Introduction
Gordon’s beautiful and deep partition theorem extends the combinatorial form of the Rogers–Ramanujan identities to arbitrary odd moduli 2 k + 3 2k+3 2 k + 3 with k ≥ 1 k \ge1 k ≥ 1 [18 ] . Its frequency conditions are the starting point for the arrays studied here. A partition π = ( π 1 , … , π ℓ ) \pi= (\pi_1,\ldots,\pi_\ell) π = ( π 1 , … , π ℓ ) is a finite weakly decreasing sequence of positive integers. Its size is ∣ π ∣ = ∑ i π i |\pi| = \sum_i \pi_i ∣ π ∣ = ∑ i π i , its length is ℓ ( π ) = ℓ \ell(\pi) = \ell ℓ ( π ) = ℓ , and the frequency f j ( π ) f_j(\pi) f j ( π ) is the number of parts equal to j j j . Thus ∣ π ∣ = ∑ j ≥ 1 j f j ( π ) |\pi| = \sum_{j\ge1} j f_j(\pi) ∣ π ∣ = ∑ j ≥ 1 j f j ( π ) and ℓ ( π ) = ∑ j ≥ 1 f j ( π ) \ell(\pi) = \sum_{j\ge1} f_j(\pi) ℓ ( π ) = ∑ j ≥ 1 f j ( π ) . We include the empty partition, of size and length zero.
For integers k ≥ 1 k \ge1 k ≥ 1 and 0 ≤ a ≤ k 0 \le a \le k 0 ≤ a ≤ k , let G k , a \mathcal{G}_{k,a} G k , a consist of the partitions satisfying
f 1 ( π ) ≤ a , f j ( π ) + f j + 1 ( π ) ≤ k ( j ≥ 1 ) . f_1(\pi) \le a,\qquad f_j(\pi) + f_{j+1}(\pi) \le k\quad(j \ge1). f 1 ( π ) ≤ a , f j ( π ) + f j + 1 ( π ) ≤ k ( j ≥ 1 ) .
Gordon’s theorem [18 ] states that, for every integer n ≥ 0 n \ge0 n ≥ 0 , the number of partitions of n n n in G k , a \mathcal{G}_{k,a} G k , a equals the number of partitions of n n n having no part congruent to 0 0 0 or ± ( a + 1 ) \pm(a+1) ± ( a + 1 ) modulo 2 k + 3 2k+3 2 k + 3 . Consequently,
∑ π ∈ G k , a q ∣ π ∣ = ( q a + 1 , q 2 k − a + 2 , q 2 k + 3 ; q 2 k + 3 ) ∞ ( q ; q ) ∞ . (1) \sum_{\pi\in\mathcal{G}_{k,a}} q^{|\pi|}
=
\frac{(q^{a+1},q^{2k-a+2},q^{2k+3};q^{2k+3})_\infty}{(q;q)_\infty}. \tag*{(1)} π ∈ G k , a ∑ q ∣ π ∣ = ( q ; q ) ∞ ( q a + 1 , q 2 k − a + 2 , q 2 k + 3 ; q 2 k + 3 ) ∞ . (1)
We use the standard notation [17 Chapter 1]
( u ; q ) m = ∏ j = 0 m − 1 ( 1 − u q j ) , ( u 1 , … , u s ; Q ) ∞ = ∏ h = 1 s ∏ j ≥ 0 ( 1 − u h Q j ) , (u;q)_m=\prod_{j=0}^{m-1}(1-uq^j),\qquad(u_1,\ldots,u_s;Q)_\infty=\prod_{h=1}^{s}\prod_{j\ge0}(1-u_hQ^j), ( u ; q ) m = j = 0 ∏ m − 1 ( 1 − u q j ) , ( u 1 , … , u s ; Q ) ∞ = h = 1 ∏ s j ≥ 0 ∏ ( 1 − u h Q j ) ,
with empty products equal to 1.
Andrews obtained the analytic multiple-series form of Gordon’s odd-modulus identities [3 Theorem 1] , and Bressoud established the corresponding even-modulus identities [8 ] . These two families are collectively known as the Andrews–Gordon–Bressoud (AGB) identities. The Bailey lattice of Agarwal, Andrews and Bressoud [2 ] provides a systematic analytic approach to these identities. For background on partition identities and Bailey methods, see Andrews’ monograph [5 ] and Warnaar’s survey [32 ] .
The multisums also admit several partition-theoretic interpretations. Andrews interpreted the odd-modulus sums using Durfee dissections into successive squares and rectangles [4 ] . Warnaar obtained polynomial refinements of the Andrews–Gordon identities and interpreted the associated q q q -multinomial coefficients through Durfee dissections [31 ] . Dousse, Jouhet and Konan gave bijective interpretations of both the Andrews–Gordon and Bressoud multisums by constructing an explicit weight- and length-preserving bijection [11 Theorem 1.7 and Corollary 1.8] . A complementary commutative-algebraic approach was developed by Afsharijoo, Dousse, Jouhet and Mourtada, who used graded quotient rings, new Durfee dissections and q q q -series identities to prove companion identities to the Andrews–Gordon identities [1 ] .
Andrews, Schilling and Warnaar extended the Bailey lemma to type A 2 A_2 A 2 and used it to derive higher-rank analogues of the AGB identities [6 ] . Warnaar subsequently gave a Hall–Littlewood derivation of the A 2 A_2 A 2 Rogers–Ramanujan identities [33 ] and developed an A 2 A_2 A 2 Bailey tree yielding further A 2 ( 1 ) A_2^{(1)} A 2 ( 1 ) analogues of the Andrews–Gordon identities [34 ] .
In the extension of Gordon’s conditions proposed by Capparelli, Meurman, Primc and Primc (CMPP) [10 ] , parts of the same size may have different colours. Parts of the same size and colour are indistinguishable. A coloured partition is therefore specified by its finitely supported frequencies f j ( c ) f_j^{(c)} f j ( c ) , where f j ( c ) f_j^{(c)} f j ( c ) counts the parts of size j > 0 j>0 j > 0 and colour c c c . For the type C C C family studied here, the bound on f j + f j + 1 f_j+f_{j+1} f j + f j + 1 becomes a bound on the sum of entries along a path. Section 7 discusses the related families of types A A A and D D D , and Appendices A and B recall the two arrays.
The type C C C array has 2 n + 1 2n+1 2 n + 1 rows [10 ] . We use the row colours and parity convention of Kanade, Russell, Tsuchioka and Warnaar (KRTW) [21 §4, model $(a')$] . For the array definition fix n ≥ 1 n\geq1 n ≥ 1 , put N = 2 n N=2n N = 2 n , and let k = ( k 0 , … , k n ) ∈ Z ≥ 0 n + 1 \boldsymbol{k}=(k_0,\ldots,k_n)\in\mathbb{Z}_{\geq0}^{n+1} k = ( k 0 , … , k n ) ∈ Z ≥ 0 n + 1 have total k = ∑ i k i k=\sum_i k_i k = ∑ i k i . Write f p ( c ) f_p^{(c)} f p ( c ) for the frequency of a part of size p > 0 p>0 p > 0 in colour c ∈ { 1 , … , N + 1 } c\in\{1,\ldots,N+1\} c ∈ { 1 , … , N + 1 } , where p + c p+c p + c is even. Prescribe
f − 1 ( 2 i + 1 ) = k i ( 0 ≤ i ≤ n ) , f 0 ( 2 i ) = 0 ( 1 ≤ i ≤ n ) . f_{-1}^{(2i+1)}=k_i \quad(0\leq i\leq n),\qquad f_0^{(2i)}=0 \quad(1\leq i\leq n). f − 1 ( 2 i + 1 ) = k i ( 0 ≤ i ≤ n ) , f 0 ( 2 i ) = 0 ( 1 ≤ i ≤ n ) .
A full path takes one entry from each row, with successive part indices differing by 1. The type C C C admissibility condition requires the sum of the entries on every such path to be at most k k k .
For the proofs, it is convenient to reverse the row index without changing the part size. Set r = N + 1 − c r=N+1-c r = N + 1 − c and define
Ω = { ( r , p ) ∈ Z 2 : 0 ≤ r ≤ N , p ≥ − 1 , r + p ≡ 1 ( m o d 2 ) } , Ω + = { ( r , p ) ∈ Ω : p > 0 } . (2) \Omega=\{(r,p)\in\mathbb{Z}^2:0\leq r\leq N,\ p\geq-1,\ r+p\equiv1\pmod2\},\qquad\Omega_+=\{(r,p)\in\Omega:p>0\}. \tag*{(2)} Ω = {( r , p ) ∈ Z 2 : 0 ≤ r ≤ N , p ≥ − 1 , r + p ≡ 1 ( mod 2 )} , Ω + = {( r , p ) ∈ Ω : p > 0 } . (2)
There is a directed edge ( r , p ) ⟶ ( r + 1 , p ± 1 ) (r,p)\longrightarrow(r+1,p\pm1) ( r , p ) ⟶ ( r + 1 , p ± 1 ) whenever both vertices belong to Ω \Omega Ω . An array of frequencies of positive parts is a finitely supported function f : Ω + → Z ≥ 0 f:\Omega_+\to\mathbb{Z}_{\geq0} f : Ω + → Z ≥ 0 . Its augmentation F F F is defined by
F ( r , p ) = f ( r , p ) ( p > 0 ) , F ( B i ) = k i , B i = ( N − 2 i , − 1 ) , F ( r , 0 ) = 0. (3) F(r,p)=f(r,p)\ (p>0),\qquad F(B_i)=k_i,\quad B_i=(N-2i,-1),\qquad F(r,0)=0. \tag*{(3)} F ( r , p ) = f ( r , p ) ( p > 0 ) , F ( B i ) = k i , B i = ( N − 2 i , − 1 ) , F ( r , 0 ) = 0. (3)
Thus f ( r , p ) = f p ( N + 1 − r ) f(r,p)=f_p^{(N+1-r)} f ( r , p ) = f p ( N + 1 − r ) . Reversing a path reverses its list of rows and preserves the sum of its entries. The original CMPP conditions are therefore exactly
∑ v ∈ P F ( v ) ≤ k for every full path P ⊂ Ω . (4) \sum_{v\in P}F(v)\leq k\qquad\text{for every full path }P\subset\Omega. \tag*{(4)} v ∈ P ∑ F ( v ) ≤ k for every full path P ⊂ Ω. (4)
Write S k S_k S k for these admissible arrays, and put
∣ f ∣ p r = ∑ ( r , p ) ∈ Ω + p f ( r , p ) , ℓ ( f ) = ∑ ( r , p ) ∈ Ω + f ( r , p ) , Z k ( z , q ) = ∑ f ∈ S k z ℓ ( f ) q ∣ f ∣ p r . (5) |f|_{\mathrm{pr}}=\sum_{(r,p)\in\Omega_+}pf(r,p),\qquad\ell(f)=\sum_{(r,p)\in\Omega_+}f(r,p),\qquad Z_k(z,q)=\sum_{f\in S_k}z^{\ell(f)}q^{|f|_{\mathrm{pr}}}. \tag*{(5)} ∣ f ∣ pr = ( r , p ) ∈ Ω + ∑ p f ( r , p ) , ℓ ( f ) = ( r , p ) ∈ Ω + ∑ f ( r , p ) , Z k ( z , q ) = f ∈ S k ∑ z ℓ ( f ) q ∣ f ∣ pr . (5)
The boundary entries are omitted from the statistics. For each d d d , only finitely many arrays have size d d d , so Z k ( z , q ) ∈ Z [ z ] [ [ q ] ] Z_k(z,q)\in\mathbb{Z}[z][[q]] Z k ( z , q ) ∈ Z [ z ] [[ q ]] .
For example, let n = 2 n=2 n = 2 and k = e 1 = ( 0 , 1 , 0 ) \boldsymbol{k}=\boldsymbol{e}_1=(0,1,0) k = e 1 = ( 0 , 1 , 0 ) , so that k = 1 k=1 k = 1 and F ( 2 , − 1 ) = 1 F(2,-1)=1 F ( 2 , − 1 ) = 1 . The path in Figure 1 gives
Figure 1. The type C 2 C_2 C 2 frequency array with reversed row order.
f ( 0 , 1 ) + f ( 1 , 2 ) + f ( 2 , 1 ) + f ( 3 , 2 ) + f ( 4 , 1 ) ≤ 1. f(0,1)+f(1,2)+f(2,1)+f(3,2)+f(4,1)\leq1. f ( 0 , 1 ) + f ( 1 , 2 ) + f ( 2 , 1 ) + f ( 3 , 2 ) + f ( 4 , 1 ) ≤ 1.
The paths with column sequences ( 1 , 0 , − 1 , 0 , 1 ) (1,0,-1,0,1) ( 1 , 0 , − 1 , 0 , 1 ) and ( − 1 , 0 , 1 , 0 , − 1 ) (-1,0,1,0,-1) ( − 1 , 0 , 1 , 0 , − 1 ) give, respectively,
f ( 0 , 1 ) + 1 + f ( 4 , 1 ) ≤ 1 , f ( 2 , 1 ) ≤ 1. f(0,1)+1+f(4,1)\leq1,\qquad f(2,1)\leq1. f ( 0 , 1 ) + 1 + f ( 4 , 1 ) ≤ 1 , f ( 2 , 1 ) ≤ 1.
Thus f ( 0 , 1 ) = f ( 4 , 1 ) = 0 f(0,1)=f(4,1)=0 f ( 0 , 1 ) = f ( 4 , 1 ) = 0 , whereas f ( 2 , 1 ) f(2,1) f ( 2 , 1 ) may equal one.
Two vertices ( r , p ) , ( r ′ , p ′ ) ∈ Ω (r,p),(r',p')\in\Omega ( r , p ) , ( r ′ , p ′ ) ∈ Ω lie on a common full path precisely when
∣ p − p ′ ∣ ≤ ∣ r − r ′ ∣ . (6) |p-p'|\leq|r-r'|. \tag*{(6)} ∣ p − p ′ ∣ ≤ ∣ r − r ′ ∣. (6)
Indeed, suppose r ≤ r ′ r\leq r' r ≤ r ′ and put d = r ′ − r d=r'-r d = r ′ − r . A path joining the two vertices has d d d steps, so ∣ p ′ − p ∣ ≤ d |p'-p|\leq d ∣ p ′ − p ∣ ≤ d is necessary. Conversely, grid parity makes u = ( d + p ′ − p ) / 2 u=(d+p'-p)/2 u = ( d + p ′ − p ) /2 an integer, and the bound gives 0 ≤ u ≤ d 0\leq u\leq d 0 ≤ u ≤ d . Starting at ( r , p ) (r,p) ( r , p ) , take u u u steps that increase the column index and then d − u d-u d − u steps that decrease it. This reaches ( r ′ , p ′ ) (r',p') ( r ′ , p ′ ) without going below min ( p , p ′ ) ≥ − 1 \min(p,p')\geq-1 min ( p , p ′ ) ≥ − 1 .
For n ≥ 2 n\geq2 n ≥ 2 , let g ^ \widehat{\mathfrak{g}} g be the untwisted affine Lie algebra of type C n ( 1 ) C_n^{(1)} C n ( 1 ) . For n = 1 n=1 n = 1 we use A 1 ( 1 ) A_1^{(1)} A 1 ( 1 ) . In either case write the simple roots as α 0 , … , α n \alpha_0,\ldots,\alpha_n α 0 , … , α n , the fundamental weights as Λ 0 , … , Λ n \Lambda_0,\ldots,\Lambda_n Λ 0 , … , Λ n , and the negative subalgebra as n ^ − \widehat{\mathfrak{n}}_- n − . Our normalizations and concrete realization are given in Section 2. Put
Λ = ∑ i = 0 n k i Λ i , \Lambda=\sum_{i=0}^{n}k_i\Lambda_i, Λ = i = 0 ∑ n k i Λ i ,
and let L ( Λ ) L(\Lambda) L ( Λ ) be the integrable irreducible highest weight module with highest vector v Λ v_\Lambda v Λ . This weight has level k k k .
Write Q + = ∑ i = 0 n Z ≥ 0 α i Q_+=\sum_{i=0}^{n}\mathbb{Z}_{\geq0}\alpha_i Q + = ∑ i = 0 n Z ≥ 0 α i . The normalized character is
e − Λ ch L ( Λ ) = ∑ β ∈ Q + dim L ( Λ ) Λ − β e − β . (7) e^{-\Lambda}\operatorname{ch}L(\Lambda)=\sum_{\beta\in Q_+}\dim L(\Lambda)_{\Lambda-\beta}e^{-\beta}. \tag*{(7)} e − Λ ch L ( Λ ) = β ∈ Q + ∑ dim L ( Λ ) Λ − β e − β . (7)
The principal specialization ps \operatorname{ps} ps sends each e − α i e^{-\alpha_i} e − α i to q q q . Thus the term indexed by β = ∑ i b i α i \beta=\sum_i b_i\alpha_i β = ∑ i b i α i contributes to degree ∑ i b i \sum_i b_i ∑ i b i . Once a partition series has been identified with such a specialization, a character formula can be used to evaluate it. Lepowsky and Milne established this connection for the Rogers–Ramanujan identities [24 ] .
CMPP formulated the following conjecture for the principally specialized normalized characters of standard C n ( 1 ) C_n^{(1)} C n ( 1 ) -modules [10 Conjecture 3.3] .
Conjecture 1. For n ≥ 2 n \ge2 n ≥ 2 and every k = ( k 0 , … , k n ) ∈ Z ≥ 0 n + 1 \boldsymbol{k}=(k_0,\ldots,k_n)\in\mathbb{Z}_{\geq0}^{n+1} k = ( k 0 , … , k n ) ∈ Z ≥ 0 n + 1 , put Λ = ∑ i = 0 n k i Λ i \Lambda=\sum_{i=0}^{n}k_i\Lambda_i Λ = ∑ i = 0 n k i Λ i . Then
Z k ( 1 , q ) = ps ( e − Λ ch L ( Λ ) ) . (8) Z_{\boldsymbol{k}}(1,q)=\operatorname{ps}(e^{-\Lambda}\operatorname{ch}L(\Lambda)). \tag*{(8)} Z k ( 1 , q ) = ps ( e − Λ ch L ( Λ )) . (8)
To see how a frequency array can index a vector of L ( Λ ) L(\Lambda) L ( Λ ) , choose an ordered basis { x b } \{x_b\} { x b } of n − \mathfrak{n}_- n − . The Poincaré–Birkhoff–Witt (PBW) theorem states that 1 1 1 and the ordered products
x b 1 ⋯ x b r , b 1 ≤ ⋯ ≤ b r , x_{b_1}\cdots x_{b_r},\qquad b_1\leq\cdots\leq b_r, x b 1 ⋯ x b r , b 1 ≤ ⋯ ≤ b r ,
form a basis of the universal enveloping algebra U ( n − ) U(\mathfrak{n}_-) U ( n − ) . The terms x b 1 ⋯ x b r v Λ x_{b_1}\cdots x_{b_r}v_\Lambda x b 1 ⋯ x b r v Λ therefore span L ( Λ ) L(\Lambda) L ( Λ ) . Our task is to identify a basis among these vectors using the CMPP path conditions.
To state the basis theorem, we translate an array f f f into a word in the generators of U ( n − ) U(\mathfrak{n}_-) U ( n − ) . Write X a b ( m ) = X a b ⊗ t m X_{ab}(m)=X_{ab}\otimes t^m X ab ( m ) = X ab ⊗ t m : the indices a , b a,b a , b specify the colour, and the integer m m m is the loop mode. The dictionary in Proposition 3.1 associates exactly one such generator x ( r , p ) x(r,p) x ( r , p ) with each positive vertex ( r , p ) (r,p) ( r , p ) . More explicitly, if x ( r , p ) = X a b ( − j ) x(r,p)=X_{ab}(-j) x ( r , p ) = X ab ( − j ) , then
p = N j + a + b − N − 1 , r = { b − a , j even , N − b + a , j odd , 1 ≤ a ≤ b ≤ N . p=Nj+a+b-N-1,\qquad r=\begin{cases}b-a,&j\text{ even},\\N-b+a,&j\text{ odd},\end{cases}\qquad1\leq a\leq b\leq N. p = N j + a + b − N − 1 , r = { b − a , N − b + a , j even , j odd , 1 ≤ a ≤ b ≤ N .
Here j ≥ 1 j\geq1 j ≥ 1 , or j = 0 j=0 j = 0 and a + b > N + 1 a+b>N+1 a + b > N + 1 . In particular, the part size p p p and the loop mode − j -j − j are different labels.
List the vertices as v 1 , … , v s v_1,\ldots,v_s v 1 , … , v s with positive exponent f f f , in the order x ( v 1 ) ≺ ⋯ ≺ x ( v s ) x(v_1)\prec\cdots\prec x(v_s) x ( v 1 ) ≺ ⋯ ≺ x ( v s ) of their generators. To obtain this order, first read the modes m m m in increasing order. At a fixed mode, read b b b in decreasing order and, for equal b b b , read a a a in decreasing order. We use
X f = x ( v 1 ) f ( v 1 ) ⋯ x ( v s ) f ( v s ) in U ( n − ) , X 0 = 1. X^f=x(v_1)^{f(v_1)}\cdots x(v_s)^{f(v_s)}\quad\text{in }U(\mathfrak{n}_-),\qquad X^0=1. X f = x ( v 1 ) f ( v 1 ) ⋯ x ( v s ) f ( v s ) in U ( n − ) , X 0 = 1.
Thus f ( v ) f(v) f ( v ) counts copies of a factor. Boundary entries constrain f f f but supply no factors. In the vector X f v Λ X^fv_\Lambda X f v Λ the rightmost factor acts first. If x ( v ) x(v) x ( v ) has weight − γ v -\gamma_v − γ v , its principal degree is ht ( γ v ) = p \operatorname{ht}(\gamma_v)=p ht ( γ v ) = p for v = ( r , p ) v=(r,p) v = ( r , p ) , and the total lowering weight is
β ( f ) = ∑ v ∈ Ω + f ( v ) γ v . \beta(f)=\sum_{v\in\Omega_+}f(v)\gamma_v. β ( f ) = v ∈ Ω + ∑ f ( v ) γ v .
To compare entire monomials we use the order of Primc and Šikić (the PS order [28 ] ), defined in § 3.1 by successive comparisons of length, total loop degree, modes and colours. An exponent is essential if its vector is not a finite linear combination of vectors with strictly larger exponents in this order. Denote the set of essential exponents by E Λ \mathcal{E}_\Lambda E Λ . Our main result identifies the admissible and essential exponents.
Theorem 2. For n ≥ 2 n\geq2 n ≥ 2 and every k ∈ Z ≥ 0 n + 1 \boldsymbol{k}\in\mathbb{Z}_{\geq0}^{n+1} k ∈ Z ≥ 0 n + 1 , we have E Λ = S k \mathcal{E}_\Lambda=\mathcal{S}_{\boldsymbol{k}} E Λ = S k . Consequently, the vectors
{ X f v Λ : f ∈ S k } \{X^fv_\Lambda:f\in\mathcal{S}_{\boldsymbol{k}}\} { X f v Λ : f ∈ S k } form a basis of L ( Λ ) L(\Lambda) L ( Λ ) . Every inadmissible monomial vector is a finite linear combination of admissible monomial vectors of the same affine weight with strictly larger exponents.
The following example shows how the factors of X f X^f X f are read from the array.
Example 1.1. Take n = 2 n=2 n = 2 and ( k 0 , k 1 , k 2 ) = ( 1 , 2 , 1 ) (k_0,k_1,k_2)=(1,2,1) ( k 0 , k 1 , k 2 ) = ( 1 , 2 , 1 ) , so that N = k = 4 N=k=4 N = k = 4 . Figure 2 gives both the generator and the frequency at each positive vertex shown. At each positive vertex, X a b ( m ) [ h ] X_{ab}(m)\ [h] X ab ( m ) [ h ] denotes a PBW generator with frequency h h h . A frame indicates h > 0 h>0 h > 0 . All frequencies beyond p = 5 p=5 p = 5 are zero.
Figure 2. A type C 2 ( 1 ) C_2^{(1)} C 2 ( 1 ) array with boundary ( k 0 , k 1 , k 2 ) = ( 1 , 2 , 1 ) (k_0,k_1,k_2)=(1,2,1) ( k 0 , k 1 , k 2 ) = ( 1 , 2 , 1 ) . The heavy path has successive frequencies 1 , 0 , 1 , 0 , 2 1,0,1,0,2 1 , 0 , 1 , 0 , 2 , which sum to the level 4.
For instance, X 33 ( − 1 ) X_{33}(-1) X 33 ( − 1 ) occupies ( 4 , 5 ) (4,5) ( 4 , 5 ) : j = 1 j=1 j = 1 gives r = 4 − 3 + 3 = 4 r=4-3+3=4 r = 4 − 3 + 3 = 4 and p = 4 + 3 + 3 − 4 − 1 = 5 p=4+3+3-4-1=5 p = 4 + 3 + 3 − 4 − 1 = 5 . Its frequency is two, so it contributes X 33 ( − 1 ) 2 X_{33}(-1)^2 X 33 ( − 1 ) 2 to the word and two parts of size five to the partition. Reading just the occupied vertices, in the order specified above, gives
X f = X 11 ( − 2 ) X 24 ( − 1 ) X 33 ( − 1 ) 2 X 12 ( − 1 ) X 34 ( 0 ) X 24 ( 0 ) X 33 ( 0 ) . X^f=X_{11}(-2)\,X_{24}(-1)X_{33}(-1)^2X_{12}(-1)\,X_{34}(0)X_{24}(0)X_{33}(0). X f = X 11 ( − 2 ) X 24 ( − 1 ) X 33 ( − 1 ) 2 X 12 ( − 1 ) X 34 ( 0 ) X 24 ( 0 ) X 33 ( 0 ) .
To state the corresponding product formula, set
κ = k + n + 1 , λ i = ∑ j = i n k j , a i = λ i + n − i + 1 ( 1 ≤ i ≤ n ) . (9) \kappa=k+n+1,\qquad\lambda_i=\sum_{j=i}^{n}k_j,\qquad a_i=\lambda_i+n-i+1\quad(1\leq i\leq n). \tag*{(9)} κ = k + n + 1 , λ i = j = i ∑ n k j , a i = λ i + n − i + 1 ( 1 ≤ i ≤ n ) . (9)
We write θ ( u ; Q ) = ( u , Q / u ; Q ) ∞ \theta(u;Q)=(u,Q/u;Q)_\infty θ ( u ; Q ) = ( u , Q / u ; Q ) ∞ .
Theorem 3. Conjecture 1 holds. For n ≥ 1 n\geq1 n ≥ 1 and every k = ( k 0 , … , k n ) ∈ Z ≥ 0 n + 1 \mathbf{k}=(k_0,\ldots,k_n)\in\mathbb{Z}_{\geq0}^{n+1} k = ( k 0 , … , k n ) ∈ Z ≥ 0 n + 1 , put Λ = ∑ i = 0 n k i Λ i \Lambda=\sum_{i=0}^{n}k_i\Lambda_i Λ = ∑ i = 0 n k i Λ i , with type A 1 ( 1 ) A_1^{(1)} A 1 ( 1 ) understood when n = 1 n=1 n = 1 . Then
Z k ( 1 , q ) = ps ( e − Λ ch L ( Λ ) ) (10) Z_{\mathbf{k}}(1,q)=\operatorname{ps}(e^{-\Lambda}\operatorname{ch}L(\Lambda)) \tag*{(10)} Z k ( 1 , q ) = ps ( e − Λ ch L ( Λ )) (10) = ( q k ; q 2 κ ) ∞ ( q 2 κ ; q 2 κ ) ∞ n ( q ; q 2 ) ∞ ( q ; q ) ∞ n ∏ i = 1 n θ ( q a i ; q κ ) ∏ 1 ≤ i < j ≤ n θ ( q a i − a j ; q 2 κ ) θ ( q a i + a j ; q 2 κ ) . (11) =\frac{(q^k;q^{2\kappa})_\infty(q^{2\kappa};q^{2\kappa})_\infty^n}{(q;q^2)_\infty(q;q)_\infty^n}\prod_{i=1}^{n}\theta(q^{a_i};q^\kappa)\prod_{1\leq i<j\leq n}\theta(q^{a_i-a_j};q^{2\kappa})\theta(q^{a_i+a_j};q^{2\kappa}). \tag*{(11)} = ( q ; q 2 ) ∞ ( q ; q ) ∞ n ( q k ; q 2 κ ) ∞ ( q 2 κ ; q 2 κ ) ∞ n i = 1 ∏ n θ ( q a i ; q κ ) 1 ≤ i < j ≤ n ∏ θ ( q a i − a j ; q 2 κ ) θ ( q a i + a j ; q 2 κ ) . (11)
The explicit product form of Conjecture 1 is given in [21 Conjecture 4.1] . The product in (1.11) is a principal-character formula arising from Lepowsky’s numerator formula [22 ] ; see also [9 Theorems 3.1 and 3.3] . It agrees with [19 (3.19)] after taking λ 0 = k \lambda_0=k λ 0 = k and replacing the parameter κ \kappa κ there by 2 κ 2\kappa 2 κ , and with [21 (2.4)] in the present notation.
The rank-one case of n = 1 n=1 n = 1 , recalled in § 4.2 , follows from Meurman–Primc [26 Theorem 6.5.5 and §11.2] and connects directly with the classical partition identities. For s ≥ 1 s\geq1 s ≥ 1 , 0 ≤ a ≤ s 0\leq a\leq s 0 ≤ a ≤ s and ( k 0 , k 1 ) = ( 2 s + 1 − a , a ) (k_0,k_1)=(2s+1-a,a) ( k 0 , k 1 ) = ( 2 s + 1 − a , a ) , the affine level is 2 s + 1 2s+1 2 s + 1 , and (1.11) gives
( q ; q 2 ) ∞ Z ( 2 s + 1 − a , a ) ( 1 , q ) = ( q a + 1 , q 2 s − a + 2 , q 2 s + 3 ; q 2 s + 3 ) ∞ ( q ; q ) ∞ = ∑ π ∈ G s , a q ∣ π ∣ . (q;q^2)_\infty Z_{(2s+1-a,a)}(1,q)=\frac{(q^{a+1},q^{2s-a+2},q^{2s+3};q^{2s+3})_\infty}{(q;q)_\infty}=\sum_{\pi\in\mathcal{G}_{s,a}}q^{|\pi|}. ( q ; q 2 ) ∞ Z ( 2 s + 1 − a , a ) ( 1 , q ) = ( q ; q ) ∞ ( q a + 1 , q 2 s − a + 2 , q 2 s + 3 ; q 2 s + 3 ) ∞ = π ∈ G s , a ∑ q ∣ π ∣ .
Thus, after removing the factor ( q ; q 2 ) ∞ − 1 (q;q^2)_{\infty}^{-1} ( q ; q 2 ) ∞ − 1 , one obtains the Gordon product and its Andrews–Gordon multiple-series expression [3 , 18 ] ; s = 1 s=1 s = 1 gives the two Rogers–Ramanujan identities. This is the rank-one principal-character connection developed by Lepowsky and Milne [23 , 24 ] .
For arbitrary rank n n n , the specialization Λ = k Λ 0 \Lambda=k\Lambda_0 Λ = k Λ 0 has a Hall–Littlewood interpretation. Griffin, Ono and Warnaar’s identity [19 Theorem 1.2] , with their parameter m = k m=k m = k , gives
Z ( k , 0 , … , 0 ) ( 1 , q ) = ∑ μ : μ 1 ≤ k q ∣ μ ∣ P 2 μ ( 1 , q , q 2 , … ; q 2 n ) . Z_{(k,0,\ldots,0)}(1,q)=\sum_{\mu:\,\mu_1\leq k}q^{|\mu|}P_{2\mu}(1,q,q^2,\ldots;q^{2n}). Z ( k , 0 , … , 0 ) ( 1 , q ) = μ : μ 1 ≤ k ∑ q ∣ μ ∣ P 2 μ ( 1 , q , q 2 , … ; q 2 n ) .
Here P ν ( x ; t ) P_\nu(x;t) P ν ( x ; t ) is the Hall–Littlewood symmetric function in the usual P P P -normalization [25 Chapter III] , 2 μ = ( 2 μ 1 , 2 μ 2 , … ) 2\mu=(2\mu_1,2\mu_2,\ldots) 2 μ = ( 2 μ 1 , 2 μ 2 , … ) , and the sum includes the empty partition. The same formula holds for Λ = k Λ n \Lambda=k\Lambda_n Λ = k Λ n by reversal of the boundary data [21 (4.5a)] . The underlying expansion of the unspecialized character of L ( k Λ 0 ) L(k\Lambda_0) L ( k Λ 0 ) in modified Hall–Littlewood functions was proved by Bartlett and Warnaar [7 Theorem 1.1] and obtained from bounded Littlewood identities by Rains and Warnaar [30 Theorem 5.4] . These formulas connect the admissible PBW arrays with Hall–Littlewood character expansions.
The same basis records the number of positive parts: ℓ ( f ) = ∑ v ∈ Ω + f ( v ) \ell(f)=\sum_{v\in\Omega_+}f(v) ℓ ( f ) = ∑ v ∈ Ω + f ( v ) is the number of factors in X f X^f X f . To keep track of this statistic, let U ≤ m ( n ^ − ) U_{\leq m}(\widehat{\mathfrak n}_{-}) U ≤ m ( n − ) be the span of products of at most m m m elements of n ^ − \widehat{\mathfrak n}_{-} n − , and set
F m L ( Λ ) = U ≤ m ( n ^ − ) v Λ , F − 1 L ( Λ ) = 0. F_mL(\Lambda)=U_{\leq m}(\widehat{\mathfrak n}_{-})v_\Lambda,\qquad F_{-1}L(\Lambda)=0. F m L ( Λ ) = U ≤ m ( n − ) v Λ , F − 1 L ( Λ ) = 0.
For m ≥ 0 m\geq0 m ≥ 0 , the component of degree m m m in the associated graded module is
gr m P B W L ( Λ ) = F m L ( Λ ) / F m − 1 L ( Λ ) . \operatorname{gr}^{\mathrm{PBW}}_m L(\Lambda)=F_mL(\Lambda)/F_{m-1}L(\Lambda). gr m PBW L ( Λ ) = F m L ( Λ ) / F m − 1 L ( Λ ) .
For w ∈ F m L ( Λ ) w\in F_mL(\Lambda) w ∈ F m L ( Λ ) , write
[ w ] m : = w + F m − 1 L ( Λ ) ∈ gr m P B W L ( Λ ) . [w]_m:=w+F_{m-1}L(\Lambda)\in\operatorname{gr}^{\mathrm{PBW}}_mL(\Lambda). [ w ] m := w + F m − 1 L ( Λ ) ∈ gr m PBW L ( Λ ) .
This is the PBW symbol of w w w in degree m m m . It is nonzero precisely when w ∉ F m − 1 L ( Λ ) w\notin F_{m-1}L(\Lambda) w ∈ / F m − 1 L ( Λ ) . For an admissible monomial we take m = ℓ ( f ) m=\ell(f) m = ℓ ( f ) . In the following refinement, z z z records this PBW degree and q q q records the principal degree.
Corollary 4. The principal symbols of the admissible monomials form a basis of gr P B W L ( Λ ) \operatorname{gr}_{\mathrm{PBW}}L(\Lambda) gr PBW L ( Λ ) . In particular,
Z k ( z , q ) = ∑ m , d ≥ 0 dim ( F m L ( Λ ) / F m − 1 L ( Λ ) ) z m q d . (12) Z_k(z,q)=\sum_{m,d\geq0}\dim\left(F_mL(\Lambda)/F_{m-1}L(\Lambda)\right)z^m q^d. \tag*{(12)} Z k ( z , q ) = m , d ≥ 0 ∑ dim ( F m L ( Λ ) / F m − 1 L ( Λ ) ) z m q d . (12)
Example 1.1 illustrates the statistics in (12) . There are two parts of size one, two of size two and four of size five. Thus ℓ ( f ) = 8 \ell(f)=8 ℓ ( f ) = 8 , ∣ f ∣ p r = 26 |f|_{\mathrm{pr}}=26 ∣ f ∣ pr = 26 , and this array contributes z 8 q 26 z^8q^{26} z 8 q 26 to Z k ( z , q ) Z_k(z,q) Z k ( z , q ) .
We now outline the proof of Theorem 2, where n ≥ 2 n\geq2 n ≥ 2 . At level one, Dousse and Konan obtained the required enumeration from perfect crystals and partition bijections [12 Theorem 1.15] . At arbitrary positive level, Primc and Trupčević proved the basis theorem for the vacuum module L ( k Λ 0 ) L(k\Lambda_0) L ( k Λ 0 ) [29 Theorem 2.1] . Here we allow every weight Λ = ∑ i = 0 n k i Λ i \Lambda=\sum_{i=0}^{n}k_i\Lambda_i Λ = ∑ i = 0 n k i Λ i with k i ≥ 0 k_i\geq0 k i ≥ 0 , and hence every choice of boundary data k k k .
For each inadmissible exponent u u u , the proof of our main theorem gives a finite straightening relation
X u v Λ = ∑ v ≻ u β ( v ) = β ( u ) c v X v v Λ . X^u v_\Lambda=\sum_{\substack{v\succ u\\ \beta(v)=\beta(u)}}c_vX^v v_\Lambda. X u v Λ = v ≻ u β ( v ) = β ( u ) ∑ c v X v v Λ .
where X v X^v X v is the ordered product of the chosen negative root space generators with exponent v v v . At level one these relations show the spanning property, and the enumeration of Dousse and Konan identifies its cardinality in each principal degree with the corresponding dimension. This proves that the family is a basis. At higher levels we decompose each admissible array into marked layers. A layer with boundary e i e_i e i gives an essential exponent for L ( Λ i ) L(\Lambda_i) L ( Λ i ) . Essential exponents add under the tensor product construction, so the admissible monomials are linearly independent at every level.
