ I W has window in [ 1 − N , 2 N ] [1-N,2N] [ 1 − N , 2 N ] by the same observation, so it is some c ω ′ c_{\omega'} c ω ′ by Lemma 3.7. This proves the first two assertions.
An interchange of letters at positions i i i , i + 1 i+1 i + 1 of one block is right multiplication by s i ∈ Q s_i\in Q s i ∈ Q , and by the proof of Lemma 3.9 it changes the length by one and the inversion number by one in the opposite direction. Sorting ω ′ \omega' ω ′ within blocks by interchanges AB → \to → BA, AC → \to → CA, BC → \to → CB ends at ω Q \omega^{Q} ω Q , and each step raises the length by one; sorting in the reverse direction ends at ω Q \omega_{Q} ω Q and lowers it by one. So c ω Q ≤ c ω ′ ≤ c ω Q c_{\omega_{Q}}\leq c_{\omega'}\leq c_{\omega^{Q}} c ω Q ≤ c ω ′ ≤ c ω Q in Bruhat order. □
Figure 4. The element c ω Q = [ 12 , 13 , 14 , 15 , 6 , − 4 , 16 , − 3 , − 2 , − 1 , 0 ] c_{\omega^{Q}}=[12,13,14,15,6,-4,16,-3,-2,-1,0] c ω Q = [ 12 , 13 , 14 , 15 , 6 , − 4 , 16 , − 3 , − 2 , − 1 , 0 ] for N = 11 N=11 N = 11 , with ω Q = C 4 B A C A 4 \omega^{Q}=\mathrm{C}^{4}\mathrm{BA}\,\mathrm{CA}^{4} ω Q = C 4 BA CA 4 . It is the longest element of O Q ( c ω ) \mathcal{O}_{Q}(c_{\omega}) O Q ( c ω ) in two cases: for c ω = X 0 c_{\omega}=X_{0} c ω = X 0 (Figure 2) and Q Q Q with position blocks { 1 , … , 6 } , { 7 , … , 11 } \{1,\ldots,6\},\{7,\ldots,11\} { 1 , … , 6 } , { 7 , … , 11 } (black braces), and for c ω = β 5 c_{\omega}=\beta_{5} c ω = β 5 (Figure 1) and Q Q Q with blocks { 1 , … , 4 } , { 5 , 6 } , { 7 , 8 } , { 9 , 10 , 11 } \{1,\ldots,4\},\{5,6\},\{7,8\},\{9,10,11\} { 1 , … , 4 } , { 5 , 6 } , { 7 , 8 } , { 9 , 10 , 11 } (green braces). In each case c ω Q c_{\omega^{Q}} c ω Q is obtained by permuting the bottom nodes inside the braces.
We shall use the sets O E ( x ) \mathcal{O}_{E}(x) O E ( x ) for E = { s 3 , … , s N − 1 } E=\{s_{3},\ldots,s_{N-1}\} E = { s 3 , … , s N − 1 } , whose position blocks are { 1 } \{1\} { 1 } , { 2 } \{2\} { 2 } and { 3 , … , N } \{3,\ldots,N\} { 3 , … , N } . By Lemma 3.11 such a set is determined by the type of ω \omega ω and its first two letters, it consists of the rearrangements of the suffix ω 3 ⋯ ω N \omega_{3}\cdots\omega_{N} ω 3 ⋯ ω N , and its longest element has the suffix sorted in the order C < < < B < < < A. In every product B S j B S_{j} B S j or C B S j C B S_{j} C B S j the first two letters are A or B (Lemma 3.19), so only four prefixes occur.
Definition 3.12. A triple ( t , a , b ) (t,a,b) ( t , a , b ) with t ∈ { 0 , … , r } t \in\{0,\ldots,r\} t ∈ { 0 , … , r } and a , b ∈ { 0 , 1 } a,b \in\{0,1\} a , b ∈ { 0 , 1 } is feasible if t ≥ a + b t \ge a+b t ≥ a + b and N − 2 t − 2 + a + b ≥ 0 N-2t-2+a+b \ge0 N − 2 t − 2 + a + b ≥ 0 . For such a triple put
ω ( t , a , b ) = A a B 1 − a A b B 1 − b C t B N − 2 t − 2 + a + b A t − a − b , z ( t , a , b ) = c ω ( t , a , b ) ∈ W . \omega(t,a,b)=A^{a}B^{1-a}A^{b}B^{1-b}C^{t}B^{N-2t-2+a+b}A^{t-a-b},\qquad z(t,a,b)=c_{\omega(t,a,b)}\in W. ω ( t , a , b ) = A a B 1 − a A b B 1 − b C t B N − 2 t − 2 + a + b A t − a − b , z ( t , a , b ) = c ω ( t , a , b ) ∈ W . The pairs ( a , b ) = ( 0 , 0 ) , ( 0 , 1 ) , ( 1 , 0 ) , ( 1 , 1 ) (a,b)=(0,0),(0,1),(1,0),(1,1) ( a , b ) = ( 0 , 0 ) , ( 0 , 1 ) , ( 1 , 0 ) , ( 1 , 1 ) correspond to the prefixes BB, BA, AB, AA. After the prefix, the remaining letters are t t t letters C, N − 2 t − 2 + a + b N-2t-2+a+b N − 2 t − 2 + a + b letters B and t − a − b t-a-b t − a − b letters A; feasibility says that these counts are nonnegative, so ω ( t , a , b ) \omega(t,a,b) ω ( t , a , b ) is admissible of type t t t . Its suffix is sorted in the order C < B < A, so z ( t , a , b ) z(t,a,b) z ( t , a , b ) is the longest element of O E ( z ( t , a , b ) ) \mathcal{O}_{E}(z(t,a,b)) O E ( z ( t , a , b )) by Lemma 3.11.
Lemma 3.13. For a feasible triple ( t , a , b ) (t,a,b) ( t , a , b ) ,
λ ( t , a , b ) : = ℓ ( z ( t , a , b ) ) = 2 t ( N − t ) − 2 t − ( a + b ) ( N − 2 t ) + b ( 1 − a ) . \lambda(t,a,b):=\ell(z(t,a,b))=2t(N-t)-2t-(a+b)(N-2t)+b(1-a). λ ( t , a , b ) := ℓ ( z ( t , a , b )) = 2 t ( N − t ) − 2 t − ( a + b ) ( N − 2 t ) + b ( 1 − a ) . Moreover ( 0 , 0 , 0 ) (0,0,0) ( 0 , 0 , 0 ) and ( r , 1 , 0 ) (r,1,0) ( r , 1 , 0 ) are feasible, and
z ( 0 , 0 , 0 ) = e , z ( r , 1 , 0 ) = X 0 . (27) z(0,0,0)=e,\qquad z(r,1,0)=X_{0}. \tag*{(27)} z ( 0 , 0 , 0 ) = e , z ( r , 1 , 0 ) = X 0 . (27) Proof. The suffix of ω ( t , a , b ) \omega(t,a,b) ω ( t , a , b ) is sorted, so every inversion has its first index equal to 1 or 2. The first letter forms inversions with the t t t letters C, and also with the N − 2 t N-2t N − 2 t letters B if it is A: this contributes t + a ( N − 2 t ) t+a(N-2t) t + a ( N − 2 t ) . The second letter precedes all t t t letters C; if it is A it also precedes the N − 2 t − ( 1 − a ) N-2t-(1-a) N − 2 t − ( 1 − a ) letters B after it (the first letter is a B exactly when a = 0 a=0 a = 0 ). It contributes t + b ( N − 2 t ) − b ( 1 − a ) t+b(N-2t)-b(1-a) t + b ( N − 2 t ) − b ( 1 − a ) . These sets of inversions are disjoint, so ι ( ω ( t , a , b ) ) = 2 t + ( a + b ) ( N − 2 t ) − b ( 1 − a ) \iota(\omega(t,a,b))=2t+(a+b)(N-2t)-b(1-a) ι ( ω ( t , a , b )) = 2 t + ( a + b ) ( N − 2 t ) − b ( 1 − a ) , and Lemma 3.9 gives the length.
Finally ω ( 0 , 0 , 0 ) = B N \omega(0,0,0)=B^{N} ω ( 0 , 0 , 0 ) = B N and ω ( r , 1 , 0 ) = A B C r A r − 1 \omega(r,1,0)=ABC^{r}A^{r-1} ω ( r , 1 , 0 ) = A B C r A r − 1 give, by the replacement rule of Definition 3.6, the elements e e e and X 0 X_{0} X 0 of (2.4). □
Corollary 3.14. ℓ ( X 0 ) = 2 r 2 − 1 \ell(X_{0})=2r^{2}-1 ℓ ( X 0 ) = 2 r 2 − 1 and ℓ ( β r ) = 2 r 2 + 2 r − 3 \ell(\beta_{r})=2r^{2}+2r-3 ℓ ( β r ) = 2 r 2 + 2 r − 3 .
Proof. By (3.8) and Lemma 3.13, ℓ ( X 0 ) = λ ( r , 1 , 0 ) = 2 r ( r + 1 ) − 2 r − 1 = 2 r 2 − 1 \ell(X_{0})=\lambda(r,1,0)=2r(r+1)-2r-1=2r^{2}-1 ℓ ( X 0 ) = λ ( r , 1 , 0 ) = 2 r ( r + 1 ) − 2 r − 1 = 2 r 2 − 1 . The length-additive path of Lemma 2.4 from X 0 X_{0} X 0 to β r \beta_{r} β r has 2 r − 2 2r-2 2 r − 2 steps. □
Example 3.15. Let N = 11 N=11 N = 11 and r = 5 r=5 r = 5 . The word ω = A B C 5 A 4 \omega=ABC^{5}A^{4} ω = A B C 5 A 4 is admissible of type 5, and the replacement rule gives
(the letters A receive A 5 = ( − 4 , … , 0 ) A_{5}=(-4,\ldots,0) A 5 = ( − 4 , … , 0 ) , the B receives B 5 = ( 6 ) B_{5}=(6) B 5 = ( 6 ) and the letters C receive C 5 = ( 12 , … , 16 ) C_{5}=(12,\ldots,16) C 5 = ( 12 , … , 16 ) ), so c ω = X 0 c_{\omega}=X_{0} c ω = X 0 is the element drawn in Figure 2 , and ω = ω ( 5 , 1 , 0 ) \omega=\omega(5,1,0) ω = ω ( 5 , 1 , 0 ) . The prefix BB is not feasible for t = 5 t=5 t = 5 , since N − 2 t − 2 = − 1 N-2t-2=-1 N − 2 t − 2 = − 1 ; the three feasible triples of type 5 are
The orbit O E ( X 0 ) \mathcal{O}_{E}(X_{0}) O E ( X 0 ) consists of the ( 9 4 ) = 126 \binom{9}{4}=126 ( 4 9 ) = 126 elements attached to the words A B ω ′ AB\omega' A B ω ′ with ω ′ \omega' ω ′ a rearrangement of C 5 A 4 C^{5}A^{4} C 5 A 4 . Each exchange AC → \to → CA in the suffix raises the length by one (Lemma 3.9), so the lengths run from 29, at A B A 4 C 5 ABA^{4}C^{5} A B A 4 C 5 , up to 49, at X 0 X_{0} X 0 .
Averaging over a parabolic subgroup
We now describe right multiplication by C w Q C_{w_Q} C w Q on the orbits O Q ( x ) \mathcal{O}_{Q}(x) O Q ( x ) : on each orbit the result is a scalar multiple of a fixed weighted sum of standard basis elements, so its coefficient at the maximum determines all others.
Lemma 3.16. Let Q ⊆ I Q\subseteq I Q ⊆ I , let ω \omega ω be admissible, and put y = c ω Q y=c_{\omega_Q} y = c ω Q , z = c ω Q z=c_{\omega_Q} z = c ω Q and
L = { s i ∈ Q : y ω Q has equal letters in positions i , i + 1 } . L=\{s_i\in Q:y\omega_Q\text{ has equal letters in positions }i,i+1\}. L = { s i ∈ Q : y ω Q has equal letters in positions i , i + 1 } . (i) L = { s ∈ Q : y s y − 1 ∈ I } L=\{s\in Q:ysy^{-1}\in I\} L = { s ∈ Q : y s y − 1 ∈ I } , and for s ∈ L s\in L s ∈ L we have y s = s ′ y ys=s'y y s = s ′ y with s ′ ∈ I s'\in I s ′ ∈ I and ℓ ( y s ) = ℓ ( y ) + 1 \ell(ys)=\ell(y)+1 ℓ ( y s ) = ℓ ( y ) + 1 .
(ii) The map u ↦ y u u\mapsto yu u ↦ y u is a bijection W Q → O Q ( c ω ) W_Q\to\mathcal{O}_Q(c_{\omega}) W Q → O Q ( c ω ) , and ℓ ( y u ) = ℓ ( y ) + ℓ ( u ) \ell(yu)=\ell(y)+\ell(u) ℓ ( y u ) = ℓ ( y ) + ℓ ( u ) .
(iii) ℓ ( z ) − ℓ ( y ) = ℓ ( ω Q ) − ℓ ( ω L ) \ell(z)-\ell(y)=\ell(\omega_Q)-\ell(\omega_L) ℓ ( z ) − ℓ ( y ) = ℓ ( ω Q ) − ℓ ( ω L ) , and this is the number of pairs of positions in a common block of Q Q Q carrying different letters.
Proof. (i) If s i ∈ L s_i \in L s i ∈ L , Lemma 3.8 gives the second statement. If ω Q \omega_Q ω Q has different letters at i , i + 1 i,i+1 i , i + 1 , then y s i y − 1 ys_i y^{-1} y s i y − 1 exchanges y ( i ) y(i) y ( i ) and y ( i + 1 ) y(i+1) y ( i + 1 ) (and their translates), which lie in different intervals [ k N + 1 , ( k + 1 ) N ] [kN+1,(k+1)N] [ k N + 1 , ( k + 1 ) N ] , so it is not in W I W_I W I . (ii) is the standard parametrization of the cosets in a double coset, see [14 Lemmas 2.12 and 2.14] . (iii) By Lemma 3.9, ℓ ( z ) − ℓ ( y ) = ι ( ω Q ) − ι ( ω Q ) \ell(z)-\ell(y)=\iota(\omega_Q)-\iota(\omega^Q) ℓ ( z ) − ℓ ( y ) = ι ( ω Q ) − ι ( ω Q ) . Inversions between different blocks are the same in both words; inside a block, ω Q \omega_Q ω Q is sorted in the order A < B < C A<B<C A < B < C and so has an inversion for every pair of different letters, and ω Q \omega^Q ω Q has none. Finally W Q W_Q W Q and W L W_L W L are products of symmetric groups on the blocks of Q Q Q and on the runs of equal letters of ω Q \omega_Q ω Q , so ℓ ( w Q ) − ℓ ( w L ) \ell(w_Q)-\ell(w_L) ℓ ( w Q ) − ℓ ( w L ) is the same count.
Lemma 3.17. Let ω \omega ω be admissible, x = c ω x=c_\omega x = c ω and Q ⊆ I Q\subseteq I Q ⊆ I . Let y , z y,z y , z be the minimum and maximum of O = O Q ( x ) \mathcal{O}=\mathcal{O}_Q(x) O = O Q ( x ) and L L L as in Lemma 3.16. Then for every c ∈ O c\in\mathcal{O} c ∈ O
M c C w Q = π L v ℓ ( y ) − ℓ ( c ) ∑ d ∈ O v ℓ ( z ) − ℓ ( d ) M d . M_cC_{w_Q}=\pi_L v^{\ell(y)-\ell(c)}\sum_{d\in\mathcal{O}}v^{\ell(z)-\ell(d)}M_d. M c C w Q = π L v ℓ ( y ) − ℓ ( c ) d ∈ O ∑ v ℓ ( z ) − ℓ ( d ) M d .
Proof. For s ∈ L s\in L s ∈ L , Lemma 3.16(i) and (2.2) give M y H s = C w I H s ′ H y = v − 1 M y M_yH_s=C_{w_I}H_{s'}H_y=v^{-1}M_y M y H s = C w I H s ′ H y = v − 1 M y , hence M y C w L = π L M y M_yC_{w_L}=\pi_LM_y M y C w L = π L M y .
By (3.2) and Lemma 3.16(ii),
M y C w Q = π L ∑ u ∈ W P v ℓ ( w Q ) − ℓ ( w L ) − ℓ ( u ) M y u = π L ∑ d ∈ O v ℓ ( z ) − ℓ ( d ) M d , M_yC_{w_Q}=\pi_L\sum_{u\in W_P}v^{\ell(w_Q)-\ell(w_L)-\ell(u)}M_{yu}=\pi_L\sum_{d\in\mathcal{O}}v^{\ell(z)-\ell(d)}M_d, M y C w Q = π L u ∈ W P ∑ v ℓ ( w Q ) − ℓ ( w L ) − ℓ ( u ) M y u = π L d ∈ O ∑ v ℓ ( z ) − ℓ ( d ) M d ,
because ℓ ( z ) − ℓ ( y u ) = ℓ ( w Q ) − ℓ ( w L ) − ℓ ( u ) \ell(z)-\ell(yu)=\ell(w_Q)-\ell(w_L)-\ell(u) ℓ ( z ) − ℓ ( y u ) = ℓ ( w Q ) − ℓ ( w L ) − ℓ ( u ) by (iii). Finally, if c = y u c=yu c = y u then M c = M y H u M_c=M_yH_u M c = M y H u and H u C w Q = v − ℓ ( u ) C w Q H_uC_{w_Q}=v^{-\ell(u)}C_{w_Q} H u C w Q = v − ℓ ( u ) C w Q , so M c C w Q = v ℓ ( y ) − ℓ ( c ) M y C w Q M_cC_{w_Q}=v^{\ell(y)-\ell(c)}M_yC_{w_Q} M c C w Q = v ℓ ( y ) − ℓ ( c ) M y C w Q .
Corollary 3.18. In the setting of Lemma 3.17, let f c f_c f c (c ∈ I W c\in I_W c ∈ I W ) be Laurent polynomials, almost all zero, and put X = ( ∑ c f c M c ) C w Q X=\left(\sum_c f_cM_c\right)C_{w_Q} X = ( ∑ c f c M c ) C w Q . Then for every d ∈ O d\in\mathcal{O} d ∈ O
[ M d ] X = π L v ℓ ( y ) ( ∑ c ∈ O f c v − ℓ ( c ) ) v ℓ ( z ) − ℓ ( d ) , (28) [M_d]X=\pi_Lv^{\ell(y)}\left(\sum_{c\in\mathcal{O}}f_cv^{-\ell(c)}\right)v^{\ell(z)-\ell(d)}, \tag*{(28)} [ M d ] X = π L v ℓ ( y ) ( c ∈ O ∑ f c v − ℓ ( c ) ) v ℓ ( z ) − ℓ ( d ) , (28) and in particular
[ M d ] X = v ℓ ( z ) − ℓ ( d ) [ M z ] X . (29) [M_d]X=v^{\ell(z)-\ell(d)}[M_z]X. \tag*{(29)} [ M d ] X = v ℓ ( z ) − ℓ ( d ) [ M z ] X . (29) Suppose moreover that f c ∼ v A − ℓ ( c ) f_c\sim v^{A-\ell(c)} f c ∼ v A − ℓ ( c ) for every c ∈ O c\in\mathcal{O} c ∈ O with f c ≠ 0 f_c\ne0 f c = 0 , that u u u is the unique longest such c c c , and that G = X / π D ∈ M G=X/\pi_D\in\mathcal{M} G = X / π D ∈ M for some D ⊆ Q D\subseteq Q D ⊆ Q . Then, with g = ℓ ( z ) − ℓ ( u ) g=\ell(z)-\ell(u) g = ℓ ( z ) − ℓ ( u ) ,
[ M d ] G ∼ v A − ℓ ( d ) − ( ℓ ( w Q ) − ℓ ( w D ) ) + 2 g ( d ∈ O ) . (30) [M_d]G\sim v^{A-\ell(d)-(\ell(w_Q)-\ell(w_D))+2g}\qquad(d\in\mathcal{O}). \tag*{(30)} [ M d ] G ∼ v A − ℓ ( d ) − ( ℓ ( w Q ) − ℓ ( w D )) + 2 g ( d ∈ O ) . (30) In particular every coefficient of G G G on O \mathcal{O} O is nonzero.
Proof. C w Q C_{w_Q} C w Q lies in the subalgebra generated by the H s H_s H s , s ∈ Q s\in Q s ∈ Q , and each M c H s M_cH_s M c H s is supported on O Q ( c ) \mathcal{O}_Q(c) O Q ( c ) . These sets partition I W I_W I W , so only c ∈ O c\in\mathcal{O} c ∈ O contribute on O \mathcal{O} O , and (3.9) follows from Lemma 3.17; (3.10) is immediate.
For (3.11): each nonzero summand f c v − ℓ ( c ) f_cv^{-\ell(c)} f c v − ℓ ( c ) is ∼ v A − 2 ℓ ( c ) \sim v^{A-2\ell(c)} ∼ v A − 2 ℓ ( c ) , with a uniquely smallest exponent at c = u c=u c = u , so the sum in (3.9) is ∼ v A − 2 ℓ ( u ) \sim v^{A-2\ell(u)} ∼ v A − 2 ℓ ( u ) . Now use π L ∼ v − ℓ ( w L ) \pi_L\sim v^{-\ell(w_L)} π L ∼ v − ℓ ( w L ) , ℓ ( y ) − ℓ ( w L ) = ℓ ( z ) − ℓ ( w Q ) \ell(y)-\ell(w_L)=\ell(z)-\ell(w_Q) ℓ ( y ) − ℓ ( w L ) = ℓ ( z ) − ℓ ( w Q ) (Lemma 3.16(iii)) and ℓ ( u ) = ℓ ( z ) − g \ell(u)=\ell(z)-g ℓ ( u ) = ℓ ( z ) − g : this gives [ M d ] X ∼ v A − ℓ ( d ) − ℓ ( w Q ) + 2 g [M_d]X\sim v^{A-\ell(d)-\ell(w_Q)+2g} [ M d ] X ∼ v A − ℓ ( d ) − ℓ ( w Q ) + 2 g . Finally π D ∼ v − ℓ ( w D ) \pi_D\sim v^{-\ell(w_D)} π D ∼ v − ℓ ( w D ) , and [ M d ] G [M_d]G [ M d ] G is a Laurent polynomial because G ∈ M G\in\mathcal{M} G ∈ M , so dividing gives (3.11).
The expansions of BSj _j j and CBSs _s s , and proof of Proposition 3.5
We now compute the expansions of the (singular) Bott–Samelson elements B S j BS_j B S j and C B S s CBS_s C B S s in the basis M c M_c M c , using the combinatorics of admissible words from the previous section. We use these to prove Proposition 3.5.
Lemma 3.19. Let 1 ≤ j ≤ r 1\le j\le r 1 ≤ j ≤ r and 1 ≤ s < N 1\le s<N 1 ≤ s < N . Then
B S ( J j ) = ∑ ω v j ( N − j − 1 ) − ℓ ( c ω ) M c ω , B S ( P s ) = ∑ ω ′ v s ( N − s ) − ℓ ( c ω ′ ) M c ω ′ , BS(J_j)=\sum_\omega v^{j(N-j-1)-\ell(c_\omega)}M_{c_\omega},\qquad BS(P_s)=\sum_{\omega'}v^{s(N-s)-\ell(c_{\omega'})}M_{c_{\omega'}}, B S ( J j ) = ω ∑ v j ( N − j − 1 ) − ℓ ( c ω ) M c ω , B S ( P s ) = ω ′ ∑ v s ( N − s ) − ℓ ( c ω ′ ) M c ω ′ , where ω \omega ω runs over the admissible words with letters in { A , B } \{A,B\} { A , B } in positions ≤ j \le j ≤ j , the letter B B B in position j + 1 j+1 j + 1 and letters in { B , C } \{B,C\} { B , C } in positions > j + 2 >j+2 > j + 2 , and ω ′ \omega' ω ′ runs over the admissible words with letters in { A , B } \{A,B\} { A , B } in positions ≤ s \le s ≤ s and in { B , C } \{B,C\} { B , C } in positions > s >s > s .
