Introduction
Let G G G be a connected reductive algebraic group over an algebraically closed field K \mathbb{K} K of positive characteristic p p p . The Weyl modules Δ ( λ ) \Delta(\lambda) Δ ( λ ) are finite-dimensional rational G G G -modules indexed by dominant weights. Their structure is a central problem in the rational representation theory of G G G and even the homomorphism spaces Hom G ( Δ ( λ ) , Δ ( μ ) ) \operatorname{Hom}_G(\Delta(\lambda),\Delta(\mu)) Hom G ( Δ ( λ ) , Δ ( μ )) are far from understood.
For the general linear group G L n \mathrm{GL}_n GL n , nonzero homomorphisms Δ ( λ ) → Δ ( μ ) \Delta(\lambda) \to\Delta(\mu) Δ ( λ ) → Δ ( μ ) under suitable conditions on λ \lambda λ , μ \mu μ and p p p , where μ \mu μ is obtained by raising cells from a row of λ \lambda λ , were obtained by Carter and Lusztig [3 ] and by Carter and Payne [4 ] . Extensions of the Carter-Payne theorem were obtained by Fayers and Martin [9 ] . In the special case where a single cell is raised, it is known that the corresponding Hom-space has dimension at most one; see Kulkarni [15 ] and Ellers and Murray [7 ] . On the other hand, there are examples of Hom-spaces that have dimension greater than one and even arbitrarily large dimensions, see Dodge [6 ] and Lyle [16 ] . Stability properties of their dimensions were studied in [8 ] .
For G L 3 \mathrm{GL}_3 GL 3 , the situation is better understood. When μ \mu μ is obtained from λ \lambda λ by moving cells from the third row to the first, the dimension of Hom G ( Δ ( λ ) , Δ ( μ ) ) \operatorname{Hom}_G(\Delta(\lambda),\Delta(\mu)) Hom G ( Δ ( λ ) , Δ ( μ )) follows from the computation of the corresponding integral Ext 1 \operatorname{Ext}^1 Ext 1 group [10 , 2 ] . Cox and Parker [5 ] classified all homomorphisms between Weyl modules for G L 3 \mathrm{GL}_3 GL 3 when p > 2 p>2 p > 2 and proved that the homomorphism spaces have dimension at most one. In this paper, we treat the remaining case p = 2 p=2 p = 2 , thus completing the classification for G L 3 \mathrm{GL}_3 GL 3 .
Our main result is a recursive classification of the homomorphism spaces between Weyl modules for G L 3 \mathrm{GL}_3 GL 3 in characteristic 2 2 2 (Theorem 4.1). For any pair of dominant weights λ , μ \lambda,\mu λ , μ , the procedure determines whether Hom G ( Δ ( λ ) , Δ ( μ ) ) \operatorname{Hom}_G(\Delta(\lambda),\Delta(\mu)) Hom G ( Δ ( λ ) , Δ ( μ )) is nonzero and, when it is, constructs an explicit generator in terms of semistandard tableaux. In particular, we have that dim K Hom G ( Δ ( λ ) , Δ ( μ ) ) ≤ 1 \dim_{\mathbb{K}}\operatorname{Hom}_G(\Delta(\lambda),\Delta(\mu))\leq1 dim K Hom G ( Δ ( λ ) , Δ ( μ )) ≤ 1 for all λ , μ \lambda,\mu λ , μ (Corollary 4.2).
In Section 2, we discuss preliminary reductions and use results from [17 ] to identify the homomorphism spaces with the common kernels of two explicit matrices. In Section 3, we use Lucas’ theorem to relate these kernels to those associated with smaller pairs of partitions. We also give explicit formulas for recovering the coefficients of a homomorphism from those of the smaller pair.
In § 4 , we prove Theorem 4.1. In § 4.2 , we combine our results with McDowell’s theorem on the inverse Schur functor [18 ] to determine the dimensions of certain homomorphism spaces between Specht modules in characteristic 2. This includes all pairs of partitions of r ≥ 3 r \ge3 r ≥ 3 for which the source has at most two parts and the target has at most three parts (Corollary 4.4 ). § 5 contains the proofs of two lemmas used in § 3 .
Preliminaries
Throughout this paper, K \mathbb{K} K denotes an algebraically closed field of characteristic 2 and G G G denotes the general linear group G L 3 ( K ) \mathrm{GL}_{3}(\mathbb{K}) GL 3 ( K ) . We identify the weights of G G G with triples α = ( α 1 , α 2 , α 3 ) \alpha= (\alpha_{1}, \alpha_{2}, \alpha_{3}) α = ( α 1 , α 2 , α 3 ) of integers and the dominant weights of G G G with triples λ = ( λ 1 , λ 2 , λ 3 ) \lambda= (\lambda_{1}, \lambda_{2}, \lambda_{3}) λ = ( λ 1 , λ 2 , λ 3 ) of integers such that λ 1 ≥ λ 2 ≥ λ 3 \lambda_{1} \ge\lambda_{2} \ge\lambda_{3} λ 1 ≥ λ 2 ≥ λ 3 . For a dominant weight λ \lambda λ we denote the corresponding Weyl module of G G G by Δ ( λ ) \Delta(\lambda) Δ ( λ ) , see [13 II.2.13] where the notation V ( λ ) V(\lambda) V ( λ ) is used. In the case where λ \lambda λ is a partition, i.e. when λ 1 ≥ λ 2 ≥ λ 3 ≥ 0 \lambda_{1} \ge\lambda_{2} \ge\lambda_{3} \ge0 λ 1 ≥ λ 2 ≥ λ 3 ≥ 0 , we use properties of Δ ( λ ) \Delta(\lambda) Δ ( λ ) described in [19 Section 2] . In Weyman’s book, Δ ( λ ) \Delta(\lambda) Δ ( λ ) is denoted by K λ E K_{\lambda}E K λ E , where E E E is the natural G L 3 ( K ) \mathrm{GL}_{3}(\mathbb{K}) GL 3 ( K ) -module.
Reductions
If α = ( α 1 , α 2 , α 3 ) \alpha= (\alpha_{1}, \alpha_{2}, \alpha_{3}) α = ( α 1 , α 2 , α 3 ) and β = ( β 1 , β 2 , β 3 ) \beta= (\beta_{1}, \beta_{2}, \beta_{3}) β = ( β 1 , β 2 , β 3 ) are weights, we write α ⊴ β \alpha\unlhd\beta α ⊴ β if α 1 ≤ β 1 \alpha_{1} \le\beta_{1} α 1 ≤ β 1 , α 1 + α 2 ≤ β 1 + β 2 \alpha_{1} + \alpha_{2} \le\beta_{1} + \beta_{2} α 1 + α 2 ≤ β 1 + β 2 and α 1 + α 2 + α 3 = β 1 + β 2 + β 3 \alpha_{1} + \alpha_{2} + \alpha_{3} = \beta_{1} + \beta_{2} + \beta_{3} α 1 + α 2 + α 3 = β 1 + β 2 + β 3 . Since every weight α \alpha α of Δ ( μ ) \Delta(\mu) Δ ( μ ) satisfies α ⊴ μ \alpha\unlhd\mu α ⊴ μ , see [13 II.2.2, Proposition (b)] for the corresponding statement involving the duals of Weyl modules, we have that Hom G ( Δ ( λ ) , Δ ( μ ) ) = 0 \operatorname{Hom}_{G}(\Delta(\lambda), \Delta(\mu)) = 0 Hom G ( Δ ( λ ) , Δ ( μ )) = 0 unless λ ⊴ μ \lambda\unlhd\mu λ ⊴ μ . Hence we may assume μ 3 ≤ λ 3 \mu_{3} \le\lambda_{3} μ 3 ≤ λ 3 .
For an integer m m m let det m \det^{m} det m denote the m m m th power of the determinant representation. If λ = ( λ 1 , λ 2 , λ 3 ) \lambda= (\lambda_{1}, \lambda_{2}, \lambda_{3}) λ = ( λ 1 , λ 2 , λ 3 ) is a dominant weight, then Δ ( λ ) ⊗ det m ≅ Δ ( λ 1 + m , λ 2 + m , λ 3 + m ) \Delta(\lambda) \otimes\det^{m} \cong\Delta(\lambda_{1} + m, \lambda_{2} + m, \lambda_{3} + m) Δ ( λ ) ⊗ det m ≅ Δ ( λ 1 + m , λ 2 + m , λ 3 + m ) . For any dominant weights λ \lambda λ , μ \mu μ we have Hom G ( Δ ( λ ) , Δ ( μ ) ) ≅ Hom G ( Δ ( λ ) ⊗ det m , Δ ( μ ) ⊗ det m ) \operatorname{Hom}_{G}(\Delta(\lambda), \Delta(\mu)) \cong\operatorname{Hom}_{G}(\Delta(\lambda) \otimes\det^{m}, \Delta(\mu) \otimes\det^{m}) Hom G ( Δ ( λ ) , Δ ( μ )) ≅ Hom G ( Δ ( λ ) ⊗ det m , Δ ( μ ) ⊗ det m ) . Taking m = − μ 3 m = -\mu_{3} m = − μ 3 , we may assume λ \lambda λ and μ \mu μ are partitions and μ \mu μ has at most two parts.
Lemma 2.1. Let λ = ( λ 1 , λ 2 , λ 3 ) \lambda= (\lambda_{1}, \lambda_{2}, \lambda_{3}) λ = ( λ 1 , λ 2 , λ 3 ) and μ = ( μ 1 , μ 2 ) \mu= (\mu_{1}, \mu_{2}) μ = ( μ 1 , μ 2 ) be partitions of r r r with μ 2 > λ 1 \mu_{2} > \lambda_{1} μ 2 > λ 1 . Then Hom G ( Δ ( λ ) , Δ ( μ ) ) \operatorname{Hom}_{G}(\Delta(\lambda), \Delta(\mu)) Hom G ( Δ ( λ ) , Δ ( μ )) is isomorphic to Hom G ( Δ ( μ 1 − λ 3 , μ 1 − λ 2 , μ 1 − λ 1 ) , Δ ( μ 1 , μ 1 − μ 2 ) ) \operatorname{Hom}_{G}(\Delta(\mu_{1} - \lambda_{3}, \mu_{1} - \lambda_{2}, \mu_{1} - \lambda_{1}), \Delta(\mu_{1}, \mu_{1} - \mu_{2})) Hom G ( Δ ( μ 1 − λ 3 , μ 1 − λ 2 , μ 1 − λ 1 ) , Δ ( μ 1 , μ 1 − μ 2 )) .
Proof. Consider the permutation matrix w 0 ∈ G w_{0} \in G w 0 ∈ G corresponding to the longest element ( 13 ) (13) ( 13 ) of the symmetric group S 3 S_{3} S 3 and the automorphism θ \theta θ of G G G given by θ ( g ) = w 0 ( g − 1 ) T w 0 \theta(g) = w_{0}(g^{-1})^{T}w_{0} θ ( g ) = w 0 ( g − 1 ) T w 0 , where g ∈ G g \in G g ∈ G and ( g − 1 ) T (g^{-1})^{T} ( g − 1 ) T denotes the transpose of g − 1 g^{-1} g − 1 . For a G G G -module M M M , let θ M {}^{\theta}M θ M denote the twist of M M M by the map θ \theta θ . From [13 II.2.13, after equation (1)] , we have θ Δ ( ν ) ≅ Δ ( − ν 3 , − ν 2 , − ν 1 ) {}^{\theta}\Delta(\nu) \cong\Delta(-\nu_{3}, -\nu_{2}, -\nu_{1}) θ Δ ( ν ) ≅ Δ ( − ν 3 , − ν 2 , − ν 1 ) for every dominant weight ν \nu ν . Thus,
Hom G ( Δ ( λ ) , Δ ( μ ) ) ≅ Hom G ( Δ ( − λ 3 , − λ 2 , − λ 1 ) , Δ ( 0 , − μ 2 , − μ 1 ) ) \operatorname{Hom}_{G}(\Delta(\lambda), \Delta(\mu)) \cong\operatorname{Hom}_{G}(\Delta(-\lambda_{3}, -\lambda_{2}, -\lambda_{1}), \Delta(0, -\mu_{2}, -\mu_{1})) Hom G ( Δ ( λ ) , Δ ( μ )) ≅ Hom G ( Δ ( − λ 3 , − λ 2 , − λ 1 ) , Δ ( 0 , − μ 2 , − μ 1 ))
≅ Hom G ( Δ ( μ 1 − λ 3 , μ 1 − λ 2 , μ 1 − λ 1 ) , Δ ( μ 1 , μ 1 − μ 2 ) ) , \cong\operatorname{Hom}_{G}(\Delta(\mu_{1} - \lambda_{3}, \mu_{1} - \lambda_{2}, \mu_{1} - \lambda_{1}), \Delta(\mu_{1}, \mu_{1} - \mu_{2})), ≅ Hom G ( Δ ( μ 1 − λ 3 , μ 1 − λ 2 , μ 1 − λ 1 ) , Δ ( μ 1 , μ 1 − μ 2 )) ,
where the last isomorphism comes from tensoring both modules with det μ 1 \det^{\mu_{1}} det μ 1 .
In the previous lemma, both ( μ 1 − λ 3 , μ 1 − λ 2 , μ 1 − λ 1 ) (\mu_{1} - \lambda_{3}, \mu_{1} - \lambda_{2}, \mu_{1} - \lambda_{1}) ( μ 1 − λ 3 , μ 1 − λ 2 , μ 1 − λ 1 ) and ( μ 1 , μ 1 − μ 2 ) (\mu_{1}, \mu_{1} - \mu_{2}) ( μ 1 , μ 1 − μ 2 ) are partitions and we have μ 1 − μ 2 ≤ μ 1 − λ 3 \mu_{1} - \mu_{2} \le\mu_{1} - \lambda_{3} μ 1 − μ 2 ≤ μ 1 − λ 3 , since λ 3 ≤ λ 1 < μ 2 \lambda_{3} \le\lambda_{1} < \mu_{2} λ 3 ≤ λ 1 < μ 2 . Hence, when studying Hom G ( Δ ( λ ) , Δ ( μ ) ) \operatorname{Hom}_{G}(\Delta(\lambda), \Delta(\mu)) Hom G ( Δ ( λ ) , Δ ( μ )) , we may assume λ \lambda λ and μ \mu μ are partitions such that
∣ λ ∣ = ∣ μ ∣ , λ = ( λ 1 , λ 2 , λ 3 ) , μ = ( μ 1 , μ 2 ) , μ 2 ≤ λ 1 , (1) |\lambda| = |\mu|,\quad\lambda= (\lambda_{1}, \lambda_{2}, \lambda_{3}),\quad\mu= (\mu_{1}, \mu_{2}),\quad\mu_{2} \le\lambda_{1}, \tag*{(1)} ∣ λ ∣ = ∣ μ ∣ , λ = ( λ 1 , λ 2 , λ 3 ) , μ = ( μ 1 , μ 2 ) , μ 2 ≤ λ 1 , (1)
where ∣ ν ∣ |\nu| ∣ ν ∣ denotes the sum of the parts of a partition ν \nu ν . We use throughout the notation
ℓ = max ( 0 , μ 2 − λ 3 ) , L = min ( λ 2 , μ 2 ) . (2) \ell= \max(0, \mu_{2} - \lambda_{3}),\quad L = \min(\lambda_{2}, \mu_{2}). \tag*{(2)} ℓ = max ( 0 , μ 2 − λ 3 ) , L = min ( λ 2 , μ 2 ) . (2)
Basis of Hom G ( Δ ( λ ) , Δ ( μ ) ) \operatorname{Hom}_{G}(\Delta(\lambda), \Delta(\mu)) Hom G ( Δ ( λ ) , Δ ( μ ))
We assume throughout (2.1) and (2.2).
Let E = K 3 E = \mathbb{K}^{3} E = K 3 be the natural G L 3 ( K ) \mathrm{GL}_{3}(\mathbb{K}) GL 3 ( K ) -module. The divided power algebra D E = ⨁ i ≥ 0 D i E DE = \bigoplus_{i \ge0} D_{i}E D E = ⨁ i ≥ 0 D i E of E E E is defined as the graded dual of the bialgebra S ( E ∗ ) S(E^{*}) S ( E ∗ ) , where E ∗ E^{*} E ∗ is the linear dual of E E E and S ( E ∗ ) S(E^{*}) S ( E ∗ ) is the symmetric algebra of E ∗ E^{*} E ∗ , see [19 Section 1.1] . For a sequence α = ( α 1 , α 2 , α 3 ) \alpha= (\alpha_{1}, \alpha_{2}, \alpha_{3}) α = ( α 1 , α 2 , α 3 ) of nonnegative integers, let D ( α ) = D α 1 E ⊗ D α 2 E ⊗ D α 3 E D(\alpha) = D_{\alpha_{1}}E \otimes D_{\alpha_{2}}E \otimes D_{\alpha_{3}}E D ( α ) = D α 1 E ⊗ D α 2 E ⊗ D α 3 E . As a G G G -module, D ( α ) D(\alpha) D ( α ) is cyclic, generated by e 1 ( α 1 ) ⊗ e 2 ( α 2 ) ⊗ e 3 ( α 3 ) e_{1}^{(\alpha_{1})} \otimes e_{2}^{(\alpha_{2})} \otimes e_{3}^{(\alpha_{3})} e 1 ( α 1 ) ⊗ e 2 ( α 2 ) ⊗ e 3 ( α 3 ) , where { e 1 , e 2 , e 3 } \{e_{1}, e_{2}, e_{3}\} { e 1 , e 2 , e 3 } is the usual basis of E E E [14 isomorphism (8.3.17)] . For a partition λ \lambda λ , we recall that there is a natural surjection π λ : D ( λ ) → Δ ( λ ) \pi_\lambda: D(\lambda) \to\Delta(\lambda) π λ : D ( λ ) → Δ ( λ ) , unique up to a nonzero scalar multiple [19 Proposition 2.1.15(a)] . A map ϕ ∈ Hom G ( D ( λ ) , Δ ( μ ) ) \phi\in\operatorname{Hom}_{G}(D(\lambda), \Delta(\mu)) ϕ ∈ Hom G ( D ( λ ) , Δ ( μ )) descends to a map ϕ ‾ ∈ Hom G ( Δ ( λ ) , Δ ( μ ) ) \overline{\phi} \in\operatorname{Hom}_{G}(\Delta(\lambda), \Delta(\mu)) ϕ ∈ Hom G ( Δ ( λ ) , Δ ( μ )) if and only if ϕ ( ker π λ ) = 0 \phi(\ker\pi_\lambda) = 0 ϕ ( ker π λ ) = 0 . In this case, ϕ ‾ \overline{\phi} ϕ is uniquely determined by ϕ ‾ = ϕ ∘ π λ \overline{\phi} = \phi\circ\pi_\lambda ϕ = ϕ ∘ π λ .