To prove spanning, we split a violating path into maximal segments on which p > 0 p>0 p > 0 . At least one segment exceeds its boundary capacity. If it reaches both outer rows, an affine annihilating field gives the required relation. Otherwise the segment lies in a finite Levi subalgebra, where we use the cardinality and integer decomposition properties of the Feigin–Fourier–Littelmann–Vinberg (FFLV) sets. In either case the relation preserves the affine weight and strictly increases the PS exponent. There are only finitely many exponents in each principal degree, so straightening terminates. Principal specialization gives Theorem 3, and keeping track of the number of factors gives Corollary 4.
Sections 2 and 3 fix the root conventions, identify PBW generators with grid vertices, and derive the straightening relations from annihilating fields. In Section 4 we shall prove the basis theorem at level one and linear independence at every level. Section 5 proves spanning for mixed boundaries by associating a straightening relation to each violated path inequality. Section 6 evaluates the character and proves the refinement by PBW degree. Section 7 explains which parts of the argument would need to change for the other CMPP families.
Realization of C n ( 1 ) C_n^{(1)} C n ( 1 ) and the ordered PBW basis
We first fix the root vectors and invariant form for the finite Lie algebra, then pass to its affine extension. These choices determine the PBW generators and the degree of the part corresponding to each generator.
The finite Lie algebra and the basis of Primc and Šikić
Let E a b E_{ab} E ab be the matrix units in Mat 2 n ( C ) \operatorname{Mat}_{2n}(\mathbb{C}) Mat 2 n ( C ) , and set
J = ( 0 I n − I n 0 ) , H i = E i i − E n + i , n + i , J=\begin{pmatrix}0&I_n\\-I_n&0\end{pmatrix},\qquad H_i=E_{ii}-E_{n+i,n+i}, J = ( 0 − I n I n 0 ) , H i = E ii − E n + i , n + i ,
g = s p 2 n ( C ) = { A ∈ Mat 2 n ( C ) : A T J + J A = 0 } . \mathfrak{g}=\mathfrak{sp}_{2n}(\mathbb{C})=\{A\in\operatorname{Mat}_{2n}(\mathbb{C}):A^{\mathsf{T}}J+JA=0\}. g = sp 2 n ( C ) = { A ∈ Mat 2 n ( C ) : A T J + J A = 0 } .
Thus A ∈ g A\in\mathfrak{g} A ∈ g has block form
( P Q R − P T ) , \begin{pmatrix}P&Q\\R&-P^{\mathsf{T}}\end{pmatrix}, ( P R Q − P T ) ,
where Q Q Q and R R R are symmetric. In particular, dim g = n ( 2 n + 1 ) \dim\mathfrak{g}=n(2n+1) dim g = n ( 2 n + 1 ) . Take the Cartan subalgebra h = ⨁ i = 1 n C H i \mathfrak{h}=\bigoplus_{i=1}^{n}\mathbb{C}H_i h = ⨁ i = 1 n C H i and define ε i ∈ h ∗ \varepsilon_i\in\mathfrak{h}^{*} ε i ∈ h ∗ by ε i ( H j ) = δ i j \varepsilon_i(H_j)=\delta_{ij} ε i ( H j ) = δ ij . A nonzero functional β ∈ h ∗ \beta\in\mathfrak{h}^{*} β ∈ h ∗ is a root if the space
g β = { x ∈ g : [ H , x ] = β ( H ) x for all H ∈ h } \mathfrak{g}_{\beta}=\{x\in\mathfrak{g}:[H,x]=\beta(H)x\text{ for all }H\in\mathfrak{h}\} g β = { x ∈ g : [ H , x ] = β ( H ) x for all H ∈ h }
is nonzero. Its nonzero elements are the root vectors of weight β \beta β . To find them explicitly, write
H = ∑ k = 1 n t k H k = diag ( t 1 , … , t n , − t 1 , … , − t n ) . H=\sum_{k=1}^{n}t_kH_k=\operatorname{diag}(t_1,\ldots,t_n,-t_1,\ldots,-t_n). H = k = 1 ∑ n t k H k = diag ( t 1 , … , t n , − t 1 , … , − t n ) .
Since [ diag ( d 1 , … , d 2 n ) , E i j ] = ( d i − d j ) E i j [\operatorname{diag}(d_1,\ldots,d_{2n}),E_{ij}]=(d_i-d_j)E_{ij} [ diag ( d 1 , … , d 2 n ) , E ij ] = ( d i − d j ) E ij , we obtain
[ H , E i j ] = ( t i − t j ) E i j , [ H , E n + i , n + j ] = ( t i − t j ) E n + i , n + j , [ H , E i , n + j ] = ( t i + t j ) E i , n + j , [ H , E n + i , j ] = ( − t i − t j ) E n + i , j ; (13) \begin{aligned}
[H,E_{ij}]&=(t_i-t_j)E_{ij}, & [H,E_{n+i,n+j}]&=(t_i-t_j)E_{n+i,n+j},\\
[H,E_{i,n+j}]&=(t_i+t_j)E_{i,n+j}, & [H,E_{n+i,j}]&=(-t_i-t_j)E_{n+i,j}; \tag*{(13)}
\end{aligned} [ H , E ij ] [ H , E i , n + j ] = ( t i − t j ) E ij , = ( t i + t j ) E i , n + j , [ H , E n + i , n + j ] [ H , E n + i , j ] = ( t i − t j ) E n + i , n + j , = ( − t i − t j ) E n + i , j ; (13)
Thus we may choose the following root vectors:
X ε i − ε j = E i j − E n + j , n + i ( i ≠ j ) , X ε i + ε j = E i , n + j + E j , n + i , X − ε i − ε j = E n + i , j + E n + j , i ( i < j ) , X 2 ε i = E i , n + i , X − 2 ε i = E n + i , i . (14) \begin{aligned}
X_{\varepsilon_i-\varepsilon_j} &= E_{ij}-E_{n+j,n+i} && (i\ne j),\\
X_{\varepsilon_i+\varepsilon_j} &= E_{i,n+j}+E_{j,n+i},\qquad
X_{-\varepsilon_i-\varepsilon_j}=E_{n+i,j}+E_{n+j,i} && (i<j),\\
X_{2\varepsilon_i} &= E_{i,n+i},\qquad
X_{-2\varepsilon_i}=E_{n+i,i}. \tag*{(14)}
\end{aligned} X ε i − ε j X ε i + ε j X 2 ε i = E ij − E n + j , n + i = E i , n + j + E j , n + i , X − ε i − ε j = E n + i , j + E n + j , i = E i , n + i , X − 2 ε i = E n + i , i . ( i = j ) , ( i < j ) , (14)
The subscript of X β X_\beta X β denotes its root. These vectors give the root space decomposition
g = h ⊕ ⨁ β ∈ R C X β , R = { ± ( ε i ± ε j ) : i < j } ∪ { ± 2 ε i : 1 ≤ i ≤ n } . \mathfrak{g}=\mathfrak{h}\oplus\bigoplus_{\beta\in R}\mathbb{C}X_\beta,\qquad
R=\{\pm(\varepsilon_i\pm\varepsilon_j):i<j\}\cup\{\pm2\varepsilon_i:1\le i\le n\}. g = h ⊕ β ∈ R ⨁ C X β , R = { ± ( ε i ± ε j ) : i < j } ∪ { ± 2 ε i : 1 ≤ i ≤ n } .
The weight space of weight zero is h \mathfrak{h} h , and each root space has dimension one. We use the same root conventions as Primc and Šikić [28 §5] .
Choose simple roots
α i = ε i − ε i + 1 ( 1 ≤ i < n ) , α n = 2 ε n . (15) \alpha_i=\varepsilon_i-\varepsilon_{i+1}\quad(1\le i<n),\qquad\alpha_n=2\varepsilon_n. \tag*{(15)} α i = ε i − ε i + 1 ( 1 ≤ i < n ) , α n = 2 ε n . (15)
The positive roots are ε i − ε j \varepsilon_i-\varepsilon_j ε i − ε j and ε i + ε j \varepsilon_i+\varepsilon_j ε i + ε j for i < j i<j i < j , together with 2 ε i 2\varepsilon_i 2 ε i . Write R + R_+ R + for this set. In terms of simple roots, we have
ε i − ε j = α i + ⋯ + α j − 1 , ε i + ε j = α i + ⋯ + α j − 1 + 2 α j + ⋯ + 2 α n − 1 + α n , 2 ε i = 2 α i + ⋯ + 2 α n − 1 + α n . (16) \begin{aligned}
\varepsilon_i-\varepsilon_j&=\alpha_i+\cdots+\alpha_{j-1},\\
\varepsilon_i+\varepsilon_j&=\alpha_i+\cdots+\alpha_{j-1}+2\alpha_j+\cdots+2\alpha_{n-1}+\alpha_n,\\
2\varepsilon_i&=2\alpha_i+\cdots+2\alpha_{n-1}+\alpha_n. \tag*{(16)}
\end{aligned} ε i − ε j ε i + ε j 2 ε i = α i + ⋯ + α j − 1 , = α i + ⋯ + α j − 1 + 2 α j + ⋯ + 2 α n − 1 + α n , = 2 α i + ⋯ + 2 α n − 1 + α n . (16)
Thus the highest root is
θ = 2 ε 1 = 2 α 1 + ⋯ + 2 α n − 1 + α n . \theta=2\varepsilon_1=2\alpha_1+\cdots+2\alpha_{n-1}+\alpha_n. θ = 2 ε 1 = 2 α 1 + ⋯ + 2 α n − 1 + α n .
We use the invariant bilinear form ( x , y ) = tr ( x y ) (x,y)=\operatorname{tr}(xy) ( x , y ) = tr ( x y ) . Since ( H i , H j ) = 2 δ i j (H_i,H_j)=2\delta_{ij} ( H i , H j ) = 2 δ ij , the induced identification ι : h ∗ → h \iota:\mathfrak{h}^{*}\to\mathfrak{h} ι : h ∗ → h sends ε i \varepsilon_i ε i to H i / 2 H_i/2 H i /2 . Thus ( ε i , ε j ) = δ i j / 2 (\varepsilon_i,\varepsilon_j)=\delta_{ij}/2 ( ε i , ε j ) = δ ij /2 . For each root β \beta β , its coroot is β ∨ = 2 ι ( β ) / ( β , β ) \beta^\vee=2\iota(\beta)/(\beta,\beta) β ∨ = 2 ι ( β ) / ( β , β ) , so
( ε i − ε j ) ∨ = H i − H j , ( ε i + ε j ) ∨ = H i + H j , ( 2 ε i ) ∨ = H i . (\varepsilon_i-\varepsilon_j)^\vee=H_i-H_j,\qquad
(\varepsilon_i+\varepsilon_j)^\vee=H_i+H_j,\qquad
(2\varepsilon_i)^\vee=H_i. ( ε i − ε j ) ∨ = H i − H j , ( ε i + ε j ) ∨ = H i + H j , ( 2 ε i ) ∨ = H i .
In particular, h i : = α i ∨ = H i − H i + 1 h_i:=\alpha_i^\vee=H_i-H_{i+1} h i := α i ∨ = H i − H i + 1 for 1 ≤ i < n 1\le i<n 1 ≤ i < n , and h n = H n h_n=H_n h n = H n . The long roots have squared length 2 and the short roots have squared length 1. With the root vectors in (2.1), we obtain
[ X β , X − β ] = β ∨ , ( X β , X − β ) = 2 ( β , β ) . [X_\beta,X_{-\beta}]=\beta^\vee,\qquad
(X_\beta,X_{-\beta})=\frac{2}{(\beta,\beta)}. [ X β , X − β ] = β ∨ , ( X β , X − β ) = ( β , β ) 2 .
The fundamental weights, characterized by ω i ( h j ) = δ i j \omega_i(h_j)=\delta_{ij} ω i ( h j ) = δ ij , are
ω i = ε 1 + ⋯ + ε i ( 1 ≤ i ≤ n ) . (17) \omega_i=\varepsilon_1+\cdots+\varepsilon_i\quad(1\le i\le n). \tag*{(17)} ω i = ε 1 + ⋯ + ε i ( 1 ≤ i ≤ n ) . (17)
We now label the root vectors and the Cartan basis using the alphabet 1 , … , n , n ˉ , … , 1 ˉ 1,\ldots,n,\bar n,\ldots,\bar1 1 , … , n , n ˉ , … , 1 ˉ of Primc and Šikić [28 §5] . Number its positions 1 , … , N 1,\ldots,N 1 , … , N , where N = 2 n N=2n N = 2 n , and set
η a = { ε a , 1 ≤ a ≤ n , − ε N + 1 − a , n < a ≤ N . \eta_a=
\begin{cases}
\varepsilon_a, & 1\le a\le n,\\
-\varepsilon_{N+1-a}, & n<a\le N.
\end{cases} η a = { ε a , − ε N + 1 − a , 1 ≤ a ≤ n , n < a ≤ N .
For 1 ≤ a ≤ b ≤ N 1\le a\le b\le N 1 ≤ a ≤ b ≤ N , define
X a b = { X η a + η b , a + b ≠ N + 1 , h a , a + b = N + 1. (18) X_{ab}=
\begin{cases}
X_{\eta_a+\eta_b}, & a+b\ne N+1,\\
h_a, & a+b=N+1.
\end{cases} \tag*{(18)} X ab = { X η a + η b , h a , a + b = N + 1 , a + b = N + 1. (18)
In the second case a ≤ n a \le n a ≤ n . The set B = { X a b : 1 ≤ a ≤ b ≤ N } \mathcal{B} = \{X_{ab}: 1 \le a \le b \le N\} B = { X ab : 1 ≤ a ≤ b ≤ N } consists of one vector for each root and the n n n Cartan elements h 1 , … , h n h_1,\ldots,h_n h 1 , … , h n . It is therefore a basis of g \mathfrak{g} g , and each X a b X_{ab} X ab has weight η a + η b \eta_a+\eta_b η a + η b .
Following [28 §4] , we call X a b X_{ab} X ab the colour of the loop generator X a b ( m ) X_{ab}(m) X ab ( m ) , whose mode is m m m . We refer to B \mathcal{B} B as the colour basis . Proposition 3.1 will relate these colours and modes to the rows and part sizes in the CMPP array.
For example, in rank two the basis takes the following form.
Table 2.6.
The affine Lie algebra of type C n ( 1 ) C_n^{(1)} C n ( 1 )
For x ∈ g x \in\mathfrak{g} x ∈ g and m ∈ Z m \in\mathbb{Z} m ∈ Z , write x ( m ) = x ⊗ t m x(m)=x\otimes t^m x ( m ) = x ⊗ t m and call m m m its mode. Adjoining the central element c c c and the degree derivation d d d gives the affine Lie algebra and its Cartan subalgebra:
g ^ = ( g ⊗ C [ t , t − 1 ] ) ⊕ C c ⊕ C d , h ^ = h ⊕ C c ⊕ C d . \widehat{\mathfrak{g}}=(\mathfrak{g}\otimes\mathbb{C}[t,t^{-1}])\oplus\mathbb{C}c\oplus\mathbb{C}d,\qquad\widehat{\mathfrak{h}}=\mathfrak{h}\oplus\mathbb{C}c\oplus\mathbb{C}d. g = ( g ⊗ C [ t , t − 1 ]) ⊕ C c ⊕ C d , h = h ⊕ C c ⊕ C d .
with bracket
[ x ( r ) , y ( s ) ] = [ x , y ] ( r + s ) + r δ r + s , 0 ( x , y ) c , [ d , x ( r ) ] = r x ( r ) , [ c , g ^ ] = 0. (19) [x(r),y(s)]=[x,y](r+s)+r\delta_{r+s,0}(x,y)c,\qquad[d,x(r)]=rx(r),\qquad[c,\widehat{\mathfrak{g}}]=0. \tag*{(19)} [ x ( r ) , y ( s )] = [ x , y ] ( r + s ) + r δ r + s , 0 ( x , y ) c , [ d , x ( r )] = r x ( r ) , [ c , g ] = 0. (19)
We follow the untwisted loop realization of [28 §2] (see also [20 Chapter 7] ). Extend each finite weight by zero on c c c and d d d . Let δ \delta δ vanish on h ⊕ C c \mathfrak{h}\oplus\mathbb{C}c h ⊕ C c and satisfy δ ( d ) = 1 \delta(d)=1 δ ( d ) = 1 . For H ^ = H + u c + v d \widehat{H}=H+uc+vd H = H + u c + v d , with H ∈ h H\in\mathfrak{h} H ∈ h , we have
[ H ^ , X β ( m ) ] = [ H , X β ( m ) ] + u [ c , X β ( m ) ] + v [ d , X β ( m ) ] = ( β ( H ) + v m ) X β ( m ) = ( β + m δ ) ( H ^ ) X β ( m ) . \begin{aligned}
[\widehat{H},X_\beta(m)]&=[H,X_\beta(m)]+u[c,X_\beta(m)]+v[d,X_\beta(m)]\\
&=(\beta(H)+vm)X_\beta(m)\\
&=(\beta+m\delta)(\widehat{H})X_\beta(m).
\end{aligned} [ H , X β ( m )] = [ H , X β ( m )] + u [ c , X β ( m )] + v [ d , X β ( m )] = ( β ( H ) + v m ) X β ( m ) = ( β + m δ ) ( H ) X β ( m ) .
Thus X β ( m ) X_\beta(m) X β ( m ) has weight β + m δ \beta+m\delta β + m δ . The roots of this form, with β ∈ R \beta\in R β ∈ R and m ∈ Z m\in\mathbb{Z} m ∈ Z , are the real roots , and their root spaces have dimension one. The remaining roots are the imaginary roots m δ m\delta m δ for m ≠ 0 m\ne0 m = 0 [20 Chapters 5 and 7] . The root space of m δ m\delta m δ is h ⊗ t m \mathfrak{h}\otimes t^m h ⊗ t m , of dimension n n n . At mode zero, h ⊗ t 0 \mathfrak{h}\otimes t^0 h ⊗ t 0 belongs to the Cartan subalgebra.
The additional simple root and the Chevalley generators are
α 0 = δ − θ , e 0 = X − θ ( 1 ) , f 0 = X θ ( − 1 ) , e i = X α i ( 0 ) , f i = X − α i ( 0 ) ( 1 ≤ i ≤ n ) . \alpha_0=\delta-\theta,\qquad e_0=X_{-\theta}(1),\qquad f_0=X_\theta(-1),\qquad e_i=X_{\alpha_i}(0),\qquad f_i=X_{-\alpha_i}(0)\quad(1\le i\le n). α 0 = δ − θ , e 0 = X − θ ( 1 ) , f 0 = X θ ( − 1 ) , e i = X α i ( 0 ) , f i = X − α i ( 0 ) ( 1 ≤ i ≤ n ) .
Since θ \theta θ is long, ( X − θ , X θ ) = 1 (X_{-\theta},X_\theta)=1 ( X − θ , X θ ) = 1 , and therefore
h 0 = [ e 0 , f 0 ] = [ X − θ , X θ ] ( 0 ) + ( X − θ , X θ ) c = c − H 1 = c − ∑ i = 1 n h i . h_0=[e_0,f_0]=[X_{-\theta},X_\theta](0)+(X_{-\theta},X_\theta)c=c-H_1=c-\sum_{i=1}^{n}h_i. h 0 = [ e 0 , f 0 ] = [ X − θ , X θ ] ( 0 ) + ( X − θ , X θ ) c = c − H 1 = c − i = 1 ∑ n h i .
Consequently
c = h 0 + h 1 + ⋯ + h n , δ = α 0 + 2 α 1 + ⋯ + 2 α n − 1 + α n . (20) c=h_0+h_1+\cdots+h_n,\qquad\delta=\alpha_0+2\alpha_1+\cdots+2\alpha_{n-1}+\alpha_n. \tag*{(20)} c = h 0 + h 1 + ⋯ + h n , δ = α 0 + 2 α 1 + ⋯ + 2 α n − 1 + α n . (20)
The affine fundamental weights are uniquely determined by
Λ i ( h j ) = δ i j ( 0 ≤ i , j ≤ n ) , Λ i ( d ) = 0. (21) \Lambda_i(h_j)=\delta_{ij}\quad(0\le i,j\le n),\qquad\Lambda_i(d)=0. \tag*{(21)} Λ i ( h j ) = δ ij ( 0 ≤ i , j ≤ n ) , Λ i ( d ) = 0. (21)
On h \mathfrak{h} h , the weight Λ 0 \Lambda_0 Λ 0 restricts to zero and Λ i \Lambda_i Λ i restricts to ω i \omega_i ω i for i > 0 i>0 i > 0 . By (20) , every Λ i \Lambda_i Λ i has level Λ i ( c ) = 1 \Lambda_i(c)=1 Λ i ( c ) = 1 . Thus the boundary parameters k i k_i k i determine the weight
Λ = ∑ i = 0 n k i Λ i , k i ∈ Z ≥ 0 , k = Λ ( c ) = ∑ i = 0 n k i . (22) \Lambda=\sum_{i=0}^{n}k_i\Lambda_i,\qquad k_i\in\mathbb{Z}_{\geq0},\qquad k=\Lambda(c)=\sum_{i=0}^{n}k_i. \tag*{(22)} Λ = i = 0 ∑ n k i Λ i , k i ∈ Z ≥ 0 , k = Λ ( c ) = i = 0 ∑ n k i . (22)
The integers k i = Λ ( h i ) ≥ 0 k_i=\Lambda(h_i)\geq0 k i = Λ ( h i ) ≥ 0 make Λ \Lambda Λ dominant integral. The normalization Λ ( d ) = 0 \Lambda(d)=0 Λ ( d ) = 0 fixes the freedom to add a multiple of δ \delta δ . For 1 ≤ r ≤ n 1\leq r\leq n 1 ≤ r ≤ n , put
λ r = Λ ( H r ) = Λ ( ∑ i = r n h i ) = ∑ i = r n k i , Λ ∣ h = ∑ i = 1 n k i ω i = ∑ r = 1 n λ r ε r . \lambda_r=\Lambda(H_r)=\Lambda\left(\sum_{i=r}^{n}h_i\right)=\sum_{i=r}^{n}k_i,\qquad\left.\Lambda\right|_{\mathfrak{h}}=\sum_{i=1}^{n}k_i\omega_i=\sum_{r=1}^{n}\lambda_r\varepsilon_r. λ r = Λ ( H r ) = Λ ( i = r ∑ n h i ) = i = r ∑ n k i , Λ ∣ h = i = 1 ∑ n k i ω i = r = 1 ∑ n λ r ε r .
For a real root γ = β + m δ \gamma=\beta+m\delta γ = β + m δ , the bracket in (19) gives
γ ∨ = [ X β ( m ) , X − β ( − m ) ] = β ∨ + 2 m ( β , β ) c . (23) \gamma^\vee=[X_\beta(m),X_{-\beta}(-m)]=\beta^\vee+\frac{2m}{(\beta,\beta)}c. \tag*{(23)} γ ∨ = [ X β ( m ) , X − β ( − m )] = β ∨ + ( β , β ) 2 m c . (23)
Writing β = ∑ r = 1 n b r ε r \beta=\sum_{r=1}^{n}b_r\varepsilon_r β = ∑ r = 1 n b r ε r , we have β ∨ = ( ∑ r = 1 n b r H r ) / ( β , β ) \beta^\vee=(\sum_{r=1}^{n}b_rH_r)/(\beta,\beta) β ∨ = ( ∑ r = 1 n b r H r ) / ( β , β ) . Applying Λ \Lambda Λ to (23) therefore gives
⟨ Λ , γ ∨ ⟩ : = Λ ( γ ∨ ) = ∑ r = 1 n b r λ r + 2 m k ( β , β ) . (24) \langle\Lambda,\gamma^\vee\rangle:=\Lambda(\gamma^\vee)=\frac{\sum_{r=1}^{n}b_r\lambda_r+2mk}{(\beta,\beta)}. \tag*{(24)} ⟨ Λ , γ ∨ ⟩ := Λ ( γ ∨ ) = ( β , β ) ∑ r = 1 n b r λ r + 2 mk . (24)
For β = 2 ε a \beta=2\varepsilon_a β = 2 ε a , this pairing is λ a + m k \lambda_a+mk λ a + mk . For β = ε a ± ε b \beta=\varepsilon_a\pm\varepsilon_b β = ε a ± ε b with a ≠ b a\ne b a = b , it is λ a ± λ b + 2 m k \lambda_a\pm\lambda_b+2mk λ a ± λ b + 2 mk .
Highest weight modules and PBW monomials
Let n ± = ⨁ β ∈ R + g ± β \mathfrak{n}_{\pm}=\bigoplus_{\beta\in R_+}\mathfrak{g}_{\pm\beta} n ± = ⨁ β ∈ R + g ± β . We use the triangular decomposition
n ^ − = n − ⊕ ( g ⊗ t − 1 C [ t − 1 ] ) , \widehat{\mathfrak{n}}_-=\mathfrak{n}_-\oplus(\mathfrak{g}\otimes t^{-1}\mathbb{C}[t^{-1}]), n − = n − ⊕ ( g ⊗ t − 1 C [ t − 1 ]) ,
n ^ + = n + ⊕ ( g ⊗ t C [ t ] ) , \widehat{\mathfrak{n}}_+=\mathfrak{n}_+\oplus(\mathfrak{g}\otimes t\mathbb{C}[t]), n + = n + ⊕ ( g ⊗ t C [ t ]) ,
g ^ = n ^ − ⊕ h ^ ⊕ n ^ + . (25) \widehat{\mathfrak{g}}=\widehat{\mathfrak{n}}_-\oplus\widehat{\mathfrak{h}}\oplus\widehat{\mathfrak{n}}_+. \tag*{(25)} g = n − ⊕ h ⊕ n + . (25)
The irreducible highest weight module L ( Λ ) L(\Lambda) L ( Λ ) is generated by a vector v Λ v_\Lambda v Λ satisfying n ^ + v Λ = 0 \widehat{\mathfrak{n}}_+v_\Lambda=0 n + v Λ = 0 and H v Λ = Λ ( H ) v Λ Hv_\Lambda=\Lambda(H)v_\Lambda H v Λ = Λ ( H ) v Λ for H ∈ h ^ H\in\widehat{\mathfrak{h}} H ∈ h . It is the unique irreducible quotient of the Verma module M ( Λ ) M(\Lambda) M ( Λ ) [20 Chapter 9] .
For a Lie algebra a \mathfrak{a} a , its universal enveloping algebra is
U ( a ) = T ( a ) / ⟨ x ⊗ y − y ⊗ x − [ x , y ] : x , y ∈ a ⟩ , U(\mathfrak{a})=T(\mathfrak{a})/\langle x\otimes y-y\otimes x-[x,y]:x,y\in\mathfrak{a}\rangle, U ( a ) = T ( a ) / ⟨ x ⊗ y − y ⊗ x − [ x , y ] : x , y ∈ a ⟩ ,
where T ( a ) T(\mathfrak{a}) T ( a ) is the tensor algebra and the brackets in the denominator denote the two-sided ideal generated by the displayed elements. Let F s U ( a ) = U ≤ s ( a ) F_sU(\mathfrak{a})=U_{\leq s}(\mathfrak{a}) F s U ( a ) = U ≤ s ( a ) be the span of products of at most s s s elements of a \mathfrak{a} a , with F s U ( a ) = 0 F_sU(\mathfrak{a})=0 F s U ( a ) = 0 for s < 0 s<0 s < 0 . This is the PBW filtration. We recall the basis theorem in the form used below [20 Chapters 1 and 9] . For its affine application, see [28 §4, (4.1)–(4.3)] .
Theorem 2.1 (Poincaré–Birkhoff–Witt theorem). Let ( x b ) b ∈ B (x_b)_{b\in B} ( x b ) b ∈ B be a basis of a Lie algebra a \mathfrak{a} a , with a total order on index set B B B . The products
x b 1 ⋯ x b r , r ≥ 0 , b 1 ≤ ⋯ ≤ b r , x_{b_1}\cdots x_{b_r},\qquad r\geq0,\qquad b_1\leq\cdots\leq b_r, x b 1 ⋯ x b r , r ≥ 0 , b 1 ≤ ⋯ ≤ b r , including the empty product 1 1 1 , form a basis of U ( a ) U(\mathfrak{a}) U ( a ) . The index set B B B may be infinite, although each product is finite.
In particular, the map u ↦ u v Λ u \mapsto uv_{\Lambda} u ↦ u v Λ identifies U ( n ^ − ) U(\widehat{\mathfrak n}_{-}) U ( n − ) with M ( Λ ) M(\Lambda) M ( Λ ) as a vector space, where v Λ v_{\Lambda} v Λ here denotes the Verma highest vector. The basis of n ^ − \widehat{\mathfrak n}_{-} n − induced by B \mathcal{B} B is
B − = { X a b ( − j ) : 1 ≤ a ≤ b ≤ N , j ≥ 1 } ∪ { X a b ( 0 ) : 1 ≤ a ≤ b ≤ N , a + b > N + 1 } . (26) \mathcal{B}_{-}=\{X_{ab}(-j):1\leq a\leq b\leq N,\ j\geq1\}\cup\{X_{ab}(0):1\leq a\leq b\leq N,\ a+b>N+1\}. \tag*{(26)} B − = { X ab ( − j ) : 1 ≤ a ≤ b ≤ N , j ≥ 1 } ∪ { X ab ( 0 ) : 1 ≤ a ≤ b ≤ N , a + b > N + 1 } . (26)
To check the condition at mode zero, suppose first that a ≤ n < b a\leq n<b a ≤ n < b , and put q = N + 1 − b q=N+1-b q = N + 1 − b . Then
η a + η b = ε a − ε q , ε a − ε q ∈ R + ⟺ a < q ⟺ a + b < N + 1. \eta_a+\eta_b=\varepsilon_a-\varepsilon_q,\qquad\varepsilon_a-\varepsilon_q\in R_{+}\Longleftrightarrow a<q\Longleftrightarrow a+b<N+1. η a + η b = ε a − ε q , ε a − ε q ∈ R + ⟺ a < q ⟺ a + b < N + 1.
Equality a = q a=q a = q gives a Cartan case, and a > q a>q a > q gives a negative root. The remaining cases b ≤ n b\leq n b ≤ n and a > n a>n a > n give positive and negative sums of two coordinate weights, respectively. Thus, in all cases,
η a + η b ∈ R + ⟺ a + b < N + 1 , \eta_a+\eta_b\in R_{+}\Longleftrightarrow a+b<N+1, η a + η b ∈ R + ⟺ a + b < N + 1 ,
η a + η b = 0 ⟺ a + b = N + 1 , \eta_a+\eta_b=0\Longleftrightarrow a+b=N+1, η a + η b = 0 ⟺ a + b = N + 1 ,
η a + η b ∈ − R + ⟺ a + b > N + 1. \eta_a+\eta_b\in-R_{+}\Longleftrightarrow a+b>N+1. η a + η b ∈ − R + ⟺ a + b > N + 1.
A weight module is integrable if every e i e_i e i and f i f_i f i acts locally nilpotently: for each vector v v v , some power of the given operator annihilates v v v [20 Chapter 3] . An integrable highest weight module is also called a standard module .
Theorem 2.2 (Integrable highest weight presentation [20 Corollary 10.4] ). For a dominant integral weight Λ \Lambda Λ , L ( Λ ) L(\Lambda) L ( Λ ) is integrable: every e i e_i e i and f i f_i f i acts locally nilpotently. It is generated by v Λ v_{\Lambda} v Λ with the highest weight relations and
f i k i + 1 v Λ = 0 ( 0 ≤ i ≤ n ) . (27) f_i^{k_i+1}v_{\Lambda}=0\qquad(0\leq i\leq n). \tag*{(27)} f i k i + 1 v Λ = 0 ( 0 ≤ i ≤ n ) . (27) These are defining relations for L ( Λ ) L(\Lambda) L ( Λ ) . In particular, every real root vector acts locally nilpotently on L ( Λ ) L(\Lambda) L ( Λ ) [20 Chapter 3] .
Let ( e , f , h ) (e,f,h) ( e , f , h ) be the s l 2 \mathfrak{sl}_2 sl 2 triple for a real root. If e v = 0 ev=0 e v = 0 and h v = a v hv=av h v = a v for a nonzero vector v v v , then for s ≥ 1 s\geq1 s ≥ 1
h f s v = ( a − 2 s ) f s v , e f s v = s ( a − s + 1 ) f s − 1 v . (28) hf^{s}v=(a-2s)f^{s}v,\qquad ef^{s}v=s(a-s+1)f^{s-1}v. \tag*{(28)} h f s v = ( a − 2 s ) f s v , e f s v = s ( a − s + 1 ) f s − 1 v . (28)
Indeed,
[ e , f s ] v = ∑ t = 0 s − 1 f t h f s − 1 − t v = ∑ t = 0 s − 1 ( a − 2 ( s − 1 − t ) ) f s − 1 v = s ( a − s + 1 ) f s − 1 v . [e,f^{s}]v=\sum_{t=0}^{s-1}f^{t}hf^{s-1-t}v=\sum_{t=0}^{s-1}(a-2(s-1-t))f^{s-1}v=s(a-s+1)f^{s-1}v. [ e , f s ] v = t = 0 ∑ s − 1 f t h f s − 1 − t v = t = 0 ∑ s − 1 ( a − 2 ( s − 1 − t )) f s − 1 v = s ( a − s + 1 ) f s − 1 v .
If f f f is locally nilpotent on v v v , let r ≥ 1 r\geq1 r ≥ 1 be minimal with f r v = 0 f^{r}v=0 f r v = 0 . Then
0 = e f r v = r ( a − r + 1 ) f r − 1 v , f r − 1 v ≠ 0 , r = a + 1. 0=ef^{r}v=r(a-r+1)f^{r-1}v,\qquad f^{r-1}v\ne0,\qquad r=a+1. 0 = e f r v = r ( a − r + 1 ) f r − 1 v , f r − 1 v = 0 , r = a + 1.
Apply this calculation to a positive real root γ \gamma γ and v = v Λ v=v_{\Lambda} v = v Λ , with e = X γ e=X_{\gamma} e = X γ , f = X − γ f=X_{-\gamma} f = X − γ and h = γ ∨ h=\gamma^{\vee} h = γ ∨ . Since e v Λ = 0 ev_{\Lambda}=0 e v Λ = 0 and h v Λ = ⟨ Λ , γ ∨ ⟩ v Λ hv_{\Lambda}=\langle\Lambda,\gamma^{\vee}\rangle v_{\Lambda} h v Λ = ⟨ Λ , γ ∨ ⟩ v Λ , we obtain
X − γ ⟨ Λ , γ ∨ ⟩ + 1 v Λ = 0 , ⟨ Λ , γ ∨ ⟩ = 0 ⟹ X − γ v Λ = 0. (29) X_{-\gamma}^{\langle\Lambda,\gamma^{\vee}\rangle+1}v_{\Lambda}=0,\qquad\langle\Lambda,\gamma^{\vee}\rangle=0\Longrightarrow X_{-\gamma}v_{\Lambda}=0. \tag*{(29)} X − γ ⟨ Λ , γ ∨ ⟩ + 1 v Λ = 0 , ⟨ Λ , γ ∨ ⟩ = 0 ⟹ X − γ v Λ = 0. (29)
Moreover, all powers with exponent between 0 0 0 and ⟨ Λ , γ ∨ ⟩ \langle\Lambda,\gamma^{\vee}\rangle ⟨ Λ , γ ∨ ⟩ are nonzero on v Λ v_{\Lambda} v Λ . For the rank two example in (2.6), Table 1 illustrates (29) for the three simple lowering operators and one further real root.
Table 1. Lowering operators and their action on v Λ 1 v_{\Lambda_1} v Λ 1 .
Lowering weights and principal degree
For γ = ∑ i = 0 n q i α i ∈ Q + : = ∑ i = 0 n Z ≥ 0 α i \gamma=\sum_{i=0}^{n}q_i\alpha_i\in Q_+:=\sum_{i=0}^{n}\mathbb{Z}_{\geq0}\alpha_i γ = ∑ i = 0 n q i α i ∈ Q + := ∑ i = 0 n Z ≥ 0 α i , define its height by ht ( γ ) = ∑ i = 0 n q i \operatorname{ht}(\gamma)=\sum_{i=0}^{n}q_i ht ( γ ) = ∑ i = 0 n q i . We assign a nonzero vector in a negative root space g ^ − γ \widehat{\mathfrak{g}}_{-\gamma} g − γ the positive principal degree ht ( γ ) \operatorname{ht}(\gamma) ht ( γ ) .
Set
ρ ∨ = ∑ i = 1 n ( n − i + 1 2 ) H i , D p r = N d + ρ ∨ . (30) \rho^\vee=\sum_{i=1}^{n}\left(n-i+\frac{1}{2}\right)H_i,\qquad D_{\mathrm{pr}}=Nd+\rho^\vee. \tag*{(30)} ρ ∨ = i = 1 ∑ n ( n − i + 2 1 ) H i , D pr = N d + ρ ∨ . (30)
Every simple root takes the value one on D p r D_{\mathrm{pr}} D pr , since
α i ( ρ ∨ ) = { ( n − i + 1 2 ) − ( n − i − 1 + 1 2 ) = 1 , i < n , 2 ⋅ 1 2 = 1 , i = n , ( 1 ≤ i ≤ n ) , \alpha_i(\rho^\vee)=
\begin{cases}
\left(n-i+\frac{1}{2}\right)-\left(n-i-1+\frac{1}{2}\right)=1, & i<n,\\
2\cdot\frac{1}{2}=1, & i=n,
\end{cases}
\qquad(1\leq i\leq n), α i ( ρ ∨ ) = { ( n − i + 2 1 ) − ( n − i − 1 + 2 1 ) = 1 , 2 ⋅ 2 1 = 1 , i < n , i = n , ( 1 ≤ i ≤ n ) ,
while
α 0 ( D p r ) = N − θ ( ρ ∨ ) = N − 2 ( n − 1 2 ) = 1. \alpha_0(D_{\mathrm{pr}})=N-\theta(\rho^\vee)=N-2\left(n-\frac{1}{2}\right)=1. α 0 ( D pr ) = N − θ ( ρ ∨ ) = N − 2 ( n − 2 1 ) = 1.