Proof. Let P ⊆ S P\subseteq S P ⊆ S and L = I ∩ P L=I\cap P L = I ∩ P . By (3.2), C w P = C w L ∑ u ∈ W P v ℓ ( w P ) − ℓ ( w L ) − ℓ ( u ) H u C_{w_P}=C_{w_L}\sum_{u\in W_P}v^{\ell(w_P)-\ell(w_L)-\ell(u)}H_u C w P = C w L ∑ u ∈ W P v ℓ ( w P ) − ℓ ( w L ) − ℓ ( u ) H u , and C w I C w L = π L C w I C_{w_I}C_{w_L}=\pi_LC_{w_I} C w I C w L = π L C w I by (2.2). The left descents of u ∈ W P u\in W_P u ∈ W P lie in P P P , so W P ∩ I W = W P ∩ I W W_P\cap I_W=W_P\cap I_W W P ∩ I W = W P ∩ I W , and
B S ( P ) = ∑ u ∈ W P ∩ I W v ℓ ( w P ) − ℓ ( w L ) − ℓ ( u ) M u . (31) BS(P)=\sum_{u\in W_P\cap I_W}v^{\ell(w_P)-\ell(w_L)-\ell(u)}M_u. \tag*{(31)} B S ( P ) = u ∈ W P ∩ I W ∑ v ℓ ( w P ) − ℓ ( w L ) − ℓ ( u ) M u . (31)
::: For P = J j P=J_j P = J j and P = P s P=P_s P = P s the pair ( W P , W L ) (W_P,W_L) ( W P , W L ) is ( S N − 1 , S j × S N − j − 1 ) (S_{N-1},S_j \times S_{N-j-1}) ( S N − 1 , S j × S N − j − 1 ) and ( S N , S s × S N − s ) (S_N,S_s \times S_{N-s}) ( S N , S s × S N − s ) , so ℓ ( w P ) − ℓ ( w L ) \ell(w_P)-\ell(w_L) ℓ ( w P ) − ℓ ( w L ) equals j ( N − j − 1 ) j(N-j-1) j ( N − j − 1 ) and s ( N − s ) s(N-s) s ( N − s ) respectively.
It remains to identify W P s ∩ I W W_{P_s}\cap{}^I W W P s ∩ I W . As permutations of Z \mathbb{Z} Z , W P s W_{P_s} W P s consists of the x x x with x ( [ s + 1 , s + N ] ) = [ s + 1 , s + N ] x([s+1,s+N])=[s+1,s+N] x ([ s + 1 , s + N ]) = [ s + 1 , s + N ] , i.e.
x ( i ) ∈ [ s + 1 − N , s ] ( 1 ≤ i ≤ s ) , x ( i ) ∈ [ s + 1 , s + N ] ( s < i ≤ N ) . (32) x(i)\in[s+1-N,s]\quad(1\leq i\leq s),\qquad x(i)\in[s+1,s+N]\quad(s<i\leq N). \tag*{(32)} x ( i ) ∈ [ s + 1 − N , s ] ( 1 ≤ i ≤ s ) , x ( i ) ∈ [ s + 1 , s + N ] ( s < i ≤ N ) . (32)
and W J j W_{J_j} W J j consists of the x ∈ W P j x\in W_{P_j} x ∈ W P j with x ( j + 1 ) = j + 1 x(j+1)=j+1 x ( j + 1 ) = j + 1 . If x ∈ I W x\in{}^I W x ∈ I W satisfies (32) , its window lies in [ 1 − N , 2 N ] [1-N,2N] [ 1 − N , 2 N ] , so x = c ω ′ x=c_{\omega'} x = c ω ′ by Lemma 3.7 , and (32) says that ω ′ \omega' ω ′ has no C in positions ≤ s \leq s ≤ s and no A in positions > s >s > s . Conversely, let ω ′ \omega' ω ′ be such a word, of type t t t . Then t ≤ min ( s , N − s ) t\leq\min(s,N-s) t ≤ min ( s , N − s ) , and the letters B in positions ≤ s \leq s ≤ s are the first s − t s-t s − t of them, so c ω ′ c_{\omega'} c ω ′ has values in [ 1 − t , s ] [1-t,s] [ 1 − t , s ] in positions ≤ s \leq s ≤ s and in [ s + 1 , N + t ] [s+1,N+t] [ s + 1 , N + t ] in positions > s >s > s . Thus c ω ′ c_{\omega'} c ω ′ satisfies (32) , and it lies in I W {}^I W I W by Lemma 3.7 . For J j J_j J j take s = j s=j s = j : then c ω ( j + 1 ) = j + 1 c_\omega(j+1)=j+1 c ω ( j + 1 ) = j + 1 exactly when ω j + 1 = B \omega_{j+1}=\mathrm{B} ω j + 1 = B , because this B is the ( j − t + 1 ) (j-t+1) ( j − t + 1 ) -st one and has value t + ( j − t + 1 ) = j + 1 t+(j-t+1)=j+1 t + ( j − t + 1 ) = j + 1 , whereas a C there has value > N >N > N .
Example 3.20. Let N = 11 N=11 N = 11 and j = 5 j=5 j = 5 . The product B S ( J 5 ) \mathcal{B}S(J_5) B S ( J 5 ) is the sum over words with letters in { A , B } \{\mathrm{A},\mathrm{B}\} { A , B } in positions 1 , … , 5 1,\ldots,5 1 , … , 5 , a B in position 6, letters in { B , C } \{\mathrm{B},\mathrm{C}\} { B , C } in positions 7 , … , 11 7,\ldots,11 7 , … , 11 , and equally many A’s and C’s. There are ∑ t ( 5 t ) 2 = 252 \sum_t \binom{5}{t}^2=252 ∑ t ( t 5 ) 2 = 252 of them. The coefficient of M c M_c M c is v 25 − ℓ ( c ) v^{25-\ell(c)} v 25 − ℓ ( c ) : it equals v 25 v^{25} v 25 at c = e c=e c = e (the word B 11 \mathrm{B}^{11} B 11 ) and 1 at the longest element, A 5 B C 5 \mathrm{A}^5\mathrm{B}\mathrm{C}^5 A 5 B C 5 , of length 25.
We can now compute the shape of the standard expansions of B S j \mathcal{B}S_j B S j and C B S s C\mathcal{B}S_s C B S s . Recall that f ∼ v b f\sim v^b f ∼ v b means f ∈ v b ( 1 + v 2 Z [ v 2 ] ) f\in v^b(1+v^2\mathbb{Z}[v^2]) f ∈ v b ( 1 + v 2 Z [ v 2 ]) .
Proposition 3.21. The expansions of B S j \mathcal{B}S_j B S j (1 ≤ j ≤ r 1\leq j\leq r 1 ≤ j ≤ r ) and C B S s C\mathcal{B}S_s C B S s (2 ≤ s ≤ r − 1 2\leq s\leq r-1 2 ≤ s ≤ r − 1 ) in the standard basis are
B S j = ∑ ( t , a , b ) f e a s i b l e a + b ≤ t ≤ j + a − 1 v d j ( t , a , b ) ρ j ; t , a , b ( v 2 ) ∑ c ∈ O E ( z ( t , a , b ) ) v λ ( t , a , b ) − ℓ ( c ) M c , (33) \mathcal{B}S_j=
\sum_{\substack{(t,a,b)\ \mathrm{feasible}\\a+b\leq t\leq j+a-1}}
v^{d_j(t,a,b)}\rho_{j;t,a,b}(v^2)
\sum_{c\in O_E(z(t,a,b))}
v^{\lambda(t,a,b)-\ell(c)}M_c, \tag*{(33)} B S j = ( t , a , b ) feasible a + b ≤ t ≤ j + a − 1 ∑ v d j ( t , a , b ) ρ j ; t , a , b ( v 2 ) c ∈ O E ( z ( t , a , b )) ∑ v λ ( t , a , b ) − ℓ ( c ) M c , (33) C B S s = ∑ ( t , a , b ) f e a s i b l e a + b ≤ t ≤ s − 2 + a + b v d s ′ ( t , a , b ) ρ s ; t , a , b ′ ( v 2 ) ∑ c ∈ O E ( z ( t , a , b ) ) v λ ( t , a , b ) − ℓ ( c ) M c , C\mathcal{B}S_s=
\sum_{\substack{(t,a,b)\ \mathrm{feasible}\\a+b\leq t\leq s-2+a+b}}
v^{d'_s(t,a,b)}\rho'_{s;t,a,b}(v^2)
\sum_{c\in O_E(z(t,a,b))}
v^{\lambda(t,a,b)-\ell(c)}M_c, C B S s = ( t , a , b ) feasible a + b ≤ t ≤ s − 2 + a + b ∑ v d s ′ ( t , a , b ) ρ s ; t , a , b ′ ( v 2 ) c ∈ O E ( z ( t , a , b )) ∑ v λ ( t , a , b ) − ℓ ( c ) M c , where ρ j ; t , a , b ( q ) , ρ s ; t , a , b ′ ( q ) ∈ Z [ q ] \rho_{j;t,a,b}(q),\rho'_{s;t,a,b}(q)\in\mathbb{Z}[q] ρ j ; t , a , b ( q ) , ρ s ; t , a , b ′ ( q ) ∈ Z [ q ] have constant term 1, and the exponents are
In other words, on each orbit O E ( z ( t , a , b ) ) O_E(z(t,a,b)) O E ( z ( t , a , b )) the coefficient at c c c is v λ ( t , a , b ) − ℓ ( c ) v^{\lambda(t,a,b)-\ell(c)} v λ ( t , a , b ) − ℓ ( c ) times the coefficient at the maximum z ( t , a , b ) z(t,a,b) z ( t , a , b ) , and the latter is ∼ v d j ( t , a , b ) \sim v^{d_j(t,a,b)} ∼ v d j ( t , a , b ) , respectively ∼ v d s ′ ( t , a , b ) \sim v^{d'_s(t,a,b)} ∼ v d s ′ ( t , a , b ) .
Proof. Throughout, an orbit O E ( z ( t , a , b ) ) O_E(z(t,a,b)) O E ( z ( t , a , b )) is determined by t t t and the first two letters, and E E E does not move them.
The case of B S j \mathcal{B}S_j B S j . By Lemma 3.19 , B S ( J j ) \mathcal{B}S(J_j) B S ( J j ) has coefficient v j ( N − j − 1 ) − ℓ ( c ) v^{j(N-j-1)-\ell(c)} v j ( N − j − 1 ) − ℓ ( c ) at each M c M_c M c in its support. The position blocks of K j K_j K j are { 1 } \{1\} { 1 } , { 2 , … , j + 1 } \{2,\ldots,j+1\} { 2 , … , j + 1 } and { j + 2 , … , N } \{j+2,\ldots,N\} { j + 2 , … , N } . For a word of type t t t with first letter of indicator a a a , these blocks carry respectively the first letter, t − a t-a t − a letters A and j − t + a j-t+a j − t + a letters B, and t t t letters C and N − j − 1 − t N-j-1-t N − j − 1 − t letters B. The orbit O K j ( c ) O_{K_j}(c) O K j ( c ) meets the support of B S ( J j ) \mathcal{B}S(J_j) B S ( J j ) exactly when the second block contains a B (to fill the fixed position j + 1 j+1 j + 1 ), i.e. a ≤ t ≤ j + a − 1 a\leq t\leq j+a-1 a ≤ t ≤ j + a − 1 . In such an orbit the unique longest element of the support has each block sorted in the order C < < < B < < < A subject to that fixed B; the t − a t-a t − a letters A before it give a length gap g = t − a g=t-a g = t − a from the maximum. Since ℓ ( w K j ) − ℓ ( w J j ∩ K j ) = j − 1 \ell(w_{K_j})-\ell(w_{J_j\cap K_j})=j-1 ℓ ( w K j ) − ℓ ( w J j ∩ K j ) = j − 1 , Corollary 3.18 (with D = J j ∩ K j D=J_j\cap K_j D = J j ∩ K j ) gives, for every c c c with these block counts,
[ M c ] B S ( J j ) ∼ v A K − ℓ ( c ) , A K = j ( N − j − 1 ) − ( j − 1 ) + 2 ( t − a ) . [M_c]\mathcal{B}S(J_j)\sim v^{A_K-\ell(c)},\qquad A_K=j(N-j-1)-(j-1)+2(t-a). [ M c ] B S ( J j ) ∼ v A K − ℓ ( c ) , A K = j ( N − j − 1 ) − ( j − 1 ) + 2 ( t − a ) .
In particular every such c c c occurs.
Now apply O E O_E O E and divide by π K j ∩ E \pi_{K_j\cap E} π K j ∩ E . Fix O E ( z ( t , a , b ) ) O_E(z(t,a,b)) O E ( z ( t , a , b )) and write k = t − a − b k=t-a-b k = t − a − b and u = j − t + a + b − 1 u=j-t+a+b-1 u = j − t + a + b − 1 . The indices occurring in B S ( J j , K j ) \mathcal{B}S(J_j,K_j) B S ( J j , K j ) and lying in this orbit have the following letter counts: They exist exactly when all counts are nonnegative and t ≤ j + a − 1 t\leq j+a-1 t ≤ j + a − 1 , that is, when a + b ≤ t ≤ j + a − 1 a+b\leq t\leq j+a-1 a + b ≤ t ≤ j + a − 1 (and the triple is feasible). Sorting each of the two blocks in the order C < B < A gives the unique longest of them, and its length gap from z ( t , a , b ) z(t,a,b) z ( t , a , b ) is g = k ( N − j − 1 ) + u t g=k(N-j-1)+ut g = k ( N − j − 1 ) + u t : each of the k k k letters A precedes all N − j − 1 N-j-1 N − j − 1 letters of the second block, and each of the u u u letters B precedes the t t t letters C. Moreover ℓ ( w E ) − ℓ ( w K j ∩ E ) = ( j − 1 ) ( N − j − 1 ) \ell(w_E)-\ell(w_{K_j\cap E})=(j-1)(N-j-1) ℓ ( w E ) − ℓ ( w K j ∩ E ) = ( j − 1 ) ( N − j − 1 ) , and B S j = B S ( J j , K j ) C w E / π K j ∩ E ∈ M BS_j=BS(J_j,K_j)C_{w_E}/\pi_{K_j\cap E}\in\mathcal{M} B S j = B S ( J j , K j ) C w E / π K j ∩ E ∈ M by Lemma 3.3 . Corollary 3.18 therefore gives
Table 3.
[ M z ( t , a , b ) ] B S j ∼ v A K − λ ( t , a , b ) − ( j − 1 ) ( N − j − 1 ) + 2 ( k ( N − j − 1 ) + u t ) . [M_{z(t,a,b)}]BS_j\sim v^{A_K-\lambda(t,a,b)-(j-1)(N-j-1)+2(k(N-j-1)+ut)}. [ M z ( t , a , b ) ] B S j ∼ v A K − λ ( t , a , b ) − ( j − 1 ) ( N − j − 1 ) + 2 ( k ( N − j − 1 ) + u t ) .
By Lemma 3.13 this exponent simplifies to d j ( t , a , b ) = ( 1 − a − b ) ( N − 2 j ) + b ( 1 + a ) d_j(t,a,b)=(1-a-b)(N-2j)+b(1+a) d j ( t , a , b ) = ( 1 − a − b ) ( N − 2 j ) + b ( 1 + a ) , which gives the first column of the table. By (29) , [ M c ] B S j = v λ ( t , a , b ) − ℓ ( c ) [ M z ( t , a , b ) ] B S j [M_c]BS_j=v^{\lambda(t,a,b)-\ell(c)}[M_{z(t,a,b)}]BS_j [ M c ] B S j = v λ ( t , a , b ) − ℓ ( c ) [ M z ( t , a , b ) ] B S j on each orbit, and the definition of ∼ \sim ∼ provides the polynomial ρ j ; t , a , b \rho_{j;t,a,b} ρ j ; t , a , b with ρ j ; t , a , b ( 0 ) = 1 \rho_{j;t,a,b}(0)=1 ρ j ; t , a , b ( 0 ) = 1 . All indices of B S j BS_j B S j lie in the orbits listed, since the support of B S ( J j ) BS(J_j) B S ( J j ) is moved only by C w K j C_{w_{K_j}} C w K j and then by C w E C_{w_E} C w E , each of which stays within orbits. Summing over the orbits proves (33) .
The case of CBSs. The product C B S s = B S ( P s ) C w E / π P s ∩ E CBS_s=BS(P_s)C_{w_E}/\pi_{P_s\cap E} C B S s = B S ( P s ) C w E / π P s ∩ E has only one averaging step. By Lemma 3.19 , B S ( P s ) BS(P_s) B S ( P s ) has coefficient v s ( N − s ) − ℓ ( c ) v^{s(N-s)-\ell(c)} v s ( N − s ) − ℓ ( c ) , and its support consists of the words with letters in { A , B } \{A,B\} { A , B } in positions ≤ s \leq s ≤ s and in { B , C } \{B,C\} { B , C } in positions > s >s > s . Since s ≥ 2 s\geq2 s ≥ 2 , the first two letters are A or B. In O E ( z ( t , a , b ) ) O_E(z(t,a,b)) O E ( z ( t , a , b )) the indices of the support have the following letter counts, where k = t − a − b k=t-a-b k = t − a − b :
Table 4.
All counts are nonnegative exactly when a + b ≤ t ≤ s − 2 + a + b a+b\leq t\leq s-2+a+b a + b ≤ t ≤ s − 2 + a + b (note N − s − t ≥ 0 N-s-t\geq0 N − s − t ≥ 0 since t ≤ s t\leq s t ≤ s and s ≤ r − 1 s\leq r-1 s ≤ r − 1 ). The longest of them has each block sorted, with length gap g = k ( N − s ) + ( s − 2 − k ) t g=k(N-s)+(s-2-k)t g = k ( N − s ) + ( s − 2 − k ) t from z ( t , a , b ) z(t,a,b) z ( t , a , b ) , and ℓ ( w E ) − ℓ ( w P s ∩ E ) = ( s − 2 ) ( N − s ) \ell(w_E)-\ell(w_{P_s\cap E})=(s-2)(N-s) ℓ ( w E ) − ℓ ( w P s ∩ E ) = ( s − 2 ) ( N − s ) . Corollary 3.18 gives [ M z ( t , a , b ) ] C B S s ∼ v s ( N − s ) − λ ( t , a , b ) − ( s − 2 ) ( N − s ) + 2 g = v 2 ( N − s ) − λ ( t , a , b ) + 2 g [M_{z(t,a,b)}]CBS_s\sim v^{s(N-s)-\lambda(t,a,b)-(s-2)(N-s)+2g}=v^{2(N-s)-\lambda(t,a,b)+2g} [ M z ( t , a , b ) ] C B S s ∼ v s ( N − s ) − λ ( t , a , b ) − ( s − 2 ) ( N − s ) + 2 g = v 2 ( N − s ) − λ ( t , a , b ) + 2 g , which simplifies, again by Lemma 3.13 , to the exponent d s ′ ( t , a , b ) d'_s(t,a,b) d s ′ ( t , a , b ) of the table. The rest is as for B S j BS_j B S j . □ \square □
Example 3.22. Let N = 11 N=11 N = 11 , r = 5 r=5 r = 5 , j = 5 j=5 j = 5 . In the standard basis B S 5 = B S ( J 5 , K 5 , E ) BS_5=BS(J_5,K_5,E) B S 5 = B S ( J 5 , K 5 , E ) has 11 131 terms. By Proposition 3.21 they lie in 18 orbits O E ( z ( t , a , b ) ) O_E(z(t,a,b)) O E ( z ( t , a , b )) , namely those with
0 ≤ t ≤ 4 ( B B ) , 1 ≤ t ≤ 4 ( B A ) , 1 ≤ t ≤ 5 ( A B ) , 2 ≤ t ≤ 5 ( A A ) , 0\leq t\leq4\ (\mathrm{BB}),\qquad1\leq t\leq4\ (\mathrm{BA}),\qquad1\leq t\leq5\ (\mathrm{AB}),\qquad2\leq t\leq5\ (\mathrm{AA}), 0 ≤ t ≤ 4 ( BB ) , 1 ≤ t ≤ 4 ( BA ) , 1 ≤ t ≤ 5 ( AB ) , 2 ≤ t ≤ 5 ( AA ) ,
and on each of them the coefficient of M c M_c M c is v d 5 + λ − ℓ ( c ) ρ ( v 2 ) v^{d_5+\lambda-\ell(c)}\rho(v^2) v d 5 + λ − ℓ ( c ) ρ ( v 2 ) , with d 5 = 1 , 1 , 0 , 1 d_5=1,1,0,1 d 5 = 1 , 1 , 0 , 1 for the four prefixes. So 18 polynomials describe all 11 131 coefficients.
The orbit of X 0 = z ( 5 , 1 , 0 ) X_0=z(5,1,0) X 0 = z ( 5 , 1 , 0 ) (Example 3.15 ) has t = 5 t=5 t = 5 . Since B S 4 BS_4 B S 4 and the C B S s CBS_s C B S s only occur for t ≤ 4 t\leq4 t ≤ 4 , none of them meets this orbit, so there C X 0 C_{X_0} C X 0 agrees with B S 5 BS_5 B S 5 . As [ M X 0 ] C X 0 = 1 [M_{X_0}]C_{X_0}=1 [ M X 0 ] C X 0 = 1 and d 5 ( 1 , 0 ) = 0 d_5(1,0)=0 d 5 ( 1 , 0 ) = 0 , we get ρ 5 ; 5 , 1 , 0 = 1 \rho_{5;5,1,0}=1 ρ 5 ; 5 , 1 , 0 = 1 , and Corollary 3.18 gives
[ M c ] C X 0 = v 49 − ℓ ( c ) ( c ∈ O E ( X 0 ) ) , [M_c]C_{X_0}=v^{49-\ell(c)}\qquad(c\in O_E(X_0)), [ M c ] C X 0 = v 49 − ℓ ( c ) ( c ∈ O E ( X 0 )) ,
since ℓ ( X 0 ) = 2 r 2 − 1 = 49 \ell(X_0)=2r^2-1=49 ℓ ( X 0 ) = 2 r 2 − 1 = 49 . The orbit has 126 elements with ℓ ( c ) \ell(c) ℓ ( c ) from 29 to 49: the coefficient is 1 at X 0 X_0 X 0 , then v v v at the elements of length 48, and so on down to v 20 v^{20} v 20 at A B A 4 C 5 \mathrm{AB}\,\mathrm{A}^4\mathrm{C}^5 AB A 4 C 5 .
Proof of Proposition 3.5. Put F = B S r − B S r − 1 + ( v + v − 1 ) ∑ s = 2 r − 1 C B S s F=BS_r-BS_{r-1}+(v+v^{-1})\sum_{s=2}^{r-1}CBS_s F = B S r − B S r − 1 + ( v + v − 1 ) ∑ s = 2 r − 1 C B S s . By Lemma 3.3 , F ∈ M F\in\mathcal{M} F ∈ M is bar-invariant, so by Lemma 3.1 it suffices to show F − M X 0 ∈ ⨁ c ∈ E v Z [ v ] M c F-M_{X_0}\in\bigoplus_{c\in E}v\mathbb{Z}[v]M_c F − M X 0 ∈ ⨁ c ∈ E v Z [ v ] M c .
By Proposition 3.21, F F F is supported on the orbits O E ( z ) O_E(z) O E ( z ) , z = z ( t , a , b ) z=z(t,a,b) z = z ( t , a , b ) , and on each of them [ M c ] F = v λ ( t , a , b ) − ℓ ( c ) [ M z ] F [M_c]F=v^{\lambda(t,a,b)-\ell(c)}[M_z]F [ M c ] F = v λ ( t , a , b ) − ℓ ( c ) [ M z ] F . Since z z z is the unique longest element of its orbit (Lemma 3.11 ), the exponent is positive for c ≠ z c\ne z c = z . It therefore suffices to show
[ M z ] F ∈ { 1 + v Z [ v ] , z = X 0 , v Z [ v ] , z ≠ X 0 , (34) [M_z]F\in
\begin{cases}
1+v\mathbb{Z}[v], & z=X_0,\\
v\mathbb{Z}[v], & z\ne X_0,
\end{cases} \tag*{(34)} [ M z ] F ∈ { 1 + v Z [ v ] , v Z [ v ] , z = X 0 , z = X 0 , (34)
for the maxima z = z ( t , a , b ) z=z(t,a,b) z = z ( t , a , b ) , where z ( r , 1 , 0 ) = X 0 z(r,1,0)=X_0 z ( r , 1 , 0 ) = X 0 by (27) .
By Proposition 3.21, the coefficients of B S j \mathcal{B}S_j B S j and C B S s \mathcal{C}BS_s C B S s at z ( t , a , b ) z(t,a,b) z ( t , a , b ) are v d j ρ j ( v 2 ) v^{dj}\rho_j(v^2) v d j ρ j ( v 2 ) and v d ′ s ρ s ′ ( v 2 ) v^{d's}\rho'_s(v^2) v d ′ s ρ s ′ ( v 2 ) with ρ ( 0 ) = 1 \rho(0)=1 ρ ( 0 ) = 1 , so only the lowest exponents matter. For B S r \mathcal{B}S_r B S r , B S r − 1 \mathcal{B}S_{r-1} B S r − 1 and ( v + v − 1 ) C B S s (v+v^{-1})\mathcal{C}BS_s ( v + v − 1 ) C B S s (whose exponents are those of C B S s \mathcal{C}BS_s C B S s lowered by one) they are, in the columns ( a , b ) = ( 0 , 0 ) , ( 0 , 1 ) , ( 1 , 0 ) , ( 1 , 1 ) (a,b)=(0,0),(0,1),(1,0),(1,1) ( a , b ) = ( 0 , 0 ) , ( 0 , 1 ) , ( 1 , 0 ) , ( 1 , 1 ) , that is BB, BA, AB, AA:
Table 1 (PDF p. 21).