We use the ordered alphabet 1 < 2 < 3 1 < 2 < 3 1 < 2 < 3 . For a partition μ = ( μ 1 , μ 2 ) \mu= (\mu_1,\mu_2) μ = ( μ 1 , μ 2 ) , a tableau of shape μ \mu μ is a filling of the diagram of μ \mu μ with entries from { 1 , 2 , 3 } \{1,2,3\} { 1 , 2 , 3 } . Such a tableau is called semistandard if the entries are weakly increasing in the rows from left to right and strictly increasing in the columns from top to bottom. The weight of a tableau T T T is the tuple α = ( α 1 , α 2 , α 3 ) \alpha= (\alpha_1,\alpha_2,\alpha_3) α = ( α 1 , α 2 , α 3 ) , where α i \alpha_i α i is the number of appearances of i i i in T T T . To a tableau T T T of shape μ = ( μ 1 , μ 2 ) \mu= (\mu_1,\mu_2) μ = ( μ 1 , μ 2 ) we may associate an element x T = x T ( 1 ) ⊗ x T ( 2 ) ∈ D ( μ 1 , μ 2 ) x_T = x_T(1) \otimes x_T(2) \in D(\mu_1,\mu_2) x T = x T ( 1 ) ⊗ x T ( 2 ) ∈ D ( μ 1 , μ 2 ) , where x T ( i ) = e 1 ( a i 1 ) e 2 ( a i 2 ) e 3 ( a i 3 ) x_T(i) = e_1^{(a_{i1})} e_2^{(a_{i2})} e_3^{(a_{i3})} x T ( i ) = e 1 ( a i 1 ) e 2 ( a i 2 ) e 3 ( a i 3 ) and a i j a_{ij} a ij is equal to the number of appearances of j j j in the i i i th row of T T T , 1 ≤ i ≤ 2 1 \le i \le2 1 ≤ i ≤ 2 , 1 ≤ j ≤ 3 1 \le j \le3 1 ≤ j ≤ 3 .
A basis of Δ ( μ ) \Delta(\mu) Δ ( μ ) is given by the elements π μ ( x T ) \pi_\mu(x_T) π μ ( x T ) , where T T T runs over the semistandard tableaux of shape μ \mu μ [19 Proposition 2.1.15(b)] .
A basis of Hom G ( D ( λ ) , Δ ( μ ) ) \operatorname{Hom}_{G}(D(\lambda),\Delta(\mu)) Hom G ( D ( λ ) , Δ ( μ )) is parameterized by the semistandard tableaux of shape μ \mu μ that have weight λ \lambda λ [14 Lemma 8.3.18 and Corollary 8.4.14] . For such a tableau, let u u u denote the number of entries equal to 2 2 2 in its second row.
Lemma 2.2. If ℓ > L \ell> L ℓ > L , then Hom G ( D ( λ ) , Δ ( μ ) ) = 0 \operatorname{Hom}_{G}(D(\lambda),\Delta(\mu)) = 0 Hom G ( D ( λ ) , Δ ( μ )) = 0 . If ℓ ≤ L \ell\le L ℓ ≤ L , then a basis of Hom G ( D ( λ ) , Δ ( μ ) ) \operatorname{Hom}_{G}(D(\lambda),\Delta(\mu)) Hom G ( D ( λ ) , Δ ( μ )) is { ϕ ℓ , … , ϕ L } \{\phi_\ell,\ldots,\phi_L\} { ϕ ℓ , … , ϕ L } , where for each ℓ ≤ u ≤ L \ell\le u \le L ℓ ≤ u ≤ L , the map ϕ u : D ( λ ) → Δ ( μ ) \phi_u : D(\lambda) \to\Delta(\mu) ϕ u : D ( λ ) → Δ ( μ ) is defined by
ϕ u ( e 1 ( λ 1 ) ⊗ e 2 ( λ 2 ) ⊗ e 3 ( λ 3 ) ) = π μ ( e 1 ( λ 1 ) e 2 ( λ 2 − u ) e 3 ( λ 3 − μ 2 + u ) ⊗ e 2 ( u ) e 3 ( μ 2 − u ) ) . \phi_u(e_1^{(\lambda_1)} \otimes e_2^{(\lambda_2)} \otimes e_3^{(\lambda_3)}) = \pi_\mu(e_1^{(\lambda_1)} e_2^{(\lambda_2-u)} e_3^{(\lambda_3-\mu_2+u)} \otimes e_2^{(u)} e_3^{(\mu_2-u)}). ϕ u ( e 1 ( λ 1 ) ⊗ e 2 ( λ 2 ) ⊗ e 3 ( λ 3 ) ) = π μ ( e 1 ( λ 1 ) e 2 ( λ 2 − u ) e 3 ( λ 3 − μ 2 + u ) ⊗ e 2 ( u ) e 3 ( μ 2 − u ) ) .
Proof. This follows from [17 Lemma 2.3 and Remark 3.3(1)] for m = 3 m = 3 m = 3 and n = 3 n = 3 n = 3 .
The matrices A A A and B B B
We assume throughout (1) and (2) . Suppose ℓ ≤ L \ell\le L ℓ ≤ L . In this subsection, we define integer matrices A A A and B B B whose reductions modulo 2 2 2 determine Hom G ( Δ ( λ ) , Δ ( μ ) ) \operatorname{Hom}_{G}(\Delta(\lambda),\Delta(\mu)) Hom G ( Δ ( λ ) , Δ ( μ )) . The columns of both are indexed by u u u satisfying ℓ ≤ u ≤ L \ell\le u \le L ℓ ≤ u ≤ L . The rows are indexed by the ordered sets
R A = { ( t , k ) : 1 ≤ t ≤ λ 2 , ℓ ≤ k ≤ min ( λ 2 − t , μ 2 ) } , (3) R_A = \{(t,k) : 1 \le t \le\lambda_2,\ \ell\le k \le\min(\lambda_2-t,\mu_2)\}, \tag*{(3)} R A = {( t , k ) : 1 ≤ t ≤ λ 2 , ℓ ≤ k ≤ min ( λ 2 − t , μ 2 )} , (3)
R B = { ( t , k ) : 1 ≤ t ≤ λ 3 , max ( 0 , μ 2 − λ 3 + t ) ≤ k ≤ min ( λ 2 + t , μ 2 ) } , (4) R_B = \{(t,k) : 1 \le t \le\lambda_3,\ \max(0,\mu_2-\lambda_3+t) \le k \le\min(\lambda_2+t,\mu_2)\}, \tag*{(4)} R B = {( t , k ) : 1 ≤ t ≤ λ 3 , max ( 0 , μ 2 − λ 3 + t ) ≤ k ≤ min ( λ 2 + t , μ 2 )} , (4)
where we consider the lexicographic ordering on pairs ( t , k ) (t,k) ( t , k ) .
For binomial coefficients we use the convention ( a b ) = 0 \binom{a}{b} = 0 ( b a ) = 0 if b < 0 b < 0 b < 0 or a < b a < b a < b .
Definition 2.3. Assume ℓ ≤ L \ell\le L ℓ ≤ L . For ℓ ≤ u ≤ L \ell\le u \le L ℓ ≤ u ≤ L define
A ( t , k ) , u = ( − 1 ) u − k ( μ 2 − k λ 1 + t − u ) ( t + k − u t ) , ( t , k ) ∈ R A , A_{(t,k),u} = (-1)^{u-k} \binom{\mu_2-k}{\lambda_1+t-u}\binom{t+k-u}{t},\qquad(t,k) \in R_A, A ( t , k ) , u = ( − 1 ) u − k ( λ 1 + t − u μ 2 − k ) ( t t + k − u ) , ( t , k ) ∈ R A , B ( t , k ) , u = ( k u ) ( λ 2 + t − k t + u − k ) , ( t , k ) ∈ R B . B_{(t,k),u} = \binom{k}{u}\binom{\lambda_2+t-k}{t+u-k},\qquad(t,k) \in R_B. B ( t , k ) , u = ( u k ) ( t + u − k λ 2 + t − k ) , ( t , k ) ∈ R B .
We denote the reductions mod 2 2 2 of A A A and B B B again by A A A and B B B . Let [ A B ] \begin{bmatrix} A \\ B \end{bmatrix} [ A B ] be the matrix whose rows are those of A A A and B B B , and let ker K [ A B ] \ker_{\mathbb{K}}\begin{bmatrix} A \\ B \end{bmatrix} ker K [ A B ] denote its kernel over K \mathbb{K} K . In the next proposition we use the notation of Section 2.2.
Proposition 2.4. The map ker K [ A B ] → Hom G ( Δ ( λ ) , Δ ( μ ) ) \ker_{\mathbb{K}}\begin{bmatrix} A \\ B \end{bmatrix} \to\operatorname{Hom}_{G}(\Delta(\lambda),\Delta(\mu)) ker K [ A B ] → Hom G ( Δ ( λ ) , Δ ( μ )) is an isomorphism, where
c = ( c ℓ , … , c L ) T ↦ ϕ ‾ c , ϕ ‾ c ∘ π λ = ∑ u = ℓ L c u ϕ u . c = (c_\ell,\ldots,c_L)^{T} \mapsto\overline{\phi}_c,\qquad\overline{\phi}_c \circ\pi_\lambda= \sum_{u=\ell}^{L} c_u\phi_u. c = ( c ℓ , … , c L ) T ↦ ϕ c , ϕ c ∘ π λ = u = ℓ ∑ L c u ϕ u .
Proof. Let ϕ = ∑ u = ℓ L c u ϕ u ∈ Hom G ( D ( λ ) , Δ ( μ ) ) \phi= \sum_{u=\ell}^{L} c_u\phi_u \in\operatorname{Hom}_{G}(D(\lambda),\Delta(\mu)) ϕ = ∑ u = ℓ L c u ϕ u ∈ Hom G ( D ( λ ) , Δ ( μ )) , c u ∈ K c_u \in\mathbb{K} c u ∈ K . From the second conclusion of [1 Theorem II.3.16] and the fact that D ( α ) D(\alpha) D ( α ) is a cyclic G G G -module for any sequence α = ( α 1 , α 2 , α 3 ) \alpha= (\alpha_1,\alpha_2,\alpha_3) α = ( α 1 , α 2 , α 3 ) of nonnegative integers, it follows that the kernel of the natural projection π λ : D ( λ ) → Δ ( λ ) \pi_\lambda: D(\lambda) \to\Delta(\lambda) π λ : D ( λ ) → Δ ( λ ) is generated as a G G G -module by
x t = e 1 ( λ 1 ) ⊗ e 2 ( t ) e 2 ( λ 2 − t ) ⊗ e 3 ( λ 3 ) , 1 ≤ t ≤ λ 2 , x_t=e_1^{(\lambda_1)}\otimes e_2^{(t)}e_2^{(\lambda_2-t)}\otimes e_3^{(\lambda_3)}, \quad1\leq t\leq\lambda_2, x t = e 1 ( λ 1 ) ⊗ e 2 ( t ) e 2 ( λ 2 − t ) ⊗ e 3 ( λ 3 ) , 1 ≤ t ≤ λ 2 ,
y t = e 1 ( λ 1 ) ⊗ e 2 ( λ 2 ) ⊗ e 2 ( t ) e 3 ( λ 3 − t ) , 1 ≤ t ≤ λ 3 . y_t=e_1^{(\lambda_1)}\otimes e_2^{(\lambda_2)}\otimes e_2^{(t)}e_3^{(\lambda_3-t)}, \quad1\leq t\leq\lambda_3. y t = e 1 ( λ 1 ) ⊗ e 2 ( λ 2 ) ⊗ e 2 ( t ) e 3 ( λ 3 − t ) , 1 ≤ t ≤ λ 3 .
Thus ϕ \phi ϕ descends to a map defined on Δ ( λ ) \Delta(\lambda) Δ ( λ ) if and only if ϕ ( x t ) = 0 \phi(x_t)=0 ϕ ( x t ) = 0 for 1 ≤ t ≤ λ 2 1\leq t\leq\lambda_2 1 ≤ t ≤ λ 2 and ϕ ( y t ) = 0 \phi(y_t)=0 ϕ ( y t ) = 0 for 1 ≤ t ≤ λ 3 1\leq t\leq\lambda_3 1 ≤ t ≤ λ 3 .
If ( t , k ) ∈ R A (t,k)\in R_A ( t , k ) ∈ R A , let T t , k A T^A_{t,k} T t , k A be the semistandard tableau of shape μ \mu μ and weight ( λ 1 + t , λ 2 − t , λ 3 ) (\lambda_1+t,\lambda_2-t,\lambda_3) ( λ 1 + t , λ 2 − t , λ 3 ) with second row 2 ( k ) 3 ( μ 2 − k ) 2^{(k)}3^{(\mu_2-k)} 2 ( k ) 3 ( μ 2 − k ) . Likewise, if ( t , k ) ∈ R B (t,k)\in R_B ( t , k ) ∈ R B , let T t , k B T^B_{t,k} T t , k B be the semistandard tableau of shape μ \mu μ and weight ( λ 1 , λ 2 + t , λ 3 − t ) (\lambda_1,\lambda_2+t,\lambda_3-t) ( λ 1 , λ 2 + t , λ 3 − t ) with second row 2 ( k ) 3 ( μ 2 − k ) 2^{(k)}3^{(\mu_2-k)} 2 ( k ) 3 ( μ 2 − k ) . The last formula for φ T ( x ) \varphi_T(x) φ T ( x ) on p. 279 and the formula for φ T ( y ) \varphi_T(y) φ T ( y ) immediately preceding (2) on p. 280 of [17 ] , specialized to the case m = 3 m=3 m = 3 parts, give
ϕ ( x t ) = ∑ ( t , k ) ∈ R A ( A c ) ( t , k ) π μ ( x T t , k A ) , ϕ ( y t ) = ∑ ( t , k ) ∈ R B ( B c ) ( t , k ) π μ ( x T t , k B ) . \phi(x_t)=\sum_{(t,k)\in R_A}(Ac)_{(t,k)}\pi_\mu(x_{T^A_{t,k}}),\qquad\phi(y_t)=\sum_{(t,k)\in R_B}(Bc)_{(t,k)}\pi_\mu(x_{T^B_{t,k}}). ϕ ( x t ) = ( t , k ) ∈ R A ∑ ( A c ) ( t , k ) π μ ( x T t , k A ) , ϕ ( y t ) = ( t , k ) ∈ R B ∑ ( B c ) ( t , k ) π μ ( x T t , k B ) .
Therefore, the map ϕ : D ( λ ) → Δ ( μ ) \phi:D(\lambda)\to\Delta(\mu) ϕ : D ( λ ) → Δ ( μ ) descends to a map Δ ( λ ) → Δ ( μ ) \Delta(\lambda)\to\Delta(\mu) Δ ( λ ) → Δ ( μ ) if and only if A c = B c = 0 Ac=Bc=0 A c = B c = 0 . Since the maps ϕ u \phi_u ϕ u , ℓ ≤ u ≤ L \ell\leq u\leq L ℓ ≤ u ≤ L , form a basis of Hom G ( D ( λ ) , Δ ( μ ) ) \operatorname{Hom}_G(D(\lambda),\Delta(\mu)) Hom G ( D ( λ ) , Δ ( μ )) , the result follows. □
Binomial congruences
We will use the following special case of Lucas’ theorem.
Lemma 2.5. Let N ≥ 0 N\geq0 N ≥ 0 , M ∈ Z M\in\mathbb{Z} M ∈ Z and ϵ , η ∈ { 0 , 1 } \epsilon,\eta\in\{0,1\} ϵ , η ∈ { 0 , 1 } . Then
( 2 N + ϵ 2 M + η ) ≡ ( N M ) ( ϵ η ) ( m o d 2 ) . \binom{2N+\epsilon}{2M+\eta}\equiv\binom{N}{M}\binom{\epsilon}{\eta}\pmod2. ( 2 M + η 2 N + ϵ ) ≡ ( M N ) ( η ϵ ) ( mod 2 ) .
For a positive integer n n n , let v 2 ( n ) v_2(n) v 2 ( n ) be the exponent of 2 in the prime factorization of n n n .
Lemma 2.6. [12 Corollary 22.5] Let n , d n,d n , d be nonnegative integers. Then ( n + t t ) ≡ 0 ( m o d 2 ) \binom{n+t}{t}\equiv0\pmod2 ( t n + t ) ≡ 0 ( mod 2 ) for every 1 ≤ t ≤ d 1\leq t\leq d 1 ≤ t ≤ d if and only if d < 2 v 2 ( n + 1 ) d<2^{v_2(n+1)} d < 2 v 2 ( n + 1 ) .
The recursion
Throughout this section we assume (2.1) and (2.2). Also we assume ℓ ≤ L \ell\leq L ℓ ≤ L .
Reduction to (0, 0, 0) and (0, 1, 1)
Suppose ℓ < L \ell<L ℓ < L . Let c = ( c ℓ , … , c L ) ∈ ker K [ A B ] c=(c_\ell,\ldots,c_L)\in\ker_K\begin{bmatrix}A\\B\end{bmatrix} c = ( c ℓ , … , c L ) ∈ ker K [ A B ] . First we establish that ( 1 , u ) ∈ R A (1,u)\in R_A ( 1 , u ) ∈ R A and ( 1 , u + 1 ) ∈ R B (1,u+1)\in R_B ( 1 , u + 1 ) ∈ R B for every ℓ ≤ u < L \ell\leq u<L ℓ ≤ u < L . Since ℓ < L \ell<L ℓ < L , we have λ 3 > 0 \lambda_3>0 λ 3 > 0 . For ℓ ≤ u < L \ell\leq u<L ℓ ≤ u < L , we have ℓ ≤ u ≤ L − 1 ≤ min ( λ 2 − 1 , μ 2 ) \ell\leq u\leq L-1\leq\min(\lambda_2-1,\mu_2) ℓ ≤ u ≤ L − 1 ≤ min ( λ 2 − 1 , μ 2 ) and
max ( 0 , μ 2 − λ 3 + 1 ) ≤ ℓ + 1 ≤ u + 1 ≤ L ≤ min ( λ 2 + 1 , μ 2 ) . \max(0,\mu_2-\lambda_3+1)\leq\ell+1\leq u+1\leq L\leq\min(\lambda_2+1,\mu_2). max ( 0 , μ 2 − λ 3 + 1 ) ≤ ℓ + 1 ≤ u + 1 ≤ L ≤ min ( λ 2 + 1 , μ 2 ) .