Consequently, ht ( γ ) = γ ( D p r ) \operatorname{ht}(\gamma)=\gamma(D_{\mathrm{pr}}) ht ( γ ) = γ ( D pr ) . In particular, the weight by which X a b ( − j ) ∈ B − X_{ab}(-j)\in\mathcal{B}_{-} X ab ( − j ) ∈ B − lowers a vector is
γ a b ; j : = j δ − η a − η b ∈ Q + ∖ { 0 } . (31) \gamma_{ab;j}:=j\delta-\eta_a-\eta_b\in Q_+\setminus\{0\}. \tag*{(31)} γ ab ; j := j δ − η a − η b ∈ Q + ∖ { 0 } . (31)
Since η a ( ρ ∨ ) = ( N + 1 ) / 2 − a \eta_a(\rho^\vee)=(N+1)/2-a η a ( ρ ∨ ) = ( N + 1 ) /2 − a , the principal degree of this generator is
p a b ; j : = γ a b ; j ( D p r ) = N j + a + b − N − 1. (32) p_{ab;j}:=\gamma_{ab;j}(D_{\mathrm{pr}})=Nj+a+b-N-1. \tag*{(32)} p ab ; j := γ ab ; j ( D pr ) = N j + a + b − N − 1. (32)
In particular each simple lowering operator has degree one, while each of the n n n Cartan colours at mode − j -j − j has degree N j Nj N j .
To make all affine coordinates explicit, write η a + η b = ∑ h = 1 n u h ε h \eta_a+\eta_b=\sum_{h=1}^{n}u_h\varepsilon_h η a + η b = ∑ h = 1 n u h ε h and γ a b ; j = ∑ i = 0 n q i α i \gamma_{ab;j}=\sum_{i=0}^{n}q_i\alpha_i γ ab ; j = ∑ i = 0 n q i α i . Then
q 0 = j , q i = 2 j − ∑ h = 1 i u h ( 1 ≤ i < n ) , q n = j − 1 2 ∑ h = 1 n u h . (33) q_0=j,\qquad q_i=2j-\sum_{h=1}^{i}u_h\ (1\leq i<n),\qquad q_n=j-\frac{1}{2}\sum_{h=1}^{n}u_h. \tag*{(33)} q 0 = j , q i = 2 j − h = 1 ∑ i u h ( 1 ≤ i < n ) , q n = j − 2 1 h = 1 ∑ n u h . (33)
To verify these formulas, first use (2.2) to write
∑ h = 1 n u h ε h = ∑ i = 1 n − 1 ( ∑ h = 1 i u h ) α i + 1 2 ( ∑ h = 1 n u h ) α n . \sum_{h=1}^{n}u_h\varepsilon_h=\sum_{i=1}^{n-1}\left(\sum_{h=1}^{i}u_h\right)\alpha_i+\frac{1}{2}\left(\sum_{h=1}^{n}u_h\right)\alpha_n. h = 1 ∑ n u h ε h = i = 1 ∑ n − 1 ( h = 1 ∑ i u h ) α i + 2 1 ( h = 1 ∑ n u h ) α n .
Subtract this expression from j δ j\delta j δ and use (2.8) to obtain (2.21). Since η a + η b \eta_a+\eta_b η a + η b is the sum of two signed coordinate vectors,
∑ h = 1 n u h ∈ { − 2 , 0 , 2 } , ∑ h = 1 i u h ≤ 2 , q i ≥ 2 j − 2 ( 1 ≤ i < n ) , q n ≥ j − 1. \sum_{h=1}^{n}u_h\in\{-2,0,2\},\qquad\sum_{h=1}^{i}u_h\leq2,\qquad q_i\geq2j-2\ (1\leq i<n),\qquad q_n\geq j-1. h = 1 ∑ n u h ∈ { − 2 , 0 , 2 } , h = 1 ∑ i u h ≤ 2 , q i ≥ 2 j − 2 ( 1 ≤ i < n ) , q n ≥ j − 1.
Thus q 0 , … , q n q_0,\ldots,q_n q 0 , … , q n are nonnegative integers for j ≥ 1 j\geq1 j ≥ 1 , with q 0 > 0 q_0>0 q 0 > 0 . For j = 0 j=0 j = 0 ,
( q 0 , … , q n ) ∈ Z ≥ 0 n + 1 ∖ { 0 } ⟺ − η a − η b ∈ R + ⟺ a + b > N + 1 , (34) (q_0,\ldots,q_n)\in\mathbb{Z}_{\geq0}^{n+1}\setminus\{0\}\Longleftrightarrow-\eta_a-\eta_b\in R_+\Longleftrightarrow a+b>N+1, \tag*{(34)} ( q 0 , … , q n ) ∈ Z ≥ 0 n + 1 ∖ { 0 } ⟺ − η a − η b ∈ R + ⟺ a + b > N + 1 , (34)
which is precisely the condition on mode zero in (2.14). In every case q 0 = j q_0=j q 0 = j recovers the negative of the loop degree.
An exponent is a finitely supported map u : B − ⟶ Z ≥ 0 u:\mathcal{B}_{-}\longrightarrow\mathbb{Z}_{\geq0} u : B − ⟶ Z ≥ 0 . Write X u X^u X u for its ordered PBW product in the order of Section 3.1, and put
ℓ ( u ) = ∑ a , b , j u a b ; j , β ( u ) = ∑ a , b , j u a b ; j γ a b ; j , ∣ u ∣ p r = ht β ( u ) = ∑ a , b , j p a b ; j u a b ; j . \ell(u)=\sum_{a,b,j}u_{ab;j},\qquad\beta(u)=\sum_{a,b,j}u_{ab;j}\gamma_{ab;j},\qquad|u|_{\mathrm{pr}}=\operatorname{ht}\beta(u)=\sum_{a,b,j}p_{ab;j}u_{ab;j}. ℓ ( u ) = a , b , j ∑ u ab ; j , β ( u ) = a , b , j ∑ u ab ; j γ ab ; j , ∣ u ∣ pr = ht β ( u ) = a , b , j ∑ p ab ; j u ab ; j .
Whenever nonzero, X u v Λ X^u v_\Lambda X u v Λ has affine weight Λ − β ( u ) \Lambda-\beta(u) Λ − β ( u ) and normalized principal degree ∣ u ∣ p r |u|_{\mathrm{pr}} ∣ u ∣ pr . On L ( Λ ) L(\Lambda) L ( Λ ) this degree is the eigenvalue of Λ ( D p r ) Id − D p r \Lambda(D_{\mathrm{pr}})\operatorname{Id}-D_{\mathrm{pr}} Λ ( D pr ) Id − D pr . In particular, v Λ v_\Lambda v Λ has degree zero.
Lemma 2.3. For fixed P ≥ 0 P\geq0 P ≥ 0 , the set of exponents u u u on B − \mathcal{B}_{-} B − with ∣ u ∣ p r = P |u|_{\mathrm{pr}}=P ∣ u ∣ pr = P is finite.
Proof. For a generator in B − \mathcal{B}_{-} B − , (2.20) gives
p a b ; 0 ≥ 1 , p a b ; j ≥ N ( j − 1 ) + 1 ( j ≥ 1 ) . p_{ab;0}\geq1,\qquad p_{ab;j}\geq N(j-1)+1\qquad(j\geq1). p ab ; 0 ≥ 1 , p ab ; j ≥ N ( j − 1 ) + 1 ( j ≥ 1 ) . If P ≥ 1 P\geq1 P ≥ 1 and u u u has principal degree P P P , then u a b ; j > 0 u_{ab;j}>0 u ab ; j > 0 implies
p a b ; j ≤ P , j ≤ 1 + ⌊ P − 1 N ⌋ ( j ≥ 1 ) , 0 ≤ u a b ; j ≤ ⌊ P p a b ; j ⌋ . p_{ab;j}\leq P,\qquad j\leq1+\left\lfloor\frac{P-1}{N}\right\rfloor\ (j\geq1),\qquad0\leq u_{ab;j}\leq\left\lfloor\frac{P}{p_{ab;j}}\right\rfloor. p ab ; j ≤ P , j ≤ 1 + ⌊ N P − 1 ⌋ ( j ≥ 1 ) , 0 ≤ u ab ; j ≤ ⌊ p ab ; j P ⌋ . Only finitely many generators can occur, each with bounded multiplicity. In particular, ℓ ( u ) ≤ P \ell(u)\leq P ℓ ( u ) ≤ P . When P = 0 P=0 P = 0 , the empty exponent is the only possibility. □ \square □
It follows that principal specialization is a well-defined substitution on the normalized character e − Λ ch L ( Λ ) e^{-\Lambda}\operatorname{ch}L(\Lambda) e − Λ ch L ( Λ ) :
ps ( e − α i ) = q ( 0 ≤ i ≤ n ) , ps ( e − Λ ch L ( Λ ) ) = ∑ P ≥ 0 dim L ( Λ ) P q P . \operatorname{ps}(e^{-\alpha_i})=q\quad(0\leq i\leq n),\qquad\operatorname{ps}(e^{-\Lambda}\operatorname{ch}L(\Lambda))=\sum_{P\geq0}\dim L(\Lambda)_Pq^P. ps ( e − α i ) = q ( 0 ≤ i ≤ n ) , ps ( e − Λ ch L ( Λ )) = P ≥ 0 ∑ dim L ( Λ ) P q P .
where L ( Λ ) P = ⨁ ht γ = P L ( Λ ) Λ − γ L(\Lambda)_P=\bigoplus_{\operatorname{ht}\gamma=P}L(\Lambda)_{\Lambda-\gamma} L ( Λ ) P = ⨁ ht γ = P L ( Λ ) Λ − γ , with γ ∈ Q + \gamma\in Q_+ γ ∈ Q + . PBW spanning and Lemma 2.3 give
dim L ( Λ ) P ≤ # { u : supp u ⊂ B − , ∣ u ∣ p r = P } < ∞ . \dim L(\Lambda)_P\leq\#\{u:\operatorname{supp}u\subset\mathcal{B}_{-},\ |u|_{\mathrm{pr}}=P\}<\infty. dim L ( Λ ) P ≤ # { u : supp u ⊂ B − , ∣ u ∣ pr = P } < ∞.
Thus every coefficient of the specialized character is finite.
The grid dictionary and homogeneous straightening
We now identify the generators in (2.14) with the positive vertices of (1.2). Under this correspondence, the leading terms of Primc and Šikić give finite straightening relations indexed by paths.
The grid bijection and the order on PBW monomials
Proposition 3.1. For 1 ≤ a ≤ b ≤ N 1\leq a\leq b\leq N 1 ≤ a ≤ b ≤ N , the assignment
X a b ( − j ) ⟼ ( r , p ) , p = N j + a + b − N − 1 , r = { b − a , j even , N − b + a , j odd , (35) X_{ab}(-j)\longmapsto(r,p),\qquad p=Nj+a+b-N-1,\qquad r=\begin{cases}b-a,&j\ \text{even},\\N-b+a,&j\ \text{odd},\end{cases} \tag*{(35)} X ab ( − j ) ⟼ ( r , p ) , p = N j + a + b − N − 1 , r = { b − a , N − b + a , j even , j odd , (35) is a bijection from B − \mathcal{B}_{-} B − onto Ω + \Omega_+ Ω + . It identifies the principal degree of a generator with the size p p p of its coloured part.
Proof. The formula gives 0 ≤ r ≤ N 0 \le r \le N 0 ≤ r ≤ N and r + p ≡ 1 ( m o d 2 ) r+p \equiv1 \pmod{2} r + p ≡ 1 ( mod 2 ) . Lemma 2.3 shows that p > 0 p>0 p > 0 for each generator. Fix a row r r r . For even j j j , b = a + r b=a+r b = a + r with 1 ≤ a ≤ N − r 1 \le a \le N-r 1 ≤ a ≤ N − r , and the possible p p p form the interval
[ N j − N + r + 1 , N j + N − r − 1 ] 2 , (36) [Nj-N+r+1, Nj+N-r-1]_2, \tag*{(36)} [ N j − N + r + 1 , N j + N − r − 1 ] 2 , (36)
where [ u , v ] 2 = { u , u + 2 , … , v } [u,v]_2=\{u,u+2,\ldots,v\} [ u , v ] 2 = { u , u + 2 , … , v } when the endpoints have the same parity, and the interval is empty if u > v u>v u > v . At j = 0 j=0 j = 0 we retain only positive values of p p p . For odd j j j , we have b = a + N − r b=a+N-r b = a + N − r with 1 ≤ a ≤ r 1 \le a \le r 1 ≤ a ≤ r , and the corresponding interval is
[ N j − r + 1 , N j + r − 1 ] 2 . (37) [Nj-r+1, Nj+r-1]_2. \tag*{(37)} [ N j − r + 1 , N j + r − 1 ] 2 . (37)
Denote these intervals by I j ( r ) I_j(r) I j ( r ) . For 0 < r < N 0<r<N 0 < r < N , their endpoints satisfy
min I 2 t + 1 ( r ) − max I 2 t ( r ) = 2 , min I 2 t + 2 ( r ) − max I 2 t + 1 ( r ) = 2. \min I_{2t+1}(r)-\max I_{2t}(r)=2,\qquad\min I_{2t+2}(r)-\max I_{2t+1}(r)=2. min I 2 t + 1 ( r ) − max I 2 t ( r ) = 2 , min I 2 t + 2 ( r ) − max I 2 t + 1 ( r ) = 2.
Thus the intervals alternate without overlap or gaps in the prescribed parity class, and
⋃ j ≥ 0 ( I j ( r ) ∩ Z > 0 ) = { p ∈ Z > 0 : p + r ≡ 1 ( m o d 2 ) } . \bigcup_{j\geq0}\left(I_j(r)\cap\mathbb{Z}_{>0}\right)=\{p\in\mathbb{Z}_{>0}:p+r\equiv1\pmod{2}\}. j ≥ 0 ⋃ ( I j ( r ) ∩ Z > 0 ) = { p ∈ Z > 0 : p + r ≡ 1 ( mod 2 )} .
At r = 0 r=0 r = 0 only even j j j occur, and at r = N r=N r = N only odd j j j occur. In either case the successive nonempty intervals satisfy min I j + 2 ( r ) − max I j ( r ) = 2 \min I_{j+2}(r)-\max I_j(r)=2 min I j + 2 ( r ) − max I j ( r ) = 2 . At mode zero,
p = a + b − N − 1 > 0 ⟺ a + b > N + 1. p=a+b-N-1>0\quad\Longleftrightarrow\quad a+b>N+1. p = a + b − N − 1 > 0 ⟺ a + b > N + 1.
The inverse is as follows. Given ( r , p ) ∈ Ω + (r,p)\in\Omega_+ ( r , p ) ∈ Ω + , set
t = ⌊ p + N − r − 1 2 N ⌋ , j = { 2 t , p ≤ 2 N t + N − r − 1 , 2 t + 1 , p > 2 N t + N − r − 1 , (38) t=\left\lfloor\frac{p+N-r-1}{2N}\right\rfloor,\qquad
j=
\begin{cases}
2t, & p\leq2Nt+N-r-1,\\
2t+1, & p>2Nt+N-r-1,
\end{cases} \tag*{(38)} t = ⌊ 2 N p + N − r − 1 ⌋ , j = { 2 t , 2 t + 1 , p ≤ 2 N t + N − r − 1 , p > 2 N t + N − r − 1 , (38)
s = p − N j + N + 1 , d r = { r , j even , N − r , j odd , a = s − d r 2 , b = s + d r 2 . s=p-Nj+N+1,\qquad
d_r=
\begin{cases}
r, & j\text{ even},\\
N-r, & j\text{ odd},
\end{cases}
\qquad
a=\frac{s-d_r}{2},\qquad b=\frac{s+d_r}{2}. s = p − N j + N + 1 , d r = { r , N − r , j even , j odd , a = 2 s − d r , b = 2 s + d r .
Indeed, the defining inequality for t t t is
2 N t − N + r + 1 ≤ p < 2 N ( t + 1 ) − N + r + 1. 2Nt-N+r+1\leq p<2N(t+1)-N+r+1. 2 N t − N + r + 1 ≤ p < 2 N ( t + 1 ) − N + r + 1.
The threshold in (3.4) selects the even interval I 2 t ( r ) I_{2t}(r) I 2 t ( r ) or the following odd interval I 2 t + 1 ( r ) I_{2t+1}(r) I 2 t + 1 ( r ) . Since p + r p+r p + r is odd, s ≡ d r ( m o d 2 ) s\equiv d_r\pmod{2} s ≡ d r ( mod 2 ) . The endpoints of the selected interval in (3.2) and (3.3) give 1 ≤ ( s − d r ) / 2 ≤ ( s + d r ) / 2 ≤ N 1\leq(s-d_r)/2\leq(s+d_r)/2\leq N 1 ≤ ( s − d r ) /2 ≤ ( s + d r ) /2 ≤ N . Thus a a a and b b b are integers in the required range, and when j = 0 j=0 j = 0 they satisfy a + b > N + 1 a+b>N+1 a + b > N + 1 .
For n = 2 n=2 n = 2 , the generators of mode 0 and mode − 1 -1 − 1 are as follows:
Table 3.5.
This bijection identifies a frequency function f f f on Ω + \Omega_+ Ω + with the exponent u ( f ) u(f) u ( f ) defined by u ( f ) ( X a b ( − j ) ) = f ( r , p ) u(f)(X_{ab}(-j))=f(r,p) u ( f ) ( X ab ( − j )) = f ( r , p ) for the pair ( r , p ) (r,p) ( r , p ) in (3.1). Thus the size of the partition is the principal degree of the monomial, and the number of parts is its PBW length.
We next order these monomials. On the finite colour basis we use the order of [28 §5, (5.1)] :
X a b ≺ X a ′ b ′ ⟺ b > b ′ or ( b = b ′ and a > a ′ ) . (39) X_{ab} \prec X_{a'b'} \Longleftrightarrow b>b' \text{ or } (b=b' \text{ and } a>a'). \tag*{(39)} X ab ≺ X a ′ b ′ ⟺ b > b ′ or ( b = b ′ and a > a ′ ) . (39)
Extend this order to the loop generators by setting
X ( m ) ≺ Y ( m ′ ) ⟺ m < m ′ or ( m = m ′ and X ≺ Y ) . X^{(m)} \prec Y^{(m')} \Longleftrightarrow m<m' \text{ or } (m=m' \text{ and } X\prec Y). X ( m ) ≺ Y ( m ′ ) ⟺ m < m ′ or ( m = m ′ and X ≺ Y ) .
Let P \mathcal{P} P be the additive monoid of finitely supported nonnegative exponents on the generators X ( m ) X^{(m)} X ( m ) , where X ∈ B X\in\mathcal{B} X ∈ B and m ∈ Z m\in\mathbb{Z} m ∈ Z . An exponent u ∈ P u\in\mathcal{P} u ∈ P with length ℓ ( u ) \ell(u) ℓ ( u ) has a unique nondecreasing list
X 1 ( m 1 ) ⪯ ⋯ ⪯ X ℓ ( u ) ( m ℓ ( u ) ) . X_1^{(m_1)}\preceq\cdots\preceq X_{\ell(u)}^{(m_{\ell(u)})}. X 1 ( m 1 ) ⪯ ⋯ ⪯ X ℓ ( u ) ( m ℓ ( u ) ) .
Its shape is ( m 1 , … , m ℓ ( u ) ) (m_1,\ldots,m_{\ell(u)}) ( m 1 , … , m ℓ ( u ) ) and its loop degree is deg l o o p u = ∑ i m i \operatorname{deg}_{\mathrm{loop}}u=\sum_i m_i deg loop u = ∑ i m i . In particular − j -j − j is the loop degree of X a b ( − j ) X_{ab}^{(-j)} X ab ( − j ) . The order of Primc and Šikić (the PS order) on P \mathcal{P} P is determined by the following successive comparisons [28 §6] .
The exponent of greater length is smaller.
At equal length, the exponent of smaller loop degree is smaller.
At equal length and loop degree, read the shapes from right to left. The smaller mode at the first difference gives the smaller exponent.
At equal shape, compare the corresponding colour lists from right to left, using (39) at the first unequal position.
We denote the ordered product by X u X^u X u and write u ≺ v u\prec v u ≺ v when X u ≺ X v X^u\prec X^v X u ≺ X v . For example, ( − 2 , − 2 ) ≺ ( − 3 , − 1 ) (-2,-2)\prec(-3,-1) ( − 2 , − 2 ) ≺ ( − 3 , − 1 ) as shapes. At a fixed mode in rank two, the colour comparison gives X 34 X 12 ≺ X 44 X 11 X_{34}X_{12}\prec X_{44}X_{11} X 34 X 12 ≺ X 44 X 11 , since X 12 ≺ X 11 X_{12}\prec X_{11} X 12 ≺ X 11 at the rightmost position.
Lemma 3.2. The order is strictly compatible with addition: u ≺ v u\prec v u ≺ v implies u + w ≺ v + w u+w\prec v+w u + w ≺ v + w for all u , v , w ∈ P u,v,w\in\mathcal{P} u , v , w ∈ P . Reordering a finite word of exponent u u u gives X u X^u X u plus shorter terms, all of whose exponents are strictly larger than u u u . Among exponents supported on B − \mathcal{B}_{-} B − , every sequence of strict increases at a fixed principal degree terminates.
Proof. Write L ( u ) = ℓ ( u ) L(u)=\ell(u) L ( u ) = ℓ ( u ) , M ( u ) = deg l o o p u M(u)=\operatorname{deg}_{\mathrm{loop}}u M ( u ) = deg loop u , and
d m ( u ) = ∑ X ∈ B u ( X ( m ) ) . d_m(u)=\sum_{X\in\mathcal{B}}u\bigl(X^{(m)}\bigr). d m ( u ) = X ∈ B ∑ u ( X ( m ) ) .
The first two rules are preserved because
L ( u + w ) − L ( v + w ) = L ( u ) − L ( v ) , M ( u + w ) − M ( v + w ) = M ( u ) − M ( v ) . L(u+w)-L(v+w)=L(u)-L(v),\qquad M(u+w)-M(v+w)=M(u)-M(v). L ( u + w ) − L ( v + w ) = L ( u ) − L ( v ) , M ( u + w ) − M ( v + w ) = M ( u ) − M ( v ) .
Suppose that length and loop degree agree but the shapes differ, and put m ∗ = max { m : d m ( u ) ≠ d m ( v ) } m_*=\max\{m:d_m(u)\ne d_m(v)\} m ∗ = max { m : d m ( u ) = d m ( v )} . Reading the sorted mode lists from the right gives
u ≺ v ⟺ d m ∗ ( u ) < d m ∗ ( v ) . u\prec v\Longleftrightarrow d_{m_*}(u)<d_{m_*}(v). u ≺ v ⟺ d m ∗ ( u ) < d m ∗ ( v ) .
For every m m m ,
d m ( u + w ) − d m ( v + w ) = d m ( u ) − d m ( v ) , d_m(u+w)-d_m(v+w)=d_m(u)-d_m(v), d m ( u + w ) − d m ( v + w ) = d m ( u ) − d m ( v ) ,
so both m ∗ m_* m ∗ and this inequality are unchanged. Finally, if the shapes agree, let z ∗ z_* z ∗ be the largest generator X ( m ) X^{(m)} X ( m ) , in the order ≺ \prec ≺ , at which the exponents differ. The colour comparison is
u ≺ v ⟺ u ( z ∗ ) < v ( z ∗ ) , ( u + w ) ( z ) − ( v + w ) ( z ) = u ( z ) − v ( z ) . u\prec v\Longleftrightarrow u(z_*)<v(z_*),\qquad(u+w)(z)-(v+w)(z)=u(z)-v(z). u ≺ v ⟺ u ( z ∗ ) < v ( z ∗ ) , ( u + w ) ( z ) − ( v + w ) ( z ) = u ( z ) − v ( z ) .
The shapes still agree after adding w w w , and z ∗ z_\ast z ∗ is unchanged. This proves strict compatibility in all four cases.
Let
U k = U ( ( g ⊗ C [ t , t − 1 ] ) ⊕ C c ) / ⟨ c − k 1 ⟩ . U_k = U\left((\mathfrak{g} \otimes\mathbb{C}[t,t^{-1}]) \oplus\mathbb{C}c\right)/\langle c-k1\rangle. U k = U ( ( g ⊗ C [ t , t − 1 ]) ⊕ C c ) / ⟨ c − k 1 ⟩ .
For s ≥ 0 s \ge0 s ≥ 0 , define the PBW filtration by
F s U k : = span C { x 1 ( m 1 ) ⋯ x r ( m r ) : 0 ≤ r ≤ s , x i ∈ g , m i ∈ Z } , F − 1 U k : = 0. F_sU_k := \operatorname{span}_{\mathbb{C}}\{x_1(m_1)\cdots x_r(m_r): 0 \le r \le s,\ x_i \in\mathfrak{g},\ m_i \in\mathbb{Z}\},\qquad F_{-1}U_k := 0. F s U k := span C { x 1 ( m 1 ) ⋯ x r ( m r ) : 0 ≤ r ≤ s , x i ∈ g , m i ∈ Z } , F − 1 U k := 0.
The products are taken in U k U_k U k , and the empty product (r = 0 r=0 r = 0 ) is 1.
Now let W W W be a word of exponent u u u and length L L L . If x ( r ) , y ( s ) x(r),y(s) x ( r ) , y ( s ) are adjacent factors in the wrong order, write W = W 1 x ( r ) y ( s ) W 2 W=W_1x(r)y(s)W_2 W = W 1 x ( r ) y ( s ) W 2 . In U k U_k U k ,
W = W 1 y ( s ) x ( r ) W 2 + W 1 [ x , y ] ( r + s ) W 2 + r δ r + s , 0 ( x , y ) k W 1 W 2 . (40) W=W_1y(s)x(r)W_2+W_1[x,y](r+s)W_2+r\delta_{r+s,0}(x,y)kW_1W_2. \tag*{(40)} W = W 1 y ( s ) x ( r ) W 2 + W 1 [ x , y ] ( r + s ) W 2 + r δ r + s , 0 ( x , y ) k W 1 W 2 . (40)
where the central element acts as the scalar k k k . The first term still has L L L factors but one fewer inversion. Expanding [ x , y ] [x,y] [ x , y ] in the chosen basis gives the second term of length L − 1 L-1 L − 1 , while the third term has length L − 2 L-2 L − 2 . Induction on length and then on the number of inversions gives the finite expansion
W − X u ∈ F L ( u ) − 1 U k , W = X u + ∑ L ( v ) < L ( u ) a v X v . (41) W-X^u\in F_{L(u)-1}U_k,\qquad W=X^u+\sum_{L(v)<L(u)}a_vX^v. \tag*{(41)} W − X u ∈ F L ( u ) − 1 U k , W = X u + L ( v ) < L ( u ) ∑ a v X v . (41)
This is the filtration calculation in [28 Remark 4.2, (4.3)–(4.4)] . By the first rule of the PS order, the exponent of every correction is strictly larger than u u u .
If A A A and B B B are affine weight vectors of weights α \alpha α and β \beta β , then, for H ∈ h ^ H\in\widehat{\mathfrak{h}} H ∈ h , the Jacobi identity gives
[ H , [ A , B ] ] = [ [ H , A ] , B ] + [ A , [ H , B ] ] = ( α ( H ) + β ( H ) ) [ A , B ] . [H,[A,B]]=[[H,A],B]+[A,[H,B]]=(\alpha(H)+\beta(H))[A,B]. [ H , [ A , B ]] = [[ H , A ] , B ] + [ A , [ H , B ]] = ( α ( H ) + β ( H )) [ A , B ] .
Thus each nonzero commutator has weight α + β \alpha+\beta α + β , the weight of A B AB A B . Consequently PBW reordering preserves total affine weight and hence principal degree. Finally, for every P ≥ 0 P\ge0 P ≥ 0 , Lemma 2.3 makes
{ u : supp u ⊂ B − , ∣ u ∣ p r = P } \{u:\operatorname{supp}u\subset B_-,\ |u|_{\mathrm{pr}}=P\} { u : supp u ⊂ B − , ∣ u ∣ pr = P }
finite. A sequence of strict increases in this set must terminate.
Annihilating fields and their leading terms
We next recall the relations whose leading terms will be compared with paths in the grid. Fix k ∈ Z > 0 k\in\mathbb{Z}_{>0} k ∈ Z > 0 and put ℓ = k + 1 \ell=k+1 ℓ = k + 1 . For g = s p 2 n \mathfrak{g}=\mathfrak{sp}_{2n} g = sp 2 n , let
p = ( g ⊗ C [ t ] ) ⊕ C c ⊕ C d . \mathfrak{p}=(\mathfrak{g}\otimes\mathbb{C}[t])\oplus\mathbb{C}c\oplus\mathbb{C}d. p = ( g ⊗ C [ t ]) ⊕ C c ⊕ C d .
Following [28 §2] , make the one-dimensional space C 1 \mathbb{C}1 C 1 a p \mathfrak{p} p -module by
x ( m ) 1 = 0 ( x ∈ g , m ≥ 0 ) , d 1 = 0 , c 1 = k 1. x(m)1=0\quad(x\in\mathfrak{g},\ m\ge0),\qquad d1=0,\qquad c1=k1. x ( m ) 1 = 0 ( x ∈ g , m ≥ 0 ) , d 1 = 0 , c 1 = k 1.
The universal vacuum module is the induced module
N ( k Λ 0 ) = U ( g ^ ) ⊗ U ( p ) C 1 ≅ U ( g ⊗ t − 1 C [ t − 1 ] ) 1. N(k\Lambda_0)=U(\widehat{\mathfrak{g}})\otimes_{U(\mathfrak{p})}\mathbb{C}1\cong U(\mathfrak{g}\otimes t^{-1}\mathbb{C}[t^{-1}])1. N ( k Λ 0 ) = U ( g ) ⊗ U ( p ) C 1 ≅ U ( g ⊗ t − 1 C [ t − 1 ]) 1.
A formal field on a vector space M M M is a series A ( z ) = ∑ s ∈ Z A s z − s − 1 A(z)=\sum_{s\in\mathbb{Z}}A_sz^{-s-1} A ( z ) = ∑ s ∈ Z A s z − s − 1 with A s ∈ End ( M ) A_s\in\operatorname{End}(M) A s ∈ End ( M ) such that, for each v ∈ M v\in M v ∈ M , A s v = 0 A_sv=0 A s v = 0 for all sufficiently large s s s . The universal vacuum module N ( k Λ 0 ) N(k\Lambda_0) N ( k Λ 0 ) carries the structure of an affine vertex operator algebra. Its state–field correspondence assigns a field Y ( v , z ) Y(v,z) Y ( v , z ) to each v ∈ N ( k Λ 0 ) v \in N(k\Lambda_0) v ∈ N ( k Λ 0 ) . We use the following identities from [28 §§2–3] :
Y ( X ( − 1 ) 1 , z ) = X ( z ) : = ∑ m ∈ Z X ( m ) z − m − 1 ( X ∈ g ) , v ∗ = ( X θ ( − 1 ) ) ℓ 1 , Y ( v ∗ , z ) = ( X θ ( z ) ) ℓ , Y ( X ( 0 ) v , z ) = [ X ( 0 ) , Y ( v , z ) ] ( X ∈ g ) . (42) \begin{aligned}
Y(X(-1)1,z) &= X(z) := \sum_{m\in\mathbb{Z}} X(m)z^{-m-1} \qquad(X\in\mathfrak{g}),\\
v_* &= (X_\theta(-1))^\ell1,\qquad Y(v_*,z)=(X_\theta(z))^\ell,\\
Y(X(0)v,z)&=[X(0),Y(v,z)]\qquad(X\in\mathfrak{g}). \tag*{(42)}
\end{aligned} Y ( X ( − 1 ) 1 , z ) v ∗ Y ( X ( 0 ) v , z ) = X ( z ) := m ∈ Z ∑ X ( m ) z − m − 1 ( X ∈ g ) , = ( X θ ( − 1 ) ) ℓ 1 , Y ( v ∗ , z ) = ( X θ ( z ) ) ℓ , = [ X ( 0 ) , Y ( v , z )] ( X ∈ g ) . (42)
Here v ∗ v_* v ∗ is a singular vector of the universal vacuum module. The zero modes X ( 0 ) X(0) X ( 0 ) for X ∈ g X\in\mathfrak{g} X ∈ g generate from it the subspace
R k = U ( g ) v ∗ . R_k=U(\mathfrak{g})v_*. R k = U ( g ) v ∗ .
As a g \mathfrak{g} g -module, R k R_k R k is irreducible and finite dimensional, of highest weight ℓ θ \ell\theta ℓ θ [28 §3] . We use the subscript to distinguish it from the finite root system R R R .
For r ∈ R k r\in R_k r ∈ R k , write
Y ( r , z ) = ∑ m ∈ Z R r ( m ) z − m − ℓ . Y(r,z)=\sum_{m\in\mathbb{Z}}R_r(m)z^{-m-\ell}. Y ( r , z ) = m ∈ Z ∑ R r ( m ) z − m − ℓ .
In the usual notation Y ( r , z ) = ∑ s r s z − s − 1 Y(r,z)=\sum_s r_s z^{-s-1} Y ( r , z ) = ∑ s r s z − s − 1 , this means R r ( m ) = r m + ℓ − 1 R_r(m)=r_{m+\ell-1} R r ( m ) = r m + ℓ − 1 . In particular, the coefficient for v ∗ v_* v ∗ is
R v ∗ ( m ) = ∑ t 1 + ⋯ + t ℓ = m X θ ( t 1 ) ⋯ X θ ( t ℓ ) . (43) R_{v_*}(m)=\sum_{t_1+\cdots+t_\ell=m}X_\theta(t_1)\cdots X_\theta(t_\ell). \tag*{(43)} R v ∗ ( m ) = t 1 + ⋯ + t ℓ = m ∑ X θ ( t 1 ) ⋯ X θ ( t ℓ ) . (43)
since
( X θ ( z ) ) ℓ = ∏ i = 1 ℓ ( ∑ t i ∈ Z X θ ( t i ) z − t i − 1 ) = ∑ m ∈ Z ∑ t 1 + ⋯ + t ℓ = m X θ ( t 1 ) ⋯ X θ ( t ℓ ) z − m − ℓ . (X_\theta(z))^\ell=\prod_{i=1}^{\ell}\left(\sum_{t_i\in\mathbb{Z}}X_\theta(t_i)z^{-t_i-1}\right)=\sum_{m\in\mathbb{Z}}\sum_{t_1+\cdots+t_\ell=m}X_\theta(t_1)\cdots X_\theta(t_\ell)z^{-m-\ell}. ( X θ ( z ) ) ℓ = i = 1 ∏ ℓ ( t i ∈ Z ∑ X θ ( t i ) z − t i − 1 ) = m ∈ Z ∑ t 1 + ⋯ + t ℓ = m ∑ X θ ( t 1 ) ⋯ X θ ( t ℓ ) z − m − ℓ .
The sum over modes is interpreted on the module: the X θ ( t i ) X_\theta(t_i) X θ ( t i ) commute and, on each fixed vector, a sufficiently large mode acts as zero. Thus only finitely many tuples with fixed sum m m m contribute. In general, every r ∈ R k r\in R_k r ∈ R k is a finite sum
r = ∑ ν = 1 A c ν X ν , 1 ( 0 ) ⋯ X ν , q ν ( 0 ) v ∗ , X ν , j ∈ g , c ν ∈ C . r=\sum_{\nu=1}^{A}c_\nu X_{\nu,1}(0)\cdots X_{\nu,q_\nu}(0)v_*,\qquad X_{\nu,j}\in\mathfrak{g},\quad c_\nu\in\mathbb{C}. r = ν = 1 ∑ A c ν X ν , 1 ( 0 ) ⋯ X ν , q ν ( 0 ) v ∗ , X ν , j ∈ g , c ν ∈ C .
Put
D ν , j ( T ) = [ X ν , j ( 0 ) , T ] , D ν = D ν , 1 ∘ ⋯ ∘ D ν , q ν , D_{\nu,j}(T)=[X_{\nu,j}(0),T],\qquad D_\nu=D_{\nu,1}\circ\cdots\circ D_{\nu,q_\nu}, D ν , j ( T ) = [ X ν , j ( 0 ) , T ] , D ν = D ν , 1 ∘ ⋯ ∘ D ν , q ν ,
with D ν D_\nu D ν the identity when q ν = 0 q_\nu=0 q ν = 0 . The last identity in (3.9) gives
Y ( r , z ) = ∑ ν = 1 A c ν D ν ( ( X θ ( z ) ) ℓ ) . Y(r,z)=\sum_{\nu=1}^{A}c_\nu D_\nu((X_\theta(z))^\ell). Y ( r , z ) = ν = 1 ∑ A c ν D ν (( X θ ( z ) ) ℓ ) .
Hence
R r ( m ) = ∑ ν = 1 A c ν D ν ( ∑ t 1 + ⋯ + t ℓ = m X θ ( t 1 ) ⋯ X θ ( t ℓ ) ) . (44) R_r(m)=\sum_{\nu=1}^{A}c_\nu D_\nu\left(\sum_{t_1+\cdots+t_\ell=m}X_\theta(t_1)\cdots X_\theta(t_\ell)\right). \tag*{(44)} R r ( m ) = ν = 1 ∑ A c ν D ν ( t 1 + ⋯ + t ℓ = m ∑ X θ ( t 1 ) ⋯ X θ ( t ℓ ) ) . (44)
Let R ‾ k \overline{R}_k R k be the span of all coefficients R r ( m ) R_r(m) R r ( m ) , with r ∈ R k r\in R_k r ∈ R k and m ∈ Z m\in\mathbb{Z} m ∈ Z .