The supports are a + b ≤ t ≤ r + a − 1 a+b\le t\le r+a-1 a + b ≤ t ≤ r + a − 1 for B S r \mathcal{B}S_r B S r , t ≤ r + a − 2 t\le r+a-2 t ≤ r + a − 2 for B S r − 1 \mathcal{B}S_{r-1} B S r − 1 and t ≤ s − 2 + a + b t\le s-2+a+b t ≤ s − 2 + a + b for C B S s \mathcal{C}BS_s C B S s . In these ranges all entries are positive except the zeros in the column AB and the − 1 -1 − 1 in the column AA, which occurs for B S r − 1 \mathcal{B}S_{r-1} B S r − 1 when 2 ≤ t ≤ r − 1 2\le t\le r-1 2 ≤ t ≤ r − 1 and for ( v + v − 1 ) C B S s (v+v^{-1})\mathcal{C}BS_s ( v + v − 1 ) C B S s when t = s t=s t = s . Hence:
for BB and BA the coefficient lies in v Z [ v ] v\mathbb{Z}[v] v Z [ v ] ;
for AB the coefficients of B S r \mathcal{B}S_r B S r and B S r − 1 \mathcal{B}S_{r-1} B S r − 1 both lie in 1 + v Z [ v ] 1+v\mathbb{Z}[v] 1 + v Z [ v ] and cancel for 1 ≤ t ≤ r − 1 1\le t\le r-1 1 ≤ t ≤ r − 1 ; at t = r t=r t = r only B S r \mathcal{B}S_r B S r contributes, and this is z = X 0 z=X_0 z = X 0 ;
for AA, the term − B S r − 1 -\mathcal{B}S_{r-1} − B S r − 1 contributes − v − 1 + v Z [ v ] -v^{-1}+v\mathbb{Z}[v] − v − 1 + v Z [ v ] at z ( t , 1 , 1 ) z(t,1,1) z ( t , 1 , 1 ) for 2 ≤ t ≤ r − 1 2\le t\le r-1 2 ≤ t ≤ r − 1 , and the term ( v + v − 1 ) C B S t (v+v^{-1})\mathcal{C}BS_t ( v + v − 1 ) C B S t contributes v − 1 + v Z [ v ] v^{-1}+v\mathbb{Z}[v] v − 1 + v Z [ v ] at the same maximum. These cancel to leave v Z [ v ] v\mathbb{Z}[v] v Z [ v ] , as do all other contributions.
This proves (34) , and Lemma 3.1 gives F = C X 0 F=\mathcal{C}X_0 F = C X 0 . For r = 2 r=2 r = 2 all ranges involving 2 ≤ s ≤ r − 1 2\le s\le r-1 2 ≤ s ≤ r − 1 are empty.
Example 3.23. For the smallest case in which the correction term is present, r = 3 r=3 r = 3 and N = 7 N=7 N = 7 , we have C X 0 = B S 3 − B S 2 + ( v + v − 1 ) C B S 2 \mathcal{C}X_0=\mathcal{B}S_3-\mathcal{B}S_2+(v+v^{-1})\mathcal{C}BS_2 C X 0 = B S 3 − B S 2 + ( v + v − 1 ) C B S 2 . The coefficients at four of the maxima are
Table 2.
At X 0 X_0 X 0 only B S 3 \mathcal{B}S_3 B S 3 contributes. At z ( 2 , 1 , 0 ) z(2,1,0) z ( 2 , 1 , 0 ) , and likewise at z ( 1 , 1 , 0 ) z(1,1,0) z ( 1 , 1 , 0 ) , the constants of B S 3 \mathcal{B}S_3 B S 3 and B S 2 \mathcal{B}S_2 B S 2 cancel. At z ( 2 , 1 , 1 ) z(2,1,1) z ( 2 , 1 , 1 ) the term − B S 2 -\mathcal{B}S_2 − B S 2 would leave a coefficient − v − 1 -v^{-1} − v − 1 , and it is removed by ( v + v − 1 ) C B S 2 (v+v^{-1})\mathcal{C}BS_2 ( v + v − 1 ) C B S 2 . The last row is the coefficient of M e M_e M e , that is h X 0 h_{X_0} h X 0 ; it begins v + v 3 + 2 v 5 v+v^3+2v^5 v + v 3 + 2 v 5 , with coefficients p ( 0 ) , p ( 1 ) , p ( 2 ) = 1 , 1 , 2 p(0),p(1),p(2)=1,1,2 p ( 0 ) , p ( 1 ) , p ( 2 ) = 1 , 1 , 2 as in Proposition 2.6.
Coefficient estimates
We now read off coefficients of C X 0 \mathrm{C}X_0 C X 0 and of C γ j \mathrm{C}_{\gamma_j} C γ j from the products above. The identity coefficient is h X 0 = ε ( C X 0 ) h_{X_0}=\varepsilon(\mathrm{C}X_0) h X 0 = ε ( C X 0 ) , and in each product B S j \mathcal{B}S_j B S j and C B S s \mathcal{C}BS_s C B S s every factor after the initial one lies in I I I , so by (23) it acts on ε \varepsilon ε by a scalar. The computation reduces to Gaussian binomial coefficients, and this proves Proposition 2.6. We then record one further coefficient of C X 0 \mathrm{C}X_0 C X 0 , needed in Theorem 5.1, and the identity coefficient of C γ j \mathrm{C}_{\gamma_j} C γ j .
Proof of Proposition 2.6. By Lemma 3.19, ε ( B S ( J j ) ) = v j ( N − j − 1 ) \varepsilon(\mathcal{B}S(J_j))=v^{j(N-j-1)} ε ( B S ( J j )) = v j ( N − j − 1 ) and ε ( B S ( P s ) ) = v s ( N − s ) \varepsilon(\mathcal{B}S(P_s))=v^{s(N-s)} ε ( B S ( P s )) = v s ( N − s ) (the coefficient at M e M_e M e , the word B N \mathrm{B}^N B N ). The blocks of K j , J j ∩ K j , E , K j ∩ E K_j,J_j\cap K_j,E,K_j\cap E K j , J j ∩ K j , E , K j ∩ E have sizes ( 1 , j , N − j − 1 ) , ( 1 , j − 1 , 1 , N − j − 1 ) , ( 1 , 1 , N − 2 ) , ( 1 , 1 , j − 1 , N − j − 1 ) (1,j,N-j-1),(1,j-1,1,N-j-1),(1,1,N-2),(1,1,j-1,N-j-1) ( 1 , j , N − j − 1 ) , ( 1 , j − 1 , 1 , N − j − 1 ) , ( 1 , 1 , N − 2 ) , ( 1 , 1 , j − 1 , N − j − 1 ) , and those of E , P s ∩ E E,P_s\cap E E , P s ∩ E have sizes ( 1 , 1 , N − 2 ) , ( 1 , 1 , s − 2 , N − s ) (1,1,N-2),(1,1,s-2,N-s) ( 1 , 1 , N − 2 ) , ( 1 , 1 , s − 2 , N − s ) . So (3.4) and (3.3) give
ε ( B S j ) = v j ( N − j − 1 ) ∏ j ∏ j − 1 ∏ N − 2 ∏ j − 1 ∏ N − j − 1 = v N − 2 j ( N − 2 j ) q , \varepsilon(\mathcal{B}S_j)=v^{j(N-j-1)}\frac{\prod_j}{\prod_{j-1}}\frac{\prod_{N-2}}{\prod_{j-1}\prod_{N-j-1}}=v^{N-2j}\binom{N-2}{j}_q, ε ( B S j ) = v j ( N − j − 1 ) ∏ j − 1 ∏ j ∏ j − 1 ∏ N − j − 1 ∏ N − 2 = v N − 2 j ( j N − 2 ) q ,
ε ( C B S s ) = v s ( N − s ) ∏ N − 2 ∏ s − 2 ∏ N − s = v 2 ( N − s ) ( N − 2 s − 2 ) q . \varepsilon(\mathcal{C}BS_s)=v^{s(N-s)}\frac{\prod_{N-2}}{\prod_{s-2}\prod_{N-s}}=v^{2(N-s)}\binom{N-2}{s-2}_q. ε ( C B S s ) = v s ( N − s ) ∏ s − 2 ∏ N − s ∏ N − 2 = v 2 ( N − s ) ( s − 2 N − 2 ) q .
In particular ( v + v − 1 ) ε ( C B S s ) = O ( v 2 r + 3 ) (v+v^{-1})\varepsilon(\mathcal{C}BS_s)=O(v^{2r+3}) ( v + v − 1 ) ε ( C B S s ) = O ( v 2 r + 3 ) for 2 ≤ s ≤ r − 1 2\le s\le r-1 2 ≤ s ≤ r − 1 . Applying ε \varepsilon ε to Proposition 3.5 and using the Gaussian identity
[ r ] q [ 2 r − 1 r − 1 ] q − q [ r − 1 ] q [ 2 r − 1 r − 2 ] q = [ 2 r − 1 r − 1 ] q + ( 1 + q ) q r [ 2 r − 1 r − 2 ] q [r]_q\begin{bmatrix}2r-1\\r-1\end{bmatrix}_q-q[r-1]_q\begin{bmatrix}2r-1\\r-2\end{bmatrix}_q=\begin{bmatrix}2r-1\\r-1\end{bmatrix}_q+(1+q)q^r\begin{bmatrix}2r-1\\r-2\end{bmatrix}_q [ r ] q [ 2 r − 1 r − 1 ] q − q [ r − 1 ] q [ 2 r − 1 r − 2 ] q = [ 2 r − 1 r − 1 ] q + ( 1 + q ) q r [ 2 r − 1 r − 2 ] q
we obtain
h X 0 = v ( [ r ] q [ 2 r − 1 r − 1 ] q − q [ r − 1 ] q [ 2 r − 1 r − 2 ] q ) + O ( v 2 r + 3 ) = v [ 2 r − 1 r − 1 ] q + v ( 1 + q ) q r [ 2 r − 1 r − 2 ] q + O ( v 2 r + 3 ) = v [ 2 r − 1 r − 1 ] q + O ( v 2 r + 1 ) . \begin{aligned}
h_{X_0}=v\left([r]_q\begin{bmatrix}2r-1\\r-1\end{bmatrix}_q-q[r-1]_q\begin{bmatrix}2r-1\\r-2\end{bmatrix}_q\right)+O(v^{2r+3}) \\
&=v\begin{bmatrix}2r-1\\r-1\end{bmatrix}_q+v(1+q)q^r\begin{bmatrix}2r-1\\r-2\end{bmatrix}_q+O(v^{2r+3}) \\
&=v\begin{bmatrix}2r-1\\r-1\end{bmatrix}_q+O(v^{2r+1}).
\end{aligned} h X 0 = v ( [ r ] q [ 2 r − 1 r − 1 ] q − q [ r − 1 ] q [ 2 r − 1 r − 2 ] q ) + O ( v 2 r + 3 ) = v [ 2 r − 1 r − 1 ] q + v ( 1 + q ) q r [ 2 r − 1 r − 2 ] q + O ( v 2 r + 3 ) = v [ 2 r − 1 r − 1 ] q + O ( v 2 r + 1 ) .
Finally [ 2 r − 1 r − 1 ] q \begin{bmatrix}2r-1\\r-1\end{bmatrix}_q [ 2 r − 1 r − 1 ] q counts partitions in a ( r − 1 ) × r (r-1)\times r ( r − 1 ) × r box by size [38 Proposition 1.7.3] , so
[ 2 r − 1 r − 1 ] q ≡ ∑ d = 0 r − 1 p ( d ) q d ( m o d q r ) . \begin{bmatrix}2r-1\\r-1\end{bmatrix}_q\equiv\sum_{d=0}^{r-1}\mathrm{p}(d)q^d\pmod{q^r}. [ 2 r − 1 r − 1 ] q ≡ d = 0 ∑ r − 1 p ( d ) q d ( mod q r ) .
Substituting q = v 2 q=v^2 q = v 2 proves (2.5). □
Lemma 3.24. Let r ≥ 2 r\ge2 r ≥ 2 , N = 2 r + 1 N=2r+1 N = 2 r + 1 , and put z = z ( 1 , 1 , 0 ) = c A B C B N − 3 = [ 0 , 2 , N + 1 , 3 , … , N − 1 ] z=z(1,1,0)=c_{\mathrm{ABCB}}N-3=[0,2,N+1,3,\ldots,N-1] z = z ( 1 , 1 , 0 ) = c ABCB N − 3 = [ 0 , 2 , N + 1 , 3 , … , N − 1 ] . For 1 ≤ j ≤ r 1\le j\le r 1 ≤ j ≤ r ,
[ M z ] B S j = [ j ] q [ N − 3 j − 1 ] q (35) [M_z]\mathcal{B}S_j=[j]_q\begin{bmatrix}N-3\\j-1\end{bmatrix}_q \tag*{(35)} [ M z ] B S j = [ j ] q [ N − 3 j − 1 ] q (35) Consequently
[ M z ] C X 0 = O ( v N − 3 ) . (36) [M_z]\mathcal{C}X_0=O(v^{N-3}). \tag*{(36)} [ M z ] C X 0 = O ( v N − 3 ) . (36) Proof. Let y = c A B N − 2 C ( = s 0 ) y=c_{\mathrm{AB}}N-2C(=s_0) y = c AB N − 2 C ( = s 0 ) , the minimum of O E ( z ) \mathcal{O}_E(z) O E ( z ) , which consists of the type-one words with prefix AB. The maximum is z z z , of length N − 2 N-2 N − 2 , since moving C one position to the left across a B raises the length by one. Here L = { s 3 , … , s N − 2 } L=\{s_3,\ldots,s_{N-2}\} L = { s 3 , … , s N − 2 } and Lemma 3.17 gives [ M z ] ( M y C w E ) = π L [M_z](M_yCw_E)=\pi_L [ M z ] ( M y C w E ) = π L .
Fix j j j and put H = K j ∩ E H=K_j\cap E H = K j ∩ E . In B S ( J j ) \mathcal{B}S(J_j) B S ( J j ) only the type-one words with first letter A can contribute to M z M_z M z , because K j K_j K j and E E E preserve the type and the first letter. By Lemma 3.19 these words are A B p − 2 C B N − p \mathrm{AB}^{p-2}\mathrm{CB}^{N-p} AB p − 2 CB N − p with j + 2 ≤ p ≤ N j+2\le p\le N j + 2 ≤ p ≤ N ; they form the orbit O H ( y ) \mathcal{O}_H(y) O H ( y ) and the longest of them has length N − j − 1 N-j-1 N − j − 1 . Their part of B S ( J j ) \mathcal{B}S(J_j) B S ( J j ) is therefore, by Lemma 3.17 for Q = H Q=H Q = H ,
F j = ∑ c ∈ O H ( y ) v j ( N − j − 1 ) − ℓ ( c ) M c = v ( j − 1 ) ( N − j − 1 ) π H ∩ L M y C w H . F_j=\sum_{c\in\mathcal{O}_H(y)}v^{j(N-j-1)-\ell(c)}M_c=\frac{v^{(j-1)(N-j-1)}}{\pi_{H\cap L}}M_yCw_H. F j = c ∈ O H ( y ) ∑ v j ( N − j − 1 ) − ℓ ( c ) M c = π H ∩ L v ( j − 1 ) ( N − j − 1 ) M y C w H . We claim F j H s = v − 1 F j F_jH_s=v^{-1}F_j F j H s = v − 1 F j for every s ∈ K j s\in K_j s ∈ K j . For s ∈ H s\in H s ∈ H this follows from (3.1), since F j F_j F j is a multiple of M y C w H M_yCw_H M y C w H . The only generator of K j K_j K j outside H H H is s 2 s_2 s 2 (present when j ≥ 2 j\ge2 j ≥ 2 ), and in every word of the support positions 2 and 3 both carry B, so M c H s 2 = v − 1 M c M_cH_{s_2}=v^{-1}M_c M c H s 2 = v − 1 M c by Lemma 3.8. Hence F j C w K j = π K j F j F_jCw_{K_j}=\pi_{K_j}F_j F j C w K j = π K j F j . The normalization of B S j \mathcal{B}S_j B S j is π J j ∩ K j π H \pi_{J_j\cap K_j}\pi_H π J j ∩ K j π H , and C w H C w E = π H C w E Cw_HCw_E=\pi_HCw_E C w H C w E = π H C w E by (3.1), so
[ M z ] B S j = v ( j − 1 ) ( N − j − 1 ) π K j π J j ∩ K j π L π H ∩ L = v ( j − 1 ) ( N − j − 1 ) Π j Π j − 1 Π N − 3 Π j − 1 Π N − j − 2 . [M_z]\mathcal{B}S_j=v^{(j-1)(N-j-1)}\frac{\pi_{K_j}}{\pi_{J_j\cap K_j}}\frac{\pi_L}{\pi_{H\cap L}}=v^{(j-1)(N-j-1)}\frac{\Pi_j}{\Pi_{j-1}}\frac{\Pi_{N-3}}{\Pi_{j-1}\Pi_{N-j-2}}. [ M z ] B S j = v ( j − 1 ) ( N − j − 1 ) π J j ∩ K j π K j π H ∩ L π L = v ( j − 1 ) ( N − j − 1 ) Π j − 1 Π j Π j − 1 Π N − j − 2 Π N − 3 . Here we used (3.3): the blocks of K j K_j K j and J j ∩ K j J_j\cap K_j J j ∩ K j have sizes ( 1 , j , N − j − 1 ) (1,j,N-j-1) ( 1 , j , N − j − 1 ) and ( 1 , j − 1 , 1 , N − j − 1 ) (1,j-1,1,N-j-1) ( 1 , j − 1 , 1 , N − j − 1 ) , while W L ≅ S N − 3 W_L\cong S_{N-3} W L ≅ S N − 3 and W H ∩ L ≅ S j − 1 × S N − j − 2 W_{H\cap L}\cong S_{j-1}\times S_{N-j-2} W H ∩ L ≅ S j − 1 × S N − j − 2 . The two quotients equal v − ( j − 1 ) [ j ] q v^{-(j-1)}[j]_q v − ( j − 1 ) [ j ] q and v − ( j − 1 ) ( N − j − 2 ) [ N − 3 j − 1 ] q v^{-(j-1)(N-j-2)}\begin{bmatrix}N-3\\j-1\end{bmatrix}_q v − ( j − 1 ) ( N − j − 2 ) [ N − 3 j − 1 ] q , and their powers of v v v cancel against v ( j − 1 ) ( N − j − 1 ) v^{(j-1)(N-j-1)} v ( j − 1 ) ( N − j − 1 ) . This proves (3.16).
Put G 0 = [ 2 r − 2 r − 1 ] q G_0=\begin{bmatrix}2r-2\\r-1\end{bmatrix}_q G 0 = [ 2 r − 2 r − 1 ] q and G 1 = [ 2 r − 2 r − 2 ] q G_1=\begin{bmatrix}2r-2\\r-2\end{bmatrix}_q G 1 = [ 2 r − 2 r − 2 ] q . Then [ r − 1 ] q G 0 = [ r ] q G 1 [r-1]_qG_0=[r]_qG_1 [ r − 1 ] q G 0 = [ r ] q G 1 , so G 1 = [ r − 1 ] q G 0 / [ r ] q G_1=[r-1]_qG_0/[r]_q G 1 = [ r − 1 ] q G 0 / [ r ] q and [ r ] q − [ r − 1 ] q = q r − 1 [r]_q-[r-1]_q=q^{r-1} [ r ] q − [ r − 1 ] q = q r − 1 give
[ M z ] ( B S r − B S r − 1 ) = [ r ] q G 0 − [ r − 1 ] q G 1 = q r − 1 ( G 0 + G 1 ) . [M_z](\mathcal{B}S_r-\mathcal{B}S_{r-1})=[r]_qG_0-[r-1]_qG_1=q^{r-1}(G_0+G_1). [ M z ] ( B S r − B S r − 1 ) = [ r ] q G 0 − [ r − 1 ] q G 1 = q r − 1 ( G 0 + G 1 ) .
This is divisible by q r − 1 = v N − 3 q^{r-1}=v^{N-3} q r − 1 = v N − 3 . For 2 ≤ s ≤ r − 1 2\le s\le r-1 2 ≤ s ≤ r − 1 , the case ( t , a , b ) = ( 1 , 1 , 0 ) (t,a,b)=(1,1,0) ( t , a , b ) = ( 1 , 1 , 0 ) of Proposition 3.21 gives [ M z ] C B S s = O ( v N − 2 ) [M_z]_{CBS_s}=O(v^{N-2}) [ M z ] C B S s = O ( v N − 2 ) , so ( v + v − 1 ) [ M z ] C B S s = O ( v N − 3 ) (v+v^{-1})[M_z]_{CBS_s}=O(v^{N-3}) ( v + v − 1 ) [ M z ] C B S s = O ( v N − 3 ) as well. Proposition 3.5 now proves (3.17). For r = 2 r=2 r = 2 the sum over s s s is empty and [ M z ] C X 0 = q ( 2 + q ) = 2 v 2 + v 4 [M_z]_{CX_0}=q(2+q)=2v^2+v^4 [ M z ] C X 0 = q ( 2 + q ) = 2 v 2 + v 4 .
Lemma 3.25. Let r ≥ 2 r\ge2 r ≥ 2 , 1 ≤ j ≤ r 1\le j\le r 1 ≤ j ≤ r and Q = { s 2 , … , s N − 1 } Q=\{s_2,\ldots,s_{N-1}\} Q = { s 2 , … , s N − 1 } . Then
B S ( P N − j , Q ) = C γ j , h γ j = v j [ N − 1 j ] q . BS(P_{N-j},Q)=C_{\gamma_j},\qquad h_{\gamma_j}=v^j\begin{bmatrix}N-1\\j\end{bmatrix}_q. B S ( P N − j , Q ) = C γ j , h γ j = v j [ N − 1 j ] q . Proof. By Lemma 3.19, the coefficient at M c M_c M c in B S ( P N − j ) BS(P_{N-j}) B S ( P N − j ) is v j ( N − j ) − ℓ ( c ) v^{j(N-j)-\ell(c)} v j ( N − j ) − ℓ ( c ) whenever c c c is in its support, and the admissible words there have only A, B in the first N − j N-j N − j positions and only B, C in the last j j j .
The position blocks of Q Q Q are { 1 } \{1\} { 1 } and { 2 , … , N } \{2,\ldots,N\} { 2 , … , N } , so a set O Q ( c ) O_Q(c) O Q ( c ) meeting the support is determined by its type u ≤ j u\le j u ≤ j and its first letter (B or A). Its maximum is obtained by sorting the suffix in the order C < B < A; the longest index in the support is likewise unique (sort within the two blocks of the initial support, keeping the first letter fixed, as in Lemma 3.11). Let λ \lambda λ be the length of the maximum and g g g its length gap from the longest index in the support. By Lemma 3.9:
For instance, in the first row the longest word in the support is B N − j − u A u C u B j − u B^{N-j-u}A^uC^uB^{j-u} B N − j − u A u C u B j − u , of length u ( N − u ) u(N-u) u ( N − u ) , and sorting its suffix gives the displayed g g g . In the second row the first A is fixed and the longest word is A B N − j − u A u − 1 C u B j − u AB^{N-j-u}A^{u-1}C^uB^{j-u} A B N − j − u A u − 1 C u B j − u , of length u ( N − u ) + u + j − N u(N-u)+u+j-N u ( N − u ) + u + j − N .
The normalizing length difference is ℓ ( w Q ) − ℓ ( w P N − j ∩ Q ) = j ( N − j − 1 ) \ell(w_Q)-\ell(w_{P_{N-j}\cap Q})=j(N-j-1) ℓ ( w Q ) − ℓ ( w P N − j ∩ Q ) = j ( N − j − 1 ) , so Corollary 3.18 (with A = j ( N − j ) A=j(N-j) A = j ( N − j ) ) gives lowest exponents j − λ + 2 g j-\lambda+2g j − λ + 2 g , that is j − u j-u j − u and N − j − u N-j-u N − j − u at the two kinds of maxima. The latter is positive since N − 2 j ≥ 1 N-2j\ge1 N − 2 j ≥ 1 . The unique constant coefficient is therefore at the first-row maximum with u = j u=j u = j , whose word is B C j B N − 2 j − 1 A j BC^jB^{N-2j-1}A^j B C j B N − 2 j − 1 A j , namely γ j \gamma_j γ j . All other coefficients acquire a positive power of v v v by (3.10). Integrality, bar-invariance (Lemma 3.3) and Lemma 3.1 now give B S ( P N − j , Q ) = C γ j BS(P_{N-j},Q)=C_{\gamma_j} B S ( P N − j , Q ) = C γ j . Finally, (3.4) and Lemma 3.19 give
h γ j = v j ( N − j ) Π N − 1 Π N − j − 1 Π j = v j [ N − 1 j ] q , h_{\gamma_j}=v^{j(N-j)}\frac{\Pi_{N-1}}{\Pi_{N-j-1}\Pi_j}=v^j\begin{bmatrix}N-1\\j\end{bmatrix}_q, h γ j = v j ( N − j ) Π N − j − 1 Π j Π N − 1 = v j [ N − 1 j ] q ,
as required. □
:::
Common tools for the two chains
This section collects the tools used in the multiplication formulas for both chains (Theorems 2.8 and 2.11). After the general KL rule and descent restrictions, we record a positivity bound for standard coefficients along finite increasing steps. We then classify the possible correction indices and determine the cover contributions. The expansions in Subsection 4.4 separate these known contributions from the remaining multiplicities. Finally, affine cancellation reduces the required coefficients by deleting two common positions at a time. Section 5 uses these tools in an induction along the first chain, with the rank fixed; Section 6 then treats the second chain without induction on the rank. Both sections also invoke results in smaller rank, but only ones already established for every r r r .