Thus we obtain ( 1 , u ) ∈ R A (1,u)\in R_A ( 1 , u ) ∈ R A and ( 1 , u + 1 ) ∈ R B (1,u+1)\in R_B ( 1 , u + 1 ) ∈ R B by (2.3) and (2.4), respectively.
For ℓ ≤ u < L \ell\leq u<L ℓ ≤ u < L , the rows ( 1 , u ) (1,u) ( 1 , u ) of A A A and ( 1 , u + 1 ) (1,u+1) ( 1 , u + 1 ) of B B B give the following equations
( λ 1 + 1 − u ) c u + ( μ 2 − u ) c u + 1 = 0 , (5) (\lambda_1+1-u)c_u+(\mu_2-u)c_{u+1}=0, \tag*{(5)} ( λ 1 + 1 − u ) c u + ( μ 2 − u ) c u + 1 = 0 , (5)
( u + 1 ) c u + ( λ 2 − u ) c u + 1 = 0. (6) (u+1)c_u+(\lambda_2-u)c_{u+1}=0. \tag*{(6)} ( u + 1 ) c u + ( λ 2 − u ) c u + 1 = 0. (6)
Adding these and using that the characteristic is 2, we have
λ 1 c u + ( λ 2 + μ 2 ) c u + 1 = 0. (7) \lambda_1c_u+(\lambda_2+\mu_2)c_{u+1}=0. \tag*{(7)} λ 1 c u + ( λ 2 + μ 2 ) c u + 1 = 0. (7)
In the next lemma, we consider all possibilities modulo 2 of the triple ( λ 1 , λ 2 , μ 2 ) (\lambda_1,\lambda_2,\mu_2) ( λ 1 , λ 2 , μ 2 ) .
Lemma 3.1. Suppose ℓ < L \ell< L ℓ < L and A c = B c = 0 Ac = Bc = 0 A c = B c = 0 , where c = ( c ℓ , … , c L ) c = (c_{\ell}, \ldots, c_{L}) c = ( c ℓ , … , c L ) . Then c c c satisfies the following conditions, according to the parity of ( λ 1 , λ 2 , μ 2 ) (\lambda_{1}, \lambda_{2}, \mu_{2}) ( λ 1 , λ 2 , μ 2 ) .
Thus, dim K Hom G ( Δ ( λ ) , Δ ( μ ) ) ≤ 1 \dim_{K}\operatorname{Hom}_{G}(\Delta(\lambda),\Delta(\mu)) \le1 dim K Hom G ( Δ ( λ ) , Δ ( μ )) ≤ 1 in all cases except possibly when ( λ 1 , λ 2 , μ 2 ) ≡ ( 0 , 0 , 0 ) (\lambda_{1},\lambda_{2},\mu_{2}) \equiv(0,0,0) ( λ 1 , λ 2 , μ 2 ) ≡ ( 0 , 0 , 0 ) or ( 0 , 1 , 1 ) (0,1,1) ( 0 , 1 , 1 ) mod 2.
Proof. For the triples ( 1 , 0 , 0 ) (1,0,0) ( 1 , 0 , 0 ) and ( 1 , 1 , 1 ) (1,1,1) ( 1 , 1 , 1 ) in the table, we have c = 0 c = 0 c = 0 . Indeed, (7) gives c u = 0 c_{u} = 0 c u = 0 for ℓ ≤ u < L \ell\le u < L ℓ ≤ u < L . Since L = min ( λ 2 , μ 2 ) L = \min(\lambda_{2},\mu_{2}) L = min ( λ 2 , μ 2 ) and λ 2 ≡ μ 2 ( m o d 2 ) \lambda_{2} \equiv\mu_{2} \pmod{2} λ 2 ≡ μ 2 ( mod 2 ) , we have λ 2 − L + 1 ≡ 1 ( m o d 2 ) \lambda_{2} - L + 1 \equiv1 \pmod{2} λ 2 − L + 1 ≡ 1 ( mod 2 ) . Thus eq. (6) , with u = L − 1 u = L - 1 u = L − 1 , gives c L = 0 c_{L} = 0 c L = 0 .
For the triple ( 1 , 0 , 1 ) (1,0,1) ( 1 , 0 , 1 ) , eq. (7) gives c u = c u + 1 c_{u} = c_{u+1} c u = c u + 1 for ℓ ≤ u < L \ell\le u < L ℓ ≤ u < L . Substituting in eq. (5) gives c u = 0 c_{u} = 0 c u = 0 for ℓ ≤ u < L \ell\le u < L ℓ ≤ u < L , and hence c = 0 c = 0 c = 0 .
For the triple ( 0 , 0 , 0 ) (0,0,0) ( 0 , 0 , 0 ) , if L L L is even, equation (5) with u = L − 1 u = L - 1 u = L − 1 gives c L = 0 c_{L} = 0 c L = 0 .
The remaining assertions in the table follow from (3.1)–(3.3).
The final conclusion of the lemma follows from Proposition 2.4 . □
Halving procedure
According to the previous lemma, we have dim K Hom G ( Δ ( λ ) , Δ ( μ ) ) ≤ 1 \dim_{K}\operatorname{Hom}_{G}(\Delta(\lambda),\Delta(\mu)) \le1 dim K Hom G ( Δ ( λ ) , Δ ( μ )) ≤ 1 in all cases except possibly the cases ( 0 , 0 , 0 ) (0,0,0) ( 0 , 0 , 0 ) and ( 0 , 1 , 1 ) (0,1,1) ( 0 , 1 , 1 ) for ( λ 1 , λ 2 , μ 2 ) (\lambda_{1},\lambda_{2},\mu_{2}) ( λ 1 , λ 2 , μ 2 ) mod 2. For these we will apply a recursion that relates Hom G ( Δ ( λ ) , Δ ( μ ) ) \operatorname{Hom}_{G}(\Delta(\lambda),\Delta(\mu)) Hom G ( Δ ( λ ) , Δ ( μ )) to Hom G ( Δ ( λ ′ ) , Δ ( μ ′ ) ) \operatorname{Hom}_{G}(\Delta(\lambda'),\Delta(\mu')) Hom G ( Δ ( λ ′ ) , Δ ( μ ′ )) for certain smaller partitions λ ′ , μ ′ \lambda',\mu' λ ′ , μ ′ .
Let us give a roadmap of the present subsection. For each pair of partitions ( λ , μ ) (\lambda,\mu) ( λ , μ ) that belong to the above cases with ℓ < L \ell< L ℓ < L , we define a new pair of partitions ( λ ′ , μ ′ ) (\lambda',\mu') ( λ ′ , μ ′ ) (Definition 3.2 ). Let A ′ , B ′ A',B' A ′ , B ′ be their corresponding matrices of Definition 2.3 . In Lemma 3.3 we show that ( λ ′ , μ ′ ) (\lambda',\mu') ( λ ′ , μ ′ ) again satisfies (2.1), that ℓ ′ ≤ L ′ \ell' \le L' ℓ ′ ≤ L ′ and that λ 3 ′ < λ 3 \lambda'_{3} < \lambda_{3} λ 3 ′ < λ 3 . In Lemma 3.4 we relate certain row indices of the matrices A , B A,B A , B with those of A ′ , B ′ A',B' A ′ , B ′ , to be used in the proof of Proposition 3.5 . In that proposition we apply column operations to A , B A,B A , B and, using Lucas’ theorem, obtain a pair of matrices A ~ , B ~ \widetilde{A},\widetilde{B} A , B such that A ~ \widetilde{A} A and A ′ A' A ′ have the same nonzero rows (without counting repetitions) and likewise for B ~ \widetilde{B} B and B ′ B' B ′ . Finally, in Proposition 3.7 we use Lemma 3.1 and the matrices A ~ , B ~ \widetilde{A},\widetilde{B} A , B to establish an isomorphism ker K [ A B ] ≅ ker K [ A ′ B ′ ] \ker_{K}\begin{bmatrix} A \\ B \end{bmatrix} \cong\ker_{K}\begin{bmatrix} A' \\ B' \end{bmatrix} ker K [ A B ] ≅ ker K [ A ′ B ′ ] , which by Proposition 2.4 gives the desired isomorphism Hom G ( Δ ( λ ) , Δ ( μ ) ) ≅ Hom G ( Δ ( λ ′ ) , Δ ( μ ′ ) ) \operatorname{Hom}_{G}(\Delta(\lambda),\Delta(\mu)) \cong\operatorname{Hom}_{G}(\Delta(\lambda'),\Delta(\mu')) Hom G ( Δ ( λ ) , Δ ( μ )) ≅ Hom G ( Δ ( λ ′ ) , Δ ( μ ′ )) .
For a rational number a a a , let ⌊ a ⌋ \lfloor a\rfloor ⌊ a ⌋ be the largest integer m m m such that m ≤ a m \le a m ≤ a .
Definition 3.2. Suppose ℓ < L \ell< L ℓ < L , λ 1 \lambda_{1} λ 1 is even, and λ 2 ≡ μ 2 ≡ ε ( m o d 2 ) \lambda_{2} \equiv\mu_{2} \equiv\varepsilon\pmod{2} λ 2 ≡ μ 2 ≡ ε ( mod 2 ) , where ε ∈ { 0 , 1 } \varepsilon\in\{0,1\} ε ∈ { 0 , 1 } . Define
λ 1 ′ = λ 1 / 2 − 1 , λ 2 ′ = ⌊ ( λ 2 − 1 ) / 2 ⌋ , λ 3 ′ = ⌊ ( λ 3 − 1 + ε ) / 2 ⌋ , \lambda'_{1} = \lambda_{1}/2 - 1,\qquad
\lambda'_{2} = \left\lfloor(\lambda_{2}-1)/2\right\rfloor,\qquad
\lambda'_{3} = \left\lfloor(\lambda_{3}-1+\varepsilon)/2\right\rfloor, λ 1 ′ = λ 1 /2 − 1 , λ 2 ′ = ⌊ ( λ 2 − 1 ) /2 ⌋ , λ 3 ′ = ⌊ ( λ 3 − 1 + ε ) /2 ⌋ , μ 2 ′ = ⌊ ( μ 2 − 1 ) / 2 ⌋ , μ 1 ′ = λ 1 ′ + λ 2 ′ + λ 3 ′ − μ 2 ′ . \mu'_{2} = \left\lfloor(\mu_{2}-1)/2\right\rfloor,\qquad
\mu'_{1} = \lambda'_{1}+\lambda'_{2}+\lambda'_{3}-\mu'_{2}. μ 2 ′ = ⌊ ( μ 2 − 1 ) /2 ⌋ , μ 1 ′ = λ 1 ′ + λ 2 ′ + λ 3 ′ − μ 2 ′ . We say that λ ′ = ( λ 1 ′ , λ 2 ′ , λ 3 ′ ) \lambda' = (\lambda'_{1},\lambda'_{2},\lambda'_{3}) λ ′ = ( λ 1 ′ , λ 2 ′ , λ 3 ′ ) and μ ′ = ( μ 1 ′ , μ 2 ′ ) \mu' = (\mu'_{1},\mu'_{2}) μ ′ = ( μ 1 ′ , μ 2 ′ ) are obtained from λ , μ \lambda,\mu λ , μ by halving .
Also define ℓ ′ = max ( 0 , μ 2 ′ − λ 3 ′ ) \ell' = \max(0,\mu'_{2}-\lambda'_{3}) ℓ ′ = max ( 0 , μ 2 ′ − λ 3 ′ ) and L ′ = min ( λ 2 ′ , μ 2 ′ ) L' = \min(\lambda'_{2},\mu'_{2}) L ′ = min ( λ 2 ′ , μ 2 ′ ) .
To maintain the flow of the argument, proofs of Lemma 3.3 and Lemma 3.4 below are deferred to § 5 .
Lemma 3.3. Under the assumptions of Definition 3.2 , the following hold.
(1) λ ′ \lambda' λ ′ and μ ′ \mu' μ ′ are partitions with ∣ λ ′ ∣ = ∣ μ ′ ∣ |\lambda'| = |\mu'| ∣ λ ′ ∣ = ∣ μ ′ ∣ . Also 0 ≤ μ 2 ′ ≤ λ 1 ′ 0 \le\mu'_{2} \le\lambda'_{1} 0 ≤ μ 2 ′ ≤ λ 1 ′ and 0 ≤ λ 3 ′ < λ 3 0 \le\lambda'_{3} < \lambda_{3} 0 ≤ λ 3 ′ < λ 3 .
(2) ℓ ′ = ⌊ ℓ / 2 ⌋ \ell' = \lfloor\ell/2\rfloor ℓ ′ = ⌊ ℓ /2 ⌋ and L ′ = ⌊ ( L − 1 ) / 2 ⌋ L' = \lfloor(L-1)/2\rfloor L ′ = ⌊( L − 1 ) /2 ⌋ . In particular, ℓ ′ ≤ L ′ \ell' \le L' ℓ ′ ≤ L ′ .
(3) We have ℓ ≤ 2 i + 1 ≤ L \ell\le2i+1 \le L ℓ ≤ 2 i + 1 ≤ L if and only if ℓ ′ ≤ i ≤ L ′ \ell' \le i \le L' ℓ ′ ≤ i ≤ L ′ .
For the partitions λ ′ , μ ′ \lambda^{\prime},\mu^{\prime} λ ′ , μ ′ of Lemma 3.3 , let A ′ A^{\prime} A ′ and B ′ B^{\prime} B ′ be the corresponding matrices of Definition 2.3 and let R A ′ R_{A^{\prime}} R A ′ and R B ′ R_{B^{\prime}} R B ′ be the corresponding sets of row indices given by (3) and (4) .
Lemma 3.4. Under the assumptions of Definition 3.2 , the following hold. Let h ≥ 1 h \ge1 h ≥ 1 and j ≥ 0 j \ge0 j ≥ 0 .
(1) ( 2 h , 2 j + 1 ) ∈ R A ⟺ ( h , j ) ∈ R A ′ (2h,2j+1) \in R_A \Longleftrightarrow(h,j) \in R_{A^{\prime}} ( 2 h , 2 j + 1 ) ∈ R A ⟺ ( h , j ) ∈ R A ′ , and ( 2 h , 2 j + 1 ) ∈ R B ⟺ ( h , j ) ∈ R B ′ (2h,2j+1) \in R_B \Longleftrightarrow(h,j) \in R_{B^{\prime}} ( 2 h , 2 j + 1 ) ∈ R B ⟺ ( h , j ) ∈ R B ′ .
(2) Let ε = 1 \varepsilon=1 ε = 1 . Then
( 2 h , 2 j ) ∈ R A ⇒ ( 2 h , 2 j + 1 ) ∈ R A , (2h,2j) \in R_A \Rightarrow(2h,2j+1) \in R_A, ( 2 h , 2 j ) ∈ R A ⇒ ( 2 h , 2 j + 1 ) ∈ R A , ( 2 h , 2 j ) ∈ R B ⇒ ( 2 h , 2 j + 1 ) ∈ R B . (2h,2j) \in R_B \Rightarrow(2h,2j+1) \in R_B. ( 2 h , 2 j ) ∈ R B ⇒ ( 2 h , 2 j + 1 ) ∈ R B . In particular, if ( 2 h , 2 j ) ∈ R A (2h,2j) \in R_A ( 2 h , 2 j ) ∈ R A , then ( h , j ) ∈ R A ′ (h,j) \in R_{A^{\prime}} ( h , j ) ∈ R A ′ ; if ( 2 h , 2 j ) ∈ R B (2h,2j) \in R_B ( 2 h , 2 j ) ∈ R B , then ( h , j ) ∈ R B ′ (h,j) \in R_{B^{\prime}} ( h , j ) ∈ R B ′ .
Proposition 3.5. Under the assumptions of Definition 3.2 , perform the following operations on each A , B A,B A , B .
(1) If ε = 0 \varepsilon=0 ε = 0 , delete all columns with even index.
(2) If ε = 1 \varepsilon=1 ε = 1 , add every column with even index to the column on the right, then delete the columns with even indices. An odd-indexed column is left unchanged if the preceding even-indexed column is absent.
Reindex the remaining columns so that the column originally indexed by 2 i + 1 2i+1 2 i + 1 is now indexed by i i i . Denote the resulting matrices by A ~ , B ~ \widetilde{A},\widetilde{B} A , B . Then every row of A ~ \widetilde{A} A is zero or a row of A ′ A^{\prime} A ′ , and every row of A ′ A^{\prime} A ′ occurs in A ~ \widetilde{A} A . The same holds for B ~ \widetilde{B} B and B ′ B^{\prime} B ′ .
Proof. (1) Suppose ε = 0 \varepsilon=0 ε = 0 .