Theorem 3.3 ([28 Theorem 3.1] ). Let g \mathfrak{g} g be a finite-dimensional complex simple Lie algebra and k k k a positive integer. Let R k R_k R k and R ‾ k \overline{R}_k R k be the spaces constructed above from a highest-root vector of g \mathfrak{g} g . A highest weight module of its untwisted affine algebra at level k k k is standard if and only if R ‾ k \overline{R}_k R k annihilates it. In particular, R ‾ k L ( Λ ) = 0 \overline{R}_kL(\Lambda)=0 R k L ( Λ ) = 0 for each dominant integral weight Λ \Lambda Λ of level $k.
To describe the leading terms of these relations, we use the triangular array of colours from (18) . A cascade is a finite staircase in the set $1 \leq a \leq b \leq N$, with steps ( a , b ) ↦ ( a − 1 , b ) (a,b) \mapsto(a-1,b) ( a , b ) ↦ ( a − 1 , b ) or ( a , b ) ↦ ( a , b + 1 ) (a,b) \mapsto(a,b+1) ( a , b ) ↦ ( a , b + 1 ) . Each position has a nonnegative multiplicity. We may insert positions of multiplicity zero to complete a staircase. For s ∈ { 1 , … , N } s \in\{1,\ldots,N\} s ∈ { 1 , … , N } , put
T s u p = { ( a , b ) : 1 ≤ a ≤ b ≤ s } , T s d o w n = { ( a , b ) : s ≤ a ≤ b ≤ N } . (45) T_s^{\mathrm{up}}=\{(a,b):1\leq a\leq b\leq s\},\qquad T_s^{\mathrm{down}}=\{(a,b):s\leq a\leq b\leq N\}. \tag*{(45)} T s up = {( a , b ) : 1 ≤ a ≤ b ≤ s } , T s down = {( a , b ) : s ≤ a ≤ b ≤ N } . (45)
A pair of cascades ( B , A ) (B,A) ( B , A ) is admissible if B ⊂ T s u p B \subset T_s^{\mathrm{up}} B ⊂ T s up and A ⊂ T s d o w n A \subset T_s^{\mathrm{down}} A ⊂ T s down for some s s s . When only one mode occurs, we take B = ∅ B=\varnothing B = ∅ and A ⊂ T 1 d o w n A\subset T_1^{\mathrm{down}} A ⊂ T 1 down . These are the cascades and admissible pairs of [28 §6, (6.1)–(6.2)] , with any longer steps filled in by positions of multiplicity zero.
Theorem 3.4 ([28 Theorem 6.1] ). Suppose n ≥ 2 n\geq2 n ≥ 2 , k ∈ Z > 0 k\in\mathbb{Z}_{>0} k ∈ Z > 0 and j ∈ Z j\in\mathbb{Z} j ∈ Z . Let ( B , A ) (B,A) ( B , A ) be an admissible pair of cascades with total multiplicities b 0 b_0 b 0 and a 0 a_0 a 0 , respectively, where a 0 + b 0 = k + 1 a_0+b_0=k+1 a 0 + b 0 = k + 1 . If only one mode occurs, take B = ∅ B=\varnothing B = ∅ and a 0 = k + 1 a_0=k+1 a 0 = k + 1 . Then the exponent represented by
∏ ( a , b ) ∈ B X a b ( − j − 1 ) m a b ∏ ( a , b ) ∈ A X a b ( − j ) n a b (46) \prod_{(a,b)\in B}X_{ab}(-j-1)^{m_{ab}}\prod_{(a,b)\in A}X_{ab}(-j)^{n_{ab}} \tag*{(46)} ( a , b ) ∈ B ∏ X ab ( − j − 1 ) m ab ( a , b ) ∈ A ∏ X ab ( − j ) n ab (46) is the smallest exponent with nonzero coefficient in an element R r ( m ) R_r(m) R r ( m ) of R ‾ k \overline{R}_k R k , where m = − ( j + 1 ) b 0 − j a 0 m=-(j+1)b_0-ja_0 m = − ( j + 1 ) b 0 − j a 0 .
Here the products specify an exponent u u u , and X u X^u X u denotes the corresponding product in PBW order. We use only the existence of a relation with this leading term. We do not need the converse assertion that these cascades account for every leading term (see [28 Remark 6.2] ).
From paths to cascades
Lemma 3.5. Let S ⊂ Ω + S\subset\Omega_+ S ⊂ Ω + be a nonempty finite set of vertices lying on a common full path. Under (35) , S S S occupies at most two adjacent modes. At one mode its image can be completed to a cascade. At two modes − j -j − j and − j − 1 -j-1 − j − 1 , its images can be completed to an admissible pair as in Theorem 3.4, with mode − j − 1 -j-1 − j − 1 in the upper triangle. All inserted positions may be assigned multiplicity zero.
Proof. For two vertices at the same mode index j j j , substitution in (35) gives
( p − p ′ ) 2 − ( r − r ′ ) 2 = 4 ( a − a ′ ) ( b − b ′ ) . (47) (p-p')^2-(r-r')^2=4(a-a')(b-b'). \tag*{(47)} ( p − p ′ ) 2 − ( r − r ′ ) 2 = 4 ( a − a ′ ) ( b − b ′ ) . (47)
Since the two vertices lie on a common path, (6) implies ( a − a ′ ) ( b − b ′ ) ≤ 0 (a-a')(b-b')\leq0 ( a − a ′ ) ( b − b ′ ) ≤ 0 . Order the support by increasing b b b , breaking ties by decreasing a a a . The resulting sequence of a a a ’s is nonincreasing. For consecutive positions ( a i , b i ) (a_i,b_i) ( a i , b i ) and ( a i + 1 , b i + 1 ) (a_{i+1},b_{i+1}) ( a i + 1 , b i + 1 ) , insert
( a i , b i ) , ( a i − 1 , b i ) , … , ( a i + 1 , b i ) , ( a i + 1 , b i + 1 ) , … , ( a i + 1 , b i + 1 ) . (a_i,b_i),(a_i-1,b_i),\ldots,(a_{i+1},b_i),(a_{i+1},b_i+1),\ldots,(a_{i+1},b_{i+1}). ( a i , b i ) , ( a i − 1 , b i ) , … , ( a i + 1 , b i ) , ( a i + 1 , b i + 1 ) , … , ( a i + 1 , b i + 1 ) .
Since a i + 1 ≤ a i ≤ b i ≤ b i + 1 a_{i+1}\leq a_i\leq b_i\leq b_{i+1} a i + 1 ≤ a i ≤ b i ≤ b i + 1 , these unit steps stay in the triangle. Give the inserted positions multiplicity zero.
Next let ( a , b ) (a,b) ( a , b ) have mode index j j j and ( a ′ , b ′ ) (a',b') ( a ′ , b ′ ) mode index j + 1 j+1 j + 1 . For either parity of j j j , substitution gives
( p ′ − p ) 2 − ( r ′ − r ) 2 = 4 ( b ′ − a ) ( N + a ′ − b ) . (48) (p'-p)^2-(r'-r)^2=4(b'-a)(N+a'-b). \tag*{(48)} ( p ′ − p ) 2 − ( r ′ − r ) 2 = 4 ( b ′ − a ) ( N + a ′ − b ) . (48)
Since N + a ′ − b ≥ 1 N+a'-b\geq1 N + a ′ − b ≥ 1 , two such vertices are on a common path exactly when b ′ ≤ a b'\leq a b ′ ≤ a . Write B 0 B_0 B 0 and A 0 A_0 A 0 for the sets of ( a ′ , b ′ ) (a',b') ( a ′ , b ′ ) and ( a , b ) (a,b) ( a , b ) . Applying (6) to every pair from different modes gives
max ( a ′ , b ′ ) ∈ B 0 b ′ ≤ min ( a , b ) ∈ A 0 a . \max_{(a',b')\in B_0} b'\leq\min_{(a,b)\in A_0}a. ( a ′ , b ′ ) ∈ B 0 max b ′ ≤ ( a , b ) ∈ A 0 min a .
Choose an integer s s s between these bounds. Then
B 0 ⊂ T s u p , A 0 ⊂ T s d o w n . B_0 \subset T_s^{\mathrm{up}}, \qquad A_0 \subset T_s^{\mathrm{down}}. B 0 ⊂ T s up , A 0 ⊂ T s down .
The staircase construction above uses only coordinates between its endpoints, so it completes each set within its own triangle.
It remains to exclude nonadjacent modes. For indices j j j and j + d j+d j + d with d ≥ 2 d \ge2 d ≥ 2 , we have
p ′ − p = d N + a ′ + b ′ − a − b ≥ ( d − 2 ) N + 2 > 0. p' - p = dN + a' + b' - a - b \ge(d-2)N + 2 > 0. p ′ − p = d N + a ′ + b ′ − a − b ≥ ( d − 2 ) N + 2 > 0.
If d d d is even, then
p ′ − p − ∣ r ′ − r ∣ = min { d N + 2 ( a ′ − a ) , d N + 2 ( b ′ − b ) } ≥ ( d − 2 ) N + 2 > 0. p' - p - \lvert r' - r\rvert= \min\{dN + 2(a' - a),\ dN + 2(b' - b)\} \ge(d-2)N + 2 > 0. p ′ − p − ∣ r ′ − r ∣ = min { d N + 2 ( a ′ − a ) , d N + 2 ( b ′ − b )} ≥ ( d − 2 ) N + 2 > 0.
If d ≥ 3 d \ge3 d ≥ 3 is odd, it is
min { ( d − 1 ) N + 2 ( b ′ − a ) , ( d + 1 ) N + 2 ( a ′ − b ) } ≥ ( d − 3 ) N + 2 > 0. \min\{(d-1)N + 2(b' - a),\ (d+1)N + 2(a' - b)\} \ge(d-3)N + 2 > 0. min {( d − 1 ) N + 2 ( b ′ − a ) , ( d + 1 ) N + 2 ( a ′ − b )} ≥ ( d − 3 ) N + 2 > 0.
Such a pair violates (6) , so nonadjacent modes cannot occur. If only one mode occurs, take B = ∅ B = \varnothing B = ∅ , let A A A be its cascade completion, and choose s = 1 s = 1 s = 1 .
For example, in rank two, X 12 ( − 1 ) X_{12}(-1) X 12 ( − 1 ) and X 24 ( 0 ) X_{24}(0) X 24 ( 0 ) have colour indices ( 1 , 2 ) (1,2) ( 1 , 2 ) and ( 2 , 4 ) (2,4) ( 2 , 4 ) and correspond to ( r , p ) = ( 3 , 2 ) (r,p) = (3,2) ( r , p ) = ( 3 , 2 ) and ( 2 , 1 ) (2,1) ( 2 , 1 ) , respectively. They lie on a common path, satisfy b ′ = a = 2 b' = a = 2 b ′ = a = 2 in (48) , and admit the choice s = 2 s = 2 s = 2 . At a single mode, X 22 ( − 1 ) X_{22}(-1) X 22 ( − 1 ) and X 13 ( − 1 ) X_{13}(-1) X 13 ( − 1 ) correspond to ( 4 , 3 ) (4,3) ( 4 , 3 ) and ( 2 , 3 ) (2,3) ( 2 , 3 ) . The position ( 1 , 2 ) (1,2) ( 1 , 2 ) , assigned multiplicity zero, completes the staircase ( 2 , 2 ) , ( 1 , 2 ) , ( 1 , 3 ) (2,2),(1,2),(1,3) ( 2 , 2 ) , ( 1 , 2 ) , ( 1 , 3 ) . By contrast, X 11 ( − 1 ) X_{11}(-1) X 11 ( − 1 ) and X 22 ( − 1 ) X_{22}(-1) X 22 ( − 1 ) correspond to distinct part sizes in the same row and therefore cannot lie on one path.
Figures 3 and 4 illustrate these two cases. Filled positions have multiplicity one, and open positions have multiplicity zero. In the grid, the lines extend or join the selected portions of the path.
Figure 3. A cascade at a single mode for n = 2 n = 2 n = 2 . The filled colours 22 and 13 correspond to ( r , p ) = ( 4 , 3 ) (r,p) = (4,3) ( r , p ) = ( 4 , 3 ) and ( 2 , 3 ) (2,3) ( 2 , 3 ) , respectively. The inserted colour 12 corresponds to ( 3 , 2 ) (3,2) ( 3 , 2 ) . The full path has heights ( 1 , 2 , 3 , 2 , 3 ) (1,2,3,2,3) ( 1 , 2 , 3 , 2 , 3 ) . At level one, the ordered leading monomial is X 13 ( − 1 ) X 22 ( − 1 ) X_{13}(-1)X_{22}(-1) X 13 ( − 1 ) X 22 ( − 1 ) .
Finite straightening relations on the highest vector
We now evaluate the coefficients of the annihilating fields on v Λ v_\Lambda v Λ . The resulting relations are finite and replace a monomial by larger ones in the PS order.
Lemma 3.6. Let n ≥ 2 n \ge2 n ≥ 2 , and let Λ \Lambda Λ be a dominant integral weight of positive level k k k . Suppose that ℓ ( u ) = k + 1 \ell(u) = k + 1 ℓ ( u ) = k + 1 and that all vertices in the support of u u u lie in Ω + \Omega_+ Ω + on a common full path. Then
X u v Λ = ∑ v ≻ u β ( v ) = β ( u ) c v X v v Λ , (49) X^u v_\Lambda= \sum_{\substack{v \succ u\\ \beta(v)=\beta(u)}} c_v X^v v_\Lambda, \tag*{(49)} X u v Λ = v ≻ u β ( v ) = β ( u ) ∑ c v X v v Λ , (49)
Figure 4. Cascades at adjacent modes for n = 3 n = 3 n = 3 . The shaded triangles share ( 3 , 3 ) (3, 3) ( 3 , 3 ) . Their filled sites map to four vertices of the path with heights ( 11 , 10 , 9 , 8 , 7 , 6 , 7 ) (11, 10, 9, 8, 7, 6, 7) ( 11 , 10 , 9 , 8 , 7 , 6 , 7 ) . The inequality b ′ ≤ 3 ≤ a b' \le3 \le a b ′ ≤ 3 ≤ a holds for every pair of positions from different modes. At level three, the ordered leading monomial is X 33 ( − 2 ) X 13 ( − 2 ) X 36 ( − 1 ) X 44 ( − 1 ) X_{33}(-2)X_{13}(-2)X_{36}(-1)X_{44}(-1) X 33 ( − 2 ) X 13 ( − 2 ) X 36 ( − 1 ) X 44 ( − 1 ) .
where the sum is finite and every exponent v v v on the right is supported on B − \mathcal{B}_{-} B − .
Proof. Put ℓ = k + 1 \ell= k + 1 ℓ = k + 1 and write
F m = R v ∗ ( m ) = ∑ t 1 + ⋯ + t ℓ = m X θ ( t 1 ) ⋯ X θ ( t ℓ ) . F_m = R_{v^*}(m) = \sum_{t_1+\cdots+t_\ell=m} X_{\theta(t_1)} \cdots X_{\theta(t_\ell)}. F m = R v ∗ ( m ) = t 1 + ⋯ + t ℓ = m ∑ X θ ( t 1 ) ⋯ X θ ( t ℓ ) .
Equation (44) can be written as
R r ( m ) = ∑ ν = 1 A c ν D ν ( F m ) , D ν = ad X ν , 1 ( 0 ) ⋯ ad X ν , q ν ( 0 ) . R_r(m) = \sum_{\nu=1}^{A} c_\nu D_\nu(F_m), \qquad D_\nu= \operatorname{ad} X_{\nu,1}(0) \cdots\operatorname{ad} X_{\nu,q_\nu}(0). R r ( m ) = ν = 1 ∑ A c ν D ν ( F m ) , D ν = ad X ν , 1 ( 0 ) ⋯ ad X ν , q ν ( 0 ) .
By Lemma 3.5 and Theorem 3.4, we may choose r ∈ R k r \in R_k r ∈ R k and m ∈ Z m \in\mathbb{Z} m ∈ Z so that R r ( m ) R_r(m) R r ( m ) has least PS exponent u u u , with coefficient c u ≠ 0 c_u \ne0 c u = 0 .
We first evaluate this relation on v Λ v_\Lambda v Λ . The X θ ( t ) X_\theta(t) X θ ( t ) commute, so we may group the terms of F m F_m F m according to nondecreasing mode tuples t 1 ≤ ⋯ ≤ t ℓ t_1 \le\cdots\le t_\ell t 1 ≤ ⋯ ≤ t ℓ . Since [ x ( 0 ) , y ( t ) ] = [ x , y ] ( t ) [x(0),y(t)] = [x,y](t) [ x ( 0 ) , y ( t )] = [ x , y ] ( t ) , each D ν D_\nu D ν replaces a word by a sum of words with the same sequence of modes. If t ℓ > 0 t_\ell> 0 t ℓ > 0 , the last factor in each such word annihilates v Λ v_\Lambda v Λ . Otherwise m ≤ t i ≤ 0 m \le t_i \le0 m ≤ t i ≤ 0 for every i i i . There are only finitely many such tuples. Let W W W be the resulting finite sum of words before applying it to v Λ v_\Lambda v Λ .
In each word of W W W , use X Y = Y X + [ X , Y ] XY = YX + [X,Y] X Y = Y X + [ X , Y ] to move the raising operators at mode zero to the right and the zero-mode Cartan factors H ( 0 ) H(0) H ( 0 ) , H ∈ h H \in\mathfrak{h} H ∈ h , immediately to their left. A word ending in a raising operator annihilates v Λ v_\Lambda v Λ . If no such factor remains, each H ( 0 ) H(0) H ( 0 ) on the right acts by the scalar Λ ( H ) \Lambda(H) Λ ( H ) . All commutator terms have smaller PBW length, as does every term in which an H ( 0 ) H(0) H ( 0 ) factor has been evaluated. Finally, order the remaining lowering factors in the fixed PS order.
At length ℓ \ell ℓ , these operations discard the exponents containing a factor outside B − \mathcal{B}_{-} B − . The terms previously omitted because t ℓ > 0 t_\ell> 0 t ℓ > 0 cannot contribute to any exponent on B − \mathcal{B}_{-} B − at this length either: an interchange preserves the multiset of factors, while a commutator shortens the word. Thus X u X^u X u retains its original nonzero coefficient c u c_u c u , and every other surviving exponent of length ℓ \ell ℓ is larger than u u u . By Theorem 3.3, W v Λ = 0 Wv_\Lambda= 0 W v Λ = 0 , so
0 = W v Λ = c u X u v Λ + ∑ z > u ℓ ( z ) = ℓ a z X z v Λ + ∑ ℓ ( z ) < ℓ b z X z v Λ . (50) 0 = Wv_\Lambda= c_uX^uv_\Lambda+ \sum_{\substack{z>u\\ \ell(z)=\ell}} a_zX^zv_\Lambda+ \sum_{\ell(z)<\ell} b_zX^zv_\Lambda. \tag*{(50)} 0 = W v Λ = c u X u v Λ + z > u ℓ ( z ) = ℓ ∑ a z X z v Λ + ℓ ( z ) < ℓ ∑ b z X z v Λ . (50)
Every exponent in the last sum is also greater than u u u , by the length rule of the PS order. Since X z v Λ X^zv_\Lambda X z v Λ has affine weight Λ − β ( z ) \Lambda-\beta(z) Λ − β ( z ) , the component of this equality in weight Λ − β ( u ) \Lambda-\beta(u) Λ − β ( u ) is
0 = c u X u v Λ + ∑ z ≻ u β ( z ) = β ( u ) d z X z v Λ . 0=c_uX^uv_\Lambda+\sum_{\substack{z\succ u\\\beta(z)=\beta(u)}}d_zX^zv_\Lambda. 0 = c u X u v Λ + z ≻ u β ( z ) = β ( u ) ∑ d z X z v Λ .
The sums are finite because ∑ i t i = m \sum_i t_i=m ∑ i t i = m is fixed, the surviving modes satisfy m ≤ t i ≤ 0 m\leq t_i\leq0 m ≤ t i ≤ 0 , and each remaining word has a finite PBW expansion. Dividing by c u c_u c u gives (3.16).
Bases at level one and linear independence
Lemma 3.6 applies when the sum of the positive frequencies along a path exceeds k k k . A path may also violate (1.4) because of its boundary entries, even when its positive frequencies sum to at most k k k . We first treat these boundary relations at level one and use the results to prove independence at every level. Spanning at higher levels will be proved in Section 5.
Throughout this section, exponents have finite support in B − \mathcal{B}_- B − . For an exponent u u u , write β ( u ) \beta(u) β ( u ) for its lowering weight. All spans consist of finite linear combinations. We use the notion of an essential monomial from [14 Definition 2.3] , with the direction of comparison determined by the order in Section 3.
Definition 4.1. An exponent u u u is essential for L ( λ ) L(\lambda) L ( λ ) if
X u v λ ∉ span { X a v λ : a ≻ u } . (51) X^uv_\lambda\notin\operatorname{span}\{X^av_\lambda:a\succ u\}. \tag*{(51)} X u v λ ∈ / span { X a v λ : a ≻ u } . (51) Write E λ \mathcal{E}_\lambda E λ for the set of essential exponents.
Since X u v λ X^uv_\lambda X u v λ has weight λ − β ( u ) \lambda-\beta(u) λ − β ( u ) , projection onto that weight space gives
u ∈ E λ ⟺ X u v λ ∉ span { X a v λ : a ≻ u , β ( a ) = β ( u ) } . u\in\mathcal{E}_\lambda\quad\Longleftrightarrow\quad X^uv_\lambda\notin\operatorname{span}\{X^av_\lambda:a\succ u,\ \beta(a)=\beta(u)\}. u ∈ E λ ⟺ X u v λ ∈ / span { X a v λ : a ≻ u , β ( a ) = β ( u )} .
Fix a lowering weight β \beta β . The set { u : β ( u ) = β } \{u:\beta(u)=\beta\} { u : β ( u ) = β } is finite by Lemma 2.3. List its elements as u 1 ≻ ⋯ ≻ u m u_1\succ\cdots\succ u_m u 1 ≻ ⋯ ≻ u m and set
W 0 = 0 , W j = W j − 1 + C X u j v λ ( 1 ≤ j ≤ m ) . (52) W_0=0,\qquad W_j=W_{j-1}+\mathbb{C}X^{u_j}v_\lambda\quad(1\leq j\leq m). \tag*{(52)} W 0 = 0 , W j = W j − 1 + C X u j v λ ( 1 ≤ j ≤ m ) . (52)
Then
u j ∈ E λ ⟺ dim W j − dim W j − 1 = 1 , u_j\in\mathcal{E}_\lambda\quad\Longleftrightarrow\quad\dim W_j-\dim W_{j-1}=1, u j ∈ E λ ⟺ dim W j − dim W j − 1 = 1 ,
L ( λ ) λ − β = W m = ⨁ u ∈ E λ β ( u ) = β C X u v λ , L(\lambda)_{\lambda-\beta}=W_m=\bigoplus_{\substack{u\in\mathcal{E}_\lambda\\\beta(u)=\beta}}\mathbb{C}X^uv_\lambda, L ( λ ) λ − β = W m = u ∈ E λ β ( u ) = β ⨁ C X u v λ ,
dim L ( λ ) λ − β = # { u ∈ E λ : β ( u ) = β } . (53) \dim L(\lambda)_{\lambda-\beta}=\#\{u\in\mathcal{E}_\lambda:\beta(u)=\beta\}. \tag*{(53)} dim L ( λ ) λ − β = # { u ∈ E λ : β ( u ) = β } . (53)
Indeed, at each step an essential exponent adds one basis vector, whereas a nonessential exponent leaves the span unchanged. Taking the direct sum over β \beta β shows that the essential monomial vectors form a basis of L ( λ ) L(\lambda) L ( λ ) .
Boundary relations and the count at level one
For L ( Λ i ) L(\Lambda_i) L ( Λ i ) the unique nonzero boundary mark is at B i B_i B i . We first show that a positive vertex in Ω + \Omega_+ Ω + on a path through this particular mark corresponds to a generator that annihilates v Λ i v_{\Lambda_i} v Λ i .
Lemma 4.2. Let 0 ≤ i ≤ n 0\leq i\leq n 0 ≤ i ≤ n . If ( r , p ) ∈ Ω + (r,p)\in\Omega_+ ( r , p ) ∈ Ω + lies on a path through B i = ( N − 2 i , − 1 ) B_i=(N-2i,-1) B i = ( N − 2 i , − 1 ) , then its corresponding generator satisfies X a b ( − j ) v Λ i = 0 X_{ab}(-j)v_{\Lambda_i}=0 X ab ( − j ) v Λ i = 0 .
Proof. Write ω i = Λ i ∣ h = ε 1 + ⋯ + ε i \omega_i=\Lambda_i|_{\mathfrak h}=\varepsilon_1+\cdots+\varepsilon_i ω i = Λ i ∣ h = ε 1 + ⋯ + ε i , with ω 0 = 0 \omega_0=0 ω 0 = 0 . For a positive real root γ \gamma γ , (29) gives
⟨ Λ i , γ ∨ ⟩ = 0 ⟹ X − γ v Λ i = 0. \langle\Lambda_i,\gamma^\vee\rangle=0 \quad\Longrightarrow\quad X_{-\gamma}v_{\Lambda_i}=0. ⟨ Λ i , γ ∨ ⟩ = 0 ⟹ X − γ v Λ i = 0.
It suffices to show that any generator on such a path belongs to a real root space whose coroot pairs to zero with Λ i \Lambda_i Λ i . We do this by considering its mode.
Mode zero. Substitution of j = 0 j=0 j = 0 in (35) and application of the criterion (6) for two vertices to lie on a common path give
( p + 1 ) 2 − ( r − N + 2 i ) 2 = 4 ( a − i ) ( b − N + i ) ≤ 0. (54) (p+1)^2-(r-N+2i)^2=4(a-i)(b-N+i)\le0. \tag*{(54)} ( p + 1 ) 2 − ( r − N + 2 i ) 2 = 4 ( a − i ) ( b − N + i ) ≤ 0. (54)
If a ≤ n a\le n a ≤ n , put q = N + 1 − b q=N+1-b q = N + 1 − b . Then q < a q<a q < a because p = a + b − N − 1 > 0 p=a+b-N-1>0 p = a + b − N − 1 > 0 . Since b − N + i = i − q + 1 b-N+i=i-q+1 b − N + i = i − q + 1 and i i i is integral,
( a − i ) ( i − q + 1 ) ≤ 0 ⟺ i < q or i ≥ a . (a-i)(i-q+1)\le0 \quad\Longleftrightarrow\quad i<q\ \text{or}\ i\ge a. ( a − i ) ( i − q + 1 ) ≤ 0 ⟺ i < q or i ≥ a .
Let 1 t ≤ i \mathbf{1}_{t\le i} 1 t ≤ i be 1 if t ≤ i t\le i t ≤ i and 0 otherwise. By (17) , ω i ( H t ) = 1 t ≤ i \omega_i(H_t)=\mathbf{1}_{t\le i} ω i ( H t ) = 1 t ≤ i , so
γ = ε q − ε a , ⟨ Λ i , γ ∨ ⟩ = ω i ( H q − H a ) = 1 q ≤ i − 1 a ≤ i = 1 q ≤ i < a = 0. \gamma=\varepsilon_q-\varepsilon_a,\qquad\langle\Lambda_i,\gamma^\vee\rangle=\omega_i(H_q-H_a)=\mathbf{1}_{q\le i}-\mathbf{1}_{a\le i}=\mathbf{1}_{q\le i<a}=0. γ = ε q − ε a , ⟨ Λ i , γ ∨ ⟩ = ω i ( H q − H a ) = 1 q ≤ i − 1 a ≤ i = 1 q ≤ i < a = 0.
If instead a > n a>n a > n , set q = N + 1 − b ≤ q ′ = N + 1 − a ≤ n q=N+1-b\le q'=N+1-a\le n q = N + 1 − b ≤ q ′ = N + 1 − a ≤ n . Then
γ = ε q + ε q ′ , ⟨ Λ i , γ ∨ ⟩ = { ω i ( H q + H q ′ ) = 1 q ≤ i + 1 q ′ ≤ i , q < q ′ , ω i ( H q ) = 1 q ≤ i , q = q ′ . \gamma=\varepsilon_q+\varepsilon_{q'},\qquad
\langle\Lambda_i,\gamma^\vee\rangle=
\begin{cases}
\omega_i(H_q+H_{q'})=\mathbf{1}_{q\le i}+\mathbf{1}_{q'\le i}, & q<q',\\
\omega_i(H_q)=\mathbf{1}_{q\le i}, & q=q'.
\end{cases} γ = ε q + ε q ′ , ⟨ Λ i , γ ∨ ⟩ = { ω i ( H q + H q ′ ) = 1 q ≤ i + 1 q ′ ≤ i , ω i ( H q ) = 1 q ≤ i , q < q ′ , q = q ′ .
Now a > n ≥ i a>n\ge i a > n ≥ i , so (54) is equivalent to i < q i<q i < q . Since q ≤ q ′ q\le q' q ≤ q ′ , all the displayed indicators vanish.
Mode − 1 -1 − 1 . The same path criterion gives
( p + 1 ) 2 − ( r − N + 2 i ) 2 = 4 ( b − i ) ( a + i ) ≤ 0. (55) (p+1)^2-(r-N+2i)^2=4(b-i)(a+i)\le0. \tag*{(55)} ( p + 1 ) 2 − ( r − N + 2 i ) 2 = 4 ( b − i ) ( a + i ) ≤ 0. (55)
Thus a ≤ b ≤ i ≤ n a\le b\le i\le n a ≤ b ≤ i ≤ n . With β = ε a + ε b \beta=\varepsilon_a+\varepsilon_b β = ε a + ε b and γ = δ − β \gamma=\delta-\beta γ = δ − β , (23) gives
γ ∨ = { c − β ∨ = c − H a , a = b , 2 c − β ∨ = 2 c − H a − H b , a < b , ⟨ Λ i , γ ∨ ⟩ = { 1 − 1 = 0 , a = b , 2 − 2 = 0 , a < b . \gamma^\vee=
\begin{cases}
c-\beta^\vee=c-H_a, & a=b,\\
2c-\beta^\vee=2c-H_a-H_b, & a<b,
\end{cases}
\qquad
\langle\Lambda_i,\gamma^\vee\rangle=
\begin{cases}
1-1=0, & a=b,\\
2-2=0, & a<b.
\end{cases} γ ∨ = { c − β ∨ = c − H a , 2 c − β ∨ = 2 c − H a − H b , a = b , a < b , ⟨ Λ i , γ ∨ ⟩ = { 1 − 1 = 0 , 2 − 2 = 0 , a = b , a < b .
Finally, if j ≥ 2 j\ge2 j ≥ 2 , then p + 1 ≥ N + 2 > ∣ r − ( N − 2 i ) ∣ p+1\ge N+2>|r-(N-2i)| p + 1 ≥ N + 2 > ∣ r − ( N − 2 i ) ∣ , which excludes a path through B i B_i B i . □
Suppose Y Y Y is a factor of X f X^f X f covered by Lemma 4.2 , and let e Y e_Y e Y denote its unit exponent. Commuting this occurrence of Y Y Y to the right gives
X f = X f − e Y Y + R , R ∈ F ℓ ( f ) − 1 U ( n ^ − ) . X^f=X^{f-e_Y}Y+R,\qquad R\in F_{\ell(f)-1}U(\widehat{\mathfrak n}_{-}). X f = X f − e Y Y + R , R ∈ F ℓ ( f ) − 1 U ( n − ) .
Since Y v Λ i = 0 Yv_{\Lambda_i}=0 Y v Λ i = 0 , reordering R R R yields the finite relation
X f v Λ i = ∑ ℓ ( h ) < ℓ ( f ) β ( h ) = β ( f ) c h X h v Λ i , ℓ ( h ) < ℓ ( f ) ⟹ h ≻ f . (56) X^fv_{\Lambda_i}=
\sum_{\substack{\ell(h)<\ell(f)\\ \beta(h)=\beta(f)}}c_hX^hv_{\Lambda_i},
\qquad\ell(h)<\ell(f)\Longrightarrow h\succ f. \tag*{(56)} X f v Λ i = ℓ ( h ) < ℓ ( f ) β ( h ) = β ( f ) ∑ c h X h v Λ i , ℓ ( h ) < ℓ ( f ) ⟹ h ≻ f . (56)
The weight equality follows from the homogeneity of the Lie bracket. An exponent is admissible for k k k if it belongs to S k S_k S k , and inadmissible otherwise.
Theorem 4.3 (Basis at level one). For n ≥ 2 n \ge2 n ≥ 2 and 0 ≤ i ≤ n 0 \le i \le n 0 ≤ i ≤ n , the vectors
{ X f v Λ i : f ∈ S e i } \{X^{f}v_{\Lambda_i}: f \in\mathcal{S}_{e_i}\} { X f v Λ i : f ∈ S e i } form a basis of L ( Λ i ) L(\Lambda_i) L ( Λ i ) , and
E Λ i = S e i . (57) \mathcal{E}_{\Lambda_i}=\mathcal{S}_{e_i}. \tag*{(57)} E Λ i = S e i . (57) Every inadmissible PBW vector is a linear combination of admissible PBW vectors with larger exponents and the same affine weight:
X g v Λ i ∈ span { X h v Λ i : h ≻ g , h ∈ S e i , β ( h ) = β ( g ) } ( g ∉ S e i ) . X^{g}v_{\Lambda_i} \in\operatorname{span}\{X^{h}v_{\Lambda_i}:h \succ g,\ h \in\mathcal{S}_{e_i},\ \beta(h)=\beta(g)\}\qquad(g \notin\mathcal{S}_{e_i}). X g v Λ i ∈ span { X h v Λ i : h ≻ g , h ∈ S e i , β ( h ) = β ( g )} ( g ∈ / S e i ) .
Proof. Ordered spanning. Let P P P be a violating path and let p ( P ) p(P) p ( P ) be the sum of the positive frequencies on P P P . If p ( P ) ≥ 2 p(P) \ge2 p ( P ) ≥ 2 , choose two occurrences on P P P , and apply Lemma 3.6. If p ( P ) < 2 p(P)<2 p ( P ) < 2 , then p ( P ) = 1 p(P)=1 p ( P ) = 1 and P P P passes through B i B_i B i : otherwise the total path sum would be at most one. Now (4.6) applies. In either case, an inadmissible exponent f f f satisfies a finite relation
X f v Λ i = ∑ h ≻ f β ( h ) = β ( f ) c h X h v Λ i . (58) X^{f}v_{\Lambda_i}=\sum_{\substack{h \succ f\\ \beta(h)=\beta(f)}}c_hX^{h}v_{\Lambda_i}. \tag*{(58)} X f v Λ i = h ≻ f β ( h ) = β ( f ) ∑ c h X h v Λ i . (58)
Fix a lowering weight β \beta β and list the finitely many exponents of that weight as u 1 ≻ ⋯ ≻ u m u_1 \succ\cdots\succ u_m u 1 ≻ ⋯ ≻ u m . In the notation of (4.2), with λ = Λ i \lambda=\Lambda_i λ = Λ i , an inadmissible u j u_j u j satisfies
X u j v Λ i = ∑ t < j c j , t X u t v Λ i , X^{u_j}v_{\Lambda_i}=\sum_{t<j}c_{j,t}X^{u_t}v_{\Lambda_i}, X u j v Λ i = t < j ∑ c j , t X u t v Λ i ,
so W j = W j − 1 W_j=W_{j-1} W j = W j − 1 . Induction on j j j gives
W j = span { X u t v Λ i : 1 ≤ t ≤ j , u t ∈ S e i } . W_j=\operatorname{span}\{X^{u_t}v_{\Lambda_i}:1\le t\le j,\ u_t\in\mathcal{S}_{e_i}\}. W j = span { X u t v Λ i : 1 ≤ t ≤ j , u t ∈ S e i } .
Applying the same induction to the terms of (4.8) proves the final assertion of the theorem. In particular, for each principal degree P P P ,
L ( Λ i ) P = span { X f v Λ i : f ∈ S e i , ∣ f ∣ p r = P } . L(\Lambda_i)_P=\operatorname{span}\{X^fv_{\Lambda_i}:f\in\mathcal{S}_{e_i},\ |f|_{\mathrm{pr}}=P\}. L ( Λ i ) P = span { X f v Λ i : f ∈ S e i , ∣ f ∣ pr = P } .
This spanning set is finite by Lemma 2.3.
The dimension count. A coloured integer ℓ d \ell_d ℓ d is a part of size ℓ \ell ℓ and colour d d d . Put
E a = { ℓ d : ℓ ≥ a , 0 ≤ d ≤ 2 n , ℓ − d ≡ 1 ( m o d 2 ) } . E_a=\{\ell_d:\ell\ge a,\ 0\le d\le2n,\ \ell-d\equiv1\pmod{2}\}. E a = { ℓ d : ℓ ≥ a , 0 ≤ d ≤ 2 n , ℓ − d ≡ 1 ( mod 2 )} .