General tools for multiplication and Bruhat order
For w ∈ W w\in W w ∈ W and x ∈ I W x\in{}^IW x ∈ I W , write
R ( w ) = { s ∈ S : w s < w } , R I ( x ) = R ( w x r ) . R(w)=\{s\in S: ws<w\},\qquad R_I(x)=R(wxr). R ( w ) = { s ∈ S : w s < w } , R I ( x ) = R ( w x r ) .
The following is the standard KL multiplication rule [25 (2.3.a)–(2.3.c)] , expressed in the spherical module via the identification in [37 Proposition 3.4(1)] .
Proposition 4.1. Let s ∈ S s\in S s ∈ S and x , x s ∈ I W x,xs\in{}^IW x , x s ∈ I W with x s > x xs>x x s > x . Then
C x C s = C x s + ∑ c ∈ I W c < x , s ∈ R I ( c ) μ I ( c , x ) C c . (37) C_xC_s=C_{xs}+\sum_{\substack{c\in{}^IW\\c<x,\ s\in R_I(c)}}\mu_I(c,x)C_c. \tag*{(37)} C x C s = C x s + c ∈ I W c < x , s ∈ R I ( c ) ∑ μ I ( c , x ) C c . (37)
For c , x ∈ I W c,x \in{}^{I}W c , x ∈ I W , write c ◃ x c \triangleleft x c ◃ x if c < x c < x c < x and ℓ ( x ) = ℓ ( c ) + 1 \ell(x)=\ell(c)+1 ℓ ( x ) = ℓ ( c ) + 1 . Such a cover satisfies μ I ( c , x ) = 1 \mu_I(c,x)=1 μ I ( c , x ) = 1 , since w I c ≺ w I x w_Ic \prec w_Ix w I c ≺ w I x . We will use the following precise form of the descent restriction, including its consequence for non-covers.
Lemma 4.2. Let c , x ∈ I W c,x \in{}^{I}W c , x ∈ I W satisfy c < x c < x c < x and μ I ( c , x ) ≠ 0 \mu_I(c,x) \ne0 μ I ( c , x ) = 0 . If s ∈ R I ( x ) ∖ R I ( c ) s \in R_I(x) \setminus R_I(c) s ∈ R I ( x ) ∖ R I ( c ) , then
x = c s . (38) x=cs. \tag*{(38)} x = cs . (38) In particular, if ℓ ( x ) − ℓ ( c ) > 1 \ell(x)-\ell(c)>1 ℓ ( x ) − ℓ ( c ) > 1 , then
R I ( x ) ⊆ R I ( c ) . (39) R_I(x) \subseteq R_I(c). \tag*{(39)} R I ( x ) ⊆ R I ( c ) . (39)
Proof. Apply the right-handed form of [9 Corollary 5.3] to w I c < w I x w_Ic < w_Ix w I c < w I x . It gives w I x = w I c s w_Ix=w_Ics w I x = w I cs when the leading coefficient is nonzero and the upper element has a right descent missing from the lower one. Cancelling w I w_I w I proves (4.2). This equality has length difference one, giving (4.3) for non-covers. □
We collect the nonzero correction indices in (4.1) as follows.
Definition 4.3. Let s ∈ S s \in S s ∈ S and x , x s ∈ I W x,xs \in{}^{I}W x , x s ∈ I W satisfy x s > x xs>x x s > x . Define
C ( x , s ) = { c ∈ I W : c < x , s ∈ R I ( c ) , μ I ( c , x ) ≠ 0 } , \mathcal{C}(x,s)=\{c \in{}^{I}W:c<x,\ s \in R_I(c),\ \mu_I(c,x)\ne0\}, C ( x , s ) = { c ∈ I W : c < x , s ∈ R I ( c ) , μ I ( c , x ) = 0 } , C c o v ( x , s ) = { c ∈ C ( x , s ) : c ◃ x } , \mathcal{C}_{\mathrm{cov}}(x,s)=\{c \in\mathcal{C}(x,s):c\triangleleft x\}, C cov ( x , s ) = { c ∈ C ( x , s ) : c ◃ x } , C n c ( x , s ) = C ( x , s ) ∖ C c o v ( x , s ) . \mathcal{C}_{\mathrm{nc}}(x,s)=\mathcal{C}(x,s)\setminus\mathcal{C}_{\mathrm{cov}}(x,s). C nc ( x , s ) = C ( x , s ) ∖ C cov ( x , s ) .
These are the correction candidates in (4.1), separated into covers and non-covers.
We also use the positivity of ordinary KL polynomials [15 Corollary 1.2(1)] . The standard coefficients of C x C_x C x are the ordinary coefficients h w I c , w I x h_{w_Ic,w_Ix} h w I c , w I x ; thus they, and the integers μ I ( c , x ) \mu_I(c,x) μ I ( c , x ) , are nonnegative. For Laurent polynomials, write f ⪯ g f\preceq g f ⪯ g when g − f ∈ Z ≥ 0 [ v , v − 1 ] g-f\in\mathbb{Z}_{\ge0}[v,v^{-1}] g − f ∈ Z ≥ 0 [ v , v − 1 ] . We write ord v ( f ) \operatorname{ord}_v(f) ord v ( f ) for the smallest exponent of a nonzero Laurent polynomial or series, and put ord v ( 0 ) = + ∞ \operatorname{ord}_v(0)=+\infty ord v ( 0 ) = + ∞ .
Lemma 4.4. Let s ∈ I s \in I s ∈ I and c ∈ I W c \in{}^{I}W c ∈ I W satisfy c s ∉ I W cs\notin{}^{I}W cs ∈ / I W . For every F ∈ M F \in\mathcal{M} F ∈ M ,
[ M c ] ( F C s ) = ( v + v − 1 ) [ M c ] F . (40) [M_c](FC_s)=(v+v^{-1})[M_c]F. \tag*{(40)} [ M c ] ( F C s ) = ( v + v − 1 ) [ M c ] F . (40) If x , x s ∈ I W x,xs \in{}^{I}W x , x s ∈ I W and x s > x xs>x x s > x , then
0 ⪯ [ M c ] C x s ⪯ ( v + v − 1 ) [ M c ] C x , 0\preceq[M_c]C_{xs}\preceq(v+v^{-1})[M_c]C_x, 0 ⪯ [ M c ] C x s ⪯ ( v + v − 1 ) [ M c ] C x , ord v ( [ M c ] C x s ) ≥ ord v ( [ M c ] C x ) − 1. (41) \operatorname{ord}_v([M_c]C_{xs})\ge\operatorname{ord}_v([M_c]C_x)-1. \tag*{(41)} ord v ([ M c ] C x s ) ≥ ord v ([ M c ] C x ) − 1. (41)
Proof. Since c c c is minimal and c s cs cs is not, c s = t c cs=tc cs = t c for some t ∈ I t\in I t ∈ I . The standard spherical multiplication rule gives M c C s = ( v + v − 1 ) M c M_cC_s=(v+v^{-1})M_c M c C s = ( v + v − 1 ) M c . No distinct minimal index can contribute to M c M_c M c : its only possible partner under right multiplication by s s s would be c s cs cs , which is not minimal. This proves (4.4) for arbitrary F F F .
By (4.1), C x C s − C x s C_xC_s-C_{xs} C x C s − C x s is a sum of canonical elements with nonnegative multiplicities. Their standard coefficients are nonnegative by KL positivity. Taking the coefficient of M c M_c M c and using (4.4) gives the first line of (4.5). Multiplication by v + v − 1 v+v^{-1} v + v − 1 lowers the smallest possible exponent by one, proving the second line. □
Corollary 4.5. Let u 0 , … , u L ∈ I W u_0,\ldots,u_L\in{}^{I}W u 0 , … , u L ∈ I W satisfy u i = u i − 1 t i > u i − 1 u_i=u_{i-1}t_i>u_{i-1} u i = u i − 1 t i > u i − 1 with t i ∈ I t_i\in I t i ∈ I . Then
ord v ( h u L ) ≥ ord v ( h u 0 ) − L . (42) \operatorname{ord}_v(h_{u_L})\ge\operatorname{ord}_v(h_{u_0})-L. \tag*{(42)} ord v ( h u L ) ≥ ord v ( h u 0 ) − L . (42)
Proof. Take c = e c=e c = e in Lemma 4.4 at each step. Since t i ∈ I t_i\in I t i ∈ I , the element e t i et_i e t i is not a minimal left representative. Iteration gives (4.6). □
We test Bruhat comparisons using the following standard criterion.
Lemma 4.6. For w ∈ W w\in W w ∈ W and m , M ∈ Z m,M\in\mathbb{Z} m , M ∈ Z , define
R w ( m , M ) = # { i ∈ Z : i ≤ m , w ( i ) ≥ M } , R_w(m,M)=\#\{i\in\mathbb{Z}:i\le m,\ w(i)\ge M\}, R w ( m , M ) = # { i ∈ Z : i ≤ m , w ( i ) ≥ M } , R ~ w ( m , M ) = # { i ∈ Z : i > m , w ( i ) ≤ M } . \widetilde{R}_w(m,M)=\#\{i\in\mathbb{Z}:i>m,\ w(i)\le M\}. R w ( m , M ) = # { i ∈ Z : i > m , w ( i ) ≤ M } . These numbers are finite. For y , w ∈ W y,w\in W y , w ∈ W ,
y ≤ w ⟺ R y ( m , M ) ≤ R w ( m , M ) for all m , M ∈ Z y\le w\quad\Longleftrightarrow\quad R_y(m,M)\le R_w(m,M)\quad\text{for all }m,M\in\mathbb{Z} y ≤ w ⟺ R y ( m , M ) ≤ R w ( m , M ) for all m , M ∈ Z ⟺ R ~ y ( m , M ) ≤ R ~ w ( m , M ) for all m , M ∈ Z . \Longleftrightarrow\quad\widetilde{R}_y(m,M)\le\widetilde{R}_w(m,M)\quad\text{for all }m,M\in\mathbb{Z}. ⟺ R y ( m , M ) ≤ R w ( m , M ) for all m , M ∈ Z .
Proof. The criterion involving R R R is [7 Theorem 8.3.7] ; see also [24 Proposition 4.1] . Since inversion preserves Bruhat order, applying this criterion to inverses and using
R ~ w ( m , M ) = R w − 1 ( M , m + 1 ) \widetilde{R}_{w}(m,M)=R_{w^{-1}}(M,m+1) R w ( m , M ) = R w − 1 ( M , m + 1 )
gives the equivalent criterion involving R ~ \widetilde{R} R . □
Possible correction indices for the two chains
We now apply the general tools above to our two chains. We first introduce the elements Θ j \Theta_j Θ j , defined below, associated to the fully sorted admissible words. The element Θ r \Theta_r Θ r will serve as a common Bruhat upper bound for both chains, as verified in the proof of Lemma 4.12. The family Θ j \Theta_j Θ j will also appear among the possible correction indices. The next two lemmas describe the minimal representatives below Θ r \Theta_r Θ r by admissible words and give the descent and prefix-count tests used to restrict those words.
Definition 4.7. For 0 ≤ j ≤ r 0 \le j \le r 0 ≤ j ≤ r , let Θ j \Theta_j Θ j be the element associated to the admissible word of type j j j
C j B N − 2 j A j . C^jB^{N-2j}A^j. C j B N − 2 j A j .
Its window is
Θ j = [ N + 1 , … , N + j , j + 1 , … , N − j , 1 − j , … , 0 ] . \Theta_j=[N+1,\ldots,N+j,j+1,\ldots,N-j,1-j,\ldots,0]. Θ j = [ N + 1 , … , N + j , j + 1 , … , N − j , 1 − j , … , 0 ] .
In particular, Θ 0 = e \Theta_0=e Θ 0 = e .
Lemma 4.8. If c ∈ I W c \in{}^{I}W c ∈ I W satisfies c ≤ Θ r c \le\Theta_r c ≤ Θ r , then c = c ω c=c_\omega c = c ω for an admissible word of some type 0 ≤ t ≤ r 0 \le t \le r 0 ≤ t ≤ r .
Proof. We have
R Θ r ( N , 2 N + 1 ) = R ~ Θ r ( 0 , − N ) = 0. R_{\Theta_r}(N,2N+1)=\widetilde{R}_{\Theta_r}(0,-N)=0. R Θ r ( N , 2 N + 1 ) = R Θ r ( 0 , − N ) = 0.
Indeed, if i ≤ N i \le N i ≤ N , write i = a + q N i=a+qN i = a + q N with 1 ≤ a ≤ N 1 \le a \le N 1 ≤ a ≤ N and q ≤ 0 q \le0 q ≤ 0 . Since the largest entry of the window of Θ r \Theta_r Θ r is N + r N+r N + r , periodicity gives
Θ r ( i ) = Θ r ( a ) + q N ≤ N + r < 2 N + 1. \Theta_r(i)=\Theta_r(a)+qN \le N+r<2N+1. Θ r ( i ) = Θ r ( a ) + q N ≤ N + r < 2 N + 1.
Thus { i ∈ Z ∣ i ≤ N , Θ r ( i ) ≥ 2 N + 1 } \{i \in\mathbb{Z}\mid i \le N,\Theta_r(i)\ge2N+1\} { i ∈ Z ∣ i ≤ N , Θ r ( i ) ≥ 2 N + 1 } is empty and R Θ r ( N , 2 N + 1 ) = 0 R_{\Theta_r}(N,2N+1)=0 R Θ r ( N , 2 N + 1 ) = 0 . The other equality follows similarly.
Since c ≤ Θ r c \le\Theta_r c ≤ Θ r , Lemma 4.6 now gives
R c ( N , 2 N + 1 ) = R ~ c ( 0 , − N ) = 0. R_c(N,2N+1)=\widetilde{R}_c(0,-N)=0. R c ( N , 2 N + 1 ) = R c ( 0 , − N ) = 0.
Consequently, 1 − N ≤ c ( i ) ≤ 2 N 1-N \le c(i) \le2N 1 − N ≤ c ( i ) ≤ 2 N for 1 ≤ i ≤ N 1 \le i \le N 1 ≤ i ≤ N . In other words, all the entries of the window of c c c belong to the integer interval [ 1 − N , 2 N ] [1-N,2N] [ 1 − N , 2 N ] .
Since c ∈ I W c \in{}^{I}W c ∈ I W , the converse direction of Lemma 3.7 now gives a unique admissible word ω \omega ω such that c = c ω c=c_\omega c = c ω . Its type t t t satisfies 0 ≤ t ≤ r 0 \le t \le r 0 ≤ t ≤ r by Definition 3.6. □
Lemma 4.9. Let ω = ω 1 ⋯ ω N \omega=\omega_1\cdots\omega_N ω = ω 1 ⋯ ω N be an admissible word. For 1 ≤ i < N 1 \le i<N 1 ≤ i < N ,
s i ∈ R I ( c ω ) ⟺ ω i ≤ ω i + 1 ( C < B < A ) . (43) s_i \in R_I(c_\omega)\quad\Longleftrightarrow\quad\omega_i\le\omega_{i+1}\qquad(C<B<A). \tag*{(43)} s i ∈ R I ( c ω ) ⟺ ω i ≤ ω i + 1 ( C < B < A ) . (43)
If η = η 1 ⋯ η N \eta=\eta_1\cdots\eta_N η = η 1 ⋯ η N is another admissible word with c η ≤ c ω c_\eta\le c_\omega c η ≤ c ω , then, for 1 ≤ m ≤ N 1 \le m\le N 1 ≤ m ≤ N ,
# { i ≤ m : η i = C } ≤ # { i ≤ m : ω i = C } . (44) \#\{i\le m:\eta_i=C\}\le\#\{i\le m:\omega_i=C\}. \tag*{(44)} # { i ≤ m : η i = C } ≤ # { i ≤ m : ω i = C } . (44)
Proof. We first prove (4.7). Fix 1 ≤ i < N 1 \le i<N 1 ≤ i < N . If ω i ≠ ω i + 1 \omega_i\ne\omega_{i+1} ω i = ω i + 1 , let η \eta η be obtained by interchanging these two letters. Then c η = c ω s i c_\eta=c_\omega s_i c η = c ω s i , and Lemma 3.9 gives
ℓ ( c η ) − ℓ ( c ω ) = ι ( ω ) − ι ( η ) = { − 1 , ω i < ω i + 1 , 1 , ω i > ω i + 1 , ( C < B < A ) . \ell(c_\eta)-\ell(c_\omega)=\iota(\omega)-\iota(\eta)=
\begin{cases}
-1, & \omega_i<\omega_{i+1},\\
1, & \omega_i>\omega_{i+1},
\end{cases}
\qquad(C<B<A). ℓ ( c η ) − ℓ ( c ω ) = ι ( ω ) − ι ( η ) = { − 1 , 1 , ω i < ω i + 1 , ω i > ω i + 1 , ( C < B < A ) .
Since c ω , c η ∈ I W c_\omega,c_\eta\in{}^{I}W c ω , c η ∈ I W , length additivity after left multiplication by w I w_I w I shows that s i ∈ R I ( c ω ) s_i\in R_I(c_\omega) s i ∈ R I ( c ω ) exactly when ω i < ω i + 1 \omega_i<\omega_{i+1} ω i < ω i + 1 .
If ω i = ω i + 1 \omega_i=\omega_{i+1} ω i = ω i + 1 , the corresponding window entries are consecutive entries of the same list in (3.7). They lie in an interval [ q N + 1 , ( q + 1 ) N ] [qN+1,(q+1)N] [ q N + 1 , ( q + 1 ) N ] for some q ∈ Z q\in\mathbb{Z} q ∈ Z , so c ω s i = s j c ω c_\omega s_i=s_jc_\omega c ω s i = s j c ω for some s j ∈ I s_j\in I s j ∈ I . Hence
ℓ ( w I c ω s i ) = ℓ ( w I s j ) + ℓ ( c ω ) = ℓ ( w I c ω ) − 1. \ell(w_Ic_\omega s_i)=\ell(w_Is_j)+\ell(c_\omega)=\ell(w_Ic_\omega)-1. ℓ ( w I c ω s i ) = ℓ ( w I s j ) + ℓ ( c ω ) = ℓ ( w I c ω ) − 1.
Thus s i ∈ R I ( c ω ) s_i\in R_I(c_\omega) s i ∈ R I ( c ω ) also when the letters are equal, proving (4.7).
Finally, let ω \omega ω be an admissible word of type t t t . Its largest window entry is N + t N+t N + t , so periodicity gives c ω ( i ) ≤ t < N + 1 c_\omega(i)\le t<N+1 c ω ( i ) ≤ t < N + 1 for every i ≤ 0 i\le0 i ≤ 0 . Hence, for 1 ≤ m ≤ N 1\le m\le N 1 ≤ m ≤ N ,
R c ω ( m , N + 1 ) = # { 1 ≤ i ≤ m : ω i = C } . R_{c_\omega}(m,N+1)=\#\{1\le i\le m:\omega_i=C\}. R c ω ( m , N + 1 ) = # { 1 ≤ i ≤ m : ω i = C } .
Applying this identity to η \eta η and ω \omega ω , the inequality c η ≤ c ω c_\eta\le c_\omega c η ≤ c ω and Lemma 4.6 give (44) . □
The first-chain candidates include the following family.
Definition 4.10. For 1 ≤ j ≤ r 1 \le j \le r 1 ≤ j ≤ r , let κ j = c ω ∈ I W \kappa_j=c_\omega\in{}^{I}W κ j = c ω ∈ I W for the admissible word of type j j j
ω = A C j B N − 2 j A j − 1 . \omega=\mathrm{AC}^{j}\mathrm{B}^{N-2j}\mathrm{A}^{j-1}. ω = AC j B N − 2 j A j − 1 . Its window is
κ j = [ 1 − j , N + 1 , … , N + j , j + 1 , … , N − j , 2 − j , … , 0 ] . \kappa_j=[1-j,N+1,\ldots,N+j,j+1,\ldots,N-j,2-j,\ldots,0]. κ j = [ 1 − j , N + 1 , … , N + j , j + 1 , … , N − j , 2 − j , … , 0 ] .
We record four properties of κ 1 \kappa_1 κ 1 that will be used to prove the vanishing of the κ j \kappa_j κ j corrections in Theorem 5.1. Recall the element z ( 1 , 1 , 0 ) z(1,1,0) z ( 1 , 1 , 0 ) from Theorem 3.12.
Lemma 4.11. For r ≥ 2 r \ge2 r ≥ 2 , we have
κ 1 s 2 = z ( 1 , 1 , 0 ) ∈ I W , κ 1 s 2 < κ 1 , κ 1 ≰ X 0 , κ 1 s i ∉ I W ( 3 ≤ i ≤ r ) . \kappa_1s_2=z(1,1,0)\in{}^{I}W,\qquad\kappa_1s_2<\kappa_1,\qquad\kappa_1\not\le X_0,\qquad\kappa_1s_i\notin{}^{I}W\quad(3\le i\le r). κ 1 s 2 = z ( 1 , 1 , 0 ) ∈ I W , κ 1 s 2 < κ 1 , κ 1 ≤ X 0 , κ 1 s i ∈ / I W ( 3 ≤ i ≤ r ) .
Proof. Right multiplication by s 2 s_2 s 2 exchanges the second and third entries of the window of κ 1 \kappa_1 κ 1 , giving
κ 1 s 2 = [ 0 , 2 , N + 1 , 3 , … , N − 1 ] = z ( 1 , 1 , 0 ) ∈ I W . \kappa_1s_2=[0,2,N+1,3,\ldots,N-1]=z(1,1,0)\in{}^{I}W. κ 1 s 2 = [ 0 , 2 , N + 1 , 3 , … , N − 1 ] = z ( 1 , 1 , 0 ) ∈ I W .
Since κ 1 ( 2 ) > κ 1 ( 3 ) \kappa_1(2)>\kappa_1(3) κ 1 ( 2 ) > κ 1 ( 3 ) , we get κ 1 s 2 < κ 1 \kappa_1s_2<\kappa_1 κ 1 s 2 < κ 1 . The admissible words of κ 1 \kappa_1 κ 1 and X 0 X_0 X 0 begin with AC and AB, respectively, so the Bruhat prefix bound (44) at m = 2 m=2 m = 2 gives κ 1 ≰ X 0 \kappa_1\not\le X_0 κ 1 ≤ X 0 . Finally, right multiplication by s i s_i s i exchanges the entries in positions i , i + 1 i,i+1 i , i + 1 , which in the window of κ 1 \kappa_1 κ 1 are i − 1 , i i-1,i i − 1 , i . This is also the effect of left multiplication by s i − 1 s_{i-1} s i − 1 , which exchanges these two values. Hence κ 1 s i = s i − 1 κ 1 \kappa_1s_i=s_{i-1}\kappa_1 κ 1 s i = s i − 1 κ 1 for 3 ≤ i ≤ r 3\le i\le r 3 ≤ i ≤ r . Since s i − 1 ∈ I s_{i-1}\in I s i − 1 ∈ I and κ 1 ∈ I W \kappa_1\in{}^{I}W κ 1 ∈ I W , it follows that κ 1 s i ∉ I W \kappa_1s_i\notin{}^{I}W κ 1 s i ∈ / I W . □
Lemma 4.12. Assume r ≥ 2 r\ge2 r ≥ 2 .
(i) For 0 ≤ k ≤ r − 2 0\le k\le r-2 0 ≤ k ≤ r − 2 , the non-cover candidates for the first-chain step x k → x k + 1 x_k\to x_{k+1} x k → x k + 1 satisfy
C n c ( x k , s k + 2 ) ⊆ { e } ∪ { γ j , κ j : 1 ≤ j ≤ k } . C_{\mathrm{nc}}(x_k,s_{k+2})\subseteq\{e\}\cup\{\gamma_j,\kappa_j:1\le j\le k\}. C nc ( x k , s k + 2 ) ⊆ { e } ∪ { γ j , κ j : 1 ≤ j ≤ k } . (ii) For 0 ≤ t ≤ r − 2 0\le t\le r-2 0 ≤ t ≤ r − 2 , put b = r − 1 − t b=r-1-t b = r − 1 − t . Every element of C n c ( y t + 1 , s b ) C_{\mathrm{nc}}(y_{t+1},s_b) C nc ( y t + 1 , s b ) has the form c ω c_\omega c ω , where ω = ω 1 ⋯ ω N \omega=\omega_1\cdots\omega_N ω = ω 1 ⋯ ω N is an admissible word satisfying
ω 1 ≤ ⋯ ≤ ω r + 1 , ω r + 2 ≤ ⋯ ≤ ω N , (45) \omega_1\le\cdots\le\omega_{r+1},\qquad\omega_{r+2}\le\cdots\le\omega_N, \tag*{(45)} ω 1 ≤ ⋯ ≤ ω r + 1 , ω r + 2 ≤ ⋯ ≤ ω N , (45) and
ω b ≠ C , (46) \omega_b\ne\mathrm{C}, \tag*{(46)} ω b = C , (46) where the letters are ordered by C < B < A \mathrm{C}<\mathrm{B}<\mathrm{A} C < B < A .