Case 1A. We show that every nonzero row of A ~ \widetilde{A} A is a row of A ′ A^{\prime} A ′ . Consider such a row, indexed by ( t , k ) ∈ R A (t,k) \in R_A ( t , k ) ∈ R A . Throughout the proof, i i i ranges over all integers with ℓ ′ ≤ i ≤ L ′ \ell^{\prime} \le i \le L^{\prime} ℓ ′ ≤ i ≤ L ′ , which index the columns of A ~ \widetilde{A} A and B ~ \widetilde{B} B by Lemma 3.3 (3). Recall from Definition 3.2 that
λ 1 = 2 λ 1 ′ + 2 , λ 2 = 2 λ 2 ′ + 2 , μ 2 = 2 μ 2 ′ + 2. (8) \lambda_1 = 2\lambda_1^{\prime} + 2,\qquad\lambda_2 = 2\lambda_2^{\prime} + 2,\qquad\mu_2 = 2\mu_2^{\prime} + 2. \tag*{(8)} λ 1 = 2 λ 1 ′ + 2 , λ 2 = 2 λ 2 ′ + 2 , μ 2 = 2 μ 2 ′ + 2. (8)
From the definition of A ~ \widetilde{A} A and Definition 2.3 we have
A ~ ( t , k ) , i = A ( t , k ) , 2 i + 1 = ( μ 2 − k 2 i + 1 − k ) ( λ 1 + t − 2 i − 1 t + k − 2 i − 1 ) . \widetilde{A}_{(t,k),i}=A_{(t,k),2i+1}=\binom{\mu_2-k}{2i+1-k}\binom{\lambda_1+t-2i-1}{t+k-2i-1}. A ( t , k ) , i = A ( t , k ) , 2 i + 1 = ( 2 i + 1 − k μ 2 − k ) ( t + k − 2 i − 1 λ 1 + t − 2 i − 1 ) .
If k k k is even, Lemma 2.5 gives ( μ 2 − k 2 i + 1 − k ) = 0 \binom{\mu_2-k}{2i+1-k}=0 ( 2 i + 1 − k μ 2 − k ) = 0 , so the row is zero. Hence k k k is odd. If t t t is odd, Lemma 2.5 gives ( λ 1 + t − 2 i − 1 t + k − 2 i − 1 ) = 0 \binom{\lambda_1+t-2i-1}{t+k-2i-1}=0 ( t + k − 2 i − 1 λ 1 + t − 2 i − 1 ) = 0 , so the row is again zero. So we may assume t = 2 h t=2h t = 2 h and k = 2 j + 1 k=2j+1 k = 2 j + 1 , where h ≥ 1 h \ge1 h ≥ 1 and j ≥ 0 j \ge0 j ≥ 0 .
Since ( 2 h , 2 j + 1 ) ∈ R A (2h,2j+1) \in R_A ( 2 h , 2 j + 1 ) ∈ R A , from Lemma 3.4 (1) we get ( h , j ) ∈ R A ′ (h,j) \in R_{A^{\prime}} ( h , j ) ∈ R A ′ . By Lemma 2.5 and (8) ,
A ~ ( 2 h , 2 j + 1 ) , i = ( μ 2 − 2 j − 1 2 i − 2 j ) ( λ 1 + 2 h − 2 i − 1 2 h + 2 j − 2 i ) = ( μ 2 ′ − j i − j ) ( λ 1 ′ + h − i h + j − i ) = A ( h , j ) , i ′ . (9) \widetilde{A}_{(2h,2j+1),i}=\binom{\mu_2-2j-1}{2i-2j}\binom{\lambda_1+2h-2i-1}{2h+2j-2i}=\binom{\mu_2^{\prime}-j}{i-j}\binom{\lambda_1^{\prime}+h-i}{h+j-i}=A^{\prime}_{(h,j),i}. \tag*{(9)} A ( 2 h , 2 j + 1 ) , i = ( 2 i − 2 j μ 2 − 2 j − 1 ) ( 2 h + 2 j − 2 i λ 1 + 2 h − 2 i − 1 ) = ( i − j μ 2 ′ − j ) ( h + j − i λ 1 ′ + h − i ) = A ( h , j ) , i ′ . (9)
This proves Case 1A.
Case 1B. We show that every nonzero row of B ~ \widetilde{B} B is a row of B ′ B^{\prime} B ′ . Consider such a row, indexed by ( t , k ) ∈ R B (t,k) \in R_B ( t , k ) ∈ R B . We have
B ~ ( t , k ) , i = B ( t , k ) , 2 i + 1 = ( k 2 i + 1 ) ( λ 2 + t − k t + 2 i + 1 − k ) . \widetilde{B}_{(t,k),i}=B_{(t,k),2i+1}=\binom{k}{2i+1}\binom{\lambda_2+t-k}{t+2i+1-k}. B ( t , k ) , i = B ( t , k ) , 2 i + 1 = ( 2 i + 1 k ) ( t + 2 i + 1 − k λ 2 + t − k ) .
If k k k is even, Lemma 2.5 gives ( k 2 i + 1 ) = 0 \binom{k}{2i+1}=0 ( 2 i + 1 k ) = 0 , so the row is zero. Hence k k k is odd. If t t t is odd, Lemma 2.5 gives ( λ 2 + t − k t + 2 i + 1 − k ) = 0 \binom{\lambda_2+t-k}{t+2i+1-k}=0 ( t + 2 i + 1 − k λ 2 + t − k ) = 0 , so the row is again zero. Thus t = 2 h t=2h t = 2 h and k = 2 j + 1 k=2j+1 k = 2 j + 1 , where h ≥ 1 h \ge1 h ≥ 1 and j ≥ 0 j \ge0 j ≥ 0 .
Since ( 2 h , 2 j + 1 ) ∈ R B (2h,2j+1) \in R_B ( 2 h , 2 j + 1 ) ∈ R B from Lemma 3.4 (1) we get ( h , j ) ∈ R B ′ (h,j) \in R_{B^{\prime}} ( h , j ) ∈ R B ′ . Again from Lemma 2.5 we get
B ~ ( 2 h , 2 j + 1 ) , i = ( 2 j + 1 2 i + 1 ) ( λ 2 + 2 h − 2 j − 1 2 h + 2 i − 2 j ) = ( j i ) ( λ 2 ′ + h − j h + i − j ) = B ( h , j ) , i ′ . (10) \widetilde{B}_{(2h,2j+1),i}=\binom{2j+1}{2i+1}\binom{\lambda_2+2h-2j-1}{2h+2i-2j}=\binom{j}{i}\binom{\lambda_2^{\prime}+h-j}{h+i-j}=B^{\prime}_{(h,j),i}. \tag*{(10)} B ( 2 h , 2 j + 1 ) , i = ( 2 i + 1 2 j + 1 ) ( 2 h + 2 i − 2 j λ 2 + 2 h − 2 j − 1 ) = ( i j ) ( h + i − j λ 2 ′ + h − j ) = B ( h , j ) , i ′ . (10)
This proves Case 1B.
Conclusion of (1). Every row of A ′ A^{\prime} A ′ occurs in A ~ \widetilde{A} A , and every row of B ′ B^{\prime} B ′ occurs in B ~ \widetilde{B} B . Indeed, let ( h , j ) ∈ R A ′ (h,j) \in R_{A^{\prime}} ( h , j ) ∈ R A ′ . By Lemma 3.4 (1) we have ( 2 h , 2 j + 1 ) ∈ R A (2h,2j+1) \in R_A ( 2 h , 2 j + 1 ) ∈ R A , and (9) gives A ~ ( 2 h , 2 j + 1 ) , i = A ( h , j ) , i ′ \widetilde{A}_{(2h,2j+1),i}=A^{\prime}_{(h,j),i} A ( 2 h , 2 j + 1 ) , i = A ( h , j ) , i ′ . If ( h , j ) ∈ R B ′ (h,j) \in R_{B^{\prime}} ( h , j ) ∈ R B ′ , the same lemma gives ( 2 h , 2 j + 1 ) ∈ R B (2h,2j+1) \in R_B ( 2 h , 2 j + 1 ) ∈ R B , and (10) gives B ~ ( 2 h , 2 j + 1 ) , i = B ( h , j ) , i ′ \widetilde{B}_{(2h,2j+1),i}=B^{\prime}_{(h,j),i} B ( 2 h , 2 j + 1 ) , i = B ( h , j ) , i ′ . (2) Suppose ε = 1 \varepsilon= 1 ε = 1 . From Definition 3.2 we have
λ 1 = 2 λ 1 ′ + 2 , λ 2 = 2 λ 2 ′ + 1 , μ 2 = 2 μ 2 ′ + 1. (11) \lambda_1 = 2\lambda'_1 + 2,\qquad\lambda_2 = 2\lambda'_2 + 1,\qquad\mu_2 = 2\mu'_2 + 1. \tag*{(11)} λ 1 = 2 λ 1 ′ + 2 , λ 2 = 2 λ 2 ′ + 1 , μ 2 = 2 μ 2 ′ + 1. (11)
Case 2A. We show that every nonzero row of A ~ \widetilde{A} A is a row of A ′ A' A ′ . Consider such a row, indexed by ( t , k ) ∈ R A (t,k) \in R_A ( t , k ) ∈ R A . Since L = min ( λ 2 , μ 2 ) L = \min(\lambda_2,\mu_2) L = min ( λ 2 , μ 2 ) is odd, every even-indexed column of A A A or B B B has its following odd-indexed column.
If column 2 i 2i 2 i is absent, then 2 i = ℓ − 1 2i = \ell- 1 2 i = ℓ − 1 and ℓ > 0 \ell> 0 ℓ > 0 , since the original column indices are consecutive. By (2) , ℓ = μ 2 − λ 3 \ell= \mu_2 - \lambda_3 ℓ = μ 2 − λ 3 . Since ( t , k ) ∈ R A (t,k) \in R_A ( t , k ) ∈ R A , equation (3) gives k ≥ ℓ k \ge\ell k ≥ ℓ , so
2 i − k = ℓ − 1 − k < 0 , ( μ 2 − k 2 i − k ) = 0. 2i-k = \ell-1-k < 0,\qquad\binom{\mu_2-k}{2i-k}=0. 2 i − k = ℓ − 1 − k < 0 , ( 2 i − k μ 2 − k ) = 0.
Thus, whether or not column 2 i 2i 2 i is present, Definition 2.3 gives
A ~ ( t , k ) , i = ( μ 2 − k 2 i − k ) ( λ 1 + t − 2 i t + k − 2 i ) + ( μ 2 − k 2 i + 1 − k ) ( λ 1 + t − 2 i − 1 t + k − 2 i − 1 ) . (12) \widetilde{A}_{(t,k),i} =
\binom{\mu_2-k}{2i-k}\binom{\lambda_1+t-2i}{t+k-2i}
+
\binom{\mu_2-k}{2i+1-k}\binom{\lambda_1+t-2i-1}{t+k-2i-1}. \tag*{(12)} A ( t , k ) , i = ( 2 i − k μ 2 − k ) ( t + k − 2 i λ 1 + t − 2 i ) + ( 2 i + 1 − k μ 2 − k ) ( t + k − 2 i − 1 λ 1 + t − 2 i − 1 ) . (12)
Write k = 2 j k = 2j k = 2 j or k = 2 j + 1 k = 2j+1 k = 2 j + 1 , where j ≥ 0 j \ge0 j ≥ 0 is an integer. If k = 2 j k = 2j k = 2 j , Lemma 2.5 gives
( μ 2 − 2 j 2 i − 2 j ) = ( μ 2 − 2 j 2 i + 1 − 2 j ) = ( μ 2 ′ − j i − j ) . \binom{\mu_2-2j}{2i-2j}
=
\binom{\mu_2-2j}{2i+1-2j}
=
\binom{\mu'_2-j}{i-j}. ( 2 i − 2 j μ 2 − 2 j ) = ( 2 i + 1 − 2 j μ 2 − 2 j ) = ( i − j μ 2 ′ − j ) .
From Pascal’s identity in characteristic 2 and (12) we have
A ~ ( t , 2 j ) , i = ( μ 2 ′ − j i − j ) [ ( λ 1 + t − 2 i t + 2 j − 2 i ) + ( λ 1 + t − 2 i − 1 t + 2 j − 2 i − 1 ) ] = ( μ 2 ′ − j i − j ) ( λ 1 + t − 2 i − 1 t + 2 j − 2 i ) . \widetilde{A}_{(t,2j),i}
=
\binom{\mu'_2-j}{i-j}
\left[
\binom{\lambda_1+t-2i}{t+2j-2i}
+
\binom{\lambda_1+t-2i-1}{t+2j-2i-1}
\right]
=
\binom{\mu'_2-j}{i-j}
\binom{\lambda_1+t-2i-1}{t+2j-2i}. A ( t , 2 j ) , i = ( i − j μ 2 ′ − j ) [ ( t + 2 j − 2 i λ 1 + t − 2 i ) + ( t + 2 j − 2 i − 1 λ 1 + t − 2 i − 1 ) ] = ( i − j μ 2 ′ − j ) ( t + 2 j − 2 i λ 1 + t − 2 i − 1 ) .
If k = 2 j + 1 k = 2j+1 k = 2 j + 1 , Lemma 2.5 and a similar computation give
( μ 2 − 2 j − 1 2 i − 2 j − 1 ) = 0 , ( μ 2 − 2 j − 1 2 i − 2 j ) = ( μ 2 ′ − j i − j ) , A ~ ( t , 2 j + 1 ) , i = ( μ 2 ′ − j i − j ) ( λ 1 + t − 2 i − 1 t + 2 j − 2 i ) . \binom{\mu_2-2j-1}{2i-2j-1}=0,\qquad
\binom{\mu_2-2j-1}{2i-2j}
=
\binom{\mu'_2-j}{i-j},\qquad
\widetilde{A}_{(t,2j+1),i}
=
\binom{\mu'_2-j}{i-j}
\binom{\lambda_1+t-2i-1}{t+2j-2i}. ( 2 i − 2 j − 1 μ 2 − 2 j − 1 ) = 0 , ( 2 i − 2 j μ 2 − 2 j − 1 ) = ( i − j μ 2 ′ − j ) , A ( t , 2 j + 1 ) , i = ( i − j μ 2 ′ − j ) ( t + 2 j − 2 i λ 1 + t − 2 i − 1 ) .
Now we conclude that for both cases of k k k , if t t t is odd, Lemma 2.5 gives ( λ 1 + t − 2 i − 1 t + 2 j − 2 i ) = 0 \binom{\lambda_1+t-2i-1}{t+2j-2i}=0 ( t + 2 j − 2 i λ 1 + t − 2 i − 1 ) = 0 and thus the row is zero. Thus we have t = 2 h t = 2h t = 2 h , where h ≥ 1 h \ge1 h ≥ 1 .
If k ∈ { 2 j , 2 j + 1 } k \in\{2j,2j+1\} k ∈ { 2 j , 2 j + 1 } with ( 2 h , k ) ∈ R A (2h,k) \in R_A ( 2 h , k ) ∈ R A , then we have ( h , j ) ∈ R A ′ (h,j) \in R_{A'} ( h , j ) ∈ R A ′ by Lemma 3.4 (1) when k = 2 j + 1 k = 2j+1 k = 2 j + 1 , and by Lemma 3.4 (2) when k = 2 j k = 2j k = 2 j . Since λ 1 + 2 h − 2 i − 1 = 2 ( λ 1 ′ + h − i ) + 1 \lambda_1 + 2h - 2i - 1 = 2(\lambda'_1+h-i)+1 λ 1 + 2 h − 2 i − 1 = 2 ( λ 1 ′ + h − i ) + 1 is odd, Lemma 2.5 gives
A ~ ( 2 h , k ) , i = ( μ 2 ′ − j i − j ) ( λ 1 + 2 h − 2 i − 1 2 h + 2 j − 2 i ) = ( μ 2 ′ − j i − j ) ( λ 1 ′ + h − i h + j − i ) = A ( h , j ) , i ′ . (13) \widetilde{A}_{(2h,k),i}
=
\binom{\mu'_2-j}{i-j}
\binom{\lambda_1+2h-2i-1}{2h+2j-2i}
=
\binom{\mu'_2-j}{i-j}
\binom{\lambda'_1+h-i}{h+j-i}
=
A'_{(h,j),i}. \tag*{(13)} A ( 2 h , k ) , i = ( i − j μ 2 ′ − j ) ( 2 h + 2 j − 2 i λ 1 + 2 h − 2 i − 1 ) = ( i − j μ 2 ′ − j ) ( h + j − i λ 1 ′ + h − i ) = A ( h , j ) , i ′ . (13)
and we have proven the claim of Case 2A.
Case 2B. We show that every nonzero row of B ~ \widetilde{B} B is a row of B ′ B' B ′ . Consider such a row, indexed by ( t , k ) ∈ R B (t,k) \in R_B ( t , k ) ∈ R B , and write k = 2 j k = 2j k = 2 j or k = 2 j + 1 k = 2j+1 k = 2 j + 1 , where j ≥ 0 j \ge0 j ≥ 0 is an integer. If column 2 i 2i 2 i is absent, then 2 i = ℓ − 1 2i = \ell- 1 2 i = ℓ − 1 and ℓ = μ 2 − λ 3 > 0 \ell= \mu_2-\lambda_3 > 0 ℓ = μ 2 − λ 3 > 0 , as above. Equation (4) gives k ≥ ℓ + t k \ge\ell+t k ≥ ℓ + t , so
t + 2 i − k = t + ℓ − 1 − k < 0 , ( λ 2 + t − k t + 2 i − k ) = 0. t+2i-k = t+\ell-1-k < 0,\qquad\binom{\lambda_2+t-k}{t+2i-k}=0. t + 2 i − k = t + ℓ − 1 − k < 0 , ( t + 2 i − k λ 2 + t − k ) = 0.
Thus Definition 2.3 gives, whether or not column 2 i 2i 2 i is present,
B ~ ( t , k ) , i = ( k 2 i ) ( λ 2 + t − k t + 2 i − k ) + ( k 2 i + 1 ) ( λ 2 + t − k t + 2 i + 1 − k ) . (14) \widetilde{B}_{(t,k),i}
=
\binom{k}{2i}\binom{\lambda_2+t-k}{t+2i-k}
+
\binom{k}{2i+1}\binom{\lambda_2+t-k}{t+2i+1-k}. \tag*{(14)} B ( t , k ) , i = ( 2 i k ) ( t + 2 i − k λ 2 + t − k ) + ( 2 i + 1 k ) ( t + 2 i + 1 − k λ 2 + t − k ) . (14)
If k = 2 j k = 2j k = 2 j , Lemma 2.5 gives
( 2 j 2 i + 1 ) = 0 , ( 2 j 2 i ) = ( j i ) , B ~ ( t , 2 j ) , i = ( j i ) ( λ 2 + t − 2 j t + 2 i − 2 j ) . \binom{2j}{2i+1}=0,\qquad
\binom{2j}{2i}=\binom{j}{i},\qquad
\widetilde{B}_{(t,2j),i}
=
\binom{j}{i}\binom{\lambda_2+t-2j}{t+2i-2j}. ( 2 i + 1 2 j ) = 0 , ( 2 i 2 j ) = ( i j ) , B ( t , 2 j ) , i = ( i j ) ( t + 2 i − 2 j λ 2 + t − 2 j ) .