A path in E − 1 E_{-1} E − 1 uses colours 0 , 1 , … , 2 n 0,1,\ldots,2n 0 , 1 , … , 2 n in order, changing size by 1 1 1 or − 1 -1 − 1 at each step. Let P i \mathcal{P}_i P i consist of partitions into parts of E 1 E_1 E 1 with frequency sum at most one on every such path, after assigning fictitious frequency one to ( − 1 ) 2 n − 2 i (-1)_{2n-2i} ( − 1 ) 2 n − 2 i and zero to the other elements of E − 1 ∖ E 1 E_{-1}\setminus E_1 E − 1 ∖ E 1 . Dousse–Konan [12 Theorem 1.15 and §5.4] prove, for n ≥ 2 n\ge2 n ≥ 2 , that
∑ π ∈ P i q ∣ π ∣ = ps ( e − Λ i ch L ( Λ i ) ) = ( q 2 n + 4 , q 2 i + 2 , q 2 n − 2 i + 2 ; q 2 n + 4 ) ∞ ( q ; q ) ∞ . (59) \sum_{\pi\in\mathcal{P}_i}q^{|\pi|}=\operatorname{ps}(e^{-\Lambda_i}\operatorname{ch}L(\Lambda_i))=\frac{(q^{2n+4},q^{2i+2},q^{2n-2i+2};q^{2n+4})_\infty}{(q;q)_\infty}. \tag*{(59)} π ∈ P i ∑ q ∣ π ∣ = ps ( e − Λ i ch L ( Λ i )) = ( q ; q ) ∞ ( q 2 n + 4 , q 2 i + 2 , q 2 n − 2 i + 2 ; q 2 n + 4 ) ∞ . (59)
where fictitious boundary parts do not contribute to ∣ π ∣ |\pi| ∣ π ∣ . Identifying a part ℓ d \ell_d ℓ d with the vertex ( d , ℓ ) (d,\ell) ( d , ℓ ) gives a bijection preserving size
P i ⟶ S e i , π ⟼ f π , f π ( d , ℓ ) = mult π ( ℓ d ) , ∣ π ∣ = ∣ f π ∣ p r . \mathcal{P}_i\longrightarrow\mathcal{S}_{e_i},\qquad\pi\longmapsto f_\pi,\qquad f_\pi(d,\ell)=\operatorname{mult}_\pi(\ell_d),\qquad|\pi|=|f_\pi|_{\mathrm{pr}}. P i ⟶ S e i , π ⟼ f π , f π ( d , ℓ ) = mult π ( ℓ d ) , ∣ π ∣ = ∣ f π ∣ pr .
This enumeration uses the character formula from perfect crystals and the partition bijections of [12 §5.4] . However, it does not assume linear independence of the PBW vectors. Comparing coefficients in (59) , we obtain
# { f ∈ S e i : ∣ f ∣ p r = P } = [ q P ] ps ( e − Λ i ch L ( Λ i ) ) = dim L ( Λ i ) P . \#\{f \in\mathcal{S}_{e_i} : |f|_{\mathrm{pr}} = P\} = [q^P]\operatorname{ps}(e^{-\Lambda_i}\operatorname{ch} L(\Lambda_i)) = \dim L(\Lambda_i)_P. # { f ∈ S e i : ∣ f ∣ pr = P } = [ q P ] ps ( e − Λ i ch L ( Λ i )) = dim L ( Λ i ) P .
The spanning family of admissible vectors is therefore a basis of L ( Λ i ) P L(\Lambda_i)_P L ( Λ i ) P .
Essentiality. For f ∉ S e i f \notin\mathcal{S}_{e_i} f ∈ / S e i , relation (58) shows that f ∉ E Λ i f \notin\mathcal{E}_{\Lambda_i} f ∈ / E Λ i . Thus E Λ i ⊆ S e i \mathcal{E}_{\Lambda_i} \subseteq\mathcal{S}_{e_i} E Λ i ⊆ S e i . Conversely, suppose an admissible exponent f f f were nonessential. Expressing X f v Λ i X^f v_{\Lambda_i} X f v Λ i in terms of larger exponents, and then applying the ordered spanning result to each inadmissible term, would give
X f v Λ i ∈ span { X h v Λ i : h ∈ S e i , h ≻ f } . X^f v_{\Lambda_i} \in\operatorname{span}\{X^h v_{\Lambda_i}: h \in\mathcal{S}_{e_i},\ h \succ f\}. X f v Λ i ∈ span { X h v Λ i : h ∈ S e i , h ≻ f } .
This contradicts the basis property. Hence S e i ⊆ E Λ i \mathcal{S}_{e_i} \subseteq\mathcal{E}_{\Lambda_i} S e i ⊆ E Λ i , proving (57) .
Rank one and level zero
For n = 1 n = 1 n = 1 we use the A 1 ( 1 ) A_1^{(1)} A 1 ( 1 ) basis theorem of Meurman and Primc [26 Theorem 6.5.5 and (11.2.1)–(11.2.3)] . In their notation, x = X 11 x = X_{11} x = X 11 , h = X 12 h = X_{12} h = X 12 and y = X 22 y = X_{22} y = X 22 . Let A j A_j A j , B j B_j B j , C j C_j C j denote the exponents of x ( − j ) x(-j) x ( − j ) , h ( − j ) h(-j) h ( − j ) , y ( − j ) y(-j) y ( − j ) , respectively, and set A 0 = B 0 = 0 A_0 = B_0 = 0 A 0 = B 0 = 0 . The three generators have principal degrees 2 j − 1 2j - 1 2 j − 1 , 2 j 2j 2 j and 2 j + 1 2j + 1 2 j + 1 , respectively. For j ≥ 1 j \ge1 j ≥ 1 , the four types of positive paths give the following bounds:
Table 1 (PDF p. 24).
For odd j j j the heights in the first and last rows are read in reverse order, since (35) interchanges the two outer rows when the parity of j j j changes. At j = 0 j = 0 j = 0 , the last two rows of the table still describe positive paths. The two boundary paths with exactly one positive vertex give
( − 1 , 0 , 1 ) : k 1 + A 1 ≤ k ⟺ A 1 ≤ k 0 , (-1,0,1):\quad k_1 + A_1 \le k \quad\Longleftrightarrow\quad A_1 \le k_0, ( − 1 , 0 , 1 ) : k 1 + A 1 ≤ k ⟺ A 1 ≤ k 0 ,
( 1 , 0 , − 1 ) : C 0 + k 0 ≤ k ⟺ C 0 ≤ k 1 . (1,0,-1):\quad C_0 + k_0 \le k \quad\Longleftrightarrow\quad C_0 \le k_1. ( 1 , 0 , − 1 ) : C 0 + k 0 ≤ k ⟺ C 0 ≤ k 1 .
These are the two initial conditions in [26 (11.2.3)] . They also give the remaining bounds at j = 0 j = 0 j = 0 , including the one for ( 1 , 0 , 1 ) (1,0,1) ( 1 , 0 , 1 ) :
A 1 ≤ k 0 ≤ k , A 1 + C 0 ≤ k 0 + k 1 = k . A_1 \le k_0 \le k,\qquad A_1 + C_0 \le k_0 + k_1 = k. A 1 ≤ k 0 ≤ k , A 1 + C 0 ≤ k 0 + k 1 = k .
The boundary path ( − 1 , 0 , − 1 ) (-1,0,-1) ( − 1 , 0 , − 1 ) has total frequency k 0 + k 1 = k k_0 + k_1 = k k 0 + k 1 = k .
To identify essential exponents we must also check the order. In rank one, (39) reads y ≺ h ≺ x y \prec h \prec x y ≺ h ≺ x . The successive comparisons by length, total loop degree, shape and colour also agree with the order of [26 §6.2] , including the direction of the last two comparisons. Write U ( π ) U(\pi) U ( π ) for the PBW product of an exponent π \pi π in this order. By the filtration in [26 §6.3] , π ∈ lt ( L ( Λ ) ) \pi\in\operatorname{lt}(L(\Lambda)) π ∈ lt ( L ( Λ )) precisely when U ( π ) v Λ U(\pi)v_\Lambda U ( π ) v Λ is not in the span of the PBW vectors with strictly larger exponents. This is Definition 4.1 . The four bounds in the table, read from bottom to top, are [26 (11.2.2)] with mode − j -j − j . Together with the initial conditions they give the required statement in rank one.
Theorem 4.4 (Meurman–Primc [26 Corollary 6.5.4 and Theorem 6.5.5] ). Let Λ = k 0 Λ 0 + k 1 Λ 1 \Lambda= k_0\Lambda_0 + k_1\Lambda_1 Λ = k 0 Λ 0 + k 1 Λ 1 , where k 0 , k 1 ∈ Z ≥ 0 k_0, k_1 \in\mathbb{Z}_{\geq0} k 0 , k 1 ∈ Z ≥ 0 and k = k 0 + k 1 > 0 k = k_0 + k_1 > 0 k = k 0 + k 1 > 0 . In the coordinates above,
lt ( L ( Λ ) ) = E Λ = S ( k 0 , k 1 ) . \operatorname{lt}(L(\Lambda)) = \mathcal{E}_{\Lambda} = S_{(k_0,k_1)}. lt ( L ( Λ )) = E Λ = S ( k 0 , k 1 ) . The vectors U ( π ) v Λ U(\pi)v_{\Lambda} U ( π ) v Λ indexed by this set form a basis of L ( Λ ) L(\Lambda) L ( Λ ) .
At level zero, a path through any positive occurrence has total frequency at least one, so S 0 = { 0 } S_0 = \{0\} S 0 = { 0 } and L ( 0 ) = C v 0 L(0) = \mathbb{C}v_0 L ( 0 ) = C v 0 .
Decomposition into marked layers
To pass from level one to higher levels, we first decompose an admissible array into layers of level one. Extend f ∈ S k f \in S_k f ∈ S k from Ω + \Omega_+ Ω + to Ω \Omega Ω by
F ( r , p ) = { f ( r , p ) , p > 0 , 0 , p = 0 , k i , ( r , p ) = B i = ( N − 2 i , − 1 ) . (60) F(r,p) =
\begin{cases}
f(r,p), & p > 0,\\
0, & p = 0,\\
k_i, & (r,p) = B_i = (N - 2i,-1).
\end{cases} \tag*{(60)} F ( r , p ) = ⎩ ⎨ ⎧ f ( r , p ) , 0 , k i , p > 0 , p = 0 , ( r , p ) = B i = ( N − 2 i , − 1 ) . (60)
For a vertex v v v in row r r r , let M ( v ) M(v) M ( v ) be the maximum sum of F F F along a path from row 0 to v v v , including F ( v ) F(v) F ( v ) . There are at most 2 r 2^r 2 r such paths, and the maximum satisfies
M − ( v ) = { 0 , r = 0 , max w → v M ( w ) , r > 0 , M ( v ) = M − ( v ) + F ( v ) , (61) M^{-}(v) =
\begin{cases}
0, & r = 0,\\
\max_{w \to v} M(w), & r > 0,
\end{cases}
\qquad
M(v) = M^{-}(v) + F(v), \tag*{(61)} M − ( v ) = { 0 , max w → v M ( w ) , r = 0 , r > 0 , M ( v ) = M − ( v ) + F ( v ) , (61)
where w → v w \to v w → v means that w w w is an immediate predecessor of v v v . Every path from row 0 to v v v extends to a path through all rows. Since F ≥ 0 F \geq0 F ≥ 0 , admissibility gives
0 ≤ M ( v ) = max Q ending at v ∑ w ∈ Q F ( w ) ≤ max P full ∑ w ∈ P F ( w ) ≤ k . 0 \leq M(v) = \max_{\substack{Q\ \text{ending at}\ v}} \sum_{w \in Q} F(w) \leq\max_{P\ \text{full}} \sum_{w \in P} F(w) \leq k. 0 ≤ M ( v ) = Q ending at v max w ∈ Q ∑ F ( w ) ≤ P full max w ∈ P ∑ F ( w ) ≤ k .
Proposition 4.5 (Marked layers and Minkowski sums of integer points). For 1 ≤ t ≤ k 1 \leq t \leq k 1 ≤ t ≤ k , define
F [ t ] ( v ) = { 1 , M − ( v ) < t ≤ M ( v ) , 0 , otherwise . (62) F^{[t]}(v) =
\begin{cases}
1, & M^{-}(v) < t \leq M(v),\\
0, & \text{otherwise}.
\end{cases} \tag*{(62)} F [ t ] ( v ) = { 1 , 0 , M − ( v ) < t ≤ M ( v ) , otherwise . (62) Then F = ∑ t = 1 k F [ t ] F = \sum_{t=1}^{k} F^{[t]} F = ∑ t = 1 k F [ t ] pointwise. Every full path meets at most one vertex at which F [ t ] F^{[t]} F [ t ] is nonzero. Each layer has exactly one boundary mark, with precisely k i k_i k i layers marked at B i B_i B i . Consequently,
S k = ∑ i = 0 n k i S e i , (63) S_k = \sum_{i=0}^{n} k_i S_{e_i}, \tag*{(63)} S k = i = 0 ∑ n k i S e i , (63) where k i S e i k_iS_{e_i} k i S e i denotes the Minkowski sum of k i k_i k i copies of S e i S_{e_i} S e i . Addition is pointwise and 0 S = { 0 } 0S = \{0\} 0 S = { 0 } .
Proof. At each vertex v v v , the integer interval defining the layers has size F ( v ) F(v) F ( v ) :
∑ t = 1 k F [ t ] ( v ) = M ( v ) − M − ( v ) = F ( v ) < ∞ . \sum_{t=1}^{k} F^{[t]}(v) = M(v) - M^{-}(v) = F(v) < \infty. t = 1 ∑ k F [ t ] ( v ) = M ( v ) − M − ( v ) = F ( v ) < ∞.
Along a path P = ( v 0 , … , v N ) P = (v_0,\ldots,v_N) P = ( v 0 , … , v N ) , the recurrence gives
M − ( v r + 1 ) ≥ M ( v r ) ≥ M − ( v r ) . (64) M^{-}(v_{r+1}) \geq M(v_r) \geq M^{-}(v_r). \tag*{(64)} M − ( v r + 1 ) ≥ M ( v r ) ≥ M − ( v r ) . (64)
If r r r is the first index with F [ t ] ( v r ) = 1 F^{[t]}(v_r)=1 F [ t ] ( v r ) = 1 , then M ( v r ) ≥ t M(v_r)\ge t M ( v r ) ≥ t , and hence M − ( v s ) ≥ t M^{-}(v_s)\ge t M − ( v s ) ≥ t for all s > r s>r s > r . It follows that
F [ t ] ( v r ) = 1 ⟹ F [ t ] ( v s ) = 0 ( s > r ) , ∑ v ∈ P F [ t ] ( v ) ≤ 1. F^{[t]}(v_r)=1 \quad\Longrightarrow\quad F^{[t]}(v_s)=0\ (s>r),\qquad\sum_{v\in P}F^{[t]}(v)\le1. F [ t ] ( v r ) = 1 ⟹ F [ t ] ( v s ) = 0 ( s > r ) , v ∈ P ∑ F [ t ] ( v ) ≤ 1.
Thus each path meets at most one nonzero entry of a given layer. Set K i = ∑ j = i n k j K_i=\sum_{j=i}^{n}k_j K i = ∑ j = i n k j and K n + 1 = 0 K_{n+1}=0 K n + 1 = 0 . The boundary zigzag from row 0 to B i B_i B i visits B n , B n − 1 , … , B i B_n,B_{n-1},\ldots,B_i B n , B n − 1 , … , B i , and therefore has total frequency K i K_i K i . Conversely, extend a path attaining M ( B i ) M(B_i) M ( B i ) along the boundary zigzag from B i B_i B i to B 0 B_0 B 0 . The added frequencies sum to k − K i k-K_i k − K i . Admissibility then gives
M ( B i ) ≥ K i , M ( B i ) + ( k − K i ) ≤ k . M(B_i)\ge K_i,\qquad M(B_i)+(k-K_i)\le k. M ( B i ) ≥ K i , M ( B i ) + ( k − K i ) ≤ k .
Thus
M ( B i ) = K i , M − ( B i ) = K i − k i = K i + 1 . (65) M(B_i)=K_i,\qquad M^{-}(B_i)=K_i-k_i=K_{i+1}. \tag*{(65)} M ( B i ) = K i , M − ( B i ) = K i − k i = K i + 1 . (65)
In particular,
F [ t ] ( B i ) = 1 ⟺ K i + 1 < t ≤ K i , { 1 , … , k } = ⋃ i = 0 n { K i + 1 + 1 , … , K i } . F^{[t]}(B_i)=1\quad\Longleftrightarrow\quad K_{i+1}<t\le K_i,\qquad\{1,\ldots,k\}=\bigcup_{i=0}^{n}\{K_{i+1}+1,\ldots,K_i\}. F [ t ] ( B i ) = 1 ⟺ K i + 1 < t ≤ K i , { 1 , … , k } = i = 0 ⋃ n { K i + 1 + 1 , … , K i } .
Every layer therefore has exactly one boundary mark, and k i k_i k i layers have type e i e_i e i . Restricting to Ω + \Omega_+ Ω + yields
f = ∑ t = 1 k f [ t ] , f [ t ] ∈ S e i ( K i + 1 < t ≤ K i ) , f=\sum_{t=1}^{k}f^{[t]},\qquad f^{[t]}\in S_{e_i}\quad(K_{i+1}<t\le K_i), f = t = 1 ∑ k f [ t ] , f [ t ] ∈ S e i ( K i + 1 < t ≤ K i ) ,
which proves one inclusion in (4.13). Conversely, take arrays g i , j ∈ S e i g_{i,j}\in S_{e_i} g i , j ∈ S e i , 1 ≤ j ≤ k i 1\le j\le k_i 1 ≤ j ≤ k i , and let G i , j G_{i,j} G i , j be their boundary extensions. For every path P P P ,
∑ v ∈ P ∑ i = 0 n ∑ j = 1 k i G i , j ( v ) = ∑ i = 0 n ∑ j = 1 k i ∑ v ∈ P G i , j ( v ) ≤ ∑ i = 0 n k i = k . \sum_{v\in P}\sum_{i=0}^{n}\sum_{j=1}^{k_i}G_{i,j}(v)=\sum_{i=0}^{n}\sum_{j=1}^{k_i}\sum_{v\in P}G_{i,j}(v)\le\sum_{i=0}^{n}k_i=k. v ∈ P ∑ i = 0 ∑ n j = 1 ∑ k i G i , j ( v ) = i = 0 ∑ n j = 1 ∑ k i v ∈ P ∑ G i , j ( v ) ≤ i = 0 ∑ n k i = k .
The sum has boundary marks k i k_i k i , so ∑ i , j g i , j ∈ S k \sum_{i,j}g_{i,j}\in S_k ∑ i , j g i , j ∈ S k .
Example 4.6. Let n = 2 n=2 n = 2 and k = ( 1 , 2 , 1 ) \mathbf{k}=(1,2,1) k = ( 1 , 2 , 1 ) . Put multiplicity two at ( 4 , 5 ) (4,5) ( 4 , 5 ) , multiplicity one at
( 0 , 1 ) , ( 1 , 2 ) , ( 2 , 1 ) , ( 2 , 5 ) , ( 3 , 2 ) , (0,1),\quad(1,2),\quad(2,1),\quad(2,5),\quad(3,2), ( 0 , 1 ) , ( 1 , 2 ) , ( 2 , 1 ) , ( 2 , 5 ) , ( 3 , 2 ) ,
and zero elsewhere in Ω + \Omega_+ Ω + . The following table lists all vertices with nonzero F ( v ) F(v) F ( v ) and shows how (4.11) assigns each occurrence to a layer.
Table 2.
The recurrence gives
( M ( 4 , − 1 ) , M ( 4 , 1 ) , M ( 4 , 3 ) , M ( 4 , 5 ) , M ( 4 , 7 ) ) = ( 4 , 4 , 4 , 4 , 1 ) , M ( 4 , p ) = 0 ( p ≥ 9 ) . \bigl(M(4,-1),M(4,1),M(4,3),M(4,5),M(4,7)\bigr)=(4,4,4,4,1),\qquad M(4,p)=0\quad(p\ge9). ( M ( 4 , − 1 ) , M ( 4 , 1 ) , M ( 4 , 3 ) , M ( 4 , 5 ) , M ( 4 , 7 ) ) = ( 4 , 4 , 4 , 4 , 1 ) , M ( 4 , p ) = 0 ( p ≥ 9 ) .
Figure 5. The four marked layers of Example 4.6. Rows are numbered r = 0 , … , 4 r = 0,\ldots,4 r = 0 , … , 4 from top to bottom. The horizontal labels give the heights p p p . Blue vertices mark the boundary entries, and red vertices mark positive occurrences. All unlabelled frequencies are zero.
Essential exponents in tensor products
To use the decomposition in (63) , we show that essential exponents add under the natural embedding of highest weight modules. On the tensor product of two g ^ \widehat{\mathfrak{g}} g -modules V V V and W W W , the action is given by
x ( u ⊗ v ) = x u ⊗ v + u ⊗ x v ( x ∈ g ^ ) . x(u \otimes v) = xu \otimes v + u \otimes xv \qquad(x \in\widehat{\mathfrak{g}}). x ( u ⊗ v ) = xu ⊗ v + u ⊗ xv ( x ∈ g ) .
It is induced by the coproduct on the universal enveloping algebra:
Δ ( x ) = x ⊗ 1 + 1 ⊗ x , Δ ( a b ) = Δ ( a ) Δ ( b ) . \Delta(x) = x \otimes1 + 1 \otimes x,\qquad\Delta(ab) = \Delta(a)\Delta(b). Δ ( x ) = x ⊗ 1 + 1 ⊗ x , Δ ( ab ) = Δ ( a ) Δ ( b ) .
Indeed, the multiplication rule ( a ⊗ b ) ( c ⊗ d ) = a c ⊗ b d (a \otimes b)(c \otimes d) = ac \otimes bd ( a ⊗ b ) ( c ⊗ d ) = a c ⊗ b d gives
[ Δ ( x ) , Δ ( y ) ] = [ x , y ] ⊗ 1 + 1 ⊗ [ x , y ] = Δ ( [ x , y ] ) , [\Delta(x),\Delta(y)] = [x,y] \otimes1 + 1 \otimes[x,y] = \Delta([x,y]), [ Δ ( x ) , Δ ( y )] = [ x , y ] ⊗ 1 + 1 ⊗ [ x , y ] = Δ ([ x , y ]) ,
so the assignment on generators respects the defining relations of U ( g ^ ) U(\widehat{\mathfrak{g}}) U ( g ) .
For a finitely supported exponent s s s , write X s = x 1 s 1 ⋯ x m s m X^s = x_1^{s_1}\cdots x_m^{s_m} X s = x 1 s 1 ⋯ x m s m in the fixed PBW order. Since x i ⊗ 1 x_i \otimes1 x i ⊗ 1 and 1 ⊗ x i 1 \otimes x_i 1 ⊗ x i commute,
Δ ( X s ) = ∏ i = 1 m ( x i ⊗ 1 + 1 ⊗ x i ) s i = ∑ 0 ≤ a i ≤ s i ( ∏ i = 1 m ( s i a i ) ) x 1 a 1 ⋯ x m a m ⊗ x 1 s 1 − a 1 ⋯ x m s m − a m . \begin{aligned}
\Delta(X^s) = \prod_{i=1}^{m}(x_i \otimes1 + 1 \otimes x_i)^{s_i} \\
&= \sum_{0\leq a_i\leq s_i}\left(\prod_{i=1}^{m}\binom{s_i}{a_i}\right)x_1^{a_1}\cdots x_m^{a_m}\otimes x_1^{s_1-a_1}\cdots x_m^{s_m-a_m}.
\end{aligned} Δ ( X s ) = i = 1 ∏ m ( x i ⊗ 1 + 1 ⊗ x i ) s i = 0 ≤ a i ≤ s i ∑ ( i = 1 ∏ m ( a i s i ) ) x 1 a 1 ⋯ x m a m ⊗ x 1 s 1 − a 1 ⋯ x m s m − a m .
Thus
Δ ( X s ) = ∑ a + b = s ( s a ) X a ⊗ X b , ( s a ) = ∏ x ( s x a x ) . (66) \Delta(X^s) = \sum_{a+b=s}\binom{s}{a}X^a\otimes X^b,\qquad\binom{s}{a}=\prod_x\binom{s_x}{a_x}. \tag*{(66)} Δ ( X s ) = a + b = s ∑ ( a s ) X a ⊗ X b , ( a s ) = x ∏ ( a x s x ) . (66)
For cyclic modules of finite dimension, the addition property for essential exponents is proved in [14 Proposition 2.11] . We give the argument for affine highest weight modules with the order used here.
Theorem 4.7 (Highest-weight tensor components and essential addition). For dominant integral weights λ \lambda λ and μ \mu μ , there is an isomorphism
L ( λ + μ ) → ∼ C : = U ( g ^ ) ( v λ ⊗ v μ ) , v λ + μ ⟼ v λ ⊗ v μ . L(\lambda+\mu)\xrightarrow{\sim} C:=U(\widehat{\mathfrak{g}})(v_\lambda\otimes v_\mu),\qquad v_{\lambda+\mu}\longmapsto v_\lambda\otimes v_\mu. L ( λ + μ ) ∼ C := U ( g ) ( v λ ⊗ v μ ) , v λ + μ ⟼ v λ ⊗ v μ . For the fixed additive comparison order,
E λ + E μ ⊆ E λ + μ . (67) E_\lambda+E_\mu\subseteq E_{\lambda+\mu}. \tag*{(67)} E λ + E μ ⊆ E λ + μ . (67)
Proof. Put w = v λ ⊗ v μ w=v_{\lambda}\otimes v_{\mu} w = v λ ⊗ v μ , and let e i , f i e_i,f_i e i , f i denote the Chevalley generators. For each i i i and each H ∈ h ^ H\in\widehat{\mathfrak h} H ∈ h ,
e i w = ( e i v λ ) ⊗ v μ + v λ ⊗ ( e i v μ ) = 0 , e_iw=(e_iv_{\lambda})\otimes v_{\mu}+v_{\lambda}\otimes(e_iv_{\mu})=0, e i w = ( e i v λ ) ⊗ v μ + v λ ⊗ ( e i v μ ) = 0 ,
H w = ( H v λ ) ⊗ v μ + v λ ⊗ ( H v μ ) = ( λ ( H ) + μ ( H ) ) w . Hw=(Hv_{\lambda})\otimes v_{\mu}+v_{\lambda}\otimes(Hv_{\mu})=(\lambda(H)+\mu(H))w. H w = ( H v λ ) ⊗ v μ + v λ ⊗ ( H v μ ) = ( λ ( H ) + μ ( H )) w .
We next check the simple root relations on w w w . If z = e i z=e_i z = e i or f i f_i f i , a ∈ L ( λ ) a\in L(\lambda) a ∈ L ( λ ) and b ∈ L ( μ ) b\in L(\mu) b ∈ L ( μ ) , integrability of the two factors gives A , B ≥ 1 A,B\geq1 A , B ≥ 1 such that z A a = z B b = 0 z^Aa=z^Bb=0 z A a = z B b = 0 . In every summand of
Δ ( z ) A + B − 1 ( a ⊗ b ) = ∑ q = 0 A + B − 1 ( A + B − 1 q ) z q a ⊗ z A + B − 1 − q b \Delta(z)^{A+B-1}(a\otimes b)=\sum_{q=0}^{A+B-1}\binom{A+B-1}{q}z^qa\otimes z^{A+B-1-q}b Δ ( z ) A + B − 1 ( a ⊗ b ) = q = 0 ∑ A + B − 1 ( q A + B − 1 ) z q a ⊗ z A + B − 1 − q b
either q ≥ A q\geq A q ≥ A or A + B − 1 − q ≥ B A+B-1-q\geq B A + B − 1 − q ≥ B . Thus the sum vanishes. Since every tensor is a finite sum of pure tensors, Δ ( z ) \Delta(z) Δ ( z ) acts locally nilpotently. In particular, with z = f i z=f_i z = f i , a = v λ a=v_{\lambda} a = v λ , b = v μ b=v_{\mu} b = v μ , A = λ ( h i ) + 1 A=\lambda(h_i)+1 A = λ ( h i ) + 1 and B = μ ( h i ) + 1 B=\mu(h_i)+1 B = μ ( h i ) + 1 , we obtain
f i ( λ + μ ) ( h i ) + 1 w = 0. f_i^{(\lambda+\mu)(h_i)+1}w=0. f i ( λ + μ ) ( h i ) + 1 w = 0.
The vector w w w therefore satisfies the defining relations for v λ + μ v_{\lambda+\mu} v λ + μ in Theorem 2.2 . Equivalently, if Ann ( v ) = { a ∈ U ( g ^ ) : a v = 0 } \operatorname{Ann}(v)=\{a\in U(\widehat{\mathfrak g}):av=0\} Ann ( v ) = { a ∈ U ( g ) : a v = 0 } denotes the annihilator left ideal, then
Ann ( v λ + μ ) ⊆ Ann ( w ) . \operatorname{Ann}(v_{\lambda+\mu})\subseteq\operatorname{Ann}(w). Ann ( v λ + μ ) ⊆ Ann ( w ) .
Hence
ι : L ( λ + μ ) ⟶ C , a v λ + μ ⟼ a w ( a ∈ U ( g ^ ) ) \iota:L(\lambda+\mu)\longrightarrow\mathcal{C},\qquad av_{\lambda+\mu}\longmapsto aw\quad(a\in U(\widehat{\mathfrak g})) ι : L ( λ + μ ) ⟶ C , a v λ + μ ⟼ a w ( a ∈ U ( g ))
is a well-defined surjective homomorphism. Since w ≠ 0 w\neq0 w = 0 and its source is irreducible, ι \iota ι is injective as well. The triangular decomposition also gives C = U ( n ^ − ) w \mathcal{C}=U(\widehat{\mathfrak n}_{-})w C = U ( n − ) w .
Now fix u ∈ E λ u\in\mathcal{E}_{\lambda} u ∈ E λ and v ∈ E μ v\in\mathcal{E}_{\mu} v ∈ E μ . In the weight space W u = L ( λ ) λ − β ( u ) W_u=L(\lambda)_{\lambda-\beta(u)} W u = L ( λ ) λ − β ( u ) , let
T u = span { X a v λ : a ≻ u , β ( a ) = β ( u ) } . T_u=\operatorname{span}\{X^av_{\lambda}:a\succ u,\ \beta(a)=\beta(u)\}. T u = span { X a v λ : a ≻ u , β ( a ) = β ( u )} .
Essentiality means X u v λ ∉ T u X^uv_{\lambda}\notin T_u X u v λ ∈ / T u . Thus there is a linear functional on W u W_u W u which takes value 1 1 1 on X u v λ X^uv_{\lambda} X u v λ and vanishes on T u T_u T u . Extend it by zero on all other weight spaces, and denote the resulting functional by ϕ \phi ϕ . Then
ϕ ( X u v λ ) = 1 , ϕ ( X a v λ ) = 0 ( a ≻ u ) . \phi(X^uv_{\lambda})=1,\qquad\phi(X^av_{\lambda})=0\quad(a\succ u). ϕ ( X u v λ ) = 1 , ϕ ( X a v λ ) = 0 ( a ≻ u ) .
Applying the same construction in the second factor gives a linear functional ψ ∈ L ( μ ) ∗ \psi\in L(\mu)^* ψ ∈ L ( μ ) ∗ with
ψ ( X v v μ ) = 1 , ψ ( X b v μ ) = 0 ( b ≻ v ) . \psi(X^vv_{\mu})=1,\qquad\psi(X^bv_{\mu})=0\quad(b\succ v). ψ ( X v v μ ) = 1 , ψ ( X b v μ ) = 0 ( b ≻ v ) .
Put Φ = ( ϕ ⊗ ψ ) ∘ ι \Phi=(\phi\otimes\psi)\circ\iota Φ = ( ϕ ⊗ ψ ) ∘ ι , where ( ϕ ⊗ ψ ) ( a ⊗ b ) = ϕ ( a ) ψ ( b ) (\phi\otimes\psi)(a\otimes b)=\phi(a)\psi(b) ( ϕ ⊗ ψ ) ( a ⊗ b ) = ϕ ( a ) ψ ( b ) on pure tensors. By (66) ,
Φ ( X s v λ + μ ) = ∑ a + b = s ( s a ) ϕ ( X a v λ ) ψ ( X b v μ ) . \Phi(X^sv_{\lambda+\mu})=\sum_{a+b=s}\binom{s}{a}\phi(X^av_{\lambda})\psi(X^bv_{\mu}). Φ ( X s v λ + μ ) = a + b = s ∑ ( a s ) ϕ ( X a v λ ) ψ ( X b v μ ) .
A nonzero summand requires a ⪯ u a\preceq u a ⪯ u and b ⪯ v b\preceq v b ⪯ v . By additivity of the comparison order, it therefore requires s = a + b ⪯ u + v s=a+b\preceq u+v s = a + b ⪯ u + v . In particular,
Φ ( X s v λ + μ ) = 0 ( s ≻ u + v ) . \Phi(X^sv_{\lambda+\mu})=0\qquad(s\succ u+v). Φ ( X s v λ + μ ) = 0 ( s ≻ u + v ) .
When s = u + v s=u+v s = u + v , a nonzero summand with ( a , b ) ≠ ( u , v ) (a,b)\neq(u,v) ( a , b ) = ( u , v ) would have either a ≺ u a\prec u a ≺ u or b ≺ v b\prec v b ≺ v , and additivity would give a + b ≺ u + v a+b\prec u+v a + b ≺ u + v . Thus only ( a , b ) = ( u , v ) (a,b)=(u,v) ( a , b ) = ( u , v ) contributes, and
Φ ( X u + v v λ + μ ) = ( u + v u ) ≠ 0. \Phi(X^{u+v}v_{\lambda+\mu})=\binom{u+v}{u}\neq0. Φ ( X u + v v λ + μ ) = ( u u + v ) = 0.
If u + v u+v u + v were not essential, X u + v v λ + μ X^{u+v}v_{\lambda+\mu} X u + v v λ + μ would be a finite linear combination of vectors X f v λ + μ X^f v_{\lambda+\mu} X f v λ + μ with f ≻ u + v f \succ u+v f ≻ u + v . Applying Φ \Phi Φ would make the left side nonzero and the right side zero, a contradiction. Therefore u + v ∈ E λ + μ u+v \in E_{\lambda+\mu} u + v ∈ E λ + μ .
Theorem 4.8. For Λ = ∑ i = 0 n k i Λ i \Lambda=\sum_{i=0}^{n}k_i\Lambda_i Λ = ∑ i = 0 n k i Λ i , the vectors X f v Λ X^f v_\Lambda X f v Λ , f ∈ S k f\in S_k f ∈ S k , are linearly independent.
Proof. For k > 0 k>0 k > 0 , apply Proposition 4.5 and the basis at level one (Theorem 4.3 for n ≥ 2 n\ge2 n ≥ 2 , Theorem 4.4 for n = 1 n=1 n = 1 ). Repeated application of (67) gives
S k = ∑ i k i S e i = ∑ i k i E Λ i ⊆ E Λ . (68) S_k=\sum_i k_i S_{e_i}=\sum_i k_i E_{\Lambda_i}\subseteq E_\Lambda. \tag*{(68)} S k = i ∑ k i S e i = i ∑ k i E Λ i ⊆ E Λ . (68) The vectors indexed by E Λ E_\Lambda E Λ are linearly independent by (53) . At k = 0 k=0 k = 0 the claim is immediate.
Finite Levi subalgebras and spanning for mixed boundaries
The preceding section proves independence. To prove spanning, we must also straighten monomials that violate an inequality because of a boundary entry. We cut the offending path into maximal positive subpaths and use the corresponding finite Levi modules to derive the required relations in the original Primc–Šikić (PS) order.
Throughout this section, n ≥ 2 n\ge2 n ≥ 2 , N = 2 n N=2n N = 2 n , and Λ = ∑ i = 0 n k i Λ i \Lambda=\sum_{i=0}^{n}k_i\Lambda_i Λ = ∑ i = 0 n k i Λ i , where k i ∈ Z ≥ 0 k_i\in\mathbb{Z}_{\ge0} k i ∈ Z ≥ 0 and k = ∑ i k i k=\sum_i k_i k = ∑ i k i . Recall that vertices satisfy 0 ≤ r ≤ N 0\le r\le N 0 ≤ r ≤ N , p ≥ − 1 p\ge-1 p ≥ − 1 and r + p r+p r + p odd. The edges are ( r , p ) → ( r + 1 , p ± 1 ) (r,p)\to(r+1,p\pm1) ( r , p ) → ( r + 1 , p ± 1 ) , and B i = ( N − 2 i , − 1 ) B_i=(N-2i,-1) B i = ( N − 2 i , − 1 ) carries the mark k i k_i k i . Each positive vertex indexes a basis vector of n ^ − \hat{\mathfrak{n}}_- n ^ − . Its height p p p is the height of the corresponding lowering root. We shall use two properties of the PS order:
u ≺ v ⟹ u + w ≺ v + w , ℓ ( u ) > ℓ ( v ) ⟹ u ≺ v . (69) u\prec v\Longrightarrow u+w\prec v+w,\qquad\ell(u)>\ell(v)\Longrightarrow u\prec v. \tag*{(69)} u ≺ v ⟹ u + w ≺ v + w , ℓ ( u ) > ℓ ( v ) ⟹ u ≺ v . (69)
We already know that E Λ j = S e j E_{\Lambda_j}=S_{e_j} E Λ j = S e j and S k ⊆ E Λ S_k\subseteq E_\Lambda S k ⊆ E Λ .
Fix an arbitrary inadmissible exponent f ∉ S k f\notin S_k f ∈ / S k . Choose a path P P P along which the sum of the positive frequencies and boundary entries exceeds k k k , and write p 0 , p 1 , … , p N p_0,p_1,\ldots,p_N p 0 , p 1 , … , p N for its heights.