Proof. For either chain, let u u u be a non-cover candidate. Then u ∈ I W u\in{}^{I}W u ∈ I W and u u u lies below the corresponding chain element. The chain words are
x k : A C k B C r − k A r − 1 , y t + 1 : C b − 1 A C t + 1 B C A r − 1 , b = r − 1 − t . x_k:\ \mathrm{AC}^{k}\mathrm{BC}^{r-k}\mathrm{A}^{r-1},\qquad
y_{t+1}:\ \mathrm{C}^{b-1}\mathrm{AC}^{t+1}\mathrm{BCA}^{r-1},\qquad b=r-1-t. x k : AC k BC r − k A r − 1 , y t + 1 : C b − 1 AC t + 1 BCA r − 1 , b = r − 1 − t .
We arrange the letters in each displayed word in the order C < B < A \mathrm{C}<\mathrm{B}<\mathrm{A} C < B < A by repeatedly applying the adjacent interchanges
A B ⟼ B A , A C ⟼ C A , B C ⟼ C B . \mathrm{AB}\longmapsto\mathrm{BA},\qquad
\mathrm{AC}\longmapsto\mathrm{CA},\qquad
\mathrm{BC}\longmapsto\mathrm{CB}. AB ⟼ BA , AC ⟼ CA , BC ⟼ CB .
By Lemma 4.9 and length additivity for minimal representatives, each of these interchanges gives a Bruhat cover. Both displayed words have type r r r , so the process ends at C r B A r \mathrm{C}^{r}\mathrm{BA}^{r} C r BA r , the word of Θ r \Theta_r Θ r . Hence x k ≤ Θ r x_k\le\Theta_r x k ≤ Θ r and y t + 1 ≤ Θ r y_{t+1}\le\Theta_r y t + 1 ≤ Θ r . It follows that u ≤ Θ r u\le\Theta_r u ≤ Θ r , and Lemma 4.8 gives an admissible word ω \omega ω with u = c ω u=c_\omega u = c ω . We now apply Lemmas 4.2 and 4.9 to restrict this word.
(i) Applying (43) to the word for x k x_k x k displayed above shows that the finite generators missing from R I ( x k ) R_I(x_k) R I ( x k ) are s 1 s_1 s 1 and s k + 2 s_{k+2} s k + 2 . Lemma 4.2 gives R I ( x k ) ⊆ R I ( c ω ) R_I(x_k)\subseteq R_I(c_\omega) R I ( x k ) ⊆ R I ( c ω ) , and the candidate condition adds s k + 2 ∈ R I ( c ω ) s_{k+2}\in R_I(c_\omega) s k + 2 ∈ R I ( c ω ) . We therefore have I ∖ { s 1 } ⊆ R I ( c ω ) I\setminus\{s_1\}\subseteq R_I(c_\omega) I ∖ { s 1 } ⊆ R I ( c ω ) . Equation (43) gives
ω 2 ≤ ω 3 ≤ ⋯ ≤ ω N ( C < B < A ) . \omega_2\le\omega_3\le\cdots\le\omega_N\qquad(\mathrm{C}<\mathrm{B}<\mathrm{A}). ω 2 ≤ ω 3 ≤ ⋯ ≤ ω N ( C < B < A ) .
These descent conditions leave the first letter ω 1 \omega_1 ω 1 unrestricted. The suffix ω 2 ⋯ ω N \omega_2\cdots\omega_N ω 2 ⋯ ω N consists of three (possibly empty) blocks of C, B, and A, in that order.
Let j j j be the type of ω \omega ω . If j = 0 j=0 j = 0 , then ω = B N \omega=\mathrm{B}^{N} ω = B N and c ω = e c_\omega=e c ω = e .
Suppose j > 0 j>0 j > 0 . We notice that first letter and the prescribed letter counts determine the sorted suffix, giving the following three cases.
::: If ω 1 = C \omega_1 = \mathrm{C} ω 1 = C , the suffix has j − 1 j - 1 j − 1 letters C, N − 2 j N - 2j N − 2 j letters B, and j j j letters A. Thus
ω = C j B N − 2 j A j , c ω = Θ j . \omega= \mathrm{C}^j\mathrm{B}^{N-2j}\mathrm{A}^j,\qquad c_\omega= \Theta_j. ω = C j B N − 2 j A j , c ω = Θ j .
Thus u = c ω = Θ j u = c_\omega= \Theta_j u = c ω = Θ j . Since u u u is a candidate for the step x k → x k + 1 x_k \to x_{k+1} x k → x k + 1 , we must also have u < x k u < x_k u < x k . Applying the prefix bound (44) to Θ j < x k \Theta_j < x_k Θ j < x k at m = 1 m = 1 m = 1 would give 1 ≤ 0 1 \le0 1 ≤ 0 : the word of Θ j \Theta_j Θ j begins with C, whereas the word of x k x_k x k begins with A. This contradiction excludes the case ω 1 = C \omega_1 = \mathrm{C} ω 1 = C .
If ω 1 = B \omega_1 = \mathrm{B} ω 1 = B , the suffix has j j j letters C, N − 2 j − 1 N - 2j - 1 N − 2 j − 1 letters B, and j j j letters A. Hence
ω = B C j B N − 2 j − 1 A j , c ω = γ j . \omega= \mathrm{B}\mathrm{C}^j\mathrm{B}^{N-2j-1}\mathrm{A}^j,\qquad c_\omega= \gamma_j. ω = B C j B N − 2 j − 1 A j , c ω = γ j .
If ω 1 = A \omega_1 = \mathrm{A} ω 1 = A , the suffix has j j j letters C, N − 2 j N - 2j N − 2 j letters B, and j − 1 j - 1 j − 1 letters A. Hence
ω = A C j B N − 2 j A j − 1 , c ω = κ j . \omega= \mathrm{A}\mathrm{C}^j\mathrm{B}^{N-2j}\mathrm{A}^{j-1},\qquad c_\omega= \kappa_j. ω = A C j B N − 2 j A j − 1 , c ω = κ j .
For either of the last two words, the first k + 2 k + 2 k + 2 positions contain min ( j , k + 1 ) \min(j, k + 1) min ( j , k + 1 ) letters C, whereas the word of x k x_k x k has k k k such letters. Applying (44) at m = k + 2 m = k + 2 m = k + 2 gives
min ( j , k + 1 ) ≤ k , \min(j, k + 1) \le k, min ( j , k + 1 ) ≤ k ,
and therefore j ≤ k j \le k j ≤ k . This proves (i).
(ii) Applying (43) to the word for y t + 1 y_{t+1} y t + 1 displayed above shows that the finite generators missing from R I ( y t + 1 ) R_I(y_{t+1}) R I ( y t + 1 ) are s b , s r + 1 s_b, s_{r+1} s b , s r + 1 . Lemma 4.2 gives R I ( y t + 1 ) ⊆ R I ( c ω ) R_I(y_{t+1}) \subseteq R_I(c_\omega) R I ( y t + 1 ) ⊆ R I ( c ω ) , and the candidate condition adds s b ∈ R I ( c ω ) s_b \in R_I(c_\omega) s b ∈ R I ( c ω ) . Equation (43) therefore gives (45) . The word of y t + 1 y_{t+1} y t + 1 begins with C b − 1 A \mathrm{C}^{b-1}\mathrm{A} C b − 1 A , so the prefix bound (44) at m = b m = b m = b gives
# { 1 ≤ i ≤ b : ω i = C } ≤ b − 1. \#\{1 \le i \le b : \omega_i = \mathrm{C}\} \le b - 1. # { 1 ≤ i ≤ b : ω i = C } ≤ b − 1.
If ω b = C \omega_b = \mathrm{C} ω b = C , the first ordered block would begin with at least b b b letters C, contradicting this inequality. This proves (46) and completes (ii). □
Covers contribute only the preceding chain element
We first record the local changes that can produce a cover between minimal representatives.
For integers p < q p < q p < q with p ≢ q ( m o d N ) p \not\equiv q \pmod N p ≡ q ( mod N ) we write t p , q ∈ W t_{p,q} \in W t p , q ∈ W for the transposition that exchanges positions p + k N p + kN p + k N and q + k N q + kN q + k N for every k ∈ Z k \in\mathbb{Z} k ∈ Z .
Lemma 4.13. Let x = c ω ≤ Θ r x = c_\omega\le\Theta_r x = c ω ≤ Θ r , and suppose that z ∈ I W z \in{}^I W z ∈ I W satisfies z ⊲ x z \lhd x z ⊲ x . Then z = c ω ′ z = c_{\omega'} z = c ω ′ for an admissible word ω ′ \omega' ω ′ obtained from ω \omega ω by one of the following operations:
Interchange C at position p p p with B at q > p q > p q > p , with only letters A between them.
Interchange B at p p p with A at q > p q > p q > p , with only letters C between them.
Interchange an adjacent pair CA.
Replace the first A and the last C of ω \omega ω by B, provided the first A precedes the last C and every occurrence of B lies strictly between them.
Proof. By Lemma 4.8, we have z = c ω ′ z = c_{\omega'} z = c ω ′ for an admissible word ω ′ \omega' ω ′ . Since z ⊲ x z \lhd x z ⊲ x , we can write z = x t p , q z = xt_{p,q} z = x t p , q for integers p < q p < q p < q with 1 ≤ p ≤ N 1 \le p \le N 1 ≤ p ≤ N and p ≢ q ( m o d N ) p \not\equiv q \pmod N p ≡ q ( mod N ) . The cover criterion in [27 Lemma 2.2(2)] gives
x ( p ) > x ( q ) , x(p) > x(q), x ( p ) > x ( q ) ,
x ( i ) ∉ ( x ( q ) , x ( p ) ) for every integer p < i < q . (47) x(i) \notin\left(x(q),x(p)\right)\qquad\text{for every integer }p < i < q. \tag*{(47)} x ( i ) ∈ / ( x ( q ) , x ( p ) ) for every integer p < i < q . (47)
Write q = q 0 + d N q = q_0 + dN q = q 0 + d N with 1 ≤ q 0 ≤ N 1 \le q_0 \le N 1 ≤ q 0 ≤ N . Since 1 ≤ p ≤ N 1 \le p \le N 1 ≤ p ≤ N and p < q p < q p < q , we have d ≥ 0 d \ge0 d ≥ 0 . Let 0 ≤ t ≤ r 0 \le t \le r 0 ≤ t ≤ r be the type of ω \omega ω . The window entries of x x x lie between 1 − t 1 - t 1 − t and N + t N + t N + t , and 2 t ≤ 2 r < N 2t \le2r < N 2 t ≤ 2 r < N . Thus
d N < x ( p ) − x ( q 0 ) ≤ ( N + t ) − ( 1 − t ) = N + 2 t − 1 < 2 N . dN < x(p) - x(q_0) \le(N+t) - (1-t) = N + 2t - 1 < 2N. d N < x ( p ) − x ( q 0 ) ≤ ( N + t ) − ( 1 − t ) = N + 2 t − 1 < 2 N .
We conclude that d = 0 d = 0 d = 0 or d = 1 d = 1 d = 1 .
If d = 0 d = 0 d = 0 , both positions p p p and q q q lie in the window. The exchange preserves the set of window entries, so ω ′ \omega' ω ′ also has type t t t . Since x ( p ) > x ( q ) x(p) > x(q) x ( p ) > x ( q ) , the letters ω p \omega_p ω p and ω q \omega_q ω q must be distinct and satisfy ω p < ω q \omega_p < \omega_q ω p < ω q in the letter order C < B < A. An occurrence of either letter ω p \omega_p ω p or ω q \omega_q ω q strictly between p p p and q q q would put entries of that list out of increasing order after the exchange, contradicting the admissibility of ω ′ \omega' ω ′ . Thus only the third letter can occur between positions p p p and q q q . Consequently, the subword ω p ⋯ ω q \omega_p \cdots\omega_q ω p ⋯ ω q must be one of C A k B \mathrm{C}\mathrm{A}^k\mathrm{B} C A k B , B C k A \mathrm{B}\mathrm{C}^k\mathrm{A} B C k A , or C B k A \mathrm{C}\mathrm{B}^k\mathrm{A} C B k A , for some k ≥ 0 k \ge0 k ≥ 0 . The first two options are (i) and (ii), respectively. The option C B k A \mathrm{C}\mathrm{B}^k\mathrm{A} C B k A with k ≥ 1 k \ge1 k ≥ 1 is excluded by (47) : any occurrence of B at a position p < i < q p < i < q p < i < q gives x ( p ) > x ( i ) > x ( q ) x(p) > x(i) > x(q) x ( p ) > x ( i ) > x ( q ) . The case k = 0 k = 0 k = 0 corresponds to (iii). If d = 1 d=1 d = 1 , then q = q 0 + N q=q_{0}+N q = q 0 + N . By periodicity, for 1 ≤ i ≤ N 1\leq i\leq N 1 ≤ i ≤ N we have
z ( i ) = { x ( i ) , if i ∉ { p , q 0 } , x ( q 0 ) + N , if i = p , x ( p ) − N , if i = q 0 . (48) z(i)=\begin{cases}
x(i), & \text{if }i\notin\{p,q_{0}\},\\
x(q_{0})+N, & \text{if }i=p,\\
x(p)-N, & \text{if }i=q_{0}.
\end{cases} \tag*{(48)} z ( i ) = ⎩ ⎨ ⎧ x ( i ) , x ( q 0 ) + N , x ( p ) − N , if i ∈ / { p , q 0 } , if i = p , if i = q 0 . (48)
Since x ( p ) > x ( q ) = x ( q 0 ) + N x(p)>x(q)=x(q_{0})+N x ( p ) > x ( q ) = x ( q 0 ) + N , we have x ( p ) − x ( q 0 ) > N x(p)-x(q_{0})>N x ( p ) − x ( q 0 ) > N . The ranges in (26) therefore force ω p = C \omega_{p}=\mathrm{C} ω p = C and ω q 0 = A \omega_{q_{0}}=\mathrm{A} ω q 0 = A .
It follows from (48) that
z ( q 0 ) = x ( p ) − N ∈ { 1 , … , t } , z ( p ) = x ( q 0 ) + N ∈ { N + 1 − t , … , N } . z(q_{0})=x(p)-N\in\{1,\ldots,t\},\qquad z(p)=x(q_{0})+N\in\{N+1-t,\ldots,N\}. z ( q 0 ) = x ( p ) − N ∈ { 1 , … , t } , z ( p ) = x ( q 0 ) + N ∈ { N + 1 − t , … , N } .
Thus the A-entry at q 0 q_{0} q 0 and the C-entry at p p p are both replaced by entries in { 1 , … , N } \{1,\ldots,N\} { 1 , … , N } . All other window entries remain unchanged, so z z z has exactly t − 1 t-1 t − 1 entries at most 0 0 0 and t − 1 t-1 t − 1 entries greater than N N N . Since z = c ω z=c_{\omega} z = c ω and ω ′ \omega' ω ′ is admissible, its type is t − 1 t-1 t − 1 . Its A-entries are therefore 2 − t , … , 0 2-t,\ldots,0 2 − t , … , 0 , and its C-entries are N + 1 , … , N + t − 1 N+1,\ldots,N+t-1 N + 1 , … , N + t − 1 . Compared with the lists for ω \omega ω , the missing entries are precisely 1 − t 1-t 1 − t and N + t N+t N + t . Hence
x ( q 0 ) = 1 − t , x ( p ) = N + t . (49) x(q_{0})=1-t,\qquad x(p)=N+t. \tag*{(49)} x ( q 0 ) = 1 − t , x ( p ) = N + t . (49)
Since each list occurs in increasing order in x x x , position q 0 q_{0} q 0 contains the first A of ω \omega ω , and position p p p contains its last C. By (48) and (49) , the two new entries are
z ( q 0 ) = t , z ( p ) = N + 1 − t . z(q_{0})=t,\qquad z(p)=N+1-t. z ( q 0 ) = t , z ( p ) = N + 1 − t .
These are the smallest and largest entries of the B-list for type t − 1 t-1 t − 1 . The original B-entries, with values t + 1 , … , N − t t+1,\ldots,N-t t + 1 , … , N − t , remain in their original positions. Since the B-entries of z z z occur in increasing order, we obtain
q 0 < i < p whenever ω i = B . q_{0}<i<p\qquad\text{whenever }\omega_{i}=\mathrm{B}. q 0 < i < p whenever ω i = B .
Since ω \omega ω contains N − 2 t ≥ 1 N-2t\geq1 N − 2 t ≥ 1 occurrences of B, we also have q 0 < p q_{0}<p q 0 < p . Thus every occurrence of B in ω \omega ω lies strictly between its first A, at q 0 q_{0} q 0 , and its last C, at p p p . Replacing these two letters by B gives ω ′ \omega' ω ′ as in (iv). □
Example 4.14. Let N = 7 N=7 N = 7 , r = 3 r=3 r = 3 , and consider the admissible word of type 3 3 3
ω = C A C A B C A . \omega=\mathrm{CACABCA}. ω = CACABCA .
The three lists are A 3 = ( − 2 , − 1 , 0 ) A_{3}=(-2,-1,0) A 3 = ( − 2 , − 1 , 0 ) , B 3 = ( 4 ) B_{3}=(4) B 3 = ( 4 ) and C 3 = ( 8 , 9 , 10 ) C_{3}=(8,9,10) C 3 = ( 8 , 9 , 10 ) , so
x = c ω = [ 8 , − 2 , 9 , − 1 , 4 , 10 , 0 ] . x=c_{\omega}=[8,-2,9,-1,4,10,0]. x = c ω = [ 8 , − 2 , 9 , − 1 , 4 , 10 , 0 ] .
Sorting the letters into the order C < B < A by adjacent interchanges increases length at each step and ends at C 3 B A 3 \mathrm{C}^{3}\mathrm{BA}^{3} C 3 BA 3 , the word of Θ 3 \Theta_{3} Θ 3 . Thus x ≤ Θ 3 x\leq\Theta_{3} x ≤ Θ 3 . The following table gives one cover of each type in Lemma 4.13 ; the positions refer to the original word ω \omega ω .
Table 4.14.
The reflections in the first three rows are t 3 , 5 t_{3,5} t 3 , 5 , t 5 , 7 t_{5,7} t 5 , 7 and s 3 s_{3} s 3 , respectively. In the fourth, the only B lies between the first A and the last C, at positions 2 and 6. Here z = x t 6 , 9 z=xt_{6,9} z = x t 6 , 9 , with z ( 2 ) = x ( 6 ) − 7 = 3 z(2)=x(6)-7=3 z ( 2 ) = x ( 6 ) − 7 = 3 and z ( 6 ) = x ( 2 ) + 7 = 5 z(6)=x(2)+7=5 z ( 6 ) = x ( 2 ) + 7 = 5 , and the type drops from 3 to 2.
The inversion numbers of ω \omega ω and the four resulting words are 6 and 7, 7, 7, 3, respectively. Using their types, Lemma 3.9 gives ℓ ( x ) = 18 \ell(x)=18 ℓ ( x ) = 18 and ℓ ( z ) = 17 \ell(z)=17 ℓ ( z ) = 17 in every row. Thus all four satisfy z < x z<x z < x .
Lemma 4.15. Assume r ≥ 2 r\geq2 r ≥ 2 .
For 0 ≤ k ≤ r − 2 0\leq k\leq r-2 0 ≤ k ≤ r − 2 , the cover candidates for the first chain are
C c o v ( x k , s k + 2 ) = { ∅ , k = 0 , { x k − 1 } , 1 ≤ k ≤ r − 2. \mathcal{C}_{\mathrm{cov}}(x_{k},s_{k+2})=
\begin{cases}
\varnothing, & k=0,\\
\{x_{k-1}\}, & 1\leq k\leq r-2.
\end{cases} C cov ( x k , s k + 2 ) = { ∅ , { x k − 1 } , k = 0 , 1 ≤ k ≤ r − 2.
(ii) For 0 ≤ t ≤ r − 2 0 \le t \le r - 2 0 ≤ t ≤ r − 2 , the cover candidates for the second chain are
C cov ( y t + 1 , s r − 1 − t ) = { { y t + 2 } , 0 ≤ t ≤ r − 3 , ∅ , t = r − 2. C_{\operatorname{cov}}(y_{t+1},s_{r-1-t}) =
\begin{cases}
\{y_{t+2}\}, & 0 \le t \le r - 3,\\
\varnothing, & t = r - 2.
\end{cases} C cov ( y t + 1 , s r − 1 − t ) = { { y t + 2 } , ∅ , 0 ≤ t ≤ r − 3 , t = r − 2.
Every candidate displayed here contributes with coefficient one.
Proof. For a step u → u s j > u u \to us_j > u u → u s j > u in either chain, a cover c = c η ⊲ u c = c_\eta\lhd u c = c η ⊲ u belongs to C cov ( u , s j ) C_{\operatorname{cov}}(u,s_j) C cov ( u , s j ) exactly when s j ∈ R I ( c ) s_j \in R_I(c) s j ∈ R I ( c ) : its coefficient μ I ( c , u ) \mu_I(c,u) μ I ( c , u ) is already one. By the descent test (43) ,
s j ∈ R I ( c η ) ⟺ η j η j + 1 ∈ { A A , B B , C C , C B , C A , B A } . (50) s_j \in R_I(c_\eta) \Longleftrightarrow\eta_j\eta_{j+1} \in\{\mathrm{AA},\mathrm{BB},\mathrm{CC},\mathrm{CB},\mathrm{CA},\mathrm{BA}\}. \tag*{(50)} s j ∈ R I ( c η ) ⟺ η j η j + 1 ∈ { AA , BB , CC , CB , CA , BA } . (50)
Thus equal letters and the pairs CB, CA, and BA give the required descent, while the pairs AB, AC, and BC do not. The proof of Lemma 4.12 shows that the chain elements lie below Θ r \Theta_r Θ r . We may therefore apply Lemma 4.13 to their covers. We treat its four cases for each chain and test the resulting consecutive letters.
(i) The step is x k → x k + 1 = x k s k + 2 x_k \to x_{k+1} = x_ks_{k+2} x k → x k + 1 = x k s k + 2 . The word of x k x_k x k is
A C k B C r − k A r − 1 . \mathrm{AC}^{k}\mathrm{BC}^{r-k}\mathrm{A}^{r-1}. AC k BC r − k A r − 1 .
Its letters at positions k + 2 k+2 k + 2 , k + 3 k+3 k + 3 are BC; the step interchanges them to CB. For a lower cover c η c_\eta c η to contribute to this multiplication, its letters at those same positions must satisfy (50) . There is only one B. In cases (i) and (ii) of Lemma 4.13 , it is exchanged with the C immediately to its left, when k ≥ 1 k \ge1 k ≥ 1 , or with the first A to its right, respectively. The only adjacent CA is at positions r + 2 r+2 r + 2 , r + 3 r+3 r + 3 , and case (iv) replaces the letters at positions 1, r + 2 r+2 r + 2 . Hence every possible cover word is listed below. The last two columns record its pair at k + 2 k+2 k + 2 , k + 3 k+3 k + 3 and whether it gives the required right descent.
Table 5.
Here k ≤ r − 2 k \le r - 2 k ≤ r − 2 gives r − k − 1 ≥ 1 r-k-1 \ge1 r − k − 1 ≥ 1 , including at the boundary. Only the first row satisfies the descent test, and its word is that of x k − 1 x_{k-1} x k − 1 . This element is indeed a cover:
x k − 1 = x k s k + 1 ⊲ x k ( k ≥ 1 ) , x_{k-1} = x_ks_{k+1} \lhd x_k \qquad(k \ge1), x k − 1 = x k s k + 1 ⊲ x k ( k ≥ 1 ) ,
since the adjacent interchange CB ↦ \mapsto ↦ BC decreases length by one, by Lemma 3.9 . When k = 0 k = 0 k = 0 , the first row is absent and there are no cover candidates. This proves (i).
(ii) Put b = r − 1 − t b = r - 1 - t b = r − 1 − t , so 1 ≤ b ≤ r − 1 1 \le b \le r - 1 1 ≤ b ≤ r − 1 . The step is y t + 1 → y t = y t + 1 s b y_{t+1} \to y_t = y_{t+1}s_b y t + 1 → y t = y t + 1 s b , and the word of y t + 1 y_{t+1} y t + 1 is
C b − 1 A C r − b B C A r − 1 . \mathrm{C}^{b-1}\mathrm{AC}^{r-b}\mathrm{BC}\mathrm{A}^{r-1}. C b − 1 AC r − b BC A r − 1 .
The letters at positions b b b , b + 1 b+1 b + 1 are AC; the step interchanges them to CA. A cover contributes precisely when its letters at b b b , b + 1 b+1 b + 1 satisfy (50) . In cases (i) and (ii) of Lemma 4.13 , the exchanged positions are r r r , r + 1 r+1 r + 1 and r + 1 r+1 r + 1 , r + 3 r+3 r + 3 , respectively. In case (iii), the adjacent CA pairs are at b − 1 b-1 b − 1 , b b b , when b ≥ 2 b \ge2 b ≥ 2 , and at r + 2 r+2 r + 2 , r + 3 r+3 r + 3 . Case (iv) replaces the letters at b b b , r + 2 r+2 r + 2 . The possible cover words and their pairs at b b b , b + 1 b+1 b + 1 are therefore as follows.