If k = 2 j + 1 k = 2j+1 k = 2 j + 1 , Lemma 2.5 gives ( 2 j + 1 2 i ) = ( 2 j + 1 2 i + 1 ) = ( j i ) \binom{2j+1}{2i}=\binom{2j+1}{2i+1}=\binom{j}{i} ( 2 i 2 j + 1 ) = ( 2 i + 1 2 j + 1 ) = ( i j ) and from Pascal’s identity and (14) we get
B ~ ( t , 2 j + 1 ) , i = ( j i ) [ ( λ 2 + t − 2 j − 1 t + 2 i − 2 j − 1 ) + ( λ 2 + t − 2 j − 1 t + 2 i − 2 j ) ] = ( j i ) ( λ 2 + t − 2 j t + 2 i − 2 j ) . \widetilde{B}_{(t,2j+1),i}
=
\binom{j}{i}
\left[
\binom{\lambda_2+t-2j-1}{t+2i-2j-1}
+
\binom{\lambda_2+t-2j-1}{t+2i-2j}
\right]
=
\binom{j}{i}\binom{\lambda_2+t-2j}{t+2i-2j}. B ( t , 2 j + 1 ) , i = ( i j ) [ ( t + 2 i − 2 j − 1 λ 2 + t − 2 j − 1 ) + ( t + 2 i − 2 j λ 2 + t − 2 j − 1 ) ] = ( i j ) ( t + 2 i − 2 j λ 2 + t − 2 j ) .
For either value of k k k , if t t t is odd, Lemma 2.5 gives ( λ 2 + t − 2 j t + 2 i − 2 j ) = 0 \binom{\lambda_2+t-2j}{t+2i-2j}=0 ( t + 2 i − 2 j λ 2 + t − 2 j ) = 0 , so the row is zero. Thus we have t = 2 h t = 2h t = 2 h , where h ≥ 1 h \ge1 h ≥ 1 . If k ∈ { 2 j , 2 j + 1 } k \in\{2j,2j+1\} k ∈ { 2 j , 2 j + 1 } with ( 2 h , k ) ∈ R B (2h,k) \in R_B ( 2 h , k ) ∈ R B , then we have ( h , j ) ∈ R B ′ (h,j) \in R_{B'} ( h , j ) ∈ R B ′ by Lemma 3.4(1) when k = 2 j + 1 k=2j+1 k = 2 j + 1 , and by Lemma 3.4(2) when k = 2 j k=2j k = 2 j . Since λ 2 + 2 h − 2 j = 2 ( λ 2 ′ + h − j ) + 1 \lambda_2+2h-2j=2(\lambda'_2+h-j)+1 λ 2 + 2 h − 2 j = 2 ( λ 2 ′ + h − j ) + 1 is odd, Lemma 2.5 gives
B ~ ( 2 h , k ) , i = ( j i ) ( λ 2 + 2 h − 2 j 2 h + 2 i − 2 j ) = ( j i ) ( λ 2 ′ + h − j h + i − j ) = B ( h , j ) , i ′ . (15) \widetilde{B}_{(2h,k),i}=\binom{j}{i}\binom{\lambda_2+2h-2j}{2h+2i-2j}=\binom{j}{i}\binom{\lambda'_2+h-j}{h+i-j}=B'_{(h,j),i}. \tag*{(15)} B ( 2 h , k ) , i = ( i j ) ( 2 h + 2 i − 2 j λ 2 + 2 h − 2 j ) = ( i j ) ( h + i − j λ 2 ′ + h − j ) = B ( h , j ) , i ′ . (15)
and we have proven the claim of Case 2B.
Conclusion of (2). Every row of A ′ A' A ′ occurs in A ~ \widetilde{A} A , and every row of B ′ B' B ′ occurs in B ~ \widetilde{B} B . Let ( h , j ) ∈ R A ′ (h,j) \in R_{A'} ( h , j ) ∈ R A ′ . By Lemma 3.4(1) we have ( 2 h , 2 j + 1 ) ∈ R A (2h,2j+1) \in R_A ( 2 h , 2 j + 1 ) ∈ R A , and (3.9) gives A ~ ( 2 h , 2 j + 1 ) , i = A ( h , j ) , i ′ \widetilde{A}_{(2h,2j+1),i}=A'_{(h,j),i} A ( 2 h , 2 j + 1 ) , i = A ( h , j ) , i ′ . If ( h , j ) ∈ R B ′ (h,j) \in R_{B'} ( h , j ) ∈ R B ′ , the same lemma gives ( 2 h , 2 j + 1 ) ∈ R B (2h,2j+1) \in R_B ( 2 h , 2 j + 1 ) ∈ R B , and from (3.11) we get B ~ ( 2 h , 2 j + 1 ) , i = B ( h , j ) , i ′ \widetilde{B}_{(2h,2j+1),i}=B'_{(h,j),i} B ( 2 h , 2 j + 1 ) , i = B ( h , j ) , i ′ .
Example 3.6. Let λ = ( 10 , 7 , 2 ) \lambda=(10,7,2) λ = ( 10 , 7 , 2 ) and μ = ( 14 , 5 ) \mu=(14,5) μ = ( 14 , 5 ) . Then ℓ = 3 \ell=3 ℓ = 3 , L = 5 L=5 L = 5 , and ε = 1 \varepsilon=1 ε = 1 . For the corresponding matrices (Definition 2.3) we have
R A = { ( 1 , 3 ) , ( 1 , 4 ) , ( 1 , 5 ) , ( 2 , 3 ) , ( 2 , 4 ) , ( 2 , 5 ) , ( 3 , 3 ) , ( 3 , 4 ) , ( 4 , 3 ) } , R_A=\{(1,3),(1,4),(1,5),(2,3),(2,4),(2,5),(3,3),(3,4),(4,3)\}, R A = {( 1 , 3 ) , ( 1 , 4 ) , ( 1 , 5 ) , ( 2 , 3 ) , ( 2 , 4 ) , ( 2 , 5 ) , ( 3 , 3 ) , ( 3 , 4 ) , ( 4 , 3 )} ,
R B = { ( 1 , 4 ) , ( 1 , 5 ) , ( 2 , 5 ) } . R_B=\{(1,4),(1,5),(2,5)\}. R B = {( 1 , 4 ) , ( 1 , 5 ) , ( 2 , 5 )} .
The columns are indexed by 3 , 4 , 5 3,4,5 3 , 4 , 5 . Thus
A = ( ( 8 1 ) ( 2 1 ) 0 0 ( 7 1 ) ( 1 1 ) 0 0 ( 6 1 ) ( 9 2 ) ( 2 1 ) ( 8 1 ) ( 2 2 ) 0 ( 8 2 ) ( 1 1 ) ( 7 1 ) 0 0 ( 7 2 ) ( 10 3 ) ( 2 1 ) ( 9 2 ) ( 2 2 ) ( 8 1 ) 0 ( 9 3 ) ( 1 1 ) ( 8 2 ) ( 11 4 ) ( 2 1 ) ( 10 3 ) ( 2 2 ) ( 9 2 ) ) , B = ( ( 4 3 ) ( 4 4 ) ( 4 4 ) 0 0 ( 5 4 ) ( 5 5 ) ( 3 1 ) ( 5 3 ) ( 5 4 ) ( 4 1 ) ( 5 5 ) ( 5 4 ) ( 4 2 ) ) . A=
\begin{pmatrix}
\binom{8}{1} & \binom{2}{1} & 0\\
0 & \binom{7}{1} & \binom{1}{1}\\
0 & 0 & \binom{6}{1}\\
\binom{9}{2} & \binom{2}{1}\binom{8}{1} & \binom{2}{2}\\
0 & \binom{8}{2} & \binom{1}{1}\binom{7}{1}\\
0 & 0 & \binom{7}{2}\\
\binom{10}{3} & \binom{2}{1}\binom{9}{2} & \binom{2}{2}\binom{8}{1}\\
0 & \binom{9}{3} & \binom{1}{1}\binom{8}{2}\\
\binom{11}{4} & \binom{2}{1}\binom{10}{3} & \binom{2}{2}\binom{9}{2}
\end{pmatrix},
\qquad
B=
\begin{pmatrix}
\binom{4}{3} & \binom{4}{4} & \binom{4}{4} & 0\\
0 & \binom{5}{4} & \binom{5}{5} & \binom{3}{1}\\
\binom{5}{3} & \binom{5}{4} & \binom{4}{1} & \binom{5}{5}\\
& & \binom{5}{4} & \binom{4}{2}
\end{pmatrix}. A = ( 1 8 ) 0 0 ( 2 9 ) 0 0 ( 3 10 ) 0 ( 4 11 ) ( 1 2 ) ( 1 7 ) 0 ( 1 2 ) ( 1 8 ) ( 2 8 ) 0 ( 1 2 ) ( 2 9 ) ( 3 9 ) ( 1 2 ) ( 3 10 ) 0 ( 1 1 ) ( 1 6 ) ( 2 2 ) ( 1 1 ) ( 1 7 ) ( 2 7 ) ( 2 2 ) ( 1 8 ) ( 1 1 ) ( 2 8 ) ( 2 2 ) ( 2 9 ) , B = ( 3 4 ) 0 ( 3 5 ) ( 4 4 ) ( 4 5 ) ( 4 5 ) ( 4 4 ) ( 5 5 ) ( 1 4 ) ( 4 5 ) 0 ( 1 3 ) ( 5 5 ) ( 2 4 ) .
According to Proposition 3.5(2), in each matrix, replace the column indexed by 5 with the sum of the columns indexed by 4 and 5, and delete the column indexed by 4. Reindexing the two remaining columns by 3 ↦ 1 3 \mapsto1 3 ↦ 1 and 5 ↦ 2 5 \mapsto2 5 ↦ 2 gives
A ~ = ( 0 0 0 0 0 0 0 1 0 1 0 1 0 0 0 0 0 0 ) , B ~ = ( 0 0 0 0 0 0 ) . \widetilde{A}=
\begin{pmatrix}
0&0\\
0&0\\
0&0\\
0&1\\
0&1\\
0&1\\
0&0\\
0&0\\
0&0
\end{pmatrix},
\qquad
\widetilde{B}=
\begin{pmatrix}
0&0\\
0&0\\
0&0
\end{pmatrix}. A = 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 0 0 0 , B = 0 0 0 0 0 0 .
On the other hand, Definition 3.2 gives the partitions λ ′ = ( 4 , 3 , 1 ) \lambda'=(4,3,1) λ ′ = ( 4 , 3 , 1 ) and μ ′ = ( 6 , 2 ) \mu'=(6,2) μ ′ = ( 6 , 2 ) . Their corresponding matrices (Definition 2.3) are
A ′ = ( ( 4 1 ) ( 1 1 ) 0 ( 3 1 ) ( 5 2 ) ( 1 1 ) ( 4 1 ) ) = ( 0 1 0 1 0 0 ) , B ′ = ( ( 2 1 ) ( 2 2 ) ( 2 1 ) ) = ( 0 0 ) . A'=
\begin{pmatrix}
\binom{4}{1} & \binom{1}{1}\\
0 & \binom{3}{1}\\
\binom{5}{2} & \binom{1}{1}\binom{4}{1}
\end{pmatrix}
=
\begin{pmatrix}
0&1\\
0&1\\
0&0
\end{pmatrix},
\qquad
B'=
\left(\binom{2}{1}\quad\binom{2}{2}\ \binom{2}{1}\right)
=(0\quad0). A ′ = ( 1 4 ) 0 ( 2 5 ) ( 1 1 ) ( 1 3 ) ( 1 1 ) ( 1 4 ) = 0 0 0 1 1 0 , B ′ = ( ( 1 2 ) ( 2 2 ) ( 1 2 ) ) = ( 0 0 ) .
By Case 2A and Case 2B in the proof of Proposition 3.5 , row ( 2 , 3 ) (2,3) ( 2 , 3 ) of A ~ \widetilde{A} A corresponds to row ( 1 , 1 ) (1,1) ( 1 , 1 ) of A ′ A' A ′ , rows ( 2 , 4 ) (2,4) ( 2 , 4 ) and ( 2 , 5 ) (2,5) ( 2 , 5 ) correspond to row ( 1 , 2 ) (1,2) ( 1 , 2 ) , and row ( 4 , 3 ) (4,3) ( 4 , 3 ) corresponds to row ( 2 , 1 ) (2,1) ( 2 , 1 ) . Similarly, row ( 2 , 5 ) (2,5) ( 2 , 5 ) of B ~ \widetilde{B} B corresponds to row ( 1 , 2 ) (1,2) ( 1 , 2 ) of B ′ B' B ′ .
Proposition 3.7. Under the assumptions of Definition 3.2 , the map
Φ : ker K [ A B ] ⟶ ker K [ A ′ B ′ ] , ( c u ) ℓ ≤ u ≤ L ↦ ( c 2 i + 1 ) ℓ ′ ≤ i ≤ L ′ \Phi:\ker_{\mathbb{K}}\begin{bmatrix} A \\ B \end{bmatrix}\longrightarrow\ker_{\mathbb{K}}\begin{bmatrix} A' \\ B' \end{bmatrix},\qquad(c_u)_{\ell\leq u\leq L}\mapsto(c_{2i+1})_{\ell'\leq i\leq L'} Φ : ker K [ A B ] ⟶ ker K [ A ′ B ′ ] , ( c u ) ℓ ≤ u ≤ L ↦ ( c 2 i + 1 ) ℓ ′ ≤ i ≤ L ′
is an isomorphism. Its inverse sends c ′ c' c ′ to c = ( c u ) ℓ ≤ u ≤ L c=(c_u)_{\ell\leq u\leq L} c = ( c u ) ℓ ≤ u ≤ L given by
c u = { c ( u − 1 ) / 2 ′ , u odd , 0 , u even and ε = 0 , c u / 2 ′ , u even and ε = 1. (16) c_u=\begin{cases}
c'_{(u-1)/2}, & u\ \text{odd},\\
0, & u\ \text{even and }\varepsilon=0,\\
c'_{u/2}, & u\ \text{even and }\varepsilon=1.
\end{cases} \tag*{(16)} c u = ⎩ ⎨ ⎧ c ( u − 1 ) /2 ′ , 0 , c u /2 ′ , u odd , u even and ε = 0 , u even and ε = 1. (16)
Proof. By Proposition 3.5 we know that ker K [ A ~ B ~ ] = ker K [ A ′ B ′ ] \ker_{\mathbb{K}}\begin{bmatrix}\widetilde{A}\\\widetilde{B}\end{bmatrix}=\ker_{\mathbb{K}}\begin{bmatrix}A'\\B'\end{bmatrix} ker K [ A B ] = ker K [ A ′ B ′ ] .
(1) We show that Φ \Phi Φ is well defined. Let c = ( c u ) ℓ ≤ u ≤ L ∈ ker K [ A B ] c=(c_u)_{\ell\leq u\leq L}\in\ker_{\mathbb{K}}\begin{bmatrix}A\\B\end{bmatrix} c = ( c u ) ℓ ≤ u ≤ L ∈ ker K [ A B ] . For even u u u with ℓ ≤ u ≤ L \ell\leq u\leq L ℓ ≤ u ≤ L , Lemma 3.1 gives c u = 0 c_u=0 c u = 0 if ε = 0 \varepsilon=0 ε = 0 and c u = c u + 1 c_u=c_{u+1} c u = c u + 1 if ε = 1 \varepsilon=1 ε = 1 . Thus from the definition of A ~ \widetilde{A} A and B ~ \widetilde{B} B in Proposition 3.5 we have A ~ Φ ( c ) = A c = 0 \widetilde{A}\Phi(c)=Ac=0 A Φ ( c ) = A c = 0 , B ~ Φ ( c ) = B c = 0 \widetilde{B}\Phi(c)=Bc=0 B Φ ( c ) = B c = 0 . Hence im Φ ⊂ ker K [ A ′ B ′ ] \operatorname{im}\Phi\subset\ker_{\mathbb{K}}\begin{bmatrix}A'\\B'\end{bmatrix} im Φ ⊂ ker K [ A ′ B ′ ] .
(2) We show that Φ \Phi Φ is injective. Suppose Φ ( c ) = 0 \Phi(c)=0 Φ ( c ) = 0 . By Lemma 3.3 (3), we have c u = 0 c_u=0 c u = 0 for every odd u u u with ℓ ≤ u ≤ L \ell\leq u\leq L ℓ ≤ u ≤ L . Lemma 3.1 (first and third rows of the table) then gives c = 0 c=0 c = 0 .