A maximal positive subpath of P P P is a maximal consecutive string of vertices of height at least one. For example, the heights 2 , 1 , 0 , 1 2,1,0,1 2 , 1 , 0 , 1 give two positive subpaths, separated by the vertex of height zero. For a maximal positive subpath B B B occupying rows [ a , b ] [a,b] [ a , b ] , set
I B = [ n − b / 2 , n − a / 2 ] ∩ Z , m ( B ) = ∑ r = a b f ( r , p r ) , c ( B ) = ∑ i ∈ I B k i . I_B=[n-b/2,n-a/2]\cap\mathbb{Z},\qquad m(B)=\sum_{r=a}^{b}f(r,p_r),\qquad c(B)=\sum_{i\in I_B}k_i. I B = [ n − b /2 , n − a /2 ] ∩ Z , m ( B ) = r = a ∑ b f ( r , p r ) , c ( B ) = i ∈ I B ∑ k i .
We call c ( B ) c(B) c ( B ) the capacity of B B B .
Maximality and the step condition give
a > 0 ⟹ ( p a − 1 , p a ) = ( 0 , 1 ) , b < N ⟹ ( p b , p b + 1 ) = ( 1 , 0 ) . a>0\Longrightarrow(p_{a-1},p_a)=(0,1),\qquad b<N\Longrightarrow(p_b,p_{b+1})=(1,0). a > 0 ⟹ ( p a − 1 , p a ) = ( 0 , 1 ) , b < N ⟹ ( p b , p b + 1 ) = ( 1 , 0 ) .
Both endpoint rows are even: an interior endpoint has height 1 1 1 , and the outer rows 0 , N 0,N 0 , N are even. It follows that
a , b ∈ 2 Z , ∣ I B ∣ = b − a 2 + 1 ≥ 1. a,b\in2\mathbb{Z},\qquad|I_B|=\frac{b-a}{2}+1\ge1. a , b ∈ 2 Z , ∣ I B ∣ = 2 b − a + 1 ≥ 1.
At the even row N − 2 i N-2i N − 2 i , the height is either − 1 -1 − 1 or positive. In the latter case the row belongs to a unique maximal positive subpath B B B , and a ≤ N − 2 i ≤ b a\le N-2i\le b a ≤ N − 2 i ≤ b is equivalent to i ∈ I B i\in I_B i ∈ I B when B B B occupies rows [ a , b ] [a,b] [ a , b ] . The intervals I B I_B I B are disjoint and
⋃ B I B = { 0 , … , n } ∖ { i : B i ∈ P } . (70) \bigcup_B I_B=\{0,\ldots,n\}\setminus\{i:B_i\in P\}. \tag*{(70)} B ⋃ I B = { 0 , … , n } ∖ { i : B i ∈ P } . (70)
where B i = ( N − 2 i , − 1 ) B_i=(N-2i,-1) B i = ( N − 2 i , − 1 ) are the marked boundary vertices. Since the marks have total weight k k k , the assumed violation gives
∑ B ( m ( B ) − c ( B ) ) = ∑ B m ( B ) + ∑ i : B i ∈ P k i − k = ∑ v ∈ P F ( v ) − k > 0. (71) \sum_B \bigl(m(B)-c(B)\bigr)=\sum_B m(B)+\sum_{i:B_i\in P}k_i-k=\sum_{v\in P}F(v)-k>0. \tag*{(71)} B ∑ ( m ( B ) − c ( B ) ) = B ∑ m ( B ) + i : B i ∈ P ∑ k i − k = v ∈ P ∑ F ( v ) − k > 0. (71)
Therefore m ( B ) > c ( B ) m(B)>c(B) m ( B ) > c ( B ) for at least one positive subpath.
If such a subpath B B B meets both outer rows, then c ( B ) = k c(B)=k c ( B ) = k and B B B contains at least k + 1 k+1 k + 1 positive occurrences. Choose k + 1 k+1 k + 1 of these occurrences, counted with multiplicity, and let u u u be their exponent. For k > 0 k>0 k > 0 , Lemma 3.6 gives
X u v Λ = ∑ z ≻ u c z X z v Λ , β ( z ) = β ( u ) , (72) X^{u}v_{\Lambda}=\sum_{z\succ u}c_zX^{z}v_{\Lambda},\qquad\beta(z)=\beta(u), \tag*{(72)} X u v Λ = z ≻ u ∑ c z X z v Λ , β ( z ) = β ( u ) , (72)
with finitely many terms.
Otherwise I B I_B I B is a proper interval. The total positive frequency on P P P need not exceed k k k , but the inequality m ( B ) > c ( B ) m(B)>c(B) m ( B ) > c ( B ) will yield a straightening relation in the finite Levi subalgebra on I B I_B I B . We first explain how a relation in the corresponding finite module gives one in L ( Λ ) L(\Lambda) L ( Λ ) with the same order on monomials.
Restriction to finite Levi subalgebras
Let α 0 , … , α n \alpha_0,\ldots,\alpha_n α 0 , … , α n and h 0 , … , h n h_0,\ldots,h_n h 0 , … , h n be the affine simple roots and coroots. In the affine root lattice
Q = ⨁ i = 0 n Z α i , Q + = ⨁ i = 0 n Z ≥ 0 α i Q=\bigoplus_{i=0}^{n}\mathbb{Z}\alpha_i,\qquad Q_+=\bigoplus_{i=0}^{n}\mathbb{Z}_{\geq0}\alpha_i Q = i = 0 ⨁ n Z α i , Q + = i = 0 ⨁ n Z ≥ 0 α i
the simple roots are linearly independent. For a positive root γ = ∑ i q i α i ∈ Q + \gamma=\sum_iq_i\alpha_i\in Q_+ γ = ∑ i q i α i ∈ Q + , put supp ( γ ) = { i : q i ≠ 0 } \operatorname{supp}(\gamma)=\{i:q_i\ne0\} supp ( γ ) = { i : q i = 0 } .
Let I = [ s , t ] ∩ Z I=[s,t]\cap\mathbb{Z} I = [ s , t ] ∩ Z be a nonempty proper interval in { 0 , … , n } \{0,\ldots,n\} { 0 , … , n } . The subalgebra g I \mathfrak{g}_I g I generated by e i , f i , h i e_i,f_i,h_i e i , f i , h i for i ∈ I i\in I i ∈ I is the derived algebra of the corresponding standard Levi subalgebra. Write Δ I + \Delta_I^+ Δ I + for its positive roots and n I − \mathfrak{n}_I^- n I − for the sum of its negative root spaces. Its Cartan subalgebra is h I = span { h i : i ∈ I } \mathfrak{h}_I=\operatorname{span}\{h_i:i\in I\} h I = span { h i : i ∈ I } , with dual h I ∗ = Hom C ( h I , C ) \mathfrak{h}_I^*=\operatorname{Hom}_{\mathbb{C}}(\mathfrak{h}_I,\mathbb{C}) h I ∗ = Hom C ( h I , C ) . The type of g I \mathfrak{g}_I g I is
{ A t − s + 1 , 0 < s ≤ t < n , C t − s + 1 , s = 0 or t = n , \begin{cases}
A_{t-s+1}, & 0<s\leq t<n,\\
C_{t-s+1}, & s=0\text{ or }t=n,
\end{cases} { A t − s + 1 , C t − s + 1 , 0 < s ≤ t < n , s = 0 or t = n ,
where A 1 = C 1 A_1=C_1 A 1 = C 1 . The fundamental weights ω i ∈ h I ∗ \omega_i\in\mathfrak{h}_I^* ω i ∈ h I ∗ satisfy ω i ( h j ) = δ i j \omega_i(h_j)=\delta_{ij} ω i ( h j ) = δ ij . Set Q I = ∑ i ∈ I Z α i Q_I=\sum_{i\in I}\mathbb{Z}\alpha_i Q I = ∑ i ∈ I Z α i , Q I + = ∑ i ∈ I Z ≥ 0 α i Q_I^+=\sum_{i\in I}\mathbb{Z}_{\geq0}\alpha_i Q I + = ∑ i ∈ I Z ≥ 0 α i and
M I = { u : supp ( γ ) ⊆ I whenever u γ , a > 0 } . (73) \mathcal{M}_I=\{u:\operatorname{supp}(\gamma)\subseteq I\text{ whenever }u_{\gamma,a}>0\}. \tag*{(73)} M I = { u : supp ( γ ) ⊆ I whenever u γ , a > 0 } . (73)
where a a a distinguishes basis vectors in the same root space, and β ( u ) = ∑ γ , a u γ , a γ \beta(u)=\sum_{\gamma,a}u_{\gamma,a}\gamma β ( u ) = ∑ γ , a u γ , a γ . The monoid Q I + Q_I^+ Q I + consists of the lattice points in the following face of Q R , + = ∑ i R ≥ 0 α i Q_{\mathbb{R},+}=\sum_i\mathbb{R}_{\geq0}\alpha_i Q R , + = ∑ i R ≥ 0 α i :
Q R , + ∩ span R { α i : i ∈ I } = { ∑ i q i α i : q i ≥ 0 , q j = 0 ( j ∉ I ) } . Q_{\mathbb{R},+}\cap\operatorname{span}_{\mathbb{R}}\{\alpha_i:i\in I\}
=
\left\{\sum_iq_i\alpha_i:q_i\geq0,\ q_j=0\ (j\notin I)\right\}. Q R , + ∩ span R { α i : i ∈ I } = { i ∑ q i α i : q i ≥ 0 , q j = 0 ( j ∈ / I ) } .
Every nonzero iterated Lie bracket of the simple lowering generators has lowering weight equal to the sum of their simple roots. Hence a Lie bracket with lowering weight in Q I + Q_I^+ Q I + can be nonzero only if it uses generators f i f_i f i with i ∈ I i\in I i ∈ I . Since n − \mathfrak{n}^- n − is generated by the f i f_i f i , the root spaces with support in I I I are precisely those of n I − \mathfrak{n}_I^- n I − . Thus M I \mathcal{M}_I M I is the set of exponents for the inherited PBW generators of n I − \mathfrak{n}_I^- n I − .
Let V I ( μ ) V_I(\mu) V I ( μ ) be the irreducible finite-dimensional g I \mathfrak{g}_I g I -module of highest weight μ = ∑ i ∈ I k i ω i I \mu=\sum_{i\in I}k_i\omega_i^I μ = ∑ i ∈ I k i ω i I . Existence and uniqueness follow from the classification of modules of highest weight [20 Chapter 10] . Choose a highest weight vector v μ v_\mu v μ . Write E I ( μ ) \mathcal{E}_I(\mu) E I ( μ ) for its essential exponents with respect to these generators and the restricted PS order.
Proposition 5.1 (Cyclic Levi modules and restriction of essential exponents). For μ = ∑ i ∈ I k i ω i I \mu=\sum_{i\in I}k_i\omega_i^I μ = ∑ i ∈ I k i ω i I , the assignment v μ ↦ v Λ v_\mu\mapsto v_\Lambda v μ ↦ v Λ extends to an isomorphism of g I \mathfrak{g}_I g I -modules
V I ( μ ) → ∼ C I ( Λ ) : = U ( g I ) v Λ = U ( n I − ) v Λ ⊆ L ( Λ ) . (74) V_I(\mu)\xrightarrow{\sim}C_I(\Lambda):=U(\mathfrak{g}_I)v_\Lambda=U(\mathfrak{n}_I^{-})v_\Lambda\subseteq L(\Lambda). \tag*{(74)} V I ( μ ) ∼ C I ( Λ ) := U ( g I ) v Λ = U ( n I − ) v Λ ⊆ L ( Λ ) . (74) Moreover,
β ( u ) ∈ Q I + ⟺ u ∈ M I . \beta(u)\in Q_I^+\quad\Longleftrightarrow\quad u\in M_I. β ( u ) ∈ Q I + ⟺ u ∈ M I . With the same exponent order on both sides,
E I ( μ ) = E Λ ∩ M I . \mathcal{E}_I(\mu)=\mathcal{E}_\Lambda\cap M_I. E I ( μ ) = E Λ ∩ M I . In particular, for j ∈ I j\in I j ∈ I ,
E I ( ω j I ) = S e j ∩ M I . (75) \mathcal{E}_I(\omega_j^I)=S_{e_j}\cap M_I. \tag*{(75)} E I ( ω j I ) = S e j ∩ M I . (75)
Proof. The cyclic module. Since I I I is a proper connected interval, g I \mathfrak{g}_I g I is of finite type. The defining relations of L ( Λ ) L(\Lambda) L ( Λ ) restrict, for i ∈ I i\in I i ∈ I , to
e i v Λ = 0 , h i v Λ = k i v Λ , f i k i + 1 v Λ = 0. (76) e_i v_\Lambda=0,\qquad h_i v_\Lambda=k_i v_\Lambda,\qquad f_i^{k_i+1}v_\Lambda=0. \tag*{(76)} e i v Λ = 0 , h i v Λ = k i v Λ , f i k i + 1 v Λ = 0. (76)
The triangular decomposition gives
C I ( Λ ) = U ( n I − ) U ( h I ) U ( n I + ) v Λ = U ( n I − ) v Λ . C_I(\Lambda)=U(\mathfrak{n}_I^-)U(\mathfrak{h}_I)U(\mathfrak{n}_I^+)v_\Lambda=U(\mathfrak{n}_I^-)v_\Lambda. C I ( Λ ) = U ( n I − ) U ( h I ) U ( n I + ) v Λ = U ( n I − ) v Λ .
Let M I ( μ ) M_I(\mu) M I ( μ ) be the Verma module with highest weight vector m μ m_\mu m μ . Its universal property gives a surjection
π : M I ( μ ) ⟶ C I ( Λ ) , x m μ ⟼ x v Λ . \pi:M_I(\mu)\longrightarrow C_I(\Lambda),\qquad xm_\mu\longmapsto xv_\Lambda. π : M I ( μ ) ⟶ C I ( Λ ) , x m μ ⟼ x v Λ .
By (76) ,
J I : = ∑ i ∈ I U ( n I − ) f i k i + 1 m μ ⊆ ker π . J_I:=\sum_{i\in I}U(\mathfrak{n}_I^-)f_i^{k_i+1}m_\mu\subseteq\ker\pi. J I := i ∈ I ∑ U ( n I − ) f i k i + 1 m μ ⊆ ker π .
The highest weight presentation [20 Corollary 10.4] identifies M I ( μ ) / J I M_I(\mu)/J_I M I ( μ ) / J I with V I ( μ ) V_I(\mu) V I ( μ ) . Thus π \pi π factors as
ι I : V I ( μ ) ↠ C I ( Λ ) , x v μ ⟼ x v Λ . (77) \iota_I:V_I(\mu)\twoheadrightarrow C_I(\Lambda),\qquad xv_\mu\longmapsto xv_\Lambda. \tag*{(77)} ι I : V I ( μ ) ↠ C I ( Λ ) , x v μ ⟼ x v Λ . (77)
Since V I ( μ ) V_I(\mu) V I ( μ ) is irreducible and ι I ( v μ ) = v Λ ≠ 0 \iota_I(v_\mu)=v_\Lambda\ne0 ι I ( v μ ) = v Λ = 0 , the map ι I \iota_I ι I is injective. It is therefore an isomorphism, proving (74) .
The face determined by I I I . Write γ = ∑ j q j ( γ ) α j \gamma=\sum_jq_j(\gamma)\alpha_j γ = ∑ j q j ( γ ) α j for each positive lowering root. All coefficients are nonnegative. Hence, for each j ∉ I j\notin I j ∈ / I ,
[ α j ] β ( u ) = ∑ γ , a u γ , a q j ( γ ) = 0 ⟺ u γ , a q j ( γ ) = 0 for every ( γ , a ) . [\alpha_j]\beta(u)=\sum_{\gamma,a}u_{\gamma,a}q_j(\gamma)=0\quad\Longleftrightarrow\quad u_{\gamma,a}q_j(\gamma)=0\text{ for every }(\gamma,a). [ α j ] β ( u ) = γ , a ∑ u γ , a q j ( γ ) = 0 ⟺ u γ , a q j ( γ ) = 0 for every ( γ , a ) .
Applying this equivalence for every j ∉ I j\notin I j ∈ / I gives
β ( u ) ∈ Q I + ⟺ u ∈ M I . (78) \beta(u)\in Q_I^+\quad\Longleftrightarrow\quad u\in M_I. \tag*{(78)} β ( u ) ∈ Q I + ⟺ u ∈ M I . (78)
Weights and essential exponents. For q = ( q j ) j ∈ I ∈ Z I q=(q_j)_{j\in I}\in\mathbb{Z}^I q = ( q j ) j ∈ I ∈ Z I , set γ = ∑ j ∈ I q j α j ∈ Q I \gamma=\sum_{j\in I}q_j\alpha_j\in Q_I γ = ∑ j ∈ I q j α j ∈ Q I and let A I = ( α j ( h i ) ) i , j ∈ I A_I=(\alpha_j(h_i))_{i,j\in I} A I = ( α j ( h i ) ) i , j ∈ I be the Cartan matrix. Define
res I : Q I ⟶ h I ∗ , γ ⟼ γ ∣ h I . (79) \operatorname{res}_I:Q_I\longrightarrow\mathfrak{h}_I^*,\qquad\gamma\longmapsto\gamma|_{\mathfrak{h}_I}. \tag*{(79)} res I : Q I ⟶ h I ∗ , γ ⟼ γ ∣ h I . (79)
Since h I = span { h i : i ∈ I } \mathfrak{h}_I=\operatorname{span}\{h_i:i\in I\} h I = span { h i : i ∈ I } ,
res I ( γ ) = 0 ⟺ ( γ ( h i ) ) i ∈ I = A I q = 0. \operatorname{res}_I(\gamma)=0\Longleftrightarrow(\gamma(h_i))_{i\in I}=A_Iq=0. res I ( γ ) = 0 ⟺ ( γ ( h i ) ) i ∈ I = A I q = 0.
For m = ∣ I ∣ m=|I| m = ∣ I ∣ , the determinant of A I A_I A I is m + 1 m+1 m + 1 in type A m A_m A m and 2 2 2 in type C m C_m C m . Thus A I A_I A I is non-singular and res I \operatorname{res}_I res I is injective. In particular,
res I ( β ) = res I ( β ′ ) ⟺ β = β ′ ( β , β ′ ∈ Q I ) . (80) \operatorname{res}_I(\beta)=\operatorname{res}_I(\beta')\Longleftrightarrow\beta=\beta'\qquad(\beta,\beta'\in Q_I). \tag*{(80)} res I ( β ) = res I ( β ′ ) ⟺ β = β ′ ( β , β ′ ∈ Q I ) . (80)
Fix u ∈ M I u\in\mathcal{M}_I u ∈ M I and put μ = Λ ∣ h I = ∑ i ∈ I k i ω i I \mu=\left.\Lambda\right|_{\mathfrak{h}_I}=\sum_{i\in I}k_i\omega_i^I μ = Λ ∣ h I = ∑ i ∈ I k i ω i I . The vectors X a v Λ X^{a}v_\Lambda X a v Λ and X a v μ X^{a}v_\mu X a v μ , for a ∈ M I a\in\mathcal{M}_I a ∈ M I , have weights
Λ − β ( a ) and μ − res I ( β ( a ) ) , \Lambda-\beta(a)\qquad\text{and}\qquad\mu-\operatorname{res}_I(\beta(a)), Λ − β ( a ) and μ − res I ( β ( a )) ,
respectively. To test essentiality, it suffices in either module to compare vectors of the same weight with strictly larger exponents. With the common order we obtain
u ∈ E Λ ⟺ X u v Λ ∉ span { X a v Λ : a ≻ u , β ( a ) = β ( u ) } ⟺ X u v Λ ∉ span { X a v Λ : a ∈ M I , a ≻ u , β ( a ) = β ( u ) } ⟺ X u v μ ∉ span { X a v μ : a ∈ M I , a ≻ u , β ( a ) = β ( u ) } ⟺ X u v μ ∉ span { X a v μ : a ∈ M I , a ≻ u , res I β ( a ) = res I β ( u ) } ⟺ u ∈ E I ( μ ) . \begin{aligned}
u\in E_\Lambda
&\Longleftrightarrow X^{u}v_\Lambda\notin\operatorname{span}\{X^{a}v_\Lambda:a\succ u,\ \beta(a)=\beta(u)\}\\
&\Longleftrightarrow X^{u}v_\Lambda\notin\operatorname{span}\{X^{a}v_\Lambda:a\in\mathcal{M}_I,\ a\succ u,\ \beta(a)=\beta(u)\}\\
&\Longleftrightarrow X^{u}v_\mu\notin\operatorname{span}\{X^{a}v_\mu:a\in\mathcal{M}_I,\ a\succ u,\ \beta(a)=\beta(u)\}\\
&\Longleftrightarrow X^{u}v_\mu\notin\operatorname{span}\{X^{a}v_\mu:a\in\mathcal{M}_I,\ a\succ u,\ \operatorname{res}_I\beta(a)=\operatorname{res}_I\beta(u)\}\\
&\Longleftrightarrow u\in E_I(\mu).
\end{aligned} u ∈ E Λ ⟺ X u v Λ ∈ / span { X a v Λ : a ≻ u , β ( a ) = β ( u )} ⟺ X u v Λ ∈ / span { X a v Λ : a ∈ M I , a ≻ u , β ( a ) = β ( u )} ⟺ X u v μ ∈ / span { X a v μ : a ∈ M I , a ≻ u , β ( a ) = β ( u )} ⟺ X u v μ ∈ / span { X a v μ : a ∈ M I , a ≻ u , res I β ( a ) = res I β ( u )} ⟺ u ∈ E I ( μ ) .
The first equivalence uses the affine weight space of weight Λ − β ( u ) \Lambda-\beta(u) Λ − β ( u ) . In the second, every a a a with β ( a ) = β ( u ) \beta(a)=\beta(u) β ( a ) = β ( u ) lies in M I \mathcal{M}_I M I by (78) . The third uses the isomorphism ι I \iota_I ι I , and the fourth uses (80) . The last equivalence is the definition of essentiality in the h I \mathfrak{h}_I h I -weight space of weight μ − res I ( β ( u ) ) \mu-\operatorname{res}_I(\beta(u)) μ − res I ( β ( u )) . We conclude that
E I ( μ ) = E Λ ∩ M I . E_I(\mu)=E_\Lambda\cap\mathcal{M}_I. E I ( μ ) = E Λ ∩ M I .
For j ∈ I j\in I j ∈ I , apply this equality with Λ = Λ j \Lambda=\Lambda_j Λ = Λ j . Then μ = ω j I \mu=\omega_j^I μ = ω j I and E Λ j = S e j E_{\Lambda_j}=S_{e_j} E Λ j = S e j , giving
E I ( ω j I ) = S e j ∩ M I . E_I(\omega_j^I)=S_{e_j}\cap\mathcal{M}_I. E I ( ω j I ) = S e j ∩ M I .
The same count by weight spaces as in (53) also gives
∣ E I ( μ ) ∣ = dim V I ( μ ) . (81) |E_I(\mu)|=\dim V_I(\mu). \tag*{(81)} ∣ E I ( μ ) ∣ = dim V I ( μ ) . (81)
Lattice points in the finite FFLV polytopes
The Feigin–Fourier–Littelmann–Vinberg (FFLV) polytopes are defined by path inequalities on directed graphs of positive roots [15 , 16 ] . These inequalities do not depend on a choice of order on PBW exponents. The connection between the symplectic Dyck paths and finite portions of the CMPP arrays was already observed in [10 Remark 5.4, p. 180] . We recall the inequalities explicitly, since we need to compare their paths with those of the affine grid.
Type A m A_m A m . Let β A , B = α A + ⋯ + α B \beta_{A,B}=\alpha_A+\cdots+\alpha_B β A , B = α A + ⋯ + α B , 1 ≤ A ≤ B ≤ m 1\le A\le B\le m 1 ≤ A ≤ B ≤ m . These are all the positive roots of A m A_m A m . A Dyck path starts at β i , i = α i \beta_{i,i}=\alpha_i β i , i = α i , ends at β j , j = α j \beta_{j,j}=\alpha_j β j , j = α j for some j ≥ i j\ge i j ≥ i , and at each step changes the indices by
( A , B ) ⟼ ( A , B + 1 ) or ( A + 1 , B ) , (A,B)\longmapsto(A,B+1)\qquad\text{or}\qquad(A+1,B), ( A , B ) ⟼ ( A , B + 1 ) or ( A + 1 , B ) ,
with the new pair still satisfying $1 \le A \le B \le m$. A single simple root is also a path. For a dominant integral weight μ = ∑ i = 1 m m i ω i \mu= \sum_{i=1}^{m} m_i\omega_i μ = ∑ i = 1 m m i ω i , with m i ∈ Z ≥ 0 m_i \in\mathbb{Z}_{\ge0} m i ∈ Z ≥ 0 , set
F A m ( μ ) = { x ∈ Z ≥ 0 Δ + : ∑ β ∈ D x β ≤ ∑ h = i j m h for every path D : α i ⇝ α j } . (82) \mathcal{F}_{A_m}(\mu)=\left\{x\in\mathbb{Z}_{\ge0}^{\Delta_{+}}:\sum_{\beta\in D}x_\beta\le\sum_{h=i}^{j}m_h\text{ for every path }D:\alpha_i\leadsto\alpha_j\right\}. \tag*{(82)} F A m ( μ ) = ⎩ ⎨ ⎧ x ∈ Z ≥ 0 Δ + : β ∈ D ∑ x β ≤ h = i ∑ j m h for every path D : α i ⇝ α j ⎭ ⎬ ⎫ . (82)
Taking a path consisting only of α i \alpha_i α i gives x α i ≤ m i x_{\alpha_i}\le m_i x α i ≤ m i .
Type C m C_m C m . Use weight coordinates ε 1 , … , ε m \varepsilon_1,\ldots,\varepsilon_m ε 1 , … , ε m , with simple roots α i = ε i − ε i + 1 \alpha_i=\varepsilon_i-\varepsilon_{i+1} α i = ε i − ε i + 1 for i < m i<m i < m and α m = 2 ε m \alpha_m=2\varepsilon_m α m = 2 ε m . Index all m 2 m^2 m 2 positive roots by
1 ≤ A ≤ m , A ≤ B ≤ 2 m − A , β A , B = { ε A − ε B + 1 , B < m , ε A + ε 2 m − B , B ≥ m . 1\le A\le m,\qquad A\le B\le2m-A,\qquad
\beta_{A,B}=
\begin{cases}
\varepsilon_A-\varepsilon_{B+1}, & B<m,\\
\varepsilon_A+\varepsilon_{2m-B}, & B\ge m.
\end{cases} 1 ≤ A ≤ m , A ≤ B ≤ 2 m − A , β A , B = { ε A − ε B + 1 , ε A + ε 2 m − B , B < m , B ≥ m .
Thus β A , A = α A \beta_{A,A}=\alpha_A β A , A = α A , including A = m A=m A = m , and β A , 2 m − A = 2 ε A \beta_{A,2m-A}=2\varepsilon_A β A , 2 m − A = 2 ε A is a long root. Here column B B B is position B B B in the ordered alphabet 1 , … , m , m − 1 ‾ , … , 1 ‾ 1,\ldots,m,\overline{m-1},\ldots,\overline{1} 1 , … , m , m − 1 , … , 1 of [16 Definition 1.2] . A symplectic Dyck path starts at β i , i \beta_{i,i} β i , i , follows the same two steps
( A , B ) ⟼ ( A , B + 1 ) or ( A + 1 , B ) (A,B)\longmapsto(A,B+1)\quad\text{or}\quad(A+1,B) ( A , B ) ⟼ ( A , B + 1 ) or ( A + 1 , B )
within this domain, and ends either
at β j , j = α j \beta_{j,j}=\alpha_j β j , j = α j for i ≤ j ≤ m i\le j\le m i ≤ j ≤ m , or
at β j , 2 m − j = 2 ε j \beta_{j,2m-j}=2\varepsilon_j β j , 2 m − j = 2 ε j for i ≤ j ≤ m i\le j\le m i ≤ j ≤ m .
Again a single simple root is allowed. The two endpoint conditions agree at α m \alpha_m α m , which is both simple and long. For a dominant integral weight μ = ∑ i = 1 m m i ω i \mu=\sum_{i=1}^{m}m_i\omega_i μ = ∑ i = 1 m m i ω i , let F C m ( μ ) \mathcal{F}_{C_m}(\mu) F C m ( μ ) consist of the nonnegative integer vectors satisfying
∑ β ∈ D x β ≤ m i + ⋯ + m j , D : α i ⇝ α j , (83) \sum_{\beta\in D}x_\beta\le m_i+\cdots+m_j,\qquad D:\alpha_i\leadsto\alpha_j, \tag*{(83)} β ∈ D ∑ x β ≤ m i + ⋯ + m j , D : α i ⇝ α j , (83)
∑ β ∈ D x β ≤ m i + ⋯ + m m , D : α i ⇝ 2 ε j . (84) \sum_{\beta\in D}x_\beta\le m_i+\cdots+m_m,\qquad D:\alpha_i\leadsto2\varepsilon_j. \tag*{(84)} β ∈ D ∑ x β ≤ m i + ⋯ + m m , D : α i ⇝ 2 ε j . (84)
To place these coordinates on I = [ s , t ] I=[s,t] I = [ s , t ] , put m = ∣ I ∣ m=|I| m = ∣ I ∣ . Write T = A m T=A_m T = A m if 0 < s ≤ t < n 0<s\le t<n 0 < s ≤ t < n , and T = C m T=C_m T = C m if s = 0 s=0 s = 0 or t = n t=n t = n . Denote its standard simple roots, fundamental weights and set of positive roots by α ˉ a \bar{\alpha}_a α ˉ a , ω ˉ a \bar{\omega}_a ω ˉ a and Δ ˉ + \bar{\Delta}_+ Δ ˉ + , respectively. Define
σ I ( a ) = { s + a − 1 , I = [ s , t ] , s > 0 , t + 1 − a , I = [ 0 , t ] , 1 ≤ a ≤ m . \sigma_I(a)=
\begin{cases}
s+a-1, & I=[s,t],\ s>0,\\
t+1-a, & I=[0,t],
\end{cases}
\qquad1\le a\le m. σ I ( a ) = { s + a − 1 , t + 1 − a , I = [ s , t ] , s > 0 , I = [ 0 , t ] , 1 ≤ a ≤ m .
The first case preserves the order of the nodes. The second reverses it, so that the long node of the standard C m C_m C m diagram maps to node 0.
Extending α ˉ a ↦ α σ I ( a ) \bar{\alpha}_a\mapsto\alpha_{\sigma_I(a)} α ˉ a ↦ α σ I ( a ) linearly gives a map of root lattices
ϕ I : ∑ a = 1 m Z α ˉ a ⟶ Q I . \phi_I:\sum_{a=1}^{m}\mathbb{Z}\bar{\alpha}_a\longrightarrow Q_I. ϕ I : a = 1 ∑ m Z α ˉ a ⟶ Q I .
The numbering respects the Cartan matrix, so ϕ I ( Δ ˉ + ) = Δ I + \phi_I(\bar{\Delta}_+)=\Delta_I^+ ϕ I ( Δ ˉ + ) = Δ I + . For μ = ∑ i ∈ I k i ω i I \mu=\sum_{i\in I}k_i\omega_i^I μ = ∑ i ∈ I k i ω i I , put μ ˉ = ∑ a = 1 m k σ I ( a ) ω ˉ a I \bar{\mu}=\sum_{a=1}^{m}k_{\sigma_I(a)}\bar{\omega}_a^I μ ˉ = ∑ a = 1 m k σ I ( a ) ω ˉ a I and define
F I ( μ ) = { x ∈ Z ≥ 0 Δ I + : ( x ϕ I ( β ˉ ) ) β ˉ ∈ Δ ˉ + ∈ F T ( μ ˉ ) } . \mathcal{F}_I(\mu)=\left\{x\in\mathbb{Z}_{\ge0}^{\Delta_I^+}:(x_{\phi_I(\bar{\beta})})_{\bar{\beta}\in\bar{\Delta}_+}\in\mathcal{F}_T(\bar{\mu})\right\}. F I ( μ ) = { x ∈ Z ≥ 0 Δ I + : ( x ϕ I ( β ˉ ) ) β ˉ ∈ Δ ˉ + ∈ F T ( μ ˉ ) } .
Thus we impose each path inequality on its image under φ I \varphi_I φ I , replacing m a m_a m a on the right by k σ I ( a ) k_{\sigma_I(a)} k σ I ( a ) .
For example, when n = 3 n=3 n = 3 and I = [ 0 , 1 ] I=[0,1] I = [ 0 , 1 ] , the local type is C 2 C_2 C 2 and σ I ( 1 ) = 1 \sigma_I(1)=1 σ I ( 1 ) = 1 , σ I ( 2 ) = 0 \sigma_I(2)=0 σ I ( 2 ) = 0 . The standard path
α ˉ 1 ⟶ α ˉ 1 + α ˉ 2 ⟶ α ˉ 2 \bar{\alpha}_1 \longrightarrow\bar{\alpha}_1+\bar{\alpha}_2 \longrightarrow\bar{\alpha}_2 α ˉ 1 ⟶ α ˉ 1 + α ˉ 2 ⟶ α ˉ 2
therefore becomes
α 1 ⟶ α 1 + α 0 ⟶ α 0 , \alpha_1 \longrightarrow\alpha_1+\alpha_0 \longrightarrow\alpha_0, α 1 ⟶ α 1 + α 0 ⟶ α 0 ,
and its inequality is
x α 1 + x α 1 + α 0 + x α 0 ≤ k 1 + k 0 . x_{\alpha_1}+x_{\alpha_1+\alpha_0}+x_{\alpha_0}\leq k_1+k_0. x α 1 + x α 1 + α 0 + x α 0 ≤ k 1 + k 0 .
Theorem 5.2 (Feigin–Fourier–Littelmann). Let F ( μ ) \mathcal{F}(\mu) F ( μ ) denote either F A m ( μ ) \mathcal{F}_{A_m}(\mu) F A m ( μ ) or F C m ( μ ) \mathcal{F}_{C_m}(\mu) F C m ( μ ) , where μ = ∑ i = 1 m m i ω i \mu=\sum_{i=1}^{m}m_i\omega_i μ = ∑ i = 1 m m i ω i is dominant integral.
The FFLV set has cardinality
∣ F ( μ ) ∣ = dim V ( μ ) . |\mathcal{F}(\mu)|=\dim V(\mu). ∣ F ( μ ) ∣ = dim V ( μ ) . Every lattice point is a sum of lattice points for the fundamental weights, in the same positive root coordinates:
F ( ∑ i = 1 m m i ω i ) = F ( ω 1 ) + ⋯ + F ( ω 1 ) + ⋯ + F ( ω m ) + ⋯ + F ( ω m ) . (85) \mathcal{F}\left(\sum_{i=1}^{m}m_i\omega_i\right)=\mathcal{F}(\omega_1)+\cdots+\mathcal{F}(\omega_1)+\cdots+\mathcal{F}(\omega_m)+\cdots+\mathcal{F}(\omega_m). \tag*{(85)} F ( i = 1 ∑ m m i ω i ) = F ( ω 1 ) + ⋯ + F ( ω 1 ) + ⋯ + F ( ω m ) + ⋯ + F ( ω m ) . (85) Here a sum with no summands is { 0 } \{0\} { 0 } , and vectors are added coordinatewise.
Proof. In type A m A_m A m , the cardinality formula is [15 Corollary 4.1] . In type C m C_m C m , it follows from the basis in [16 Theorem 1.3] , since passing to the associated graded module preserves the dimension.
For the second assertion, [15 Proposition 4.2] gives, in type A m A_m A m ,
F ( λ + ν ) = F ( λ ) + F ( ν ) \mathcal{F}(\lambda+\nu)=\mathcal{F}(\lambda)+\mathcal{F}(\nu) F ( λ + ν ) = F ( λ ) + F ( ν )
and iteration gives (5.17). In type C m C_m C m , write μ = ∑ i m i ω i \mu=\sum_i m_i\omega_i μ = ∑ i m i ω i . If μ ≠ 0 \mu\ne0 μ = 0 , let i i i be the least index with m i > 0 m_i>0 m i > 0 . For every x ∈ F ( μ ) x\in\mathcal{F}(\mu) x ∈ F ( μ ) , [16 Lemma 3.5] supplies an exponent vector y y y with entries in { 0 , 1 } \{0,1\} { 0 , 1 } such that
y ∈ F ( ω i ) , x − y ∈ F ( μ − ω i ) . y\in\mathcal{F}(\omega_i),\qquad x-y\in\mathcal{F}(\mu-\omega_i). y ∈ F ( ω i ) , x − y ∈ F ( μ − ω i ) .
Induction on ∑ i m i \sum_i m_i ∑ i m i , with F ( 0 ) = { 0 } \mathcal{F}(0)=\{0\} F ( 0 ) = { 0 } , gives
F ( μ ) ⊆ ∑ i m i F ( ω i ) . \mathcal{F}(\mu)\subseteq\sum_i m_i\mathcal{F}(\omega_i). F ( μ ) ⊆ i ∑ m i F ( ω i ) .
Conversely, let
x = ∑ i = 1 m ∑ t = 1 m i z i , t , z i , t ∈ F ( ω i ) . x=\sum_{i=1}^{m}\sum_{t=1}^{m_i}z_{i,t},\qquad z_{i,t}\in\mathcal{F}(\omega_i). x = i = 1 ∑ m t = 1 ∑ m i z i , t , z i , t ∈ F ( ω i ) .
For each defining symplectic Dyck path D D D , let b D ( ν ) b_D(\nu) b D ( ν ) denote the bound in its inequality at weight ν \nu ν . Since this bound is additive in ν \nu ν ,
∑ β ∈ D x β = ∑ i = 1 m ∑ t = 1 m i ∑ β ∈ D ( z i , t ) β ≤ ∑ i = 1 m m i b D ( ω i ) = b D ( μ ) . \sum_{\beta\in D}x_\beta=\sum_{i=1}^{m}\sum_{t=1}^{m_i}\sum_{\beta\in D}(z_{i,t})_\beta\leq\sum_{i=1}^{m}m_i b_D(\omega_i)=b_D(\mu). β ∈ D ∑ x β = i = 1 ∑ m t = 1 ∑ m i β ∈ D ∑ ( z i , t ) β ≤ i = 1 ∑ m m i b D ( ω i ) = b D ( μ ) .