Table 6.
In the first row, the pair is AC when b ≤ r − 2 b \le r-2 b ≤ r − 2 and AB when b = r − 1 b=r-1 b = r − 1 ; neither gives a descent. The two rows from case (iii) correspond, in order, to the adjacent pairs at b − 1 , b b-1,b b − 1 , b and r + 2 , r + 3 r+2,r+3 r + 2 , r + 3 . Only the first of these rows passes the test. Its word is that of y t + 2 y_{t+2} y t + 2 , and
y t + 2 = y t + 1 s b − 1 ◃ y t + 1 ( b ≥ 2 ) , y_{t+2}=y_{t+1}s_{b-1}\triangleleft y_{t+1}\qquad(b\ge2), y t + 2 = y t + 1 s b − 1 ◃ y t + 1 ( b ≥ 2 ) ,
since the adjacent interchange CA ↦ \mapsto ↦ AC decreases length by one. This row exists exactly when b ≥ 2 b\ge2 b ≥ 2 , or equivalently t ≤ r − 3 t\le r-3 t ≤ r − 3 . When b = 1 b=1 b = 1 , all remaining pairs fail the test, so there are no cover candidates. This proves (ii). □
The multiplication formulas at this stage
Lemmas 4.12 and 4.15 locate every correction that can be nonzero in (4.1). The cover contributions are now completely determined, each with coefficient one. For the non-covers, however, we have only necessary restrictions on the indices: an element satisfying these restrictions may still have coefficient μ I \mu_I μ I equal to zero.
For k = 0 k=0 k = 0 , the first-chain formula (4.1) becomes
C x 0 C s 2 = C x 1 + μ I ( e , x 0 ) C e . (51) C_{x_0}C_{s_2}=C_{x_1}+\mu_I(e,x_0)C_e. \tag*{(51)} C x 0 C s 2 = C x 1 + μ I ( e , x 0 ) C e . (51)
For 1 ≤ k ≤ r − 2 1\le k\le r-2 1 ≤ k ≤ r − 2 , it becomes
C x k C s k + 2 = C x k + 1 + C x k − 1 + μ I ( e , x k ) C e + ∑ j = 1 k ( μ I ( γ j , x k ) C γ j + μ I ( κ j , x k ) C κ j ) . (52) C_{x_k}C_{s_{k+2}}=C_{x_{k+1}}+C_{x_{k-1}}+\mu_I(e,x_k)C_e+\sum_{j=1}^{k}\bigl(\mu_I(\gamma_j,x_k)C_{\gamma_j}+\mu_I(\kappa_j,x_k)C_{\kappa_j}\bigr). \tag*{(52)} C x k C s k + 2 = C x k + 1 + C x k − 1 + μ I ( e , x k ) C e + j = 1 ∑ k ( μ I ( γ j , x k ) C γ j + μ I ( κ j , x k ) C κ j ) . (52)
The coefficient-one term in the second formula is the cover contribution; all terms multiplied by μ I \mu_I μ I are possible non-cover contributions.
For the second chain, fix 0 ≤ t ≤ r − 2 0\le t\le r-2 0 ≤ t ≤ r − 2 , put b = r − 1 − t b=r-1-t b = r − 1 − t , and define
N t = { c ω : ω is admissible and satisfies (4.9) and (4.10), c ω < y t + 1 , ℓ ( y t + 1 ) − ℓ ( c ω ) > 1 } . (53) \mathcal{N}_t=\left\{c_\omega:
\begin{array}{l}
\omega\text{ is admissible and satisfies (4.9) and (4.10),}\\
c_\omega<y_{t+1},\quad\ell(y_{t+1})-\ell(c_\omega)>1
\end{array}
\right\}. \tag*{(53)} N t = { c ω : ω is admissible and satisfies (4.9) and (4.10), c ω < y t + 1 , ℓ ( y t + 1 ) − ℓ ( c ω ) > 1 } . (53)
The order within the first block gives s b ∈ R I ( c ) s_b\in R_I(c) s b ∈ R I ( c ) for every c ∈ N t c\in\mathcal{N}_t c ∈ N t , by (4.7). Thus (4.1) becomes
C y t + 1 C s r − 1 − t = C y t + { C y t + 2 , 0 ≤ t ≤ r − 3 , 0 , t = r − 2 , + ∑ c ∈ N t μ I ( c , y t + 1 ) C c . (54) C_{y_{t+1}}C_{s_{r-1-t}}=C_{y_t}+
\begin{cases}
C_{y_{t+2}}, & 0\le t\le r-3,\\
0, & t=r-2,
\end{cases}
+\sum_{c\in\mathcal{N}_t}\mu_I(c,y_{t+1})C_c. \tag*{(54)} C y t + 1 C s r − 1 − t = C y t + { C y t + 2 , 0 , 0 ≤ t ≤ r − 3 , t = r − 2 , + c ∈ N t ∑ μ I ( c , y t + 1 ) C c . (54)
Here C n c ( y t + 1 , s r − 1 − t ) ⊆ N t C_{nc}(y_{t+1},s_{r-1-t})\subseteq\mathcal{N}_t C n c ( y t + 1 , s r − 1 − t ) ⊆ N t , but membership in N t \mathcal{N}_t N t does not assert that the corresponding coefficient is nonzero.
Cancellation and reduction to smaller rank
We now use affine cancellation to express certain correction coefficients in terms of Kazhdan–Lusztig polynomials in smaller rank.
Notation. For r ≥ 2 r\ge2 r ≥ 2 , a superscript ( r ) (r) ( r ) indicates that the symbol is taken with rank parameter r r r , so that N = 2 r + 1 N=2r+1 N = 2 r + 1 . In particular, W ( r ) W^{(r)} W ( r ) is the affine Weyl group of type A ~ 2 r \widetilde{A}_{2r} A 2 r , realized by affine permutations of period 2 r + 1 2r+1 2 r + 1 , whose windows have 2 r + 1 2r+1 2 r + 1 entries. The elements x k ( r ) x_k^{(r)} x k ( r ) , γ j ( r ) \gamma_j^{(r)} γ j ( r ) , κ j ( r ) \kappa_j^{(r)} κ j ( r ) and Θ j ( r ) \Theta_j^{(r)} Θ j ( r ) are defined by the same formulas as before, with this value of N N N .
The same convention applies to the associated basis elements and coefficients. Thus
h x k ( r ) = [ M ( r ) ] C x k ( r ) , h_{x_k}^{(r)}=[M^{(r)}]C_{x_k}^{(r)}, h x k ( r ) = [ M ( r ) ] C x k ( r ) ,
and [ M γ j ( r ) ] C x k ( r ) [M_{\gamma_j}^{(r)}]C_{x_k}^{(r)} [ M γ j ( r ) ] C x k ( r ) is computed in the spherical module for W ( r ) W^{(r)} W ( r ) , with indices γ j ( r ) \gamma_j^{(r)} γ j ( r ) and x k ( r ) x_k^{(r)} x k ( r ) . In particular, h x k − j ( r − j ) h_{x_{k-j}}^{(r-j)} h x k − j ( r − j ) uses the index x k − j ( r − j ) x_{k-j}^{(r-j)} x k − j ( r − j ) in period 2 ( r − j ) + 1 2(r-j)+1 2 ( r − j ) + 1 . We omit the superscript when the rank parameter is clear.
Example 4.16. We illustrate the elements x 1 x_1 x 1 and γ 1 \gamma_1 γ 1 in rank parameters 2 and 3:
Table 4.16.
Definition 4.17. Let N ≥ 5 N \ge5 N ≥ 5 , and let w w w be an affine permutation of period N N N satisfying
w ( a ) = 2 N , w ( N ) = 1 − N for some 1 ≤ a < N . w(a) = 2N,\qquad w(N) = 1-N \qquad\text{for some } 1 \le a < N. w ( a ) = 2 N , w ( N ) = 1 − N for some 1 ≤ a < N . We define del a ( w ) \operatorname{del}_{a}(w) del a ( w ) by deleting the position classes a + N Z a+N\mathbb{Z} a + N Z and N + N Z N+N\mathbb{Z} N + N Z , together with their image classes (that are 1 + N Z 1+N\mathbb{Z} 1 + N Z and N + N Z N+N\mathbb{Z} N + N Z ), and relabelling the remaining positions and values increasingly. Explicitly, list the surviving window positions as
{ p 1 < ⋯ < p N − 2 } = { 1 , … , N } ∖ { a , N } , \{p_{1}<\cdots<p_{N-2}\}=\{1,\ldots,N\}\setminus\{a,N\}, { p 1 < ⋯ < p N − 2 } = { 1 , … , N } ∖ { a , N } , and write w ( p i ) = b i + q i N w(p_i)=b_i+q_iN w ( p i ) = b i + q i N , where 2 ≤ b i ≤ N − 1 2\le b_i\le N-1 2 ≤ b i ≤ N − 1 and q i ∈ Z q_i\in\mathbb{Z} q i ∈ Z . Set
del a ( w ) ( i ) = b i − 1 + q i ( N − 2 ) ( 1 ≤ i ≤ N − 2 ) , \operatorname{del}_{a}(w)(i)=b_i-1+q_i(N-2)\qquad(1\le i\le N-2), del a ( w ) ( i ) = b i − 1 + q i ( N − 2 ) ( 1 ≤ i ≤ N − 2 ) , and extend periodically with period N − 2 N-2 N − 2 . The removed residue classes have shifts + 1 +1 + 1 and − 1 -1 − 1 , so ∑ i q i = 0 \sum_iq_i=0 ∑ i q i = 0 . Thus del a ( w ) \operatorname{del}_{a}(w) del a ( w ) is an affine permutation of period N − 2 N-2 N − 2 , with window sum ( N − 2 ) ( N − 1 ) / 2 (N-2)(N-1)/2 ( N − 2 ) ( N − 1 ) /2 .
Proposition 4.18. Let N ≥ 5 N\ge5 N ≥ 5 , and let Z ≤ Y Z\le Y Z ≤ Y be affine permutations of period N N N . Suppose that
1 − N ≤ Z ( i ) , Y ( i ) ≤ 2 N ( 1 ≤ i ≤ N ) , 1-N\le Z(i),Y(i)\le2N\qquad(1\le i\le N), 1 − N ≤ Z ( i ) , Y ( i ) ≤ 2 N ( 1 ≤ i ≤ N ) , and that, for some 1 ≤ a < N 1\le a<N 1 ≤ a < N ,
Z ( a ) = Y ( a ) = 2 N , Z ( N ) = Y ( N ) = 1 − N . Z(a)=Y(a)=2N,\qquad Z(N)=Y(N)=1-N. Z ( a ) = Y ( a ) = 2 N , Z ( N ) = Y ( N ) = 1 − N . Then
del a ( Z ) ≤ del a ( Y ) , h Z , Y = h del a ( Z ) , del a ( Y ) . \operatorname{del}_{a}(Z)\le\operatorname{del}_{a}(Y),\qquad h_{Z,Y}=h_{\operatorname{del}_{a}(Z),\operatorname{del}_{a}(Y)}. del a ( Z ) ≤ del a ( Y ) , h Z , Y = h del a ( Z ) , del a ( Y ) . Proof. This follows by two applications of [24 Proposition 4.2] . □
Example 4.19. For γ 1 ( 3 ) \gamma_{1}^{(3)} γ 1 ( 3 ) and x 1 ( 3 ) x_{1}^{(3)} x 1 ( 3 ) , consider the ordinary endpoints
Z = w I ( 3 ) γ 1 ( 3 ) = [ 6 , 14 , 5 , 4 , 3 , 2 , − 6 ] , Z=w_{I}^{(3)}{}_{\gamma_{1}^{(3)}}=[6,14,5,4,3,2,-6], Z = w I ( 3 ) γ 1 ( 3 ) = [ 6 , 14 , 5 , 4 , 3 , 2 , − 6 ] , Y = w I ( 3 ) x 1 ( 3 ) = [ − 4 , 14 , 4 , 13 , 12 , − 5 , − 6 ] . Y=w_{I}^{(3)}{}_{x_{1}^{(3)}}=[-4,14,4,13,12,-5,-6]. Y = w I ( 3 ) x 1 ( 3 ) = [ − 4 , 14 , 4 , 13 , 12 , − 5 , − 6 ] . The rank criterion gives Z ≤ Y Z\le Y Z ≤ Y . Their windows lie in [ − 6 , 14 ] [-6,14] [ − 6 , 14 ] and share the extreme entries 14 14 14 and − 6 -6 − 6 at positions 2 2 2 and 7 7 7 . Thus Proposition 4.18 applies with N = 7 N=7 N = 7 and a = 2 a=2 a = 2 . The surviving positions are 1 , 3 , 4 , 5 , 6 1,3,4,5,6 1 , 3 , 4 , 5 , 6 , and Definition 4.17 gives
del 2 ( Z ) = [ 5 , 4 , 3 , 2 , 1 ] = w I ( 2 ) , \operatorname{del}_{2}(Z)=[5,4,3,2,1]=w_{I}^{(2)}, del 2 ( Z ) = [ 5 , 4 , 3 , 2 , 1 ] = w I ( 2 ) , del 2 ( Y ) = [ − 3 , 3 , 10 , 9 , − 4 ] = w I ( 2 ) x 0 ( 2 ) . \operatorname{del}_{2}(Y)=[-3,3,10,9,-4]=w_{I}^{(2)}{}_{x_{0}^{(2)}}. del 2 ( Y ) = [ − 3 , 3 , 10 , 9 , − 4 ] = w I ( 2 ) x 0 ( 2 ) . At the word level, deleting the second and last letters gives
B C B B B B A ⟶ B B B B B , A C B C C A A ⟶ A B C C A . \mathrm{BCBBBBA}\longrightarrow\mathrm{BBBBB},\qquad\mathrm{ACBCCAA}\longrightarrow\mathrm{ABCCA}. BCBBBBA ⟶ BBBBB , ACBCCAA ⟶ ABCCA . Consequently,
[ M γ 1 ( 3 ) ( 3 ) ] C x 1 ( 3 ) ( 3 ) = h Z , Y = h del 2 ( Z ) , del 2 ( Y ) = h x 0 ( 2 ) ( 2 ) . [M_{\gamma_{1}^{(3)}}^{(3)}]C_{x_{1}^{(3)}}^{(3)}=h_{Z,Y}=h_{\operatorname{del}_{2}(Z),\operatorname{del}_{2}(Y)}=h_{x_{0}^{(2)}}^{(2)}. [ M γ 1 ( 3 ) ( 3 ) ] C x 1 ( 3 ) ( 3 ) = h Z , Y = h del 2 ( Z ) , del 2 ( Y ) = h x 0 ( 2 ) ( 2 ) . Thus a coefficient indexed by γ 1 ( 3 ) \gamma_{1}^{(3)} γ 1 ( 3 ) in period 7 7 7 becomes an identity coefficient in period 5 5 5 , with upper element x 0 ( 2 ) x_{0}^{(2)} x 0 ( 2 ) . This is the case r = 3 r=3 r = 3 , j = k = 1 j=k=1 j = k = 1 of the first-chain reduction in Lemma 4.20 below.
We apply cancellation to common extreme entries. For w ∈ I W w\in IW w ∈ I W associated to an admissible word, the entries of w I w w_{I}w w I w lie in [ 1 − N , 2 N ] [1-N,2N] [ 1 − N , 2 N ] . The following reductions therefore require only two common positions at each stage.
Lemma 4.20. Assume r ≥ 3 r\ge3 r ≥ 3 . Then:
(i) For $1\le j\le k\le r-2,
[ M γ j ( r ) ] C x k ( r ) = h x k − j ( r − j ) . (55) [M_{\gamma_j}^{(r)}]C_{x_k}^{(r)}=h_{x_{k-j}}^{(r-j)}. \tag*{(55)} [ M γ j ( r ) ] C x k ( r ) = h x k − j ( r − j ) . (55) (ii) For 0 ≤ t ≤ r − 3 0\le t\le r-3 0 ≤ t ≤ r − 3 and $1\le j\le r-t-2,
[ M ϑ j ( r ) ] C y t + 1 ( r ) = h y t + 1 ( r − j ) . [M_{\vartheta_j}^{(r)}]C_{y_{t+1}}^{(r)}=h_{y_{t+1}}^{(r-j)}. [ M ϑ j ( r ) ] C y t + 1 ( r ) = h y t + 1 ( r − j ) . (iii) If 1 ≤ j ≤ k ≤ r − 2 1\le j\le k\le r-2 1 ≤ j ≤ k ≤ r − 2 and κ j < x k \kappa_j<x_k κ j < x k , put m = r − j + 1 m=r-j+1 m = r − j + 1 . Then
[ M κ j ( r ) ] C x k ( r ) = [ M κ 1 ( m ) ] C x k − j + 1 ( m ) [M_{\kappa_j}^{(r)}]C_{x_k}^{(r)}=[M_{\kappa_1}^{(m)}]C_{x_{k-j+1}}^{(m)} [ M κ j ( r ) ] C x k ( r ) = [ M κ 1 ( m ) ] C x k − j + 1 ( m )
Proof. We only prove (i); the other two identities are proved in a similar fashion. For u , v ∈ I W u,v\in I^W u , v ∈ I W , we have
[ M u ] C v = h w I u , w I v . [M_u]C_v=h_{w_Iu,w_Iv}. [ M u ] C v = h w I u , w I v .
Thus (55) is equivalent to
h w I ( r ) γ j ( r ) , w I ( r ) x k ( r ) = h w I ( r − j ) γ 0 ( r − j ) , w I ( r − j ) x k − j ( r − j ) . (56) h_{w_I^{(r)}\gamma_j^{(r)},w_I^{(r)}x_k^{(r)}}=h_{w_I^{(r-j)}\gamma_0^{(r-j)},w_I^{(r-j)}x_{k-j}^{(r-j)}}. \tag*{(56)} h w I ( r ) γ j ( r ) , w I ( r ) x k ( r ) = h w I ( r − j ) γ 0 ( r − j ) , w I ( r − j ) x k − j ( r − j ) . (56)
Put γ 0 = e \gamma_0=e γ 0 = e and set
Z = w I ( r ) γ j ( r ) , Y = w I ( r ) x k ( r ) . Z=w_I^{(r)}\gamma_j^{(r)},\qquad Y=w_I^{(r)}x_k^{(r)}. Z = w I ( r ) γ j ( r ) , Y = w I ( r ) x k ( r ) .
By Theorems 2.5 and 2.7 , the windows of x k ( r ) x_k^{(r)} x k ( r ) and γ j ( r ) \gamma_j^{(r)} γ j ( r ) both have N + 1 N+1 N + 1 at position 2 and 0 at position N N N . Since w I ( r ) ( N + 1 ) = 2 N w_I^{(r)}(N+1)=2N w I ( r ) ( N + 1 ) = 2 N and w I ( r ) ( 0 ) = 1 − N w_I^{(r)}(0)=1-N w I ( r ) ( 0 ) = 1 − N , we obtain
Z ( 2 ) = Y ( 2 ) = 2 N , Z ( N ) = Y ( N ) = 1 − N . Z(2)=Y(2)=2N,\qquad Z(N)=Y(N)=1-N. Z ( 2 ) = Y ( 2 ) = 2 N , Z ( N ) = Y ( N ) = 1 − N .
The lists in (26) also show that both endpoint windows lie in [ 1 − N , 2 N ] [1-N,2N] [ 1 − N , 2 N ] . The admissible words of the two minimal indices are
B C j B N − 2 j − 1 A j , A C k B C r − k A r − 1 . \mathrm{BC}^{j}\mathrm{B}^{N-2j-1}\mathrm{A}^{j},\qquad\mathrm{AC}^{k}\mathrm{BC}^{r-k}\mathrm{A}^{r-1}. BC j B N − 2 j − 1 A j , AC k BC r − k A r − 1 .
Deleting the letters in positions 2 and N N N gives the words of γ j − 1 ( r − 1 ) \gamma_{j-1}^{(r-1)} γ j − 1 ( r − 1 ) and x k − 1 ( r − 1 ) x_{k-1}^{(r-1)} x k − 1 ( r − 1 ) . Using the value relabelling in Definition 4.17 , we therefore obtain
Z ^ : = del 2 ( Z ) = w I ( r − 1 ) γ j − 1 ( r − 1 ) , \widehat{Z}:=\operatorname{del}_2(Z)=w_I^{(r-1)}\gamma_{j-1}^{(r-1)}, Z := del 2 ( Z ) = w I ( r − 1 ) γ j − 1 ( r − 1 ) ,
Y ^ : = del 2 ( Y ) = w I ( r − 1 ) x k − 1 ( r − 1 ) . \widehat{Y}:=\operatorname{del}_2(Y)=w_I^{(r-1)}x_{k-1}^{(r-1)}. Y := del 2 ( Y ) = w I ( r − 1 ) x k − 1 ( r − 1 ) .
The deleted periodic strands contribute equally to the rank functions of Z Z Z and Y Y Y , and the increasing relabellings in Definition 4.17 identify the remaining cutoffs. Hence Lemma 4.6 gives Z ≤ Y Z\leq Y Z ≤ Y if and only if Z ^ ≤ Y ^ \widehat{Z}\leq\widehat{Y} Z ≤ Y . Iterating this comparison j j j times reduces it to
w I ( r − j ) = w I ( r − j ) γ 0 ( r − j ) ≤ w I ( r − j ) x k − j ( r − j ) , w_I^{(r-j)}=w_I^{(r-j)}\gamma_0^{(r-j)}\leq w_I^{(r-j)}x_{k-j}^{(r-j)}, w I ( r − j ) = w I ( r − j ) γ 0 ( r − j ) ≤ w I ( r − j ) x k − j ( r − j ) ,
which follows from length additivity for the minimal representative x k − j ( r − j ) x_{k-j}^{(r-j)} x k − j ( r − j ) . The parameter bounds ensure that all intermediate indices are defined and r − j ≥ 2 r-j\geq2 r − j ≥ 2 .
We can now apply Proposition 4.18 to obtain
h Z , Y ( r ) = h Z ^ , Y ^ ( r − 1 ) . h_{Z,Y}^{(r)}=h_{\widehat{Z},\widehat{Y}}^{(r-1)}. h Z , Y ( r ) = h Z , Y ( r − 1 ) .
For j = 1 j=1 j = 1 , this proves (56) , since γ 0 = e \gamma_0=e γ 0 = e . For general j j j , replacing the pair ( Z , Y ) (Z,Y) ( Z , Y ) by ( Z ^ , Y ^ ) (\widehat{Z},\widehat{Y}) ( Z , Y ) and repeating the argument j − 1 j-1 j − 1 more times yields (56) . This proves (i). □
Proof of the first multiplication formulas
In this section we simultaneously prove the multiplication formulas in Theorem 2.8 and the coefficient estimates in Theorem 2.9 for the first chain x 0 = X 0 , x 1 , … , x r − 1 x_0=X_0,x_1,\ldots,x_{r-1} x 0 = X 0 , x 1 , … , x r − 1 from Theorem 2.5. Throughout this section, we let r ≥ 2 r\geq2 r ≥ 2 and put η = v + v − 1 \eta=v+v^{-1} η = v + v − 1 .
Recall the elements γ j \gamma_j γ j and κ j \kappa_j κ j from Theorems 2.7 and 4.10. To obtain Proposition 2.8 from the candidate expansions (51) and (52) , it remains to prove the following identities:
μ I ( e , x 0 ) = 1 , \mu_I(e,x_0)=1, μ I ( e , x 0 ) = 1 ,
μ I ( e , x k ) = 0 ( 1 ≤ k ≤ r − 2 ) , \mu_I(e,x_k)=0\qquad(1\leq k\leq r-2), μ I ( e , x k ) = 0 ( 1 ≤ k ≤ r − 2 ) ,
μ I ( κ j , x k ) = 0 ( 1 ≤ j ≤ k ≤ r − 2 ) , \mu_I(\kappa_j,x_k)=0\qquad(1\leq j\leq k\leq r-2), μ I ( κ j , x k ) = 0 ( 1 ≤ j ≤ k ≤ r − 2 ) ,
μ I ( γ j , x k ) = { 0 , j < k , 1 , j = k , ( 1 ≤ j ≤ k ≤ r − 2 ) . \mu_I(\gamma_j,x_k)=
\begin{cases}
0,&j<k,\\
1,&j=k,
\end{cases}
\qquad(1\leq j\leq k\leq r-2). μ I ( γ j , x k ) = { 0 , 1 , j < k , j = k , ( 1 ≤ j ≤ k ≤ r − 2 ) .
The first identity follows immediately from the initial estimate in Theorem 2.6 . We prove the third identity in Theorem 5.1, and then establish a recurrence for the numbers A a , b A_{a,b} A a , b in Theorem 5.2. With these two lemmas in hand, we prove Theorems 2.8 and 2.9 simultaneously, establishing the remaining identities along the way.