(3) We show that Φ \Phi Φ is surjective. Let c ′ ∈ ker K [ A ′ B ′ ] c'\in\ker_{\mathbb{K}}\begin{bmatrix}A'\\B'\end{bmatrix} c ′ ∈ ker K [ A ′ B ′ ] . For ℓ ≤ u ≤ L \ell\leq u\leq L ℓ ≤ u ≤ L , Lemma 3.3 (3) gives ℓ ′ ≤ ( u − 1 ) / 2 ≤ L ′ \ell'\leq(u-1)/2\leq L' ℓ ′ ≤ ( u − 1 ) /2 ≤ L ′ if u u u is odd. If ε = 1 \varepsilon=1 ε = 1 and u u u is even, then L L L is odd, so ℓ ≤ u + 1 ≤ L \ell\leq u+1\leq L ℓ ≤ u + 1 ≤ L ; the same lemma gives ℓ ′ ≤ u / 2 ≤ L ′ \ell'\leq u/2\leq L' ℓ ′ ≤ u /2 ≤ L ′ . Define c = ( c u ) ℓ ≤ u ≤ L c=(c_u)_{\ell\leq u\leq L} c = ( c u ) ℓ ≤ u ≤ L by (3.12). The column operations in Proposition 3.5 show that A c = A c ~ ′ Ac=A\widetilde{c}' A c = A c ′ and B c = B c ~ Bc=B\widetilde{c} B c = B c . By the same proposition, A c ~ ′ = B c ~ ′ = 0 A\widetilde{c}'=B\widetilde{c}'=0 A c ′ = B c ′ = 0 . Hence c ∈ ker K [ A B ] c\in\ker_{\mathbb{K}}\begin{bmatrix}A\\B\end{bmatrix} c ∈ ker K [ A B ] . Since c 2 i + 1 = c i ′ c_{2i+1}=c'_i c 2 i + 1 = c i ′ by (3.12), we have Φ ( c ) = c ′ \Phi(c)=c' Φ ( c ) = c ′ . □
Remark 3.8. In Definition 3.2 suppose ε = 1 \varepsilon=1 ε = 1 and λ 3 \lambda_3 λ 3 is even. Then an isomorphism Hom G ( Δ ( λ ) , Δ ( μ ) ) ≅ Hom G ( Δ ( λ ′ ) , Δ ( μ ′ ) ) \operatorname{Hom}_G(\Delta(\lambda),\Delta(\mu))\cong\operatorname{Hom}_G(\Delta(\lambda'),\Delta(\mu')) Hom G ( Δ ( λ ) , Δ ( μ )) ≅ Hom G ( Δ ( λ ′ ) , Δ ( μ ′ )) can be obtained using the Steinberg module. Indeed, let ρ = ( 2 , 1 , 0 ) \rho=(2,1,0) ρ = ( 2 , 1 , 0 ) and S t = Δ ( ρ ) \mathrm{St}=\Delta(\rho) St = Δ ( ρ ) be the Steinberg module. Then λ = ρ + 2 λ ′ \lambda=\rho+2\lambda' λ = ρ + 2 λ ′ and μ = ρ + 2 μ ′ \mu=\rho+2\mu' μ = ρ + 2 μ ′ . The Steinberg tensor identity [13 II.3.18, 3.19] gives Δ ( λ ) ≅ S t ⊗ Δ ( λ ′ ) ( 1 ) \Delta(\lambda)\cong\mathrm{St}\otimes\Delta(\lambda')^{(1)} Δ ( λ ) ≅ St ⊗ Δ ( λ ′ ) ( 1 ) , Δ ( μ ) ≅ S t ⊗ Δ ( μ ′ ) ( 1 ) \Delta(\mu)\cong\mathrm{St}\otimes\Delta(\mu')^{(1)} Δ ( μ ) ≅ St ⊗ Δ ( μ ′ ) ( 1 ) , where M ( 1 ) M^{(1)} M ( 1 ) denotes the first Frobenius twist of a G G G -module M M M . The functor M ↦ S t ⊗ M ( 1 ) M\mapsto\mathrm{St}\otimes M^{(1)} M ↦ St ⊗ M ( 1 ) is fully faithful [13 II.10.5(1)] and thus we obtain the isomorphism.
Main result
A recursive nonvanishing criterion.
The following is the main result of the paper.
Theorem 4.1. Let λ , μ \lambda,\mu λ , μ satisfy (2.1) and let ℓ , L \ell,L ℓ , L be as in (2.2). If ℓ > L \ell>L ℓ > L , then Hom G ( Δ ( λ ) , Δ ( μ ) ) = 0 \operatorname{Hom}_G(\Delta(\lambda),\Delta(\mu))=0 Hom G ( Δ ( λ ) , Δ ( μ )) = 0 . If ℓ ≤ L \ell\leq L ℓ ≤ L , then exactly one of the following cases applies.
(1) Suppose ℓ = L \ell=L ℓ = L or λ 1 \lambda_1 λ 1 is odd. Then Hom G ( Δ ( λ ) , Δ ( μ ) ) ≠ 0 \operatorname{Hom}_G(\Delta(\lambda),\Delta(\mu))\ne0 Hom G ( Δ ( λ ) , Δ ( μ )) = 0 if and only if
λ 2 − ℓ < 2 v 2 ( λ 1 − μ 2 + 1 ) , λ 3 < 2 v 2 ( λ 2 + 1 ) . (17) \lambda_2-\ell<2^{v_2(\lambda_1-\mu_2+1)},\qquad\lambda_3<2^{v_2(\lambda_2+1)}. \tag*{(17)} λ 2 − ℓ < 2 v 2 ( λ 1 − μ 2 + 1 ) , λ 3 < 2 v 2 ( λ 2 + 1 ) . (17)
If (4.1) holds, then dim K Hom G ( Δ ( λ ) , Δ ( μ ) ) = 1 \dim_{\mathbb{K}}\operatorname{Hom}_G(\Delta(\lambda),\Delta(\mu))=1 dim K Hom G ( Δ ( λ ) , Δ ( μ )) = 1 and a generator is induced by the map ∑ u = ℓ L ϕ u : D ( λ ) → Δ ( μ ) \sum_{u=\ell}^{L}\phi_u:D(\lambda)\to\Delta(\mu) ∑ u = ℓ L ϕ u : D ( λ ) → Δ ( μ ) .
(2) Suppose ℓ < L \ell<L ℓ < L , λ 1 \lambda_1 λ 1 is even, and λ 2 , μ 2 \lambda_2,\mu_2 λ 2 , μ 2 have opposite parity. Then Hom G ( Δ ( λ ) , Δ ( μ ) ) ≠ 0 \operatorname{Hom}_G(\Delta(\lambda),\Delta(\mu))\ne0 Hom G ( Δ ( λ ) , Δ ( μ )) = 0 if and only if
λ 2 − ℓ < 2 v 2 ( λ 1 − ℓ + 1 ) , λ 3 < 2 v 2 ( ℓ + 1 ) . (18) \lambda_2-\ell<2^{v_2(\lambda_1-\ell+1)},\qquad\lambda_3<2^{v_2(\ell+1)}. \tag*{(18)} λ 2 − ℓ < 2 v 2 ( λ 1 − ℓ + 1 ) , λ 3 < 2 v 2 ( ℓ + 1 ) . (18)
If (4.2) holds, then dim K Hom G ( Δ ( λ ) , Δ ( μ ) ) = 1 \dim_{\mathbb{K}}\operatorname{Hom}_G(\Delta(\lambda),\Delta(\mu))=1 dim K Hom G ( Δ ( λ ) , Δ ( μ )) = 1 and a generator is induced by the map ϕ ℓ : D ( λ ) → Δ ( μ ) \phi_\ell:D(\lambda)\to\Delta(\mu) ϕ ℓ : D ( λ ) → Δ ( μ ) . (3) Suppose ℓ < L \ell< L ℓ < L , λ 1 \lambda_1 λ 1 is even, and λ 2 , μ 2 \lambda_2,\mu_2 λ 2 , μ 2 have the same parity. For the partitions λ ′ , μ ′ \lambda',\mu' λ ′ , μ ′ of Definition 3.2 we have Hom G ( Δ ( λ ) , Δ ( μ ) ) ≅ Hom G ( Δ ( λ ′ ) , Δ ( μ ′ ) ) \operatorname{Hom}_{G}(\Delta(\lambda),\Delta(\mu)) \cong\operatorname{Hom}_{G}(\Delta(\lambda'),\Delta(\mu')) Hom G ( Δ ( λ ) , Δ ( μ )) ≅ Hom G ( Δ ( λ ′ ) , Δ ( μ ′ )) . Moreover, if ∑ i = ℓ ′ L ′ c i ′ ϕ i \sum_{i=\ell'}^{L'}c'_i\phi_i ∑ i = ℓ ′ L ′ c i ′ ϕ i induces a generator of Hom G ( Δ ( λ ′ ) , Δ ( μ ′ ) ) \operatorname{Hom}_{G}(\Delta(\lambda'),\Delta(\mu')) Hom G ( Δ ( λ ′ ) , Δ ( μ ′ )) , where ϕ i \phi_i ϕ i are the maps of Lemma 2.2 for ( λ ′ , μ ′ ) (\lambda',\mu') ( λ ′ , μ ′ ) , then, with c u c_u c u defined by (3.12), ∑ u = ℓ L c u ϕ u \sum_{u=\ell}^{L}c_u\phi_u ∑ u = ℓ L c u ϕ u induces a generator of Hom G ( Δ ( λ ) , Δ ( μ ) ) \operatorname{Hom}_{G}(\Delta(\lambda),\Delta(\mu)) Hom G ( Δ ( λ ) , Δ ( μ )) .
Repeated application of (3) terminates in (1) or (2). Thus, dim K Hom G ( Δ ( λ ) , Δ ( μ ) ) ≤ 1 \dim_{\mathbb{K}}\operatorname{Hom}_{G}(\Delta(\lambda),\Delta(\mu))\leq1 dim K Hom G ( Δ ( λ ) , Δ ( μ )) ≤ 1 .
Proof. By Lemma 2.2, we have Hom G ( Δ ( λ ) , Δ ( μ ) ) = 0 \operatorname{Hom}_{G}(\Delta(\lambda),\Delta(\mu))=0 Hom G ( Δ ( λ ) , Δ ( μ )) = 0 if ℓ > L \ell>L ℓ > L . Let ℓ ≤ L \ell\leq L ℓ ≤ L .
(1) Suppose ℓ = L \ell=L ℓ = L or λ 1 \lambda_1 λ 1 is odd. If ℓ = L \ell=L ℓ = L , the matrix [ A B ] \begin{bmatrix} A \\ B \end{bmatrix} [ A B ] has one column. If ℓ < L \ell<L ℓ < L , then λ 1 \lambda_1 λ 1 is odd, and Lemma 3.1 shows that every vector in its kernel has equal coordinates (possibly all zero). Thus in both cases the kernel is contained in the subspace of K L − ℓ + 1 \mathbb{K}^{L-\ell+1} K L − ℓ + 1 spanned by ( 1 , … , 1 ) T (1,\ldots,1)^{T} ( 1 , … , 1 ) T . By Proposition 2.4, Hom G ( Δ ( λ ) , Δ ( μ ) ) \operatorname{Hom}_{G}(\Delta(\lambda),\Delta(\mu)) Hom G ( Δ ( λ ) , Δ ( μ )) is nonzero if and only if every row of A A A and B B B has sum zero. For ( t , k ) ∈ R A (t,k)\in R_A ( t , k ) ∈ R A , we compute
∑ u = ℓ L A ( t , k ) , u = ∑ u = ℓ L ( − 1 ) u − k ( μ 2 − k u − k ) ( λ 1 + t − u t + k − u ) = ∑ j = 0 t ( − 1 ) j ( μ 2 − k j ) ( λ 1 + t − k − j t − j ) = ( λ 1 − μ 2 + t t ) . (19) \sum_{u=\ell}^{L}A_{(t,k),u}=\sum_{u=\ell}^{L}(-1)^{u-k}\binom{\mu_2-k}{u-k}\binom{\lambda_1+t-u}{t+k-u}=\sum_{j=0}^{t}(-1)^j\binom{\mu_2-k}{j}\binom{\lambda_1+t-k-j}{t-j}=\binom{\lambda_1-\mu_2+t}{t}. \tag*{(19)} u = ℓ ∑ L A ( t , k ) , u = u = ℓ ∑ L ( − 1 ) u − k ( u − k μ 2 − k ) ( t + k − u λ 1 + t − u ) = j = 0 ∑ t ( − 1 ) j ( j μ 2 − k ) ( t − j λ 1 + t − k − j ) = ( t λ 1 − μ 2 + t ) . (19)
In (19) we used the change of variable j = u − k j=u-k j = u − k . So ℓ − k ≤ j ≤ L − k \ell-k\leq j\leq L-k ℓ − k ≤ j ≤ L − k . For the new end points, by (2.3), we have ℓ ≤ k \ell\leq k ℓ ≤ k and t ≤ λ 2 − k t\leq\lambda_2-k t ≤ λ 2 − k . Hence
ℓ − k ≤ j ≤ L − k , 0 ≤ j ≤ t ⟺ 0 ≤ j ≤ min ( t , L − k ) ⟺ 0 ≤ j ≤ min ( t , μ 2 − k ) , \ell-k\leq j\leq L-k,\quad0\leq j\leq t\quad\Longleftrightarrow\quad0\leq j\leq\min(t,L-k)\quad\Longleftrightarrow\quad0\leq j\leq\min(t,\mu_2-k), ℓ − k ≤ j ≤ L − k , 0 ≤ j ≤ t ⟺ 0 ≤ j ≤ min ( t , L − k ) ⟺ 0 ≤ j ≤ min ( t , μ 2 − k ) ,
where in the last equivalence we used L = min ( λ 2 , μ 2 ) L=\min(\lambda_2,\mu_2) L = min ( λ 2 , μ 2 ) . Since the summand ( − 1 ) j ( μ 2 − k j ) ( λ 1 + t − k − j t − j ) (-1)^j\binom{\mu_2-k}{j}\binom{\lambda_1+t-k-j}{t-j} ( − 1 ) j ( j μ 2 − k ) ( t − j λ 1 + t − k − j ) vanishes for j < 0 j<0 j < 0 , j > t j>t j > t , or j > μ 2 − k j>\mu_2-k j > μ 2 − k we may sum over 0 ≤ j ≤ t 0\leq j\leq t 0 ≤ j ≤ t . The last equality in (4.3) uses [17 Lemma 4.1(2)] ` with a = λ 1 − k a=\lambda_1-k a = λ 1 − k , b = μ 2 − k b=\mu_2-k b = μ 2 − k , and c = t c=t c = t .
For ( t , k ) ∈ R B (t,k)\in R_B ( t , k ) ∈ R B , a similar computation yields
∑ u = ℓ L B ( t , k ) , u = ∑ u = ℓ L ( k u ) ( λ 2 + t − k t + u − k ) = ∑ j = 0 t ( k j ) ( λ 2 + t − k t − j ) = ( λ 2 + t t ) . (20) \sum_{u=\ell}^{L}B_{(t,k),u}=\sum_{u=\ell}^{L}\binom{k}{u}\binom{\lambda_2+t-k}{t+u-k}=\sum_{j=0}^{t}\binom{k}{j}\binom{\lambda_2+t-k}{t-j}=\binom{\lambda_2+t}{t}. \tag*{(20)} u = ℓ ∑ L B ( t , k ) , u = u = ℓ ∑ L ( u k ) ( t + u − k λ 2 + t − k ) = j = 0 ∑ t ( j k ) ( t − j λ 2 + t − k ) = ( t λ 2 + t ) . (20)
Indeed, we used j = k − u j=k-u j = k − u , so k − L ≤ j ≤ k − ℓ k-L\leq j\leq k-\ell k − L ≤ j ≤ k − ℓ . By (2.4), we have k ≤ μ 2 k\leq\mu_2 k ≤ μ 2 and k − ℓ ≥ min ( k , t ) k-\ell\geq\min(k,t) k − ℓ ≥ min ( k , t ) . Hence
k − L ≤ j ≤ k − ℓ , 0 ≤ j ≤ min ( k , t ) ⟺ max ( 0 , k − L ) ≤ j ≤ min ( k , t ) k-L\leq j\leq k-\ell,\quad0\leq j\leq\min(k,t)\quad\Longleftrightarrow\quad\max(0,k-L)\leq j\leq\min(k,t) k − L ≤ j ≤ k − ℓ , 0 ≤ j ≤ min ( k , t ) ⟺ max ( 0 , k − L ) ≤ j ≤ min ( k , t )
⟺ max ( 0 , k − λ 2 ) ≤ j ≤ min ( k , t ) , \Longleftrightarrow\quad\max(0,k-\lambda_2)\leq j\leq\min(k,t), ⟺ max ( 0 , k − λ 2 ) ≤ j ≤ min ( k , t ) ,
where the last equivalence uses L = min ( λ 2 , μ 2 ) L=\min(\lambda_2,\mu_2) L = min ( λ 2 , μ 2 ) and k ≤ μ 2 k\leq\mu_2 k ≤ μ 2 . The summand ( k j ) ( λ 2 + t − k t − j ) \binom{k}{j}\binom{\lambda_2+t-k}{t-j} ( j k ) ( t − j λ 2 + t − k ) vanishes for j < 0 j<0 j < 0 , j > k j>k j > k , j > t j>t j > t , or j < k − λ 2 j<k-\lambda_2 j < k − λ 2 . Thus we may sum over 0 ≤ j ≤ t 0\leq j\leq t 0 ≤ j ≤ t . The last equality in (20) is Vandermonde’s identity.
By (2.3) and (2.4), the ranges of t t t are 1 ≤ t ≤ λ 2 − ℓ 1\leq t\leq\lambda_2-\ell 1 ≤ t ≤ λ 2 − ℓ and 1 ≤ t ≤ λ 3 1\leq t\leq\lambda_3 1 ≤ t ≤ λ 3 , respectively. Thus equations (19) and (20) show that every row of A A A and B B B has sum zero if and only if
( λ 1 − μ 2 + t t ) ≡ 0 ( m o d 2 ) ( 1 ≤ t ≤ λ 2 − ℓ ) , ( λ 2 + t t ) ≡ 0 ( m o d 2 ) ( 1 ≤ t ≤ λ 3 ) . \binom{\lambda_1-\mu_2+t}{t}\equiv0\pmod{2}\quad(1\leq t\leq\lambda_2-\ell),\qquad\binom{\lambda_2+t}{t}\equiv0\pmod{2}\quad(1\leq t\leq\lambda_3). ( t λ 1 − μ 2 + t ) ≡ 0 ( mod 2 ) ( 1 ≤ t ≤ λ 2 − ℓ ) , ( t λ 2 + t ) ≡ 0 ( mod 2 ) ( 1 ≤ t ≤ λ 3 ) .
By Lemma 2.6, these conditions are equivalent to (4.1). When they hold, the vector ( 1 , … , 1 ) T (1,\ldots,1)^{T} ( 1 , … , 1 ) T spans ker K [ A B ] \ker_{\mathbb{K}}\begin{bmatrix} A \\ B \end{bmatrix} ker K [ A B ] . Then Proposition 2.4 shows that the map ∑ u = ℓ L ϕ u : D ( λ ) → Δ ( μ ) \sum_{u=\ell}^{L}\phi_u:D(\lambda)\to\Delta(\mu) ∑ u = ℓ L ϕ u : D ( λ ) → Δ ( μ ) induces a generator of Hom G ( Δ ( λ ) , Δ ( μ ) ) \operatorname{Hom}_{G}(\Delta(\lambda),\Delta(\mu)) Hom G ( Δ ( λ ) , Δ ( μ )) .