Thus x ∈ F ( μ ) x\in\mathcal{F}(\mu) x ∈ F ( μ ) , proving the reverse inclusion in type C m C_m C m . ∎
Finite roots in the affine grid
Proposition 5.3 (Coordinates for a finite root subsystem). For a proper interval I = [ s , t ] I=[s,t] I = [ s , t ] , the positive roots supported on I I I correspond to the following set of grid vertices:
T I = { ( r , p ) ∈ Ω + : p + r ≤ N − 2 s + 1 if s > 0 , p − r ≤ 2 t − N + 1 if t < n } . (86) T_I=\left\{(r,p)\in\Omega_+:\begin{array}{ll}p+r\le N-2s+1&\text{if }s>0,\\ p-r\le2t-N+1&\text{if }t<n\end{array}\right\}. \tag*{(86)} T I = { ( r , p ) ∈ Ω + : p + r ≤ N − 2 s + 1 p − r ≤ 2 t − N + 1 if s > 0 , if t < n } . (86)
Proof. We use A , B A,B A , B for the indices of finite positive roots and a , b a,b a , b for the PS colour indices. A positive lowering root indexes a generator in the corresponding negative root space. At loop index zero, Proposition 3.1 gives
X a b ( 0 ) ⟼ ( r , p ) = ( b − a , a + b − N − 1 ) , N = 2 n . (87) X_{ab}(0)\longmapsto(r,p)=(b-a,a+b-N-1),\qquad N=2n. \tag*{(87)} X ab ( 0 ) ⟼ ( r , p ) = ( b − a , a + b − N − 1 ) , N = 2 n . (87)
First suppose I = [ s , t ] I=[s,t] I = [ s , t ] with 0 < s ≤ t < n 0<s\le t<n 0 < s ≤ t < n . Its positive roots are
β A , B = α A + ⋯ + α B = ε A − ε B + 1 , s ≤ A ≤ B ≤ t . \beta_{A,B}=\alpha_A+\cdots+\alpha_B=\varepsilon_A-\varepsilon_{B+1},\qquad s\le A\le B\le t. β A , B = α A + ⋯ + α B = ε A − ε B + 1 , s ≤ A ≤ B ≤ t .
By (18) , the root vector of weight − β A , B = ε B + 1 − ε A -\beta_{A,B}=\varepsilon_{B+1}-\varepsilon_A − β A , B = ε B + 1 − ε A has PS colour indices
a = B + 1 , b = N + 1 − A . a=B+1,\qquad b=N+1-A. a = B + 1 , b = N + 1 − A .
Substitution in (87) gives
β A , B ⟼ ( r , p ) = ( N − A − B , B − A + 1 ) . (88) \beta_{A,B}\longmapsto(r,p)=(N-A-B,B-A+1). \tag*{(88)} β A , B ⟼ ( r , p ) = ( N − A − B , B − A + 1 ) . (88)
In particular,
p + r = N − 2 A + 1 , p − r = 2 B − N + 1. p+r=N-2A+1,\qquad p-r=2B-N+1. p + r = N − 2 A + 1 , p − r = 2 B − N + 1.
Thus A ≥ s A\ge s A ≥ s and B ≤ t B\le t B ≤ t imply the two bounds in (86) . Conversely, if a positive grid vertex satisfies those bounds, set
A = N − r − p + 1 2 , B = N − r + p − 1 2 . A=\frac{N-r-p+1}{2},\qquad B=\frac{N-r+p-1}{2}. A = 2 N − r − p + 1 , B = 2 N − r + p − 1 .
Grid parity makes A , B A,B A , B integers. The condition p ≥ 1 p\ge1 p ≥ 1 gives A ≤ B A\le B A ≤ B , while the two bounds give A ≥ s A\ge s A ≥ s and B ≤ t B\le t B ≤ t .
Next suppose I = [ s , n ] I=[s,n] I = [ s , n ] , where s > 0 s>0 s > 0 , and put m = n − s + 1 m=n-s+1 m = n − s + 1 . Use the C m C_m C m local roots β A , B \beta_{A,B} β A , B , indexed by 1 ≤ A ≤ m 1\le A\le m 1 ≤ A ≤ m and A ≤ B ≤ 2 m − A A\le B\le2m-A A ≤ B ≤ 2 m − A .
If B < m B<m B < m , the corresponding positive root of C n C_n C n is
β A , B = ε s + A − 1 − ε s + B . \beta_{A,B}=\varepsilon_{s+A-1}-\varepsilon_{s+B}. β A , B = ε s + A − 1 − ε s + B .
A generator of the root space of weight − β A , B -\beta_{A,B} − β A , B has colour indices
a = s + B , b = N − s − A + 2. a=s+B,\qquad b=N-s-A+2. a = s + B , b = N − s − A + 2.
Equation (87) therefore gives
r = 2 m − A − B , p = B − A + 1. r=2m-A-B,\qquad p=B-A+1. r = 2 m − A − B , p = B − A + 1.
If B ≥ m B\ge m B ≥ m , set D = 2 m − B D=2m-B D = 2 m − B , so A ≤ D ≤ m A\le D\le m A ≤ D ≤ m . Now
β A , B = ε s + A − 1 + ε s + D − 1 . \beta_{A,B}=\varepsilon_{s+A-1}+\varepsilon_{s+D-1}. β A , B = ε s + A − 1 + ε s + D − 1 .
Its negative-root generator has colour indices
a = N − s − D + 2 , b = N − s − A + 2. a = N - s - D + 2,\qquad b = N - s - A + 2. a = N − s − D + 2 , b = N − s − A + 2.
Hence
r = b − a = D − A = 2 m − A − B , p = a + b − N − 1 = B − A + 1. r = b - a = D - A = 2m - A - B,\qquad p = a + b - N - 1 = B - A + 1. r = b − a = D − A = 2 m − A − B , p = a + b − N − 1 = B − A + 1.
Thus both kinds of C m C_m C m roots satisfy the single formula
β A , B ⟼ ( r , p ) = ( 2 m − A − B , B − A + 1 ) . (89) \beta_{A,B} \longmapsto(r,p) = (2m - A - B,\; B - A + 1). \tag*{(89)} β A , B ⟼ ( r , p ) = ( 2 m − A − B , B − A + 1 ) . (89)
It follows that
p + r = 2 m − 2 A + 1 ≤ 2 m − 1 = N − 2 s + 1 , p + r = 2m - 2A + 1 \le2m - 1 = N - 2s + 1, p + r = 2 m − 2 A + 1 ≤ 2 m − 1 = N − 2 s + 1 ,
the bound in (86) for I = [ s , n ] I = [s,n] I = [ s , n ] . Conversely, a positive grid vertex satisfying this bound determines
A = 2 m − r − p + 1 2 , B = 2 m − r + p − 1 2 . A = \frac{2m-r-p+1}{2},\qquad B = \frac{2m-r+p-1}{2}. A = 2 2 m − r − p + 1 , B = 2 2 m − r + p − 1 .
These are integers by grid parity. The bound gives A ≥ 1 A \ge1 A ≥ 1 , the condition p ≥ 1 p \ge1 p ≥ 1 gives B ≥ A B \ge A B ≥ A , and r ≥ 0 r \ge0 r ≥ 0 gives B ≤ 2 m − A B \le2m-A B ≤ 2 m − A . The last two inequalities also imply A ≤ m A \le m A ≤ m . Hence ( A , B ) (A,B) ( A , B ) is a unique local C m C_m C m root index.
It remains to consider I = [ 0 , t ] I = [0,t] I = [ 0 , t ] , t < n t < n t < n . On the real span of the affine roots define
σ ( δ ) = δ , σ ( ε i ) = δ 2 − ε n + 1 − i . \sigma(\delta) = \delta,\qquad\sigma(\varepsilon_i) = \frac{\delta}{2} - \varepsilon_{n+1-i}. σ ( δ ) = δ , σ ( ε i ) = 2 δ − ε n + 1 − i .
Then σ ( α i ) = α n − i \sigma(\alpha_i) = \alpha_{n-i} σ ( α i ) = α n − i for 0 ≤ i ≤ n 0 \le i \le n 0 ≤ i ≤ n . Thus σ \sigma σ exchanges I I I with [ n − t , n ] [n-t,n] [ n − t , n ] . On the relevant real roots it corresponds to the grid reflection ( r , p ) ⟼ ( N − r , p ) (r,p) \longmapsto(N-r,p) ( r , p ) ⟼ ( N − r , p ) . Indeed,
σ ( ε i − ε j ) = ε n + 1 − j − ε n + 1 − i , σ ( ε i + ε j ) = δ − ε n + 1 − i − ε n + 1 − j . \sigma(\varepsilon_i-\varepsilon_j) = \varepsilon_{n+1-j} - \varepsilon_{n+1-i},\qquad
\sigma(\varepsilon_i+\varepsilon_j) = \delta- \varepsilon_{n+1-i} - \varepsilon_{n+1-j}. σ ( ε i − ε j ) = ε n + 1 − j − ε n + 1 − i , σ ( ε i + ε j ) = δ − ε n + 1 − i − ε n + 1 − j .
Substitution in the grid dictionary gives p p p unchanged and r r r replaced by N − r N-r N − r in both cases. Reflecting the bound for [ n − t , n ] [n-t,n] [ n − t , n ] now gives
p + ( N − r ) ≤ N − 2 ( n − t ) + 1 ⟺ p − r ≤ 2 t − N + 1 , p + (N-r) \le N - 2(n-t) + 1
\quad\Longleftrightarrow\quad
p-r \le2t-N+1, p + ( N − r ) ≤ N − 2 ( n − t ) + 1 ⟺ p − r ≤ 2 t − N + 1 ,
which is precisely (86) for I = [ 0 , t ] I = [0,t] I = [ 0 , t ] . This proves the claimed correspondence for every proper connected interval I I I .
For s > 0 s > 0 s > 0 , formulas (88) and (89) send the steps ( A , B ) ⟼ ( A , B + 1 ) (A,B) \longmapsto(A,B+1) ( A , B ) ⟼ ( A , B + 1 ) and ( A , B ) ⟼ ( A + 1 , B ) (A,B) \longmapsto(A+1,B) ( A , B ) ⟼ ( A + 1 , B ) , respectively, to
( r , p ) ⟼ ( r − 1 , p + 1 ) or ( r − 1 , p − 1 ) . (90) (r,p) \longmapsto(r-1,p+1)\quad\text{or}\quad(r-1,p-1). \tag*{(90)} ( r , p ) ⟼ ( r − 1 , p + 1 ) or ( r − 1 , p − 1 ) . (90)
For I = [ 0 , t ] I = [0,t] I = [ 0 , t ] the row coordinate is reflected, so the corresponding steps are ( r , p ) ⟼ ( r + 1 , p + 1 ) (r,p) \longmapsto(r+1,p+1) ( r , p ) ⟼ ( r + 1 , p + 1 ) and ( r , p ) ⟼ ( r + 1 , p − 1 ) (r,p) \longmapsto(r+1,p-1) ( r , p ) ⟼ ( r + 1 , p − 1 ) .
Definition 5.4 (Path completion). Let D D D be a finite Dyck or symplectic Dyck path for a proper interval I I I . Read its grid image in increasing row order as
( a , p a ) , ( a + 1 , p a + 1 ) , … , ( b , p b ) . (a,p_a),(a+1,p_{a+1}),\ldots,(b,p_b). ( a , p a ) , ( a + 1 , p a + 1 ) , … , ( b , p b ) . The completion of D D D is the path P D = ( ( r , p ~ r ) ) r = 0 N P_D=((r,\widetilde p_r))_{r=0}^{N} P D = (( r , p r ) ) r = 0 N defined by
p ~ r = { p r , a ≤ r ≤ b , − 1 , r ∉ [ a , b ] and r is even , 0 , r ∉ [ a , b ] and r is odd . (91) \widetilde p_r =
\begin{cases}
p_r, & a \le r \le b,\\
-1, & r \notin[a,b]\text{ and }r\text{ is even},\\
0, & r \notin[a,b]\text{ and }r\text{ is odd}.
\end{cases} \tag*{(91)} p r = ⎩ ⎨ ⎧ p r , − 1 , 0 , a ≤ r ≤ b , r ∈ / [ a , b ] and r is even , r ∈ / [ a , b ] and r is odd . (91) Thus outside D D D the path follows the boundary zigzag.
A simple-root endpoint lies in an even row and has height 1; a long-root endpoint lies on an outer row and requires no extension on that side. Hence every join in (91) is a step between heights 1 and 0. Thus P D P_D P D is a full path, also when D D D consists of one root.
Set
A D = { h ∈ { 0 , … , n } : a ≤ N − 2 h ≤ b } . A_D=\{h\in\{0,\ldots,n\}:a\le N-2h\le b\}. A D = { h ∈ { 0 , … , n } : a ≤ N − 2 h ≤ b } .
If D D D starts at local simple root α ˉ i \bar{\alpha}_i α ˉ i , its initial row is N − 2 σ I ( i ) N-2\sigma_I(i) N − 2 σ I ( i ) . Its terminal row is N − 2 σ I ( j ) N-2\sigma_I(j) N − 2 σ I ( j ) for a simple-root endpoint α ˉ j \bar{\alpha}_j α ˉ j , and N − 2 σ I ( m ) N-2\sigma_I(m) N − 2 σ I ( m ) for a long-root endpoint, where m = ∣ I ∣ m=|I| m = ∣ I ∣ . Since each step changes the row by one in the same direction, it follows that
A D = { { σ I ( h ) : i ≤ h ≤ j } , for an endpoint α ˉ j , { σ I ( h ) : i ≤ h ≤ m } , for a long endpoint in type C m . (92) A_D=
\begin{cases}
\{\sigma_I(h):i\le h\le j\},&\text{for an endpoint }\bar{\alpha}_j,\\
\{\sigma_I(h):i\le h\le m\},&\text{for a long endpoint in type }C_m.
\end{cases} \tag*{(92)} A D = { { σ I ( h ) : i ≤ h ≤ j } , { σ I ( h ) : i ≤ h ≤ m } , for an endpoint α ˉ j , for a long endpoint in type C m . (92)
The formulas agree when the endpoint is α ˉ m \bar{\alpha}_m α ˉ m . In either case the finite-path bound is ∑ h ∈ A D k h \sum_{h\in A_D} k_h ∑ h ∈ A D k h . By (91) ,
B h ∈ P D ⟺ N − 2 h ∉ [ a , b ] ⟺ h ∉ A D . B_h\in P_D\Longleftrightarrow N-2h\notin[a,b]\Longleftrightarrow h\notin A_D. B h ∈ P D ⟺ N − 2 h ∈ / [ a , b ] ⟺ h ∈ / A D .
Figure 6 shows the completion of
D : α 1 ⟶ α 1 + α 2 ⟶ 2 α 1 + α 2 . D:\quad\alpha_1\longrightarrow\alpha_1+\alpha_2\longrightarrow2\alpha_1+\alpha_2. D : α 1 ⟶ α 1 + α 2 ⟶ 2 α 1 + α 2 .
Figure 6. Path completion for n = 2 n=2 n = 2 and I = [ 1 , 2 ] I=[1,2] I = [ 1 , 2 ] . The finite C 2 C_2 C 2 path occupies ( 0 , 3 ) , ( 1 , 2 ) , ( 2 , 1 ) (0,3),(1,2),(2,1) ( 0 , 3 ) , ( 1 , 2 ) , ( 2 , 1 ) . The completion adds ( 3 , 0 ) (3,0) ( 3 , 0 ) and B 0 = ( 4 , − 1 ) B_0=(4,-1) B 0 = ( 4 , − 1 ) . Thus A D = { 1 , 2 } A_D=\{1,2\} A D = { 1 , 2 } and the completed path visits precisely the boundary vertex with index outside A D A_D A D .
Transfer to the restricted PS order
The path completion allows us to compare the finite FFLV inequalities with admissibility in the affine grid. We first make this comparison for a fundamental weight, then use (85) .
For an exponent x x x supported on I I I , set
β ( x ) = ∑ γ ∈ Δ I + x γ γ ∈ Q I + , β I ( x ) : = β ( x ) ∣ h I ∈ h I ∗ . \beta(x)=\sum_{\gamma\in\Delta_I^+}x_\gamma\gamma\in Q_I^+,\qquad\beta_I(x):=\beta(x)|_{\mathfrak h_I}\in\mathfrak h_I^*. β ( x ) = γ ∈ Δ I + ∑ x γ γ ∈ Q I + , β I ( x ) := β ( x ) ∣ h I ∈ h I ∗ .
using the restriction map in (79) . Thus X x v μ X^xv_\mu X x v μ has h I \mathfrak h_I h I -weight μ − β I ( x ) \mu-\beta_I(x) μ − β I ( x ) .
Theorem 5.5 (Essential exponents for the restricted PS order). For every nonempty proper connected interval I I I and every dominant integral weight μ = ∑ j ∈ I k j ω j I \mu= \sum_{j\in I} k_j\omega_j^I μ = ∑ j ∈ I k j ω j I
F I ( μ ) = E I ( μ ) (93) \mathcal{F}_I(\mu)=\mathcal{E}_I(\mu) \tag*{(93)} F I ( μ ) = E I ( μ ) (93) for the restriction of the original PS order. Consequently, if an exponent u u u supported on I I I does not belong to F I ( μ ) \mathcal{F}_I(\mu) F I ( μ ) , there is a finite relation
X u v μ = ∑ z ≻ u c z X z v μ . (94) X^u v_\mu= \sum_{z\succ u} c_z X^z v_\mu. \tag*{(94)} X u v μ = z ≻ u ∑ c z X z v μ . (94) where every z z z is supported on I I I , β ( z ) = β ( u ) \beta(z)=\beta(u) β ( z ) = β ( u ) and ℓ ( z ) ≤ ℓ ( u ) \ell(z)\leq\ell(u) ℓ ( z ) ≤ ℓ ( u ) .
Proof. Fix I I I and first take μ = ω j I \mu=\omega_j^I μ = ω j I . The finite PBW exponents lie in M I \mathcal{M}_I M I , and both sets in (93) use the same root coordinates. Take f ∈ S e j ∩ M I f\in S_{e_j}\cap\mathcal{M}_I f ∈ S e j ∩ M I . For any finite path D D D , the completed path of Definition 5.4 has sum of positive frequencies ∑ x ∈ D f ( x ) \sum_{x\in D}f(x) ∑ x ∈ D f ( x ) and boundary contribution 1 j ∉ A D \mathbf{1}_{j\notin A_D} 1 j ∈ / A D . Level-one admissibility therefore gives
∑ x ∈ D f ( x ) + 1 j ∉ A D ≤ 1 , hence ∑ x ∈ D f ( x ) ≤ 1 j ∈ A D . \sum_{x\in D} f(x)+\mathbf{1}_{j\notin A_D}\leq1,\qquad\text{hence}\qquad\sum_{x\in D} f(x)\leq\mathbf{1}_{j\in A_D}. x ∈ D ∑ f ( x ) + 1 j ∈ / A D ≤ 1 , hence x ∈ D ∑ f ( x ) ≤ 1 j ∈ A D .
By (92) , this is precisely the FFLV inequality for D D D at ω j I \omega_j^I ω j I . Thus (75) gives
E I ( ω j I ) = S e j ∩ M I ⊆ F I ( ω j I ) . \mathcal{E}_I(\omega_j^I)=S_{e_j}\cap\mathcal{M}_I\subseteq\mathcal{F}_I(\omega_j^I). E I ( ω j I ) = S e j ∩ M I ⊆ F I ( ω j I ) .
The cardinalities satisfy
∣ E I ( ω j I ) ∣ = dim V I ( ω j I ) = ∣ F I ( ω j I ) ∣ \lvert\mathcal{E}_I(\omega_j^I)\rvert=\dim V_I(\omega_j^I)=\lvert\mathcal{F}_I(\omega_j^I)\rvert ∣ E I ( ω j I )∣ = dim V I ( ω j I ) = ∣ F I ( ω j I )∣
by (81) and Theorem 5.2 . Hence the inclusion is an equality.
For a general dominant integral weight μ = ∑ j ∈ I k j ω j I \mu=\sum_{j\in I}k_j\omega_j^I μ = ∑ j ∈ I k j ω j I , Theorem 5.2 gives
F I ( μ ) = ∑ j ∈ I k j F I ( ω j I ) = ∑ j ∈ I k j E I ( ω j I ) , \mathcal{F}_I(\mu)=\sum_{j\in I}k_j\mathcal{F}_I(\omega_j^I)=\sum_{j\in I}k_j\mathcal{E}_I(\omega_j^I), F I ( μ ) = j ∈ I ∑ k j F I ( ω j I ) = j ∈ I ∑ k j E I ( ω j I ) ,
by the equality just proved for each fundamental weight. The set M I \mathcal{M}_I M I contains zero and is closed under addition. We may therefore apply Proposition 5.1 to each summand and use essential addition (Theorem 4.7 ) in the ambient affine modules:
F I ( μ ) = ∑ j ∈ I k j ( E Λ j ∩ M I ) ⊆ ( ∑ j ∈ I k j E Λ j ) ∩ M I ⊆ E ∑ j ∈ I k j Λ j ∩ M I = E I ( μ ) . \mathcal{F}_I(\mu)=\sum_{j\in I}k_j(\mathcal{E}_{\Lambda_j}\cap\mathcal{M}_I)
\subseteq\left(\sum_{j\in I}k_j\mathcal{E}_{\Lambda_j}\right)\cap\mathcal{M}_I
\subseteq\mathcal{E}_{\sum_{j\in I}k_j\Lambda_j}\cap\mathcal{M}_I
=\mathcal{E}_I(\mu). F I ( μ ) = j ∈ I ∑ k j ( E Λ j ∩ M I ) ⊆ j ∈ I ∑ k j E Λ j ∩ M I ⊆ E ∑ j ∈ I k j Λ j ∩ M I = E I ( μ ) .
Both sets have cardinality dim V I ( μ ) \dim V_I(\mu) dim V I ( μ ) by (81) and Theorem 5.2 . The inclusion is therefore an equality, proving (93) .
If u ∉ F I ( μ ) = E I ( μ ) u\notin\mathcal{F}_I(\mu)=\mathcal{E}_I(\mu) u ∈ / F I ( μ ) = E I ( μ ) , the definition of essentiality gives a finite relation
X u v μ = ∑ z ∈ M I z ≻ u c z X z v μ . X^u v_\mu=\sum_{\substack{z\in\mathcal{M}_I\\z\succ u}}c_zX^zv_\mu. X u v μ = z ∈ M I z ≻ u ∑ c z X z v μ .
Projecting the relation onto the weight μ − β I ( u ) \mu-\beta_I(u) μ − β I ( u ) therefore removes every term for which β I ( z ) ≠ β I ( u ) \beta_I(z)\neq\beta_I(u) β I ( z ) = β I ( u ) . For each remaining term,
β I ( z ) = β I ( u ) ⟹ β ( z ) = β ( u ) , \beta_I(z)=\beta_I(u)\quad\Longrightarrow\quad\beta(z)=\beta(u), β I ( z ) = β I ( u ) ⟹ β ( z ) = β ( u ) ,
because β ( z ) , β ( u ) ∈ Q I \beta(z),\beta(u)\in Q_I β ( z ) , β ( u ) ∈ Q I and the restriction Q I → h I ∗ Q_I\to\mathfrak{h}_I^* Q I → h I ∗ is injective by (80) . Finally, z ≻ u z\succ u z ≻ u implies ℓ ( z ) ≤ ℓ ( u ) \ell(z)\leq\ell(u) ℓ ( z ) ≤ ℓ ( u ) by (69) . This proves all assertions concerning (94) .
Example 5.6. For n = 2 n=2 n = 2 and I = [ 1 , 2 ] I=[1,2] I = [ 1 , 2 ] , denote the roots α 1 \alpha_1 α 1 , α 1 + α 2 \alpha_1+\alpha_2 α 1 + α 2 , 2 α 1 + α 2 2\alpha_1+\alpha_2 2 α 1 + α 2 , α 2 \alpha_2 α 2 by a , b , c , d a,b,c,d a , b , c , d , respectively. They occur at ( 2 , 1 ) , ( 1 , 2 ) , ( 0 , 3 ) , ( 0 , 1 ) (2,1),(1,2),(0,3),(0,1) ( 2 , 1 ) , ( 1 , 2 ) , ( 0 , 3 ) , ( 0 , 1 ) , as shown in Figure 7 . Their path inequalities are
Figure 7. The positive roots and Dyck paths of C 2 C_2 C 2 in the affine grid.
x a ≤ m 1 , x d ≤ m 2 , x a + x b + x c ≤ m 1 + m 2 , x a + x b + x d ≤ m 1 + m 2 . x_a \le m_1,\qquad x_d \le m_2,\qquad x_a+x_b+x_c \le m_1+m_2,\qquad x_a+x_b+x_d \le m_1+m_2. x a ≤ m 1 , x d ≤ m 2 , x a + x b + x c ≤ m 1 + m 2 , x a + x b + x d ≤ m 1 + m 2 .
Thus
F ( ω 1 ) = { 0 , e a , e b , e c } , F ( ω 2 ) = { 0 , e b , e c , e d , e c + e d } . \mathcal{F}(\omega_1)=\{0,e_a,e_b,e_c\},\qquad\mathcal{F}(\omega_2)=\{0,e_b,e_c,e_d,e_c+e_d\}. F ( ω 1 ) = { 0 , e a , e b , e c } , F ( ω 2 ) = { 0 , e b , e c , e d , e c + e d } .
For μ = ω 1 + ω 2 \mu=\omega_1+\omega_2 μ = ω 1 + ω 2 , the possibilities are
Table 3.
There are 6 + 4 + 1 + 4 + 1 = 16 6+4+1+4+1=16 6 + 4 + 1 + 4 + 1 = 16 lattice points. Hence Theorem 5.2 gives dim V C 2 ( ω 1 + ω 2 ) = 16 \dim V_{C_2}(\omega_1+\omega_2)=16 dim V C 2 ( ω 1 + ω 2 ) = 16 .
Localizing a violated path inequality
We can now identify the finite path inequality for each maximal positive subpath whose interval I B I_B I B is proper. Its type and orientation depend on which outer rows the subpath meets.
Lemma 5.7 (Types of maximal positive subpaths). Let B B B be a maximal positive subpath of P P P occupying rows [ a , b ] [a,b] [ a , b ] .
If 0 < a ≤ b < N 0<a\le b<N 0 < a ≤ b < N , its vertices lie in T I B T_{I_B} T I B and, read in reverse row order, form an A ∣ I B ∣ A_{|I_B|} A ∣ I B ∣ Dyck path whose capacity is c ( B ) c(B) c ( B ) .
If a = 0 a=0 a = 0 and b < N b<N b < N , reading B B B in decreasing row order gives a symplectic Dyck path for I B = [ n − b / 2 , n ] I_B=[n-b/2,n] I B = [ n − b /2 , n ] with bound c ( B ) c(B) c ( B ) .
If a > 0 a>0 a > 0 and b = N b=N b = N , reading B B B in increasing row order gives a symplectic Dyck path for I B = [ 0 , n − a / 2 ] I_B=[0,n-a/2] I B = [ 0 , n − a /2 ] with the same bound c ( B ) c(B) c ( B ) .
If a = 0 a=0 a = 0 and b = N b=N b = N , then B = P ⊆ Ω + B=P\subseteq\Omega_+ B = P ⊆ Ω + and I B = { 0 , … , n } I_B=\{0,\ldots,n\} I B = { 0 , … , n } . Lemma 3.6 applies when m ( B ) > c ( B ) m(B)>c(B) m ( B ) > c ( B ) and k > 0 k>0 k > 0 .
Singleton subpaths are included in the first three cases.
Proof. In the first case both endpoints have height 1. By the step condition (6) , for each a ≤ r ≤ b a \le r \le b a ≤ r ≤ b we have
p r − 1 ≤ r − a , p r − 1 ≤ b − r . p_r - 1 \le r-a,\qquad p_r - 1 \le b-r. p r − 1 ≤ r − a , p r − 1 ≤ b − r .
If s = n − b / 2 s=n-b/2 s = n − b /2 , t = n − a / 2 t=n-a/2 t = n − a /2 , these are exactly
p r − r ≤ 1 − a = 2 t − N + 1 , p r + r ≤ b + 1 = N − 2 s + 1. p_r-r \le1-a=2t-N+1,\qquad p_r+r \le b+1=N-2s+1. p r − r ≤ 1 − a = 2 t − N + 1 , p r + r ≤ b + 1 = N − 2 s + 1.
Thus every vertex lies in T [ s , t ] T_{[s,t]} T [ s , t ] . The inverse coordinates in the proof of Proposition 5.3 identify each step in decreasing row order with one of the steps in (90) . By (88) , ( b , 1 ) (b,1) ( b , 1 ) corresponds to α s \alpha_s α s and ( a , 1 ) (a,1) ( a , 1 ) to α t \alpha_t α t . The resulting path has bound k s + ⋯ + k t = c ( B ) k_s+\cdots+k_t=c(B) k s + ⋯ + k t = c ( B ) .
In the second case, p b = 1 p_b=1 p b = 1 and p r − 1 ≤ b − r p_r-1\le b-r p r − 1 ≤ b − r , so every vertex satisfies the inequality in (86) for I B = [ s , n ] I_B=[s,n] I B = [ s , n ] , where s = n − b / 2 s=n-b/2 s = n − b /2 . Put m = ∣ I B ∣ = b / 2 + 1 m=|I_B|=b/2+1 m = ∣ I B ∣ = b /2 + 1 . In reverse row order, the path starts at ( b , 1 ) = ( 2 m − 2 , 1 ) (b,1)=(2m-2,1) ( b , 1 ) = ( 2 m − 2 , 1 ) , the first local simple root, and ends at ( 0 , p 0 ) (0,p_0) ( 0 , p 0 ) . The inverse coordinates of the latter vertex are
A = 2 m − p 0 + 1 2 , B = 2 m − A , β A , B = 2 ε s + A − 1 A=\frac{2m-p_0+1}{2},\qquad B=2m-A,\qquad\beta_{A,B}=2\varepsilon_{s+A-1} A = 2 2 m − p 0 + 1 , B = 2 m − A , β A , B = 2 ε s + A − 1
in the ambient root system. The bounds 1 ≤ p 0 ≤ b + 1 = 2 m − 1 1\le p_0\le b+1=2m-1 1 ≤ p 0 ≤ b + 1 = 2 m − 1 , together with grid parity, give A ∈ { 1 , … , m } A\in\{1,\ldots,m\} A ∈ { 1 , … , m } . Thus the path ends at a long root. Since it starts at the first local simple root, (84) gives the bound c ( B ) c(B) c ( B ) .
In the third case p a = 1 p_a=1 p a = 1 and p r − r ≤ 1 − a p_r-r\le1-a p r − r ≤ 1 − a for a ≤ r ≤ N a\le r\le N a ≤ r ≤ N . Reflection in the rows, r ↦ N − r r\mapsto N-r r ↦ N − r , now reduces the assertion to the second case and reverses the reading direction.
In the fourth case, B = P B=P B = P is contained in Ω + \Omega_+ Ω + , so c ( B ) = k c(B)=k c ( B ) = k and the path meets no boundary mark. If m ( B ) > c ( B ) m(B)>c(B) m ( B ) > c ( B ) , it contains k + 1 k+1 k + 1 positive occurrences, to which Lemma 3.6 applies. □
Example 5.8. Take n = 3 n=3 n = 3 and k = ( 1 , 1 , 0 , 1 ) \boldsymbol{k}=(1,1,0,1) k = ( 1 , 1 , 0 , 1 ) . Consider the path with heights
( p 0 , … , p 6 ) = ( 1 , 0 , 1 , 2 , 1 , 0 , − 1 ) . (p_0,\ldots,p_6)=(1,0,1,2,1,0,-1). ( p 0 , … , p 6 ) = ( 1 , 0 , 1 , 2 , 1 , 0 , − 1 ) .
Put f ( 0 , 1 ) = 1 f(0,1)=1 f ( 0 , 1 ) = 1 , f ( 3 , 2 ) = 2 f(3,2)=2 f ( 3 , 2 ) = 2 , and set the other positive frequencies to zero. Then
∑ v ∈ P ∩ Ω + f ( v ) = 1 + 2 = 3 = k , ∑ v ∈ P F ( v ) = 3 + k 0 = 4 > k . \sum_{v\in P\cap\Omega_+}f(v)=1+2=3=k,\qquad\sum_{v\in P}F(v)=3+k_0=4>k. v ∈ P ∩ Ω + ∑ f ( v ) = 1 + 2 = 3 = k , v ∈ P ∑ F ( v ) = 3 + k 0 = 4 > k .
The maximal positive subpaths occupy rows [ 0 , 0 ] [0,0] [ 0 , 0 ] and [ 2 , 4 ] [2,4] [ 2 , 4 ] . Their index sets are { 3 } \{3\} { 3 } and [ 1 , 2 ] [1,2] [ 1 , 2 ] , respectively. The first has total frequency and capacity 1. The second has total frequency 2 but capacity 1 (Figure 8 ). Read in decreasing row order, the second is the A 2 A_2 A 2 path through α 1 , α 1 + α 2 , α 2 \alpha_1,\alpha_1+\alpha_2,\alpha_2 α 1 , α 1 + α 2 , α 2 . The relation (29) gives
Figure 8. A mixed boundary for n = 3 n=3 n = 3 and ( k 0 , k 1 , k 2 , k 3 ) = ( 1 , 1 , 0 , 1 ) (k_0,k_1,k_2,k_3)=(1,1,0,1) ( k 0 , k 1 , k 2 , k 3 ) = ( 1 , 1 , 0 , 1 ) .
⟨ Λ , ( α 1 + α 2 ) ∨ ⟩ = k 1 + k 2 = 1 ⟹ X − ( α 1 + α 2 ) 2 v Λ = 0. \langle\Lambda,(\alpha_1+\alpha_2)^\vee\rangle=k_1+k_2=1\quad\Longrightarrow\quad X_{-(\alpha_1+\alpha_2)}^2v_\Lambda=0. ⟨ Λ , ( α 1 + α 2 ) ∨ ⟩ = k 1 + k 2 = 1 ⟹ X − ( α 1 + α 2 ) 2 v Λ = 0.
Proof of Theorem 2
Suppose k > 0 k>0 k > 0 and fix f ∉ S k f\notin S_k f ∈ / S k . By (71) , some maximal positive subpath B B B satisfies m ( B ) > c ( B ) m(B)>c(B) m ( B ) > c ( B ) . If it meets both outer rows, then c ( B ) = k c(B)=k c ( B ) = k . Choose k + 1 k+1 k + 1 parts on B B B , counted with multiplicity, and denote their exponent by u ≤ f u\le f u ≤ f . Lemma 3.6 gives (72) .
Otherwise I B I_B I B is proper. Define
u ( r , p ) = { f ( r , p ) , ( r , p ) ∈ B , 0 , ( r , p ) ∉ B , μ B = ∑ i ∈ I B k i ω i I B . (95) u(r,p)=
\begin{cases}
f(r,p), & (r,p)\in B,\\
0, & (r,p)\notin B,
\end{cases}
\qquad
\mu_B=\sum_{i\in I_B}k_i\omega_i^{I_B}. \tag*{(95)} u ( r , p ) = { f ( r , p ) , 0 , ( r , p ) ∈ B , ( r , p ) ∈ / B , μ B = i ∈ I B ∑ k i ω i I B . (95)
By Lemma 5.7 ,
u ∈ M I B , ∑ ( r , p ) ∈ B u ( r , p ) = m ( B ) > c ( B ) , u ∉ F I B ( μ B ) . u\in\mathcal{M}_{I_B},\qquad\sum_{(r,p)\in B}u(r,p)=m(B)>c(B),\qquad u\notin\mathcal{F}_{I_B}(\mu_B). u ∈ M I B , ( r , p ) ∈ B ∑ u ( r , p ) = m ( B ) > c ( B ) , u ∈ / F I B ( μ B ) .
Transporting the relation of Theorem 5.5 through the isomorphism (5.6) gives
X u v Λ = ∑ z ≻ u c z X z v Λ . X^u v_\Lambda=\sum_{z\succ u}c_zX^zv_\Lambda. X u v Λ = z ≻ u ∑ c z X z v Λ .
In both cases the sum is finite, u ≤ f u\leq f u ≤ f coordinatewise, and
β ( z ) = β ( u ) , ℓ ( z ) ≤ ℓ ( u ) . \beta(z)=\beta(u),\qquad\ell(z)\leq\ell(u). β ( z ) = β ( u ) , ℓ ( z ) ≤ ℓ ( u ) .
The next lemma extends the relation for u u u to the original exponent f f f .
Lemma 5.9. Let u ≤ f u\leq f u ≤ f coordinatewise and suppose X u v Λ X^u v_\Lambda X u v Λ is a finite linear combination of X z v Λ X^zv_\Lambda X z v Λ with z ≻ u z\succ u z ≻ u and β ( z ) = β ( u ) \beta(z)=\beta(u) β ( z ) = β ( u ) . Then there is a finite relation
X f v Λ = ∑ h ≻ f β ( h ) = β ( f ) d h X h v Λ . (96) X^f v_\Lambda=\sum_{\substack{h\succ f\\\beta(h)=\beta(f)}}d_hX^hv_\Lambda. \tag*{(96)} X f v Λ = h ≻ f β ( h ) = β ( f ) ∑ d h X h v Λ . (96)
Proof. Put U = U ( n ^ − ) U=U(\widehat{\mathfrak n}_{-}) U = U ( n − ) and w = f − u w=f-u w = f − u . Multiply the given relation on the left by X w X^w X w . By Lemma 3.2, reordering the products gives
X w X u = X f + R L , R L ∈ F ℓ ( f ) − 1 U , X^wX^u=X^f+R_L,\qquad R_L\in F_{\ell(f)-1}U, X w X u = X f + R L , R L ∈ F ℓ ( f ) − 1 U ,
X w X z = X w + z + R z , R z ∈ F ℓ ( w ) + ℓ ( z ) − 1 U ⊆ F ℓ ( f ) − 1 U . X^wX^z=X^{w+z}+R_z,\qquad R_z\in F_{\ell(w)+\ell(z)-1}U\subseteq F_{\ell(f)-1}U. X w X z = X w + z + R z , R z ∈ F ℓ ( w ) + ℓ ( z ) − 1 U ⊆ F ℓ ( f ) − 1 U .