Lemma 5.1. For 1 ≤ j ≤ k ≤ r − 2 1\leq j\leq k\leq r-2 1 ≤ j ≤ k ≤ r − 2 ,
μ I ( κ j , x k ) = 0. \mu_I(\kappa_j,x_k)=0. μ I ( κ j , x k ) = 0.
Proof. We first prove that
ord v ( [ M κ 1 ] C x k ) ≥ N − k − 2 ( 1 ≤ k ≤ r − 1 ) . (57) \operatorname{ord}_{v}([M_{\kappa_1}]C_{x_k}) \ge N-k-2 \qquad(1 \le k \le r-1). \tag*{(57)} ord v ([ M κ 1 ] C x k ) ≥ N − k − 2 ( 1 ≤ k ≤ r − 1 ) . (57)
Combining Theorems 3.24 and 4.11, we get
[ M κ 1 s 2 ] C x 0 = O ( v N − 3 ) , [ M κ 1 ] C x 0 = 0. [M_{\kappa_1s_2}]C_{x_0}=O(v^{N-3}),\qquad[M_{\kappa_1}]C_{x_0}=0. [ M κ 1 s 2 ] C x 0 = O ( v N − 3 ) , [ M κ 1 ] C x 0 = 0.
Since κ 1 , κ 1 s 2 ∈ I W \kappa_1,\kappa_1s_2\in I W κ 1 , κ 1 s 2 ∈ I W , the standard multiplication rules give, for every F ∈ M F\in\mathcal{M} F ∈ M ,
[ M κ 1 ] ( F C s 2 ) = [ M κ 1 s 2 ] F + v − 1 [ M κ 1 ] F . [M_{\kappa_1}](F C_{s_2})=[M_{\kappa_1s_2}]F+v^{-1}[M_{\kappa_1}]F. [ M κ 1 ] ( F C s 2 ) = [ M κ 1 s 2 ] F + v − 1 [ M κ 1 ] F .
Applying this to (51) , and using [ M κ 1 ] C e = 0 [M_{\kappa_1}]C_e=0 [ M κ 1 ] C e = 0 , yields
[ M κ 1 ] C x 1 = [ M κ 1 s 2 ] C x 0 = O ( v N − 3 ) . [M_{\kappa_1}]C_{x_1}=[M_{\kappa_1s_2}]C_{x_0}=O(v^{N-3}). [ M κ 1 ] C x 1 = [ M κ 1 s 2 ] C x 0 = O ( v N − 3 ) .
For 1 ≤ k ≤ r − 2 1\le k\le r-2 1 ≤ k ≤ r − 2 , Theorem 4.11 gives κ 1 s k + 2 ∉ I W \kappa_1s_{k+2}\notin I W κ 1 s k + 2 ∈ / I W . Thus Theorem 4.4 implies
ord v ( [ M κ 1 ] C x k + 1 ) ≥ ord v ( [ M κ 1 ] C x k ) − 1. \operatorname{ord}_{v}([M_{\kappa_1}]C_{x_{k+1}})\ge\operatorname{ord}_{v}([M_{\kappa_1}]C_{x_k})-1. ord v ([ M κ 1 ] C x k + 1 ) ≥ ord v ([ M κ 1 ] C x k ) − 1.
Iterating proves the claimed bound (57) . Since N − k − 2 ≥ r ≥ 2 N-k-2\ge r\ge2 N − k − 2 ≥ r ≥ 2 , we get μ I ( κ 1 , x k ) = 0 \mu_I(\kappa_1,x_k)=0 μ I ( κ 1 , x k ) = 0 by Theorem 2.1.
For 2 ≤ j ≤ k ≤ r − 2 2\le j\le k\le r-2 2 ≤ j ≤ k ≤ r − 2 suppose that μ I ( κ j , x k ) ≠ 0 \mu_I(\kappa_j,x_k)\ne0 μ I ( κ j , x k ) = 0 , so that κ j < x k \kappa_j<x_k κ j < x k . Then part (iii) of Theorem 4.20 gives
[ M κ j ] C x k = [ M κ 1 ( m ) ] C x a ( m ) , m = r − j + 1 , a = k − j + 1. [M_{\kappa_j}]C_{x_k}=[M_{\kappa_1}^{(m)}]C_{x_a}^{(m)},\qquad m=r-j+1,\quad a=k-j+1. [ M κ j ] C x k = [ M κ 1 ( m ) ] C x a ( m ) , m = r − j + 1 , a = k − j + 1.
Here 1 ≤ a ≤ m − 2 1\le a\le m-2 1 ≤ a ≤ m − 2 and m ≥ 3 m\ge3 m ≥ 3 . Applying the bound (57) with rank parameter m m m gives
ord v ( [ M κ 1 ( m ) ] C x a ( m ) ) ≥ 2 m − a − 1 ≥ m + 1 ≥ 4. \operatorname{ord}_{v}([M_{\kappa_1}^{(m)}]C_{x_a}^{(m)})\ge2m-a-1\ge m+1\ge4. ord v ([ M κ 1 ( m ) ] C x a ( m ) ) ≥ 2 m − a − 1 ≥ m + 1 ≥ 4.
The coefficient of v v v therefore vanishes, contradicting μ I ( κ j , x k ) ≠ 0 \mu_I(\kappa_j,x_k)\ne0 μ I ( κ j , x k ) = 0 .
We now record the recurrence satisfied by the partition counts A a , b A_{a,b} A a , b defined in (6) .
Lemma 5.2. Write p ≤ a ( n ) p_{\le a}(n) p ≤ a ( n ) for the number of partitions of n n n with at most a a a parts. Then A 0 , b = p ( b ) A_{0,b}=p(b) A 0 , b = p ( b ) , A a , 0 = a + 1 A_{a,0}=a+1 A a , 0 = a + 1 and
A a + 1 , b = A a , b + A a , b + 1 − A a − 1 , b + 1 − p ≤ a ( b + 1 ) . (58) A_{a+1,b}=A_{a,b}+A_{a,b+1}-A_{a-1,b+1}-p_{\le a}(b+1). \tag*{(58)} A a + 1 , b = A a , b + A a , b + 1 − A a − 1 , b + 1 − p ≤ a ( b + 1 ) . (58)
Proof. The first two identities are immediate. The difference d a , b + 1 − d a + 1 , b d_{a,b+1}-d_{a+1,b} d a , b + 1 − d a + 1 , b counts partitions of a + b + 1 a+b+1 a + b + 1 with exactly a a a parts. Subtracting one from each part of these partitions and discarding zeros gives a bijection with the partitions of b + 1 b+1 b + 1 having at most a a a parts. Its inverse consists of padding the partition with zeros to obtain exactly a a a entries, then adding one to each entry. Hence
d a , b + 1 = d a + 1 , b + p ≤ a ( b + 1 ) . (59) d_{a,b+1}=d_{a+1,b}+p_{\le a}(b+1). \tag*{(59)} d a , b + 1 = d a + 1 , b + p ≤ a ( b + 1 ) . (59)
On the other hand, d a , b = A a , b − A a − 1 , b d_{a,b}=A_{a,b}-A_{a-1,b} d a , b = A a , b − A a − 1 , b by definition. By replacing this in (59) and rearranging the terms we obtain (58) .
We are now in position to complete the proof of Theorem 2.8 and Theorem 2.9 .
Proof of Proposition 2.8 and Lemma 2.9. Fix r ≥ 2 r\ge2 r ≥ 2 . We prove the multiplication formulas and the triangular estimate (7) simultaneously, by induction on k k k . The scalar calculations use the initial estimate in Theorem 2.6 , the partition recurrence in Theorem 5.2, and the formula for h γ j h_{\gamma_j} h γ j in Theorem 3.25.
Denote the multiplication identity at step k k k by ( M k ) (M_k) ( M k ) and the approximation to h x k h_{x_k} h x k by ( N k ) (N_k) ( N k ) :
C x k C s k + 2 = { C x k + 1 + C e , k = 0 , C x k + 1 + C x k − 1 + C γ k , k > 0 , ( 0 ≤ k ≤ r − 2 ) . (60) C_{x_k}C_{s_{k+2}}=
\begin{cases}
C_{x_{k+1}}+C_e, & k=0,\\
C_{x_{k+1}}+C_{x_{k-1}}+C_{\gamma_k}, & k>0,
\end{cases}
\qquad(0\le k\le r-2). \tag*{(60)} C x k C s k + 2 = { C x k + 1 + C e , C x k + 1 + C x k − 1 + C γ k , k = 0 , k > 0 , ( 0 ≤ k ≤ r − 2 ) . (60)
h x k = ∑ b = 0 r − 1 − k A k , b v k + 1 + 2 b + O ( v 2 r + 1 − k ) ( 0 ≤ k ≤ r − 1 ) . (61) h_{x_k}=\sum_{b=0}^{r-1-k}A_{k,b}v^{k+1+2b}+O(v^{2r+1-k})
\qquad(0\le k\le r-1). \tag*{(61)} h x k = b = 0 ∑ r − 1 − k A k , b v k + 1 + 2 b + O ( v 2 r + 1 − k ) ( 0 ≤ k ≤ r − 1 ) . (61)
Starting from ( N 0 ) (N_0) ( N 0 ) we prove ( M 0 ) (M_0) ( M 0 ) and then ( N 1 ) (N_1) ( N 1 ) . At each subsequent step, ( N k − 1 ) (N_{k-1}) ( N k − 1 ) and ( N k ) (N_k) ( N k ) will give ( M k ) (M_k) ( M k ) ; applying ε \varepsilon ε to that multiplication identity will then give ( N k + 1 ) (N_{k+1}) ( N k + 1 ) .
Theorem 2.6 and the identity A 0 , b = p ( b ) A_{0,b}=p(b) A 0 , b = p ( b ) give ( N 0 ) (N_0) ( N 0 ) . This implies that [ v ] h x 0 = 1 [v]h_{x_0}=1 [ v ] h x 0 = 1 , which in turn gives ( M 0 ) (M_0) ( M 0 ) by formula (51) . Applying ε \varepsilon ε and Theorem 2.2 gives
h x 1 = η h x 0 − 1 , η = v + v − 1 . h_{x_1}=\eta h_{x_0}-1,\qquad\eta=v+v^{-1}. h x 1 = η h x 0 − 1 , η = v + v − 1 .
To prove (N1 _1 1 ), we substitute (N0 _0 0 ) into this scalar recurrence, obtaining
h x 1 = ∑ d = 0 r − 1 p ( d ) v 2 d + 2 + ∑ d = 0 r − 1 p ( d ) v 2 d − 1 + O ( v 2 r ) . h_{x_1}=\sum_{d=0}^{r-1}p(d)v^{2d+2}+\sum_{d=0}^{r-1}p(d)v^{2d}-1+O(v^{2r}). h x 1 = d = 0 ∑ r − 1 p ( d ) v 2 d + 2 + d = 0 ∑ r − 1 p ( d ) v 2 d − 1 + O ( v 2 r ) .
The constant term p ( 0 ) = 1 p(0)=1 p ( 0 ) = 1 cancels with the − 1 -1 − 1 . For 0 ≤ b ≤ r − 2 0\le b\le r-2 0 ≤ b ≤ r − 2 , the coefficient of v 2 + 2 b v^{2+2b} v 2 + 2 b receives p ( b ) p(b) p ( b ) from the term with d = b d=b d = b in the first sum and p ( b + 1 ) p(b+1) p ( b + 1 ) from the term with d = b + 1 d=b+1 d = b + 1 in the second sum. The error term does not contribute, since 2 + 2 b < 2 r 2+2b<2r 2 + 2 b < 2 r . By definition, d 0 , b = p ( b ) d_{0,b}=p(b) d 0 , b = p ( b ) and d 1 , b = p ( b + 1 ) d_{1,b}=p(b+1) d 1 , b = p ( b + 1 ) . Hence,
[ v 2 + 2 b ] h x 1 = p ( b ) + p ( b + 1 ) = d 0 , b + d 1 , b = A 1 , b . [v^{2+2b}]h_{x_1}=p(b)+p(b+1)=d_{0,b}+d_{1,b}=A_{1,b}. [ v 2 + 2 b ] h x 1 = p ( b ) + p ( b + 1 ) = d 0 , b + d 1 , b = A 1 , b .
Absorbing the last term p ( r − 1 ) v 2 r p(r-1)v^{2r} p ( r − 1 ) v 2 r of the first sum into the error O ( v 2 r ) O(v^{2r}) O ( v 2 r ) gives (N1 _1 1 ):
h x 1 = ∑ b = 0 r − 2 A 1 , b v 2 + 2 b + O ( v 2 r ) . h_{x_1}=\sum_{b=0}^{r-2}A_{1,b}v^{2+2b}+O(v^{2r}). h x 1 = b = 0 ∑ r − 2 A 1 , b v 2 + 2 b + O ( v 2 r ) .
Now let 1 ≤ k ≤ r − 2 1\le k\le r-2 1 ≤ k ≤ r − 2 and assume (Nk − 1 _{k-1} k − 1 ) and (Nk _k k ). We first prove (Mk _k k ). We have
h x k − 1 = k v k + O ( v k + 2 ) , h x k = ( k + 1 ) v k + 1 + O ( v k + 3 ) . h_{x_{k-1}}=kv^k+O(v^{k+2}),\qquad h_{x_k}=(k+1)v^{k+1}+O(v^{k+3}). h x k − 1 = k v k + O ( v k + 2 ) , h x k = ( k + 1 ) v k + 1 + O ( v k + 3 ) .
Applying ε \varepsilon ε to (52) and using Theorem 5.1 gives
η h x k − h x k − 1 = h x k + 1 + μ I ( e , x k ) + ∑ j = 1 k μ I ( γ j , x k ) h γ j . \eta h_{x_k}-h_{x_{k-1}}=h_{x_{k+1}}+\mu_I(e,x_k)+\sum_{j=1}^{k}\mu_I(\gamma_j,x_k)h_{\gamma_j}. η h x k − h x k − 1 = h x k + 1 + μ I ( e , x k ) + j = 1 ∑ k μ I ( γ j , x k ) h γ j .
The left-hand side has no terms below degree k k k . Every summand on the right is coefficient-wise nonnegative by KL positivity, and h γ j h_{\gamma_j} h γ j begins with v j v^j v j by Theorem 3.25. Consequently
μ I ( e , x k ) = 0 , μ I ( γ j , x k ) = 0 ( 1 ≤ j < k ) . \mu_I(e,x_k)=0,\qquad\mu_I(\gamma_j,x_k)=0\quad(1\le j<k). μ I ( e , x k ) = 0 , μ I ( γ j , x k ) = 0 ( 1 ≤ j < k ) .
For j = k j=k j = k , part (i) of Theorem 4.20 combined with Theorem 2.6 , Theorem 2.1 and r − k ≥ 2 r-k\ge2 r − k ≥ 2 , yields
μ I ( γ k , x k ) = [ v ] ( [ M γ k ] C x k ) = [ v ] h x 0 ( r − k ) = 1. \mu_I(\gamma_k,x_k)=[v]([M_{\gamma_k}]C_{x_k})=[v]h_{x_0}^{(r-k)}=1. μ I ( γ k , x k ) = [ v ] ([ M γ k ] C x k ) = [ v ] h x 0 ( r − k ) = 1.
Substituting these μ I \mu_I μ I -values into (52) gives
C x k C s k + 2 = C x k + 1 + C x k − 1 + C γ k , C_{x_k}C_{s_{k+2}}=C_{x_{k+1}}+C_{x_{k-1}}+C_{\gamma_k}, C x k C s k + 2 = C x k + 1 + C x k − 1 + C γ k ,
which proves (Mk _k k ). Applying ε \varepsilon ε and Theorem 2.2 to this identity gives
h x k + 1 = η h x k − h x k − 1 − h γ k . (62) h_{x_{k+1}}=\eta h_{x_k}-h_{x_{k-1}}-h_{\gamma_k}. \tag*{(62)} h x k + 1 = η h x k − h x k − 1 − h γ k . (62)
We now prove (Nk + 1 _{k+1} k + 1 ) by computing (62) modulo v 2 r − k v^{2r-k} v 2 r − k , the error order required in (Nk + 1 _{k+1} k + 1 ). The partition interpretation in [38 Proposition 1.7.3] implies
[ q n ] ( [ N − 1 k ] q ) = p ≤ k ( n ) ( 0 ≤ n ≤ N − 1 − k ) . [q^n]\left(\begin{bmatrix}N-1\\k\end{bmatrix}_q\right)=p_{\le k}(n)\qquad(0\le n\le N-1-k). [ q n ] ( [ N − 1 k ] q ) = p ≤ k ( n ) ( 0 ≤ n ≤ N − 1 − k ) .
Since q = v 2 q=v^2 q = v 2 , Theorem 3.25 then shows that
h γ k = ∑ n = 0 r − 1 − k p ≤ k ( n ) v 2 n + k + O ( v 2 r − k ) . h_{\gamma_k}=\sum_{n=0}^{r-1-k}p_{\le k}(n)v^{2n+k}+O(v^{2r-k}). h γ k = n = 0 ∑ r − 1 − k p ≤ k ( n ) v 2 n + k + O ( v 2 r − k ) .
Substituting these expansions into (62) , we compute the remaining coefficients. By (Nk _k k ) and (Nk − 1 _{k-1} k − 1 ), the error terms of η h x k \eta h_{x_k} η h x k and h x k − 1 h_{x_{k-1}} h x k − 1 have orders at least 2 r − k 2r-k 2 r − k and 2 r + 2 − k 2r+2-k 2 r + 2 − k , respectively. Since no terms of degree less than k k k occur, it remains to compute [ v i ] h x k + 1 [v^i]h_{x_{k+1}} [ v i ] h x k + 1 for k ≤ i < 2 r − k k\le i<2r-k k ≤ i < 2 r − k . Moreover, these coefficients vanish unless i i i has parity k k k , by Theorem 2.1 together with Theorem 2.4 and Theorem 3.14. The coefficient in degree k k k vanishes:
[ v k ] h x k + 1 = A k , 0 − A k − 1 , 0 − p ≤ k ( 0 ) = ( k + 1 ) − k − 1 = 0. [v^k]h_{x_{k+1}}=A_{k,0}-A_{k-1,0}-p_{\le k}(0)=(k+1)-k-1=0. [ v k ] h x k + 1 = A k , 0 − A k − 1 , 0 − p ≤ k ( 0 ) = ( k + 1 ) − k − 1 = 0.
For 0 ≤ b ≤ r − k − 2 0\le b\le r-k-2 0 ≤ b ≤ r − k − 2 , we obtain
[ v k + 2 + 2 b ] h x k + 1 = A k , b + A k , b + 1 − A k − 1 , b + 1 − p ≤ k ( b + 1 ) = A k + 1 , b , [v^{k+2+2b}]h_{x_{k+1}}=A_{k,b}+A_{k,b+1}-A_{k-1,b+1}-p_{\le k}(b+1)=A_{k+1,b}, [ v k + 2 + 2 b ] h x k + 1 = A k , b + A k , b + 1 − A k − 1 , b + 1 − p ≤ k ( b + 1 ) = A k + 1 , b ,
where the last equality follows from Theorem 5.2. This proves (Nk + 1 _{k+1} k + 1 ) and completes the induction.
We have proved Proposition 2.8 and equation (7) of Theorem 2.9 . Now let 0 ≤ a ≤ r − 1 0 \le a \le r - 1 0 ≤ a ≤ r − 1 . Since A a , 0 = a + 1 A_{a,0} = a + 1 A a , 0 = a + 1 and a + 3 ≤ 2 r + 1 − a a + 3 \le2r + 1 - a a + 3 ≤ 2 r + 1 − a , (Na _a a ) gives
h x a = ( a + 1 ) v a + 1 + O ( v a + 3 ) . h_{x_a} = (a + 1)v^{a+1} + O(v^{a+3}). h x a = ( a + 1 ) v a + 1 + O ( v a + 3 ) .
In particular, h x r − 1 = r v r + O ( v r + 2 ) h_{x_{r-1}} = rv^r + O(v^{r+2}) h x r − 1 = r v r + O ( v r + 2 ) , completing the proof of Theorem 2.9 .
Proof of the second multiplication formulas
In this section we prove the multiplication formulas in Theorem 2.11 for the second chain. Throughout, r ≥ 2 r \ge2 r ≥ 2 . Comparing Proposition 2.11 with the candidate expansion (54) , it suffices to prove the following order estimate:
ord v ( ∑ z ∈ N t μ I ( z , y t + 1 ) h z ) ≥ t + 3 ( 0 ≤ t ≤ r − 2 ) . \operatorname{ord}_v\left(\sum_{z \in\mathcal{N}_t}\mu_I(z,y_{t+1})h_z\right) \ge t + 3 \qquad(0 \le t \le r - 2). ord v ( z ∈ N t ∑ μ I ( z , y t + 1 ) h z ) ≥ t + 3 ( 0 ≤ t ≤ r − 2 ) .
Fix 0 ≤ t ≤ r − 2 0 \le t \le r - 2 0 ≤ t ≤ r − 2 and put b = r − 1 − t b = r - 1 - t b = r − 1 − t . Recall from (53) that z ∈ N t z \in\mathcal{N}_t z ∈ N t if and only if z = c ω z = c_\omega z = c ω for an admissible word ω \omega ω whose first r + 1 r + 1 r + 1 and last r r r positions are separately ordered by C < B < A C < B < A C < B < A , with ω b ≠ C \omega_b \ne C ω b = C , z < y t + 1 z < y_{t+1} z < y t + 1 , and ℓ ( y t + 1 ) − ℓ ( z ) > 1 \ell(y_{t+1}) - \ell(z) > 1 ℓ ( y t + 1 ) − ℓ ( z ) > 1 .
Given c ω ∈ N t c_\omega\in\mathcal{N}_t c ω ∈ N t with ω \omega ω of type j j j , let c c c and a a a be the numbers of C’s and A’s in the first block, respectively. We have
ω = C c B r + 1 − c − a A a ∣ C j − c B r − 2 j + c + a A j − a , \omega= C^cB^{r+1-c-a}A^a \mid C^{j-c}B^{r-2j+c+a}A^{j-a}, ω = C c B r + 1 − c − a A a ∣ C j − c B r − 2 j + c + a A j − a ,
where the bar separates positions r + 1 r + 1 r + 1 and r + 2 r + 2 r + 2 . All exponents are nonnegative. Since each block is ordered, inversions can only occur between the two blocks. We distinguish the following cases:
(i) Ordered case : a = 0 a = 0 a = 0 and c = j c = j c = j . Here, c ω = Θ j c_\omega= \Theta_j c ω = Θ j .
(ii) Pure BC case : a = 0 a = 0 a = 0 and c < j c < j c < j .
(iii) Pure AB case : a > 0 a > 0 a > 0 and c = j c = j c = j .
(iv) Mixed AC case : a > 0 a > 0 a > 0 and c < j c < j c < j .
The labels BC, AB, and AC indicate the pair of letters adjacent to the bar. The terminology refers to inversions for C < B < A C < B < A C < B < A : in the ordered case there are no inversions; in each pure case, all inversions have the same type, BC or AB, respectively; and in the mixed case, AC inversions occur together with at least one of the other two types, since N − 2 j ≥ 1 N - 2j \ge1 N − 2 j ≥ 1 . In all four cases, the condition ω b ≠ C \omega_b \ne C ω b = C is equivalent to c ≤ b − 1 = r − t − 2 c \le b - 1 = r - t - 2 c ≤ b − 1 = r − t − 2 .
For z ∈ N t z \in\mathcal{N}_t z ∈ N t , the coefficients μ I ( z , y t + 1 ) \mu_I(z,y_{t+1}) μ I ( z , y t + 1 ) may still vanish. We prove that they do vanish in the ordered and mixed cases in Theorems 6.1 and 6.5, respectively. In the pure cases, we prove that ord v ( h z ) ≥ t + 3 \operatorname{ord}_v(h_z) \ge t + 3 ord v ( h z ) ≥ t + 3 in Theorem 6.2.
Lemma 6.1. Let z ∈ N t z \in\mathcal{N}_t z ∈ N t belong to the ordered case. Then μ I ( z , y t + 1 ) = 0 \mu_I(z,y_{t+1}) = 0 μ I ( z , y t + 1 ) = 0 .