(2) Suppose ℓ < L \ell<L ℓ < L , λ 1 \lambda_1 λ 1 is even, and λ 2 , μ 2 \lambda_2,\mu_2 λ 2 , μ 2 have opposite parity. By Lemma 3.1, every vector in ker K [ A B ] \ker_{\mathbb{K}}\begin{bmatrix} A \\ B \end{bmatrix} ker K [ A B ] has all coordinates zero except possibly the one indexed by ℓ \ell ℓ . Hence Hom G ( Δ ( λ ) , Δ ( μ ) ) ≠ 0 \operatorname{Hom}_{G}(\Delta(\lambda),\Delta(\mu))\neq0 Hom G ( Δ ( λ ) , Δ ( μ )) = 0 if and only if column ℓ \ell ℓ of both matrices is zero. We consider two cases according to ℓ \ell ℓ .
(2a) Let ℓ = 0 \ell=0 ℓ = 0 . Then μ 2 ≤ λ 3 \mu_2\leq\lambda_3 μ 2 ≤ λ 3 by (2.2), and μ 2 ≥ 1 \mu_2\geq1 μ 2 ≥ 1 since ℓ < L \ell<L ℓ < L . Thus ( 1 , 1 ) ∈ R B (1,1)\in R_B ( 1 , 1 ) ∈ R B by (2.4), and
B ( 1 , 1 ) , 0 = ( 1 0 ) ( λ 2 0 ) = 1. B_{(1,1),0}=\binom{1}{0}\binom{\lambda_2}{0}=1. B ( 1 , 1 ) , 0 = ( 0 1 ) ( 0 λ 2 ) = 1.
Hence Hom G ( Δ ( λ ) , Δ ( μ ) ) = 0 \operatorname{Hom}_{G}(\Delta(\lambda),\Delta(\mu))=0 Hom G ( Δ ( λ ) , Δ ( μ )) = 0 in agreement with the criterion in (4.2), since λ 3 ≥ 2 ν 2 ( ℓ + 1 ) \lambda_3\geq2^{\nu_2(\ell+1)} λ 3 ≥ 2 ν 2 ( ℓ + 1 ) . (2b) Let ℓ > 0 \ell> 0 ℓ > 0 . Then ℓ = μ 2 − λ 3 \ell= \mu_{2} - \lambda_{3} ℓ = μ 2 − λ 3 .
Let ( t , k ) ∈ R A (t,k) \in R_{A} ( t , k ) ∈ R A . If k > ℓ k > \ell k > ℓ , then ( μ 2 − k ℓ − k ) = 0 \binom{\mu_{2}-k}{\ell-k} = 0 ( ℓ − k μ 2 − k ) = 0 , and A ( t , k ) , ℓ = 0 A_{(t,k),\ell} = 0 A ( t , k ) , ℓ = 0 by Definition 2.3. We conclude that column ℓ \ell ℓ of A A A is zero if and only if
A ( t , ℓ ) , ℓ = ( λ 1 − ℓ + t t ) = 0 , 1 ≤ t ≤ λ 2 − ℓ . A_{(t,\ell),\ell} = \binom{\lambda_{1}-\ell+t}{t} = 0,\quad1 \leq t \leq\lambda_{2} - \ell. A ( t , ℓ ) , ℓ = ( t λ 1 − ℓ + t ) = 0 , 1 ≤ t ≤ λ 2 − ℓ .
Similarly, if ( t , k ) ∈ R B (t,k) \in R_{B} ( t , k ) ∈ R B , then we have k ≥ ℓ + t k \geq\ell+ t k ≥ ℓ + t by (4) . If k > ℓ + t k > \ell+ t k > ℓ + t , then ( λ 2 + t − k t + ℓ − k ) = 0 \binom{\lambda_{2}+t-k}{t+\ell-k} = 0 ( t + ℓ − k λ 2 + t − k ) = 0 , so B ( t , k ) , ℓ = 0 B_{(t,k),\ell} = 0 B ( t , k ) , ℓ = 0 by Definition 2.3. We conclude that column ℓ \ell ℓ of B B B is zero if and only if
B ( t , ℓ + t ) , ℓ = ( ℓ + t t ) = 0 , 1 ≤ t ≤ λ 3 . B_{(t,\ell+t),\ell} = \binom{\ell+t}{t} = 0,\quad1 \leq t \leq\lambda_{3}. B ( t , ℓ + t ) , ℓ = ( t ℓ + t ) = 0 , 1 ≤ t ≤ λ 3 .
By Lemma 2.6, the above conditions are equivalent to λ 2 − ℓ < 2 v 2 ( λ 1 − ℓ + 1 ) \lambda_{2} - \ell< 2^{v_{2}(\lambda_{1}-\ell+1)} λ 2 − ℓ < 2 v 2 ( λ 1 − ℓ + 1 ) , λ 3 < 2 v 2 ( ℓ + 1 ) \lambda_{3} < 2^{v_{2}(\ell+1)} λ 3 < 2 v 2 ( ℓ + 1 ) , which is (18) . When these inequalities hold, Proposition 2.4 shows that ϕ ℓ : D ( λ ) → Δ ( μ ) \phi_{\ell}: D(\lambda) \to\Delta(\mu) ϕ ℓ : D ( λ ) → Δ ( μ ) induces a generator of Hom G ( Δ ( λ ) , Δ ( μ ) ) \operatorname{Hom}_{G}(\Delta(\lambda), \Delta(\mu)) Hom G ( Δ ( λ ) , Δ ( μ )) .
(3) From Proposition 3.7 and Proposition 2.4 we have the claimed isomorphism and generator of Hom G ( Δ ( λ ) , Δ ( μ ) ) \operatorname{Hom}_{G}(\Delta(\lambda), \Delta(\mu)) Hom G ( Δ ( λ ) , Δ ( μ )) .
It remains to prove that only a finite number of halvings can be applied starting with the pair ( λ , μ ) (\lambda,\mu) ( λ , μ ) . By Lemma 3.3(1),[2 ] , each application of (3) produces a pair satisfying (1) and ℓ ′ ≤ L ′ \ell' \leq L' ℓ ′ ≤ L ′ , with 0 ≤ λ 3 ′ < λ 3 0 \leq\lambda'_{3} < \lambda_{3} 0 ≤ λ 3 ′ < λ 3 .
Corollary 4.2. We have dim K Hom G ( Δ ( λ ) , Δ ( μ ) ) ≤ 1 \dim_{K} \operatorname{Hom}_{G}(\Delta(\lambda), \Delta(\mu)) \leq1 dim K Hom G ( Δ ( λ ) , Δ ( μ )) ≤ 1 for all dominant weights λ , μ \lambda,\mu λ , μ of G G G .
Proof. This follows from the reductions in § 2.1 and Theorem 4.1.
Example 4.3. Let q ≥ 2 q \geq2 q ≥ 2 be an integer and consider the partitions
λ = ( 8 q − 6 , 4 q + 1 , 4 q − 5 ) , μ = ( 12 q − 7 , 4 q − 3 ) . \lambda= (8q - 6, 4q + 1, 4q - 5),\quad\mu= (12q - 7, 4q - 3). λ = ( 8 q − 6 , 4 q + 1 , 4 q − 5 ) , μ = ( 12 q − 7 , 4 q − 3 ) . For this pair we have ℓ = 2 \ell= 2 ℓ = 2 , L = 4 q − 3 L = 4q - 3 L = 4 q − 3 . Thus ℓ < L \ell< L ℓ < L . Since λ 1 \lambda_{1} λ 1 is even and λ 2 , μ 2 \lambda_{2},\mu_{2} λ 2 , μ 2 are both odd, we apply Definition 3.2 with ε = 1 \varepsilon= 1 ε = 1 . This gives the pair ( λ ′ , μ ′ ) (\lambda',\mu') ( λ ′ , μ ′ ) , where
λ ′ = ( 4 q − 4 , 2 q , 2 q − 3 ) , μ ′ = ( 6 q − 5 , 2 q − 2 ) . \lambda' = (4q - 4, 2q, 2q - 3),\quad\mu' = (6q - 5, 2q - 2). λ ′ = ( 4 q − 4 , 2 q , 2 q − 3 ) , μ ′ = ( 6 q − 5 , 2 q − 2 ) . For the pair ( λ ′ , μ ′ ) (\lambda',\mu') ( λ ′ , μ ′ ) we have ℓ ′ = 1 \ell' = 1 ℓ ′ = 1 , L ′ = 2 q − 2 L' = 2q - 2 L ′ = 2 q − 2 . Again ℓ ′ < L ′ \ell' < L' ℓ ′ < L ′ . Now λ 1 ′ \lambda'_{1} λ 1 ′ is even and λ 2 ′ , μ 2 ′ \lambda'_{2},\mu'_{2} λ 2 ′ , μ 2 ′ are both even. We apply Definition 3.2 a second time, but with ε = 0 \varepsilon= 0 ε = 0 to obtain the pair ( λ ′ ′ , μ ′ ′ ) (\lambda'',\mu'') ( λ ′′ , μ ′′ ) , where
λ ′ ′ = ( 2 q − 3 , q − 1 , q − 2 ) , μ ′ ′ = ( 3 q − 4 , q − 2 ) . \lambda'' = (2q - 3, q - 1, q - 2),\quad\mu'' = (3q - 4, q - 2). λ ′′ = ( 2 q − 3 , q − 1 , q − 2 ) , μ ′′ = ( 3 q − 4 , q − 2 ) . For ( λ ′ ′ , μ ′ ′ ) (\lambda'',\mu'') ( λ ′′ , μ ′′ ) we have ℓ ′ ′ = 0 \ell'' = 0 ℓ ′′ = 0 and L ′ ′ = q − 2 L'' = q - 2 L ′′ = q − 2 . Since λ 1 ′ ′ = 2 q − 3 \lambda''_{1} = 2q - 3 λ 1 ′′ = 2 q − 3 is odd, we apply Theorem 4.1[1 ] to conclude Hom G ( Δ ( λ ′ ′ ) , Δ ( μ ′ ′ ) ) ≠ 0 \operatorname{Hom}_{G}(\Delta(\lambda''), \Delta(\mu'')) \ne0 Hom G ( Δ ( λ ′′ ) , Δ ( μ ′′ )) = 0 if and only if q − 1 < 2 v 2 ( q ) q - 1 < 2^{v_{2}(q)} q − 1 < 2 v 2 ( q ) and q − 2 < 2 v 2 ( q ) q - 2 < 2^{v_{2}(q)} q − 2 < 2 v 2 ( q ) . These inequalities hold if and only if q q q is a power of 2. By Theorem 4.1[3 ] , we conclude that
dim K Hom G ( Δ ( λ ) , Δ ( μ ) ) = { 1 , if q is a power of 2 , 0 , otherwise. \dim_{K} \operatorname{Hom}_{G}(\Delta(\lambda), \Delta(\mu)) =
\begin{cases}
1, & \text{if } q \text{ is a power of } 2,\\
0, & \text{otherwise.}
\end{cases} dim K Hom G ( Δ ( λ ) , Δ ( μ )) = { 1 , 0 , if q is a power of 2 , otherwise. Suppose now that q q q is a power of 2. We determine a generator of Hom G ( Δ ( λ ) , Δ ( μ ) ) \operatorname{Hom}_{G}(\Delta(\lambda), \Delta(\mu)) Hom G ( Δ ( λ ) , Δ ( μ )) as follows. Since λ ′ ′ = ( 2 q − 3 , q − 1 , q − 2 ) \lambda'' = (2q - 3, q - 1, q - 2) λ ′′ = ( 2 q − 3 , q − 1 , q − 2 ) and 2 q − 3 2q - 3 2 q − 3 is odd, by Theorem 4.1[1 ] a generator of Hom G ( Δ ( λ ′ ′ ) , Δ ( μ ′ ′ ) ) \operatorname{Hom}_{G}(\Delta(\lambda''), \Delta(\mu'')) Hom G ( Δ ( λ ′′ ) , Δ ( μ ′′ )) has coefficients c j ′ ′ = 1 c''_{j} = 1 c j ′′ = 1 , ( 0 ≤ j ≤ q − 2 ) (0 \leq j \leq q - 2) ( 0 ≤ j ≤ q − 2 ) .
For the pair ( λ ′ , μ ′ ) (\lambda',\mu') ( λ ′ , μ ′ ) we have ε = 0 \varepsilon= 0 ε = 0 . By (16) a generator of Hom G ( Δ ( λ ′ ) , Δ ( μ ′ ) ) \operatorname{Hom}_{G}(\Delta(\lambda'), \Delta(\mu')) Hom G ( Δ ( λ ′ ) , Δ ( μ ′ )) has coefficients c 2 j + 1 ′ = c j ′ ′ = 1 c'_{2j+1} = c''_{j} = 1 c 2 j + 1 ′ = c j ′′ = 1 , c 2 j + 2 ′ = 0 c'_{2j+2} = 0 c 2 j + 2 ′ = 0 ( 0 ≤ j ≤ q − 2 ) (0 \leq j \leq q - 2) ( 0 ≤ j ≤ q − 2 ) .
For the pair ( λ , μ ) (\lambda,\mu) ( λ , μ ) we have ε = 1 \varepsilon= 1 ε = 1 . By (16) a generator of Hom G ( Δ ( λ ) , Δ ( μ ) ) \operatorname{Hom}_{G}(\Delta(\lambda), \Delta(\mu)) Hom G ( Δ ( λ ) , Δ ( μ )) has coefficients c 2 i = c 2 i + 1 = c i ′ c_{2i} = c_{2i+1} = c'_{i} c 2 i = c 2 i + 1 = c i ′ ( 1 ≤ i ≤ 2 q − 2 ) (1 \leq i \leq2q - 2) ( 1 ≤ i ≤ 2 q − 2 ) . In other words, a generator of Hom G ( Δ ( λ ) , Δ ( μ ) ) \operatorname{Hom}_{G}(\Delta(\lambda), \Delta(\mu)) Hom G ( Δ ( λ ) , Δ ( μ )) is induced by the map ∑ j = 0 q − 2 ( ϕ 4 j + 2 + ϕ 4 j + 3 ) : D ( λ ) → Δ ( μ ) \sum_{j=0}^{q-2}(\phi_{4j+2}+\phi_{4j+3}): D(\lambda) \to\Delta(\mu) ∑ j = 0 q − 2 ( ϕ 4 j + 2 + ϕ 4 j + 3 ) : D ( λ ) → Δ ( μ ) .
Homomorphisms between Specht modules
For a partition ν \nu ν of r r r , let S ν S^{\nu} S ν be the corresponding Specht module of the symmetric group S r \mathfrak{S}_r S r . Recall that ν \nu ν is called 2-regular if its positive parts are distinct.
Corollary 4.4. Let λ , μ \lambda,\mu λ , μ be partitions of an integer r ≥ 3 r \ge3 r ≥ 3 , each with at most three parts. Suppose that μ \mu μ is 2-regular, or that μ 1 = μ 2 ≥ μ 3 + 2 \mu_1 = \mu_2 \ge\mu_3 + 2 μ 1 = μ 2 ≥ μ 3 + 2 . Then
dim K Hom K S r ( S μ , S λ ) = dim G L 3 ( K ) Hom ( Δ ( λ ) , Δ ( μ ) ) . \dim_{\mathbb{K}}\operatorname{Hom}_{\mathbb{K}\mathfrak{S}_r}(S^{\mu},S^{\lambda})=\dim_{\mathop{\rm GL}\nolimits_3(\mathbb{K})}\operatorname{Hom}(\Delta(\lambda),\Delta(\mu)). dim K Hom K S r ( S μ , S λ ) = dim GL 3 ( K ) Hom ( Δ ( λ ) , Δ ( μ )) . We note that dim K Hom G L 3 ( K ) ( Δ ( λ ) , Δ ( μ ) ) \dim_{\mathbb{K}}\operatorname{Hom}_{\mathop{\rm GL}\nolimits_3(\mathbb{K})}(\Delta(\lambda),\Delta(\mu)) dim K Hom GL 3 ( K ) ( Δ ( λ ) , Δ ( μ )) is at most one and is determined by Theorem 4.1, after the reductions of § 2.1 .
Proof. Let d ≥ r d \ge r d ≥ r and let f f f be the Schur functor from the category of homogeneous polynomial representations of G L d ( K ) \mathop{\rm GL}\nolimits_d(\mathbb{K}) GL d ( K ) of degree r r r to the category of K S r \mathbb{K}\mathfrak{S}_r K S r -modules [11 Section 6.1] . Let G ⊗ G_{\otimes} G ⊗ be the left adjoint of f f f [18 Definition 3.3] . Let Δ d ( ν ) \Delta_d(\nu) Δ d ( ν ) and ∇ d ( ν ) \nabla_d(\nu) ∇ d ( ν ) be the Weyl and dual Weyl modules for G L d ( K ) \mathop{\rm GL}\nolimits_d(\mathbb{K}) GL d ( K ) . In the second case, the partition ( μ 2 , μ 3 ) (\mu_2,\mu_3) ( μ 2 , μ 3 ) is 2-regular, since μ 2 ≥ μ 3 + 2 \mu_2 \ge\mu_3 + 2 μ 2 ≥ μ 3 + 2 . Thus, in both cases, G ⊗ ( S μ ) ≅ ∇ d ( μ ) G_{\otimes}(S^{\mu}) \cong\nabla_d(\mu) G ⊗ ( S μ ) ≅ ∇ d ( μ ) by [18 Theorem 1.2] . We also have f ( ∇ d ( λ ) ) ≅ S λ f(\nabla_d(\lambda)) \cong S^{\lambda} f ( ∇ d ( λ )) ≅ S λ [11 Section 6.3] . Hence Hom K S r ( S μ , S λ ) ≅ Hom G L d ( K ) ( ∇ d ( μ ) , ∇ d ( λ ) ) ≅ Hom G L d ( K ) ( Δ d ( λ ) , Δ d ( μ ) ) \operatorname{Hom}_{\mathbb{K}\mathfrak{S}_r}(S^{\mu},S^{\lambda}) \cong\operatorname{Hom}_{\mathop{\rm GL}\nolimits_d(\mathbb{K})}(\nabla_d(\mu),\nabla_d(\lambda)) \cong\operatorname{Hom}_{\mathop{\rm GL}\nolimits_d(\mathbb{K})}(\Delta_d(\lambda),\Delta_d(\mu)) Hom K S r ( S μ , S λ ) ≅ Hom GL d ( K ) ( ∇ d ( μ ) , ∇ d ( λ )) ≅ Hom GL d ( K ) ( Δ d ( λ ) , Δ d ( μ )) , where we used contravariant duality in the second isomorphism. But for d ≥ 3 d \ge3 d ≥ 3 and partitions with at most three parts, it is well known that Hom G L d ( K ) ( Δ d ( λ ) , Δ d ( μ ) ) ≅ Hom G L 3 ( K ) ( Δ 3 ( λ ) , Δ 3 ( μ ) ) \operatorname{Hom}_{\mathop{\rm GL}\nolimits_d(\mathbb{K})}(\Delta_d(\lambda),\Delta_d(\mu)) \cong\operatorname{Hom}_{\mathop{\rm GL}\nolimits_3(\mathbb{K})}(\Delta_3(\lambda),\Delta_3(\mu)) Hom GL d ( K ) ( Δ d ( λ ) , Δ d ( μ )) ≅ Hom GL 3 ( K ) ( Δ 3 ( λ ) , Δ 3 ( μ )) , see, for example, the proof of [15 Proposition 1.1] . □
Proofs of Lemmas 3.3 and 3.4
In this section, we prove Lemma 3.3 and Lemma 3.4
We adopt throughout the assumptions and notation of Definition 3.2 .