Here each commutator replaces two factors by one and ℓ ( z ) ≤ ℓ ( u ) \ell(z)\leq\ell(u) ℓ ( z ) ≤ ℓ ( u ) . The order properties imply
w + z ≻ w + u = f , ℓ ( h ) < ℓ ( f ) ⟹ h ≻ f . w+z\succ w+u=f,\qquad\ell(h)<\ell(f)\Longrightarrow h\succ f. w + z ≻ w + u = f , ℓ ( h ) < ℓ ( f ) ⟹ h ≻ f .
Thus
X f v Λ = ∑ z ≻ u c z X w + z v Λ + ∑ z ≻ u c z R z v Λ − R L v Λ X^fv_\Lambda=\sum_{z\succ u}c_zX^{w+z}v_\Lambda+\sum_{z\succ u}c_zR_zv_\Lambda-R_Lv_\Lambda X f v Λ = z ≻ u ∑ c z X w + z v Λ + z ≻ u ∑ c z R z v Λ − R L v Λ
has only exponents strictly larger than f f f after PBW reordering. To check the weights, let x , y ∈ n ^ − x,y\in\widehat{\mathfrak n}_{-} x , y ∈ n − have lowering roots γ , η \gamma,\eta γ , η . For every H ∈ h ^ H\in\widehat{\mathfrak h} H ∈ h , the Jacobi identity gives
[ H , [ x , y ] ] = [ [ H , x ] , y ] + [ x , [ H , y ] ] = − ( γ + η ) ( H ) [ x , y ] . [H,[x,y]]=[[H,x],y]+[x,[H,y]]=-(\gamma+\eta)(H)[x,y]. [ H , [ x , y ]] = [[ H , x ] , y ] + [ x , [ H , y ]] = − ( γ + η ) ( H ) [ x , y ] .
Thus each nonzero commutator term has the same lowering weight as the product x y xy x y in the enveloping algebra. Reordering each finite product takes only finitely many steps, so the resulting sum is finite and homogeneous. □
Proof of Theorem 2 . Suppose first that k > 0 k>0 k > 0 . By (96) , every exponent f ∉ S k f\notin S_k f ∈ / S k has a finite relation
X f v Λ = ∑ z ≻ f β ( z ) = β ( f ) d z X z v Λ . X^f v_\Lambda=\sum_{\substack{z\succ f\\ \beta(z)=\beta(f)}}d_zX^zv_\Lambda. X f v Λ = z ≻ f β ( z ) = β ( f ) ∑ d z X z v Λ .
For a fixed principal degree d d d , set
M d = { u : ∣ u ∣ p r = d } . \mathcal{M}_d=\{u:|u|_{\mathrm{pr}}=d\}. M d = { u : ∣ u ∣ pr = d } .
If u ∈ M d u\in\mathcal{M}_d u ∈ M d , then
u ( r , p ) = 0 ( p > d ) , 0 ≤ u ( r , p ) ≤ ⌊ d p ⌋ ( 1 ≤ p ≤ d ) . u(r,p)=0\quad(p>d),\qquad0\leq u(r,p)\leq\left\lfloor\frac{d}{p}\right\rfloor\quad(1\leq p\leq d). u ( r , p ) = 0 ( p > d ) , 0 ≤ u ( r , p ) ≤ ⌊ p d ⌋ ( 1 ≤ p ≤ d ) .
There are finitely many rows and finitely many vertices with 1 ≤ p ≤ d 1\leq p\leq d 1 ≤ p ≤ d . Hence M d \mathcal{M}_d M d is finite. Fix a lowering weight β ∈ Q + \beta\in Q_+ β ∈ Q + and put d = ht ( β ) d=\operatorname{ht}(\beta) d = ht ( β ) . The set
M β = { u : β ( u ) = β } ⊆ M d \mathcal{M}_\beta=\{u:\beta(u)=\beta\}\subseteq\mathcal{M}_d M β = { u : β ( u ) = β } ⊆ M d
is finite. List its elements in decreasing comparison order as u 1 ≻ ⋯ ≻ u m u_1\succ\cdots\succ u_m u 1 ≻ ⋯ ≻ u m , and define
W 0 = 0 , W j = W j − 1 + C X u j v Λ ( 1 ≤ j ≤ m ) . W_0=0,\qquad W_j=W_{j-1}+\mathbb{C}X^{u_j}v_\Lambda\quad(1\leq j\leq m). W 0 = 0 , W j = W j − 1 + C X u j v Λ ( 1 ≤ j ≤ m ) .
If u j u_j u j is inadmissible, the straightening relation involves only exponents among u 1 , … , u j − 1 u_1,\ldots,u_{j-1} u 1 , … , u j − 1 . Hence X u j v Λ ∈ W j − 1 X^{u_j}v_\Lambda\in W_{j-1} X u j v Λ ∈ W j − 1 and W j = W j − 1 W_j=W_{j-1} W j = W j − 1 . If u j u_j u j is admissible, we adjoin its vector to the previous span. Induction on j j j now gives
W j = span { X u t v Λ : 1 ≤ t ≤ j , u t ∈ S k } . W_j=\operatorname{span}\{X^{u_t}v_\Lambda:1\leq t\leq j,\ u_t\in S_k\}. W j = span { X u t v Λ : 1 ≤ t ≤ j , u t ∈ S k } .
In particular, for an inadmissible f = u j f=u_j f = u j , the equality X u j v Λ ∈ W j − 1 X^{u_j}v_\Lambda\in W_{j-1} X u j v Λ ∈ W j − 1 expresses its vector as a finite linear combination of admissible vectors with exponents u t ≻ f u_t\succ f u t ≻ f and lowering weight β ( f ) \beta(f) β ( f ) . PBW spanning gives W m = L ( Λ ) Λ − β W_m=L(\Lambda)_{\Lambda-\beta} W m = L ( Λ ) Λ − β . Since β \beta β was arbitrary, the admissible monomial vectors span L ( Λ ) L(\Lambda) L ( Λ ) .
Their linear independence follows from S k ⊆ E Λ S_k\subseteq E_\Lambda S k ⊆ E Λ in (68) . Conversely, (96) shows that every inadmissible exponent is nonessential. Hence E Λ = S k E_\Lambda=S_k E Λ = S k .
If k = 0 k=0 k = 0 , every positive vertex lies on a full path and all path sums are at most zero. Thus S 0 = { 0 } S_0=\{0\} S 0 = { 0 } and L ( 0 ) = C v 0 L(0)=\mathbb{C}v_0 L ( 0 ) = C v 0 . Every nonempty monomial annihilates v 0 v_0 v 0 , so the zero exponent is the only essential exponent and { v 0 } \{v_0\} { v 0 } is a basis of L ( 0 ) L(0) L ( 0 ) .
From the PBW basis to the partition identity
Theorem 2 gives a basis of weight vectors for n ≥ 2 n\geq2 n ≥ 2 . At rank one and positive level we use the basis of Meurman and Primc in Theorem 4.4 , while at level zero the module is one dimensional. Counting these basis vectors by weight and applying principal specialization gives the partition series Z k ( 1 , q ) Z_k(1,q) Z k ( 1 , q ) . To recover Z k ( z , q ) Z_k(z,q) Z k ( z , q ) , we also keep track of their PBW degrees. Throughout this section normalized characters are formal series in
Z [ [ e − α 0 , … , e − α n ] ] . \mathbb{Z}\left[\left[e^{-\alpha_0},\ldots,e^{-\alpha_n}\right]\right]. Z [ [ e − α 0 , … , e − α n ] ] .
The character and its principal specialization
For a positive vertex v = ( r , p ) v=(r,p) v = ( r , p ) , let γ v = ∑ i = 0 n c i ( v ) α i \gamma_v=\sum_{i=0}^{n}c_i(v)\alpha_i γ v = ∑ i = 0 n c i ( v ) α i be the lowering root of its PBW generator. For a finitely supported array f f f , write
β ( f ) = ∑ v ∈ Ω + f ( v ) γ v = ∑ i = 0 n b i ( f ) α i . \beta(f)=\sum_{v\in\Omega_+}f(v)\gamma_v=\sum_{i=0}^{n}b_i(f)\alpha_i. β ( f ) = v ∈ Ω + ∑ f ( v ) γ v = i = 0 ∑ n b i ( f ) α i .
Thus X f v Λ X^fv_\Lambda X f v Λ has weight Λ − β ( f ) \Lambda-\beta(f) Λ − β ( f ) . Counting the weight vectors in the appropriate basis gives
dim L ( Λ ) Λ − β = # { f ∈ S k : β ( f ) = β } ( β ∈ Q + ) . \dim L(\Lambda)_{\Lambda-\beta}=\#\{f\in S_k:\beta(f)=\beta\}\qquad(\beta\in Q_+). dim L ( Λ ) Λ − β = # { f ∈ S k : β ( f ) = β } ( β ∈ Q + ) .
Here e ν e^\nu e ν denotes a formal weight symbol satisfying e ν e ν ′ = e ν + ν ′ e^\nu e^{\nu'}=e^{\nu+\nu'} e ν e ν ′ = e ν + ν ′ . Expanding the character by weights and then using the preceding multiplicity formula gives
e − Λ ch L ( Λ ) = ∑ β ∈ Q + dim L ( Λ ) Λ − β e − β = ∑ β ∈ Q + ∑ f ∈ S k β ( f ) = β e − β = ∑ f ∈ S k e − β ( f ) = ∑ f ∈ S k ∏ i = 0 n ( e − α i ) b i ( f ) . (97) \begin{aligned}
e^{-\Lambda}\operatorname{ch}L(\Lambda)=\sum_{\beta\in Q_+}\dim L(\Lambda)_{\Lambda-\beta}e^{-\beta}=\sum_{\beta\in Q_+}\sum_{\substack{f\in S_k\\\beta(f)=\beta}}e^{-\beta} \\
&=\sum_{f\in S_k}e^{-\beta(f)}=\sum_{f\in S_k}\prod_{i=0}^{n}(e^{-\alpha_i})^{b_i(f)}. \tag*{(97)}
\end{aligned} e − Λ ch L ( Λ ) = β ∈ Q + ∑ dim L ( Λ ) Λ − β e − β = β ∈ Q + ∑ f ∈ S k β ( f ) = β ∑ e − β = f ∈ S k ∑ e − β ( f ) = f ∈ S k ∑ i = 0 ∏ n ( e − α i ) b i ( f ) . (97)
By (32) , ht ( γ ( r , p ) ) = p \operatorname{ht}(\gamma_{(r,p)})=p ht ( γ ( r , p ) ) = p . Hence
ht β ( f ) = ∑ i = 0 n ∑ v f ( v ) c i ( v ) = ∑ v f ( v ) ht ( γ v ) = ∑ ( r , p ) ∈ Ω + p f ( r , p ) = ∣ f ∣ p r . \begin{aligned}
\operatorname{ht}\beta(f)=\sum_{i=0}^{n}\sum_v f(v)c_i(v)=\sum_v f(v)\operatorname{ht}(\gamma_v) \\
&=\sum_{(r,p)\in\Omega_+}p f(r,p)=|f|_{\mathrm{pr}}.
\end{aligned} ht β ( f ) = i = 0 ∑ n v ∑ f ( v ) c i ( v ) = v ∑ f ( v ) ht ( γ v ) = ( r , p ) ∈ Ω + ∑ p f ( r , p ) = ∣ f ∣ pr .
Applying the principal specialization e − α i ↦ q e^{-\alpha_i}\mapsto q e − α i ↦ q to (97) gives
ps ( e − Λ ch L ( Λ ) ) = ∑ f ∈ S k q ∑ i = 0 n b i ( f ) = ∑ f ∈ S k q ∑ ( r , p ) ∈ Ω + p f ( r , p ) = Z k ( 1 , q ) . (98) \operatorname{ps}(e^{-\Lambda}\operatorname{ch}L(\Lambda))=\sum_{f\in S_k}q^{\sum_{i=0}^{n}b_i(f)}=\sum_{f\in S_k}q^{\sum_{(r,p)\in\Omega_+}pf(r,p)}=Z_k(1,q). \tag*{(98)} ps ( e − Λ ch L ( Λ )) = f ∈ S k ∑ q ∑ i = 0 n b i ( f ) = f ∈ S k ∑ q ∑ ( r , p ) ∈ Ω + p f ( r , p ) = Z k ( 1 , q ) . (98)
Proof of Theorem 3. For n ≥ 2 n\geq2 n ≥ 2 , the principal specialization of the normalized character is given by Lepowsky’s product formula [22 ] , see [9 Theorem 7.4] and the explicit theta-product form [21 (2.4)] . With λ i = ∑ j = i n k j \lambda_i=\sum_{j=i}^{n}k_j λ i = ∑ j = i n k j , κ = k + n + 1 \kappa=k+n+1 κ = k + n + 1 , and a i = λ i + n − i + 1 a_i=\lambda_i+n-i+1 a i = λ i + n − i + 1 , the formula is exactly the product on the right-hand side of (1.11). Combining it with (6.2) proves (1.11) for n ≥ 2 n\geq2 n ≥ 2 . At level zero both sides equal one:
S 0 = { 0 } , L ( 0 ) = C v 0 , Z 0 ( 1 , q ) = ps ( ch L ( 0 ) ) = 1. S_0=\{0\},\qquad L(0)=\mathbb{C}v_0,\qquad Z_0(1,q)=\operatorname{ps}(\operatorname{ch}L(0))=1. S 0 = { 0 } , L ( 0 ) = C v 0 , Z 0 ( 1 , q ) = ps ( ch L ( 0 )) = 1.
For rank one, put
s 0 = k 0 + 1 , s 1 = k 1 + 1 , κ = s 0 + s 1 = k + 2. s_0=k_0+1,\qquad s_1=k_1+1,\qquad\kappa=s_0+s_1=k+2. s 0 = k 0 + 1 , s 1 = k 1 + 1 , κ = s 0 + s 1 = k + 2.
Since the Cartan matrix A = ( 2 − 2 − 2 2 ) A=\begin{pmatrix}2&-2\\-2&2\end{pmatrix} A = ( 2 − 2 − 2 2 ) of A 1 ( 1 ) A_1^{(1)} A 1 ( 1 ) satisfies A T = A A^{\mathsf T}=A A T = A , Lepowsky’s formula uses the A 1 ( 1 ) A_1^{(1)} A 1 ( 1 ) root product in both the numerator and denominator, with specialization parameters ( k 0 + 1 , k 1 + 1 ) (k_0+1,k_1+1) ( k 0 + 1 , k 1 + 1 ) and ( 1 , 1 ) (1,1) ( 1 , 1 ) , respectively. The positive roots are
α 0 + m δ , α 1 + m δ ( m ≥ 0 ) , m δ ( m ≥ 1 ) , δ = α 0 + α 1 . \alpha_0+m\delta,\qquad\alpha_1+m\delta\quad(m\geq0),\qquad m\delta\quad(m\geq1),\qquad\delta=\alpha_0+\alpha_1. α 0 + m δ , α 1 + m δ ( m ≥ 0 ) , m δ ( m ≥ 1 ) , δ = α 0 + α 1 .
The real-root spaces are one-dimensional. The imaginary root space for m δ m\delta m δ is spanned by the Cartan loop mode h ⊗ t m h\otimes t^m h ⊗ t m , so it is also one-dimensional.
In the numerator specialization e − α i ↦ q s i e^{-\alpha_i} \mapsto q^{s_i} e − α i ↦ q s i , one has deg δ = s 0 + s 1 = κ \deg\delta= s_0+s_1=\kappa deg δ = s 0 + s 1 = κ . Then
deg ( α 0 + m δ ) = s 0 + m κ , deg ( α 1 + m δ ) = s 1 + m κ , deg ( m δ ) = m κ . \deg(\alpha_0+m\delta)=s_0+m\kappa,\qquad\deg(\alpha_1+m\delta)=s_1+m\kappa,\qquad\deg(m\delta)=m\kappa. deg ( α 0 + m δ ) = s 0 + mκ , deg ( α 1 + m δ ) = s 1 + mκ , deg ( m δ ) = mκ .
Since every root listed above has multiplicity one, the definition of the specialized root product gives
Q A 1 ( 1 ) ( k 0 + 1 , k 1 + 1 ; q ) = ∏ m ≥ 0 ( 1 − q s 0 + m κ ) ∏ m ≥ 0 ( 1 − q s 1 + m κ ) ∏ m ≥ 1 ( 1 − q m κ ) = ( q k 0 + 1 , q k 1 + 1 , q κ ; q κ ) ∞ . \begin{aligned}
Q_{A_1^{(1)}}(k_0+1,k_1+1;q)=\prod_{m\geq0}(1-q^{s_0+m\kappa})\prod_{m\geq0}(1-q^{s_1+m\kappa})\prod_{m\geq1}(1-q^{m\kappa}) \\
&=(q^{k_0+1},q^{k_1+1},q^\kappa;q^\kappa)_\infty.
\end{aligned} Q A 1 ( 1 ) ( k 0 + 1 , k 1 + 1 ; q ) = m ≥ 0 ∏ ( 1 − q s 0 + mκ ) m ≥ 0 ∏ ( 1 − q s 1 + mκ ) m ≥ 1 ∏ ( 1 − q mκ ) = ( q k 0 + 1 , q k 1 + 1 , q κ ; q κ ) ∞ .
For the principal denominator, set s 0 = s 1 = 1 s_0=s_1=1 s 0 = s 1 = 1 in the same calculation,
Q A 1 ( 1 ) ( 1 , 1 ; q ) = ( q ; q 2 ) ∞ 2 ( q 2 ; q 2 ) ∞ = ( q ; q 2 ) ∞ ( q ; q ) ∞ . Q_{A_1^{(1)}}(1,1;q)=(q;q^2)_\infty^2(q^2;q^2)_\infty=(q;q^2)_\infty(q;q)_\infty. Q A 1 ( 1 ) ( 1 , 1 ; q ) = ( q ; q 2 ) ∞ 2 ( q 2 ; q 2 ) ∞ = ( q ; q 2 ) ∞ ( q ; q ) ∞ .
Finally, at n = 1 n=1 n = 1 the parameters in (1.9) are κ = k + 2 \kappa=k+2 κ = k + 2 and a 1 = k 1 + 1 a_1=k_1+1 a 1 = k 1 + 1 . As
( q κ ; q κ ) ∞ = ( q κ ; q 2 κ ) ∞ ( q 2 κ ; q 2 κ ) ∞ , θ ( q a 1 ; q κ ) = ( q k 1 + 1 , q k 0 + 1 ; q κ ) ∞ , (q^\kappa;q^\kappa)_\infty=(q^\kappa;q^{2\kappa})_\infty(q^{2\kappa};q^{2\kappa})_\infty,\qquad\theta(q^{a_1};q^\kappa)=(q^{k_1+1},q^{k_0+1};q^\kappa)_\infty, ( q κ ; q κ ) ∞ = ( q κ ; q 2 κ ) ∞ ( q 2 κ ; q 2 κ ) ∞ , θ ( q a 1 ; q κ ) = ( q k 1 + 1 , q k 0 + 1 ; q κ ) ∞ ,
the quotient of the two root products is precisely (1.11) at n = 1 n=1 n = 1 .
The refinement by PBW degree
Recall the cyclic PBW filtration [13 , 15 , 16 ] :
F m L ( Λ ) = U ≤ m ( n ^ − ) v Λ , F − 1 L ( Λ ) = 0. F_mL(\Lambda)=U_{\leq m}(\widehat{\mathfrak n}_{-})v_\Lambda,\qquad F_{-1}L(\Lambda)=0. F m L ( Λ ) = U ≤ m ( n − ) v Λ , F − 1 L ( Λ ) = 0.
Proof of Corollary 4. PBW expresses F m L ( Λ ) F_mL(\Lambda) F m L ( Λ ) as the span of X u v Λ X^u v_\Lambda X u v Λ with ℓ ( u ) ≤ m \ell(u)\leq m ℓ ( u ) ≤ m . For n ≥ 2 n\geq2 n ≥ 2 , Theorem 2 expresses each such vector in terms of admissible vectors with exponents f ⪰ u f\succeq u f ⪰ u . In rank one at positive level, Theorem 4.4 identifies the admissible exponents with the essential ones. By their definition, a nonessential exponent can be replaced by strictly larger ones. Projecting to its weight space and iterating terminates, since there are finitely many exponents of the given principal degree. At level zero only the empty monomial is needed. The first rule of the PS order implies ℓ ( f ) ≤ ℓ ( u ) \ell(f)\leq\ell(u) ℓ ( f ) ≤ ℓ ( u ) , so
F m L ( Λ ) = span { X f v Λ : f ∈ S κ , ℓ ( f ) ≤ m } . (99) F_mL(\Lambda)=\operatorname{span}\{X^f v_\Lambda:f\in S_\kappa,\ \ell(f)\leq m\}. \tag*{(99)} F m L ( Λ ) = span { X f v Λ : f ∈ S κ , ℓ ( f ) ≤ m } . (99)
These vectors belong to the basis in the respective rank and level, hence they form a basis. Passing from m − 1 m-1 m − 1 to m m m shows that
{ X f v Λ + F m − 1 L ( Λ ) : f ∈ S κ , ℓ ( f ) = m } \{X^f v_\Lambda+F_{m-1}L(\Lambda):f\in S_\kappa,\ \ell(f)=m\} { X f v Λ + F m − 1 L ( Λ ) : f ∈ S κ , ℓ ( f ) = m }
is a basis of F m / F m − 1 F_m/F_{m-1} F m / F m − 1 . Each principal symbol is nonzero, with lowering weight β ( f ) \beta(f) β ( f ) and principal degree ∣ f ∣ p r |f|_{\mathrm{pr}} ∣ f ∣ pr . In particular,
dim ( F m / F m − 1 ) d = # { f ∈ S κ : ℓ ( f ) = m , ∣ f ∣ p r = d } . \dim(F_m/F_{m-1})_d=\#\{f\in S_\kappa:\ell(f)=m,\ |f|_{\mathrm{pr}}=d\}. dim ( F m / F m − 1 ) d = # { f ∈ S κ : ℓ ( f ) = m , ∣ f ∣ pr = d } .
Multiplying by z m q d z^m q^d z m q d and summing proves (12) .
By (99) , an admissible monomial of length m m m lies in F m ∖ F m − 1 F_m\setminus F_{m-1} F m ∖ F m − 1 . An inadmissible monomial of length m m m may reduce to a linear combination of shorter monomials and hence have zero symbol in F m / F m − 1 F_m/F_{m-1} F m / F m − 1 .
PBW degree is not determined by affine weight. In C 2 ( 1 ) C_2^{(1)} C 2 ( 1 ) , take Λ = Λ 1 + Λ 2 \Lambda=\Lambda_1+\Lambda_2 Λ = Λ 1 + Λ 2 . The arrays supported on { ( 1 , 2 ) } \{(1,2)\} {( 1 , 2 )} and on { ( 0 , 1 ) , ( 2 , 1 ) } \{(0,1),(2,1)\} {( 0 , 1 ) , ( 2 , 1 )} , with frequency one at each listed vertex, are both admissible. The corresponding ordered monomials are
X 34 ( 0 ) v Λ , X 24 ( 0 ) X 33 ( 0 ) v Λ . X_{34}(0)v_\Lambda,\qquad X_{24}(0)X_{33}(0)v_\Lambda. X 34 ( 0 ) v Λ , X 24 ( 0 ) X 33 ( 0 ) v Λ .
Indeed, in the second array each of rows 0 and 2 contributes at most one, including the boundary entries. In the first array the vertex ( 1 , 2 ) (1,2) ( 1 , 2 ) cannot share a path with either nonzero boundary entry, by (6) . Since X 24 ≺ X 33 X_{24} \prec X_{33} X 24 ≺ X 33 , the second product is in PBW order. Their lowering weights are both α 1 + α 2 \alpha_1+\alpha_2 α 1 + α 2 , but their nonzero principal symbols have PBW degrees one and two. Thus (97) records their common weight, whereas (12) distinguishes their filtration degrees. Setting z = 1 z=1 z = 1 forgets that distinction and recovers the product formula.
Concluding remarks
Two features of the path conditions play distinct roles in the proof. First, decomposing arrays into marked layers reduces linear independence to level one. Indeed, essential exponents add under tensor products, so
S k = ∑ i = 0 n k i S e i = ∑ i = 0 n k i E Λ i ⊆ E Λ . \mathcal{S}_{\boldsymbol{k}}=\sum_{i=0}^{n}k_i\mathcal{S}_{\boldsymbol{e}_i}=\sum_{i=0}^{n}k_i\mathcal{E}_{\Lambda_i}\subseteq\mathcal{E}_{\Lambda}. S k = i = 0 ∑ n k i S e i = i = 0 ∑ n k i E Λ i ⊆ E Λ .
Second, a violated path inequality provides a straightening relation. The full path case follows from Lemma 3.6 . For a proper path, Lemma 5.7 identifies the finite Levi subsystem, and Theorem 5.5 gives the relation there. Inserting either relation into the original monomial gives
X f v Λ = ∑ v ≻ f β ( v ) = β ( f ) c v X v v Λ . X^{\boldsymbol{f}}v_{\Lambda}=\sum_{\substack{\boldsymbol{v}\succ\boldsymbol{f}\\ \beta(\boldsymbol{v})=\beta(\boldsymbol{f})}}c_{\boldsymbol{v}}X^{\boldsymbol{v}}v_{\Lambda}. X f v Λ = v ≻ f β ( v ) = β ( f ) ∑ c v X v v Λ .
Each replacement preserves principal degree and strictly increases the exponent. Finiteness in each degree therefore forces the process to terminate at admissible monomials. This proves spanning and completes the basis argument. The partition identity follows by principal specialization.
The type A A A and D D D families have different coloured arrays, described in Appendices § 8 and § 9 , respectively. The family introduced by Primc [27 ] is related to type D n + 1 ( 2 ) D^{(2)}_{n+1} D n + 1 ( 2 ) in [21 Conjecture 4.3 and Proposition A.2] . Extending the proof to either family would require a correspondence between its arrays and monomials, together with suitable straightening relations. The character specialization also changes. For example, the specialization of type A A A in (A.4) sends e − α 0 e^{-\alpha_0} e − α 0 to − 1 -1 − 1 , so a lowering weight ∑ i ≥ 1 b i α i \sum_{i\geq1}b_i\alpha_i ∑ i ≥ 1 b i α i contributes ( − 1 ) b 0 q ∑ i ≥ 1 b i (-1)^{b_0}q^{\sum_{i\geq1}b_i} ( − 1 ) b 0 q ∑ i ≥ 1 b i . An extension may therefore account for the signs as well as the new path conditions.
The type A CMPP array
Let n ≥ 1 n\geq1 n ≥ 1 and k = ( k 0 , … , k n ) ∈ Z ≥ 0 n + 1 \boldsymbol{k}=(k_0,\ldots,k_n)\in\mathbb{Z}_{\geq0}^{n+1} k = ( k 0 , … , k n ) ∈ Z ≥ 0 n + 1 , with k = ∑ i = 0 n k i k=\sum_{i=0}^{n}k_i k = ∑ i = 0 n k i . Place the frequencies of colour c c c , for 1 ≤ c ≤ n 1\leq c\leq n 1 ≤ c ≤ n , in two rows: row 2 c − 1 2c-1 2 c − 1 contains the even part sizes, and row 2 c 2c 2 c the odd part sizes. Complete the leftmost two columns by
f 0 ( c ) = k 0 δ c , 1 , f − 1 ( c ) = k c ( 1 ≤ c ≤ n ) . (100) f^{(c)}_0=k_0\delta_{c,1},\qquad f^{(c)}_{-1}=k_c\qquad(1\leq c\leq n). \tag*{(100)} f 0 ( c ) = k 0 δ c , 1 , f − 1 ( c ) = k c ( 1 ≤ c ≤ n ) . (100)
The resulting frequency array is shown in Figure 9 for n = 2 n=2 n = 2 . A full path takes one entry from each row and its successive part indices differ by 1. Thus its column indices j 1 , … , j 2 n j_1,\ldots,j_{2n} j 1 , … , j 2 n satisfy j s ≥ − 1 j_s\geq-1 j s ≥ − 1 , j s + s ≡ 1 ( m o d 2 ) j_s+s\equiv1\pmod{2} j s + s ≡ 1 ( mod 2 ) , and ∣ j s + 1 − j s ∣ = 1 \lvert j_{s+1}-j_s\rvert=1 ∣ j s + 1 − j s ∣ = 1 . The admissible set A k \mathcal{A}_{\boldsymbol{k}} A k consists of the coloured partitions whose arrays satisfy the path inequalities
Figure 9. The type A CMPP array for n = 2 n=2 n = 2 . Blue and red entries have colours 2 and 1, respectively. The solid and dashed lines show two full paths.
∑ s = 1 2 n f ⌊ s / 2 ⌋ ( s ) ≤ k , for every full path. (101) \sum_{s=1}^{2n}f^{(s)}_{\lfloor s/2\rfloor}\leq k,\qquad\text{for every full path.} \tag*{(101)} s = 1 ∑ 2 n f ⌊ s /2 ⌋ ( s ) ≤ k , for every full path. (101)
The prescribed entries at indices − 1 -1 − 1 and 0 0 0 contribute to the path sums but they count neither as parts of π \pi π nor towards its size.
CMPP conjectured a product formula for this family [10 Conjecture 4.1] . Kanade, Russell, Tsuchioka and Warnaar expressed that product as a specialization of an A 2 n ( 2 ) A_{2n}^{(2)} A 2 n ( 2 ) character [21 Conjecture 1.3] . With the simple roots and fundamental weights of that algebra, put
Λ A = 2 k 0 Λ 0 A + ∑ i = 1 n k i Λ i A , χ Λ A = e − Λ A ch L ( Λ A ) . (102) \Lambda^A=2k_0\Lambda_0^A+\sum_{i=1}^{n}k_i\Lambda_i^A,\qquad\chi_{\Lambda^A}=e^{-\Lambda^A}\operatorname{ch}L(\Lambda^A). \tag*{(102)} Λ A = 2 k 0 Λ 0 A + i = 1 ∑ n k i Λ i A , χ Λ A = e − Λ A ch L ( Λ A ) . (102)
This weight has level 2 k 2k 2 k . Define
φ A ( e − α 0 A ) = − 1 , φ A ( e − α i A ) = q ( 1 ≤ i ≤ n ) . (103) \varphi_A(e^{-\alpha_0^A})=-1,\qquad\varphi_A(e^{-\alpha_i^A})=q\quad(1\le i\le n). \tag*{(103)} φ A ( e − α 0 A ) = − 1 , φ A ( e − α i A ) = q ( 1 ≤ i ≤ n ) . (103)
Conjecture A.1. With A k \mathcal{A}_k A k , Λ A \Lambda^A Λ A and φ A \varphi_A φ A as defined above,
∑ π ∈ A k q ∣ π ∣ = φ A ( χ Λ A ) . (104) \sum_{\pi\in\mathcal{A}_k}q^{|\pi|}=\varphi_A(\chi_{\Lambda^A}). \tag*{(104)} π ∈ A k ∑ q ∣ π ∣ = φ A ( χ Λ A ) . (104)
For the product evaluation of this specialization, see [21 Proposition A.1] .
The type D partition array
We recall Primc’s odd-row companion to the CMPP arrays [27 ] , using the row colours of [21 §4, model (b′)] . Fix n ≥ 2 n\ge2 n ≥ 2 , set m = 2 n − 1 m=2n-1 m = 2 n − 1 , and let k = ( k 0 , … , k n ) ∈ Z ≥ 0 n + 1 \mathbf{k}=(k_0,\ldots,k_n)\in\mathbb{Z}_{\ge0}^{n+1} k = ( k 0 , … , k n ) ∈ Z ≥ 0 n + 1 , with k = ∑ i = 0 n k i k=\sum_{i=0}^{n}k_i k = ∑ i = 0 n k i .
A positive part of size p p p and colour c ∈ { 1 , … , m } c\in\{1,\ldots,m\} c ∈ { 1 , … , m } is allowed when p + c p+c p + c is odd. Write f p ( c ) f_p^{(c)} f p ( c ) for its frequency, and prescribe the boundary entries
f − 1 ( 2 i ) = k i ( 1 ≤ i ≤ n − 1 ) , f 0 ( 2 i − 1 ) = k 0 δ i , 1 + k n δ i , n ( 1 ≤ i ≤ n ) . (105) \begin{aligned}
f_{-1}^{(2i)}&=k_i &&(1\le i\le n-1),\\
f_0^{(2i-1)}&=k_0\delta_{i,1}+k_n\delta_{i,n} &&(1\le i\le n). \tag*{(105)}
\end{aligned} f − 1 ( 2 i ) f 0 ( 2 i − 1 ) = k i = k 0 δ i , 1 + k n δ i , n ( 1 ≤ i ≤ n − 1 ) , ( 1 ≤ i ≤ n ) . (105)
The frequency array for n = 2 n=2 n = 2 is shown in Figure 10 . A full path has column indices p 1 , … , p m p_1,\ldots,p_m p 1 , … , p m satisfying p c ≥ − 1 p_c\ge-1 p c ≥ − 1 , p c + c ≡ 1 ( m o d 2 ) p_c+c\equiv1\pmod{2} p c + c ≡ 1 ( mod 2 ) , and ∣ p c + 1 − p c ∣ = 1 \lvert p_{c+1}-p_c\rvert=1 ∣ p c + 1 − p c ∣ = 1 . Let D k \mathcal{D}_k D k be the coloured partitions whose augmented arrays satisfy
Figure 10. The type D D D array for n = 2 n = 2 n = 2 . Red, blue and green entries have colours 1, 2 and 3, respectively. The solid and dashed lines show two full paths, the latter through the boundary entries k 2 k_2 k 2 , k 1 k_1 k 1 , k 0 k_0 k 0 .
∑ c = 1 m f p c ( c ) ≤ k for every full path . (106) \sum_{c=1}^{m}f_{p_c}^{(c)}\le k\qquad\text{for every full path}. \tag*{(106)} c = 1 ∑ m f p c ( c ) ≤ k for every full path . (106)
As in type A, the boundary entries contribute to the path sums but not to the size of the partition.
Let α 0 D , … , α n D \alpha_0^D,\ldots,\alpha_n^D α 0 D , … , α n D and Λ 0 D , … , Λ n D \Lambda_0^D,\ldots,\Lambda_n^D Λ 0 D , … , Λ n D be the simple roots and the usual fundamental weights of D n + 1 ( 2 ) D_{n+1}^{(2)} D n + 1 ( 2 ) , labelled as in [21 §2.1] . Set
Λ D = 2 k 0 Λ 0 D + ∑ i = 1 n − 1 k i Λ i D + 2 k n Λ n D . (107) \Lambda^D = 2k_0\Lambda_0^D + \sum_{i=1}^{n-1} k_i\Lambda_i^D + 2k_n\Lambda_n^D . \tag*{(107)} Λ D = 2 k 0 Λ 0 D + i = 1 ∑ n − 1 k i Λ i D + 2 k n Λ n D . (107)
This weight has level 2 k 2k 2 k . For the normalized character χ Λ D = e − Λ D ch L ( Λ D ) \chi_{\Lambda^D}=e^{-\Lambda^D}\operatorname{ch}L(\Lambda^D) χ Λ D = e − Λ D ch L ( Λ D ) , use the specialization
φ D ( e − α i D ) = { − 1 , i ∈ { 0 , n } , q , 1 ≤ i ≤ n − 1. (108) \varphi_D(e^{-\alpha_i^D}) =
\begin{cases}
-1, & i\in\{0,n\},\\
q, & 1\le i\le n-1.
\end{cases} \tag*{(108)} φ D ( e − α i D ) = { − 1 , q , i ∈ { 0 , n } , 1 ≤ i ≤ n − 1. (108)
KRTW give the following character formulation of Primc’s conjecture [21 Conjecture 4.3] .
Conjecture B.1. For D k \mathcal{D}_k D k and Λ D \Lambda^D Λ D as above,
∑ π ∈ D k q ∣ π ∣ = φ D ( χ Λ D ) . (109) \sum_{\pi\in\mathcal{D}_k}q^{|\pi|}=\varphi_D(\chi_{\Lambda^D}). \tag*{(109)} π ∈ D k ∑ q ∣ π ∣ = φ D ( χ Λ D ) . (109)
The character on the right has the product evaluation in [21 Proposition A.2] .
Acknowledgments
This work is supported by the National Natural Science Foundation of China (Grant No. 12071235) and the Fundamental Research Funds for the Central Universities of China (Grant Nos. 63261129 and 63261210).
AI Declaration
OpenAI ChatGPT (GPT 6 pro) and OpenAI Codex (GPT-6 Astra Ultra) were used to develop, check and revise the tensor product argument and the comparison with finite Levi modules. These tools also assisted with literature searches, calculations of examples and English exposition. The author remains responsible for the mathematical content, references and final decision to submit.
Xiangyu Ding
Center for Combinatorics, LPMC, Nankai University
Tianjin 300071, P.R. China
Corresponding author. Email: dingmath@mail.nankai.edu.cn