Proof. Let 0 ≤ s ≤ r − 1 0 \le s \le r - 1 0 ≤ s ≤ r − 1 . By Theorem 2.9 , we have ord v ( h x r − 1 ) = r \operatorname{ord}_v(h_{x_{r-1}}) = r ord v ( h x r − 1 ) = r . The factorization y s = x r − 1 ( s 1 ⋯ s r − 1 − s ) y_s = x_{r-1}(s_1 \cdots s_{r-1-s}) y s = x r − 1 ( s 1 ⋯ s r − 1 − s ) from (8) gives a path from x r − 1 x_{r-1} x r − 1 to y s y_s y s with r − 1 − s r - 1 - s r − 1 − s finite ascents, all of whose elements lie in 1 W {}^{1}W 1 W by Theorem 2.4. Theorem 4.5 therefore gives
ord v ( h y s ) ≥ r − ( r − 1 − s ) = s + 1. (63) \operatorname{ord}_v(h_{y_s}) \ge r - (r - 1 - s) = s + 1. \tag*{(63)} ord v ( h y s ) ≥ r − ( r − 1 − s ) = s + 1. (63) Since z z z belongs to the ordered case, we have z = Θ j z = \Theta_j z = Θ j with $0 \le j \le b - 1 = r - t - 2$. If j = 0 j = 0 j = 0 , then z = e z = e z = e , and Theorem 2.1 together with equation (63) give
μ I ( e , y t + 1 ) = [ v ] h y t + 1 = 0. \mu_I(e,y_{t+1}) = [v]h_{y_{t+1}} = 0. μ I ( e , y t + 1 ) = [ v ] h y t + 1 = 0. Suppose now that j ≥ 1 j \ge1 j ≥ 1 . Then part (ii) of Theorem 4.20 yields
μ I ( Θ j , y t + 1 ) = [ v ] ( [ M Θ j ] C y t + 1 ) = [ v ] h y t + 1 ( r − j ) . \mu_I(\Theta_j,y_{t+1}) = [v]\left([M_{\Theta_j}]C_{y_{t+1}}\right) = [v]h_{y_{t+1}^{(r-j)}}. μ I ( Θ j , y t + 1 ) = [ v ] ( [ M Θ j ] C y t + 1 ) = [ v ] h y t + 1 ( r − j ) . Since r − j ≥ t + 2 r - j \ge t + 2 r − j ≥ t + 2 , the element y t + 1 ( r − j ) y_{t+1}^{(r-j)} y t + 1 ( r − j ) is well-defined, and thus ord v ( h y t + 1 ( r − j ) ) ≥ 2 \operatorname{ord}_v(h_{y_{t+1}^{(r-j)}}) \ge2 ord v ( h y t + 1 ( r − j ) ) ≥ 2 , by equation (63) . Its coefficient of v v v therefore vanishes, proving μ I ( z , y t + 1 ) = 0 \mu_I(z,y_{t+1}) = 0 μ I ( z , y t + 1 ) = 0 . □
We now treat the pure cases.
Lemma 6.2. Let z ∈ N t z \in\mathcal{N}_t z ∈ N t belong to one of the pure cases. Then ord v ( h z ) ≥ t + 3 \operatorname{ord}_v(h_z) \ge t + 3 ord v ( h z ) ≥ t + 3 .
Proof. Let z = c ω z = c_{\omega} z = c ω with ω \omega ω of type j j j , and put n = N − 2 j n = N - 2j n = N − 2 j . In either pure case, 1 ≤ j ≤ r 1 \le j \le r 1 ≤ j ≤ r . Our strategy is to find a length-additive factorization
z = d s i 1 ⋯ s i L , s i k ∈ I , z = d s_{i_1} \cdots s_{i_L}, \qquad s_{i_k} \in I, z = d s i 1 ⋯ s i L , s i k ∈ I ,
with d ∈ I W d \in{}^{I}W d ∈ I W for which both h d h_d h d and L L L can be computed explicitly. Since z ∈ I W z \in{}^{I}W z ∈ I W , all intermediate elements also lie in I W {}^{I}W I W , and Theorem 4.5 will then give
ord v ( h z ) ≥ ord v ( h d ) − L . (64) \operatorname{ord}_{v}(h_z) \ge\operatorname{ord}_{v}(h_d) - L. \tag*{(64)} ord v ( h z ) ≥ ord v ( h d ) − L . (64)
We first construct the starting elements. For 1 ≤ s < N 1 \le s < N 1 ≤ s < N , let
d s = w I ∩ P s w P s , P s = S ∖ { s s } . d_s = w_{I \cap P_s}w_{P_s}, \qquad P_s = S \setminus\{s_s\}. d s = w I ∩ P s w P s , P s = S ∖ { s s } .
This is the unique longest element of I ∩ P s W P s {}^{I \cap P_s}W_{P_s} I ∩ P s W P s . We have ℓ ( d s ) = s ( N − s ) \ell(d_s) = s(N-s) ℓ ( d s ) = s ( N − s ) , since the factorization w P s = w I ∩ P s d s w_{P_s} = w_{I \cap P_s}d_s w P s = w I ∩ P s d s is length-additive. By equation (31) , we have the expansion
B S ( P s ) = M d s + ∑ u ∈ I ∩ P s W P s u ≠ d s v s ( N − s ) − ℓ ( u ) M u , \mathcal{B}S(P_s) = M_{d_s} + \sum_{\substack{u \in{}^{I \cap P_s}W_{P_s} \\ u \ne d_s}} v^{s(N-s)-\ell(u)}M_u, B S ( P s ) = M d s + u ∈ I ∩ P s W P s u = d s ∑ v s ( N − s ) − ℓ ( u ) M u ,
where every exponent in the sum is positive. Theorem 3.3 gives bar-invariance, so the canonical basis criterion in Lemma 3.1 yields
B S ( P s ) = C d s , h d s = [ M e ] B S ( P s ) = v s ( N − s ) . \mathcal{B}S(P_s) = C_{d_s}, \qquad h_{d_s} = [M_e]\mathcal{B}S(P_s) = v^{s(N-s)}. B S ( P s ) = C d s , h d s = [ M e ] B S ( P s ) = v s ( N − s ) .
Now consider the words
B n A j C j , A j C j B n . B^nA^jC^j, \qquad A^jC^jB^n. B n A j C j , A j C j B n .
Both are admissible of type j j j , and have inversion number j ( N − j ) j(N-j) j ( N − j ) for the order C < B < A C < B < A C < B < A . By Theorem 3.19, the corresponding elements belong to the supports of B S ( P N − j ) \mathcal{B}S(P_{N-j}) B S ( P N − j ) and B S ( P j ) \mathcal{B}S(P_j) B S ( P j ) , respectively. By Theorem 3.9, both have length j ( N − j ) j(N-j) j ( N − j ) , so they are the unique longest elements d N − j d_{N-j} d N − j and d j d_j d j of these supports.
We take d = d N − j d = d_{N-j} d = d N − j in the pure BC case and d = d j d = d_j d = d j in the pure AB case. In either case, h d = v j ( N − j ) h_d = v^{j(N-j)} h d = v j ( N − j ) . It remains to construct the path from d d d to z z z and determine its length L L L . Each adjacent interchange
A B ⟼ B A , A C ⟼ C A , B C ⟼ C B AB \longmapsto BA, \qquad AC \longmapsto CA, \qquad BC \longmapsto CB A B ⟼ B A , A C ⟼ C A , B C ⟼ C B
reduces the inversion number by one. By Theorems 3.7, 3.9 and 4.9, it corresponds to right multiplication by a finite simple reflection and increases length by one. We construct the path using these interchanges. Recall that c ≤ b − 1 = r − t − 2 c \le b - 1 = r - t - 2 c ≤ b − 1 = r − t − 2 .
Pure BC case. Here a = 0 a = 0 a = 0 and c < j c < j c < j . We have
B n A j C j ⟼ B n C j A j ⟼ B r + 1 − c C j B r − 2 j + c A j ⟼ C c B r + 1 − c ∣ C j − c B r − 2 j + c A j , B^nA^jC^j \longmapsto B^nC^jA^j \longmapsto B^{r+1-c}C^jB^{r-2j+c}A^j \longmapsto C^cB^{r+1-c}\mid C^{j-c}B^{r-2j+c}A^j, B n A j C j ⟼ B n C j A j ⟼ B r + 1 − c C j B r − 2 j + c A j ⟼ C c B r + 1 − c ∣ C j − c B r − 2 j + c A j ,
where the first and last words correspond to d = d N − j d = d_{N-j} d = d N − j and z z z , respectively. Each arrow is a succession of adjacent interchanges of the types listed above, so we get a length-additive factorization z = d s i 1 ⋯ s i L z = d s_{i_1} \cdots s_{i_L} z = d s i 1 ⋯ s i L with s i k ∈ I s_{i_k} \in I s i k ∈ I and
L = j 2 + j ( r − 2 j + c ) + c ( r + 1 − c ) = j ( N − j ) − ( r + 1 − c ) ( j − c ) . L = j^2 + j(r - 2j + c) + c(r + 1 - c) = j(N-j) - (r + 1 - c)(j-c). L = j 2 + j ( r − 2 j + c ) + c ( r + 1 − c ) = j ( N − j ) − ( r + 1 − c ) ( j − c ) .
In particular,
j ( N − j ) − L ≥ r + 1 − c ≥ t + 3. j(N-j) - L \ge r + 1 - c \ge t + 3. j ( N − j ) − L ≥ r + 1 − c ≥ t + 3.
Pure AB case. Here a > 0 a > 0 a > 0 and c = j c = j c = j . Similarly, we have
A j C j B n ⟼ C j A j B n ⟼ C j B r + 1 − j − a A j B r − j + a ⟼ C j B r + 1 − j − a A a ∣ B r − j + a A j − a , A^jC^jB^n \longmapsto C^jA^jB^n \longmapsto C^jB^{r+1-j-a}A^jB^{r-j+a} \longmapsto C^jB^{r+1-j-a}A^a\mid B^{r-j+a}A^{j-a}, A j C j B n ⟼ C j A j B n ⟼ C j B r + 1 − j − a A j B r − j + a ⟼ C j B r + 1 − j − a A a ∣ B r − j + a A j − a ,
where the first and last words correspond to d = d j d = d_j d = d j and z z z , respectively. We thus get a length-additive factorization z = d s i 1 ⋯ s i L z = d s_{i_1} \cdots s_{i_L} z = d s i 1 ⋯ s i L with s i k ∈ I s_{i_k} \in I s i k ∈ I and
L = j 2 + j ( r + 1 − j − a ) + ( j − a ) ( r − j + a ) = j ( N − j ) − a ( r − j + a ) . L = j^2 + j(r + 1 - j - a) + (j-a)(r-j+a) = j(N-j) - a(r-j+a). L = j 2 + j ( r + 1 − j − a ) + ( j − a ) ( r − j + a ) = j ( N − j ) − a ( r − j + a ) .
In particular,
j ( N − j ) − L ≥ r − j + a ≥ r − j + 1 ≥ t + 3. j(N-j) - L \ge r-j+a \ge r-j+1 \ge t+3. j ( N − j ) − L ≥ r − j + a ≥ r − j + 1 ≥ t + 3.
Finally, combining h d = v j ( N − j ) h_d = v^{j(N-j)} h d = v j ( N − j ) and equation (64) , we get
ord v ( h z ) ≥ j ( N − j ) − L ≥ t + 3 , \operatorname{ord}_{v}(h_z) \ge j(N-j) - L \ge t+3, ord v ( h z ) ≥ j ( N − j ) − L ≥ t + 3 ,
as required.
To treat the mixed case, we recall right star operations and the invariance property needed below.
Definition 6.3. For distinct simple reflections s , t s,t s , t with s t s = t s t sts=tst s t s = t s t , let
D R ( s , t ) = { w ∈ W : ∣ R ( w ) ∩ { s , t } ∣ = 1 } . D_R(s,t)=\{w\in W:|R(w)\cap\{s,t\}|=1\}. D R ( s , t ) = { w ∈ W : ∣ R ( w ) ∩ { s , t } ∣ = 1 } . For w ∈ D R ( s , t ) w\in D_R(s,t) w ∈ D R ( s , t ) , exactly one of w s ws w s , w t wt w t belongs to D R ( s , t ) D_R(s,t) D R ( s , t ) ; that element is denoted by w ∗ w^* w ∗ . Equivalently, in the right parabolic factor from W { s , t } W_{\{s,t\}} W { s , t } this involution exchanges s ↔ s t s\leftrightarrow st s ↔ s t and t ↔ t s t\leftrightarrow ts t ↔ t s .
Proposition 6.4. Let Z , Y ∈ D R ( s , t ) Z,Y\in D_R(s,t) Z , Y ∈ D R ( s , t ) be such that Z < Y Z<Y Z < Y in the Bruhat order. Suppose that ℓ ( Z ∗ ) = ℓ ( Z ) + 1 \ell(Z^*)=\ell(Z)+1 ℓ ( Z ∗ ) = ℓ ( Z ) + 1 , ℓ ( Y ∗ ) = ℓ ( Y ) − 1 \ell(Y^*)=\ell(Y)-1 ℓ ( Y ∗ ) = ℓ ( Y ) − 1 , and ℓ ( Y ) − ℓ ( Z ) ≥ 3 \ell(Y)-\ell(Z)\ge3 ℓ ( Y ) − ℓ ( Z ) ≥ 3 . Then
μ ( Z , Y ) = μ ( Z ∗ , Y ∗ ) , \mu(Z,Y)=\mu(Z^*,Y^*), μ ( Z , Y ) = μ ( Z ∗ , Y ∗ ) , where the right side is zero if the starred pair is not Bruhat-comparable.
Proof. Since ℓ ( Y ) − ℓ ( Z ) ≥ 3 \ell(Y)-\ell(Z)\ge3 ℓ ( Y ) − ℓ ( Z ) ≥ 3 , Z , Y Z,Y Z , Y lie in different right W { s , t } W_{\{s,t\}} W { s , t } -cosets: within one such coset elements of D R ( s , t ) D_R(s,t) D R ( s , t ) have parabolic factors of length one or two, so their lengths differ by at most one. By [9 Theorem 6.2] , we have Z ≺ Y Z\prec Y Z ≺ Y if and only if Z ∗ ≺ Y ∗ Z^*\prec Y^* Z ∗ ≺ Y ∗ , where u ≺ v u\prec v u ≺ v means u < v u<v u < v in Bruhat order and μ ( u , v ) ≠ 0 \mu(u,v)\ne0 μ ( u , v ) = 0 . When these conditions hold, the same theorem gives μ ( Z , Y ) = μ ( Z ∗ , Y ∗ ) \mu(Z,Y)=\mu(Z^*,Y^*) μ ( Z , Y ) = μ ( Z ∗ , Y ∗ ) . Otherwise, both sides vanish, with the stated convention, since
ℓ ( Y ∗ ) − ℓ ( Z ∗ ) = ℓ ( Y ) − ℓ ( Z ) − 2 ≥ 1 \ell(Y^*)-\ell(Z^*)=\ell(Y)-\ell(Z)-2\ge1 ℓ ( Y ∗ ) − ℓ ( Z ∗ ) = ℓ ( Y ) − ℓ ( Z ) − 2 ≥ 1
rules out Y ∗ < Z ∗ Y^*<Z^* Y ∗ < Z ∗ .
Lemma 6.5. Let z ∈ N t z\in\mathcal{N}_t z ∈ N t belong to the mixed case. Then μ I ( z , y t + 1 ) = 0 \mu_I(z,y_{t+1})=0 μ I ( z , y t + 1 ) = 0 .
Proof. Suppose that μ I ( z , y t + 1 ) ≠ 0 \mu_I(z,y_{t+1})\ne0 μ I ( z , y t + 1 ) = 0 , and write z = c ω z=c\omega z = c ω . Since z z z belongs to the mixed case, we have ω r + 1 ω r + 2 = A C \omega_{r+1}\omega_{r+2}=AC ω r + 1 ω r + 2 = A C . Let j ≤ r j\le r j ≤ r be the type of ω \omega ω . We first check that the letters at positions r , … , r + 3 r,\ldots,r+3 r , … , r + 3 have the form
X A C T , X ∈ { A , B } , T ∈ { B , C } . XACT,\qquad X\in\{A,B\},\qquad T\in\{B,C\}. X A C T , X ∈ { A , B } , T ∈ { B , C } . If X = C X=C X = C , the first ordered block contains at least r r r letters C C C , with another at position r + 2 r+2 r + 2 . This would force j ≥ r + 1 j\ge r+1 j ≥ r + 1 . If T = A T=A T = A , the last ordered block is C A r − 1 CA^{r-1} C A r − 1 . Together with the A at position r + 1 r+1 r + 1 , this forces j = r j=r j = r . Therefore,
ω = C r − 1 B A ∣ C A r − 1 . \omega=C^{r-1}BA\mid CA^{r-1}. ω = C r − 1 B A ∣ C A r − 1 . This contradicts ω b ≠ C \omega_b\ne C ω b = C , since 1 ≤ b ≤ r − 1 1\le b\le r-1 1 ≤ b ≤ r − 1 .
Set Z = w I z Z=w_Iz Z = w I z and Y = w I y t + 1 Y=w_Iy_{t+1} Y = w I y t + 1 . We will apply a right star operation and then use the descent restriction to force an equality incompatible with the ordered first block of ω \omega ω .
By Theorem 2.10, the word of y t + 1 y_{t+1} y t + 1 is
C b − 1 A C t + 1 B C A r − 1 . C^{b-1}AC^{t+1}BCA^{r-1}. C b − 1 A C t + 1 B C A r − 1 . Since
w I ( k + q N ) = N + 1 − k + q N ( 1 ≤ k ≤ N , q ∈ Z ) , w_I(k+qN)=N+1-k+qN\qquad(1\le k\le N,\ q\in\mathbb{Z}), w I ( k + q N ) = N + 1 − k + q N ( 1 ≤ k ≤ N , q ∈ Z ) , left multiplication by w I w_I w I reverses the numerical order within each list in (26) . Entries corresponding to A remain smaller than those corresponding to B, which remain smaller than those corresponding to C. Since the letters of z z z and y t + 1 y_{t+1} y t + 1 at positions r , … , r + 3 r,\ldots,r+3 r , … , r + 3 are, respectively,
X A C T and C B C A , XACT\qquad\text{and}\qquad CBCA, X A C T and C B C A , we must have
Z ( r + 1 ) < Z ( r ) < Z ( r + 2 ) , Y ( r + 1 ) < Y ( r + 2 ) < Y ( r ) . Z(r+1)<Z(r)<Z(r+2),\qquad Y(r+1)<Y(r+2)<Y(r). Z ( r + 1 ) < Z ( r ) < Z ( r + 2 ) , Y ( r + 1 ) < Y ( r + 2 ) < Y ( r ) . The first inequality holds also when X = A X=A X = A , since successive occurrences of A have decreasing entries in Z Z Z . The triples of entries at positions r , r + 1 , r + 2 r,r+1,r+2 r , r + 1 , r + 2 have relative order 213 and 312, respectively. Thus Z , Y ∈ D R ( s r , s r + 1 ) Z,Y\in D_R(s_r,s_{r+1}) Z , Y ∈ D R ( s r , s r + 1 ) and
Z ∗ = Z s r + 1 , Y ∗ = Y s r , ℓ ( Z ∗ ) = ℓ ( Z ) + 1 , ℓ ( Y ∗ ) = ℓ ( Y ) − 1. Z^*=Zs_{r+1},\qquad Y^*=Ys_r,\qquad\ell(Z^*)=\ell(Z)+1,\qquad\ell(Y^*)=\ell(Y)-1. Z ∗ = Z s r + 1 , Y ∗ = Y s r , ℓ ( Z ∗ ) = ℓ ( Z ) + 1 , ℓ ( Y ∗ ) = ℓ ( Y ) − 1. We have Z < Y Z<Y Z < Y and μ ( Z , Y ) = μ I ( z , y t + 1 ) ≠ 0 \mu(Z,Y)=\mu_I(z,y_{t+1})\ne0 μ ( Z , Y ) = μ I ( z , y t + 1 ) = 0 , so ℓ ( Y ) − ℓ ( Z ) \ell(Y)-\ell(Z) ℓ ( Y ) − ℓ ( Z ) is odd. Moreover,
ℓ ( Y ) − ℓ ( Z ) = ℓ ( y t + 1 ) − ℓ ( z ) > 1 \ell(Y)-\ell(Z)=\ell(y_{t+1})-\ell(z)>1 ℓ ( Y ) − ℓ ( Z ) = ℓ ( y t + 1 ) − ℓ ( z ) > 1 because z ∈ N t z\in\mathcal{N}_t z ∈ N t . Thus this difference is at least three. We may therefore apply Theorem 6.4, obtaining
Z ∗ < Y ∗ , μ ( Z ∗ , Y ∗ ) ≠ 0. Z^*<Y^*,\qquad\mu(Z^*,Y^*)\ne0. Z ∗ < Y ∗ , μ ( Z ∗ , Y ∗ ) = 0.
Put z ′ = z s r + 1 z^{\prime}=zs_{r+1} z ′ = z s r + 1 and y ′ = y t + 1 s r y^{\prime}=y_{t+1}s_r y ′ = y t + 1 s r . In the words of z z z and y t + 1 y_{t+1} y t + 1 , these swaps exchange adjacent distinct letters, so the resulting admissible words correspond to the elements z ′ z^{\prime} z ′ and y ′ y^{\prime} y ′ . Thus, z ′ , y ′ ∈ I W z^{\prime},y^{\prime}\in I W z ′ , y ′ ∈ I W by Theorem 3.7. Since Z ∗ = w I z ′ Z^*=w_Iz^{\prime} Z ∗ = w I z ′ and Y ∗ = w I y ′ Y^*=w_Iy^{\prime} Y ∗ = w I y ′ , we have
z ′ < y ′ , μ I ( z ′ , y ′ ) ≠ 0. z^{\prime}<y^{\prime},\qquad\mu_I(z^{\prime},y^{\prime})\ne0. z ′ < y ′ , μ I ( z ′ , y ′ ) = 0.
Their words at positions r , … , r + 3 r,\ldots,r+3 r , … , r + 3 are now
X C A T and B C C A , \mathrm{XCAT}\qquad\text{and}\qquad\mathrm{BCCA}, XCAT and BCCA ,
respectively. Since T ∈ { B , C } T\in\{\mathrm{B},\mathrm{C}\} T ∈ { B , C } , the descent test Theorem 4.9 gives
s r + 2 ∈ R I ( y ′ ) ∖ R I ( z ′ ) . s_{r+2}\in R_I(y^{\prime})\setminus R_I(z^{\prime}). s r + 2 ∈ R I ( y ′ ) ∖ R I ( z ′ ) .
Then, Theorem 4.2 yields
y ′ = z ′ s r + 2 = z s r + 1 s r + 2 . y^{\prime}=z^{\prime}s_{r+2}=zs_{r+1}s_{r+2}. y ′ = z ′ s r + 2 = z s r + 1 s r + 2 .
The two swaps on the right exchange distinct letters and leave the first r r r letters of ω \omega ω unchanged. Thus the word of z s r + 1 s r + 2 zs_{r+1}s_{r+2} z s r + 1 s r + 2 has an ordered prefix of length r r r . However, the word of y ′ y^{\prime} y ′ has prefix C b − 1 A C t B C^{b-1}AC^{t}B C b − 1 A C t B , which has length r r r and is not ordered, since its A precedes its B. This contradicts the uniqueness of the admissible word in Theorem 3.7. □
We are now in a position to complete the proof of Theorem 2.11.
Proof of Theorem 2.11. Fix r ≥ 2 r\ge2 r ≥ 2 and 0 ≤ t ≤ r − 2 0\le t\le r-2 0 ≤ t ≤ r − 2 , and define
E t = ∑ z ∈ N t μ I ( z , y t + 1 ) C z . \mathcal{E}_t=\sum_{z\in N_t}\mu_I(z,y_{t+1})C_z. E t = z ∈ N t ∑ μ I ( z , y t + 1 ) C z .
The candidate expansion (54) gives (10) , with the cover term C y t + 2 C_{y_{t+2}} C y t + 2 omitted when t = r − 2 t=r-2 t = r − 2 . It remains to prove that ord v ( ε ( E t ) ) ≥ t + 3 \operatorname{ord}_v(\varepsilon(\mathcal{E}_t))\ge t+3 ord v ( ε ( E t )) ≥ t + 3 , as explained in the beginning of this section.
By Theorems 6.1 and 6.5, the coefficients μ I ( z , y t + 1 ) \mu_I(z,y_{t+1}) μ I ( z , y t + 1 ) vanish in the ordered and mixed cases. Hence
ε ( E t ) = ∑ z ∈ N t z belongs to a pure case μ I ( z , y t + 1 ) h z . \varepsilon(\mathcal{E}_t)=\sum_{\substack{z\in N_t\\ z\text{ belongs to a pure case}}}\mu_I(z,y_{t+1})h_z. ε ( E t ) = z ∈ N t z belongs to a pure case ∑ μ I ( z , y t + 1 ) h z .
By Theorem 6.2, every h z h_z h z in this sum has order at least t + 3 t+3 t + 3 . Since the coefficients μ I ( z , y t + 1 ) \mu_I(z,y_{t+1}) μ I ( z , y t + 1 ) are integers, it follows that
ord v ( ε ( E t ) ) ≥ t + 3 , \operatorname{ord}_v\bigl(\varepsilon(\mathcal{E}_t)\bigr)\ge t+3, ord v ( ε ( E t ) ) ≥ t + 3 ,
as required. □
References
Department of Mathematics, University of York, Heslington, York, UK
Email address: Chris.Bowman-Scargill@york.ac.uk
LACIM, Université du Québec à Montréal, Montreal, Canada.
Email address: damiandlfa@gmail.com
Departamento de Matemáticas, Facultad de Ciencias, Universidad de Chile, Chile
Email address: nlibedinsky@u.uchile.cl
Instituto de Matemáticas, Universidad de Talca, Talca, Chile
Email address: dplaza@utalca.cl