Proof of Lemma 3.3
(1) By (2) and the assumption ℓ < L \ell< L ℓ < L , we have
0 ≤ ℓ < L ≤ μ 2 , μ 2 − λ 3 ≤ ℓ < L ≤ μ 2 . 0 \le\ell< L \le\mu_2,\qquad\mu_2-\lambda_3 \le\ell< L \le\mu_2. 0 ≤ ℓ < L ≤ μ 2 , μ 2 − λ 3 ≤ ℓ < L ≤ μ 2 .
Hence μ 2 , λ 3 ≥ 1 \mu_2,\lambda_3 \ge1 μ 2 , λ 3 ≥ 1 , so Definition 3.2 gives μ 2 ′ , λ 3 ′ ≥ 0 \mu'_2,\lambda'_3 \ge0 μ 2 ′ , λ 3 ′ ≥ 0 .
We first show that λ ′ \lambda' λ ′ is a partition. Since λ 1 \lambda_1 λ 1 is even and λ 2 , μ 2 ≤ λ 1 \lambda_2,\mu_2 \le\lambda_1 λ 2 , μ 2 ≤ λ 1 , we have λ 2 ′ , μ 2 ′ ≤ ⌊ ( λ 1 − 1 ) / 2 ⌋ = λ 1 2 − 1 = λ 1 ′ \lambda'_2,\mu'_2 \le\lfloor(\lambda_1-1)/2\rfloor=\frac{\lambda_1}{2}-1=\lambda'_1 λ 2 ′ , μ 2 ′ ≤ ⌊( λ 1 − 1 ) /2 ⌋ = 2 λ 1 − 1 = λ 1 ′ . Moreover, since λ 3 ≤ λ 2 \lambda_3 \le\lambda_2 λ 3 ≤ λ 2 and λ 2 ≡ ε ( m o d 2 ) \lambda_2 \equiv\varepsilon\pmod2 λ 2 ≡ ε ( mod 2 ) , we have
λ 3 ′ = ⌊ ( λ 3 − 1 + ε ) / 2 ⌋ ≤ ⌊ ( λ 2 − 1 + ε ) / 2 ⌋ = ⌊ ( λ 2 − 1 ) / 2 ⌋ = λ 2 ′ . \lambda'_3=\lfloor(\lambda_3-1+\varepsilon)/2\rfloor\le\lfloor(\lambda_2-1+\varepsilon)/2\rfloor=\lfloor(\lambda_2-1)/2\rfloor=\lambda'_2. λ 3 ′ = ⌊( λ 3 − 1 + ε ) /2 ⌋ ≤ ⌊( λ 2 − 1 + ε ) /2 ⌋ = ⌊( λ 2 − 1 ) /2 ⌋ = λ 2 ′ .
Thus λ 1 ′ ≥ λ 2 ′ ≥ λ 3 ′ ≥ 0 \lambda'_1 \ge\lambda'_2 \ge\lambda'_3 \ge0 λ 1 ′ ≥ λ 2 ′ ≥ λ 3 ′ ≥ 0 , so λ ′ \lambda' λ ′ is a partition. We also have 0 ≤ μ 2 ′ ≤ λ 1 ′ 0 \le\mu'_2 \le\lambda'_1 0 ≤ μ 2 ′ ≤ λ 1 ′ .
We now show that μ ′ \mu' μ ′ is a partition. By (2) and ℓ < L \ell< L ℓ < L we get μ 2 − λ 3 ≤ ℓ < L ≤ λ 2 \mu_2-\lambda_3 \le\ell< L \le\lambda_2 μ 2 − λ 3 ≤ ℓ < L ≤ λ 2 , so μ 2 − λ 2 < λ 3 \mu_2-\lambda_2 < \lambda_3 μ 2 − λ 2 < λ 3 . Since λ 2 ≡ μ 2 ( m o d 2 ) \lambda_2 \equiv\mu_2\pmod2 λ 2 ≡ μ 2 ( mod 2 ) , the formulas in Definition 3.2 give 2 ( μ 2 ′ − λ 2 ′ ) = μ 2 − λ 2 2(\mu'_2-\lambda'_2)=\mu_2-\lambda_2 2 ( μ 2 ′ − λ 2 ′ ) = μ 2 − λ 2 . Also, the formula for λ 3 ′ \lambda'_3 λ 3 ′ gives λ 3 ≤ 2 λ 3 ′ + 2 \lambda_3 \le2\lambda'_3+2 λ 3 ≤ 2 λ 3 ′ + 2 . Hence 2 ( μ 2 ′ − λ 2 ′ ) = μ 2 − λ 2 < λ 3 ≤ 2 λ 3 ′ + 2 2(\mu'_2-\lambda'_2)=\mu_2-\lambda_2<\lambda_3\le2\lambda'_3+2 2 ( μ 2 ′ − λ 2 ′ ) = μ 2 − λ 2 < λ 3 ≤ 2 λ 3 ′ + 2 . Since μ 2 ′ − λ 2 ′ \mu'_2-\lambda'_2 μ 2 ′ − λ 2 ′ and λ 3 ′ \lambda'_3 λ 3 ′ are integers, we obtain μ 2 ′ − λ 2 ′ ≤ λ 3 ′ \mu'_2-\lambda'_2\le\lambda'_3 μ 2 ′ − λ 2 ′ ≤ λ 3 ′ . Therefore,
μ 1 ′ = λ 1 ′ + λ 2 ′ + λ 3 ′ − μ 2 ′ ≥ λ 1 ′ ≥ μ 2 ′ \mu'_1=\lambda'_1+\lambda'_2+\lambda'_3-\mu'_2\ge\lambda'_1\ge\mu'_2 μ 1 ′ = λ 1 ′ + λ 2 ′ + λ 3 ′ − μ 2 ′ ≥ λ 1 ′ ≥ μ 2 ′
and μ ′ \mu' μ ′ is a partition. We have ∣ λ ′ ∣ = ∣ μ ′ ∣ |\lambda'|=|\mu'| ∣ λ ′ ∣ = ∣ μ ′ ∣ by the definition of μ 1 ′ \mu'_1 μ 1 ′ .
Finally, since λ 3 ≥ 1 \lambda_3 \ge1 λ 3 ≥ 1 and ε ∈ { 0 , 1 } \varepsilon\in\{0,1\} ε ∈ { 0 , 1 } , Definition 3.2 gives 0 ≤ λ 3 ′ = ⌊ ( λ 3 − 1 + ε ) / 2 ⌋ ≤ λ 3 2 < λ 3 0 \le\lambda'_3=\lfloor(\lambda_3-1+\varepsilon)/2\rfloor\le\frac{\lambda_3}{2}<\lambda_3 0 ≤ λ 3 ′ = ⌊( λ 3 − 1 + ε ) /2 ⌋ ≤ 2 λ 3 < λ 3 .
(2) Since μ 2 ≡ ε ( m o d 2 ) \mu_2\equiv\varepsilon\pmod2 μ 2 ≡ ε ( mod 2 ) , the formulas for μ 2 ′ \mu'_2 μ 2 ′ and λ 3 ′ \lambda'_3 λ 3 ′ in Definition 3.2 give μ 2 ′ − λ 3 ′ = ⌊ ( μ 2 − λ 3 ) / 2 ⌋ \mu'_2-\lambda'_3=\lfloor(\mu_2-\lambda_3)/2\rfloor μ 2 ′ − λ 3 ′ = ⌊( μ 2 − λ 3 ) /2 ⌋ . Thus ℓ ′ = max ( 0 , μ 2 ′ − λ 3 ′ ) = ⌊ ℓ / 2 ⌋ \ell'=\max(0,\mu'_2-\lambda'_3)=\lfloor\ell/2\rfloor ℓ ′ = max ( 0 , μ 2 ′ − λ 3 ′ ) = ⌊ ℓ /2 ⌋ . Similarly, L ′ = ⌊ ( L − 1 ) / 2 ⌋ L'=\lfloor(L-1)/2\rfloor L ′ = ⌊( L − 1 ) /2 ⌋ . Since ℓ < L \ell<L ℓ < L , we have ℓ ′ ≤ L ′ \ell'\le L' ℓ ′ ≤ L ′ .
(3) For every integer i i i , part (2) gives
ℓ ≤ 2 i + 1 ≤ L ⟺ ⌊ ℓ / 2 ⌋ ≤ i ≤ ⌊ ( L − 1 ) / 2 ⌋ ⟺ ℓ ′ ≤ i ≤ L ′ . \ell\le2i+1\le L\quad\Longleftrightarrow\quad\lfloor\ell/2\rfloor\le i\le\lfloor(L-1)/2\rfloor\quad\Longleftrightarrow\quad\ell'\le i\le L'. ℓ ≤ 2 i + 1 ≤ L ⟺ ⌊ ℓ /2 ⌋ ≤ i ≤ ⌊( L − 1 ) /2 ⌋ ⟺ ℓ ′ ≤ i ≤ L ′ .
Proof of Lemma 3.4
(1) Let h ≥ 1 h \ge1 h ≥ 1 and j ≥ 0 j \ge0 j ≥ 0 .
(1A) We first consider R A R_A R A . Suppose ( 2 h , 2 j + 1 ) ∈ R A (2h, 2j+1) \in R_A ( 2 h , 2 j + 1 ) ∈ R A . By (3) , ℓ ≤ 2 j + 1 ≤ min ( λ 2 − 2 h , μ 2 ) \ell\le2j+1 \le\min(\lambda_2-2h,\mu_2) ℓ ≤ 2 j + 1 ≤ min ( λ 2 − 2 h , μ 2 ) . By Lemma 3.3(2) and Definition 3.2, these inequalities give
ℓ ′ ≤ j ≤ min ( λ 2 ′ − h , μ 2 ′ ) . (21) \ell' \le j \le\min(\lambda'_2-h,\mu'_2). \tag*{(21)} ℓ ′ ≤ j ≤ min ( λ 2 ′ − h , μ 2 ′ ) . (21)
Thus ( h , j ) ∈ R A ′ (h,j) \in R_{A'} ( h , j ) ∈ R A ′ by (3) .
Conversely, suppose ( h , j ) ∈ R A ′ (h,j) \in R_{A'} ( h , j ) ∈ R A ′ . Then we have 1 ≤ h ≤ λ 2 ′ 1 \le h \le\lambda'_2 1 ≤ h ≤ λ 2 ′ and (21) . Lemma 3.3(2) and Definition 3.2 yield
ℓ ≤ 2 j + 1 ≤ min ( λ 2 − 2 h , μ 2 ) . \ell\le2j+1 \le\min(\lambda_2-2h,\mu_2). ℓ ≤ 2 j + 1 ≤ min ( λ 2 − 2 h , μ 2 ) .
Moreover, Definition 3.2 gives 1 ≤ 2 h ≤ λ 2 1 \le2h \le\lambda_2 1 ≤ 2 h ≤ λ 2 . Hence ( 2 h , 2 j + 1 ) ∈ R A (2h,2j+1) \in R_A ( 2 h , 2 j + 1 ) ∈ R A .
(1B) We next consider R B R_B R B . Suppose ( 2 h , 2 j + 1 ) ∈ R B (2h,2j+1) \in R_B ( 2 h , 2 j + 1 ) ∈ R B . By (4) ,
max ( 0 , μ 2 − λ 3 + 2 h ) ≤ 2 j + 1 ≤ min ( λ 2 + 2 h , μ 2 ) . \max(0,\mu_2-\lambda_3+2h) \le2j+1 \le\min(\lambda_2+2h,\mu_2). max ( 0 , μ 2 − λ 3 + 2 h ) ≤ 2 j + 1 ≤ min ( λ 2 + 2 h , μ 2 ) .
Using Definition 3.2 we see that these inequalities give
max ( 0 , μ 2 ′ − λ 3 ′ + h ) ≤ j ≤ min ( λ 2 ′ + h , μ 2 ′ ) . \max(0,\mu'_2-\lambda'_3+h) \le j \le\min(\lambda'_2+h,\mu'_2). max ( 0 , μ 2 ′ − λ 3 ′ + h ) ≤ j ≤ min ( λ 2 ′ + h , μ 2 ′ ) .
In particular, μ 2 ′ − λ 3 ′ + h ≤ j ≤ μ 2 ′ \mu'_2-\lambda'_3+h \le j \le\mu'_2 μ 2 ′ − λ 3 ′ + h ≤ j ≤ μ 2 ′ , so 1 ≤ h ≤ λ 3 ′ 1 \le h \le\lambda'_3 1 ≤ h ≤ λ 3 ′ . Thus ( h , j ) ∈ R B ′ (h,j) \in R_{B'} ( h , j ) ∈ R B ′ by (4) .
Conversely, suppose ( h , j ) ∈ R B ′ (h,j) \in R_{B'} ( h , j ) ∈ R B ′ . Then
max ( 0 , μ 2 ′ − λ 3 ′ + h ) ≤ j ≤ min ( λ 2 ′ + h , μ 2 ′ ) . (22) \max(0,\mu'_2-\lambda'_3+h) \le j \le\min(\lambda'_2+h,\mu'_2). \tag*{(22)} max ( 0 , μ 2 ′ − λ 3 ′ + h ) ≤ j ≤ min ( λ 2 ′ + h , μ 2 ′ ) . (22)
By (22) and Definition 3.2 we get max ( 0 , μ 2 − λ 3 + 2 h ) ≤ 2 j + 1 ≤ min ( λ 2 + 2 h , μ 2 ) \max(0,\mu_2-\lambda_3+2h) \le2j+1 \le\min(\lambda_2+2h,\mu_2) max ( 0 , μ 2 − λ 3 + 2 h ) ≤ 2 j + 1 ≤ min ( λ 2 + 2 h , μ 2 ) . Moreover, Definition 3.2 gives 1 ≤ 2 h ≤ λ 3 1 \le2h \le\lambda_3 1 ≤ 2 h ≤ λ 3 . Hence ( 2 h , 2 j + 1 ) ∈ R B (2h,2j+1) \in R_B ( 2 h , 2 j + 1 ) ∈ R B .
(2) Suppose ε = 1 \varepsilon=1 ε = 1 , so that λ 2 \lambda_2 λ 2 and μ 2 \mu_2 μ 2 are odd. Suppose ( 2 h , 2 j ) ∈ R A (2h,2j) \in R_A ( 2 h , 2 j ) ∈ R A . (3) implies ℓ ≤ 2 j ≤ min ( λ 2 − 2 h , μ 2 ) \ell\le2j \le\min(\lambda_2-2h,\mu_2) ℓ ≤ 2 j ≤ min ( λ 2 − 2 h , μ 2 ) . Since this upper bound is odd,
ℓ ≤ 2 j + 1 ≤ min ( λ 2 − 2 h , μ 2 ) . \ell\le2j+1 \le\min(\lambda_2-2h,\mu_2). ℓ ≤ 2 j + 1 ≤ min ( λ 2 − 2 h , μ 2 ) .
Thus ( 2 h , 2 j + 1 ) ∈ R A (2h,2j+1) \in R_A ( 2 h , 2 j + 1 ) ∈ R A . By part (1) of the lemma we have ( h , j ) ∈ R A ′ (h,j) \in R_{A'} ( h , j ) ∈ R A ′ .
Finally, suppose ( 2 h , 2 j ) ∈ R B (2h,2j) \in R_B ( 2 h , 2 j ) ∈ R B . By (4) we have max ( 0 , μ 2 − λ 3 + 2 h ) ≤ 2 j ≤ min ( λ 2 + 2 h , μ 2 ) \max(0,\mu_2-\lambda_3+2h) \le2j \le\min(\lambda_2+2h,\mu_2) max ( 0 , μ 2 − λ 3 + 2 h ) ≤ 2 j ≤ min ( λ 2 + 2 h , μ 2 ) . Again the upper bound is odd, and thus max ( 0 , μ 2 − λ 3 + 2 h ) ≤ 2 j + 1 ≤ min ( λ 2 + 2 h , μ 2 ) \max(0,\mu_2-\lambda_3+2h) \le2j+1 \le\min(\lambda_2+2h,\mu_2) max ( 0 , μ 2 − λ 3 + 2 h ) ≤ 2 j + 1 ≤ min ( λ 2 + 2 h , μ 2 ) . Hence ( 2 h , 2 j + 1 ) ∈ R B (2h,2j+1) \in R_B ( 2 h , 2 j + 1 ) ∈ R B . From part (1) of the lemma we conclude that ( h , j ) ∈ R B ′ (h,j) \in R_{B'} ( h , j ) ∈ R B ′ .
Department of Mathematics, National and Kapodistrian University of Athens, Athens, Greece Email address: mmaliak@math.uoa.gr
Department of Mathematics, University of Thessaly, Lamia, Greece Email address: dstergiop@math.uoa.gr Email address: dstergiopoulou@uth.gr